doc/manual/R-ints.texi

\input texinfo
@c %**start of header
@setfilename R-ints.info
@settitle R Internals
@setchapternewpage on
@c %**end of header

@c @documentencoding ISO-8859-1

@syncodeindex fn vr

@dircategory Programming
@direntry
* R Internals: (R-ints).      R Internals.
@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 R Internals
@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, R Internal Structures, (dir), (dir)
@top R Internals

This is a guide to the internal structures of @R{} and coding standards for
the core team working on @R{} itself.

@insertcopying

@end ifnottex

@menu
* R Internal Structures::       
* .Internal vs .Primitive::     
* Internationalization in the R sources::  
* Package Structure::           
* Files::                       
* Graphics Devices::            
* GUI consoles::                
* Tools::                       
* R coding standards::          
* Testing R code::              
* Use of TeX dialects::         
* Current and future directions::  
* Function and variable index::  
* Concept index::               
@end menu
@c  Could have (autogenerated) @detailmenu here ..

@node R Internal Structures, .Internal vs .Primitive, Top, Top
@chapter R Internal Structures

This chapter is the beginnings of documentation about @R{} internal
structures.  It is written for the core team and others studying the
code in the @file{src/main} directory.

It is a work-in-progress and should be checked against the current
version of the source code.  Versions for @R{} 2.x.y contain historical
comments about when features were introduced: this version is for the
3.x.y series.

@menu
* SEXPs::                       
* Environments and variable lookup::  
* Attributes::                  
* Contexts::                    
* Argument evaluation::         
* Autoprinting::                
* The write barrier::           
* Serialization Formats::       
* Encodings for CHARSXPs::      
* The CHARSXP cache::           
* Warnings and errors::         
* S4 objects::                  
* Memory allocators::           
* Internal use of global and base environments::  
* Modules::                     
* Visibility::                  
* Lazy loading::                
@end menu

@node SEXPs, Environments and variable lookup, R Internal Structures, R Internal Structures
@section SEXPs

@cindex SEXP
@cindex SEXPRREC
What @R{} users think of as @emph{variables} or @emph{objects} are
symbols which are bound to a value.  The value can be thought of as
either a @code{SEXP} (a pointer), or the structure it points to, a
@code{SEXPREC} (and there are alternative forms used for vectors, namely
@code{VECSXP} pointing to @code{VECTOR_SEXPREC} structures).
So the basic building blocks of @R{} objects are often called
@emph{nodes}, meaning @code{SEXPREC}s or @code{VECTOR_SEXPREC}s.

Note that the internal structure of the @code{SEXPREC} is not made
available to @R{} Extensions: rather @code{SEXP} is an opaque pointer,
and the internals can only be accessed by the functions provided.

@cindex node
Both types of node structure have as their first three fields a 64-bit
@code{sxpinfo} header and then three pointers (to the attributes and the
previous and next node in a doubly-linked list), and then some further
fields.  On a 32-bit platform a node@footnote{strictly, a @code{SEXPREC}
node; @code{VECTOR_SEXPREC} nodes are slightly smaller but followed by
data in the node.} occupies 32 bytes: on a 64-bit platform typically 56
bytes (depending on alignment constraints).

The first five bits of the @code{sxpinfo} header specify one of up to 32
@code{SEXPTYPE}s.

@menu
* SEXPTYPEs::                   
* Rest of header::              
* The 'data'::                  
* Allocation classes::          
@end menu

@node SEXPTYPEs, Rest of header, SEXPs, SEXPs
@subsection SEXPTYPEs

@cindex SEXPTYPE
Currently @code{SEXPTYPE}s 0:10 and 13:25 are in use.  Values 11 and 12
were used for internal factors and ordered factors and have since been
withdrawn.  Note that the @code{SEXPTYPE} numbers are stored in
@code{save}d objects and that the ordering of the types is used, so the
gap cannot easily be reused.

@cindex SEXPTYPE table
@quotation
@multitable {no} {SPECIALSXPXXX} {S4 classes not of simple type}
@headitem no @tab  SEXPTYPE@tab  Description
@item @code{0}   @tab @code{NILSXP}      @tab @code{NULL}
@item @code{1}   @tab @code{SYMSXP}      @tab symbols
@item @code{2}   @tab @code{LISTSXP}     @tab pairlists
@item @code{3}   @tab @code{CLOSXP}      @tab closures
@item @code{4}   @tab @code{ENVSXP}      @tab environments
@item @code{5}   @tab @code{PROMSXP}     @tab promises
@item @code{6}   @tab @code{LANGSXP}     @tab language objects
@item @code{7}   @tab @code{SPECIALSXP}  @tab special functions
@item @code{8}   @tab @code{BUILTINSXP}  @tab builtin functions
@item @code{9}   @tab @code{CHARSXP}     @tab internal character strings
@item @code{10}   @tab @code{LGLSXP}     @tab logical vectors
@item @code{13}   @tab @code{INTSXP}     @tab integer vectors
@item @code{14}   @tab @code{REALSXP}    @tab numeric vectors
@item @code{15}   @tab @code{CPLXSXP}    @tab complex vectors
@item @code{16}   @tab @code{STRSXP}     @tab character vectors
@item @code{17}   @tab @code{DOTSXP}     @tab dot-dot-dot object
@item @code{18}   @tab @code{ANYSXP}     @tab make ``any'' args work
@item @code{19}   @tab @code{VECSXP}     @tab list (generic vector)
@item @code{20}   @tab @code{EXPRSXP}    @tab expression vector
@item @code{21}   @tab @code{BCODESXP}   @tab byte code
@item @code{22}   @tab @code{EXTPTRSXP}  @tab external pointer
@item @code{23}   @tab @code{WEAKREFSXP} @tab weak reference
@item @code{24}   @tab @code{RAWSXP}     @tab raw vector
@item @code{25}   @tab @code{S4SXP}      @tab S4 classes not of simple type
@end multitable
@end quotation

@cindex atomic vector type
Many of these will be familiar from @R{} level: the atomic vector types
are @code{LGLSXP}, @code{INTSXP}, @code{REALSXP}, @code{CPLXSP},
@code{STRSXP} and @code{RAWSXP}.  Lists are @code{VECSXP} and names
(also known as symbols) are @code{SYMSXP}.  Pairlists (@code{LISTSXP},
the name going back to the origins of @R{} as a Scheme-like language)
are rarely seen at @R{} level, but are for example used for argument
lists.  Character vectors are effectively lists all of whose elements
are @code{CHARSXP}, a type that is rarely visible at @R{} level.

@cindex language object
@cindex argument list
Language objects (@code{LANGSXP}) are calls (including formulae and so
on).  Internally they are pairlists with first element a
reference@footnote{a pointer to a function or a symbol to look up the
function by name, or a language object to be evaluated to give a
function.} to the function to be called with remaining elements the
actual arguments for the call (and with the tags if present giving the
specified argument names).  Although this is not enforced, many places
in the code assume that the pairlist is of length one or more, often
without checking.

@cindex expression
Expressions are of type @code{EXPRSXP}: they are a vector of (usually
language) objects most often seen as the result of @code{parse()}.

@cindex function
The functions are of types @code{CLOSXP}, @code{SPECIALSXP} and
@code{BUILTINSXP}: where @code{SEXPTYPE}s are stored in an integer
these are sometimes lumped into a pseudo-type @code{FUNSXP} with code
99.  Functions defined via @code{function} are of type @code{CLOSXP} and
have formals, body and environment.

@cindex S4 type
The @code{SEXPTYPE} @code{S4SXP} is for S4 objects which do not consist
solely of a simple type such as an atomic vector or function.


@node Rest of header, The 'data', SEXPTYPEs, SEXPs
@subsection Rest of header

Note that the size and structure of the header changed in @R{} 3.5.0:
see earlier editions of this manual for the previous layout.

The @code{sxpinfo} header is defined as a 64-bit C structure by

@example
#define NAMED_BITS 16
struct sxpinfo_struct @{
    SEXPTYPE type      :  5;  /* @r{discussed above} */
    unsigned int scalar:  1;  /* @r{is this a numeric vector of length 1?}
    unsigned int obj   :  1;  /* @r{is this an object with a class attribute?} */
    unsigned int alt   :  1;  /* @r{is this an @code{ALTREP} object?} */
    unsigned int gp    : 16;  /* @r{general purpose, see below} */
    unsigned int mark  :  1;  /* @r{mark object as `in use' in GC} */
    unsigned int debug :  1;
    unsigned int trace :  1;
    unsigned int spare :  1;  /* @r{debug once and with reference counting} */
    unsigned int gcgen :  1;  /* @r{generation for GC} */
    unsigned int gccls :  3;  /* @r{class of node for GC} */
    unsigned int named : NAMED_BITS; /* @r{used to control copying} */
    unsigned int extra : 32 - NAMED_BITS;
@}; /*		    Tot: 64 */
@end example

@findex debug bit
The @code{debug} bit is used for closures and environments.  For
closures it is set by @code{debug()} and unset by @code{undebug()}, and
indicates that evaluations of the function should be run under the
browser.  For environments it indicates whether the browsing is in
single-step mode.

@findex trace bit
The @code{trace} bit is used for functions for @code{trace()} and for
other objects when tracing duplications (see @code{tracemem}).

@findex spare bit
The @code{spare} bit is used for closures to mark them for one-time
debugging.

@findex named bits
@findex NAMED
@findex SET_NAMED
@cindex copying semantics
The @code{named} field is set and accessed by the @code{SET_NAMED} and
@code{NAMED} macros, and take values @code{0}, @code{1} and @code{2}, or
possibly higher if @code{NAMEDMAX} is set to a higher value.
@R{} has a `call by value' illusion, so an assignment like
@example
b <- a
@end example
[The @code{NAMED} mechanism has been replaced by reference counting.]

@noindent
appears to make a copy of @code{a} and refer to it as @code{b}.
However, if neither @code{a} nor @code{b} are subsequently altered there
is no need to copy.  What really happens is that a new symbol @code{b}
is bound to the same value as @code{a} and the @code{named} field on the
value object is set (in this case to @code{2}).  When an object is about
to be altered, the @code{named} field is consulted.  A value of @code{2}
or more means that the object must be duplicated before being changed.  (Note
that this does not say that it is necessary to duplicate, only that it
should be duplicated whether necessary or not.)  A value of @code{0}
means that it is known that no other @code{SEXP} shares data with this
object, and so it may safely be altered.  A value of @code{1} is used
for situations like

@example
dim(a) <- c(7, 2)
@end example

@noindent
where in principle two copies of @code{a} exist for the duration of the
computation as (in principle)

@example
a <- `dim<-`(a, c(7, 2))
@end example

@noindent
but for no longer, and so some primitive functions can be optimized to
avoid a copy in this case.  [This mechanism is scheduled to be replaced
in @R{} 4.0.0.]

The @code{gp} bits are by definition `general purpose'.  We label these
from 0 to 15.  Bits 0--5 and bits 14--15 have been used as described below
(mainly from detective work on the sources).

@findex gp bits
@findex LEVELS
@findex SETLEVELS
The bits can be accessed and set by the @code{LEVELS} and
@code{SETLEVELS} macros, which names appear to date back to the internal
factor and ordered types and are now used in only a few places in the
code.  The @code{gp} field is serialized/unserialized for the
@code{SEXPTYPE}s other than @code{NILSXP}, @code{SYMSXP} and
@code{ENVSXP}.

Bits 14 and 15 of @code{gp} are used for `fancy bindings'.  Bit 14 is
used to lock a binding or an environment, and bit 15 is used to indicate
an active binding.  (For the definition of an `active binding' see the
header comments in file @file{src/main/envir.c}.)  Bit 15 is used for an
environment to indicate if it participates in the global cache.

@findex ARGSUSED
@findex SET_ARGUSED
The macros @code{ARGUSED} and @code{SET_ARGUSED} are used when matching
actual and formal function arguments, and take the values 0, 1 and 2.

@findex MISSING
@findex SET_MISSING
The macros @code{MISSING} and @code{SET_MISSING} are used for pairlists
of arguments.  Four bits are reserved, but only two are used (and
exactly what for is not explained).  It seems that bit 0 is used by
@code{matchArgs_NR} to mark missingness on the returned argument list, and
bit 1 is used to mark the use of a default value for an argument copied
to the evaluation frame of a closure.

@findex DDVAL
@findex SET_DDVAL
@cindex ... argument
Bit 0 is used by macros @code{DDVAL} and @code{SET_DDVAL}.  This
indicates that a @code{SYMSXP} is one of the symbols @code{..n} which
are implicitly created when @code{...} is processed, and so indicates
that it may need to be looked up in a @code{DOTSXP}.

@findex PRSEEN
@cindex promise
Bit 0 is used for @code{PRSEEN}, a flag to indicate if a promise has
already been seen during the evaluation of the promise (and so to avoid
recursive loops).

Bit 0 is used for @code{HASHASH}, on the @code{PRINTNAME} of the
@code{TAG} of the frame of an environment. (This bit is not serialized
for @code{CHARSXP} objects.)

Bits 0 and 1 are used for weak references (to indicate `ready to
finalize', `finalize on exit').

Bit 0 is used by the condition handling system (on a @code{VECSXP}) to
indicate a calling handler.

Bit 4 is turned on to mark S4 objects.

Bits 1, 2, 3, 5 and 6 are used for a @code{CHARSXP} to denote its
encoding.  Bit 1 indicates that the @code{CHARSXP} should be treated as
a set of bytes, not necessarily representing a character in any known
encoding.  Bits 2, 3 and 6 are used to indicate that it is known to be
in Latin-1, UTF-8 or @acronym{ASCII} respectively.

Bit 5 for a @code{CHARSXP} indicates that it is hashed by its address,
that is @code{NA_STRING} or is in the @code{CHARSXP} cache (this is not
serialized).  Only exceptionally is a @code{CHARSXP} not hashed, and
this should never happen in end-user code.

@node The 'data', Allocation classes, Rest of header, SEXPs
@subsection The `data'

A @code{SEXPREC} is a C structure containing the 64-bit header as
described above, three pointers (to the attributes, previous and next
node) and the node data, a union

@example
union @{
    struct primsxp_struct primsxp;
    struct symsxp_struct symsxp;
    struct listsxp_struct listsxp;
    struct envsxp_struct envsxp;
    struct closxp_struct closxp;
    struct promsxp_struct promsxp;
@} u;
@end example

@noindent
All of these alternatives apart from the first (an @code{int}) are three
pointers, so the union occupies three words.

@cindex vector type
The vector types are @code{RAWSXP}, @code{CHARSXP}, @code{LGLSXP},
@code{INTSXP}, @code{REALSXP}, @code{CPLXSXP}, @code{STRSXP},
@code{VECSXP}, @code{EXPRSXP} and @code{WEAKREFSXP}.  Remember that such
types are a @code{VECTOR_SEXPREC}, which again consists of the header
and the same three pointers, but followed by two integers giving the
length and `true length'@footnote{The only current use is for hash tables of
environments (@code{VECSXP}s), where @code{length} is the size of the table
and @code{truelength} is the number of primary slots in use, for the
reference hash tables in serialization (@code{VECSXP}s), and for `growable'
vectors (atomic vectors, @code{VECSXP}s and @code{EXPRSXP}s) which are
created by slightly over-committing when enlarging a vector during
subassignment, so that some number of the following enlargements during
subassignment can be performed in place), where @code{truelength} is the
number of slots in use.  } of the vector, and then followed by the data
(aligned as required: on most 32-bit systems with a 24-byte
@code{VECTOR_SEXPREC} node the data can follow immediately after the node). 
The data are a block of memory of the appropriate length to store `true
length' elements (rounded up to a multiple of 8 bytes, with the 8-byte
blocks being the `Vcells' referred in the documentation for @code{gc()}).

The `data' for the various types are given in the table below.  A lot of
this is interpretation, i.e.@: the types are not checked.

@table @code
@item NILSXP
There is only one object of type @code{NILSXP}, @code{R_NilValue}, with
no data.

@item SYMSXP
Pointers to three nodes, the name, value and internal, accessed by
@code{PRINTNAME} (a @code{CHARSXP}), @code{SYMVALUE} and
@code{INTERNAL}.  (If the symbol's value is a @code{.Internal} function,
the last is a pointer to the appropriate @code{SEXPREC}.)  Many symbols
have @code{SYMVALUE} @code{R_UnboundValue}.

@item LISTSXP
Pointers to the CAR, CDR (usually a @code{LISTSXP} or @code{NULL}) and
TAG (a @code{SYMSXP} or @code{NULL}).

@item CLOSXP
Pointers to the formals (a pairlist), the body and the environment.

@item ENVSXP
Pointers to the frame, enclosing environment and hash table (@code{NULL} or a
@code{VECSXP}).  A frame is a tagged pairlist with tag the symbol and
CAR the bound value.

@item PROMSXP
Pointers to the value, expression and environment (in which to evaluate
the expression).  Once an promise has been evaluated, the environment is
set to @code{NULL}.

@item LANGSXP
A special type of @code{LISTSXP} used for function calls.  (The CAR
references the function (perhaps via a symbol or language object), and
the CDR the argument list with tags for named arguments.)  @R{}-level
documentation references to `expressions' / `language objects' are
mainly @code{LANGSXP}s, but can be symbols (@code{SYMSXP}s) or
expression vectors (@code{EXPRSXP}s).

@item SPECIALSXP
@itemx BUILTINSXP
An integer giving the offset into the table of
primitives/@code{.Internal}s.

@item CHARSXP
@code{length}, @code{truelength} followed by a block of bytes (allowing
for the @code{nul} terminator).

@item LGLSXP
@itemx INTSXP
@code{length}, @code{truelength} followed by a block of C @code{int}s
(which are 32 bits on all @R{} platforms).

@item REALSXP
@code{length}, @code{truelength} followed by a block of C @code{double}s.

@item CPLXSXP
@code{length}, @code{truelength} followed by a block of C99 @code{double
complex}s.

@item STRSXP
@code{length}, @code{truelength} followed by a block of pointers
(@code{SEXP}s pointing to @code{CHARSXP}s).

@item DOTSXP
A special type of @code{LISTSXP} for the value bound to a @code{...}
symbol: a pairlist of promises.

@item ANYSXP
This is used as a place holder for any type: there are no actual objects
of this type.

@item VECSXP
@itemx EXPRSXP
@code{length}, @code{truelength} followed by a block of pointers.  These
are internally identical (and identical to @code{STRSXP}) but differ in
the interpretations placed on the elements.

@item BCODESXP
For the `byte-code' objects generated by the compiler.

@item EXTPTRSXP
Has three pointers, to the pointer, the protection value (an @R{} object
which if alive protects this object) and a tag (a @code{SYMSXP}?).

@item WEAKREFSXP
A @code{WEAKREFSXP} is a special @code{VECSXP} of length 4, with
elements @samp{key}, @samp{value}, @samp{finalizer} and @samp{next}.
The @samp{key} is @code{NULL}, an environment or an external pointer,
and the @samp{finalizer} is a function or @code{NULL}.

@item RAWSXP
@code{length}, @code{truelength} followed by a block of bytes.

@item S4SXP
two unused pointers and a tag.
@end table

@node Allocation classes,  , The 'data', SEXPs
@subsection Allocation classes

@cindex allocation classes
As we have seen, the field @code{gccls} in the header is three bits to
label up to 8 classes of nodes.  Non-vector nodes are of class 0, and
`small' vector nodes are of classes 1 to 5, with a class for custom
allocator vector nodes 6 and `large' vector nodes being of class 7.  The
`small' vector nodes are able to store vector data of up to 8, 16, 32,
64 and 128 bytes: larger vectors are @code{malloc}-ed individually
whereas the `small' nodes are allocated from pages of about 2000
bytes. Vector nodes allocated using custom allocators (via
@code{allocVector3}) are not counted in the gc memory usage statistics
since their memory semantics is not under R's control and may be
non-standard (e.g., memory could be partially shared across nodes).


@node Environments and variable lookup, Attributes, SEXPs, R Internal Structures
@section Environments and variable lookup

@cindex environment
@cindex variable lookup
What users think of as `variables' are symbols which are bound to
objects in `environments'.  The word `environment' is used ambiguously
in @R{} to mean @emph{either} the frame of an @code{ENVSXP} (a pairlist
of symbol-value pairs) @emph{or} an @code{ENVSXP}, a frame plus an
enclosure.

@cindex user databases
There are additional places that `variables' can be looked up, called
`user databases' in comments in the code.  These seem undocumented in
the @R{} sources, but apparently refer to the @pkg{RObjectTable} package
formerly available at @uref{https://www.omegahat.net/RObjectTables/}.

@cindex base environment
@cindex environment, base
The base environment is special.  There is an @code{ENVSXP} environment
with enclosure the empty environment @code{R_EmptyEnv}, but the frame of
that environment is not used.  Rather its bindings are part of the
global symbol table, being those symbols in the global symbol table
whose values are not @code{R_UnboundValue}.  When @R{} is started the
internal functions are installed (by C code) in the symbol table, with
primitive functions having values and @code{.Internal} functions having
what would be their values in the field accessed by the @code{INTERNAL}
macro.  Then @code{.Platform} and @code{.Machine} are computed and the
base package is loaded into the base environment followed by the system
profile.

The frames of environments (and the symbol table) are normally hashed
for faster access (including insertion and deletion).

By default @R{} maintains a (hashed) global cache of `variables' (that
is symbols and their bindings) which have been found, and this refers
only to environments which have been marked to participate, which
consists of the global environment (aka the user workspace), the base
environment plus environments@footnote{Remember that attaching a list or
a saved image actually creates and populates an environment and attaches
that.} which have been @code{attach}ed.  When an environment is either
@code{attach}ed or @code{detach}ed, the names of its symbols are flushed
from the cache.  The cache is used whenever searching for variables from
the global environment (possibly as part of a recursive search).

@menu
* Search paths::                
* Namespaces::                  
* Hash table::                  
@end menu

@node Search paths, Namespaces, Environments and variable lookup, Environments and variable lookup
@subsection Search paths

@cindex search path
@Sl{} has the notion of a `search path': the lookup for a `variable'
leads (possibly through a series of frames) to the `session frame' the
`working directory' and then along the search path.  The search path is
a series of databases (as returned by @code{search()}) which contain the
system functions (but not necessarily at the end of the path, as by
default the equivalent of packages are added at the end).

@R{} has a variant on the @Sl{} model.  There is a search path (also
returned by @code{search()}) which consists of the global environment
(aka user workspace) followed by environments which have been attached
and finally the base environment.  Note that unlike @Sl{} it is not
possible to attach environments before the workspace nor after the base
environment.

However, the notion of variable lookup is more general in @R{}, hence
the plural in the title of this subsection.  Since environments have
enclosures, from any environment there is a search path found by looking
in the frame, then the frame of its enclosure and so on.  Since loops
are not allowed, this process will eventually terminate: it can
terminate at either the base environment or the empty environment.  (It
can be conceptually simpler to think of the search always terminating at
the empty environment, but with an optimization to stop at the base
environment.)  So the `search path' describes the chain of environments
which is traversed once the search reaches the global environment.

@node Namespaces, Hash table, Search paths, Environments and variable lookup
@subsection Namespaces

@cindex namespace
Namespaces are environments associated with packages (and once again
the base package is special and will be considered separately).  A
package @code{@var{pkg}} defines two environments
@code{namespace:@var{pkg}} and @code{package:@var{pkg}}: it is
@code{package:@var{pkg}} that can be @code{attach}ed and form part of
the search path.

The objects defined by the @R{} code in the package are symbols with
bindings in the @code{namespace:@var{pkg}} environment.  The
@code{package:@var{pkg}} environment is populated by selected symbols
from the @code{namespace:@var{pkg}} environment (the exports).  The
enclosure of this environment is an environment populated with the
explicit imports from other namespaces, and the enclosure of
@emph{that} environment is the base namespace.  (So the illusion of the
imports being in the namespace environment is created via the
environment tree.)  The enclosure of the base namespace is the global
environment, so the search from a package namespace goes via the
(explicit and implicit) imports to the standard `search path'.

@cindex base namespace
@cindex namespace, base
@findex R_BaseNamespace
The base namespace environment @code{R_BaseNamespace} is another
@code{ENVSXP} that is special-cased.  It is effectively the same thing
as the base environment @code{R_BaseEnv} @emph{except} that its
enclosure is the global environment rather than the empty environment:
the internal code diverts lookups in its frame to the global symbol
table.

@node Hash table,  , Namespaces, Environments and variable lookup
@subsection Hash table

Environments in @R{} usually have a hash table, and nowadays that is the
default in @code{new.env()}.  It is stored as a @code{VECSXP} where
@code{length} is used for the allocated size of the table and
@code{truelength} is the number of primary slots in use---the pointer to
the @code{VECSXP} is part of the header of a @code{SEXP} of type
@code{ENVSXP}, and this points to @code{R_NilValue} if the environment
is not hashed.

For the pros and cons of hashing, see a basic text on Computer Science.

The code to implement hashed environments is in @file{src/main/envir.c}.
Unless set otherwise (e.g.@: by the @code{size} argument of
@code{new.env()}) the initial table size is @code{29}.  The table will
be resized by a factor of 1.2 once the load factor (the proportion of
primary slots in use) reaches 85%.

The hash chains are stored as pairlist elements of the @code{VECSXP}:
items are inserted at the front of the pairlist.  Hashing is principally
designed for fast searching of environments, which are from time to time
added to but rarely deleted from, so items are not actually deleted but
have their value set to @code{R_UnboundValue}.


