doc/html/NEWS.3.html - R - Git at Google

 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><title>R: R News</title>
 <meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
 <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes" />
 <link rel="stylesheet" type="text/css" href="R.css" />
 </head><body><div class="container">


 <table width="100%" summary="page for NEWS"><tr><td>NEWS</td><td style="text-align: right;">R Documentation</td></tr></table>

 <h2>R News</h2>

 <h3>LATER NEWS</h3>


 <ul>
 <li><p> News for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 4.x.y can be found in HTML format at
 <a href="NEWS.html">NEWS.html</a>,
 and in text format in file
 <a href="../NEWS">NEWS</a>
 in the &lsquo;<span class="file">doc</span>&rsquo; directory.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.6.3</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> The included LAPACK has been updated to version 3.9.0 (for
 the included routines, just bug fixes).
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> Fixed a C level integer overflow in <code>rhyper()</code>; reported
 by Benjamin Tyner in <a href="https://bugs.R-project.org/show_bug.cgi?id=17694">PR#17694</a>.
 </p>
 </li>
 <li><p> Uses of <code>url(gzcon(.))</code> needing to extend buffer size
 have failed (with HTTP/2 servers), reported by Gábor Csárdi.
 </p>
 </li>
 <li> <p><code>predict(loess(..), se=TRUE)</code> now errors out (instead of
 seg.faulting etc) for large sample sizes, thanks to a report and
 patch by Benjamin Tyner in <a href="https://bugs.R-project.org/show_bug.cgi?id=17121">PR#17121</a>.
 </p>
 </li>
 <li> <p><code>tools:assertCondition(., "error")</code> and hence
 <code>assertError()</code> no longer return errors twice (invisibly).
 </p>
 </li>
 <li> <p><code>update(form, new)</code> in the case of a long <code>new</code>
 formula sometimes wrongly eliminated the intercept from <code>form</code>,
 or (more rarely) added a garbage term (or seg.faulted !); the fix
 happened by simplifying the C-level logic of <code>terms.formula()</code>.
 Reported by Mathias Ambühl in <a href="https://bugs.R-project.org/show_bug.cgi?id=16326">PR#16326</a>.
 </p>
 </li>
 <li><p> The error message from <code>stopifnot(.., &lt;error producing call&gt;)</code>
 again contains the full <code>"stopifnot(.......)"</code> call: Its
 attempted suppression did not work consistently.
 </p>
 </li>
 <li><p> On Windows, <code>download.file(., , "wininet", headers=character())</code>
 would fail; reported with patch proposal by Kevin Ushey in <a href="https://bugs.R-project.org/show_bug.cgi?id=17710">PR#17710</a>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.6.2</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>runmed(x, *)</code> gains a new option <code>na.action</code>
 determining <em>how</em> to handle <code>NaN</code> or <code>NA</code> in <code>x</code>.
 </p>
 </li>
 <li> <p><code>dotchart()</code> gains new options <code>ann</code>, <code>xaxt</code>,
 <code>frame.plot</code> and <code>log</code>.
 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> Detection of the C stack direction has been moved from
 run-time to <code>configure</code>: this is safer with LTO builds and
 allows the detection to be overridden &ndash; see file
 &lsquo;<span class="file">config.site</span>&rsquo;.
 </p>
 </li>
 <li><p> Source-code changes enable installation on platforms using
 <code>gcc -fno-common</code> (the expected default for
 <code>gcc 10.x</code>).
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li> <p><code>installTrChar</code> (which is nowadays is wrapped by
 <code>installChar</code>) is defined in &lsquo;<span class="file">Rinternals.h</span>&rsquo;.  (Neither
 are part of the API.)
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> Header &lsquo;<span class="file">Rconfig.h</span>&rsquo; contains the value of
 <code>FC_LEN_T</code> deduced at installation which is used by the
 prototypes in headers &lsquo;<span class="file">R_ext/BLAS.h</span>&rsquo; and
 &lsquo;<span class="file">R_ext/Lapack.h</span>&rsquo; but to avoid extensive breakage this is
 only exposed when <code>USE_FC_LEN_T</code> is defined.
 </p>
 <p>If a package's C/C++ calls to BLAS/LAPACK allow for the
 &lsquo;hidden&rsquo; arguments used by most Fortran compilers to pass
 the lengths of Fortran character arguments, define
 <code>USE_FC_LEN_T</code> and include &lsquo;<span class="file">Rconfig.h</span>&rsquo; (possibly
 <em>via</em> &lsquo;<span class="file">R.h</span>&rsquo;) before including &lsquo;<span class="file">R_ext/BLAS.h</span>&rsquo; or
 &lsquo;<span class="file">R_ext/Lapack.h</span>&rsquo;.
 </p>
 </li>
 <li><p> A package with Fortran source code and perhaps C (but not
 C++) sources can request for its shared object/DLL to be linked by
 the Fortran compiler by including a line <code>USE_FC_TO_LINK=</code> in
 &lsquo;<span class="file">src/Makevars[.win]</span>&rsquo; and using <code>$(SHLIB_OPENMP_FFLAGS)</code>
 as part of <code>PKG_LIBS</code>.
 </p>
 <p>The known reason for doing so is a package which uses Fortran
 (only) OpenMP on a platform where the Fortran OpenMP runtime is
 incompatible with the C one (e.g. <code>gfortran</code> 9.x
 with <code>clang</code>).
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> has a new option to mitigate checks
 leaving files/directories in &lsquo;<span class="file">/tmp</span>&rsquo;.  See the &lsquo;R
 Internals&rsquo; manual &ndash; this is part of <code>--as-cran</code>.
 </p>
 </li></ul>


 <h4>Windows</h4>


 <ul>
 <li><p> The default standard for C++ in package installation is
 C++11 (as it has been on other platforms where available since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.6.0: the default toolchain on Windows was defaulting to C++98).
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> Support for specifying C++98 in package installation is
 deprecated.
 </p>
 </li>
 <li><p> Support in <code>R CMD config</code> for <span class="samp">F77</span>,
 <span class="samp">FCPIFCPLAGS</span>, <span class="samp">CPP</span>, <span class="samp">CXXCPP</span> and <span class="samp">CXX98</span> and
 similar is deprecated.  (<span class="samp">CPP</span> is found from the system
 <code>make</code> and may well not be set.)
 </p>
 <p>Use <span class="samp">$CC -E</span> and <span class="samp">$CXX -E</span> instead of <span class="samp">CPP</span> and
 <span class="samp">CXXCPP</span>.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>runmed(x, *)</code> when <code>x</code> contains missing values
 now works consistently for both <code>algorithm="Stuetzle"</code> and
 <code>"Turlach"</code>, and no longer segfaults for <code>"Turlach"</code>, as
 reported by Hilmar Berger.
 </p>
 </li>
 <li> <p><code>apply(diag(3), 2:3, mean)</code> now gives a helpful error message.
 </p>
 </li>
 <li> <p><code>dgamma(x, shape, log=TRUE)</code> now longer overflows
 to <code>Inf</code> for shape &lt; 1 and very small <code>x</code>, fixing
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17577">PR#17577</a>, reported by Jonathan Rougier.
 </p>
 </li>
 <li><p> Buffer overflow in building error messages fixed. Reported
 by Benjamin Tremblay.
 </p>
 </li>
 <li> <p><code>options(str = .)</code> is correctly initialized at
 package <span class="pkg">utils</span> load time, now.  A consequence is that
 <code>str()</code> in scripts now is more consistent to interactive use,
 e.g., when displaying function(**) argument lists.
 </p>
 </li>
 <li> <p><code>as.numeric(&lt;call&gt;)</code> now gives correct error message.
 </p>
 </li>
 <li><p> Printing <code>ls.str()</code> no longer wrongly shows
 <code>"&lt;missing&gt;"</code> in rare cases.
 </p>
 </li>
 <li><p> Auto-printing S4 objects no longer duplicates the object,
 for faster speed and reduced memory consumption. Reported by Aaron
 Lun.
 </p>
 </li>
 <li> <p><code>pchisq(&lt;LRG&gt;, &lt;LRG&gt;, ncp=100)</code> no longer takes
 practically forever in some cases.  Hence ditto for corresponding
 <code>qchisq()</code> calls.
 </p>
 </li>
 <li> <p><code>x %% L</code> for finite <code>x</code> no longer returns
 <code>NaN</code> when <code>L</code> is infinite, nor suffers from cancellation
 for large finite <code>L</code>, thanks to Long Qu's <a href="https://bugs.R-project.org/show_bug.cgi?id=17611">PR#17611</a>.
 </p>
 <p>Analogously, <code>x %/% L</code> and <code>L %/% x</code> suffer less from
 cancellation and return values corresponding to limits for large
 <code>L</code>.
 </p>
 </li>
 <li> <p><code>grepl(NA, *)</code> now returns <code>logical</code> as documented.
 </p>
 </li>
 <li> <p><code>options(warn=1e11)</code> is an error now, instead of later
 leading to C stack overflow because of infinite recursion.
 </p>
 </li>
 <li> <p><code>R_tryCatch</code> no longer transfers control for all
 conditions. Reported and patch provided by Lionel Henry in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17617">PR#17617</a>.
 </p>
 </li>
 <li> <p><code>format(object.size(.), digits=NULL)</code> now works, fixing
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17628">PR#17628</a> reported by Jonathan Carroll.
 </p>
 </li>
 <li> <p><code>get_all_vars(f, d)</code> now also works for cases, e.g. where
 <code>d</code> contains a matrix.  Reported by Simon Wood in 2009 and patch
 provided by Ben Bolker in <a href="https://bugs.R-project.org/show_bug.cgi?id=13624">PR#13624</a>.
 </p>
 <p>Additionally, it now also works when some variables are data frames,
 fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=14905">PR#14905</a>, reported by Patrick Breheny.
 </p>
 </li>
 <li> <p><code>barplot()</code> could get spacings wrong if there were exactly two
 bars <a href="https://bugs.R-project.org/show_bug.cgi?id=15522">PR#15522</a>.  Patch by Michael Chirico.
 </p>
 </li>
 <li> <p><code>power.t.test()</code> works in more cases when returning values
 of n smaller than 2.
 </p>
 </li>
 <li> <p><code>dotchart(*, pch=., groups=.)</code> now works better.  Reported
 by Robert and confirmed by Nic Rochette in <a href="https://bugs.R-project.org/show_bug.cgi?id=16953">PR#16953</a>.
 </p>
 </li>
 <li> <p><code>canCoerce(obj, cl)</code> no longer assumes
 <code>length(class(obj)) == 1</code>.
 </p>
 </li>
 <li> <p><code>plot.formula(*, subset = *)</code> now also works in a boundary
 case reported by Robert Schlicht (TU Dresden).
 </p>
 </li>
 <li> <p><code>readBin()</code> and <code>writeBin()</code> of a
 <code>rawConnection()</code> now also work in large cases, thanks to a
 report and proposal by Taeke Harkema in <a href="https://bugs.R-project.org/show_bug.cgi?id=17665">PR#17665</a>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.6.1</h3>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> The default detection of the shell variable <code>libNN</code> is
 overridden for derivatives of Debian Linux, some of which have
 started to have a &lsquo;<span class="file">/usr/lib64</span>&rsquo; directory.
 (E.g. Ubuntu 19.04.)  As before, it can be specified in
 &lsquo;<span class="file">config.site</span>&rsquo;.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD config</code> knows the values of <code>AR</code> and
 <code>RANLIB</code>, often set for LTO builds.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> The use of a character vector with <code>.Fortran()</code> is
 formally deprecated and gives a non-portability warning.  (It has
 long been strongly discouraged in &lsquo;Writing R Extensions&rsquo;.)
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> On Windows, GUI package installation via <code>menuInstallPkgs()</code>
 works again, thanks to Len Weil's and Duncan Murdoch's <a href="https://bugs.R-project.org/show_bug.cgi?id=17556">PR#17556</a>.
 </p>
 </li>
 <li> <p><code>R CMD check</code> on <code>data()</code> fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17558">PR#17558</a>
 thanks to Duncan Murdoch.
 </p>
 </li>
 <li> <p><code>quasi(*, variance = list(..))</code> now works more
 efficiently, and should work in all cases fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17560">PR#17560</a>.
 Further, <code>quasi(var = mu(1-mu))</code> and <code>quasi(var = "mu ^ 3")</code>
 now work, and <code>quasi(variance = "log(mu)")</code> now gives a correct
 error message.
 </p>
 </li>
 <li><p> Creation of lazy loading database during package installation is
 again robust to Rprofile changing the current working directory
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17559">PR#17559</a>).
 </p>
 </li>
 <li> <p><code>boxplot(y ~ f, horizontal=TRUE)</code> now produces correct x-
 and y-labels.
 </p>
 </li>
 <li> <p><code>rbind.data.frame()</code> allows to keep <code>&lt;NA&gt;</code> levels
 from factor columns (<a href="https://bugs.R-project.org/show_bug.cgi?id=17562">PR#17562</a>) via new option <code>factor.exclude</code>.
 </p>
 <p>Additionally, it works in one more case with matrix-columns which
 had been reported on 2017-01-16 by Krzysztof Banas.
 </p>
 </li>
 <li><p> Correct messaging in C++ pragma checks in <span class="pkg">tools</span> code for
 <code>R CMD check</code>, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17566">PR#17566</a> thanks to Xavier Robin.
 </p>
 </li>
 <li><p>print()ing and auto-printing no longer differs for
 <code>function</code>s with a user defined <code>print.function</code>, thanks to
 Bill Dunlap's report.
 </p>
 </li>
 <li><p> On Windows, <code>writeClipboard(.., format = &lt;n&gt;)</code> now
 does correctly pass <code>format</code> to the underlying C code, thanks to
 a bug report (with patch) by Jenny Bryan.
 </p>
 </li>
 <li> <p><code>as.data.frame()</code> treats 1D arrays the same as vectors,
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17570">PR#17570</a>.
 </p>
 </li>
 <li><p> Improvements in <code>smoothEnds(x, *)</code> working with
 <code>NA</code>s (towards <code>runmed()</code> working in that case, in the next
 version of R).
