\input texinfo
@c %**start of header
@setfilename R-exts.info
@settitle Writing R Extensions
@setchapternewpage on
@c %**end of header

@c @documentencoding ISO-8859-1

@c Put the functions in the variable index
@syncodeindex fn vr

@dircategory Programming
@direntry
* R Extensions: (R-exts).      Writing R Extensions.
@end direntry

@finalout

@include R-defs.texi
@include version.texi

@copying
This manual is for R, version @value{VERSION}.

@Rcopyright{1999}

@quotation
@permission{}
@end quotation
@end copying

@titlepage
@title Writing R Extensions
@subtitle Version @value{VERSION}
@author R Core Team
@page
@vskip 0pt plus 1filll
@insertcopying
@end titlepage

@ifplaintext
@insertcopying
@end ifplaintext

@c @ifnothtml
@contents
@c @end ifnothtml

@ifnottex
@node Top, Acknowledgements, (dir), (dir)
@top Writing R Extensions

This is a guide to extending @R{}, describing the process of creating
@R{} add-on packages, writing @R{} documentation, @R{}'s system and
foreign language interfaces, and the @R{} @acronym{API}.

@insertcopying

@end ifnottex

@menu
* Acknowledgements::
* Creating R packages::
* Writing R documentation files::
* Tidying and profiling R code::
* Debugging::
* System and foreign language interfaces::
* The R API::
* Generic functions and methods::
* Linking GUIs and other front-ends to R::
* Function and variable index::
* Concept index::
@end menu

@node Acknowledgements, Creating R packages, Top, Top
@unnumbered Acknowledgements


The contributions to early versions of this manual by Saikat DebRoy
(who wrote the first draft of a guide to using @code{.Call} and
@code{.External}) and Adrian Trapletti (who provided information on the
C++ interface) are gratefully acknowledged.

@node Creating R packages, Writing R documentation files, Acknowledgements, Top
@chapter Creating R packages
@cindex Packages
@cindex Creating packages

Packages provide a mechanism for loading optional code, data and
documentation as needed.  The @R{} distribution itself includes about 30
packages.

In the following, we assume that you know the @code{library()} command,
including its @code{lib.loc} argument, and we also assume basic
knowledge of the @command{R CMD INSTALL} utility.  Otherwise, please
look at @R{}'s help pages on

@example
?library
?INSTALL
@end example

@noindent
before reading on.

For packages which contain code to be compiled, a computing environment
including a number of tools is assumed; the ``R Installation and
Administration'' manual describes what is needed for each OS.

Once a source package is created, it must be installed by
the command @code{R CMD INSTALL}.
@ifset UseExternalXrefs
@xref{Add-on packages, , Add-on-packages,
R-admin, R Installation and Administration}.
@end ifset

Other types of extensions are supported (but rare): @xref{Package types}.

Some notes on terminology complete this introduction.  These will help
with the reading of this manual, and also in describing concepts
accurately when asking for help.

A @emph{package} is a directory of files which extend @R{}, a
@emph{source package} (the master files of a package), or a tarball
containing the files of a source package, or an @emph{installed}
package, the result of running @command{R CMD INSTALL} on a source
package.  On some platforms (notably macOS and Windows) there are also
@emph{binary packages}, a zip file or tarball containing the files of an
installed package which can be unpacked rather than installing from
sources.

A package is @strong{not}@footnote{although this is a persistent
mis-usage.  It seems to stem from S, whose analogues of @R{}'s packages
were officially known as @emph{library sections} and later as
@emph{chapters}, but almost always referred to as @emph{libraries}.} a
@emph{library}.  The latter is used in two senses in @R{} documentation.

@itemize

@item
A directory into which packages are installed, e.g.@:
@file{/usr/lib/R/library}: in that sense it is sometimes referred to as
a @emph{library directory} or @emph{library tree} (since the library is
a directory which contains packages as directories, which themselves
contain directories).

@item
That used by the operating system, as a shared, dynamic or static
library or (especially on Windows) a DLL, where the second L stands for
`library'.  Installed packages may contain compiled code in what is
known on Unix-alikes as a @emph{shared object} and on Windows as a DLL.
The concept of a @emph{shared library} (@emph{dynamic library} on macOS)
as a collection of compiled code to which a package might link is also
used, especially for @R{} itself on some platforms. On most platforms
these concepts are interchangeable (shared objects and DLLs can both be
loaded into the @R{} process and be linked against), but macOS
distinguishes between shared objects (extension @file{.so}) and dynamic
libraries (extension @file{.dylib}).

@end itemize

There are a number of well-defined operations on source packages.

@itemize

@item
The most common is @emph{installation} which takes a source package and
installs it in a library using @command{R CMD INSTALL} or
@code{install.packages}.

@item
Source packages can be @emph{built}.  This involves taking a source
directory and creating a tarball ready for distribution, including
cleaning it up and creating PDF documentation from any @emph{vignettes}
it may contain.  Source packages (and most often tarballs) can be
@emph{checked}, when a test installation is done and tested (including
running its examples); also, the contents of the package are tested in
various ways for consistency and portability.

@item
@emph{Compilation} is not a correct term for a package.  Installing a
source package which contains C, C++ or Fortran code will involve
compiling that code.  There is also the possibility of `byte' compiling
the @R{} code in a package (using the facilities of package
@pkg{compiler}): nowadays this is enabled by default for all
packages. So @emph{compiling} a package may come to mean byte-compiling
its @R{} code.

@item
It used to be unambiguous to talk about @emph{loading} an installed
package using @code{library()}, but since the advent of package
namespaces this has been less clear: people now often talk about
@emph{loading} the package's namespace and then @emph{attaching} the
package so it becomes visible on the search path.  Function
@code{library} performs both steps, but a package's namespace can be
loaded without the package being attached (for example by calls like
@code{splines::ns}).

@end itemize

The concept of @emph{lazy loading} of code or data is mentioned at
several points.  This is part of the installation, always selected for
@R{} code but optional for data.  When used the @R{} objects of the
package are created at installation time and stored in a database in the
@file{R} directory of the installed package, being loaded into the
session at first use.  This makes the @R{} session start up faster and
use less (virtual) memory.
@ifset UseExternalXrefs
(For technical details,
@pxref{Lazy loading, , Lazy loading, R-ints,  R Internals}.)
@end ifset

@cindex CRAN
@acronym{CRAN} is a network of WWW sites holding the @R{} distributions
and contributed code, especially @R{} packages.  Users of @R{} are
encouraged to join in the collaborative project and to submit their own
packages to @acronym{CRAN}: current instructions are linked from
@uref{https://CRAN.R-project.org/banner.shtml#submitting}.


@menu
* Package structure::
* Configure and cleanup::
* Checking and building packages::
* Writing package vignettes::
* Package namespaces::
* Writing portable packages::
* Diagnostic messages::
* Internationalization::
* CITATION files::
* Package types::
* Services::
@end menu

@node Package structure, Configure and cleanup, Creating R packages, Creating R packages
@section Package structure
@cindex Package structure

The sources of an @R{} package consist of a subdirectory containing the
files @file{DESCRIPTION} and @file{NAMESPACE}, and the subdirectories
@file{R}, @file{data}, @file{demo}, @file{exec}, @file{inst},
@file{man}, @file{po}, @file{src}, @file{tests}, @file{tools} and
@file{vignettes} (some of which can be missing, but which should not be
empty).  The package subdirectory may also contain files @file{INDEX},
@file{configure}, @file{cleanup}, @file{LICENSE}, @file{LICENCE} and
@file{NEWS}.  Other files such as @file{INSTALL} (for non-standard
installation instructions), @file{README}/@file{README.md}@footnote{This
seems to be commonly used for a file in `markdown' format.  Be aware
that most users of @R{} will not know that, nor know how to view such a
file: platforms such as macOS and Windows do not have a default viewer
set in their file associations.  The @acronym{CRAN} package web pages
render such files in @HTML{}: the converter used expects the file to be
encoded in UTF-8.}, or @file{ChangeLog} will be ignored by @R{}, but may
be useful to end users.  The utility @command{R CMD build} may add files
in a @file{build} directory (but this should not be used for other
purposes).

Except where specifically mentioned,@footnote{currently, top-level files
@file{.Rbuildignore} and @file{.Rinstignore}, and
@file{vignettes/.install_extras}.} packages should not contain
Unix-style `hidden' files/directories (that is, those whose name starts
with a dot).

The @file{DESCRIPTION} and @file{INDEX} files are described in the
subsections below.  The @file{NAMESPACE} file is described in the
section on @ref{Package namespaces}.

@cindex configure file
@cindex cleanup file

The optional files @file{configure} and @file{cleanup} are (Bourne)
shell scripts which are, respectively, executed before and (if option
@option{--clean} was given) after installation on Unix-alikes, see
@ref{Configure and cleanup}.  The analogues on Windows are
@file{configure.win} and @file{cleanup.win}. Since @R{} 4.2.0 on Windows,
@file{configure.ucrt} and @file{cleanup.ucrt} are supported and take
precedence over @file{configure.win} and @file{cleanup.win}. They can
hence be used to provide content specific for UCRT or Rtools42, if needed,
but the support for @file{.ucrt} files may be removed in the future when
building packages from source on the older versions of R will no longer
be needed, and hence the files may be renamed back to @file{.win}.

For the conventions for files @file{NEWS} and @file{ChangeLog} in the
@acronym{GNU} project see
@uref{https://www.gnu.org/prep/standards/standards.html#Documentation}.

The package subdirectory should be given the same name as the package.
Because some file systems (e.g., those on Windows and by default on
macOS) are not case-sensitive, to maintain portability it is strongly
recommended that case distinctions not be used to distinguish different
packages.  For example, if you have a package named @file{foo}, do not
also create a package named @file{Foo}.

To ensure that file names are valid across file systems and supported
operating systems, the @acronym{ASCII} control characters as well as the
characters @samp{"}, @samp{*}, @samp{:}, @samp{/}, @samp{<}, @samp{>},
@samp{?}, @samp{\}, and @samp{|} are not allowed in file names.  In
addition, files with names @samp{con}, @samp{prn}, @samp{aux},
@samp{clock$}, @samp{nul}, @samp{com1} to @samp{com9}, and @samp{lpt1}
to @samp{lpt9} after conversion to lower case and stripping possible
``extensions'' (e.g., @samp{lpt5.foo.bar}), are disallowed.  Also, file
names in the same directory must not differ only by case (see the
previous paragraph).  In addition, the basenames of @samp{.Rd} files may
be used in URLs and so must be @acronym{ASCII} and not contain @code{%}.
For maximal portability filenames should only contain only
@acronym{ASCII} characters not excluded already (that is
@code{A-Za-z0-9._!#$%&+,;=@@^()@{@}'[]} --- we exclude space as many
utilities do not accept spaces in file paths): non-English alphabetic
characters cannot be guaranteed to be supported in all locales.  It
would be good practice to avoid the shell metacharacters
@code{()@{@}'[]$~}: @code{~} is also used as part of `8.3' filenames on
Windows.  In addition, packages are normally distributed as tarballs,
and these have a limit on path lengths: for maximal portability 100
bytes.

A source package if possible should not contain binary executable files:
they are not portable, and a security risk if they are of the
appropriate architecture.  @command{R CMD check} will warn about
them@footnote{false positives are possible, but only a handful have been
seen so far.} unless they are listed (one filepath per line) in a file
@file{BinaryFiles} at the top level of the package.  Note that
@acronym{CRAN} will not accept submissions containing binary files
even if they are listed.

The @R{} function @code{package.skeleton} can help to create the
structure for a new package: see its help page for details.

@menu
* The DESCRIPTION file::
* Licensing::
* Package Dependencies::
* The INDEX file::
* Package subdirectories::
* Data in packages::
* Non-R scripts in packages::
* Specifying URLs::
@end menu

@node The DESCRIPTION file, Licensing, Package structure, Package structure
@subsection The @file{DESCRIPTION} file
@cindex DESCRIPTION file

The @file{DESCRIPTION} file contains basic information about the package
in the following format:

@quotation
@cartouche
@smallexample
Package: pkgname
Version: 0.5-1
Date: 2015-01-01
Title: My First Collection of Functions
Authors@@R: c(person("Joe", "Developer", role = c("aut", "cre"),
                     email = "Joe.Developer@@some.domain.net"),
              person("Pat", "Developer", role = "aut"),
              person("A.", "User", role = "ctb",
                     email = "A.User@@whereever.net"))
Author: Joe Developer [aut, cre],
  Pat Developer [aut],
  A. User [ctb]
Maintainer: Joe Developer <Joe.Developer@@some.domain.net>
Depends: R (>= 3.1.0), nlme
Suggests: MASS
Description: A (one paragraph) description of what
  the package does and why it may be useful.
License: GPL (>= 2)
URL: https://www.r-project.org, http://www.another.url
BugReports: https://pkgname.bugtracker.url
@end smallexample
@end cartouche
@end quotation

@noindent
The format is that of a version of a `Debian Control File' (see the help
for @samp{read.dcf} and
@uref{https://www.debian.org/doc/debian-policy/ch-controlfields.html}:
@R{} does not require encoding in UTF-8 and does not support comments
starting with @samp{#}).  Fields start with an @acronym{ASCII} name
immediately followed by a colon: the value starts after the colon and a
space.  Continuation lines (for example, for descriptions longer than
one line) start with a space or tab.  Field names are case-sensitive:
all those used by @R{} are capitalized.

For maximal portability, the @file{DESCRIPTION} file should be written
entirely in @acronym{ASCII} --- if this is not possible it must contain
an @samp{Encoding} field (see below).

Several optional fields take @emph{logical values}: these can be
specified as @samp{yes}, @samp{true}, @samp{no} or @samp{false}:
capitalized values are also accepted.

The @samp{Package}, @samp{Version}, @samp{License}, @samp{Description},
@samp{Title}, @samp{Author}, and @samp{Maintainer} fields are mandatory,
all other fields are optional.  Fields @samp{Author} and
@samp{Maintainer} can be auto-generated from @samp{Authors@@R}, and may
be omitted if the latter is provided: however if they are not
@acronym{ASCII} we recommend that they are provided.

@c DESCRIPTION field Package
The mandatory @samp{Package} field gives the name of the package.  This
should contain only (@acronym{ASCII}) letters, numbers and dot, have at
least two characters and start with a letter and not end in a dot.  If
it needs explaining, this should be done in the @samp{Description} field
(and not the @samp{Title} field).

@c DESCRIPTION field Version
The mandatory @samp{Version} field gives the version of the package.
This is a sequence of at least @emph{two} (and usually three)
non-negative integers separated by single @samp{.} or @samp{-}
characters.  The canonical form is as shown in the example, and a
version such as @samp{0.01} or @samp{0.01.0} will be handled as if it
were @samp{0.1-0}.  It is @strong{not} a decimal number, so for example
@code{0.9 < 0.75} since @code{9 < 75}.

The mandatory @samp{License} field is discussed in the next subsection.

@c DESCRIPTION field Title
The mandatory @samp{Title} field should give a @emph{short} description
of the package.  Some package listings may truncate the title to 65
characters.  It should use @emph{title case} (that is, use capitals for
the principal words: @code{tools::toTitleCase} can help you with this),
not use any markup, not have any continuation lines, and not end in a
period (unless part of @dots{}).  Do not repeat the package name: it is
often used prefixed by the name.  Refer to other packages and external
software in single quotes, and to book titles (and similar) in double
quotes.

@c DESCRIPTION field Description
The mandatory @samp{Description} field should give a
@emph{comprehensive} description of what the package does.  One can use
several (complete) sentences, but only one paragraph. It should be
intelligible to all the intended readership (e.g.@: for a @acronym{CRAN}
package to all @acronym{CRAN} users).  It is good practice not to start
with the package name, `This package' or similar.  As with the
@samp{Title} field, double quotes should be used for quotations
(including titles of books and articles), and single quotes for
non-English usage, including names of other packages and external
software.  This field should also be used for explaining the package
name if necessary.  URLs should be enclosed in angle brackets, e.g.@:
@samp{<https://www.r-project.org>}: see also @ref{Specifying URLs}.

@c DESCRIPTION field Author
@c DESCRIPTION field Authors@R
The mandatory @samp{Author} field describes who wrote @emph{the
package}.  It is a plain text field intended for human readers, but not
for automatic processing (such as extracting the email addresses of all
listed contributors: for that use @samp{Authors@@R}).  Note that all
significant contributors must be included: if you wrote an @R{} wrapper
for the work of others included in the @file{src} directory, you are not
the sole (and maybe not even the main) author.

@c DESCRIPTION field Maintainer
The mandatory @samp{Maintainer} field should give a @emph{single} name
followed by a @emph{valid} (RFC 2822) email address in angle brackets.  It
should not end in a period or comma.  This field is what is reported by
the @code{maintainer} function and used by @code{bug.report}.  For a
@acronym{CRAN} package it should be a @emph{person}, not a mailing list
and not a corporate entity: do ensure that it is valid and will remain
valid for the lifetime of the package.

Note that the @emph{display name} (the part before the address in angle
brackets) should be enclosed in double quotes if it contains
non-alphanumeric characters such as comma or period.  (The current
standard, RFC 5322, allows periods but RFC 2822 did not.)

Both @samp{Author} and @samp{Maintainer} fields can be omitted if a
suitable @samp{Authors@@R} field is given.  This field can be used to
provide a refined and machine-readable description of the package
``authors'' (in particular specifying their precise @emph{roles}),
@emph{via} suitable @R{} code. It should create an object of class
@code{"person"}, by either a call to @code{person} or a series of calls
(one per ``author'') concatenated by @code{c()}: see the example
@file{DESCRIPTION} file above.  The roles can include @samp{"aut"}
(author) for full authors, @samp{"cre"} (creator) for the package
maintainer, and @samp{"ctb"} (contributor) for other contributors,
@samp{"cph"} (copyright holder, which should be the legal name for an
institution or corporate body), among others.  See @code{?person} for
more information.  Note that no role is assumed by default.
Auto-generated package citation information takes advantage of this
specification. The @samp{Author} and @samp{Maintainer} fields are
auto-generated from it if needed when building@footnote{at least if this
is done in a locale which matches the package encoding.}  or installing.

@findex COPYRIGHTS
@c DESCRIPTION field Copyright
An optional @samp{Copyright} field can be used where the copyright
holder(s) are not the authors.  If necessary, this can refer to an
installed file: the convention is to use file @file{inst/COPYRIGHTS}.

@c DESCRIPTION field Date
The optional @samp{Date} field gives the @emph{release date} of the
current version of the package.  It is strongly recommended@footnote{and
required by @acronym{CRAN}, so checked by @command{R CMD check
--as-cran}.} to use the @samp{yyyy-mm-dd} format conforming to the ISO
8601 standard.

The @samp{Depends}, @samp{Imports}, @samp{Suggests}, @samp{Enhances},
@samp{LinkingTo} and @samp{Additional_repositories} fields are discussed
in a later subsection.

@c DESCRIPTION field SystemRequirements
Dependencies external to the @R{} system should be listed in the
@samp{SystemRequirements} field, possibly amplified in a separate
@file{README} file.  This includes specifying a non-default C++ standard
and the need for GNU @command{make}.

@c DESCRIPTION field URL
The @samp{URL} field may give a list of @acronym{URL}s
separated by commas or whitespace, for example the homepage of the
author or a page where additional material describing the software can
be found.  These @acronym{URL}s are converted to active hyperlinks in
@acronym{CRAN} package listings.  @xref{Specifying URLs}.

@c DESCRIPTION field BugReports
The @samp{BugReports} field may contain a single @acronym{URL} to which
bug reports about the package should be submitted.  This @acronym{URL}
will be used by @code{bug.report} instead of sending an email to the
maintainer.  A browser is opened for a @samp{http://} or @samp{https://}
@acronym{URL}.  To specify another email address for bug reports, use
@samp{Contact} instead: however @code{bug.report} will try to extract an
email address (preferably from a @samp{mailto:} URL or enclosed in angle
brackets) from @samp{BugReports}.

@c DESCRIPTION field Priority
Base and recommended packages (i.e., packages contained in the @R{}
source distribution or available from @acronym{CRAN} and recommended to
be included in every binary distribution of @R{}) have a @samp{Priority}
field with value @samp{base} or @samp{recommended}, respectively.  These
priorities must not be used by other packages.

@c DESCRIPTION field Collate
@c DESCRIPTION field Collate.unix
@c DESCRIPTION field Collate.windows
A @samp{Collate} field can be used for controlling the collation order
for the @R{} code files in a package when these are processed for
package installation.  The default is to collate according to the
@samp{C} locale.  If present, the collate specification must list
@emph{all} @R{} code files in the package (taking possible OS-specific
subdirectories into account, see @ref{Package subdirectories}) as a
whitespace separated list of file paths relative to the @file{R}
subdirectory.
@c % double quotes are not allowed in path names, for Windows
Paths containing white space or quotes need to be quoted.  An
OS-specific collation field (@samp{Collate.unix} or
@samp{Collate.windows}) will be used in preference to @samp{Collate}.

@c DESCRIPTION field LazyData
@c DESCRIPTION field LazyLoad
The @samp{LazyData} logical field controls whether the @R{} datasets use
lazy-loading.  A @samp{LazyLoad} field was used in versions prior to
2.14.0, but now is ignored.

@c DESCRIPTION field KeepSource
The @samp{KeepSource} logical field controls if the package code is sourced
using @code{keep.source = TRUE} or @code{FALSE}: it might be needed
exceptionally for a package designed to always be used with
@code{keep.source = TRUE}.

@c DESCRIPTION field ByteCompile
The @samp{ByteCompile} logical field controls if the package R code is to
be byte-compiled on installation: the default is to byte-compile.  This
can be overridden by installing with flag @option{--no-byte-compile}.

@c DESCRIPTION field UseLTO
The @samp{UseLTO} logical field is used on a Unix-alike to indicate if
source code in the package is to be compiled with Link-Time Optimization
(@pxref{Using Link-time Optimization}) if @R{} was installed with
@option{--enable-lto} (default true) or @option{--enable-lto=R} (default
false). This can be overridden by by the flags @option{--use-LTO} and
@option{--no-use-LTO}.  LTO is said to give most size and performance
improvements for large and complex (heavily templated) C++ projects.

@c DESCRIPTION field StagedInstall
The @samp{StagedInstall} logical field controls if package installation
is `staged', that is done to a temporary location and moved to the final
location when successfully completed.  This field was introduced in @R{}
3.6.0 and it true by default: it is considered to be a temporary measure
which may be withdrawn in future.

@c DESCRIPTION field ZipData
The @samp{ZipData} logical field has been ignored since @R{} 2.13.0.

@c DESCRIPTION field Biarch
The @samp{Biarch} logical field is used on Windows to select the
@command{INSTALL} option @option{--force-biarch} for this package.

@c DESCRIPTION field BuildVignettes
The @samp{BuildVignettes} logical field can be set to a false value to
stop @command{R CMD build} from attempting to build the vignettes, as
well as preventing@footnote{But it is checked for Open Source packages
by @command{R CMD check --as-cran}.}  @command{R CMD check} from testing
this.  This should only be used exceptionally, for example if the PDFs
include large figures which are not part of the package sources (and
hence only in packages which do not have an Open Source license).

@c DESCRIPTION field VignetteBuilder
The @samp{VignetteBuilder} field names (in a comma-separated list)
packages that provide an engine for building vignettes.  These may
include the current package, or ones listed in @samp{Depends},
@samp{Suggests} or @samp{Imports}. The @pkg{utils} package is always
implicitly appended.  See @ref{Non-Sweave vignettes} for details.  Note
that if, for example, a vignette has engine @samp{knitr::rmarkdown},
then @CRANpkg{knitr} provides the engine but both @pkg{knitr} and
@CRANpkg{rmarkdown} are needed for using it, so @emph{both} these
packages need to be in the @samp{VignetteBuilder} field and at least
suggested (as @pkg{rmarkdown} is only suggested by @pkg{knitr}, and
hence not available automatically along with it).  Many packages using
@CRANpkg{knitr} also need the package @CRANpkg{formatR} which it
suggests and so the user package needs to do so too and include this in
@samp{VignetteBuilder}.

@c DESCRIPTION field Encoding
If the @file{DESCRIPTION} file is not entirely in @acronym{ASCII} it
should contain an @samp{Encoding} field specifying an encoding.  This is
used as the encoding of the @file{DESCRIPTION} file itself and of the
@file{R} and @file{NAMESPACE} files, and as the default encoding of
@file{.Rd} files.  The examples are assumed to be in this encoding when
running @command{R CMD check}, and it is used for the encoding of the
@code{CITATION} file.  Only encoding names @code{latin1}, @code{latin2}
and @code{UTF-8} are known to be portable.  (Do not specify an encoding
unless one is actually needed: doing so makes the package @emph{less}
portable.  If a package has a specified encoding, you should run
@command{R CMD build} etc in a locale using that encoding.)

@c DESCRIPTION NeedsCompilation
The @samp{NeedsCompilation} field should be set to @code{"yes"} if the
package contains native code which needs to be compiled, otherwise @code{"no"} (when
the package could be installed from source on any platform without
additional tools).  This is used by @code{install.packages(type =
"both")} in @R{} >= 2.15.2 on platforms where binary packages are the
norm: it is normally set by @command{R CMD build} or the repository
assuming compilation is required if and only if the package has a
@file{src} directory.

@c DESCRIPTION field OS_type
The @samp{OS_type} field specifies the OS(es) for which the
package is intended.  If present, it should be one of @code{unix} or
@code{windows}, and indicates that the package can only be installed
on a platform with @samp{.Platform$OS.type} having that value.

@c DESCRIPTION field Type
The @samp{Type} field specifies the type of the package:
@pxref{Package types}.

@c DESCRIPTION field Classification/ACM
@c DESCRIPTION field Classification/ACM-2012
@c DESCRIPTION field Classification/JEL
@c DESCRIPTION field Classification/MSC
@c DESCRIPTION field Classification/MSC-2010
One can add subject classifications for the content of the package using
the fields @samp{Classification/ACM} or @samp{Classification/ACM-2012}
(using the Computing Classification System of the Association for
Computing Machinery, @uref{https://www.acm.org/publications/class-2012}; the former refers
to the 1998 version), @samp{Classification/JEL} (the Journal of Economic
Literature Classification System,
@uref{https://www.aeaweb.org/econlit/jelCodes.php}, or
@samp{Classification/MSC} or @samp{Classification/MSC-2010} (the
Mathematics Subject Classification of the American Mathematical Society,
@uref{https://mathscinet.ams.org/msc/msc2010.html}; the former refers to the 2000 version).
The subject classifications should be comma-separated lists of the
respective classification codes, e.g., @samp{Classification/ACM: G.4,
H.2.8, I.5.1}.

@c DESCRIPTION field Language
A @samp{Language} field can be used to indicate if the package
documentation is not in English: this should be a comma-separated list
of standard (not private use or grandfathered) IETF language tags as
currently defined by RFC 5646
(@uref{https://tools.ietf.org/html/rfc5646}, see also
@uref{https://en.wikipedia.org/wiki/IETF_language_tag}), i.e., use
language subtags which in essence are 2-letter ISO 639-1
(@uref{https://en.wikipedia.@/org/wiki/ISO_639-1}) or 3-letter ISO
639-3 (@uref{https://en.wikipedia.@/org/wiki/ISO_639-3}) language
codes.

@c DESCRIPTION field RdMacros
An @samp{RdMacros} field can be used to hold a comma-separated list of
packages from which the current package will import @file{Rd} macro
definitions.  These package should also be listed in @samp{Imports}
(or @samp{Depends}).  The macros in these packages will be 
imported after the system macros, in the 
order listed in the @samp{RdMacros} field, before any macro definitions
in the current package are loaded.  Macro definitions in individual
@file{.Rd} files in the @file{man} directory are loaded last, and are
local to later parts of that file.  In case of duplicates, the last
loaded definition will be used.@footnote{Duplicate definitions may
trigger a warning: see @ref{User-defined macros}.}  Both @command{R CMD
Rd2pdf} and @command{R CMD Rdconv} have an optional flag
@option{--RdMacros=pkglist}.  The option is also a comma-separated list
of package names, and has priority over the value given in
@file{DESCRIPTION}.  Packages using @file{Rd} macros should depend on
@R{} 3.2.0 or later.

@c DESCRIPTION field Built
@c DESCRIPTION field Packaged
@quotation Note
There should be no @samp{Built} or @samp{Packaged} fields, as these are
added by the package management tools.
@end quotation

@c DESCRIPTION field Note
@c DESCRIPTION field Contact
@c DESCRIPTION field MailingList
There is no restriction on the use of other fields not mentioned here
(but using other capitalizations of these field names would cause
confusion).  Fields @code{Note}, @code{Contact} (for contacting the
authors/developers@footnote{@code{bug.report} will try to extract an
email address from a @code{Contact} field if there is no
@code{BugReports} field.}) and @code{MailingList} are in common
use. Some repositories (including @acronym{CRAN} and R-forge) add their
own fields.


@node Licensing, Package Dependencies, The DESCRIPTION file, Package structure
@subsection Licensing

Licensing for a package which might be distributed is an important but
potentially complex subject.

It is very important that you include license information!  Otherwise,
it may not even be legally correct for others to distribute copies of
the package, let alone use it.

The package management tools use the concept of
`free or open source software'
(FOSS, e.g., @uref{https://en.wikipedia.org/wiki/FOSS})
licenses: the idea being that some users of @R{} and its packages want
to restrict themselves to such software.  Others need to ensure that
there are no restrictions stopping them using a package, e.g.@:
forbidding commercial or military use.  It is a central tenet of FOSS
software that there are no restrictions on users nor usage.

Do not use the @samp{License} field for information on copyright
holders: if needed, use a @samp{Copyright} field.

@c DESCRIPTION field License
@c DESCRIPTION field License_is_FOSS
@c DESCRIPTION field License_restricts_use
The mandatory @samp{License} field in the @file{DESCRIPTION} file should
specify the license of the package in a standardized form.  Alternatives
are indicated @emph{via} vertical bars.  Individual specifications must
be one of
@itemize @bullet
@item
One of the ``standard'' short specifications
@example
GPL-2 GPL-3 LGPL-2 LGPL-2.1 LGPL-3 AGPL-3 Artistic-2.0
BSD_2_clause BSD_3_clause MIT
@end example
@noindent
as made available @emph{via} @uref{https://www.R-project.org/Licenses/} and
contained in subdirectory @file{share/licenses} of the @R{} source or home
directory.
@item
The names or abbreviations of other licenses contained in the license
data base in file @file{share/licenses/license.db} in the @R{} source or
home directory, possibly (for versioned licenses) followed by a version
restriction of the form @samp{(@var{op} @var{v})} with @samp{@var{op}} one of
the comparison operators @samp{<}, @samp{<=}, @samp{>}, @samp{>=},
@samp{==}, or @samp{!=} and @samp{@var{v}} a numeric version specification
(strings of non-negative integers separated by @samp{.}), possibly
combined @emph{via} @samp{,} (see below for an example).  For versioned
licenses, one can also specify the name followed by the version, or
combine an existing abbreviation and the version with a @samp{-}.

Abbreviations @code{GPL} and @code{LGPL} are ambiguous and
usually@footnote{@acronym{CRAN} expands them to e.g.@: @code{GPL-2
| GPL-3}.} taken to mean any version of the license: but it is better
not to use them.
@item
One of the strings @samp{file LICENSE} or @samp{file LICENCE} referring
to a file named @file{LICENSE} or @file{LICENCE} in the package (source
and installation) top-level directory.
@item
The string @samp{Unlimited}, meaning that there are no restrictions on
distribution or use other than those imposed by relevant laws (including
copyright laws).
@end itemize

@noindent
Multiple licences can be specified separated by @samp{|} (surrounded by
spaces) in which case the user can choose any of the alternatives.

If a package license @emph{restricts} a base license (where permitted,
e.g., using GPL-3 or AGPL-3 with an attribution clause), the additional
terms should be placed in file @file{LICENSE} (or @file{LICENCE}), and
the string @samp{+ file LICENSE} (or @samp{+ file LICENCE},
respectively) should be appended to the corresponding individual license
specification (preferably with the @samp{+} surrounded by spaces).  Note
that several commonly used licenses do not permit restrictions: this
includes GPL-2 and hence any specification which includes it.

Examples of standardized specifications include
@example
License: GPL-2
License: LGPL (>= 2.0, < 3) | Mozilla Public License
License: GPL-2 | file LICENCE
License: GPL (>= 2) | BSD_3_clause + file LICENSE
License: Artistic-2.0 | AGPL-3 + file LICENSE
@end example
@noindent
Please note in particular that ``Public domain'' is not a valid license,
since it is not recognized in some jurisdictions.

Please ensure that the license you choose also covers any dependencies
(including system dependencies) of your package: it is particularly
important that any restrictions on the use of such dependencies are
evident to people reading your @file{DESCRIPTION} file.

Fields @samp{License_is_FOSS} and @samp{License_restricts_use} may be
added by repositories where information cannot be computed from the name
of the license.  @samp{License_is_FOSS: yes} is used for licenses which
are known to be FOSS, and @samp{License_restricts_use} can have values
@samp{yes} or @samp{no} if the @file{LICENSE} file is known to restrict
users or usage, or known not to.  These are used by, e.g.@:, the
@code{available.packages} filters.


@cindex LICENSE file
@cindex LICENCE file
The optional file @file{LICENSE}/@file{LICENCE} contains a copy of the
license of the package.  To avoid any confusion only include such a file
if it is referred to in the @samp{License} field of the
@file{DESCRIPTION} file.

Whereas you should feel free to include a license file in your
@emph{source} distribution, please do not arrange to @emph{install} yet
another copy of the @acronym{GNU} @file{COPYING} or @file{COPYING.LIB}
files but refer to the copies on
@uref{https://www.R-project.org/Licenses/} and included in the @R{}
distribution (in directory @file{share/licenses}).  Since files named
@file{LICENSE} or @file{LICENCE} @emph{will} be installed, do not use
these names for standard license files.  To include comments about the
licensing rather than the body of a license, use a file named something
like @file{LICENSE.note}.

A few ``standard'' licenses are rather license templates which need
additional information to be completed @emph{via} @samp{+ file LICENSE}
(with the @samp{+} surrounded by spaces)

@node Package Dependencies, The INDEX file, Licensing, Package structure
@subsection Package Dependencies

@c DESCRIPTION field Depends
The @samp{Depends} field gives a comma-separated list of package names
which this package depends on.  Those packages will be attached before
the current package when @code{library} or @code{require} is called.
Each package name may be optionally followed by a comment in parentheses
specifying a version requirement.  The comment should contain a
comparison operator, whitespace and a valid version number,
e.g.@: @samp{MASS (>= 3.1-20)}.

The @samp{Depends} field can also specify a dependence on a certain
version of @R{} --- e.g., if the package works only with @R{} version
4.0.0 or later, include @samp{R (>= 4.0)} in the @samp{Depends}
field. (As here, trailing zeroes can be dropped and it is recommended
that they are.)  You can also require a certain SVN revision for R-devel
or R-patched, e.g.@: @samp{R (>= 2.14.0), R (>= r56550)} requires a
version later than R-devel of late July 2011 (including released
versions of 2.14.0).

It makes no sense to declare a dependence on @code{R} without a version
specification, nor on the package @pkg{base}: this is an @R{} package
and package @pkg{base} is always available.

A package or @samp{R} can appear more than once in the @samp{Depends}
field, for example to give upper and lower bounds on acceptable
versions.

It is inadvisable to use a dependence on @R{} with patchlevel (the third
digit) other than zero.  Doing so with packages which others depend on
will cause the other packages to become unusable under earlier versions
in the series, and e.g.@: versions 4.x.1 are widely used throughout the
Northern Hemisphere academic year.

Both @code{library} and the @R{} package checking facilities use this
field: hence it is an error to use improper syntax or misuse the
@samp{Depends} field for comments on other software that might be
needed.  The @R{} @command{INSTALL} facilities check if the version of
@R{} used is recent enough for the package being installed, and the list
of packages which is specified will be attached (after checking version
requirements) before the current package.

@c DESCRIPTION field Imports
The @samp{Imports} field lists packages whose namespaces are imported
from (as specified in the @file{NAMESPACE} file) but which do not need
to be attached.  Namespaces accessed by the @samp{::} and @samp{:::}
operators must be listed here, or in @samp{Suggests} or @samp{Enhances}
(see below).  Ideally this field will include all the standard packages
that are used, and it is important to include S4-using packages (as
their class definitions can change and the @file{DESCRIPTION} file is
used to decide which packages to re-install when this happens).
Packages declared in the @samp{Depends} field should not also be in the
@samp{Imports} field.  Version requirements can be specified and are
checked when the namespace is loaded.

@c DESCRIPTION field Suggests
The @samp{Suggests} field uses the same syntax as @samp{Depends} and
lists packages that are not necessarily needed.  This includes packages
used only in examples, tests or vignettes (@pxref{Writing package
vignettes}), and packages loaded in the body of functions.  E.g.,
suppose an example@footnote{even one wrapped in @code{\donttest}.} from
package @pkg{foo} uses a dataset from package @pkg{bar}. Then it is not
necessary to have @pkg{bar} use @pkg{foo} unless one wants to execute
all the examples/tests/vignettes: it is useful to have @pkg{bar}, but
not necessary.  Version requirements can be specified but should be
checked by the code which uses the package.

@c DESCRIPTION field Enhances
Finally, the @samp{Enhances} field lists packages ``enhanced'' by the
package at hand, e.g., by providing methods for classes from these
packages, or ways to handle objects from these packages (so several
packages have @samp{Enhances: chron} because they can handle datetime
objects from @CRANpkg{chron} even though they prefer @R{}'s native
datetime functions).  Version requirements can be specified, but are
currently not used.  Such packages cannot be required to check the
package: any tests which use them must be conditional on the presence
of the package.  (If your tests use e.g.@: a dataset from another
package it should be in @samp{Suggests} and not @samp{Enhances}.)

The general rules are

@itemize @bullet
@item
A package should be listed in only one of these fields.
@item
Packages whose namespace only is needed to load the package using
@code{library(@var{pkgname})} should be listed in the @samp{Imports} field
and not in the @samp{Depends} field.  Packages listed in @code{import}
or @code{importFrom} directives in the @file{NAMESPACE} file should
almost always be in @samp{Imports} and not @samp{Depends}.
@item
Packages that need to be attached to successfully load the package using
@code{library(@var{pkgname})} must be listed in the @samp{Depends}
field.
@item
All packages that are needed@footnote{This includes all packages
directly called by @code{library} and @code{require} calls, as well as
data obtained @emph{via} @code{data(theirdata, package = "somepkg")}
calls: @command{R CMD check} will warn about all of these.  But there
are subtler uses which it may not detect: e.g.@: if package A uses
package B and makes use of functionality in package B which uses package
C which package B suggests or enhances, then package C needs to be in
the @samp{Suggests} list for package A.  Nor will undeclared uses in
included files be reported, nor unconditional uses of packages listed
under @samp{Enhances}.  @command{R CMD check --as-cran} will detect more
of the subtler uses.} to successfully run @code{R CMD check} on the
package must be listed in one of @samp{Depends} or @samp{Suggests} or
@samp{Imports}.  Packages used to run examples or tests conditionally
(e.g.@: @emph{via} @code{if(require(@var{pkgname}))}) should be listed
in @samp{Suggests} or @samp{Enhances}.  (This allows checkers to ensure
that all the packages needed for a complete check are installed.)

@item
Packages needed to use datasets from the package should be in
@samp{Imports}: this includes those needed to define S4 classes used.
@end itemize

@noindent
In particular, packages providing ``only'' data for examples or
vignettes should be listed in @samp{Suggests} rather than @samp{Depends}
in order to make lean installations possible.

Version dependencies in the @samp{Depends} and @samp{Imports} fields are
used by @code{library} when it loads the package, and
@code{install.packages} checks versions for the @samp{Depends},
@samp{Imports} and (for @code{dependencies = TRUE}) @samp{Suggests}
fields.

It is important that the information in these fields is complete and
accurate: it is for example used to compute which packages depend on an
updated package and which packages can safely be installed in parallel.

This scheme was developed before all packages had namespaces (@R{}
2.14.0 in October 2011), and good practice changed once that was in
place.

Field @samp{Depends} should nowadays be used rarely, only for packages
which are intended to be put on the search path to make their facilities
available to the end user (and not to the package itself): for example
it makes sense that a user of package @CRANpkg{latticeExtra} would want
the functions of package @CRANpkg{lattice} made available.

Almost always packages mentioned in @samp{Depends} should also be
imported from in the @file{NAMESPACE} file: this ensures that any needed
parts of those packages are available when some other package imports
the current package.

The @samp{Imports} field should not contain packages which are not
imported from (@emph{via} the @file{NAMESPACE} file or @code{::} or
@code{:::} operators), as all the packages listed in that field need to
be installed for the current package to be installed.  (This is checked
by @command{R CMD check}.)

@R{} code in the package should call @code{library} or @code{require}
only exceptionally.  Such calls are never needed for packages listed in
@samp{Depends} as they will already be on the search path.  It used to
be common practice to use @code{require} calls for packages listed in
@samp{Suggests} in functions which used their functionality, but
nowadays it is better to access such functionality @emph{via} @code{::}
calls.

@c DESCRIPTION field LinkingTo
A package that wishes to make use of header files in other packages to
compile its C/C++ code needs to declare them as a comma-separated list
in the field @samp{LinkingTo} in the @file{DESCRIPTION} file.  For
example

@example
LinkingTo: link1, link2
@end example

@noindent
The @samp{LinkingTo} field can have a version requirement which is
checked at installation.

Specifying a package in @samp{LinkingTo} suffices if these are C/C++
headers containing source code or static linking is done at
installation: the packages do not need to be (and usually should not be)
listed in the @samp{Depends} or @samp{Imports} fields.  This includes
@acronym{CRAN} package @CRANpkg{BH} and almost all users of
@CRANpkg{RcppArmadillo} and @CRANpkg{RcppEigen}.  Note that
@samp{LinkingTo} applies only to installation: if a packages wishes to
use headers to compile code in tests or vignettes the package providing
them needs to be listed in @samp{Suggests} or perhaps @samp{Depends}.

For another use of @samp{LinkingTo} see @ref{Linking to native routines
in other packages}.

@c DESCRIPTION field Additional_repositories
The @samp{Additional_repositories} field is a comma-separated list of
repository URLs where the packages named in the other fields may be
found.  It is currently used by @command{R CMD check} to check that the
packages can be found, at least as source packages (which can be
installed on any platform).

@menu
* Suggested packages::
@end menu

@node Suggested packages,  , Package Dependencies, Package Dependencies
@subsubsection Suggested packages

Note that someone wanting to run the examples/tests/vignettes may not
have a suggested package available (and it may not even be possible to
install it for that platform).  The recommendation used to be to make
their use conditional @emph{via} @code{if(require("@var{pkgname}"))}:
this is OK if that conditioning is done in examples/tests/vignettes,
although using @code{if(requireNamespace("@var{pkgname}"))} is
preferred, if possible.

However, using @code{require} for conditioning @emph{in package code} is
not good practice as it alters the search path for the rest of the
session and relies on functions in that package not being masked by
other @code{require} or @code{library} calls.  It is better practice to
use code like
@example
   if (requireNamespace("rgl", quietly = TRUE)) @{
      rgl::plot3d(...)
   @} else @{
      ## do something else not involving rgl.
   @}
@end example
@noindent
Note the use of @code{rgl::} as that object would not necessarily be
visible (and if it is, it need not be the one from that namespace:
@code{plot3d} occurs in several other packages).  If the intention is to
give an error if the suggested package is not available, simply use
e.g.@: @code{rgl::plot3d}.

If the conditional code produces @code{print} output, function
@code{withAutoprint} can be useful.

Note that the recommendation to use suggested packages conditionally in
tests does also apply to packages used to manage test suites: a
notorious example was @CRANpkg{testthat} which in version 1.0.0 contained
illegal C++ code and hence could not be installed on standards-compliant
platforms.

Some people have assumed that a `recommended' package in @samp{Suggests}
can safely be used unconditionally, but this is not so.  (@R{} can be
installed without recommended packages, and which packages are
`recommended' may change.)

As noted above, packages in @samp{Enhances} @emph{must} be used
conditionally and hence objects within them should always be accessed
@emph{via} @code{::}.

On most systems, @command{R CMD check} can be run with only those
packages declared in @samp{Depends} and @samp{Imports} by setting
environment variable @env{_R_CHECK_DEPENDS_ONLY_=true}, whereas setting
@env{_R_CHECK_SUGGESTS_ONLY_=true} also allows suggested packages, but
not those in @samp{Enhances} nor those not mentioned in the
@file{DESCRIPTION} file.  It is recommended that a package is checked
with each of these set, as well as with neither.

@strong{WARNING:} Be extremely careful if you do things which would be
run at installation time depending on whether suggested packages are
available or not---this includes top-level code in @R{} code files,
@code{.onLoad} functions and the definitions of S4 classes and methods.
The problem is that once a namespace of a suggested package is loaded,
references to it may be captured in the installed package (most commonly
in S4 methods), but the suggested package may not be available when the
installed package is used (which especially for binary packages might be
on a different machine).  Even worse, the problems might not be confined
to your package, for the namespaces of your suggested packages will also
be loaded whenever any package which imports yours is installed and so
may be captured there.

@node The INDEX file, Package subdirectories, Package Dependencies, Package structure
@subsection The @file{INDEX} file
@cindex INDEX file

The optional file @file{INDEX} contains a line for each sufficiently
interesting object in the package, giving its name and a description
(functions such as print methods not usually called explicitly might not
be included).  Normally this file is missing and the corresponding
information is automatically generated from the documentation sources
(using @code{tools::Rdindex()}) when installing from source.

The file is part of the information given by @code{library(help =
@var{pkgname})}.

Rather than editing this file, it is preferable to put customized
information about the package into an overview help page
(@pxref{Documenting packages}) and/or a vignette (@pxref{Writing package
vignettes}).

@node Package subdirectories, Data in packages, The INDEX file, Package structure
@subsection Package subdirectories
@cindex Package subdirectories

The @file{R} subdirectory contains @R{} code files, only.  The code
files to be installed must start with an @acronym{ASCII} (lower or upper
case) letter or digit and have one of the extensions@footnote{Extensions
@file{.S} and @file{.s} arise from code originally written for S(-PLUS),
but are commonly used for assembler code.  Extension @file{.q} was used
for S, which at one time was tentatively called QPE.} @file{.R},
@file{.S}, @file{.q}, @file{.r}, or @file{.s}.  We recommend using
@file{.R}, as this extension seems to be not used by any other software.
It should be possible to read in the files using @code{source()}, so
@R{} objects must be created by assignments.  Note that there need be no
connection between the name of the file and the @R{} objects created by
it.  Ideally, the @R{} code files should only directly assign @R{}
objects and definitely should not call functions with side effects such
as @code{require} and @code{options}.  If computations are required to
create objects these can use code `earlier' in the package (see the
@samp{Collate} field) plus functions in the @samp{Depends} packages
provided that the objects created do not depend on those packages except
@emph{via} namespace imports.

Extreme care is needed if top-level computations are made to depend on
availability or not of other packages.  In particular this applies to
@code{setMethods} and @code{setClass} calls.

Two exceptions are allowed: if the @file{R} subdirectory contains a file
@file{sysdata.rda} (a saved image of one or more @R{} objects: please
use suitable compression as suggested by @code{tools::resaveRdaFiles},
and see also the @samp{SysDataCompression} @file{DESCRIPTION} field.)
this will be lazy-loaded into the namespace environment -- this is
intended for system datasets that are not intended to be user-accessible
@emph{via} @code{data}.  Also, files ending in @samp{.in} will be
allowed in the @file{R} directory to allow a @file{configure} script to
generate suitable files.

Only @acronym{ASCII} characters (and the control characters tab,
formfeed, LF and CR) should be used in code files.  Other characters are
accepted in comments@footnote{but they should be in the encoding
declared in the @file{DESCRIPTION} file.}, but then the comments may not
be readable in e.g.@: a UTF-8 locale.  Non-@acronym{ASCII} characters in
object names will normally@footnote{This is true for OSes which
implement the @samp{C} locale: Windows' idea of the @samp{C} locale uses
the WinAnsi charset.} fail when the package is installed.  Any byte will
be allowed in a quoted character string but @samp{\uxxxx} escapes should
be used for non-@acronym{ASCII} characters.  However,
non-@acronym{ASCII} character strings may not be usable in some locales
and may display incorrectly in others.


@findex library.dynam
Various @R{} functions in a package can be used to initialize and
clean up.  @xref{Load hooks}.

The @file{man} subdirectory should contain (only) documentation files
for the objects in the package in @dfn{R documentation} (Rd) format.
The documentation filenames must start with an @acronym{ASCII} (lower or
upper case) letter or digit and have the extension @file{.Rd} (the
default) or @file{.rd}.  Further, the names must be valid in
@samp{file://} URLs, which means@footnote{More precisely, they can
contain the English alphanumeric characters and the symbols
@samp{$ - _ . + ! ' ( ) , ; @ = &}.}
they must be entirely @acronym{ASCII} and not contain @samp{%}.
@xref{Writing R documentation files}, for more information.  Note that
all user-level objects in a package should be documented; if a package
@var{pkg} contains user-level objects which are for ``internal'' use
only, it should provide a file @file{@var{pkg}-internal.Rd} which
documents all such objects, and clearly states that these are not meant
to be called by the user.  See e.g.@: the sources for package @pkg{grid}
in the @R{} distribution.  Note that packages which use internal objects
extensively should not export those objects from their namespace, when
they do not need to be documented (@pxref{Package namespaces}).

Having a @file{man} directory containing no documentation files may give
an installation error.

The @file{man} subdirectory may contain a subdirectory named @file{macros};
this will contain source for user-defined Rd macros.
(See @ref{User-defined macros}.)  These use the Rd format, but may
not contain anything but macro definitions, comments and whitespace.

The @file{R} and @file{man} subdirectories may contain OS-specific
subdirectories named @file{unix} or @file{windows}.

The sources and headers for the compiled code are in @file{src}, plus
optionally a file @file{Makevars} or @file{Makefile}.  When a package is
installed using @code{R CMD INSTALL}, @command{make} is used to control
compilation and linking into a shared object for loading into @R{}.
There are default @command{make} variables and rules for this
(determined when @R{} is configured and recorded in
@file{@var{R_HOME}/etc@var{R_ARCH}/Makeconf}), providing support for C,
C++, fixed- or free-form Fortran, Objective C and Objective
C++@footnote{either or both of which may not be supported on particular
platforms.  Their main use is on macOS, but unfortunately recent
versions of the macOS SDK have removed much of the support for Objective
C v1.0 and Objective C++.} with associated extensions @file{.c},
@file{.cc} or @file{.cpp}, @file{.f}, @file{.f90} or @file{.f95},
@file{.m}, and @file{.mm}, respectively.  We recommend using @file{.h}
for headers, also for C++@footnote{Using @file{.hpp} is not guaranteed
to be portable.} or Fortran 9x include files.  (Use of extension
@file{.C} for C++ is no longer supported.)  Files in the @file{src}
directory should not be hidden (start with a dot), and hidden files will
under some versions of @R{} be ignored.

It is not portable (and may not be possible at all) to mix all these
languages in a single package.  Because @R{} itself uses it, we know that
C and fixed-form Fortran can be used together, and mixing C, C++ and
Fortran usually work for the platform's native compilers.

If your code needs to depend on the platform there are certain defines
which can used in C or C++.  On all Windows builds (even 64-bit ones)
@samp{_WIN32} will be defined: on 64-bit Windows builds also
@samp{_WIN64}.  On macOS @samp{__APPLE__} is defined@footnote{There is
also @samp{__APPLE_CC__}, but that indicates a compiler with
Apple-specific features not the OS, although for historical reasons is
is defined by LLVM @command{clang}.  It is used in
@file{Rinlinedfuns.h}.}; for an `Apple Silicon' platform, test for both
@samp{__APPLE__} and @samp{__arm64__}.

The default rules can be tweaked by setting macros@footnote{the POSIX
terminology, called `make variables' by GNU make.} in a file
@file{src/Makevars} (@pxref{Using Makevars}).  Note that this mechanism
should be general enough to eliminate the need for a package-specific
@file{src/Makefile}.  If such a file is to be distributed, considerable
care is needed to make it general enough to work on all @R{} platforms.
If it has any targets at all, it should have an appropriate first target
named @samp{all} and a (possibly empty) target @samp{clean} which
removes all files generated by running @command{make} (to be used by
@samp{R CMD INSTALL --clean} and @samp{R CMD INSTALL --preclean}).
There are platform-specific file names on Windows:
@file{src/Makevars.win} takes precedence over @file{src/Makevars} and
@file{src/Makefile.win} must be used. Since @R{} 4.2.0,
@file{src/Makevars.ucrt} takes precedence over
@file{src/Makevars.win} and @file{src/Makefile.ucrt} takes precedence
over @file{src/Makefile.win}. @file{src/Makevars.ucrt} and
@file{src/Makefile.ucrt} will be ignored by earlier versions of @R{}, and
hence can be used to provide content specific for UCRT or Rtools42,
but the support for @file{.ucrt} files may be removed in the future when
building packages from source on the older versions of R will no longer
be needed, and hence the files may be renamed back to @file{.win}.
Some @command{make} programs
require makefiles to have a complete final line, including a newline.

A few packages use the @file{src} directory for purposes other than
making a shared object (e.g.@: to create executables).  Such packages
should have files @file{src/Makefile} and @file{src/Makefile.win} or 
@file{src/Makefile.ucrt}
(unless intended for only Unix-alikes or only Windows).

In very special cases packages may create binary files other than the
shared objects/DLLs in the @file{src} directory.  Such files will not be
installed in a multi-architecture setting since @code{R CMD INSTALL
--libs-only} is used to merge multiple sub-architectures and it only
copies shared objects/DLLs.  If a package wants to install other
binaries (for example executable programs), it should provide an @R{}
script @file{src/install.libs.R} which will be run as part of the
installation in the @code{src} build directory @emph{instead of} copying
the shared objects/DLLs. The script is run in a separate @R{}
environment containing the following variables: @code{R_PACKAGE_NAME}
(the name of the package), @code{R_PACKAGE_SOURCE} (the path to the
source directory of the package), @code{R_PACKAGE_DIR} (the path of the
target installation directory of the package), @code{R_ARCH} (the
arch-dependent part of the path, often empty), @code{SHLIB_EXT} (the
extension of shared objects) and @code{WINDOWS} (@code{TRUE} on Windows,
@code{FALSE} elsewhere).  Something close to the default behavior could
be replicated with the following @file{src/install.libs.R} file:

@example
files <- Sys.glob(paste0("*", SHLIB_EXT))
dest <- file.path(R_PACKAGE_DIR, paste0('libs', R_ARCH))
dir.create(dest, recursive = TRUE, showWarnings = FALSE)
file.copy(files, dest, overwrite = TRUE)
if(file.exists("symbols.rds"))
    file.copy("symbols.rds", dest, overwrite = TRUE)
@end example
@noindent
On the other hand, executable programs could be installed along the
lines of
@example
execs <- c("one", "two", "three")
if(WINDOWS) execs <- paste0(execs, ".exe")
if ( any(file.exists(execs)) ) @{
  dest <- file.path(R_PACKAGE_DIR,  paste0('bin', R_ARCH))
  dir.create(dest, recursive = TRUE, showWarnings = FALSE)
  file.copy(execs, dest, overwrite = TRUE)
@}
@end example

@noindent
Note the use of architecture-specific subdirectories of @file{bin} where
needed.

The @file{data} subdirectory is for data files: @xref{Data in packages}.

The @file{demo} subdirectory is for @R{} scripts (for running @emph{via}
@code{demo()}) that demonstrate some of the functionality of the
package.  Demos may be interactive and are not checked automatically, so
if testing is desired use code in the @file{tests} directory to achieve
this.  The script files must start with a (lower or upper case) letter
and have one of the extensions @file{.R} or @file{.r}.  If present, the
@file{demo} subdirectory should also have a @file{00Index} file with one
line for each demo, giving its name and a description separated by a tab
or at least three spaces. (This index file is not generated
automatically.)  Note that a demo does not have a specified encoding and
so should be an @acronym{ASCII} file (@pxref{Encoding issues}). Function
@code{demo()} will use the package encoding if there is one, but this is
mainly useful for non-@acronym{ASCII} comments.

@cindex .Rinstignore file
The contents of the @file{inst} subdirectory will be copied recursively
to the installation directory.  Subdirectories of @file{inst} should not
interfere with those used by @R{} (currently, @file{R}, @file{data},
@file{demo}, @file{exec}, @file{libs}, @file{man}, @file{help},
@file{html} and @file{Meta}, and earlier versions used @file{latex},
@file{R-ex}).  The copying of the @file{inst} happens after @file{src}
is built so its @file{Makefile} can create files to be installed.  To
exclude files from being installed, one can specify a list of exclude
patterns in file @file{.Rinstignore} in the top-level source directory.
These patterns should be Perl-like regular expressions (see the help for
@code{regexp} in @R{} for the precise details), one per line, to be
matched case-insensitively against the file and directory paths, e.g.@:
@file{doc/.*[.]png$} will exclude all PNG files in @file{inst/doc} based
on the extension.

Note that with the exceptions of @file{INDEX},
@file{LICENSE}/@file{LICENCE} and @file{NEWS}, information files at the
top level of the package will @emph{not} be installed and so not be
known to users of Windows and macOS compiled packages (and not seen
by those who use @command{R CMD INSTALL} or @code{install.packages()}
on the tarball).  So any information files you wish an end user to see
should be included in @file{inst}.  Note that if the named exceptions
also occur in @file{inst}, the version in @file{inst} will be that seen
in the installed package.

@findex CITATION
@cindex citation
@findex NEWS.Rd
@cindex news
Things you might like to add to @file{inst} are a @file{CITATION} file
for use by the @code{citation} function, and a @file{NEWS.Rd} file for
use by the @code{news} function.  See its help page for the specific
format restrictions of the @file{NEWS.Rd} file.

@findex AUTHORS
@findex COPYRIGHTS
Another file sometimes needed in @file{inst} is @file{AUTHORS} or
@file{COPYRIGHTS} to specify the authors or copyright holders when this
is too complex to put in the @file{DESCRIPTION} file.

Subdirectory @file{tests} is for additional package-specific test code,
similar to the specific tests that come with the @R{} distribution.
Test code can either be provided directly in a @file{.R} (or @file{.r}
as from @R{} 3.4.0) file, or @emph{via} a @file{.Rin} file containing
code which in turn creates the corresponding @file{.R} file (e.g., by
collecting all function objects in the package and then calling them
with the strangest arguments).  The results of running a @file{.R} file
are written to a @file{.Rout} file.  If there is a
corresponding@footnote{The best way to generate such a file is to copy
the @file{.Rout} from a successful run of @command{R CMD check}.  If you
want to generate it separately, do run @R{} with options
@option{--vanilla --no-echo} and with environment variable
@env{LANGUAGE=en} set to get messages in English.  Be careful not to use
output with the option @option{--timings} (and note that
@option{--as-cran} sets it).} @file{.Rout.save} file, these two are
compared, with differences being reported but not causing an error.  The
directory @file{tests} is copied to the check area, and the tests are
run with the copy as the working directory and with @code{R_LIBS} set to
ensure that the copy of the package installed during testing will be
found by @code{library(@var{pkg_name})}.  Note that the package-specific
tests are run in a vanilla @R{} session without setting the
random-number seed, so tests which use random numbers will need to set
the seed to obtain reproducible results (and it can be helpful to do so
in all cases, to avoid occasional failures when tests are run).

If directory @file{tests} has a subdirectory @file{Examples} containing
a file @code{@var{pkg}-Ex.Rout.save}, this is compared to the output
file for running the examples when the latter are checked.  Reference
output should be produced without having the @option{--timings} option
set (and note that @option{--as-cran} sets it).

If reference output is included for examples, tests or vignettes do make
sure that it is fully reproducible, as it will be compared verbatim to
that produced in a check run, unless the @samp{IGNORE_RDIFF} markup is
used.  Things which trip up maintainers include displayed version
numbers from loading other packages, printing numerical results to an
unreproducibly high precision and printing timings.  Another trap is
small values which are in fact rounding error from zero: consider using
@code{zapsmall}.

Subdirectory @file{exec} could contain additional executable scripts the
package needs, typically scripts for interpreters such as the shell,
Perl, or Tcl.  NB: only files (and not directories) under @file{exec} are
installed (and those with names starting with a dot are ignored), and
they are all marked as executable (mode @code{755}, moderated by
@samp{umask}) on POSIX platforms.  Note too that this is not suitable
for executable @emph{programs} since some platforms (including Windows)
support multiple architectures using the same installed package
directory.

Subdirectory @file{po} is used for files related to @emph{localization}:
@pxref{Internationalization}.

Subdirectory @file{tools} is the preferred place for auxiliary files
needed during configuration, and also for sources need to re-create
scripts (e.g.@: M4 files for @command{autoconf}: some prefer to put
those in a subdirectory @file{m4} of @file{tools}).


@node Data in packages, Non-R scripts in packages, Package subdirectories, Package structure
@subsection Data in packages

The @file{data} subdirectory is for data files, either to be made
available @emph{via} lazy-loading or for loading using @code{data()}.
(The choice is made by the @samp{LazyData} field in the
@file{DESCRIPTION} file: the default is not to do so.)  It should not be
used for other data files needed by the package, and the convention has
grown up to use directory @file{inst/extdata} for such files.

Data files can have one of three types as indicated by their extension:
plain @R{} code (@file{.R} or @file{.r}), tables (@file{.tab},
@file{.txt}, or @file{.csv}, see @code{?data} for the file formats, and
note that @file{.csv} is @strong{not} the standard@footnote{e.g.@:
@uref{https://tools.ietf.org/html/rfc4180}.} CSV format), or
@code{save()} images (@file{.RData} or @file{.rda}).  The files should
not be hidden (have names starting with a dot).  Note that @R{} code
should be if possible ``self-sufficient'' and not make use of extra
functionality provided by the package, so that the data file can also be
used without having to load the package or its namespace: it should run
as silently as possible and not change the @code{search()} path by
attaching packages or other environments.

Images (extensions @file{.RData}@footnote{People who have trouble with
case are advised to use @file{.rda} as a common error is to refer to
@file{abc.RData} as @file{abc.Rdata}!} or @file{.rda}) can contain
references to the namespaces of packages that were used to create them.
Preferably there should be no such references in data files, and in any
case they should only be to packages listed in the @code{Depends} and
@code{Imports} fields, as otherwise it may be impossible to install the
package.  To check for such references, load all the images into a
vanilla @R{} session, run @code{str()} on all the datasets, and look at
the output of @code{loadedNamespaces()}.

Particular care is needed where a dataset or one of its components is of
an S4 class, especially if the class is defined in a different package.
First, the package containing the class definition has to be available
to do useful things with the dataset, so that package must be listed in
@code{Imports} or @code{Depends} (even if this gives a check warning
about unused imports).  Second, the definition of an S4 class can
change, and often is unnoticed when in a package with a different
author.  So it may be wiser to use the @file{.R} form and use that to
create the dataset object when needed (loading package namespaces but
not attaching them by using @code{requireNamespace(@var{pkg}, quietly =
TRUE)} and using @code{@var{pkg}::} to refer to objects in the
namespace).

If you are not using @samp{LazyData} and either your data files are large
or e.g., you use @file{data/foo.R} scripts to produce your data, loading
your namespace, you 
can speed up installation by providing a file @file{datalist} in the
@file{data} subdirectory.  This should have one line per topic that
@code{data()} will find, in the format @samp{foo} if @code{data(foo)}
provides @samp{foo}, or @samp{foo: bar bah} if @code{data(foo)} provides
@samp{bar} and @samp{bah}.  @command{R CMD build} will automatically add
a @file{datalist} file to @file{data} directories of over 1Mb, using the
function @code{tools::add_datalist}.

Tables (@file{.tab}, @file{.txt}, or @file{.csv} files) can be
compressed by @command{gzip}, @command{bzip2} or @command{xz},
optionally with additional extension @file{.gz}, @file{.bz2} or
@file{.xz}.

If your package is to be distributed, do consider the resource
implications of large datasets for your users: they can make packages
very slow to download and use up unwelcome amounts of storage space, as
well as taking many seconds to load.  It is normally best to distribute
large datasets as @file{.rda} images prepared by @code{save(, compress =
TRUE)} (the default).  Using @command{bzip2} or @command{xz} compression
will usually reduce the size of both the package tarball and the
installed package, in some cases by a factor of two or more.

Package @pkg{tools} has a couple of functions to help with data images:
@code{checkRdaFiles} reports on the way the image was saved, and
@code{resaveRdaFiles} will re-save with a different type of compression,
including choosing the best type for that particular image.

@c DESCRIPTION field LazyDataCompression
Many packages using @samp{LazyData} will benefit from using a form of
compression other than @command{gzip} in the installed lazy-loading
database.  This can be selected by the @option{--data-compress} option
to @command{R CMD INSTALL} or by using the @samp{LazyDataCompression}
field in the @file{DESCRIPTION} file.  Useful values are @code{bzip2},
@code{xz} and the default, @code{gzip}: value @code{none} is also
accepted.  The only way to discover which is best is to try them all and
look at the size of the @file{@var{pkgname}/data/Rdata.rdb} file.  A
function to do that (quoting sizes in KB) is
@example
CheckLazyDataCompression <- function(pkg)
@{
    pkg_name <- sub("_.*", "", pkg)
    lib <- tempfile(); dir.create(lib)
    zs <- c("gzip", "bzip2", "xz")
    res <- integer(3); names(res) <- zs
    for (z in zs) @{
        opts <- c(paste0("--data-compress=", z),
                  "--no-libs", "--no-help", "--no-demo", "--no-exec", "--no-test-load")
        install.packages(pkg, lib, INSTALL_opts = opts, repos = NULL, quiet = TRUE)
        res[z] <- file.size(file.path(lib, pkg_name, "data", "Rdata.rdb"))
    @}
    ceiling(res/1024)
@}
@end example

@noindent
(applied to a source package without any @samp{LazyDataCompression}
field).  @command{R CMD check} will warn if it finds a
@file{@var{pkgname}/data/Rdata.rdb} file of more than 5MB without
@samp{LazyDataCompression} being set.  If you see that, run
@code{CheckLazyDataCompression()} and set the field -- to @code{gzip} in
the unlikely event@footnote{For all the @acronym{CRAN} packages tested,
either @code{gz} or @code{bzip2} provided a very substantial reduction
in installed size.} that is the best choice.

@c DESCRIPTION field SysDataCompression
The analogue for @file{sysdata.rda} is field @samp{SysDataCompression}:
the default is @code{xz} for files bigger than 1MB otherwise
@code{gzip}.

Lazy-loading is not supported for very large datasets (those which when
serialized exceed 2GB, the limit for the format on 32-bit platforms).

@node Non-R scripts in packages, Specifying URLs, Data in packages, Package structure
@subsection Non-R scripts in packages

Code which needs to be compiled (C, C++, Fortran @dots{})
is included in the @file{src} subdirectory and discussed elsewhere in
this document.

Subdirectory @file{exec} could be used for scripts for interpreters such
as the shell, BUGS, JavaScript, Matlab, Perl, php (@CRANpkg{amap}),
Python or Tcl (@CRANpkg{Simile}), or even @R{}.  However, it seems more
common to use the @file{inst} directory, for example
@file{WriteXLS/inst/Perl}, @file{NMF/inst/m-files},
@file{RnavGraph/inst/tcl}, @file{RProtoBuf/inst/python} and
@file{emdbook/inst/BUGS} and @file{gridSVG/inst/js}.

Java code is a special case: except for very small programs,
@file{.java} files should be byte-compiled (to a @file{.class} file) and
distributed as part of a @file{.jar} file: the conventional location for
the @file{.jar} file(s) is @file{inst/java}.  It is desirable (and
required under an Open Source license) to make the Java source files
available: this is best done in a top-level @file{java} directory in the
package---the source files should not be installed.

If your package requires one of these interpreters or an extension then
this should be declared in the @samp{SystemRequirements} field of its
@file{DESCRIPTION} file.  (Users of Java most often do so @emph{via}
@CRANpkg{rJava}, when depending on/importing that suffices unless there
is a version requirement on Java code in the package.)

Windows and Mac users should be aware that the Tcl extensions
@samp{BWidget} and @samp{Tktable} (which have sometimes been included in
the Windows@footnote{@samp{BWidget} still is on Windows but
@samp{Tktable} was not in @R{} 4.0.0.} and macOS @R{} installers)
@emph{are} extensions and do need to be declared (and that
@samp{Tktable} is less widely available than it used to be, including
not in the main repositories for major Linux distributions).
@c Not in Fedora since 17
@c
@c In Sep 2021 Tktable declared by DALY PhViD RSurvey forensim
@c rriskDistributions tcltk2
@c
@c In Sep 2021 BWidget declared by FFD GGEBiplotGUI PBSmodelling asbio
@c biplotbootGUI cncaGUI multibiplotGUI
@c
@samp{BWidget} needs to be installed by the user on other OSes.  This is
fairly easy to do: first find the Tcl search path:

@example
library(tcltk)
strsplit(tclvalue('auto_path'), " ")[[1]]
@end example

@noindent
then download the sources from
@uref{https://sourceforge.net/projects/tcllib/files/BWidget/}
and in a terminal run something like

@example
tar xf bwidget-1.9.14.tar.gz
sudo mv bwidget-1.9.14 /usr/local/lib
@end example

@noindent
substituting a location on the Tcl search path for @file{/usr/local/lib} if
needed.  (If no location on that search path is writeable, you will need
to add one each time BWidget is to be used with @code{tcltk::addTclPath()}.)

To (silently) test for the presence of @samp{Tktable} one can use

@example
library(tcltk)
have_tktable <- !isFALSE(suppressWarnings(tclRequire('Tktable')))
@end example

@noindent
Installing @samp{Tktable} needs a C compiler and the Tk headers (not
necessarily installed with Tcl/Tk).  At the time of writing the latest
sources (from 2008) were available from
@uref{https://sourceforge.net/projects/tktable/files/tktable/2.10/Tktable2.10.tar.gz},
but needed patching for current Tk (8.6.11, but not 8.6.10) -- a patch
can be found at @uref{https://www.stats.ox.ac.uk/pub/bdr/Tktable/}. For
a system installation of Tk you may need to install Tktable as
@samp{root} as on e.g.@: Fedora all the locations on @code{auto_path}
are owned by @samp{root}.

@node Specifying URLs,  , Non-R scripts in packages, Package structure
@subsection Specifying URLs

URLs in many places in the package documentation will be converted to
clickable hyperlinks in at least some of their renderings.  So care is
needed that their forms are correct and portable.

The full URL should be given, including the scheme (often @samp{http://}
or @samp{https://}) and a final @samp{/} for references to directories.

Spaces in URLs are not portable and how they are handled does vary by
HTTP server and by client.  There should be no space in the host part of
an @samp{http://} URL, and spaces in the remainder should be encoded,
with each space replaced by @samp{%20}.

Other characters may benefit from being encoded: see the help on
@code{URLencode()}.

The canonical URL for a @acronym{CRAN} package is
@example
https://cran.r-project.org/package=@var{pkgname}
@end example

@noindent
and not a version starting
@samp{https://cran.r-project.org/web/packages/@var{pkgname}}.

@node Configure and cleanup, Checking and building packages, Package structure, Creating R packages
@section Configure and cleanup

Note that most of this section is specific to Unix-alikes: see the
comments later on about the Windows port of @R{}.

If your package needs some system-dependent configuration before
installation you can include an executable (Bourne@footnote{The script
should only assume a POSIX-compliant @command{/bin/sh} -- see
@uref{https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html}.
In particular @command{bash} extensions must not be used, and not all
@R{} platforms have a @command{bash} command, let alone one at
@file{/bin/bash}.  All known shells used with @R{} support the use of
backticks, but not all support @samp{$(@var{cmd})}.  However, real-world
shells are not fully POSIX-compliant and omissions and idiosyncrasies
need to be worked around---which Autoconf will do for you.  Arithmetic
expansion is a known issue: see
@uref{https://www.gnu.org/software/autoconf/manual/autoconf.html#Portable-Shell}
for this and others.  Some checks can be done by the
@code{checkbashisms} Perl script at
@c 'checkbaskisms' really is the correct path
@uref{https://sourceforge.net/projects/checkbaskisms/files}, also
available in most Linux distributions in a package named either
@samp{devscripts} or @samp{devscripts-checkbashisms}: a later version
can be extracted from Debian sources such as the most recent
@file{tar.xz} in
@uref{https://deb.debian.org/debian/pool/main/d/devscripts/} and has
been needed for recent versions of Perl.}  shell script @file{configure}
in your package which (if present) is executed by @code{R CMD INSTALL}
before any other action is performed.  This can be a script created by
the Autoconf mechanism, but may also be a script written by yourself.
Use this to detect if any nonstandard libraries are present such that
corresponding code in the package can be disabled at install time rather
than giving error messages when the package is compiled or used.  To
summarize, the full power of Autoconf is available for your extension
package (including variable substitution, searching for libraries,
etc.).  Background and useful tips on Autoconf and related tools
(including @command{pkg-config} described below) can be found at
@uref{https://autotools.io/}.

@vindex R_PACKAGE_DIR
@vindex R_PACKAGE_NAME
A @command{configure} script is run in an environment which has all the
environment variables set for an @R{} session (see
@file{@var{R_HOME}/etc/Renviron}) plus @code{R_PACKAGE_NAME} (the name of
the package), @code{R_PACKAGE_DIR} (the path of the target installation
directory of the package, a temporary location for staged installs) and
@code{R_ARCH} (the arch-dependent part of the path, often empty).

Under a Unix-alike only, an executable (Bourne shell) script
@command{cleanup} is executed as the last thing by @code{R CMD INSTALL} if
option @option{--clean} was given, and by @code{R CMD build} when
preparing the package for building from its source.

As an example consider we want to use functionality provided by a (C or
Fortran) library @code{foo}.  Using Autoconf, we can create a configure
script which checks for the library, sets variable @code{HAVE_FOO} to
@code{TRUE} if it was found and to @code{FALSE} otherwise, and then
substitutes this value into output files (by replacing instances of
@samp{@@HAVE_FOO@@} in input files with the value of @code{HAVE_FOO}).
For example, if a function named @code{bar} is to be made available by
linking against library @code{foo} (i.e., using @option{-lfoo}), one
could use

@example
@group
AC_CHECK_LIB(foo, @var{fun}, [HAVE_FOO=TRUE], [HAVE_FOO=FALSE])
AC_SUBST(HAVE_FOO)
......
AC_CONFIG_FILES([foo.R])
AC_OUTPUT
@end group
@end example

@noindent
in @file{configure.ac} (assuming Autoconf 2.50 or later).

The definition of the respective @R{} function in @file{foo.R.in} could be

@example
@group
foo <- function(x) @{
    if(!@@HAVE_FOO@@)
      stop("Sorry, library 'foo' is not available")
    ...
@end group
@end example

@noindent
From this file @command{configure} creates the actual @R{} source file
@file{foo.R} looking like

@example
@group
foo <- function(x) @{
    if(!FALSE)
      stop("Sorry, library 'foo' is not available")
    ...
@end group
@end example

@noindent
if library @code{foo} was not found (with the desired functionality).
In this case, the above @R{} code effectively disables the function.

One could also use different file fragments for available and missing
functionality, respectively.

You will very likely need to ensure that the same C compiler and
compiler flags are used in the @file{configure} tests as when compiling
@R{} or your package.  Under a Unix-alike, you can achieve this by
including the following fragment early in @file{configure.ac}
(@emph{before} calling @code{AC_PROG_CC} or anything which calls it)

@example
@group
: $@{R_HOME=`R RHOME`@}
if test -z "$@{R_HOME@}"; then
  echo "could not determine R_HOME"
  exit 1
fi
CC=`"$@{R_HOME@}/bin/R" CMD config CC`
CFLAGS=`"$@{R_HOME@}/bin/R" CMD config CFLAGS`
CPPFLAGS=`"$@{R_HOME@}/bin/R" CMD config CPPFLAGS`
@end group
@end example

@noindent
(Using @samp{$@{R_HOME@}/bin/R} rather than just @samp{R} is necessary
in order to use the correct version of @R{} when running the script as
part of @code{R CMD INSTALL}, and the quotes since @samp{$@{R_HOME@}}
might contain spaces.)

If your code does load checks (for example, to check for an entry point in
a library or to run code) then you will also need
@example
LDFLAGS=`"$@{R_HOME@}/bin/R" CMD config LDFLAGS`
@end example

Packages written with C++ need to pick up the details for the C++
compiler and switch the current language to C++ by something like
@example
CXX=`"$@{R_HOME@}/bin/R" CMD config CXX`
if test -z "$CXX"; then
  AC_MSG_ERROR([No C++ compiler is available])
fi
CXXFLAGS=`"$@{R_HOME@}/bin/R" CMD config CXXFLAGS`
CPPFLAGS=`"$@{R_HOME@}/bin/R" CMD config CPPFLAGS`
AC_LANG(C++)
@end example

@noindent
The latter is important, as for example C headers may not be available
to C++ programs or may not be written to avoid C++ name-mangling.  Note
that an @R{} installation is not required to have a C++ compiler so
@samp{CXX} may be empty.  If the package specifies a non-default C++
standard, use the @command{config} variable names (such as @code{CXX17})
appropriate to the standard, but still set @code{CXX} and
@code{CXXFLAGS}.

@findex R CMD config
You can use @code{R CMD config} to get the value of the basic
configuration variables, and also the header and library flags necessary
for linking a front-end executable program against @R{}, see @kbd{R CMD
config --help} for details.  If you do, it is essential that you use
both the command and the appropriate flags, so that for example
@samp{CC} must always be used with @samp{CFLAGS} and (for code to be
linked into a shared library) @samp{CPICFLAGS}.  For Fortran, be careful
to use @samp{FC FFLAGS FPICFLAGS} for fixed-form Fortran and 
@samp{FC FCFLAGS FPICFLAGS} for free-form Fortran.

To check for an external BLAS library using the @code{AX_BLAS} macro
from the official Autoconf Macro
Archive@footnote{@uref{https://www.gnu.org/software/autoconf-archive/ax_blas.html}. If
you include macros from that archive you need to arrange for them to be
included in the package sources for use by @command{autoreconf}.}, one
can use

@example
@group
FC=`"$@{R_HOME@}/bin/R" CMD config FC`
FCLAGS=`"$@{R_HOME@}/bin/R" CMD config FFLAGS`
AC_PROG_FC
FLIBS=`"$@{R_HOME@}/bin/R" CMD config FLIBS`
AX_BLAS([], AC_MSG_ERROR([could not find your BLAS library], 1))
@end group
@end example

Note that @code{FLIBS} as determined by @R{} must be used to ensure that
Fortran code works on all @R{} platforms.  
@c Calls to the Autoconf macro
@c @code{AC_F77_LIBRARY_LDFLAGS}, which would overwrite @code{FLIBS}, must
@c not be used (and hence e.g.@: removed from @code{ACX_BLAS}).  (Recent
@c versions of Autoconf in fact allow an already set @code{FLIBS} to
@c override the test for the Fortran linker flags.)