@node Attributes, Contexts, Environments and variable lookup, R Internal Structures
@section Attributes

@cindex attributes
@findex ATTRIB
@findex SET_ATTRIB
@findex DUPLICATE_ATTRIB
As we have seen, every @code{SEXPREC} has a pointer to the attributes of
the node (default @code{R_NilValue}).  The attributes can be
accessed/set by the macros/functions @code{ATTRIB} and
@code{SET_ATTRIB}, but such direct access is normally only used to check
if the attributes are @code{NULL} or to reset them.  Otherwise access
goes through the functions @code{getAttrib} and @code{setAttrib} which
impose restrictions on the attributes.  One thing to watch is that if
you copy attributes from one object to another you may (un)set the
@code{"class"} attribute and so need to copy the object and S4 bits as
well.  There is a macro/function @code{DUPLICATE_ATTRIB} to automate
this.

Note that the `attributes' of a @code{CHARSXP} are used as part of the
management of the @code{CHARSXP} cache: of course @code{CHARSXP}'s are
not user-visible but C-level code might look at their attributes.

The code assumes that the attributes of a node are either
@code{R_NilValue} or a pairlist of non-zero length (and this is checked
by @code{SET_ATTRIB}).  The attributes are named (via tags on the
pairlist).  The replacement function @code{attributes<-} ensures that
@code{"dim"} precedes @code{"dimnames"} in the pairlist.  Attribute
@code{"dim"} is one of several that is treated specially: the values are
checked, and any @code{"names"} and @code{"dimnames"} attributes are
removed.  Similarly, you cannot set @code{"dimnames"} without having set
@code{"dim"}, and the value assigned must be a list of the correct
length and with elements of the correct lengths (and all zero-length
elements are replaced by @code{NULL}).

The other attributes which are given special treatment are
@code{"names"}, @code{"class"}, @code{"tsp"}, @code{"comment"} and
@code{"row.names"}.  For pairlist-like objects the names are not stored
as an attribute but (as symbols) as the tags: however the @R{} interface
makes them look like conventional attributes, and for one-dimensional
arrays they are stored as the first element of the @code{"dimnames"}
attribute.  The C code ensures that the @code{"tsp"} attribute is an
@code{REALSXP}, the frequency is positive and the implied length agrees
with the number of rows of the object being assigned to.  Classes and
comments are restricted to character vectors, and assigning a
zero-length comment or class removes the attribute.  Setting or removing
a @code{"class"} attribute sets the object bit appropriately.  Integer
row names are converted to and from the internal compact representation.

@cindex copying semantics
Care needs to be taken when adding attributes to objects of the types
with non-standard copying semantics.  There is only one object of type
@code{NILSXP}, @code{R_NilValue}, and that should never have attributes
(and this is enforced in @code{installAttrib}).  For environments,
external pointers and weak references, the attributes should be relevant
to all uses of the object: it is for example reasonable to have a name
for an environment, and also a @code{"path"} attribute for those
environments populated from @R{} code in a package.

@cindex attributes, preserving
@cindex preserving attributes
When should attributes be preserved under operations on an object?
Becker, Chambers & Wilks (1988, pp. 144--6) give some guidance.  Scalar
functions (those which operate element-by-element on a vector and whose
output is similar to the input) should preserve attributes (except
perhaps class, and if they do preserve class they need to preserve the
@code{OBJECT} and S4 bits).  Binary operations normally call
@findex copyMostAttrib
@code{copyMostAttrib} to copy most attributes from the longer
argument (and if they are of the same length from both, preferring the
values on the first).  Here `most' means all except the @code{names},
@code{dim} and @code{dimnames} which are set appropriately by the code
for the operator.

Subsetting (other than by an empty index) generally drops all attributes
except @code{names}, @code{dim} and @code{dimnames} which are reset as
appropriate.  On the other hand, subassignment generally preserves such
attributes even if the length is changed.  Coercion drops all
attributes. For example:

@example
> x <- structure(1:8, names=letters[1:8], comm="a comment")
> x[]
a b c d e f g h
1 2 3 4 5 6 7 8
attr(,"comm")
[1] "a comment"
> x[1:3]
a b c
1 2 3
> x[3] <- 3
> x
a b c d e f g h
1 2 3 4 5 6 7 8
attr(,"comm")
[1] "a comment"
> x[9] <- 9
> x
a b c d e f g h
1 2 3 4 5 6 7 8 9
attr(,"comm")
[1] "a comment"
@end example


@node Contexts, Argument evaluation, Attributes, R Internal Structures
@section Contexts

@cindex context
@emph{Contexts} are the internal mechanism used to keep track of where a
computation has got to (and from where), so that control-flow constructs
can work and reasonable information can be produced on error conditions
(such as @emph{via} traceback), and otherwise (the @code{sys.@var{xxx}}
functions).

Execution contexts are a stack of C @code{structs}:

@example
typedef struct RCNTXT @{
    struct RCNTXT *nextcontext; /* @r{The next context up the chain} */
    int callflag;               /* @r{The context `type'} */
    JMP_BUF cjmpbuf;            /* @r{C stack and register information} */
    int cstacktop;              /* @r{Top of the pointer protection stack} */
    int evaldepth;              /* @r{Evaluation depth at inception} */
    SEXP promargs;              /* @r{Promises supplied to closure} */
    SEXP callfun;               /* @r{The closure called} */
    SEXP sysparent;             /* @r{Environment the closure was called from} */
    SEXP call;                  /* @r{The call that effected this context} */
    SEXP cloenv;                /* @r{The environment} */
    SEXP conexit;               /* @r{Interpreted @code{on.exit} code} */
    void (*cend)(void *);       /* @r{C @code{on.exit} thunk} */
    void *cenddata;             /* @r{Data for C @code{on.exit} thunk} */
    char *vmax;                 /* @r{Top of the @code{R_alloc} stack} */
    int intsusp;                /* @r{Interrupts are suspended} */
    SEXP handlerstack;          /* @r{Condition handler stack} */
    SEXP restartstack;          /* @r{Stack of available restarts} */
    struct RPRSTACK *prstack;   /* @r{Stack of pending promises} */
@} RCNTXT, *context;
@end example

@noindent
plus additional fields for the byte-code compiler.  The `types'
are from

@example
enum @{
    CTXT_TOPLEVEL = 0,  /* @r{toplevel context} */
    CTXT_NEXT     = 1,  /* @r{target for @code{next}} */
    CTXT_BREAK    = 2,  /* @r{target for @code{break}} */
    CTXT_LOOP     = 3,  /* @r{@code{break} or @code{next} target} */
    CTXT_FUNCTION = 4,  /* @r{function closure} */
    CTXT_CCODE    = 8,  /* @r{other functions that need error cleanup} */
    CTXT_RETURN   = 12, /* @r{@code{return()} from a closure} */
    CTXT_BROWSER  = 16, /* @r{return target on exit from browser} */
    CTXT_GENERIC  = 20, /* @r{rather, running an S3 method} */
    CTXT_RESTART  = 32, /* @r{a call to @code{restart} was made from a closure} */
    CTXT_BUILTIN  = 64  /* @r{builtin internal function} */
@};
@end example

@noindent
where the @code{CTXT_FUNCTION} bit is on wherever function closures are
involved.

Contexts are created by a call to @code{begincontext} and ended by a
call to @code{endcontext}: code can search up the stack for a
particular type of context via @code{findcontext} (and jump there) or
jump to a specific context via @code{R_JumpToContext}.
@code{R_ToplevelContext} is the `idle' state (normally the command
prompt), and @code{R_GlobalContext} is the top of the stack.

Note that whilst calls to closures set a context, internal functions never
do and primitive builtins only set it when profiling or when they are
interfaces to foreign functions.

The byte-code compiler generates a map of instructions to source references
and expressions at compile time, which allows to produce information on
error conditions.  As an optimization, the byte-code interpreter then does
not set a context in some cases, such as in simple loops or when inlining
simple builtins or wrappers for internal functions.

@findex UseMethod
@cindex method dispatch
Dispatching from a S3 generic (via @code{UseMethod} or its internal
equivalent) or calling @code{NextMethod} sets the context type to
@code{CTXT_GENERIC}.  This is used to set the @code{sysparent} of the
method call to that of the @code{generic}, so the method appears to have
been called in place of the generic rather than from the generic.

The @R{} @code{sys.frame} and @code{sys.call} functions work by counting
calls to closures (type @code{CTXT_FUNCTION}) from either end of the
context stack.

Note that the @code{sysparent} element of the structure is not the same
thing as @code{sys.parent()}.  Element @code{sysparent} is primarily
used in managing changes of the function being evaluated, i.e.@: by
@code{Recall} and method dispatch.

@code{CTXT_CCODE} contexts are currently used in @code{cat()},
@code{load()}, @code{scan()} and @code{write.table()} (to close the
connection on error), by @code{PROTECT}, serialization (to recover from
errors, e.g.@: free buffers) and within the error handling code (to
raise the C stack limit and reset some variables).


@node Argument evaluation, Autoprinting, Contexts, R Internal Structures
@section Argument evaluation

@cindex argument evaluation
As we have seen, functions in @R{} come in three types, closures
(@code{SEXPTYPE} @code{CLOSXP}), specials (@code{SPECIALSXP}) and
builtins (@code{BUILTINSXP}).  In this section we consider when (and if)
the actual arguments of function calls are evaluated.  The rules are
different for the internal (special/builtin) and @R{}-level functions
(closures).

For a call to a closure, the actual and formal arguments are matched and
a matched call (another @code{LANGSXP}) is constructed.  This process
first replaces the actual argument list by a list of promises to the
values supplied.  It then constructs a new environment which contains
the names of the formal parameters matched to actual or default values:
all the matched values are promises, the defaults as promises to be
evaluated in the environment just created.  That environment is then
used for the evaluation of the body of the function, and promises will
be forced (and hence actual or default arguments evaluated) when they
are encountered.
@findex NAMED
(Evaluating a promise sets @code{NAMED = NAMEDMAX} on its value, so if the
argument was a symbol its binding is regarded as having multiple
references during the evaluation of the closure call.)
[The @code{NAMED} mechanism has been replaced by reference counting.]

If the closure is an S3 generic (that is, contains a call to
@code{UseMethod}) the evaluation process is the same until the
@code{UseMethod} call is encountered.  At that point the argument on
which to do dispatch (normally the first) will be evaluated if it has
not been already.  If a method has been found which is a closure, a new
evaluation environment is created for it containing the matched
arguments of the method plus any new variables defined so far during the
evaluation of the body of the generic.  (Note that this means changes to
the values of the formal arguments in the body of the generic are
discarded when calling the method, but @emph{actual} argument promises
which have been forced retain the values found when they were forced.
On the other hand, missing arguments have values which are promises to
use the default supplied by the method and not by the generic.)  If the
method found is a primitive it is called with the matched argument list
of promises (possibly already forced) used for the generic.

@cindex builtin function
@cindex special function
@cindex primitive function
@cindex .Internal function
The essential difference@footnote{There is currently one other
difference: when profiling builtin functions are counted as function
calls but specials are not.} between special and builtin functions is
that the arguments of specials are not evaluated before the C code is
called, and those of builtins are.  Note that being a special/builtin is
separate from being primitive or @code{.Internal}: @code{quote} is a
special primitive, @code{+} is a builtin primitive, @code{cbind} is a
special @code{.Internal} and @code{grep} is a builtin @code{.Internal}.

@cindex generic, internal
@findex DispatchOrEval
Many of the internal functions are internal generics, which for specials
means that they do not evaluate their arguments on call, but the C code
starts with a call to @code{DispatchOrEval}.  The latter evaluates the
first argument, and looks for a method based on its class.  (If S4
dispatch is on, S4 methods are looked for first, even for S3 classes.)
If it finds a method, it dispatches to that method with a call based on
promises to evaluate the remaining arguments.  If no method is found,
the remaining arguments are evaluated before return to the internal
generic.

@cindex generic, generic
@findex DispatchGeneric
The other way that internal functions can be generic is to be group
generic.  Most such functions are builtins (so immediately evaluate all
their arguments), and all contain a call to the C function
@code{DispatchGeneric}.  There are some peculiarities over the number of
arguments for the @code{"Math"} group generic, with some members
allowing only one argument, some having two (with a default for the
second) and @code{trunc} allows one or more but the default method only
accepts one.

@menu
* Missingness::                 
* Dot-dot-dot arguments::       
@end menu

@node Missingness, Dot-dot-dot arguments, Argument evaluation, Argument evaluation
@subsection Missingness

@cindex missingness
Actual arguments to (non-internal) @R{} functions can be fewer than are
required to match the formal arguments of the function.  Having
unmatched formal arguments will not matter if the argument is never used
(by lazy evaluation),  but when the argument is evaluated, either its
default value is evaluated (within the evaluation environment of the
function) or an error is thrown with a message along the lines of

@example
argument "foobar" is missing, with no default
@end example

@findex MISSING
@findex R_MissingArg
Internally missingness is handled by two mechanisms. The object
@code{R_MissingArg} is used to indicate that a formal argument has no
(default) value.  When matching the actual arguments to the formal
arguments, a new argument list is constructed from the formals all of
whose values are @code{R_MissingArg} with the first @code{MISSING} bit
set.  Then whenever a formal argument is matched to an actual argument,
the corresponding member of the new argument list has its value set to
that of the matched actual argument, and if that is not
@code{R_MissingArg} the missing bit is unset.

This new argument list is used to form the evaluation frame for the
function, and if named arguments are subsequently given a new value
(before they are evaluated) the missing bit is cleared.

Missingness of arguments can be interrogated via the @code{missing()}
function.  An argument is clearly missing if its missing bit is set or
if the value is @code{R_MissingArg}.  However, missingness can be passed
on from function to function, for using a formal argument as an actual
argument in a function call does not count as evaluation.  So
@code{missing()} has to examine the value (a promise) of a
non-yet-evaluated formal argument to see if it might be missing, which
might involve investigating a promise and so on @dots{}.

Special primitives also need to handle missing arguments, and in some
case (e.g.@: @code{log}) that is why they are special and not
builtin.  This is usually done by testing if an argument's value is
@code{R_MissingArg}.

@node Dot-dot-dot arguments,  , Missingness, Argument evaluation
@subsection Dot-dot-dot arguments

@cindex ... argument
Dot-dot-dot arguments are convenient when writing functions, but
complicate the internal code for argument evaluation.

The formals of a function with a @code{...} argument represent that as a
single argument like any other argument, with tag the symbol
@code{R_DotsSymbol}.  When the actual arguments are matched to the
formals, the value of the @code{...} argument is of @code{SEXPTYPE}
@code{DOTSXP}, a pairlist of promises (as used for matched arguments)
but distinguished by the @code{SEXPTYPE}.

Recall that the evaluation frame for a function initially contains the
@code{@var{name}=@var{value}} pairs from the matched call, and hence
this will be true for @code{...} as well.  The value of @code{...} is a
(special) pairlist whose elements are referred to by the special symbols
@code{..1}, @code{..2}, @dots{} which have the @code{DDVAL} bit set:
when one of these is encountered it is looked up (via @code{ddfindVar})
in the value of the @code{...} symbol in the evaluation frame.

Values of arguments matched to a @code{...} argument can be missing.

Special primitives may need to handle @code{...} arguments: see for
example the internal code of @code{switch} in file
@file{src/main/builtin.c}.

@node Autoprinting, The write barrier, Argument evaluation, R Internal Structures
@section Autoprinting

@cindex autoprinting
@findex R_Visible

Whether the returned value of a top-level @R{} expression is printed is
controlled by the global boolean variable @code{R_Visible}.  This is set
(to true or false) on entry to all primitive and internal functions
based on the @code{eval} column of the table in file
@file{src/main/names.c}: the appropriate setting can be extracted by the
macro @code{PRIMPRINT}.
@findex PRIMPRINT

@findex invisible
The @R{} primitive function @code{invisible} makes use of this
mechanism: it just sets @code{R_Visible = FALSE} before entry and
returns its argument.

For most functions the intention will be that the setting of
@code{R_Visible} when they are entered is the setting used when they
return, but there need to be exceptions.  The @R{} functions
@code{identify}, @code{options}, @code{system} and @code{writeBin}
determine whether the result should be visible from the arguments or
user action.  Other functions themselves dispatch functions which may
change the visibility flag: examples@footnote{the other current example
is left brace, which is implemented as a primitive.} are
@code{.Internal}, @code{do.call}, @code{eval}, @code{withVisible},
@code{if}, @code{NextMethod}, @code{Recall}, @code{recordGraphics},
@code{standardGeneric}, @code{switch} and @code{UseMethod}.

`Special' primitive and internal functions evaluate their arguments
internally @emph{after} @code{R_Visible} has been set, and evaluation of
the arguments (e.g.@: an assignment as in PR#9263) can change the value
of the flag.

The @code{R_Visible} flag can also get altered during the evaluation of
a function, with comments in the code about @code{warning},
@code{writeChar} and graphics functions calling @code{GText} (PR#7397).
(Since the C-level function @code{eval} sets @code{R_Visible}, this
could apply to any function calling it.  Since it is called when
evaluating promises, even object lookup can change @code{R_Visible}.)
Internal and primitive functions force the documented setting of
@code{R_Visible} on return, unless the C code is allowed to change it
(the exceptions above are indicated by @code{PRIMPRINT} having value 2).

The actual autoprinting is done by @code{PrintValueEnv} in file
@file{print.c}.  If the object to be printed has the S4 bit set and S4
methods dispatch is on, @code{show} is called to print the object.
Otherwise, if the object bit is set (so the object has a
@code{"class"} attribute), @code{print} is called to dispatch methods:
for objects without a class the internal code of @code{print.default}
is called.


@node The write barrier, Serialization Formats, Autoprinting, R Internal Structures
@section The write barrier and the garbage collector

@cindex write barrier
@cindex garbage collector
@R{} has long had a generational garbage collector, and bit @code{gcgen}
in the @code{sxpinfo} header is used in the implementation of this.
This is used in conjunction with the @code{mark} bit to identify two
previous generations.

There are three levels of collections.  Level 0 collects only the
youngest generation, level 1 collects the two youngest generations and
level 2 collects all generations.  After 20 level-0 collections the next
collection is at level 1, and after 5 level-1 collections at level 2.
Further, if a level-@var{n} collection fails to provide 20% free space
(for each of nodes and the vector heap), the next collection will be at
level @var{n+1}.  (The @R{}-level function @code{gc()} performs a
level-2 collection.)

A generational collector needs to efficiently `age' the objects,
especially list-like objects (including @code{STRSXP}s).  This is done
by ensuring that the elements of a list are regarded as at least as old
as the list @emph{when they are assigned}.  This is handled by the
functions @code{SET_VECTOR_ELT} and @code{SET_STRING_ELT}, which is why
they are functions and not macros.  Ensuring the integrity of such
operations is termed the @dfn{write barrier} and is done by making the
@code{SEXP} opaque and only providing access via functions (which cannot
be used as lvalues in assignments in C).

All code in @R{} extensions is behind the write barrier.  @R{}
extensions cannot directly access the internals of the @code{SEXPREC}s.
Base code can access internals if @samp{USE_RINTERNALS} is defined.
This is normally defined in @file{Defn.h} when @R{} is compiled.  To
enable a check on the way that the access is used, @R{} can be compiled
with flag @option{--enable-strict-barrier} which ensures that header
@file{Defn.h} does not define @samp{USE_RINTERNALS} and hence that
@code{SEXP} is opaque in most of @R{} itself.  (There are some necessary
exceptions: foremost in file @file{memory.c} where the accessor
functions are defined and also in file @file{size.c} which needs access
to the sizes of the internal structures.)

For background papers see
@uref{https://homepage.stat.uiowa.edu/~luke/R/barrier.html} and
@uref{https://homepage.stat.uiowa.edu/~luke/R/gengcnotes.html}.

@node Serialization Formats, Encodings for CHARSXPs, The write barrier, R Internal Structures
@section Serialization Formats

@cindex serialization
Serialized versions of @R{} objects are used by @code{load}/@code{save}
and also at a slightly lower level by @code{saveRDS}/@code{readRDS} (and
their earlier `internal' dot-name versions) and
@code{serialize}/@code{unserialize}.  These differ in what they
serialize to (a file, a connection, a raw vector) and whether they are
intended to serialize a single object or a collection of objects
(typically the workspace).  @code{save} writes a header at the beginning
of the file (a single LF-terminated line) which the lower-level versions
do not.

@code{save} and @code{saveRDS} allow various forms of compression, and
@command{gzip} compression is the default (except for @acronym{ASCII}
saves).  Compression is applied to the whole file stream, including the
headers, so serialized files can be uncompressed or re-compressed by
external programs.  Both @code{load} and @code{readRDS} can read
@command{gzip}, @command{bzip2} and @command{xz} forms of compression
when reading from a file, and @command{gzip} compression when reading
from a connection.

@R{} has used the same serialization format called `version 2' from @R{}
1.4.0 in December 2001 until @R{} 3.5.3 in March 2019.  It has been expanded
in back-compatible ways since its inception, for example to support
additional @code{SEXPTYPE}s.  Earlier formats are still supported via
@code{load} and @code{save} but such formats are not described here.  The
current default serialization format is called `version 3', and has been
introduced in @R{} 3.5.0.

@code{save} works by writing a single-line header (typically
@code{RDX2\n} for a binary save: the only other current value is
@code{RDA2\n} for @code{save(files=TRUE)}), then creating a tagged
pairlist of the objects to be saved and serializing that single object.
@code{load} reads the header line, unserializes a single object (a
pairlist or a vector list) and assigns the elements of the object in the
specified environment.  The header line serves two purposes in @R{}: it
identifies the serialization format so @code{load} can switch to the
appropriate reader code, and the newline @code{\n} allows the detection of files
which have been subjected to a non-binary transfer which re-mapped line
endings.  It can also be thought of as a `magic number' in the sense
used by the @command{file} program (although @R{} save files are not yet
by default known to that program).

Serialization in @R{} needs to take into account that objects may
contain references to environments, which then have enclosing
environments and so on.  (Environments recognized as package or name
space environments are saved by name.)  There are `reference objects'
which are not duplicated on copy and should remain shared on
unserialization.  These are weak references, external pointers and
environments other than those associated with packages, namespaces and
the global environment.  These are handled via a hash table, and
references after the first are written out as a reference marker indexed
by the table entry.

Version-2 serialization first writes a header indicating the format
(normally @samp{X\n} for an XDR format binary save, but @samp{A\n},
ASCII, and @samp{B\n}, native word-order binary, can also occur) and
then three integers giving the version of the format and two @R{}
versions (packed by the @code{R_Version} macro from @file{Rversion.h}).
(Unserialization interprets the two versions as the version of @R{}
which wrote the file followed by the minimal version of @R{} needed to
read the format.)  Serialization then writes out the object recursively
using function @code{WriteItem} in file @file{src/main/serialize.c}.

Some objects are written as if they were @code{SEXPTYPE}s: such
pseudo-@code{SEXPTYPE}s cover @code{R_NilValue}, @code{R_EmptyEnv},
@code{R_BaseEnv}, @code{R_GlobalEnv}, @code{R_UnboundValue},
@code{R_MissingArg} and @code{R_BaseNamespace}.

For all @code{SEXPTYPE}s except @code{NILSXP}, @code{SYMSXP} and
@code{ENVSXP} serialization starts with an integer with the
@code{SEXPTYPE} in bits 0:7@footnote{only bits 0:4 are currently used
for @code{SEXPTYPE}s but values 241:255 are used for
pseudo-@code{SEXPTYPE}s.} followed by the object bit, two bits
indicating if there are any attributes and if there is a tag (for the
pairlist types), an unused bit and then the @code{gp}
field@footnote{Currently the only relevant bits are 0:1, 4, 14:15.} in
bits 12:27.  Pairlist-like objects write their attributes (if any), tag
(if any), CAR and then CDR (using tail recursion): other objects write
their attributes after themselves.  Atomic vector objects write their
length followed by the data: generic vector-list objects write their
length followed by a call to @code{WriteItem} for each element.  The
code for @code{CHARSXP}s special-cases @code{NA_STRING} and writes it as
length @code{-1} with no data.  Lengths no more than @code{2^31 - 1} are
written in that way and larger lengths (which only occur on 64-bit
systems) as @code{-1} followed by the upper and lower 32-bits as integers
(regarded as unsigned).

Environments are treated in several ways: as we have seen, some are
written as specific pseudo-@code{SEXPTYPE}s.  Package and namespace
environments are written with pseudo-@code{SEXPTYPE}s followed by the
name.  `Normal' environments are written out as @code{ENVSXP}s with an
integer indicating if the environment is locked followed by the
enclosure, frame, `tag' (the hash table) and attributes.

In the `XDR' format integers and doubles are written in bigendian order:
however the format is not fully XDR (as defined in RFC 1832) as byte
quantities (such as the contents of @code{CHARSXP} and @code{RAWSXP}
types) are written as-is and not padded to a multiple of four bytes.

The `ASCII' format writes 7-bit characters.  Integers are formatted with
@code{%d} (except that @code{NA_integer_} is written as @code{NA}),
doubles formatted with @code{%.16g} (plus @code{NA}, @code{Inf} and
@code{-Inf}) and bytes with @code{%02x}.  Strings are written using
standard escapes (e.g.@: @code{\t} and @code{\013}) for non-printing and
non-@acronym{ASCII} bytes.