 </p>
 </li>
 <li> <p><code>vcov(glm(&lt;quasi&gt;), dispersion = *)</code> works correctly
 again, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17571">PR#17571</a> thanks to Pavel Krivitsky.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> of binary packages on Windows now
 works also with per-directory locking.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> and <code>install.packages()</code> on Windows
 are now more robust against a locked file in an earlier installation
 of the package to be installed.  The default value of option
 <code>install.lock</code> on Windows has been changed to <code>TRUE</code>.
 </p>
 </li>
 <li><p> On Unix alikes (when <code>readline</code> is active), only
 expand tilde (<code>~</code>) file names starting with a tilde, instead of
 almost all tildes.
 </p>
 </li>
 <li><p> In R documentation (&lsquo;<span class="file">*.Rd</span>&rsquo;) files, <span class="samp">\item [..]</span> is no
 longer treated specially when rendered in LaTeX and hence pdf, but
 rather shows the brackets in all cases.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.6.0</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li><p> Serialization format version 3 becomes the default for
 serialization and saving of the workspace (<code>save()</code>,
 <code>serialize()</code>, <code>saveRDS()</code>, <code>compiler::cmpfile()</code>).
 Serialized data in format 3 cannot be read by versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> prior
 to version 3.5.0.  Serialization format version 2 is still
 supported and can be selected by <code>version = 2</code> in the
 save/serialization functions.  The default can be changed back for
 the whole <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session by setting environment variables
 <span class="env">R_DEFAULT_SAVE_VERSION</span> and <span class="env">R_DEFAULT_SERIALIZE_VERSION</span>
 to <code>2</code>.  For maximal back-compatibility, files
 &lsquo;<span class="file">vignette.rds</span>&rsquo; and &lsquo;<span class="file">partial.rdb</span>&rsquo; generated by <code>R
 	 CMD build</code> are in serialization format version 2, and resave by
 default produces files in serialization format version 2 (unless
 the original is already in format version 3).
 </p>
 </li>
 <li><p> The default method for generating from a discrete uniform
 distribution (used in <code>sample()</code>, for instance) has been
 changed.  This addresses the fact, pointed out by Ottoboni and
 Stark, that the previous method made <code>sample()</code> noticeably
 non-uniform on large populations.  See <a href="https://bugs.R-project.org/show_bug.cgi?id=17494">PR#17494</a> for a
 discussion.  The previous method can be requested using
 <code>RNGkind()</code> or <code>RNGversion()</code> if necessary for
 reproduction of old results.  Thanks to Duncan Murdoch for
 contributing the patch and Gabe Becker for further assistance.
 </p>
 <p>The output of <code>RNGkind()</code> has been changed to also return the
 &lsquo;kind&rsquo; used by <code>sample()</code>.
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>Sys.setFileTime()</code> has been vectorized so arguments
 <code>path</code> and <code>time</code> of length greater than one are now
 supported.
 </p>
 </li>
 <li> <p><code>axis()</code> gets new option <code>gap.axis = NA</code> for
 specifying a multiplication factor for the minimal &ldquo;gap&rdquo;
 (distance) between axis labels drawn.   Its default is <code>1</code>
 for labels <em>parallel</em> to the axis, and <code>0.25</code> for
 perpendicular ones.
 </p>
 <p>Perpendicular labels no longer overlap, fixing bug <a href="https://bugs.R-project.org/show_bug.cgi?id=17384">PR#17384</a>.
 </p>
 </li>
 <li><p> The default method of <code>plot()</code> gains new arguments
 <code>xgap.axis = NA</code> and <code>ygap.axis = NA</code> to be passed to
 the x&ndash; and y&ndash; <code>axis(.., gap.axis=*)</code> calls.
 </p>
 </li>
 <li> <p><code>removeSource()</code> now works not only for functions but
 also for some language objects.
 </p>
 </li>
 <li> <p><code>as.call()</code>, <code>rep.int()</code>, <code>rep_len()</code> and
 <code>nchar()</code> dispatch internally.
 </p>
 </li>
 <li> <p><code>is(object, class2)</code> looks for <code>class2</code> in the
 calling namespace after looking in the namespace of
 <code>class(object)</code>.
 </p>
 </li>
 <li> <p><code>extendrange(.., f)</code> with a length-2 <code>f</code> now
 extends separately to the left and the right.
 </p>
 </li>
 <li> <p><code>lengths()</code> dispatches internally to S4 methods.
 </p>
 </li>
 <li> <p><code>download.file()</code> on Windows now uses <code>URLdecode()</code>
 to determine the file extension, and uses binary transfer
 (<code>mode = "wb"</code>) also for file extension &lsquo;<span class="file">.rds</span>&rsquo;.
 </p>
 <p>The help page for <code>download.file()</code> now contains the same
 information on all platforms.
 </p>
 </li>
 <li><p> Setting <span class="samp">C</span> locale for collation <em>via</em> environment
 variables <span class="env">LC_ALL</span> and <span class="env">LC_COLLATE</span> and <em>via</em> a call
 to <code>Sys.setlocale()</code> now takes precedence over environment
 variable <span class="env">R_ICU_LOCALE</span>.
 </p>
 </li>
 <li><p> There is a new function, <code>nullfile()</code>, to give the file
 name of the null system device (e.g., &lsquo;<span class="file">/dev/null</span>&rsquo;) on the
 current platform.
 </p>
 </li>
 <li><p> There are two new options, <code>keep.parse.data</code> and
 <code>keep.parse.data.pkgs</code>, which control whether parse data are
 included into sources when <code>keep.source</code> or
 <code>keep.source.pkgs</code> is <code>TRUE</code>.  By default,
 <code>keep.parse.data.pkgs</code> is now <code>FALSE</code>, which changes
 previous behavior and significantly reduces space and time overhead
 when sources are kept when installing packages.
 </p>
 </li>
 <li><p> In <code>rapply(x, ..)</code>, <code>x</code> can also be
 &ldquo;list-like&rdquo; and of length <i>&gt;= 2^{31}</i>.
 </p>
 </li>
 <li> <p><code>trimws()</code> gets new optional <code>whitespace</code>
 argument, allowing more extensive definitions of &ldquo;space&rdquo;,
 such as including Unicode spaces (as wished in <a href="https://bugs.R-project.org/show_bug.cgi?id=17431">PR#17431</a>).
 </p>
 </li>
 <li> <p><code>weighted.mean()</code> no longer coerces the weights to a
 double/numeric vector, since <code>sum()</code> now handles integer
 overflow. This makes <code>weighted.mean()</code> more polymorphic and
 endomorphic, but be aware that the results are no longer
 guaranteed to be a vector of type <span class="samp">double</span>.
 </p>
 </li>
 <li><p> When loading namespaces, S3 method registrations which
 overwrite previous registrations are now noted by default (using
 <code>packageStartupMessage()</code>).
 </p>
 </li>
 <li> <p><code>compiler::cmpfile()</code> gains a <code>version</code> argument, for
 use when the output file should be saved in serialization format 2.
 </p>
 </li>
 <li><p> The axis labeling in the default method of <code>pairs()</code>
 may now be toggled by new options <code>horOdd</code> and <code>verOdd</code>.
 </p>
 </li>
 <li><p> (Not Windows nor macOS.) Package <span class="pkg">tcltk</span> now supports an
 environment variable <span class="env">R_DONT_USE_TK</span> which if set disables Tk
 initialization.  This is intended for use to circumvent errors in
 loading the package, e.g. with recent Linux running under
 an address sanitizer.
 </p>
 </li>
 <li><p> The <code>numeric</code> method of <code>all.equal()</code> gets
 optional arguments <code>countEQ</code> and <code>formatFUN</code>.  If
 <code>countEQ</code> is true, the mean error is more sensible when many
 entries are <b>eq</b>ual.
 </p>
 </li>
 <li> <p><code>outer(x,y, FUN = "*")</code> is more efficient using
 <code>tcrossprod(u,v)</code> instead of <code>u %*% t(v)</code>.
 </p>
 </li>
 <li> <p><code>vcov(&lt;mlm&gt;)</code> is more efficient via new optional
 arguments in <code>summary.mlm()</code>.
 </p>
 </li>
 <li><p> The default method of <code>summary()</code> gets an option to
 choose the <em>kind</em> of <code>quantile()</code>s to use; wish of
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17438">PR#17438</a>.
 </p>
 </li>
 <li><p> Fitting multiple linear models <em>via</em> <code>lm()</code> does
 work with <em>matrix</em> offsets, as suggested in <a href="https://bugs.R-project.org/show_bug.cgi?id=17407">PR#17407</a>.
 </p>
 </li>
 <li><p> The new functions <code>mem.maxVSize()</code> and
 <code>mem.maxMSize()</code> allow the maximal size of the vector heap
 and the maximal number of nodes allowed in the current <code>R</code>
 process to be queried and set.
 </p>
 </li>
 <li> <p><code>news()</code> gains support for &lsquo;<span class="file">NEWS.md</span>&rsquo; files.
 </p>
 </li>
 <li><p> An effort has been started to have our reference manuals,
 i.e., all help pages. show platform-independent information (rather
 than Windows or Unix-alike specifics visible only on that platform).
 Consequently, the Windows version of <code>X11()</code> / <code>x11()</code>
 got identical formal arguments to the Unix one.
 </p>
 </li>
 <li> <p><code>sessionInfo()$running</code> has been factored out in a new
 variable <code>osVersion</code>.
 </p>
 </li>
 <li> <p><code>slice.index()</code> now also works for multi-dimensional
 margins.
 </p>
 </li>
 <li> <p><code>untar()</code> used with an external <code>tar</code> command
 assumes this supports decompression including <code>xz</code> and
 automagically detecting the compression type.  This has been true
 of all mainstream implementations since 2009 (for GNU
 <code>tar</code>, since version 1.22): older implementations are
 still supported <em>via</em> the new argument
 <code>support_old_tars</code> whose default is controlled by environment
 variable <span class="env">R_SUPPORT_OLD_TARS</span>.  (It looks like NetBSD and
 OpenBSD have &lsquo;older&rsquo; <code>tar</code> commands for this
 purpose.)
 </p>
 </li>
 <li><p> The new function <code>asplit()</code> allow splitting an array or
 matrix by its margins.
 </p>
 </li>
 <li><p> New functions <code>errorCondition()</code> and
 <code>warningCondition()</code> provide a convenient way to create
 structured error and warning objects.
 </p>
 <p><code>.Deprecated()</code> now signals a warning of class
 <code>"deprecatedWarning"</code>, and <code>.Defunct()</code> now signals an
 error of class <code>"defunctError"</code>.
 </p>
 </li>
 <li><p> Many &lsquo;package not found&rsquo; errors are now signaled as
 errors of class <code>"packageNotFoundError"</code>.
 </p>
 </li>
 <li><p> As an experimental feature, when <code>loadNamespace()</code> fails
 because the requested package is not available the error is
 initially signaled with a <code>retry_loadNamespace</code> restart
 available.  This allows a calling handler to try to install the
 package and continue.
 </p>
 </li>
 <li> <p><code>S3method()</code> directives in &lsquo;<span class="file">NAMESPACE</span>&rsquo; can now
 also be used to perform <em>delayed</em> S3 method registration.
 </p>
 </li>
 <li><p> Experimentally, setting environment variable
 <span class="env">_R_CHECK_LENGTH_1_LOGIC2_</span> will lead to warnings (or errors
 if the variable is set to a &lsquo;true&rsquo; value) when <code>&amp;&amp;</code> or
 <code>||</code> encounter and use arguments of length more than one.
 </p>
 </li>
 <li><p> Added <code>"lines"</code> and <code>"chars"</code> coordinate systems
 to <code>grconvertX()</code> and <code>grconvertY()</code>.
 </p>
 </li>
 <li> <p><code>getOption()</code> is more efficient notably for the rare
 case when called with two arguments, from several contributors in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17394">PR#17394</a>.
 </p>
 </li>
 <li><p> In <code>.col(dim)</code> and <code>.row(dim)</code>, <code>dim</code> now may
 also be an integer-valued <code>"double"</code>.
 </p>
 </li>
 <li> <p><code>sQuote()</code> and <code>dQuote()</code> get an explicit <code>q</code>
 argument with obvious default instead of using
 <code>getOption("fancyQuotes")</code> implicitly and unconditionally.
 </p>
 </li>
 <li> <p><code>unzip()</code> can list archives with comments and with
 spaces in file names even using an external <code>unzip</code> command.
 </p>
 </li>
 <li><p> Command line completion has a new setting
 <code>rc.settings(dots = FALSE)</code> to remove <code>...</code> from the
 list of possible function arguments.
 </p>
 </li>
 <li> <p><code>library()</code> no longer checks packages with compiled
 code match <span class="samp">R.version$platform</span>.  <code>loadNamespace()</code>
 never has, and increasingly the &lsquo;canonical name&rsquo; does not
 reflect the important characteristics of compiled code.

 </p>
 </li>
 <li><p> The primitive functions <code>drop()</code> and <code>unclass()</code>
 now avoid duplicating their data for atomic vectors that are large
 enough, by returning <code>ALTREP</code> wrapper objects with adjusted
 attributes.  <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>-level assignments to change attributes will also
 use wrapper objects to avoid duplicating data for larger atomic
 vectors. <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> functions like <code>structure()</code> and <code>unname()</code>
 will therefore not duplicate data in these settings.  Generic
 vectors as produced by <code>list()</code> are not yet covered by this
 optimization but may be in due course.
 </p>
 </li>
 <li><p> In <code>formals()</code>, <code>envir</code> becomes an optional
 argument instead of being hardwired.
 </p>
 </li>
 <li><p> Instead of signalling an error for an invalid S4 object
 <code>x</code>, <code>str(x)</code> now gives a warning and subsequently still
 shows most parts of <code>x</code>, e.g., when slots are missing.
 </p>
 </li>
 <li> <p><code>gamma(x)</code> and <code>lgamma(x)</code> no longer warn when
 correctly returning <code>Inf</code> or underflowing to zero.  This
 helps maximum likelihood and similar computations.
 </p>
 </li>
 <li> <p><code>convertColor()</code> is now vectorized, so a lot faster for
 converting many colours at once.  The new argument
 <code>vectorized</code> to <code>colorConverter()</code> ensures that
 non-vectorized colour converters still work.  (Thanks to Brodie
 Gaslam.)