@strong{N.B.}: If the @command{configure} script creates files, e.g.@:
@file{src/Makevars}, you do need a @command{cleanup} script to remove
them.  Otherwise @command{R CMD build} may ship the files that are
created.  For example, package @CRANpkg{RODBC} has

@example
#!/bin/sh

rm -f config.* src/Makevars src/config.h
@end example

@noindent
As this example shows, @command{configure} often creates working files
such as @file{config.log}.

If your configure script needs auxiliary files, it is recommended that
you ship them in a @file{tools} directory (as @R{} itself does).

You should bear in mind that the configure script will not be used on
Windows systems.  If your package is to be made publicly available,
please give enough information for a user on a non-Unix-alike platform
to configure it manually, or provide a @file{configure.win} script
(or @file{configure.ucrt}) to be
used on that platform.  (Optionally, there can be a @file{cleanup.win}
script (or @file{cleanup.ucrt}).
Both should be shell scripts to be executed by @command{ash},
which is a minimal version of Bourne-style @command{sh}.)  When
@file{configure.win} (or @file{configure.ucrt}) is run the environment variables @env{R_HOME}
(which uses @samp{/} as the file separator), @env{R_ARCH} and
@env{R_ARCH_BIN} will be set.  Use @env{R_ARCH} to decide if this is a
64-bit build (its value there is @samp{/x64}) and to install DLLs to the
correct place (@file{$@{R_HOME@}/libs$@{R_ARCH@}}).  Use
@env{R_ARCH_BIN} to find the correct place under the @file{bin}
directory, e.g.@: @file{$@{R_HOME@}/bin$@{R_ARCH_BIN@}/Rscript.exe}.

In some rare circumstances, the configuration and cleanup scripts need
to know the location into which the package is being installed.  An
example of this is a package that uses C code and creates two shared
object/DLLs.  Usually, the object that is dynamically loaded by @R{}
is linked against the second, dependent, object.  On some systems, we
can add the location of this dependent object to the object that is
dynamically loaded by @R{}.  This means that each user does not have to
set the value of the @env{LD_LIBRARY_PATH} (or equivalent) environment
variable, but that the secondary object is automatically resolved.
Another example is when a package installs support files that are
required at run time, and their location is substituted into an @R{}
data structure at installation time.
@vindex R_LIBRARY_DIR
@vindex R_PACKAGE_DIR
@vindex R_PACKAGE_NAME
The names of the top-level library directory (i.e., specifiable
@emph{via} the @samp{-l} argument) and the directory of the package
itself are made available to the installation scripts @emph{via} the two
shell/environment variables @env{R_LIBRARY_DIR} and @env{R_PACKAGE_DIR}.
Additionally, the name of the package (e.g.@: @samp{survival} or
@samp{MASS}) being installed is available from the environment variable
@env{R_PACKAGE_NAME}.  (Currently the value of @env{R_PACKAGE_DIR} is
always @code{$@{R_LIBRARY_DIR@}/$@{R_PACKAGE_NAME@}}, but this used not to
be the case when versioned installs were allowed.  Its main use is in
@file{configure.win} (or @file{configure.ucrt}) scripts for the installation path of external
software's DLLs.)  Note that the value of @env{R_PACKAGE_DIR} may
contain spaces and other shell-unfriendly characters, and so should be
quoted in makefiles and configure scripts.

One of the more tricky tasks can be to find the headers and libraries of
external software.  One tool which is increasingly available on
Unix-alikes (but not by default@footnote{but it is available on the
machines used to produce the @acronym{CRAN} binary packages: however as
Apple does not ship @file{.pc} files for its system libraries such as
@code{expat}, @code{libcurl}, @code{libxml2}, @code{sqlite3} and
@samp{zlib}, it may well not find information on these.  Some
substitutes are available from
@url{https://github.com/R-macos/recipes/tree/master/stubs/pkgconfig-darwin}
and are installed on the @acronym{CRAN} package builders.}  on macOS) to
do this is @command{pkg-config}.  The @file{configure} script will need
to test for the presence of the command itself@footnote{It is not wise
to check the version of @command{pkg-config} as it is sometimes a link
to
@c in 2020 it was for Fedora but not Debian/Ubuntu
@command{pkgconf}, a separate project with a different version series.}
(see for example package @CRANpkg{tiff}), and if present it can be
asked if the software is installed, of a suitable version and for
compilation/linking flags by e.g.@:

@c packages using pkg-config in 2021-11:
@c Cairo R2SWF RProtoBuf Rmixmod Rmpi RoBMA Rpoppler SuperGauss XML
@c audio cairoDevice covafillr devEMF diseq fftw fftwtools gdaltools
@c git2r imager lwgeom nloptr pbdMPI pcaL1 polyclip qs rgdal rgl rjags
@c runjags sf showtext stringfish stringi sysfonts tiff tiledb vapour
@example
$ pkg-config --exists 'libtiff-4 >= 4.1.0' --print-errors # check the status
$ pkg-config --modversion libtiff-4
4.3.0
$ pkg-config --cflags libtiff-4
-I/usr/local/include
$ pkg-config --libs libtiff-4
-L/usr/local/lib -ltiff
$ pkg-config --static --libs libtiff-4
-L/usr/local/lib -ltiff -lwebp -llzma -ljpeg -lz
@end example

@noindent
Note that @command{pkg-config --libs} gives the information required to
link against the default version@footnote{but not all projects get this
right when only a static library is installed, so it is often necessary
to try in turn @command{pkg-config --libs} and @command{pkg-config
--static --libs}.} of that library (usually the dynamic one), and
@command{pkg-config --static --libs} may be needed if the static library is
to be used.

@c package libproj repeatedly got this wrong in 2020.
Static libraries are commonly used on macOS (and Windows) to facilitate
bundling external software with binary distributions of packages.  This
means that portable (source) packages need to allow for this.  It is
@emph{not} safe to just use @command{pkg-config --static --libs}, as
that will often include further libraries that are not necessarily
installed on the user's system (or maybe only the versioned library such
as @file{libjbig.so.2.1} is installed and not @file{libjbig.so} which
would be needed to use @code{-ljbig} sometimes included in
@command{pkg-config --static --libs libtiff-4}).

Another issue is that @command{pkg-config --exists} may not be reliable.
It checks not only that the `module' is available but @emph{all} of the
dependencies, including those in principle needed for static linking.
(XQuartz 2.8.x only distributed dynamic libraries and not some of the
@file{.pc} files needed for @code{--exists}.)

Sometimes the name by which the software is known to
@command{pkg-config} is not what one might expect (e.g.@:
@samp{gtk+-2.0} even for 2.22).  To get a complete list use

@example
pkg-config --list-all | sort
@end example

Some external software provides a @file{-config} command to do a similar
job to @command{pkg-config}, including
@example
curl-config   freetype-config  gdal-config    geos-config
gsl-config    iodbc-config     libpng-config  nc-config
pcre-config   pcre2-config     xml2-config    xslt-config
@end example
@noindent
(@command{curl-config} is for @code{libcurl} not @command{curl}.
@command{nc-config} is for @code{netcdf}.)  Most have an option to use
static libraries.

If using Autoconf it is good practice to include all the Autoconf
sources in the the package (and required for an Open Source package and
tested by @command{R CMD check --as-cran}).  This will include the file
@file{configure.ac}@footnote{a decade ago Autoconf used
@file{configure.in}: this is still accepted but should be renamed and
@command{autoreconf} as used by @command{R CMD check --as-cran} will
report as such.} in the top-level directory of the package.  If
extensions written in @command{m4} are needed, these should be included
under the directory @file{tools} and included from @file{configure.ac}
@emph{via} e.g.,
@example
m4_include([tools/ax_pthread.m4])
@end example
@noindent
Alternatively, Autoconf can be asked to search all @file{.m4} files in a
directory by including something like@footnote{For those using
@command{autoconf} 2.70 or later there is also
@code{AC_CONFIG_MACRO_DIRS} which allows multiple directories to be
specified.}
@example
AC_CONFIG_MACRO_DIR([tools/m4])
@end example
@noindent
One source of such extensions is the `Autoconf Archive'
(@uref{https://www.gnu.org/software/autoconf-archive/}.  It is not
safe to assume this is installed on users' machines, so the extension
should be shipped with the package (taking care to comply with its
licence).

@menu
* Using Makevars::
* Configure example::
* Using F9x code::
* Using C++ code::
* Using @command{cmake}::
@end menu

@node Using Makevars, Configure example, Configure and cleanup, Configure and cleanup
@subsection Using @file{Makevars}

@menu
* OpenMP support::
* Using pthreads::
* Compiling in sub-directories::
@end menu

Sometimes writing your own @file{configure} script can be avoided by
supplying a file @file{Makevars}: also one of the most common uses of a
@file{configure} script is to make @file{Makevars} from
@file{Makevars.in}.

A @file{Makevars} file is a makefile and is used as one of several
makefiles by @command{R CMD SHLIB} (which is called by @command{R CMD
INSTALL} to compile code in the @file{src} directory).  It should be
written if at all possible in a portable style, in particular (except
for @file{Makevars.win} and @file{Makevars.ucrt}) without the use of GNU
extensions.

The most common use of a @file{Makevars} file is to set additional
preprocessor options (for example include paths and definitions) for
C/C++ files @emph{via} @code{PKG_CPPFLAGS}, and additional compiler
flags by setting @code{PKG_CFLAGS}, @code{PKG_CXXFLAGS} or
@code{PKG_FFLAGS}, for C, C++ or Fortran respectively (@pxref{Creating
shared objects}).

@strong{N.B.}: Include paths are preprocessor options, not compiler
options, and @strong{must} be set in @code{PKG_CPPFLAGS} as otherwise
platform-specific paths (e.g.@: @samp{-I/usr/local/include}) will take
precedence.  @code{PKG_CPPFLAGS} should contain @samp{-I}, @samp{-D},
@samp{-U} and (where supported) @samp{-include} and @samp{-pthread}
options: everything else should be a compiler flag.  The order of flags
matters, and using @samp{-I} in @code{PKG_CFLAGS} or @code{PKG_CXXFLAGS}
has led to hard-to-debug platform-specific errors.

@file{Makevars} can also be used to set flags for the linker, for
example @samp{-L} and @samp{-l} options, @emph{via} @code{PKG_LIBS}.

When writing a @file{Makevars} file for a package you intend to
distribute, take care to ensure that it is not specific to your
compiler: flags such as @option{-O2 -Wall -pedantic} (and all other
@option{-W} flags: for the Oracle compilers these are used to pass
arguments to compiler phases) are all specific to GCC (and compilers such
as @command{clang} which aim to be options-compatible with it).

Also, do not set variables such as @code{CPPFLAGS}, @code{CFLAGS} etc.:
these should be settable by users (sites) through appropriate personal
(site-wide) @file{Makevars} files.
@ifset UseExternalXrefs
@xref{Customizing package compilation, , Customizing package compilation,
R-admin, R Installation and Administration},
@end ifset

There are some macros@footnote{in POSIX parlance: GNU @command{make}
calls these `make variables'.}  which are set whilst configuring the
building of @R{} itself and are stored in
@file{@var{R_HOME}/etc@var{R_ARCH}/Makeconf}.  That makefile is included
as a @file{Makefile} @emph{after} @file{Makevars[.win]}, and the macros
it defines can be used in macro assignments and make command lines in
the latter.  These include

@table @code
@item FLIBS
@vindex FLIBS
A macro containing the set of libraries need to link Fortran code.  This
may need to be included in @code{PKG_LIBS}: it will normally be included
automatically if the package contains Fortran source files in the
@file{src} directory.

@item BLAS_LIBS
@vindex BLAS_LIBS
A macro containing the BLAS libraries used when building @R{}.  This may
need to be included in @code{PKG_LIBS}.  Beware that if it is empty then
the @R{} executable will contain all the double-precision and
double-complex BLAS routines, but no single-precision nor complex
routines.  If @code{BLAS_LIBS} is included, then @code{FLIBS} also needs
to be@footnote{at least on Unix-alikes: the Windows build currently
resolves such dependencies to a static Fortran library when
@file{Rblas.dll} is built.  Also, not if @code{USE_FC_TO_LINK} is used.}
included following it, as most BLAS libraries are written at least
partially in Fortran.

@item LAPACK_LIBS
@vindex LAPACK_LIBS
A macro containing the LAPACK libraries (and paths where appropriate)
used when building @R{}.  This may need to be included in
@code{PKG_LIBS}.  It may point to a dynamic library @code{libRlapack}
which contains the main double-precision LAPACK routines as well as
those double-complex LAPACK routines needed to build @R{}, or it may
point to an external LAPACK library, or may be empty if an external BLAS
library also contains LAPACK.

[@code{libRlapack} includes all the double-precision LAPACK routines
which were current in 2003: a list of which routines are included is in
file @file{src/modules/lapack/README}. Note that an external LAPACK/BLAS
library need not do so, as some were `deprecated' (and not compiled by
default) in LAPACK 3.6.0 in late 2015.]

For portability, the macros @code{BLAS_LIBS} and @code{FLIBS} should
always be included @emph{after} @code{LAPACK_LIBS} (and in that order).

@item SAFE_FFLAGS
@vindex SAFE_FFLAGS
A macro containing flags which are needed to circumvent
over-optimization of FORTRAN code: it is might be @samp{-g -O2
-ffloat-store} or @samp{-g -O2 -msse2 -mfpmath=sse} on @cputype{ix86}
platforms using @command{gfortran}.  Note that this is @strong{not} an
additional flag to be used as part of @code{PKG_FFLAGS}, but a
replacement for @code{FFLAGS}.  See the example later in this section.
@end table

@vindex OBJECTS
Setting certain macros in @file{Makevars} will prevent @command{R CMD
SHLIB} setting them: in particular if @file{Makevars} sets
@samp{OBJECTS} it will not be set on the @command{make} command line.
This can be useful in conjunction with implicit rules to allow other
types of source code to be compiled and included in the shared object.
It can also be used to control the set of files which are compiled,
either by excluding some files in @file{src} or including some files in
subdirectories.  For example

@example
OBJECTS = 4dfp/endianio.o 4dfp/Getifh.o R4dfp-object.o
@end example


Note that @file{Makevars} should not normally contain targets, as it is
included before the default makefile and @command{make} will call the
first target, intended to be @code{all} in the default makefile.  If you
really need to circumvent that, use a suitable (phony) target @code{all}
before any actual targets in @file{Makevars.[win]}: for example package
@CRANpkg{fastICA} used to have

@example
PKG_LIBS = @@BLAS_LIBS@@

SLAMC_FFLAGS=$(R_XTRA_FFLAGS) $(FPICFLAGS) $(SHLIB_FFLAGS) $(SAFE_FFLAGS)

all: $(SHLIB)

slamc.o: slamc.f
        $(FC) $(SLAMC_FFLAGS) -c -o slamc.o slamc.f
@end example

@noindent
needed to ensure that the LAPACK routines find some constants without
infinite looping.  The Windows equivalent was

@example
all: $(SHLIB)

slamc.o: slamc.f
        $(FC) $(SAFE_FFLAGS) -c -o slamc.o slamc.f
@end example

@noindent
(since the other macros are all empty on that platform, and @R{}'s
internal BLAS was not used).  Note that the first target in
@file{Makevars} will be called, but for back-compatibility it is best
named @code{all}.

If you want to create and then link to a library, say using code in a
subdirectory, use something like

@example
.PHONY: all mylibs

all: $(SHLIB)
$(SHLIB): mylibs

mylibs:
        (cd subdir; $(MAKE))
@end example

@noindent
Be careful to create all the necessary dependencies, as there is no
guarantee that the dependencies of @code{all} will be run in a
particular order (and some of the @acronym{CRAN} build machines use
multiple CPUs and parallel makes).  In particular,

@example
all: mylibs
@end example

@noindent
does @strong{not} suffice.  GNU make does allow the construct
@example
.NOTPARALLEL: all
all: mylibs $(SHLIB)
@end example

@noindent
but that is not portable.  @command{dmake} and @command{pmake} allow the
similar @code{.NO_PARALLEL}, also not portable: some variants of
@command{pmake} accept @code{.NOTPARALLEL} as an alias for
@code{.NO_PARALLEL}.

Note that on Windows it is required that @file{Makevars[.win, .ucrt]} does
create a DLL: this is needed as it is the only reliable way to ensure
that building a DLL succeeded.  If you want to use the @file{src}
directory for some purpose other than building a DLL, use a
@file{Makefile.win} or @file{Makefile.ucrt} file.

It is sometimes useful to have a target @samp{clean} in @file{Makevars},
@file{Makevars.win} or @file{Makevars.ucrt}:
this will be used by @command{R CMD build} to
clean up (a copy of) the package sources.  When it is run by
@command{build} it will have fewer macros set, in particular not
@code{$(SHLIB)}, nor @code{$(OBJECTS)} unless set in the file itself.
It would also be possible to add tasks to the target @samp{shlib-clean}
which is run by @command{R CMD INSTALL} and @command{R CMD SHLIB} with
options @option{--clean} and @option{--preclean}.

An unfortunately common error is to have

@example
all: $(SHLIB) clean
@end example

@noindent
which asks @command{make} to clean in parallel with compiling the code.
Not only does this lead to hard-to-debug installation errors, it wipes
out all the evidence of any error (from a parallel make or not).  It is
much better to leave cleaning to the end user using the facilities in
the previous paragraph.

If you want to run @R{} code in @file{Makevars}, e.g.@: to find
configuration information, please do ensure that you use the correct
copy of @code{R} or @code{Rscript}: there might not be one in the path
at all, or it might be the wrong version or architecture.  The correct
way to do this is @emph{via}

@example
"$(R_HOME)/bin$(R_ARCH_BIN)/Rscript" @var{filename}
"$(R_HOME)/bin$(R_ARCH_BIN)/Rscript" -e '@var{R expression}'
@end example

@noindent
where @code{$(R_ARCH_BIN)} is only needed currently on Windows.

Environment or make variables can be used to select different macros for
32- and 64-bit code, for example (GNU @command{make} syntax, allowed on
Windows)

@example
ifeq "$(WIN)" "64"
PKG_LIBS = @var{value for 64-bit Windows}
else
PKG_LIBS = @var{value for 32-bit Windows}
endif
@end example

On Windows there is normally a choice between linking to an import
library or directly to a DLL.  Where possible, the latter is much more
reliable: import libraries are tied to a specific toolchain, and in
particular on 64-bit Windows two different conventions have been
commonly used.  So for example instead of

@example
PKG_LIBS = -L$(XML_DIR)/lib -lxml2
@end example

@noindent
one can use

@example
PKG_LIBS = -L$(XML_DIR)/bin -lxml2
@end example

@noindent
since on Windows @code{-lxxx} will look in turn for

@example
libxxx.dll.a
xxx.dll.a
libxxx.a
xxx.lib
libxxx.dll
xxx.dll
@end example

@noindent
where the first and second are conventionally import libraries, the
third and fourth often static libraries (with @code{.lib} intended for
Visual C++), but might be import libraries.  See for example
@uref{https://sourceware.org/binutils/docs-2.20/ld/WIN32.html#WIN32}.

The fly in the ointment is that the DLL might not be named
@file{libxxx.dll}, and in fact on 32-bit Windows there is a
@file{libxml2.dll} whereas on one build for 64-bit Windows the DLL is
called @file{libxml2-2.dll}.  Using import libraries can cover over
these differences but can cause equal difficulties.

If static libraries are available they can save a lot of problems with
run-time finding of DLLs, especially when binary packages are to be
distributed and even more when these support both architectures.  Where
using DLLs is unavoidable we normally arrange (@emph{via}
@file{configure.win} or @file{configure.ucrt}) to ship them in the same directory as the package
DLL.

@node OpenMP support, Using pthreads, Using Makevars, Using Makevars
@subsubsection OpenMP support

@cindex OpenMP

There is some support for packages which wish to use
OpenMP@footnote{@uref{https://www.openmp.org/},
@uref{https://en.wikipedia.org/wiki/OpenMP},
@uref{https://hpc.llnl.gov/training/tutorials/openmp-tutorial}}.  The
@command{make} macros

@example
SHLIB_OPENMP_CFLAGS
SHLIB_OPENMP_CXXFLAGS
SHLIB_OPENMP_FFLAGS
@end example

@noindent
are available for use in @file{src/Makevars}, @file{src/Makevars.win}
or @file{Makevars.ucrt}.
Include the appropriate macro in @code{PKG_CFLAGS}, @code{PKG_CXXFLAGS}
and so on, and also in @code{PKG_LIBS} (but see below for Fortran).
C/C++ code that needs to be conditioned on the use of OpenMP can be used
inside @code{#ifdef _OPENMP}: note that some toolchains used for @R{}
(including Apple's for macOS and some others using
@command{clang}@footnote{Default builds of @command{clang} 3.8.0 and
later have support for OpenMP, but the @code{libomp} run-time library
may not be installed.}) have no OpenMP support at all, not even
@file{omp.h}.

For example, a package with C code written for OpenMP should have in
@file{src/Makevars} the lines

@example
PKG_CFLAGS = $(SHLIB_OPENMP_CFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CFLAGS)
@end example

Note that the macro @code{SHLIB_OPENMP_CXXFLAGS} applies to the default
C++ compiler and not necessarily to the C++17/20 compiler: users of the
latter should do their own @command{configure} checks.  If you do use
your own checks, make sure that OpenMP support is complete by compiling
and linking an OpenMP-using program: on some platforms the runtime
library is optional and on others that library depends on other optional
libraries.
@c For clang pre-7, libomp.so depended on libatomic.

Some care is needed when compilers are from different families which may
use different OpenMP runtimes (e.g.@: @command{clang} @emph{vs} GCC
including @command{gfortran}, although it is often possible to use the
@command{clang} runtime with GCC but not @emph{vice versa}: however
@command{gfortran} >= 9 may generate calls not in the @command{clang}
runtime).  For a package with Fortran code using OpenMP the appropriate
lines are

@example
PKG_FFLAGS = $(SHLIB_OPENMP_FFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CFLAGS)
@end example

@noindent
as the C compiler will be used to link the package code. There are
platforms on which this does not work @emph{for some OpenMP-using code}
and installation will fail.  Since @R{} >= 3.6.2 the best alternative
for a package with only Fortran sources using OpenMP is to use
@example
USE_FC_TO_LINK =
PKG_FFLAGS = $(SHLIB_OPENMP_FFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_FFLAGS)
@end example
in @file{src/Makevars}, @file{src/Makevars.win} or @file{Makevars.ucrt}.
Note however, that
when this is used @code{$(FLIBS)} should not be included in
@code{PKG_LIBS} since it is for linking Fortran-compiled code by the C
compiler.

@c Most often seen with clang and sanitizer flags.
Common platforms may inline all OpenMP calls and so tolerate the
omission of the OpenMP flag from @code{PKG_LIBS}, but this usually
results in an installation failure with a different compiler or
compilation flags.  So cross-check that e.g.@: @code{-fopenmp} appears
in the linking line in the installation logs.

It is not portable to use OpenMP with more than one of C, C++ and
Fortran in a single package since it is not uncommon that the compilers
are of different families.

For portability, any C/C++ code using the @code{omp_*} functions should
include the @file{omp.h} header: some compilers (but not all) include it
when OpenMP mode is switched on (e.g.@: @emph{via} flag
@option{-fopenmp}).

@c http://openmp.org/wp/openmp-compilers-tools/
@c clang 3.8.x reports 201307 but has full support only for 3.1 (201111)
@c clang 3.9.x reports 201111 but has all but offloading support of 4.0.
There is nothing@footnote{In most implementations the @code{_OPENMP}
macro has value a date which can be mapped to an OpenMP version: for
example, value @code{201307} is the date of version 4.0 (July
2013). However this may be used to denote the latest version which is
partially supported, not that which is fully implemented.} to say what
version of OpenMP is supported: version 4.0 (and much of 4.5 or 5.0) is
supported by recent versions of the Linux, Windows and Solaris
platforms, but portable packages cannot assume that end users have
recent versions.  Apple @command{clang} on macOS has no OpenMP support.
@uref{https://www.openmp.org/resources/openmp-compilers-tools/} gives
some idea of what compilers support what versions.

Rarely, using OpenMP with @command{clang} on Linux generates calls in
@code{libatomic}, resulting in loading messages like
@example
  undefined symbol: __atomic_compare_exchange
  undefined symbol: __atomic_load
@end example
@noindent
The workaround is to link with @code{-latomic} (having checked it exists).

The performance of OpenMP varies substantially between platforms.  The
Windows implementation has substantial overheads, so is only beneficial
if quite substantial tasks are run in parallel.  Also, on Windows new
threads are started with the default@footnote{Windows default, not
MinGW-w64 default.} FPU control word, so computations done on OpenMP
threads will not make use of extended-precision arithmetic which is the
default for the main process.
@c mingw64-public, 2015-02-02.
@c https://stackoverflow.com/questions/2553725/is-the-fpu-control-word-setting-per-thread-or-per-process

Do not include these macros unless your code does make use of OpenMP
(possibly for C++ via included external headers): this can result in the
OpenMP runtime being linked in, threads being started, @dots{}.

Calling any of the @R{} API from threaded code is `for experts only' and
strongly discouraged.  Many functions in the @R{} API modify internal
@R{} data structures and might corrupt these data structures if called
simultaneously from multiple threads.  Most @R{} API functions can
signal errors, which must only happen on the @R{} main thread.  Also,
external libraries (e.g.@: LAPACK) may not be thread-safe.

Packages are not standard-alone programs, and an @R{} process could
contain more than one OpenMP-enabled package as well as other components
(for example, an optimized BLAS) making use of OpenMP.  So careful
consideration needs to be given to resource usage.  OpenMP works with
parallel regions, and for most implementations the default is to use as
many threads as `CPUs' for such regions.  Parallel regions can be
nested, although it is common to use only a single thread below the
first level.  The correctness of the detected number of `CPUs' and the
assumption that the @R{} process is entitled to use them all are both
dubious assumptions.  One way to limit resources is to limit the overall
number of threads available to OpenMP in the @R{} process: this can be
done @emph{via} environment variable @env{OMP_THREAD_LIMIT}, where
implemented.@footnote{Which it was at the time of writing with GCC,
Oracle, Intel and Clang compilers.  The count may include the thread
running the main process.}  Alternatively, the number of threads per
region can be limited by the environment variable @env{OMP_NUM_THREADS}
or API call @code{omp_set_num_threads}, or, better, for the regions in
your code as part of their specification. E.g.@: @R{} uses@footnote{Be
careful not to declare @code{nthreads} as @code{const int}: the Oracle
compiler requires it to be `an lvalue'.}
@example
#pragma omp parallel for num_threads(nthreads) @dots{}
@end example
@noindent
That way you only control your own code and not that of other OpenMP users.

Note that setting environment variables to control OpenMP is
implementation-dependent and may need to be done outside the @R{}
process or before any use of OpenMP (which might be by another process
or @R{} itself).   Also, implementation-specific variables such as
@env{KMP_THREAD_LIMIT} might take precedence.

@node Using pthreads, Compiling in sub-directories, OpenMP support, Using Makevars
@subsubsection Using pthreads

There is no direct support for the POSIX threads (more commonly known as
@code{pthreads}): by the time we considered adding it several packages
were using it unconditionally so it seems that nowadays it is
universally available on POSIX operating systems (hence not Windows).

For reasonably recent versions of @command{gcc} and @command{clang} the
correct specification is

@example
PKG_CPPFLAGS = -pthread
PKG_LIBS = -pthread
@end example

@noindent
(and the plural version is also accepted on some systems/versions).  For
other platforms the specification is

@example
PKG_CPPFLAGS = -D_REENTRANT
PKG_LIBS = -lpthread
@end example
@noindent
(and note that the library name is singular).  This is what
@option{-pthread} does on all known current platforms (although earlier
versions of OpenBSD used a different library name).

For a tutorial see
@uref{https://hpc-tutorials.llnl.gov/posix/}.

POSIX threads are not normally used on Windows, which has its own native
concepts of threads.  However, there are two projects implementing
@code{pthreads} on top of Windows, @code{pthreads-w32} and
@code{winpthreads} (part of the MinGW-w64 project).

Whether Windows toolchains implement @code{pthreads} is up to the
toolchain provider.  A @command{make} variable
@code{SHLIB_PTHREAD_FLAGS} is available for use in
@file{src/Makevars.win} or @file{Makevars.ucrt}: this should be included in both
@code{PKG_CPPFLAGS} (or the Fortran compiler flags) and @code{PKG_LIBS}.

The presence of a working @code{pthreads} implementation cannot be
unambiguously determined without testing for yourself: however, that
@samp{_REENTRANT} is defined@footnote{some Windows toolchains had the
typo @samp{_REENTRANCE} instead.} in C/C++ code is a good indication.

Note that not all @code{pthreads} implementations are equivalent as parts
are optional (see
@uref{https://pubs.opengroup.org/onlinepubs/009695399/basedefs/pthread.h.html}):
for example, macOS lacks the `Barriers' option.

See also the comments on thread-safety and performance under OpenMP: on
all known @R{} platforms OpenMP is implemented @emph{via}
@code{pthreads} and the known performance issues are in the latter.

@node Compiling in sub-directories,  , Using pthreads, Using Makevars
@subsubsection Compiling in sub-directories

Package authors fairly often want to organize code in sub-directories of
@file{src}, for example if they are including a separate piece of
external software to which this is an @R{} interface.

One simple way is simply to set @code{OBJECTS} to be all the objects
that need to be compiled, including in sub-directories.  For example,
@acronym{CRAN} package @CRANpkg{RSiena} has

@smallexample
SOURCES = $(wildcard data/*.cpp network/*.cpp utils/*.cpp model/*.cpp model/*/*.cpp model/*/*/*.cpp)

OBJECTS = siena07utilities.o siena07internals.o siena07setup.o siena07models.o $(SOURCES:.cpp=.o)
@end smallexample

@noindent
One problem with that approach is that unless GNU make extensions are
used, the source files need to be listed and kept up-to-date.  As in the
following from @acronym{CRAN} package @CRANpkg{lossDev}:

@smallexample
OBJECTS.samplers = samplers/ExpandableArray.o samplers/Knots.o \
  samplers/RJumpSpline.o samplers/RJumpSplineFactory.o \
  samplers/RealSlicerOV.o samplers/SliceFactoryOV.o samplers/MNorm.o
OBJECTS.distributions = distributions/DSpline.o \
  distributions/DChisqrOV.o distributions/DTOV.o \
  distributions/DNormOV.o distributions/DUnifOV.o distributions/RScalarDist.o
OBJECTS.root = RJump.o

OBJECTS = $(OBJECTS.samplers) $(OBJECTS.distributions) $(OBJECTS.root)
@end smallexample

Where the subdirectory is self-contained code with a suitable makefile,
the best approach is something like

@smallexample
PKG_LIBS = -LCsdp/lib -lsdp $(LAPACK_LIBS) $(BLAS_LIBS) $(FLIBS)

$(SHLIB): Csdp/lib/libsdp.a

Csdp/lib/libsdp.a:      
        @@(cd Csdp/lib && $(MAKE) libsdp.a \
          CC="$(CC)" CFLAGS="$(CFLAGS) $(CPICFLAGS)" AR="$(AR)" RANLIB="$(RANLIB)")
@end smallexample

@noindent
Note the quotes: the macros can contain spaces, e.g.@: @code{CC = "gcc
-m64 -std=gnu99"}.  Several authors have forgotten about parallel makes:
the static library in the subdirectory must be made before the shared
object (@code{$(SHLIB)}) and so the latter must depend on the former.
Others forget the need@footnote{A few OSes (AIX, Windows) do not
need special flags for such code, but most do---although compilers will
often generate PIC code when not asked to do so.} for
position-independent code.

We really do not recommend using @file{src/Makefile} instead of
@file{src/Makevars}, and as the example above shows, it is not
necessary.

@node Configure example, Using F9x code, Using Makevars, Configure and cleanup
@subsection Configure example

It may be helpful to give an extended example of using a
@file{configure} script to create a @file{src/Makevars} file: this is
based on that in the @CRANpkg{RODBC} package.

The @file{configure.ac} file follows: @file{configure} is created from
this by running @command{autoconf} in the top-level package directory
(containing @file{configure.ac}).

@quotation
@c @cartouche
@smallexample
AC_INIT([RODBC], 1.1.8) dnl package name, version

dnl A user-specifiable option
odbc_mgr=""
AC_ARG_WITH([odbc-manager],
            AC_HELP_STRING([--with-odbc-manager=MGR],
                           [specify the ODBC manager, e.g. odbc or iodbc]),
            [odbc_mgr=$withval])

if test "$odbc_mgr" = "odbc" ; then
  AC_PATH_PROGS(ODBC_CONFIG, odbc_config)
fi

dnl Select an optional include path, from a configure option
dnl or from an environment variable.
AC_ARG_WITH([odbc-include],
            AC_HELP_STRING([--with-odbc-include=INCLUDE_PATH],
                           [the location of ODBC header files]),
            [odbc_include_path=$withval])
RODBC_CPPFLAGS="-I."
if test [ -n "$odbc_include_path" ] ; then
   RODBC_CPPFLAGS="-I. -I$@{odbc_include_path@}"
else
  if test [ -n "$@{ODBC_INCLUDE@}" ] ; then
     RODBC_CPPFLAGS="-I. -I$@{ODBC_INCLUDE@}"
  fi
fi

dnl ditto for a library path
AC_ARG_WITH([odbc-lib],
            AC_HELP_STRING([--with-odbc-lib=LIB_PATH],
                           [the location of ODBC libraries]),
            [odbc_lib_path=$withval])
if test [ -n "$odbc_lib_path" ] ; then
   LIBS="-L$odbc_lib_path $@{LIBS@}"
else
  if test [ -n "$@{ODBC_LIBS@}" ] ; then
     LIBS="-L$@{ODBC_LIBS@} $@{LIBS@}"
  else
    if test -n "$@{ODBC_CONFIG@}"; then
      odbc_lib_path=`odbc_config --libs | sed s/-lodbc//`
      LIBS="$@{odbc_lib_path@} $@{LIBS@}"
    fi
  fi
fi

dnl Now find the compiler and compiler flags to use
: $@{R_HOME=`R RHOME`@}
if test -z "$@{R_HOME@}"; then
  echo "could not determine R_HOME"
  exit 1
fi
CC=`"$@{R_HOME@}/bin/R" CMD config CC`
CFLAGS=`"$@{R_HOME@}/bin/R" CMD config CFLAGS`
CPPFLAGS=`"$@{R_HOME@}/bin/R" CMD config CPPFLAGS`

if test -n "$@{ODBC_CONFIG@}"; then
  RODBC_CPPFLAGS=`odbc_config --cflags`
fi
CPPFLAGS="$@{CPPFLAGS@} $@{RODBC_CPPFLAGS@}"

dnl Check the headers can be found
AC_CHECK_HEADERS(sql.h sqlext.h)
if test "$@{ac_cv_header_sql_h@}" = no ||
   test "$@{ac_cv_header_sqlext_h@}" = no; then
   AC_MSG_ERROR("ODBC headers sql.h and sqlext.h not found")
fi

dnl search for a library containing an ODBC function
if test [ -n "$@{odbc_mgr@}" ] ; then
  AC_SEARCH_LIBS(SQLTables, $@{odbc_mgr@}, ,
      AC_MSG_ERROR("ODBC driver manager $@{odbc_mgr@} not found"))
else
  AC_SEARCH_LIBS(SQLTables, odbc odbc32 iodbc, ,
      AC_MSG_ERROR("no ODBC driver manager found"))
fi

dnl for 64-bit ODBC need SQL[U]LEN, and it is unclear where they are defined.
AC_CHECK_TYPES([SQLLEN, SQLULEN], , , [# include <sql.h>])
dnl for unixODBC header
AC_CHECK_SIZEOF(long, 4)

dnl substitute RODBC_CPPFLAGS and LIBS
AC_SUBST(RODBC_CPPFLAGS)
AC_SUBST(LIBS)
AC_CONFIG_HEADERS([src/config.h])
dnl and do substitution in the src/Makevars.in and src/config.h
AC_CONFIG_FILES([src/Makevars])
AC_OUTPUT
@end smallexample
@c @end cartouche
@end quotation

@noindent
where @file{src/Makevars.in} would be simply

@quotation
@example
PKG_CPPFLAGS = @@RODBC_CPPFLAGS@@
PKG_LIBS = @@LIBS@@
@end example
@end quotation

A user can then be advised to specify the location of the ODBC driver
manager files by options like (lines broken for easier reading)

@example
R CMD INSTALL \
  --configure-args='--with-odbc-include=/opt/local/include \
  --with-odbc-lib=/opt/local/lib --with-odbc-manager=iodbc' \
  RODBC
@end example

@noindent
or by setting the environment variables @code{ODBC_INCLUDE} and
@code{ODBC_LIBS}.

@node Using F9x code, Using C++ code, Configure example, Configure and cleanup
@subsection Using F9x code

@R{} assumes that source files with extension @file{.f} are fixed-form
Fortran 90 (which includes Fortran 77), and passes them to the compiler
specified by macro @samp{FC}.  On known platforms the Fortran compiler
will also accept free-form Fortran 90/95 code with extension @file{.f90}
or @file{.f95}, but those are not used by @R{} itself so this is not
required.

@vindex PKG_FCFLAGS
The same compiler is used@footnote{for versions of @R{} since 3.6.0.}
for both fixed-form and free-form Fortran code (with different file
extensions and possibly different flags).  Macro @code{PKG_FFLAGS} can
be used for package-specific flags: for the un-encountered case that both
are included in a single package and that different flags are needed for
the two forms, macro @code{PKG_FCFLAGS} is also available for free-form
Fortran.

The code used to build @R{} allows a `Fortran 90' compiler to be
selected as @samp{FC}, so platforms might be encountered which only
support Fortran 90.  However, Fortran 95 is widely supported.

Some compilers specified by @samp{FC} will accept Fortran 2003, 2008 or
2018 code: such code should still use file extension @file{.f90} or
@file{.f95}.  Most platforms use @command{gfortran} where you may need
to include @option{-std=f2003}, @option{-std=f2008} or (from version 8)
@option{-std=f2018} in @code{PKG_FFLAGS} or @code{PKG_FCFLAGS}: the
default is `GNU Fortran', Fortran 95 with non-standard extensions.  The
Oracle @command{f95} compiler `accepts some Fortran 2003/8 features'
(search for `Oracle Developer Studio 12.6: Fortran User's Guide' and
look for §4.6).  Intel Fortran has full Fortran 2008 support from
version 17.0, and some 2018 support in version 16.0 and more in version
19.0.

Modern versions of Fortran support modules, whereby compiling one source
file creates a module file which is then included in others.  (Module
files typically have a @file{.mod} extension: they do depend on the
compiler used and so should never be included in a package.)  This
creates a dependence which @command{make} will not know about and often
causes installation with a parallel make to fail.  Thus it is necessary
to add explicit dependencies to @file{src/Makevars} to tell
@command{make} the constraints on the order of compilation.  For
example, if file @file{iface.f90} creates a module @samp{iface} used by
files @file{cmi.f90} and @file{dmi.f90} then @file{src/Makevars} needs
to contain something like

@example
cmi.o dmi.o: iface.o
@end example

@noindent
Note that it is not portable (although some platforms do accept it) to
define a module of the same name in multiple source files.
@c As was done by frailtypack in 2018-12: gfortran accepted this, ODS
@c on Solaris did not.

@node Using C++ code, Using @command{cmake}, Using F9x code, Configure and cleanup
@subsection Using C++ code

@R{} can be built without a C++ compiler although one is available (but
not necessarily installed) on all known @R{} platforms.  As from @R{}
4.0.0 a C++ compiler will be selected only if it conforms to the 2011
standard (`C++11').  A minor update@footnote{Some changes are linked
from
@uref{https://isocpp.org/std/standing-documents/sd-6-sg10-feature-test-recommendations}:
there were also additional deprecations.} (`C++14') was published in
December 2014 and will be used by default as from @R{} 4.1.0 if
supported.  Further revisions `C++17' (in December 2017), and `C++20'
(with many new features in December 2020) have been published since.

What standard a C++ compiler aims to support can be hard to determine:
the value@footnote{Values @code{201103L}, @code{201402L} and
@code{201703L} are most commonly used for C++11, C++14 and C++17
respectively, but some compilers set @code{1L}.  At the time of writing
there was no official value for C++20, but some compilers are using
@code{202002L}, others @code{201703L}.} of @code{__cplusplus} may help
but some compilers use it to denote a standard which is partially
supported and some the latest standard which is (almost) fully
supported.

The webpage
@uref{https://en.cppreference.com/w/cpp/compiler_support} gives
some information on which compilers are known to support recent C++
features.

Different versions of @R{} have specified different minimum C++
standards, so for maximal portability a package should specify the
standard it requires.  In order to specify C++11 code in a package,
@file{Makevars} file (or @file{Makevars.win} or @file{Makevars.ucrt} on Windows) should include
the line

@example
CXX_STD = CXX11
@end example
@noindent
Compilation and linking will then be done with the C++11 compiler (if any).

Packages without a @file{src/Makevars} or @file{src/Makefile} file may
specify that they require C++11 for code in the @file{src} directory by
including @samp{C++11} in the @samp{SystemRequirements} field of the
@file{DESCRIPTION} file, e.g.

@example
SystemRequirements: C++11
@end example

If a package does have a @file{src/Makevars[.win]} file then setting the
make variable @samp{CXX_STD} is preferred, as it allows @command{R CMD
SHLIB} to work correctly in the package's @file{src} directory.

If a package using C++ has a @command{configure} script it is essential
that it selects the correct C++ standard, @emph{via} something like

@example
CXX11=`"$@{R_HOME@}/bin/R" CMD config CXX11`
if test -z "$CXX11"; then
  AC_MSG_ERROR([No C++11 compiler is available])
fi
CXX11STD=`"$@{R_HOME@}/bin/R" CMD config CXX11STD`
CXX="$@{CXX11@} $@{CXX11STD@}"
CXXFLAGS=`"$@{R_HOME@}/bin/R" CMD config CXX11FLAGS`
AC_LANG(C++)
@end example

@noindent
if C++11 was specified, but using @code{CXX} instead of @code{CXX11} if
no standard was specified.

If you want to compile C++ code in a subdirectory, make sure you pass
down the macros to specify the appropriate compiler, e.g.@: in
@file{src/Makevars}
@example
sublibs:
         @@(cd libs && $(MAKE) \
            CXX="$(CXX11) $(CXX11STD)" CXXFLAGS="$(CXX11FLAGS) $(CXX11PICFLAGS)")
@end example

Note that the mechanisms described here specify C++11 for code compiled
by @command{R CMD SHLIB} as used by default by @command{R CMD INSTALL}.
They do not necessarily apply if there is a @file{src/Makefile} file,
nor to compilation done in vignettes or @emph{via} other packages.

Support for a C++14 compiler (where available) has been in @R{}
since version 3.4.0.  Similar considerations to C++11 apply, with the
variables associated with the C++14 compiler using the prefix
@samp{CXX14} instead of @samp{CXX11}. For example, to use C++14 code in a
package, the package's @file{Makevars} file (or @file{Makevars.win} or @file{Makevars.ucrt} on
Windows) should include the line
@example
CXX_STD = CXX14
@end example

Essentially complete C++14 support is available from GCC 5, LLVM
@command{clang} 3.4 and currently-used versions of Apple
@command{clang}.

Code needing C++14 features can check for their presence @emph{via}
`SD-6 feature tests'@footnote{See
@uref{https://isocpp.org/std/standing-documents/sd-6-sg10-feature-test-recommendations}
or
@uref{https://en.cppreference.com/w/cpp/experimental/feature_test}.
It seems a reasonable assumption that any compiler promising some C++14
conformance will provide these---e.g.@: @command{g++} 4.9.x did but
4.8.5 did not.}.  Such a check could be

@example
#include <memory> // header where this is defined
#if defined(__cpp_lib_make_unique) && (__cpp_lib_make_unique >= 201304)
using std::make_unique;
#else
// your emulation
#endif
@end example

Note that Windows builds prior to @R{} 4.0.0 used @command{g++} 4.9.x
which had only partial C++14 support, and the flag to obtain that
support was not included in the default Windows build of @R{} --- one
could try something like

@example
CXX14="$(BINPREF)g++ $(M_ARCH)"
CXX14FLAGS="-O2 -Wall" 
CXX14STD=-std=gnu1y 
@end example

@noindent
in @file{@var{HOME}/.R/Makevars.win}.  The @command{g++} version used
from @R{} 4.0.0 supports C++14 with flag @option{-std=gnu14} and for
back-compatibility @option{-std=gnu1y}.

@c Ubuntu 16.04LTS (5.4, ESM support until 2024-04)
@c Ubuntu 14.04LTS (4.8, ESM support until 2022-04)
@c RHEL 7 (4.8, end of all support in 2028) are examples.
C++17 (as from @R{} 3.4.0) and C++20 (as from @R{} 4.0.0) can be
specified in an analogous way (replacing @code{14} by @code{17} or
@code{20}) but compiler/OS support is platform-dependent.  Some C++17
and C++20 support is available with the default builds of @R{} on macOS
and Windows as from @R{} 4.0.0.  Much of @command{g++}'s support for
C++17 needs version 7 or later: that is more recent than some
still-current Linux distributions and the OpenCSW compilers for
Solaris.

Note that C++17 or C++20 `support' does not mean complete support: use
feature tests as well as resources such as
@uref{https://en.cppreference.com/w/cpp/compiler_support},
@uref{https://gcc.gnu.org/projects/cxx-status.html} and
@uref{https://clang.llvm.org/cxx_status.html} to see if the
features you want to use are widely implemented.

A requirement of C++17 or later should always be declared in the
@samp{SystemRequirements} field (as well as in @file{src/Makevars} or
@file{src/Makefile}) so this is shown on the package's summary pages on
@acronym{CRAN} or similar.


@node Using @command{cmake},  , Using C++ code, Configure and cleanup
@subsection Using @command{cmake}

Packages often wish to include the sources of other software and compile
that for inclusion in their @file{.so} or @file{.dll}, which is normally
done by including (or unpacking) the sources in a subdirectory of
@file{src}, as considered above.

Further issues arise when the external software uses another build
system such as @command{cmake}, principally to ensure that that
@emph{all} the settings for compilers, include and load paths @emph{etc}
are made.  This section has already mentioned the need to set
at least some of
@example
CC CFLAGS CXX CXXFLAGS CPPFLAGS LDFLAGS
@end example
@noindent
@code{CFLAGS} and @code{CXXFLAGS} will need to include @code{CPICFLAGS}
and @code{CXXPICFLAGS} respectively unless (as below) @command{cmake} is
asked to generate PIC code.

Setting these (and more) as environment variables controls the behaviour
of @command{cmake}
(@uref{https://cmake.org/cmake/help/latest/manual/cmake-env-variables.7.html#manual:cmake-env-variables(7)}),
but it may be desirable to translate these into native settings such as
@example
CMAKE_C_COMPILER
CMAKE_C_FLAGS
CMAKE_CXX_COMPILER
CMAKE_CXX_FLAGS
CMAKE_INCLUDE_PATH
CMAKE_LIBRARY_PATH
CMAKE_SHARED_LINKER_FLAGS_INIT
CMAKE_OSX_DEPLOYMENT_TARGET
@end example
@noindent
and it is often necessary to ensure a static library of PIC code is built by
@example
-DBUILD_SHARED_LIBS:bool=OFF
-DCMAKE_POSITION_INDEPENDENT_CODE:bool=ON
@end example

If @R{} is to be detected or used, this must be the build being used for
package installation -- @command{"$@{R_HOME@}"/bin/R}.

To fix ideas, consider a package with sources for a library @file{myLib}
under @file{src/libs}.  Two approaches have been used.  It is often most
convenient to build the external software in a directory other than its
sources (particularly during development when the build directory can be
removed between builds rather than attempting to clean the sources) --
this is illustrated in the first approach.

@enumerate
@c taken by osqp and rbedrock
@item
Use the package's @file{configure} script to create a static library
@file{src/build/libmyLib.a}.  This can then be treated in the same way
as external software, for example having in @file{src/Makevars}
@example
PKG_CPPFLAGS = -Ilibs/include
PKG_LIBS = build/libmyLib.a
@end example
@noindent
(@code{-Lbuild -lmyLib} could also be used but this explicit
specification avoids any confusion with dynamic libraries of the same
name.)

The @file{configure} script will need to contain something like (for C
code)
@example
: $@{R_HOME=`R RHOME`@}
if test -z "$@{R_HOME@}"; then
  echo "could not determine R_HOME"
  exit 1
fi
CC=`"$@{R_HOME@}/bin/R" CMD config CC`
CFLAGS=`"$@{R_HOME@}/bin/R" CMD config CFLAGS`
CPPFLAGS=`"$@{R_HOME@}/bin/R" CMD config CPPFLAGS`
LDFLAGS=`"$@{R_HOME@}/bin/R" CMD config LDFLAGS`

cd src
mkdir build && cd build
cmake ../libs \
  -DCMAKE_BUILD_TYPE=Release \
  -DBUILD_SHARED_LIBS:bool=OFF \
  -DCMAKE_POSITION_INDEPENDENT_CODE:bool=ON
$@{MAKE@}
@end example

@c similar to symengine
@item
Use @file{src/Makevars} (or @file{src/Makevars.win} or @file{Makevars.ucrt}) to build within the
subdirectory.  This could be something like (for C code)
@example
PKG_CPPFLAGS = -Ilibs/include
PKG_LIBS = libs/libmyLib.a

$(SHLIB): mylibs

mylibs:
        (cd libs; \
          CC="$(CC)" CFLAGS="$(CFLAGS)" \
          CPPFLAGS="$(CPPFLAGS)" LDFLAGS="$(LDFLAGS)" \
	  cmake . \
	    -DCMAKE_BUILD_TYPE=Release \
	    -DBUILD_SHARED_LIBS:bool=OFF \
	    -DCMAKE_POSITION_INDEPENDENT_CODE:bool=ON; \
	  $(MAKE))

@end example
@noindent
the compiler and other settings having been set as Make variables by an
@R{} makefile included by @command{INSTALL} before @file{src/Makevars}.

@end enumerate

A complication is that on macOS @command{cmake} (where installed) is
commonly not on the path but at
@file{/Applications/CMake.app/Contents/bin/cmake}.  One way to work
around this is for the package's @file{configure} script to include
@example
if test -z "$CMAKE"; then CMAKE="`which cmake`"; fi
if test -z "$CMAKE"; then CMAKE=/Applications/CMake.app/Contents/bin/cmake; fi
if test -f "$CMAKE"; then echo "no 'cmake' command found"; exit 1; fi
@end example
@noindent
and for the second approach to substitute @env{CMAKE} into
@file{src//Makevars}.

@node Checking and building packages, Writing package vignettes, Configure and cleanup, Creating R packages
@section Checking and building packages

Before using these tools, please check that your package can be
installed.  @code{R CMD check} will @emph{inter alia} do this, but you
may get more detailed error messages doing the install directly.

@menu
* Checking packages::
* Building package tarballs::
* Building binary packages::
@end menu

If your package specifies an encoding in its @file{DESCRIPTION} file,
you should run these tools in a locale which makes use of that encoding:
they may not work at all or may work incorrectly in other locales
(although UTF-8 locales will most likely work).

@quotation Note
@code{R CMD check} and @code{R CMD build} run @R{} processes with
@option{--vanilla} in which none of the user's startup files are read.
If you need @env{R_LIBS} set (to find packages in a non-standard
library) you can set it in the environment: also you can use the check
and build environment files (as specified by the environment variables
@env{R_CHECK_ENVIRON} and @env{R_BUILD_ENVIRON}; if unset,
files@footnote{On systems which use sub-architectures,
architecture-specific versions such as @file{~/.R/check.Renviron.i386}
take precedence.}  @file{~/.R/check.Renviron} and
@file{~/.R/build.Renviron} are used) to set environment variables when
using these utilities.
@end quotation

@quotation Note to Windows users
@code{R CMD build} may make use of the Windows toolset (see the ``R
Installation and Administration'' manual) if present and in your path,
and it is required for packages which need it to install (including
those with @file{configure.win},  @file{cleanup.win}, @file{configure.ucrt}
or @file{cleanup.ucrt} scripts or a
@file{src} directory) and e.g.@: need vignettes built.

You may need to set the environment variable @env{TMPDIR} to point to a
suitable writable directory with a path not containing spaces -- use
forward slashes for the separators.  Also, the directory needs to be on
a case-honouring file system (some network-mounted file systems are
not).
@end quotation


@node Checking packages, Building package tarballs, Checking and building packages, Checking and building packages
@subsection Checking packages
@cindex Checking packages

@findex R CMD check
Using @code{R CMD check}, the @R{} package checker, one can test whether
@emph{source} @R{} packages work correctly.  It can be run on one or
more directories, or compressed package @command{tar} archives with
extension @file{.tar.gz}, @file{.tgz}, @file{.tar.bz2} or
@file{.tar.xz}.

It is strongly recommended that the final checks are run on a
@command{tar} archive prepared by @command{R CMD build}.

This runs a series of checks, including

@enumerate
@item
The package is installed.  This will warn about missing cross-references
and duplicate aliases in help files.

@item
The file names are checked to be valid across file systems and supported
operating system platforms.

@item
The files and directories are checked for sufficient permissions
(Unix-alikes only).

@item
The files are checked for binary executables, using a suitable version
of @command{file} if available@footnote{A suitable @command{file.exe} is
part of the Windows toolset: it checks for @command{gfile} if a suitable
@command{file} is not found: the latter is available in the OpenCSW
collection for Solaris at @uref{https://www.opencsw.org/}.  The source
repository is @uref{http://ftp.astron.com/pub/file/}.}.  (There may be
rare false positives.)

@item
The @file{DESCRIPTION} file is checked for completeness, and some of its
entries for correctness.  Unless installation tests are skipped,
checking is aborted if the package dependencies cannot be resolved at
run time.  (You may need to set @env{R_LIBS} in the environment if
dependent packages are in a separate library tree.) One check is that
the package name is not that of a standard package, nor one of the
defunct standard packages (@samp{ctest}, @samp{eda}, @samp{lqs},
@samp{mle}, @samp{modreg}, @samp{mva}, @samp{nls}, @samp{stepfun} and
@samp{ts}).  Another check is that all packages mentioned in
@code{library} or @code{require}s or from which the @file{NAMESPACE}
file imports or are called @emph{via} @code{::} or @code{:::} are listed
(in @samp{Depends}, @samp{Imports}, @samp{Suggests}): this is not an
exhaustive check of the actual imports.

@item
Available index information (in particular, for demos and vignettes) is
checked for completeness.

@item
The package subdirectories are checked for suitable file names and for
not being empty.  The checks on file names are controlled by the option
@option{--check-subdirs=@var{value}}.  This defaults to @samp{default},
which runs the checks only if checking a tarball: the default can be
overridden by specifying the value as @samp{yes} or @samp{no}.  Further,
the check on the @file{src} directory is only run if the package
does not contain a @file{configure} script (which corresponds to the
value @samp{yes-maybe}) and there is no @file{src/Makefile} or
@file{src/Makefile.in}.

To allow a @file{configure} script to generate suitable files, files
ending in @samp{.in} will be allowed in the @file{R} directory.

A warning is given for directory names that look like @R{} package check
directories -- many packages have been submitted to @acronym{CRAN}
containing these.

@item
The @R{} files are checked for syntax errors.  Bytes which are
non-@acronym{ASCII} are reported as warnings, but these should be
regarded as errors unless it is known that the package will always be
used in the same locale.

@item
It is checked that the package can be loaded, first with the usual
default packages and then only with package @pkg{base} already
loaded. It is checked that the namespace can be loaded in an empty
session with only the @pkg{base} namespace loaded.  (Namespaces and
packages can be loaded very early in the session, before the default
packages are available, so packages should work then.)

@item
The @R{} files are checked for correct calls to @code{library.dynam}.
Package startup functions are checked for correct argument lists and
(incorrect) calls to functions which modify the search path or
inappropriately generate messages.  The @R{} code is checked for
possible problems using @CRANpkg{codetools}.  In addition, it is checked
whether S3 methods have all the arguments of the corresponding generic, and
whether the final argument of replacement functions is called
@samp{value}.  All foreign function calls (@code{.C}, @code{.Fortran},
@code{.Call} and @code{.External} calls) are tested to see if they have
a @code{PACKAGE} argument, and if not, whether the appropriate DLL might
be deduced from the namespace of the package.  Any other calls are
reported.  (The check is generous, and users may want to supplement this
by examining the output of @code{tools::checkFF("mypkg", verbose=TRUE)},
especially if the intention were to always use a @code{PACKAGE}
argument)

@item
The @file{Rd} files are checked for correct syntax and metadata,
including the presence of the mandatory fields (@code{\name}, @code{\alias},
@code{\title} and @code{\description}).  The @file{Rd} name and
title are checked for being non-empty, and there is a check for missing
cross-references (links).

@item
A check is made for missing documentation entries, such as undocumented
user-level objects in the package.

@item
Documentation for functions, data sets, and S4 classes is checked for
consistency with the corresponding code.

@item
It is checked whether all function arguments given in @code{\usage}
sections of @file{Rd} files are documented in the corresponding
@code{\arguments} section.

@item
The @file{data} directory is checked for non-@acronym{ASCII} characters
and for the use of reasonable levels of compression.

@item
C, C++ and Fortran source and header files@footnote{An exception is made
for subdirectories with names starting @samp{win} or @samp{Win}.} are
tested for portable (LF-only) line endings.  If there is a
@file{Makefile} or @file{Makefile.in} or @file{Makevars} or
@file{Makevars.in} file under the @file{src} directory, it is checked
for portable line endings and the correct use of @samp{$(BLAS_LIBS)} and
@samp{$(LAPACK_LIBS)}

Compiled code is checked for symbols corresponding to functions which
might terminate @R{} or write to @file{stdout}/@file{stderr} instead of
the console.  Note that the latter might give false positives in that
the symbols might be pulled in with external libraries and could never
be called.  Windows@footnote{on most other platforms such runtime
libraries are dynamic, but static libraries are currently used on
Windows because the toolchain is not a standard part of the OS.} users
should note that the Fortran and C++ runtime libraries are examples of
such external libraries.

@item
Some checks are made of the contents of the @file{inst/doc} directory.
These always include checking for files that look like leftovers, and if
suitable tools (such as @command{qpdf}) are available, checking that the
PDF documentation is of minimal size.

@item
The examples provided by the package's documentation are run.
(@pxref{Writing R documentation files}, for information on using
@code{\examples} to create executable example code.)  If there is a file
@file{tests/Examples/@var{pkg}-Ex.Rout.save}, the output of running the
examples is compared to that file.

Of course, released packages should be able to run at least their own
examples.  Each example is run in a `clean' environment (so earlier
examples cannot be assumed to have been run), and with the variables
@code{T} and @code{F} redefined to generate an error unless they are set
in the example: @xref{Logical vectors, , Logical vectors, R-intro, An
Introduction to R}.

@item
If the package sources contain a @file{tests} directory then the tests
specified in that directory are run.  (Typically they will consist of a
set of @file{.R} source files and target output files
@file{.Rout.save}.)  Please note that the comparison will be done in the
end user's locale, so the target output files should be @acronym{ASCII}
if at all possible.  (The command line option @code{--test-dir=foo} may
be used to specify tests in a non-standard location.  For example,
unusually slow tests could be placed in @file{inst/slowTests} and then
@code{R CMD check --test-dir=inst/slowTests} would be used to run them.
Other names that have been suggested are, for example,
@file{inst/testWithOracle} for tests that require Oracle to be installed,
@file{inst/randomTests} for tests which use random values and may
occasionally fail by chance, etc.)

@item
The code in package vignettes (@pxref{Writing package vignettes}) is
executed, and the vignette PDFs re-made from their sources as a check of
completeness of the sources (unless there is a @samp{BuildVignettes}
field in the package's @file{DESCRIPTION} file with a false value).  If
there is a target output file @file{.Rout.save} in the vignette source
directory, the output from running the code in that vignette is compared
with the target output file and any differences are reported (but not
recorded in the log file).  (If the vignette sources are in the
deprecated location @file{inst/doc}, do mark such target output files to
not be installed in @file{.Rinstignore}.)

If there is an error@footnote{or if option @option{--use-valgrind} is
used or environment variable @env{_R_CHECK_ALWAYS_LOG_VIGNETTE_OUTPUT_}
is set to a true value or if there are differences from a target output
file} in executing the @R{} code in vignette @file{@var{foo.ext}}, a log
file @file{@var{foo.ext}.log} is created in the check directory.  The
vignette PDFs are re-made in a copy of the package sources in the
@file{vign_test} subdirectory of the check directory, so for further
information on errors look in directory
@file{@var{pkgname}/vign_test/vignettes}.  (It is only retained if there
are errors or if environment variable @env{_R_CHECK_CLEAN_VIGN_TEST_} is
set to a false value.)

@item
The PDF version of the package's manual is created (to check that the
@file{Rd} files can be converted successfully).  This needs @LaTeX{} and
suitable fonts and @LaTeX{} packages to be installed.
@ifset UseExternalXrefs
@xref{Making the manuals, , Making the manuals,
R-admin, R Installation and Administration}.
@end ifset
@ifclear UseExternalXrefs
See the section `Making the manuals' in the `R Installation and
Administration' manual' for further details.
@end ifclear

@end enumerate

All these tests are run with collation set to the @code{C} locale, and
for the examples and tests with environment variable @env{LANGUAGE=en}:
this is to minimize differences between platforms.

Use @kbd{R CMD check --help} to obtain more information about the usage
of the @R{} package checker.  A subset of the checking steps can be
selected by adding command-line options. It also allows customization by
setting environment variables @w{@env{_R_CHECK_*_}} as described in
@ifset UseExternalXrefs
@ref{Tools, , Tools, R-ints, R Internals}:
@end ifset
@ifclear UseExternalXrefs
`R Internals':
@end ifclear
a set of these customizations similar to those used by @acronym{CRAN}
can be selected by the option @option{--as-cran} (which works best if
Internet access is available).  Some Windows users may
need to set environment variable @env{R_WIN_NO_JUNCTIONS} to a non-empty
value.  The test of cyclic declarations@footnote{For example, in early
2014 @CRANpkg{gdata} declared @samp{Imports: gtools} and @CRANpkg{gtools}
declared @samp{Imports: gdata}.}in @file{DESCRIPTION} files needs
repositories (including @acronym{CRAN}) set: do this in
@file{~/.Rprofile}, by e.g.@:
@example
options(repos = c(CRAN="https://cran.r-project.org"))
@end example

One check customization which can be revealing is
@example
_R_CHECK_CODETOOLS_PROFILE_="suppressLocalUnused=FALSE"
@end example
@noindent
which reports unused local assignments.  Not only does this point out
computations which are unnecessary because their results are unused, it
also can uncover errors.  (Two such are to intend to update an object by
assigning a value but mistype its name or assign in the wrong scope,
for example using @code{<-} where @code{<<-} was intended.)  This can
give false positives, most commonly because of non-standard evaluation
for formulae and because the intention is to return objects in the
environment of a function for later use.

Complete checking of a package which contains a file @file{README.md}
needs a reasonably current version of  @command{pandoc} installed: see
@uref{https://pandoc.org/installing.html}.

You do need to ensure that the package is checked in a suitable locale
if it contains non-@acronym{ASCII} characters.  Such packages are likely
to fail some of the checks in a @code{C} locale, and @command{R CMD
check} will warn if it spots the problem.  You should be able to check
any package in a UTF-8 locale (if one is available).  Beware that
although a @code{C} locale is rarely used at a console, it may be the
default if logging in remotely or for batch jobs.

@quotation Multiple sub-architectures
On systems which support multiple sub-architectures (principally
Windows), @command{R CMD check} will install and check a package which
contains compiled code under all available sub-architectures. (Use
option @option{--force-multiarch} to force this for packages without
compiled code, which are otherwise only checked under the main
sub-architecture.)  This will run the loading tests, examples and
@file{tests} directory under each installed sub-architecture in turn,
and give an error if any fail.  Where environment variables (including
perhaps @env{PATH}) need to be set differently for each
sub-architecture, these can be set in architecture-specific files such
as @file{@var{R_HOME}/etc/i386/Renviron.site}.

An alternative approach is to use @command{R CMD check --no-multiarch}
to check the primary sub-architecture, and then to use something like
@command{R --arch=x86_64 CMD check --extra-arch} or (Windows)
@command{/path/to/R/bin/x64/Rcmd check --extra-arch} to run for each
additional sub-architecture just the checks@footnote{loading, examples,
tests, running vignette code} which differ by sub-architecture.  (This
approach is required for packages which are installed by @command{R CMD
INSTALL --merge-multiarch}.)

Where packages need additional commands to install all the
sub-architectures these can be supplied by e.g.@:
@option{--install-args=--force-biarch}.

@end quotation


@node Building package tarballs, Building binary packages, Checking packages, Checking and building packages
@subsection Building package tarballs
@cindex Building source packages

@findex R CMD build
@cindex Package builder
@cindex tarballs
Packages may be distributed in source form as ``tarballs''
(@file{.tar.gz} files) or in binary form. The source form can be
installed on all platforms with suitable tools and is the usual form for
Unix-like systems; the binary form is platform-specific, and is the more
common distribution form for the Windows and macOS platforms.

Using @command{R CMD build}, the @R{} package builder, one can build
@R{} package tarballs from their sources (for example, for subsequent
release).  It is recommended that packages are built for release by the
current release version of @R{} or @samp{r-patched}, to avoid
inadvertently picking up new features of a development version of @R{}.

Prior to actually building the package in the standard gzipped tar file
format, a few diagnostic checks and cleanups are performed.  In
particular, it is tested whether object indices exist and can be assumed
to be up-to-date, and C, C++ and Fortran source files and relevant
makefiles in a @file{src} directory are tested and converted to LF
line-endings if necessary.

Run-time checks whether the package works correctly should be performed
using @command{R CMD check} prior to invoking the final build procedure.

@cindex .Rbuildignore file
To exclude files from being put into the package, one can specify a list
of exclude patterns in file @file{.Rbuildignore} in the top-level source
directory.  These patterns should be Perl-like regular expressions (see
the help for @code{regexp} in @R{} for the precise details), one per
line, to be matched case-insensitively against the file and directory
names relative to the top-level package source directory.  In addition,
directories from source control systems@footnote{called @file{CVS} or
@file{.svn} or @file{.arch-ids} or @file{.bzr} or @file{.git} (but not
files called @file{.git}) or @file{.hg}.} or from
@command{eclipse}@footnote{called @file{.metadata}.}, directories with
names @file{check}, @file{chm}, or ending @file{.Rcheck} or @file{Old}
or @file{old} and files
@file{GNUMakefile}@footnote{which is an error: GNU make uses
@file{GNUmakefile}.}, @file{Read-and-delete-me} or with base names
starting with @samp{.#}, or starting and ending with @samp{#}, or ending
in @samp{~}, @samp{.bak} or @samp{.swp}, are excluded by
default@footnote{see @code{tools:::.hidden_file_exclusions} and
@code{tools:::get_exclude_patterns()} for further
excluded files and file patterns, respectively.}.  In addition,
same-package tarballs (from previous builds) and their binary forms will
be excluded from the top-level directory, as well as
those files in the @file{R}, @file{demo} and @file{man}
directories which are flagged by @command{R CMD check} as having invalid
names.

Use @kbd{R CMD build --help} to obtain more information about the usage
of the @R{} package builder.

@c DESCRIPTION field BuildVignettes
Unless @kbd{R CMD build} is invoked with the
@option{--no-build-vignettes} option (or the package's
@file{DESCRIPTION} contains @samp{BuildVignettes: no} or similar), it
will attempt to (re)build the vignettes (@pxref{Writing package
vignettes}) in the package.  To do so it installs the current package
into a temporary library tree, but any dependent packages need to be
installed in an available library tree (see the Note: at the top of this
section).

@c DESCRIPTION field BuildManual
Similarly, if the @file{.Rd} documentation files contain any
@code{\Sexpr} macros (@pxref{Dynamic pages}), the package will be
temporarily installed to execute them.  Post-execution binary copies of
those pages containing build-time macros will be saved in
@file{build/partial.rdb}.  If there are any install-time or render-time
macros, a @file{.pdf} version of the package manual will be built and
installed in the @file{build} subdirectory.  (This allows
@acronym{CRAN} or other repositories to display the manual even if they
are unable to install the package.)  This can be suppressed by the
option @option{--no-manual} or if package's @file{DESCRIPTION} contains
@samp{BuildManual: no} or similar.

@c DESCRIPTION field BuildKeepEmpty
One of the checks that @command{R CMD build} runs is for empty source
directories.  These are in most (but not all) cases unintentional, if
they are intentional use the option @option{--keep-empty-dirs} (or set
the environment variable @env{_R_BUILD_KEEP_EMPTY_DIRS_} to @samp{TRUE},
or have a @samp{BuildKeepEmpty} field with a true value in the
@file{DESCRIPTION} file).

@c DESCRIPTION field BuildResaveData
The @option{--resave-data} option allows saved images (@file{.rda} and
@file{.RData} files) in the @file{data} directory to be optimized for
size.  It will also compress tabular files and convert @file{.R} files
to saved images.  It can take values @code{no}, @code{gzip} (the default
if this option is not supplied, which can be changed by setting the
environment variable @env{_R_BUILD_RESAVE_DATA_}) and @code{best}
(equivalent to giving it without a value), which chooses the most
effective compression.  Using @code{best} adds a dependence on @code{R
(>= 2.10)} to the @file{DESCRIPTION} file if @command{bzip2} or
@command{xz} compression is selected for any of the files.  If this is
thought undesirable, @option{--resave-data=gzip} (which is the default
if that option is not supplied) will do what compression it can with
@command{gzip}.  A package can control how its data is resaved by
supplying a @samp{BuildResaveData} field (with one of the values given
earlier in this paragraph) in its @file{DESCRIPTION} file.

The @option{--compact-vignettes} option will run
@code{tools::compactPDF} over the PDF files in @file{inst/doc} (and its
subdirectories) to losslessly compress them.  This is not enabled by
default (it can be selected by environment variable
@env{_R_BUILD_COMPACT_VIGNETTES_}) and needs @command{qpdf}
(@uref{http://qpdf.sourceforge.net/}) to be available.

It can be useful to run @command{R CMD check --check-subdirs=yes} on the
built tarball as a final check on the contents.

Where a non-POSIX file system is in use which does not utilize execute
permissions, some care is needed with permissions.  This applies on
Windows and to e.g.@: FAT-formatted drives and SMB-mounted file systems
on other OSes.  The `mode' of the file recorded in the tarball will be
whatever @code{file.info()} returns.  On Windows this will record only
directories as having execute permission and on other OSes it is likely
that all files have reported `mode' @code{0777}.  A particular issue is
packages being built on Windows which are intended to contain executable
scripts such as @file{configure} and @file{cleanup}: @command{R CMD
build} ensures those two are recorded with execute permission.

Directory @file{build} of the package sources is reserved for use by
@command{R CMD build}: it contains information which may not easily be
created when the package is installed, including index information on
the vignettes and, rarely, information on the help pages and perhaps a
copy of the PDF reference manual (see above).

@node Building binary packages,  , Building package tarballs, Checking and building packages
@subsection Building binary packages
@cindex Building binary packages

Binary packages are compressed copies of installed versions of
packages.  They contain compiled shared libraries rather than C, C++ or
Fortran source code, and the R functions are included in their installed
form.  The format and filename are platform-specific; for example, a
binary package for Windows is usually supplied as a @file{.zip} file,
and for the macOS platform the default binary package file extension is
@file{.tgz}.

The recommended method of building binary packages is to use

@command{R CMD INSTALL --build pkg}
@noindent
where @file{pkg} is either the name of a source tarball (in the usual
@file{.tar.gz} format) or the location of the directory of the package
source to be built.  This operates by first installing the package and
then packing the installed binaries into the appropriate binary package
file for the particular platform.

By default, @command{R CMD INSTALL --build} will attempt to install the
package into the default library tree for the local installation of
@R{}. This has two implications:

@itemize @bullet
@item
If the installation is successful, it will overwrite any existing installation
of the same package.

@item
The default library tree must have write permission; if not, the package will
not install and the binary will not be created.

@end itemize

@noindent
To prevent changes to the present working installation or to provide an
install location with write access, create a suitably located directory
with write access and use the @command{-l} option to build the package
in the chosen location.  The usage is then

@command{R CMD INSTALL -l location --build pkg}

@noindent
where @file{location} is the chosen directory with write access. The package
will be installed as a subdirectory of @file{location}, and the package binary
will be created in the current directory.

Other options for @command{R CMD INSTALL} can be found using @command{R
CMD INSTALL --help}, and platform-specific details for special cases are
discussed in the platform-specific FAQs.


@c In much earlier versions of @R{}, @command{R CMD build --binary} could
@c build a binary version of a package, but this approach is now deprecated
@c in favour of @command{R CMD INSTALL --build}.


Finally, at least one web-based service is available for building binary
packages from (checked) source code: WinBuilder (see
@uref{https://win-builder.R-project.org/}) is able to build Windows
binaries. Note that this is intended for developers on other platforms
who do not have access to Windows but wish to provide binaries for the
Windows platform.

@node Writing package vignettes, Package namespaces, Checking and building packages, Creating R packages
@section Writing package vignettes
@cindex vignettes
@cindex Sweave

@menu
* Encodings and vignettes::
* Non-Sweave vignettes::
@end menu

In addition to the help files in @file{Rd} format, @R{} packages allow
the inclusion of documents in arbitrary other formats.  The standard
location for these is subdirectory @file{inst/doc} of a source package,
the contents will be copied to subdirectory @file{doc} when the package
is installed.  Pointers from package help indices to the installed
documents are automatically created.  Documents in @file{inst/doc} can
be in arbitrary format, however we strongly recommend providing them in
PDF format, so users on almost all platforms can easily read them.  To
ensure that they can be accessed from a browser (as an @HTML{} index is
provided), the file names should start with an @acronym{ASCII} letter
and be comprised entirely of @acronym{ASCII} letters or digits or hyphen
or underscore.

A special case is @emph{package vignettes}.  Vignettes are documents in
PDF or @HTML{} format obtained from plain-text literate source files
from which @R{} knows how to extract @R{} code and create output (in
PDF/@HTML{} or intermediate @LaTeX{}).  Vignette engines do this work,
using ``tangle'' and ``weave'' functions respectively.  Sweave, provided
by the R distribution, is the default engine.  Other vignette engines
besides Sweave are supported; see @ref{Non-Sweave vignettes}.

Package vignettes have their sources in subdirectory @file{vignettes} of
the package sources.  Note that the location of the vignette sources
only affects @command{R CMD build} and @command{R CMD check}: the
tarball built by @command{R CMD build} includes in @file{inst/doc} the
components intended to be installed.

Sweave vignette sources are normally given the file extension
@file{.Rnw} or @file{.Rtex}, but for historical reasons
extensions@footnote{and to avoid problems with case-insensitive file
systems, lower-case versions of all these extensions.} @file{.Snw} and
@file{.Stex} are also recognized.  Sweave allows the integration of
@LaTeX{} documents: see the @code{Sweave} help page in @R{} and the
@code{Sweave} vignette in package @pkg{utils} for details on the
source document format.

Package vignettes are tested by @code{R CMD check} by executing all @R{}
code chunks they contain (except those marked for non-evaluation, e.g.,
with option @code{eval=FALSE} for Sweave).  The @R{} working directory
for all vignette tests in @code{R CMD check} is a @emph{copy} of the
vignette source directory.  Make sure all files needed to run the @R{}
code in the vignette (data sets, @dots{}) are accessible by either
placing them in the @file{inst/doc} hierarchy of the source package or
by using calls to @code{system.file()}.  All other files needed to
re-make the vignettes (such as @LaTeX{} style files, Bib@TeX{} input
files and files for any figures not created by running the code in the
vignette) must be in the vignette source directory.  @code{R CMD check} 
will check that vignette production has succeeded by comparing
modification times of output files in @file{inst/doc} with
the source in @file{vignettes}.

@code{R CMD build} will automatically@footnote{unless inhibited by using
@samp{BuildVignettes: no} in the @file{DESCRIPTION} file.} create the
(PDF or @HTML{} versions of the) vignettes in @file{inst/doc} for
distribution with the package sources.  By including the vignette
outputs in the package sources it is not necessary that these can be
re-built at install time, i.e., the package author can use private @R{}
packages, screen snapshots and @LaTeX{} extensions which are only
available on their machine.@footnote{provided the conditions of the
package's license are met: many, including @acronym{CRAN}, see the
omission of source components as incompatible with an Open Source
license.}

By default @code{R CMD build} will run @code{Sweave} on all Sweave
vignette source files in @file{vignettes}.  If @file{Makefile} is found
in the vignette source directory, then @code{R CMD build} will try to
run @command{make} after the @code{Sweave} runs, otherwise
@code{texi2pdf} is run on each @file{.tex} file produced.

The first target in the @file{Makefile} should take care of both
creation of PDF/@HTML{} files and cleaning up afterwards (including
after @code{Sweave}), i.e., delete all files that shall not appear in
the final package archive.  Note that if the @code{make} step runs @R{}
it needs to be careful to respect the environment values of @env{R_LIBS}
and @env{R_HOME}@footnote{@code{R_HOME/bin} is prepended to the
@env{PATH} so that references to @command{R} or @command{Rscript} in the
@file{Makefile} do make use of the currently running version of @R{}.}.
Finally, if there is a @file{Makefile} and it has a @samp{clean:}
target, @command{make clean} is run.

All the usual @emph{caveats} about including a @file{Makefile} apply.
It must be portable (no @acronym{GNU} extensions), use LF line endings
and must work correctly with a parallel @command{make}: too many authors
have written things like

@example
## BAD EXAMPLE
all: pdf clean

pdf: ABC-intro.pdf ABC-details.pdf

%.pdf:  %.tex
        texi2dvi --pdf $*

clean:
        rm *.tex ABC-details-*.pdf
@end example

@noindent
which will start removing the source files whilst @command{pdflatex} is
working.

Metadata lines can be placed in the source file, preferably in @LaTeX{}
comments in the preamble.  One such is a @code{\VignetteIndexEntry} of
the form
@example
%\VignetteIndexEntry@{Using Animal@}
@end example
@noindent
Others you may see are @code{\VignettePackage} (currently ignored),
@code{\VignetteDepends} and @code{\VignetteKeyword} (which replaced
@code{\VignetteKeywords}).  These are processed at package installation
time to create the saved data frame @file{Meta/vignette.rds}, but only
the @code{\VignetteIndexEntry} and @code{\VignetteKeyword} statements
are currently used.  The @code{\VignetteEngine} statement
is described in @ref{Non-Sweave vignettes}.

At install time an @HTML{} index for all vignettes in the package is
automatically created from the @code{\VignetteIndexEntry} statements
unless a file @file{index.html} exists in directory
@file{inst/doc}. This index is linked from the @HTML{} help index for
the package.  If you do supply a @file{inst/doc/index.html} file it
should contain relative links only to files under the installed
@file{doc} directory, or perhaps (not really an index) to @HTML{} help
files or to the @file{DESCRIPTION} file, and be valid @HTML{} as
confirmed @emph{via} the @uref{https://validator.w3.org, W3C Markup
Validation Service} or @uref{https://validator.nu/, Validator.nu}.

Sweave/Stangle allows the document to specify the @code{split=TRUE}
option to create a single @R{} file for each code chunk: this will not
work for vignettes where it is assumed that each vignette source
generates a single file with the vignette extension replaced by
@file{.R}.