Version-3 serialization extends version-2 by support for custom
serialization of @code{ALTREP} framework objects.  It also stores the
current native encoding at serialization time, so that unflagged strings can
be converted if unserialized in R running under different native encoding.

@node Encodings for CHARSXPs, The CHARSXP cache, Serialization Formats, R Internal Structures
@section Encodings for CHARSXPs

Character data in @R{} are stored in the sexptype @code{CHARSXP}.

There is support for encodings other than that of the current locale, in
particular UTF-8 and the multi-byte encodings used on Windows for CJK
languages. A limited means to indicate the encoding of a @code{CHARSXP}
is @emph{via} two of the `general purpose' bits which are used to declare
the encoding to be either Latin-1 or UTF-8.  (Note that it is possible
for a character vector to contain elements in different encodings.)
Both printing and plotting notice the declaration and convert the string
to the current locale (possibly using @code{<xx>} to display in
hexadecimal bytes that are not valid in the current locale).  Many (but
not all) of the character manipulation functions will either preserve
the declaration or re-encode the character string.

Strings that refer to the OS such as file names need to be passed
through a wide-character interface on some OSes (e.g.@: Windows).

When are character strings declared to be of known encoding?  One way is
to do so directly via @code{Encoding}.  The parser declares the encoding
if this is known, either via the @code{encoding} argument to
@code{parse} or from the locale within which parsing is being done at
the @R{} command line.  (Other ways are recorded on the help page for
@code{Encoding}.)

It is not necessary to declare the encoding of @acronym{ASCII} strings
as they will work in any locale.  @acronym{ASCII} strings should never
have a marked encoding, as any encoding will be ignored when entering
such strings into the @code{CHARSXP} cache.

The rationale behind considering only UTF-8 and Latin-1 was that most
systems are capable of producing UTF-8 strings and this is the nearest
we have to a universal format.  For those that do not (for example those
lacking a powerful enough @code{iconv}), it is likely that they work in
Latin-1, the old @R{} assumption. Then the parser can return a
UTF-8-encoded string if it encounters a @samp{\uxxxx} escape for a
Unicode point that cannot be represented in the current charset.  (This
needs MBCS support, and in the past was only enabled@footnote{See define
@code{USE_UTF8_IF_POSSIBLE} in old versions of file @file{src/main/gram.c}.} on
Windows.)  Now this is enabled for all platforms, and a @samp{\uxxxx} or
@samp{\Uxxxxxxxx} escape ensures that the parsed string will be marked
as UTF-8.

Most of the character manipulation functions now preserve UTF-8
encodings: there are some notes as to which at the top of file
@file{src/main/character.c} and in file
@file{src/library/base/man/Encoding.Rd}.

Graphics devices are offered the possibility of handing UTF-8-encoded
strings without re-encoding to the native character set, by setting
@code{hasTextUTF8} to be @samp{TRUE} and supplying functions
@code{textUTF8} and @code{strWidthUTF8} that expect UTF-8-encoded
inputs.  Normally the symbol font is encoded in Adobe Symbol encoding,
but that can be re-encoded to UTF-8 by setting @code{wantSymbolUTF8} to
@samp{TRUE}.  The Windows' port of cairographics has a rather peculiar
assumption: it wants the symbol font to be encoded in UTF-8 as if it
were encoded in Latin-1 rather than Adobe Symbol: this is selected by
@code{wantSymbolUTF8 = NA_LOGICAL}.

Windows with MSVCRT as the C runtime has no UTF-8 locales, but rather expects to work with
UCS-2@footnote{or UTF-16 if support for surrogates is enabled in the OS,
which it used not to be when encoding support was added to @R{}.}
strings.  @R{} (being written in standard C) would not work internally
with UCS-2 without extensive changes.  The @file{Rgui}
console@footnote{but not the GraphApp toolkit.} uses UCS-2 internally,
but communicates with the @R{} engine in the native encoding.  To allow
UTF-8 strings to be printed in UTF-8 in @file{Rgui.exe}, an escape
convention is used (see header file @file{rgui_UTF8.h}) by
@code{cat}, @code{print} and autoprinting.

`Unicode' (UCS-2LE) files are common in the Windows world, and
@code{readLines} and @code{scan} will read them into UTF-8 strings on
Windows if the encoding is declared explicitly on an unopened
connection passed to those functions.

Windows have multiple notions of the current locale encoding, one is in the
C runtime (C library) and another is the active code page (system locale). 
The active code page is used when calling non-UTF-16 variants of Windows API
functions (earlier referred to as ANSI calls), either directly or indirectly
via POSIX wrappers inside MinGW-w64, from R, R packages and libraries they
link to.  While R has handled many cases by calling directly the UTF-16
variants of the Windows API, it still may sometimes use the non-UTF-16 ones,
and external libraries also primarily developed for POSIX systems typically
do that.  Therefore, for R to reliably work with (non-ASCII) strings on
Windows, both the C locale encoding and the active code page on Windows must
be the same, and by default they are.

The Windows UCRT C runtime supports UTF-8 locales.  Historically, the active
code page was a system-wide setting, changing which required a reboot, and
UTF-8 was not supported.  Later a "BETA: Use Unicode UTF-8 for worldwide
language support" feature has been added to set the active code page to
UTF-8, but this still required a reboot and impacted all applications, many
of which would not work correctly with that unexpected setting, so it could
not be used in practice.

Windows since Windows 10 (version 1903), Windows Server 2022 (LTSC), and
Windows Server 1903 (semi-annual channel) allow setting the active code page
to UTF-8 in the application manifest.  This changes the active code page
only for the given application and does so together with changing also the
UCRT C locale to UTF-8. R 4.2 for Windows uses this feature to get UTF-8 as
the native encoding on Windows. To make that possible, R had to switch to
UCRT, which in turn required creation of Rtools42.

Older versions of Windows still rely on the previous encoding support where
the native encoding cannot be UTF-8. R 4.2 requires UCRT to work, but UCRT
can be installed on Windows since Vista SP2 and Windows Server 2008 SP2. It
is shipped with Windows since Windows 10 and Windows Server 2016.

@node The CHARSXP cache, Warnings and errors, Encodings for CHARSXPs, R Internal Structures
@section The CHARSXP cache

@findex mkChar
There is a global cache for @code{CHARSXP}s created by @code{mkChar} ---
the cache ensures that most @code{CHARSXP}s with the same contents share
storage (`contents' including any declared encoding).  Not all
@code{CHARSXP}s are part of the cache -- notably @samp{NA_STRING} is
not. @code{CHARSXP}s reloaded from the @code{save} formats of @R{} prior
to 0.99.0 are not cached (since the code used is frozen and very few
examples still exist).

@findex mkCharLenCE
The cache records the encoding of the string as well as the bytes: all
requests to create a @code{CHARSXP} should be @emph{via} a call to
@code{mkCharLenCE}.  Any encoding given in @code{mkCharLenCE} call will
be ignored if the string's bytes are all @acronym{ASCII} characters.


@node Warnings and errors, S4 objects, The CHARSXP cache, R Internal Structures
@section Warnings and errors

@findex warning
@findex warningcall
@findex error
@findex errorcall

Each of @code{warning} and @code{stop} have two C-level equivalents,
@code{warning}, @code{warningcall}, @code{error} and @code{errorcall}.
The relationship between the pairs is similar: @code{warning} tries to
fathom out a suitable call, and then calls @code{warningcall} with that
call as the first argument if it succeeds, and with @code{call =
R_NilValue} if it does not.  When @code{warningcall} is called, it
includes the deparsed call in its printout unless @code{call =
R_NilValue}.

@code{warning} and @code{error} look at the context stack.  If the
topmost context is not of type @code{CTXT_BUILTIN}, it is used to
provide the call, otherwise the next context provides the call.
This means that when these functions are called from a primitive or
@code{.Internal}, the imputed call will not be to
primitive/@code{.Internal} but to the function calling the
primitive/@code{.Internal} .  This is exactly what one wants for a
@code{.Internal}, as this will give the call to the closure wrapper.
(Further, for a @code{.Internal}, the call is the argument to
@code{.Internal}, and so may not correspond to any @R{} function.)
However, it is unlikely to be what is needed for a primitive.

The upshot is that that @code{warningcall} and @code{errorcall} should
normally be used for code called from a primitive, and @code{warning}
and @code{error} should be used for code called from a @code{.Internal}
(and necessarily from @code{.Call}, @code{.C} and so on, where the call
is not passed down).  However, there are two complications.  One is that
code might be called from either a primitive or a @code{.Internal}, in
which case probably @code{warningcall} is more appropriate.  The other
involves replacement functions, where the call was once of the form
@example
> length(x) <- y ~ x
Error in "length<-"(`*tmp*`, value = y ~ x) : invalid value
@end example

@noindent
which is unpalatable to the end user.  For replacement functions there
will be a suitable context at the top of the stack, so @code{warning}
should be used.  (The results for @code{.Internal} replacement functions
such as @code{substr<-} are not ideal.)



@node S4 objects, Memory allocators, Warnings and errors, R Internal Structures
@section S4 objects

[This section is currently a preliminary draft and should not be taken
as definitive.  The description assumes that @env{R_NO_METHODS_TABLES}
has not been set.]

@menu
* Representation of S4 objects::  
* S4 classes::                  
* S4 methods::                  
* Mechanics of S4 dispatch::    
@end menu

@node Representation of S4 objects, S4 classes, S4 objects, S4 objects
@subsection Representation of S4 objects

S4 objects can be of any @code{SEXPTYPE}.  They are either an object of
a simple type (such as an atomic vector or function) with S4 class
information or of type @code{S4SXP}.  In all cases, the `S4 bit' (bit 4
of the `general purpose' field) is set, and can be tested by the
macro/function @code{IS_S4_OBJECT}.

S4 objects are created via @code{new()}@footnote{This can also create
non-S4 objects, as in @code{new("integer")}.} and thence via the C
function @code{R_do_new_object}.  This duplicates the prototype of the
class, adds a class attribute and sets the S4 bit.  All S4 class
attributes should be character vectors of length one with an attribute
giving (as a character string) the name of the package (or
@code{.GlobalEnv}) containing the class definition.  Since S4 objects
have a class attribute, the @code{OBJECT} bit is set.

It is currently unclear what should happen if the class attribute is
removed from an S4 object, or if this should be allowed.

@node S4 classes, S4 methods, Representation of S4 objects, S4 objects
@subsection S4 classes

S4 classes are stored as @R{} objects in the environment in which they
are created, with names @code{.__C__@var{classname}}: as such they are
not listed by default by @code{ls}.

The objects are S4 objects of class @code{"classRepresentation"} which
is defined in the @pkg{methods} package.

Since these are just objects, they are subject to the normal scoping
rules and can be imported and exported from namespaces like other
objects.  The directives @code{importClassesFrom} and
@code{exportClasses} are merely convenient ways to refer to class
objects without needing to know their internal `metaname' (although
@code{exportClasses} does a little sanity checking via @code{isClass}).

@node S4 methods, Mechanics of S4 dispatch, S4 classes, S4 objects
@subsection S4 methods

Details of the methods are stored in environments (typically hidden in the
respective namespace) with a non-syntactic name of the form
@code{.__T__@var{generic}:@var{package}} containing objects of class
@code{MethodDefinition} for all methods defined in the current environment
for the named generic derived from a specific package (which might be @code{.GlobalEnv}).
This is sometimes referred to as a `methods table'.

For example,
@example
 length(nM <- asNamespace("Matrix") )                    # 941 for Matrix 1.2-6
 length(meth <- grep("^[.]__T__", names(nM), value=TRUE))# 107 generics with methods
 length(meth.Ops <- nM$`.__T__Ops:base`) # 71 methods for the 'Ops' (group)generic
 head(sort(names(meth.Ops))) ## "abIndex#abIndex" ... "ANY#ddiMatrix" "ANY#ldiMatrix" "ANY#Matrix"
@end example

During an @R{} session there is an environment associated with each
non-primitive generic containing objects @code{.AllMTable},
@code{.Generic}, @code{.Methods}, @code{.MTable}, @code{.SigArgs} and
@code{.SigLength}.  @code{.MTable} and @code{AllMTable} are merged
methods tables containing all the methods defined directly and via
inheritance respectively.  @code{.Methods} is a merged methods list.

Exporting methods from a namespace is more complicated than exporting a
class.  Note first that you do not export a method, but rather the
directive @code{exportMethods} will export all the methods defined in
the namespace for a specified generic: the code also adds to the list
of generics any that are exported directly.  For generics which are
listed via @code{exportMethods} or exported themselves, the
corresponding environment is exported and so
will appear (as hidden object) in the package environment.

Methods for primitives which are internally S4 generic (see below) are
always exported, whether mentioned in the @file{NAMESPACE} file or not.

Methods can be imported either via the directive
@code{importMethodsFrom} or via importing a namespace by @code{import}.
Also, if a generic is imported via @code{importFrom}, its methods are
also imported.  In all cases the generic will be imported if it is in
the namespace, so @code{importMethodsFrom} is most appropriate for
methods defined on generics in other packages.  Since methods for a
generic could be imported from several different packages, the methods
tables are merged.

When a package is attached
@code{methods:::cacheMetaData} is called to update the internal tables:
only the visible methods will be cached.


@node Mechanics of S4 dispatch,  , S4 methods, S4 objects
@subsection Mechanics of S4 dispatch

This subsection does not discuss how S4 methods are chosen: see
@uref{https://developer.@/r-project.org/howMethodsWork.pdf}.

For all but primitive functions, setting a method on an existing
function that is not itself S4 generic creates a new object in the
current environment which is a call to @code{standardGeneric} with the
old definition as the default method.  Such S4 generics can also be
created @emph{via} a call to @code{setGeneric}@footnote{although this is
not recommended as it is less future-proof.} and are standard closures
in the @R{} language, with environment the environment within which they
are created.  With the advent of namespaces this is somewhat
problematic: if @code{myfn} was previously in a package with a name
space there will be two functions called @code{myfn} on the search
paths, and which will be called depends on which search path is in use.
This is starkest for functions in the base namespace, where the
original will be found ahead of the newly created function from any
other package.

Primitive functions are treated quite differently, for efficiency
reasons: this results in different semantics.  @code{setGeneric} is
disallowed for primitive functions.  The @pkg{methods} namespace
contains a list @code{.BasicFunsList} named by primitive functions:
the entries are either @code{FALSE} or a standard S4 generic showing
the effective definition.  When @code{setMethod} (or
@code{setReplaceMethod}) is called, it either fails (if the list entry
is @code{FALSE}) or a method is set on the effective generic given in
the list.

Actual dispatch of S4 methods for almost all primitives piggy-backs on
the S3 dispatch mechanism, so S4 methods can only be dispatched for
primitives which are internally S3 generic.  When a primitive that is
internally S3 generic is called with a first argument which is an S4
object and S4 dispatch is on (that is, the @pkg{methods} namespace is
loaded), @code{DispatchOrEval} calls @code{R_possible_dispatch} (defined
in file @file{src/main/objects.c}).  (Members of the S3 group generics,
which includes all the generic operators, are treated slightly
differently: the first two arguments are checked and
@code{DispatchGroup} is called.)  @code{R_possible_dispatch} first
checks an internal table to see if any S4 methods are set for that
generic (and S4 dispatch is currently enabled for that generic), and if
so proceeds to S4 dispatch using methods stored in another internal
table.  All primitives are in the base namespace, and this mechanism
means that S4 methods can be set for (some) primitives and will always
be used, in contrast to setting methods on non-primitives.

The exception is @code{%*%}, which is S4 generic but not S3 generic as
its C code contains a direct call to @code{R_possible_dispatch}.

The primitive @code{as.double} is special, as @code{as.numeric} and
@code{as.real} are copies of it.  The @pkg{methods} package code partly
refers to generics by name and partly by function, and maps
@code{as.double} and @code{as.real} to @code{as.numeric} (since that is
the name used by packages exporting methods for it).

Some elements of the language are implemented as primitives, for example
@code{@}}.  This includes the subset and subassignment `functions' and
they are S4 generic, again piggybacking on S3 dispatch.

@code{.BasicFunsList} is generated when @pkg{methods} is installed, by
computing all primitives, initially disallowing methods on all and then
setting generics for members of @code{.GenericArgsEnv}, the S4 group
generics and a short exceptions list in file @file{BasicFunsList.R}: this
currently contains the subsetting and subassignment operators and an
override for @code{c}.

@node Memory allocators, Internal use of global and base environments, S4 objects, R Internal Structures
@section Memory allocators

@R{}'s memory allocation is almost all done via routines in file
@file{src/main/memory.c}. 

The rest of @R{} should where possible make use of the allocators made
available by file @file{src/main/memory.c}, which are also the methods
recommended in
@ifset UseExternalXrefs
@ref{Memory allocation, , Memory allocation, R-exts, Writing R Extensions}
@end ifset
@ifclear UseExternalXrefs
`Writing R Extensions'
@end ifclear
@findex R_alloc
@findex R_Calloc
@findex R_Realloc
@findex R_Free
for use in @R{} packages, namely the use of @code{R_alloc},
@code{R_Calloc}, @code{R_Realloc} and @code{R_Free}.  Memory allocated by
@code{R_alloc} is freed by the garbage collector once the `watermark'
has been reset by calling
@findex vmaxset
@code{vmaxset}.  This is done automatically by the wrapper code calling
primitives and @code{.Internal} functions (and also by the wrapper code
to @code{.Call} and @code{.External}), but
@findex vmaxget
@code{vmaxget} and @code{vmaxset} can be used to reset the watermark
from within internal code if the memory is only required for a short
time.

@findex alloca
All of the methods of memory allocation mentioned so far are relatively
expensive.  All @R{} platforms support @code{alloca}, and in almost all
cases@footnote{but apparently not on Windows.} this is managed by the
compiler, allocates memory on the C stack and is very efficient.

There are two disadvantages in using @code{alloca}.  First, it is
fragile and care is needed to avoid writing (or even reading) outside
the bounds of the allocation block returned.  Second, it increases the
danger of overflowing the C stack.   It is suggested that it is only
used for smallish allocations (up to tens of thousands of bytes), and
that

@findex R_CheckStack
@example
    R_CheckStack();
@end example

@noindent
is called immediately after the allocation (as @R{}'s stack checking
mechanism will warn far enough from the stack limit to allow for modest
use of alloca).  (@code{do_makeunique} in file @file{src/main/unique.c}
provides an example of both points.)

There is an alternative check,
@findex R_CheckStack2
@example
    R_CheckStack2(size_t extra);
@end example

@noindent
to be called immediately @emph{before} trying an allocation of
@code{extra} bytes.

An alternative strategy has been used for various functions which
require intermediate blocks of storage of varying but usually small
size, and this has been consolidated into the routines in the header
file @file{src/main/RBufferUtils.h}.  This uses a structure which
contains a buffer, the current size and the default size. A call to
@findex R_AllocStringBuffer
@example
    R_AllocStringBuffer(size_t blen, R_StringBuffer *buf);
@end example

@noindent
sets @code{buf->data} to a memory area of at least @code{blen+1} bytes.
At least the default size is used, which means that for small
allocations the same buffer can be reused.  A call to
@findex R_FreeStringBufferL
@findex R_FreeStringBuffer
@code{R_FreeStringBufferL} releases memory if more than the default has
been allocated whereas a call to @code{R_FreeStringBuffer} frees any
memory allocated.

The @code{R_StringBuffer} structure needs to be initialized, for example by

@example
static R_StringBuffer ex_buff = @{NULL, 0, MAXELTSIZE@};
@end example

@noindent
which uses a default size of @code{MAXELTSIZE = 8192} bytes.  Most
current uses have a static @code{R_StringBuffer} structure, which
allows the (default-sized) buffer to be shared between calls to e.g.@:
@code{grep} and even between functions: this will need to be changed if
@R{} ever allows concurrent evaluation threads.  So the idiom is

@example
static R_StringBuffer ex_buff = @{NULL, 0, MAXELTSIZE@};
...
    char *buf;
    for(i = 0; i < n; i++) @{
        compute len
        buf = R_AllocStringBuffer(len, &ex_buff);
        use buf
    @}
    /*  free allocation if larger than the default, but leave
        default allocated for future use */
   R_FreeStringBufferL(&ex_buff);
@end example


@menu
* Internals of R_alloc::        
@end menu

@node Internals of R_alloc,  , Memory allocators, Memory allocators
@subsection Internals of R_alloc

The memory used by @code{R_alloc} is allocated as @R{} vectors, of type
@code{RAWSXP}.  Thus the allocation is in units of 8 bytes, and is
rounded up.  A request for zero bytes currently returns @code{NULL} (but
this should not be relied on).  For historical reasons, in all other
cases 1 byte is added before rounding up so the allocation is always
1--8 bytes more than was asked for: again this should not be relied on.

The vectors allocated are protected via the setting of @code{R_VStack},
as the garbage collector marks everything that can be reached from that
location.  When a vector is @code{R_alloc}ated, its @code{ATTRIB}
pointer is set to the current @code{R_VStack}, and @code{R_VStack} is
set to the latest allocation.  Thus @code{R_VStack} is a single-linked
chain of the vectors currently allocated via @code{R_alloc}.  Function
@code{vmaxset} resets the location @code{R_VStack}, and should be to a
value that has previously be obtained @emph{via} @code{vmaxget}:
allocations after the value was obtained will no longer be protected and
hence available for garbage collection.

@node Internal use of global and base environments, Modules, Memory allocators, R Internal Structures
@section Internal use of global and base environments

This section notes known use by the system of these environments: the
intention is to minimize or eliminate such uses.

@menu
* Base environment::            
* Global environment::          
@end menu

@node Base environment, Global environment, Internal use of global and base environments, Internal use of global and base environments
@subsection Base environment

@cindex base environment
@cindex environment, base
@findex .Device
@findex .Devices
The graphics devices system maintains two variables @code{.Device} and
@code{.Devices} in the base environment: both are always set.  The
variable @code{.Devices} gives a list of character vectors of the names
of open devices, and @code{.Device} is the element corresponding to the
currently active device.  The null device will always be open.

@findex .Options
There appears to be a variable @code{.Options}, a pairlist giving the
current options settings.  But in fact this is just a symbol with a
value assigned, and so shows up as a base variable.

@findex .Last.value
Similarly, the evaluator creates a symbol @code{.Last.value} which
appears as a variable in the base environment.

@findex .Traceback
@findex last.warning
Errors can give rise to objects @code{.Traceback} and
@code{last.warning} in the base environment.

@node Global environment,  , Base environment, Internal use of global and base environments
@subsection Global environment

@cindex global environment
@cindex environment, global
@findex .Random.seed
The seed for the random number generator is stored in object
@code{.Random.seed} in the global environment.

@findex dump.frames
Some error handlers may give rise to objects in the global environment:
for example @code{dump.frames} by default produces @code{last.dump}.

@findex .SavedPlots
The @code{windows()} device makes use of a variable @code{.SavedPlots}
to store display lists of saved plots for later display.  This is
regarded as a variable created by the user.


@node Modules, Visibility, Internal use of global and base environments, R Internal Structures
@section Modules

@cindex modules
@R{} makes use of a number of shared objects/DLLs stored in the
@file{modules} directory.  These are parts of the code which have been
chosen to be loaded `on demand' rather than linked as dynamic libraries
or incorporated into the main executable/dynamic library.

For the remaining modules the motivation has been the amount of (often
optional) code they will bring in @emph{via} libraries to which they are
linked.

@table @asis

@item @code{internet}
The internal HTTP and FTP clients and socket support, which link to
system-specific support libraries.  This may load @code{libcurl} and on
Windows will load @file{wininet.dll} and @file{ws2_32.dll}.

@item @code{lapack}
The code which makes use of the LAPACK library, and is linked to
@file{libRlapack} or an external LAPACK library.

@item @code{X11}
(Unix-alikes only.)  The @code{X11()}, @code{jpeg()}, @code{png()} and
@code{tiff()} devices. These are optional, and links to some or all of
the @code{X11}, @code{pango}, @code{cairo}, @code{jpeg}, @code{libpng}
and @code{libtiff} libraries.
@end table

@node Visibility, Lazy loading, Modules, R Internal Structures
@section Visibility
@cindex visibility

@menu
* Hiding C entry points::       
* Variables in Windows DLLs::   
@end menu

@node Hiding C entry points, Variables in Windows DLLs, Visibility, Visibility
@subsection Hiding C entry points

We make use of the visibility mechanisms discussed in
@ifset UseExternalXrefs
@ref{Controlling visibility, , Controlling visibility, R-exts, Writing R Extensions},
@end ifset
@ifclear UseExternalXrefs
section `Controlling Visibility' in `Writing R Extensions',
@end ifclear
C entry points not needed outside the main @R{} executable/dynamic
library (and in particular in no package nor module) should be prefixed
by @code{attribute_hidden}.
@findex attribute_hidden
Minimizing the visibility of symbols in the @R{} dynamic library will
speed up linking to it (which packages will do) and reduce the
possibility of linking to the wrong entry points of the same name.  In
addition, on some platforms reducing the number of entry points allows
more efficient versions of PIC to be used: somewhat over half the entry
points are hidden.  A convenient way to hide variables (as distinct from
functions) is to declare them @code{extern0} in header file @file{Defn.h}.