 </p>
 </li>
 <li> <p><code>download.file()</code> and <code>url()</code> get new argument
 <code>headers</code> for custom HTTP headers, e.g., allowing to perform
 basic http authentication, thanks to a patch contributed by Gábor
 Csárdi.
 </p>
 </li>
 <li><p> File-based connection functions <code>file()</code>,
 <code>gzfile()</code>, <code>bzfile()</code> and <code>xzfile()</code> now signal an
 error when used on a directory.
 </p>
 </li>
 <li><p> For <code>approx()</code>, <code>splinefun()</code> <em>etc</em>, a new
 setting <code>ties = c("ordered", &lt;fun&gt;)</code> allows skipping the
 sorting and still treat ties.
 </p>
 </li>
 <li> <p><code>format(x)</code> gives a more user friendly error message in
 the case where no method is defined.  A minimal method is provided
 in <code>format.default(x)</code> when <code>isS4(x)</code> is true.
 </p>
 </li>
 <li> <p><code>which(x)</code> now also works when <code>x</code> is a long
 vector, thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17201">PR#17201</a>.  <strong>NB</strong>: this
 may return a double result, breaking the previous guarantee of an
 integer result.
 </p>
 </li>
 <li> <p><code>seq.default()</code> is more careful to return an
 <code>integer</code> (as opposed to <code>double</code>) result when its
 arguments are large and/or classed objects; see comment #9 of
 Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17497">PR#17497</a>.
 </p>
 </li>
 <li><p> The <code>plot()</code> method for <code>lm</code> and <code>glm</code> fits,
 <code>plot.lm()</code>, gains a new option <code>iter.smooth</code> with a
 default of <code>0</code> for binomial fits, no longer down-weighting
 when smoothing the residuals.
 </p>
 </li>
 <li> <p><code>zip()</code> passes its list of files <em>via</em> standard
 input to the external command when too long for the command line
 (on some platforms).
 </p>
 </li>
 <li> <p><code>data()</code> gains an <code>overwrite</code> argument.
 </p>
 </li>
 <li> <p><code>t.test()</code> now also returns the standard error (in
 list component <code>stderr</code>).
 </p>
 </li>
 <li> <p><code>model.matrix(*, contrasts.arg = CC)</code> now warns about
 invalid <code>contrasts.arg</code>s.
 </p>
 </li>
 <li><p> Performance of <code>substr()</code> and <code>substring()</code> has been
 improved.
 </p>
 </li>
 <li> <p><code>stopifnot()</code> has been simplified thanks to Suharto
 Anggono's proposals to become considerably faster for cheap
 expressions.
 </p>
 </li>
 <li><p> The default &lsquo;user agent&rsquo; has been changed when
 accessing <span class="samp">http://</span> and <span class="samp">https://</span> sites using
 <span class="samp">libcurl</span>.  (A site was found which caused <span class="samp">libcurl</span> to
 infinite-loop with the previous default.)
 </p>
 </li>
 <li> <p><code>sessionInfo()</code> now also contains <code>RNGkind()</code> and
 prints it when it differs from the default; based on a proposal and
 patch by Gabe Becker in <a href="https://bugs.R-project.org/show_bug.cgi?id=17535">PR#17535</a>.
 Also, <code>RNGversion(getRversion())</code> works directly.
 </p>
 </li>
 <li> <p><code>library()</code> and <code>require()</code> now allow more control
 over handling search path conflicts when packages are
 attached. The policy is controlled by the new
 <code>conflicts.policy</code> option.
 </p>
 </li>
 <li> <p><code>barplot()</code> gets a formula method, thanks to a patch
 proposal by Arni Magnusson in <a href="https://bugs.R-project.org/show_bug.cgi?id=17521">PR#17521</a>.
 </p>
 </li>
 <li> <p><code>pmax()</code> and <code>pmin(x)</code> now also work for long
 vectors, thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17533">PR#17533</a>.
 </p>
 </li>
 <li> <p><code>bxp()</code> now warns when omitting duplicated arguments.
 </p>
 </li>
 <li><p> New <code>hcl.colors()</code> function to provide wide range of
 HCL-based colour palettes with much better perceptual properties
 than the existing RGB/HSV-based palettes like <code>rainbow()</code>.
 </p>
 <p>Also a new <code>hcl.pals()</code> function to list available palette
 names for <code>hcl.colors()</code>.
 </p>
 <p>Contributed by Achim Zeileis.
 </p>
 </li>
 <li><p> The default colours for <code>image()</code> and
 <code>filled.contour()</code> are now based on <code>hcl.colors()</code>.
 </p>
 </li>
 <li><p> The palette-generating functions <code>rainbow()</code>,
 <code>gray.colors()</code>, etc. get a new <code>rev</code> argument
 to facilitate reversing the order of colors.
 </p>
 </li>
 <li><p> New <code>str2lang()</code> and <code>str2expression()</code> as streamlined
 versions of <code>parse(text=., keep.source=FALSE)</code> allow to abstract
 typical call constructions, e.g., in <code>formula</code> manipulations.
 (Somewhat experimental)
 </p>
 </li>
 <li><p> Add <code>update_PACKAGES()</code> for incrementally updating a
 package repository index, instead of rebuilding the index from
 scratch. Thanks to Gabe Becker in <a href="https://bugs.R-project.org/show_bug.cgi?id=17544">PR#17544</a> for the patch, based
 on part of his <span class="pkg">switchr</span> package.
 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> The options selected for the C++ compiler default to the
 C++11 standard if supported, otherwise to the C++98 standard.
 </p>
 </li>
 <li><p> Visibility macros such as <span class="samp">C_VISIBILITY</span> can now be
 user-set (including to empty), e.g. in &lsquo;<span class="file">config.site</span>&rsquo;.
 </p>
 </li>
 <li><p> Macro <span class="samp">FCLIBS</span>, which has sometimes been needed on
 Solaris, has been renamed to <span class="samp">FCLIBS_XTRA</span>.
 </p>
 </li>
 <li><p> Macro <span class="samp">F77</span> is always set to the value of <span class="samp">FC</span>, so
 the latter should be set to user-select the Fortran compiler for
 both fixed-form and free-form Fortran.  In particular,
 <code>gfortran</code> is now the first choice for <span class="samp">F77</span>, not
 <code>f95</code>.
 </p>
 <p>Macros <span class="samp">FFLAGS</span> and <span class="samp">FCFLAGS</span> remain distinct to allow
 for a compiler which needs a flag to select free- or fixed-form
 Fortran (most use the source-file extension to choose: &lsquo;<span class="file">.f</span>&rsquo;
 is fixed-form and &lsquo;<span class="file">.f90</span>&rsquo; and &lsquo;<span class="file">.f95</span>&rsquo; are free-form).
 </p>
 <p>If only one of them is set, its value is used for both.
 </p>
 </li>
 <li><p> The special-casing of <span class="samp">CFLAGS</span>, <span class="samp">CXXFLAGS</span> and
 <span class="samp">FFLAGS</span> for Intel compilers on Linux has been removed: we do
 not have recent experience but the generic defaults now chosen are
 the same as those previously special-cased for <span class="samp">x86_64</span>.
 </p>
 <p>If necessary, override the defaults on the <code>configure</code>
 command line or in file &lsquo;<span class="file">config.site</span>&rsquo;.
 </p>
 </li>
 <li><p> Long-untested <code>configure</code> support for HP-UX and very
 old versions of Linux has been removed.
 </p>
 </li>
 <li> <p><code>configure --with-blas</code> (without specifying a value)
 includes OpenBLAS in its search (before ATLAS and a generic
 BLAS).  This follows recent versions of the <span class="samp">ax_blas</span>
 autoconf macro.
 </p>
 </li>
 <li><p> The <code>configure</code> macro <span class="samp">MAKEINFO</span> has been
 updated to <span class="samp">TEXI2ANY</span>.
 </p>
 </li>
 <li><p> Support for <code>make install-strip</code> has been enhanced.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> Source package installation is by default &lsquo;staged&rsquo;:
 the package is installed into a temporary location under the final
 library directory and moved into place once the installation is
 complete.  The benefit is that partially-installed packages are
 hidden from other <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> sessions.
 </p>
 <p>The overall default is set by environment variable
 <span class="env">R_INSTALL_STAGED</span>.  <code>R CMD INSTALL</code> has new options
 <span class="option">--staged-install</span> and <span class="option">--no-staged-install</span>, and
 packages can use the <span class="samp">StagedInstall</span> field in their
 &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file to opt out.  (That opt-out is a temporary
 measure which may be withdrawn in future.)
 </p>
 <p>Staged installation requires either <span class="option">--pkglock</span> or
 <span class="option">--lock</span>, one of which is used by default.
 </p>
 </li>
 <li><p> The interpretation of source code with extension &lsquo;<span class="file">.f</span>&rsquo;
 is changing.  Previously this denoted FORTRAN 77 code, but current
 compilers no longer have a FORTRAN 77 mode and interpret it as
 &lsquo;fixed-form&rsquo; Fortran 90 (or later where supported) code.
 Extensions &lsquo;<span class="file">.f90</span>&rsquo; and &lsquo;<span class="file">.f95</span>&rsquo; continue to indicate
 &lsquo;free-form&rsquo; Fortran code.
 </p>
 <p>Legal FORTRAN 77 code is also legal fixed-form Fortran 9x; however
 this change legitimizes the use of later features, in particular
 to replace features marked &lsquo;obsolescent&rsquo; in Fortran 90 and
 &lsquo;deleted&rsquo; in Fortran 2018 which <code>gfortran</code> 8.x and
 later warn about.
 </p>
 </li>
 <li><p> Packages containing files in the &lsquo;<span class="file">src</span>&rsquo; directory with
 extensions &lsquo;<span class="file">.f90</span>&rsquo; or &lsquo;<span class="file">.f95</span>&rsquo; are now linked using the C
 or C++ compiler rather than the Fortran 9x compiler.  This is
 consistent with fixed-form Fortran code and allows mixing of C++
 and free-form Fortran on most platforms.
 </p>
 <p>Consequentially, a package which includes free-form Fortran 9x
 code which uses OpenMP should include <span class="samp">SHLIB_OPENMP_CFLAGS</span>
 (or the <span class="samp">CXXFLAGS</span> version if they also include C++ code) in
 <span class="samp">PKG_LIBS</span> rather than <span class="samp">SHLIB_OPENMP_FCFLAGS</span> &mdash;
 fortunately on almost all current platforms they are the same
 flag.
 </p>
 </li>
 <li><p> Macro <span class="samp">PKG_FFLAGS</span> will be used for the compilation of
 both fixed-form and free-form Fortran code unless
 <span class="samp">PKG_FCFLAGS</span> is also set (in &lsquo;<span class="file">src/Makevars</span>&rsquo; or
 &lsquo;<span class="file">src/Makevars.win</span>&rsquo;).
 </p>
 </li>
 <li><p> The make macro <span class="samp">F_VISIBILITY</span> is now preferred for
 both fixed-form and free-form Fortran, for use in
 &lsquo;<span class="file">src/Makevars</span>&rsquo; and similar.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> gains a new option <span class="option">--strip</span>
 which (where supported) strips installed shared object(s): this
 can also be achieved by setting the environment variable
 <span class="env">_R_SHLIB_STRIP_</span> to a true value.
 </p>
 <p>The new option <span class="option">--strip-lib</span> attempts stripping of
 static and shared libraries installed under &lsquo;<span class="file">lib</span>&rsquo;.
 </p>
 <p>These are most useful on platforms using GNU <code>binutils</code> (such
 as Linux) and compiling with <span class="option">-g</span> flags.
 </p>
 </li>
 <li><p> There is more support for installing UTF-8-encoded packages
 in a strict Latin-1 locale (and probably for other Latin locales):
 non-ASCII comments in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> code (and &lsquo;<span class="file">NAMESPACE</span>&rsquo; files) are
 worked around better.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> now optionally checks makefiles for
 correct and portable use of the <span class="samp">SHLIB_OPENMP_*FLAGS</span> macros.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now evaluates <code>\Sexpr{}</code>
 expressions (including those in macros) before checking the
 contents of &lsquo;<span class="file">Rd</span>&rsquo; files and so detects issues both in
 evaluating the expressions and in the expanded contents.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now lists missing packages separated
 by commas and with regular quotes such as to be useful as argument
 in calling <code>install.packages(c(..))</code>; from a suggestion by
 Marcel Ramos.
 </p>
 </li>
 <li> <p><code>tools::Rd2latex()</code> now uses UTF-8 as its default
 output encoding.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now checks line endings of files with
 extension &lsquo;<span class="file">.hpp</span>&rsquo; and those under &lsquo;<span class="file">inst/include</span>&rsquo;.  The
 check now includes that a non-empty file is terminated with a
 newline.
 </p>
 <p><code>R CMD build</code> will correct line endings in such files.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now tries re-building all vignettes
 rather than stopping at the first error: whilst doing so it adds
 &lsquo;bookmarks&rsquo; to the log.  By default (see the &lsquo;R
 Internals&rsquo; manual) it re-builds each vignette in a separate
 process.
 </p>
 <p>It now checks for duplicated vignette titles (also known as
 &lsquo;index entries&rsquo;): they are used as hyperlinks on
 <acronym><span class="acronym">CRAN</span></acronym> package pages and so do need to be unique.
 </p>
 </li>
 <li> <p><code>R CMD check</code> has more comprehensive checks on the
 &lsquo;<span class="file">data</span>&rsquo; directory and the functioning of <code>data()</code> in a
 package.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now checks
 <code>autoconf</code>-generated &lsquo;<span class="file">configure</span>&rsquo; files have their
 corresponding source files, including optionally attempting to
 regenerate them on platforms with <code>autoreconf</code>.
 </p>
 </li>
 <li> <p><code>R CMD build</code> has a new option
 <span class="option">--compression</span> to select the compression used for the tarball.
 </p>
 </li>
 <li> <p><code>R CMD build</code> now removes &lsquo;<span class="file">src/*.mod</span>&rsquo; files on
 all platforms.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> New pointer protection C functions <code>R_PreserveInMSet</code> and
 <code>R_ReleaseFromMSet</code> have been introduced to replace
 <code>UNPROTECT_PTR</code>, which is not safe to mix with <code>UNPROTECT</code>
 (and with <code>PROTECT_WITH_INDEX</code>). Intended for use in parsers
 only.