Do watch that PDFs are not too large -- one in a @acronym{CRAN} package
was 72MB!  This is usually caused by the inclusion of overly detailed
figures, which will not render well in PDF viewers.  Sometimes it is
much better to generate fairly high resolution bitmap (PNG, JPEG)
figures and include those in the PDF document.

@cindex .install_extras file
When @command{R CMD build} builds the vignettes, it copies these and
the vignette sources from directory @file{vignettes} to @file{inst/doc}.
To install any other files from the @file{vignettes} directory, include
a file @file{vignettes/.install_extras} which specifies these as
Perl-like regular expressions on one or more lines.  (See the
description of the @file{.Rinstignore} file for full details.)


@node Encodings and vignettes, Non-Sweave vignettes, Writing package vignettes, Writing package vignettes
@subsection Encodings and vignettes

Vignettes will in general include descriptive text, @R{} input, @R{}
output and figures, @LaTeX{} include files and bibliographic references.
As any of these may contain non-@acronym{ASCII} characters, the handling
of encodings can become very complicated.

The vignette source file should be written in @acronym{ASCII} or contain
a declaration of the encoding (see below).  This applies even to
comments within the source file, since vignette engines process comments
to look for options and metadata lines.  When an engine's weave and
tangle functions are called on the vignette source, it will be converted
to the encoding of the current @R{} session.

@code{Stangle()} will produce an @R{} code file in the current locale's
encoding: for a non-@acronym{ASCII} vignette what that is is recorded in a
comment at the top of the file.

@code{Sweave()} will produce a @file{.tex} file in the current
encoding, or in UTF-8 if that is declared.  Non-@acronym{ASCII} encodings
need to be declared to @LaTeX{} via a line like
@example
\usepackage[utf8]@{inputenc@}
@end example
@noindent
(It is also possible to use the more recent @samp{inputenx} @LaTeX{}
package.) For files where this line is not needed (e.g.@: chapters
included within the body of a larger document, or non-Sweave
vignettes), the encoding may be declared using a comment like
@example
%\VignetteEncoding@{UTF-8@}
@end example
@noindent
If the encoding is UTF-8, this can also be declared using
the declaration
@example
%\SweaveUTF8
@end example
@noindent
If no declaration is given in the vignette, it will be assumed to be
in the encoding declared for the package.  If there is no encoding
declared in either place, then it is an error to use non-@acronym{ASCII}
characters in the vignette.

In any case, be aware that @LaTeX{} may require the @samp{usepackage}
declaration.

@code{Sweave()} will also parse and evaluate the @R{} code in each
chunk.  The @R{} output will also be in the current locale (or @acronym{UTF-8}
if so declared), and should
be covered by the @samp{inputenc} declaration.  One thing people often
forget is that the @R{} output may not be @acronym{ASCII} even for
@acronym{ASCII} @R{} sources, for many possible reasons.  One common one
is the use of `fancy' quotes: see the @R{} help on @code{sQuote}: note
carefully that it is not portable to declare UTF-8 or CP1252 to cover
such quotes, as their encoding will depend on the locale used to run
@code{Sweave()}:  this can be circumvented by setting
@code{options(useFancyQuotes="UTF-8")} in the vignette.

The final issue is the encoding of figures -- this applies only to PDF
figures and not PNG etc.  The PDF figures will contain declarations for
their encoding, but the Sweave option @code{pdf.encoding} may need to be
set appropriately: see the help for the @code{pdf()} graphics device.

As a real example of the complexities, consider the @CRANpkg{fortunes}
package version @samp{1.4-0}.  That package did not have a declared
encoding, and its vignette was in @acronym{ASCII}.  However, the data it
displays are read from a UTF-8 CSV file and will be assumed to be in the
current encoding, so @file{fortunes.tex} will be in UTF-8 in any locale.
Had @code{read.table} been told the data were UTF-8, @file{fortunes.tex}
would have been in the locale's encoding.

@node Non-Sweave vignettes,  , Encodings and vignettes, Writing package vignettes
@subsection Non-Sweave vignettes

Vignettes in formats other than Sweave are supported @emph{via}
``vignette engines''.  For example @CRANpkg{knitr} version 1.1 or later
can create @file{.tex} files from a variation on Sweave format, and
@file{.html} files from a variation on ``markdown'' format. These
engines replace the @code{Sweave()} function with other functions to
convert vignette source files into @LaTeX{} files for processing into
@file{.pdf}, or directly into @file{.pdf} or @file{.html} files. The
@code{Stangle()} function is replaced with a function that extracts the
@R{} source from a vignette.

@R{} recognizes non-Sweave vignettes using filename extensions specified
by the engine.  For example, the @CRANpkg{knitr} package supports
the extension @file{.Rmd} (standing for
``R markdown'').  The user indicates the vignette engine
within the vignette source using a @code{\VignetteEngine} line, for example
@example
%\VignetteEngine@{knitr::knitr@}
@end example
@noindent
This specifies the name of a package and an engine to use in place of
Sweave in processing the vignette.  As @code{Sweave} is the only engine
supplied with the @R{} distribution, the package providing any other
engine must be specified in the @samp{VignetteBuilder} field of the
package @file{DESCRIPTION} file, and also specified in the
@samp{Suggests}, @samp{Imports} or @samp{Depends} field (since its
namespace must be available to build or check your package).  If more
than one package is specified as a builder, they will be searched in the
order given there.  The @pkg{utils} package is always implicitly
appended to the list of builder packages, but may be included earlier
to change the search order.

Note that a package with non-Sweave vignettes should always have a
@samp{VignetteBuilder} field in the @file{DESCRIPTION} file, since this
is how @command{R CMD check} recognizes that there are vignettes to be
checked: packages listed there are required when the package is checked.

The vignette engine can produce @file{.tex}, @file{.pdf}, or @file{.html}
files as output.  If it produces @file{.tex} files, @R{} will
call @code{texi2pdf} to convert them to @file{.pdf} for display
to the user (unless there is a @file{Makefile} in the @file{vignettes}
directory).

Package writers who would like to supply vignette engines need
to register those engines in the package @code{.onLoad} function.
For example, that function could make the call
@example
tools::vignetteEngine("knitr", weave = vweave, tangle = vtangle,
                      pattern = "[.]Rmd$", package = "knitr")
@end example
@noindent
(The actual registration in @CRANpkg{knitr} is more complicated, because
it supports other input formats.)  See the @code{?tools::vignetteEngine}
help topic for details on engine registration.


@node Package namespaces, Writing portable packages, Writing package vignettes, Creating R packages
@section Package namespaces
@cindex namespaces

@R{} has a namespace management system for code in packages.  This
system allows the package writer to specify which variables in the
package should be @emph{exported} to make them available to package
users, and which variables should be @emph{imported} from other
packages.

The namespace for a package is specified by the
@file{NAMESPACE} file in the top level package directory.  This file
contains @emph{namespace directives} describing the imports and exports
of the namespace.  Additional directives register any shared objects to
be loaded and any S3-style methods that are provided.  Note that
although the file looks like @R{} code (and often has @R{}-style
comments) it is not processed as @R{} code.  Only very simple
conditional processing of @code{if} statements is implemented.

Packages are loaded and attached to the search path by calling
@code{library} or @code{require}.  Only the exported variables are
placed in the attached frame.  Loading a package that imports variables
from other packages will cause these other packages to be loaded as well
(unless they have already been loaded), but they will @emph{not} be
placed on the search path by these implicit loads.  Thus code in the
package can only depend on objects in its own namespace and its imports
(including the @pkg{base} namespace) being visible@footnote{Note that
lazy-loaded datasets are @emph{not} in the package's namespace so need
to be accessed @emph{via} @code{::}, e.g.@:
@code{survival::survexp.us}.}.

Namespaces are @emph{sealed} once they are loaded.  Sealing means that
imports and exports cannot be changed and that internal variable
bindings cannot be changed.  Sealing allows a simpler implementation
strategy for the namespace mechanism and allows code analysis and
compilation tools to accurately identify the definition corresponding to
a global variable reference in a function body.

The namespace controls the search strategy for variables used by
functions in the package.  If not found locally, @R{} searches the
package namespace first, then the imports, then the base namespace and
then the normal search path (so the base namespace precedes the normal
search rather than being at the end of it).


@menu
* Specifying imports and exports::
* Registering S3 methods::
* Load hooks::
* useDynLib::
* An example::
* Namespaces with S4 classes and methods::
@end menu


@node Specifying imports and exports, Registering S3 methods, Package namespaces, Package namespaces
@subsection Specifying imports and exports

Exports are specified using the @code{export} directive in the
@file{NAMESPACE} file.  A directive of the form

@findex export
@example
export(f, g)
@end example

@noindent
specifies that the variables @code{f} and @code{g} are to be exported.
(Note that variable names may be quoted, and reserved words and
non-standard names such as @code{[<-.fractions} must be.)

For packages with many variables to export it may be more convenient to
specify the names to export with a regular expression using
@code{exportPattern}.  The directive

@findex exportPattern
@example
exportPattern("^[^\\.]")
@end example

@noindent
exports all variables that do not start with a period.  However, such
broad patterns are not recommended for production code: it is better to
list all exports or use narrowly-defined groups.  (This pattern applies
to S4 classes.)  Beware of patterns which include names starting with a
period: some of these are internal-only variables and should never be
exported, e.g.@: @samp{.__S3MethodsTable__.} (and loading excludes known
cases).

Packages implicitly import the base namespace.
Variables exported from other packages with namespaces need to be
imported explicitly using the directives @code{import} and
@code{importFrom}.  The @code{import} directive imports all exported
variables from the specified package(s).  Thus the directives

@findex import
@example
import(foo, bar)
@end example

@noindent
specifies that all exported variables in the packages @pkg{foo} and
@pkg{bar} are to be imported.  If only some of the exported variables
from a package are needed, then they can be imported using
@code{importFrom}.  The directive

@findex importFrom
@example
importFrom(foo, f, g)
@end example

@noindent
specifies that the exported variables @code{f} and @code{g} of the
package @pkg{foo} are to be imported.  Using @code{importFrom}
selectively rather than @code{import} is good practice and recommended
notably when importing from packages with more than a dozen exports and
especially from those written by others (so what they export can change
in future).

To import every symbol from a package but for a few exceptions,
pass the @code{except} argument to @code{import}. The directive

@example
import(foo, except=c(bar, baz))
@end example

@noindent
imports every symbol from @pkg{foo} except @code{bar} and
@code{baz}. The value of @code{except} should evaluate to something
coercible to a character vector, after substituting each symbol for
its corresponding string.

It is possible to export variables from a namespace which it has
imported from other namespaces: this has to be done explicitly and not
@emph{via} @code{exportPattern}.

If a package only needs a few objects from another package it can use a
fully qualified variable reference in the code instead of a formal
import.  A fully qualified reference to the function @code{f} in package
@pkg{foo} is of the form @code{foo::f}.  This is slightly less efficient
than a formal import and also loses the advantage of recording all
dependencies in the @file{NAMESPACE} file (but they still need to be
recorded in the @file{DESCRIPTION} file).  Evaluating @code{foo::f} will
cause package @pkg{foo} to be loaded, but not attached, if it was not
loaded already---this can be an advantage in delaying the loading of a
rarely used package.

Using @code{foo:::f} instead of @code{foo::f} allows access to
unexported objects.  This is generally not recommended, as the
semantics of unexported objects may be changed by the package author
in routine maintenance.

@node Registering S3 methods, Load hooks, Specifying imports and exports, Package namespaces
@subsection Registering S3 methods

The standard method for S3-style @code{UseMethod} dispatching might fail
to locate methods defined in a package that is imported but not attached
to the search path.  To ensure that these methods are available the
packages defining the methods should ensure that the generics are
imported and register the methods using @code{S3method} directives.  If
a package defines a function @code{print.foo} intended to be used as a
@code{print} method for class @code{foo}, then the directive

@findex S3method
@example
S3method(print, foo)
@end example

@noindent
ensures that the method is registered and available for @code{UseMethod}
dispatch, and the function @code{print.foo} does not need to be exported.
Since the generic @code{print} is defined in @pkg{base} it does not need
to be imported explicitly.

(Note that function and class names may be quoted, and reserved words
and non-standard names such as @code{[<-} and @code{function} must
be.)

It is possible to specify a third argument to S3method, the function to
be used as the method, for example

@example
S3method(print, check_so_symbols, .print.via.format)
@end example

@noindent
when @code{print.check_so_symbols} is not needed.

As from @R{} 3.6.0 one can also use @code{S3method()} directives to
perform @emph{delayed} registration.  With
@example
if(getRversion() >= "3.6.0") @{
    S3method(pkg::gen, cls)
@}
@end example
@noindent
function @code{gen.cls} will get registered as an S3 method for class
@code{cls} and generic @code{gen} from package @code{pkg} only when the
namespace of @code{pkg} is loaded.  This can be employed to deal with
situations where the method is not ``immediately'' needed, and having to
pre-load the namespace of @code{pkg} (and all its strong dependencies)
in order to perform immediate registration is considered too onerous.

@node Load hooks, useDynLib, Registering S3 methods, Package namespaces
@subsection Load hooks

@findex .onLoad
@findex .onAttach
There are a number of hooks called as packages are loaded, attached,
detached, and unloaded.  See @code{help(".onLoad")} for more details.

Since loading and attaching are distinct operations, separate hooks are
provided for each.  These hook functions are called @code{.onLoad} and
@code{.onAttach}.  They both take arguments@footnote{they will be called
with two unnamed arguments, in that order.} @code{libname} and
@code{pkgname}; they should be defined in the namespace but not
exported.

@findex .onUnload
@findex .onDetach
@findex .Last.lib
Packages can use a @code{.onDetach} or @code{.Last.lib} function
(provided the latter is exported from the namespace) when @code{detach}
is called on the package.  It is called with a single argument, the full
path to the installed package.  There is also a hook @code{.onUnload}
which is called when the namespace is unloaded (@emph{via} a call to
@code{unloadNamespace}, perhaps called by @code{detach(unload = TRUE)})
with argument the full path to the installed package's directory.
Functions @code{.onUnload} and @code{.onDetach} should be defined in the
namespace and not exported, but @code{.Last.lib} does need to be
exported.

Packages are not likely to need @code{.onAttach} (except perhaps for a
start-up banner); code to set options and load shared objects should be
placed in a @code{.onLoad} function, or use made of the @code{useDynLib}
directive described next.

User-level hooks are also available: see the help on function
@code{setHook}.

These hooks are often used incorrectly.  People forget to export
@code{.Last.lib}.   Compiled code should be loaded in @code{.onLoad} (or
@emph{via} a @code{useDynLb} directive: see below) and unloaded in
@code{.onUnload}.  Do remember that a package's namespace can be loaded
without the namespace being attached (e.g.@: by @code{pkgname::fun}) and
that a package can be detached and re-attached whilst its namespace
remains loaded.

It is good practice for these functions to be quiet.  Any messages
should use @code{packageStartupMessage} so users (include check scripts)
can suppress them if desired.

@node useDynLib, An example, Load hooks, Package namespaces
@subsection useDynLib

A @file{NAMESPACE} file can contain one or more @code{useDynLib}
directives which allows shared objects that need to be
loaded.@footnote{NB: this will only be read in all versions of @R{} if
the package contains @R{} code in a @file{R} directory.}  The directive

@findex useDynLib
@example
useDynLib(foo)
@end example

@noindent
registers the shared object @code{foo}@footnote{Note that this is the
basename of the shared object, and the appropriate extension (@file{.so}
or @file{.dll}) will be added.} for loading with @code{library.dynam}.
Loading of registered object(s) occurs after the package code has been
loaded and before running the load hook function.  Packages that would
only need a load hook function to load a shared object can use the
@code{useDynLib} directive instead.

The @code{useDynLib} directive also accepts the names of the native
routines that are to be used in @R{} @emph{via} the @code{.C}, @code{.Call},
@code{.Fortran} and @code{.External} interface functions.  These are given as
additional arguments to the directive, for example,

@example
useDynLib(foo, myRoutine, myOtherRoutine)
@end example

By specifying these names in the @code{useDynLib} directive, the native
symbols are resolved when the package is loaded and @R{} variables
identifying these symbols are added to the package's namespace with
these names.  These can be used in the @code{.C}, @code{.Call},
@code{.Fortran} and @code{.External} calls in place of the name of the
routine and the @code{PACKAGE} argument.  For instance, we can call the
routine @code{myRoutine} from @R{} with the code

@example
 .Call(myRoutine, x, y)
@end example

@noindent
rather than

@example
 .Call("myRoutine", x, y, PACKAGE = "foo")
@end example

There are at least two benefits to this approach.  Firstly, the symbol
lookup is done just once for each symbol rather than each time the
routine is invoked.  Secondly, this removes any ambiguity in resolving
symbols that might be present in more than one DLL.  However, this
approach is nowadays deprecated in favour of supplying registration
information (see below).

In some circumstances, there will already be an @R{} variable in the
package with the same name as a native symbol. For example, we may have
an @R{} function in the package named @code{myRoutine}.  In this case,
it is necessary to map the native symbol to a different @R{} variable
name. This can be done in the @code{useDynLib} directive by using named
arguments. For instance, to map the native symbol name @code{myRoutine}
to the @R{} variable @code{myRoutine_sym}, we would use

@example
useDynLib(foo, myRoutine_sym = myRoutine, myOtherRoutine)
@end example

We could then call that routine from @R{} using the command

@example
 .Call(myRoutine_sym, x, y)
@end example

Symbols without explicit names are assigned to the @R{} variable with
that name.

In some cases, it may be preferable not to create @R{} variables in the
package's namespace that identify the native routines.  It may be too
costly to compute these for many routines when the package is loaded
if many of these routines are not likely to be used.  In this case,
one can still perform the symbol resolution correctly using the DLL,
but do this each time the routine is called.  Given a reference to the
DLL as an @R{} variable, say @code{dll}, we can call the routine
@code{myRoutine} using the expression

@example
 .Call(dll$myRoutine, x, y)
@end example

The @code{$} operator resolves the routine with the given name in the
DLL using a call to @code{getNativeSymbol}.  This is the same
computation as above where we resolve the symbol when the package is
loaded. The only difference is that this is done each time in the case
of @code{dll$myRoutine}.

In order to use this dynamic approach (e.g., @code{dll$myRoutine}), one
needs the reference to the DLL as an @R{} variable in the package.  The
DLL can be assigned to a variable by using the @code{variable =
dllName} format used above for mapping symbols to @R{} variables.  For
example, if we wanted to assign the DLL reference for the DLL
@code{foo} in the example above to the variable @code{myDLL}, we would
use the following directive in the @file{NAMESPACE} file:

@example
myDLL = useDynLib(foo, myRoutine_sym = myRoutine, myOtherRoutine)
@end example

Then, the @R{} variable @code{myDLL} is in the package's namespace and
available for calls such as @code{myDLL$dynRoutine} to access routines
that are not explicitly resolved at load time.

If the package has registration information (see @ref{Registering native
routines}), then we can use that directly rather than specifying the
list of symbols again in the @code{useDynLib} directive in the
@file{NAMESPACE} file.  Each routine in the registration information is
specified by giving a name by which the routine is to be specified along
with the address of the routine and any information about the number and
type of the parameters.  Using the @code{.registration} argument of
@code{useDynLib}, we can instruct the namespace mechanism to create
@R{} variables for these symbols.  For example, suppose we have the
following registration information for a DLL named @code{myDLL}:

@example
static R_NativePrimitiveArgType foo_t[] = @{
    REALSXP, INTSXP, STRSXP, LGLSXP
@};

static const R_CMethodDef cMethods[] = @{
   @{"foo", (DL_FUNC) &foo, 4, foo_t@},
   @{"bar_sym", (DL_FUNC) &bar, 0@},
   @{NULL, NULL, 0, NULL@}
@};

static const R_CallMethodDef callMethods[] = @{
   @{"R_call_sym", (DL_FUNC) &R_call, 4@},
   @{"R_version_sym", (DL_FUNC) &R_version, 0@},
   @{NULL, NULL, 0@}
@};
@end example

Then, the directive in the @file{NAMESPACE} file

@example
useDynLib(myDLL, .registration = TRUE)
@end example

@noindent
causes the DLL to be loaded and also for the @R{} variables @code{foo},
@code{bar_sym}, @code{R_call_sym} and @code{R_version_sym} to be
defined in the package's namespace.

Note that the names for the @R{} variables are taken from the entry in
the registration information and do not need to be the same as the name
of the native routine.  This allows the creator of the registration
information to map the native symbols to non-conflicting variable names
in @R{}, e.g.@: @code{R_version} to @code{R_version_sym} for use in an
@R{} function such as

@example
R_version <- function()
@{
  .Call(R_version_sym)
@}
@end example

Using argument @code{.fixes} allows an automatic prefix to be added to
the registered symbols, which can be useful when working with an
existing package.  For example, package @CRANpkg{KernSmooth} has

@example
useDynLib(KernSmooth, .registration = TRUE, .fixes = "F_")
@end example

@noindent
which makes the @R{} variables corresponding to the Fortran symbols
@code{F_bkde} and so on, and so avoid clashes with @R{} code in the
namespace.

@strong{NB}: Using these arguments for a package which does not register
native symbols merely slows down the package loading (although many
@acronym{CRAN} packages have done so).  Once symbols are registered,
check that the corresponding @R{} variables are not accidentally
exported by a pattern in the @file{NAMESPACE} file.


@node An example, Namespaces with S4 classes and methods, useDynLib, Package namespaces
@subsection An example

As an example consider two packages named @pkg{foo} and @pkg{bar}.  The
@R{} code for package @pkg{foo} in file @file{foo.R} is

@quotation
@cartouche
@example
x <- 1
f <- function(y) c(x,y)
foo <- function(x) .Call("foo", x, PACKAGE="foo")
print.foo <- function(x, ...) cat("<a foo>\n")
@end example
@end cartouche
@end quotation

@noindent
Some C code defines a C function compiled into DLL @code{foo} (with an
appropriate extension).  The @file{NAMESPACE} file for this package is

@quotation
@cartouche
@example
useDynLib(foo)
export(f, foo)
S3method(print, foo)
@end example
@end cartouche
@end quotation

@noindent
The second package @pkg{bar} has code file @file{bar.R}

@quotation
@cartouche
@example
c <- function(...) sum(...)
g <- function(y) f(c(y, 7))
h <- function(y) y+9
@end example
@end cartouche
@end quotation

@noindent
and @file{NAMESPACE} file

@quotation
@cartouche
@example
import(foo)
export(g, h)
@end example
@end cartouche
@end quotation

@noindent
Calling @code{library(bar)} loads @pkg{bar} and attaches its exports to
the search path.  Package @pkg{foo} is also loaded but not attached to
the search path.  A call to @code{g} produces

@example
> g(6)
[1]  1 13
@end example

@noindent
This is consistent with the definitions of @code{c} in the two settings:
in @pkg{bar} the function @code{c} is defined to be equivalent to
@code{sum}, but in @pkg{foo} the variable @code{c} refers to the
standard function @code{c} in @pkg{base}.

@node Namespaces with S4 classes and methods,  , An example, Package namespaces
@subsection Namespaces with S4 classes and methods

Some additional steps are needed for packages which make use of formal
(S4-style) classes and methods (unless these are purely used
internally).  The package should have @code{Depends: methods} in its
@file{DESCRIPTION} and @code{import(methods)} or
@code{importFrom(methods, ...)} plus any classes and methods which are
to be exported need to be declared in the @file{NAMESPACE} file.  For
example, the @pkg{stats4} package has

@findex exportClasses
@findex exportMethods

@example
export(mle) # exporting methods implicitly exports the generic
importFrom("stats", approx, optim, pchisq, predict, qchisq, qnorm, spline)
## For these, we define methods or (AIC, BIC, nobs) an implicit generic:
importFrom("stats", AIC, BIC, coef, confint, logLik, nobs, profile,
           update, vcov)
exportClasses(mle, profile.mle, summary.mle)
## All methods for imported generics:
exportMethods(coef, confint, logLik, plot, profile, summary,
              show, update, vcov)
## implicit generics which do not have any methods here
export(AIC, BIC, nobs)
@end example

@findex exportPattern
@findex exportClassPattern
@noindent
All S4 classes to be used outside the package need to be listed in an
@code{exportClasses} directive. Alternatively, they can be specified
using @code{exportClassPattern}@footnote{This defaults to the same
pattern as @code{exportPattern}: use something like
@code{exportClassPattern("^$")} to override this.} in the same style as
for @code{exportPattern}.  To export methods for generics from other
packages an @code{exportMethods} directive can be used.

Note that exporting methods on a generic in the namespace will also
export the generic, and exporting a generic in the namespace will also
export its methods.  If the generic function is not local to this
package, either because it was imported as a generic function or because
the non-generic version has been made generic solely to add S4 methods
to it (as for functions such as @code{coef} in the example above), it
can be declared @emph{via} either or both of @code{export} or
@code{exportMethods}, but the latter is clearer (and is used in the
@pkg{stats4} example above).  In particular, for primitive functions
there is no generic function, so @code{export} would export the
primitive, which makes no sense.  On the other hand, if the generic is
local to this package, it is more natural to export the function itself
using @code{export()}, and this @emph{must} be done if an implicit
generic is created without setting any methods for it (as is the case
for @code{AIC} in @pkg{stats4}).

A non-local generic function is only exported to ensure that calls to
the function will dispatch the methods from this package (and that is
not done or required when the methods are for primitive functions).  For
this reason, you do not need to document such implicitly created generic
functions, and @code{undoc} in package @pkg{tools} will not report them.

If a package uses S4 classes and methods exported from another package,
but does not import the entire namespace of the other
package@footnote{if it does, there will be opaque warnings about
replacing imports if the classes/methods are also imported.}, it needs
to import the classes and methods explicitly, with directives

@findex importClassesFrom
@findex importMethodsFrom

@example
importClassesFrom(package, ...)
importMethodsFrom(package, ...)
@end example

@noindent
listing the classes and functions with methods respectively.  Suppose we
had two small packages @pkg{A} and @pkg{B} with @pkg{B} using @pkg{A}.
Then they could have @code{NAMESPACE} files

@quotation
@cartouche
@example
export(f1, ng1)
exportMethods("[")
exportClasses(c1)
@end example
@end cartouche
@end quotation

@noindent
and

@quotation
@cartouche
@example
importFrom(A, ng1)
importClassesFrom(A, c1)
importMethodsFrom(A, f1)
export(f4, f5)
exportMethods(f6, "[")
exportClasses(c1, c2)
@end example
@end cartouche
@end quotation

@noindent
respectively.

Note that @code{importMethodsFrom} will also import any generics defined
in the namespace on those methods.

It is important if you export S4 methods that the corresponding generics
are available.  You may for example need to import @code{coef} from
@pkg{stats} to make visible a function to be converted into its
implicit generic.  But it is better practice to make use of the generics
exported by @pkg{stats4} as this enables multiple packages to
unambiguously set methods on those generics.

@node Writing portable packages, Diagnostic messages, Package namespaces, Creating R packages
@section Writing portable packages

This section contains advice on writing packages to be used on multiple
platforms or for distribution (for example to be submitted to a package
repository such as @acronym{CRAN}).

@menu
* PDF size::
* Check timing::
* Encoding issues::
* Portable C and C++ code::
* Binary distribution::
@end menu

Portable packages should have simple file names: use only alphanumeric
@acronym{ASCII} characters and period (@code{.}), and avoid those names
not allowed under Windows (@pxref{Package structure}).

Many of the graphics devices are platform-specific: even @code{X11()}
(aka @code{x11()}) which although emulated on Windows may not be
available on a Unix-alike (and is not the preferred screen device on OS
X).  It is rarely necessary for package code or examples to open a new
device, but if essential,@footnote{People use @code{dev.new()} to open a
device at a particular size: that is not portable but using
@code{dev.new(noRStudioGD = TRUE)} helps.} use @code{dev.new()}.

Use @command{R CMD build} to make the release @file{.tar.gz} file.

@command{R CMD check} provides a basic set of checks, but often further
problems emerge when people try to install and use packages submitted to
@acronym{CRAN} -- many of these involve compiled code.  Here are some
further checks that you can do to make your package more portable.

@itemize

@item
If your package has a @file{configure} script, provide a
@file{configure.win} or @file{configure.ucrt} script to be used on Windows (an
empty @file{configure.win} file if no actions are needed).

@item
If your package has a @file{Makevars} or @file{Makefile} file, make sure
that you use only portable make features.  Such files should be
LF-terminated@footnote{Solaris @command{make} does not accept
CRLF-terminated Makefiles; Solaris warns about and some other
@command{make}s ignore incomplete final lines.} (including the final
line of the file) and not make use of GNU extensions.  (The POSIX
specification is available at
@uref{https://pubs.opengroup.org/onlinepubs/9699919799/utilities/make.html};
anything not documented there should be regarded as an extension to be
avoided.  Further advice can be found at
@uref{https://www.gnu.org/software/autoconf/manual/autoconf.html#Portable-Make}.                                )
Commonly misused GNU extensions are conditional inclusions (@code{ifeq}
and the like), @code{$@{shell ...@}}, @code{$@{wildcard ...@}} and
similar, and the use of @code{+=}@footnote{This was apparently
introduced in SunOS 4, and is available elsewhere @emph{provided} it is
surrounded by spaces.} and @code{:=}.  Also, the use of @code{$<} other
than in implicit rules is a GNU extension, as is the @code{$^} macro.
As is the use of  @code{.PHONY} (some other makes ignore it).
Unfortunately makefiles which use GNU extensions often run on other
platforms but do not have the intended results.

Note that the @option{-C} flag for @command{make} is not included in the
POSIX specification and is not implemented by some of the
@command{make}s used with @R{}.

The use of @code{$@{shell ...@}} can be avoided by using backticks, e.g.@:

@example
PKG_CPPFLAGS = `gsl-config --cflags`
@end example

@noindent
which works in all versions of @command{make} known@footnote{GNU make,
BSD make and other variants of @command{pmake} in FreeBSD, NetBSD and
formerly in macOS, AT&T make as implemented on Solaris and `Distributed
Make' (@code{dmake}), part of Oracle Developer Studio and available in
other versions including from Apache OpenOffice.} to be used with @R{}.

If you really must require GNU make, declare it in the @file{DESCRIPTION}
file by

@example
SystemRequirements: GNU make
@end example

@noindent
and ensure that you use the value of environment variable @env{MAKE}
(and not just @command{make}) in your scripts.  (On some platforms GNU
make is available under a name such as @command{gmake}, and there
@code{SystemRequirements} is used to set @env{MAKE}.)

If you only need GNU make for parts of the package which are rarely
needed (for example to create bibliography files under
@file{vignettes}), use a file called @file{GNUmakefile} rather than
@file{Makefile} as GNU make (only) will use the former.

macOS has used GNU make for many years (it previously used BSD make),
but the version has been frozen at 3.81 (from 2006).

Since the only viable make for Windows is GNU make, it is permissible to
use GNU extensions in files @file{Makevars.win}, @file{Makevars.ucrt},
@file{Makefile.win} or @file{Makefile.ucrt}.

@item
If you use @file{src/Makevars} to compile code in a subdirectory,
ensure that you have followed all the advice above.  In particular
@itemize
@item
Anticipate a parallel @command{make}.  @xref{Using Makevars}.

@item
Pass macros down to the makefile in the subdirectory, including
@strong{all} the needed compiler flags (including PIC and visibility
flags).  If they are used in the subdirectory's Makefile, this includes
macros @samp{AR} and @samp{RANLIB}.  @xref{Compiling in
sub-directories}, which has a C example.  A C++ example:

@example
pkg/libpkg.a:
        (cd pkg && $(MAKE) -f make_pkg libpkg.a \
         CXX="$(CXX)" CXXFLAGS="$(CXXFLAGS) $(CXXPICFLAGS) $(C_VISIBILITY)" \
         AR="$(AR)" RANLIB="$(RANLIB)")
@end example

@item
Ensure that cleanup will be performed by @command{R CMD build}, for
example in a @code{cleanup} script or a @samp{clean} target.
@end itemize

@item
If your package uses a @file{src/Makefile} file to compile code to be
linked into @R{}, ensure that it uses exactly the same compiler and flag
settings that @R{} uses when compiling such code: people often forget
@samp{PIC} flags.  If @code{R CMD config} is used, this needs something
like (for C++)

@example
RBIN = `"$@{R_HOME@}/bin/R"`
CXX = `"$@{RBIN@}" CMD config CXX`
CXXFLAGS = `"$@{RBIN@}" CMD config CXXFLAGS` `"$@{RBIN@}" CMD config CXXPICFLAGS`
@end example

@item
Names of source files including @file{=} (such as
@c from TCIU in 2020-09
@file{src/complex_Sig=gen.c}) will confuse some @command{make} programs
and should be avoided.

@item
Bash extensions also need to be avoided in shell scripts, including
expressions in Makefiles (which are passed to the shell for processing).
Some @R{} platforms use strict@footnote{For example, @command{test}
options @option{-a} and @option{-e} are not portable, and not supported
in the AT&T Bourne shell used on Solaris 10/11, even though they are in
the POSIX standard.  Nor does Solaris support @samp{$(@var{cmd})}.}
Bourne shells: an earlier @R{} toolset on Windows@footnote{as from
@R{} 4.0.0 the default is @command{bash}.} and some Unix-alike OSes use
@command{ash} (@uref{https://en.wikipedia.org/wiki/Almquist_shell}),
a `slim' shell with few builtins or variants such as @command{dash}.
Beware of assuming that all the POSIX command-line utilities are
available, especially on Windows where only a subset (which has changed
by version of @file{Rtools}) is provided for use with @R{}.  One
particular issue is the use of @command{echo}, for which two behaviours
are allowed
(@uref{https://pubs.opengroup.org/onlinepubs/9699919799/utilities/echo.html})
and both have occurred as defaults on @R{} platforms: portable
applications should use neither @option{-n} (as the first argument) nor
escape sequences.  The recommended replacement for @command{echo -n} is
the command @command{printf}.  Another common issue is the construction
@example
export FOO=value
@end example
@noindent
which is @command{bash}-specific (first set the variable then export it
by name).

Using @code{test -e} (or @code{[ -e ]}) in shell scripts is not fully
portable@footnote{it was not in the Bourne shell, and is not supported
by Solaris 10.}: @code{-f} is normally what is intended.  Flags
@option{-a} and @option{-o} are nowadays declared obsolescent by POSIX
and should not be used.

Use of `brace expansion', e.g.,
@example
rm -f src/*.@{o,so,d@}
@end example
@noindent
is not portable.

The @option{-o} flag for @command{set} in shell scripts is optional in
POSIX and not supported on all the platforms @R{} is used on.

The variable @samp{OSTYPE} is shell-specific and its values are
rather unpredictable and may include a version such as
@samp{darwin19.0}: @command{`uname`} is often what is intended (with
common values @samp{Darwin}, @samp{Linux} and @samp{SunOS}).

On macOS which shell @file{/bin/sh} invokes is user- and
platform-dependent: it might be @command{bash} version 3.2,
@command{dash} or @command{zsh} (for new accounts it is @command{zsh},
for accounts ported from Mojave or earlier it is usually
@command{bash}).

@item
Make use of the abilities of your compilers to check the
standards-conformance of your code.  For example, @command{gcc} and
@command{gfortran}@footnote{@uref{http://fortranwiki.org/fortran/show/Modernizing+Old+Fortran}
may help explain some of the warnings from @command{gfortran -Wall
-pedantic}.}  can be used with options @option{-Wall -pedantic} to alert
you to potential problems.  This is particularly important for C++,
where @code{g++ -Wall -pedantic} will alert you to the use of some of
the GNU extensions which fail to compile on most other C++ compilers. If
@R{} was not configured accordingly, one can achieve this @emph{via}
personal @file{Makevars} files.
@ifset UseExternalXrefs
@xref{Customizing package compilation, , Customizing package compilation,
R-admin, R Installation and Administration},
@end ifset

Portable C++ code needs to follow both the 2011 and 2014 standards or to
specify C+11/14/17/20 where available (which is not the case on all @R{}
platforms).  Currently C++17/20 support is patchy across @R{} platforms.

If using Fortran with the GNU compiler, use the flags
@option{-std=f95 -Wall -pedantic} which reject most GNU extensions and
features from later standards.  (Although @R{} only requires Fortran 90,
@command{gfortran} does not have a way to specify that standard.)

@R{} has tested that @code{DOUBLE COMPLEX} works and so is preferred to
@code{COMPLEX*16}.  (One can also use something like
@code{COMPLEX(KIND=KIND(0.0D0))}.)


@c https://gcc.gnu.org/onlinedocs/gfortran/KIND-Type-Parameters.html
@c https://stackoverflow.com/questions/838310/fortran-90-kind-parameter
The use of Fortran types such as @code{REAL(KIND=8)} is very far from
portable.  According of the standards this merely enumerates different
supported types, so @code{DOUBLE PRECISION} might be @code{REAL(KIND=3)}
(and is on an actual compiler).  Even if for a particular compiler the
value indicates the size in bytes, which values are supported is
platform-specific --- for example @command{gfortran} supports values of 4
and 8 on all current platforms and 10 and 16 on a few (but not for
example on @cputype{arm} CPUs).

Not all common @R{} platforms conform to the expected standards, e.g.@:
C99 for C code.  One common area of problems is the @code{*printf}
functions where Windows did not support @code{%lld}, @code{%Lf} and
similar formats (and has its own formats such as @code{%I64d} for 64-bit
integers).  It is very rare to need to output such types, and 64-bit
integers can usually be converted to doubles for output.  However, the
C11 standard (section 7.8.1) includes @code{PRIxNN}
macros@footnote{These are optional because the corresponding types are,
but must be provided if the types are.} in C header @file{inttypes.h}
(for example @code{PRId64}) so the portable approach is to test for
these and if not available provide emulations in the package.

As from macOS 11 (late 2020), its C compiler sets the flag
@option{-Werror=implicit-function-declaration} by default which forces
stricter conformance to C99.  This can be used on other platforms with
@command{gcc} or @command{clang}.  If your package has a
(@command{autoconf}-generated) @command{configure script}, try
installing it whilst using this flag, and read through the
@file{config.log} file --- compilation warnings and errors can lead to
features which are present not being detected.  (If possible do this on
several platforms.)

@item
@command{R CMD check} performs some checks for non-portable
compiler/linker flags in @file{src/Makevars}.  However, it cannot check
the meaning of such flags, and some are commonly accepted but with
compiler-specific meanings.  There are other non-portable flags which
are not checked, nor are @file{src/Makefile} files and makefiles in
sub-directories.  As a comment in the code says

@quotation
It is hard to think of anything apart from @option{-I*} and @option{-D*}
that is safe for general use @dots{}
@end quotation

@noindent
although @option{-pthread} is pretty close to portable.  (Option
@option{-U} is portable but little use on the command line as it will
only cancel built-in defines (not portable) and those defined earlier on
the command line (@R{} does not use any).)

People have used @command{configure} to customize @file{src/Makevars},
including for specific compilers.  This is unsafe for several reasons.
First, unintended compilers might meet the check---for example, several
compilers other than GCC identify themselves as `GCC' whilst being only
partially conformant.  Second, future versions of compilers may behave
differently (including updates to quite old series) so for example
@option{-Werror} (and specializations) can make a package
non-installable under a future version.  Third, using flags to suppress
diagnostic messages can hide important information for debugging on a
platform not tested by the package maintainer.  (@command{R CMD check}
can optionally report on unsafe flags which were used.)

Avoid the use of @option{-march} and especially @option{-march=native}.
This allows the compiler to generate code that will only run on a
particular class of CPUs (that of the compiling machine for
@samp{native}).  People assume this is a `minimum' CPU specification,
but that is not how it is documented for @command{gcc} (it is accepted
by @command{clang} but apparently it is undocumented what precisely it
does, and it can be accepted and may be ignored for other compilers).
(For personal use @option{-mtune} is safer, but still not portable
enough to be used in a public package.)  Not even @command{gcc} supports
@samp{native} for all CPUs, and it can do surprising things if it finds
a CPU released later than its version.

@item
Do be very careful with passing arguments between @R{}, C and Fortran
code.  In particular, @code{long} in C will be 32-bit on some @R{}
platforms (including 64-bit Windows), but 64-bit on most modern Unix and
Linux platforms.  It is rather unlikely that the use of @code{long} in C
code has been thought through: if you need a longer type than @code{int}
you should use a configure test for a C99/C++11 type such as
@code{int_fast64_t} (and failing that, @code{long long}) and typedef
your own type, or use another suitable type (such as @code{size_t}, but
beware that is unsigned and @code{ssize_t} is not portable).
@c https://en.cppreference.com/w/cpp/language/types claims long long is
@c >= 64-bit, but that is not obvious in the standard.

It is not safe to assume that @code{long} and pointer types are the same
size, and they are not on 64-bit Windows.  If you need to convert
pointers to and from integers use the C99/C++11 integer types
@code{intptr_t} and @code{uintptr_t} (in the headers @code{<stdint.h>}
and @code{<cstdint>}: they are not required to be implemented by the
standards but are used in C code by @R{} itself).

Note that @code{integer} in Fortran corresponds to @code{int} in C on
all @R{} platforms.

@item
Under no circumstances should your compiled code ever call @code{abort}
or @code{exit}@footnote{or where supported the variants @code{_Exit} and
@code{_exit}.}: these terminate the user's @R{} process, quite possibly
losing all unsaved work.  One usage that could call @code{abort} is the
@code{assert} macro in C or C++ functions, which should never be active
in production code.  The normal way to ensure that is to define the
macro @code{NDEBUG}, and @command{R CMD INSTALL} does so as part of the
compilation flags.  Beware of including headers (including from other
packages) which could undefine it, now or in future versions.  If you
wish to use @code{assert} during development. you can include
@code{-UNDEBUG} in @code{PKG_CPPFLAGS} or @code{#undef} it in your
headers or code files.  Note that your own @file{src/Makefile} or
makefiles in sub-directories may also need to define @code{NDEBUG}.
@c RcppArmadillo undefined NDEBUG and compiled in assert calls
@c on macOS in 2022-03.

This applies not only to your own code but to any external software you
compile in or link to.

@item
Compiled code should not write to @file{stdout} or @file{stderr} and C++
and Fortran I/O should not be used.  As with the previous item such
calls may come from external software and may never be called, but
package authors are often mistaken about that.

@item
Compiled code should not call the system random number generators such
as @code{rand}, @code{drand48} and @code{random}@footnote{This and
@code{srandom} are in any case not portable.  They are in POSIX but not
in the C99 standard, and not available on Windows.}, but rather use the
interfaces to @R{}'s RNGs described in @ref{Random numbers}.  In
particular, if more than one package initializes a system RNG (e.g.@:
@emph{via} @code{srand}), they will interfere with each other.

Nor should the C++11 random number library be used, nor any other
third-party random number generators such as those in GSL.

@item
Errors in memory allocation and reading/writing outside arrays are very
common causes of crashes (e.g., segfaults) on some machines.
See @ref{Checking memory access} for tools which can be used to look for this.

@item
Many platforms will allow unsatisfied entry points in compiled code, but
will crash the application (here @R{}) if they are ever used.  Some
(notably Windows) will not.  Looking at the output of

@example
nm -pg mypkg.so
@end example

@noindent
and checking if any of the symbols marked @code{U} is unexpected is a
good way to avoid this.

@item
Linkers have a lot of freedom in how to resolve entry points in
dynamically-loaded code, so the results may differ by platform.  One
area that has caused grief is packages including copies of standard
system software such as @code{libz} (especially those already linked
into @R{}).  In the case in point, entry point @code{gzgets} was
sometimes resolved against the old version compiled into the package,
sometimes against the copy compiled into @R{} and sometimes against the
system dynamic library.  The only safe solution is to rename the entry
points in the copy in the package.  We have even seen problems with
entry point name @code{myprintf}, which is a system entry
point@footnote{in @file{libselinux}.} on some Linux systems.

@c example fron package fst on M1 mac in 2022-02.
A related issue is the naming of libraries built as part of the package
installation.  macOS and Windows have case-insensitive file systems, so
using
@example
-L. -lLZ4
@end example
@noindent
in @code{PKG_LIBS} will match @code{liblz4}.  And @code{-L.} only
appends to the list of searched locations, and @code{liblz4} might be
found in an earlier-searched location (and has been).  The only safe way
is to give an explicit path, for example
@example
./libLZ4.a
@end example

@item
Conflicts between symbols in DLLs are handled in very platform-specific
ways.  Good ways to avoid trouble are to make as many symbols as
possible static (check with @code{nm -pg}), and to use names which are
clearly tied to your package (which also helps users if anything does go
wrong).  Note that symbol names starting with @code{R_} are regarded as
part of @R{}'s namespace and should not be used in packages.

@item
It is good practice for DLLs to register their symbols
(@pxref{Registering native routines}), restrict visibility
(@pxref{Controlling visibility}) and not allow symbol search
(@pxref{Registering native routines}).  It should be possible for a DLL
to have only one visible symbol, @code{R_init_@var{pkgname}}, on
suitable platforms@footnote{At least Linux and Windows, but not macOS.},
which would completely avoid symbol conflicts.

@item
It is not portable to call compiled code in @R{} or other packages
@emph{via} @code{.Internal}, @code{.C}, @code{.Fortran}, @code{.Call} or
@code{.External}, since such interfaces are subject to change without
notice and will probably result in your code terminating the @R{}
process.

@item
Do not use (hard or symbolic) file links in your package sources.
Where possible @command{R CMD build} will replace them by copies.

@item
If you do not yourself have a Windows system, consider submitting your
source package to WinBuilder (@uref{https://win-builder.r-project.org/})
before distribution.  If you need to check on an M1 Mac, there is a
check service at
@uref{https://mac.r-project.org/macbuilder/submit.html}.

@item
It is bad practice for package code to alter the search path using
@code{library}, @code{require} or @code{attach} and this often does not
work as intended.  For alternatives, see @ref{Suggested packages} and
@code{with()}.

@item
Examples can be run interactively @emph{via} @code{example} as well as
in batch mode when checking.  So they should behave appropriately in
both scenarios, conditioning by @code{interactive()} the parts which
need an operator or observer.  For instance, progress
bars@footnote{except perhaps the simplest kind as used by
@code{download.file()} in non-interactive use.} are only appropriate in
interactive use, as is displaying help pages or calling @code{View()}
(see below).

@item
Be careful with the order of entries in macros such as @code{PKG_LIBS}.
Some linkers will re-order the entries, and behaviour can differ between
dynamic and static libraries.  Generally @option{-L} options should
precede@footnote{Whereas the GNU linker reorders so @option{-L} options
are processed first, the Solaris one does not.} the libraries (typically
specified by @option{-l} options) to be found from those directories,
and libraries are searched once in the order they are specified.  Not
all linkers allow a space after @option{-L} .

@item
Care is needed with the use of @code{LinkingTo}.  This puts one or more
directories on the include search path ahead of system headers but
(prior to @R{} 3.4.0) after those specified in the @code{CPPFLAGS} macro
of the @R{} build (which normally includes @code{-I/usr/local/include},
but most platforms ignore that and include it with the system headers).

Any confusion would be avoided by having @code{LinkingTo} headers in a
directory named after the package.  In any case, name conflicts of
headers and directories under package @file{include} directories should
be avoided, both between packages and between a package and system and
third-party software.

@item
The @command{ar} utility is often used in makefiles to make static
libraries.  Its modifier @code{u} is defined by POSIX but is disabled in
GNU @command{ar} on some Linux distributions which use
`deterministic mode'.  The safest way to make a static library is to first
remove any existing file of that name then use @command{ar -cr} and then
@command{ranlib} if needed (which is system-dependent: on most
systems@footnote{some versions of macOS did not.} @command{ar} always
maintains a symbol table).  The POSIX standard says options should be
preceded by a hyphen (as in @option{-cr}), although most OSes accept
them without.
@c flowWorkspace failed on macOS in Mar 2016 because a wildcard spec was empty
Note that on some systems @command{ar -cr} must have at least one file
specified.

The @code{s} modifier (to replace a separate call to @command{ranlib})
is required by X/OPEN but not POSIX, so @command{ar -crs} is not
portable.

@item
The @command{strip} utility is platform-specific (and @acronym{CRAN}
prohibits removing debug symbols).  For example the options
@option{--strip-debug} and @option{--strip-unneeded} of the GNU version
are not supported on macOS nor Solaris@footnote{which is
@file{/usr/ccs/bin/strip}, not @file{/usr/bin/strip}.}: the POSIX
standard for @command{strip} does not mention any options, and what
calling it without options does is platform-dependent.  Stripping a
@file{.so} file could even prevent it being dynamically loaded into @R{}
on an untested platform.

@command{ld -S} invokes @command{strip --strip-debug} on GNU
@command{ld} but is not portable: in particular on Solaris it does
something completely different and takes an argument.

@item
Some people have a need to set a locale.  Locale names are not portable,
and e.g.@: @samp{fr_FR.utf8} is commonly used on Linux but not accepted on
either Solaris or macOS.  @samp{fr_FR.UTF-8} is more portable, being
accepted on recent Linux, AIX, FreeBSD, macOS and Solaris (at least).
However, some Linux distributions micro-package, so locales defined by
@pkg{glibc} (including these examples) may not be installed.

@item
Avoid spaces in file names, not least as they can cause difficulties for
external tools.  An example was a package with a @CRANpkg{knitr}
vignette that used spaces in plot names: this caused some older versions
of @command{pandoc} to fail with a baffling error message.
@c msmtools in June 2016 failed with pandoc 1.12 but not 1.16.

Non-ASCII filenames can also cause problems (particularly in non-UTF-8
locales).

@item
Take care in naming @LaTeX{} macros (also known as `commands') in
vignette sources: if these are also defined in a future version of one
of the @LaTeX{} packages used there will be a fatal error.  One instance
in 2021 was package @samp{hyperref} newly defining @samp{\C}, @samp{\F},
@samp{\G}, @samp{\U} and @samp{\textapprox}.  If you are confident that
your definitions will be the only ones relevant you can use
@samp{\renewcommand} but it is better to use names clearly associated
with your package.

@item
Make sure that any version requirement for Java code is both declared in
the @samp{SystemRequirements} field@footnote{If a Java interpreter is
required directly (not @emph{via} @CRANpkg{rJava}) this must be declared
and its presence tested like any other external command.} and tested at
runtime (not least as the Java installation when the package is
installed might not be the same as when the package is run and will not
be for binary packages).  Java 8 is available for fewer platforms than
Java 7 was, and Java 11 for fewer still (at the time of writing, only
@cputype{x86_64} Linux, macOS, 64-bit Windows and 64-bit Solaris 11 from
Oracle; Linux on several 64-bit CPUs, Intel-based macOS, 32- and 64-bit
Windows and AIX from from @uref{https://adoptium.net/} and
@cputype{arm64} macOS from Zulu -- several OSes provide builds of
OpenJDK including FreeBSD and most Linux distributions).

When specifying a minimum Java version please use the official version
names, which are (confusingly)
@example
1.1 1.2 1.3 1.4 5.0 6 7 8 9 10 11 12 13 14 15 16 17 18 (announced 19 20)
@end example
@noindent
and as from 2018 a year.month scheme such as @samp{18.9} is also in
use.  Fortunately only the integer values are likely to be relevant.

A suitable test for Java at least version 8 for packages using
@CRANpkg{rJava} would be something like
@example
.jinit()
jv <- .jcall("java/lang/System", "S", "getProperty", "java.runtime.version")
if(substr(jv, 1L, 2L) == "1.") @{
  jvn <- as.numeric(paste0(strsplit(jv, "[.]")[[1L]][1:2], collapse = "."))
  if(jvn < 1.8) stop("Java >= 8 is needed for this package but not available")
@}
@end example
@noindent
Java 9 changed the format of this string (which used to be something
like @samp{1.8.0_292-b10}); Java 11 gave @code{jv} as @samp{11+28}
whereas Java 11.0.11 gives @samp{11.0.11+9}.
(@uref{https://openjdk.java.net/jeps/322} details the current scheme.
Note that it is necessary to allow for pre-releases like
@samp{11-ea+22}.)

Note too that the compiler used to produce a @code{jar} can impose a minimum
Java version, often resulting in an arcane message like

@example
java.lang.UnsupportedClassVersionError: ... Unsupported major.minor version 52.0
@end example
@noindent
(Where @uref{https://en.wikipedia.org/wiki/Java_class_file} maps
class-file version numbers to Java versions.)  Compile with something
like @command{javac -target 1.6} to ensure this is avoided. (As from
Java 8, @command{javac} defaults to compiling for Java 8.  Versions as
old as @samp{1.6} are already deprecated and will give a warning with
Java 10's @command{javac}.)  Note this also applies to packages
distributing (or even downloading) compiled Java code produced by
others, so their requirements need to be checked (they are often not
documented accurately) and accounted for.  It should be possible to
check the class-file version @emph{via} command-line utility
@command{javap}, if necessary after extracting the @file{.class} files
from a @file{.jar} archive.

Some packages have stated a requirement on a particular JDK, but a
package should only be requiring a JRE unless providing its own Java
interface.

@c https://endoflife.date/java
Java 8 is still in widespread use (and may remain so because of licence
changes: although Oracle ended its support even for personal use at the
end of 2020, OpenJDK has full support until 2022-03), but Java 7 was the
latest provided by Oracle for several platforms.

@item
A package with a hard-to-satisfy system requirement is by definition not
portable, annoyingly so if this is not declared in the
@samp{SystemRequirements} field.  The most common example is the use of
@command{pandoc}, which is only available for a very limited range of
platforms (and has onerous requirements to install from source) and has
capabilities@footnote{For example, the ability to handle @samp{https://}
URLs.} that vary by build but are not documented.  Several recent
versions of @command{pandoc} for macOS did not work on @R{}'s target of
High Sierra (and this too was undocumented): at the time of writing
@samp{2.14.0} to @samp{2.16.2} did, but not @samp{2.17.0.1}.  (Currently
@command{pandoc} is only available from the official site for Intel
macOS but that build works well enough on @cputype{arm64} machines.)
Another example is the Rust compilation system (@command{cargo} and
@command{rustc}).

Usage of external commands should always be conditional on a test for
presence (perhaps using @code{Sys.which}), as well as declared in the
@samp{SystemRequirements} field.  A package should pass its checks
without warnings nor errors without the external command being present.

An external command can be a (possibly optional) requirement for an
imported or suggested package but needed for examples, tests or
vignettes in the package itself.  Such usages should always be declared
and conditional.

Interpreters for scripting languages such as Perl, Python and Ruby need
to be declared as system requirements and used conditionally: for
example macOS 10.16 was announced not to have them (but released as
macOS 11 with them); later it was announced that macOS 12.3 does not
have Python 2 and only a minimal install of Python 3 is included.
Python 2 has passed end-of-life and been removed from many major
distributions.  Support for Rust cannot be assumed, and it does not even
support all the @R{} platforms used by @acronym{CRAN}.

Command @command{cmake} is not commonly installed, and where it is, it
might not be on the path.  In particular, the most common location on
macOS is @file{/Applications/CMake.app/Contents/bin/cmake} and that
should be looked for if @command{cmake} is not found on the path.

@item
Be sure to use portable encoding names: none of @code{utf8}, @code{mac}
and @code{macroman} is portable.  See the help for @code{file} for more
details.


@item
Do not invoke @R{} by plain @command{R}, @command{Rscript} or (on
Windows) @command{Rterm} in your examples, tests, vignettes, makefiles
or other scripts.  As pointed out in several places earlier in this
manual, use something like
@example
"$(R_HOME)/bin/Rscript"
"$(R_HOME)/bin$(R_ARCH_BIN)/Rterm"
@end example
with appropriate quotes (as, although not recommended, @env{R_HOME} can
contain spaces).

@item
Do not use @env{R_HOME} in makefiles except when passing them to the shell.
Specifically, do not use @env{R_HOME} in the argument to @code{include},
as @env{R_HOME} can contain spaces. Quoting the argument to @code{include}
does not help. A portable and the recommended way to avoid the problem of spaces in
@code{$@{R_HOME@}} is using option @code{-f} of @command{make}.  This is
easy to do with recursive invocation of @command{make}, which is also the
only usual situation when @env{R_HOME} is needed in the argument for
@code{include}.

@example
$(MAKE) -f "$@{R_HOME@}/etc$@{R_ARCH@}/Makeconf" -f Makefile.inner
@end example
@end itemize


Do be careful in what your tests (and examples) actually test.  Bad
practice seen in distributed packages include:

@itemize

@item
It is not reasonable to test the time taken by a command: you cannot
know how fast or how heavily loaded an @R{} platform might be.  At best
you can test a ratio of times, and even that is fraught with
difficulties and not advisable: for example, the garbage collector may
trigger at unpredictable times following heuristics that may change
without notice.

@item
Do not test the exact format of @R{} messages (from @R{} itself or from
other packages): They change, and they can be translated.

Packages have even tested the exact format of system error messages,
which are platform-dependent and perhaps locale-dependent.  For example,
in late 2021 @code{libcurl} changed its warning/error messages,
including when URLs are not found.

@item
If you use functions such as @code{View}, remember that in testing there
is no one to look at the output.  It is better to use something like one of
@example
if(interactive()) View(obj) else print(head(obj))
if(interactive()) View(obj) else str(obj)
@end example

@item
Be careful when comparing file paths.  There can be multiple paths to a
single file, and some of these can be very long character strings.  If
possible canonicalize paths before comparisons, but study
@code{?normalizePath} to be aware of the pitfalls.

@item
Only test the accuracy of results if you have done a formal error
analysis.  Things such as checking that probabilities numerically sum to
one are silly: numerical tests should always have a tolerance.  That the
tests on your platform achieve a particular tolerance says little about
other platforms.  @R{} is configured by default to make use of long
doubles where available, but they may not be available or be too slow
for routine use.  Most @R{} platforms use @cputype{ix86} or
@cputype{x86_64} CPUs: these may use extended precision registers on
some but not all of their FPU instructions.  Thus the achieved precision
can depend on the compiler version and optimization flags---our
experience is that 32-bit builds tend to be less precise than 64-bit
ones.  But not all platforms use those CPUs, and not all@footnote{Not
doing so is the default on Windows, overridden for the @R{} executables.
It is also the default on some Solaris compilers.} which use them
configure them to allow the use of extended precision.  In particular,
current ARM CPUs do not have extended precision nor long doubles, and
@command{clang} currently has long double the same as double on all ARM
CPUs.  On the other hand some CPUs have higher-precision modes which may
be used for @code{long double}, notably 64-bit PowerPC and Sparc.

If you must try to establish a tolerance empirically, configure and
build @R{} with @option{--disable-long-double} and use appropriate
compiler flags (such as @option{-ffloat-store} and
@option{-fexcess-precision=standard} for @command{gcc}, depending on the
CPU type@footnote{These are not needed for the default compiler settings
on @cputype{x86_64} but are likely to be needed on @cputype{ix86}.}) to
mitigate the effects of extended-precision calculations.

Tests which involve random inputs or non-deterministic algorithms should
normally set a seed or be tested for many seeds.

@item
Tests should use @code{options(warn = 1)} as reporting

@example
There were 22 warnings (use warnings() to see them)
@end example

@noindent
is pointless, especially for automated checking systems.

@item
If your package uses dates/times, ensure that it works in all timezones,
especially those near boundaries (problems have most often be seen in
@samp{Europe/London} (zero offset in Winter) and
@samp{Pacific/Auckland}, near enough the International Date line) and
with offsets not in whole hours (Adelaide, Chatham Islands, ...).

@end itemize

@node PDF size, Check timing, Writing portable packages, Writing portable packages
@subsection PDF size

There are a several tools available to reduce the size of PDF files:
often the size can be reduced substantially with no or minimal loss in
quality. Not only do large files take up space: they can stress the PDF
viewer and take many minutes to print (if they can be printed at all).

@command{qpdf} (@uref{http://qpdf.sourceforge.net/}) can compress
losslessly.  It is fairly readily available (e.g.@: it has binaries for
Windows and packages in Debian/Ubuntu/Fedora, and is installed as part
of the @acronym{CRAN} macOS distribution of @R{}).  @command{R CMD
build} has an option to run @command{qpdf} over PDF files under
@file{inst/doc} and replace them if at least 10Kb and 10% is saved.  The
full path to the @command{qpdf} command can be supplied as environment
variable @env{R_QPDF} (and is on the @acronym{CRAN} binary of @R{} for
macOS).  It seems MiKTeX does not use PDF object compression and so
@command{qpdf} can reduce considerably the files it outputs: MiKTeX's
defaults can be overridden by code in the preamble of an Sweave or
@LaTeX{} file --- see how this is done for the @R{} reference manual at
@uref{https://svn.r-project.org/R/trunk/doc/manual/refman.top}.
(Although earlier versions of @command{qpdf} are supported, versions
6.0.0 and later in some cases achieve considerably better compression.)

Other tools can reduce the size of PDFs containing bitmap images at
excessively high resolution.  These are often best re-generated (for
example @code{Sweave} defaults to 300 ppi, and 100--150 is more
appropriate for a package manual).  These tools include Adobe Acrobat
(not Reader), Apple's Preview@footnote{Select `Save as', and select
`Reduce file size' from the `Quartz filter' menu': this can be accessed
in other ways, for example by Automator.}  and Ghostscript (which
converts PDF to PDF by

@example
ps2pdf @var{options} -dAutoRotatePages=/None -dPrinted=false @var{in}.pdf @var{out}.pdf
@end example

@noindent
and suitable options might be

@example
-dPDFSETTINGS=/ebook
-dPDFSETTINGS=/screen
@end example

@noindent
See @uref{https://www.ghostscript.com/doc/current/VectorDevices.htm} for
more such and consider all the options for image downsampling).  There
have been examples in @acronym{CRAN} packages for which current versions
of Ghostscript produced much bigger reductions than earlier ones (e.g.@:
at the upgrade from @code{9.50} to @code{9.52} in March 2020).

We come across occasionally large PDF files containing excessively
complicated figures using PDF vector graphics: such figures are often
best redesigned or failing that, output as PNG files.

Option @option{--compact-vignettes} to @command{R CMD build} defaults to
value @samp{qpdf}: use @samp{both} to try harder to reduce the size,
provided you have Ghostscript available (see the help for
@code{tools::compactPDF}).

@node Check timing, Encoding issues, PDF size, Writing portable packages
@subsection Check timing

There are several ways to find out where time is being spent in the
check process.  Start by setting the environment variable
@env{_R_CHECK_TIMINGS_} to @samp{0}.  This will report the total CPU
times (not Windows) and elapsed times for installation and running
examples, tests and vignettes, under each sub-architecture if
appropriate.  For tests and vignettes, it reports the time for each as
well as the total.

Setting @env{_R_CHECK_TIMINGS_} to a positive value sets a threshold (in
seconds elapsed time) for reporting timings.

If you need to look in more detail at the timings for examples, use
option @option{--timings} to @command{R CMD check} (this is set by
@option{--as-cran}).  This adds a summary to the check output for all
the examples with CPU or elapsed time of more than 5 seconds.  It
produces a file @file{@var{mypkg}.Rcheck/@var{mypkg}-Ex.timings}
containing timings for each help file: it is a tab-delimited file which
can be read into @R{} for further analysis.

Timings for the tests and vignette runs are given at the bottom of the
corresponding log file: note that log files for successful vignette runs
are only retained if environment variable
@env{_R_CHECK_ALWAYS_LOG_VIGNETTE_OUTPUT_} is set to a true value.


@node Encoding issues, Portable C and C++ code, Check timing, Writing portable packages
@subsection Encoding issues

Care is needed if your package contains non-@acronym{ASCII} text, and in
particular if it is intended to be used in more than one locale.  It is
possible to mark the encoding used in the @file{DESCRIPTION} file and in
@file{.Rd} files, as discussed elsewhere in this manual.

First, consider carefully if you really need non-@acronym{ASCII} text.
Many users of @R{} will only be able to view correctly text in their
native language group (e.g.@: Western European, Eastern European,
Simplified Chinese) and @acronym{ASCII}.@footnote{except perhaps some
special characters such as backslash and hash which may be taken over
for currency symbols.}.  Other characters may not be rendered at all,
rendered incorrectly, or cause your @R{} code to give an error.  For
@file{.Rd} documentation, marking the encoding and including
@acronym{ASCII} transliterations is likely to do a reasonable job.  The
set of characters which is commonly supported is wider than it used to
be around 2000, but non-Latin alphabets (Greek, Russian, Georgian,
@dots{}) are still often problematic and those with double-width
characters (Chinese, Japanese, Korean, emoji) often need specialist
fonts to render correctly.

Several @acronym{CRAN} packages have messages in their @R{} code in French (and a
few in German).  A better way to tackle this is to use the
internationalization facilities discussed elsewhere in this manual.

Function @code{showNonASCIIfile} in package @pkg{tools} can help in
finding non-@acronym{ASCII} bytes in files.

There is a portable way to have arbitrary text in character strings
(only) in your @R{} code, which is to supply them in Unicode as
@samp{\uxxxx} escapes.  If there are any characters not in the current
encoding the parser will encode the character string as UTF-8 and mark
it as such.  This applies also to character strings in datasets: they
can be prepared using @samp{\uxxxx} escapes or encoded in UTF-8 in a
UTF-8 locale, or even converted to UTF-8 @emph{via} @code{iconv()}.  If
you do this, make sure you have @samp{R (>= 2.10)} (or later) in the
@samp{Depends} field of the @file{DESCRIPTION} file.

@R{} sessions running in non-UTF-8 locales will if possible re-encode
such strings for display (and this is done by @command{RGui} on Windows,
for example).  Suitable fonts will need to be selected or made
available@footnote{Typically on a Unix-alike this is done by telling
@command{fontconfig} where to find suitable fonts to select glyphs
from.}  both for the console/terminal and graphics devices such as
@samp{X11()} and @samp{windows()}.  Using @samp{postscript} or
@samp{pdf} will choose a default 8-bit encoding depending on the
language of the UTF-8 locale, and your users would need to be told how
to select the @samp{encoding} argument.

Note that the previous two paragraphs only apply to character strings in
@R{} code.  Non-ASCII characters are particularly prevalent in comments
(in the @R{} code of the package, in examples, tests, vignettes and even
in the @file{NAMESPACE} file) but should be avoided there.  Most commonly
people use the Windows extensions to Latin-1 (often directional single
and double quotes, ellipsis, bullet and en and em dashes) which are not
supported in strict Latin-1 locales nor in CJK locales on Windows.  A
surprisingly common misuse is to use a right quote in @samp{don't}
instead of the correct apostrophe.

If you want to run @command{R CMD check} on a Unix-alike over a package
that sets a package encoding in its @file{DESCRIPTION} file @emph{and do
not use a UTF-8 locale} you may need to specify a suitable locale
@emph{via} environment variable @env{R_ENCODING_LOCALES}.  The default
is equivalent to the value

@example
"latin1=en_US:latin2=pl_PL:UTF-8=en_US.UTF-8:latin9=fr_FR.iso885915@@euro"
@end example

@noindent
(which is appropriate for a system based on @code{glibc}: macOS requires
@code{latin9=fr_FR.ISO8859-15}) except that if the current locale is
UTF-8 then the package code is translated to UTF-8 for syntax checking,
so it is strongly recommended to check in a UTF-8 locale.

@node Portable C and C++ code, Binary distribution, Encoding issues, Writing portable packages
@subsection Portable C and C++ code

@menu
* Common symbols::
@end menu

Writing portable C and C++ code is mainly a matter of observing the
standards (C99, C++11 or where declared C++14/17/20) and testing that
extensions (such as POSIX functions) are supported.

@strong{C++ standards}: From version 3.6.0 (3.6.2 on Windows), @R{}
defaulted to C++11 where available@footnote{which it is on all known
platforms, and is required as from @R{} 4.0.0}; @R{} 4.1.0 defaults to
C++14 (where available).  However, in earlier
versions the default standard was that of the compiler used, often C++98
or C++14, and the default is likely to change in future.  For maximal
portability a package should either specify a standard (@pxref{Using C++
code}) or be tested under all of C++11, C++98 and C++14.  (Specifying
C++14 will limit portability.)

Note that the `TR1' C++ extensions are not part of any of these
standards and the @code{<tr1/@var{name}>} headers are not supplied by some of
the compilers used for @R{}, including on macOS.  (Use the C++11
versions instead.)

As noted elsewhere, the C++17 standard was finalized only recently (Dec
2017) and support for it is patchy or absent on several platforms used
for @R{}.  So portable code should not require C++17 (let alone C++20).

@c Centos had 10 years of support, for Centos 7 expiring in June 2024.
@c RHEL has 10 years, with more extended support, up to 4 years.
@c Ubuntu LTS has 5 years' general support and 8 extended support.
@c Ubuntu 16.04 (EOL 2021-04) came with GCC 5.4 but 7 was available
@c Ubuntu 18.04 came with GCC 7.3 or 7.4.
@c RHEL/Centos 6 came with GCC 4.4.
@c RHEL/Centos 7 came with GCC 4.8, with 4.4 available.
@c RHEL/Centos 8 has GCC 8.
A common error is to assume recent versions of compilers or OSes.  In
production environments `long term support' versions of OSes may be in
use for many years,@footnote{Ubuntu provides 5 years of support (but
people were running 14.04 after 7 years) and RHEL provides 10 years full
support and up to 14 with extended support.} and their compilers may not
be updated during that time.  For example, GCC 4.8 was still in use in
2021 and could be (in RHEL 7) until 2028: that supports neither C++14
nor C++17.

The POSIX standards only require recently-defined functions to be
declared if certain macros are defined with large enough values, and on
some compiler/OS combinations@footnote{This is seen on Linux, Solaris
and FreeBSD, although each has other ways to turn on all extensions,
e.g.@: defining @code{_GNU_SOURCE}, @code{__EXTENSIONS__} or
@code{_BSD_SOURCE}: the GCC compilers by default define
@code{_GNU_SOURCE} unless a strict standard such as @option{-std=c99} is
used.  On macOS extensions are declared unless one of these macros is
given too small a value.} they are not declared otherwise.  So you may
need to include something like one of @footnote{Solaris 10 does not
recognize this value of @code{_POSIX_C_SOURCE}, nor values of
@code{_XOPEN_SOURCE} beyond 600 (700 corresponds to POSIX 2008).
Further, the value of 500 is not allowed in C99 mode, @R{}'s default for
C code.}
@example
#define _XOPEN_SOURCE 600
@end example
@noindent
or
@example
#ifdef __GLIBC__
# define _POSIX_C_SOURCE 200809L
#endif
@end example
@noindent
before @emph{any} headers.  (@code{strdup}, @code{strncasecmp} and
@code{strnlen} are such functions --  there are several older platforms
which do not have the POSIX 2008 function @code{strnlen}.)

However, some common errors are worth pointing out here.  It can be
helpful to look up functions at
@uref{https://www.cplusplus.com/reference/} or
@uref{https://en.cppreference.com/w/} and compare what is defined in the
various standards.

More care is needed for functions such as @code{mallinfo} which are not
specified by any of these standards---hopefully the @command{man} page
on your system will tell you so.  Searching online for such pages for
various OSes (preferably at least Linux, macOS and Solaris, and the
FreeBSD manual pages at @uref{https://www.freebsd.org/cgi/man.cgi} allow
you to select many OSes@footnote{`SunOS 5.10' is the most current for
Solaris.}) should reveal useful information but a @file{configure}
script is likely to be needed to check availability and functionality.

Both the compiler and OS (@emph{via} system header files, which may
differ by architecture even for nominally the same OS) affect the
compilability of C/C++ code.  Compilers from the GCC, @command{clang},
Intel and Oracle Developer Studio suites are routinely used with @R{},
and both @command{clang} and Oracle have more than one implementation of
C++ headers and library.  The range of possibilities makes comprehensive
empirical checking impossible, and regrettably compilers are patchy at
best on warning about non-standard code.

@itemize
@item
Mathematical functions such as @code{sqrt} are defined in C++11 for
floating-point arguments: @code{float}, @code{double}, @code{long
double} and possibly more.  The standard specifies what happens with an
argument of integer type but this is not always implemented, resulting
in a report of `overloading ambiguity': this is commonly seen on
Solaris, but for @code{pow} also seen on macOS (and other platforms
using @command{clang++}).

A not-uncommonly-seen problem is to mistakenly call @code{floor(x/y)} or
@code{ceil(x/y)} for @code{int} arguments @code{x} and @code{y}.  Since
@code{x/y} does integer division, the result is of type @code{int} and
`overloading ambiguity' may be reported.  Some people have (pointlessly)
called @code{floor} and @code{ceil} on arguments of integer type, which
may have an `overloading ambiguity'.

A surprising common misuse is things like @code{pow(10, -3)}: this
should be the constant @code{1e-3}.  Note that there are constants such
as @code{M_SQRT2} defined in @file{Rmath.h}@footnote{often taken from
the toolchain's headers.} for @code{sqrt(2.0)}, frequently mis-coded as
@code{sqrt(2)}.

@item
Function @code{fabs} is defined only for floating-point types, except in
C++11 which has overloads for @code{std::fabs} in @file{<cmath>} for
integer types.  Function @code{abs} is defined in C99's
@file{<stdlib.h>} for @code{int} and in C++'s @file{<cstdlib>} for
integer types, overloaded in @file{<cmath>} for floating-point types.
C++11 has additional overloads for @code{std::abs} in @file{<cmath>} for
integer types.  The effect of calling @code{abs} with a floating-point
type is implementation-specific: it may truncate to an integer.  For
clarity and to avoid compiler warnings, use @code{abs} for integer types
and @code{fabs} for double values, and when using C++11 include
@file{<cmath>} and use the @code{std::} prefix.

@item
It is an error (and make little sense, although has been seen) to call
macros/functions @code{isnan}, @code{isinf} and @code{isfinite} for integer
arguments: a few compilers give a compilation error.  Function
@code{finite} is obsolete, and some compilers will warn about its use.

@item
The GNU C/C++ compilers support a large number of non-portable
extensions.  For example, @code{INFINITY} (which is a @emph{float} value
in C99 and C++11), for which @R{} provides the portable double value
@code{R_PosInf} (and @code{R_NegInf} for @code{-INFINITY}).  And
@code{NAN}@footnote{also part of C++11 and later.} is just one NaN
@emph{float} value: for use with @R{}, @code{NA_REAL} is often what is
intended, but @code{R_NaN} is also available.

Some (but not all) extensions are listed at
@uref{https://gcc.gnu.org/onlinedocs/gcc/C-Extensions.html} and
@uref{https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Extensions.html}.

Other GNU extensions which have bitten package writers are the use of
non-portable characters such as @samp{$} in identifiers and use of C++
headers under @file{ext}.

The GNU Fortran compiler also supports a large number of non-portable
extensions, the most commonly encountered one being
@code{ISNAN}@footnote{There is a portable way to do this in Fortran 2003
(@code{ieee_is_nan()} in module @code{ieee_arithmetic}), but ironically
that is not supported in the versions 4.x of GNU Fortran.  A pretty
robust alternative is to test @code{if(my_var /= my_var)}.}.  Some are
listed at
@uref{https://gcc.gnu.org/onlinedocs/gfortran/Extensions-implemented-in-GNU-Fortran.html}.
One that frequently catches package writers is that it allows
out-of-order declarations: in standard-conformant Fortran variables must
be declared (explicitly or implicitly) before use in other declarations
such as dimensions.

GNU Fortran 10 and later give a compilation error for the previously
widespread practice of passing a Fortran array element where an array is
expected, or a scalar instead of a length-one array.  See
@uref{https://gcc.gnu.org/gcc-10/porting_to.html}.

@item
Including C-style headers in C++ code is not portable.  Including the
legacy header@footnote{which often is the same as the header included by
the C compiler, but some compilers have wrappers for some of the C
headers.} @file{math.h} in C++ code may conflict with @file{cmath} which
may be included by other headers.  In C++11, functions like @code{sqrt}
and @code{isnan} are defined for @code{double} arguments in
@file{math.h} and for a range of types including @code{double} in
@file{cmath}.  Similar issues have been seen for @file{stdlib.h} and
@file{cstdlib}.  Including the C++ header first used to be a sufficient
workaround but for some 2016 compilers only one could be included.

@item
Be careful to include the headers which define the functions you use.
Some compilers/OSes include other system headers in their headers which
are not required by the standards, and so code may compile on such
systems and not on others.  (A prominent example is the C++ header
@code{<random>} which is indirectly included by @code{<algorithm>} by
@command{g++}.  Another issue is the C header @code{<time.h>} which is
included by other headers on Linux and Windows but not macOS nor
Solaris.)  @command{g++}@tie{}11 often needs explicit inclusion of the
C++ headers @code{<limits>} (for @code{numeric_limits}) or
@code{<exception>} (for @code{set_terminate} and similar), whereas
earlier versions included these in other headers.

Note that @code{malloc}, @code{calloc}, @code{realloc} and @code{free}
are defined by C99 in the header @file{stdlib.h} and (in the
@code{std::} namespace) by C++ header @file{cstdlib}.  Some earlier
implementations used a header @file{malloc.h}, but that is not portable
and does not exist on macOS.

This also applies to types such as @code{ssize_t}.  The POSIX standards
say that is declared in headers @code{unistd.h} and @code{sys/types.h},
and the latter is often included indirectly by other headers on some
but not all systems.

Similarly for constants: for example @code{SIZE_MAX} is defined in
@code{stdint.h} alongside @code{size_t}.

@item
Some headers are not portable: we have just mentioned @file{malloc.h}
and often @acronym{CRAN} submissions attempt to use @file{endian.h}.
The latter is a @code{glibc} extension: some OSs have
@file{machine/endian.h} or @file{sys/endian.h} but some have neither.
@c macOS has machine/endian.h.  Solaris has none.

@item
Use @code{#include "my.h"} not @code{#include <my.h>} for headers in
your package.  The second form is intended for system headers and the
search order for such headers is platform-dependent (and may not include
the current directory).  For extra safety, name headers in a way that
cannot be confused with a system header so not, for example,
@file{types.h}.

@item
For C++ code, be careful to specify namespaces where needed.  Many
functions are defined by the standards to be in the @code{std}
namespace, but @command{g++} puts many such also in the C++ main
namespace.  One way to do so is to use declarations such as
@example
using std::floor;
@end example
@noindent
but it is usually preferable to use explicit namespace prefixes in the code.

Examples seen in @acronym{CRAN} packages include
@example
abs acos atan bind calloc ceil div exp fabs floor fmod free log malloc
memcpy memset pow printf qsort round sin sprintf sqrt strcmp strcpy
strerror strlen strncmp strtol tan trunc
@end example
@noindent
This problem is less common than it used to be, but in 2019
@command{clang} did not have @code{bind} in the main namespace.  Also
seen has been type @code{size_t} defined only in the @code{std} namespace.
@c with g++ on Solaris 10.

@item
@c including clang as from 4.0.0
Some C++ compilers refuse to compile constructs such as
@example
      if(ptr > 0) @{ ....@}
@end example
@noindent
which compares a pointer to the integer @code{0}.  This could just use
@code{if(ptr)} (pointer addresses cannot be negative) but if needed
pointers can be tested against @code{nullptr} (C++11) or @code{NULL}.