The visibility mechanism used is only available with some compilers and
platforms, and in particular not on Windows, where an alternative
mechanism is used.  Entry points will not be made available in
@file{R.dll} if they are listed in the file
@file{src/gnuwin32/Rdll.hide}.
@findex Rdll.hide
Entries in that file start with a space and must be strictly in
alphabetic order in the C locale (use @command{sort} on the file to
ensure this if you change it).  It is possible to hide Fortran as well
as C entry points via this file: the former are lower-cased and have an
underline as suffix, and the suffixed name should be included in the
file.  Some entry points exist only on Windows or need to be visible
only on Windows, and some notes on these are provided in file
@file{src/gnuwin32/Maintainters.notes}.

Because of the advantages of reducing the number of visible entry
points, they should be declared @code{attribute_hidden} where possible.
Note that this only has an effect on a shared-R-library build, and so
care is needed not to hide entry points that are legitimately used by
packages.  So it is best if the decision on visibility is made when a
new entry point is created, including the decision if it should be
included in header file @file{Rinternals.h}.  A list of the visible
entry points on shared-R-library build on a reasonably standard
Unix-alike can be made by something like

@example
nm -g libR.so | grep ' [BCDT] ' | cut -b20-
@end example

@node Variables in Windows DLLs,  , Hiding C entry points, Visibility
@subsection Variables in Windows DLLs

Windows is unique in that it conventionally treats importing variables
differently from functions: variables that are imported from a DLL need
to be specified  by a prefix (often @samp{_imp_}) when being linked to
(`imported') but not when being linked from (`exported').  The details
depend on the compiler system, and have changed for MinGW during the
lifetime of that port.  They are in the main hidden behind some macros
defined in header file @file{R_ext/libextern.h}.

A (non-function) variable in the main @R{} sources that needs to be
referred to outside @file{R.dll} (in a package, module or another DLL
such as @file{Rgraphapp.dll}) should be declared with prefix
@code{LibExtern}.  The main use is in @file{Rinternals.h}, but it needs
to be considered for any public header and also @file{Defn.h}.

It would nowadays be possible to make use of the `auto-import' feature
of the MinGW port of @command{ld} to fix up imports from DLLs (and if
@R{} is built for the Cygwin platform this is what happens).  However,
this was not possible when the MinGW build of @R{} was first constructed
in ca 1998, allows less control of visibility and would not work for
other Windows compiler suites.

It is only possible to check if this has been handled correctly by
compiling the @R{} sources on Windows.

@node Lazy loading,  , Visibility, R Internal Structures
@section Lazy loading

Lazy loading is always used for code in packages but is optional
(selected by the package maintainer) for datasets in packages.  When a
package/namespace which uses it is loaded, the package/namespace
environment is populated with promises for all the named objects: when
these promises are evaluated they load the actual code from a database.

There are separate databases for code and data, stored in the @file{R}
and @file{data} subdirectories.  The database consists of two files,
@file{@var{name}.rdb} and @file{@var{name}.rdx}.  The @file{.rdb} file
is a concatenation of serialized objects, and the @file{.rdx} file
contains an index.  The objects are stored in (usually) a
@command{gzip}-compressed format with a 4-byte header giving the
uncompressed serialized length (in XDR, that is big-endian, byte order)
and read by a call to the primitive @code{lazyLoadDBfetch}.  (Note that
this makes lazy-loading unsuitable for really large objects: the
unserialized length of an @R{} object can exceed 4GB.)

The index or `map' file @file{@var{name}.rdx} is a compressed serialized
@R{} object to be read by @code{readRDS}.  It is a list with three
elements @code{variables}, @code{references} and @code{compressed}.  The
first two are named lists of integer vectors of length 2 giving the
offset and length of the serialized object in the @file{@var{name}.rdb}
file.  Element @code{variables} has an entry for each named object:
@code{references} serializes a temporary environment used when named
environments are added to the database.  @code{compressed} is a logical
indicating if the serialized objects were compressed: compression is
always used nowadays. We later added the values @code{compressed = 2}
and @code{3} for @command{bzip2} and @command{xz} compression (with the
possibility of future expansion to other methods): these formats add a
fifth byte to the header for the type of compression, and store
serialized objects uncompressed if compression expands them.

Source references are treated specially for performance reasons: bindings
@code{lines} and @code{parseData} from @code{srcfile} environments are
loaded lazily.  This uses a mechanism that allows loading selected bindings
from an environment lazily.  The key for such environment is a list with two
elements: @code{eagerKey} gives the length-two integer key for the bindings
loaded eagerly and @code{lazyKeys} gives a vector of length-two integer
keys, one for each lazily loaded binding.

The loader for a lazy-load database of code or data is function
@code{lazyLoad} in the @pkg{base} package, but note that there is a
separate copy to load @pkg{base} itself in file
@file{R_HOME/base/R/base}.

Lazy-load databases are created by the code in
@file{src/library/tools/R/makeLazyLoad.R}: the main tool is the
unexported function @code{makeLazyLoadDB} and the insertion of database
entries is done by calls to @code{.Call("R_lazyLoadDBinsertValue",
...)}.

Lazy-load databases of less than 10MB are cached in memory at first use:
this was found necessary when using file systems with high latency
(removable devices and network-mounted file systems on Windows).

Lazy-load databases are loaded into the exports for a package, but not
into the namespace environment itself.  Thus they are visible when the
package is @emph{attached}, and also @emph{via} the @code{::} operator.
This was a deliberate design decision, as packages mostly make datasets
available for use by the end user (or other packages), and they should
not be found preferentially from functions in the package, surprising
users who expected the normal search path to be used.  (There is an
alternative mechanism, @file{sysdata.rda}, for `system datasets' that
are intended primarily to be used within the package.)

The same database mechanism is used to store parsed @file{Rd} files.
One or all of the parsed objects is fetched by a call to
@code{tools:::fetchRdDB}.

@node .Internal vs .Primitive, Internationalization in the R sources, R Internal Structures, Top
@chapter @code{.Internal} vs @code{.Primitive}

@findex .Internal
@findex .Primitive
C code compiled into @R{} at build time can be called directly in what
are termed @emph{primitives} or via the @code{.Internal} interface,
which is very similar to the @code{.External} interface except in
syntax.  More precisely, @R{} maintains a table of @R{} function names and
corresponding C functions to call, which by convention all start with
@samp{do_} and return a @code{SEXP}.  This table (@code{R_FunTab} in
file @file{src/main/names.c}) also specifies how many arguments to a
function are required or allowed, whether or not the arguments are to be
evaluated before calling, and whether the function is `internal' in
the sense that it must be accessed via the @code{.Internal} interface,
or directly accessible in which case it is printed in @R{} as
@code{.Primitive}.

Functions using @code{.Internal()} wrapped in a closure are in general
preferred as this ensures standard handling of named and default
arguments.  For example, @code{grep} is defined as

@example
@group
grep <-
function (pattern, x, ignore.case = FALSE, perl = FALSE, value = FALSE,
         fixed = FALSE, useBytes = FALSE, invert = FALSE)
@{
    if (!is.character(x)) x <- structure(as.character(x), names = names(x))
    .Internal(grep(as.character(pattern), x, ignore.case, value,
                   perl, fixed, useBytes, invert))
@}

@end group
@end example
@noindent
and the use of @code{as.character} allows methods to be dispatched (for
example, for factors).

However, for reasons of convenience and also efficiency (as there is
some overhead in using the @code{.Internal} interface wrapped in a
function closure), the primitive functions are exceptions that can be
accessed directly.  And of course, primitive functions are needed for
basic operations---for example @code{.Internal} is itself a primitive.
Note that primitive functions make no use of @R{} code, and hence are
very different from the usual interpreted functions.  In particular,
@code{formals} and @code{body} return @code{NULL} for such objects, and
argument matching can be handled differently.  For some primitives
(including @code{call}, @code{switch}, @code{.C} and @code{.subset})
positional matching is important to avoid partial matching of the first
argument.

The list of primitive functions is subject to change; currently, it
includes the following.

@enumerate

@item
``Special functions'' which really are @emph{language} elements, but
implemented as primitive functions:

@example
@group
@{       (         if     for      while  repeat  break  next
return  function  quote  switch
@end group
@end example

@item
Language elements and basic @emph{operator}s (i.e., functions usually
@emph{not} called as @code{foo(a, b, ...)}) for subsetting, assignment,
arithmetic, comparison and logic:

@example
@group
               [    [[    $    @@
<-   <<-  =    [<-  [[<-  $<-  @@<-

+    -    *    /     ^    %%   %*%  %/%
<    <=   ==   !=    >=   >
|    ||   &    &&    !
@end group
@end example

@noindent
When the arithmetic, comparison and logical operators are called as
functions, any argument names are discarded so positional matching is used.

@item
``Low level'' 0-- and 1--argument functions which belong to one of the
following groups of functions:

@enumerate a
@item
Basic mathematical functions with a single argument, i.e.,

@example
@group
abs     sign    sqrt
floor   ceiling
@end group

@group
exp     expm1
log2    log10   log1p
cos     sin     tan
acos    asin    atan
cosh    sinh    tanh
acosh   asinh   atanh
cospi   sinpi   tanpi
@end group

@group
gamma   lgamma  digamma trigamma
@end group

@group
cumsum  cumprod cummax  cummin
@end group

@group
Im  Re  Arg  Conj  Mod
@end group
@end example

@code{log} is a primitive function of one or two arguments with named
argument matching.

@code{trunc} is a difficult case: it is a primitive that can have one
or more arguments: the default method handled in the primitive has
only one.

@item
Functions rarely used outside of ``programming'' (i.e., mostly used
inside other functions), such as

@example
@group
nargs          missing        on.exit        interactive
as.call        as.character   as.complex     as.double
as.environment as.integer     as.logical     as.raw
is.array       is.atomic      is.call        is.character
is.complex     is.double      is.environment is.expression
is.finite      is.function    is.infinite    is.integer
is.language    is.list        is.logical     is.matrix
is.na          is.name        is.nan         is.null
is.numeric     is.object      is.pairlist    is.raw
is.real        is.recursive   is.single      is.symbol
baseenv        emptyenv       globalenv      pos.to.env
unclass        invisible      seq_along      seq_len
@end group
@end example

@item
The programming and session management utilities

@example
@group
browser  proc.time  gc.time tracemem retracemem untracemem
@end group
@end example

@end enumerate

@item
The following basic replacement and extractor functions

@example
@group
length      length<-
class       class<-
oldClass    oldClass<-
attr        attr<-
attributes  attributes<-
names       names<-
dim         dim<-
dimnames    dimnames<-
            environment<-
            levels<-
            storage.mode<-
@end group
@end example

@findex NAMED
@noindent
Note that optimizing @code{NAMED = 1} is only effective within a
primitive (as the closure wrapper of a @code{.Internal} will set
@code{NAMED = NAMEDMAX} when the promise to the argument is evaluated) and
hence replacement functions should where possible be primitive to avoid
copying (at least in their default methods).
[The @code{NAMED} mechanism has been replaced by reference counting.]

@item
The following functions are primitive for efficiency reasons:

@example
@group
:           ~           c           list
call        expression  substitute
UseMethod   standardGeneric
.C          .Fortran   .Call        .External
round       signif      rep         seq.int
@end group
@end example

@noindent
as well as the following internal-use-only functions

@example
@group
.Primitive      .Internal
.Call.graphics  .External.graphics
.subset         .subset2
.primTrace      .primUntrace
lazyLoadDBfetch
@end group
@end example

@end enumerate


The multi-argument primitives
@example
@group
call       switch
.C         .Fortran   .Call       .External
@end group
@end example

@noindent
intentionally use positional matching, and need to do so to avoid
partial matching to their first argument.  They do check that the first
argument is unnamed or for the first two, partially matches the formal
argument name.  On the other hand,

@example
@group
attr       attr<-     browser     rememtrace substitute  UseMethod
log        round      signif      rep        seq.int
@end group
@end example

@noindent
manage their own argument matching and do work in the standard way.

All the one-argument primitives check that if they are called with a
named argument that this (partially) matches the name given in the
documentation: this is also done for replacement functions with one
argument plus @code{value}.

The net effect is that argument matching for primitives intended for
end-user use @emph{as functions} is done in the same way as for
interpreted functions except for the six exceptions where positional
matching is required.

@menu
* Special primitives::          
* Special internals::           
* Prototypes for primitives::   
* Adding a primitive::          
@end menu

@node Special primitives, Special internals, .Internal vs .Primitive, .Internal vs .Primitive
@section Special primitives

A small number of primitives are @emph{specials} rather than
@emph{builtins}, that is they are entered with unevaluated arguments.
This is clearly necessary for the language constructs and the assignment
operators, as well as for @code{&&} and @code{||} which conditionally
evaluate their second argument, and @code{~}, @code{.Internal},
@code{call}, @code{expression}, @code{missing}, @code{on.exit},
@code{quote} and @code{substitute} which do not evaluate some of their
arguments.

@code{rep} and @code{seq.int} are special as they evaluate some of their
arguments conditional on which are non-missing.

@code{log}, @code{round} and @code{signif} are special to allow default
values to be given to missing arguments.

The subsetting, subassignment and @code{@@} operators are all special.
(For both extraction and replacement forms, @code{$} and @code{@@}
take a symbol argument, and @code{[} and @code{[[} allow missing
arguments.)

@code{UseMethod} is special to avoid the additional contexts added to
calls to builtins.

@node Special internals, Prototypes for primitives, Special primitives, .Internal vs .Primitive
@section Special internals

There are also special @code{.Internal} functions: @code{NextMethod},
@code{Recall}, @code{withVisible}, @code{cbind}, @code{rbind} (to allow
for the @code{deparse.level} argument), @code{eapply}, @code{lapply} and
@code{vapply}.

@node Prototypes for primitives, Adding a primitive, Special internals, .Internal vs .Primitive
@section Prototypes for primitives

Prototypes are available for the primitive functions and operators, and
these are used for printing, @code{args} and package checking (e.g.@: by
@code{tools::checkS3methods} and by package @CRANpkg{codetools}).  There are
two environments in the @pkg{base} package (and namespace),
@samp{.GenericArgsEnv} for those primitives which are internal S3
generics, and @samp{.ArgsEnv} for the rest.  Those environments contain
closures with the same names as the primitives, formal arguments derived
(manually) from the help pages, a body which is a suitable call to
@code{UseMethod} or @code{NULL} and environment the base namespace.

The C code for @code{print.default} and @code{args} uses the closures in
these environments in preference to the definitions in base (as
primitives).

The QC function @code{undoc} checks that all the functions prototyped in
these environments are currently primitive, and that the primitives not
included are better thought of as language elements (at the time of
writing

@example
$  $<-  &&  (  :  @@  @@<-  [  [[  [[<-  [<-  @{  ||  ~  <-  <<-  =
break  for function  if  next  repeat  return  while
@end example

@noindent
).  One could argue about @code{~}, but it is known to the parser and has
semantics quite unlike a normal function.  And @code{:} is documented
with different argument names in its two meanings.

The QC functions @code{codoc} and @code{checkS3methods} also make use of
these environments (effectively placing them in front of base in the
search path), and hence the formals of the functions they contain are
checked against the help pages by @code{codoc}.  However, there are two
problems with the generic primitives.  The first is that many of the
operators are part of the S3 group generic @code{Ops} and that defines
their arguments to be @code{e1} and @code{e2}: although it would be very
unusual, an operator could be called as e.g.@: @code{"+"(e1=a, e2=b)}
and if method dispatch occurred to a closure, there would be an argument
name mismatch.  So the definitions in environment @code{.GenericArgsEnv}
have to use argument names @code{e1} and @code{e2} even though the
traditional documentation is in terms of @code{x} and @code{y}:
@code{codoc} makes the appropriate adjustment via
@code{tools:::.make_S3_primitive_generic_env}.  The second discrepancy
is with the @code{Math} group generics, where the group generic is
defined with argument list @code{(x, ...)}, but most of the members only
allow one argument when used as the default method (and @code{round} and
@code{signif} allow two as default methods): again fix-ups are used.

Those primitives which are in @code{.GenericArgsEnv} are checked (via
@file{tests/primitives.R}) to be generic @emph{via} defining methods for
them, and a check is made that the remaining primitives are probably not
generic, by setting a method and checking it is not dispatched to (but
this can fail for other reasons).  However, there is no certain way to
know that if other @code{.Internal} or primitive functions are not
internally generic except by reading the source code.

@node Adding a primitive,  , Prototypes for primitives, .Internal vs .Primitive
@section Adding a primitive

[For R-core use: reverse this procedure to remove a primitive.  Most
commonly this is done by changing a @code{.Internal} to a primitive or
@emph{vice versa}.]

Primitives are listed in the table @code{R_FunTab} in
@file{src/main/names.c}: primitives have @samp{Y = 0} in the @samp{eval}
field.

There needs to be an @samp{\alias} entry in a help file in the @pkg{base}
package, and the primitive needs to be added to one of the lists at the
start of this section.

Some primitives are regarded as language elements (the current ones are
listed above).  These need to be added to two lists of exceptions,
@code{langElts} in @code{undoc()} (in file
@file{src/library/tools/R/QC.R}) and @code{lang_elements} in
@file{tests/primitives.R}.

All other primitives are regarded as functions and should be listed in
one of the environments defined in @file{src/library/base/R/zzz.R},
either @code{.ArgsEnv} or @code{.GenericArgsEnv}: internal generics also
need to be listed in the character vector @code{.S3PrimitiveGenerics}.
Note too the discussion about argument matching above: if you add a
primitive function with more than one argument by converting a
@code{.Internal} you need to add argument matching to the C code, and
for those with a single argument, add argument-name checking.

Do ensure that @command{make check-devel} has been run: that tests most
of these requirements.

@node Internationalization in the R sources, Package Structure, .Internal vs .Primitive, Top
@chapter Internationalization in the R sources

The process of marking messages (errors, warnings etc) for translation
in an @R{} package is described in
@ifset UseExternalXrefs
@ref{Internationalization, , Internationalization, R-exts, Writing R Extensions},
@end ifset
@ifclear UseExternalXrefs
`Writing R Extensions',
@end ifclear
and the standard packages included with @R{} have (with an exception in
@pkg{grDevices} for the menus of the @code{windows()} device) been
internationalized in the same way as other packages.

@menu
* R code::                      
* Main C code::                 
* Windows-GUI-specific code::   
* macOS GUI::                   
* Updating::                    
@end menu

@node R code, Main C code, Internationalization in the R sources, Internationalization in the R sources
@section R code

Internationalization for @R{} code is done in exactly the same way as
for extension packages.  As all standard packages which have @R{} code
also have a namespace, it is never necessary to specify @code{domain},
but for efficiency calls to @code{message}, @code{warning} and
@code{stop} should include @code{domain = NA} when the message is
constructed @emph{via} @code{gettextf}, @code{gettext} or
@code{ngettext}.

For each package, the extracted messages and translation sources are
stored under package directory @file{po} in the source package, and
compiled translations under @file{inst/po} for installation to package
directory @file{po} in the installed package.  This also applies to C
code in packages.

@node Main C code, Windows-GUI-specific code, R code, Internationalization in the R sources
@section Main C code

The main C code (e.g.@: that in files @file{src/*/*.c} and in
the modules) is where @R{} is closest to the sort of application for
which @samp{gettext} was written.  Messages in the main C code are in
domain @code{R} and stored in the top-level directory @file{po} with
compiled translations under @file{share/locale}.

The list of files covered by the @R{} domain is specified in file
@file{po/POTFILES.in}.

The normal way to mark messages for translation is via @code{_("msg")}
just as for packages.  However, sometimes one needs to mark passages for
translation without wanting them translated at the time, for example
when declaring string constants.  This is the purpose of the @code{N_}
macro, for example

@example
@{ ERROR_ARGTYPE,           N_("invalid argument type")@},
@end example

@noindent
from file @file{src/main/errors.c}.

The @code{P_} macro

@example
#ifdef ENABLE_NLS
#define P_(StringS, StringP, N) ngettext (StringS, StringP, N)
#else
#define P_(StringS, StringP, N) (N > 1 ? StringP: StringS)
#endif
@end example

@noindent
may be used
as a wrapper for @code{ngettext}: however in some cases the preferred
approach has been to conditionalize (on @code{ENABLE_NLS}) code using
@code{ngettext}.

The macro @code{_("msg")} can safely be used in directory
@file{src/appl}; the header for standalone @samp{nmath} skips possible
translation.  (This does not apply to @code{N_} or @code{P_}).


@node Windows-GUI-specific code, macOS GUI, Main C code, Internationalization in the R sources
@section Windows-GUI-specific code

Messages for the Windows GUI are in a separate domain @samp{RGui}.  This
was done for two reasons:

@itemize
@item
The translators for the Windows version of @R{} might be separate from
those for the rest of @R{} (familiarity with the GUI helps), and

@item
Messages for Windows are most naturally handled in the native charset
for the language, and in the case of CJK languages the charset is
Windows-specific.  (It transpires that as the @code{iconv} we ported
works well under Windows, this is less important than anticipated.)
@end itemize

Messages for the @samp{RGui} domain are marked by @code{G_("msg")}, a
macro that is defined in header file @file{src/gnuwin32/win-nls.h}.  The
list of files that are considered is hardcoded in function @code{update_RGui_po}
in file @file{src/library/tools/R/translations.R}, which is invoked via the
@code{update-RGui} target of file @file{po/Makefile.win}: note
that this includes @file{devWindows.c} as the menus on the
@code{windows} device are considered to be part of the GUI.  (There is
also @code{GN_("msg")}, the analogue of @code{N_("msg")}.)

The template and message catalogs for the @samp{RGui} domain are in the
top-level @file{po} directory of package @pkg{base}.


@node macOS GUI, Updating, Windows-GUI-specific code, Internationalization in the R sources
@section macOS GUI

This is handled separately: see
@uref{https://developer.r-project.org/Translations30.html}.


@node Updating,  , macOS GUI, Internationalization in the R sources
@section Updating

See file @file{po/README} for how to update the message templates and catalogs.

@node Package Structure, Files, Internationalization in the R sources, Top
@chapter Structure of an Installed Package

@menu
* Metadata::                    
* Help::                        
@end menu

The structure of a @emph{source} packages is described in @ref{Creating
R packages, , Creating R packages, R-exts, Writing R Extensions}: this
chapter is concerned with the structure of @emph{installed} packages.

An installed package has a top-level file @file{DESCRIPTION}, a copy of
the file of that name in the package sources with a @samp{Built} field
appended, and file @file{INDEX}, usually describing the objects on which
help is available, a file @file{NAMESPACE} if the package has a name
space, optional files such as @file{CITATION}, @file{LICENCE} and
@file{NEWS}, and any other files copied in from @file{inst}.  It will
have directories @file{Meta}, @file{help} and @file{html} (even if the
package has no help pages), almost always has a directory @file{R} and
often has a directory @file{libs} to contain compiled code.  Other
directories with known meaning to @R{} are @file{data}, @file{demo},
@file{doc} and @file{po}.

Function @code{library} looks for a namespace and if one is found
passes control to @code{loadNamespace}.  Then @code{library} or
@code{loadNamespace} looks for file @file{R/@var{pkgname}}, warns if it
is not found and otherwise sources the code (using @code{sys.source})
into the package's environment, then lazy-loads a database
@file{R/sysdata} if present.  So how @R{} code gets loaded depends on
the contents of  @file{R/@var{pkgname}}: a standard template to load
lazy-load databases are provided in @file{share/R/nspackloader.R}.

Compiled code is usually loaded when the package's namespace is loaded
by a @code{useDynlib} directive in a @file{NAMESPACE} file or by the
package's @code{.onLoad} function.  Conventionally compiled code is
loaded by a call to @code{library.dynam} and this looks in directory
@file{libs} (and in an appropriate sub-directory if sub-architectures
are in use) for a shared object (Unix-alike) or DLL (Windows).

Subdirectory @file{data} serves two purposes. In a package using
lazy-loading of data, it contains a lazy-load database @file{Rdata},
plus a file @file{Rdata.rds} which contain a named character vector used
by @code{data()} in the (unusual) event that it is used for such a
package.  Otherwise it is a copy of the @file{data} directory in the
sources, with saved images re-compressed if @command{R CMD INSTALL
--resave-data} was used.

Subdirectory @file{demo} supports the @code{demo} function, and is
copied from the sources.

Subdirectory @file{po} contains (in subdirectories) compiled message
catalogs.

@node Metadata, Help, Package Structure, Package Structure
@section Metadata

Directory @file{Meta} contains several files in @code{.rds} format, that
is serialized @R{} objects written by @code{saveRDS}.  All packages
have files @file{Rd.rds}, @file{hsearch.rds}, @file{links.rds},
@file{features.rds}, and
@file{package.rds}.  Packages with namespaces have a file
@file{nsInfo.rds}, and those with data, demos or vignettes have
@file{data.rds}, @file{demo.rds} or @file{vignette.rds} files.

The structure of these files (and their existence and names) is private
to @R{}, so the description here is for those trying to follow the @R{}
sources: there should be no reference to these files in non-base
packages.