 </p>
 </li>
 <li> <p><code>NAMEDMAX</code> has been raised to 7 to allow further
 protection of intermediate results from (usually ill-advised)
 assignments in arguments to <code>BUILTIN</code> functions. Properly
 written package code should not be affected.
 </p>
 </li>
 <li> <p><code>R_unif_index</code> is now considered to be part of the C API.
 </p>
 </li>
 <li> <p><code>R_GetCurrentEnv()</code> allows C code to retrieve the current
 environment.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> Argument <code>compressed</code> of <code>untar()</code> is
 deprecated &mdash; it is only used for external <code>tar</code> commands
 which increasingly for extraction auto-detect compression and
 ignore their <span class="option">zjJ</span> flags.
 </p>
 </li>
 <li> <p><code>var(f)</code> and hence <code>sd(f)</code> now give an error for
 <code>factor</code> arguments; they gave a deprecation warning since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.2.3, <a href="https://bugs.R-project.org/show_bug.cgi?id=16564">PR#16564</a>.
 </p>
 </li>
 <li><p> Package <span class="pkg">tools</span>' <code>vignetteDepends()</code> has been
 deprecated (it called a function deprecated since Feb 2016), being
 partly replaced by newly exported <code>vignetteInfo()</code>.
 </p>
 </li>
 <li><p> The <code>f77_f2c</code> script has been removed: it
 no longer sufficed to compile the &lsquo;<span class="file">.f</span>&rsquo; files in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li><p> The deprecated legacy support of <code>make</code> macros such
 as <span class="samp">CXX1X</span> has been removed: use the <span class="samp">CXX11</span> forms
 instead.
 </p>
 </li>
 <li><p> Make macro <span class="samp">F77_VISIBILITY</span> is deprecated in favour of
 <span class="samp">F_VISIBILITY</span>.
 </p>
 </li>
 <li><p> Make macros <span class="samp">F77</span>, <span class="samp">FCPIFCPLAGS</span> and
 <span class="samp">SHLIB_OPENMP_FCFLAGS</span> are deprecated in favour of <span class="samp">FC</span>,
 <span class="samp">FPICFLAGS</span> and <span class="samp">SHLIB_OPENMP_FFLAGS</span> respectively.
 </p>
 </li>
 <li> <p><code>$.data.frame</code> had become an expensive version of
 the default method, so has been removed. (Thanks to Radford Neal
 for picking this up and to Duncan Murdoch for providing a patch.)
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>replayPlot(r)</code> now also works in the same <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session
 when <code>r</code> has been &ldquo;reproduced&rdquo; from serialization,
 typically after saving to and reading from an RDS file.
 </p>
 </li>
 <li> <p><code>substr()</code> and <code>substring()</code> now signal an error
 when the input is invalid UTF-8.
 </p>
 </li>
 <li> <p><code>file.copy()</code> now works also when its argument <code>to</code>
 is of length greater than one.
 </p>
 </li>
 <li> <p><code>mantelhaen.test()</code> no longer suffers from integer
 overflow in largish cases, thanks to Ben Bolker's <a href="https://bugs.R-project.org/show_bug.cgi?id=17383">PR#17383</a>.
 </p>
 </li>
 <li><p> Calling <code>setGeneric("foo")</code> in a package no longer
 fails when the enclosing environment of the implicit generic
 <code>foo()</code> is <code>.GlobalEnv</code>.
 </p>
 </li>
 <li> <p><code>untar(file("&lt;some&gt;.tar.gz"), *)</code> now gives a better
 error message, suggesting to use <code>gzfile()</code> instead.
 </p>
 </li>
 <li><p> Method dispatch uses more relevant environments when looking
 up class definitions.
 </p>
 </li>
 <li><p> The documentation for <code>identify()</code> incorrectly claimed
 that the indices of identified points were returned in the order
 that the points were selected.  <code>identify()</code> now has a new
 argument <code>order</code> to allow the return value to include the
 order in which points were identified; the documentation has been
 updated.  Reported by Richard Rowe and Samuel Granjeaud.
 </p>
 </li>
 <li> <p><code>order(...., decreasing=c(TRUE, FALSE))</code> could fail in
 some cases. Reported from StackOverflow via Karl Nordström.
 </p>
 </li>
 <li><p> User macros in Rd files now accept empty and multi-line
 arguments.
 </p>
 </li>
 <li><p> Changes in <code>print.*()</code>, thanks to Lionel Henry's
 patches in <a href="https://bugs.R-project.org/show_bug.cgi?id=17398">PR#17398</a>:
 </p>

 <ul>
 <li><p> Printing lists, pairlists or attributes containing calls
 with S3 class no longer evaluate those.
 </p>
 </li>
 <li><p> Printing S4 objects within lists and pairlists dispatches
 with <code>show()</code> rather than <code>print()</code>, as with
 auto-printing.
 </p>
 </li>
 <li><p> The indexing tags (names or <code>[[&lt;n&gt;]]</code>) of recursive
 data structures are now printed correctly in complex cases.
 </p>
 </li>
 <li><p> Arguments supplied to <code>print()</code> are now properly
 forwarded to methods when printing lists, pairlists or
 attributes containing S3 objects.
 </p>
 </li>
 <li><p> The print parameters are now preserved when printing S3
 objects or deparsing symbols and calls.  Previously, printing
 lists containing S3 objects or expressions would reset these
 parameters.
 </p>
 </li>
 <li><p> Printing lists, pairlists or attributes containing
 functions now uses <code>srcref</code> attributes if present.
 </p>
 </li></ul>

 </li>
 <li><p> Calling <code>install.packages()</code> with a length zero
 <code>pkgs</code> argument now is a no-op (<a href="https://bugs.R-project.org/show_bug.cgi?id=17422">PR#17422</a>).
 </p>
 </li>
 <li> <p><code>unlist(x)</code> now returns a correct <code>factor</code> when
 <code>x</code> is a nested list with factor leaves, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=12572">PR#12572</a>
 and <a href="https://bugs.R-project.org/show_bug.cgi?id=17419">PR#17419</a>.
 </p>
 </li>
 <li><p> The documentation <code>help(family)</code> gives more details
 about the <code>aic</code> component, thanks to Ben Bolker's prompting.
 </p>
 </li>
 <li><p> The documentation for <code>attributes</code> and
 <code>`attributes&lt;-`</code> now gives <code>x</code> as name of the first and
 main argument which the implementation has been requiring, fixing
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17434">PR#17434</a>.  For consistency, the first argument name is also
 changed from <code>obj</code> to <code>x</code> for <code>`mostattributes&lt;-`</code>.
 </p>
 </li>
 <li> <p><code>strwidth()</code> now uses <code>par("font")</code> as default
 font face (<a href="https://bugs.R-project.org/show_bug.cgi?id=17352">PR#17352</a>).
 </p>
 </li>
 <li> <p><code>plot(&lt;table&gt;, log="x")</code> no longer warns about <code>log</code>.
 </p>
 </li>
 <li><p> The <code>print()</code> method for <code>"htest"</code> objects now
 formats the test statistic and parameter directly and hence no
 longer rounds to units <em>before</em> the decimal point.  Consequently,
 printing of <code>t.test()</code> results with a small number of digits
 now shows non-large <code>df</code>'s to the full precision (<a href="https://bugs.R-project.org/show_bug.cgi?id=17444">PR#17444</a>).
 </p>
 </li>
 <li> <p><code>kruskal.test()</code> and <code>fligner.test()</code> no longer
 erroneously insist on numeric <code>g</code> group arguments (<a href="https://bugs.R-project.org/show_bug.cgi?id=16719">PR#16719</a>).
 </p>
 </li>
 <li><p> Printing a news db via the browser now does a much better
 job (<a href="https://bugs.R-project.org/show_bug.cgi?id=17433">PR#17433</a>).
 </p>
 </li>
 <li> <p><code>print.aov()</code> missed column names in the multivariate case
 due to misspelling (reported by Chris Andrews).
 </p>
 </li>
 <li> <p><code>axis()</code> now creates valid <code>at</code> locations also for
 small subnormal number ranges in log scale plots.
 </p>
 </li>
 <li> <p><code>format.POSIXlt()</code> now also recycles the <code>zone</code>
 and <code>gmtoff</code> list components to full length when needed,
 and its internal C code detects <code>have_zone</code> in more cases.
 In some cases, this changes its output to become compatible with
 <code>format.POSIXct()</code>.
 </p>
 </li>
 <li><p> On Windows, <code>detectCores()</code> in package <span class="pkg">parallel</span>
 now detects processors in all processor groups, not just the group
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is running in (impacts particularly systems with more than 64
 logical processors).  Reported by Arunkumar Srinivasan.
 </p>
 </li>
 <li><p> On Windows, <code>socketSelect()</code> would hang with more than
 64 sockets, and hence <code>parallel::clusterApplyLB()</code> would hang
 with more than 64 workers.  Reported by Arunkumar Srinivasan.
 </p>
 </li>
 <li> <p><code>as(1L, "double")</code> now does coerce (<a href="https://bugs.R-project.org/show_bug.cgi?id=17457">PR#17457</a>).
 </p>
 </li>
 <li> <p><code>lm.influence()</code>, <code>influence.measures()</code>,
 <code>rstudent()</code> etc now work (more) correctly for multivariate
 models (<code>"mlm"</code>), thanks to (anonymous) stackoverflow remarks.
 </p>
 </li>
 <li> <p><code>sample.int(2.9, *, replace=TRUE)</code> again behaves as
 documented and as in R &lt; 3.0.0, namely identically to
 <code>sample.int(2, ..)</code>.
 </p>
 </li>
 <li><p> Fixes to <code>convertColor()</code> for chromatic adaptation;
 thanks to Brodie Gaslam <a href="https://bugs.R-project.org/show_bug.cgi?id=17473">PR#17473</a>.
 </p>
 </li>
 <li><p> Using <code>\Sexpr[stage=install]{..}</code> to create an
 &lsquo;<span class="file">Rd</span>&rsquo; section no longer gives a warning in <code>R CMD
 	check</code>; problem originally posted by Gábor Csárdi, then reported
 as <a href="https://bugs.R-project.org/show_bug.cgi?id=17479">PR#17479</a> with a partial patch by Duncan Murdoch.
 </p>
 </li>
 <li><p> Parse data now include a special node for equal assignment.
 </p>
 </li>
 <li> <p><code>split.default()</code> no longer relies on <code>[[&lt;-()</code>, so
 it behaves as expected when splitting an object by a factor with
 the empty string as one of its levels.  Thanks to Brad Friedman for
 the report.
 </p>
 </li>
 <li><p> Line numbers in messages about &lsquo;<span class="file">.Rd</span>&rsquo; files are now more
 reliable, thanks to a patch from Duncan Murdoch.
 </p>
 </li>
 <li><p> In the <code>numeric</code> method for <code>all.equal()</code>, a
 numeric <code>scale</code> argument is now checked to be positive and
 allowed to be of length &gt; 1. (The latter worked originally and
 with a warning in recent years).
 </p>
 </li>
 <li><p> Deferred string conversions now record the <code>OutDec</code>
 option setting when not equal to the default.  Reported by Michael
 Sannella.
 </p>
 </li>
 <li><p> When <code>y</code> is numeric and <code>f</code> a <code>factor</code>,
 <code>plot(y ~ f)</code> nicely uses &quot;y&quot; and &quot;f&quot; as y- and x-labels.
 The more direct <code>boxplot(y ~ f)</code> now does too.  The new
 argument <code>ann = FALSE</code> may be used to suppress these.
 </p>
 </li>
 <li><p> Subassignment to no/empty rows of a data frame is more
 consistent and typically a no-op in all cases instead of sometimes
 an error; part of Emil Bode's <a href="https://bugs.R-project.org/show_bug.cgi?id=17483">PR#17483</a>.
 </p>
 </li>
 <li><p> Calls like <code>formatC(*, zero.print = "&lt; 0.001")</code> no
 longer give an error and are further improved via new optional
 argument <code>replace.zero</code>.  Reported by David Hugh-Jones.
 </p>
 </li>
 <li> <p><code>methods::formalArgs("&lt;fn&gt;")</code> now finds the same
 function as <code>formals("&lt;fn&gt;")</code>, fixing Emil Bode's <a href="https://bugs.R-project.org/show_bug.cgi?id=17499">PR#17499</a>.
 </p>
 </li>
 <li><p> The <span class="pkg">methods</span> package better handles duplicated class
 names across packages.
 </p>
 </li>
 <li><p> The default method of <code>seq()</code> now avoids integer
 overflow, thanks to the report and &quot;cumsum&quot; patch of Suharto
 Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17497">PR#17497</a>.
 </p>
 </li>
 <li> <p><code>sub()</code> no longer loses encodings for non-ASCII
 replacements (<a href="https://bugs.R-project.org/show_bug.cgi?id=17509">PR#17509</a>).
 </p>
 </li>
 <li><p> Fix for rotated raster image on X11 device.  (Partial
 fix for <a href="https://bugs.R-project.org/show_bug.cgi?id=17148">PR#17148</a>;  thanks to Mikko Korpela).
 </p>
 </li>
 <li> <p><code>formula(model.frame(frml, ..))</code> now returns <code>frml</code>
 in all cases, thanks to Bill Dunlap.   The previous behavior is
 available as <code>DF2formula(&lt;model.frame&gt;)</code>.
 </p>
 </li>
 <li> <p><code>ar.ols()</code> also returns scalar <code>var.pred</code> in
 univariate case (<a href="https://bugs.R-project.org/show_bug.cgi?id=17517">PR#17517</a>).
 </p>
 </li>
 <li> <p><code>normalizePath()</code> now treats <code>NA</code> path as non-existent
 and normalizes it to <code>NA</code>.  <code>file.access()</code> treats <code>NA</code>
 file name as non-existent.  <code>file.edit()</code> and connection
 functions such as <code>file()</code> now treat <code>NA</code> file names as
 errors.
 </p>
 </li>
 <li><p> The internal <code>regularize.values()</code> auxiliary of
 <code>approx()</code>, <code>splinefun()</code> etc now warns again when there
 are ties and the caller did not specify <code>ties</code>.  Further, it no
 longer duplicates <code>x</code> and <code>y</code> unnecessarily when <code>x</code>
 is already sorted (<a href="https://bugs.R-project.org/show_bug.cgi?id=17515">PR#17515</a>).