@item
Macros defined by the compiler/OS can cause problems.  Identifiers
starting with an underscore followed by an upper-case letter or another
underscore are reserved for system macros and should not be used in
portable code (including not as guards in C/C++ headers).  Other macros,
typically upper-case, may be defined by the compiler or system headers
and can cause problems.
@c http://lists.x.org/archives/xorg-devel/2013-November/038808.html
The most common issue involves the names of the Intel CPU registers such
as @code{CS}, @code{DS}, @code{ES}, @code{FS}, @code{GS} and @code{SS}
(and more with longer abbreviations@footnote{including @code{EAX},
@code{EBP}, @code{EBX}, @code{ECX}, @code{EDI},@code{EDX}, @code{EFL},
@code{EIP}, @code{ESI} and @code{ESP} .}) defined on i586/x64 Solaris in
@file{<sys/regset.h>} and often included indirectly by @file{<stdlib.h>}
and other core headers.  Further examples are @code{ERR}, @code{VERSION},
@code{LITTLE_ENDIAN}, @code{zero} and @code{I} (which is defined in
Solaris' @file{<complex.h>} as a compiler intrinsic for the imaginary
unit).  Some of these can be avoided by defining @code{_POSIX_C_SOURCE}
before including any system headers, but it is better to only use
all-upper-case names which have a unique prefix such as the package
name.

@item
@code{typedef}s in OS headers can conflict with those in the package:
examples include @code{ulong} on several OSes, @code{index_t} and
@code{single} on Solaris and @code{thread} using @command{clang++} from
version 9.  (Note that these may conflict with other uses as
identifiers, e.g.@: defining a C++ function called @code{single}.)
@c as done by package Emcdf in June 2017.
@c and thread as a typedef by RVowalWabbit
The POSIX standard reserves (in §2.2.2) all identifiers ending in
@code{_t}.

@item
Some compilers do not allow a space between @code{-D} and the macro to
be defined.  Similarly for @code{-U}.

@item
If you use OpenMP, check carefully that you have followed the advice in
the subsection on @ref{OpenMP support}.  In particular, any use of
OpenMP in C/C++ code will need to use
@example
#ifdef _OPENMP
# include <omp.h>
#endif
@end example
@noindent
Any use of OpenMP functions, e.g.@: @code{omp_set_num_threads}, also
needs to be conditioned.  To avoid incessant warnings such as
@example
warning: ignoring #pragma omp parallel [-Wunknown-pragmas]
@end example

@noindent
uses of such pragmas should also be conditioned (or commented out if
they are used in code in a package not enabling OpenMP on any platform).

Do not hardcode @option{-lgomp}: not only is that specific to the
GCC family of compilers, using the correct linker flag often sets up the
run-time path to the library.

@item
Package authors commonly assume things are part of C/C++ when they are
not: the most common example is POSIX function @code{strdup}.  The most
common C library on Linux, @code{glibc}, will hide the declarations of
such extensions unless a `feature-test macro' is defined @strong{before}
(almost) any system header is included.  So for @code{strdup} you need
@example
#define _POSIX_C_SOURCE 200809L
...
#include <string.h>
...
strdup call(s)
@end example
@noindent
where the appropriate value can be found by @command{man strdup} on
Linux.  (Use of @code{strncasecmp} is similar.)

However, modes of @command{gcc} with `GNU EXTENSIONS' (which are the
default, either @option{-std=gnu99} or @option{-std=gnu11}) declare
enough macros to ensure that missing declarations are rarely seen.

This applies also to constants such as @code{M_PI} and @code{M_LN2},
which are part of the X/Open standard: to use these define
@code{_XOPEN_SOURCE} before including any headers, or include the @R{}
header @file{Rmath.h}.

@item
Using @code{alloca} portably is tricky: it is neither an ISO C/C++ nor a
POSIX function.  An adequately portable preamble is
@example
#ifdef __GNUC__
/* Includes GCC, clang and Intel compilers */
# undef alloca
# define alloca(x) __builtin_alloca((x))
#elif defined(__sun) || defined(_AIX)
/* this is necessary (and sufficient) for Solaris 10 and AIX 6: */
# include <alloca.h>
#endif
@end example

@item
Compiler writers feel free to implement features from later standards
than the one specified, so for example they may implement or warn on
C++14/17/20 features.  Portable code will not use such features -- it
can be hard to know what they are but the most common warnings are
@example
'register' storage class specifier is deprecated and incompatible with C++17

ISO C++11 does not allow conversion from string literal to 'char *'
@end example
@noindent
(where conversion should be to @code{const char *}).  Keyword
@code{register} was not mentioned in C++98, deprecated in C++11 and
removed in C++17.

There are quite a lot of other C++98 features deprecated in C++11 and
removed in C++17, and @command{clang} 9 and later warn about
them. Examples include @code{bind1st}/@code{bind2nd} (use
@code{std::bind} or
lambdas@footnote{@uref{https://stackoverflow.com/questions/32739018/a-replacement-for-stdbind2nd}})
@code{std::auto_ptr} (replaced by @code{std::unique_ptr}),
@code{std:;mem_fun_ref} and @code{std::ptr_fun}.

@item
Be careful about including C headers in C++ code.  Issues include
@itemize
@item
Use of the @code{register} storage class specifier (see the previous
item).
@item
The C99 keyword @code{restrict} is not part of@footnote{it is allowed
but ignored in system headers.} any C++ standard and is rejected by some
C++ compilers.
@c but package treatSens attempted to use it.
@c http://stackoverflow.com/questions/6434549/does-c11-add-the-c99-restrict-specifier-if-not-why-not
@item
Inclusion by such headers of C-style headers such as @file{math.h} (see above).
@end itemize
@noindent
The most portable way to interface to other software with a C API is to
use C code (which can normally be mixed with C++ code in a package).

@item
@code{reinterpret_cast} in C++ is not safe for pointers: for example the types
may have different alignment requirements.  Use @code{memcpy} to copy
the contents to a fresh variable of the destination type.
@c seen in 2019 for casts to both int32 and double from a byte stream

@item
Avoid platform-specific code if at all possible, but if you need to test
for a platform ensure that all platforms are covered.  For example,
@code{__unix__} is not defined on all Unix-alikes, in particular not on
macOS.  A reasonably portable way to condition code for a Unix-alike is
@example
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
#endif
@end example
@noindent
but
@example
#ifdef _WIN32
// Windows-specific code
#else
// Unix-alike code
#endif
@end example
@noindent
would be better.  For a Unix-alike it is much better to use
@command{configure} to test for the functionality needed than make
assumptions about OSes (and people all too frequently forget @R{} is
used on platforms other than Linux, Windows and macOS --- and some
forget macOS).

@item
Headers in subdirectories are often not portable.  For C++, this
includes @file{bits/}, @file{tr1/} and @file{tr2/}, none of which exist
on macOS (and @file{ext/} exists there but with different content from
@command{g++}-based platforms).  Header @file{bits/stdc++.h} is both not
portable and not recommended for end-user code even on platforms which
include it.

@item
Be careful if using @code{malloc} or @code{calloc}.  First, their return
value must always be checked to see if the allocation succeeded -- it is
almost always easier to use @R{}'s @code{Calloc}, which does check.
Second, the first argument is of type @code{size_t} and recent compilers
are warning about passing signed arguments (which could get promoted to
ridiculously large values).

@end itemize

Some additional information for C++ is available at
@uref{https://journal.r-project.org/archive/2011-2/RJournal_2011-2_Plummer.pdf}
by Martyn Plummer.

@node Common symbols,  , Portable C and C++ code, Portable C and C++ code
@subsubsection Common symbols

Most OSes (including all those commonly used for @R{}) have the concept
of `tentative definitions' where global C variables are defined without
an initializer.  Traditionally the linker resolves all tentative
definitions of the same variable in different object files to the same
object, or to a non-tentative definition.  However,
@command{gcc}@tie{}10@footnote{see
@uref{https://gcc.gnu.org/gcc-10/porting_to.html}.} and
LLVM @command{clang}@tie{}11@footnote{See
@uref{https://prereleases.llvm.org/11.0.0/rc2/tools/clang/docs/ReleaseNotes.html#modified-compiler-flags}.}
have changed their default so that tentative definitions cannot be
merged and the linker will give an error if the same variable is defined
in more than one object file.  To avoid this, all but one of the C
source files should declare the variable @code{extern} --- which means
that any such variables included in header files need to be declared
@code{extern}.  A commonly used idiom (including by @R{} itself) is to
define all global variables as @code{extern} in a header, say
@file{globals.h} (and nowhere else), and then in one (and one only)
source file use
@example
#define extern
# include "globals.h"
#undef extern
@end example

A cleaner approach is not to have global variables at all, but to place
in a single file common variables (declared @code{static}) followed by
all the functions which make use of them: this may result in more
efficient code.

The `modern' behaviour can be seen@footnote{In principle this could
depend on the OS, but has been checked on Linux and macOS.} by using
compiler flag @option{-fno-common} as part of @samp{CFLAGS} in earlier
versions of @command{gcc} and @command{clang}.

@option{-fno-common} is said to be particularly beneficial for ARM cpus.

This is not pertinent to C++ which does not permit tentative definitions.

@node Binary distribution,  , Portable C and C++ code, Writing portable packages
@subsection Binary distribution

If you want to distribute a binary version of a package on Windows or
macOS, there are further checks you need to do to check it is portable:
it is all too easy to depend on external software on your own machine
that other users will not have.

For Windows, check what other DLLs your package's DLL depends on
(`imports' from in the DLL tools' parlance).  A convenient GUI-based
tool to do so is `Dependency Walker'
(@uref{https://www.dependencywalker.com/}) for both 32-bit and 64-bit
DLLs -- note that this will report as missing links to @R{}'s own DLLs
such as @file{R.dll} and @file{Rblas.dll}.  The command-line tool
@command{objdump} in the appropriate toolchain will also reveal what
DLLs are imported from.  If you use a toolchain other than one provided
by the @R{} developers or use your own makefiles, watch out in
particular for dependencies on the toolchain's runtime DLLs such as
@file{libgfortran}, @file{libstdc++} and @file{libgcc_s}.

For macOS, using @code{R CMD otool -L} on the package's shared object(s)
in the @file{libs} directory will show what they depend on: watch for
any dependencies in @file{/usr/local/lib} or
@file{/usr/local/gfortran/lib}, notably @file{libgfortran.?.dylib} and
@file{libquadmath.0.dylib}.
@ifset UseExternalXrefs
(For ways to fix these,
@pxref{Building binary packages, , Building binary-packages,
R-admin, R Installation and Administration}.)
@end ifset
@ifclear UseExternalXrefs
(For ways to fix these, see the section `Building binary packages' in
the `R Installation and Administration' manual'.)
@end ifclear

Many people (including the @acronym{CRAN} package repository) will not
accept source packages containing binary files as the latter are a
security risk.  If you want to distribute a source package which needs
external software on Windows or macOS, options include
@itemize
@item
To arrange for installation of the package to download the
additional software from a URL, as e.g.@: package @CRANpkg{Cairo} does.

@item
(For @acronym{CRAN}.)
To negotiate with Uwe Ligges to host the additional components on
WinBuilder, and write a @file{configure.win} file to install them.

@end itemize

@noindent
Be aware that license requirements you may require you to supply the
sources for the additional components (and will if your package has a
GPL-like license).


@node Diagnostic messages, Internationalization, Writing portable packages, Creating R packages
@section Diagnostic messages

Diagnostic messages can be made available for translation, so it is
important to write them in a consistent style.  Using the tools
described in the next section to extract all the messages can give a
useful overview of your consistency (or lack of it).
Some guidelines follow.

@itemize
@item
Messages are sentence fragments, and not viewed in isolation.  So it is
conventional not to capitalize the first word and not to end with a
period (or other punctuation).

@item
Try not to split up messages into small pieces.  In C error messages use
a single format string containing all English words in the messages.

In @R{} error messages do not construct a message with @code{paste} (such
messages will not be translated) but @emph{via} multiple arguments to
@code{stop} or @code{warning}, or @emph{via} @code{gettextf}.

@item
Do not use colloquialisms such as ``can't'' and ``don't''.

@item
Conventionally single quotation marks are used for quotations such as

@example
'ord' must be a positive integer, at most the number of knots
@end example

@noindent
and double quotation marks when referring to an @R{} character string or
a class, such as

@example
'format' must be "normal" or "short" - using "normal"
@end example

Since @acronym{ASCII} does not contain directional quotation marks, it
is best to use @samp{'} and let the translator (including automatic
translation) use directional quotations where available.  The range of
quotation styles is immense: unfortunately we cannot reproduce them in a
portable @code{texinfo} document.  But as a taster, some languages use
`up' and `down' (comma) quotes rather than left or right quotes, and
some use guillemets (and some use what Adobe calls `guillemotleft' to
start and others use it to end).

In @R{} messages it is also possible to use @code{sQuote} or @code{dQuote} as in

@example
        stop(gettextf("object must be of class %s or %s",
                      dQuote("manova"), dQuote("maov")),
             domain = NA)
@end example

@item
Occasionally messages need to be singular or plural (and in other
languages there may be no such concept or several plural forms --
Slovenian has four).  So avoid constructions such as was once used in
@code{library}

@example
if((length(nopkgs) > 0) && !missing(lib.loc)) @{
    if(length(nopkgs) > 1)
        warning("libraries ",
                paste(sQuote(nopkgs), collapse = ", "),
                " contain no packages")
    else
        warning("library ", paste(sQuote(nopkgs)),
                " contains no package")
@}
@end example

@noindent
and was replaced by

@example
if((length(nopkgs) > 0) && !missing(lib.loc)) @{
    pkglist <- paste(sQuote(nopkgs), collapse = ", ")
    msg <- sprintf(ngettext(length(nopkgs),
                            "library %s contains no packages",
                            "libraries %s contain no packages",
                            domain = "R-base"),
                   pkglist)
    warning(msg, domain=NA)
@}
@end example

@noindent
Note that it is much better to have complete clauses as here, since
in another language one might need to say
`There is no package in library %s' or
`There are no packages in libraries %s'.

@end itemize

@node Internationalization, CITATION files, Diagnostic messages, Creating R packages
@section Internationalization

There are mechanisms to translate the @R{}- and C-level error and warning
messages.  There are only available if @R{} is compiled with NLS support
(which is requested by @command{configure} option @option{--enable-nls},
the default).

The procedures make use of @code{msgfmt} and @code{xgettext} which are
part of @acronym{GNU} @code{gettext} and this will need to be installed:
Windows users can find pre-compiled binaries at
@uref{https://www.stats.ox.ac.uk/pub/Rtools/goodies/gettext-tools.zip}.

@menu
* C-level messages::
* R messages::
* Preparing translations::
@end menu

@node C-level messages, R messages, Internationalization, Internationalization
@subsection C-level messages

The process of enabling translations is

@itemize
@item
In a header file that will be included in all the C (or C++ or Objective
C/C++) files containing messages that should be translated, declare

@example
#include <R.h>  /* to include Rconfig.h */

#ifdef ENABLE_NLS
#include <libintl.h>
#define _(String) dgettext ("@var{pkg}", String)
/* replace @var{pkg} as appropriate */
#else
#define _(String) (String)
#endif
@end example

@item
For each message that should be translated, wrap it in @code{_(...)},
for example

@example
error(_("'ord' must be a positive integer"));
@end example

If you want to use different messages for singular and plural forms, you
need to add

@example
#ifndef ENABLE_NLS
#define dngettext(pkg, String, StringP, N) (N == 1 ? String : StringP)
#endif
@end example

@noindent
and mark strings by

@example
dngettext("@var{pkg}", @var{<singular string>}, @var{<plural string>}, n)
@end example

@item
In the package's @file{src} directory run

@example
xgettext --keyword=_ -o @var{pkg}.pot *.c
@end example

@end itemize

The file @file{src/@var{pkg}.pot} is the template file, and
conventionally this is shipped as @file{po/@var{pkg}.pot}.

@node R messages, Preparing translations, C-level messages, Internationalization
@subsection R messages

Mechanisms are also available to support the automatic translation of
@R{} @code{stop}, @code{warning} and @code{message} messages.  They make
use of message catalogs in the same way as C-level messages, but using
domain @code{R-@var{pkg}} rather than @code{@var{pkg}}.  Translation of
character strings inside @code{stop}, @code{warning} and @code{message}
calls is automatically enabled, as well as other messages enclosed in
calls to @code{gettext} or @code{gettextf}.  (To suppress this, use
argument @code{domain=NA}.)

Tools to prepare the @file{R-@var{pkg}.pot} file are provided in package
@pkg{tools}: @code{xgettext2pot} will prepare a file from all strings
occurring inside @code{gettext}/@code{gettextf}, @code{stop},
@code{warning} and @code{message} calls.  Some of these are likely to be
spurious and so the file is likely to need manual editing.
@code{xgettext} extracts the actual calls and so is more useful when
tidying up error messages.

The @R{} function @code{ngettext} provides an interface to the C
function of the same name: see example in the previous section.  It is
safest to use @code{domain="R-@var{pkg}"} explicitly in calls to
@code{ngettext}, and necessary for earlier versions of @R{} unless they
are calls directly from a function in the package.


@node Preparing translations,  , R messages, Internationalization
@subsection Preparing translations

Once the template files have been created, translations can be made.
Conventional translations have file extension @file{.po} and are placed
in the @file{po} subdirectory of the package with a name that is either
@samp{@var{ll}.po} or @samp{R-@var{ll}.po} for translations of the C and @R{}
messages respectively to language with code @samp{@var{ll}}.

@ifset UseExternalXrefs
@xref{Localization of messages, , Localization of messages, R-admin,
R Installation and Administration}, for details of language codes.
@end ifset
@ifclear UseExternalXrefs
See `Localization of messages' in `R Installation and Administration',
for details of language codes.
@end ifclear

There is an @R{} function, @code{update_pkg_po} in package @pkg{tools},
to automate much of the maintenance of message translations.  See its
help for what it does in detail.

If this is called on a package with no existing translations, it creates
the directory @file{@var{pkgdir}/po}, creates a template file of @R{}
messages, @file{@var{pkgdir}/po/R-@var{pkg}.pot}, within it, creates the
@samp{en@@quot} translation and installs that.  (The @samp{en@@quot}
pseudo-language interprets quotes in their directional forms in suitable
(e.g.@: UTF-8) locales.)

If the package has C source files in its @file{src} directory
that are marked for translation, use

@example
touch @var{pkgdir}/po/@var{pkg}.pot
@end example

@noindent
to create a dummy template file, then call @code{update_pkg_po} again
(this can also be done before it is called for the first time).

When translations to new languages are added in the @file{@var{pkgdir}/po}
directory, running the same command will check and then
install the translations.

If the package sources are updated, the same command will update the
template files, merge the changes into the translation @file{.po} files
and then installed the updated translations.  You will often see that
merging marks translations as `fuzzy' and this is reported in the
coverage statistics.  As fuzzy translations are @emph{not} used, this is
an indication that the translation files need human attention.

The merged translations are run through @code{tools::checkPofile} to
check that C-style formats are used correctly: if not the mismatches are
reported and the broken translations are not installed.

This function needs the GNU @command{gettext-tools} installed and on the
path: see its help page.


@findex CITATION
@cindex citation
@node CITATION files, Package types, Internationalization, Creating R packages
@section CITATION files

An installed file named @file{CITATION} will be used by the
@code{citation()} function.  (It should be in the @file{inst}
subdirectory of the package sources.)

The @file{CITATION} file is parsed as @R{} code (in the package's
declared encoding, or in @acronym{ASCII} if none is declared).
It will contain calls to function @code{bibentry}.
Here is that for @CRANpkg{nlme}:

@example
## R package reference generated from DESCRIPTION metadata
citation(auto = meta)

## NLME book
bibentry(bibtype = "Book",
         title = "Mixed-Effects Models in S and S-PLUS",
         author = c(person(c("Jos@'e", "C."), "Pinheiro"),
                    person(c("Douglas", "M."), "Bates")),
         year = "2000", publisher = "Springer", address = "New York",
         doi = "10.1007/b98882")
@end example

Note how the first call auto-generates citation information
from object @code{meta}, a parsed version of the @file{DESCRIPTION} file
-- it is tempting to hardcode such information, but it normally then
gets outdated.  How the first entry would look like as a @code{bibentry}
call can be seen from
@code{print(citation("@var{pkgname}", auto = TRUE), style = "R")}
for any installed package.  Auto-generated information is
returned by default if no @file{CITATION} file is present.

See @code{?bibentry} for further details of the
information which can be provided.
In case a bibentry contains @LaTeX{} markup (e.g., for accented
characters or mathematical symbols), it may be necessary to provide a
text representation to be used for printing @emph{via} the
@code{textVersion} argument to @code{bibentry}.  E.g., earlier versions
of @CRANpkg{nlme} additionally used something like

@example
         textVersion =
         paste0("Jose Pinheiro, Douglas Bates, Saikat DebRoy, ",
                "Deepayan Sarkar and the R Core Team (",
                sub("-.*", "", meta$Date),
                "). nlme: Linear and Nonlinear Mixed Effects Models. ",
                sprintf("R package version %s", meta$Version), ".")
@end example

The @file{CITATION} file should itself produce no output when
@code{source}-d.

It is desirable (and essential for @acronym{CRAN}) that the
@file{CITATION} file does not contain calls to functions such as
@code{packageDescription} which assume the package is installed in a
library tree on the package search path.

@node Package types, Services, CITATION files, Creating R packages
@section Package types

The @file{DESCRIPTION} file has an optional field @code{Type} which if
missing is assumed to be @samp{Package}, the sort of extension discussed
so far in this chapter.  Currently one other type is recognized; there
used also to be a @samp{Translation} type.

@menu
* Frontend::
@end menu

@node Frontend,  , Package types, Package types
@subsection Frontend

This is a rather general mechanism, designed for adding new front-ends
such as the former @pkg{gnomeGUI} package (see the @file{Archive} area on
@acronym{CRAN}).  If a @file{configure} file is found in the top-level
directory of the package it is executed, and then if a @file{Makefile}
is found (often generated by @file{configure}), @code{make} is called.
If @code{R CMD INSTALL --clean} is used @code{make clean} is called.  No
other action is taken.

@code{R CMD build} can package up this type of extension, but @code{R
CMD check} will check the type and skip it.

Many packages of this type need write permission for the @R{}
installation directory.

@node Services,  , Package types, Creating R packages
@section Services

Several members of the @R{} project have set up services to assist those
writing @R{} packages, particularly those intended for public
distribution.

@uref{https://win-builder.r-project.org, win-builder.r-project.org}
offers the automated preparation of (32/64-bit) Windows binaries from
well-tested source packages.

R-Forge (@uref{https://R-Forge.r-project.org, R-Forge.r-project.org}) and
RForge (@uref{https://www.rforge.net, www.rforge.net}) are similar
services with similar names.  Both provide source-code management
through SVN, daily building and checking, mailing lists and a repository
that can be accessed @emph{via} @code{install.packages} (they can be
selected by @code{setRepositories} and the GUI menus that use it).
Package developers have the opportunity to present their work on the
basis of project websites or news announcements.  Mailing lists, forums
or wikis provide useRs with convenient instruments for discussions and
for exchanging information between developers and/or interested useRs.


@node Writing R documentation files, Tidying and profiling R code, Creating R packages, Top
@chapter Writing R documentation files
@cindex Documentation, writing

@menu
* Rd format::
* Sectioning::
* Marking text::
* Lists and tables::
* Cross-references::
* Mathematics::
* Figures::
* Insertions::
* Indices::
* Platform-specific sections::
* Conditional text::
* Dynamic pages::
* User-defined macros::
* Encoding::
* Processing documentation files::
* Editing Rd files::
@end menu

@node Rd format, Sectioning, Writing R documentation files, Writing R documentation files
@section Rd format

@R{} objects are documented in files written in ``@R{} documentation''
(Rd) format, a simple markup language much of which closely resembles
(La)@TeX{}, which can be processed into a variety of formats,
including @LaTeX{}, @HTML{} and plain text.  The translation is
carried out by functions in the @pkg{tools} package called by the
script @command{Rdconv} in @file{@var{R_HOME}/bin} and by the
installation scripts for packages.

@c 1324 as of 2011-01-16
The @R{} distribution contains more than 1300 such files which can be
found in the @file{src/library/@var{pkg}/man} directories of the @R{}
source tree, where @var{pkg} stands for one of the standard packages
which are included in the @R{} distribution.

As an example, let us look at a simplified version of
@file{src/library/base/man/load.Rd} which documents the @R{} function
@code{load}.

@quotation
@cartouche
@smallexample
% File src/library/base/man/load.Rd
\name@{load@}
\alias@{load@}
\title@{Reload Saved Datasets@}
\description@{
  Reload the datasets written to a file with the function
  \code@{save@}.
@}
\usage@{
load(file, envir = parent.frame())
@}
\arguments@{
  \item@{file@}@{a connection or a character string giving the
    name of the file to load.@}
  \item@{envir@}@{the environment where the data should be
    loaded.@}
@}
\seealso@{
  \code@{\link@{save@}@}.
@}
\examples@{
## save all data
save(list = ls(), file= "all.RData")

## restore the saved values to the current environment
load("all.RData")

## restore the saved values to the workspace
load("all.RData", .GlobalEnv)
@}
\keyword@{file@}
@end smallexample
@end cartouche
@end quotation

An @file{Rd} file consists of three parts.  The header gives basic
information about the name of the file, the topics documented, a title,
a short textual description and @R{} usage information for the objects
documented.  The body gives further information (for example, on the
function's arguments and return value, as in the above example).
Finally, there is an optional footer with keyword information.  The
header is mandatory.

Information is given within a series of @emph{sections} with standard
names (and user-defined sections are also allowed).  Unless otherwise
specified@footnote{e.g.@: @code{\alias}, @code{\keyword} and
@code{\note} sections.} these should occur only once in an @file{Rd}
file (in any order), and the processing software will retain only the
first occurrence of a standard section in the file, with a warning.

See @uref{https://developer.r-project.org/Rds.html, ``Guidelines for Rd
files''} for guidelines for writing documentation in @file{Rd} format
which should be useful for package writers.
@findex prompt
The @R{}
generic function @code{prompt} is used to construct a bare-bones @file{Rd}
file ready for manual editing.  Methods are defined for documenting
functions (which fill in the proper function and argument names) and
data frames.  There are also functions @code{promptData},
@code{promptPackage}, @code{promptClass}, and @code{promptMethods} for
other types of @file{Rd} file.

The general syntax of @file{Rd} files is summarized below.  For a detailed
technical discussion of current @file{Rd} syntax, see
@uref{https://developer.r-project.org/parseRd.pdf, ``Parsing Rd files''}.

@file{Rd} files consist of four types of text input.  The most common
is @LaTeX{}-like, with the backslash used as a prefix on markup
(e.g.@: @code{\alias}), and braces used to indicate arguments
(e.g.@: @code{@{load@}}).  The least common type of text is `verbatim'
text, where no markup other than the comment marker (@code{%}) is
processed.  There is also a rare variant of `verbatim' text
(used in @code{\eqn}, @code{\deqn}, @code{\figure},
and @code{\newcommand}) where comment markers need not be escaped.
The final type is @R{}-like, intended for @R{} code, but allowing some
embedded macros.  Quoted strings within @R{}-like text are handled
specially: regular character escapes such as @code{\n} may be entered
as-is.  Only markup starting with @code{\l} (e.g.@: @code{\link}) or
@code{\v} (e.g.@: @code{\var}) will be recognized within quoted strings.
The rarely used vertical tab @code{\v} must be entered as @code{\\v}.

Each macro defines the input type for its argument.  For example, the
file initially uses @LaTeX{}-like syntax, and this is also used in the
@code{\description} section, but the @code{\usage} section uses
@R{}-like syntax, and the @code{\alias} macro uses `verbatim' syntax.
Comments run from a percent symbol @code{%} to the end of the line in
all types of text except the rare `verbatim' variant
(as on the first line of the @code{load} example).

Because backslashes, braces and percent symbols have special meaning, to
enter them into text sometimes requires escapes using a backslash.  In
general balanced braces do not need to be escaped, but percent symbols
always do, except in the `verbatim' variant.
For the complete list of macros and rules for escapes, see
@uref{https://developer.r-project.org/parseRd.pdf, ``Parsing Rd files''}.

@menu
* Documenting functions::
* Documenting data sets::
* Documenting S4 classes and methods::
* Documenting packages::
@end menu

@node Documenting functions, Documenting data sets, Rd format, Rd format
@subsection Documenting functions

The basic markup commands used for documenting @R{} objects (in
particular, functions) are given in this subsection.

@table @code
@item \name@{@var{name}@}
@findex \name
@var{name} typically@footnote{There can be exceptions: for example
@file{Rd} files are not allowed to start with a dot, and have to be
uniquely named on a case-insensitive file system.} is the basename of
the @file{Rd} file containing the documentation.  It is the ``name'' of
the @file{Rd} object represented by the file and has to be unique in a
package.  To avoid problems with indexing the package manual, it may not
@c Problems seen in 2.13.x but not 2.14.0
contain @samp{!}  @samp{|} nor @samp{@@}, and to avoid possible problems
with the @HTML{} help system it should not contain @samp{/} nor a space.
(@LaTeX{} special characters are allowed, but may not be collated
correctly in the index.)  There can only be one @code{\name} entry in a
file, and it must not contain any markup.  Entries in the package manual
will be in alphabetic@footnote{in the current locale, and with special
treatment for @LaTeX{} special characters and with any
@samp{@var{pkgname}-package} topic moved to the top of the list.} order
of the @code{\name} entries.

@item \alias@{@var{topic}@}
@findex \alias
The @code{\alias} sections specify all ``topics'' the file documents.
This information is collected into index data bases for lookup by the
on-line (plain text and @HTML{}) help systems.  The @var{topic} can
contain spaces, but (for historical reasons) leading and trailing spaces
will be stripped.  Percent and left brace need to be escaped by
a backslash.

There may be several @code{\alias} entries.  Quite often it is
convenient to document several @R{} objects in one file.  For example,
file @file{Normal.Rd} documents the density, distribution function,
quantile function and generation of random variates for the normal
distribution, and hence starts with

@example
@group
\name@{Normal@}
\alias@{Normal@}
\alias@{dnorm@}
\alias@{pnorm@}
\alias@{qnorm@}
\alias@{rnorm@}
@end group
@end example

@noindent
Also, it is often convenient to have several different ways to refer to
an @R{} object, and an @code{\alias} does not need to be the name of an
object.

Note that the @code{\name} is not necessarily a topic documented, and if
so desired it needs to have an explicit @code{\alias} entry (as in this
example).

@item \title@{@var{Title}@}
@findex \title
Title information for the @file{Rd} file.  This should be capitalized
and not end in a period; try to limit its length to at most 65
characters for widest compatibility.

Markup is supported in the text, but use of characters other than
English text and punctuation (e.g., @samp{<}) may limit portability.

There must be one (and only one) @code{\title} section in a help file.

@item \description@{@dots{}@}
@findex \description
A short description of what the function(s) do(es) (one paragraph, a few
lines only).  (If a description is too long and cannot easily be
shortened, the file probably tries to document too much at once.)
This is mandatory except for package-overview files.

@item \usage@{@var{fun}(@var{arg1}, @var{arg2}, @dots{})@}
@findex \usage
One or more lines showing the synopsis of the function(s) and variables
documented in the file.  These are set in typewriter font.  This is an
@R{}-like command.

The usage information specified should match the function definition
@emph{exactly} (such that automatic checking for consistency between
code and documentation is possible).

To indicate that a function can be used in several different ways,
depending on the named arguments specified, use section @code{\details}.
E.g., @file{abline.Rd} contains

@example
@group
\details@{
  Typical usages are
\preformatted@{abline(a, b, ...)
......
@}
@end group
@end example

@findex \method
Use @code{\method@{@var{generic}@}@{@var{class}@}} to indicate the name
of an S3 method for the generic function @var{generic} for objects
inheriting from class @code{"@var{class}"}.  In the printed versions,
this will come out as @var{generic} (reflecting the understanding that
methods should not be invoked directly but @emph{via} method dispatch), but
@code{codoc()} and other QC tools always have access to the full name.

For example, @file{print.ts.Rd} contains

@example
@group
\usage@{
\method@{print@}@{ts@}(x, calendar, \dots)
@}
@end group
@end example

@noindent
which will print as

@example
@group
Usage:

     ## S3 method for class 'ts':
     print(x, calendar, ...)
@end group
@end example

Usage for replacement functions should be given in the style of
@code{dim(x) <- value} rather than explicitly indicating the name of the
replacement function (@w{@code{"dim<-"}} in the above).  Similarly, one
can use @code{\method@{@var{generic}@}@{@var{class}@}(@var{arglist}) <-
value} to indicate the usage of an S3 replacement method for the generic
replacement function @code{"@var{generic}<-"} for objects inheriting
from class @code{"@var{class}"}.

Usage for S3 methods for extracting or replacing parts of an object, S3
methods for members of the Ops group, and S3 methods for user-defined
(binary) infix operators (@samp{%@var{xxx}%}) follows the above rules,
using the appropriate function names.  E.g., @file{Extract.factor.Rd}
contains

@example
@group
\usage@{
\method@{[@}@{factor@}(x, \dots, drop = FALSE)
\method@{[[@}@{factor@}(x, \dots)
\method@{[@}@{factor@}(x, \dots) <- value
@}
@end group
@end example

@noindent
which will print as

@example
@group
Usage:

     ## S3 method for class 'factor':
     x[..., drop = FALSE]
     ## S3 method for class 'factor':
     x[[...]]
     ## S3 replacement method for class 'factor':
     x[...] <- value
@end group
@end example

@findex \S3method
@code{\S3method} is accepted as an alternative to @code{\method}.

@item \arguments@{@dots{}@}
@findex \arguments
Description of the function's arguments, using an entry of the form

@example
\item@{@var{arg_i}@}@{@var{Description of arg_i}.@}
@end example

@noindent for each element of the argument list.  (Note that there is
no whitespace between the three parts of the entry.) There may be
optional text outside the @code{\item} entries, for example to give
general information about groups of parameters.


@item \details@{@dots{}@}
@findex \details
A detailed if possible precise description of the functionality
provided, extending the basic information in the @code{\description}
slot.

@item \value@{@dots{}@}
@findex \value
Description of the function's return value.

If a list with multiple values is returned, you can use entries of the
form

@example
\item@{@var{comp_i}@}@{@var{Description of comp_i}.@}
@end example

@noindent
for each component of the list returned.  Optional text may
precede@footnote{Text between or after list items is discouraged.} this
list (see for example the help for @code{rle}).  Note that @code{\value}
is implicitly a @code{\describe} environment, so that environment should
not be used for listing components, just individual @code{\item@{@}@{@}}
entries.

@item \references@{@dots{}@}
@findex \references
A section with references to the literature.  Use @code{\url@{@}} or
@code{\href@{@}@{@}} for web pointers, and @code{\doi@{@}} for DOIs
(this needs @R{} >= 3.3, see @ref{User-defined macros} for more info).

@item \note@{...@}
@findex \note
Use this for a special note you want to have pointed out.  Multiple
@code{\note} sections are allowed, but might be confusing to the end users.

For example, @file{pie.Rd} contains

@example
@group
\note@{
  Pie charts are a very bad way of displaying information.
  The eye is good at judging linear measures and bad at
  judging relative areas.
  ......
@}
@end group
@end example

@item \author@{@dots{}@}
@findex \author
Information about the author(s) of the @file{Rd} file.  Use
@code{\email@{@}} without extra delimiters (such as @samp{( )} or
@samp{< >}) to specify email addresses, or @code{\url@{@}} or
@code{\href@{@}@{@}} for web pointers.

@item \seealso@{@dots{}@}
@findex \seealso
Pointers to related @R{} objects, using @code{\code@{\link@{...@}@}} to
refer to them (@code{\code} is the correct markup for @R{} object names,
and @code{\link} produces hyperlinks in output formats which support
this.  @xref{Marking text}, and @ref{Cross-references}).

@findex \examples
@item \examples@{@dots{}@}
Examples of how to use the function.  Code in this section is set
in typewriter font without reformatting and is run by
@code{example()} unless marked otherwise (see below).

Examples are not only useful for documentation purposes, but also
provide test code used for diagnostic checking of @R{} code.  By
default, text inside @code{\examples@{@}} will be displayed in the
output of the help page and run by @code{example()} and by @code{R CMD
check}.  You can use @code{\dontrun@{@}}
@findex \dontrun
for text that should only be shown, but not run, and
@code{\dontshow@{@}}
@findex \dontshow
for extra commands for testing that should not be shown to users, but
will be run by @code{example()}.  (Previously this was called
@code{\testonly}, and that is still accepted.)

Text inside @code{\dontrun@{@}} is `verbatim', but the other parts
of the @code{\examples} section are @R{}-like text.

For example,

@example
@group
x <- runif(10)       # @r{Shown and run.}
\dontrun@{plot(x)@}    # @r{Only shown.}
\dontshow@{log(x)@}    # @r{Only run.}
@end group
@end example

Thus, example code not included in @code{\dontrun} must be executable!
In addition, it should not use any system-specific features or require
special facilities (such as Internet access or write permission to
specific directories).  Text included in @code{\dontrun} is indicated by
comments in the processed help files: it need not be valid @R{} code but
the escapes must still be used for @code{%}, @code{\} and unpaired
braces as in other `verbatim' text.

Example code must be capable of being run by @code{example}, which uses
@code{source}.  This means that it should not access @file{stdin},
e.g.@: to @code{scan()} data from the example file.

Data needed for making the examples executable can be obtained by random
number generation (for example, @code{x <- rnorm(100)}), or by using
standard data sets listed by @code{data()} (see @code{?data} for more
info).

Finally, there is @code{\donttest}, used (at the beginning of a separate
line) to mark code that should be run by @code{example()} but not by
@code{R CMD check} (by default: the option @option{--run-donttest} can
be used).  This should be needed only occasionally but can be used for
code which might fail in circumstances that are hard to test for, for
example in some locales.  (Use e.g.@: @code{capabilities()} or
@code{nzchar(Sys.which("someprogram"))} to test for features needed in
the examples wherever possible, and you can also use @code{try()} or
@code{tryCatch()}.  Use @code{interactive()} to condition examples which
need someone to interact with.)  Note that code included in
@code{\donttest} must be correct @R{} code, and any packages used should
be declared in the @file{DESCRIPTION} file.  It is good practice to
include a comment in the @code{\donttest} section explaining why it is
needed.

Output from code between comments
@example
## IGNORE_RDIFF_BEGIN
## IGNORE_RDIFF_END
@end example
@noindent
is ignored when comparing check output to reference output (a
@file{-Ex.Rout.save} file).  This markup can also be used for scripts
under @file{tests}.

@findex \keyword
@item \keyword@{@var{key}@}
There can be zero or more @code{\keyword} sections per file.
Each @code{\keyword} section should specify a single keyword, preferably
one of the standard keywords as listed in file @file{KEYWORDS} in the
@R{} documentation directory (default @file{@var{R_HOME}/doc}).  Use
e.g.@: @code{RShowDoc("KEYWORDS")} to inspect the standard keywords from
within @R{}.  There can be more than one @code{\keyword} entry if the @R{}
object being documented falls into more than one category, or none.

Do strongly consider using @code{\concept} (@pxref{Indices}) instead of
@code{\keyword} if you are about to use more than very few non-standard
keywords.

The special keyword @samp{internal} marks a page of internal objects
that are not part of the package's API.  If the help page for object
@code{foo} has keyword @samp{internal}, then @code{help(foo)} gives this
help page, but @code{foo} is excluded from several object indices,
including the alphabetical list of objects in the @HTML{} help system.

@code{help.search()} can search by keyword, including user-defined
values: however the `Search Engine & Keywords' @HTML{} page accessed
@emph{via} @code{help.start()} provides single-click access only to a
pre-defined list of keywords.
@end table


@node Documenting data sets, Documenting S4 classes and methods, Documenting functions, Rd format
@subsection Documenting data sets

The structure of @file{Rd} files which document @R{} data sets is slightly
different.  Sections such as @code{\arguments} and @code{\value} are not
needed but the format and source of the data should be explained.

As an example, let us look at @file{src/library/datasets/man/rivers.Rd}
which documents the standard @R{} data set @code{rivers}.

@quotation
@cartouche
@smallexample
\name@{rivers@}
\docType@{data@}
\alias@{rivers@}
\title@{Lengths of Major North American Rivers@}
\description@{
  This data set gives the lengths (in miles) of 141 \dQuote@{major@}
  rivers in North America, as compiled by the US Geological
  Survey.
@}
\usage@{rivers@}
\format@{A vector containing 141 observations.@}
\source@{World Almanac and Book of Facts, 1975, page 406.@}
\references@{
  McNeil, D. R. (1977) \emph@{Interactive Data Analysis@}.
  New York: Wiley.
@}
\keyword@{datasets@}
@end smallexample
@end cartouche
@end quotation

This uses the following additional markup commands.

@table @code
@item \docType@{@dots{}@}
Indicates the ``type'' of the documentation object.  Always @samp{data}
for data sets, and @samp{package} for @file{@var{pkg}-package.Rd}
overview files.  Documentation for S4 methods and classes uses
@samp{methods} (from @code{promptMethods()}) and @samp{class} (from
@code{promptClass()}).

@item \format@{@dots{}@}
@findex \format
A description of the format of the data set (as a vector, matrix, data
frame, time series, @dots{}).  For matrices and data frames this should
give a description of each column, preferably as a list or table.
@xref{Lists and tables}, for more information.

@item \source@{@dots{}@}
@findex \source
Details of the original source (a reference or @acronym{URL},
@pxref{Specifying URLs}).  In addition, section @code{\references} could
give secondary sources and usages.
@end table

Note also that when documenting data set @var{bar},

@itemize @bullet
@item
The @code{\usage} entry is always @code{@var{bar}} or (for packages
which do not use lazy-loading of data) @code{data(@var{bar})}.  (In
particular, only document a @emph{single} data object per @file{Rd} file.)
@item
The @code{\keyword} entry should always be @samp{datasets}.
@end itemize

If @code{@var{bar}} is a data frame, documenting it as a data set can
be initiated @emph{via} @code{prompt(@var{bar})}.  Otherwise, the @code{promptData}
function may be used.

@node Documenting S4 classes and methods, Documenting packages, Documenting data sets, Rd format
@subsection Documenting S4 classes and methods

There are special ways to use the @samp{?}  operator, namely
@samp{class?@var{topic}} and @samp{methods?@var{topic}}, to access
documentation for S4 classes and methods, respectively.  This mechanism
depends on conventions for the topic names used in @code{\alias}
entries.  The topic names for S4 classes and methods respectively are of
the form

@example
@var{class}-class
@var{generic},@var{signature_list}-method
@end example

@noindent
where @var{signature_list} contains the names of the classes in the
signature of the method (without quotes) separated by @samp{,} (without
whitespace), with @samp{ANY} used for arguments without an explicit
specification.  E.g., @samp{genericFunction-class} is the topic name for
documentation for the S4 class @code{"genericFunction"}, and
@samp{coerce,ANY,NULL-method} is the topic name for documentation for
the S4 method for @code{coerce} for signature @code{c("ANY", "NULL")}.

Skeletons of documentation for S4 classes and methods can be generated
by using the functions @code{promptClass()} and @code{promptMethods()}
from package @pkg{methods}.  If it is necessary or desired to provide an
explicit function declaration (in a @code{\usage} section) for an S4
method (e.g., if it has ``surprising arguments'' to be mentioned
explicitly), one can use the special markup

@example
\S4method@{@var{generic}@}@{@var{signature_list}@}(@var{argument_list})
@end example

@noindent
(e.g., @samp{\S4method@{coerce@}@{ANY,NULL@}(from, to)}).

To make full use of the potential of the on-line documentation system,
all user-visible S4 classes and methods in a package should at least
have a suitable @code{\alias} entry in one of the package's @file{Rd} files.
If a package has methods for a function defined originally somewhere
else, and does not change the underlying default method for the
function, the package is responsible for documenting the methods it
creates, but not for the function itself or the default method.

An S4 replacement method is documented in the same way as an S3 one: see
the description of  @code{\method} in @ref{Documenting functions}.


See @code{help("Documentation", package = "methods")} for more
information on using and creating on-line documentation for S4 classes and
methods.

@node Documenting packages,  , Documenting S4 classes and methods, Rd format
@subsection Documenting packages

Packages may have an overview help page with an @code{\alias}
@code{@var{pkgname}-package}, e.g.@: @samp{utils-package} for the
@pkg{utils} package, when @code{package?@var{pkgname}} will open that
help page.  If a topic named @code{@var{pkgname}} does not exist in
another @file{Rd} file, it is helpful to use this as an additional
@code{\alias}.

Skeletons of documentation for a package can be generated using the
function @code{promptPackage()}.  If the @code{final = LIBS} argument
is used, then the @file{Rd} file will be generated in final form, containing
the information that would be produced up to
@code{library(help = @var{pkgname})}.  Otherwise (the default) comments
will be inserted giving suggestions for content.

Apart from the mandatory @code{\name} and @code{\title} and the
@code{@var{pkgname}-package} alias, the only requirement for the package
overview page is that it include a @code{\docType@{package@}} statement.
All other content is optional.  We suggest that it should be a short
overview, to give a reader unfamiliar with the package enough
information to get started.  More extensive documentation is better
placed into a package vignette (@pxref{Writing package vignettes}) and
referenced from this page, or into individual man pages for the
functions, datasets, or classes.

@node Sectioning, Marking text, Rd format, Writing R documentation files
@section Sectioning

To begin a new paragraph or leave a blank line in an example, just
insert an empty line (as in (La)@TeX{}).  To break a line, use
@code{\cr}.
@findex \cr

In addition to the predefined sections (such as @code{\description@{@}},
@code{\value@{@}}, etc.), you can ``define'' arbitrary ones by
@code{\section@{@var{section_title}@}@{@dots{}@}}.
@findex \section
For example

@example
\section@{Warning@}@{
  You must not call this function unless @dots{}
@}
@end example

@noindent
For consistency with the pre-assigned sections, the section name (the
first argument to @code{\section}) should be capitalized (but not all
upper case).  Whitespace between the first and second braced expressions
is not allowed.  Markup (e.g.@: @code{\code}) within the section title
may cause problems with the latex conversion (depending on the version
of macro packages such as @samp{hyperref}) and so should be avoided.

The @code{\subsection} macro takes arguments in the same format as
@code{\section}, but is used within a section, so it may be used to
nest subsections within sections or other subsections.  There is no
predefined limit on the nesting level, but formatting is not designed
for more than 3 levels (i.e.@: subsections within subsections within
sections).

Note that additional named sections are always inserted at a fixed
position in the output (before @code{\note}, @code{\seealso} and the
examples), no matter where they appear in the input (but in the same
order amongst themselves as in the input).


@node Marking text, Lists and tables, Sectioning, Writing R documentation files
@section Marking text
@cindex Marking text in documentation

The following logical markup commands are available for emphasizing or
quoting text.

@table @code
@item \emph@{@var{text}@}
@findex \emph
@itemx \strong@{@var{text}@}
@findex \strong
Emphasize @var{text} using @emph{italic} and @strong{bold} font if
possible; @code{\strong} is regarded as stronger (more emphatic).

@item \bold@{@var{text}@}
@findex \bold
Set @var{text} in @b{bold} font where possible.

@item \sQuote@{@var{text}@}
@findex \sQuote
@itemx \dQuote@{@var{text}@}
@findex \dQuote
Portably single or double quote @var{text} (without hard-wiring the
characters used for quotation marks).
@end table

Each of the above commands takes @LaTeX{}-like input, so other macros
may be used within @var{text}.

The following logical markup commands are available for indicating
specific kinds of text.  Except as noted, these take `verbatim' text
input, and so other macros may not be used within them.  Some characters
will need to be escaped (@pxref{Insertions}).

@table @code
@item \code@{@var{text}@}
@findex \code
Indicate text that is a literal example of a piece of an @R{} program,
e.g., a fragment of @R{} code or the name of an @R{} object.  Text is
entered in @R{}-like syntax, and displayed using @code{typewriter} font
where possible.  Macros @code{\var} and @code{\link} are interpreted within
@var{text}.

@item \preformatted@{@var{text}@}
@findex \preformatted
Indicate text that is a literal example of a piece of a program.  Text
is displayed using @code{typewriter} font where possible.  Formatting,
e.g.@: line breaks, is preserved.  (Note that this includes a line break
after the initial @{, so typically text should start on the same line as
the command.)

Due to limitations in @LaTeX{} as of this writing, this macro may not be
nested within other markup macros other than @code{\dQuote} and
@code{\sQuote}, as errors or bad formatting may result.

@item \kbd@{@var{keyboard-characters}@}
@findex \kbd
Indicate keyboard input, using @kbd{slanted typewriter} font if
possible, so users can distinguish the characters they are supposed to
type from computer output.  Text is entered `verbatim'.

@item \samp@{@var{text}@}
@findex \samp
Indicate text that is a literal example of a sequence of characters,
entered `verbatim'.  No wrapping or reformatting will occur.  Displayed
using @code{typewriter} font where possible.


@item \verb@{@var{text}@}
@findex \verb
Indicate text that is a literal example of a sequence of characters,
with no interpretation of e.g.@: @code{\var}, but which will be included
within word-wrapped text.  Displayed using @code{typewriter} font if
possible.

@item \pkg@{@var{package_name}@}
@findex \pkg
Indicate the name of an @R{} package.  @LaTeX{}-like.

@item \file@{@var{file_name}@}
@findex \file
Indicate the name of a file.  Text is @LaTeX{}-like, so backslash needs
to be escaped.  Displayed using a distinct font where possible.

@item \email@{@var{email_address}@}
@findex \email
Indicate an electronic mail address.  @LaTeX{}-like, will be rendered as
a hyperlink in @HTML{} and PDF conversion.  Displayed using
@code{typewriter} font where possible.

@item \url@{@var{uniform_resource_locator}@}
@findex \url
Indicate a uniform resource locator (@acronym{URL}) for the World Wide
Web.  The argument is handled as `verbatim' text (with percent and
braces escaped by backslash), and rendered as a hyperlink in @HTML{} and
PDF conversion.  Linefeeds are removed, and leading and trailing
whitespace@footnote{as defined by the @R{} function @code{trimws}.} is
removed. @xref{Specifying URLs}.

Displayed using @code{typewriter} font where possible.

@item \href@{@var{uniform_resource_locator}@}@{@var{text}@}
@findex \href
Indicate a hyperlink to the World Wide Web. The first argument is
handled as `verbatim' text (with percent and braces escaped by
backslash) and is used as the @acronym{URL} in the hyperlink, with the
second argument of @LaTeX{}-like text displayed to the user.  Linefeeds
are removed from the first argument, and leading and trailing whitespace
is removed.

Note that RFC3986-encoded URLs (e.g.@: using @samp{%28VS.85%29} in
place of @samp{(VS.85)}) may not work correctly in versions of @R{}
before 3.1.3 and are best avoided---use @code{URLdecode()} to decode
them.

@item \var@{@var{metasyntactic_variable}@}
@findex \var
Indicate a metasyntactic variable.  In some cases this will be rendered
distinctly, e.g.@: in italic, but not in all@footnote{Currently it is
rendered differently only in @HTML{} conversions, and @LaTeX{} conversion
outside @samp{\usage} and @samp{\examples} environments.}. @LaTeX{}-like.
@item \env@{@var{environment_variable}@}
@findex \env
Indicate an environment variable. `Verbatim'.
Displayed using @code{typewriter} font where possible
@item \option@{@var{option}@}
@findex \option
Indicate a command-line option.  `Verbatim'.
Displayed using @code{typewriter} font where possible.
@item \command@{@var{command_name}@}
@findex \command
Indicate the name of a command. @LaTeX{}-like, so @code{\var} is
interpreted.  Displayed using @code{typewriter} font where possible.
@item \dfn@{@var{term}@}
@findex \dfn
Indicate the introductory or defining use of a term. @LaTeX{}-like.
@item \cite@{@var{reference}@}
@findex \cite
Indicate a reference without a direct cross-reference @emph{via} @code{\link}
(@pxref{Cross-references}), such as the name of a book. @LaTeX{}-like.
@item \acronym@{@var{acronym}@}
@findex \acronym
Indicate an acronym (an abbreviation written in all capital letters),
such as @acronym{GNU}. @LaTeX{}-like.
@end table


@node Lists and tables, Cross-references, Marking text, Writing R documentation files
@section Lists and tables
@cindex Lists and tables in documentation

@findex \itemize
@findex \enumerate
The @code{\itemize} and @code{\enumerate} commands take a single
argument, within which there may be one or more @code{\item} commands.
The text following each @code{\item} is formatted as one or more
paragraphs, suitably indented and with the first paragraph marked with a
bullet point (@code{\itemize}) or a number (@code{\enumerate}).

Note that unlike argument lists, @code{\item} in these formats is
followed by a space and the text (not enclosed in braces).  For example

@example
  \enumerate@{
    \item A database consists of one or more records, each with one or
    more named fields.
    \item Regular lines start with a non-whitespace character.
    \item Records are separated by one or more empty lines.
  @}
@end example

@code{\itemize} and @code{\enumerate} commands may be nested.

@findex \describe
The @code{\describe} command is similar to @code{\itemize} but allows
initial labels to be specified.  Each @code{\item} takes two arguments,
the label and the body of the item, in exactly the same way as an
argument or value @code{\item}.  @code{\describe} commands are mapped to
@code{<DL>} lists in @HTML{} and @code{\description} lists in @LaTeX{}.

@c \itemize did to LaTeX in 2021-11
Using these without any @code{\item}s may cause problems with some
conversions and makes little sense.

@findex \tabular
The @code{\tabular} command takes two arguments.  The first gives for
each of the columns the required alignment (@samp{l} for
left-justification, @samp{r} for right-justification or @samp{c} for
centring.)  The second argument consists of an arbitrary number of
lines separated by @code{\cr}, and with fields separated by @code{\tab}.
For example:

@example
@group
  \tabular@{rlll@}@{
    [,1] \tab Ozone   \tab numeric \tab Ozone (ppb)\cr
    [,2] \tab Solar.R \tab numeric \tab Solar R (lang)\cr
    [,3] \tab Wind    \tab numeric \tab Wind (mph)\cr
    [,4] \tab Temp    \tab numeric \tab Temperature (degrees F)\cr
    [,5] \tab Month   \tab numeric \tab Month (1--12)\cr
    [,6] \tab Day     \tab numeric \tab Day of month (1--31)
  @}
@end group
@end example

@noindent
There must be the same number of fields on each line as there are
alignments in the first argument, and they must be non-empty (but can
contain only spaces).  (There is no whitespace between @code{\tabular}
and the first argument, nor between the two arguments.)

@node Cross-references, Mathematics, Lists and tables, Writing R documentation files
@section Cross-references
@cindex Cross-references in documentation

@findex \link
The markup @code{\link@{@var{foo}@}} (usually in the combination
@code{\code@{\link@{@var{foo}@}@}}) produces a hyperlink to the help for
@var{foo}.  Here @var{foo} is a @emph{topic}, that is the argument of
@code{\alias} markup in another @file{Rd} file (possibly in another package).
Hyperlinks are supported in some of the formats to which @file{Rd} files are
converted, for example @HTML{} and PDF, but ignored in others, e.g.@:
the text format.

One main usage of @code{\link} is in the @code{\seealso} section of the
help page, @pxref{Rd format}.

Note that whereas leading and trailing spaces are stripped when
extracting a topic from a @code{\alias}, they are not stripped when
looking up the topic of a @code{\link}.

@cindex \linkS4class
You can specify a link to a different topic than its name by
@code{\link[=@var{dest}]@{@var{name}@}} which links to topic @var{dest}
with name @var{name}.  This can be used to refer to the documentation
for S3/4 classes, for example @code{\code@{"\link[=abc-class]@{abc@}"@}}
would be a way to refer to the documentation of an S4 class @code{"abc"}
defined in your package, and
@code{\code@{"\link[=terms.object]@{terms@}"@}} to the S3 @code{"terms"}
class (in package @pkg{stats}).  To make these easy to read in the
source file, @code{\code@{"\linkS4class@{abc@}"@}} expands to the form
given above.

There are two other forms with an optional argument, specified as
@code{\link[@var{pkg}]@{@var{foo}@}} and
@code{\link[@var{pkg:bar}]@{@var{foo}@}}, to link to topics
@code{@var{foo}} and @code{@var{bar}} respectively in the package
@pkg{@var{pkg}}. They are currently only used in @HTML{} help (and
ignored for hyperlinks in @LaTeX{} conversions of help pages). One
should be careful about topics containing special characters (such as
arithmetic operators) as they may result in unresolvable links, and
preferably use a safer alias in the same help page.

Historically (before @R{} version 4.1.0), links of the form
@code{\link[@var{pkg}]@{@var{foo}@}} and
@code{\link[@var{pkg:bar}]@{@var{foo}@}} used to be interpreted as links
to @emph{files} @file{@var{foo}.html} and @file{@var{bar}.html} in
package @pkg{@var{pkg}}, respectively. For this reason, the @HTML{} help
system looks for file @file{@var{foo}.html} in package @pkg{@var{pkg}}
if it does not find topic @code{@var{foo}}, and then searches for the
topic in other installed packages. To test that links work both with
both old and new systems, the pre-4.1.0 behaviour can be restored by
setting the environment variable @env{_R_HELP_LINKS_TO_TOPICS_=false}.

Packages referred to by these `other forms' should be declared in the
@file{DESCRIPTION} file, in the @samp{Depends}, @samp{Imports},
@samp{Suggests} or @samp{Enhances} fields.

@node Mathematics, Figures, Cross-references, Writing R documentation files
@section Mathematics
@cindex Mathematics in documentation
@findex \eqn
@findex \deqn

Mathematical formulae should be set beautifully for printed
documentation yet we still want something useful for text and @HTML{}
online help.  To this end, the two commands
@code{\eqn@{@var{latex}@}@{@var{ascii}@}} and
@code{\deqn@{@var{latex}@}@{@var{ascii}@}} are used.  Whereas @code{\eqn}
is used for ``inline'' formulae (corresponding to @TeX{}'s
@code{$@dots{}$}), @code{\deqn} gives ``displayed equations'' (as in
@LaTeX{}'s @code{displaymath} environment, or @TeX{}'s
@code{$$@dots{}$$}).  Both arguments are treated as `verbatim' text.

Both commands can also be used as @code{\eqn@{@var{latexascii}@}} (only
@emph{one} argument) which then is used for both @var{latex} and
@var{ascii}.  No whitespace is allowed between command and the first
argument, nor between the first and second arguments.

The following example is from @file{Poisson.Rd}:

@example
@group
  \deqn@{p(x) = \frac@{\lambda^x e^@{-\lambda@}@}@{x!@}@}@{%
        p(x) = \lambda^x exp(-\lambda)/x!@}
  for \eqn@{x = 0, 1, 2, \ldots@}.
@end group
@end example

@iftex
For the @LaTeX{} manual, this becomes
@c: Name-and-shame for Brian Diggs:
@c: this is TeXinfo markup, not the result of the conversions.
@quotation
@cartouche
@tex
$$ p(x) = \lambda^x\ {e^{-\lambda} \over x!} $$
for $x = 0, 1, 2, \ldots$.
@end tex
@end cartouche
@end quotation
@end iftex

For text on-line help we get

@quotation
@cartouche
@example
    p(x) = lambda^x exp(-lambda)/x!

for x = 0, 1, 2, ....
@end example
@end cartouche
@end quotation

Greek letters (both cases) will be rendered in @HTML{} if preceded by a
backslash, @code{\dots} and @code{\ldots} will be rendered as ellipses
and @code{\sqrt}, @code{\ge} and @code{\le} as mathematical symbols.

Note that only basic @LaTeX{} can be used, there being no provision to
specify @LaTeX{} style files such as the AMS extensions.

@node Figures, Insertions, Mathematics, Writing R documentation files
@section Figures
@cindex Figures in documentation
@findex \figure

To include figures in help pages, use the @code{\figure} markup.  There
are three forms.

The two commonly used simple forms are @code{\figure@{@var{filename}@}}
and @code{\figure@{@var{filename}@}@{@var{alternate text}@}}.  This will
include a copy of the figure in either @HTML{} or @LaTeX{} output.  In text
output, the alternate text will be displayed instead.  (When the second
argument is omitted, the filename will be used.)  Both the filename and
the alternate text will be parsed verbatim, and should not include
special characters that are significant in @HTML{} or @LaTeX{}.

The expert form is @code{\figure@{@var{filename}@}@{options:
@var{string}@}}.  (The word @samp{options:} must be typed exactly as
shown and followed by at least one space.)  In this form, the
@var{string} is copied into the @HTML{} @code{img} tag as attributes
following the @code{src} attribute, or into the second argument of the
@code{\Figure} macro in @LaTeX{}, which by default is used as options to
an @code{\includegraphics} call.  As it is unlikely that any single
string would suffice for both display modes, the expert form would
normally be wrapped in conditionals.  It is up to the author to make
sure that legal @HTML{}/@LaTeX{} is used.  For example, to include a
logo in both @HTML{} (using the simple form) and @LaTeX{} (using the
expert form), the following could be used:

@example
\if@{html@}@{\figure@{Rlogo.svg@}@{options: width=100 alt="R logo"@}@}
\if@{latex@}@{\figure@{Rlogo.pdf@}@{options: width=0.5in@}@}
@end example

The files containing the figures should be stored in the directory
@file{man/figures}.  Files with extensions @file{.jpg}, @file{.jpeg},
@file{.pdf}, @file{.png} and @file{.svg} from that directory will be
copied to the @file{help/figures} directory at install time. (Figures in
PDF format will not display in most @HTML{} browsers, but might be the
best choice in reference manuals.)  Specify the filename relative to
@file{man/figures} in the @code{\figure} directive.

@node Insertions, Indices, Figures, Writing R documentation files
@section Insertions

@findex \R
Use @code{\R} for the @R{} system itself.  Use @code{\dots}
@findex \dots
for the dots in function argument lists @samp{@dots{}}, and
@code{\ldots}
@findex \ldots
for ellipsis dots in ordinary text.@footnote{There is only a fine
distinction between @code{\dots} and @code{\ldots}.  It is technically
incorrect to use @code{\ldots} in code blocks and @code{tools::checkRd}
will warn about this---on the other hand the current converters treat
them the same way in code blocks, and elsewhere apart from the small
distinction between the two in @LaTeX{}.}  These can be followed by
@code{@{@}}, and should be unless followed by whitespace.

After an unescaped @samp{%}, you can put your own comments regarding the
help text.  The rest of the line (but not the newline at the end) will
be completely disregarded.  Therefore, you can also use it to make part
of the ``help'' invisible.

You can produce a backslash (@samp{\}) by escaping it by another
backslash.  (Note that @code{\cr} is used for generating line breaks.)

The ``comment'' character @samp{%} and unpaired braces@footnote{See the
examples section in the file @file{Paren.Rd} for an example.}
@emph{almost always} need to be escaped by @samp{\}, and @samp{\\} can
be used for backslash and needs to be when there are two or more adjacent
backslashes.  In @R{}-like code quoted strings are handled slightly
differently; see @uref{https://developer.r-project.org/parseRd.pdf,
``Parsing Rd files''} for details -- in particular braces should not be
escaped in quoted strings.

All of @samp{% @{ @} \} should be escaped in @LaTeX{}-like text.

@findex \enc
Text which might need to be represented differently in different
encodings should be marked by @code{\enc}, e.g.@:
@code{\enc@{J@"oreskog@}@{Joreskog@}} (with no whitespace between the
braces) where the first argument will be used where encodings are
allowed and the second should be @acronym{ASCII} (and is used for e.g.@:
the text conversion in locales that cannot represent the encoded form).
(This is intended to be used for individual words, not whole sentences
or paragraphs.)

@node Indices, Platform-specific sections, Insertions, Writing R documentation files
@section Indices
@cindex Indices

The @code{\alias} command (@pxref{Documenting functions}) is used to
specify the ``topics'' documented, which should include @emph{all} @R{}
objects in a package such as functions and variables, data sets, and S4
classes and methods (@pxref{Documenting S4 classes and methods}).  The
on-line help system searches the index data base consisting of all
alias topics.

@findex \concept
In addition, it is possible to provide ``concept index entries'' using
@code{\concept}, which can be used for @code{help.search()} lookups.
E.g., file @file{cor.test.Rd} in the standard package @pkg{stats}
contains

@example
\concept@{Kendall correlation coefficient@}
\concept@{Pearson correlation coefficient@}
\concept@{Spearman correlation coefficient@}
@end example

@noindent
so that e.g.@: @kbd{??Spearman} will succeed in finding the
help page for the test for association between paired samples using
Spearman's @eqn{\rho, rho}.

(Note that @code{help.search()} only uses ``sections'' of documentation
objects with no additional markup.)

Each @code{\concept} entry should give a @emph{single} index term (word
or phrase), and not use any Rd markup.

If you want to cross reference such items from other help files @emph{via}
@code{\link}, you need to use @code{\alias} and not @code{\concept}.


@node  Platform-specific sections, Conditional text, Indices, Writing R documentation files
@section Platform-specific documentation
@cindex Platform-specific documentation

Sometimes the documentation needs to differ by platform.  Currently two
OS-specific options are available, @samp{unix} and @samp{windows}, and
lines in the help source file can be enclosed in

@example
@group
#ifdef @var{OS}
   ...
#endif
@end group
@end example

@noindent
or

@example
@group
#ifndef @var{OS}
   ...
#endif
@end group
@end example

@noindent
for OS-specific inclusion or exclusion.  Such blocks should not be
nested, and should be entirely within a block (that, is between the
opening and closing brace of a section or item), or at top-level contain
one or more complete sections.

If the differences between platforms are extensive or the @R{} objects
documented are only relevant to one platform, platform-specific @file{Rd} files
can be put in a @file{unix} or @file{windows} subdirectory.

@node  Conditional text, Dynamic pages, Platform-specific sections, Writing R documentation files
@section Conditional text
@cindex conditionals
@findex \if
@findex \ifelse
@findex \out

Occasionally the best content for one output format is different from
the best content for another.  For this situation, the
@code{\if@{@var{format}@}@{@var{text}@}} or
@code{\ifelse@{@var{format}@}@{@var{text}@}@{@var{alternate}@}} markup
is used.  Here @var{format} is a comma separated list of formats in
which the @var{text} should be rendered.  The @var{alternate} will be
rendered if the format does not match.  Both @var{text} and
@var{alternate} may be any sequence of text and markup.

Currently the following formats are recognized:  @code{example},
@code{html}, @code{latex} and @code{text}.  These select output for
the corresponding targets. (Note that @code{example} refers to
extracted example code rather than the displayed example in some other
format.)  Also accepted are @code{TRUE} (matching all formats) and
@code{FALSE} (matching no formats).  These could be the output
of the @code{\Sexpr} macro (@pxref{Dynamic pages}).

The @code{\out@{@var{literal}@}} macro would usually be used within
the @var{text} part of @code{\if@{@var{format}@}@{@var{text}@}}.  It
causes the renderer to output the literal text exactly, with no
attempt to escape special characters.  For example, use
the following to output the markup necessary to display the Greek letter in
@LaTeX{} or @HTML{}, and the text string @code{alpha} in other formats:
@example
\ifelse@{latex@}@{\out@{$\alpha$@}@}@{\ifelse@{html@}@{\out@{&alpha;@}@}@{alpha@}@}
@end example

@node Dynamic pages, User-defined macros, Conditional text, Writing R documentation files
@section Dynamic pages
@cindex dynamic pages
@findex \Sexpr
@findex \RdOpts

Two macros supporting dynamically generated man pages are @code{\Sexpr}
and @code{\RdOpts}.  These are modelled after Sweave, and are intended
to contain executable @R{} expressions in the @file{Rd} file.

The main argument to @code{\Sexpr} must be valid @R{} code that can be
executed. It may also take options in square brackets before the main
argument. Depending on the options, the code may be executed at
package build time, package install time, or man page rendering time.

The options follow the same format as in Sweave, but different options
are supported.  Currently the allowed options and their defaults are:

@itemize @bullet
@item @code{eval=TRUE}
Whether the @R{} code should be evaluated.

@item @code{echo=FALSE}
Whether the @R{} code should be echoed.  If @code{TRUE}, a display will
be given in a preformatted block.  For example,
@code{\Sexpr[echo=TRUE]@{ x <- 1 @}} will be displayed as
@example
> x <- 1
@end example

@item @code{keep.source=TRUE}
Whether to keep the author's formatting when displaying the
code, or throw it away and use a deparsed version.

@item @code{results=text}
How should the results be displayed?  The possibilities
are:

@itemize @minus
@item @code{results=text}
Apply @code{as.character()} to the result of the code, and insert it
as a text element.

@item @code{results=verbatim}
Print the results of the code just as if it was executed at the console,
and include the printed results verbatim.  (Invisible results will not print.)

@item @code{results=rd}
The result is assumed to be a character vector containing markup to be
passed to @code{parse_Rd()}, with the result inserted in place.  This
could be used to insert computed aliases, for instance.
@code{parse_Rd()} is called first with @code{fragment = FALSE} to allow
a single Rd section macro to be inserted.  If that fails, it is called
again with @code{fragment = TRUE}, the older behavior.

@item @code{results=hide}
Insert no output.
@end itemize

@item @code{strip.white=TRUE}
Remove leading and trailing white space from each line of
output if @code{strip.white=TRUE}.  With
@code{strip.white=all}, also remove blank lines.

@item @code{stage=install}
Control when this macro is run.  Possible values are
@itemize @minus
@item @code{stage=build}
The macro is run when building a source tarball.

@item @code{stage=install}
The macro is run when installing from source.

@item @code{stage=render}
The macro is run when displaying the help page.
@end itemize

Conditionals such as @code{#ifdef}
(@pxref{Platform-specific sections}) are applied after the
@code{build} macros but before the @code{install} macros.  In some
situations (e.g.@: installing directly from a source directory without a
tarball, or building a binary package) the above description is not
literally accurate, but authors can rely on the sequence being
@code{build}, @code{#ifdef}, @code{install}, @code{render}, with all
stages executed.

Code is only run once in each stage, so a @code{\Sexpr[results=rd]}
macro can output an @code{\Sexpr} macro designed for a later stage,
but not for the current one or any earlier stage.

@item @code{width, height, fig}
These options are currently allowed but ignored.
@end itemize

The @code{\RdOpts} macro is used to set new defaults for options to apply
to following uses of @code{\Sexpr}.

For more details, see the online document
@uref{https://developer.r-project.org/parseRd.pdf, ``Parsing Rd files''}.

@node User-defined macros, Encoding, Dynamic pages, Writing R documentation files
@section User-defined macros
@cindex user-defined macros
@findex \newcommand
@findex \renewcommand

The @code{\newcommand} and @code{\renewcommand} macros allow new macros
to be defined within an Rd file.  These are similar but not identical to
the same-named @LaTeX{} macros.

They each take two arguments which are parsed verbatim.  The first is
the name of the new macro including the initial backslash, and the
second is the macro definition.  As in @LaTeX{}, @code{\newcommand}
requires that the new macro not have been previously defined, whereas
@code{\renewcommand} allows existing macros (including all built-in
ones) to be replaced.  (This test is disabled by default, but may be
enabled by setting the environment variable
@env{_R_WARN_DUPLICATE_RD_MACROS_} to a true value.)

Also as in @LaTeX{}, the new macro may be defined to take arguments,
and numeric placeholders such as @code{#1} are used in the macro
definition. However, unlike @LaTeX{}, the number of arguments is
determined automatically from the highest placeholder number seen in
the macro definition.  For example, a macro definition containing
@code{#1} and @code{#3} (but no other placeholders) will define a
three argument macro (whose second argument will be ignored). As in
@LaTeX{}, at most 9 arguments may be defined. If the @code{#}
character is followed by a non-digit it will have no special
significance.  All arguments to user-defined macros will be parsed as
verbatim text, and simple text-substitution will be used to replace
the place-holders, after which the replacement text will be parsed.

A number of macros are defined in the file
@file{share/Rd/macros/system.Rd} of the @R{} source or home directory,
and these will normally be available in all @file{.Rd} files.  For
example, that file contains the definition
@example
\newcommand@{\PR@}@{\Sexpr[results=rd]@{tools:::Rd_expr_PR(#1)@}@}
@end example
@noindent
which defines @code{\PR} to be a single argument macro; then code
(typically used in the @file{NEWS.Rd} file) like
@example
\PR@{1234@}
@end example
@noindent
will expand to
@example
\Sexpr[results=rd]@{tools:::Rd_expr_PR(1234)@}
@end example
@noindent
when parsed.  

Some macros that might be of general use are:
@ftable @code
@item \CRANpkg@{@var{pkg}@}
A package on CRAN

@item \sspace
A single space (used after a period that does not end a sentence).

@item \doi@{@var{numbers}@}
A digital object identifier (DOI).
@end ftable
See the @file{system.Rd} file in @file{share/Rd/macros} for more details
and macro definitions, including macros @code{\packageTitle},
@code{\packageDescription}, @code{\packageAuthor}, @code{\packageMaintainer},
@code{\packageDESCRIPTION} and @code{\packageIndices}.
@findex @code{\packageTitle}
@findex @code{\packageDescription}
@findex @code{\packageAuthor}
@findex @code{\packageMaintainer}
@findex @code{\packageDESCRIPTION}
@findex @code{\packageIndices}


Packages may also define their own common macros; these would be stored
in an @file{.Rd} file in @file{man/macros} in the package source and
will be installed into @file{help/macros} when the package is installed.
A package may also use the macros from a different package by listing
the other package in the @samp{RdMacros} field in the @file{DESCRIPTION}
file.


@node Encoding, Processing documentation files, User-defined macros, Writing R documentation files
@section Encoding
@cindex encoding

Rd files are text files and so it is impossible to deduce the encoding
they are written in unless @acronym{ASCII}: files with 8-bit characters
could be UTF-8, Latin-1, Latin-9, KOI8-R, EUC-JP, @emph{etc}.  So an
@code{\encoding@{@}} section must be used to specify the encoding if it
is not @acronym{ASCII}.  (The @code{\encoding@{@}} section must be on a
line by itself, and in particular one containing no non-@acronym{ASCII}
characters.  The encoding declared in the @file{DESCRIPTION} file will
be used if none is declared in the file.)  The @file{Rd} files are
converted to UTF-8 before parsing and so the preferred encoding for the
files themselves is now UTF-8.

Wherever possible, avoid non-@acronym{ASCII} chars in @file{Rd} files, and
even symbols such as @samp{<}, @samp{>}, @samp{$}, @samp{^}, @samp{&},
@samp{|}, @samp{@@}, @samp{~}, and @samp{*} outside `verbatim'
environments (since they may disappear in fonts designed to render
text).  (Function @code{showNonASCIIfile} in package @pkg{tools} can help
in finding non-@acronym{ASCII} bytes in the files.)

For convenience, encoding names @samp{latin1} and @samp{latin2} are
always recognized: these and @samp{UTF-8} are likely to work fairly
widely.  However, this does not mean that all characters in UTF-8 will
be recognized, and the coverage of non-Latin characters@footnote{@R{}
2.9.0 added support for UTF-8 Cyrillic characters in @LaTeX{}, but on
some OSes this will need Cyrillic support added to @LaTeX{}, so
environment variable @env{_R_CYRILLIC_TEX_} may need to be set to a
non-empty value to enable this.} is fairly low.  Using @LaTeX{}
@code{inputenx} (see @code{?Rd2pdf} in @R{}) will give greater coverage
of UTF-8.

The @code{\enc} command (@pxref{Insertions}) can be used to provide
transliterations which will be used in conversions that do not support
the declared encoding.

The @LaTeX{} conversion converts the file to UTF-8 from the declared
encoding, and includes a

@example
\inputencoding@{utf8@}
@end example

@noindent
command, and this needs to be matched by a suitable invocation of the
@command{\usepackage@{inputenc@}} command.  The @R{} utility @command{R
CMD Rd2pdf} looks at the converted code and includes the encodings used:
it might for example use

@example
\usepackage[utf8]@{inputenc@}
@end example

@noindent
(Use of @code{utf8} as an encoding requires @LaTeX{} dated 2003/12/01 or
later.  Also, the use of Cyrillic characters in @samp{UTF-8} appears to
also need @samp{\usepackage[T2A]@{fontenc@}}, and @command{R CMD Rd2pdf}
includes this conditionally on the file @file{t2aenc.def} being present
and environment variable @env{_R_CYRILLIC_TEX_} being set.)

Note that this mechanism works best with Latin letters: the coverage of
UTF-8 in @LaTeX{} is quite low.


@node Processing documentation files, Editing Rd files, Encoding, Writing R documentation files
@section Processing documentation files
@cindex Processing Rd format

There are several commands to process Rd files from the system command
line.

@findex R CMD Rdconv
Using @code{R CMD Rdconv} one can convert @R{} documentation format to
other formats, or extract the executable examples for run-time testing.
The currently supported conversions are to plain text, @HTML{} and
@LaTeX{} as well as extraction of the examples.

@findex R CMD Rd2pdf
@code{R CMD Rd2pdf} generates PDF output from documentation in @file{Rd}
files, which can be specified either explicitly or by the path to a
directory with the sources of a package.  In the latter case, a
reference manual for all documented objects in the package is created,
including the information in the @file{DESCRIPTION} files.

@findex R CMD Sweave
@findex R CMD Stangle
@code{R CMD Sweave} and @code{R CMD Stangle} process vignette-like
documentation files (e.g.@: Sweave vignettes with extension
@samp{.Snw} or @samp{.Rnw}, or other non-Sweave vignettes).
@code{R CMD Stangle} is used to extract the @R{} code fragments.

The exact usage and a detailed list of available options for all of
these commands can be obtained by running @code{R CMD @var{command}
--help}, e.g., @kbd{R CMD Rdconv --help}.  All available commands can be
listed using @kbd{R --help} (or @kbd{Rcmd --help} under Windows).

All of these work under Windows.  You may need to have installed the
the tools to build packages from source as described in the ``R
Installation and Administration'' manual, although typically all that is
needed is a @LaTeX{} installation.

@node Editing Rd files,  , Processing documentation files, Writing R documentation files
@section Editing Rd files
@cindex Editing Rd files

It can be very helpful to prepare @file{.Rd} files using a editor which
knows about their syntax and will highlight commands, indent to show the
structure and detect mis-matched braces, and so on.

The system most commonly used for this is some version of
@command{Emacs} (including @command{XEmacs}) with the @acronym{ESS}
package (@uref{https://ESS.R-project.org/}: it is often is installed with
@command{Emacs} but may need to be loaded, or even installed,
separately).

Another is the Eclipse IDE with the Stat-ET plugin
(@uref{https://projects.eclipse.org/projects/science.statet}), and (on
Windows only) Tinn-R
(@uref{https://sourceforge.net/projects/tinn-r/}).

People have also used @LaTeX{} mode in a editor, as @file{.Rd} files are
rather similar to @LaTeX{} files.

Some @R{} front-ends provide editing support for  @file{.Rd} files, for
example RStudio (@uref{https://www.rstudio.com/}).

@node Tidying and profiling R code, Debugging, Writing R documentation files, Top
@chapter Tidying and profiling R code

@menu
* Tidying R code::
* Profiling R code for speed::
* Profiling R code for memory use::
* Profiling compiled code::
@end menu

@R{} code which is worth preserving in a package and perhaps making
available for others to use is worth documenting, tidying up and perhaps
optimizing. The last two of these activities are the subject of this
chapter.

@node Tidying R code, Profiling R code for speed, Tidying and profiling R code, Tidying and profiling R code
@section Tidying R code
@cindex Tidying R code

@R{} treats function code loaded from packages and code entered by users
differently.  By default code entered by users has the source code stored
internally, and when the function is listed, the original source is
reproduced.  Loading code from a package (by default) discards the
source code, and the function listing is re-created from the parse tree
of the function.