File @file{package.rds} is a dump of information extracted from the
@file{DESCRIPTION} file.  It is a list of several components.  The
first, @samp{DESCRIPTION}, is a character vector, the @file{DESCRIPTION}
file as read by @code{read.dcf}.  Further elements @samp{Depends},
@samp{Suggests}, @samp{Imports}, @samp{Rdepends} and @samp{Rdepends2}
record the @samp{Depends}, @samp{Suggests} and @samp{Imports} fields.
These are all lists, and can be empty.  The first three have an entry
for each package named, each entry being a list of length 1 or 3, which
element @samp{name} (the package name) and optional elements @samp{op}
(a character string) and @samp{version} (an object of class
@samp{"package_version"}).  Element @samp{Rdepends} is used for the
first version dependency on @R{}, and @samp{Rdepends2} is a list of zero
or more @R{} version dependencies---each is a three-element list of the
form described for packages.  Element @samp{Rdepends} is no longer used,
but it is still potentially needed so @R{} < 2.7.0 can detect that the
package was not installed for it.

File @file{nsInfo.rds} records a list, a parsed version of the
@file{NAMESPACE} file.

File @file{Rd.rds} records a data frame with one row for each help file.
The columns are @samp{File} (the file name with extension), @samp{Name}
(the @samp{\name} section), @samp{Type} (from the optional
@samp{\docType} section), @samp{Title}, @samp{Encoding}, @samp{Aliases},
@samp{Concepts} and @samp{Keywords}.  All columns are character vectors
apart from @samp{Aliases}, which is a list of character vectors.

File @file{hsearch.rds} records the information to be used by
@samp{help.search}.  This is a list of four unnamed elements which are
character matrices for help files, aliases, keywords and concepts.  All
the matrices have columns @samp{ID} and @samp{Package} which are used to
tie the aliases, keywords and concepts (the remaining column of the last
three elements) to a particular help file.  The first element has
further columns @samp{LibPath} (stored as @code{""} and filled in what
the file is loaded), @samp{name}, @samp{title}, @samp{topic} (the first
alias, used when presenting the results as
@samp{@var{pkgname}::@var{topic}}) and @samp{Encoding}.

File @file{links.rds} records a named character vector, the names being
aliases and the values character strings of the form
@example
"../../@var{pkgname}/html/@var{filename}.html"
@end example

File @file{data.rds} records a two-column character matrix with columns
of dataset names and titles from the corresponding help file.  File
@file{demo.rds} has the same structure for package demos.

File @file{vignette.rds} records a data frame with one row for each
`vignette' (@file{.[RS]nw} file in @file{inst/doc}) and with columns
@samp{File} (the full file path in the sources), @samp{Title},
@samp{PDF} (the pathless file name of the installed PDF version, if
present), @samp{Depends}, @samp{Keywords} and @samp{R} (the pathless
file name of the installed @R{} code, if present).


@node Help,  , Metadata, Package Structure
@section Help

All installed packages, whether they had any @file{.Rd} files or not,
have @file{help} and @file{html} directories. The latter normally only
contains the single file @file{00Index.html}, the package index which
has hyperlinks to the help topics (if any).

Directory @file{help} contains files @file{AnIndex}, @file{paths.rds}
and @file{@var{pkgname}.rd[bx]}.  The latter two files are a lazy-load
database of parsed @file{.Rd} files, accessed by
@code{tools:::fetchRdDB}.  File @file{paths.rds} is a saved character
vector of the original path names of the @file{.Rd} files, used when
updating the database.

File @file{AnIndex} is a two-column tab-delimited file: the first column
contains the aliases defined in the help files and the second the
basename (without the @file{.Rd} or @file{.rd} extension) of the file
containing that alias.  It is read by @code{utils:::index.search} to
search for files matching a topic (alias), and read by @code{scan} in
@code{utils:::matchAvailableTopics}, part of the completion system.

File @file{aliases.rds} is the same information as @file{AnIndex} as a
named character vector (names the topics, values the file basename), for
faster access.

@node Files, Graphics Devices, Package Structure, Top
@chapter Files

@R{} provides many functions to work with files and directories: many of
these have been added relatively recently to facilitate scripting in
@R{} and in particular the replacement of Perl scripts by @R{} scripts
in the management of @R{} itself.

These functions are implemented by standard C/POSIX library calls,
except on Windows.  That means that filenames must be encoded in the
current locale as the OS provides no other means to access the file
system: increasingly filenames are stored in UTF-8 and the OS will
translate filenames to UTF-8 in other locales.  So using a UTF-8 locale
gives transparent access to the whole file system.

Windows is another story.  There the internal view of filenames is in
UTF-16LE (so-called `Unicode'), and standard C library calls can only
access files whose names can be expressed in the current codepage.  To
circumvent that restriction, there is a parallel set of Windows-specific
calls which take wide-character arguments for filepaths.  Much of the
file-handling in @R{} has been moved over to using these functions, so
filenames can be manipulated in @R{} as UTF-8 encoded character strings,
converted to wide characters (which on Windows are UTF-16LE) and passed
to the OS.  The utilities @code{RC_fopen} and @code{filenameToWchar}
help this process.  Currently @code{file.copy} to a directory,
@code{list.files}, @code{list.dirs} and @code{path.expand} work only
with filepaths encoded in the current codepage.

All these functions do tilde expansion, in the same way as
@code{path.expand}, with the deliberate exception of @code{Sys.glob}.

File names may be case sensitive or not: the latter is the norm on
Windows and macOS, the former on other Unix-alikes.  Note that this
is a property of both the OS and the file system: it is often possible
to map names to upper or lower case when mounting the file system.  This
can affect the matching of patterns in @code{list.files} and
@code{Sys.glob}.

File names commonly contain spaces on Windows and macOS but not
elsewhere.  As file names are handled as character strings by @R{},
spaces are not usually a concern unless file names are passed to other
process, e.g.@: by a @code{system} call.

Windows has another couple of peculiarities.  Whereas a POSIX file
system has a single root directory (and other physical file systems are
mounted onto logical directories under that root), Windows has separate
roots for each physical or logical file system (`volume'), organized
under @emph{drives} (with file paths starting @code{D:} for an
@acronym{ASCII} letter, case-insensitively) and @emph{network shares}
(with paths like @code{\netname\topdir\myfiles\a file}).  There is a
current drive, and path names without a drive part are relative to the
current drive.  Further, each drive has a current directory, and
relative paths are relative to that current directory, on a particular
drive if one is specified.  So @file{D:dir\file} and @file{D:} are valid
path specifications (the last being the current directory on drive
@file{D:}).

@c basename        Wchar   na
@c dir.create      Wchar   ~
@c dirname         Wchar   ~
@c getwd
@c file.access     Wchar   ~
@c file.append     RC_fopen
@c file.copy       no      ~ (+ file.append)
@c file.create     RC_fopen
@c file.edit       UTF-8   in R code
@c file.exists     Wchar   ~
@c file.info       Wchar   ~
@c file.link       8-bit   ~
@c file.remove     Wchar   ~
@c file.rename     Wchar   ~
@c file.show       UTF-8   in R code
@c file.symlink    not     ~
@c file_test
@c list.dirs       no      ~
@c list.files      no      ~
@c normalizePath   Wchar   ~
@c path.expand     no
@c setwd           Wchar   ~
@c Sys.chmod       Wchar   ~
@c Sys.glob        Wchar   not
@c Sys.readlink    not     ~
@c Sys.umask
@c unlink          Wchar   ~


@node Graphics Devices, GUI consoles, Files, Top
@chapter Graphics

@R{}'s graphics internals were re-designed to enable multiple graphics
systems to be installed on top on the graphics `engine' -- currently
there are two such systems, one supporting `base' graphics (based on
that in S and whose @R{} code@footnote{The C code is in files
@file{base.c}, @file{graphics.c}, @file{par.c}, @file{plot.c} and
@file{plot3d.c} in directory @file{src/main}.} is in package
@pkg{graphics}) and one implemented in package @pkg{grid}.

Some notes on the historical changes can be found at
@uref{https://www.stat.auckland.ac.nz/~paul/R/basegraph.html} and
@uref{https://www.stat.auckland.ac.nz/~paul/R/graphicsChanges.html}.

At the lowest level is a graphics device, which manages a plotting
surface (a screen window or a representation to be written to a file).
This implements a set of graphics primitives, to `draw'

@itemize
@item a circle, optionally filled
@item a rectangle, optionally filled
@item a line
@item a set of connected lines
@item a polygon, optionally filled
@item a paths, optionally filled using a winding rule
@item text
@item a raster image (optional)
@item and to set a clipping rectangle
@end itemize

@noindent
as well as requests for information such as

@itemize
@item the width of a string if plotted
@item the metrics (width, ascent, descent) of a single character
@item the current size of the plotting surface
@end itemize

@noindent
and requests/opportunities to take action such as

@itemize
@item start a new `page', possibly after responding to a request to ask
the user for confirmation.
@item return the position of the device pointer (if any).
@item when a device become the current device or stops being the current
device (this is usually used to change the window title on a screen
device).
@item when drawing starts or finishes (e.g.@: used to flush graphics to
the screen when drawing stops).
@item wait for an event, for example a mouse click or keypress.
@item an `onexit' action, to clean up if plotting is interrupted (by an
error or by the user).
@item capture the current contents of the device as a raster image.
@item close the device.
@end itemize



The device also sets a number of variables, mainly Boolean flags
indicating its capabilities.  Devices work entirely in `device units'
which are up to its developer: they can be in pixels, big points (1/72
inch), twips, @dots{}, and can differ@footnote{although that needs to be
handled carefully, as for example the @code{circle} callback is given a
radius (and that should be interpreted as in the x units).} in the
@samp{x} and @samp{y} directions.

@c think of the engine as colors.c, devices.c, engine.c, plotmath.c, vfonts.c
The next layer up is the graphics `engine' that is the main interface to
the device (although the graphics subsystems do talk directly to
devices).  This is responsible for clipping lines, rectangles and
polygons, converting the @code{pch} values @code{0...26} to sets of
lines/circles, centring (and otherwise adjusting) text, rendering
mathematical expressions (`plotmath') and mapping colour descriptions
such as names to the internal representation.

@c graphics.c looks at device dimensions, locator, metricinfo
@c par.c looks at various device pars
@c plot3d.c looks at useRotatedTextInContour
@c grid looks at size, clipping, locator, ipr

Another function of the engine is to manage display lists and snapshots.
Some but not all instances of graphics devices maintain display lists, a
`list' of operations that have been performed on the device to produce
the current plot (since the device was opened or the plot was last
cleared, e.g.@: by @code{plot.new}).  Screen devices generally maintain
a display list to handle repaint and resize events whereas file-based
formats do not---display lists are also used to implement
@code{dev.copy()} and friends.  The display list is a pairlist of
@code{.Internal} (base graphics) or @code{.Call.graphics} (grid
graphics) calls, which means that the C code implementing a graphics
operation will be re-called when the display list is replayed: apart
from the part which records the operation if successful.

Snapshots of the current graphics state are taken by
@code{GEcreateSnapshot} and replayed later in the session by
@code{GEplaySnapshot}.  These are used by @code{recordPlot()},
@code{replayPlot()} and the GUI menus of the @code{windows()} device.
The `state' includes the display list.


The top layer comprises the graphics subsystems. Although there is
provision for 24 subsystems since about 2001, currently still only two
exist, `base' and
`grid'.  The base subsystem is registered with the engine when @R{} is
initialized, and unregistered (via @code{KillAllDevices}) when an @R{}
session is shut down.  The grid subsystem is registered in its
@code{.onLoad} function and unregistered in the @code{.onUnload}
function.  The graphics subsystem may also have `state' information
saved in a snapshot (currently base does and grid does not).

Package @pkg{grDevices} was originally created to contain the basic
graphics devices (although @code{X11} is in a separate load-on-demand
module because of the volume of external libraries it brings in).  Since
then it has been used for other functionality that was thought desirable
for use with @pkg{grid}, and hence has been transferred from package
@pkg{graphics} to @pkg{grDevices}.  This is principally concerned with
the handling of colours and recording and replaying plots.

@menu
* Graphics devices::            
* Colours::                     
* Base graphics::               
* Grid graphics::               
@end menu

@node Graphics devices, Colours, Graphics Devices, Graphics Devices
@section Graphics Devices

@R{} ships with several graphics devices, and there is support for
third-party packages to provide additional devices---several packages
now do.  This section describes the device internals from the viewpoint
of a would-be writer of a graphics device.

@menu
* Device structures::           
* Device capabilities::         
* Handling text::               
* Conventions::                 
* 'Mode'::                      
* Graphics events::             
* Specific devices::            
@end menu

@node Device structures, Device capabilities, Graphics devices, Graphics devices
@subsection Device structures

There are two types used internally which are pointers to structures
related to graphics devices.

The @code{DevDesc} type is a structure defined in the header file
@file{R_ext/GraphicsDevice.h} (which is included by
@file{R_ext/GraphicsEngine.h}).  This describes the physical
characteristics of a device, the capabilities of the device driver and
contains a set of callback functions that will be used by the graphics
engine to obtain information about the device and initiate actions
(e.g.@: a new page, plotting a line or some text).  Type @code{pDevDesc}
is a pointer to this type.

The following callbacks can be omitted (or set to the null pointer,
their default value) when appropriate default behaviour will be taken by
the graphics engine: @code{activate}, @code{cap}, @code{deactivate},
@code{locator}, @code{holdflush} (API version 9), @code{mode},
@code{newFrameConfirm}, @code{path}, @code{raster} and @code{size}.

The relationship of device units to physical dimensions is set by the
element @code{ipr} of the @code{DevDesc} structure: a @samp{double}
array of length 2.


The @code{GEDevDesc} type is a structure defined in
@file{R_ext/GraphicsEngine.h} (with comments in the file) as

@example
typedef struct _GEDevDesc GEDevDesc;
struct _GEDevDesc @{
    pDevDesc dev;
    Rboolean displayListOn;
    SEXP displayList;
    SEXP DLlastElt;
    SEXP savedSnapshot;
    Rboolean dirty;
    Rboolean recordGraphics;
    GESystemDesc *gesd[MAX_GRAPHICS_SYSTEMS];
    Rboolean ask;
@}
@end example

@noindent
So this is essentially a device structure plus information about the
device maintained by the graphics engine and normally@footnote{It is
possible for the device to find the @code{GEDevDesc} which points to its
@code{DevDesc}, and this is done often enough that there is a
convenience function @code{desc2GEDesc} to do so.} visible to the engine
and not to the device.  Type @code{pGEDevDesc} is a pointer to this
type.

The graphics engine maintains an array of devices, as pointers to
@code{GEDevDesc} structures.  The array is of size 64 but the first
element is always occupied by the @code{"null device"} and the final
element is kept as NULL as a sentinel.@footnote{Calling
@code{R_CheckDeviceAvailable()} ensures there is a free slot or throws
an error.}  This array is reflected in the @R{} variable
@samp{.Devices}.  Once a device is killed its element becomes available
for reallocation (and its name will appear as @code{""} in
@samp{.Devices}).  Exactly one of the devices is `active': this is the
the null device if no other device has been opened and not killed.

Each instance of a graphics device needs to set up a @code{GEDevDesc}
structure by code very similar to

@example
    pGEDevDesc gdd;

    R_GE_checkVersionOrDie(R_GE_version);
    R_CheckDeviceAvailable();
    BEGIN_SUSPEND_INTERRUPTS @{
        pDevDesc dev;
        /* Allocate and initialize the device driver data */
        if (!(dev = (pDevDesc) calloc(1, sizeof(DevDesc))))
            return 0; /* or error() */
        /* set up device driver or free 'dev' and error() */
        gdd = GEcreateDevDesc(dev);
        GEaddDevice2(gdd, "dev_name");
    @} END_SUSPEND_INTERRUPTS;
@end example

The @code{DevDesc} structure contains a @code{void *} pointer
@samp{deviceSpecific} which is used to store data specific to the
device.  Setting up the device driver includes initializing all the
non-zero elements of the @code{DevDesc} structure.

Note that the device structure is zeroed when allocated: this provides
some protection against future expansion of the structure since the
graphics engine can add elements that need to be non-NULL/non-zero to be
`on' (and the structure ends with 64 reserved bytes which will be zeroed
and allow for future expansion).

Rather more protection is provided by the version number of the
engine/device API, @code{R_GE_version} defined in
@file{R_ext/GraphicsEngine.h} together with access functions

@example
int R_GE_getVersion(void);
void R_GE_checkVersionOrDie(int version);
@end example

@noindent
If a graphics device calls @code{R_GE_checkVersionOrDie(R_GE_version)}
it can ensure it will only be used in versions of @R{} which provide the
API it was designed for and compiled against.

The @code{DevDesc} structure also contains an @code{int} @samp{deviceVersion}
to indicate which version of the engine/device API that the device supports.
If the device driver sets this correctly, 
there is no need for a device driver to use 
@code{R_GE_checkVersionOrDie(R_GE_version)} because the graphics engine will
not make use of callbacks from an API version above the version that is
supported by the device.

@node Device capabilities, Handling text, Device structures, Graphics devices
@subsection Device capabilities

The following `capabilities' can be defined for the device's
@code{DevDesc} structure.

@itemize
@item @code{canChangeGamma} --
@code{Rboolean}: can the display gamma be adjusted?  This is now
ignored, as gamma support has been removed.
@item @code{canHadj} --
@code{integer}: can the device do horizontal adjustment of text
@emph{via} the @code{text} callback, and if so, how precisely? 0 = no
adjustment, 1 = @{0, 0.5, 1@} (left, centre, right justification) or 2 =
continuously variable (in [0,1]) between left and right justification.
@item @code{canGenMouseDown} --
@code{Rboolean}: can the device handle mouse down events?  This
flag and the next three are not currently used by R, but are maintained
for back compatibility.
@item @code{canGenMouseMove} --
@code{Rboolean}: ditto for mouse move events.
@item @code{canGenMouseUp} --
@code{Rboolean}: ditto for mouse up events.
@item @code{canGenKeybd} --
@code{Rboolean}: ditto for keyboard events.
@item @code{hasTextUTF8} --
@code{Rboolean}: should non-symbol text be sent (in UTF-8) to the
@code{textUTF8} and @code{strWidthUTF8} callbacks, and sent as Unicode
points (negative values) to the @code{metricInfo} callback?
@item @code{wantSymbolUTF8} --
@code{Rboolean}: should symbol text be handled in UTF-8 in the same way
as other text?  Requires @code{textUTF8 = TRUE}.
@item @code{haveTransparency}:
does the device support semi-transparent colours?
@item @code{haveTransparentBg}:
can the background be fully or semi-transparent?
@item @code{haveRaster}:
is there support for rendering raster images?
@item @code{haveCapture}:
is there support for @code{grid::grid.cap}?
@item @code{haveLocator}:
is there an interactive locator?
@item @code{deviceClip}:
should the engine leave @emph{all} clipping to the device?
@end itemize

@code{haveRaster},
@code{haveCapture}, and @code{haveLocator}
 can often be deduced to be false from the presence of
@code{NULL} entries instead of the corresponding functions.

In addition, the @code{capabilities} callback allows the device driver
to provide more detailed information, especially related to callbacks
in the engine/device API version 13 or higher.

The @code{capabilities} callback is called with a list of integer
vectors that represent the best guess that the graphics engine
can make, based on the flags in the @code{DevDesc} structure and
the @samp{deviceVersion}.  For some capabilities, the integer vector
is length 1 with @code{0} for no support, @code{1} for support, or
@code{NA} for unknown support.  For capabilities where support
can be more nuanced, the integer vector may either take higher values
or it may have length greater than 1,
though length 1 and @code{0} still means no support and @code{NA} still means 
unknown support.

The following components of 
this list are likely to need modifying (for these, the graphics engine 
can only guess @code{0}
if @samp{deviceVersion} is too low or @code{NA}
otherwise):

@itemize
@item The @code{patterns} component reports what sort of pattern fills are 
supported.  If the device supports one or more pattern types,
this component should be replaced with an integer vector containing
a value for each supported pattern type;
the graphics engine provides 
constants @code{R_GE_linearGradientPattern},
@code{R_GE_radialGradientPattern}, and 
@code{R_GE_tilingPattern}.
If the device does not provide support, this component should
be set to 0.

@item The @code{clippingPaths} component reports whether 
arbitrary clipping paths are supported. 
If the device supports clipping paths, this component should be set to 1.
If the device does not provide support, this component should
be set to 0.

@item The @code{masks} component reports what sort of masks are supported.
If the device supports one or more mask types,
this component should be replaced with an integer vector containing
a value for each supported mask type;
the graphics engine provides 
constants @code{R_GE_alphaMask} and @code{R_GE_luminanceMask}.
If the device does not provide support, this component should
be set to 0.

@item The @code{compositing} component reports which compositing operators 
are supported. 
If the device supports one or more compositing operators,
this component should be replaced with an integer vector containing
a value for each supported operator;
The list of possible operators is long, 
encompassing Porter-Duff operators and Adobe PDF Blend Modes; 
the graphics engine provides constants @code{R_GE_compositeClear}, etc.
If the device does not provide support, this component should
be set to 0.

@item The @code{transformations} component reports whether 
affine transformations are supported.
If the device supports transformations, this component should be set to 1.
If the device does not provide support, this component should
be set to 0.

@item The @code{paths} component reports whether stroking and 
filling of paths composed of multiple shapes is supported. 
If the device supports stroking and filling paths, 
this component should be set to 1.
If the device does not provide support, this component should
be set to 0.

@end itemize

The graphics engine provides constants like 
@code{R_GE_capability_patterns} for selecting the appropriate
component of the list of capabilities.

It is valid (if unhelpful) for the device driver to return the list
of capabilities unaltered.

@node Handling text, Conventions, Device capabilities, Graphics devices
@subsection Handling text

Handling text is probably the hardest task for a graphics device, and
the design allows for the device to optionally indicate that it has
additional capabilities.  (If the device does not, these will if
possible be handled in the graphics engine.)

The three callbacks for handling text that must be in all graphics
devices are @code{text}, @code{strWidth} and @code{metricInfo} with
declarations

@example
void text(double x, double y, const char *str, double rot, double hadj,
          pGgcontext gc, pDevDesc dd);

double strWidth(const char *str, pGEcontext gc, pDevDesc dd);

void metricInfo(int c, pGEcontext gc,
               double* ascent, double* descent, double* width,
               pDevDesc dd);
@end example

@noindent
The @samp{gc} parameter provides the graphics context, most importantly
the current font and fontsize, and @samp{dd} is a pointer to the active
device's structure.

The @code{text} callback should plot @samp{str} at @samp{(x,
y)}@footnote{in device coordinates} with an anti-clockwise rotation of
@samp{rot} degrees.  (For @samp{hadj} see below.)  The interpretation
for horizontal text is that the baseline is at @code{y} and the start is
a @code{x}, so any left bearing for the first character will start at
@code{x}.

The @code{strWidth} callback computes the width of the string which it
would occupy if plotted horizontally in the current font.  (Width here
is expected to include both (preferably) or neither of left and right
bearings.)

The @code{metricInfo} callback computes the size of a single
character: @code{ascent} is the distance it extends above the baseline
and @code{descent} how far it extends below the baseline.
@code{width} is the amount by which the cursor should be advanced when
the character is placed.  For @code{ascent} and @code{descent} this is
intended to be the bounding box of the `ink' put down by the glyph and
not the box which might be used when assembling a line of conventional
text (it needs to be for e.g.@: @code{hat(beta)} to work correctly).
However, the @code{width} is used in plotmath to advance to the next
character, and so needs to include left and right bearings.

The @emph{interpretation} of @samp{c} depends on the locale.  In a
single-byte locale values @code{32...255} indicate the corresponding
character in the locale (if present).  For the symbol font (as used by
@samp{graphics::par(font=5)}, @samp{grid::gpar(fontface=5}) and by
`plotmath'), values @code{32...126, 161...239, 241...254} indicate
glyphs in the Adobe Symbol encoding.  In a multibyte locale, @code{c}
represents a Unicode point (except in the symbol font).  So the function
needs to include code like

@example
    Rboolean Unicode = mbcslocale && (gc->fontface != 5);
    if (c < 0) @{ Unicode = TRUE; c = -c; @}
    if(Unicode) UniCharMetric(c, ...); else CharMetric(c, ...);
@end example

@noindent
In addition, if device capability @code{hasTextUTF8} (see below) is
true, Unicode points will be passed as negative values: the code snippet
above shows how to handle this.  (This applies to the symbol font only
if device capability @code{wantSymbolUTF8} is true.)

If possible, the graphics device should handle clipping of text.  It
indicates this by the structure element @code{canClip} which if true
will result in calls to the callback @code{clip} to set the clipping
region. If this is not done, the engine will clip very crudely (by
omitting any text that does not appear to be wholly inside the clipping
region).

The device structure has an integer element @code{canHadj}, which
indicates if the device can do horizontal alignment of text.  If this is
one, argument @samp{hadj} to @code{text} will be called as @code{0 ,0.5,
1} to indicate left-, centre- and right-alignment at the indicated
position.  If it is two, continuous values in the range @code{[0, 1]}
are assumed to be supported.

Capability @code{hasTextUTF8} if true, it has two consequences.
First, there are callbacks @code{textUTF8} and @code{strWidthUTF8} that
should behave identically to @code{text} and @code{strWidth} except that
@samp{str} is assumed to be in UTF-8 rather than the current locale's
encoding.  The graphics engine will call these for all text except in
the symbol font.  Second, Unicode points will be passed to the
@code{metricInfo} callback as negative integers.  If your device would
prefer to have UTF-8-encoded symbols, define @code{wantSymbolUTF8} as
well as @code{hasTextUTF8}.  In that case text in the symbol font is
sent to @code{textUTF8} and @code{strWidthUTF8}.