 </p>
 </li>
 <li> <p><code>strtoi("", base)</code> now gives <code>NA</code> on all platforms,
 following its documentation.  Reported by Michael Chirico.
 </p>
 </li>
 <li><p> In the definition of an S4 class, prototype elements are
 checked against the slots of the class, with giving a prototype
 for an undefined slot now being an error.  (Reported by Bill
 Dunlap.)
 </p>
 </li>
 <li><p> From <code>setClassUnion()</code>, if environment variable
 <span class="env">_R_METHODS_SHOW_CHECKSUBCLASSES</span> is set
 to true, the internal <code>.checkSubclasses()</code> utility prints
 debugging info to see where it is used.
 </p>
 </li>
 <li> <p><code>max.col(m)</code> with an <code>m</code> of zero columns now
 returns integer <code>NA</code> (instead of <code>1</code>).
 </p>
 </li>
 <li> <p><code>axTicks()</code> no longer returns small &ldquo;almost zero&rdquo;
 numbers (in exponential format) instead of zero, fixing Ilario
 Gelmetti's <a href="https://bugs.R-project.org/show_bug.cgi?id=17534">PR#17534</a>.
 </p>
 </li>
 <li> <p><code>isSymmetric(matrix(0, dimnames=list("A","b")))</code> is
 <code>FALSE</code> again, as always documented.
 </p>
 </li>
 <li><p> The <code>cairo_pdf</code> graphics device (and other Cairo-based
 devices) now clip correctly to the right and bottom border.
 </p>
 <p>There was an off-by-one-pixel bug, reported by Lee Kelvin.
 </p>
 </li>
 <li> <p><code>as.roman(3) &lt;= 2:4</code> and all other comparisons now work,
 as do group <code>"Summary"</code> function calls such as
 <code>max(as.roman(sample(20)))</code> and <code>as.roman(NA)</code>.
 (Partly reported by Bill Dunlap in <a href="https://bugs.R-project.org/show_bug.cgi?id=17542">PR#17542</a>.)
 </p>
 </li>
 <li> <p><code>reformulate("x", response = "sin(y)")</code> no longer produces
 extra back quotes, <a href="https://bugs.R-project.org/show_bug.cgi?id=17359">PR#17359</a>, and gains new optional argument
 <code>env</code>.
 </p>
 </li>
 <li><p> When reading console input from &lsquo;<span class="file">stdin</span>&rsquo; with re-encoding
 (<code>R --encoding=enc &lt; input</code>) the code on a Unix-alike now
 ensures that each converted input line is terminated with a
 newline even if re-encoding fails.
 </p>
 </li>
 <li> <p><code>as.matrix.data.frame()</code> now produces better strings from
 logicals, thanks to <a href="https://bugs.R-project.org/show_bug.cgi?id=17548">PR#17548</a> from Gabe Becker.
 </p>
 </li>
 <li><p> The S4 generic signature of <code>rowSums()</code>,
 <code>rowMeans()</code>, <code>colSums()</code> and <code>colMeans()</code> is
 restricted to <code>"x"</code>.
 </p>
 </li>
 <li> <p><code>match(x, tab)</code> now works for long <em>character</em>
 vectors <code>x</code>, thanks to <a href="https://bugs.R-project.org/show_bug.cgi?id=17552">PR#17552</a> by Andreas Kersting.
 </p>
 </li>
 <li><p> Class unions are unloaded when their namespace is
 unloaded (<a href="https://bugs.R-project.org/show_bug.cgi?id=17531">PR#17531</a>, adapted from a patch by Brodie Gaslam).
 </p>
 </li>
 <li> <p><code>selectMethod()</code> is robust to ANY-truncation of method
 signatures (thanks to Herve Pages for the report).
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.5.3</h3>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> Detection of flags for C++98/11/14/17 has been improved: in
 particular if CXX??STD is set, it is tried first with no
 additional flags.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> New macro <span class="samp">F_VISIBILITY</span> as an alternative to
 <span class="samp">F77_VISIBILITY</span>.  This will become the preferred form in
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.6.0.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>writeLines(readLines(fnam), fnam)</code> now works as expected,
 thanks to Peter Meissner's <a href="https://bugs.R-project.org/show_bug.cgi?id=17528">PR#17528</a>.
 </p>
 </li>
 <li> <p><code>setClassUnion()</code> no longer warns, but uses <code>message()</code>
 for now, when encountering &ldquo;non local&rdquo; subclasses of class members.
 </p>
 </li>
 <li> <p><code>stopifnot(exprs = T)</code> no longer fails.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.5.2</h3>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> New macro <span class="samp">CXX_VISIBILITY</span> analogous to
 <span class="samp">C_VISIBILITY</span> (which several packages have been misusing for
 C++ code) for the default C++ compiler (but not necessarily one used
 for non-default C++ dialects like C++14).
 </p>
 </li></ul>


 <h4>TESTING</h4>


 <ul>
 <li><p> The random number generator tests in
 &lsquo;<span class="file">tests/p-r-random-tests.R</span>&rsquo; no longer fail occasionally as
 they now randomly sample from &ldquo;certified&rdquo; random seeds.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> The <code>"glm"</code> method of <code>drop1()</code> miscalculated the
 score test (<code>test="Rao"</code>) when the model contained an offset.
 </p>
 </li>
 <li><p> Linear multiple empty models such as <code>lm(y ~ 0)</code>
 now have a correctly dimensioned empty coefficient matrix;
 reported by Brett Presnell.
 </p>
 </li>
 <li> <p><code>vcov(&lt;empty mlm&gt;)</code> and hence <code>confint()</code> now
 work (via a consistency change in <code>summary.lm()</code>).
 </p>
 </li>
 <li> <p><code>confint(&lt;multiple lm()&gt;)</code> now works correctly;
 reported on R-devel by Steven Pav.
 </p>
 </li>
 <li> <p><code>quade.test()</code> now also works correctly when its
 arguments are not yet sorted along <code>groups</code>, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=15842">PR#15842</a>.
 </p>
 </li>
 <li><p> Installation on a Unix-alike tries harder to link to the
 <span class="samp">pthread</span> library where required (rather than relying on
 OpenMP to provide it: configuring with <span class="option">--disable-openmp</span>
 was failing on some Linux systems).
 </p>
 </li>
 <li><p> The <code>data.frame</code> method for <code>print(x)</code> is fast now
 also for large data frames <code>x</code> and got an optional argument
 <code>max</code>, thanks to suggestions by Juan Telleria.
 </p>
 </li>
 <li> <p><code>hist()</code> no longer integer overflows in very rare
 cases, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17450">PR#17450</a>.
 </p>
 </li>
 <li> <p><code>untar()</code> ignored a character <code>compressed</code>
 argument: however many external <code>tar</code> programs ignore
 the flags which should have been set and automagically choose the
 compression type, and if appropriate <code>gzip</code> or
 <code>bzip2</code> compression would have been chosen from the magic
 header of the tarball.
 </p>
 </li>
 <li> <p><code>zapsmall(x)</code> now works for more &ldquo;number-like&rdquo; objects.
 </p>
 </li>
 <li><p> The tools-internal function called from <code>R CMD
 	INSTALL</code> now gets a <code>warnOption = 1</code> argument and only sets
 <code>options(warn = warnOption)</code> when that increases the warning
 level (<a href="https://bugs.R-project.org/show_bug.cgi?id=17453">PR#17453</a>).
 </p>
 </li>
 <li><p> Analogously, the tools-internal function called from <code>R CMD
 	check</code> gets a <code>warnOption = 1</code> argument and uses the larger
 of that and <code>getOption("warn")</code>, also allowing to be run with
 increased warning level.
 </p>
 </li>
 <li><p> Parse data now have deterministic parent nodes (<a href="https://bugs.R-project.org/show_bug.cgi?id=16041">PR#16041</a>).
 </p>
 </li>
 <li><p> Calling <code>match()</code> with length one <code>x</code> and POSIXlt
 <code>table</code> gave a segfault (<a href="https://bugs.R-project.org/show_bug.cgi?id=17459">PR#17459</a>).
 </p>
 </li>
 <li><p> Fork clusters could hang due to a race condition in cluster
 initialization (<code>makeCluster()</code>).
 </p>
 </li>
 <li> <p><code>nextn(n)</code> now also works for larger <code>n</code> and no
 longer loops infinitely for e.g, <code>n &lt;- 214e7</code>.
 </p>
 </li>
 <li> <p><code>cooks.distance()</code> and <code>rstandard()</code> now work
 correctly for multiple linear models (<code>"mlm"</code>).
 </p>
 </li>
 <li> <p><code>polym()</code> and corresponding <code>lm()</code> prediction now
 also work for a boundary &quot;vector&quot; case fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17474">PR#17474</a>,
 reported by Alexandre Courtiol.
 </p>
 </li>
 <li><p> With a very large number of variables <code>terms()</code> could
 segfault (<a href="https://bugs.R-project.org/show_bug.cgi?id=17480">PR#17480</a>).
 </p>
 </li>
 <li> <p><code>cut(rep(0, 7))</code> now works, thanks to Joey Reid and
 Benjamin Tyner (<a href="https://bugs.R-project.org/show_bug.cgi?id=16802">PR#16802</a>).
 </p>
 </li>
 <li> <p><code>download.file(*, method = "curl", cacheOK = FALSE)</code>
 should work now on Windows, thanks to Kevin Ushey's patch in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17323">PR#17323</a>.
 </p>
 </li>
 <li> <p><code>duplicated(&lt;dataframe with 'f'&gt;)</code> now works, too,
 thanks to Andreas Kersting's <a href="https://bugs.R-project.org/show_bug.cgi?id=17485">PR#17485</a>; ditto for
 <code>anyDuplicated()</code>.
 </p>
 </li>
 <li> <p><code>legend(*, cex = 1:2)</code> now works less badly.
 </p>
 </li>
 <li><p> The <code>print()</code> method for <code>POSIXct</code> and
 <code>POSIXlt</code> now correctly obeys <code>getOption("max.print")</code>,
 fixing a long-standing typo, and it also gets a corresponding
 optional <code>max</code> argument.
 </p>
 </li>
 <li><p> Unserialization of raw vectors serialized in ASCII
 representation now works correctly.
 </p>
 </li>
 <li> <p><code>&lt;data frame&gt;[TRUE, &lt;new&gt;] &lt;- list(c1, c2)</code> now works
 correctly, thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=15362">PR#15362</a> and Emil Bode's
 patch in <a href="https://bugs.R-project.org/show_bug.cgi?id=17504">PR#17504</a>.
 </p>
 </li>
 <li> <p><code>seq.int(*, by=by, length=n)</code> no longer wrongly
 &ldquo;drops fractional parts&rdquo; when <code>by</code> is integer, thanks
 to Suharto Anggono's report <a href="https://bugs.R-project.org/show_bug.cgi?id=17506">PR#17506</a>.
 </p>
 </li>
 <li><p> Buffering is disabled for <code>file()</code> connections to
 non-regular files (like sockets), as well as <code>fifo()</code> and
 <code>pipe()</code> connections.  Fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=17470">PR#17470</a>, reported by Chris
 Culnane.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.5.1</h3>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>file("stdin")</code> is no longer considered seekable.
 </p>
 </li>
 <li> <p><code>dput()</code> and <code>dump()</code> are no longer truncating
 when <code>options(deparse.max.lines = *)</code> is set.
 </p>
 </li>
 <li><p> Calls with an S3 class are no longer evaluated when printed,
 fixing part of <a href="https://bugs.R-project.org/show_bug.cgi?id=17398">PR#17398</a>, thanks to a patch from Lionel Henry.
 </p>
 </li>
 <li><p> Allow <code>file</code> argument of <code>Rscript</code> to include space
 even when it is first on the command line.
 </p>
 </li>
 <li> <p><code>callNextMethod()</code> uses the generic from the
 environment of the calling method. Reported by Hervé Pagès with
 well documented examples.
 </p>
 </li>
 <li><p> Compressed file connections are marked as blocking.
 </p>
 </li>
 <li> <p><code>optim(*, lower = c(-Inf, -Inf))</code> no longer warns (and
 switches the method), thanks to a suggestion by John Nash.
 </p>
 </li>
 <li> <p><code>predict(fm, newdata)</code> is now correct also for models
 where the formula has terms such as <code>splines::ns(..)</code> or
 <code>stats::poly(..)</code>, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17414">PR#17414</a>, based on a patch from
 Duncan Murdoch.
 </p>
 </li>
 <li> <p><code>simulate.lm(glm(*, gaussian(link = &lt;non-default&gt;)))</code>
 has been corrected, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17415">PR#17415</a> thanks to Alex Courtiol.
 </p>
 </li>
 <li> <p><code>unlist(x)</code> no longer fails in some cases of nested
 empty lists.  Reported by Steven Nydick.
 </p>
 </li>
 <li> <p><code>qr.coef(qr(&lt;all 0, w/ colnames&gt;))</code> now works.
 Reported by Kun Ren.
 </p>
 </li>
 <li><p> The radix sort is robust to vectors with &gt;1 billion elements
 (but long vectors are still unsupported). Thanks to Matt Dowle for
 the fix.
 </p>
 </li>
 <li><p> Terminal connections (e.g., stdin) are no longer
 buffered. Fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=17432">PR#17432</a>.
 </p>
 </li>
 <li> <p><code>deparse(x)</code>, <code>dput(x)</code> and <code>dump()</code> now
 respect <code>c()</code>'s argument names <code>recursive</code> and
 <code>use.names</code>, e.g., for <code>x &lt;- setNames(0, "recursive")</code>,
 thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17427">PR#17427</a>.
 </p>
 </li>
 <li><p> Unbuffered connections now work with encoding conversion.
 Reported by Stephen Berman.
 </p>
 </li>
 <li> <p>&lsquo;<span class="file">.Renviron</span>&rsquo; on Windows with <code>Rgui</code> is again by
 default searched for in user documents directory when invoked
 <em>via</em> the launcher icon.  Reported by Jeroen Ooms.