Normally keeping the source code is a good idea, and in particular it
avoids comments being removed from the source.  However, we can make
use of the ability to re-create a function listing from its parse tree
to produce a tidy version of the function, for example with consistent
indentation and spaces around operators.  If the original source
does not follow the standard format this tidied version can be much
easier to read.

We can subvert the keeping of source in two ways.

@enumerate
@item
The option @code{keep.source} can be set to @code{FALSE} before the code
is loaded into @R{}.
@item
The stored source code can be removed by calling the @code{removeSource()}
function, for example by

@example
myfun <- removeSource(myfun)
@end example

@end enumerate

@noindent
In each case if we then list the function we will get the standard
layout.

Suppose we have a file of functions @file{myfuns.R} that we want to
tidy up.  Create a file @file{tidy.R} containing

@example
@group
source("myfuns.R", keep.source = FALSE)
dump(ls(all.names = TRUE), file = "new.myfuns.R")
@end group
@end example

@noindent
and run @R{} with this as the source file, for example by @kbd{R
--vanilla < tidy.R} or by pasting into an @R{} session.  Then the file
@file{new.myfuns.R} will contain the functions in alphabetical order in
the standard layout.  Warning:  comments in your functions will be lost.

The standard format provides a good starting point for further tidying.
Although the deparsing cannot do so, we recommend the consistent use of
the preferred assignment operator @samp{<-} (rather than @samp{=}) for
assignment.  Many package authors use a version of Emacs (on a
Unix-alike or Windows) to edit @R{} code, using the ESS[S] mode of the
@acronym{ESS} Emacs package.  See @ref{R coding standards, , R coding
standards, R-ints, R Internals} for style options within the ESS[S] mode
recommended for the source code of @R{} itself.


@node Profiling R code for speed, Profiling R code for memory use, Tidying R code, Tidying and profiling R code
@section Profiling R code for speed
@cindex Profiling
@findex Rprof

It is possible to profile @R{} code on Windows and most@footnote{@R{}
has to be built to enable this, but the option
@option{--enable-R-profiling} is the default.} Unix-alike versions of
@R{}.

The command @command{Rprof} is used to control profiling, and its help
page can be consulted for full details.  Profiling works by recording
at fixed intervals@footnote{For Unix-alikes these are intervals of CPU
time, and for Windows of elapsed time.} (by default every 20 msecs)
which line in which @R{} function is being used, and recording the
results in a file (default @file{Rprof.out} in the working directory).
Then the function @code{summaryRprof} or the command-line utility
@code{R CMD Rprof @var{Rprof.out}} can be used to summarize the
activity.

As an example, consider the following code (from Venables & Ripley,
2002, pp. 225--6).

@smallexample
@group
library(MASS); library(boot)
storm.fm <- nls(Time ~ b*Viscosity/(Wt - c), stormer,
                start = c(b=30.401, c=2.2183))
st <- cbind(stormer, fit=fitted(storm.fm))
storm.bf <- function(rs, i) @{
    st$Time <-  st$fit + rs[i]
    tmp <- nls(Time ~ (b * Viscosity)/(Wt - c), st,
               start = coef(storm.fm))
    tmp$m$getAllPars()
@}
rs <- scale(resid(storm.fm), scale = FALSE) # remove the mean
Rprof("boot.out")
storm.boot <- boot(rs, storm.bf, R = 4999) # slow enough to profile
Rprof(NULL)
@end group
@end smallexample

@noindent
Having run this we can summarize the results by

@smallexample
@group
R CMD Rprof boot.out

Each sample represents 0.02 seconds.
Total run time: 22.52 seconds.

Total seconds: time spent in function and callees.
Self seconds: time spent in function alone.
@end group

@group
   %       total       %        self
 total    seconds     self    seconds    name
 100.0     25.22       0.2      0.04     "boot"
  99.8     25.18       0.6      0.16     "statistic"
  96.3     24.30       4.0      1.02     "nls"
  33.9      8.56       2.2      0.56     "<Anonymous>"
  32.4      8.18       1.4      0.36     "eval"
  31.8      8.02       1.4      0.34     ".Call"
  28.6      7.22       0.0      0.00     "eval.parent"
  28.5      7.18       0.3      0.08     "model.frame"
  28.1      7.10       3.5      0.88     "model.frame.default"
  17.4      4.38       0.7      0.18     "sapply"
  15.0      3.78       3.2      0.80     "nlsModel"
  12.5      3.16       1.8      0.46     "lapply"
  12.3      3.10       2.7      0.68     "assign"
 ...
@end group

@group
   %        self        %      total
  self    seconds     total   seconds    name
   5.7      1.44       7.5      1.88     "inherits"
   4.0      1.02      96.3     24.30     "nls"
   3.6      0.92       3.6      0.92     "$"
   3.5      0.88      28.1      7.10     "model.frame.default"
   3.2      0.80      15.0      3.78     "nlsModel"
   2.8      0.70       9.8      2.46     "qr.coef"
   2.7      0.68      12.3      3.10     "assign"
   2.5      0.64       2.5      0.64     ".Fortran"
   2.5      0.62       7.1      1.80     "qr.default"
   2.2      0.56      33.9      8.56     "<Anonymous>"
   2.1      0.54       5.9      1.48     "unlist"
   2.1      0.52       7.9      2.00     "FUN"
  ...
@end group
@end smallexample

@noindent
This often produces
surprising results and can be used to identify bottlenecks or pieces of
@R{} code that could benefit from being replaced by compiled code.

Two warnings: profiling does impose a small performance penalty, and the
output files can be very large if long runs are profiled at the default
sampling interval.

Profiling short runs can sometimes give misleading results.  @R{} from
time to time performs @emph{garbage collection} to reclaim unused
memory, and this takes an appreciable amount of time which profiling
will charge to whichever function happens to provoke it.  It may be
useful to compare profiling code immediately after a call to @code{gc()}
with a profiling run without a preceding call to @code{gc}.

More detailed analysis of the output can be achieved by the tools in the
@acronym{CRAN} packages @CRANpkg{proftools} and @CRANpkg{profr}: in
particular these allow call graphs to be studied.

@node Profiling R code for memory use, Profiling compiled code, Profiling R code for speed, Tidying and profiling R code
@section Profiling R code for memory use
@cindex Profiling
@cindex Memory use

Measuring memory use in @R{} code is useful either when the code takes
more memory than is conveniently available or when memory allocation and
copying of objects is responsible for slow code. There are three ways to
profile memory use over time in @R{} code. The second and third require
@R{} to have been compiled with @option{--enable-memory-profiling},
which is not the default, but is currently used for the macOS and
Windows binary distributions. All can be misleading, for different
reasons.

In understanding the memory profiles it is useful to know a little more
about @R{}'s memory allocation. Looking at the results of @code{gc()}
shows a division of memory into @code{Vcells} used to store the contents
of vectors and @code{Ncells} used to store everything else, including
all the administrative overhead for vectors such as type and length
information.  In fact the vector contents are divided into two
pools. Memory for small vectors (by default 128 bytes or less) is
obtained in large chunks and then parcelled out by @R{}; memory for
larger vectors is obtained directly from the operating system.

Some memory allocation is obvious in interpreted code, for example,

@smallexample
y <- x + 1
@end smallexample

@noindent
allocates memory for a new vector @code{y}. Other memory allocation is
less obvious and occurs because @code{R} is forced to make good on its
promise of `call-by-value' argument passing.  When an argument is
passed to a function it is not immediately copied. Copying occurs (if
necessary) only when the argument is modified.  This can lead to
surprising memory use. For example, in the `survey' package we have

@smallexample
print.svycoxph <- function (x, ...)
@{
    print(x$survey.design, varnames = FALSE, design.summaries = FALSE, ...)
    x$call <- x$printcall
    NextMethod()
@}
@end smallexample

@noindent
It may not be obvious that the assignment to @code{x$call} will cause
the entire object @code{x} to be copied.  This copying to preserve the
call-by-value illusion is usually done by the internal C function
@code{duplicate}.

The main reason that memory-use profiling is difficult is garbage
collection. Memory is allocated at well-defined times in an @R{}
program, but is freed whenever the garbage collector happens to run.

@menu
* Memory statistics from Rprof::
* Tracking memory allocations::
* Tracing copies of an object::
@end menu

@node Memory statistics from Rprof, Tracking memory allocations, Profiling R code for memory use, Profiling R code for memory use
@subsection Memory statistics from @code{Rprof}
@findex Rprof
@findex summaryRprof

The sampling profiler @code{Rprof} described in the previous section can
be given the option @code{memory.profiling=TRUE}. It then writes out the
total @R{} memory allocation in small vectors, large vectors, and cons
cells or nodes at each sampling interval. It also writes out the number
of calls to the internal function @code{duplicate}, which is called to
copy @R{} objects. @code{summaryRprof} provides summaries of this
information.  The main reason that this can be misleading is that the
memory use is attributed to the function running at the end of the
sampling interval. A second reason is that garbage collection can make
the amount of memory in use decrease, so a function appears to use
little memory.  Running under @code{gctorture} helps with both problems:
it slows down the code to effectively increase the sampling frequency
and it makes each garbage collection release a smaller amount of memory.

@node Tracking memory allocations, Tracing copies of an object, Memory statistics from Rprof, Profiling R code for memory use
@subsection Tracking memory allocations
@findex Rprofmem

The second method of memory profiling uses a memory-allocation
profiler, @code{Rprofmem()}, which writes out a stack trace to an
output file every time a large vector is allocated (with a
user-specified threshold for `large') or a new page of memory is
allocated for the @R{} heap. Summary functions for this output are still
being designed.

Running the example from the previous section with

@smallexample
> Rprofmem("boot.memprof",threshold=1000)
> storm.boot <- boot(rs, storm.bf, R = 4999)
> Rprofmem(NULL)
@end smallexample

@noindent
shows that apart from some initial and final work in @code{boot} there
are no vector allocations over 1000 bytes.

@node  Tracing copies of an object,  , Tracking memory allocations, Profiling R code for memory use
@subsection Tracing copies of an object
@findex tracemem
@findex untracemem

The third method of memory profiling involves tracing copies made of a
specific (presumably large) @R{} object. Calling @code{tracemem} on an
object marks it so that a message is printed to standard output when
the object is copied @emph{via} @code{duplicate} or coercion to another type,
or when a new object of the same size is created in arithmetic
operations. The main reason that this can be misleading is that
copying of subsets or components of an object is not tracked. It may
be helpful to use @code{tracemem} on these components.


In the example above we can run @code{tracemem} on the data frame
@code{st}

@smallexample
> tracemem(st)
[1] "<0x9abd5e0>"
> storm.boot <- boot(rs, storm.bf, R = 4)
memtrace[0x9abd5e0->0x92a6d08]: statistic boot
memtrace[0x92a6d08->0x92a6d80]: $<-.data.frame $<- statistic boot
memtrace[0x92a6d80->0x92a6df8]: $<-.data.frame $<- statistic boot
memtrace[0x9abd5e0->0x9271318]: statistic boot
memtrace[0x9271318->0x9271390]: $<-.data.frame $<- statistic boot
memtrace[0x9271390->0x9271408]: $<-.data.frame $<- statistic boot
memtrace[0x9abd5e0->0x914f558]: statistic boot
memtrace[0x914f558->0x914f5f8]: $<-.data.frame $<- statistic boot
memtrace[0x914f5f8->0x914f670]: $<-.data.frame $<- statistic boot
memtrace[0x9abd5e0->0x972cbf0]: statistic boot
memtrace[0x972cbf0->0x972cc68]: $<-.data.frame $<- statistic boot
memtrace[0x972cc68->0x972cd08]: $<-.data.frame $<- statistic boot
memtrace[0x9abd5e0->0x98ead98]: statistic boot
memtrace[0x98ead98->0x98eae10]: $<-.data.frame $<- statistic boot
memtrace[0x98eae10->0x98eae88]: $<-.data.frame $<- statistic boot
@end smallexample

@noindent
The object is duplicated fifteen times, three times for each of the
@code{R+1} calls to @code{storm.bf}.  This is surprising, since none of the duplications happen inside @code{nls}. Stepping through @code{storm.bf} in the debugger shows that all three happen in the line

@smallexample
st$Time <- st$fit + rs[i]
@end smallexample

Data frames are slower than matrices and this is an example of why.
Using @code{tracemem(st$Viscosity)} does not reveal any additional
copying.

@node Profiling compiled code,  , Profiling R code for memory use, Tidying and profiling R code
@section Profiling compiled code
@cindex Profiling

Profiling compiled code is highly system-specific, but this section
contains some hints gleaned from various @R{} users.  Some methods need
to be different for a compiled executable and for dynamic/shared
libraries/objects as used by @R{} packages.  We know of no good way to
profile DLLs on Windows.

This chapter is based on reports from users and the information may not
be current.

@menu
* Linux::
* macOS::
@end menu

@node Linux, macOS, Profiling compiled code, Profiling compiled code
@subsection Linux

Options include using @command{sprof} for a shared object, and
@command{oprofile} (see @uref{https://oprofile.sourceforge.io/news/}) and
@command{perf} (see
@uref{https://perf.wiki.kernel.org/index.php/Tutorial}) for any
executable or shared object.

@subsubsection sprof

You can select shared objects to be profiled with @command{sprof} by
setting the environment variable @env{LD_PROFILE}.  For example

@example
% setenv LD_PROFILE /path/to/R_HOME/library/stats/libs/stats.so
R
... run the boot example
% sprof /path/to/R_HOME/library/stats/libs/stats.so \
  /var/tmp/path/to/R_HOME/library/stats/libs/stats.so.profile

Flat profile:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total
 time   seconds   seconds    calls  us/call  us/call  name
 76.19      0.32     0.32        0     0.00           numeric_deriv
 16.67      0.39     0.07        0     0.00           nls_iter
  7.14      0.42     0.03        0     0.00           getListElement

rm /var/tmp/path/to/R_HOME/library/stats/libs/stats.so.profile
... to clean up ...
@end example

It is possible that root access is needed to create the directories used
for the profile data.

@subsubsection oprofile and operf

The @command{oprofile} project has two modes of operation.  In what is
now called `legacy' mode, it is uses a daemon to collect information on
a process (see below).  Since version 0.9.8 (August 2012), the preferred
mode is to use @command{operf}, so we discuss that first.  The modes
differ in how the profiling data is collected: it is analysed by tools
such as @command{opreport} and @command{oppannote} in both.

Here is an example on @code{x86_64} Linux using @R{} 3.0.2.  File
@file{pvec.R} contains the part of the examples from @code{pvec} in
package @pkg{parallel}:
@example
library(parallel)
N <- 1e6
dates <- sprintf('%04d-%02d-%02d', as.integer(2000+rnorm(N)),
                 as.integer(runif(N, 1, 12)), as.integer(runif(N, 1, 28)))
system.time(a <- as.POSIXct(dates, format = "%Y-%m-%d"))
@end example
@noindent
with timings from the final step
@example
   user  system elapsed
  0.371   0.237   0.612
@end example

@R{}-level profiling by @code{Rprof} shows
@example
                     self.time self.pct total.time total.pct
"strptime"                1.70    41.06       1.70     41.06
"as.POSIXct.POSIXlt"      1.40    33.82       1.42     34.30
"sprintf"                 0.74    17.87       0.98     23.67
...
@end example
@noindent
so the conversion from character to @code{POSIXlt} takes most of the
time.

This can be run under @command{operf} and analysed by
@example
operf R -f pvec.R
opreport
opreport -l /path/to/R_HOME/bin/exec/R
opannotate --source /path/to/R_HOME/bin/exec/R
## And for the system time
opreport -l /lib64/libc.so.6
@end example
@noindent
The first report shows where (which library etc) the time was spent:
@example
CPU_CLK_UNHALT...|
  samples|      %|
------------------
   166761 99.9161 Rdev
        CPU_CLK_UNHALT...|
          samples|      %|
        ------------------
            70586 42.3276 no-vmlinux
            56963 34.1585 libc-2.16.so
            36922 22.1407 R
             1584  0.9499 stats.so
              624  0.3742 libm-2.16.so
...
@end example

@noindent
The rest of the output is voluminous, and only extracts are shown below.

Most of the time within @R{} is spent in
@example
samples  %        image name symbol name
10397    28.5123  R           R_gc_internal
5683     15.5848  R           do_sprintf
3036      8.3258  R           do_asPOSIXct
2427      6.6557  R           do_strptime
2421      6.6392  R           Rf_mkCharLenCE
1480      4.0587  R           w_strptime_internal
1202      3.2963  R           Rf_qnorm5
1165      3.1948  R           unif_rand
675       1.8511  R           mktime0
617       1.6920  R           makelt
617       1.6920  R           validate_tm
584       1.6015  R           day_of_the_week
...
@end example
@noindent
@command{opannotate} shows that 31% of the time in @R{} is spent in
@file{memory.c}, 21% in @file{datetime.c} and 7% in @file{Rstrptime.h}.
The analysis for @file{libc} showed that calls to @code{wcsftime}
dominated, so those calls were cached for @R{} 3.0.3: the time spent in
@code{no-vmlinux} (the kernel) was reduced dramatically.

On platforms which support it, call graphs can be produced by
@command{opcontrol --callgraph} if collected @emph{via} @command{operf
--callgraph}.

The profiling data is by default stored in sub-directory
@file{oprofile_data} of the current directory, which can be removed at
the end of the session.

Another example, from @CRANpkg{sm} version 2.2-5.4.  The example for
@code{sm.variogram} took a long time:
@example
system.time(example(sm.variogram))
...
   user  system elapsed
  5.543   3.202   8.785
@end example

@noindent
including a lot of system time.  Profiling just the slow part, the
second plot, showed

@example
  samples|      %|
------------------
   381845 99.9885 R
        CPU_CLK_UNHALT...|
          samples|      %|
        ------------------
           187484 49.0995 sm.so
           169627 44.4230 no-vmlinux
            12636  3.3092 libgfortran.so.3.0.0
             6455  1.6905 R
@end example

@noindent
so the system time was almost all in the Linux kernel.  It is possible
to dig deeper if you have a matching uncompressed kernel with debug
symbols to specify @emph{via} @option{--vmlinux}: we did not.

In `legacy' mode @code{oprofile} works by running a daemon which
collects information.  The daemon must be started as root, e.g.

@example
% su
% opcontrol --no-vmlinux
% (optional, some platforms) opcontrol --callgraph=5
% opcontrol --start
% exit
@end example

Then as a user

@example
% R
... run the boot example
% opcontrol --dump
% opreport -l /path/to/R_HOME/library/stats/libs/stats.so
...
samples  %        symbol name
1623     75.5939  anonymous symbol from section .plt
349      16.2552  numeric_deriv
113       5.2632  nls_iter
62        2.8878  getListElement
% opreport -l /path/to/R_HOME/bin/exec/R
...
samples  %        symbol name
76052    11.9912  Rf_eval
54670     8.6198  Rf_findVarInFrame3
37814     5.9622  Rf_allocVector
31489     4.9649  Rf_duplicate
28221     4.4496  Rf_protect
26485     4.1759  Rf_cons
23650     3.7289  Rf_matchArgs
21088     3.3250  Rf_findFun
19995     3.1526  findVarLocInFrame
14871     2.3447  Rf_evalList
13794     2.1749  R_Newhashpjw
13522     2.1320  R_gc_internal
...
@end example

Shutting down the profiler and clearing the records needs to be done as
root.

@node macOS,  , Linux, Profiling compiled code
@subsection macOS

Developers have recommended @command{sample} (or @command{Sampler.app},
which is a GUI version), @command{Shark} (in version of @code{Xcode}
up to those for Snow Leopard), and @command{Instruments} (part of
@code{Xcode}, see
@uref{https://help.apple.com/instruments/mac/current/}).


@node Debugging, System and foreign language interfaces, Tidying and profiling R code, Top
@chapter Debugging

This chapter covers the debugging of @R{} extensions, starting with the
ways to get useful error information and moving on to how to deal with
errors that crash @R{}.

@menu
* Browsing::
* Debugging R code::
* Checking memory access::
* Debugging compiled code::
* Using Link-time Optimization::
@end menu


@node Browsing, Debugging R code, Debugging, Debugging
@section Browsing

@findex browser
Most of the @R{}-level debugging facilities are based around the
built-in browser.  This can be used directly by inserting a call to
@code{browser()} into the code of a function (for example, using
@code{fix(my_function)} ).  When code execution reaches that point in
the function, control returns to the @R{} console with a special prompt.
For example

@example
> fix(summary.data.frame) ## insert browser() call after for() loop
> summary(women)
Called from: summary.data.frame(women)
Browse[1]> ls()
 [1] "digits" "i"      "lbs"    "lw"     "maxsum" "nm"     "nr"     "nv"
 [9] "object" "sms"    "z"
Browse[1]> maxsum
[1] 7
Browse[1]>
     height         weight
 Min.   :58.0   Min.   :115.0
 1st Qu.:61.5   1st Qu.:124.5
 Median :65.0   Median :135.0
 Mean   :65.0   Mean   :136.7
 3rd Qu.:68.5   3rd Qu.:148.0
 Max.   :72.0   Max.   :164.0
> rm(summary.data.frame)
@end example

@noindent
At the browser prompt one can enter any @R{} expression, so for example
@code{ls()} lists the objects in the current frame, and entering the
name of an object will@footnote{With the exceptions of the commands
listed below: an object of such a name can be printed @emph{via} an
explicit call to @code{print}.} print it.  The following commands are
also accepted

@itemize @bullet
@item @code{n}

Enter `step-through' mode.  In this mode, hitting return executes the
next line of code (more precisely one line and any continuation lines).
Typing @code{c} will continue to the end of the current context, e.g.@:
to the end of the current loop or function.

@item @code{c}

In normal mode, this quits the browser and continues execution, and just
return works in the same way.  @code{cont} is a synonym.

@item @code{where}

This prints the call stack.  For example

@example
> summary(women)
Called from: summary.data.frame(women)
Browse[1]> where
where 1: summary.data.frame(women)
where 2: summary(women)

Browse[1]>
@end example

@item @code{Q}

Quit both the browser and the current expression, and return to the
top-level prompt.
@end itemize

Errors in code executed at the browser prompt will normally return
control to the browser prompt.  Objects can be altered by assignment,
and will keep their changed values when the browser is exited.  If
really necessary, objects can be assigned to the workspace from the
browser prompt (by using @code{<<-} if the name is not already in
scope).

@node Debugging R code, Checking memory access, Browsing, Debugging
@section Debugging R code

@findex traceback
Suppose your @R{} program gives an error message.  The first thing to
find out is what @R{} was doing at the time of the error, and the most
useful tool is @code{traceback()}.  We suggest that this is run whenever
the cause of the error is not immediately obvious.  Errors are often
reported to the @R{} mailing lists as being in some package when
@code{traceback()} would show that the error was being reported by some
other package or base @R{}.  Here is an example from the regression
suite.

@smallexample
> success <- c(13,12,11,14,14,11,13,11,12)
> failure <- c(0,0,0,0,0,0,0,2,2)
> resp <- cbind(success, failure)
> predictor <- c(0, 5^(0:7))
> glm(resp ~ 0+predictor, family = binomial(link="log"))
Error: no valid set of coefficients has been found: please supply starting values
> traceback()
3: stop("no valid set of coefficients has been found: please supply
         starting values", call. = FALSE)
2: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart,
       mustart = mustart, offset = offset, family = family, control = control,
       intercept = attr(mt, "intercept") > 0)
1: glm(resp ~ 0 + predictor, family = binomial(link ="log"))
@end smallexample

@noindent
The calls to the active frames are given in reverse order (starting with
the innermost).  So we see the error message comes from an explicit
check in @code{glm.fit}.  (@code{traceback()} shows you all the lines of
the function calls, which can be limited by setting @code{option}
@option{"deparse.max.lines"}.)

Sometimes the traceback will indicate that the error was detected inside
compiled code, for example (from @code{?nls})

@smallexample
Error in nls(y ~ a + b * x, start = list(a = 0.12345, b = 0.54321), trace = TRUE) :
        step factor 0.000488281 reduced below 'minFactor' of 0.000976563
>  traceback()
2: .Call(R_nls_iter, m, ctrl, trace)
1: nls(y ~ a + b * x, start = list(a = 0.12345, b = 0.54321), trace = TRUE)
@end smallexample

@noindent
This will be the case if the innermost call is to @code{.C},
@code{.Fortran}, @code{.Call}, @code{.External} or @code{.Internal}, but
as it is also possible for such code to evaluate @R{} expressions, this
need not be the innermost call, as in

@smallexample
> traceback()
9: gm(a, b, x)
8: .Call(R_numeric_deriv, expr, theta, rho, dir)
7: numericDeriv(form[[3]], names(ind), env)
6: getRHS()
5: assign("rhs", getRHS(), envir = thisEnv)
4: assign("resid", .swts * (lhs - assign("rhs", getRHS(), envir = thisEnv)),
       envir = thisEnv)
3: function (newPars)
   @{
       setPars(newPars)
       assign("resid", .swts * (lhs - assign("rhs", getRHS(), envir = thisEnv)),
           envir = thisEnv)
       assign("dev", sum(resid^2), envir = thisEnv)
       assign("QR", qr(.swts * attr(rhs, "gradient")), envir = thisEnv)
       return(QR$rank < min(dim(QR$qr)))
   @}(c(-0.00760232418963883, 1.00119632515036))
2: .Call(R_nls_iter, m, ctrl, trace)
1: nls(yeps ~ gm(a, b, x), start = list(a = 0.12345, b = 0.54321))
@end smallexample

Occasionally @code{traceback()} does not help, and this can be the case
if S4 method dispatch is involved.  Consider the following example

@example
> xyd <- new("xyloc", x=runif(20), y=runif(20))
Error in as.environment(pkg) : no item called "package:S4nswv"
on the search list
Error in initialize(value, ...) : S language method selection got
an error when called from internal dispatch for function 'initialize'
> traceback()
2: initialize(value, ...)
1: new("xyloc", x = runif(20), y = runif(20))
@end example

@noindent
which does not help much, as there is no call to @code{as.environment}
in @code{initialize} (and the note ``called from internal dispatch''
tells us so).  In this case we searched the @R{} sources for the quoted
call, which occurred in only one place,
@code{methods:::.asEnvironmentPackage}.  So now we knew where the
error was occurring.  (This was an unusually opaque example.)

The error message

@example
evaluation nested too deeply: infinite recursion / options(expressions=)?
@end example

@noindent
can be hard to handle with the default value (5000).  Unless you know
that there actually is deep recursion going on, it can help to set
something like

@example
options(expressions=500)
@end example

@noindent 
and re-run the example showing the error.

Sometimes there is warning that clearly is the precursor to some later
error, but it is not obvious where it is coming from.  Setting
@command{options(warn = 2)} (which turns warnings into errors) can help here.

Once we have located the error, we have some choices.  One way to proceed
is to find out more about what was happening at the time of the crash by
looking a @emph{post-mortem} dump.  To do so, set
@findex dump.frames
@command{options(error=dump.frames)} and run the code again.  Then invoke
@command{debugger()} and explore the dump.  Continuing our example:

@smallexample
> options(error = dump.frames)
> glm(resp ~ 0 + predictor, family = binomial(link ="log"))
Error: no valid set of coefficients has been found: please supply starting values
@end smallexample

@noindent
which is the same as before, but an object called @code{last.dump} has
appeared in the workspace.  (Such objects can be large, so remove it
when it is no longer needed.)  We can examine this at a later time by
calling the function @code{debugger}.
@findex debugger

@smallexample
> debugger()
Message:  Error: no valid set of coefficients has been found: please supply starting values
Available environments had calls:
1: glm(resp ~ 0 + predictor, family = binomial(link = "log"))
2: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart, mus
3: stop("no valid set of coefficients has been found: please supply starting values
Enter an environment number, or 0 to exit  Selection:
@end smallexample

@noindent
which gives the same sequence of calls as @code{traceback}, but in
outer-first order and with only the first line of the call, truncated to
the current width.  However, we can now examine in more detail what was
happening at the time of the error.  Selecting an environment opens the
browser in that frame.  So we select the function call which spawned the
error message, and explore some of the variables (and execute two
function calls).

@smallexample
Enter an environment number, or 0 to exit  Selection: 2
Browsing in the environment with call:
   glm.fit(x = X, y = Y, weights = weights, start = start, etas
Called from: debugger.look(ind)
Browse[1]> ls()
 [1] "aic"        "boundary"   "coefold"    "control"    "conv"
 [6] "dev"        "dev.resids" "devold"     "EMPTY"      "eta"
[11] "etastart"   "family"     "fit"        "good"       "intercept"
[16] "iter"       "linkinv"    "mu"         "mu.eta"     "mu.eta.val"
[21] "mustart"    "n"          "ngoodobs"   "nobs"       "nvars"
[26] "offset"     "start"      "valideta"   "validmu"    "variance"
[31] "varmu"      "w"          "weights"    "x"          "xnames"
[36] "y"          "ynames"     "z"
Browse[1]> eta
            1             2             3             4             5
 0.000000e+00 -2.235357e-06 -1.117679e-05 -5.588393e-05 -2.794197e-04
            6             7             8             9
-1.397098e-03 -6.985492e-03 -3.492746e-02 -1.746373e-01
Browse[1]> valideta(eta)
[1] TRUE
Browse[1]> mu
        1         2         3         4         5         6         7         8
1.0000000 0.9999978 0.9999888 0.9999441 0.9997206 0.9986039 0.9930389 0.9656755
        9
0.8397616
Browse[1]> validmu(mu)
[1] FALSE
Browse[1]> c
Available environments had calls:
1: glm(resp ~ 0 + predictor, family = binomial(link = "log"))
2: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart
3: stop("no valid set of coefficients has been found: please supply starting v

Enter an environment number, or 0 to exit  Selection: 0
> rm(last.dump)
@end smallexample

Because @code{last.dump} can be looked at later or even in another @R{}
session, post-mortem debugging is possible even for batch usage of @R{}.
We do need to arrange for the dump to be saved: this can be done either
using the command-line flag @option{--save} to save the workspace at the
end of the run, or @emph{via} a setting such as

@example
> options(error = quote(@{dump.frames(to.file=TRUE); q()@}))
@end example

@noindent
See the help on @code{dump.frames} for further options and a worked
example.

@findex recover
An alternative error action is to use the function @command{recover()}:

@smallexample
> options(error = recover)
> glm(resp ~ 0 + predictor, family = binomial(link = "log"))
Error: no valid set of coefficients has been found: please supply starting values

Enter a frame number, or 0 to exit

1: glm(resp ~ 0 + predictor, family = binomial(link = "log"))
2: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart

Selection:
@end smallexample

@noindent
which is very similar to @code{dump.frames}.  However, we can examine
the state of the program directly, without dumping and re-loading the
dump.  As its help page says, @code{recover} can be routinely used as
the error action in place of @code{dump.calls} and @code{dump.frames},
since it behaves like @code{dump.frames} in non-interactive use.


@findex debug
Post-mortem debugging is good for finding out exactly what went wrong,
but not necessarily why.  An alternative approach is to take a closer
look at what was happening just before the error, and a good way to do
that is to use @command{debug}.  This inserts a call to the browser
at the beginning of the function, starting in step-through mode.  So in
our example we could use

@smallexample
> debug(glm.fit)
> glm(resp ~ 0 + predictor, family = binomial(link ="log"))
debugging in: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart,
    mustart = mustart, offset = offset, family = family, control = control,
    intercept = attr(mt, "intercept") > 0)
debug: @{
## lists the whole function
Browse[1]>
debug: x <- as.matrix(x)
...
Browse[1]> start
[1] -2.235357e-06
debug: eta <- drop(x %*% start)
Browse[1]> eta
            1             2             3             4             5
 0.000000e+00 -2.235357e-06 -1.117679e-05 -5.588393e-05 -2.794197e-04
            6             7             8             9
-1.397098e-03 -6.985492e-03 -3.492746e-02 -1.746373e-01
Browse[1]>
debug: mu <- linkinv(eta <- eta + offset)
Browse[1]> mu
        1         2         3         4         5         6         7         8
1.0000000 0.9999978 0.9999888 0.9999441 0.9997206 0.9986039 0.9930389 0.9656755
        9
0.8397616
@end smallexample

@noindent
(The prompt @code{Browse[1]>} indicates that this is the first level of
browsing: it is possible to step into another function that is itself
being debugged or contains a call to @code{browser()}.)

@code{debug} can be used for hidden functions and S3 methods by
e.g.@: @code{debug(stats:::predict.Arima)}.  (It cannot be used for S4
methods, but an alternative is given on the help page for @code{debug}.)
Sometimes you want to debug a function defined inside another function,
e.g.@: the function @code{arimafn} defined inside @code{arima}.  To do so,
set @code{debug} on the outer function (here @code{arima}) and
step through it until the inner function has been defined.  Then
call @code{debug} on the inner function (and use @code{c} to get out of
step-through mode in the outer function).

@findex undebug
To remove debugging of a function, call @code{undebug} with the argument
previously given to @code{debug}; debugging otherwise lasts for the rest
of the @R{} session (or until the function is edited or otherwise
replaced).

@findex trace
@code{trace} can be used to temporarily insert debugging code into a
function, for example to insert a call to @code{browser()} just before
the point of the error.  To return to our running example

@example
## first get a numbered listing of the expressions of the function
> page(as.list(body(glm.fit)), method="print")
> trace(glm.fit, browser, at=22)
Tracing function "glm.fit" in package "stats"
[1] "glm.fit"
> glm(resp ~ 0 + predictor, family = binomial(link ="log"))
Tracing glm.fit(x = X, y = Y, weights = weights, start = start,
   etastart = etastart,  .... step 22
Called from: eval(expr, envir, enclos)
Browse[1]> n
## and single-step from here.
> untrace(glm.fit)
@end example
@noindent
For your own functions, it may be as easy to use @code{fix} to insert
temporary code, but @code{trace} can help with functions in a namespace
(as can @code{fixInNamespace}).  Alternatively, use
@code{trace(,edit=TRUE)} to insert code visually.


@node Checking memory access, Debugging compiled code, Debugging R code, Debugging
@section Checking memory access

Errors in memory allocation and reading/writing outside arrays are very
common causes of crashes (e.g.,@: segfaults) on some machines.  Often
the crash appears long after the invalid memory access: in particular
damage to the structures which @R{} itself has allocated may only become
apparent at the next garbage collection (or even at later garbage
collections after objects have been deleted).

Note that memory access errors may be seen with LAPACK, BLAS, OpenMP and
Java-using packages: some at least of these seem to be intentional, and
some are related to passing characters to Fortran.

Some of these tools can detect mismatched allocation and deallocation.
C++ programmers should note that memory allocated by @code{new []} must
be freed by @code{delete []}, other uses of @code{new} by @code{delete},
and memory allocated by @code{malloc}, @code{calloc} and @code{realloc}
by @code{free}.  Some platforms will tolerate mismatches (perhaps with
memory leaks) but others will segfault.

@menu
* Using gctorture::
* Using valgrind::
* Using Address Sanitizer::
* Using Undefined Behaviour Sanitizer::
* Other analyses with `clang'::
* Other analyses with `gcc'::
* Using `Dr. Memory'::
* Fortran array bounds checking::
@end menu

@node Using gctorture, Using valgrind, Checking memory access, Checking memory access
@subsection Using gctorture

@findex gctorture
We can help to detect memory problems in @R{} objects earlier by running
garbage collection as often as possible.  This is achieved by
@code{gctorture(TRUE)}, which as described on its help page

@quotation
Provokes garbage collection on (nearly) every memory allocation.
Intended to ferret out memory protection bugs.  Also makes @R{} run
@emph{very} slowly, unfortunately.
@end quotation

@noindent
The reference to `memory protection' is to missing C-level calls to
@code{PROTECT}/@code{UNPROTECT} (@pxref{Garbage Collection}) which if
missing allow @R{} objects to be garbage-collected when they are still
in use.  But it can also help with other memory-related errors.

Normally running under @code{gctorture(TRUE)} will just produce a crash
earlier in the @R{} program, hopefully close to the actual cause. See
the next section for how to decipher such crashes.

It is possible to run all the examples, tests and vignettes covered by
@code{R CMD check} under @code{gctorture(TRUE)} by using the option
@option{--use-gct}.

The function @code{gctorture2} provides more refined control over the GC
torture process.  Its arguments @code{step}, @code{wait} and
@code{inhibit_release} are documented on its help page.  Environment
variables can also be used at the start of the @R{} session to turn on
GC torture: @env{R_GCTORTURE} corresponds to the @code{step} argument to
@code{gctorture2}, @env{R_GCTORTURE_WAIT} to @code{wait}, and
@env{R_GCTORTURE_INHIBIT_RELEASE} to @code{inhibit_release}.

If @R{} is configured with @option{--enable-strict-barrier} then a
variety of tests for the integrity of the write barrier are enabled.  In
addition tests to help detect protect issues are enabled:

@itemize @bullet

@item
All GCs are full GCs.

@item
New nodes in small node pages are marked as @code{NEWSXP} on creation.

@item
After a GC all free nodes that are not of type @code{NEWSXP} are marked
as type @code{FREESXP} and their previous type is recorded.

@item
Most calls to accessor functions check their @code{SEXP} inputs and
@code{SEXP} outputs and signal an error if a @code{FREESXP} is found.
The address of the node and the old type are included in the error
message.

@end itemize

@code{R CMD check --use-gct} can be set to use
@code{gctorture2(@var{n})} rather than @code{gctorture(TRUE)} by setting
environment variable @env{_R_CHECK_GCT_N_} to a positive integer value
to be used as @code{@var{n}}.

Used with a debugger and with @code{gctorture} or @code{gctorture2} this
mechanism can be helpful in isolating memory protect problems.


@node Using valgrind, Using Address Sanitizer, Using gctorture, Checking memory access
@subsection Using valgrind

If you have access to Linux on a common CPU type or supported versions
of FreeBSD or Solaris@footnote{The macOS support is for long-obsolete
versions.} you can use @code{valgrind}
(@uref{https://www.valgrind.org/}, pronounced to rhyme with `tinned') to
check for possible problems.  To run some examples under @code{valgrind}
use something like

@example
R -d valgrind --vanilla < mypkg-Ex.R
R -d "valgrind --tool=memcheck --leak-check=full" --vanilla < mypkg-Ex.R
@end example

@noindent
where @file{mypkg-Ex.R} is a set of examples, e.g.@: the file created in
@file{mypkg.Rcheck} by @code{R CMD check}.  Occasionally this reports
memory reads of `uninitialised values' that are the result of compiler
optimization, so can be worth checking under an unoptimized compile: for
maximal information use a build with debugging symbols.  We know there
will be some small memory leaks from @code{readline} and @R{} itself ---
these are memory areas that are in use right up to the end of the @R{}
session.  Expect this to run around 20x slower than without
@code{valgrind}, and in some cases much slower than that.  Several
versions of @code{valgrind} were not happy with some optimized BLASes
that use @acronym{CPU}-specific instructions so you may need to build a
version of @R{} specifically to use with @code{valgrind}.

On platforms where @code{valgrind} is installed you can build a version
of @R{} with extra instrumentation to help @code{valgrind} detect errors
in the use of memory allocated from the @R{} heap.  The
@command{configure} option is
@option{--with-valgrind-instrumentation=@var{level}}, where @var{level}
is 0, 1 or 2.  Level 0 is the default and does not add anything.
Level 1 will detect some uses@footnote{Those in some numeric, logical,
integer, raw, complex vectors and in memory allocated by
@code{R_alloc}.} of uninitialised memory and has little impact on speed
(compared to level 0). Level 2 will detect many other memory-use
bugs@footnote{including using the data sections of @R{} vectors after
they are freed.}  but make @R{} much slower when running under
@code{valgrind}.  Using this in conjunction with @code{gctorture} can be
even more effective (and even slower).

An example of @code{valgrind} output is
@smallexample
==12539== Invalid read of size 4
==12539==    at 0x1CDF6CBE: csc_compTr (Mutils.c:273)
==12539==    by 0x1CE07E1E: tsc_transpose (dtCMatrix.c:25)
==12539==    by 0x80A67A7: do_dotcall (dotcode.c:858)
==12539==    by 0x80CACE2: Rf_eval (eval.c:400)
==12539==    by 0x80CB5AF: R_execClosure (eval.c:658)
==12539==    by 0x80CB98E: R_execMethod (eval.c:760)
==12539==    by 0x1B93DEFA: R_standardGeneric (methods_list_dispatch.c:624)
==12539==    by 0x810262E: do_standardGeneric (objects.c:1012)
==12539==    by 0x80CAD23: Rf_eval (eval.c:403)
==12539==    by 0x80CB2F0: Rf_applyClosure (eval.c:573)
==12539==    by 0x80CADCC: Rf_eval (eval.c:414)
==12539==    by 0x80CAA03: Rf_eval (eval.c:362)
==12539==  Address 0x1C0D2EA8 is 280 bytes inside a block of size 1996 alloc'd
==12539==    at 0x1B9008D1: malloc (vg_replace_malloc.c:149)
==12539==    by 0x80F1B34: GetNewPage (memory.c:610)
==12539==    by 0x80F7515: Rf_allocVector (memory.c:1915)
...
@end smallexample
@noindent
This example is from an instrumented version of @R{}, while tracking
down a bug in the @CRANpkg{Matrix} package in 2006.  The first line
indicates that @R{} has tried to read 4 bytes from a memory address that
it does not have access to. This is followed by a C stack trace showing
where the error occurred. Next is a description of the memory that was
accessed. It is inside a block allocated by @code{malloc}, called from
@code{GetNewPage}, that is, in the internal @R{} heap.  Since this
memory all belongs to @R{}, @code{valgrind} would not (and did not)
detect the problem in an uninstrumented build of @R{}.  In this example
the stack trace was enough to isolate and fix the bug, which was in
@code{tsc_transpose}, and in this example running under
@code{gctorture()} did not provide any additional information.
@c Was removed: see https://sourceforge.net/p/valgrind/mailman/message/34306867/
@c When the stack trace is not sufficiently informative the option
@c @option{--db-attach=yes} to @code{valgrind} may be helpful.  This starts
@c a post-mortem debugger (by default @code{gdb}) so that variables in the
@c C code can be inspected (@pxref{Inspecting R objects}).

@command{valgrind} is good at spotting the use of uninitialized values:
use option @option{--track-origins=yes} to show where these originated
from.  What it cannot detect is the misuse of arrays allocated on the
stack: this includes C automatic variables and some@footnote{small
fixed-size arrays by default in @command{gfortran}, for example.}
Fortran arrays.

It is possible to run all the examples, tests and vignettes covered by
@code{R CMD check} under @code{valgrind} by using the option
@option{--use-valgrind}.  If you do this you will need to select the
@code{valgrind} options some other way, for example by having a
@file{~/.valgrindrc} file containing

@example
--leak-check=full
--track-origins=yes
@end example

@noindent
or setting the environment variable @env{VALGRIND_OPTS}.  As from @R{}
4.2.0, @option{--use-valgrind} also uses @command{valgrind} when
re-building the vignettes.

On macOS you may need to ensure that debugging symbols are made available
(so @command{valgrind} reports line numbers in files).  This can usually
be done with the @command{valgrind} option @option{--dsymutil=yes} to
ask for the symbols to be dumped when the @file{.so} file is loaded.
This will not work where packages are installed into a system area (such
as the @file{R.framework}) and can be slow.  Installing packages with
@command{R CMD INSTALL --dsym} installs the dumped symbols.  (This can
also be done by setting environment variable @env{PKG_MAKE_DSYM} to a
non-empty value before the @command{INSTALL}.)

This section has described the use of @command{memtest}, the default
(and most useful) of @code{valgrind}'s tools.  There are others
described in its documentation: @command{helgrind} can be useful for
threaded programs.

@node Using Address Sanitizer, Using Undefined Behaviour Sanitizer, Using valgrind, Checking memory access
@subsection Using the Address Sanitizer

@c https://github.com/google/sanitizers/wiki/AddressSanitizer
@command{AddressSanitizer} (`ASan') is a tool with similar aims to the
memory checker in @command{valgrind}.  It is available with suitable
builds@footnote{currently on @cputype{x86_64}/@cputype{ix86} Linux and
FreeBSD, with some support for Intel macOS but not
with the toolchain normally used with @R{}. (There is a faster variant,
HWASAN, for @cputype{aarch64} only.)  On some platforms the runtime
library, @pkg{libasan}, needs to be installed separately, and for
checking C++ you may also need @pkg{libubsan}.} of @command{gcc} and
@command{clang} on common Linux and macOS platforms.  See
@uref{https://clang.llvm.org/docs/UsersManual.html#controlling-code-generation},
@uref{https://clang.llvm.org/docs/AddressSanitizer.html} and
@uref{https://github.com/google/sanitizers}.

More thorough checks of C++ code are done if the C++ library has been
`annotated': at the time of writing this applied to @code{std::vector}
in @code{libc++} for use with @command{clang} and gives rise to
@samp{container-overflow}@footnote{see
@uref{https://llvm.org/devmtg/2014-04/PDFs/LightningTalks/EuroLLVM%202014%20--%20container%20overflow.pdf}.}
reports.

It requires code to have been compiled @emph{and linked} with
@option{-fsanitize=address} and compiling with @code{-fno-omit-frame-pointer}
will give more legible reports.  It has a runtime penalty of 2--3x,
extended compilation times and uses substantially more memory, often
1--2GB, at run time.  On 64-bit platforms it reserves (but does not
allocate) 16--20TB of virtual memory: restrictive shell settings can
cause problems.

By comparison with @command{valgrind}, ASan can
detect misuse of stack and global variables but not the use of
uninitialized memory.

Recent versions return symbolic addresses for the location of the error
provided @command{llvm-symbolizer}@footnote{part of the LLVM project and
distributed in @code{llvm} RPMs and @code{.deb}s on Linux.  It is not
currently shipped by Apple.} is on the path: if it is available but not
on the path or has been renamed@footnote{as Ubuntu has been said to
do.}, one can use an environment variable, e.g.@:

@example
ASAN_SYMBOLIZER_PATH=/path/to/llvm-symbolizer
@end example

@noindent
An alternative is to pipe the output through
@command{asan_symbolize.py}@footnote{installed on some Linux systems as
@command{asan_symbolize}, and obtainable from
@uref{https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/asan/scripts/asan_symbolize.py}:
it makes use of @command{llvm-symbolizer} if available.} and perhaps
then (for compiled C++ code) @command{c++filt}.  (On macOS, you may need
to run @command{dsymutil} to get line-number reports.)

The simplest way to make use of this is to build a version of @R{} with
something like

@example
CC="gcc -std=gnu99 -fsanitize=address"
CFLAGS="-fno-omit-frame-pointer -g -O2 -Wall -pedantic -mtune=native"
@end example

@noindent
which will ensure that the @code{libasan} run-time library is compiled
into the @R{} executable.  However this check can be enabled on a
per-package basis by using a @file{~/.R/Makevars} file like
@example
CC = gcc -std=gnu99 -fsanitize=address -fno-omit-frame-pointer
CXX = g++ -fsanitize=address -fno-omit-frame-pointer
FC = gfortran -fsanitize=address
@end example
@noindent
(Note that @code{-fsanitize=address} has to be part of the compiler
specification to ensure it is used for linking.  These settings will not
be honoured by packages which ignore @file{~/.R/Makevars}.)  It will
be necessary to build @R{} with

@example
MAIN_LDFLAGS = -fsanitize=address
@end example

@noindent
to link the runtime libraries into the @R{} executable if it was not
specified as part of @samp{CC} when @R{} was built.  (For some builds
without OpenMP, @option{-pthread} is also required.)

For options available @emph{via} the environment variable
@env{ASAN_OPTIONS} see
@uref{https://github.com/google/sanitizers/wiki/AddressSanitizerFlags}.
With @command{gcc} additional control is available @emph{via} the
@option{--param} flag: see its @command{man} page.

For more detailed information on an error, @R{} can be run under a
debugger with a breakpoint set before the address sanitizer report is
produced: for @command{gdb} or @command{lldb} you could use
@example
break __asan_report_error
@end example
@noindent
(See
@uref{https://github.com/google/sanitizers/wiki/AddressSanitizerAndDebugger}.)

More recent versions@footnote{including @command{gcc} 7.1 and
@command{clang} 4.0.0: for @command{gcc} it is implied by
@option{-fsanitize=address}.}  added the flag
@option{-fsanitize-address-use-after-scope}: see
@uref{https://github.com/google/sanitizers/wiki/AddressSanitizerUseAfterScope}.

One of the checks done by ASan is that @code{malloc/free} and in C++
@code{new/delete} and @code{new[]/delete[]} are used consistently
(rather than say @code{free} being used to dealloc memory allocated by
@code{new[]}).  This matters on some systems but not all: unfortunately
on some of those where it does not matter, system libraries@footnote{for
example, X11/GL libraries on Linux, seen when checking package
@CRANpkg{rgl} and some others using it---a workaround is to set
environment variable @env{RGL_USE_NULL=true}.} are not consistent.  The
check can be suppressed by including @samp{alloc_dealloc_mismatch=0} in
@env{ASAN_OPTIONS}.

ASan also checks system calls and sometimes reports can refer to
problems in the system software and not the package nor @R{}.  A couple
of reports have been of `heap-use-after-free' errors in the X11
libraries called from Tcl/Tk.

@menu
* Using Leak Sanitizer::
@end menu

@node Using Leak Sanitizer,  , Using Address Sanitizer, Using Address Sanitizer
@subsubsection Using the Leak Sanitizer

For @code{x86_64} Linux there is a leak sanitizer, `LSan': see
@uref{https://github.com/google/sanitizers/wiki/AddressSanitizerLeakSanitizer}.
This is available on recent versions of @code{gcc} and @code{clang}, and
where available is compiled in as part of ASan.

One way to invoke this from an ASan-enabled build is by the environment
variable

@example
ASAN_OPTIONS='detect_leaks=1'
@end example
@noindent
However, this was made the default as from @command{clang} 3.5 and
@command{gcc} 5.1.0.

When LSan is enabled, leaks give the process a failure error status (by
default @code{23}).  For an @R{} package this means the @R{} process,
and as the parser retains some memory to the end of the process, if @R{}
itself was built against ASan all runs will have a failure error status
(which may include running @R{} as part of building @R{} itself).

To disable this, allocation-mismatch checking and some strict C++
checking use

@example
setenv ASAN_OPTIONS 'alloc_dealloc_mismatch=0:detect_leaks=0:detect_odr_violation=0'
@end example

LSan also has a `stand-alone' mode where it is compiled in using
@option{-fsanitize=leak} and avoids the run-time overhead of ASan.

@node Using Undefined Behaviour Sanitizer, Other analyses with `clang', Using Address Sanitizer, Checking memory access
@subsection Using the Undefined Behaviour Sanitizer

`Undefined behaviour' is where the language standard does not require
particular behaviour from the compiler.  Examples include division by
zero (where for doubles @R{} requires the
@acronym{ISO}/@acronym{IEC}@tie{}60559 behaviour but C/C++ do not), use
of zero-length arrays, shifts too far for signed types (e.g.@: @code{int
x, y; y = x << 31;}), out-of-range coercion, invalid C++ casts and
mis-alignment.  Not uncommon examples of out-of-range coercion in @R{}
packages are attempts to coerce a @code{NaN} or infinity to type
@code{int} or @code{NA_INTEGER} to an unsigned type such as
@code{size_t}.  Also common is @code{y[x - 1]} forgetting that @code{x}
might be @code{NA_INTEGER}.

`UBSanitizer' is a tool for C/C++ source code selected by
@option{-fsanitize=undefined} in suitable builds@footnote{On some
platforms the runtime library, @pkg{libubsan}, needs to be installed
separately.} of @command{clang} and GCC.  Its (main) runtime library is
linked into each package's DLL, so it is less often needed to be
included in @env{MAIN_LDFLAGS}.  Platforms supported by @command{clang}
are listed at
@uref{https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html#supported-platforms}:
@acronym{CRAN} uses it for C/C++ with both GCC and @command{clang} on
@cputype{x86_64} Linux: the two toolchains often highlight different
things with more reports from @command{clang} than GCC.

This sanitizer can be combined with the Address Sanitizer by
@option{-fsanitize=undefined,address} (where both are supported).

Finer control of what is checked can be achieved by other options.

For @command{clang} see
@uref{https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html#ubsan-checks}.
The current set is (on a single line):
@example
-fsanitize=alignment,bool,bounds,builtin,enum,float-cast-overflow,
float-divide-by-zero,function,implicit-unsigned-integer-truncation,
implicit-signed-integer-truncation,implicit-integer-sign-change,
integer-divide-by-zero,nonnull-attribute,null,object-size,
pointer-overflow,return,returns-nonnull-attribute,shift,
signed-integer-overflow,unreachable,unsigned-integer-overflow,
unsigned-shift-base,vla-bound,vptr
@end example

@noindent
(plus the more specific versions @code{shift-base} and
@code{shift-exponent}) a subset of which could be combined with
@code{address}, or use something like

@example
-fsanitize=undefined -fno-sanitize=float-divide-by-zero
@end example

@noindent
Options @code{function}, @code{return} and @code{vptr} apply only to C++: to
use @code{vptr} its run-time library needs to be linked into the main
@R{} executable by building the latter with something like
@example
MAIN_LD="clang++ -fsanitize=undefined"
@end example
@noindent
Option @code{float-divide-by-zero} is undesirable for use with @R{}
which allow such divisions as part of @acronym{IEC}@tie{}60559
arithmetic, and in versions of @command{clang} since June 2019 it is no
longer part of @option{-fsanitize=undefined}.

For GCC see
@uref{https://gcc.gnu.org/onlinedocs/gcc/Instrumentation-Options.html}
(or the manual for your version of GCC, installed or @emph{via}
@uref{https://gcc.gnu.org/onlinedocs/}: look for `Program
Instrumentation Options') for the options supported by GCC: 10 supported
@example
-fsanitize=alignment,bool,bounds,builtin,enum,integer-divide-by-zero,
nonnull-attribute,null,object-size,pointer-overflow,return,
returns-nonnull-attribute,shift,signed-integer-overflow,
unreachable,vla-bound,vptr
@end example
@noindent
plus the more specific versions @code{shift-base} and
@code{shift-exponent} and non-default options
@example
bound-strict,float-cast-overflow,float-divide-by-zero
@end example
@noindent
where @code{float-divide-by-zero} is not desirable for @R{} uses and
@code{bounds-strict} is an extension of @code{bounds}.

Other useful flags include
@example
-no-fsanitize-recover
@end example

@noindent
which causes the first report to be fatal (it always is for the
@code{unreachable} and @code{return} suboptions).  For more detailed
information on where the runtime error occurs, using 

@example
setenv UBSAN_OPTIONS 'print_stacktrace=1' 
@end example
@noindent will include a traceback in the report.  Beyond that, @R{} can
be run under a debugger with a breakpoint set before the sanitizer
report is produced: for @command{gdb} or @command{lldb} you could use
@example
break __ubsan_handle_float_cast_overflow
break __ubsan_handle_float_cast_overflow_abort
@end example
@noindent
or similar (there are handlers for each type of undefined behaviour).

There are also the compiler flags @option{-fcatch-undefined-behavior}
and @option{-ftrapv}, said to be more reliable in @command{clang} than
@command{gcc}.

For more details on the topic see
@uref{https://blog.regehr.org/archives/213} and
@uref{https://blog.llvm.org/2011/05/what-every-c-programmer-should-know.html}
(which has 3 parts).

It may or may not be possible to build @R{} itself with
@option{-fsanitize=undefined}: problems have been seen with OpenMP-using
code with @command{gcc} but there has been success with @command{clang}.


@node Other analyses with `clang', Other analyses with `gcc', Using Undefined Behaviour Sanitizer, Checking memory access
@subsection Other analyses with `clang'

Recent versions of @command{clang} on @cputype{x86_64} Linux have
`ThreadSanitizer' (@uref{https://github.com/google/sanitizers/wiki#threadsanitizer}),
a `data race detector for C/C++ programs', and `MemorySanitizer'
(@uref{https://clang.llvm.org/docs/MemorySanitizer.html},
@uref{https://github.com/google/sanitizers})
for the detection of uninitialized memory.  Both are based on and
provide similar functionality to tools in @command{valgrind}.

@command{clang} has a `Static Analyzer' which can be run on the source
files during compilation: see @uref{https://clang-analyzer.llvm.org/}.

@node Other analyses with `gcc', Using `Dr. Memory', Other analyses with `clang', Checking memory access
@subsection Other analyses with `gcc'

GCC 10 introduced a new flag @option{-fanalyzer} which does static
analysis during compilation, currently for C code.  It is regarded as
@emph{experimental} and it may slow down computation considerably when
problems are found (and use many GB of resident memory).  There is some
overlap with problems detected by the Undefined Behaviour sanitizer, but
some issues are only reported by this tool and as it is a static
analysis, it does not rely on code paths being exercised.

See
@uref{https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Static-Analyzer-Options.html}
(or the documentation for your version of @command{gcc} if later) and
@uref{https://developers.redhat.com/blog/2020/03/26/static-analysis-in-gcc-10}

@node Using `Dr. Memory', Fortran array bounds checking, Other analyses with `gcc', Checking memory access
@subsection Using `Dr. Memory'

`Dr. Memory' from @uref{https://drmemory.org/} is a memory checker
for (currently) 32-bit Windows, Linux and macOS with similar aims to
@command{valgrind}.  It works with unmodified executables@footnote{but
works better if inlining and frame pointer optimizations are disabled.}
and detects memory access errors, uninitialized reads and memory leaks.

@node Fortran array bounds checking,  , Using `Dr. Memory', Checking memory access
@subsection Fortran array bounds checking

Most of the Fortran compilers used with @R{} allow code to be compiled
with checking of array bounds: for example @command{gfortran} has option
@option{-fbounds-check} and Oracle Developer Studio has @option{-C}.
This will give an error when the upper or lower bound is exceeded, e.g.
@example
At line 97 of file .../src/appl/dqrdc2.f
Fortran runtime error: Index '1' of dimension 1 of array 'x' above upper bound of 0
@end example

One does need to be aware that lazy programmers often specify Fortran
dimensions as @code{1} rather than @code{*} or a real bound and these
will be reported (as may @code{*} dimensions)

It is easy to arrange to use this check on just the code in your
package: add to @file{~/.R/Makevars} something like (for
@command{gfortran})
@example
FFLAGS = -g -O2 -mtune=native -fbounds-check
@end example

@noindent
when you run @command{R CMD check}.

This may report errors with the way that Fortran character variables are
passed, particularly when Fortran subroutines are called from C code and
character lengths are not passed (@pxref{Fortran character strings}).


@node Debugging compiled code, Using Link-time Optimization, Checking memory access, Debugging
@section Debugging compiled code
@cindex Debugging


Sooner or later programmers will be faced with the need to debug
compiled code loaded into @R{}.  This section is geared to platforms
using @command{gdb} with code compiled by @code{gcc}, but similar things
are possible with other debuggers such as @command{lldb}
(@uref{https://lldb.llvm.org/}, used on macOS) and Sun's @command{dbx}:
some debuggers have graphical front-ends available.

Consider first `crashes', that is when @R{} terminated unexpectedly with
an illegal memory access (a `segfault' or `bus error'), illegal
instruction or similar.  Unix-alike versions of @R{} use a signal
handler which aims to give some basic information.  For example

@example
 *** caught segfault ***
address 0x20000028, cause 'memory not mapped'

Traceback:
 1: .identC(class1[[1]], class2)
 2: possibleExtends(class(sloti), classi, ClassDef2 = getClassDef(classi,
where = where))
 3: validObject(t(cu))
 4: stopifnot(validObject(cu <- as(tu, "dtCMatrix")), validObject(t(cu)),
validObject(t(tu)))

Possible actions:
1: abort (with core dump)
2: normal R exit
3: exit R without saving workspace
4: exit R saving workspace
Selection: 3
@end example

@noindent
Since the @R{} process may be damaged, the only really safe options are
the first or third.  (Note that a core dump is only produced where
enabled: a common default in a shell is to limit its size to 0, thereby
disabling it.)

A fairly common cause of such crashes is a package which uses @code{.C}
or @code{.Fortran} and writes beyond (at either end) one of the
arguments it is passed.  There is a good way to detect this: using
@code{options(CBoundsCheck = TRUE)} (which can be selected @emph{via}
the environment variable @env{R_C_BOUNDS_CHECK=yes)} changes the way
@code{.C} and @code{.Fortran} work to check if the compiled code writes
in the 64 bytes at either end of an argument.

Another cause of a `crash' is to overrun the C stack.  @R{} tries to
track that in its own code, but it may happen in third-party compiled
code.  For modern POSIX-compliant OSes @R{} can safely catch that and
return to the top-level prompt, so one gets something like

@example
> .C("aaa")
Error: segfault from C stack overflow
>
@end example

@noindent
However, C stack overflows are fatal under Windows and normally defeat
attempts at debugging on that platform.  Further, the size of the stack
is set when @R{} is compiled on Windows, whereas on POSIX OSes it can be
set in the shell from which @R{} is launched.

If you have a crash which gives a core dump you can use something like

@example
gdb /path/to/R/bin/exec/R core.12345
@end example

@noindent
to examine the core dump.  If core dumps are disabled or to catch errors
that do not generate a dump one can run @R{} directly under a debugger
by for example

@example
$ R -d gdb --vanilla
...
gdb> run
@end example

@noindent
at which point @R{} will run normally, and hopefully the debugger will
catch the error and return to its prompt.  This can also be used to
catch infinite loops or interrupt very long-running code.  For a simple
example

@example
> for(i in 1:1e7) x <- rnorm(100)
[hit Ctrl-C]
Program received signal SIGINT, Interrupt.
0x00397682 in _int_free () from /lib/tls/libc.so.6
(gdb) where
#0  0x00397682 in _int_free () from /lib/tls/libc.so.6
#1  0x00397eba in free () from /lib/tls/libc.so.6
#2  0xb7cf2551 in R_gc_internal (size_needed=313)
    at /users/ripley/R/svn/R-devel/src/main/memory.c:743
#3  0xb7cf3617 in Rf_allocVector (type=13, length=626)
    at /users/ripley/R/svn/R-devel/src/main/memory.c:1906
#4  0xb7c3f6d3 in PutRNGstate ()
    at /users/ripley/R/svn/R-devel/src/main/RNG.c:351
#5  0xb7d6c0a5 in do_random2 (call=0x94bf7d4, op=0x92580e8, args=0x9698f98,
    rho=0x9698f28) at /users/ripley/R/svn/R-devel/src/main/random.c:183
...
@end example

In many cases it is possible to attach a debugger to a running process:
this is helpful if an alternative front-end is in use or to investigate
a task that seems to be taking far too long.  This is done by something
like

@example
gdb -p @var{pid}
@end example

@noindent
where @code{@var{pid}} is the id of the @R{} executable or front-end
process and can be found from within a running @R{} process by calling
@code{Sys.getpid()} or from a process monitor.  This stops the process
so its state can be examined: use @code{continue} to resume execution.


Some ``tricks'' worth knowing follow:

@menu
* Finding entry points::
* Inspecting R objects::
* Debugging on macOS::
@end menu

@node Finding entry points, Inspecting R objects, Debugging compiled code, Debugging compiled code
@subsection Finding entry points in dynamically loaded code

Under most compilation environments, compiled code dynamically loaded
into @R{} cannot have breakpoints set within it until it is loaded.  To
use a symbolic debugger on such dynamically loaded code under
Unix-alikes use

@itemize @bullet
@item
Call the debugger on the @R{} executable, for example by @kbd{R -d gdb}.
@item
Start @R{}.
@item
At the @R{} prompt, use @code{dyn.load} or @code{library} to load your
shared object.
@item
Send an interrupt signal.  This will put you back to the debugger
prompt.
@item
Set the breakpoints in your code.
@item
Continue execution of @R{} by typing @kbd{signal 0@key{RET}}.
@end itemize

Under Windows signals may not be able to be used, and if so the procedure is
more complicated.  See the rw-FAQ.


@node Inspecting R objects, Debugging on macOS, Finding entry points, Debugging compiled code
@subsection Inspecting R objects when debugging
@cindex Inspecting R objects when debugging

The key to inspecting @R{} objects from compiled code is the function
@code{PrintValue(SEXP @var{s})} which uses the normal @R{} printing
mechanisms to print the @R{} object pointed to by @var{s}, or the safer
version @code{R_PV(SEXP @var{s})} which will only print `objects'.
@findex PrintValue
@findex R_PV

One way to make use of @code{PrintValue} is to insert suitable calls
into the code to be debugged.

Another way is to call @code{R_PV} from the symbolic debugger.
(@code{PrintValue} is hidden as @code{Rf_PrintValue}.)  For example,
from @code{gdb} we can use

@example
(gdb) p R_PV(ab)
@end example

@noindent
using the object @code{ab} from the convolution example, if we have
placed a suitable breakpoint in the convolution C code.

To examine an arbitrary @R{} object we need to work a little harder.
For example, let

@example
R> DF <- data.frame(a = 1:3, b = 4:6)
@end example

@noindent
By setting a breakpoint at @code{do_get} and typing @kbd{get("DF")} at
the @R{} prompt, one can find out the address in memory of @code{DF}, for
example

@example
@group
Value returned is $1 = (SEXPREC *) 0x40583e1c
(gdb) p *$1
$2 = @{
  sxpinfo = @{type = 19, obj = 1, named = 1, gp = 0,
    mark = 0, debug = 0, trace = 0, = 0@},
  attrib = 0x40583e80,
  u = @{
    vecsxp = @{
      length = 2,
      type = @{c = 0x40634700 "0>X@@D>X@@0>X@@", i = 0x40634700,
        f = 0x40634700, z = 0x40634700, s = 0x40634700@},
      truelength = 1075851272,
    @},
    primsxp = @{offset = 2@},
    symsxp = @{pname = 0x2, value = 0x40634700, internal = 0x40203008@},
    listsxp = @{carval = 0x2, cdrval = 0x40634700, tagval = 0x40203008@},
    envsxp = @{frame = 0x2, enclos = 0x40634700@},
    closxp = @{formals = 0x2, body = 0x40634700, env = 0x40203008@},
    promsxp = @{value = 0x2, expr = 0x40634700, env = 0x40203008@}
  @}
@}
@end group
@end example

@noindent
(Debugger output reformatted for better legibility).

Using @code{R_PV()} one can ``inspect'' the values of the various
elements of the SEXP, for example,

@example
@group
(gdb) p R_PV($1->attrib)
$names
[1] "a" "b"

$row.names
[1] "1" "2" "3"

$class
[1] "data.frame"

$3 = void
@end group
@end example

To find out where exactly the corresponding information is stored, one
needs to go ``deeper'':

@example
@group
(gdb) set $a = $1->attrib
(gdb) p $a->u.listsxp.tagval->u.symsxp.pname->u.vecsxp.type.c
$4 = 0x405d40e8 "names"
(gdb) p $a->u.listsxp.carval->u.vecsxp.type.s[1]->u.vecsxp.type.c
$5 = 0x40634378 "b"
(gdb) p $1->u.vecsxp.type.s[0]->u.vecsxp.type.i[0]
$6 = 1
(gdb) p $1->u.vecsxp.type.s[1]->u.vecsxp.type.i[1]
$7 = 5
@end group
@end example

Another alternative is the @code{R_inspect} function which shows the
low-level structure of the objects recursively (addresses differ from
the above as this example is created on another machine):

@example
@group
(gdb) p R_inspect($1)
@@100954d18 19 VECSXP g0c2 [OBJ,NAM(2),ATT] (len=2, tl=0)
  @@100954d50 13 INTSXP g0c2 [NAM(2)] (len=3, tl=0) 1,2,3
  @@100954d88 13 INTSXP g0c2 [NAM(2)] (len=3, tl=0) 4,5,6
ATTRIB:
  @@102a70140 02 LISTSXP g0c0 []
    TAG: @@10083c478 01 SYMSXP g0c0 [MARK,NAM(2),gp=0x4000] "names"
    @@100954dc0 16 STRSXP g0c2 [NAM(2)] (len=2, tl=0)
      @@10099df28 09 CHARSXP g0c1 [MARK,gp=0x21] "a"
      @@10095e518 09 CHARSXP g0c1 [MARK,gp=0x21] "b"
    TAG: @@100859e60 01 SYMSXP g0c0 [MARK,NAM(2),gp=0x4000] "row.names"
    @@102a6f868 13 INTSXP g0c1 [NAM(2)] (len=2, tl=1) -2147483648,-3
    TAG: @@10083c948 01 SYMSXP g0c0 [MARK,gp=0x4000] "class"
    @@102a6f838 16 STRSXP g0c1 [NAM(2)] (len=1, tl=1)
      @@1008c6d48 09 CHARSXP g0c2 [MARK,gp=0x21,ATT] "data.frame"
@end group
@end example

In general the representation of each object follows the format:

@smallexample
@@<address> <type-nr> <type-name> <gc-info> [<flags>] ...
@end smallexample

For a more fine-grained control over the depth of the recursion
and the output of vectors @code{R_inspect3} takes additional two character()
parameters: maximum depth and the maximal number of elements that will
be printed for scalar vectors. The defaults in @code{R_inspect} are
currently -1 (no limit) and 5 respectively.

@node Debugging on macOS,  , Inspecting R objects, Debugging compiled code
@subsection Debugging on macOS

To debug code in a package it is easiest to unpack it in a directory and
install it with

@example
R CMD INSTALL --dsym @var{pkgname}
@end example

@noindent
as macOS does not store debugging symbols in the @file{.so} file.  (It
is not necessary to have @R{} built with debugging symbols, although
compiling the package should be done including @option{-g} in
@code{CFLAGS} / @code{CXXFLAGS} / @code{FCFLAGS} as appropriate.)

Security measures may prevent running a @acronym{CRAN} binary
distribution of @R{} under @command{lldb} or attaching this as a
debugger
(@uref{https://cran.r-project.org/bin/macosx/RMacOSX-FAQ.html#I-cannot-attach-debugger-to-R}),
although both were possible on High Sierra.  This can also affect
locally compiled builds, where attaching to an interactive @R{} session
under Big Sur or Monterey worked in 2021 after giving administrator
permission @emph{via} a popup-up.  (To debug in what Apple deems a
non-interactive session, e.g. logged in remotely, see @command{man
DevToolsSecurity}.)

Debugging a local build of @R{} on macOS can raise additional hurdles as
environment variables such as @env{DYLD_FALLBACK_LIBRARY_PATH} are not
usually passed through@footnote{By default as a security measure: see
@command{man dyld}.} the @command{lldb} process, resulting in messages
like
@example
R -d lldb
...
(lldb) run
Process 16828 launched: '/path/to/bin/exec/R' (x86_64)
dyld: Library not loaded: libR.dylib
  Referenced from: /path/to/bin/exec/R
@end example

@noindent
A quick workaround is to symlink the dylibs under @file{R_HOME/lib} to
somewhere where they will be found such as the current working
directory.  It would be possible to do as the distribution
does@footnote{See
@uref{https://svn.r-project.org/R-dev-web/trunk/CRAN/QA/Simon/R-build/fixpathR}:
@samp{@@executable_path} could be used rather than absolute paths.}  and
use @command{install_name_tool}, but that would have to be done for all
the dylibs including those in packages.

It may be simplest to attach the debugger to a running process (see
above).  Specifically, run @R{} and when it is at the prompt just before
a command that is to be debugged, at a terminal 

@example
ps -ef | grep exec/R
# identify the pid @var{pid} for the next command
lldb -p @var{pid}
(lldb) continue
@end example

@noindent
and then return to the @R{} console.

For non-interactive use, one may need @command{lldb --batch}.


@node Using Link-time Optimization,  , Debugging compiled code, Debugging
@section Using Link-time Optimization

Where supported, @emph{link time optimization} provides a comprehensive
way to check the consistency of calls between Fortran files or between C
and Fortran.  Use this @emph{via} @command{R CMD INSTALL --use-LTO}.

To set up support on a Unix-alike,
@ifset UseExternalXrefs
@xref{Link-Time Optimization, , Link-Time Optimization,
R-admin, R Installation and Administration}.
@end ifset
On Linux using GCC without building @R{} with LTO support,
@c This will not set LTO for linking, but GCC does not seem to need that.
it should suffice to set
@example
LTO_OPT = -flto
LTO_FC_OPT = -flto
AR = gcc-ar
NM = gcc-nm
@end example
in a personal (or site) @file{Makevars} file:
@ifset UseExternalXrefs
@xref{Customizing package compilation, , Customizing package compilation,
R-admin, R Installation and Administration},
@end ifset

For Windows, first edit file @file{etc/$@{R_ARCH@}/Makeconf} to give
@code{LTO_OPT} the value @code{-flto} or do so in a personal/site
@file{Makevars} file; see also file
@file{src/gnuwin32/README.compilation} in the sources.

For example:
@example
boot.f:61: warning: type of 'ddot' does not match original declaration [-Wlto-type-mismatch]
        y(j,i)=ddot(p,x(j,1),n,b(1,j,i),1)
crq.f:1023: note: return value type mismatch
@end example
@noindent
where the package author forgot to declare
@example
      double precision ddot
      external ddot
@end example
@noindent
in @file{boot.f}.  That package had its own copy of @code{ddot}: to
detect misuse of the one in @R{}'s BLAS library would have needed @R{}
configured with @option{--enable-lto=check}.

Further examples:

@c package assist 3.1.4
@example
rkpk2.f:77:5: warning: type of 'dstup' does not match original declaration [-Wlto-type-mismatch]
      *info, wk)
rkpk1.f:2565:5: note: type mismatch in parameter 14
       subroutine dstup (s, lds, nobs, nnull, qraux, jpvt, y, q, ldqr,
rkpk1.f:2565:5: note: 'dstup' was previously declared here
@end example
@noindent
where the fourteenth argument @code{dum} was missing in the call.

@c package gss 2.1-9
@example
reg.f:78:33: warning: type of 'dqrdc' does not match original declaration [-Wlto-type-mismatch]
       call dqrdc (sr, nobs, nobs, nnull, wk, dum, dum, 0)
dstup.f:20: note: 'dqrdc' was previously declared here
       call dqrdc (s, lds, nobs, nnull, qraux, jpvt, work, 1)
@end example
@noindent
@code{dqrdc} is a LINPACK routine from @R{}, @code{jpvt} is an integer
array and @code{work} is a double precision one so @code{dum} cannot
match both.  (If @option{--enable-lto=check} had been used the
comparison would have been with the definition in @R{}.)

For Fortran files all in the package, most inconsistencies can be
detected by concatenating the Fortran files and compiling the result,
sometimes with clearer diagnostics than provided by LTO.  For our last
two examples this gives

@example
all.f:2966:72:

      *info, work1)
                                                                        1
Warning: Missing actual argument for argument 'dum' at (1)
@end example
@noindent
and
@example
all.f:1663:72:

      *ipvtwk), wk(ikwk), wk(iwork1), wk(iwork2), info)
                                                                        1
Warning: Type mismatch in argument 'jpvt' at (1); passed REAL(8) to INTEGER(4)
@end example

@node System and foreign language interfaces, The R API, Debugging, Top
@chapter System and foreign language interfaces

@menu
* Operating system access::
* Interface functions .C and .Fortran::
* dyn.load and dyn.unload::
* Registering native routines::
* Creating shared objects::
* Interfacing C++ code::
* Fortran I/O::
* Linking to other packages::
* Handling R objects in C::
* Interface functions .Call and .External::
* Evaluating R expressions from C::
* Parsing R code from C::
* External pointers and weak references::
* Vector accessor functions::
* Character encoding issues::
@end menu

@node  Operating system access, Interface functions .C and .Fortran, System and foreign language interfaces, System and foreign language interfaces
@section  Operating system access
@cindex Operating system access

Access to operating system functions is @emph{via} the @R{} functions
@code{system} and @code{system2}.
@findex system
@findex system2
The details will differ by platform (see the on-line help), and about
all that can safely be assumed is that the first argument will be a
string @code{command} that will be passed for execution (not necessarily
by a shell) and the second argument to @code{system} will be
@code{internal} which if true will collect the output of the command
into an @R{} character vector.

On POSIX-compliant OSes these commands pass a command-line to a shell:
Windows is not POSIX-compliant and there is a separate function
@code{shell} to do so.

The function @code{system.time}
@findex system.time
is available for timing.  Timing on child processes is only available on
Unix-alikes, and may not be reliable there.

@node Interface functions .C and .Fortran, dyn.load and dyn.unload, Operating system access, System and foreign language interfaces
@section Interface functions @code{.C} and @code{.Fortran}
@cindex Interfaces to compiled code

@findex .C
@findex .Fortran

These two functions provide an interface to compiled code that has been
linked into @R{}, either at build time or @emph{via} @code{dyn.load}
(@pxref{dyn.load and dyn.unload}).  They are primarily intended for
compiled C and Fortran code respectively, but the @code{.C} function can
be used with other languages which can generate C interfaces, for
example C++ (@pxref{Interfacing C++ code}).

The first argument to each function is a character string specifying the
symbol name as known@footnote{possibly after some platform-specific
translation, e.g.@: adding leading or trailing underscores.} to C or
Fortran, that is the function or subroutine name.  (That the symbol is
loaded can be tested by, for example, @code{is.loaded("cg")}.  Use the
name you pass to @code{.C} or @code{.Fortran} rather than the translated
symbol name.)

There can be up to 65 further arguments giving @R{} objects to be passed
to compiled code.  Normally these are copied before being passed in, and
copied again to an @R{} list object when the compiled code returns.  If
the arguments are given names, these are used as names for the
components in the returned list object (but not passed to the compiled
code).

The following table gives the mapping between the modes of @R{} atomic
vectors and the types of arguments to a C function or Fortran
subroutine.

@quotation
@multitable {RRR storage.mode} {RRR unsigned char * RR} {DOUBLE PRECISION}
@headitem @R{} storage mode @tab  C type  @tab   Fortran type
@item @code{logical}   @tab @code{int *}     @tab @code{INTEGER}
@item @code{integer}   @tab @code{int *}     @tab @code{INTEGER}
@item @code{double}    @tab @code{double *}  @tab @code{DOUBLE PRECISION}
@item @code{complex}   @tab @code{Rcomplex *} @tab @code{DOUBLE COMPLEX}
@item @code{character} @tab @code{char **}   @tab @code{CHARACTER(255)}
@item @code{raw}       @tab @code{unsigned char *}    @tab none
@end multitable
@end quotation

@noindent
On all @R{} platforms @code{int} and @code{INTEGER} are 32-bit.  Code
ported from S-PLUS (which uses @code{long *} for @code{logical} and
@code{integer}) will not work on all 64-bit platforms (although it may
appear to work on some, including Windows).  Note also that if your
compiled code is a mixture of C functions and Fortran subprograms the
argument types must match as given in the table above.

C type @code{Rcomplex} is a structure with @code{double} members
@code{r} and @code{i} defined in the header file @file{R_ext/Complex.h}
included by @file{R.h}.  (On most platforms this is stored in a way
compatible with the C99 @code{double complex} type: however, it may not
be possible to pass @code{Rcomplex} to a C99 function expecting a
@code{double complex} argument.  Nor need it be compatible with a C++
@code{complex} type.  Moreover, the compatibility can depend on the
optimization level set for the compiler.)