Some devices can produce high-quality rotated text, but those based on
bitmaps often cannot.  Those which can should set
@code{useRotatedTextInContour} to be true from graphics API version 4.

Several other elements relate to the precise placement of text by the
graphics engine:

@example
double xCharOffset;
double yCharOffset;
double yLineBias;
double cra[2];
@end example

@noindent
These are more than a little mysterious.  Element @code{cra} provides an
indication of the character size, @code{par("cra")} in base graphics, in
device units.  The mystery is what is meant by `character size': which
character, which font at which size?  Some help can be obtained by
looking at what this is used for.  The first element, `width', is not
used by @R{} except to set the graphical parameters.  The second,
`height', is use to set the line spacing, that is the relationship
between @code{par("mar")} and @code{par("mai")} and so on.  It is
suggested that a good choice is

@example
dd->cra[0] = 0.9 * fnsize;
dd->cra[1] = 1.2 * fnsize;
@end example

@noindent
where @samp{fnsize} is the `size' of the standard font (@code{cex=1})
on the device, in device units.  So for a 12-point font (the usual
default for graphics devices), @samp{fnsize} should be 12 points in
device units.

The remaining elements are yet more mysterious.  The @code{postscript()}
device says

@example
    /* Character Addressing Offsets */
    /* These offsets should center a single */
    /* plotting character over the plotting point. */
    /* Pure guesswork and eyeballing ... */

    dd->xCharOffset =  0.4900;
    dd->yCharOffset =  0.3333;
    dd->yLineBias = 0.2;
@end example

@noindent
It seems that @code{xCharOffset} is not currently used, and
@code{yCharOffset} is used by the base graphics system to set vertical
alignment in @code{text()} when @code{pos} is specified, and in
@code{identify()}.  It is occasionally used by the graphic engine when
attempting exact centring of text, such as character string values of
@code{pch} in @code{points()} or @code{grid.points()}---however, it is
only used when precise character metric information is not available or
for multi-line strings.

@code{yLineBias} is used in the base graphics system in @code{axis()} and
@code{mtext()} to provide a default for their @samp{padj} argument.

@node Conventions, 'Mode', Handling text, Graphics devices
@subsection Conventions

The aim is to make the (default) output from graphics devices as similar
as possible.  Generally people follow the model of the @code{postscript}
and @code{pdf} devices (which share most of their internal code).

The following conventions have become established:

@itemize

@item
The default size of a device should be 7 inches square.

@item
There should be a @samp{pointsize} argument which defaults to 12, and it
should give the pointsize in big points (1/72 inch).  How exactly this
is interpreted is font-specific, but it should use a font which works
with lines packed 1/6 inch apart, and looks good with lines 1/5 inch
apart (that is with 2pt leading).

@item
The default font family should be a sans serif font, e.g Helvetica or
similar (e.g.@: Arial on Windows).

@item
@code{lwd = 1} should correspond to a line width of 1/96 inch.  This
will be a problem with pixel-based devices, and generally there is a
minimum line width of 1 pixel (although this may not be appropriate
where anti-aliasing of lines is used, and @code{cairo} prefers a minimum
of 2 pixels).

@item
Even very small circles should be visible, e.g.@: by using a minimum
radius of 1 pixel or replacing very small circles by a single filled
pixel.

@item
How RGB colour values will be interpreted should be documented, and
preferably be sRGB.

@item
The help page should describe its policy on these conventions.

@end itemize

These conventions are less clear-cut for bitmap devices, especially
where the bitmap format does not have a design resolution.

The interpretation of the line texture (@code{par("lty"}) is described
in the header @file{GraphicsEngine.h} and in the help for @code{par}: note that the
`scale' of the pattern should be proportional to the line width (at
least for widths above the default).


@node 'Mode', Graphics events, Conventions, Graphics devices
@subsection `Mode'

One of the device callbacks is a function @code{mode}, documented in
the header as

@example
     * device_Mode is called whenever the graphics engine
     * starts drawing (mode=1) or stops drawing (mode=0)
     * GMode (in graphics.c) also says that
     * mode = 2 (graphical input on) exists.
     * The device is not required to do anything
@end example

@noindent
Since @code{mode = 2} has only recently been documented at device level.
It could be used to change the graphics cursor, but devices currently do
that in the @code{locator} callback.  (In base graphics the mode is set
for the duration of a @code{locator} call, but if @code{type != "n"} is
switched back for each point whilst annotation is being done.)

Many devices do indeed do nothing on this call, but some screen devices
ensure that drawing is flushed to the screen when called with @code{mode
= 0}.  It is tempting to use it for some sort of buffering, but note
that `drawing' is interpreted at quite a low level and a typical single
figure will stop and start drawing many times.  The buffering introduced
in the @code{X11()} device makes use of @code{mode = 0} to indicate
activity: it updates the screen after @emph{ca} 100ms of inactivity.

This callback need not be supplied if it does nothing.

@node Graphics events, Specific devices, 'Mode', Graphics devices
@subsection Graphics events

Graphics devices may be designed to handle user interaction: not all are.

Users may use @code{grDevices::setGraphicsEventEnv} to set the
@code{eventEnv} environment in the device driver to hold event
handlers. When the user calls @code{grDevices::getGraphicsEvent}, R will
take three steps.  First, it sets the device driver member
@code{gettingEvent} to @code{true} for each device with a
non-@code{NULL} @code{eventEnv} entry, and calls @code{initEvent(dd,
true)} if the callback is defined.  It then enters an event loop.  Each
time through the loop R will process events once, then check whether any
device has set the @code{result} member of @code{eventEnv} to a
non-@code{NULL} value, and will save the first such value found to be
returned.  C functions @code{doMouseEvent} and @code{doKeybd} are
provided to call the R event handlers @code{onMouseDown},
@code{onMouseMove}, @code{onMouseUp}, and @code{onKeybd} and set
@code{eventEnv$result} during this step.  Finally, @code{initEvent} is
called again with @code{init=false} to inform the devices that the
loop is done, and the result is returned to the user.

@node Specific devices,  , Graphics events, Graphics devices
@subsection Specific devices

Specific devices are mostly documented by comments in their sources,
although for devices of many years' standing those comments can be in
need of updating.  This subsection is a repository of notes on design
decisions.

@menu
* X11()::                       
* windows()::                   
@end menu

@node X11(), windows(), Specific devices, Specific devices
@subsubsection X11()

The @code{X11(type="Xlib")} device dates back to the mid 1990's and was
written then in @code{Xlib}, the most basic X11 toolkit.  It has since
optionally made use of a few features from other toolkits: @code{libXt}
is used to read X11 resources, and @code{libXmu} is used in the handling
of clipboard selections.

Using basic @code{Xlib} code makes drawing fast, but is limiting.  There
is no support of translucent colours (that came in the @code{Xrender}
toolkit of 2000) nor for rotated text (which @R{} implements by
rendering text to a bitmap and rotating the latter).

The hinting for the X11 window asks for backing store to be used, and
some windows managers may use it to handle repaints, but it seems that
most repainting is done by replaying the display list (and here the fast
drawing is very helpful).

There are perennial problems with finding fonts.  Many users fail to
realize that fonts are a function of the X server and not of the machine
that @R{} is running on.  After many difficulties, @R{} tries first to
find the nearest size match in the sizes provided for Adobe fonts in the
standard 75dpi and 100dpi X11 font packages---even that will fail to
work when users of near-100dpi screens have only the 75dpi set
installed.  The 75dpi set allows sizes down to 6 points on a 100dpi
screen, but some users do try to use smaller sizes and even 6 and 8
point bitmapped fonts do not look good.

Introduction of UTF-8 locales has caused another wave of difficulties.
X11 has very few genuine UTF-8 fonts, and produces composite fontsets
for the @code{iso10646-1} encoding.  Unfortunately these seem to have
low coverage apart from a few monospaced fonts in a few sizes (which are
not suitable for graph annotation), and where glyphs are missing what is
plotted is often quite unsatisfactory.

The current approach is to make use of more modern toolkits, namely
@code{cairo} for rendering and @code{Pango} for font
management---because these are associated with @code{Gtk+2} they are
widely available.  Cairo supports translucent colours and alpha-blending
(@emph{via} @code{Xrender}), and anti-aliasing for the display of lines
and text.  Pango's font management is based on @code{fontconfig} and
somewhat mysterious, but it seems mainly to use Type 1 and TrueType
fonts on the machine running @R{} and send grayscale bitmaps to cairo.


@node windows(),  , X11(), Specific devices
@subsubsection windows()

The @code{windows()} device is a family of devices: it supports plotting
to Windows (enhanced) metafiles, @code{BMP}, @code{JPEG}, @code{PNG} and
@code{TIFF} files as well as to Windows printers.

In most of these cases the primary plotting is to a bitmap: this is used
for the (default) buffering of the screen device, which also enables the
current plot to be saved to BMP, JPEG, PNG or TIFF (it is the internal
bitmap which is copied to the file in the appropriate format).

The device units are pixels (logical ones on a metafile device).

The code was originally written by Guido Masarotto with extensive use of
macros, which can make it hard to disentangle.

For a screen device, @code{xd->gawin} is the canvas of the screen, and
@code{xd->bm} is the off-screen bitmap.  So macro @code{DRAW} arranges
to plot to @code{xd->bm}, and if buffering is off, also to
@code{xd->gawin}.  For all other device, @code{xd->gawin} is the canvas,
a bitmap for the @code{jpeg()} and @code{png()} device, and an internal
representation of a Windows metafile for the @code{win.metafile()} and
@code{win.print} device.  Since `plotting' is done by Windows GDI calls
to the appropriate canvas, its precise nature is hidden by the GDI
system.

Buffering on the screen device is achieved by running a timer, which
when it fires copies the internal bitmap to the screen.  This is set to
fire every 500ms (by default) and is reset to 100ms after plotting
activity.

Repaint events are handled by copying the internal bitmap to the screen
canvas (and then reinitializing the timer), unless there has been a resize.
Resizes are handled by replaying the display list: this might not be
necessary if a fixed canvas with scrollbars is being used, but that is
the least popular of the three forms of resizing.

Text on the device has moved to `Unicode' (UCS-2) in recent years.
UTF-8 is requested (@code{hasTextUTF8 = TRUE}) for standard text, and
converted to UCS-2 in the plotting functions in file
@file{src/extra/graphapp/gdraw.c}.  However, GDI has no support for
Unicode symbol fonts, and symbols are handled in Adobe Symbol encoding.

There is support for translucent colours (with alpha channel between 0
and 255) was introduced on the screen device and bitmap
devices.@footnote{It is technically possible to use alpha-blending on
metafile devices such as printers, but it seems few drivers have support
for this.} This is done by drawing on a further internal bitmap,
@code{xd->bm2}, in the opaque version of the colour then alpha-blending
that bitmap to @code{xd->bm}.  The alpha-blending routine is in a
separate DLL, @file{msimg32.dll}, which is loaded on first use.  As
small a rectangular region as reasonably possible is alpha-blended (this
is rectangle @code{r} in the code), but things like mitre joins make
estimation of a tight bounding box too much work for lines and polygonal
boundaries.  Translucent-coloured lines are not common, and the
performance seems acceptable.

The support for a transparent background in @code{png()} predates full
alpha-channel support in @code{libpng} (let alone in PNG viewers), so
makes use of the limited transparency support in earlier versions of
PNG.  Where 24-bit colour is used, this is done by marking a single
colour to be rendered as transparent.  @R{} chose @samp{#fdfefd}, and
uses this as the background colour (in @code{GA_NewPage} if the
specified background colour is transparent (and all non-opaque
background colours are treated as transparent).  So this works by
marking that colour in the PNG file, and viewers without transparency
support see a slightly-off-white background, as if there were a
near-white canvas.  Where a palette is used in the PNG file (if less
than 256 colours were used) then this colour is recorded with full
transparency and the remaining colours as opaque.  If 32-bit colour were
available then we could add a full alpha channel, but this is dependent
on the graphics hardware and undocumented properties of GDI.


@node Colours, Base graphics, Graphics devices, Graphics Devices
@section Colours

Devices receive colours as a @code{typedef} @code{rcolor} (an
@code{unsigned int}) defined in the header
@file{R_ext/GraphicsEngine.h}).  The 4 bytes are @emph{R} ,@emph{G},
@emph{B} and @emph{alpha} from least to most significant. So each of RGB
has 256 levels of luminosity from 0 to 255.  The alpha byte represents
opacity, so value 255 is fully opaque and 0 fully transparent: many but
not all devices handle semi-transparent colours.

Colors can be created in C via the macro @code{R_RGBA}, and a set of
macros are defined in @file{R_ext/GraphicsDevice.h} to extract the
various components.

Colours in the base graphics system were originally adopted from S (and
before that the GRZ library from Bell Labs), with the concept of a
(variable-sized) palette of colours referenced by numbers
@samp{1...@var{N}} plus @samp{0} (the background colour of the current
device).  @R{} introduced the idea of referring to colours by character
strings, either in the forms @samp{#RRGGBB} or @samp{#RRGGBBAA}
(representing the bytes in hex) as given by function @code{rgb()} or via
names: the 657 known names are given in the character vector
@code{colors} and in a table in file @file{colors.c} in package
@pkg{grDevices}.  Note that semi-transparent colours are not
`premultiplied', so 50% transparent white is @samp{#ffffff80}.

Integer or character @code{NA} colours are mapped internally to
transparent white, as is the character string @code{"NA"}.

Negative colour numbers are an error.  Colours greater than
@samp{@var{N}} are wrapped around, so that for example with the default
palette of size 8, colour @samp{10} is colour @samp{2} in the palette.

Integer colours have been used more widely than the base graphics
sub-system, as they are supported by package @pkg{grid} and hence by
@CRANpkg{lattice} and @CRANpkg{ggplot2}.  (They are also used by package
@CRANpkg{rgl}.)  @pkg{grid} did re-define colour @samp{0} to be
transparent white, but @CRANpkg{rgl} used @code{col2rgb} and hence the
background colour of base graphics.

Note that positive integer colours refer to the current palette and
colour @samp{0} to the current device (and a device is opened if needs
be).  These are mapped to type @code{rcolor} at the time of use: this
matters when re-playing the display list, e.g.@: when a device is
resized or @code{dev.copy} is used.  The palette should be thought of as
per-session: it is stored in package @pkg{grDevices}.

The convention is that devices use the colorspace `sRGB'. This is an
industry standard: it is used by Web browsers and JPEGs from all but
high-end digital cameras.  The interpretation is a matter for graphics
devices and for code that manipulates colours, but not for the graphics
engine or subsystems.

@R{} uses a painting model similar to PostScript and PDF.  This means
that where shapes (circles, rectangles and polygons) can both be filled
and have a stroked border, the fill should be painted first and then the
border (or otherwise only half the border will be visible).  Where both
the fill and the border are semi-transparent there is some room for
interpretation of the intention.  Most devices first paint the fill and
then the border, alpha-blending at each step.  However, PDF does some
automatic grouping of objects, and @emph{when the fill and the border
have the same alpha}, they are painted onto the same layer and then
alpha-blended in one step.  (See p. 569 of the PDF Reference Sixth
Edition, version 1.7.  Unfortunately, although this is what the PDF
standard says should happen, it is not correctly implemented by some
viewers.)

The mapping from colour numbers to type @code{rcolor} is primarily done
by function @code{RGBpar3}: this is exported from the @R{} binary but
linked to code in package @pkg{grDevices}.  The first argument is a
@code{SEXP} pointing to a character, integer or double vector, and the
second is the @code{rcolor} value for colour @code{0} (or @code{"0"}).
C entry point @code{RGBpar} is a wrapper that takes @code{0} to be
transparent white: it is often used to set colour defaults for devices.
The @R{}-level wrapper is @code{col2rgb}.

There is also @code{R_GE_str2col} which takes a C string and converts to
type @code{rcolor}: @code{"0'} is converted to transparent white.

There is a @R{}-level conversion of colours to @samp{##RRGGBBAA} by
@code{image.default(useRaster = TRUE)}.

The other color-conversion entry point in the API is @code{name2col}
which takes a colour name (a C string) and returns a value of type
@code{rcolor}.  This handles @code{"NA"}, @code{"transparent"} and the
657 colours known to the @R{} function @code{colors()}.

@node Base graphics, Grid graphics, Colours, Graphics Devices
@section Base graphics

The base graphics system was migrated to package @pkg{graphics} in @R{}
3.0.0: it was previously implemented in files in @file{src/main}.

For historical reasons it is largely implemented in two layers.
Files @file{plot.c}, @file{plot3d.c} and @file{par.c} contain the code
for the around 30 @code{.External} calls that implement the basic
graphics operations.  This code then calls functions with names starting
with @code{G} and declared in header @file{Rgraphics.h} in file
@file{graphics.c}, which in turn call the graphics engine (whose
functions almost all have names starting with @code{GE}).

A large part of the infrastructure of the base graphics subsystem are
the graphics parameters (as set/read by @code{par()}).  These are stored
in a @code{GPar} structure declared in the private header
@file{Graphics.h}.  This structure has two variables (@code{state} and
@code{valid}) tracking the state of the base subsystem on the device,
and many variables recording the graphics parameters and functions of
them.

The base system state is contained in @code{baseSystemState} structure
defined in the private header @file{GraphicsBase.h}.  
This contains three @code{GPar}
structures and a Boolean variable used to record if @code{plot.new()}
(or @code{persp}) has been used successfully on the device.

The three copies of the @code{GPar} structure are used to store the
current parameters (accessed via @code{gpptr}), the `device copy'
(accessed via @code{dpptr}) and space for a saved copy of the `device
copy' parameters.  The current parameters are, clearly, those currently
in use and are copied from the `device copy' whenever @code{plot.new()}
is called (whether or not that advances to the next `page'). The saved
copy keeps the state when the device was last completely cleared (e.g.@:
when @code{plot.new()} was called with @code{par(new=TRUE)}), and is
used to replay the display list.

The separation is not completely clean: the `device copy' is altered if
a plot with log scale(s) is set up via @code{plot.window()}.

There is yet another copy of most of the graphics parameters in
@code{static} variables in @file{graphics.c} which are used to preserve
the current parameters across the processing of inline parameters in
high-level graphics calls (handled by @code{ProcessInlinePars}).

Snapshots of the base subsystem record the `saved device copy' of the
@code{GPar} structure.

@menu
* Arguments and parameters::    
@end menu

@node Arguments and parameters,  , Base graphics, Base graphics
@subsection Arguments and parameters

There is an unfortunate confusion between some of the graphical
parameters (as set by @code{par}) and arguments to base graphic
functions of the same name.  This description may help set the record
straight.

Most of the high-level plotting functions accept graphical parameters as
additional arguments, which are then often passed to lower-level
functions if not already named arguments (which is the main source of
confusion).

Graphical parameter @code{bg} is the background colour of the plot.
Argument @code{bg} refers to the fill colour for the filled symbols
@code{21} to @code{25}.  It is an argument to the function
@code{plot.xy}, but normally passed by the default method of
@code{points}, often from a @code{plot} method.

Graphics parameters @code{cex}, @code{col}, @code{lty}, @code{lwd} and
@code{pch} also appear as arguments of @code{plot.xy} and so are often
passed as arguments from higher-level plot functions such as
@code{lines}, @code{points} and @code{plot} methods.  They appear as
arguments of @code{legend}, @code{col}, @code{lty} and @code{lwd} are
arguments of @code{arrows} and @code{segments}.  When used as arguments
they can be vectors, recycled to control the various lines, points and
segments.  When set a graphical parameters they set the default
rendering: in addition @code{par(cex=)} sets the overall character
expansion which subsequent calls (as arguments or on-line graphical
parameters) multiply.

The handling of missing values differs in the two classes of uses.
Generally these are errors when used in @code{par} but cause the
corresponding element of the plot to be omitted when used as an element
of a vector argument.  Originally the interpretation of arguments was
mainly left to the device, but nowadays some of this is pre-empted in
the graphics engine (but for example the handling of @code{lwd = 0}
remains device-specific, with some interpreting it as a `thinnest
possible' line).

@node Grid graphics,  , Base graphics, Graphics Devices
@section Grid graphics

[At least pointers to documentation.]

@node GUI consoles, Tools, Graphics Devices, Top
@chapter GUI consoles

The standard @R{} front-ends are programs which run in a terminal, but
there are several ways to provide a GUI console.

This can be done by a package which is loaded from terminal-based @R{}
and launches a console as part of its startup code or by the user
running a specific function: package @CRANpkg{Rcmdr} is a well-known
example with a Tk-based GUI.

There used to be a Gtk-based console invoked by @command{R --gui=GNOME}:
this relied on special-casing in the front-end shell script to launch a
different executable.  There still is @command{R --gui=Tk}, which starts
terminal-based @R{} and runs @code{tcltk::tkStartGui()} as part of the
modified startup sequence.

However, the main way to run a GUI console is to launch a separate
program which runs embedded @R{}: this is done by @command{Rgui.exe} on
Windows and @command{R.app} on macOS.  The first is an integral part
of @R{} and the code for the console is currently in @file{R.dll}.

@menu
* R.app::                       
@end menu

@node R.app,  , GUI consoles, GUI consoles
@section R.app

@command{R.app} is a macOS application which provides a console.  Its
sources are a separate project@footnote{an Xcode project, in SVN at
@uref{https://svn.r-project.org/R-packages/trunk/Mac-GUI/}.}, and its binaries
link to an @R{} installation which it runs as a dynamic library
@file{libR.dylib}.  The standard @acronym{CRAN} distribution of @R{} for
macOS bundles the GUI and @R{} itself, but installing the GUI is optional
and either component can be updated separately.

@command{R.app} relies on @file{libR.dylib} being in a specific place,
and hence on @R{} having been built and installed as a Mac macOS
`framework'.  Specifically, it uses
@file{/Library/Frameworks/R.framework/R}.  This is a symbolic link, as
frameworks can contain multiple versions of @R{}.  It eventually
resolves to
@file{/Library/Frameworks/R.framework/Versions/Current/Resources/lib/libR.dylib},
which is (in the @acronym{CRAN} distribution) a `fat' binary containing
multiple sub-architectures.

macOS applications are directory trees: each @command{R.app} contains
a front-end written in Objective-C for one sub-architecture: in the
standard distribution there are separate applications for 32- and 64-bit
Intel architectures.

Originally the @R{} sources contained quite a lot of code used only by
the macOS GUI, but this was migrated to the @command{R.app} sources.

@command{R.app} starts @R{} as an embedded application with a
command-line which includes @option{--gui=aqua} (see below).  It uses
most of the interface pointers defined in the header
@file{Rinterface.h}, plus a private interface pointer in file
@file{src/main/sysutils.c}.  It adds an environment
it names @code{tools:RGUI} to the second position in the search path.
This contains a number of utility functions used to support the menu
items, for example @code{package.manager()}, plus functions @code{q()}
and @code{quit()} which mask those in package @pkg{base}---the custom
versions save the history in a way specific to @code{R.app}.

There is a @command{configure} option @option{--with-aqua} for @R{}
which customizes the way @R{} is built: this is distinct from the
@option{--enable-R-framework} option which causes @command{make install}
to install @R{} as the framework needed for use with @code{R.app}.  (The
option @option{--with-aqua} is the default on macOS.)  It sets the
macro @code{HAVE_AQUA} in @file{config.h} and the make variable
@code{BUILD_AQUA_TRUE}.  These have several consequences:

@itemize
@item
The @code{quartz()} device is built (other than as a stub) in package
@pkg{grDevices}: this needs an Objective-C compiler.  Then
@code{quartz()} can be used with terminal @R{} provided the latter has
access to the macOS screen.

@item
File @file{src/unix/aqua.c} is compiled.  This now only contains an
interface pointer for the @code{quartz()} device(s).

@item
@code{capabilities("aqua")} is set to @code{TRUE}.

@item
The default path for a personal library directory is set as
@file{~/Library/R/arch/x.y/library}.
@c This is done in @file{etc/Renviron}.

@item
There is support for setting a `busy' indicator whilst waiting for
@code{system()} to return.

@item
@code{R_ProcessEvents} is inhibited in a forked child from package
@pkg{parallel}.  The associated callback in @code{R.app} does things
which should not be done in a child, and forking forks the whole process
including the console.

@item
There is support for starting the embedded @R{} with the option
@option{--gui=aqua}: when this is done the global C variable
@code{useaqua} is set to a true value.  This has consequences:

@itemize
@item
The @R{} session is asserted to be interactive @emph{via} @code{R_Interactive}.

@item
@code{.Platform$GUI} is set to @code{"AQUA"}.  That has consequences:
@itemize
@item
The environment variable @env{DISPLAY} is set to @samp{:0} if not
already set.

@item
@file{/usr/local/bin} is appended to @env{PATH} since that is where
@command{gfortran} is installed.

@item
The default @HTML{} browser is switched to the one in @command{R.app}.

@item
Various widgets are switched to the versions provided in
@command{R.app}: these include graphical menus, the data editor (but not
the data viewer used by @code{View()}) and the workspace browser invoked
by @code{browseEnv()}.

@item
The @pkg{grDevices} package when loaded knows that it is being run
under @command{R.app} and so informs any @code{quartz} devices that a
Quartz event loop is already running.
@end itemize

@item
The use of the OS's @code{system} function (including by @code{system()}
and @code{system2()}, and to launch editors and pagers) is replaced by a
version in @code{R.app} (which by default just calls the OS's
@code{system} with various signal handlers reset).