 </p>
 </li>
 <li> <p><code>printCoefmat()</code> now also works with explicit
 <code>right=TRUE</code>.
 </p>
 </li>
 <li> <p><code>print.noquote()</code> now also works with explicit
 <code>quote=FALSE</code>.
 </p>
 </li>
 <li><p> The default method for <code>pairs(.., horInd=*, verInd=*)</code>
 now gets the correct order, thanks to reports by Chris Andrews and
 Gerrit Eichner.  Additionally, when <code>horInd</code> or <code>verInd</code>
 contain only a subset of variables, all the axes are labeled
 correctly now.
 </p>
 </li>
 <li> <p><code>agrep("..|..", .., fixed=FALSE)</code> now matches when it
 should, thanks to a reminder by Andreas Kolter.
 </p>
 </li>
 <li> <p><code>str(ch)</code> now works for more invalid multibyte strings.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.5.0</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li><p> All packages are by default byte-compiled on installation.
 This makes the installed packages larger (usually marginally so)
 and may affect the format of messages and tracebacks (which often
 exclude <code>.Call</code> and similar).
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>factor()</code> now uses <code>order()</code> to sort its levels,
 rather than <code>sort.list()</code>.  This allows <code>factor()</code> to
 support custom vector-like objects if methods for the appropriate
 generics are defined.  It has the side effect of making
 <code>factor()</code> succeed on empty or length-one non-atomic
 vector(-like) types (e.g., <code>"list"</code>), where it failed before.
 </p>
 </li>
 <li> <p><code>diag()</code> gets an optional <code>names</code> argument: this
 may require updates to packages defining S4 methods for it.
 </p>
 </li>
 <li> <p><code>chooseCRANmirror()</code> and <code>chooseBioCmirror()</code> no
 longer have a <code>useHTTPS</code> argument, not needed now all <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 builds support <span class="samp">https://</span> downloads.
 </p>
 </li>
 <li><p> New <code>summary()</code> method for <code>warnings()</code> with a
 (somewhat experimental) <code>print()</code> method.
 </p>
 </li>
 <li><p> (<span class="pkg">methods</span> package.) <code>.self</code> is now automatically
 registered as a global variable when registering a reference class
 method.
 </p>
 </li>
 <li> <p><code>tempdir(check = TRUE)</code> recreates the <code>tempdir()</code>
 directory if it is no longer valid (e.g. because some
 other process has cleaned up the &lsquo;<span class="file">/tmp</span>&rsquo; directory).
 </p>
 </li>
 <li><p> New <code>askYesNo()</code> function and <code>"askYesNo"</code> option
 to ask the user binary response questions in a customizable but
 consistent way.  (Suggestion of <a href="https://bugs.R-project.org/show_bug.cgi?id=17242">PR#17242</a>.)
 </p>
 </li>
 <li><p> New low level utilities <code>...elt(n)</code> and <code>...length()</code>
 for working with <code>...</code> parts inside a function.
 </p>
 </li>
 <li> <p><code>isTRUE()</code> is more tolerant and now true
 in  </p>
 <pre>   x &lt;- rlnorm(99)
    isTRUE(median(x) == quantile(x)["50%"])</pre>
 <p>New function <code>isFALSE()</code> defined analogously to <code>isTRUE()</code>.
 </p>
 </li>
 <li><p> The default symbol table size has been increased from 4119
 to 49157; this may improve the performance of symbol resolution
 when many packages are loaded.  (Suggested by Jim Hester.)
 </p>
 </li>
 <li> <p><code>line()</code> gets a new option <code>iter = 1</code>.
 </p>
 </li>
 <li><p> Reading from connections in text mode is buffered,
 significantly improving the performance of <code>readLines()</code>, as
 well as <code>scan()</code> and <code>read.table()</code>, at least when
 specifying <code>colClasses</code>.
 </p>
 </li>
 <li> <p><code>order()</code> is smarter about picking a default sort
 <code>method</code> when its arguments are objects.
 </p>
 </li>
 <li> <p><code>available.packages()</code> has two new arguments which
 control if the values from the per-session repository cache are
 used (default true, as before) and if so how old cached values can
 be used (default one hour).
 </p>
 <p>These arguments can be passed from <code>install.packages()</code>,
 <code>update.packages()</code> and functions calling that: to enable
 this <code>available.packages()</code>, <code>packageStatus()</code> and
 <code>download.file()</code> gain a <code>...</code> argument.
 </p>
 </li>
 <li> <p><code>packageStatus()</code>'s <code>upgrade()</code> method no longer
 ignores its <code>...</code> argument but passes it to
 <code>install.packages()</code>.
 </p>
 </li>
 <li> <p><code>installed.packages()</code> gains a <code>...</code> argument to
 allow arguments (including <code>noCache</code>) to be passed from
 <code>new.packages()</code>, <code>old.packages()</code>,
 <code>update.packages()</code> and <code>packageStatus()</code>.
 </p>
 </li>
 <li> <p><code>factor(x, levels, labels)</code> now allows duplicated
 <code>labels</code> (not duplicated <code>levels</code>!).  Hence you can map
 different values of <code>x</code> to the same level directly.
 </p>
 </li>
 <li><p> Attempting to use <code>names&lt;-()</code> on an S4 derivative of a
 basic type no longer emits a warning.
 </p>
 </li>
 <li><p> The <code>list</code> method of <code>within()</code> gains an option
 <code>keepAttrs = FALSE</code> for some speed-up.
 </p>
 </li>
 <li> <p><code>system()</code> and <code>system2()</code> now allow the
 specification of a maximum elapsed time (&lsquo;timeout&rsquo;).
 </p>
 </li>
 <li> <p><code>debug()</code> supports debugging of methods on any object
 of S4 class <code>"genericFunction"</code>, including group generics.
 </p>
 </li>
 <li><p> Attempting to increase the length of a variable containing
 <code>NULL</code> using <code>length()&lt;-</code> still has no effect on the
 target variable, but now triggers a warning.
 </p>
 </li>
 <li> <p><code>type.convert()</code> becomes a generic function, with
 additional methods that operate recursively over list and
 <code>data.frame</code> objects. Courtesy of Arni Magnusson
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17269">PR#17269</a>).
 </p>
 </li>
 <li> <p><code>lower.tri(x)</code> and <code>upper.tri(x)</code> only needing
 <code>dim(x)</code> now work via new functions <code>.row()</code> and
 <code>.col()</code>, so no longer call <code>as.matrix()</code> by default in
 order to work efficiently for all kind of matrix-like objects.
 </p>
 </li>
 <li> <p><code>print()</code> methods for <code>"xgettext"</code> and
 <code>"xngettext"</code> now use <code>encodeString()</code> which keeps,
 e.g. <code>"\n"</code>, visible.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=17298">PR#17298</a>.)
 </p>
 </li>
 <li> <p><code>package.skeleton()</code> gains an optional <code>encoding</code>
 argument.
 </p>
 </li>
 <li> <p><code>approx()</code>, <code>spline()</code>, <code>splinefun()</code> and
 <code>approxfun()</code> also work for long vectors.
 </p>

 </li>
 <li> <p><code>deparse()</code> and <code>dump()</code> are more useful for S4
 objects, <code>dput()</code> now using the same internal C code instead
 of its previous imperfect workaround R code.  S4 objects now
 typically deparse perfectly, i.e., can be recreated identically
 from deparsed code.
 </p>
 <p><code>dput()</code>, <code>deparse()</code> and <code>dump()</code> now print the
 <code>names()</code> information only once, using the more readable
 <code>(tag = value)</code> syntax, notably for <code>list()</code>s, i.e.,
 including data frames.
 </p>
 <p>These functions gain a new control option <code>"niceNames"</code> (see
 <code>.deparseOpts()</code>), which when set (as by default) also uses
 the <code>(tag = value)</code> syntax for atomic vectors.  On the other
 hand, without deparse options <code>"showAttributes"</code> and
 <code>"niceNames"</code>, names are no longer shown also for lists.
 <code>as.character(list(  c (one = 1)))</code> now includes the name, as
 <code>as.character(list(list(one = 1)))</code> has always done.
 </p>
 <p><code>m:n</code> now also deparses nicely when <i>m &gt; n</i>.
 </p>
 <p>The <code>"quoteExpressions"</code> option, also part of <code>"all"</code>,
 no longer <code>quote()</code>s formulas as that may not re-parse
 identically.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17378">PR#17378</a>)
 </p>
 </li>
 <li><p> If the option <code>setWidthOnResize</code> is set and <code>TRUE</code>,
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> run in a terminal using a recent <code>readline</code> library will
 set the <code>width</code> option when the terminal is
 resized.  Suggested by Ralf Goertz.
 </p>
 </li>
 <li><p> If multiple <code>on.exit()</code> expressions are set using
 <code>add = TRUE</code> then all expressions will now be run even if one
 signals an error.
 </p>
 </li>
 <li> <p><code>mclapply()</code> gets an option <code>affinity.list</code> which
 allows more efficient execution with heterogeneous processors,
 thanks to Helena Kotthaus.
 </p>
 </li>
 <li><p> The <code>character</code> methods for <code>as.Date()</code> and
 <code>as.POSIXlt()</code> are more flexible <em>via</em> new arguments
 <code>tryFormats</code> and <code>optional</code>: see their help pages.
 </p>
 </li>
 <li> <p><code>on.exit()</code> gains an optional argument <code>after</code>
 with default <code>TRUE</code>. Using <code>after = FALSE</code> with
 <code>add = TRUE</code> adds an exit expression before any
 existing ones. This way the expressions are run in a first-in
 last-out fashion. (From Lionel Henry.)
 </p>
 </li>
 <li><p> On Windows, <code>file.rename()</code> internally retries the
 operation in case of error to attempt to recover from possible
 anti-virus interference.
 </p>
 </li>
 <li><p> Command line completion on <span class="samp">::</span> now also includes
 lazy-loaded data.
 </p>
 </li>
 <li><p> If the <span class="env">TZ</span> environment variable is set when date-time
 functions are first used, it is recorded as the session default
 and so will be used rather than the default deduced from the OS
 if <span class="env">TZ</span> is subsequently unset.
 </p>
 </li>
 <li><p> There is now a <code>[</code> method for class <code>"DLLInfoList"</code>.
 </p>
 </li>
 <li> <p><code>glm()</code> and <code>glm.fit</code> get the same
 <code>singular.ok = TRUE</code> argument that <code>lm()</code> has had
 forever.  As a consequence, in <code>glm(*, method = &lt;your_own&gt;)</code>,
 user specified methods need to accept a <code>singular.ok</code>
 argument as well.
 </p>
 </li>
 <li> <p><code>aspell()</code> gains a filter for Markdown (&lsquo;<span class="file">.md</span>&rsquo; and
 &lsquo;<span class="file">.Rmd</span>&rsquo;) files.
 </p>
 </li>
 <li> <p><code>intToUtf8(multiple = FALSE)</code> gains an argument to
 allow surrogate pairs to be interpreted.
 </p>
 </li>
 <li><p> The maximum number of DLLs that can be loaded into R
 e.g. <em>via</em> <code>dyn.load()</code> has been increased up to 614
 when the OS limit on the number of open files allows.
 </p>
 </li>
 <li> <p><code>Sys.timezone()</code> on a Unix-alike caches the value at
 first use in a session: <em>inter alia</em> this means that setting
 <span class="env">TZ</span> later in the session affects only the <em>current</em> time
 zone and not the <em>system</em> one.
 </p>
 <p><code>Sys.timezone()</code> is now used to find the system timezone to
 pass to the code used when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is configured with
 <span class="option">--with-internal-tzcode</span>.
 </p>
 </li>
 <li><p> When <code>tar()</code> is used with an external command which is
 detected to be GNU <code>tar</code> or libarchive <code>tar</code> (aka
 <code>bsdtar</code>), a different command-line is generated to
 circumvent line-length limits in the shell.
 </p>
 </li>
 <li> <p><code>system(*, intern = FALSE)</code>, <code>system2()</code> (when not
 capturing output), <code>file.edit()</code> and <code>file.show()</code> now
 issue a warning when the external command cannot be executed.
 </p>
 </li>
 <li><p> The &ldquo;default&rdquo; (<code>"lm"</code> etc) methods of
 <code>vcov()</code> have gained new optional argument <code>complete =
       TRUE</code> which makes the <code>vcov()</code> methods more consistent with
 the <code>coef()</code> methods in the case of singular designs.  The
 former (back-compatible) behavior is given by <code>vcov(*,
 	complete = FALSE)</code>.
 </p>
 </li>
 <li> <p><code>coef()</code> methods (for <code>lm</code> etc) also gain a
 <code>complete = TRUE</code> optional argument for consistency with
 <code>vcov()</code>. <br />
 For <code>"aov"</code>, both <code>coef()</code> and <code>vcov()</code> methods
 remain back-compatibly consistent, using the <em>other</em> default,
 <code>complete = FALSE</code>.
 </p>
 </li>
 <li> <p><code>attach(*, pos = 1)</code> is now an error instead of a
 warning.
 </p>
 </li>
 <li><p> New function <code>getDefaultCluster()</code> in package
 <span class="pkg">parallel</span> to get the default cluster set via
 <code>setDefaultCluster()</code>.
 </p>
 </li>
 <li> <p><code>str(x)</code> for atomic objects <code>x</code> now treats both
 cases of <code>is.vector(x)</code> similarly, and hence much less often
 prints <code>"atomic"</code>.  This is a slight non-back-compatible
 change producing typically both more informative and shorter
 output.
 </p>
 </li>
 <li> <p><code>gc()</code> gets new argument <code>full</code>.
 </p>
 </li>
 <li> <p><code>write.dcf()</code> gets optional argument <code>useBytes</code>.
 </p>
 </li>
 <li><p> New, partly experimental <code>packageDate()</code> which tries to
 get a valid <code>"Date"</code> object from a package &lsquo;<span class="file">DESCRIPTION</span>&rsquo;
 file, thanks to suggestions in <a href="https://bugs.R-project.org/show_bug.cgi?id=17324">PR#17324</a>.