Only a single character string of fixed length can be passed to or from
Fortran (the length is not passed), and the success of this is
compiler-dependent: its use was formally deprecated in 2019.  Other @R{}
objects can be passed to @code{.C}, but it is much better to use one of
the other interfaces.

It is possible to pass numeric vectors of storage mode @code{double} to
C as @code{float *} or to Fortran as @code{REAL} by setting the
attribute @code{Csingle}, most conveniently by using the @R{} functions
@code{as.single}, @code{single} or @code{mode}.  This is intended only
to be used to aid interfacing existing C or Fortran code.

Logical values are sent as @code{0} (@code{FALSE}), @code{1}
(@code{TRUE}) or @code{INT_MIN = -2147483648} (@code{NA}, but only if
@code{NAOK} is true), and the compiled code should return one of these
three values.  (Non-zero values other than @code{INT_MIN} are mapped to
@code{TRUE}.)  Note that the use of @code{int *} for Fortran logical is
not guaranteed to be portable (although people have gotten away with it
for many years): it is better to pass integers and convert to/from
Fortran logical in a Fortran wrapper.

Unless formal argument @code{NAOK} is true, all the other arguments are
checked for missing values @code{NA} and for the @acronym{IEEE} special
values @code{NaN}, @code{Inf} and @code{-Inf}, and the presence of any
of these generates an error.  If it is true, these values are passed
unchecked.

Argument @code{PACKAGE} confines the search for the symbol name to a
specific shared object (or use @code{"base"} for code compiled into
@R{}).  Its use is highly desirable, as there is no way to avoid two
package writers using the same symbol name, and such name clashes are
normally sufficient to cause @R{} to crash.  (If it is not present and
the call is from the body of a function defined in a package namespace,
the shared object loaded by the first (if any) @code{useDynLib}
directive will be used.) 
@c However, prior to @R{} 2.15.2 the detection of the correct namespace is
@c unreliable and you are strongly recommended to use the @code{PACKAGE}
@c argument for packages to be used with earlier versions of @R{}.

Note that the compiled code should not return anything except through
its arguments: C functions should be of type @code{void} and Fortran
subprograms should be subroutines.

To fix ideas, let us consider a very simple example which convolves two
finite sequences. (This is hard to do fast in interpreted @R{} code, but
easy in C code.)  We could do this using @code{.C} by

@example
@group
void convolve(double *a, int *na, double *b, int *nb, double *ab)
@{
    int nab = *na + *nb - 1;

    for(int i = 0; i < nab; i++)
        ab[i] = 0.0;
    for(int i = 0; i < *na; i++)
        for(int j = 0; j < *nb; j++)
            ab[i + j] += a[i] * b[j];
@}
@end group
@end example

@noindent
called from @R{} by

@example
@group
conv <- function(a, b)
    .C("convolve",
       as.double(a),
       as.integer(length(a)),
       as.double(b),
       as.integer(length(b)),
       ab = double(length(a) + length(b) - 1))$ab
@end group
@end example

Note that we take care to coerce all the arguments to the correct @R{}
storage mode before calling @code{.C}; mistakes in matching the types
can lead to wrong results or hard-to-catch errors.

Special care is needed in handling @code{character} vector arguments in
C (or C++).  On entry the contents of the elements are duplicated and
assigned to the elements of a @code{char **} array, and on exit the
elements of the C array are copied to create new elements of a character
vector.  This means that the contents of the character strings of the
@code{char **} array can be changed, including to @code{\0} to shorten
the string, but the strings cannot be lengthened.  It is
possible@footnote{Note that this is then not checked for over-runs by
option @code{CBoundsCheck = TRUE}.} to allocate a new string @emph{via}
@code{R_alloc} and replace an entry in the @code{char **} array by the
new string.  However, when character vectors are used other than in a
read-only way, the @code{.Call} interface is much to be preferred.

Passing character strings to Fortran code needs even more care, is
deprecated and should be avoided where possible.  Only the first element
of the character vector is passed in, as a fixed-length (255) character
array.  Up to 255 characters are passed back to a length-one character
vector.  How well this works (or even if it works at all) depends on the
C and Fortran compilers on each platform (including on their options).
Often what is being passed to Fortran is one of a small set of possible
values (a factor in @R{} terms) which could alternatively be passed as
an integer code: similarly Fortran code that wants to generate
diagnostic messages could pass an integer code to a C or @R{} wrapper
which would convert it to a character string.

It is possible to pass some @R{} objects other than atomic vectors @emph{via}
@code{.C}, but this is only supported for historical compatibility: use
the @code{.Call} or @code{.External} interfaces for such objects.  Any
C/C++ code that includes @file{Rinternals.h} should be called @emph{via}
@code{.Call} or @code{.External}.

@node dyn.load and dyn.unload, Registering native routines, Interface functions .C and .Fortran, System and foreign language interfaces
@section @code{dyn.load} and @code{dyn.unload}
@cindex Dynamic loading

@findex dyn.load
@findex dyn.unload

Compiled code to be used with @R{} is loaded as a shared object
(Unix-alikes including macOS, @pxref{Creating shared objects} for more
information) or DLL (Windows).

The shared object/DLL is loaded by @code{dyn.load} and unloaded by
@code{dyn.unload}.  Unloading is not normally necessary and is not safe in
general, but it is needed to allow the DLL to be re-built on some platforms,
including Windows. Unloading a DLL and then re-loading a DLL of the same name
may not work: Solaris uses the first version loaded. A DLL that registers
C finalizers, but fails to unregister them when unloaded, may cause R to crash
after unloading.

The first argument to both functions is a character string giving the
path to the object.  Programmers should not assume a specific file
extension for the object/DLL (such as @file{.so}) but use a construction
like

@example
file.path(path1, path2, paste0("mylib", .Platform$dynlib.ext))
@end example

@noindent
for platform independence.  On Unix-alike systems the path supplied to
@code{dyn.load} can be an absolute path, one relative to the current
directory or, if it starts with @samp{~}, relative to the user's home
directory.

Loading is most often done automatically based on the @code{useDynLib()}
declaration in the @file{NAMESPACE} file, but may be done
explicitly @emph{via} a call to @code{library.dynam}.
@findex library.dynam
This has the form

@example
library.dynam("libname", package, lib.loc)
@end example

@noindent
where @code{libname} is the object/DLL name @emph{with the extension
omitted}.  Note that the first argument, @code{chname}, should
@strong{not} be @code{package} since this will not work if the package
is installed under another name.

Under some Unix-alike systems there is a choice of how the symbols are
resolved when the object is loaded, governed by the arguments
@code{local} and @code{now}.  Only use these if really necessary: in
particular using @code{now=FALSE} and then calling an unresolved symbol
will terminate @R{} unceremoniously.

@R{} provides a way of executing some code automatically when a object/DLL
is either loaded or unloaded.  This can be used, for example, to
register native routines with @R{}'s dynamic symbol mechanism, initialize
some data in the native code, or initialize a third party library.  On
loading a DLL, @R{} will look for a routine within that DLL named
@code{R_init_@var{lib}} where @var{lib} is the name of the DLL file with
the extension removed.  For example, in the command

@example
library.dynam("mylib", package, lib.loc)
@end example

@noindent
R looks for the symbol named @code{R_init_mylib}.  Similarly, when
unloading the object, @R{} looks for a routine named
@code{R_unload_@var{lib}}, e.g., @code{R_unload_mylib}.  In either case,
if the routine is present, @R{} will invoke it and pass it a single
argument describing the DLL.  This is a value of type @code{DllInfo}
which is defined in the @file{Rdynload.h} file in the @file{R_ext}
directory.

Note that there are some implicit restrictions on this mechanism as the
basename of the DLL needs to be both a valid file name and valid as part
of a C entry point (e.g.@: it cannot contain @samp{.}): for portable
code it is best to confine DLL names to be @acronym{ASCII} alphanumeric
plus underscore.  If entry point @code{R_init_@var{lib}} is not found it
is also looked for with @samp{.} replaced by @samp{_}.


The following example shows templates for the initialization and
unload routines for the @code{mylib} DLL.

@quotation
@cartouche
@example
#include <R_ext/Rdynload.h>

void
R_init_mylib(DllInfo *info)
@{
  /* Register routines,
     allocate resources. */
@}

void
R_unload_mylib(DllInfo *info)
@{
  /* Release resources. */
@}
@end example
@end cartouche
@end quotation

If a shared object/DLL is loaded more than once the most recent version
is used.@footnote{Strictly this is OS-specific, but no exceptions have
been seen for many years.}  More generally, if the same symbol name
appears in several shared objects, the most recently loaded occurrence
is used.  The @code{PACKAGE} argument and registration (see the next
section) provide good ways to avoid any ambiguity in which occurrence is
meant.

On Unix-alikes the paths used to resolve dynamically linked dependent
libraries are fixed (for security reasons) when the process is launched,
so @code{dyn.load} will only look for such libraries in the locations
set by the @file{R} shell script (@emph{via} @file{etc/ldpaths}) and in
the OS-specific defaults.

Windows allows more control (and less security) over where dependent
DLLs are looked for.  On all versions this includes the @env{PATH}
environment variable, but with lowest priority: note that it does not
include the directory from which the DLL was loaded.  It is possible to
add a single path with quite high priority @emph{via} the @code{DLLpath}
argument to @code{dyn.load}.  This is (by default) used by
@code{library.dynam} to include the package's @file{libs/i386} or
@file{libs/x64} directory in the DLL search path.


@node Registering native routines, Creating shared objects, dyn.load and dyn.unload, System and foreign language interfaces
@section Registering native routines
@cindex Registering native routines

@menu
* Speed considerations::
* Converting a package to use registration::
* Linking to native routines in other packages::
@end menu

By `native' routine, we mean an entry point in compiled code.

In calls to @code{.C}, @code{.Call}, @code{.Fortran} and
@code{.External}, @R{} must locate the specified native routine by
looking in the appropriate shared object/DLL.  By default, @R{} uses the
operating-system-specific dynamic loader to lookup the symbol in
all@footnote{For calls from within a namespace the search is confined to
the DLL loaded for that package.} loaded DLLs and the @R{} executable
or libraries it is linked to.  Alternatively, the author of the DLL can
explicitly register routines with @R{} and use a single,
platform-independent mechanism for finding the routines in the DLL.  One
can use this registration mechanism to provide additional information
about a routine, including the number and type of the arguments, and
also make it available to @R{} programmers under a different name.
@c No sign of this in 15 years ....
@c In the future, registration may be used to
@c implement a form of ``secure'' or limited native access.

Registering routines has two main advantages: it provides a
faster@footnote{For unregistered entry points the OS's @code{dlsym}
routine is used to find addresses.  Its performance varies considerably
by OS and even in the best case it will need to search a much larger
symbol table than, say, the table of @code{.Call} entry points.} way to
find the address of the entry point @emph{via} tables stored in the DLL
at compilation time, and it provides a run-time check that the entry
point is called with the right number of arguments and, optionally, the
right argument types.

@findex R_registerRoutines
To register routines with @R{}, one calls the C routine
@code{R_registerRoutines}.  This is typically done when the DLL is first
loaded within the initialization routine @code{R_init_@var{dll name}}
described in @ref{dyn.load and dyn.unload}.  @code{R_registerRoutines}
takes 5 arguments.  The first is the @code{DllInfo} object passed by
@R{} to the initialization routine. This is where @R{} stores the
information about the methods.  The remaining 4 arguments are arrays
describing the routines for each of the 4 different interfaces:
@code{.C}, @code{.Call}, @code{.Fortran} and @code{.External}.  Each
argument is a @code{NULL}-terminated array of the element types given in
the following table:

@quotation
@multitable {@code{.External  }} {@code{R_ExternalMethodDef}}
@item @code{.C} @tab @code{R_CMethodDef}
@item @code{.Call} @tab @code{R_CallMethodDef}
@item @code{.Fortran} @tab @code{R_FortranMethodDef}
@item @code{.External} @tab @code{R_ExternalMethodDef}
@end multitable
@end quotation

Currently, the @code{R_ExternalMethodDef} type is the same as
@code{R_CallMethodDef} type and contains fields for the name of the
routine by which it can be accessed in @R{}, a pointer to the actual
native symbol (i.e., the routine itself), and the number of arguments
the routine expects to be passed from @R{}. For example, if we had a
routine named @code{myCall} defined as

@example
SEXP myCall(SEXP a, SEXP b, SEXP c);
@end example

@noindent
we would describe this as

@example
static const R_CallMethodDef callMethods[]  = @{
  @{"myCall", (DL_FUNC) &myCall, 3@},
  @{NULL, NULL, 0@}
@};
@end example

@noindent
along with any other routines for the @code{.Call} interface. For
routines with a variable number of arguments invoked @emph{via} the
@code{.External} interface, one specifies @code{-1} for the number of
arguments which tells @R{} not to check the actual number passed.

Routines for use with the @code{.C} and @code{.Fortran} interfaces are
described with similar data structures, which have one optional
additional field for describing the type of each argument.  If
specified, this field should be an array with the @code{SEXP} types
describing the expected type of each argument of the routine.
(Technically, the elements of the types array are of type
@code{R_NativePrimitiveArgType} which is just an unsigned integer.)
The @R{} types and corresponding type identifiers are provided in the
following table:

@quotation
@multitable {@code{character  }} {@code{SINGLESXP}}
@item @code{numeric} @tab @code{REALSXP}
@item @code{integer} @tab @code{INTSXP}
@item @code{logical} @tab @code{LGLSXP}
@item @code{single} @tab @code{SINGLESXP}
@item @code{character} @tab @code{STRSXP}
@item @code{list} @tab @code{VECSXP}
@end multitable
@end quotation

Consider a C routine, @code{myC}, declared as

@example
void myC(double *x, int *n, char **names, int *status);
@end example

We would register it as

@example
@group
static R_NativePrimitiveArgType myC_type[] = @{
    REALSXP, INTSXP, STRSXP, LGLSXP
@};

static const R_CMethodDef cMethods[] = @{
   @{"myC", (DL_FUNC) &myC, 4, myC_type@},
   @{NULL, NULL, 0, NULL@}
@};
@end group
@end example

@c Never implemented ....
@c One can also specify whether each argument is used simply as input, or
@c as output, or as both input and output.  The style field in the
@c description of a method is used for this.  The purpose is to
@c allow@footnote{but this is not currently done.} @R{} to transfer values
@c more efficiently across the @R{}-C/Fortran interface by avoiding copying
@c values when it is not necessary. Typically, one omits this information
@c in the registration data.

If registering types, check carefully that the number of types matches
the number of arguments: as the type array (here @code{myC_type}) is
passed as a pointer in C, the registration mechanism cannot check this
for you.

Note that @code{.Fortran} entry points are mapped to lowercase, so
registration should use lowercase only.

Having created the arrays describing each routine, the last step is to
actually register them with @R{}.  We do this by calling
@code{R_registerRoutines}.  For example, if we have the descriptions
above for the routines accessed by the @code{.C} and @code{.Call}
we would use the following code:

@example
void
R_init_myLib(DllInfo *info)
@{
   R_registerRoutines(info, cMethods, callMethods, NULL, NULL);
@}
@end example

This routine will be invoked when @R{} loads the shared object/DLL named
@code{myLib}.  The last two arguments in the call to
@code{R_registerRoutines} are for the routines accessed by
@code{.Fortran} and @code{.External} interfaces.  In our example, these
are given as @code{NULL} since we have no routines of these types.

When @R{} unloads a shared object/DLL, its registrations are removed.
There is no other facility for unregistering a symbol.

Examples of registering routines can be found in the different packages
in the @R{} source tree (e.g., @pkg{stats} and @pkg{graphics}).  Also,
there is a brief, high-level introduction in @emph{R News} (volume 1/3,
September 2001, pages 20--23,
@uref{https://www.r-project.org/doc/Rnews/Rnews_2001-3.pdf}).

Once routines are registered, they can be referred to as @R{} objects if
this is arranged in the @code{useDynLib} call in the package's
@file{NAMESPACE} file (see @ref{useDynLib}).  So for example the
@pkg{stats} package has
@example
# Refer to all C/Fortran routines by their name prefixed by C_
useDynLib(stats, .registration = TRUE, .fixes = "C_")
@end example

@noindent
in its @file{NAMESPACE} file, and then @code{ansari.test}'s default
methods can contain
@example
        pansari <- function(q, m, n)
            .C(C_pansari, as.integer(length(q)), p = as.double(q),
                as.integer(m), as.integer(n))$p
@end example

@noindent
This avoids the overhead of looking up an entry point each time it is
used, and ensures that the entry point in the package is the one used
(without a @code{PACKAGE = "pkg"} argument).

@code{R_init_} routines are often of the form
@example
void attribute_visible R_init_mypkg(DllInfo *dll)
@{
    R_registerRoutines(dll, CEntries, CallEntries, FortEntries,
                       ExternalEntries);
    R_useDynamicSymbols(dll, FALSE);
    R_forceSymbols(dll, TRUE);
...
@}
@end example

@noindent
@findex R_useDynamicSymbols
@findex R_forceSymbols
The @code{R_useDynamicSymbols} call says the DLL is not to be searched
for entry points specified by character strings so @code{.C} etc calls
will only find registered symbols: the @code{R_forceSymbols} call only
allows @code{.C} etc calls which specify entry points by @R{} objects
such as @code{C_pansari} (and not by character strings).  Each provides
some protection against accidentally finding your entry points when
people supply a character string without a package, and avoids slowing
down such searches.  (For the visibility attribute @pxref{Controlling
visibility}.)

In more detail, if a package @code{mypkg} contains entry points
@code{reg} and @code{unreg} and the first is registered as a 0-argument
@code{.Call} routine, we could use (from code in the package)

@example
.Call("reg")
.Call("unreg")
@end example

@noindent
Without or with registration, these will both work.  If
@code{R_init_mypkg} calls @code{R_useDynamicSymbols(dll, FALSE)}, only
the first will work.  If in addition to registration the
@file{NAMESPACE} file contains

@example
useDynLib(mypkg, .registration = TRUE, .fixes = "C_")
@end example

@noindent
then we can call @code{.Call(C_reg)}.  Finally, if @code{R_init_mypkg}
also calls @code{R_forceSymbols(dll, TRUE)}, only @code{.Call(C_reg)}
will work (and not @code{.Call("reg")}).  This is usually what we want:
it ensures that all of our own @code{.Call} calls go directly to the
intended code in our package and that no one else accidentally finds our
entry points.  (Should someone need to call our code from outside the
package, for example for debugging, they can use
@code{.Call(mypkg:::C_reg)}.)


@node Speed considerations, Converting a package to use registration, Registering native routines, Registering native routines
@subsection Speed considerations

Sometimes registering native routines or using a @code{PACKAGE} argument
can make a large difference.  The results can depend quite markedly on
the OS (and even if it is 32- or 64-bit), on the version of @R{} and
what else is loaded into @R{} at the time.

To fix ideas, first consider @code{x86_64} OS 10.7 and @R{} 2.15.2.  A
simple @code{.Call} function might be
@example
foo <- function(x) .Call("foo", x)
@end example
@noindent
with C code
@example
@group
#include <Rinternals.h>

SEXP foo(SEXP x)
@{
    return x;
@}
@end group
@end example
If we compile with by @command{R CMD SHLIB foo.c}, load the code by
@code{dyn.load("foo.so")} and run @code{foo(pi)} it took around 22
microseconds (us). Specifying the DLL by
@example
foo2 <- function(x) .Call("foo", x, PACKAGE = "foo")
@end example
@noindent
reduced the time to 1.7 us.

Now consider making these functions part of a package whose
@file{NAMESPACE} file uses @code{useDynlib(foo)}.  This immediately
reduces the running time as @code{"foo"} will be preferentially looked
for @file{foo.dll}.  Without specifying @code{PACKAGE} it took about 5
us (it needs to fathom out the appropriate DLL each time it is invoked
but it does not need to search all DLLs), and with the @code{PACKAGE}
argument it is again about 1.7 us.

Next suppose the package has registered the native routine @code{foo}.
Then @code{foo()} still has to find the appropriate DLL but can get to
the entry point in the DLL faster, in about 4.2 us.  And @code{foo2()}
now takes about 1 us.  If we register the symbols in the
@file{NAMESPACE} file and use
@example
foo3 <- function(x) .Call(C_foo, x)
@end example
@noindent
then the address for the native routine is looked up just once when the
package is loaded, and @code{foo3(pi)} takes about 0.8 us.

Versions using @code{.C()} rather than @code{.Call()} took about 0.2 us
longer.

These are all quite small differences, but C routines are not uncommonly
invoked millions of times for run times of a few microseconds each, and
those doing such things may wish to be aware of the differences.

On Linux and Solaris there is a smaller overhead in looking up
symbols.

Symbol lookup on Windows used to be far slower, so @R{} maintains a
small cache.  If the cache is currently empty enough that the symbol can
be stored in the cache then the performance is similar to Linux and
Solaris: if not it may be slower.  @R{}'s own code always uses
registered symbols and so these never contribute to the cache: however
many other packages do rely on symbol lookup.

In more recent versions of @R{} all the standard packages register
native symbols and do not allow symbol search, so in a new session
@code{foo()} can only look in @file{foo.so} and may be as fast as
@code{foo2()}.  This will no longer apply when many contributed packages
are loaded, and generally those last loaded are searched first.  For
example, consider @R{} 3.3.2 on x86_64 Linux.  In an empty @R{} session,
both @code{foo()} and @code{foo2()} took about 0.75 us; however after
packages @CRANpkg{igraph} and @CRANpkg{spatstat} had been loaded (which
loaded another 12 DLLs), @code{foo()} took 3.6 us but @code{foo2()}
still took about 0.80 us.  Using registration in a package reduced this
to 0.55 us and @code{foo3()} took 0.40 us, times which were unchanged
when further packages were loaded.

@node Converting a package to use registration, Linking to native routines in other packages, Speed considerations, Registering native routines
@subsection Example: converting a package to use registration

The @pkg{splines} package was converted to use symbol registration in
2001, but we can use it as an example@footnote{Because it is a standard
package, one would need to rename it before attempting to reproduce the
account here.} of what needs to be done for a small package.

@itemize

@item
Find the relevant entry points.
This is somewhat OS-specific, but something like the following should be
possible at the OS command-line

@example
@group
nm -g /path/to/splines.so | grep " T "
0000000000002670 T _spline_basis
0000000000001ec0 T _spline_value
@end group
@end example

@noindent
This indicates that there are two relevant entry points. (They may or
may not have a leading underscore, as here. Fortran entry points will
have a trailing underscore.)  Check in the @R{} code that they are
called by the package and how: in this case they are used by
@code{.Call}.

Alternatively, examine the package's @R{} code for all @code{.C},
@code{.Fortran}, @code{.Call} and @code{.External} calls.

@item
Construct the registration table.  First write skeleton registration
code, conventionally in file @file{src/init.c} (or at the end of the
only C source file in the package: if included in a C++ file the
@samp{R_init} function would need to be declared @code{extern "C"}):

@example
@group
#include <stdlib.h> // for NULL
#include <R_ext/Rdynload.h>

#define CALLDEF(name, n)  @{#name, (DL_FUNC) &name, n@}

static const R_CallMethodDef R_CallDef[] = @{
   CALLDEF(spline_basis, ?),
   CALLDEF(spline_value, ?),
   @{NULL, NULL, 0@}
@};

void R_init_splines(DllInfo *dll)
@{
    R_registerRoutines(dll, NULL, R_CallDef, NULL, NULL);
@}
@end group
@end example

@noindent
and then replace the @code{?} in the skeleton with the actual numbers of
arguments.  You will need to add declarations (also known as
`prototypes') of the functions unless appending to the only C source
file.  Some packages will already have these in a header file, or you
could create one and include it in @file{init.c}, for example
@file{splines.h} containing

@smallexample
@group
#include <Rinternals.h> // for SEXP
extern SEXP spline_basis(SEXP knots, SEXP order, SEXP xvals, SEXP derivs);
extern SEXP spline_value(SEXP knots, SEXP coeff, SEXP order, SEXP x, SEXP deriv);
@end group
@end smallexample
@noindent
Tools are available to extract declarations, at least for C and C++
code: see the help file for
@code{package_native_routine_registration_skeleton} in package
@pkg{tools}. Here we could have used
@example
cproto -I/path/to/R/include -e splines.c
@end example

For examples of registering other types of calls, see packages
@pkg{graphics} and @pkg{stats}.  In particular, when registering entry
points for @code{.Fortran} one needs declarations as if called from C,
such as

@example
@group
#include <R_ext/RS.h>
void F77_NAME(supsmu)(int *n, double *x, double *y,
                      double *w, int *iper, double *span, double *alpha,
                      double *smo, double *sc, double *edf);
@end group
@end example

@noindent
@command{gfortran} 8.4, 9.2 and later can help generate such prototypes
with its flag @option{-fc-prototypes-external} (although one will need
to replace the hard-coded trailing underscore with the @code{F77_NAME}
macro).

One can get away with inaccurate argument lists in the declarations: it
is easy to specify the arguments for @code{.Call} (all @code{SEXP}) and
@code{.External} (one @code{SEXP}) and as the arguments for @code{.C}
and @code{.Fortran} are all pointers, specifying them as @code{void *}
suffices. (For most platforms one can omit all the arguments, although
link-time optimization will warn.)

Using @option{-fc-prototypes-external} will give a prototype using
@code{int_least32_t *lgl} for Fortran @code{LOGICAL LGL}, but this is
not portable and traditionally it has been assumed that the C/C++
equivalent was @code{int *lgl}.  If adding a declaration just to
register a @code{.Fortran} call, the most portable version is @code{void
*lgl}.

@item
(Optional but highly recommended.)  Restrict @code{.Call} etc to use the
symbols you chose to register by editing @file{src/init.c} to contain

@example
@group
void R_init_splines(DllInfo *dll)
@{
    R_registerRoutines(dll, NULL, R_CallDef, NULL, NULL);
    R_useDynamicSymbols(dll, FALSE);
@}
@end group
@end example

@end itemize

A skeleton for the steps so far can be made using
@code{package_native_routine_registration_skeleton} in package
@pkg{tools}.  This will optionally create declarations based on the
usage in the @R{} code.

The remaining steps are optional but recommended.

@itemize
@item
Edit the @file{NAMESPACE} file to create @R{} objects for the registered
symbols:

@example
useDynLib(splines, .registration = TRUE, .fixes = "C_")
@end example

@item
Find all the relevant calls in the @R{} code and edit them to use the
@R{} objects.  This entailed changing the lines

@smallexample
temp <- .Call("spline_basis", knots, ord, x, derivs, PACKAGE = "splines")
y[accept] <- .Call("spline_value", knots, coeff, ord, x[accept], deriv, PACKAGE = "splines")
y = .Call("spline_value", knots, coef(object), ord, x, deriv, PACKAGE = "splines")
@end smallexample
@noindent
to

@smallexample
temp <- .Call(C_spline_basis, knots, ord, x, derivs)
y[accept] <- .Call(C_spline_value, knots, coeff, ord, x[accept], deriv)
y = .Call(C_spline_value, knots, coef(object), ord, x, deriv)
@end smallexample

Check that there is no @code{exportPattern} directive which
unintentionally exports the newly created @R{} objects.

@item
Restrict @code{.Call} to use the @R{} symbols by editing
@file{src/init.c} to contain

@example
@group
void R_init_splines(DllInfo *dll)
@{
    R_registerRoutines(dll, NULL, R_CallDef, NULL, NULL);
    R_useDynamicSymbols(dll, FALSE);
    R_forceSymbols(dll, TRUE);
@}
@end group
@end example

@item
Consider visibility.  On some OSes we can hide entry points from the
loader, which precludes any possible name clashes and calling them
accidentally (usually with incorrect arguments and crashing the @R{}
process). If we repeat the first step we now see

@example
@group
nm -g /path/to/splines.so | grep " T "
0000000000002e00 T _R_init_splines
00000000000025e0 T _spline_basis
0000000000001e20 T _spline_value
@end group
@end example

@noindent
If there were any entry points not intended to be used by the package we
should try to avoid exporting them, for example by making them
@code{static}.  Now that the two relevant entry points are only accessed
@emph{via} the registration table, we can hide them.  There are two ways
to do so on some Unix-alikes.  We can hide individual entry points
@emph{via}

@example
@group
#include <R_ext/Visibility.h>

SEXP attribute_hidden
spline_basis(SEXP knots, SEXP order, SEXP xvals, SEXP derivs)
@dots{}

SEXP attribute_hidden
spline_value(SEXP knots, SEXP coeff, SEXP order, SEXP x, SEXP deriv)
@dots{}
@end group
@end example

@noindent
Alternatively, we can change the default visibility for all C symbols by
including

@example
PKG_CFLAGS = $(C_VISIBILITY)
@end example

@noindent
in @file{src/Makevars}, and then we need to allow registration by
declaring @code{R_init_splines} to be visible:

@example
@group
#include <R_ext/Visibility.h>

void attribute_visible
R_init_splines(DllInfo *dll)
@dots{}
@end group
@end example

@noindent
@xref{Controlling visibility} for more details, including using Fortran
code and ways to restrict visibility on Windows.

@item
We end up with a file @file{src/init.c} containing

@quotation
@cartouche
@example
#include <stdlib.h>
#include <R_ext/Rdynload.h>
#include <R_ext/Visibility.h>  // optional

#include "splines.h"

#define CALLDEF(name, n)  @{#name, (DL_FUNC) &name, n@}

static const R_CallMethodDef R_CallDef[] = @{
    CALLDEF(spline_basis, 4),
    CALLDEF(spline_value, 5),
    @{NULL, NULL, 0@}
@};

void
attribute_visible  // optional
R_init_splines(DllInfo *dll)
@{
    R_registerRoutines(dll, NULL, R_CallDef, NULL, NULL);
    R_useDynamicSymbols(dll, FALSE);
    R_forceSymbols(dll, TRUE);
@}
@end example
@end cartouche
@end quotation

@end itemize

@node Linking to native routines in other packages,  , Converting a package to use registration, Registering native routines
@subsection Linking to native routines in other packages

In addition to registering C routines to be called by @R{}, it can at
times be useful for one package to make some of its C routines available
to be called by C code in another package.  The interface consists of
two routines declared in header @file{R_ext/Rdynload.h} as

@findex R_RegisterCCallable
@findex R_GetCCallable
@example
void R_RegisterCCallable(const char *package, const char *name,
                         DL_FUNC fptr);
DL_FUNC R_GetCCallable(const char *package, const char *name);
@end example

A package @pkg{packA} that wants to make a C routine @code{myCfun}
available to C code in other packages would include the call

@example
R_RegisterCCallable("packA", "myCfun", myCfun);
@end example
@noindent
in its initialization function @code{R_init_packA}.  A package
@pkg{packB} that wants to use this routine would retrieve the function
pointer with a call of the form

@example
p_myCfun = R_GetCCallable("packA", "myCfun");
@end example

The author of @pkg{packB} is responsible for ensuring that
@code{p_myCfun} has an appropriate declaration. In the future @R{} may
provide some automated tools to simplify exporting larger numbers of
routines.

A package that wishes to make use of header files in other packages
needs to declare them as a comma-separated list in the field
@samp{LinkingTo} in the @file{DESCRIPTION} file.  This then arranges
for the @file{include} directories in the installed linked-to packages
to be added to the include paths for C and C++ code.

It must specify@footnote{whether or not @samp{LinkingTo} is used.}
@samp{Imports} or @samp{Depends} of those packages, for they have to be
loaded@footnote{so there needs to be a corresponding @code{import} or
@code{importFrom} entry in the @file{NAMESPACE} file.} prior to this one
(so the path to their compiled code has been registered).


@acronym{CRAN} examples of the use of this mechanism include @CRANpkg{coxme}
linking to @CRANpkg{bdsmatrix} and @CRANpkg{xts} linking to
@CRANpkg{zoo}.

@node Creating shared objects, Interfacing C++ code, Registering native routines, System and foreign language interfaces
@section Creating shared objects
@cindex Creating shared objects
@findex R CMD SHLIB

Shared objects for loading into @R{} can be created using @command{R CMD
SHLIB}.  This accepts as arguments a list of files which must be object
files (with extension @file{.o}) or sources for C, C++, Fortran,
Objective C or Objective C++ (with extensions @file{.c}, @file{.cc} or
@file{.cpp}, @file{.f} (fixed-form Fortran), @file{.f90} or @file{.f95}
(free-form), @file{.m}, and @file{.mm} or @file{.M}, respectively), or
commands to be passed to the linker.  See @kbd{R CMD SHLIB --help} (or
the @R{} help for @code{SHLIB}) for usage information.  Note that files
intended for the Fortran pre-processor with extension @file{.F} are not
accepted.

If compiling the source files does not work ``out of the box'', you can
specify additional flags by setting some of the variables
@vindex PKG_CPPFLAGS
@code{PKG_CPPFLAGS} (for the C/C++ preprocessor, mainly @samp{-I},
@samp{-D} and @samp{-U} flags),
@vindex PKG_CFLAGS
@vindex PKG_CXXFLAGS
@vindex PKG_FFLAGS
@vindex PKG_OBJCFLAGS
@vindex PKG_OBJCXXFLAGS
@code{PKG_CFLAGS}, @code{PKG_CXXFLAGS}, @code{PKG_FFLAGS},
@code{PKG_OBJCFLAGS}, and @code{PKG_OBJCXXFLAGS}
(for the C, C++, Fortran, Objective C, and Objective C++
compilers, respectively) in the file @file{Makevars} in the compilation
directory (or, of course, create the object files directly from the
command line).
@vindex PKG_LIBS
Similarly, variable @code{PKG_LIBS} in @file{Makevars} can be used for
additional @samp{-l} and @samp{-L} flags to be passed to the linker when
building the shared object. (Supplying linker commands as arguments to
@code{R CMD SHLIB} will take precedence over @code{PKG_LIBS} in
@file{Makevars}.)

@vindex OBJECTS
It is possible to arrange to include compiled code from other languages
by setting the macro @samp{OBJECTS} in file @file{Makevars}, together
with suitable rules to make the objects.

Flags that are already set (for example in file
@file{etc@var{R_ARCH}/Makeconf}) can be overridden by the environment
variable @env{MAKEFLAGS} (at least for systems using a POSIX-compliant
@code{make}), as in (Bourne shell syntax)

@example
MAKEFLAGS="CFLAGS=-O3" R CMD SHLIB *.c
@end example

It is also possible to set such variables in personal @file{Makevars}
files, which are read after the local @file{Makevars} and the system
makefiles or in a site-wide @file{Makevars.site} file.
@ifset UseExternalXrefs
@xref{Customizing package compilation, , Customizing package compilation,
R-admin, R Installation and Administration},
@end ifset


Note that as @command{R CMD SHLIB} uses Make, it will not remake a shared
object just because the flags have changed, and if @file{test.c} and
@file{test.f} both exist in the current directory

@example
R CMD SHLIB test.f
@end example

@noindent
will compile @file{test.c}!


If the @file{src} subdirectory of an add-on package contains source code
with one of the extensions listed above or a file @file{Makevars} but
@strong{not} a file @file{Makefile}, @command{R CMD INSTALL} creates a
shared object (for loading into @R{} through @code{useDynlib} in the
@file{NAMESPACE}, or in the @code{.onLoad} function of the package)
using the @command{R CMD SHLIB} mechanism.  If file @file{Makevars}
exists it is read first, then the system makefile and then any personal
@file{Makevars} files.

If the @file{src} subdirectory of package contains a file
@file{Makefile}, this is used by @command{R CMD INSTALL} in place of the
@code{R CMD SHLIB} mechanism.  @command{make} is called with makefiles
@file{@var{R_HOME}/etc@var{R_ARCH}/Makeconf}, @file{src/Makefile} and
any personal @file{Makevars} files (in that order).  The first target
found in @file{src/Makefile} is used.

It is better to make use of a @file{Makevars} file rather than a
@file{Makefile}: the latter should be needed only exceptionally.

@c Not so clearcut on case-insensitive file systems.
@c Note that whereas @code{R CMD INSTALL} makes use of a @file{Makefile},
@c @code{R CMD SHLIB} does not.  The file must be named @file{Makefile},
@c not for example @file{makefile} nor @file{GNUmakefile}.

Under Windows the same commands work, but @file{Makevars.win} will be
used in preference to @file{Makevars}, and only @file{src/Makefile.win}
will be used by @code{R CMD INSTALL} with @file{src/Makefile} being
ignored. Since @R{} 4.2.0, @file{Makevars.ucrt} will be used in preference to
@file{Makevars.win} and @file{src/Makefile.ucrt} will be used in preference
to @file{src/Makefile.win}.
For past experiences of building DLLs with a variety of
compilers, see file @samp{README.packages}.  
Under Windows you can supply an exports definitions file called
@file{@var{dllname}-win.def}: otherwise all entry points in objects (but
not libraries) supplied to @code{R CMD SHLIB} will be exported from the
DLL.  An example is @file{stats-win.def} for the @pkg{stats} package: a
@acronym{CRAN} example in package @CRANpkg{fastICA}.

If you feel tempted to read the source code and subvert these
mechanisms, please resist.  Far too much developer time has been wasted
in chasing down errors caused by failures to follow this documentation,
and even more by package authors demanding explanations as to why their
packages no longer work.
@c Jasjeet Singh Sekhon: this is your moment of infamy.
In particular, undocumented environment or @command{make} variables are
not for use by package writers and are subject to change without notice.

@node Interfacing C++ code, Fortran I/O, Creating shared objects, System and foreign language interfaces
@section Interfacing C++ code
@cindex Interfacing C++ code
@cindex C++ code, interfacing

Suppose we have the following hypothetical C++ library, consisting of
the two files @file{X.h} and @file{X.cpp}, and implementing the two
classes @code{X} and @code{Y} which we want to use in @R{}.

@quotation
@cartouche
@example
// X.h

class X @{
public: X (); ~X ();
@};

class Y @{
public: Y (); ~Y ();
@};
@end example
@end cartouche
@end quotation

@quotation
@cartouche
@example
// X.cpp

#include <R.h>
#include "X.h"

static Y y;

X::X()  @{ REprintf("constructor X\n"); @}
X::~X() @{ REprintf("destructor X\n");  @}
Y::Y()  @{ REprintf("constructor Y\n"); @}
Y::~Y() @{ REprintf("destructor Y\n");  @}
@end example
@end cartouche
@end quotation

To use with @R{}, the only thing we have to do is writing a wrapper
function and ensuring that the function is enclosed in

@example
@group
extern "C" @{

@}
@end group
@end example

For example,

@quotation
@cartouche
@example
// X_main.cpp:

#include "X.h"

extern "C" @{

void X_main () @{
  X x;
@}

@} // extern "C"
@end example
@end cartouche
@end quotation

Compiling and linking should be done with the C++ compiler-linker
(rather than the C compiler-linker or the linker itself); otherwise, the
C++ initialization code (and hence the constructor of the static
variable @code{Y}) are not called.  On a properly configured system, one
can simply use

@example
R CMD SHLIB X.cpp X_main.cpp
@end example

@noindent
to create the shared object, typically @file{X.so} (the file name
extension may be different on your platform).  Now starting @R{} yields

@example
@group
R version 2.14.1 Patched (2012-01-16 r58124)
Copyright (C) 2012 The R Foundation for Statistical Computing
...
Type    "q()" to quit R.
@end group

@group
R> dyn.load(paste("X", .Platform$dynlib.ext, sep = ""))
constructor Y
R> .C("X_main")
constructor X
destructor X
list()
R> q()
Save workspace image? [y/n/c]: y
destructor Y
@end group
@end example

The @R{} for Windows @acronym{FAQ} (@file{rw-FAQ}) contains details of how
to compile this example under Windows.

Earlier versions of this example used C++ iostreams: this is best
avoided.  There is no guarantee that the output will appear in the @R{}
console, and indeed it will not on the @R{} for Windows console.  Use
@R{} code or the C entry points (@pxref{Printing}) for all I/O if at all
possible.  Examples have been seen where merely loading a DLL that
contained calls to C++ I/O upset @R{}'s own C I/O (for example by
resetting buffers on open files).

Most @R{} header files can be included within C++ programs but they
should @strong{not} be included within an @code{extern "C"} block (as
they include system headers@footnote{Even including C system headers in
such a block has caused compilation errors.}).

@subsection External C++ code

Quite a lot of external C++ software is header-only (e.g.@: most of the
Boost `libraries' including all those supplied by package @CRANpkg{BH},
and most of Armadillo as supplied by package @CRANpkg{RcppArmadillo})
and so is compiled when an @R{} package which uses it is installed.
This causes few problems.

A small number of external libraries used in @R{} packages have a C++
interface to a library of compiled code, e.g.@: packages @CRANpkg{rgdal}
and @CRANpkg{rjags}.  This raises many more problems!  The C++ interface
uses name-mangling and the
ABI@footnote{@uref{https://en.wikipedia.org/wiki/Application_binary_interface}.}
may depend on the compiler, version and even C++ defines@footnote{For
example, @samp{_GLIBCXX_USE_CXX11_ABI} in @command{g++} 5.1 and later:
@uref{https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html}.},
so requires the package C++ code to be compiled in exactly the same way
as the library (and what that was is often undocumented).  Examples
include use of @command{g++} @emph{vs} @command{clang++} or Solaris'
@command{CC}, and the two ABIs available for C++11 in @command{g++} with
different defaults for GCC 4.9 and 5.x in some Linux distributions.

Even fewer external libraries use C++ internally but present a C
interface, such as @CRANpkg{rgeos}.  These require the C++ runtime
library to be linked into the package's shared object/DLL, and this is
best done by including a dummy C++ file in the package sources.

There is a recent trend to link to the C++ interfaces offered by C
software such as @pkg{hdf5}, @pkg{pcre} and @pkg{ImageMagick}.  Their C
interfaces are much preferred for portability (and can be used from C++
code).  Also, the C++ interfaces are often optional in the software
build or packaged separately and so users installing from package
sources are far less likely to already have them installed.


@node Fortran I/O, Linking to other packages, Interfacing C++ code, System and foreign language interfaces
@section Fortran I/O

We have already warned against the use of C++ iostreams not least
because output is not guaranteed to appear on the @R{} console, and this
warning applies equally to Fortran output to units @code{*}
and @code{6}. @xref{Printing from Fortran}, which describes workarounds.

In the past most Fortran compilers implemented I/O on top of the C I/O
system and so the two interworked successfully.  This was true of
@command{g77}, but it is less true of @command{gfortran} as used in
@command{gcc} 4 and later.  In particular, any package that makes use of
Fortran I/O will when compiled on Windows interfere with C I/O: when the
Fortran I/O support code is initialized (typically when the package is
loaded) the C @code{stdout} and @code{stderr} are switched to LF line
endings.  (Function @code{init} in file
@file{src/modules/lapack/init_win.c} shows how to mitigate this.  In a
package this would look something like
@example
#ifdef _WIN32
# include <fcntl.h>
#endif

void R_init_mypkgname(DllInfo *dll)
@{
    // Native symbol registration calls

#ifdef _WIN32
    // gfortran I/O initialization sets these to _O_BINARY
    setmode(1, _O_TEXT); /* stdout */
    setmode(2, _O_TEXT); /* stderr */
#endif
@}
@end example
@noindent
in the file used for native symbol registration.)


@node Linking to other packages, Handling R objects in C, Fortran I/O, System and foreign language interfaces
@section Linking to other packages

It is not in general possible to link a DLL in package @pkg{packA} to a
DLL provided by package @pkg{packB} (for the security reasons mentioned
in @ref{dyn.load and dyn.unload}, and also because some platforms
distinguish between shared objects and dynamic libraries), but it is on
Windows.

Note that there can be tricky versioning issues here, as package
@pkg{packB} could be re-installed after package @pkg{packA} --- it is
desirable that the API provided by package @pkg{packB} remains
backwards-compatible.

Shipping a static library in package @pkg{packB} for other packages to
link to avoids most of the difficulties.

@menu
* Unix-alikes::
* Windows::
@end menu

@node Unix-alikes, Windows, Linking to other packages, Linking to other packages
@subsection Unix-alikes

It is possible to link a shared object in package @pkg{packA} to a
library provided by package @pkg{packB} under limited circumstances
on a Unix-alike OS.  There are severe portability issues, so this is not
recommended for a distributed package.

This is easiest if @pkg{packB} provides a static library
@file{packB/lib/libpackB.a}.  (Note using directory @file{lib} rather
than @file{libs} is conventional, and architecture-specific
sub-directories may be needed and are assumed in the sample code
below. The code in the static library will need to be compiled with
@code{PIC} flags on platforms where it matters.)  Then as the code from
package @pkg{packB} is incorporated when package @pkg{packA} is
installed, we only need to find the static library at install time for
package @pkg{packA}.  The only issue is to find package @pkg{packB}, and
for that we can ask @R{} by something like (long lines broken for
display here)

@example
PKGB_PATH=`echo 'library(packB);
  cat(system.file("lib",  package="packB", mustWork=TRUE))' \
 | "$@{R_HOME@}/bin/R" --vanilla --no-echo`
PKG_LIBS="$(PKGB_PATH)$(R_ARCH)/libpackB.a"
@end example

For a dynamic library @file{packB/lib/libpackB.so}
(@file{packB/lib/libpackB.dylib} on macOS: note that you cannot link to
a shared object, @file{.so}, on that platform) we could use

@example
PKGB_PATH=`echo 'library(packB);
  cat(system.file("lib", package="packB", mustWork=TRUE))' \
 | "$@{R_HOME@}/bin/R" --vanilla --no-echo`
PKG_LIBS=-L"$(PKGB_PATH)$(R_ARCH)" -lpackB
@end example

@noindent
This will work for installation, but very likely not when package
@code{packB} is loaded, as the path to package @pkg{packB}'s @file{lib}
directory is not in the @command{ld.so}@footnote{@command{dyld} on macOS,
and @env{DYLD_LIBRARY_PATHS} below.}  search path.  You can arrange to
put it there @strong{before} @R{} is launched by setting (on some
platforms) @env{LD_RUN_PATH} or @env{LD_LIBRARY_PATH} or adding to the
@command{ld.so} cache (see @command{man ldconfig}).  On platforms that
support it, the path to the directory containing the dynamic library can
be hardcoded at install time (which assumes that the location of package
@pkg{packB} will not be changed nor the package updated to a changed
API).  On systems with the @command{gcc} or @command{clang} and the
@acronym{GNU} linker (e.g.@: Linux) and some others this can be done by
e.g.@:

@example
PKGB_PATH=`echo 'library(packB);
  cat(system.file("lib", package="packB", mustWork=TRUE)))' \
 | "$@{R_HOME@}/bin/R" --vanilla --no-echo`
PKG_LIBS=-L"$(PKGB_PATH)$(R_ARCH)" -Wl,-rpath,"$(PKGB_PATH)$(R_ARCH)" -lpackB
@end example

@noindent
Some other systems (e.g.@: Solaris with its native linker) use
@option{-Rdir} rather than @option{-rpath,dir} (and this is accepted by
the compiler as well as the linker).

It may be possible to figure out what is required semi-automatically
from the result of @command{R CMD libtool --config} (look for
@samp{hardcode}).

Making headers provided by package @pkg{packB} available to the code to
be compiled in package @pkg{packA} can be done by the @code{LinkingTo}
mechanism (@pxref{Registering native routines}).


@node Windows,  , Unix-alikes, Linking to other packages
@subsection Windows

Suppose package @pkg{packA} wants to make use of compiled code provided
by @pkg{packB} in DLL @file{packB/libs/exB.dll}, possibly the package's
DLL @file{packB/libs/packB.dll}.  (This can be extended to linking to
more than one package in a similar way.)  There are three issues to be
addressed:

@itemize

@item
Making headers provided by package @pkg{packB} available to the code to
be compiled in package @pkg{packA}.

This is done by the @code{LinkingTo} mechanism (@pxref{Registering native
routines}).

@item preparing @code{packA.dll} to link to @file{packB/libs/exB.dll}.

This needs an entry in @file{Makevars.win} or @file{Makevars.ucrt} of the form

@example
PKG_LIBS= -L<something> -lexB
@end example
@noindent
and one possibility is that @code{<something>} is the path to the
installed @file{pkgB/libs} directory.  To find that we need to ask @R{}
where it is by something like

@example
PKGB_PATH=`echo 'library(packB);
  cat(system.file("libs", package="packB", mustWork=TRUE))' \
 | rterm --vanilla --no-echo`
PKG_LIBS= -L"$(PKGB_PATH)$(R_ARCH)" -lexB
@end example

Another possibility is to use an import library, shipping with package
@pkg{packA} an exports file @file{exB.def}.  Then @file{Makevars.win} (or
@file{Makevars.ucrt})
could contain

@example
PKG_LIBS= -L. -lexB

all: $(SHLIB) before

before: libexB.dll.a
libexB.dll.a: exB.def
@end example
@noindent
and then installing package @pkg{packA} will make and use the import
library for @file{exB.dll}.  (One way to prepare the exports file is to
use @file{pexports.exe}.)

@item loading @file{packA.dll} which depends on @file{exB.dll}.

If @code{exB.dll} was used by package @pkg{packB} (because it is in fact
@file{packB.dll} or @file{packB.dll} depends on it) and @pkg{packB} has
been loaded before @pkg{packA}, then nothing more needs to be done as
@file{exB.dll} will already be loaded into the @R{} executable.  (This
is the most common scenario.)

More generally, we can use the @code{DLLpath} argument to
@code{library.dynam} to ensure that @code{exB.dll} is found, for example
by setting

@example
library.dynam("packA", pkg, lib,
              DLLpath = system.file("libs", package="packB"))
@end example

Note that @code{DLLpath} can only set one path, and so for linking to
two or more packages you would need to resort to setting environment
variable @env{PATH}.

@end itemize

@node Handling R objects in C, Interface functions .Call and .External, Linking to other packages, System and foreign language interfaces
@section Handling R objects in C
@cindex Handling R objects in C

Using C code to speed up the execution of an @R{} function is often very
fruitful.  Traditionally this has been done @emph{via} the @code{.C}
function in @R{}.  However, if a user wants to write C code using
internal @R{} data structures, then that can be done using the
@code{.Call} and @code{.External} functions.  The syntax for the calling
function in @R{} in each case is similar to that of @code{.C}, but the
two functions have different C interfaces.  Generally the @code{.Call}
interface is simpler to use, but @code{.External} is a little more
general.
@findex .Call
@findex .External

A call to @code{.Call} is very similar to @code{.C}, for example

@example
.Call("convolve2", a, b)
@end example

@noindent
The first argument should be a character string giving a C symbol name
of code that has already been loaded into @R{}.  Up to 65 @R{} objects
can passed as arguments.  The C side of the interface is

@example
@group
#include <R.h>
#include <Rinternals.h>

SEXP convolve2(SEXP a, SEXP b)
 ...
@end group
@end example

A call to @code{.External} is almost identical

@example
.External("convolveE", a, b)
@end example

@noindent
but the C side of the interface is different, having only one argument

@example
@group
#include <R.h>
#include <Rinternals.h>

SEXP convolveE(SEXP args)
 ...
@end group
@end example

@noindent
Here @code{args} is a @code{LISTSXP}, a Lisp-style pairlist from which
the arguments can be extracted.

In each case the @R{} objects are available for manipulation @emph{via}
a set of functions and macros defined in the header file
@file{Rinternals.h} or some @Sl{}-compatibility macros@footnote{That is,
similar to those defined in @Sl{} version 4 from the 1990s: these are
not kept up to date and are not recommended for new projects.}  See
@ref{Interface functions .Call and .External} for details on
@code{.Call} and @code{.External}.

Before you decide to use @code{.Call} or @code{.External}, you should
look at other alternatives.  First, consider working in interpreted @R{}
code; if this is fast enough, this is normally the best option.  You
should also see if using @code{.C} is enough.  If the task to be
performed in C is simple enough involving only atomic vectors and
requiring no call to @R{}, @code{.C} suffices. A great deal of useful
code was written using just @code{.C} before @code{.Call} and
@code{.External} were available.  These interfaces allow much more
control, but they also impose much greater responsibilities so need to
be used with care.  Neither @code{.Call} nor @code{.External} copy their
arguments: you should treat arguments you receive through these
interfaces as read-only.

To handle @R{} objects from within C code we use the macros and functions
that have been used to implement the core parts of @R{}.  A
public@footnote{ @pxref{The R API}: note that these are not all part of
the API.}  subset of these is defined in the header file
@file{Rinternals.h} in the directory @file{@var{R_INCLUDE_DIR}} (default
@file{@var{R_HOME}/include}) that should be available on any @R{}
installation.

A substantial amount of @R{}, including the standard packages, is
implemented using the functions and macros described here, so the @R{}
source code provides a rich source of examples and ``how to do it'': do
make use of the source code for inspirational examples.

It is necessary to know something about how @R{} objects are handled in
C code.  All the @R{} objects you will deal with will be handled with
the type @dfn{SEXP}@footnote{SEXP is an acronym for @emph{S}imple
@emph{EXP}ression, common in LISP-like language syntaxes.}, which is a
pointer to a structure with typedef @code{SEXPREC}.  Think of this
structure as a @emph{variant type} that can handle all the usual types
of @R{} objects, that is vectors of various modes, functions,
environments, language objects and so on.  The details are given later
in this section and in @ref{R Internal Structures, , R Internal
Structures, R-ints, R Internals}, but for most
purposes the programmer does not need to know them.  Think rather of a
model such as that used by Visual Basic, in which @R{} objects are
handed around in C code (as they are in interpreted @R{} code) as the
variant type, and the appropriate part is extracted for, for example,
numerical calculations, only when it is needed.  As in interpreted @R{}
code, much use is made of coercion to force the variant object to the
right type.

@menu
* Garbage Collection::
* Allocating storage::
* Details of R types::
* Attributes::
* Classes::
* Handling lists::
* Handling character data::
* Finding and setting variables::
* Some convenience functions::
* Named objects and copying::
@end menu

@node Garbage Collection, Allocating storage, Handling R objects in C, Handling R objects in C
@subsection Handling the effects of garbage collection
@cindex Garbage collection

@findex PROTECT
@findex UNPROTECT
@findex protect
@findex unprotect

We need to know a little about the way @R{} handles memory allocation.
The memory allocated for @R{} objects is not freed by the user; instead,
the memory is from time to time @dfn{garbage collected}.  That is, some
or all of the allocated memory not being used is freed or marked as
re-usable.

The @R{} object types are represented by a C structure defined by a
typedef @code{SEXPREC} in @file{Rinternals.h}.  It contains several
things among which are pointers to data blocks and to other
@code{SEXPREC}s.  A @code{SEXP} is simply a pointer to a @code{SEXPREC}.

If you create an @R{} object in your C code, you must tell @R{} that you
are using the object by using the @code{PROTECT} macro on a pointer to
the object. This tells @R{} that the object is in use so it is not
destroyed during garbage collection.  Notice that it is the object which
is protected, not the pointer variable.  It is a common mistake to
believe that if you invoked @code{PROTECT(@var{p})} at some point then
@var{p} is protected from then on, but that is not true once a new
object is assigned to @var{p}.

Protecting an @R{} object automatically protects all the @R{} objects
pointed to in the corresponding @code{SEXPREC}, for example all elements
of a protected list are automatically protected.

The programmer is solely responsible for housekeeping the calls to
@code{PROTECT}.  There is a corresponding macro @code{UNPROTECT} that
takes as argument an @code{int} giving the number of objects to
unprotect when they are no longer needed.  The protection mechanism is
stack-based, so @code{UNPROTECT(@var{n})} unprotects the last @var{n}
objects which were protected.  The calls to @code{PROTECT} and
@code{UNPROTECT} must balance when the user's code returns and should
balance in all functions.  @R{} will warn about
@code{"stack imbalance in .Call"} (or @code{.External}) if the 
housekeeping is wrong. 

Here is a small example of creating an @R{} numeric vector in C code:

@example
@group
#include <R.h>
#include <Rinternals.h>

    SEXP ab;
      ....
    ab = PROTECT(allocVector(REALSXP, 2));
    REAL(ab)[0] = 123.45;
    REAL(ab)[1] = 67.89;
    UNPROTECT(1);
@end group
@end example

Now, the reader may ask how the @R{} object could possibly get removed
during those manipulations, as it is just our C code that is running.
As it happens, we can do without the protection in this example, but in
general we do not know (nor want to know) what is hiding behind the @R{}
macros and functions we use, and any of them might cause memory to be
allocated, hence garbage collection and hence our object @code{ab} to be
removed. It is usually wise to err on the side of caution and assume
that any of the @R{} macros and functions might remove the object.

In some cases it is necessary to keep better track of whether protection
is really needed.  Be particularly aware of situations where a large
number of objects are generated.  The pointer protection stack has a
fixed size (default 10,000) and can become full.  It is not a good idea
then to just @code{PROTECT} everything in sight and @code{UNPROTECT}
several thousand objects at the end. It will almost invariably be
possible to either assign the objects as part of another object (which
automatically protects them) or unprotect them immediately after use.

There is a less-used macro @code{UNPROTECT_PTR(@var{s})} that unprotects the
object pointed to by the @code{SEXP} @var{s}, even if it is not the top item
on the pointer protection stack.  This macro was introduced for use in the
parser, where the code interfacing with the R heap is generated and the
generator cannot be configured to insert proper calls to @code{PROTECT} and
@code{UNPROTECT}.  However, @code{UNPROTECT_PTR} is dangerous to use in
combination with @code{UNPROTECT} when the same object has been protected
multiple times.  It has been superseded by multi-set based functions
@code{R_PreserveInMSet} and @code{R_ReleaseFromMSet}, which protect objects
in a multi-set created by @code{R_NewPreciousMSet} and typically itself
protected using @code{PROTECT}.  These functions should not be needed
outside parsers.
@findex UNPROTECT_PTR
@findex unprotect_ptr
@findex R_PreserveInMSet
@findex R_ReleaseFromMSet
@findex R_NewPreciousMSet

Sometimes an object is changed (for example duplicated, coerced or
grown) yet the current value needs to be protected.  For these cases
@code{PROTECT_WITH_INDEX} saves an index of the protection location that
can be used to replace the protected value using @code{REPROTECT}.
@findex PROTECT_WITH_INDEX
@findex REPROTECT
@findex R_ProtectWithIndex
@findex R_Reprotect
For example (from the internal code for @code{optim})

@example
    PROTECT_INDEX ipx;

    ....
    PROTECT_WITH_INDEX(s = eval(OS->R_fcall, OS->R_env), &ipx);
    REPROTECT(s = coerceVector(s, REALSXP), ipx);
@end example

Note that it is dangerous to mix @code{UNPROTECT_PTR} also with
@code{PROTECT_WITH_INDEX}, as the former changes the protection
locations of objects that were protected after the one being
unprotected.

@findex R_PreserveObject
@findex R_ReleaseObject
There is another way to avoid the effects of garbage collection: a call
to @code{R_PreserveObject} adds an object to an internal list of objects
not to be collects, and a subsequent call to @code{R_ReleaseObject}
removes it from that list.  This provides a way for objects which are
not returned as part of @R{} objects to be protected across calls to
compiled code: on the other hand it becomes the user's responsibility to
release them when they are no longer needed (and this often requires the
use of a finalizer).  It is less efficient than the normal protection
mechanism, and should be used sparingly.

For functions from packages as well as @R{} to safely co-operate in
protecting objects, certain rules have to be followed:

@itemize

@item 
Pointer-protection balance.  Calls to @code{PROTECT} and @code{UNPROTECT}
should balance in each function.  A function may only call @code{UNPROTECT} or
@code{REPROTECT} on objects it has itself protected.  Note that the pointer
protection stack balance is restored automatically on non-local transfer of
control (See @ref{Condition handling and cleanup code}.), as if a call to
@code{UNPROTECT} was invoked with the right argument.

@item
Caller protection.  It is the responsibility of the caller that all
arguments passed to a function are protected and will stay protected for the
whole execution of the callee.  Typically this is achieved by @code{PROTECT}
and @code{UNPROTECT} calls.

@item
Protecting return values. Any @R{} objects returned from a function are
unprotected (the callee must maintain pointer-protection balance), and hence
should be protected immediately by the caller. To be safe against future
code changes, assume that any @R{} object returned from any function may
need protection. Note that even when conceptually returning an existing
protected object, that object may be duplicated.

@item
All functions/macros allocate.  To be safe against future code changes,
assume that any function or macro may allocate and hence garbage collector
may run and destroy unprotected objects.

@end itemize

It is always safe and recommended to follow those rules.  In fact, several
@R{} functions and macros protect their own arguments and some functions do
not allocate or do not allocate when used in a certain way, but that is
subject to change, so relying on that may be fragile.  @code{PROTECT} and
@code{PROTECT_WITH_INDEX} can be safely called with unprotected arguments
and @code{UNPROTECT} does not allocate.

@node Allocating storage, Details of R types, Garbage Collection, Handling R objects in C
@subsection Allocating storage
@cindex Allocating storage

For many purposes it is sufficient to allocate @R{} objects and
manipulate those.  There are quite a few @code{alloc@var{Xxx}} functions
defined in @file{Rinternals.h}---you may want to explore them.

@findex allocVector
One that is commonly used is @code{allocVector}, the C-level equivalent
of @R{}-level @code{vector()} and its wrappers such as @code{integer()}
and @code{character()}.  One distinction is that whereas the @R{}
functions always initialize the elements of the vector,
@code{allocVector} only does so for lists, expressions and character
vectors (the cases where the elements are themselves @R{} objects).

If storage is required for C objects during the calculations this is
best allocating by calling @code{R_alloc}; @pxref{Memory allocation}.
All of these memory allocation routines do their own error-checking, so
the programmer may assume that they will raise an error and not return
if the memory cannot be allocated.

@node Details of R types, Attributes, Allocating storage, Handling R objects in C
@subsection Details of R types
@cindex Details of R types

Users of the @file{Rinternals.h} macros will need to know how the @R{}
types are known internally.  The different @R{} data types are
represented in C by @dfn{SEXPTYPE}.  Some of these are familiar from
@R{} and some are internal data types.  The usual @R{} object modes are
given in the table.

@quotation
@multitable {SEXPTYPE} {numeric with storage mode integer integer}
@headitem SEXPTYPE @tab @R{} equivalent
@item @code{REALSXP}  @tab numeric with storage mode @code{double}
@item @code{INTSXP}   @tab integer
@item @code{CPLXSXP}  @tab complex
@item @code{LGLSXP}   @tab logical
@item @code{STRSXP}   @tab character
@item @code{VECSXP}   @tab list (generic vector)
@item @code{LISTSXP}  @tab pairlist
@item @code{DOTSXP}   @tab a @samp{@dots{}} object
@item @code{NILSXP}   @tab NULL
@item @code{SYMSXP}   @tab name/symbol
@item @code{CLOSXP}   @tab function or function closure
@item @code{ENVSXP}   @tab environment
@end multitable
@end quotation

@noindent
Among the important internal @code{SEXPTYPE}s are @code{LANGSXP},
@code{CHARSXP}, @code{PROMSXP}, etc.  (@strong{N.B.}: although it is
possible to return objects of internal types, it is unsafe to do so as
assumptions are made about how they are handled which may be violated at
user-level evaluation.)  More details are given in @ref{R Internal
Structures, , R Internal Structures, R-ints, R Internals}.

Unless you are very sure about the type of the arguments, the code
should check the data types.  Sometimes it may also be necessary to
check data types of objects created by evaluating an @R{} expression in
the C code.  You can use functions like @code{isReal}, @code{isInteger}
and @code{isString} to do type checking.
@findex isReal
@findex isInteger
@findex isString
Other such functions declared in the header file @file{Rinternals.h}
include @code{isNull}, @code{isSymbol}, @code{isLogical},
@code{isComplex}, @code{isExpression}, and @code{isEnvironment}.
@findex isNull
@findex isSymbol
@findex isLogical
@findex isComplex
@findex isExpression
@findex isEnvironment
All of these take a @code{SEXP} as argument and return 1 or 0 to
indicate @var{TRUE} or @var{FALSE}.

What happens if the @code{SEXP} is not of the correct type?  Sometimes
you have no other option except to generate an error.  You can use the
function @code{error} for this.  It is usually better to coerce the
object to the correct type.  For example, if you find that an
@code{SEXP} is of the type @code{INTEGER}, but you need a @code{REAL}
object, you can change the type by using

@example
@var{newSexp} = PROTECT(coerceVector(@var{oldSexp}, REALSXP));
@end example
@findex coerceVector

@noindent
Protection is needed as a new @emph{object} is created; the object
formerly pointed to by the @code{SEXP} is still protected but now
unused.@footnote{If no coercion was required, @code{coerceVector} would
have passed the old object through unchanged.}

All the coercion functions do their own error-checking, and generate
@code{NA}s with a warning or stop with an error as appropriate.

Note that these coercion functions are @emph{not} the same as calling
@code{as.numeric} (and so on) in @R{} code, as they do not dispatch on
the class of the object.  Thus it is normally preferable to do the
coercion in the calling @R{} code.

So far we have only seen how to create and coerce @R{} objects from C
code, and how to extract the numeric data from numeric @R{} vectors.
These can suffice to take us a long way in interfacing @R{} objects to
numerical algorithms, but we may need to know a little more to create
useful return objects.

@node Attributes, Classes, Details of R types, Handling R objects in C
@subsection Attributes
@cindex Attributes

Many @R{} objects have attributes: some of the most useful are classes
and the @code{dim} and @code{dimnames} that mark objects as matrices or
arrays.  It can also be helpful to work with the @code{names} attribute
of vectors.

To illustrate this, let us write code to take the outer product of two
vectors (which @code{outer} and @code{%o%} already do).  As usual the
@R{} code is simple

@example
out <- function(x, y)
@{
    storage.mode(x) <- storage.mode(y) <- "double"
    .Call("out", x, y)
@}
@end example

@noindent
where we expect @code{x} and @code{y} to be numeric vectors (possibly
integer), possibly with names.  This time we do the coercion in the
calling @R{} code.

C code to do the computations is

@example
@group
#include <R.h>
#include <Rinternals.h>

SEXP out(SEXP x, SEXP y)
@{
    int nx = length(x), ny = length(y);
    SEXP ans = PROTECT(allocMatrix(REALSXP, nx, ny));
    double *rx = REAL(x), *ry = REAL(y), *rans = REAL(ans);
    for(int i = 0; i < nx; i++) @{
        double tmp = rx[i];
        for(int j = 0; j < ny; j++)
            rans[i + nx*j] = tmp * ry[j];
    @}
    UNPROTECT(1);
    return ans;
@}
@end group
@end example

@noindent
Note the way @code{REAL} is used: as it is a function call it can be
considerably faster to store the result and index that.

However, we would like to set the @code{dimnames} of the result.  We can use

@example
#include <R.h>
#include <Rinternals.h>

@group
SEXP out(SEXP x, SEXP y)
@{
    int nx = length(x), ny = length(y);
    SEXP ans = PROTECT(allocMatrix(REALSXP, nx, ny));
    double *rx = REAL(x), *ry = REAL(y), *rans = REAL(ans);

    for(int i = 0; i < nx; i++) @{
      double tmp = rx[i];
      for(int j = 0; j < ny; j++)
        rans[i + nx*j] = tmp * ry[j];
    @}

    SEXP dimnames = PROTECT(allocVector(VECSXP, 2));
    SET_VECTOR_ELT(dimnames, 0, getAttrib(x, R_NamesSymbol));
    SET_VECTOR_ELT(dimnames, 1, getAttrib(y, R_NamesSymbol));
    setAttrib(ans, R_DimNamesSymbol, dimnames);
@end group

@group
    UNPROTECT(2);
    return ans;
@}
@end group
@end example

This example introduces several new features.  The @code{getAttrib} and
@code{setAttrib}
@findex getAttrib
@findex setAttrib
functions get and set individual attributes.  Their second argument is a
@code{SEXP} defining the name in the symbol table of the attribute we
want; these and many such symbols are defined in the header file
@file{Rinternals.h}.

There are shortcuts here too: the functions @code{namesgets},
@code{dimgets} and @code{dimnamesgets} are the internal versions of the
default methods of @code{names<-}, @code{dim<-} and @code{dimnames<-}
(for vectors and arrays), and there are functions such as
@code{GetColNames}, @code{GetRowNames}, @code{GetMatrixDimnames} and
@code{GetArrayDimnames}.
@findex GetArrayDimnames
@findex GetMatrixDimnames
@findex GetColNames
@findex GetRowNames
@findex namesgets
@findex dimnamesgets
@findex dimgets

What happens if we want to add an attribute that is not pre-defined? We
need to add a symbol for it @emph{via} a call to
@findex install
@code{install}.  Suppose for illustration we wanted to add an attribute
@code{"version"} with value @code{3.0}.  We could use

@example
@group
    SEXP version;
    version = PROTECT(allocVector(REALSXP, 1));
    REAL(version)[0] = 3.0;
    setAttrib(ans, install("version"), version);
    UNPROTECT(1);
@end group
@end example

Using @code{install} when it is not needed is harmless and provides a
simple way to retrieve the symbol from the symbol table if it is already
installed. However, the lookup takes a non-trivial amount of time, so
consider code such as

@example
static SEXP VerSymbol = NULL;
...
    if (VerSymbol == NULL) VerSymbol = install("version");
@end example

@noindent
if it is to be done frequently.

This example can be simplified by another convenience function:

@example
@group
    SEXP version = PROTECT(ScalarReal(3.0));
    setAttrib(ans, install("version"), version);
    UNPROTECT(1);
@end group
@end example


@node Classes, Handling lists, Attributes, Handling R objects in C
@subsection Classes
@cindex Classes

In @R{} the class is just the attribute named @code{"class"} so it can
be handled as such, but there is a shortcut @code{classgets}.  Suppose
we want to give the return value in our example the class @code{"mat"}.
We can use

@example
@group
#include <R.h>
#include <Rinternals.h>
      ....
    SEXP ans, dim, dimnames, class;
      ....
    class = PROTECT(allocVector(STRSXP, 1));
    SET_STRING_ELT(class, 0, mkChar("mat"));
    classgets(ans, class);
    UNPROTECT(4);
    return ans;
@}
@end group
@end example
@findex classgets

@noindent
As the value is a character vector, we have to know how to create that
from a C character array, which we do using the function
@code{mkChar}.

@node Handling lists, Handling character data, Classes, Handling R objects in C
@subsection Handling lists
@cindex Handling lists

Some care is needed with lists, as @R{} moved early on from using
LISP-like lists (now called ``pairlists'') to S-like generic vectors.
As a result, the appropriate test for an object of mode @code{list} is
@code{isNewList}, and we need @code{allocVector(VECSXP, @var{n}}) and
@emph{not} @code{allocList(@var{n})}.

List elements can be retrieved or set by direct access to the elements
of the generic vector.  Suppose we have a list object

@example
a <- list(f = 1, g = 2, h = 3)
@end example

@noindent
Then we can access @code{a$g} as @code{a[[2]]} by

@example
@group
    double g;
      ....
    g = REAL(VECTOR_ELT(a, 1))[0];
@end group
@end example

This can rapidly become tedious, and the following function (based on
one in package @pkg{stats}) is very useful:

@example
@group
/* get the list element named str, or return NULL */

SEXP getListElement(SEXP list, const char *str)
@{
    SEXP elmt = R_NilValue, names = getAttrib(list, R_NamesSymbol);
@end group

@group
    for (int i = 0; i < length(list); i++)
        if(strcmp(CHAR(STRING_ELT(names, i)), str) == 0) @{
           elmt = VECTOR_ELT(list, i);
           break;
        @}
    return elmt;
@}
@end group
@end example

@noindent
and enables us to say

@example
@group
  double g;
  g = REAL(getListElement(a, "g"))[0];
@end group
@end example

@node Handling character data, Finding and setting variables, Handling lists, Handling R objects in C
@subsection Handling character data
@cindex handling character data

R character vectors are stored as @code{STRSXP}s, a vector type like
@code{VECSXP} where every element is of type @code{CHARSXP}.  The
@code{CHARSXP} elements of @code{STRSXP}s are accessed using
@code{STRING_ELT} and @code{SET_STRING_ELT}.
@findex STRING_ELT
@findex SET_STRING_ELT

@code{CHARSXP}s are read-only objects and must never be modified.  In
particular, the C-style string contained in a @code{CHARSXP} should be
treated as read-only and for this reason the @code{CHAR} function used
to access the character data of a @code{CHARSXP} returns @code{(const
char *)} (this also allows compilers to issue warnings about improper
use).  Since @code{CHARSXP}s are immutable, the same @code{CHARSXP} can
be shared by any @code{STRSXP} needing an element representing the same
string. @R{} maintains a global cache of @code{CHARSXP}s so that there
is only ever one @code{CHARSXP} representing a given string in memory.

@findex mkChar
@findex mkCharLen
You can obtain a @code{CHARSXP} by calling @code{mkChar} and providing a
nul-terminated C-style string.  This function will return a pre-existing
@code{CHARSXP} if one with a matching string already exists, otherwise
it will create a new one and add it to the cache before returning it to
you.   The variant @code{mkCharLen} can be used to create a
@code{CHARSXP} from part of a buffer and will ensure null-termination.

Note that @R{} character strings are restricted to @code{2^31 - 1}
bytes, and hence so should the input to @code{mkChar} be (C allows
longer strings on 64-bit platforms).

@node Finding and setting variables, Some convenience functions, Handling character data, Handling R objects in C
@subsection Finding and setting variables
@cindex Finding variables
@cindex Setting variables

It will be usual that all the @R{} objects needed in our C computations
are passed as arguments to @code{.Call} or @code{.External}, but it is
possible to find the values of @R{} objects from within the C given
their names.  The following code is the equivalent of @code{get(name,
envir = rho)}.

@example
@group
SEXP getvar(SEXP name, SEXP rho)
@{
    SEXP ans;

    if(!isString(name) || length(name) != 1)
        error("name is not a single string");
    if(!isEnvironment(rho))
        error("rho should be an environment");
    ans = findVar(installChar(STRING_ELT(name, 0)), rho);
    Rprintf("first value is %f\n", REAL(ans)[0]);
    return R_NilValue;
@}
@end group
@end example
@findex installChar

The main work is done by
@findex findVar
@code{findVar}, but to use it we need to install @code{name} as a name
in the symbol table.  As we wanted the value for internal use, we return
@code{NULL}.

Similar functions with syntax

@example
@group
void defineVar(SEXP symbol, SEXP value, SEXP rho)
void setVar(SEXP symbol, SEXP value, SEXP rho)
@end group
@end example
@findex defineVar
@findex setVar

@noindent
can be used to assign values to @R{} variables.  @code{defineVar}
creates a new binding or changes the value of an existing binding in the
specified environment frame; it is the analogue of @code{assign(symbol,
value, envir = rho, inherits = FALSE)}, but unlike @code{assign},
@code{defineVar} does not make a copy of the object
@code{value}.@footnote{You can assign a @emph{copy} of the object in the
environment frame @code{rho} using @code{defineVar(symbol,
duplicate(value), rho)}).}  @code{setVar} searches for an existing
binding for @code{symbol} in @code{rho} or its enclosing environments.
If a binding is found, its value is changed to @code{value}.  Otherwise,
a new binding with the specified value is created in the global
environment.  This corresponds to @code{assign(symbol, value, envir =
rho, inherits = TRUE)}.

At times it may also be useful to create a new environment frame in C code.
@code{R_NewEnv} is a C version of the @R{} function @code{new.env}:

@example
@group
SEXP R_NewEnv(SEXP enclos, int hash, ins size)
@end group
@end example
@findex R_NewEnv

@node Some convenience functions, Named objects and copying, Finding and setting variables, Handling R objects in C
@subsection Some convenience functions

Some operations are done so frequently that there are convenience
functions to handle them. (All these are provided @emph{via} the header
file @file{Rinternals.h}.)

Suppose we wanted to pass a single logical argument
@code{ignore_quotes}: we could use

@example
    int ign = asLogical(ignore_quotes);
    if(ign == NA_LOGICAL) error("'ignore_quotes' must be TRUE or FALSE");
@end example

@noindent
which will do any coercion needed (at least from a vector argument), and
return @code{NA_LOGICAL} if the value passed was @code{NA} or coercion
failed.  There are also @code{asInteger}, @code{asReal} and
@code{asComplex}.  The function @code{asChar} returns a @code{CHARSXP}.
All of these functions ignore any elements of an input vector after the
first.
@findex asInteger
@findex asLogical
@findex asReal
@findex asComplex
@findex asChar

To return a length-one real vector we can use

@example
    double x;

    ...
    return ScalarReal(x);
@end example

@noindent
and there are versions of this for all the atomic vector types (those for
a length-one character vector being @code{ScalarString} with argument a
@code{CHARSXP} and @code{mkString} with argument @code{const char *}).

@findex ScalarReal
@findex ScalarInteger
@findex ScalarLogical
@findex ScalarComplex
@findex ScalarRaw
@findex ScalarString
@findex mkString
@example
@group
SEXP ScalarReal(double);
SEXP ScalarInteger(int);
SEXP ScalarLogical(int)
SEXP ScalarRaw(Rbyte);
SEXP ScalarComplex(Rcomplex);
SEXP ScalarString(SEXP);
SEXP mkString(const char *);
@end group
@end example

Some of the @code{is@var{XXXX}} functions differ from their apparent
@R{}-level counterparts: for example @code{isVector} is true for any
atomic vector type (@code{isVectorAtomic}) and for lists and expressions
(@code{isVectorList}) (with no check on attributes).  @code{isMatrix} is
a test of a length-2 @code{"dim"} attribute.
@findex isVector
@findex isVectorAtomic
@findex isVectorList
@findex isMatrix
@findex isPairList
@findex isPrimitive
@findex isTs
@findex isNumeric
@findex isArray
@findex isFactor
@findex isObject
@findex isFunction
@findex isLanguage
@findex isNewList
@findex isList
@findex isOrdered
@findex isUnordered
@comment @findex isNumber
@comment @findex isFrame
@example
@group
Rboolean isVector(SEXP);
Rboolean isVectorAtomic(SEXP);
Rboolean isVectorList(SEXP);
Rboolean isMatrix(SEXP);
Rboolean isPairList(SEXP);
Rboolean isPrimitive(SEXP);
Rboolean isTs(SEXP);
Rboolean isNumeric(SEXP);
Rboolean isArray(SEXP);
Rboolean isFactor(SEXP);
Rboolean isObject(SEXP);
Rboolean isFunction(SEXP);
Rboolean isLanguage(SEXP);
Rboolean isNewList(SEXP);
Rboolean isList(SEXP);
Rboolean isOrdered(SEXP);
Rboolean isUnordered(SEXP);
@comment Rboolean isNumber(SEXP);
@comment Rboolean isFrame (SEXP);
@end group
@end example

There are a series of small macros/functions to help construct pairlists
and language objects (whose internal structures just differ by
@code{SEXPTYPE}).  Function @code{CONS(u, v)} is the basic building
block: it constructs a pairlist from @code{u} followed by @code{v}
(which is a pairlist or @code{R_NilValue}).  @code{LCONS} is a variant
that constructs a language object.  Functions @code{list1} to
@code{list6} construct a pairlist from one to six items, and
@code{lang1} to @code{lang6} do the same for a language object (a
function to call plus zero to five arguments).
@findex CONS
@findex cons
@findex list1
@findex list2
@findex list3
@findex list4
@findex list5
@findex list6
@findex LCONS
@findex lcons
@findex lang1
@findex lang2
@findex lang3
@findex lang4
@findex lang5
@findex lang6
Functions @code{elt} and @code{lastElt} find the @var{i}th element and
the last element of a pairlist, and @code{nthcdr} returns a pointer to
the @var{n}th position in the pairlist (whose @code{CAR} is the
@var{n}th item).
@findex elt
@findex lastElt
@findex nthcdr

Functions @code{str2type} and @code{type2str} map @R{}
length-one character strings to and from @code{SEXPTYPE} numbers, and
@code{type2char} maps numbers to C character strings.
@findex str2type
@findex type2str
@findex type2char

@comment Want to encourage use of some of the more stable and useful R_*
@comment and Rf_* functions:
@menu
* Semi-internal convenience functions::
@end menu

@node Semi-internal convenience functions,  , Some convenience functions, Some convenience functions
@subsubsection Semi-internal convenience functions

There is quite a collection of functions that may be used in your C code
@emph{if} you are willing to adapt to rare ``API'' changes.
These typically contain ``workhorses'' of their @R{} counterparts.