@end itemize

@item
If either @R{} was started by @option{--gui=aqua} or @R{} is running in
a terminal which is not of type @samp{dumb}, the standard output to
files @file{stdout} and @file{stderr} is directed through the C function
@code{Rstd_WriteConsoleEx}.  This uses ANSI terminal escapes to render
lines sent to @code{stderr} as bold on @code{stdout}.

@item
For historical reasons the startup option @code{-psn} is allowed but
ignored.  (It seems that in 2003, @samp{r27492}, this was added by Finder.)

@end itemize



@node Tools, R coding standards, GUI consoles, Top
@chapter Tools

The behavior of @command{R CMD check} can be controlled through a
variety of command line arguments and environment variables.

There is an internal @option{--install=@var{value}} command line
argument not shown by @command{R CMD check --help}, with possible values

@table @code
@item check:@var{file}
Assume that installation was already performed with stdout/stderr to
@var{file}, the contents of which need to be checked (without repeating
the installation).  This is useful for checks applied by repository
maintainers: it reduces the check time by the installation time given
that the package has already been installed.  In this case, one also
needs to specify @emph{where} the package was installed to using command
line option @option{--library}.
@item fake
Fake installation, and turn off the run-time tests.
@item skip
Skip installation, e.g., when testing recommended packages bundled with
R.
@item no
The same as @option{--no-install} : turns off installation and the tests
which require the package to be installed.
@end table

The following environment variables can be used to customize the
operation of @command{check}: a convenient place to set these is the
check environment file (default, @file{~/.R/check.Renviron}).

@vtable @code
@item _R_CHECK_ALL_NON_ISO_C_
If true, do not ignore compiler (typically GCC) warnings about non ISO C
code in @emph{system} headers.  Note that this may also show additional
ISO C++ warnings.
Default: false.
@item _R_CHECK_FORCE_SUGGESTS_
If true, give an error if suggested packages are not available.
Default: true (but false for CRAN submission checks).
@item _R_CHECK_RD_CONTENTS_
If true, check @file{Rd} files for auto-generated content which needs
editing, and missing argument documentation.
Default: true.
@item _R_CHECK_RD_LINE_WIDTHS_
If true, check @file{Rd} line widths in usage and examples sections.
Default: false (but true for CRAN submission checks).
@item _R_CHECK_RD_STYLE_
If true, check whether @file{Rd} usage entries for S3 methods use the full
function name rather than the appropriate @code{\method} markup.
Default: true.
@item _R_CHECK_RD_XREFS_
If true, check the cross-references in @file{.Rd} files.
Default: true.
@item _R_CHECK_SUBDIRS_NOCASE_
If true, check the case of directories such as @file{R} and @file{man}.
Default: true.
@item _R_CHECK_SUBDIRS_STRICT_
Initial setting for @option{--check-subdirs}.
Default: @samp{default} (which checks only tarballs, and checks in the
@file{src} only if there is no @file{configure} file).
@item _R_CHECK_USE_CODETOOLS_
If true, make use of the @CRANpkg{codetools} package, which provides a
detailed analysis of visibility of objects (but may give false
positives).
Default: true (if recommended packages are installed).
@item _R_CHECK_USE_INSTALL_LOG_
If true, record the output from installing a package as part of its
check to a log file (@file{00install.out} by default), even when running
interactively.
Default: true.
@item _R_CHECK_VIGNETTES_NLINES_
Maximum number of lines to show from the bottom of the output when
reporting errors in running or re-building vignettes. ( Value @code{0}
means all lines will be shown.)
Default: 10 for running, 25 for re-building.
@item _R_CHECK_CODOC_S4_METHODS_
Control whether @code{codoc()} testing is also performed on S4 methods.
Default: true.
@item _R_CHECK_DOT_INTERNAL_
Control whether the package code is scanned for @code{.Internal} calls,
which should only be used by base (and occasionally by recommended) packages.
Default: true.
@item _R_CHECK_EXECUTABLES_
Control checking for executable (binary) files.
Default: true.
@item _R_CHECK_EXECUTABLES_EXCLUSIONS_
Control whether checking for executable (binary) files ignores files
listed in the package's @file{BinaryFiles} file.
Default: true (but false for CRAN submission checks).
However, most likely this package-level override mechanism will be
removed eventually.
@item _R_CHECK_PERMISSIONS_
Control whether permissions of files should be checked.
Default: true iff @code{.Platform$OS.type == "unix"}.
@item _R_CHECK_FF_CALLS_
Allows turning off @code{checkFF()} testing. If set to
@samp{registration}, checks the registration information (number of
arguments, correct choice of @code{.C/.Fortran/.Call/.External}) for
such calls provided the package is installed.
Default: true.
@item _R_CHECK_FF_DUP_
Controls @code{checkFF(check_DUP)}
Default: true (and forced to be true for CRAN submission checks).
@item _R_CHECK_LICENSE_
Control whether/how license checks are performed. A possible value is
@samp{maybe} (warn in case of problems, but not about standardizable
non-standard license specs).
Default: true.
@item _R_CHECK_RD_EXAMPLES_T_AND_F_
Control whether @code{check_T_and_F()} also looks for ``bad'' (global)
@samp{T}/@samp{F} uses in examples.
Off by default because this can result in false positives.
@item _R_CHECK_RD_CHECKRD_MINLEVEL_
Controls the minimum level for reporting warnings from @code{checkRd}.
Default: -1.
@item _R_CHECK_XREFS_REPOSITORIES_
If set to a non-empty value, a space-separated list of repositories to
use to determine known packages.  Default: empty, when the CRAN
and Bioconductor repositories known to @R{} is used.
@item _R_CHECK_SRC_MINUS_W_IMPLICIT_
Control whether installation output is checked for compilation warnings
about implicit function declarations (as spotted by GCC with command
line option @option{-Wimplicit-function-declaration}, which is implied
by @option{-Wall}).  NB: implicit function declarations are errors in
some recent C compilers, including Apple @command{clang}.
Default: true from @R{} 4.2.0, previously false.
@item _R_CHECK_SRC_MINUS_W_UNUSED_
Control whether installation output is checked for compilation warnings
about unused code constituents (as spotted by GCC with command line
option @option{-Wunused}, which is implied by @option{-Wall}).
Default: true.
@item _R_CHECK_WALL_FORTRAN_
Control whether gfortran 4.0 or later @option{-Wall} warnings are used in
the analysis of installation output.
Default: false, even though the warnings are justifiable.
@item _R_CHECK_ASCII_CODE_
If true, check @R{} code for non-ascii characters.
Default: true.
@item _R_CHECK_ASCII_DATA_
If true, check data for non-ascii characters.  @emph{En route}, checks
that all the datasets can be loaded and that their components can be
accessed.
Default: true.
@item _R_CHECK_COMPACT_DATA_
If true, check data for ascii and uncompressed saves, and also check if
using @command{bzip2} or @code{xz} compression would be significantly
better.
Default: true.
@item _R_CHECK_SKIP_ARCH_
Comma-separated list of architectures that will be omitted from
checking in a multi-arch setup.
Default: none.
@item _R_CHECK_SKIP_TESTS_ARCH_
Comma-separated list of architectures that will be omitted from
running tests in a multi-arch setup.
Default: none.
@item _R_CHECK_SKIP_EXAMPLES_ARCH_
Comma-separated list of architectures that will be omitted from
running examples in a multi-arch setup.
Default: none.
@item _R_CHECK_VC_DIRS_
Should the unpacked package directory be checked for version-control
directories (@file{CVS}, @file{.svn} @dots{})?
Default: true for tarballs.
@item _R_CHECK_PKG_SIZES_
Should @command{du} be used to find the installed sizes of packages?
@command{R CMD check} does check for the availability of @command{du}.
but this option allows the check to be overruled if an unsuitable
command is found (including one that does not respect the @option{-k}
flag to report in units of 1Kb, or reports in a different format -- the
GNU, macOS and Solaris @command{du} commands have been tested).
Default: true if @command{du} is found.
@item _R_CHECK_PKG_SIZES_THRESHOLD_
Threshold used for @env{_R_CHECK_PKG_SIZES_} (in Mb).
Default: 5
@item _R_CHECK_DOC_SIZES_
Should @command{qpdf} be used to check the installed sizes of PDFs?
Default: true if @command{qpdf} is found.
@item _R_CHECK_DOC_SIZES2_
Should @command{gs} be used to check the installed sizes of PDFs?  This
is slower than (and in addition to) the previous check, but does detect
figures with excessive detail (often hidden by over-plotting) or bitmap
figures with too high a resolution.  Requires that @env{R_GSCMD} is set
to a valid program, or @command{gs} (or on Windows,
@command{gswin32.exe} or @command{gswin64c.exe}) is on the path.
Default: false (but true for CRAN submission checks).
@item _R_CHECK_ALWAYS_LOG_VIGNETTE_OUTPUT_
By default the output from running the @R{} code in the vignettes is
kept only if there is an error.  This also applies to the
@file{build_vignettes.log} log from the re-building of vignettes.
Default: false.
@item _R_CHECK_CLEAN_VIGN_TEST_
Should the @file{vign_test} directory be removed if the test is
successful?
Default: true.
@item _R_CHECK_REPLACING_IMPORTS_
Should warnings about replacing imports be reported?  These sometimes come
from auto-generated @file{NAMESPACE} files in other packages, but most
often from importing the whole of a namespace rather than using
@code{importFrom}.
Default: true.
@item _R_CHECK_UNSAFE_CALLS_
Check for calls that appear to tamper with (or allow tampering with)
already loaded code not from the current package: such calls may well
contravene CRAN policies.
Default: true.
@item _R_CHECK_TIMINGS_
Optionally report timings for installation, examples, tests and
running/re-building vignettes as part of the check log.  The format is
@samp{[as/bs]} for the total CPU time (including child processes)
@samp{a} and elapsed time @samp{b}, except on Windows, when it is
@samp{[bs]}.  In most cases timings are only given for @samp{OK} checks.
Times with an elapsed component over 10 mins are reported in minutes
(with abbreviation @samp{m}).  The value is the smallest numerical value
in elapsed seconds that should be reported: non-numerical values
indicate that no report is required, a value of @samp{0} that a report
is always required.
Default: @code{""}. (@code{10} for CRAN checks.)

@item _R_CHECK_EXAMPLE_TIMING_THRESHOLD_
If timings are being recorded, set the threshold in seconds for
reporting long-running examples (either user+system CPU time or elapsed
time).  Default: @code{"5"}.

@item _R_CHECK_EXAMPLE_TIMING_CPU_TO_ELAPSED_THRESHOLD_
For checks with timings enabled, report examples where the ratio of CPU
time to elapsed time exceeds this threshold (and the CPU time is at
least one second).  This can help detect the simultaneous use of
multiple CPU cores.
Default: @code{NA}.

@item _R_CHECK_TEST_TIMING_CPU_TO_ELAPSED_THRESHOLD_
Report for running an individual test if the ratio of CPU time to
elapsed time exceeds this threshold (and the CPU time is at least one
second).  Not supported on Windows.
Default: @code{NA}.

@item _R_CHECK_VIGNETTE_TIMING_CPU_TO_ELAPSED_THRESHOLD_
Report if when running/re-building vignettes (individually or in
aggregate) the ratio of CPU time to elapsed time exceeds this threshold
(and the CPU time is at least one second).  Not supported on
Windows.
Default: @code{NA}.

@item _R_CHECK_CODETOOLS_PROFILE_
A string with comma-separated @code{@var{name}=@var{value}} pairs (with
@var{value} a logical constant) giving additional arguments for the
@CRANpkg{codetools} functions used for analyzing package code.  E.g.,
use @code{_R_CHECK_CODETOOLS_PROFILE_="suppressLocalUnused=FALSE"} to
turn off suppressing warnings about unused local variables.  Default: no
additional arguments, corresponding to using @code{skipWith = TRUE},
@code{suppressPartialMatchArgs = FALSE} and @code{suppressLocalUnused =
TRUE}.

@item _R_CHECK_CRAN_INCOMING_
Check whether package is suitable for publication on CRAN.
Default: false, except for CRAN submission checks.

@item _R_CHECK_CRAN_INCOMING_REMOTE_
Include checks that require remote access among the above.
Default: same as @code{_R_CHECK_CRAN_INCOMING_}

@item _R_CHECK_XREFS_USE_ALIASES_FROM_CRAN_
When checking anchored Rd xrefs, use Rd aliases from the CRAN package
web areas in addition to those in the packages installed locally.
Default: false.

@item _R_SHLIB_BUILD_OBJECTS_SYMBOL_TABLES_
Make the checks of compiled code more accurate by recording the symbol
tables for objects (@file{.o} files) at installation in a file
@file{symbols.rds}.  (Only currently supported on Linux, Solaris, macOS,
Windows and FreeBSD.)
Default: true.

@item _R_CHECK_CODE_ASSIGN_TO_GLOBALENV_
Should the package code be checked for assignments to the global
environment?
Default: false (but true for CRAN submission checks).

@item _R_CHECK_CODE_ATTACH_
Should the package code be checked for calls to @code{attach()}?
Default: false (but true for CRAN submission checks).

@item _R_CHECK_CODE_DATA_INTO_GLOBALENV_
Should the package code be checked for calls to @code{data()} which load
into the global environment?
Default: false (but true for CRAN submission checks).

@item _R_CHECK_DOT_FIRSTLIB_
Should the package code be checked for the presence of the obsolete function
@code{.First.lib()}?
Default: false (but true for CRAN submission checks).

@item _R_CHECK_DEPRECATED_DEFUNCT_
Should the package code be checked for the presence of recently deprecated
or defunct functions (including completely removed functions).  Also for
platform-specific graphics devices.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_SCREEN_DEVICE_
If set to @samp{warn}, give a warning if examples etc open a screen
device.  If set to @samp{stop}, give an error.
Default: empty (but @samp{stop} for CRAN submission checks).

@item _R_CHECK_WINDOWS_DEVICE_
If set to @samp{stop}, give an error if a Windows-only device is used in
example etc.  This is only useful on Windows: the devices do not exist
elsewhere.
Default: empty (but @samp{stop} for CRAN submission checks on Windows).

@item _R_CHECK_TOPLEVEL_FILES_
Report on top-level files in the package sources that are not described
in `Writing R Extensions' nor are commonly understood (like
@file{ChangeLog}).  Variations on standard names (e.g.@:
@file{COPYRIGHT}) are also reported.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_GCT_N_
Should the @option{--use-gct} use @code{gctorture2(@var{n})} rather than
@code{gctorture(TRUE)}?  Use a positive integer to enable this.
Default: @code{0}.

@item _R_CHECK_LIMIT_CORES_
If set, check the usage of too many cores in package @pkg{parallel}.  If
set to @samp{warn} gives a warning, to @samp{false} or @samp{FALSE} the
check is skipped, and any other non-empty value gives an error when more
than 2 children are spawned.
Default: unset (but @samp{TRUE} for CRAN submission checks).

@item _R_CHECK_CODE_USAGE_VIA_NAMESPACES_
If set, check code usage (via @CRANpkg{codetools}) directly on the
package namespace without loading and attaching the package and its
suggests and enhances.
Default: true (and true for CRAN submission checks).

@item _R_CHECK_CODE_USAGE_WITH_ONLY_BASE_ATTACHED_
If set, check code usage (via @CRANpkg{codetools}) with only the base
package attached.
Default: true.

@item _R_CHECK_EXIT_ON_FIRST_ERROR_
If set to a true value, the check will exit on the first error.
Default: false.

@item _R_CHECK_S3_METHODS_NOT_REGISTERED_
If set to a true value, report (apparent) S3 methods exported but not
registered.
Default: true.

@item _R_CHECK_OVERWRITE_REGISTERED_S3_METHODS_
If set to a true value, report already registered S3 methods in
base/recommended packages which are overwritten when this package's
namespace is loaded. 
Default: false (but true for CRAN submission checks).

@item _R_CHECK_TESTS_NLINES_
Number of trailing lines of test output to reproduce in the log.  If
@code{0} all lines except the @R{} preamble are reproduced.
Default: 13.

@item _R_CHECK_NATIVE_ROUTINE_REGISTRATION_
If set to a true value, report if the entry points to register native
routines and to suppress dynamic search are not found in a package's
DLL.  (@strong{NB:} this requires system command @command{nm} to be on the
@env{PATH}. On Windows, @command{objdump.exe} is first searched for in
compiler toolchain specified via @code{Makeconf} (can be customized by
environment variable @env{BINPREF}). If not found there, it must be on the
@env{PATH}. On Unix this would be normal when using a package with compiled
code (which are the only ones this checks), but Windows' users should check.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_NO_STOP_ON_TEST_ERROR_
If set to a true value, do not stop running tests after first error (as
if command line option @option{--no-stop-on-test-error} had been given).
Default: false (but true for CRAN submission checks).

@item _R_CHECK_PRAGMAS_
Run additional checks on the pragmas in C/C++ source code and headers.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_COMPILATION_FLAGS_
If the package is installed and has C/C++/Fortran code, check the
install log for non-portable flags (for example those added to
@file{src/Makevars} during configuration).  Currently @option{-W} flags
are reported, except @option{-Wall}, @option{-Wextra} and
@option{-Weverything}, and flags which appear to be attempts to suppress
warnings are highlighted.
See
@ifset UseExternalXrefs
@ref{Writing portable packages, , Writing portable packages, R-exts, Writing R Extensions}
@end ifset
@ifclear UseExternalXrefs
`Writing R Extensions'
@end ifclear
for the rationale of this check (and why even @option{-Werror} is
unsafe).

Environment variable @env{_R_CHECK_COMPILATION_FLAGS_KNOWN_} can be set
to a space-separated set of flags which come from the @R{} build used
for testing (flags such as @option{-Wall} and @option{-Wextra} are
already known).  For example, for CRAN build of @R{} >= 4.0.0 on macOS
one could use
@example
_R_CHECK_COMPILATION_FLAGS_KNOWN_="-mmacosx-version-min=10.13" 
@end example
@noindent
Default: false (but true for CRAN submission checks).

@item _R_CHECK_R_DEPENDS_
Check that any dependence on R is not on a recent patch-level version
such as @code{R (>= 3.3.3)} since blocking installation of a package
will also block its reverse dependencies.  Possible values
@samp{"note"}, @samp{"warn"} and logical values (where currently true
values are equivalent to @samp{"note"}).
Default: false (but @samp{"warn"} for @option{--as-cran}).

@item _R_CHECK_SERIALIZATION_
Check that serialized @R{} objects in the package sources were
serialized with version 2 and there is no dependence on @samp{R >=
3.5.0}.  (Version 3 is in use as from @R{} 3.5.0 but should only be used
when necessary.)
Default: false (but true for CRAN submission checks).

@item _R_CHECK_R_ON_PATH_
This checks if the package attempts to use @command{R} or
@command{Rscript} from the path rather than that under test.
It does so by putting scripts at the head of the path which print a
message and fail.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_PACKAGES_USED_IN_TESTS_USE_SUBDIRS_
If set to a true value, also check the R code in common unit test
subdirectories of @file{tests} for undeclared package dependencies.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_SHLIB_OPENMP_FLAGS_
Check correct and portable use of @code{SHLIB_OPENMP_*FLAGS} in
@file{src/Makevars} (and similar).
Default: false (but true for CRAN submission checks).

@item _R_CHECK_CONNECTIONS_LEFT_OPEN_
When checking examples, check for each example if connections are left
open: if any are found, this is reported with a fatal error.  NB:
`connections' includes most use of files and any parallel clusters which
have not be stopped by @code{stopCluster()}.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_FUTURE_FILE_TIMESTAMPS_
Check if any of the input files has a timestamp in the future (and to do
so, checks that the system clock is correct to within 5 minutes).
Default: false (but true for CRAN submission checks).
@c _R_CHECK_SYSTEM_CLOCK_ can be used to disable the clock check, for
@c use on a check farm.

@item _R_CHECK_LENGTH_1_CONDITION_
No longer in use: conditions of length greater than one in @code{if} or
@code{while} statements are now an error.

@item _R_CHECK_LENGTH_1_LOGIC2_
Optionally check if an argument of the binary operators @code{&&} and
@code{||} has length greater than one, the right-hand side being checked
only if it is used.  For a false value (@samp{F}, @samp{False},
@samp{FALSE} or @samp{false}) or when unset, print a warning.  Any other
non-true non-empty value needs to be a list of commands separated by
commas: @samp{abort} causes R to terminate unconditionally instead of
signalling an error (which is useful in pinpointing issues in code
called from @code{try} or @code{tryCatch}), @samp{verbose} prints a very
detailed diagnostic message, @samp{package:pkg} restricts the check to
if/while statements executing in the namespace of package @samp{pkg}
(but @code{all_base} refers to all the standard packages),
@samp{package:_R_CHECK_PACKAGE_NAME_} restricts the check to if/while
statements executing in the package that is currently being checked by
@code{R CMD check}, @samp{warn} causes R to report a warning instead of
signalling an error.  (More than one package specification can be given:
a report, error or warning will be given if any are satisfied.)
Default: unset (a warning is reported), but
@samp{package:_R_CHECK_PACKAGE_NAME_,abort,verbose} for the CRAN
submission checks.

@item _R_CHECK_VIGNETTES_SKIP_RUN_MAYBE_
Should running the @R{} code in the vignettes be skipped if vignette
outputs are to be rebuilt (which will involve running that code).
Default: false (but true for CRAN checking)

@item _R_CHECK_BUILD_VIGNETTES_SEPARATELY_
Prior to @R{} 3.6.0, re-building the vignette outputs was done in a
single @R{} session which allowed accidental reliance of one vignette on
another (for example, in the loading of packages).  The current default
is to use a separate session for each vignette; this option allows
testing the older behaviour,
Default: true

@item _R_CHECK_SYSTEM_CLOCK_
As part of the `checking for future file timestamps' enabled by
@option{--as-cran}, check the system clock against an external clock to
catch errors such as the wrong day or even year.  Not necessary on
systems doing repeated checks.
Default: true (but false for CRAN checking)

@item _R_CHECK_AUTOCONF_
For packages with a @file{configure} file generated by GNU
@command{autoconf} and either @file{configure.ac} or
@file{configure,.in}, check that @command{autoreconf} can, if available,
be run in a copy of the sources (this will detect missing source files
and report @command{autoconf} warnings).  Environment variable
@env{AUTORECONf} controls the command used: it can give the full path to
@command{autoreconf} (without spaces) and can include flags such as
@option{--warnings=obsolete} (which is added for @command{autoreconf}
version 2.68 or 2.69 and is the default for later versions).
Default: false (but true for CRAN submission checks).

@item _R_CHECK_DATALIST_
Check whether file @file{data/datalist} is out-of-date.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_THINGS_IN_CHECK_DIR_
Check and report at the end of the check run if files have been left in
the check directory.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_THINGS_IN_TEMP_DIR_
Check and report at the end of tthe check run if files would have been
left in the temporary directory (usually @file{/tmp} on a Unix-alike).
It does this by setting the environment variable @env{TEMPDIR} to a
subdirectory of the @R{} session directory for the @code{check} process:
if any files or directories are left there they are removed.  Since some
of these might be out of the user's control, environment variable
@env{_R_CHECK_THINGS_IN_TEMP_DIR_EXCLUDE_} can specify an (extended
regex) pattern of file paths not to be reported -- CRAN uses
@samp{^ompi.} for directories left behind by OpenMPI.  There are rare
instances where @env{TEMPDIR} is not respected and so files are left in
@file{/tmp} (and not reported, but see the next item): one example is
@file{/tmp/boost_interprocess} on some OSes.
@c macOS is one.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_THINGS_IN_OTHER_DIRS_
Check and report at the end of the check run if new files or directories
are created in a selected set of directories during the check run.
(This is confined to files owned by the user running the check process.)
Currently the directories monitored are the home directory, @file{/tmp}
(excluding @file{RtmpXXXXXX} dirs), @file{/dav/shm}, @file{~/.cache}
(recursively) and @file{~/.local/share} (recursively) or their
equivalents on Windows and macOS (the directories in which the default
settings for @code{tools::R_user_dir()} use a @file{R} subdirectory).
Additional directories can be specified in environment variable
@env{_R_CHECK_THINGS_IN_OTHER_DIRS_XTRA_}, separated by semicolons.
Directories are reported with a trailing @samp{/} on all platforms.

Environment variable @env{_R_CHECK_THINGS_IN_OTHER_DIRS_EXCLUDE_} can
specify an (extended regex) pattern of file paths not to be reported --
this should match absolute file paths with home represented by
@file{~}.  For example, on a Linux system
@example
'^~/.cache/(mozilla/firefox|mesa_shader_cache)/'
@end example
@noindent
matches cache directories used by Firefox and OpenGL (and their
content).  If the value starts with @samp{@@} it is considered as a
filepath which is read with each line treated as a pattern to be
matched.