 </p>
 </li>
 <li> <p><code>tools::resaveRdaFiles()</code> gains a <code>version</code>
 argument, for use when packages should remain compatible with
 earlier versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li> <p><code>ar.yw(x)</code> and hence by default <code>ar(x)</code> now work
 when <code>x</code> has <code>NA</code>s, mostly thanks to a patch by Pavel
 Krivitsky in <a href="https://bugs.R-project.org/show_bug.cgi?id=17366">PR#17366</a>.   The <code>ar.yw.default()</code>'s AIC
 computations have become more efficient by using
 <code>determinant()</code>.
 </p>
 </li>
 <li><p> New <code>warnErrList()</code> utility (from package <span class="pkg">nlme</span>,
 improved).
 </p>
 </li>
 <li><p> By default the (arbitrary) signs of the loadings from
 <code>princomp()</code> are chosen so the first element is non-negative.
 </p>
 </li>
 <li><p> If <span class="option">--default-packages</span> is not used, then
 <code>Rscript</code> now checks the environment variable
 <span class="env">R_SCRIPT_DEFAULT_PACKAGES</span>. If this is set, then it takes
 precedence over <span class="env">R_DEFAULT_PACKAGES</span>. If default packages are
 not specified on the command line or by one of these environment
 variables, then <code>Rscript</code> now uses the same default
 packages as <code>R</code>. For now, the previous behavior of not
 including <span class="pkg">methods</span> can be restored by setting the environment
 variable <span class="env">R_SCRIPT_LEGACY</span> to <span class="samp">yes</span>.
 </p>
 </li>
 <li><p> When a package is found more than once, the warning from
 <code>find.package(*, verbose=TRUE)</code> lists all library locations.
 </p>
 </li>
 <li><p> POSIXt objects can now also be rounded or truncated to month
 or year.
 </p>
 </li>
 <li> <p><code>stopifnot()</code> can be used alternatively via new argument
 <code>exprs</code> which is nicer and useful when testing several
 expressions in one call.
 </p>
 </li>
 <li><p> The environment variable <span class="env">R_MAX_VSIZE</span> can now be used
 to specify the maximal vector heap size. On macOS, unless specified
 by this environment variable, the maximal vector heap size is set to
 the maximum of 16GB and the available physical memory. This is to
 avoid having the <code>R</code> process killed when macOS over-commits
 memory.
 </p>
 </li>
 <li> <p><code>sum(x)</code> and <code>sum(x1, x2, .., x&lt;N&gt;)</code> with many or
 long logical or integer vectors no longer overflows (and returns
 <code>NA</code> with a warning), but returns <code>double</code> numbers in
 such cases.
 </p>
 </li>
 <li><p> Single components of <code>"POSIXlt"</code> objects can now be
 extracted and replaced via <code>[</code> indexing with 2 indices.
 </p>
 </li>
 <li><p> S3 method lookup now searches the namespace registry after
 the top level environment of the calling environment.
 </p>
 </li>
 <li><p> Arithmetic sequences created by <code>1:n</code>,
 <code>seq_along</code>, and the like now use compact internal
 representations via the <code>ALTREP</code> framework.  Coercing integer
 and numeric vectors to character also now uses the <code>ALTREP</code>
 framework to defer the actual conversion until first use.
 </p>
 </li>
 <li><p> Finalizers are now run with interrupts suspended.
 </p>
 </li>
 <li> <p><code>merge()</code> gains new option <code>no.dups</code> and by
 default suffixes the second of two duplicated column names,
 thanks to a proposal by Scott Ritchie (and Gabe Becker).
 </p>
 </li>
 <li> <p><code>scale.default(x, center, scale)</code> now also allows
 <code>center</code> or <code>scale</code> to be &ldquo;numeric-alike&rdquo;, i.e.,
 such that <code>as.numeric(.)</code> coerces them correctly.  This also
 eliminates a wrong error message in such cases.
 </p>
 </li>
 <li> <p><code>par*apply</code> and <code>par*applyLB</code> gain an optional
 argument <code>chunk.size</code> which allows to specify the granularity of
 scheduling.
 </p>
 </li>
 <li><p> Some <code>as.data.frame()</code> methods, notably the
 <code>matrix</code> one, are now more careful in not accepting
 duplicated or <code>NA</code> row names, and by default produce unique
 non-NA row names.  This is based on new function
 <code>.rowNamesDF(x, make.names = *) &lt;- rNms</code> where the logical
 argument <code>make.names</code> allows to specify <em>how</em> invalid
 row names <code>rNms</code> are handled.  <code>.rowNamesDF()</code> is a
 &ldquo;workaround&rdquo; compatible default.
 </p>
 </li>
 <li><p> R has new serialization format (version 3) which supports
 custom serialization of <code>ALTREP</code> framework objects.  These
 objects can still be serialized in format 2, but less efficiently.
 Serialization format 3 also records the current native encoding of
 unflagged strings and converts them when de-serialized in R
 running under different native encoding.  Format 3 comes with new
 serialization magic numbers (RDA3, RDB3, RDX3).  Format 3 can be
 selected by <code>version = 3</code> in <code>save()</code>,
 <code>serialize()</code> and <code>saveRDS()</code>, but format 2 remains the
 default for all serialization and saving of the workspace.
 Serialized data in format 3 cannot be read by versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> prior
 to version 3.5.0.
 </p>
 </li>
 <li><p> The <code>"Date"</code> and &ldquo;date-time&rdquo; classes
 <code>"POSIXlt"</code> and <code>"POSIXct"</code> now have a working
 <code>length&lt;-()</code> method, as wished in <a href="https://bugs.R-project.org/show_bug.cgi?id=17387">PR#17387</a>.
 </p>
 </li>
 <li> <p><code>optim(*, control = list(warn.1d.NelderMead = FALSE))</code>
 allows to turn off the warning when applying the default
 <code>"Nelder-Mead"</code> method to 1-dimensional problems.
 </p>
 </li>
 <li> <p><code>matplot(.., panel.first = .)</code> etc now work, as
 <code>log</code> becomes explicit argument and <code>...</code> is passed to
 <code>plot()</code> unevaluated, as suggested by Sebastian Meyer in <a href="https://bugs.R-project.org/show_bug.cgi?id=17386">PR#17386</a>.
 </p>
 </li>
 <li><p> Interrupts can be suspended while evaluating an expression
 using <code>suspendInterrupts</code>.  Subexpression can be evaluated
 with interrupts enabled using <code>allowInterrupts</code>.  These
 functions can be used to make sure cleanup handlers cannot be
 interrupted.
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.5.0 includes a framework that allows packages to
 provide alternate representations of basic <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> objects
 (<code>ALTREP</code>).  The framework is still experimental and may
 undergo changes in future <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> releases as more experience is
 gained. For now, documentation is provided in
 <a href="https://svn.r-project.org/R/branches/ALTREP/ALTREP.html">https://svn.r-project.org/R/branches/ALTREP/ALTREP.html</a>.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>install.packages()</code> for source packages now has the
 possibility to set a &lsquo;timeout&rsquo; (elapsed-time limit).  For
 serial installs this uses the <code>timeout</code> argument of
 <code>system2()</code>: for parallel installs it requires the
 <code>timeout</code> utility command from GNU <span class="pkg">coreutils</span>.
 </p>
 </li>
 <li><p> It is now possible to set &lsquo;timeouts&rsquo; (elapsed-time
 limits) for most parts of <code>R CMD check</code> <em>via</em>
 environment variables documented in the &lsquo;R Internals&rsquo;
 manual.
 </p>
 </li>
 <li><p> The &lsquo;BioC extra&rsquo; repository which was dropped from
 Bioconductor 3.6 and later has been removed from
 <code>setRepositories()</code>.  This changes the mapping for 6&ndash;8 used
 by <code>setRepositories(ind=)</code>.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now also applies the settings of
 environment variables <span class="env">_R_CHECK_SUGGESTS_ONLY_</span> and
 <span class="env">_R_CHECK_DEPENDS_ONLY_</span> to the re-building of vignettes.
 </p>
 </li>
 <li> <p><code>R CMD check</code> with environment variable
 <span class="env">_R_CHECK_DEPENDS_ONLY_</span> set to a true value makes
 test-suite-management packages available and (for the time being)
 works around a common omission of <a href="https://CRAN.R-project.org/package=rmarkdown"><span class="pkg">rmarkdown</span></a> from the
 <span class="samp">VignetteBuilder</span> field.
 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> Support for a system Java on macOS has been
 removed &mdash; install a fairly recent Oracle Java (see &lsquo;R
 Installation and Administration&rsquo; §C.3.2).
 </p>
 </li>
 <li> <p><code>configure</code> works harder to set additional flags in
 <span class="samp">SAFE_FFLAGS</span> only where necessary, and to use flags which
 have little or no effect on performance.
 </p>
 <p>In rare circumstances it may be necessary to override the
 setting of <span class="samp">SAFE_FFLAGS</span>.
 </p>
 </li>
 <li><p> C99 functions <code>expm1</code>, <code>hypot</code>, <code>log1p</code> and
 <code>nearbyint</code> are now required.
 </p>
 </li>
 <li> <p><code>configure</code> sets a <span class="option">-std</span> flag for the C++
 compiler for all supported C++ standards (e.g., <span class="option">-std=gnu++11</span>
 for the C++11 compiler).   Previously this was not done in a few
 cases where the default standard passed the tests made
 (e.g. <code>clang 6.0.0</code> for C++11).
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li> <p>&lsquo;Writing R Extensions&rsquo; documents macros
 <code>MAYBE_REFERENCED</code>, <code>MAYBE_SHARED</code> and
 <code>MARK_NOT_MUTABLE</code> that should be used by package <code>C</code>
 code instead <code>NAMED</code> or <code>SET_NAMED</code>.
 </p>
 </li>
 <li><p> The object header layout has been changed to support merging
 the <code>ALTREP</code> branch. This requires re-installing packages
 that use compiled code.
 </p>
 </li>
 <li> <p>&lsquo;Writing R Extensions&rsquo; now documents the
 <code>R_tryCatch</code>, <code>R_tryCatchError</code>, and
 <code>R_UnwindProtect</code> functions.
 </p>
 </li>
 <li> <p><code>NAMEDMAX</code> has been raised to 3 to allow protection of
 intermediate results from (usually ill-advised) assignments in
 arguments to <code>BUILTIN</code> functions. Package <code>C</code> code using
 <code>SET_NAMED</code> may need to be revised.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li> <p><code>Sys.timezone(location = FALSE)</code> is defunct, and is
 ignored (with a warning).
 </p>
 </li>
 <li> <p><code>methods:::bind_activation()</code> is defunct now; it
 typically has been unneeded for years.
 </p>
 <p>The undocumented &lsquo;hidden&rsquo; objects <code>.__H__.cbind</code> and
 <code>.__H__.rbind</code> in package <span class="pkg">base</span> are deprecated (in
 favour of <code>cbind</code> and <code>rbind</code>).
 </p>
 </li>
 <li><p> The declaration of <code>pythag()</code> in &lsquo;<span class="file">Rmath.h</span>&rsquo; has
 been removed &mdash; the entry point has not been provided since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 2.14.0.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>printCoefmat()</code> now also works without column names.
 </p>
 </li>
 <li><p> The S4 methods on <code>Ops()</code> for the <code>"structure"</code> class no
 longer cause infinite recursion when the structure is not an S4 object.
 </p>
 </li>
 <li> <p><code>nlm(f, ..)</code> for the case where <code>f()</code> has a
 <code>"hessian"</code> attribute now computes
 <i>LL' = H + µI</i> correctly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17249">PR#17249</a>).
 </p>
 </li>
 <li><p> An S4 method that &ldquo;rematches&rdquo; to its generic and
 overrides the default value of a generic formal argument to
 <code>NULL</code> no longer drops the argument from its formals.
 </p>
 </li>
 <li> <p><code>Rscript</code> can now accept more than one argument given on
 the <span class="samp">#!</span> line of a script. Previously, one could only pass a
 single argument on the <span class="samp">#!</span> line in Linux.
 </p>
 </li>
 <li><p> Connections are now written correctly with encoding
 <code>"UTF-16LE"</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16737">PR#16737</a>).
 </p>
 </li>
 <li><p> Evaluation of <code>..0</code> now signals an error.  When
 <code>..1</code> is used and <code>...</code> is empty, the error message is
 more appropriate.
 </p>
 </li>
 <li><p> (Windows mainly.)  Unicode code points which require
 surrogate pairs in UTF-16 are now handled.  All systems should
 properly handle surrogate pairs, even those systems that do not
 need to make use of them.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16098">PR#16098</a>)
 </p>
 </li>
 <li> <p><code>stopifnot(e, e2, ...)</code> now evaluates the
 expressions sequentially and in case of an error or warning shows
 the relevant expression instead of the full <code>stopifnot(..)</code>
 call.
 </p>
 </li>
 <li> <p><code>path.expand()</code> on Windows now accepts paths specified as
 UTF-8-encoded character strings even if not representable in the
 current locale.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17120">PR#17120</a>)
 </p>
 </li>
 <li> <p><code>line(x, y)</code> now correctly computes the medians of the
 left and right group's x-values and in all cases reproduces
 straight lines.
 </p>
 </li>
 <li><p> Extending S4 classes with slots corresponding to special
 attributes like <code>dim</code> and <code>dimnames</code> now works.
 </p>
 </li>
 <li><p> Fix for <code>legend()</code> when <code>fill</code> has multiple values
 the first of which is <code>NA</code> (all colours used to default to
 <code>par(fg)</code>).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17288">PR#17288</a>)
 </p>
 </li>
 <li> <p><code>installed.packages()</code> did not remove the cached value
 for a library tree that had been emptied (but would not use the
 old value, just waste time checking it).
 </p>
 </li>
 <li><p> The documentation for <code>installed.packages(noCache = TRUE)</code>
 incorrectly claimed it would refresh the cache.
 </p>
 </li>
 <li> <p><code>aggregate(&lt;data.frame&gt;)</code> no longer uses spurious names
 in some cases.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17283">PR#17283</a>)
 </p>
 </li>
 <li> <p><code>object.size()</code> now also works for long vectors.
 </p>
 </li>
 <li> <p><code>packageDescription()</code> tries harder to solve
 re-encoding issues, notably seen in some Windows locales.  This
 fixes the <code>citation()</code> issue in <a href="https://bugs.R-project.org/show_bug.cgi?id=17291">PR#17291</a>.