Functions @code{any_duplicated} and @code{any_duplicated3} are fast
versions of @R{}'s @code{any(duplicated(.))}.
@findex any_duplicated
@findex any_duplicated3

Function @code{R_compute_identical} corresponds to @R{}'s @code{identical} function.
@findex R_compute_identical

@node Named objects and copying,  , Some convenience functions, Handling R objects in C
@subsection Named objects and copying
@findex duplicate
@cindex Copying objects

[The @code{NAMED} mechanism has been replaced by reference counting.]

When assignments are done in @R{} such as

@example
x <- 1:10
y <- x
@end example

@noindent
the named object is not necessarily copied, so after those two
assignments @code{y} and @code{x} are bound to the same @code{SEXPREC}
(the structure a @code{SEXP} points to).  This means that any code which
alters one of them has to make a copy before modifying the copy if the
usual @R{} semantics are to apply.  Note that whereas @code{.C} and
@code{.Fortran} do copy their arguments (unless the dangerous @code{dup
= FALSE} is used), @code{.Call} and @code{.External} do not.  So
@code{duplicate} is commonly called on arguments to @code{.Call} before
modifying them.

However, at least some of this copying is unneeded.  In the first
assignment shown, @code{x <- 1:10}, @R{} first creates an object with
value @code{1:10} and then assigns it to @code{x} but if @code{x} is
modified no copy is necessary as the temporary object with value
@code{1:10} cannot be referred to again.  @R{} distinguishes between
named and unnamed objects @emph{via} a field in a @code{SEXPREC} that
can be accessed @emph{via} the macros @code{NAMED} and @code{SET_NAMED}.  This
can take values

@table @code
@item 0
The object is not bound to any symbol
@item 1
The object has been bound to exactly one symbol
@item >= 2
The object has potentially been bound to two or more symbols, and one
should act as if another variable is currently bound to this value.
The maximal value is @code{NAMEDMAX}.
@end table

@noindent
Note the past tenses: @R{} does not do currently do full reference
counting and there may currently be fewer bindings.

It is safe to modify the value of any @code{SEXP} for which
@code{NAMED(foo)} is zero, and if @code{NAMED(foo)} is two or more, the
value should be duplicated (@emph{via} a call to @code{duplicate})
before any modification.  Note that it is the responsibility of the
author of the code making the modification to do the duplication, even
if it is @code{x} whose value is being modified after @code{y <- x}.

The case @code{NAMED(foo) == 1} allows some optimization, but it can be
ignored (and duplication done whenever @code{NAMED(foo) > 0}).  (This
optimization is not currently usable in user code.)  It is intended
for use within replacement functions.  Suppose we used

@example
x <- 1:10
foo(x) <- 3
@end example

@noindent
which is computed as

@example
x <- 1:10
x <- "foo<-"(x, 3)
@end example

@noindent
Then inside @code{"foo<-"} the object pointing to the current value of
@code{x} will have @code{NAMED(foo)} as one, and it would be safe to
modify it as the only symbol bound to it is @code{x} and that will be
rebound immediately.  (Provided the remaining code in @code{"foo<-"}
make no reference to @code{x}, and no one is going to attempt a direct
call such as @code{y <- "foo<-"(x)}.)

This mechanism was replaced in @R{} 4.0.0. To
support future changes, package code should use the macros
@code{MAYBE_REFERENCED}, @code{MAYBE_SHARED}, and
@code{MARK_NOT_MUTABLE}. These currently correspond to
@findex MAYBE_REFERENCED
@findex MAYBE_SHARED
@findex MARK_NOT_MUTABLE

@table@code
@item MAYBE_REFERENCED(x)
@code{NAMED(x) > 0}
@item MAYBE_SHARED(x)
@code{NAMED(x) > 1}
@item MARK_NOT_MUTABLE(x)
@code{SET_NAMED(x, NAMEDMAX)}
@end table

@c commented out as people misread this as general.
@c Currently all arguments to a @code{.Call} call will have @code{NAMED}
@c set to 2 or higher and so users must assume that they need to be duplicated
@c before alteration.

@node Interface functions .Call and .External, Evaluating R expressions from C, Handling R objects in C, System and foreign language interfaces
@section Interface functions @code{.Call} and @code{.External}
@cindex Interfaces to compiled code

In this section we consider the details of the @R{}/C interfaces.

These two interfaces have almost the same functionality. @code{.Call} is
based on the interface of the same name in @Sl{} version 4, and
@code{.External} is based on @R{}'s @code{.Internal}.  @code{.External}
is more complex but allows a variable number of arguments.

@menu
* Calling .Call::
* Calling .External::
* Missing and special values::
@end menu

@node Calling .Call, Calling .External, Interface functions .Call and .External, Interface functions .Call and .External
@subsection Calling @code{.Call}

@findex .Call

Let us convert our finite convolution example to use @code{.Call}.  The
calling function in @R{} is

@example
conv <- function(a, b) .Call("convolve2", a, b)
@end example

@noindent
which could hardly be simpler, but as we shall see all the type
coercion is transferred to the C code, which is

@example
@group
#include <R.h>
#include <Rinternals.h>

SEXP convolve2(SEXP a, SEXP b)
@{
    int na, nb, nab;
    double *xa, *xb, *xab;
    SEXP ab;

    a = PROTECT(coerceVector(a, REALSXP));
    b = PROTECT(coerceVector(b, REALSXP));
    na = length(a); nb = length(b); nab = na + nb - 1;
    ab = PROTECT(allocVector(REALSXP, nab));
    xa = REAL(a); xb = REAL(b); xab = REAL(ab);
    for(int i = 0; i < nab; i++) xab[i] = 0.0;
    for(int i = 0; i < na; i++)
        for(int j = 0; j < nb; j++) xab[i + j] += xa[i] * xb[j];
    UNPROTECT(3);
    return ab;
@}
@end group
@end example

@node Calling .External, Missing and special values, Calling .Call, Interface functions .Call and .External
@subsection Calling @code{.External}

@findex .External

We can use the same example to illustrate @code{.External}.  The @R{}
code changes only by replacing @code{.Call} by @code{.External}

@example
conv <- function(a, b) .External("convolveE", a, b)
@end example

@noindent
but the main change is how the arguments are passed to the C code, this
time as a single SEXP.  The only change to the C code is how we handle
the arguments.

@example
@group
#include <R.h>
#include <Rinternals.h>

SEXP convolveE(SEXP args)
@{
    int i, j, na, nb, nab;
    double *xa, *xb, *xab;
    SEXP a, b, ab;

    a = PROTECT(coerceVector(CADR(args), REALSXP));
    b = PROTECT(coerceVector(CADDR(args), REALSXP));
    ...
@}
@end group
@end example

@noindent
Once again we do not need to protect the arguments, as in the @R{} side
of the interface they are objects that are already in use.  The macros

@example
@group
  first = CADR(args);
  second = CADDR(args);
  third = CADDDR(args);
  fourth = CAD4R(args);
@end group
@end example
@findex CADR
@findex CADDR
@findex CADDDR
@findex CAD4R

@noindent
provide convenient ways to access the first four arguments.  More
generally we can use the
@findex CAR
@findex CDR
@code{CDR} and @code{CAR} macros as in

@example
@group
  args = CDR(args); a = CAR(args);
  args = CDR(args); b = CAR(args);
@end group
@end example

@noindent
which clearly allows us to extract an unlimited number of arguments
(whereas @code{.Call} has a limit, albeit at 65 not a small one).

More usefully, the @code{.External} interface provides an easy way to
handle calls with a variable number of arguments, as @code{length(args)}
will give the number of arguments supplied (of which the first is
ignored).  We may need to know the names (`tags') given to the actual
arguments, which we can by using the @code{TAG} macro and using
something like the following example, that prints the names and the first
value of its arguments if they are vector types.

@example
@group
SEXP showArgs(SEXP args)
@{
    args = CDR(args); /* skip 'name' */
    for(int i = 0; args != R_NilValue; i++, args = CDR(args)) @{
        const char *name =
            isNull(TAG(args)) ? "" : CHAR(PRINTNAME(TAG(args)));
        SEXP el = CAR(args);
        if (length(el) == 0) @{
            Rprintf("[%d] '%s' R type, length 0\n", i+1, name);
           continue;
        @}
@end group
@group
        switch(TYPEOF(el)) @{
        case REALSXP:
            Rprintf("[%d] '%s' %f\n", i+1, name, REAL(el)[0]);
            break;
@end group
@group
        case LGLSXP:
        case INTSXP:
            Rprintf("[%d] '%s' %d\n", i+1, name, INTEGER(el)[0]);
            break;
@end group
@group
        case CPLXSXP:
        @{
            Rcomplex cpl = COMPLEX(el)[0];
            Rprintf("[%d] '%s' %f + %fi\n", i+1, name, cpl.r, cpl.i);
        @}
            break;
@end group
@group
        case STRSXP:
            Rprintf("[%d] '%s' %s\n", i+1, name,
                   CHAR(STRING_ELT(el, 0)));
           break;
@end group
@group
        default:
            Rprintf("[%d] '%s' R type\n", i+1, name);
       @}
    @}
    return R_NilValue;
@}
@end group
@end example
@findex PRINTNAME
@findex TYPEOF
@findex TAG

This can be called by the wrapper function

@example
showArgs <- function(...) invisible(.External("showArgs", ...))
@end example

@noindent
Note that this style of programming is convenient but not necessary, as
an alternative style is

@example
showArgs1 <- function(...) invisible(.Call("showArgs1", list(...)))
@end example

@noindent
The (very similar) C code is in the scripts.

Additional functions for accessing pairlist components are @code{CAAR},
@code{CDAR}, @code{CDDR}, and @code{CDDDR}.
@findex CAAR
@findex CDAR
@findex CDDR
@findex CDDDR
These components can be modified with @code{SETCAR}, @code{SETCDR},
@code{SETCADR}, @code{SETCADDR}, @code{SETCADDDR}, and @code{SETCAD4R}.
@findex SETCAR
@findex SETCDR
@findex SETCADR
@findex SETCADDR
@findex SETCADDDR
@findex SETCAD4R

@node Missing and special values,  , Calling .External, Interface functions .Call and .External
@subsection Missing and special values
@cindex Missing values
@cindex IEEE special values

One piece of error-checking the @code{.C} call does (unless @code{NAOK}
is true) is to check for missing (@code{NA}) and @acronym{IEEE} special
values (@code{Inf}, @code{-Inf} and @code{NaN}) and give an error if any
are found.  With the @code{.Call} interface these will be passed to our
code.  In this example the special values are no problem, as
@acronym{IEC60559} arithmetic will handle them correctly.  In the current
implementation this is also true of @code{NA} as it is a type of
@code{NaN}, but it is unwise to rely on such details.  Thus we will
re-write the code to handle @code{NA}s using macros defined in
@file{R_ext/Arith.h} included by @file{R.h}.

The code changes are the same in any of the versions of @code{convolve2}
or @code{convolveE}:

@example
@group
    ...
  for(int i = 0; i < na; i++)
    for(int j = 0; j < nb; j++)
        if(ISNA(xa[i]) || ISNA(xb[j]) || ISNA(xab[i + j]))
            xab[i + j] = NA_REAL;
        else
            xab[i + j] += xa[i] * xb[j];
    ...
@end group
@end example

@findex ISNA
@findex ISNAN

Note that the @code{ISNA} macro, and the similar macros @code{ISNAN}
(which checks for @code{NaN} or @code{NA}) and @code{R_FINITE} (which is
false for @code{NA} and all the special values), only apply to numeric
values of type @code{double}.  Missingness of integers, logicals and
character strings can be tested by equality to the constants
@code{NA_INTEGER}, @code{NA_LOGICAL} and @code{NA_STRING}.  These and
@code{NA_REAL} can be used to set elements of @R{} vectors to @code{NA}.

The constants @code{R_NaN}, @code{R_PosInf} and @code{R_NegInf} can be
used to set @code{double}s to the special values.

@node Evaluating R expressions from C, Parsing R code from C, Interface functions .Call and .External, System and foreign language interfaces
@section Evaluating R expressions from C
@cindex Evaluating R expressions from C

The main function we will use is

@example
SEXP eval(SEXP expr, SEXP rho);
@end example
@findex eval

@noindent
the equivalent of the interpreted @R{} code @code{eval(expr, envir =
rho)} (so @code{rho} must be an environment), although we can also make
use of @code{findVar}, @code{defineVar} and @code{findFun} (which
restricts the search to functions).
@findex findFun

To see how this might be applied, here is a simplified internal version
of @code{lapply} for expressions, used as

@example
@group
a <- list(a = 1:5, b = rnorm(10), test = runif(100))
.Call("lapply", a, quote(sum(x)), new.env())
@end group
@end example

@noindent
with C code

@example
@group
SEXP lapply(SEXP list, SEXP expr, SEXP rho)
@{
    int n = length(list);
    SEXP ans;

    if(!isNewList(list)) error("'list' must be a list");
    if(!isEnvironment(rho)) error("'rho' should be an environment");
    ans = PROTECT(allocVector(VECSXP, n));
    for(int i = 0; i < n; i++) @{
        defineVar(install("x"), VECTOR_ELT(list, i), rho);
        SET_VECTOR_ELT(ans, i, eval(expr, rho));
    @}
    setAttrib(ans, R_NamesSymbol, getAttrib(list, R_NamesSymbol));
    UNPROTECT(1);
    return ans;
@}
@end group
@end example

It would be closer to @code{lapply} if we could pass in a function
rather than an expression.  One way to do this is @emph{via} interpreted
@R{} code as in the next example, but it is possible (if somewhat
obscure) to do this in C code.  The following is based on the code in
@file{src/main/optimize.c}.

@example
@group
SEXP lapply2(SEXP list, SEXP fn, SEXP rho)
@{
    int n = length(list);
    SEXP R_fcall, ans;

    if(!isNewList(list)) error("'list' must be a list");
    if(!isFunction(fn)) error("'fn' must be a function");
    if(!isEnvironment(rho)) error("'rho' should be an environment");
    R_fcall = PROTECT(lang2(fn, R_NilValue));
    ans = PROTECT(allocVector(VECSXP, n));
    for(int i = 0; i < n; i++) @{
        SETCADR(R_fcall, VECTOR_ELT(list, i));
        SET_VECTOR_ELT(ans, i, eval(R_fcall, rho));
    @}
    setAttrib(ans, R_NamesSymbol, getAttrib(list, R_NamesSymbol));
    UNPROTECT(2);
    return ans;
@}
@end group
@end example

@noindent
used by

@example
.Call("lapply2", a, sum, new.env())
@end example

@noindent
Function @code{lang2} creates an executable pairlist of two elements, but
this will only be clear to those with a knowledge of a LISP-like
language.

As a more comprehensive example of constructing an @R{} call in C code
and evaluating, consider the following fragment of
@code{printAttributes} in @file{src/main/print.c}.

@example
    /* Need to construct a call to
       print(CAR(a), digits=digits)
       based on the R_print structure, then eval(call, env).
       See do_docall for the template for this sort of thing.
    */
    SEXP s, t;
    t = s = PROTECT(allocList(3));
    SET_TYPEOF(s, LANGSXP);
    SETCAR(t, install("print")); t = CDR(t);
    SETCAR(t,  CAR(a)); t = CDR(t);
    SETCAR(t, ScalarInteger(digits));
    SET_TAG(t, install("digits"));
    eval(s, env);
    UNPROTECT(1);
@end example
@findex allocList
@findex SET_TAG

@noindent
At this point @code{CAR(a)} is the @R{} object to be printed, the
current attribute.  There are three steps: the call is constructed as
a pairlist of length 3, the list is filled in, and the expression
represented by the pairlist is evaluated.

A pairlist is quite distinct from a generic vector list, the only
user-visible form of list in @R{}.  A pairlist is a linked list (with
@code{CDR(t)} computing the next entry), with items (accessed by
@code{CAR(t)}) and names or tags (set by @code{SET_TAG}).  In this call
there are to be three items, a symbol (pointing to the function to be
called) and two argument values, the first unnamed and the second named.
Setting the type to @code{LANGSXP} makes this a call which can be evaluated.

Customarily, the evaluation environment is passed from the calling
@R{} code (see @code{rho} above). In special cases it is possible that
the C code may need to obtain the current evaluation environment
which can be done via @code{R_GetCurrentEnv()} function.
@findex R_GetCurrentEnv

@menu
* Zero-finding::
* Calculating numerical derivatives::
@end menu

@node Zero-finding, Calculating numerical derivatives, Evaluating R expressions from C, Evaluating R expressions from C
@subsection Zero-finding
@cindex Zero-finding

In this section we re-work the example of Becker, Chambers & Wilks (1988,
pp.~205--10) on finding a zero of a univariate function.  The @R{} code
and an example are

@example
zero <- function(f, guesses, tol = 1e-7) @{
    f.check <- function(x) @{
        x <- f(x)
        if(!is.numeric(x)) stop("Need a numeric result")
        as.double(x)
    @}
    .Call("zero", body(f.check), as.double(guesses), as.double(tol),
          new.env())
@}

cube1 <- function(x) (x^2 + 1) * (x - 1.5)
zero(cube1, c(0, 5))
@end example

@noindent
where this time we do the coercion and error-checking in the @R{} code.
The C code is

@example
@group
SEXP mkans(double x)
@{
    // no need for PROTECT() here, as REAL(.) does not allocate:
    SEXP ans = allocVector(REALSXP, 1);
    REAL(ans)[0] = x;
    return ans;
@}
@end group

@group
double feval(double x, SEXP f, SEXP rho)
@{
    // a version with (too) much PROTECT()ion .. "better safe than sorry"
    SEXP symbol, value;
    PROTECT(symbol = install("x"));
    PROTECT(value = mkans(x));
    defineVar(symbol, value, rho);
    UNPROTECT(2);
    return(REAL(eval(f, rho))[0]);
@}
@end group

@group
SEXP zero(SEXP f, SEXP guesses, SEXP stol, SEXP rho)
@{
    double x0 = REAL(guesses)[0], x1 = REAL(guesses)[1],
           tol = REAL(stol)[0];
    double f0, f1, fc, xc;
@end group

@group
    if(tol <= 0.0) error("non-positive tol value");
    f0 = feval(x0, f, rho); f1 = feval(x1, f, rho);
    if(f0 == 0.0) return mkans(x0);
    if(f1 == 0.0) return mkans(x1);
    if(f0*f1 > 0.0) error("x[0] and x[1] have the same sign");
@end group

@group
    for(;;) @{
        xc = 0.5*(x0+x1);
        if(fabs(x0-x1) < tol) return  mkans(xc);
        fc = feval(xc, f, rho);
        if(fc == 0) return  mkans(xc);
        if(f0*fc > 0.0) @{
            x0 = xc; f0 = fc;
        @} else @{
            x1 = xc; f1 = fc;
        @}
    @}
@}
@end group
@end example

@node Calculating numerical derivatives,  , Zero-finding, Evaluating R expressions from C
@subsection Calculating numerical derivatives
@cindex Numerical derivatives

We will use a longer example (by Saikat DebRoy) to illustrate the use of
evaluation and @code{.External}.  This calculates numerical derivatives,
something that could be done as effectively in interpreted @R{} code but
may be needed as part of a larger C calculation.

An interpreted @R{} version and an example are

@example
@group
numeric.deriv <- function(expr, theta, rho=sys.frame(sys.parent()))
@{
    eps <- sqrt(.Machine$double.eps)
    ans <- eval(substitute(expr), rho)
    grad <- matrix(, length(ans), length(theta),
                   dimnames=list(NULL, theta))
    for (i in seq_along(theta)) @{
        old <- get(theta[i], envir=rho)
        delta <- eps * max(1, abs(old))
        assign(theta[i], old+delta, envir=rho)
        ans1 <- eval(substitute(expr), rho)
        assign(theta[i], old, envir=rho)
        grad[, i] <- (ans1 - ans)/delta
    @}
    attr(ans, "gradient") <- grad
    ans
@}
omega <- 1:5; x <- 1; y <- 2
numeric.deriv(sin(omega*x*y), c("x", "y"))
@end group
@end example

@noindent
where @code{expr} is an expression, @code{theta} a character vector of
variable names and @code{rho} the environment to be used.

For the compiled version the call from @R{} will be

@example
.External("numeric_deriv", @var{expr}, @var{theta}, @var{rho})
@end example

@noindent
with example usage

@example
.External("numeric_deriv", quote(sin(omega*x*y)),
          c("x", "y"), .GlobalEnv)
@end example

@noindent
Note the need to quote the expression to stop it being evaluated in the
caller.

Here is the complete C code which we will explain section by section.

@example
@group
#include <R.h> 
#include <Rinternals.h>
#include <float.h> /* for DBL_EPSILON */

SEXP numeric_deriv(SEXP args)
@{
    SEXP theta, expr, rho, ans, ans1, gradient, par, dimnames;
    double tt, xx, delta, eps = sqrt(DBL_EPSILON), *rgr, *rans;
    int i, start;
@end group

@group
    expr = CADR(args);
    if(!isString(theta = CADDR(args)))
        error("theta should be of type character");
    if(!isEnvironment(rho = CADDDR(args)))
        error("rho should be an environment");
@end group

@group
    ans = PROTECT(coerceVector(eval(expr, rho), REALSXP));
    gradient = PROTECT(allocMatrix(REALSXP, LENGTH(ans), LENGTH(theta)));
    rgr = REAL(gradient); rans = REAL(ans);
@end group

@group
    for(i = 0, start = 0; i < LENGTH(theta); i++, start += LENGTH(ans)) @{
        par = PROTECT(findVar(installChar(STRING_ELT(theta, i)), rho));
        tt = REAL(par)[0];
        xx = fabs(tt);
        delta = (xx < 1) ? eps : xx*eps;
        REAL(par)[0] += delta;
        ans1 = PROTECT(coerceVector(eval(expr, rho), REALSXP));
        for(int j = 0; j < LENGTH(ans); j++)
            rgr[j + start] = (REAL(ans1)[j] - rans[j])/delta;
        REAL(par)[0] = tt;
        UNPROTECT(2); /* par, ans1 */
    @}
@end group

@group
    dimnames = PROTECT(allocVector(VECSXP, 2));
    SET_VECTOR_ELT(dimnames, 1,  theta);
    dimnamesgets(gradient, dimnames);
    setAttrib(ans, install("gradient"), gradient);
    UNPROTECT(3); /* ans  gradient  dimnames */
    return ans;
@}
@end group
@end example

The code to handle the arguments is

@example
@group
    expr = CADR(args);
    if(!isString(theta = CADDR(args)))
        error("theta should be of type character");
    if(!isEnvironment(rho = CADDDR(args)))
        error("rho should be an environment");
@end group
@end example

@noindent
Note that we check for correct types of @code{theta} and @code{rho} but
do not check the type of @code{expr}.  That is because @code{eval} can
handle many types of @R{} objects other than @code{EXPRSXP}.  There is
no useful coercion we can do, so we stop with an error message if the
arguments are not of the correct mode.

The first step in the code is to evaluate the expression in the
environment @code{rho}, by

@example
    ans = PROTECT(coerceVector(eval(expr, rho), REALSXP));
@end example

@noindent
We then allocate space for the calculated derivative by

@example
    gradient = PROTECT(allocMatrix(REALSXP, LENGTH(ans), LENGTH(theta)));
@end example

@noindent
The first argument to @code{allocMatrix} gives the @code{SEXPTYPE} of
the matrix: here we want it to be @code{REALSXP}.  The other two
arguments are the numbers of rows and columns.  (Note that @code{LENGTH}
is intended to be used for vectors: @code{length} is more generally
applicable.)
@findex allocMatrix

@example
@group
    for(i = 0, start = 0; i < LENGTH(theta); i++, start += LENGTH(ans)) @{
        par = PROTECT(findVar(installChar(STRING_ELT(theta, i)), rho));
@end group
@end example

@noindent
Here, we are entering a for loop.  We loop through each of the
variables.  In the @code{for} loop, we first create a symbol
corresponding to the @code{i}'th element of the @code{STRSXP}
@code{theta}.  Here, @code{STRING_ELT(theta, i)} accesses the
@code{i}'th element of the @code{STRSXP} @code{theta}.  Macro
@code{CHAR()} extracts the actual character
representation@footnote{@pxref{Character encoding issues} for why this
might not be what is required.} of it: it returns a pointer.  We then
install the name and use @code{findVar} to find its value.

@example
@group
        tt = REAL(par)[0];
        xx = fabs(tt);
        delta = (xx < 1) ? eps : xx*eps;
        REAL(par)[0] += delta;
        ans1 = PROTECT(coerceVector(eval(expr, rho), REALSXP));
@end group
@end example

@noindent
We first extract the real value of the parameter, then calculate
@code{delta}, the increment to be used for approximating the numerical
derivative.  Then we change the value stored in @code{par} (in
environment @code{rho}) by @code{delta} and evaluate @code{expr} in
environment @code{rho} again.  Because we are directly dealing with
original @R{} memory locations here, @R{} does the evaluation for the
changed parameter value.

@example
@group
        for(int j = 0; j < LENGTH(ans); j++)
            rgr[j + start] = (REAL(ans1)[j] - rans[j])/delta;
        REAL(par)[0] = tt;
        UNPROTECT(2);
    @}
@end group
@end example

@noindent
Now, we compute the @code{i}'th column of the gradient matrix.  Note how
it is accessed: @R{} stores matrices by column (like Fortran).

@example
@group
    dimnames = PROTECT(allocVector(VECSXP, 2));
    SET_VECTOR_ELT(dimnames, 1, theta);
    dimnamesgets(gradient, dimnames);
    setAttrib(ans, install("gradient"), gradient);
    UNPROTECT(3);
    return ans;
@}
@end group
@end example

@noindent
First we add column names to the gradient matrix.  This is done by
allocating a list (a @code{VECSXP}) whose first element, the row names,
is @code{NULL} (the default) and the second element, the column names,
is set as @code{theta}.  This list is then assigned as the attribute
having the symbol @code{R_DimNamesSymbol}.  Finally we set the gradient
matrix as the gradient attribute of @code{ans}, unprotect the remaining
protected locations and return the answer @code{ans}.

@node Parsing R code from C, External pointers and weak references, Evaluating R expressions from C, System and foreign language interfaces
@section Parsing R code from C
@cindex Parsing R code from C

Suppose an @R{} extension want to accept an @R{} expression from the
user and evaluate it.  The previous section covered evaluation, but the
expression will be entered as text and needs to be parsed first.  A
small part of @R{}'s parse interface is declared in header file
@file{R_ext/Parse.h}@footnote{This is only guaranteed to show the
current interface: it is liable to change.}.

An example of the usage can be found in the (example) Windows package
@pkg{windlgs} included in the @R{} source tree.  The essential part is

@example
@group
#include <R.h>
#include <Rinternals.h>
#include <R_ext/Parse.h>

SEXP menu_ttest3()
@{
    char cmd[256];
    SEXP cmdSexp, cmdexpr, ans = R_NilValue;
    ParseStatus status;
   ...
    if(done == 1) @{
        cmdSexp = PROTECT(allocVector(STRSXP, 1));
        SET_STRING_ELT(cmdSexp, 0, mkChar(cmd));
        cmdexpr = PROTECT(R_ParseVector(cmdSexp, -1, &status, R_NilValue));
        if (status != PARSE_OK) @{
            UNPROTECT(2);
            error("invalid call %s", cmd);
        @}
        /* Loop is needed here as EXPSEXP will be of length > 1 */
        for(int i = 0; i < length(cmdexpr); i++)
            ans = eval(VECTOR_ELT(cmdexpr, i), R_GlobalEnv);
        UNPROTECT(2);
    @}
    return ans;
@}
@end group
@end example
@noindent
Note that a single line of text may give rise to more than one @R{}
expression.

@findex R_ParseVector
@code{R_ParseVector} is essentially the code used to implement
@code{parse(text=)} at @R{} level.  The first argument is a character
vector (corresponding to @code{text}) and the second the maximal
number of expressions to parse (corresponding to @code{n}).  The third
argument is a pointer to a variable of an enumeration type, and it is
normal (as @code{parse} does) to regard all values other than
@code{PARSE_OK} as an error.  Other values which might be returned are
@code{PARSE_INCOMPLETE} (an incomplete expression was found) and
@code{PARSE_ERROR} (a syntax error), in both cases the value returned
being @code{R_NilValue}. The fourth argument is a length one character
vector to be used as a filename in error messages, a @code{srcfile}
object or the @R{} @code{NULL} object (as in the example above). If a
@code{srcfile} object was used, a @code{srcref} attribute would be
attached to the result, containing a list of @code{srcref} objects of
the same length as the expression, to allow it to be echoed with its
original formatting.

@menu
* Accessing source references::
@end menu

@node Accessing source references,  , Parsing R code from C, Parsing R code from C
@subsection Accessing source references

The source references added by the parser are recorded by @R{}'s evaluator
as it evaluates code. Two functions
make these available to debuggers running C code:
@findex R_Srcref
@findex R_GetCurrentSrcref
@findex R_GetSrcFilename

@example
SEXP R_GetCurrentSrcref(int skip);
@end example

This function checks @code{R_Srcref} and the current evaluation stack
for entries that contain source reference information.  The
@code{skip} argument tells how many source references to skip before
returning the @code{SEXP} of the @code{srcref} object, counting from
the top of the stack.  If @code{skip < 0}, @code{abs(skip)} locations
are counted up from the bottom of the stack. If too few or no source
references are found, @code{NULL} is returned.

@example
SEXP R_GetSrcFilename(SEXP srcref);
@end example

This function extracts the filename from the source reference for
display, returning a length 1 character vector containing the
filename.  If no name is found, @code{""} is returned.

@node External pointers and weak references, Vector accessor functions, Parsing R code from C, System and foreign language interfaces
@section External pointers and weak references

The @code{SEXPTYPE}s @code{EXTPTRSXP} and @code{WEAKREFSXP} can be
encountered at @R{} level, but are created in C code.

@cindex external pointer
External pointer @code{SEXP}s are intended to handle references to C
structures such as `handles', and are used for this purpose in package
@CRANpkg{RODBC} for example.  They are unusual in their copying semantics in
that when an @R{} object is copied, the external pointer object is not
duplicated.  (For this reason external pointers should only be used as
part of an object with normal semantics, for example an attribute or an
element of a list.)

An external pointer is created by

@example
SEXP R_MakeExternalPtr(void *p, SEXP tag, SEXP prot);
@end example
@findex R_MakeExternalPtr

@noindent
where @code{p} is the pointer (and hence this cannot portably be a
function pointer), and @code{tag} and @code{prot} are references to
ordinary @R{} objects which will remain in existence (be protected from
garbage collection) for the lifetime of the external pointer object.  A
useful convention is to use the @code{tag} field for some form of type
identification and the @code{prot} field for protecting the memory that
the external pointer represents, if that memory is allocated from the
@R{} heap.  Both @code{tag} and @code{prot} can be @code{R_NilValue},
and often are.

An alternative way to create an external pointer from a function pointer
is

@example
typedef void * (*R_DL_FUNC)();
SEXP R_MakeExternalPtrFn(R_DL_FUNC p, SEXP tag, SEXP prot);
@end example
@findex R_MakeExternalPtrFn


The elements of an external pointer can be accessed and set @emph{via}

@example
void *R_ExternalPtrAddr(SEXP s);
DL_FUNC R_ExternalPtrAddrFn(SEXP s);
SEXP R_ExternalPtrTag(SEXP s);
SEXP R_ExternalPtrProtected(SEXP s);
void R_ClearExternalPtr(SEXP s);
void R_SetExternalPtrAddr(SEXP s, void *p);
void R_SetExternalPtrTag(SEXP s, SEXP tag);
void R_SetExternalPtrProtected(SEXP s, SEXP p);
@end example
@findex R_ExternalPtrAddr
@findex R_ExternalPtrAddrFn
@findex R_ExternalPtrTag
@findex R_ExternalPtrProtected
@findex R_ClearExternalPtr
@findex R_SetExternalPtrAddr
@findex R_SetExternalPtrTag
@findex R_SetExternalPtrProtected


@noindent
Clearing a pointer sets its value to the C @code{NULL} pointer.

@cindex finalizer
An external pointer object can have a @emph{finalizer}, a piece of code
to be run when the object is garbage collected.  This can be @R{} code
or C code, and the various interfaces are, respectively.

@example
void R_RegisterFinalizer(SEXP s, SEXP fun);
void R_RegisterFinalizerEx(SEXP s, SEXP fun, Rboolean onexit);

typedef void (*R_CFinalizer_t)(SEXP);
void R_RegisterCFinalizer(SEXP s, R_CFinalizer_t fun);
void R_RegisterCFinalizerEx(SEXP s, R_CFinalizer_t fun, Rboolean onexit);
@end example
@findex R_RegisterFinalizer
@findex R_RegisterCFinalizer
@findex R_RegisterFinalizerEx
@findex R_RegisterCFinalizerEx

@noindent
The @R{} function indicated by @code{fun} should be a function of a
single argument, the object to be finalized.  @R{} does not perform a
garbage collection when shutting down, and the @code{onexit} argument of
the extended forms can be used to ask that the finalizer be run during a
normal shutdown of the @R{} session.  It is suggested that it is good
practice to clear the pointer on finalization.

The only @R{} level function for interacting with external pointers is
@code{reg.finalizer} which can be used to set a finalizer.

It is probably not a good idea to allow an external pointer to be
@code{save}d and then reloaded, but if this happens the pointer will be
set to the C @code{NULL} pointer.

Finalizers can be run at many places in the code base and much of it,
including the @R{} interpreter, is not re-entrant.  So great care is
needed in choosing the code to be run in a finalizer.  Finalizers are
marked to be run at garbage collection but only run at a somewhat safe
point thereafter.

@cindex weak reference
Weak references are used to allow the programmer to maintain information
on entities without preventing the garbage collection of the entities
once they become unreachable.

A weak reference contains a key and a value.  The value is reachable is
if it either reachable directly or @emph{via} weak references with reachable
keys.  Once a value is determined to be unreachable during garbage
collection, the key and value are set to @code{R_NilValue} and the
finalizer will be run later in the garbage collection.

Weak reference objects are created by one of

@example
SEXP R_MakeWeakRef(SEXP key, SEXP val, SEXP fin, Rboolean onexit);
SEXP R_MakeWeakRefC(SEXP key, SEXP val, R_CFinalizer_t fin,
                    Rboolean onexit);
@end example
@findex R_MakeWeakRef
@findex R_MakeWeakRefC

@noindent
where the @R{} or C finalizer are specified in exactly the same way as
for an external pointer object (whose finalization interface is
implemented @emph{via} weak references).

The parts can be accessed @emph{via}

@example
SEXP R_WeakRefKey(SEXP w);
SEXP R_WeakRefValue(SEXP w);
void R_RunWeakRefFinalizer(SEXP w);
@end example
@findex R_WeakRefKey
@findex R_WeakRefValue
@findex R_RunWeakRefFinalizer

A toy example of the use of weak references can be found at
@uref{https://homepage.stat.uiowa.edu/~luke/R/references/weakfinex.html},
but that is used to add finalizers to external pointers which can now be
done more directly.  At the time of writing no @acronym{CRAN} or
Bioconductor package uses weak references.


@menu
* An external pointer example::
@end menu

@node An external pointer example,  , External pointers and weak references, External pointers and weak references
@subsection An example

Package @CRANpkg{RODBC} uses external pointers to maintain its
@emph{channels}, connections to databases.  There can be several
connections open at once, and the status information for each is stored
in a C structure (pointed to by @code{thisHandle} in the code extract
below) that is returned @emph{via} an external pointer as part of the RODBC
`channel' (as the @code{"handle_ptr"} attribute).  The external pointer
is created by

@example
    SEXP ans, ptr;
    ans = PROTECT(allocVector(INTSXP, 1));
    ptr = R_MakeExternalPtr(thisHandle, install("RODBC_channel"), R_NilValue);
    PROTECT(ptr);
    R_RegisterCFinalizerEx(ptr, chanFinalizer, TRUE);
            ...
    /* return the channel no */
    INTEGER(ans)[0] = nChannels;
    /* and the connection string as an attribute */
    setAttrib(ans, install("connection.string"), constr);
    setAttrib(ans, install("handle_ptr"), ptr);
    UNPROTECT(3);
    return ans;
@end example

@noindent
Note the symbol given to identify the usage of the external pointer, and
the use of the finalizer.  Since the final argument when registering the
finalizer is @code{TRUE}, the finalizer will be run at the end of the
@R{} session (unless it crashes).  This is used to close and clean up
the connection to the database.  The finalizer code is simply

@example
static void chanFinalizer(SEXP ptr)
@{
    if(!R_ExternalPtrAddr(ptr)) return;
    inRODBCClose(R_ExternalPtrAddr(ptr));
    R_ClearExternalPtr(ptr); /* not really needed */
@}
@end example

@noindent
Clearing the pointer and checking for a @code{NULL} pointer avoids any
possibility of attempting to close an already-closed channel.

@R{}'s connections provide another example of using external pointers,
in that case purely to be able to use a finalizer to close and destroy the
connection if it is no longer is use.

@node Vector accessor functions, Character encoding issues, External pointers and weak references, System and foreign language interfaces
@section Vector accessor functions

The vector accessors like @code{REAL}, @code{INTEGER}, @code{LOGICAL},
@code{RAW}, @code{COMPLEX}, and @code{VECTOR_ELT} are @emph{functions}
when used in @R{} extensions.  (For efficiency they may be macros or
inline functions when used in the @R{} source code, apart from
@code{SET_STRING_ELT} and @code{SET_VECTOR_ELT} which are always
functions.)
@findex VECTOR_ELT
@findex SET_VECTOR_ELT
@findex REAL
@findex INTEGER
@findex LOGICAL
@findex RAW
@findex COMPLEX

The accessor functions check that they are being used on an appropriate
type of @code{SEXP}.

Formerly it was possible for packages to obtain internal versions of
some accessors by defining @samp{USE_RINTERNALS} before including
@file{Rinternals.h}. This is no longer the case. Defining
@samp{USE_RINTERNALS} now has no effect.


@node Character encoding issues,  , Vector accessor functions, System and foreign language interfaces
@section Character encoding issues

@findex translateChar
@findex translateCharUTF8
@code{CHARSXP}s can be marked as coming from a known encoding (Latin-1
or UTF-8).  This is mainly intended for human-readable output, and most
packages can just treat such @code{CHARSXP}s as a whole.  However, if
they need to be interpreted as characters or output at C level then it
would normally be correct to ensure that they are converted to the
encoding of the current locale: this can be done by accessing the data
in the @code{CHARSXP} by @code{translateChar} rather than by
@code{CHAR}.  If re-encoding is needed this allocates memory with
@code{R_alloc} which thus persists to the end of the
@code{.Call}/@code{.External} call unless @code{vmaxset} is used
(@pxref{Transient storage allocation}).

There is a similar function @code{translateCharUTF8} which converts to
UTF-8: this has the advantage that a faithful translation is almost
always possible (whereas only a few languages can be represented in the
encoding of the current locale unless that is UTF-8).

Both @code{translateChar} and @code{translateCharUTF8} will translate
any input, using escapes such as @samp{<A9>} and @samp{<U+0093>} to
represent untranslatable parts of the input.

@findex getCharCE
@findex mkCharCE
There is a public interface to the encoding marked on @code{CHARSXPs}
@emph{via}

@example
typedef enum @{CE_NATIVE, CE_UTF8, CE_LATIN1, CE_BYTES, CE_SYMBOL, CE_ANY@} cetype_t;
cetype_t getCharCE(SEXP);
SEXP mkCharCE(const char *, cetype_t);
@end example

@noindent
Only @code{CE_UTF8} and @code{CE_LATIN1} are marked on @code{CHARSXPs}
(and so @code{Rf_getCharCE} will only return one of the first three),
and these should only be used on non-@acronym{ASCII} strings.  Value
@code{CE_BYTES} is used to make @code{CHARSXP}s which should be regarded
as a set of bytes and not translated.  Value @code{CE_SYMBOL} is used
internally to indicate Adobe Symbol encoding.  Value @code{CE_ANY} is
used to indicate a character string that will not need re-encoding --
this is used for character strings known to be in @acronym{ASCII}, and
can also be used as an input parameter where the intention is that the
string is treated as a series of bytes.  (See the comments under
@code{mkChar} about the length of input allowed.)

Function

@findex reEnc
@example
const char *reEnc(const char *x, cetype_t ce_in, cetype_t ce_out,
                  int subst);
@end example

@noindent
can be used to re-encode character strings: like @code{translateChar} it
returns a string allocated by @code{R_alloc}.  This can translate from
@code{CE_SYMBOL} to @code{CE_UTF8}, but not conversely.  Argument
@code{subst} controls what to do with untranslatable characters or
invalid input: this is done byte-by-byte with @code{1} indicates to
output hex of the form @code{<a0>}, and @code{2} to replace by @code{.},
with any other value causing the byte to produce no output.

@findex mkCharLenCE
There is also

@example
SEXP mkCharLenCE(const char *, size_t, cetype_t);
@end example

@noindent
to create marked character strings of a given length.


@node The R API, Generic functions and methods, System and foreign language interfaces, Top
@chapter The R @acronym{API}: entry points for C code

@menu
* Memory allocation::
* Error signaling::
* Random numbers::
* Missing and IEEE values::
* Printing::
* Calling C from Fortran and vice versa::
* Numerical analysis subroutines::
* Optimization::
* Integration::
* Utility functions::
* Re-encoding::
* Condition handling and cleanup code::
* Allowing interrupts::
* Platform and version information::
* Inlining C functions::
* Controlling visibility::
* Standalone Mathlib::
* Organization of header files::
@end menu

There are a large number of entry points in the @R{} executable/DLL that
can be called from C code (and some that can be called from Fortran
code).  Only those documented here are stable enough that they will only
be changed with considerable notice.

The recommended procedure to use these is to include the header file
@file{R.h} in your C code by

@example
#include <R.h>
@end example

@noindent
This will include several other header files from the directory
@file{@var{R_INCLUDE_DIR}/R_ext}, and there are other header files
there that can be included too, but many of the features they contain
should be regarded as undocumented and unstable.

Most of these header files, including all those included by @file{R.h},
can be used from C++ code.

@quotation Note
Because @R{} re-maps many of its external names to avoid clashes with
system or user code, it is @emph{essential} to include the appropriate
header files when using these entry points.
@end quotation

This remapping can cause problems@footnote{Known problems have been
defining @code{LENGTH}, @code{error}, @code{length}, @code{vector} and
@code{warning}: whether these matter depends on the OS and toolchain,
with many problem reports involving @command{clang++}.  As from
@command{clang} 13.0.0, the remapping of @code{match} breaks the
subsequent inclusion of @file{omp.h}.}, and can be eliminated by
defining @code{R_NO_REMAP} (before including any @R{} headers) and
prepending @samp{Rf_} to @emph{all} the function names used from
@file{Rinternals.h} and @file{R_ext/Error.h}.  These problems can
usually be avoided by including other headers (such as system headers
and those for external software used by the package) before any @R{}
headers. (Headers from other packages may include @R{} headers directly
or @emph{via} inclusion from further packages, and may define
@code{R_NO_REMAP} with or without including @file{Rinternals.h}.)

We can classify the entry points as

@table @emph
@item API
Entry points which are documented in this manual and declared in an
installed header file.  These can be used in distributed packages and
will only be changed after deprecation.

@item public
Entry points declared in an installed header file that are exported
on all @R{} platforms but are not documented and subject to change
without notice.

@item private
Entry points that are used when building @R{} and exported on all @R{}
platforms but are not declared in the installed header files.
Do not use these in distributed code.

@item hidden
Entry points that are where possible (Windows and some modern Unix-alike
compilers/loaders when using @R{} as a shared library) not exported.
@end table

@node Memory allocation, Error signaling, The R API, The R API
@section Memory allocation
@cindex Memory allocation from C

@menu
* Transient storage allocation::
* User-controlled memory::
@end menu

There are two types of memory allocation available to the C programmer,
one in which @R{} manages the clean-up and the other in which user
has full control (and responsibility).

These functions are declared in header @file{R_exts/RS.h} which is
included by @file{R.h}.

@node Transient storage allocation, User-controlled memory, Memory allocation, Memory allocation
@subsection Transient storage allocation
@findex R_alloc
@findex R_allocLD
@findex S_alloc
@findex S_realloc
@findex vmaxget
@findex vmaxset
@findex nrows

Here @R{} will reclaim the memory at the end of the call to @code{.C},
@code{.Call} or @code{.External}.  Use

@example
char *R_alloc(size_t @var{n}, int @var{size})
@end example

@noindent
which allocates @var{n} units of @var{size} bytes each.  A typical usage
(from package @pkg{stats}) is

@example
x = (int *) R_alloc(nrows(merge)+2, sizeof(int));
@end example

@noindent
(@code{size_t} is defined in @file{stddef.h} which the header defining
@code{R_alloc} includes.)

There is a similar call, @code{S_alloc} (named for compatibility with older
versions of @Sl{}) which zeroes the memory allocated,

@example
char *S_alloc(long @var{n}, int @var{size})
@end example

@noindent
and

@example
char *S_realloc(char *@var{p}, long @var{new}, long @var{old}, int @var{size})
@end example

@noindent
which (for @code{@var{new} > @var{old}}) changes the allocation size
from @var{old} to @var{new} units, and zeroes the additional units.  NB:
these calls are best avoided as @code{long} is insufficient for large
memory allocations on 64-bit Windows (where it is limited to 2^31-1
bytes).

This memory is taken from the heap, and released at the end of the
@code{.C}, @code{.Call} or @code{.External} call.  Users can also manage
it, by noting the current position with a call to @code{vmaxget} and
subsequently clearing memory allocated by a call to @code{vmaxset}. An
example might be

@example
void *vmax = vmaxget()
// a loop involving the use of R_alloc at each iteration
vmaxset(vmax)
@end example

@noindent
This is only recommended for experts.

Note that this memory will be freed on error or user interrupt
(if allowed: @pxref{Allowing interrupts}).

The memory returned is only guaranteed to be aligned as required for
@code{double} pointers: take precautions if casting to a pointer which
needs more.  There is also

@example
long double *R_allocLD(size_t @var{n})
@end example

@noindent
which is guaranteed to have the 16-byte alignment needed for @code{long
double} pointers on some platforms.


These functions should only be used in code called by @code{.C} etc,
never from front-ends.  They are not thread-safe.

@node User-controlled memory,  , Transient storage allocation, Memory allocation
@subsection User-controlled memory
@findex R_Calloc
@findex R_Realloc
@findex R_Free

The other form of memory allocation is an interface to @code{malloc},
the interface providing @R{} error signaling.  This memory lasts until
freed by the user and is additional to the memory allocated for the @R{}
workspace.

The interface macros are

@example
@group
@var{type}* R_Calloc(size_t @var{n}, @var{type})
@var{type}* R_Realloc(@var{any} *@var{p}, size_t @var{n}, @var{type})
void R_Free(@var{any} *@var{p})
@end group
@end example

@noindent
providing analogues of @code{calloc}, @code{realloc} and @code{free}.
If there is an error during allocation it is handled by @R{}, so if
these return the memory has been successfully allocated or freed.
@code{R_Free} will set the pointer @var{p} to @code{NULL}.  (Some but not
all versions of @Sl{} did so.)

Users should arrange to @code{R_Free} this memory when no longer needed,
including on error or user interrupt.  This can often be done most
conveniently from an @code{on.exit} action in the calling @R{} function
-- see @code{pwilcox} for an example.

Do not assume that memory allocated by @code{R_Calloc}/@code{R_Realloc}
comes from the same pool as used by @code{malloc}:@footnote{That was not
the case on Windows prior to @R{} 4.2.0.}in particular do not use
@code{free} or @code{strdup} with it.

Memory obtained by these macros should be aligned in the same way as
@code{malloc}, that is `suitably aligned for any kind of variable'.

@findex Calloc
@findex Realloc
@findex Free
@c The R_ forms were introduced in 2016-09, hence for R 3.4.0
Historically the macros @code{Calloc}, @code{Free} and @code{Realloc}
were used, and these remain available unless  @code{STRICT_R_HEADERS}
was defined prior to the inclusion of the header.

@findex CallocCharBuf
@findex Memcpy
@findex Memzero
@example
@group
char * CallocCharBuf(size_t @var{n})
void * Memcpy(@var{p}, @var{q}, @var{n})
void * Memzero(@var{p}, @var{m})
@end group
@end example
@code{CallocCharBuf} is shorthand for @code{R_Calloc(n+1, char)} to allow
for the @code{nul} terminator.  @code{Memcpy} and @code{Memzero} take
@code{n} items from array @code{p} and copy them to array @code{p} or
zero them respectively.

@node Error signaling, Random numbers, Memory allocation, The R API
@section Error signaling
@cindex Error signaling from C

The basic error signaling routines are the equivalents of @code{stop} and
@code{warning} in @R{} code, and use the same interface.

@example
@group
void error(const char * @var{format}, ...);
void warning(const char * @var{format}, ...);
void errorcall(SEXP @var{call}, const char * @var{format}, ...);
void warningcall(SEXP @var{call}, const char * @var{format}, ...);
void warningcall_immediate(SEXP @var{call}, const char * @var{format}, ...);
@end group
@end example
@findex error
@findex warning
@findex errorcall
@findex warningcall
@findex warningcall_immediate

@noindent
These have the same call sequences as calls to @code{printf}, but in the
simplest case can be called with a single character string argument
giving the error message. (Don't do this if the string contains @samp{%}
or might otherwise be interpreted as a format.)

These are defined in header @file{R_ext/Error.h} included by
@file{R.h}.

@menu
* Error signaling from Fortran::
@end menu

@node Error signaling from Fortran,  , Error signaling, Error signaling
@subsection Error signaling from Fortran
@cindex Error signaling from Fortran

There are two interface function provided to call @code{error} and
@code{warning} from Fortran code, in each case with a simple character
string argument.  They are defined as

@example
@group
subroutine rexit(@var{message})
subroutine rwarn(@var{message})
@end group
@end example

Messages of more than 255 characters are truncated, with a warning.


@node Random numbers, Missing and IEEE values, Error signaling, The R API
@section Random number generation
@cindex Random numbers in C
@findex unif_rand
@findex norm_rand
@findex exp_rand
@findex R_unif_index
@findex GetRNGstate
@findex PutRNGstate
@findex .Random.seed
@c @findex seed_in
@c @findex seed_out

The interface to @R{}'s internal random number generation routines is

@example
@group
double unif_rand();
double norm_rand();
double exp_rand();
double R_unif_index(double);
@end group
@end example

@noindent
giving one uniform, normal or exponential pseudo-random variate.
However, before these are used, the user must call

@example
GetRNGstate();
@end example

@noindent
and after all the required variates have been generated, call

@example
PutRNGstate();
@end example

@noindent
These essentially read in (or create) @code{.Random.seed} and write it
out after use.

These are defined in header @file{R_ext/Random.h}.

The random number generator is private to @R{}; there is no way to
select the kind of RNG nor set the seed except by evaluating calls to the
@R{} functions.

The C code behind @R{}'s @code{r@var{xxx}} functions can be accessed by
including the header file @file{Rmath.h}; @xref{Distribution functions}.
Those calls should also be preceded and followed by calls to
@code{GetRNGstate} and @code{PutRNGstate}.

@node Missing and IEEE values, Printing, Random numbers, The R API
@section Missing and @acronym{IEEE} special values
@cindex Missing values
@cindex IEEE special values
@findex ISNA
@findex ISNAN
@findex R_FINITE
@findex R_IsNaN
@findex R_PosInf
@findex R_NegInf
@findex NA_REAL

A set of functions is provided to test for @code{NA}, @code{Inf},
@code{-Inf} and @code{NaN}.  These functions are accessed @emph{via} macros:

@example
@group
ISNA(@var{x})        @r{True for R's @code{NA} only}
ISNAN(@var{x})       @r{True for R's @code{NA} and @acronym{IEEE} @code{NaN}}
R_FINITE(@var{x})    @r{False for @code{Inf}, @code{-Inf}, @code{NA}, @code{NaN}}
@end group
@end example

@noindent
and @emph{via} function @code{R_IsNaN} which is true for @code{NaN} but not
@code{NA}.

Do use @code{R_FINITE} rather than @code{isfinite} or @code{finite}; the
latter is often mendacious and @code{isfinite} is only available on a
some platforms, on which @code{R_FINITE} is a macro expanding to
@code{isfinite}.

Currently in C code @code{ISNAN} is a macro calling @code{isnan}.
(Since this gives problems on some C++ systems, if the @R{} headers is
called from C++ code a function call is used.)

You can check for @code{Inf} or @code{-Inf} by testing equality to
@code{R_PosInf} or @code{R_NegInf}, and set (but not test) an @code{NA}
as @code{NA_REAL}.

All of the above apply to @emph{double} variables only.  For integer
variables there is a variable accessed by the macro @code{NA_INTEGER}
which can used to set or test for missingness.

These are defined in header @file{R_ext/Arith.h} included by @file{R.h}.

@node Printing, Calling C from Fortran and vice versa, Missing and IEEE values, The R API
@section Printing
@cindex Printing from C
@findex Rprintf
@findex REprintf
@findex Rvprintf
@findex REvprintf

The most useful function for printing from a C routine compiled into
@R{} is @code{Rprintf}.  This is used in exactly the same way as
@code{printf}, but is guaranteed to write to @R{}'s output (which might
be a @acronym{GUI} console rather than a file, and can be re-directed by
@code{sink}).  It is wise to write complete lines (including the
@code{"\n"}) before returning to @R{}.  It is defined in
@file{R_ext/Print.h}.

The function @code{REprintf} is similar but writes on the error stream
(@code{stderr}) which may or may not be different from the standard
output stream.

Functions @code{Rvprintf} and @code{REvprintf} are analogues using the
@code{vprintf} interface.  Because that is a C99@footnote{also part of
C++11.} interface, they are only defined by @file{R_ext/Print.h} in C++
code if the macro @code{R_USE_C99_IN_CXX} is defined before it is
included or (as from @R{} 4.0.0) a C++11 compiler is used.

Another circumstance when it may be important to use these functions is
when using parallel computation on a cluster of computational nodes, as
their output will be re-directed/logged appropriately.

@menu
* Printing from Fortran::
@end menu

@node Printing from Fortran,  , Printing, Printing
@subsection Printing from Fortran
@cindex Printing from Fortran

On many systems Fortran @code{write} and @code{print} statements can be
used, but the output may not interleave well with that of C, and may be
invisible on @acronym{GUI} interfaces.  They are not portable and best
avoided.

Some subroutines are provided to ease the output of information from
Fortran code.

@findex dblepr
@findex realpr
@findex intpr
@example
@group
subroutine dblepr(@var{label}, @var{nchar}, @var{data}, @var{ndata})
subroutine realpr(@var{label}, @var{nchar}, @var{data}, @var{ndata})
subroutine intpr (@var{label}, @var{nchar}, @var{data}, @var{ndata})
@end group
@end example

@noindent
and from @R{}@tie{}4.0.0,
@findex labelpr
@findex dblepr1
@findex realpr1
@findex intpr1
@example
@group
subroutine labelpr(@var{label}, @var{nchar})
subroutine dblepr1(@var{label}, @var{nchar}, @var{var})
subroutine realpr1(@var{label}, @var{nchar}, @var{var})
subroutine intpr1 (@var{label}, @var{nchar}, @var{var})
@end group
@end example

@noindent
Here @var{label} is a character label of up to 255 characters,
@var{nchar} is its length (which can be @code{-1} if the whole label is
to be used), @var{data} is an array of length at least @var{ndata} of
the appropriate type (@code{double precision}, @code{real} and
@code{integer} respectively) and @var{var} is a (scalar) variable.
These routines print the label on one line and then print @var{data} or
@var{var} as if it were an @R{} vector on subsequent line(s).  Note that
some compilers will give an error or warning unless @var{data} is an
array: others will accept a scalar when @var{ndata} has value one or
zero.  @strong{NB:} There is no check on the type of @var{data} or
@var{var}, so using @code{real} (including a real constant) instead of
@code{double precision} will give incorrect answers.

@code{intpr} works with zero @var{ndata} so can be used to print a
label in earlier versions of @R{}.

@node Calling C from Fortran and vice versa, Numerical analysis subroutines, Printing, The R API
@section Calling C from Fortran and vice versa
@cindex Calling C from Fortran and vice versa

Naming conventions for symbols generated by Fortran differ by platform:
it is not safe to assume that Fortran names appear to C with a trailing
underscore.  To help cover up the platform-specific differences there is
a set of macros@footnote{The @samp{F77_} in the names is historical and
dates back to usage in @Sl{}.} that should be used.

@table @code
@item F77_SUB(@var{name})
to define a function in C to be called from Fortran
@item F77_NAME(@var{name})
to declare a Fortran routine in C before use
@item F77_CALL(@var{name})
to call a Fortran routine from C
@item F77_COMDECL(@var{name})
to declare a Fortran common block in C
@item F77_COM(@var{name})
to access a Fortran common block from C
@end table

On most current platforms these are all the same, but it is unwise to
rely on this.  Note that names containing underscores were not legal in
Fortran 77, and are not portably handled by the above macros.  (Also,
all Fortran names for use by @R{} are lower case, but this is not
enforced by the macros.)

For example, suppose we want to call R's normal random numbers from
Fortran.  We need a C wrapper along the lines of

@cindex Random numbers in Fortran
@example
@group
#include <R.h>

void F77_SUB(rndstart)(void) @{ GetRNGstate(); @}
void F77_SUB(rndend)(void) @{ PutRNGstate(); @}
double F77_SUB(normrnd)(void) @{ return norm_rand(); @}
@end group
@end example

@noindent
to be called from Fortran as in

@example
@group
      subroutine testit()
      double precision normrnd, x
      call rndstart()
      x = normrnd()
      call dblepr("X was", 5, x, 1)
      call rndend()
      end
@end group
@end example

@noindent
Note that this is not guaranteed to be portable, for the return
conventions might not be compatible between the C and Fortran compilers
used.  (Passing values @emph{via} arguments is safer.)

The standard packages, for example @pkg{stats}, are a rich source of
further examples.

Where supported, @emph{link time optimization} provides a reliable way
to check the consistency of calls to C from Fortran or @emph{vice
versa}.
@xref{Using Link-time Optimization}.
One place where this occurs is the registration of @code{.Fortran} calls
in C code (@pxref{Registering native routines}).  For example
@example
init.c:10:13: warning: type of 'vsom_' does not match original
 declaration [-Wlto-type-mismatch]
  extern void F77_NAME(vsom)(void *, void *, void *, void *, 
    void *, void *, void *, void *, void *);
vsom.f90:20:33: note: type mismatch in parameter 9
   subroutine vsom(neurons,dt,dtrows,dtcols,xdim,ydim,alpha,train)
vsom.f90:20:33: note: 'vsom' was previously declared here
@end example
shows that a subroutine has been registered with 9 arguments (as that is
what the @code{.Fortran} call used) but only has 8.

@menu
* Fortran character strings::
* Fortran LOGICAL::
* Passing functions::
@end menu

@node Fortran character strings, Fortran LOGICAL, Calling C from Fortran and vice versa, Calling C from Fortran and vice versa
@subsection Fortran character strings

Passing character strings from C to Fortran or @emph{vice versa} is
not portable, but can be done with care.  The internal representations
are different: a character array in C (or C++) is nul-terminated so its
length can be computed by @code{strlen}.  Fortran character arrays are
typically stored as an array of bytes and a length.  This matters when
passing strings from C to Fortran or @emph{vice versa}: in many cases
one has been able to get away with passing the string but not the
length.  However, in 2019 this changed for @command{gfortran}, starting
with version 9 but backported to versions 7 and 8.  Several months
later, @command{gfortran} 9.2 introduced an option
@example
-ftail-call-workaround
@end example
@noindent
and made it the current default but said it might be withdrawn in future.

Suppose we want a function to report a message from Fortran to @R{}'s
console (one could use @code{labelpr}, or @code{intpr} with dummy data,
but this might be the basis of a custom reporting function).  Suppose the
equivalent in Fortran would be
@example
      subroutine rmsg(msg)
      character*(*) msg
      print *.msg
      end
@end example
@noindent
in file @file{rmsg.f}. Using @command{gfortran} 9.2 and later we can
extract the C view by
@example
gfortran -c -fc-prototypes-external rmsg.f
@end example
@noindent
which gives
@example
void rmsg_ (char *msg, size_t msg_len);
@end example
@noindent 
(where @code{size_t} applies to version 8 and later).  We could re-write
that portably in C as
@example
#define USE_FC_LEN_T
#include <Rconfig.h> // included by R.h, so define USE_FC_LEN_T early

void F77_NAME(rmsg)(char *msg, FC_LEN_T msg_len)
@{
    char cmsg[msg_len+1];
    strncpy(cmsg, msg, msg_len);
    cmsg[msg_len] = '\0'; // nul-terminate the string, to be sure
    // do something with 'cmsg' 
@}
@end example
@noindent
in code depending on @code{R(>= 3.6.2)}.  For earlier versions of @R{} we
could just assume that @code{msg} is nul-terminated (not guaranteed, but
people have been getting away with it for many years), so the complete C
side might be
@example
#define USE_FC_LEN_T
#include <Rconfig.h>

#ifdef FC_LEN_T
void F77_NAME(rmsg)(char *msg, FC_LEN_T msg_len)
@{
    char cmsg[msg_len+1];
    strncpy(cmsg, msg, msg_len);
    cmsg[msg_len] = '\0';
    // do something with 'cmsg' 
@}
#else
void F77_NAME(rmsg)(char *msg)
@{
    // do something with 'msg' 
@}
#endif
@end example


An alternative is to use Fortran 2003 features to set up the Fortran
routine to pass a C-compatible character string.  We could use something
like
@example
      module cfuncs
        use iso_c_binding, only: c_char, c_null_char
        interface
          subroutine cmsg(msg) bind(C, name = 'cmsg')
            use iso_c_binding, only: c_char
            character(kind = c_char):: msg(*)
          end subroutine cmsg
        end interface
      end module

      subroutine rmsg(msg)
        use cfuncs
        character(*) msg
        call cmsg(msg//c_null_char) ! need to concatenate a nul terminator
      end subroutine rmsg
@end example
@noindent
where the C side is simply
@example
void cmsg(const char *msg)
@{
    // do something with nul-terminated string 'msg'
@}
@end example

Passing a variable-length string from C to Fortran is trickier, but
@uref{https://software.intel.com/content/www/us/en/develop/documentation/fortran-compiler-oneapi-dev-guide-and-reference/top/compiler-reference/mixed-language-programming/standard-tools-for-interoperability/bind.html}
provides a recipe.  However, all the uses in BLAS and LAPACK are of a
single character, and for these we can write a wrapper in Fortran
along the lines of
@example
      subroutine c_dgemm(transa, transb, m, n, k, alpha,
     +     a, lda, b, ldb, beta, c, ldc)
     +     bind(C, name = 'Cdgemm')
        use iso_c_binding, only : c_char, c_int, c_double
        character(c_char):: transa, transb
        integer(c_int):: m, n, k, lda, ldb, ldc
        real(c_double):: alpha, beta, a(lda,*), b(ldb,*), c(ldc,*)
        call dgemm(transa, transb, m, n, k, alpha,
     +             a, lda, b, ldb, beta, c, ldc)
      end subroutine c_dgemm
@end example
@noindent
which is then called from C with declaration
@example
void
Cdgemm(const char *transa, const char *transb, const int *m,
       const int *n, const int *k, const double *alpha,
       const double *a, const int *lda, const double *b, const int *ldb,
       const double *beta, double *c, const int *ldc);
@end example

@noindent
Alternatively, do as @R{} does as from version 3.6.2 and pass
the character length(s) from C to Fortran. A portable way to do this is
@example
// before any R headers, or define in PKG_CPPFLAGS
#define USE_FC_LEN_T
#include <Rconfig.h>
#include <R_ext/BLAS.h>
#ifndef FCONE
# define FCONE
#endif
...
        F77_CALL(dgemm)("N", "T", &nrx, &ncy, &ncx, &one, x, 
                        &nrx, y, &nry, &zero, z, &nrx FCONE FCONE);
@end example
@noindent
(Note there is no comma before or between the @code{FCONE} invocations.)
It is strongly recommended that packages which call from C/C++
BLAS/LAPACK routines with character arguments adopt this approach.

@node Fortran LOGICAL, Passing functions, Fortran character strings, Calling C from Fortran and vice versa
@subsection Fortran LOGICAL

Passing Fortran LOGICAL variables to/from C/C++ is potentially
compiler-dependent.  Fortran compilers have long used a 32-bit integer
type so it is pretty portable to use @code{int *} on the C/C++ side.
However, recent versions of @command{gfortran} @emph{via} the option
@option{-fc-prototypes-external} say the C equivalent is
@code{int_least32_t *}: `Link-Time Optimization' will report @code{int
*} as a mismatch. It is possible to use @code{iso_c_binding} in Fortran
to map LOGICAL variables to the C99 type @code{_Bool}, but it is usually
simpler to pass integers to and fro.

@node Passing functions,  , Fortran LOGICAL, Calling C from Fortran and vice versa
@subsection Passing functions

A number of packages call C functions passed as arguments to Fortran
code along the lines of

@example
c         subroutine fcn(m,n,x,fvec,iflag)
c         integer m,n,iflag
c         double precision x(n),fvec(m)
...
      subroutine lmdif(fcn, ...
@end example

@noindent
where the C declaration and call are

@example
void fcn_lmdif(int *m, int *n, double *par, double *fvec, int *iflag);

void F77_NAME(lmdif)(void (*fcn_lmdif)(int *m, int *n, double *par,
                                       double *fvec, int *iflag), ...

F77_CALL(lmdif)(&fcn_lmdif, ...
@end example

@c clang and experimental gfortran on M1 macOS in late 2020.
This works on most platforms but depends on the C and Fortran compilers
agreeing on calling conventions: this have been seen to fail.  The most
portable solution seems to be to convert the Fortran code to C, perhaps
using @command{f2c}.

@node Numerical analysis subroutines, Optimization, Calling C from Fortran and vice versa, The R API
@section Numerical analysis subroutines
@cindex Numerical analysis subroutines from C

@R{} contains a large number of mathematical functions for its own use,
for example numerical linear algebra computations and special functions.

The header files @file{R_ext/BLAS.h}, @file{R_ext/Lapack.h} and
@file{R_ext/Linpack.h} contains declarations of the BLAS, LAPACK and
LINPACK linear algebra functions included in @R{}.  These are expressed
as calls to Fortran subroutines, and they will also be usable from
users' Fortran code.  Although not part of the official @acronym{API},
this set of subroutines is unlikely to change (but might be
supplemented).

The header file @file{Rmath.h} lists many other functions that are
available and documented in the following subsections. Many of these are
C interfaces to the code behind @R{} functions, so the @R{} function
documentation may give further details.

@menu
* Distribution functions::
* Mathematical functions::
* Numerical Utilities::
* Mathematical constants::
@end menu

@node Distribution functions, Mathematical functions, Numerical analysis subroutines, Numerical analysis subroutines
@subsection Distribution functions
@cindex  Distribution functions from C

The routines used to calculate densities, cumulative distribution
functions and quantile functions for the standard statistical
distributions are available as entry points.

The arguments for the entry points follow the pattern of those for the
normal distribution:

@example
@group
double dnorm(double @var{x}, double @var{mu}, double @var{sigma}, int @var{give_log});
double pnorm(double @var{x}, double @var{mu}, double @var{sigma}, int @var{lower_tail},
             int @var{give_log});
double qnorm(double @var{p}, double @var{mu}, double @var{sigma}, int @var{lower_tail},
             int @var{log_p});
double rnorm(double @var{mu}, double @var{sigma});
@end group
@end example

@noindent
That is, the first argument gives the position for the density and CDF
and probability for the quantile function, followed by the
distribution's parameters.  Argument @var{lower_tail} should be
@code{TRUE} (or @code{1}) for normal use, but can be @code{FALSE} (or
@code{0}) if the probability of the upper tail is desired or specified.

Finally, @var{give_log} should be non-zero if the result is required on
log scale, and @var{log_p} should be non-zero if @var{p} has been
specified on log scale.

Note that you directly get the cumulative (or ``integrated'')
@emph{hazard} function, @eqn{H(t) = - \log(1 - F(t)), H(t) = - log(1 -
F(t))}, by using

@example
- p@var{dist}(t, ..., /*lower_tail = */ FALSE, /* give_log = */ TRUE)
@end example

@noindent
or shorter (and more cryptic) @code{- p@var{dist}(t, ..., 0, 1)}.
@cindex cumulative hazard

The random-variate generation routine @code{rnorm} returns one normal
variate. @xref{Random numbers}, for the protocol in using the
random-variate routines.
@cindex Random numbers in C

Note that these argument sequences are (apart from the names and that
@code{rnorm} has no @var{n}) mainly the same as the corresponding @R{}
functions of the same name, so the documentation of the @R{} functions
can be used.  Note that the exponential and gamma distributions are
parametrized by @code{scale} rather than @code{rate}.


For reference, the following table gives the basic name (to be prefixed
by @samp{d}, @samp{p}, @samp{q} or @samp{r} apart from the exceptions
noted) and distribution-specific arguments for the complete set of
distributions.

@quotation
@multitable @columnfractions .28 .22 .30
@item beta @tab @code{beta} @tab @code{a}, @code{b}
@item non-central beta @tab @code{nbeta} @tab @code{a}, @code{b}, @code{ncp}
@c in R shape1, shape2, ncp
@item binomial @tab @code{binom} @tab @code{n}, @code{p}
@item Cauchy @tab @code{cauchy} @tab @code{location}, @code{scale}
@item chi-squared @tab @code{chisq} @tab @code{df}
@item non-central chi-squared @tab @code{nchisq} @tab @code{df}, @code{ncp}
@item exponential @tab @code{exp} @tab @code{scale} (and @strong{not} @code{rate})
@item F @tab @code{f} @tab @code{n1}, @code{n2}
@item non-central F @tab @code{nf} @tab @code{n1}, @code{n2}, @code{ncp}
@item gamma @tab @code{gamma} @tab @code{shape}, @code{scale}
@item geometric @tab @code{geom} @tab @code{p}
@item hypergeometric @tab @code{hyper} @tab @code{NR}, @code{NB}, @code{n}
@c in R m, n, k
@item logistic @tab @code{logis} @tab @code{location}, @code{scale}
@item lognormal @tab @code{lnorm} @tab @code{logmean}, @code{logsd}
@item negative binomial @tab @code{nbinom} @tab @code{size}, @code{prob}
@item normal @tab @code{norm} @tab @code{mu}, @code{sigma}
@item Poisson @tab @code{pois} @tab @code{lambda}
@item Student's t @tab @code{t} @tab @code{n}
@item non-central t @tab @code{nt} @tab @code{df}, @code{delta}
@item Studentized range @tab @code{tukey} (*) @tab @code{rr}, @code{cc}, @code{df}
@c in R nranges, nmeans, df
@item uniform @tab @code{unif} @tab @code{a}, @code{b}
@c in R min, max
@item Weibull @tab @code{weibull} @tab @code{shape}, @code{scale}
@item Wilcoxon rank sum @tab @code{wilcox} @tab @code{m}, @code{n}
@item Wilcoxon signed rank @tab @code{signrank} @tab @code{n}
@end multitable
@end quotation

@noindent
Entries marked with an asterisk only have @samp{p} and @samp{q}
functions available, and none of the non-central distributions have
@samp{r} functions.

@noindent
(If remapping is suppressed, the Normal distribution names are
@code{Rf_dnorm4}, @code{Rf_pnorm5} and @code{Rf_qnorm5}.)

Additionally, a @emph{multivariate} RNG for the multinomial distribution is
@example
void rmultinom(int n, double* prob, int K, int* rN)
@end example
@noindent
where @code{K = length(prob)},
@iftex @tex $\pi :=$ @code{prob[]},
@end tex
@end iftex
@eqn{\sum_{j=1}^K \pi_j = 1, sum(prob[.]) == 1}
and @code{rN} must point to a length-@code{K} integer vector
@eqn{n_1 n_2 \ldots n_K, n1 n2 .. nK} where each entry
@eqn{n_j=, nj=}@code{rN[j]} is ``filled'' by a random binomial from
@eqn{Bin(n; \pi_j), Bin(n; prob[j])},
constrained to @eqn{\sum_{j=1}^K n_j = n, sum(rN[.]) == n}.

After calls to @code{dwilcox}, @code{pwilcox} or @code{qwilcox} the
function @code{wilcox_free()} should be called, and similarly
@code{signrank_free()} for the signed rank functions.
@findex wilcox_free
@findex signrank_free
Since @code{wilcox_free()} and @code{signrank_free()} were only added to
@file{Rmath.h} in @R{}@tie{} 4.2.0, their use requires something like
@example
#include "Rmath.h"
#include "Rversion.h"

#if R_VERSION < R_Version(4, 2, 0)
extern void wilcox_free(void);
extern void signrank_free(void);
#endif
@end example

For the negative binomial distribution (@samp{nbinom}), in addition to the
@code{(size, prob)} parametrization, the alternative @code{(size, mu)}
parametrization is provided as well by functions @samp{[dpqr]nbinom_mu()},
see @kbd{?NegBinomial} in @R{}.

Functions @code{dpois_raw(x, *)} and @code{dbinom_raw(x, *)} are versions of the
Poisson and binomial probability mass functions which work continuously in
@code{x}, whereas @code{dbinom(x,*)} and @code{dpois(x,*)} only return non
zero values for integer @code{x}.
@example
@group
double dbinom_raw(double x, double n, double p, double q, int give_log)
double dpois_raw (double x, double lambda, int give_log)
@end group
@end example
Note that @code{dbinom_raw()} returns both @eqn{p, p} and @eqn{q = 1-p,
q = 1-p} which may be advantageous when one of them is close to @eqn{1,
1}.


@node Mathematical functions, Numerical Utilities, Distribution functions, Numerical analysis subroutines
@subsection Mathematical functions

@cindex Gamma function
@cindex Polygamma functions
@deftypefun double gammafn (double @var{x})
@deftypefunx double lgammafn (double @var{x})
@deftypefunx double digamma (double @var{x})
@deftypefunx double trigamma (double @var{x})
@deftypefunx double tetragamma (double @var{x})
@deftypefunx double pentagamma (double @var{x})
@deftypefunx double psigamma (double @var{x}, double @var{deriv})
@deftypefunx void dpsifn (double @var{x}, int @var{n}, int @var{kode}, int @var{m}, double* @var{ans}, int* @var{nz}, int* @var{ierr})
The Gamma function, the natural logarithm of its absolute value and
first four derivatives and the n-th derivative of Psi, the digamma
function, which is the derivative of @code{lgammafn}. In other words,
@code{digamma(x)} is the same as @code{psigamma(x,0)},
@code{trigamma(x) == psigamma(x,1)}, etc.
The underlying workhorse, @code{dpsifn()}, is useful, e.g., when several derivatives of
@eqn{\log\Gamma=, log Gamma=}@code{lgammafn} are desired.  It computes and
returns in @code{ans[]} the length-@var{m} sequence 
@eqn{(-1)^{k+1} / \Gamma(k+1) * \psi^{(k)}(x), (-1)^(k+1) / gamma(k+1) * psi(k;x)} for
@eqn{k = n \ldots n+m-1, k = n ... n+m-1}, where @eqn{\psi^{(k)}(x), psi(k;x)}
is the k-th derivative of @eqn{\psi(x), Psi(x)}, i.e.,
@code{psigamma(x,k)}.  For more details, see the comments in
@file{src/nmath/polygamma.c}.
@end deftypefun

@cindex Beta function
@deftypefun double beta (double @var{a}, double @var{b})
@deftypefunx double lbeta (double @var{a}, double @var{b})
The (complete) Beta function and its natural logarithm.
@end deftypefun

@deftypefun double choose (double @var{n}, double @var{k})
@deftypefunx double lchoose (double @var{n}, double @var{k})
The number of combinations of @var{k} items chosen from from @var{n} and
the natural logarithm of its absolute value, generalized to arbitrary real
@var{n}.  @var{k} is rounded to the nearest integer (with a warning if
needed).
@end deftypefun

@cindex Bessel functions
@deftypefun double bessel_i (double @var{x}, double @var{nu}, double @var{expo})
@deftypefunx double bessel_j (double @var{x}, double @var{nu})
@deftypefunx double bessel_k (double @var{x}, double @var{nu}, double @var{expo})
@deftypefunx double bessel_y (double @var{x}, double @var{nu})
Bessel functions of types I, J, K and Y with index @var{nu}.  For
@code{bessel_i} and @code{bessel_k} there is the option to return
@w{exp(-@var{x}) I(@var{x}; @var{nu})} or @w{exp(@var{x}) K(@var{x};
@var{nu})} if @var{expo} is 2. (Use @code{@var{expo} == 1} for unscaled
values.)
@end deftypefun


@node Numerical Utilities, Mathematical constants, Mathematical functions, Numerical analysis subroutines
@subsection Numerical Utilities
There are a few other numerical utility functions available as entry points.


@deftypefun double R_pow (double @var{x}, double @var{y})
@deftypefunx double R_pow_di (double @var{x}, int @var{i})
@code{R_pow(@var{x}, @var{y})} and @code{R_pow_di(@var{x}, @var{i})}
compute @code{@var{x}^@var{y}} and @code{@var{x}^@var{i}}, respectively
using @code{R_FINITE} checks and returning the proper result (the same
as @R{}) for the cases where @var{x}, @var{y} or @var{i} are 0 or
missing or infinite or @code{NaN}.
@end deftypefun

@deftypefun double log1p (double @var{x})
Computes @code{log(1 + @var{x})} (@emph{log 1 @b{p}lus x}), accurately
even for small @var{x}, i.e., @eqn{|x| \ll 1, |x| << 1}.

This should be provided by your platform, in which case it is not
included in @file{Rmath.h}, but is (probably) in @file{math.h} which
@file{Rmath.h} includes (except under C++, so it may not be declared for
C++98).
@end deftypefun

@deftypefun double log1pmx (double @var{x})
Computes @code{log(1 + @var{x}) - @var{x}} (@emph{log 1 @b{p}lus x @b{m}inus @b{x}}),
accurately even for small @var{x}, i.e., @eqn{|x| \ll 1, |x| << 1}.
@end deftypefun

@deftypefun double log1pexp (double @var{x})
Computes @code{log(1 + exp(@var{x}))} (@emph{log 1 @b{p}lus @b{exp}}),
accurately, notably for large @var{x}, e.g., @eqn{x > 720, x > 720}.
@end deftypefun

@deftypefun double log1mexp (double @var{x})
Computes @code{log(1 - exp(@var{-x}))} (@emph{log 1 @b{m}inus @b{exp}}),
accurately, carefully for two regions of @var{x}, optimally cutting
off at @eqn{\log 2, log 2} (= 0.693147..), using
@code{((-x) > -M_LN2 ? log(-expm1(-x)) : log1p(-exp(-x)))}.
@end deftypefun


@deftypefun double expm1 (double @var{x})
Computes @code{exp(@var{x}) - 1} (@emph{exp x @b{m}inus 1}), accurately
even for small @var{x}, i.e., @eqn{|x| \ll 1, |x| << 1}.