Note that other processes (include check runs in parallel) may create
new files in these directories which will get reported.  However, this
optional check is very useful for narrowing down possible packages which
are leaving behind unexpected files.
@c
Default: false

@item _R_CHECK_BASHISMS_
Check the top-level scripts @file{configure} (unless generated by
@file{autoconf}) and @file{cleanup} for non-Bourne-shell code, using the
Perl script @command{checkbashisms} if available.  This includes
reporting scripts using the non-portable @code{#! /bin/bash}.
(Script @command{checkbashisms} is available in most Linux distributions
in a package named either @samp{devscripts} or @samp{devscripts-checkbashisms}
and from @uref{https://sourceforge.net/projects/checkbaskisms/files}.)
Default: false (but true for CRAN submission checks except on Windows).

@item _R_CHECK_ORPHANED_
Check if dependencies are orphaned packages.  As from @R{} 4.1.0 this
checks strict dependencies recursively, so will report any orphaned
packages which are needed to attach the package by @code{library()} as
well as any orphaned packages which are suggested.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_EXCESSIVE_IMPORTS_
A positive integer.  If set, give a NOTE if the number of imports from
non-base packages exceed this threshold.  Large numbers of imports
make a package vulnerable to any of them becoming unavailable.
Default: unset (but 20 for CRAN submission checks)

@item _R_CHECK_DONTTEST_EXAMPLES_
If true and examples are found with @code{\donttest} sections, the
tests are run in one pass with these commented out and then in a
second pass including the @code{\donttest} sections, (for the main
architecture only).  Only for the first pass are the results compared
to any @file{.Rout.save} file and timings analysed.  Overridden by
@option{--run-donttest}.
Default: false unless @option{-as-cran} is specified (which can be
overridden by setting @samp{_R_CHECK_DONTTEST_EXAMPLES_=false}).

@item _R_CHECK_XREFS_PKGS_ARE_DECLARED_
Check if packages used in `anchored' cross-references in @file{.Rd}
files (those of the form @code{\link[@var{pkg}]@{@var{foo}@}} and
@code{\link[@var{pkg:bar}]@{@var{foo}@}}) are declared in the
@file{DESCRIPTION} file and so these links can be checked.
Default: false.

@item _R_CHECK_XREFS_MIND_SUSPECT_ANCHORS_
Check if package-anchored Rd cross-references are to @emph{files} (and
not aliases).
Default: false.

@item _R_CHECK_BOGUS_RETURN_
If true and @env{_R_CHECK_USE_CODETOOLS_} is also true, functions are
scanned for use of @code{return} rather than @code{return()}.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_MATRIX_DATA_
By default, the check for a mismatch between the data length and the
dimensions in a call to @code{matrix} gives a warning: setting this to a
true value gives an error with a compact traceback.
@c
Default: false (but true for CRAN submission checks).

@item _R_CHECK_CODE_CLASS_IS_STRING_
Check if package code has @code{if()} condition which compare the class
of an object to a string.
See
@uref{https://developer.r-project.org/Blog/public/2019/11/09/when-you-think-class.-think-again/index.html}
why this is a bad idea.
Default: false (but true for CRAN submission checks).

@item _R_CHECK_RD_VALIDATE_RD2HTML_
Check the validity of the package HTML help pages using HTML tidy
(@uref{https://www.html-tidy.org/}) (if available on the system path for
executables). 
Default: false (but true for CRAN submission checks).
@end vtable

The following variables control checks for undeclared/unconditional use
of other packages.  They work by setting up a temporary library
directory and setting @code{.libPaths()} to just that and
@code{.Library}, so are only effective if additional packages are
installed somewhere other than @code{.Library}.  The temporary library
is populated by symbolic links@footnote{under Windows, junction points,
or copies if environment variable @env{R_WIN_NO_JUNCTIONS} has a
non-empty value.} to installed packages not also in @code{.library}.

@vtable @code
@item _R_CHECK_INSTALL_DEPENDS_
If set to a true value and a test installation is to be done, this is
done with a temporary library populated by all the
Depends/Imports/LinkingTo packages.
Default: false (but true for CRAN submission checks).

Note that this is actually implemented in @command{R CMD INSTALL}, so it
is available to those who first install recording to a log, then call
@command{R CMD check}.

@item _R_CHECK_SUGGESTS_ONLY_
If set to a true value, running examples, tests and vignettes is done
with a temporary library directory populated by all the
Depends/Imports/Suggests packages.  (As exceptions, packages in a
@samp{VignetteBuilder} field are always made available.)
@c
Default: false (but true for CRAN submission checks: some of the regular
checks use true and some use false).

@item _R_CHECK_DEPENDS_ONLY_
As for @env{_R_CHECK_SUGGESTS_ONLY_} but using only Depends/Imports (and
the exceptions, including test-suite managers in @samp{Suggests}).
Default: false

@item _R_CHECK_DEPENDS_ONLY_DATA_
Apply @env{_R_CHECK_DEPENDS_ONLY_} only to the check of loading from
the @file{data} directory, so checks if any dataset depends on
packages which are in Suggests or undeclared.
Default: false (but true for CRAN submission checks)

@item _R_CHECK_DEPENDS_ONLY_EXAMPLES_
@itemx _R_CHECK_DEPENDS_ONLY_TESTS_
@itemx _R_CHECK_DEPENDS_ONLY_VIGNETTES_
Apply @env{_R_CHECK_DEPENDS_ONLY_} only to the checking of
examples, tests or vignettes.  These can be used on their own, or
with a false value to override @env{_R_CHECK_DEPENDS_ONLY_}.
@c
Default: the value of @env{_R_CHECK_DEPENDS_ONLY_} or false if that is unset..

@item _R_CHECK_NO_RECOMMENDED_
If set to a true value, augment the previous checks to make recommended
packages unavailable unless declared (even if installed in @code{.library}).
Default: false (but true for CRAN submission checks).

This may give false positives on code which uses
@code{grDevices::densCols} and @code{stats:::.asSparse} /
@code{stats:::.Diag} as these invoke @CRANpkg{KernSmooth} and
@CRANpkg{Matrix} respectively. (Those in @pkg{stats} are called from
various contrasts functions if @code{sparse = TRUE} is used.)
@end vtable

CRAN's submission checks use something like

@example
_R_CHECK_CRAN_INCOMING_=TRUE
_R_CHECK_CRAN_INCOMING_REMOTE_=TRUE
_R_CHECK_VC_DIRS_=TRUE
_R_CHECK_TIMINGS_=10
_R_CHECK_INSTALL_DEPENDS_=TRUE
_R_CHECK_SUGGESTS_ONLY_=TRUE
_R_CHECK_NO_RECOMMENDED_=TRUE
_R_CHECK_EXECUTABLES_EXCLUSIONS_=FALSE
_R_CHECK_DOC_SIZES2_=TRUE
_R_CHECK_CODE_ASSIGN_TO_GLOBALENV_=TRUE
_R_CHECK_CODE_ATTACH_=TRUE
_R_CHECK_CODE_DATA_INTO_GLOBALENV_=TRUE
_R_CHECK_CODE_USAGE_VIA_NAMESPACES_=TRUE
_R_CHECK_DOT_FIRSTLIB_=TRUE
_R_CHECK_DEPRECATED_DEFUNCT_=TRUE
_R_CHECK_REPLACING_IMPORTS_=TRUE
_R_CHECK_SCREEN_DEVICE_=stop
_R_CHECK_TOPLEVEL_FILES_=TRUE
_R_CHECK_S3_METHODS_NOT_REGISTERED_=TRUE
_R_CHECK_OVERWRITE_REGISTERED_S3_METHODS_=TRUE
_R_CHECK_PRAGMAS_=TRUE
_R_CHECK_COMPILATION_FLAGS_=TRUE
_R_CHECK_R_DEPENDS_=warn
_R_CHECK_SERIALIZATION_=TRUE
_R_CHECK_R_ON_PATH_=TRUE
_R_CHECK_PACKAGES_USED_IN_TESTS_USE_SUBDIRS_=TRUE
_R_CHECK_SHLIB_OPENMP_FLAGS_=TRUE
_R_CHECK_CONNECTIONS_LEFT_OPEN_=TRUE
_R_CHECK_FUTURE_FILE_TIMESTAMPS_=TRUE
_R_CHECK_LENGTH_1_CONDITION_=package:_R_CHECK_PACKAGE_NAME_,abort,verbose
_R_CHECK_LENGTH_1_LOGIC2_=package:_R_CHECK_PACKAGE_NAME_,abort,verbose
_R_CHECK_AUTOCONF_=true
_R_CHECK_DATALIST_=true
_R_CHECK_THINGS_IN_CHECK_DIR_=true
_R_CHECK_THINGS_IN_TEMP_DIR_=true
_R_CHECK_BASHISMS_=true
_R_CLASS_MATRIX_ARRARY_=true
_R_CHECK_ORPHANED_=true
_R_CHECK_BOGUS_RETURN_=true
_R_CHECK_MATRIX_DATA_=TRUE
_R_CHECK_CODE_CLASS_IS_STRING_=true
_R_CHECK_RD_VALIDATE_RD2HTML_=true
@end example

@noindent
These are turned on by @command{R CMD check --as-cran}: the incoming
checks also use
@example
_R_CHECK_FORCE_SUGGESTS_=FALSE
@end example

@noindent
since some packages do suggest other packages not available on CRAN or
other commonly-used repositories.

Several environment variables can be used to set `timeouts': limits for
the elapsed time taken by the sub-processes used for parts of the
checks.  A value of @code{0} indicates no limit, and is the default.
Character strings ending in @samp{s}, @samp{m} or @samp{h} indicate a
number of seconds, minutes or hours respectively: other values are
interpreted as a whole number of seconds (with invalid inputs being
treated as no limit).
@vtable @code
@item _R_CHECK_ELAPSED_TIMEOUT_
The default timeout for sub-processes not otherwise mentioned, and the
default value for all except @env{_R_CHECK_ONE_TEST_ELAPSED_TIMEOUT_}.
(This is also used by @code{tools::check_packages_in_dir}.)

@item _R_CHECK_INSTALL_ELAPSED_TIMEOUT_
Limit for when @command{R CMD INSTALL} is run by @command{check}.

@item _R_CHECK_EXAMPLES_ELAPSED_TIMEOUT_
Limit for running all the examples for one sub-architecture.

@item _R_CHECK_ONE_TEST_ELAPSED_TIMEOUT_
Limit for running one test for one sub-architecture.  Default
@env{_R_CHECK_TESTS_ELAPSED_TIMEOUT_}.

@item _R_CHECK_TESTS_ELAPSED_TIMEOUT_
Limit for running all the tests for one sub-architecture (and the
default limit for running one test).

@item _R_CHECK_ONE_VIGNETTE_ELAPSED_TIMEOUT_
Limit for running the @R{} code in one vignette, including for
re-building each vignette separately. 

@item _R_CHECK_BUILD_VIGNETTES_ELAPSED_TIMEOUT_
Limit for re-building all vignettes.

@item _R_CHECK_PKGMAN_ELAPSED_TIMEOUT_
Limit for each attempt at building the PDF package manual.
@end vtable

Another variable which enables stricter checks is to set
@env{R_CHECK_CONSTANTS} to @code{5}.  This checks that
nothing@footnote{The usual culprits are calls to compiled code
@emph{via} @code{.Call} or @code{.External} which alter their
arguments.} changes the values of `constants'@footnote{things which the
byte compiler assumes do not change, e.g.@: function bodies.} in @R{}
code.  This is best used in conjunction with setting
@env{R_JIT_STRATEGY} to @code{3}, which checks code on first use (by
default most code is only checked after byte-compilation on second use).
Unfortunately these checks slow down checking of examples, tests and
vignettes, typically two-fold but in the worst cases at least a
hundred-fold.

The following environment variables can be used to customize the
operation of @command{INSTALL}.

@vtable @code
@item _R_INSTALL_LIBS_ONLY_FORCE_DEPENDS_IMPORTS_
If true, give an error if installing only package libraries via
@option{--libs-only} and some package imported or depended on is not
available.
Default: true (false only for special applications, which analyze native
code of packages).
@end vtable

@node R coding standards, Testing R code, Tools, Top
@chapter R coding standards

@cindex coding standards
@R{} is meant to run on a wide variety of platforms, including Linux and
most variants of Unix as well as Windows and macOS.
Therefore, when extending @R{} by either adding to the @R{} base
distribution or by providing an add-on package, one should not rely on
features specific to only a few supported platforms, if this can be
avoided.  In particular, although most @R{} developers use @acronym{GNU}
tools, they should not employ the @acronym{GNU} extensions to standard
tools.  Whereas some other software packages explicitly rely on e.g.@:
@acronym{GNU} make or the @acronym{GNU} C++ compiler, @R{} does not.
Nevertheless, @R{} is a @acronym{GNU} project, and the spirit of the
@cite{@acronym{GNU} Coding Standards} should be followed if possible.

The following tools can ``safely be assumed'' for @R{} extensions.

@itemize @bullet
@item
An ISO C99 C compiler.  Note that extensions such as @acronym{POSIX}
1003.1 must be tested for, typically using Autoconf unless you are sure
they are supported on all mainstream @R{} platforms (including Windows
and macOS).

@item
A fixed-form Fortran compiler.

@item
A simple @command{make}, considering the features of @command{make} in
4.2 @acronym{BSD} systems as a baseline.
@findex make

@acronym{GNU} or other extensions, including pattern rules using
@samp{%}, the automatic variable @samp{$^}, the @samp{+=} syntax to
append to the value of a variable, the (``safe'') inclusion of makefiles
with no error, conditional execution, and many more, must not be used
(see Chapter ``Features'' in the @cite{@acronym{GNU} Make Manual} for
more information).  On the other hand, building @R{} in a separate
directory (not containing the sources) should work provided that
@command{make} supports the @code{VPATH} mechanism.

Windows-specific makefiles can assume @acronym{GNU} @command{make} 3.79
or later, as no other @command{make} is viable on that platform.

@item
A Bourne shell and the ``traditional'' Unix programming tools, including
@command{grep}, @command{sed}, and @command{awk}.

There are @acronym{POSIX} standards for these tools, but these may not
be fully supported.  Baseline features could be determined from a book
such as @cite{The UNIX Programming Environment} by Brian W. Kernighan &
Rob Pike.  Note in particular that @samp{|} in a regexp is an extended
regexp, and is not supported by all versions of @command{grep} or
@command{sed}.  The Open Group Base Specifications, Issue 7, which are
technically identical to  IEEE Std 1003.1 (POSIX), 2008,
are available at
@uref{https://pubs.opengroup.org/onlinepubs/9699919799/mindex.html}.
@end itemize

Under Windows, most users will not have these tools installed, and you
should not require their presence for the operation of your package.
However, users who install your package from source will have them, as
they can be assumed to have followed the instructions in ``the Windows
toolset'' appendix of the ``R Installation and Administration'' manual
to obtain them.  Redirection cannot be assumed to be available via
@command{system} as this does not use a standard shell (let alone a
Bourne shell).

@noindent
In addition, the following tools are needed for certain tasks.

@itemize @bullet
@item
Perl version 5 is needed for the maintainer-only script
@file{tools/help2man.pl}.
@findex Perl

@item
@command{texinfo} version 5.1 or later is needed to build the HTML, PDF
and Info files for the @R{} manuals written in the @acronym{GNU} Texinfo
system.  And that requires Perl.
@findex makeinfo
@end itemize

It is also important that code is written in a way that allows others to
understand it.  This is particularly helpful for fixing problems, and
includes using self-descriptive variable names, commenting the code, and
also formatting it properly.  The @R{} Core Team recommends to use a
basic indentation of 4 for @R{} and C (and most likely also Perl) code,
and 2 for documentation in Rd format.  Emacs (21 or later) users can
implement this indentation style by putting the following in one of
their startup files, and using customization to set the
@code{c-default-style} to @code{"bsd"} and @code{c-basic-offset} to
@code{4}.)
@findex emacs

@smallexample
@group
;;; ESS
(add-hook 'ess-mode-hook
          (lambda ()
            (ess-set-style 'C++ 'quiet)
            ;; Because
            ;;                                 DEF GNU BSD K&R C++
            ;; ess-indent-level                  2   2   8   5   4
            ;; ess-continued-statement-offset    2   2   8   5   4
            ;; ess-brace-offset                  0   0  -8  -5  -4
            ;; ess-arg-function-offset           2   4   0   0   0
            ;; ess-expression-offset             4   2   8   5   4
            ;; ess-else-offset                   0   0   0   0   0
            ;; ess-close-brace-offset            0   0   0   0   0
            (add-hook 'local-write-file-hooks
                      (lambda ()
                        (ess-nuke-trailing-whitespace)))))
(setq ess-nuke-trailing-whitespace-p 'ask)
;; or even
;; (setq ess-nuke-trailing-whitespace-p t)
@end group
@group
;;; Perl
(add-hook 'perl-mode-hook
          (lambda () (setq perl-indent-level 4)))
@end group
@end smallexample

@noindent
(The `GNU' styles for Emacs' C and R modes use a basic indentation of 2,
which has been determined not to display the structure clearly enough
when using narrow fonts.)

@node Testing R code, Use of TeX dialects, R coding standards, Top
@chapter Testing R code

When you (as @R{} developer) add new functions to the R base (all the
packages distributed with @R{}), be careful to check if @kbd{make
test-Specific} or particularly, @kbd{cd tests; make no-segfault.Rout}
still works (without interactive user intervention, and on a standalone
computer).  If the new function, for example, accesses the Internet, or
requires @acronym{GUI} interaction, please add its name to the ``stop
list'' in @file{tests/no-segfault.Rin}.

[To be revised: use @command{make check-devel}, check the write barrier
if you change internal structures.]

@node Use of TeX dialects, Current and future directions, Testing R code, Top
@chapter Use of TeX dialects

Various dialects of TeX are used for different purposes in @R{}.  The
policy is that manuals be written in @samp{texinfo}, and for convenience
the main and Windows FAQs are also.  This has the advantage that is is
easy to produce @HTML{} and plain text versions as well as typeset manuals.

@LaTeX{} is not used directly, but rather as an intermediate format for
typeset help documents and for vignettes.

Care needs to be taken about the assumptions made about the @R{} user's
system: it may not have either @samp{texinfo} or a TeX system
installed.  We have attempted to abstract out the cross-platform
differences, and almost all the setting of typeset documents is done by
@code{tools::texi2dvi}.  This is used for offline printing of help
documents, preparing vignettes and for package manuals via @command{R
CMD Rd2pdf}.  It is not currently used for the @R{} manuals created in
directory @file{doc/manual}.

@code{tools::texi2dvi} makes use of a system command @command{texi2dvi}
where available.  On a Unix-alike this is usually part of
@samp{texinfo}, whereas on Windows if it exists at all it would be an
executable, part of MiKTeX.  If none is available, the @R{} code runs
a sequence of @command{(pdf)latex}, @command{bibtex} and
@command{makeindex} commands.

This process has been rather vulnerable to the versions of the external
software used: particular issues have been @command{texi2dvi} and
@file{texinfo.tex} updates, mismatches between the two@footnote{Linux
distributions tend to unbundle @file{texinfo.tex} from @samp{texinfo}.},
versions of the @LaTeX{} package @samp{hyperref} and quirks in index
production.  The licenses used for @LaTeX{} and latterly @samp{texinfo}
prohibit us from including `known good' versions in the @R{}
distribution.

On a Unix-alike @command{configure} looks for the executables for TeX and
friends and if found records the absolute paths in the system
@file{Renviron} file.  This used to record @samp{false} if no command
was found, but it nowadays records the name for looking up on the path
at run time.  The latter can be important for binary distributions: one
does not want to be tied to, for example, TeX Live 2007.


@node Current and future directions, Function and variable index, Use of TeX dialects, Top
@chapter Current and future directions

This chapter is for notes about possible in-progress and future changes
to @R{}: there is no commitment to release such changes, let alone to a
timescale.

@menu
* Long vectors::                
* 64-bit types::                
* Large matrices::              
@end menu

@node Long vectors, 64-bit types, Current and future directions, Current and future directions
@section Long vectors

Vectors in @R{} 2.x.y were limited to a length of 2^31 - 1 elements
(about 2 billion), as the length is stored in the @code{SEXPREC} as a C
@code{int}, and that type is used extensively to record lengths and
element numbers, including in packages.

Note that longer vectors are effectively impossible under 32-bit
platforms because of their address limit, so this section applies only
on 64-bit platforms.  The internals are unchanged on a 32-bit build of
@R{}.

A single object with 2^31 or more elements will take up at least 8GB of
memory if integer or logical and 16GB if numeric or character, so
routine use of such objects is still some way off.

There is now some support for long vectors.  This applies to raw,
logical, integer, numeric and character vectors, and lists and
expression vectors.  (Elements of character vectors (@code{CHARSXP}s)
remain limited to 2^31 - 1 bytes.)  Some considerations:


@itemize

@item
This has been implemented by recording the length (and true length) as
@code{-1} and recording the actual length as a 64-bit field at the
beginning of the header.  Because a fair amount of code in @R{} uses a
signed type for the length, the `long length' is recorded using the
signed C99 type @code{ptrdiff_t}, which is typedef-ed to
@code{R_xlen_t}.

@item
These can in theory have 63-bit lengths, but note that current 64-bit
OSes do not even theoretically offer 64-bit address spaces and there is
currently a 52-bit limit (which exceeds the theoretical limit of current
OSes and ensures that such lengths can be stored exactly in doubles).

@item
The serialization format has been changed to accommodate longer lengths,
but vectors of lengths up to 2^31-1 are stored in the same way as
before.  Longer vectors have their length field set to @code{-1} and
followed by two 32-bit fields giving the upper and lower 32-bits of the
actual length.  There is currently a sanity check which limits lengths
to 2^48 on unserialization.

@item
The type @code{R_xlen_t} is made available to packages in C header
@file{Rinternals.h}: this should be fine in C code since C99 is
required.  People do try to use @R{} internals in C++, but C++98
compilers are not required to support these types.

@item
Indexing can be done via the use of doubles.  The internal indexing code
used to work with positive integer indices (and negative, logical and
matrix indices were all converted to positive integers): it now works
with either @code{INTSXP} or @code{REALSXP} indices.

@item
The @R{} function @code{length} returns a double value if the length
exceeds 2^31-1. Code calling @code{as.integer(length(x))} before passing
to @code{.C}/@code{.Fortran} should checks for an @code{NA} result.

@end itemize

@node 64-bit types, Large matrices, Long vectors, Current and future directions
@section 64-bit types

There is also some desire to be able to store larger integers in @R{},
although the possibility of storing these as @code{double} is often
overlooked (and e.g.@: file pointers as returned by @code{seek} are
already stored as @code{double}).

Different routes have been proposed:

@itemize

@item
Add a new type to @R{} and use that for lengths and indices---most likely
this would be a 64-bit signed type, say @code{longint}.  @R{}'s usual
implicit coercion rules would ensure that supplying an @code{integer}
vector for indexing or @code{length<-} would work.

@item
A more radical alternative is to change the existing @code{integer} type
to be 64-bit on 64-bit platforms (which was the approach taken by S-PLUS
for DEC/Compaq Alpha systems).  Or even on all platforms.

@item
Allow either @code{integer} or @code{double} values for lengths and
indices, and return @code{double} only when necessary.

@end itemize

The third has the advantages of minimal disruption to existing code and
not increasing memory requirements. In the first and third scenarios
both @R{}'s own code and user code would have to be adapted for lengths
that were not of type @code{integer}, and in the third code branches for
long vectors would be tested rarely.

Most users of the @code{.C} and @code{.Fortran} interfaces use
@code{as.integer} for lengths and element numbers, but a few omit these
in the knowledge that these were of type @code{integer}.  It may be
reasonable to assume that these are never intended to be used with long
vectors.

The remaining interfaces will need to cope with the changed
@code{VECTOR_SEXPREC} types.  It seems likely that in most cases lengths
are accessed by the @code{length} and @code{LENGTH}
functions@footnote{but @code{LENGTH} is a macro under some internal
uses.}  The current approach is to keep these returning 32-bit lengths and
introduce `long' versions @code{xlength} and @code{XLENGTH} which return
@code{R_xlen_t} values.


See also @uref{https://homepage.cs.uiowa.edu/~luke/talks/useR10.pdf}.

@node Large matrices,  , 64-bit types, Current and future directions
@section Large matrices

Matrices are stored as vectors and so were also limited to 2^31-1
elements.  Now longer vectors are allowed on 64-bit platforms, matrices
with more elements are supported provided that each of the dimensions is
no more than 2^31-1.  However, not all applications can be supported.

The main problem is linear algebra done by Fortran code compiled with
32-bit @code{INTEGER}.  Although not guaranteed, it seems that all the
compilers currently used with @R{} on a 64-bit platform allow matrices
each of whose dimensions is less than 2^31 but with 2^31 or more
elements and index them correctly, and a substantial part of the
support software (such as @acronym{BLAS} and @acronym{LAPACK}) also
work.

There are exceptions: for example some complex @acronym{LAPACK}
auxiliary routines do use a single @code{INTEGER} index and hence
overflow silently and segfault or give incorrect results.  One example
seen was @code{svd()} on a complex matrix.

Since this is implementation-dependent, it is possible that optimized
@acronym{BLAS} and @acronym{LAPACK} may have further restrictions:
a segfault have been reported from @code{svd()} using ATLAS on
@cputype{x86_64} Linux.
@c https://stat.ethz.ch/pipermail/r-devel/2021-August/081004.html

For matrix algebra on large matrices one almost certainly wants a
machine with a lot of RAM (100s of gigabytes), many cores and a
multi-threaded @acronym{BLAS}.



@node Function and variable index, Concept index, Current and future directions, 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: ***