This should be provided by your platform, in which case it is not
included in @file{Rmath.h}, but is (probably) in @file{math.h} which
@file{Rmath.h} includes (except under C++, so it may not be declared for
C++98).
@end deftypefun

@deftypefun double lgamma1p (double @var{x})
Computes @code{log(gamma(@var{x} + 1))} (@emph{log(gamma(1 @b{p}lus x))}),
accurately even for small @var{x}, i.e., @eqn{0 < x < 0.5, 0 < x < 0.5}.
@end deftypefun

@deftypefun double cospi (double @var{x})
Computes @code{cos(pi * x)} (where @code{pi} is 3.14159...),
accurately, notably for half integer @var{x}.

This might be provided by your platform@footnote{It is an optional C11
extension.}, in which case it is not included in @file{Rmath.h}, but is
in @file{math.h} which @file{Rmath.h} includes.  (Ensure that
neither @file{math.h} nor @file{cmath} is included before
@file{Rmath.h} or define
@example
#define __STDC_WANT_IEC_60559_FUNCS_EXT__ 1
@end example
@noindent
before the first inclusion.)
@end deftypefun

@deftypefun double sinpi (double @var{x})
Computes @code{sin(pi * x)} accurately, notably for (half) integer @var{x}.

This might be provided by your platform, in which case it is not
included in @file{Rmath.h}, but is in @file{math.h} which @file{Rmath.h}
includes (but see the comments for @code{cospi}).
@end deftypefun

@deftypefun double Rtanpi (double @var{x})
Computes @code{tan(pi * x)} accurately, notably for integer @var{x}, giving
@var{NaN} for half integer @var{x} and exactly +1 or -1 for (non half)
quarter integers.
@end deftypefun

@deftypefun double tanpi (double @var{x})
Computes @code{tan(pi * x)} accurately for integer @var{x} with possibly
platform dependent behavior for half (and quarter) integers.
This might be provided by your platform, in which case it is not included
in @file{Rmath.h}, but is in @file{math.h} which @file{Rmath.h} includes
(but see the comments for @code{cospi}).
@end deftypefun

@deftypefun double logspace_add (double @var{logx}, double @var{logy})
@deftypefunx double logspace_sub (double @var{logx}, double @var{logy})
@deftypefunx double logspace_sum (const double* @var{logx}, int @var{n})
Compute the log of a sum or difference from logs of terms, i.e., ``x +
y'' as @code{log (exp(@var{logx}) + exp(@var{logy}))} and ``x - y'' as
@code{log (exp(@var{logx}) - exp(@var{logy}))},
and ``sum_i x[i]'' as @code{log (sum[i = 1:@var{n} exp(@var{logx}[i])] )}
without causing unnecessary overflows or throwing away too much accuracy.
@end deftypefun

@deftypefun int imax2 (int @var{x}, int @var{y})
@deftypefunx int imin2 (int @var{x}, int @var{y})
@deftypefunx double fmax2 (double @var{x}, double @var{y})
@deftypefunx double fmin2 (double @var{x}, double @var{y})
Return the larger (@code{max}) or smaller (@code{min}) of two integer or
double numbers, respectively.  Note that @code{fmax2} and @code{fmin2}
differ from C99/C++11's @code{fmax} and @code{fmin} when one of the
arguments is a @code{NaN}: these versions return @code{NaN}.
@end deftypefun

@deftypefun double sign (double @var{x})
Compute the @emph{signum} function, where sign(@var{x}) is 1, 0, or
@math{-1}, when @var{x} is positive, 0, or negative, respectively, and
@code{NaN} if @code{x} is a @code{NaN}.
@end deftypefun

@deftypefun double fsign (double @var{x}, double @var{y})
Performs ``transfer of sign'' and is defined as @eqn{|x| *
\hbox{sign}(y), |x| * sign(y)}.
@end deftypefun

@deftypefun double fprec (double @var{x}, double @var{digits})
Returns the value of @var{x} rounded to @var{digits} decimal digits
(after the decimal point).

This is the function used by @R{}'s @code{signif()}.
@end deftypefun

@deftypefun double fround (double @var{x}, double @var{digits})
Returns the value of @var{x} rounded to @var{digits} @emph{significant}
decimal digits.

This is the function used by @R{}'s @code{round()}.  (Note that C99/C++11
provide a @code{round} function but C++98 need not.)
@end deftypefun

@deftypefun double ftrunc (double @var{x})
Returns the value of @var{x} truncated (to an integer value) towards
zero.
@end deftypefun

@node Mathematical constants,  , Numerical Utilities, Numerical analysis subroutines
@subsection Mathematical constants
@findex M_E
@findex M_PI
@c maybe not all into the index ...

@R{} has a set of commonly used mathematical constants encompassing
constants defined by POSIX and usually found in headers @file{math.h}
and @file{cmath}, as well as further ones that are used in statistical
computations.  These are defined to (at least) 30 digits accuracy in
@file{Rmath.h}.  The following definitions use @code{ln(x)} for the
natural logarithm (@code{log(x)} in @R{}).

@quotation
@multitable {Name can be long}  {Definition (needs space)}  {0.123456789012345678 ...}
@headitem Name @tab Definition (@code{ln = log}) @tab round(@emph{value}, 7)
@c SVID & X/Open Constants -- names from Solaris math.h :
@item @code{M_E} @tab @math{e} @tab 2.7182818
@item @code{M_LOG2E} @tab log2(@math{e}) @tab 1.4426950
@item @code{M_LOG10E} @tab log10(@math{e}) @tab 0.4342945
@item @code{M_LN2} @tab ln(2) @tab 0.6931472
@item @code{M_LN10} @tab ln(10) @tab 2.3025851
@item @code{M_PI} @tab @eqn{\pi, pi}   @tab 3.1415927
@item @code{M_PI_2} @tab @eqn{\pi/2, pi/2} @tab 1.5707963
@item @code{M_PI_4} @tab @eqn{\pi/4, pi/4} @tab 0.7853982
@item @code{M_1_PI} @tab @eqn{1/\pi, 1/pi} @tab 0.3183099
@item @code{M_2_PI} @tab @eqn{2/\pi, 2/pi} @tab 0.6366198
@item @code{M_2_SQRTPI} @tab 2/sqrt(@eqn{\pi, pi}) @tab 1.1283792
@item @code{M_SQRT2} @tab sqrt(2) @tab 1.4142136
@item @code{M_SQRT1_2} @tab 1/sqrt(2) @tab 0.7071068
@c R-specific ones
@item @code{M_SQRT_3} @tab sqrt(3) @tab 1.7320508
@item @code{M_SQRT_32} @tab sqrt(32) @tab 5.6568542
@item @code{M_LOG10_2} @tab log10(2) @tab 0.3010300
@item @code{M_2PI} @tab @eqn{2\pi, 2*pi} @tab 6.2831853
@item @code{M_SQRT_PI} @tab sqrt(@eqn{\pi, pi}) @tab 1.7724539
@item @code{M_1_SQRT_2PI} @tab 1/sqrt(@eqn{2\pi, 2*pi}) @tab 0.3989423
@item @code{M_SQRT_2dPI} @tab sqrt(2/@eqn{\pi, pi}) @tab 0.7978846
@item @code{M_LN_SQRT_PI} @tab ln(sqrt(@eqn{\pi, pi})) @tab 0.5723649
@item @code{M_LN_SQRT_2PI} @tab ln(sqrt(@eqn{2\pi, 2*pi})) @tab 0.9189385
@item @code{M_LN_SQRT_PId2} @tab ln(sqrt(@eqn{\pi, pi}/2)) @tab 0.2257914
@end multitable
@end quotation

There are a set of constants (@code{PI}, @code{DOUBLE_EPS}) (and so on)
defined (unless @code{STRICT_R_HEADERS} is defined) in the included
header @file{R_ext/Constants.h}, mainly for compatibility with @Sl{}.
All but @code{PI} are deprecated and should be replaced by the C99/C++11
versions used in that file.

@findex TRUE
@findex FALSE
Further, the included header @file{R_ext/Boolean.h} has enumeration
constants @code{TRUE} and @code{FALSE} of type @code{Rboolean} in
order to provide a way of using ``logical'' variables in C consistently.
This can conflict with other software: for example it conflicts with the
headers in IJG's @code{jpeg-9} (but not earlier versions).


@node Optimization, Integration, Numerical analysis subroutines, The R API
@section Optimization
@cindex optimization

The C code underlying @code{optim} can be accessed directly.  The user
needs to supply a function to compute the function to be minimized, of
the type

@findex optimfn
@example
typedef double optimfn(int n, double *par, void *ex);
@end example

@noindent
where the first argument is the number of parameters in the second
argument.  The third argument is a pointer passed down from the calling
routine, normally used to carry auxiliary information.

Some of the methods also require a gradient function

@findex optimgr
@example
typedef void optimgr(int n, double *par, double *gr, void *ex);
@end example

@noindent
which passes back the gradient in the @code{gr} argument.  No function
is provided for finite-differencing, nor for approximating the Hessian
at the result.

The interfaces (defined in header @file{R_ext/Applic.h}) are

@itemize @bullet
@item Nelder Mead:
@findex nmmin
@example
void nmmin(int n, double *xin, double *x, double *Fmin, optimfn fn,
           int *fail, double abstol, double intol, void *ex,
           double alpha, double beta, double gamma, int trace,
           int *fncount, int maxit);
@end example

@item BFGS:
@findex vmmin
@example
void vmmin(int n, double *x, double *Fmin,
           optimfn fn, optimgr gr, int maxit, int trace,
           int *mask, double abstol, double reltol, int nREPORT,
           void *ex, int *fncount, int *grcount, int *fail);
@end example

@item Conjugate gradients:
@findex cgmin
@example
void cgmin(int n, double *xin, double *x, double *Fmin,
           optimfn fn, optimgr gr, int *fail, double abstol,
           double intol, void *ex, int type, int trace,
           int *fncount, int *grcount, int maxit);
@end example

@item Limited-memory BFGS with bounds:
@findex lbfgsb
@example
void lbfgsb(int n, int lmm, double *x, double *lower,
            double *upper, int *nbd, double *Fmin, optimfn fn,
            optimgr gr, int *fail, void *ex, double factr,
            double pgtol, int *fncount, int *grcount,
            int maxit, char *msg, int trace, int nREPORT);
@end example

@item Simulated annealing:
@findex samin
@example
void samin(int n, double *x, double *Fmin, optimfn fn, int maxit,
           int tmax, double temp, int trace, void *ex);
@end example

@end itemize

@noindent
Many of the arguments are common to the various methods.  @code{n} is
the number of parameters, @code{x} or @code{xin} is the starting
parameters on entry and @code{x} the final parameters on exit, with
final value returned in @code{Fmin}.  Most of the other parameters can
be found from the help page for @code{optim}: see the source code
@file{src/appl/lbfgsb.c} for the values of @code{nbd}, which
specifies which bounds are to be used.


@node Integration, Utility functions, Optimization, The R API
@section Integration
@cindex integration

The C code underlying @code{integrate} can be accessed directly.  The
user needs to supply a @emph{vectorizing} C function to compute the
function to be integrated, of the type

@findex integr_fn
@example
typedef void integr_fn(double *x, int n, void *ex);
@end example

@noindent
where @code{x[]} is both input and output and has length @code{n}, i.e.,
a C function, say @code{fn}, of type @code{integr_fn} must basically do
@code{for(i in 1:n) x[i] := f(x[i], ex)}.  The vectorization requirement
can be used to speed up the integrand instead of calling it @code{n}
times.  Note that in the current implementation built on QUADPACK,
@code{n} will be either 15 or 21.  The @code{ex} argument is a pointer
passed down from the calling routine, normally used to carry auxiliary
information.

There are interfaces (defined in header @file{R_ext/Applic.h}) for
integrals over finite and infinite intervals (or ``ranges'' or
``integration boundaries'').

@itemize @bullet
@item Finite:
@findex Rdqags
@example
void Rdqags(integr_fn f, void *ex, double *a, double *b,
            double *epsabs, double *epsrel,
            double *result, double *abserr, int *neval, int *ier,
            int *limit, int *lenw, int *last,
            int *iwork, double *work);
@end example

@item Infinite:
@findex Rdqagi
@example
void Rdqagi(integr_fn f, void *ex, double *bound, int *inf,
            double *epsabs, double *epsrel,
            double *result, double *abserr, int *neval, int *ier,
            int *limit, int *lenw, int *last,
            int *iwork, double *work);
@end example

@end itemize

@noindent
Only the 3rd and 4th argument differ for the two integrators; for the
finite range integral using @code{Rdqags}, @code{a} and @code{b} are the
integration interval bounds, whereas for an infinite range integral using
@code{Rdqagi}, @code{bound} is the finite bound of the integration (if
the integral is not doubly-infinite) and @code{inf} is a code indicating
the kind of integration range,

@table @code
@item inf = 1
      corresponds to (bound, +Inf),
@item inf = -1
      corresponds to (-Inf, bound),
@item inf = 2
      corresponds to (-Inf, +Inf),
@end table

@code{f} and @code{ex} define the integrand function, see above;
@code{epsabs} and @code{epsrel} specify the absolute and relative
accuracy requested, @code{result}, @code{abserr} and @code{last} are the
output components @code{value}, @code{abs.err} and @code{subdivisions}
of the @R{} function integrate, where @code{neval} gives the number of
integrand function evaluations, and the error code @code{ier} is
translated to @R{}'s @code{integrate() $ message}, look at that function
definition.  @code{limit} corresponds to @code{integrate(...,
subdivisions = *)}.  It seems you should always define the two work
arrays and the length of the second one as

@example
    lenw = 4 * limit;
    iwork =   (int *) R_alloc(limit, sizeof(int));
    work = (double *) R_alloc(lenw,  sizeof(double));
@end example

The comments in the source code in @file{src/appl/integrate.c} give
more details, particularly about reasons for failure (@code{ier >= 1}).


@node Utility functions, Re-encoding, Integration, The R API
@section Utility functions
@cindex Sort functions from C

@R{} has a fairly comprehensive set of sort routines which are made
available to users' C code.
The following is declared in header file @file{Rinternals.h}.

@deftypefun void  R_orderVector (int* @var{indx}, int @var{n}, SEXP @var{arglist}, Rboolean @var{nalast}, Rboolean @var{decreasing})
@deftypefunx void R_orderVector1 (int* @var{indx}, int @var{n}, SEXP @var{x}, Rboolean @var{nalast}, Rboolean @var{decreasing})

@code{R_orderVector()} corresponds to @R{}'s @code{order(..., na.last, decreasing)}.
More specifically, @code{indx <- order(x, y, na.last, decreasing)} corresponds to
@code{R_orderVector(indx, n, Rf_lang2(x, y), nalast, decreasing)} and for
three vectors, @code{Rf_lang3(x,y,z)} is used as @var{arglist}.

Both @code{R_orderVector} and @code{R_orderVector1} assume the vector
@code{indx} to be allocated to length @eqn{\ge n, >= n}.  On return,
@code{indx[]} contains a permutation of @code{0:(n-1)}, i.e., 0-based C
indices (and not 1-based @R{} indices, as @R{}'s @code{order()}).

When ordering only one vector, @code{R_orderVector1} is faster and
corresponds (but is 0-based) to @R{}'s @code{indx <- order(x, na.last,
decreasing)}.  It was added in @R{} 3.3.0.
@end deftypefun

All other sort routines are declared in header file
@file{R_ext/Utils.h} (included by @file{R.h}) and include the following.

@deftypefun void R_isort (int* @var{x}, int @var{n})
@deftypefunx void R_rsort (double* @var{x}, int @var{n})
@deftypefunx void R_csort (Rcomplex* @var{x}, int @var{n})
@deftypefunx void rsort_with_index (double* @var{x}, int* @var{index}, int @var{n})
The first three sort integer, real (double) and complex data
respectively.  (Complex numbers are sorted by the real part first then
the imaginary part.)  @code{NA}s are sorted last.

@code{rsort_with_index} sorts on @var{x}, and applies the same
permutation to @var{index}.  @code{NA}s are sorted last.
@end deftypefun

@deftypefun void revsort (double* @var{x}, int* @var{index}, int @var{n})
Is similar to @code{rsort_with_index} but sorts into decreasing order,
and @code{NA}s are not handled.
@end deftypefun

@deftypefun void iPsort (int* @var{x}, int @var{n}, int @var{k})
@deftypefunx void rPsort (double* @var{x}, int @var{n}, int @var{k})
@deftypefunx void cPsort (Rcomplex* @var{x}, int @var{n}, int @var{k})
These all provide (very) partial sorting: they permute @var{x} so that
@code{@var{x}[@var{k}]} is in the correct place with smaller values to
the left, larger ones to the right.
@end deftypefun

@deftypefun  void R_qsort   (double *@var{v}, size_t @var{i}, size_t @var{j})
@deftypefunx void R_qsort_I (double *@var{v}, int *@var{I}, int @var{i}, int @var{j})
@deftypefunx void R_qsort_int   (int *@var{iv}, size_t @var{i}, size_t @var{j})
@deftypefunx void R_qsort_int_I (int *@var{iv}, int *@var{I}, int @var{i}, int @var{j})


These routines sort @code{@var{v}[@var{i}:@var{j}]} or
@code{@var{iv}[@var{i}:@var{j}]} (using 1-indexing, i.e.,
@code{@var{v}[1]} is the first element) calling the quicksort algorithm
as used by @R{}'s @code{sort(v, method = "quick")} and documented on the
help page for the @R{} function @code{sort}.  The @code{..._I()}
versions also return the @code{sort.index()} vector in @code{I}.  Note
that the ordering is @emph{not} stable, so tied values may be permuted.

Note that @code{NA}s are not handled (explicitly) and you should
use different sorting functions if @code{NA}s can be present.
@end deftypefun

@deftypefun subroutine qsort4 (double precision @var{v}, integer @var{indx}, integer @var{ii}, integer @var{jj})
@deftypefunx subroutine qsort3 (double precision @var{v}, integer @var{ii}, integer @var{jj})

The Fortran interface routines for sorting double precision vectors are
@code{qsort3} and @code{qsort4}, equivalent to @code{R_qsort} and
@code{R_qsort_I}, respectively.
@end deftypefun

@deftypefun void R_max_col (double* @var{matrix}, int* @var{nr}, int* @var{nc}, int* @var{maxes}, int* @var{ties_meth})
Given the @var{nr} by @var{nc} matrix @code{matrix} in column-major
(``Fortran'')
order, @code{R_max_col()} returns in @code{@var{maxes}[@var{i}-1]} the
column number of the maximal element in the @var{i}-th row (the same as
@R{}'s @code{max.col()} function).  In the case of ties (multiple maxima),
@code{*ties_meth} is an integer code in @code{1:3} determining the method:
1 = ``random'', 2 = ``first'' and 3 = ``last''.
See @R{}'s help page @code{?max.col}.
@end deftypefun

@deftypefun int findInterval (double* @var{xt}, int @var{n}, double @var{x}, Rboolean @var{rightmost_closed}, Rboolean @var{all_inside}, int @var{ilo}, int* @var{mflag})
@deftypefunx int findInterval2(double* @var{xt}, int @var{n}, double @var{x}, Rboolean @var{rightmost_closed}, Rboolean @var{all_inside}, Rboolean @var{left_open}, int @var{ilo}, int* @var{mflag})
Given the ordered vector @var{xt} of length @var{n}, return the interval
or index of @var{x} in @code{@var{xt}[]}, typically max(@math{i}; @eqn{1
\le i \le @var{n}, 1 <= i <= @var{n}} & @math{@var{xt}[i]} @eqn{\le, <=}
@var{x}) where we use 1-indexing as in @R{} and Fortran (but not C).  If
@var{rightmost_closed} is true, also returns @math{@var{n}-1} if @var{x}
equals @math{@var{xt}[@var{n}]}.  If @var{all_inside} is not 0, the
result is coerced to lie in @code{1:(@var{n}-1)} even when @var{x} is
outside the @var{xt}[] range.  On return, @code{*@var{mflag}} equals
@math{-1} if @var{x} < @var{xt}[1], @math{+1} if @var{x} >=
@var{xt}[@var{n}], and 0 otherwise.

The algorithm is particularly fast when @var{ilo} is set to the last
result of @code{findInterval()} and @var{x} is a value of a sequence which
is increasing or decreasing for subsequent calls.

@code{findInterval2()} is a generalization of @code{findInterval()},
with an extra @code{Rboolean} argument @var{left_open}.  Setting
@code{left_open = TRUE} basically replaces all left-closed right-open
intervals @eqn{[s, t)} by left-open ones @eqn{(s, t]}, see the help page
of @R{} function @code{findInterval} for details.

There is also an @code{F77_CALL(interv)()} version of
@code{findInterval()} with the same arguments, but all pointers.
@end deftypefun

A system-independent interface to produce the name of a temporary
file is provided as

@deftypefun {char *} R_tmpnam (const char *@var{prefix}, const char *@var{tmpdir})
@deftypefunx {char *} R_tmpnam2 (const char *@var{prefix}, const char *@var{tmpdir}, const char *@var{fileext})
@deftypefunx {void} R_free_tmpnam (char *@var{name})
Return a pathname for a temporary file with name beginning with
@var{prefix} and ending with @var{fileext} in directory @var{tmpdir}.
A @code{NULL} prefix or extension is replaced by @code{""}.  Note that
the return value is dynamically allocated and should be freed using
@code{R_free_tmpnam} when no longer needed (unlike the
system call @code{tmpnam}). Freeing the result using @code{free} is no
longer recommended. 
@end deftypefun

@c ----

There is also the internal function used to expand file names in several
@R{} functions, and called directly by @code{path.expand}.

@deftypefun {const char *} R_ExpandFileName (const char *@var{fn})
Expand a path name @var{fn} by replacing a leading tilde by the user's
home directory (if defined).  The precise meaning is platform-specific;
it will usually be taken from the environment variable @env{HOME} if
this is defined.
@end deftypefun

For historical reasons there are Fortran interfaces to functions
@code{D1MACH} and @code{I1MACH}.  These can be called from C code as
e.g.@: @code{F77_CALL(d1mach)(4)}.  Note that these are emulations of
the original functions by Fox, Hall and Schryer on NetLib at
@uref{https://www.netlib.org/slatec/src/} for IEC 60559 arithmetic
(required by @R{}).

@node Re-encoding, Condition handling and cleanup code, Utility functions, The R API
@section Re-encoding

@R{} has its own C-level interface to the encoding conversion
capabilities provided by @code{iconv} because there are
incompatibilities between the declarations in different implementations
of @code{iconv}.

These are declared in header file @file{R_ext/Riconv.h}.

@deftypefun {void *} Riconv_open (const char *@var{to}, const char *@var{from})
@end deftypefun
Set up a pointer to an encoding object to be used to convert between two
encodings: @code{""} indicates the current locale.

@deftypefun size_t Riconv (void *@var{cd}, const char **@var{inbuf}, size_t *@var{inbytesleft}, char  **@var{outbuf}, size_t *@var{outbytesleft})
@end deftypefun
Convert as much as possible of @code{inbuf} to @code{outbuf}.  Initially
the @code{size_t} variables indicate the number of bytes available in the
buffers, and they are updated (and the @code{char} pointers are updated
to point to the next free byte in the buffer).  The return value is the
number of characters converted, or @code{(size_t)-1} (beware:
@code{size_t} is usually an unsigned type).  It should be safe to assume
that an error condition sets @code{errno} to one of @code{E2BIG} (the
output buffer is full), @code{EILSEQ} (the input cannot be converted,
and might be invalid in the encoding specified) or @code{EINVAL} (the
input does not end with a complete multi-byte character).

@deftypefun int Riconv_close (void * @var{cd})
@end deftypefun
Free the resources of an encoding object.


@node Condition handling and cleanup code, Allowing interrupts, Re-encoding, The R API
@section Condition handling and cleanup code
@cindex Condition handling
@cindex Cleanup code
@cindex Error handling

Three functions are available for establishing condition handlers from
within C code:

@example
#include <Rinternals.h>

SEXP R_tryCatchError(SEXP (*fun)(void *data), void *data,
		     SEXP (*hndlr)(SEXP cond, void *hdata), void *hdata);

SEXP R_tryCatch(SEXP (*fun)(void *data), void *data,
		SEXP,
		SEXP (*hndlr)(SEXP cond, void *hdata), void *hdata,
		void (*clean)(void *cdata), void *cdata);
SEXP R_withCallingErrorHandler(SEXP (*fun)(void *data), void *data,
			       SEXP (*hndlr)(SEXP cond, void *hdata), void *hdata)
@end example
@findex R_tryCatchError
@findex R_tryCatch
@findex R_withCallingErrorHandler

@code{R_tryCatchError} establishes an exiting handler for conditions
inheriting form class @code{error}.

@code{R_tryCatch} can be used to establish a handler for other
conditions and to register a cleanup action. The conditions to be
handled are specified as a character vector (@code{STRSXP}).
A @code{NULL} pointer can be passed as @code{fun} or @code{clean}
if condition handling or cleanup are not needed.

These are currently implemented using the R-level @code{tryCatch}
mechanism so are subject to some overhead.

@code{R_withCallingErrorHandler} establishes a calling handler for
conditions inheriting form class @code{error}.  It establishes the
handler without calling back into @R{} and will therefore be more
efficient.

The function @code{R_UnwindProtect} can be used to ensure that a cleanup
action takes place on ordinary return as well as on a non-local transfer
of control, which R implements as a @code{longjmp}.

@example
SEXP R_UnwindProtect(SEXP (*fun)(void *data), void *data,
                     void (*clean)(void *data, Rboolean jump), void *cdata,
                     SEXP cont);
@end example
@findex R_UnwindProtect

@code{R_UnwindProtect} can be used in two ways.  The simper usage,
suitable for use in C code, passes @code{NULL} for the @code{cont}
argument.  @code{R_UnwindProtect} will call @code{fun(data)}.  If
@code{fun} returns a value, then @code{R_UnwindProtect} calls
@code{clean(cleandata, FALSE)} before returning the value returned by
@code{fun}. If @code{fun} executes a non-local transfer of control, then
@code{clean(cleandata, TRUE)} is called, and the non-local transfer of
control is resumed.

The second use pattern, suitable to support C++ stack unwinding, uses
two additional functions:

@example
SEXP R_MakeUnwindCont();
void NORET R_ContinueUnwind(SEXP cont);
@end example
@findex R_MakeUnwindCont
@findex R_ContinueUnwind

@code{R_MakeUnwindCont} allocates a @emph{continuation token}
@code{cont} to pass to @code{R_UnwindProtect}. This token should be
protected with @code{PROTECT} before calling
@code{R_UnwindProtect}. When the @code{clean} function is called with
@code{jump == TRUE}, indicating that R is executing a non-local transfer
of control, it can throw a C++ exception to a C++ @code{catch} outside
the C++ code to be unwound, and then use the continuation token in the a
call @code{R_ContinueUnwind(cont)} to resume the non-local transfer of
control within R.

@node Allowing interrupts, Platform and version information, Condition handling and cleanup code, The R API
@section Allowing interrupts
@cindex Interrupts

No part of @R{} can be interrupted whilst running long computations in
compiled code, so programmers should make provision for the code to be
interrupted at suitable points by calling from C

@example
#include <R_ext/Utils.h>

void R_CheckUserInterrupt(void);
@end example
@findex R_CheckUserInterrupt

@noindent
and from Fortran

@example
subroutine rchkusr()
@end example
@findex rchkusr

These check if the user has requested an interrupt, and if so branch to
@R{}'s error signaling functions.

Note that it is possible that the code behind one of the entry points
defined here if called from your C or Fortran code could be interruptible
or generate an error and so not return to your code.


@node Platform and version information, Inlining C functions, Allowing interrupts, The R API
@section Platform and version information
@cindex Version information from C
@cindex OpenMP
@findex R_Version

The header files define @code{USING_R}, which can be used to test if
the code is indeed being used with @R{}.

Header file @file{Rconfig.h} (included by @file{R.h}) is used to define
platform-specific macros that are mainly for use in other header files.
The macro @code{WORDS_BIGENDIAN} is defined on
big-endian@footnote{@uref{https://en.wikipedia.org/wiki/Endianness}.}
systems (e.g.@: most OSes on Sparc and PowerPC hardware) and not on
little-endian systems (nowadays all the commoner @R{} platforms).  It
can be useful when manipulating binary files.  NB: these macros apply
only to the C compiler used to build @R{}, not necessarily to another C
or C++ compiler.

Header file @file{Rversion.h} (@strong{not} included by @file{R.h})
defines a macro @code{R_VERSION} giving the version number encoded as an
integer, plus a macro @code{R_Version} to do the encoding.  This can be
used to test if the version of @R{} is late enough, or to include
back-compatibility features.  For protection against very old versions
of @R{} which did not have this macro, use a construction such as

@example
@group
#if defined(R_VERSION) && R_VERSION >= R_Version(3, 1, 0)
  ...
#endif
@end group
@end example

More detailed information is available in the macros @code{R_MAJOR},
@code{R_MINOR}, @code{R_YEAR}, @code{R_MONTH} and @code{R_DAY}: see the
header file @file{Rversion.h} for their format.  Note that the minor
version includes the patchlevel (as in @samp{2.2}).

Packages which use @code{alloca} need to ensure it is defined: as it is
part of neither C nor POSIX there is no standard way to do so.  One can
use

@example
#include <Rconfig.h> // for HAVE_ALLOCA_H
#ifdef __GNUC__
// this covers gcc, clang, icc
# undef alloca
# define alloca(x) __builtin_alloca((x))
#elif defined(HAVE_ALLOCA_H)
// needed for native compilers on Solaris and AIX
# include <alloca.h>
#endif
@end example

@noindent
(and this should be included before standard C headers such as
@file{stdlib.h}, since on some platforms these include @file{malloc.h}
which may have a conflicting definition), which suffices for known @R{}
platforms.

@node Inlining C functions, Controlling visibility, Platform and version information, The R API
@section Inlining C functions
@findex R_INLINE

The C99 keyword @code{inline} should be recognized by all compilers
nowadays used to build @R{}.  Portable code which might be used with
earlier versions of @R{} can be written using the macro @code{R_INLINE}
(defined in file @file{Rconfig.h} included by @file{R.h}), as for
example from package @CRANpkg{cluster}

@example
#include <R.h>

static R_INLINE int ind_2(int l, int j)
@{
...
@}
@end example

Be aware that using inlining with functions in more than one compilation
unit is almost impossible to do portably, see
@uref{https://www.greenend.org.uk/rjk/tech/inline.html}, so this usage
is for @code{static} functions as in the example.  All the @R{}
configure code has checked is that @code{R_INLINE} can be used in a
single C file with the compiler used to build @R{}.  We recommend that
packages making extensive use of inlining include their own configure
code.

@node Controlling visibility, Standalone Mathlib, Inlining C functions, The R API
@section Controlling visibility
@cindex Visibility

Header @file{R_ext/Visibility.h} has some definitions for controlling the
visibility of entry points.  These are only effective when
@samp{HAVE_VISIBILITY_ATTRIBUTE} is defined -- this is checked when @R{}
is configured and recorded in header @file{Rconfig.h} (included by
@file{R_ext/Visibility.h}).  It is often defined on modern Unix-alikes
with a recent compiler@footnote{It is defined by the Intel compilers,
but also hides unsatisfied references and so cannot be used with @R{}.
It is not supported by the AIX nor Solaris compilers.}, but not
supported on macOS nor Windows.  Minimizing the visibility of symbols in
a shared library will both speed up its loading (unlikely to be
significant) and reduce the possibility of linking to other entry points
of the same name.

C/C++ entry points prefixed by @code{attribute_hidden} will not be
visible in the shared object.  There is no comparable mechanism for
Fortran entry points, but there is a more comprehensive scheme used by,
for example package @pkg{stats}.  Most compilers which allow control of
visibility will allow control of visibility for all symbols @emph{via} a
flag, and where known the flag is encapsulated in the macros
@samp{C_VISIBILITY}, @samp{CXX_VISIBILITY}@footnote{This applies to the
compiler for the default C++ dialect (currently C++11) and not
necessarily to other dialects.} and @samp{F_VISIBILITY} for C, C++ and
Fortran compilers.@footnote{In some cases Fortran compilers accept the
flag but do not actually hide their symbols.}  These are defined in
@file{etc/Makeconf} and so available for normal compilation of package
code.  For example, @file{src/Makevars} could include some of

@example
PKG_CFLAGS=$(C_VISIBILITY)
PKG_CXXFLAGS=$(CXX_VISIBILITY)
PKG_FFLAGS=$(F_VISIBILITY)
@end example

This would end up with @strong{no} visible entry points, which would be
pointless.  However, the effect of the flags can be overridden by using
the @code{attribute_visible} prefix.  A shared object which registers
its entry points needs only for have one visible entry point, its
initializer, so for example package @pkg{stats} has

@example
void attribute_visible R_init_stats(DllInfo *dll)
@{
    R_registerRoutines(dll, CEntries, CallEntries, FortEntries, NULL);
    R_useDynamicSymbols(dll, FALSE);
...
@}
@end example

Because the @samp{C_VISIBILITY} mechanism is only useful in conjunction
with @code{attribute_visible}, it is not enabled unless
@samp{HAVE_VISIBILITY_ATTRIBUTE} is defined.  The usual visibility flag
is @option{-fvisibility=hidden}: some compilers also support
@option{-fvisibility-inlines-hidden} which can be used by overriding
@samp{C_VISIBILITY} and @samp{CXX_VISIBILITY} in @file{config.site} when
building @R{}, or editing @file{etc/Makeconf} in the @R{} installation.

Note that @command{configure} only checks that visibility attributes and
flags are accepted, not that they actually hide symbols.

The visibility mechanism is not available on Windows, but there is an
equally effective way to control which entry points are visible, by
supplying a definitions file
@file{@var{pkgnme}/src/@var{pkgname}-win.def}: only entry points
listed in that file will be visible.  Again using @pkg{stats} as an
example, it has

@example
LIBRARY stats.dll
EXPORTS
 R_init_stats
@end example

@node Standalone Mathlib, Organization of header files, Controlling visibility, The R API
@section Using these functions in your own C code

It is possible to build @code{Mathlib}, the @R{} set of mathematical
functions documented in @file{Rmath.h}, as a standalone library
@file{libRmath} under both Unix-alikes and Windows.  (This includes the
functions documented in @ref{Numerical analysis subroutines} as from
that header file.)

The library is not built automatically when @R{} is installed, but can
be built in the directory @file{src/nmath/standalone} in the @R{}
sources: see the file @file{README} there.  To use the code in your own
C program include

@example
@group
#define MATHLIB_STANDALONE
#include <Rmath.h>
@end group
@end example

@noindent
and link against @samp{-lRmath} (and perhaps @samp{-lm}).  There is an
example file @file{test.c}.

A little care is needed to use the random-number routines. You will
need to supply the uniform random number generator

@example
double unif_rand(void)
@end example

@noindent
or use the one supplied (and with a dynamic library or DLL you will have
to use the one supplied, which is the Marsaglia-multicarry with an entry
points

@example
set_seed(unsigned int, unsigned int)
@end example

@noindent
to set its seeds and

@example
get_seed(unsigned int *, unsigned int *)
@end example

@noindent
to read the seeds).

@node Organization of header files,  , Standalone Mathlib, The R API
@section Organization of header files

The header files which @R{} installs are in directory
@file{@var{R_INCLUDE_DIR}} (default @file{@var{R_HOME}/include}).  This
currently includes

@quotation
@multitable @columnfractions 0.30 0.55
@item @file{R.h} @tab includes many other files
@item @file{Rinternals.h} @tab definitions for using @R{}'s internal
structures
@item @file{Rdefines.h} @tab macros for an @Sl{}-like interface to the
above (no longer maintained)
@item @file{Rmath.h} @tab standalone math library
@item @file{Rversion.h} @tab @R{} version information
@item @file{Rinterface.h} @tab for add-on front-ends (Unix-alikes only)
@item @file{Rembedded.h} @tab for add-on front-ends
@item @file{R_ext/Applic.h} @tab optimization and integration
@item @file{R_ext/BLAS.h} @tab C definitions for BLAS routines
@item @file{R_ext/Callbacks.h} @tab C (and R function) top-level task
handlers
@item @file{R_ext/GetX11Image.h} @tab X11Image interface used by package
@pkg{trkplot}
@item @file{R_ext/Lapack.h} @tab C definitions for some LAPACK routines
@item @file{R_ext/Linpack.h} @tab C definitions for some LINPACK
routines, not all of which are included in @R{}
@item @file{R_ext/Parse.h} @tab a small part of @R{}'s parse interface:
not part of the stable API.
@item @file{R_ext/RStartup.h} @tab for add-on front-ends
@item @file{R_ext/Rdynload.h} @tab needed to register compiled code in
packages
@item @file{R_ext/Riconv.h} @tab interface to @code{iconv}
@item @file{R_ext/Visibility.h} @tab definitions controlling visibility
@item @file{R_ext/eventloop.h} @tab for add-on front-ends and for
packages that need to share in the @R{} event loops (not Windows)
@end multitable
@end quotation

The following headers are included by @file{R.h}:

@quotation
@multitable @columnfractions 0.30 0.55
@item @file{Rconfig.h} @tab configuration info that is made available
@item @file{R_ext/Arith.h} @tab handling for @code{NA}s, @code{NaN}s,
@code{Inf}/@code{-Inf}
@item @file{R_ext/Boolean.h} @tab @code{TRUE}/@code{FALSE} type
@item @file{R_ext/Complex.h} @tab C typedefs for @R{}'s @code{complex}
@item @file{R_ext/Constants.h} @tab constants
@item @file{R_ext/Error.h} @tab error signaling
@item @file{R_ext/Memory.h} @tab memory allocation
@item @file{R_ext/Print.h} @tab @code{Rprintf} and variations.
@item @file{R_ext/RS.h} @tab definitions common to @file{R.h} and the former
@file{S.h}, including @code{F77_CALL} etc.
@item @file{R_ext/Random.h} @tab random number generation
@item @file{R_ext/Utils.h} @tab sorting and other utilities
@item @file{R_ext/libextern.h} @tab definitions for exports from
@file{R.dll} on Windows.
@end multitable
@end quotation

The graphics systems are exposed in headers
@file{R_ext/GraphicsEngine.h}, @file{R_ext/GraphicsDevice.h} (which it
includes) and @file{R_ext/QuartzDevice.h}. Facilities for defining
custom connection implementations are provided in
@file{R_ext/Connections.h}, but make sure you consult the file before
use.

Let us re-iterate the advice to include system headers before the @R{}
header files, especially @file{Rinternals.h} (included by
@file{Rdefines.h}) and @file{Rmath.h}, which redefine names which may be
used in system headers (fewer if @samp{R_NO_REMAP} is defined before
inclusion, or @samp{R_NO_REMAP_RMATH} for @file{Rmath.h}).

@node Generic functions and methods, Linking GUIs and other front-ends to R, The R API, Top
@chapter Generic functions and methods
@cindex Generic functions
@cindex Method functions

@R{} programmers will often want to add methods for existing generic
functions, and may want to add new generic functions or make existing
functions generic.  In this chapter we give guidelines for doing so,
with examples of the problems caused by not adhering to them.

This chapter only covers the `informal' class system copied from S3,
and not with the S4 (formal) methods of package @pkg{methods}.

First, a @emph{caveat}: a function named @code{@var{gen}.@var{cl}} will
be invoked by the generic @code{@var{gen}} for class @code{@var{cl}}, so
do not name functions in this style unless they are intended to be
methods.

The key function for methods is @code{NextMethod}, which dispatches the
next method.  It is quite typical for a method function to make a few
changes to its arguments, dispatch to the next method, receive the
results and modify them a little.  An example is

@example
@group
t.data.frame <- function(x)
@{
    x <- as.matrix(x)
    NextMethod("t")
@}
@end group
@end example

@noindent
Note that the example above works because there is a @emph{next} method,
the default method, not that a new method is selected when the class is
changed.

@emph{Any} method a programmer writes may be invoked from another method
by @code{NextMethod}, @emph{with the arguments appropriate to the
previous method}.  Further, the programmer cannot predict which method
@code{NextMethod} will pick (it might be one not yet dreamt of), and the
end user calling the generic needs to be able to pass arguments to the
next method.  For this to work

@quotation
@emph{A method must have all the arguments of the generic, including
@code{@dots{}} if the generic does.}
@end quotation

It is a grave misunderstanding to think that a method needs only to
accept the arguments it needs.  The original S version of
@code{predict.lm} did not have a @code{@dots{}} argument, although
@code{predict} did.  It soon became clear that @code{predict.glm} needed
an argument @code{dispersion} to handle over-dispersion.  As
@code{predict.lm} had neither a @code{dispersion} nor a @code{@dots{}}
argument, @code{NextMethod} could no longer be used.  (The legacy, two
direct calls to @code{predict.lm}, lives on in @code{predict.glm} in
@R{}, which is based on the workaround for S3 written by Venables &
Ripley.)

Further, the user is entitled to use positional matching when calling
the generic, and the arguments to a method called by @code{UseMethod}
are those of the call to the generic.  Thus

@quotation
@emph{A method must have arguments in exactly the same order as the
generic.}
@end quotation

@noindent
To see the scale of this problem, consider the generic function
@code{scale}, defined as

@example
@group
scale <- function (x, center = TRUE, scale = TRUE)
    UseMethod("scale")
@end group
@end example

@noindent
Suppose an unthinking package writer created methods such as

@example
scale.foo <- function(x, scale = FALSE, ...) @{ @}
@end example

@noindent
Then for @code{x} of class @code{"foo"} the calls

@example
@group
scale(x, , TRUE)
scale(x, scale = TRUE)
@end group
@end example

@noindent
would most likely do different things, to the justifiable
consternation of the end user.

To add a further twist, which default is used when a user calls
@code{scale(x)} in our example?  What if

@example
scale.bar <- function(x, center, scale = TRUE) NextMethod("scale")
@end example

@noindent
and @code{x} has class @code{c("bar", "foo")}?  It is the default
specified in the method that is used, but the default
specified in the generic may be the one the user sees.
This leads to the recommendation:

@quotation
@emph{If the generic specifies defaults, all methods should use the same defaults.}
@end quotation

@noindent
An easy way to follow these recommendations is to always keep generics
simple, e.g.

@example
scale <- function(x, ...) UseMethod("scale")
@end example

Only add parameters and defaults to the generic if they make sense in
all possible methods implementing it.

@menu
* Adding new generics::
@end menu

@node Adding new generics,  , Generic functions and methods, Generic functions and methods
@section Adding new generics

When creating a new generic function, bear in mind that its argument
list will be the maximal set of arguments for methods, including those
written elsewhere years later.  So choosing a good set of arguments may
well be an important design issue, and there need to be good arguments
@emph{not} to include a @code{@dots{}} argument.

If a @code{@dots{}} argument is supplied, some thought should be given
to its position in the argument sequence.  Arguments which follow
@code{@dots{}} must be named in calls to the function, and they must be
named in full (partial matching is suppressed after @code{@dots{}}).
Formal arguments before @code{@dots{}} can be partially matched, and so
may `swallow' actual arguments intended for @code{@dots{}}.  Although it
is commonplace to make the @code{@dots{}} argument the last one, that is
not always the right choice.

Sometimes package writers want to make generic a function in the base
package, and request a change in @R{}.  This may be justifiable, but
making a function generic with the old definition as the default method
does have a small performance cost.  It is never necessary, as a package
can take over a function in the base package and make it generic by
something like

@example
@group
foo <- function(object, ...) UseMethod("foo")
foo.default <- function(object, ...) base::foo(object)
@end group
@end example

@noindent
Earlier versions of this manual suggested assigning @code{foo.default <-
base::foo}.  This is @strong{not} a good idea, as it captures the base
function at the time of installation and it might be changed as @R{} is
patched or updated.

The same idea can be applied for functions in other packages.

@node Linking GUIs and other front-ends to R, Function and variable index, Generic functions and methods, Top
@chapter Linking GUIs and other front-ends to R

There are a number of ways to build front-ends to @R{}: we take this to
mean a GUI or other application that has the ability to submit commands
to @R{} and perhaps to receive results back (not necessarily in a text
format).  There are other routes besides those described here, for
example the package @CRANpkg{Rserve} (from @acronym{CRAN}, see also
@uref{https://www.rforge.net/Rserve/}) and connections to Java in
@samp{JRI} (part of the @CRANpkg{rJava} package on @acronym{CRAN}).

Note that the APIs described in this chapter are only intended to be
used in an alternative front-end: they are not part of the API made
available for @R{} packages and can be dangerous to use in a
conventional package (although packages may contain alternative
front-ends).  Conversely some of the functions from the API (such as
@code{R_alloc}) should not be used in front-ends.

@menu
* Embedding R under Unix-alikes::
* Embedding R under Windows::
@end menu

@node Embedding R under Unix-alikes, Embedding R under Windows, Linking GUIs and other front-ends to R, Linking GUIs and other front-ends to R
@section Embedding R under Unix-alikes

@R{} can be built as a shared library@footnote{In the parlance of macOS
this is a @emph{dynamic} library, and is the normal way to build @R{} on
that platform.} if configured with @option{--enable-R-shlib}.  This
shared library can be used to run @R{} from alternative front-end
programs.  We will assume this has been done for the rest of this
section.  Also, it can be built as a static library if configured with
@option{--enable-R-static-lib}, and that can be used in a very similar
way (at least on Linux: on other platforms one needs to ensure that all
the symbols exported by @file{libR.a} are linked into the front-end).

The command-line @R{} front-end, @file{@var{R_HOME}/bin/exec/R}, is one
such example, and the former @acronym{GNOME} (see package @pkg{gnomeGUI}
on @acronym{CRAN}'s @samp{Archive} area) and macOS consoles are others.
The source for @file{@var{R_HOME}/bin/exec/R} is in file
@file{src/main/Rmain.c} and is very simple

@findex Rf_initialize_R
@findex Rf_mainloop
@example
int Rf_initialize_R(int ac, char **av); /* in ../unix/system.c */
void Rf_mainloop();                     /* in main.c */

extern int R_running_as_main_program;   /* in ../unix/system.c */

int main(int ac, char **av)
@{
    R_running_as_main_program = 1;
    Rf_initialize_R(ac, av);
    Rf_mainloop(); /* does not return */
    return 0;
@}
@end example

@noindent
indeed, misleadingly simple.  Remember that
@file{@var{R_HOME}/bin/exec/R} is run from a shell script
@file{@var{R_HOME}/bin/R} which sets up the environment for the
executable, and this is used for

@itemize @bullet
@item
Setting @env{R_HOME} and checking it is valid, as well as the path
@env{R_SHARE_DIR} and @env{R_DOC_DIR} to the installed @file{share} and
@file{doc} directory trees.  Also setting @env{R_ARCH} if needed.

@item
Setting @env{LD_LIBRARY_PATH} to include the directories used in linking
@R{}.  This is recorded as the default setting of
@env{R_LD_LIBRARY_PATH} in the shell script
@file{@var{R_HOME}/etc@var{R_ARCH}/ldpaths}.

@item
Processing some of the arguments, for example to run @R{} under a
debugger and to launch alternative front-ends to provide GUIs.
@end itemize

@noindent
The first two of these can be achieved for your front-end by running it
@emph{via} @command{R CMD}. So, for example

@example
R CMD /usr/local/lib/R/bin/exec/R
R CMD exec/R
@end example

@noindent
will both work in a standard @R{} installation. (@command{R CMD} looks
first for executables in @file{@var{R_HOME}/bin}.  These command-lines
need modification if a sub-architecture is in use.) If you do not want
to run your front-end in this way, you need to ensure that @env{R_HOME}
is set and @env{LD_LIBRARY_PATH} is suitable.  (The latter might well
be, but modern Unix/Linux systems do not normally include
@file{/usr/local/lib} (@file{/usr/local/lib64} on some architectures),
and @R{} does look there for system components.)

The other senses in which this example is too simple are that all the
internal defaults are used and that control is handed over to the
@R{} main loop.  There are a number of small examples@footnote{but these
are not part of the automated test procedures and so little tested.} in the
@file{tests/Embedding} directory.  These make use of
@code{Rf_initEmbeddedR} in @file{src/main/Rembedded.c}, and essentially
use
@findex Rf_initEmbeddedR
@findex R_ReplDLLinit
@findex R_ReplDLLdo1
@findex Rf_endEmbeddedR
@findex run_Rmainloop
@example
#include <Rembedded.h>

int main(int ac, char **av)
@{
    /* do some setup */
    Rf_initEmbeddedR(argc, argv);
    /* do some more setup */

    /* submit some code to R, which is done interactively via
        run_Rmainloop();

        A possible substitute for a pseudo-console is

        R_ReplDLLinit();
        while(R_ReplDLLdo1() > 0) @{
        /* add user actions here if desired */
       @}

     */
    Rf_endEmbeddedR(0);
    /* final tidying up after R is shutdown */
    return 0;
@}
@end example

@noindent
If you do not want to pass @R{} arguments, you can fake an @code{argv}
array, for example by

@example
    char *argv[]= @{"REmbeddedPostgres", "--silent"@};
    Rf_initEmbeddedR(sizeof(argv)/sizeof(argv[0]), argv);
@end example

However, to make a GUI we usually do want to run @code{run_Rmainloop}
after setting up various parts of @R{} to talk to our GUI, and arranging
for our GUI callbacks to be called during the @R{} mainloop.

One issue to watch is that on some platforms @code{Rf_initEmbeddedR} and
@code{Rf_endEmbeddedR} change the settings of the FPU (e.g.@: to allow
errors to be trapped and to make use of extended precision registers).

The standard code sets up a session temporary directory in the usual
way, @emph{unless} @code{R_TempDir} is set to a non-NULL value before
@code{Rf_initEmbeddedR} is called.  In that case the value is assumed to
contain an existing writable directory, and it is not
cleaned up when @R{} is shut down.

@code{Rf_initEmbeddedR} sets @R{} to be in interactive mode: you can set
@code{R_Interactive} (defined in @file{Rinterface.h}) subsequently to
change this.

Note that @R{} expects to be run with the locale category
@samp{LC_NUMERIC} set to its default value of @code{C}, and so should
not be embedded into an application which changes that.

It is the user's responsibility to attempt to initialize only once.  To
protect the @R{} interpreter, @code{Rf_initialize_R} will exit the
process if re-initialization is attempted.

@menu
* Compiling against the R library::
* Setting R callbacks::
* Registering symbols::
* Meshing event loops::
* Threading issues::
@end menu

@node Compiling against the R library, Setting R callbacks, Embedding R under Unix-alikes, Embedding R under Unix-alikes
@subsection Compiling against the R library

Suitable flags to compile and link against the @R{} (shared or static)
library can be found by

@example
R CMD config --cppflags
R CMD config --ldflags
@end example

@noindent
(These apply only to an uninstalled copy or a standard install.)

If @R{} is installed, @code{pkg-config} is available and neither
sub-architectures nor a macOS framework have been used, alternatives for
a shared @R{} library are

@example
pkg-config --cflags libR
pkg-config --libs libR
@end example

@noindent
and for a static @R{} library

@example
pkg-config --cflags libR
pkg-config --static --libs libR
@end example

@noindent
(This may work for an installed OS framework if @code{pkg-config} is
taught where to look for @file{libR.pc}: it is installed inside the
framework.)

However, a more comprehensive way is to set up a @file{Makefile} to
compile the front-end.  Suppose file @file{myfe.c} is to be compiled to
@file{myfe}. A suitable @file{Makefile} might be

@example
## WARNING: does not work when $@{R_HOME@} contains spaces
include $@{R_HOME@}/etc$@{R_ARCH@}/Makeconf
all: myfe

## The following is not needed, but avoids PIC flags.
myfe.o: myfe.c
        $(CC) $(ALL_CPPFLAGS) $(CFLAGS) -c myfe.c -o $@@

## replace $(LIBR) $(LIBS) by $(STATIC_LIBR) if R was build with a static libR
myfe: myfe.o
        $(MAIN_LINK) -o $@@ myfe.o $(LIBR) $(LIBS)
@end example

@noindent
invoked as

@example
R CMD make
R CMD myfe
@end example

Even though not recommended, @code{$@{R_HOME@}} may contain spaces.  In
that case, it cannot be passed as an argument to @code{include} in the
makefile.  Instead, one can instruct @command{make} using the @code{-f}
option to include @file{Makeconf}, for example @emph{via} recursive
invocation of @command{make}, see @ref{Writing portable packages}.

@example
all:
      $(MAKE) -f "$@{R_HOME@}/etc$@{R_ARCH@}/Makeconf" -f Makefile.inner
@end example
  
Additional flags which @code{$(MAIN_LINK)} includes are, amongst others,
those to select OpenMP and @option{--export-dynamic} for the GNU linker
on some platforms.  In principle @code{$(LIBS)} is not needed
when using a shared @R{} library as @file{libR} is linked against
those libraries, but some platforms need the executable also linked
against them.
@c E.g. it seems current Linux needs the executable linked against -lm.

@node Setting R callbacks, Registering symbols, Compiling against the R library, Embedding R under Unix-alikes
@subsection Setting R callbacks

For Unix-alikes there is a public header file @file{Rinterface.h} that
makes it possible to change the standard callbacks used by @R{} in a
documented way.  This defines pointers (if @code{R_INTERFACE_PTRS} is
defined)

@example
extern void (*ptr_R_Suicide)(const char *);
extern void (*ptr_R_ShowMessage)(const char *);
extern int  (*ptr_R_ReadConsole)(const char *, unsigned char *, int, int);
extern void (*ptr_R_WriteConsole)(const char *, int);
extern void (*ptr_R_WriteConsoleEx)(const char *, int, int);
extern void (*ptr_R_ResetConsole)();
extern void (*ptr_R_FlushConsole)();
extern void (*ptr_R_ClearerrConsole)();
extern void (*ptr_R_Busy)(int);
extern void (*ptr_R_CleanUp)(SA_TYPE, int, int);
extern int  (*ptr_R_ShowFiles)(int, const char **, const char **,
                               const char *, Rboolean, const char *);
extern int  (*ptr_R_ChooseFile)(int, char *, int);
extern int  (*ptr_R_EditFile)(const char *);
extern void (*ptr_R_loadhistory)(SEXP, SEXP, SEXP, SEXP);
extern void (*ptr_R_savehistory)(SEXP, SEXP, SEXP, SEXP);
extern void (*ptr_R_addhistory)(SEXP, SEXP, SEXP, SEXP);
extern int  (*ptr_R_EditFiles)(int, const char **, const char **, const char *);
extern SEXP (*ptr_do_selectlist)(SEXP, SEXP, SEXP, SEXP);
extern SEXP (*ptr_do_dataentry)(SEXP, SEXP, SEXP, SEXP);
extern SEXP (*ptr_do_dataviewer)(SEXP, SEXP, SEXP, SEXP);
extern void (*ptr_R_ProcessEvents)();
@end example

@noindent
which allow standard @R{} callbacks to be redirected to your GUI.  What
these do is generally documented in the file @file{src/unix/system.txt}.

@deftypefun void R_ShowMessage (char *@var{message})
This should display the message, which may have multiple lines:  it
should be brought to the user's attention immediately.
@end deftypefun

@deftypefun void R_Busy (int @var{which})
This function invokes actions (such as change of cursor) when @R{}
embarks on an extended computation (@code{@var{which}=1}) and when such
a state terminates (@code{@var{which}=0}).
@end deftypefun

@deftypefun int R_ReadConsole (const char *@var{prompt}, unsigned char *@var{buf}, @
  int @var{buflen}, int @var{hist})
@deftypefunx void R_WriteConsole (const char *@var{buf}, int @var{buflen})
@deftypefunx void R_WriteConsoleEx (const char *@var{buf}, int @var{buflen}, int @var{otype})
@deftypefunx void R_ResetConsole ()
@deftypefunx void R_FlushConsole ()
@deftypefunx void R_ClearerrConsole ()

These functions interact with a console.

@code{R_ReadConsole} prints the given prompt at the console and then
does a @code{fgets(3)}--like operation, transferring up to @var{buflen}
characters into the buffer @var{buf}. The last two bytes should be
set to @samp{"\n\0"} to preserve sanity.  If @var{hist} is non-zero,
then the line should be added to any command history which is being
maintained.  The return value is 0 is no input is available and >0
otherwise.

@code{R_WriteConsoleEx} writes the given buffer to the console,
@var{otype} specifies the output type (regular output or
warning/error). Call to @code{R_WriteConsole(buf, buflen)} is equivalent
to @code{R_WriteConsoleEx(buf, buflen, 0)}. To ensure backward
compatibility of the callbacks, @code{ptr_R_WriteConsoleEx} is used only
if @code{ptr_R_WriteConsole} is set to @code{NULL}.  To ensure that
@code{stdout()} and @code{stderr()} connections point to the console,
set the corresponding files to @code{NULL} @emph{via}
@example
      R_Outputfile = NULL;
      R_Consolefile = NULL;
@end example

@code{R_ResetConsole} is called when the system is reset after an error.
@code{R_FlushConsole} is called to flush any pending output to the
system console.  @code{R_ClearerrConsole} clears any errors associated
with reading from the console.
@end deftypefun

@deftypefun int R_ShowFiles (int @var{nfile}, const char **@var{file}, @
  const char **@var{headers}, const char *@var{wtitle}, Rboolean @var{del}, @
  const char *@var{pager})

This function is used to display the contents of files.
@end deftypefun

@deftypefun int R_ChooseFile (int @var{new}, char *@var{buf}, @
  int @var{len})

Choose a file and return its name in @var{buf} of length @var{len}.
Return value is 0 for success, > 0 otherwise.
@end deftypefun

@deftypefun int R_EditFile (const char *@var{buf})
Send a file to an editor window.
@end deftypefun

@deftypefun int R_EditFiles (int @var{nfile}, const char **@var{file}, const char **@var{title}, const char *@var{editor})
Send @var{nfile} files to an editor, with titles possibly to be used for
the editor window(s).
@end deftypefun

@deftypefun SEXP R_loadhistory (SEXP, SEXP, SEXP, SEXP);
@deftypefunx SEXP R_savehistory (SEXP, SEXP, SEXP, SEXP);
@deftypefunx SEXP R_addhistory (SEXP, SEXP, SEXP, SEXP);

@code{.Internal} functions for @code{loadhistory}, @code{savehistory}
and @code{timestamp}.

If the console has no history mechanism these can be as
simple as

@example
SEXP R_loadhistory (SEXP call, SEXP op, SEXP args, SEXP env)
@{
    errorcall(call, "loadhistory is not implemented");
    return R_NilValue;
@}
SEXP R_savehistory (SEXP call, SEXP op , SEXP args, SEXP env)
@{
    errorcall(call, "savehistory is not implemented");
    return R_NilValue;
@}
SEXP R_addhistory (SEXP call, SEXP op , SEXP args, SEXP env)
@{
    return R_NilValue;
@}
@end example

The @code{R_addhistory} function should return silently if no history
mechanism is present, as a user may be calling @code{timestamp} purely
to write the time stamp to the console.
@end deftypefun

@deftypefun void R_Suicide (const char *@var{message})
This should abort @R{} as rapidly as possible, displaying the message.
A possible implementation is

@example
void R_Suicide (const char *message)
@{
    char  pp[1024];
    snprintf(pp, 1024, "Fatal error: %s\n", message);
    R_ShowMessage(pp);
    R_CleanUp(SA_SUICIDE, 2, 0);
@}
@end example
@end deftypefun

@deftypefun void R_CleanUp (SA_TYPE @var{saveact}, int @var{status}, @
  int @var{RunLast})

This function invokes any actions which occur at system termination.
It needs to be quite complex:

@example
#include <Rinterface.h>
#include <Rembedded.h>    /* for Rf_KillAllDevices */

void R_CleanUp (SA_TYPE saveact, int status, int RunLast)
@{
    if(saveact == SA_DEFAULT) saveact = SaveAction;
    if(saveact == SA_SAVEASK) @{
       /* ask what to do and set saveact */
    @}
    switch (saveact) @{
    case SA_SAVE:
        if(runLast) R_dot_Last();
        if(R_DirtyImage) R_SaveGlobalEnv();
        /* save the console history in R_HistoryFile */
        break;
    case SA_NOSAVE:
        if(runLast) R_dot_Last();
        break;
    case SA_SUICIDE:
    default:
        break;
    @}

    R_RunExitFinalizers();
    /* clean up after the editor e.g. CleanEd() */

    R_CleanTempDir();

    /* close all the graphics devices */
    if(saveact != SA_SUICIDE) Rf_KillAllDevices();
    fpu_setup(FALSE);

    exit(status);
@}
@end example
@end deftypefun
@findex R_dot_Last
@findex R_RunExitFinalizers
@findex R_CleanTempDir
@findex R_SaveGlobalEnv
@findex Rf_KillAllDevices
@findex CleanEd
@findex fpu_setup

These callbacks should never be changed in a running @R{} session (and
hence cannot be called from an extension package).

@deftypefun SEXP R_dataentry (SEXP, SEXP, SEXP, SEXP);
@deftypefunx SEXP R_dataviewer (SEXP, SEXP, SEXP, SEXP);
@deftypefunx SEXP R_selectlist (SEXP, SEXP, SEXP, SEXP);

@code{.External} functions for @code{dataentry} (and @code{edit} on
matrices and data frames), @code{View} and @code{select.list}.  These
can be changed if they are not currently in use.
@end deftypefun


@node Registering symbols, Meshing event loops, Setting R callbacks, Embedding R under Unix-alikes
@subsection Registering symbols

An application embedding @R{} needs a different way of registering
symbols because it is not a dynamic library loaded by @R{} as would be
the case with a package.  Therefore @R{} reserves a special
@code{DllInfo} entry for the embedding application such that it can
register symbols to be used with @code{.C}, @code{.Call} etc.  This
entry can be obtained by calling @code{getEmbeddingDllInfo}, so a
typical use is

@findex R_getEmbeddingDllInfo
@example
DllInfo *info = R_getEmbeddingDllInfo();
R_registerRoutines(info, cMethods, callMethods, NULL, NULL);
@end example

The native routines defined by @code{cMethods} and @code{callMethods}
should be present in the embedding application.  See @ref{Registering
native routines} for details on registering symbols in general.


@node Meshing event loops, Threading issues, Registering symbols, Embedding R under Unix-alikes
@subsection Meshing event loops

One of the most difficult issues in interfacing @R{} to a front-end is
the handling of event loops, at least if a single thread is used.  @R{}
uses events and timers for

@itemize
@item
Running X11 windows such as the graphics device and data editor, and
interacting with them (e.g., using @code{locator()}).

@item
Supporting Tcl/Tk events for the @pkg{tcltk} package (for at least the
X11 version of Tk).

@item
Preparing input.

@item
Timing operations, for example for profiling @R{} code and
@code{Sys.sleep()}.

@item
Interrupts, where permitted.
@end itemize

@noindent
Specifically, the Unix-alike command-line version of @R{} runs separate
event loops for

@itemize
@item
Preparing input at the console command-line, in file
@file{src/unix/sys-unix.c}.

@item
Waiting for a response from a socket in the internal functions for
direct socket access in file
@file{src/@/modules/@/internet/@/Rsock.c}
and for the interface to @code{libcurl}.

@item
Mouse and window events when displaying the X11-based dataentry window,
in file @file{src/modules/X11/dataentry.c}.  This is regarded as
@emph{modal}, and no other events are serviced whilst it is active.
@end itemize

There is a protocol for adding event handlers to the first two types of
event loops, using types and functions declared in the header
@file{R_ext/eventloop.h} and described in comments in file
@file{src/unix/sys-std.c}.  It is possible to add (or remove) an input
handler for events on a particular file descriptor, or to set a polling
interval (@emph{via} @code{R_wait_usec}) and a function to be called
periodically @emph{via} @code{R_PolledEvents}: the polling mechanism is used by
the @pkg{tcltk} package.
@findex R_PolledEvents
@findex R_wait_usec

It is not intended that these facilities are used by packages, but if
they are needed exceptionally, the package should ensure that it cleans
up and removes its handlers when its namespace is unloaded.  Note that
the header @file{sys/select.h} is needed@footnote{At least according to
POSIX 2004 and later.  Earlier standards prescribed @file{sys/time.h}:
@file{R_ext/eventloop.h} will include it if @code{HAVE_SYS_TIME_H} is
defined.}: users should check this is available and define
@code{HAVE_SYS_SELECT_H} before including @file{R_ext/eventloop.h}. (It
is often the case that another header will include @file{sys/select.h}
before @file{eventloop.h} is processed, but this should not be relied
on.)

An alternative front-end needs both to make provision for other @R{}
events whilst waiting for input, and to ensure that it is not frozen out
during events of the second type.  The ability to add a polled handler
as @code{R_timeout_handler} is used by the @pkg{tcltk} package.


@node Threading issues,  , Meshing event loops, Embedding R under Unix-alikes
@subsection Threading issues

Embedded @R{} is designed to be run in the main thread, and all the
testing is done in that context.  There is a potential issue with the
stack-checking mechanism where threads are involved.  This uses two
variables declared in @file{Rinterface.h} (if @code{CSTACK_DEFNS} is
defined) as

@example
extern uintptr_t R_CStackLimit; /* C stack limit */
extern uintptr_t R_CStackStart; /* Initial stack address */
@end example

@noindent
Note that @code{uintptr_t} is an optional C99 type for which a
substitute is defined in @R{}, so your code needs to define
@code{HAVE_UINTPTR_T} appropriately.  To do so, test if the type is
defined in C header @file{stdint.h} or C++ header @file{cstdint} and if
so include the header and define @code{HAVE_UINTPTR_T} before including
@file{Rinterface.h}.  (For C code one can simply include
@file{Rconfig.h}, possibly @emph{via} @file{R.h}, and for C++11 code
@file{Rinterface.h} will include the header @file{cstdint}.)

These will be set@footnote{at least on platforms where the values are
available, that is having @code{getrlimit} and on Linux or having
@code{sysctl} supporting @code{KERN_USRSTACK}, including FreeBSD and
macOS.} when @code{Rf_initialize_R} is called, to values appropriate to
the main thread.  Stack-checking can be disabled by setting
@code{R_CStackLimit = (uintptr_t)-1} immediately after
@code{Rf_initialize_R} is called, but it is better to if possible set
appropriate values.  (What these are and how to determine them are
OS-specific, and the stack size limit may differ for secondary threads.
If you have a choice of stack size, at least 10Mb is recommended.)

You may also want to consider how signals are handled: @R{} sets signal
handlers for several signals, including @code{SIGINT}, @code{SIGSEGV},
@code{SIGPIPE}, @code{SIGUSR1} and @code{SIGUSR2}, but these can all be
suppressed by setting the variable @code{R_SignalHandlers} (declared in
@file{Rinterface.h}) to @code{0}.

Note that these variables must not be changed by an @R{}
@strong{package}: a package should not call @R{} internals which
makes use of the stack-checking mechanism on a secondary thread.

@node Embedding R under Windows,  , Embedding R under Unix-alikes, Linking GUIs and other front-ends to R
@section Embedding R under Windows

All Windows interfaces to @R{} call entry points in the DLL
@file{R.dll}, directly or indirectly.  Simpler applications may find it
easier to use the indirect route @emph{via} @acronym{(D)COM}.

@menu
* Using (D)COM::
* Calling R.dll directly::
* Finding R_HOME::
@end menu

@node Using (D)COM, Calling R.dll directly, Embedding R under Windows, Embedding R under Windows
@subsection Using (D)COM

@acronym{(D)COM} is a standard Windows mechanism used for communication
between Windows applications.  One application (here @R{}) is run as COM
server which offers services to clients, here the front-end calling
application.  The services are described in a `Type Library' and are
(more or less) language-independent, so the calling application can be
written in C or C++ or Visual Basic or Perl or Python and so on.
The `D' in (D)COM refers to `distributed', as the client and server can
be running on different machines.

The basic @R{} distribution is not a (D)COM server, but two addons are
currently available that interface directly with @R{} and provide a
(D)COM server:
@itemize
@item
There is a (D)COM server called @code{StatConnector} written by Thomas
Baier available @emph{via} @uref{https://www.autstat.com/},
which works with @R{} packages to support transfer of data to and from
@R{} and remote execution of @R{} commands, as well as embedding of an
@R{} graphics window.

Recent versions have usage restrictions.

@c @item
@c Another (D)COM server, @code{RDCOMServer}, may be available from Omegahat,
@c @uref{http://www.omegahat.net/}.  Its philosophy is discussed in
@c @uref{http://www.omegahat.net/RDCOMServer/Docs/Paradigm.html} and is
@c very different from the purpose of this section.
@end itemize
@node Calling R.dll directly, Finding R_HOME, Using (D)COM, Embedding R under Windows
@subsection Calling R.dll directly

The @code{R} DLL is mainly written in C and has @code{_cdecl} entry
points.  Calling it directly will be tricky except from C code (or C++
with a little care).

There is a version of the Unix-alike interface calling

@example
int Rf_initEmbeddedR(int ac, char **av);
void Rf_endEmbeddedR(int fatal);
@end example

@noindent
which is an entry point in @file{R.dll}.  Examples of its use (and a
suitable @file{Makefile.win}) can be found in the @file{tests/Embedding}
directory of the sources.  You may need to ensure that
@file{@var{R_HOME}/bin} is in your @env{PATH} so the @R{} DLLs are found.

Examples of calling @file{R.dll} directly are provided in the directory
@file{src/@/gnuwin32/@/front-ends}, including a simple command-line
front end @file{rtest.c} whose code is

@findex setup_Rmainloop
@findex R_setStartTime
@findex R_set_command_line_arguments
@smallexample
#define Win32
#include <windows.h>
#include <stdio.h>
#include <Rversion.h>
#define LibExtern __declspec(dllimport) extern
#include <Rembedded.h>
#include <R_ext/RStartup.h>
/* for askok and askyesnocancel */
#include <graphapp.h>

/* for signal-handling code */
#include <psignal.h>

/* simple input, simple output */

/* This version blocks all events: a real one needs to call ProcessEvents
   frequently. See rterm.c and ../system.c for one approach using
   a separate thread for input.
*/
int myReadConsole(const char *prompt, unsigned char *buf, int len, int addtohistory)
@{
    fputs(prompt, stdout);
    fflush(stdout);
    if(fgets((char *)buf, len, stdin)) return 1; else return 0;
@}

void myWriteConsole(const char *buf, int len)
@{
    printf("%s", buf);
@}

void myCallBack(void)
@{
    /* called during i/o, eval, graphics in ProcessEvents */
@}

void myBusy(int which)
@{
    /* set a busy cursor ... if which = 1, unset if which = 0 */
@}

static void my_onintr(int sig) @{ UserBreak = 1; @}

int main (int argc, char **argv)
@{
    structRstart rp;
    Rstart Rp = &rp;
    char Rversion[25], *RHome;

    sprintf(Rversion, "%s.%s", R_MAJOR, R_MINOR);
    if(strcmp(getDLLVersion(), Rversion) != 0) @{
        fprintf(stderr, "Error: R.DLL version does not match\n");
        exit(1);
    @}

    R_setStartTime();
    R_DefParamsEx(Rp, RSTART_VERSION);
    if((RHome = get_R_HOME()) == NULL) @{
        fprintf(stderr, "R_HOME must be set in the environment or Registry\n");
        exit(1);
    @}
    Rp->rhome = RHome;
    Rp->home = getRUser();
    Rp->CharacterMode = LinkDLL;
    Rp->EmitEmbeddedUTF8 = FALSE;
    Rp->ReadConsole = myReadConsole;
    Rp->WriteConsole = myWriteConsole;
    Rp->CallBack = myCallBack;
    Rp->ShowMessage = askok;
    Rp->YesNoCancel = askyesnocancel;
    Rp->Busy = myBusy;

    Rp->R_Quiet = TRUE;        /* Default is FALSE */
    Rp->R_Interactive = FALSE; /* Default is TRUE */
    Rp->RestoreAction = SA_RESTORE;
    Rp->SaveAction = SA_NOSAVE;
    R_SetParams(Rp);
    R_set_command_line_arguments(argc, argv);

    FlushConsoleInputBuffer(GetStdHandle(STD_INPUT_HANDLE));

    signal(SIGBREAK, my_onintr);
    GA_initapp(0, 0);
    readconsolecfg();
    setup_Rmainloop();
#ifdef SIMPLE_CASE
    run_Rmainloop();
#else
    R_ReplDLLinit();
    while(R_ReplDLLdo1() > 0) @{
/* add user actions here if desired */
    @}
/* only get here on EOF (not q()) */
#endif
    Rf_endEmbeddedR(0);
    return 0;
@}
@end smallexample

The ideas are

@itemize
@item
Check that the front-end and the linked @file{R.dll} match -- other
front-ends may allow a looser match.

@item
Find and set the @R{} home directory and the user's home directory.  The
former may be available from the Windows Registry: it will be in
@code{HKEY_LOCAL_MACHINE\Software\R-core\R\InstallPath} from an
administrative install and
@code{HKEY_CURRENT_USER\Software\R-core\R\InstallPath} otherwise, if
selected during installation (as it is by default).

@findex R_DefParams
@findex R_DefParamsEx
@findex R_SetParams
@item
Define startup conditions and callbacks @emph{via} the @code{Rstart} structure.
@code{R_DefParams} sets the defaults, and @code{R_SetParams} sets
updated values. @code{R_DefParamsEx} takes an extra argument, the version
number of the @code{Rstart} structure provided (@code{RSTART_VERSION} refers
to the current version) and returns a non-zero status when that version is
not supported by @R{}.

@item
Record the command-line arguments used by
@code{R_set_command_line_arguments} for use by the @R{} function
@code{commandArgs()}.

@item
Set up the signal handler and the basic user interface.

@item
Run the main @R{} loop, possibly with our actions intermeshed.

@item
Arrange to clean up.
@end itemize

An underlying theme is the need to keep the GUI `alive', and this has
not been done in this example.  The @R{} callback @code{R_ProcessEvents}
needs to be called frequently to ensure that Windows events in @R{}
windows are handled expeditiously.  Conversely, @R{} needs to allow the
GUI code (which is running in the same process) to update itself as
needed -- two ways are provided to allow this:

@itemize
@item
@code{R_ProcessEvents} calls the callback registered by
@code{Rp->callback}.  A version of this is used to run package Tcl/Tk
for @pkg{tcltk} under Windows, for the code is

@findex R_ProcessEvents
@findex onintr
@example
void R_ProcessEvents(void)
@{
    while (peekevent()) doevent(); /* Windows events for GraphApp */
    if (UserBreak) @{ UserBreak = FALSE; onintr(); @}
    R_CallBackHook();
    if(R_tcldo) R_tcldo();
@}
@end example

@item
The mainloop can be split up to allow the calling application to take
some action after each line of input has been dealt with: see the
alternative code below @code{#ifdef SIMPLE_CASE}.
@end itemize

It may be that no @R{} GraphApp windows need to be considered, although
these include pagers, the @code{windows()} graphics device, the @R{}
data and script editors and various popups such as @code{choose.file()}
and @code{select.list()}.  It would be possible to replace all of these,
but it seems easier to allow GraphApp to handle most of them.

It is possible to run @R{} in a GUI in a single thread (as
@file{RGui.exe} shows) but it will normally be easier@footnote{An
attempt to use only threads in the late 1990s failed to work correctly
under Windows 95, the predominant version of Windows at that time.} to
use multiple threads.

Note that @R{}'s own front ends use a stack size of 10Mb, whereas MinGW
executables default to 2Mb, and Visual C++ ones to 1Mb.  The latter
stack sizes are too small for a number of @R{} applications, so
general-purpose front-ends should use a larger stack size.

Applications embedding @R{} 4.2.0 and newer should use UCRT as the C runtime
and opt in for UTF-8 as the active code page in their manifest, as all
frontends shipped with @R{} do.  This will allow the embedded @R{} to use
UTF-8 as its native encoding on recent Windows systems.

@node Finding R_HOME,  , Calling R.dll directly, Embedding R under Windows
@subsection Finding R_HOME

Both applications which embed @R{} and those which use a @code{system}
call to invoke @R{} (as @command{Rscript.exe}, @command{Rterm.exe} or
@command{R.exe}) need to be able to find the @R{} @file{bin} directory.
The simplest way to do so is the ask the user to set an environment
variable @env{R_HOME} and use that, but naive users may be flummoxed as
to how to do so or what value to use.

The @R{} for Windows installers have for a long time allowed the value
of @code{R_HOME} to be recorded in the Windows Registry: this is
optional but selected by default.  @emph{Where} it is recorded has
changed over the years to allow for multiple versions of @R{} to be
installed at once, and to allow 32- and 64-bit versions of @R{} to be
installed on the same machine.

The basic Registry location is @code{Software\R-core\R}.  For an
administrative install this is under @code{HKEY_LOCAL_MACHINE} and on a
64-bit OS @code{HKEY_LOCAL_MACHINE\Software\R-core\R} is by default
redirected for a 32-bit application, so a 32-bit application will see
the information for the last 32-bit install, and a 64-bit application
that for the last 64-bit install.  For a personal install, the
information is under @code{HKEY_CURRENT_USER\Software\R-core\R} which is
seen by both 32-bit and 64-bit applications and so records the last
install of either architecture.  To circumvent this, there are locations
@code{Software\R-core\R32} and @code{Software\R-core\R64} which always
refer to one architecture.

When @R{} is installed and recording is not disabled then two string
values are written at that location for keys @code{InstallPath} and
@code{Current Version}, and these keys are removed when @R{} is
uninstalled.  To allow information about other installed versions to be
retained, there is also a key named something like @code{3.0.0} or
@code{3.0.0 patched} or @code{3.1.0 Pre-release} with a value for
@code{InstallPath}.

So a comprehensive algorithm to search for @code{R_HOME} is something
like

@itemize
@item
Decide which of personal or administrative installs should have
precedence.  There are arguments both ways: we find that with roaming
profiles that @code{HKEY_CURRENT_USER\Software} often gets reverted to
an earlier version.  Do the following for one or both of
@code{HKEY_CURRENT_USER} and @code{HKEY_LOCAL_MACHINE}.

@item
If the desired architecture is known, look in @code{Software\R-core\R32}
or @code{Software\R-core\R64}, and if that does not exist or the
architecture is immaterial, in @code{Software\R-core\R}.

@item
If key @code{InstallPath} exists then this is @code{R_HOME} (recorded
using backslashes).  If it does not, look for version-specific keys like
@code{2.11.0 alpha}, pick the latest (which is of itself a complicated
algorithm as @code{2.11.0 patched > 2.11.0 > 2.11.0 alpha > 2.8.1}) and
use its value for @code{InstallPath}.
@end itemize

@node Function and variable index, Concept index, Linking GUIs and other front-ends to R, Top
@unnumbered Function and variable index

@printindex vr

@node Concept index,  , Function and variable index, Top
@unnumbered Concept index

@printindex cp

@bye

@c Local Variables: ***
@c mode: TeXinfo ***
@c End: ***