 </p>
 </li>
 <li> <p><code>poly(&lt;matrix&gt;, 3)</code> now works, thanks to prompting by
 Marc Schwartz.
 </p>
 </li>
 <li> <p><code>readLines()</code> no longer segfaults on very large files
 with embedded <code>'\0'</code> (aka &lsquo;nul&rsquo;) characters.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17311">PR#17311</a>)
 </p>
 </li>
 <li> <p><code>ns()</code> (package <span class="pkg">splines</span>) now also works for a
 single observation.  <code>interpSpline()</code> gives a more friendly
 error message when the number of points is less than four.
 </p>
 </li>
 <li> <p><code>dist(x, method = "canberra")</code> now uses the correct
 definition; the result may only differ when <code>x</code> contains
 values of differing signs, e.g. not for 0-1 data.
 </p>
 </li>
 <li> <p><code>methods:::cbind()</code> and <code>methods:::rbind()</code> avoid
 deep recursion, thanks to Suharto Anggono via <a href="https://bugs.R-project.org/show_bug.cgi?id=17300">PR#17300</a>.
 </p>
 </li>
 <li><p> Arithmetic with zero-column data frames now works more
 consistently; issue raised by Bill Dunlap.
 </p>
 <p>Arithmetic with data frames gives a data frame for <code>^</code> (which
 previously gave a numeric matrix).
 </p>
 </li>
 <li> <p><code>pretty(x, n)</code> for large <code>n</code> or large
 <code>diff(range(x))</code> now works better (though
 it was never meant for large <code>n</code>); internally it uses the
 same rounding fuzz (1e-10) as <code>seq.default()</code> &mdash; as it did up
 to 2010-02-03 when both were 1e-7.
 </p>
 </li>
 <li><p> Internal C-level <code>R_check_class_and_super()</code> and hence
 <code>R_check_class_etc()</code> now also consider non-direct super
 classes and hence return a match in more cases.  This e.g., fixes
 behaviour of derived classes in package <a href="https://CRAN.R-project.org/package=Matrix"><span class="pkg">Matrix</span></a>.
 </p>
 </li>
 <li><p> Reverted unintended change in behavior of <code>return</code>
 calls in <code>on.exit</code> expressions introduced by stack unwinding
 changes in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0.
 </p>
 </li>
 <li><p> Attributes on symbols are now detected and prevented; attempt
 to add an attribute to a symbol results in an error.
 </p>
 </li>
 <li> <p><code>fisher.test(*, workspace = &lt;n&gt;)</code> now may also increase
 the internal stack size which allows larger problem to be solved,
 fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=1662">PR#1662</a>.
 </p>
 </li>
 <li><p> The <span class="pkg">methods</span> package no longer directly copies slots
 (attributes) into a prototype that is of an &ldquo;abnormal&rdquo;
 (reference) type, like a symbol.
 </p>
 </li>
 <li><p> The <span class="pkg">methods</span> package no longer attempts to call
 <code>length&lt;-()</code> on <code>NULL</code> (during the bootstrap process).
 </p>
 </li>
 <li><p> The <span class="pkg">methods</span> package correctly shows methods when there
 are multiple methods with the same signature for the same generic
 (still not fully supported, but at least the user can see them).
 </p>
 </li>
 <li> <p><code>sys.on.exit()</code> is now always evaluated in the right
 frame. (From Lionel Henry.)
 </p>
 </li>
 <li> <p><code>seq.POSIXt(*, by = "&lt;n&gt; DSTdays")</code> now should work
 correctly in all cases and is faster.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17342">PR#17342</a>)
 </p>
 </li>
 <li> <p><code>.C()</code> when returning a logical vector now always maps
 values other than FALSE and NA to TRUE (as documented).
 </p>
 </li>
 <li><p> Subassignment with zero length vectors now coerces as
 documented (<a href="https://bugs.R-project.org/show_bug.cgi?id=17344">PR#17344</a>).<br />
 Further, <code>x &lt;- numeric(); x[1] &lt;- character()</code> now signals
 an error &lsquo;<code style="white-space: pre;">replacement has length zero</code>&rsquo; (or a
 translation of that) instead of doing nothing.
 </p>
 </li>
 <li><p> (Package <span class="pkg">parallel</span>.)  <code>mclapply()</code>, <code>pvec()</code>
 and <code>mcparallel()</code> (when <code>mccollect()</code> is used to
 collect results) no longer leave zombie processes behind.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL &lt;pkg&gt;</code> now produces the intended error
 message when, e.g., the <code>LazyData</code> field is invalid.
 </p>
 </li>
 <li> <p><code>as.matrix(dd)</code> now works when the data frame <code>dd</code>
 contains a column which is a data frame or matrix, including a
 0-column matrix/d.f. .
 </p>
 </li>
 <li> <p><code>mclapply(X, mc.cores)</code> now follows its documentation
 and calls <code>lapply()</code> in case <code>mc.cores = 1</code> also in the
 case <code>mc.preschedule</code> is false.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17373">PR#17373</a>)
 </p>
 </li>
 <li> <p><code>aggregate(&lt;data.frame&gt;, drop=FALSE)</code> no longer calls
 the function on &lt;empty&gt; parts but sets corresponding results to
 <code>NA</code>.  (Thanks to Suharto Anggono's patches in <a href="https://bugs.R-project.org/show_bug.cgi?id=17280">PR#17280</a>).
 </p>
 </li>
 <li><p> The <code>duplicated()</code> method for data frames is now based
 on the <code>list</code> method (instead of string coercion).  Consequently
 <code>unique()</code> is better distinguishing data frame rows, fixing
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17369">PR#17369</a> and <a href="https://bugs.R-project.org/show_bug.cgi?id=17381">PR#17381</a>.  The methods for matrices and arrays
 are changed accordingly.
 </p>
 </li>
 <li><p> Calling <code>names()</code> on an S4 object derived from
 <code>"environment"</code> behaves (by default) like calling
 <code>names()</code> on an ordinary environment.
 </p>
 </li>
 <li> <p><code>read.table()</code> with a non-default separator now
 supports quotes following a non-whitespace character, matching the
 behavior of <code>scan()</code>.
 </p>
 </li>
 <li> <p><code>parLapplyLB</code> and <code>parSapplyLB</code> have been fixed to do
 load balancing (dynamic scheduling).  This also means that results of
 computations depending on random number generators will now really be
 non-reproducible, as documented.
 </p>
 </li>
 <li><p> Indexing a list using dollar and empty string (<code>l$""</code>)
 returns NULL.
 </p>
 </li>
 <li><p> Using <code> \usage{ data(&lt;name&gt;, package="&lt;pkg&gt;") } </code>
 no longer produces <code>R CMD check</code> warnings.
 </p>
 </li>
 <li> <p><code>match.arg()</code> more carefully chooses the environment
 for constructing default <code>choices</code>, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=17401">PR#17401</a> as
 proposed by Duncan Murdoch.
 </p>
 </li>
 <li><p> Deparsing of consecutive <code>!</code> calls is now consistent
 with deparsing unary <code>-</code> and <code>+</code> calls and creates code
 that can be reparsed exactly; thanks to a patch by Lionel Henry in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17397">PR#17397</a>.  (As a side effect, this uses fewer parentheses in
 some other deparsing involving <code>!</code> calls.)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.4.4</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>Sys.timezone()</code> tries more heuristics on Unix-alikes and so
 is more likely to succeed (especially on Linux).  For the slowest
 method, a warning is given recommending that <span class="env">TZ</span> is set to
 avoid the search.
 </p>
 </li>
 <li><p> The version of LAPACK included in the sources has been
 updated to 3.8.0 (for the routines used by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>, a very minor bug-fix
 change).
 </p>
 </li>
 <li> <p><code>parallel::detectCores(logical = FALSE)</code> is ignored on
 Linux systems, since the information is not available with
 virtualized OSes.
 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li> <p><code>configure</code> will use <code>pkg-config</code> to find the
 flags to link to <code>jpeg</code> if available (as it should be for the
 recently-released <code>jpeg-9c</code> and <code>libjpeg-turbo</code>).  (This
 amends the code added in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0 as the module name in
 <code>jpeg-9c</code> is not what that tested for.)
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li> <p><code>Sys.timezone(location = FALSE)</code> (which was a stop-gap
 measure for Windows long ago) is deprecated.  It no longer returns
 the value of environment variable <span class="env">TZ</span> (usually a location).
 </p>
 </li>
 <li><p> Legacy support of <code>make</code> macros such as <span class="samp">CXX1X</span>
 is formally deprecated: use the <span class="samp">CXX11</span> forms instead.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>power.prop.test()</code> now warns when it cannot solve the
 problem, typically because of impossible constraints. (<a href="https://bugs.R-project.org/show_bug.cgi?id=17345">PR#17345</a>)
 </p>
 </li>
 <li> <p><code>removeSource()</code> no longer erroneously removes
 <code>NULL</code> in certain cases, thanks to Dénes Tóth.
 </p>
 </li>
 <li> <p><code>nls(`NO [mol/l]` ~ f(t))</code> and <code>nls(y ~ a)</code> now
 work.  (Partly from <a href="https://bugs.R-project.org/show_bug.cgi?id=17367">PR#17367</a>)
 </p>
 </li>
 <li> <p><code>R CMD build</code> checks for GNU <code>cp</code> rather
 than assuming Linux has it.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17370">PR#17370</a> says &lsquo;Alpine
 Linux&rsquo; does not.)
 </p>
 </li>
 <li><p> Non-UTF-8 multibyte character handling fixed more
 permanently (<a href="https://bugs.R-project.org/show_bug.cgi?id=16732">PR#16732</a>).
 </p>
 </li>
 <li> <p><code>sum(&lt;large ints&gt;, &lt;stuff&gt;)</code> is more consistent.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17372">PR#17372</a>)
 </p>
 </li>
 <li> <p><code>rf()</code> and <code>rbeta()</code> now also work correctly when
 <code>ncp</code> is not scalar, notably when (partly) <code>NA</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17375">PR#17375</a>)
 </p>
 </li>
 <li> <p><code>is.na(NULL)</code> no longer warns.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16107">PR#16107</a>)
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> now correctly sets C++ compiler flags
 when all source files are in sub-directories of &lsquo;<span class="file">src</span>&rsquo;.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.4.3</h3>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> A workaround has been added for the changes in location of
 time-zone files in macOS 10.13 &lsquo;High Sierra&rsquo; and
 again in 10.13.1, so the default time zone is deduced
 correctly from the system setting when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is configured with
 <span class="option">--with-internal-tzcode</span> (the default on macOS).
 </p>
 </li>
 <li> <p><code>R CMD javareconf</code> has been updated to recognize the
 use of a Java 9 SDK on macOS.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>raw(0) &amp; raw(0)</code> and <code>raw(0) | raw(0)</code>
 again return <code>raw(0)</code> (rather than <code>logical(0)</code>).
 </p>
 </li>
 <li> <p><code>intToUtf8()</code> converts integers corresponding to
 surrogate code points to <code>NA</code> rather than invalid UTF-8, as
 well as values larger than the current Unicode maximum of
 <code>0x10FFFF</code>.  (This aligns with the current RFC3629.)
 </p>
 </li>
 <li><p> Fix calling of methods on S4 generics that dispatch on
 <code>...</code> when the call contains <code>...</code>.
 </p>
 </li>
 <li><p> Following Unicode &lsquo;Corrigendum 9&rsquo;, the UTF-8
 representations of U+FFFE and U+FFFF are now regarded as valid by
 <code>utf8ToInt()</code>.
 </p>
 </li>
 <li> <p><code>range(c(TRUE, NA), finite = TRUE)</code> and similar no longer
 return <code>NA</code>. (Reported by Lukas Stadler.)
 </p>
 </li>
 <li><p> The self starting function <code>attr(SSlogis, "initial")</code>
 now also works when the y values have exact minimum zero and is
 slightly changed in general, behaving symmetrically in the y range.
 </p>
 </li>
 <li><p> The printing of named raw vectors is now formatted nicely as
 for other such atomic vectors, thanks to Lukas Stadler.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.4.2</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> Setting the <code>LC_ALL</code> category in <code>Sys.setlocale()</code>
 invalidates any cached locale-specific day/month names and the
 AM/PM indicator for <code>strptime()</code> (as setting <code>LC_TIME</code>
 has since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.0).
 </p>
 </li>
 <li><p> The version of LAPACK included in the sources has been
 updated to 3.7.1, a bug-fix release.
 </p>
 </li>
 <li><p> The default for <code>tools::write_PACKAGES(rds_compress=)</code>
 has been changed to <code>"xz"</code> to match the compression used by
 <acronym><span class="acronym">CRAN</span></acronym>.
 </p>
 </li>
 <li> <p><code>c()</code> and <code>unlist()</code> are now more efficient in
 constructing the <code>names(.)</code> of their return value, thanks to
 a proposal by Suharto Anggono.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17284">PR#17284</a>)
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> checks for and <code>R CMD build</code>
 corrects CRLF line endings in shell scripts <code>configure</code>
 and <code>cleanup</code> (even on Windows).
 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> The order of selection of OpenMP flags has been changed:
 Oracle Developer Studio 12.5 accepts <span class="option">-fopenmp</span> and
 <span class="option">-xopenmp</span> but only the latter enables OpenMP so it is now
 tried first.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>within(List, rm(x1, x2))</code> works correctly again,
 including when <code>List[["x2"]]</code> is <code>NULL</code>.
 </p>
 </li>
 <li> <p><code>regexec(pattern, text, *)</code> now applies
 <code>as.character(.)</code> to its first two arguments, as documented.
 </p>
 </li>
 <li> <p><code>write.table()</code> and related functions,
 <code>writeLines()</code>, and perhaps other functions writing text to
 connections did not signal errors when the writes failed,
 e.g. due to a disk being full.  Errors will now be
 signalled if detected during the write, warnings if detected when
 the connection is closed.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17243">PR#17243</a>)
 </p>
 </li>
 <li> <p><code>rt()</code> assumed the <code>ncp</code> parameter was a scalar.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17306">PR#17306</a>)
 </p>
 </li>
 <li> <p><code>menu(choices)</code> with more than 10 choices which easily
 fit into one <code>getOption("width")</code>-line no longer erroneously
 repeats choices.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17312">PR#17312</a>)
 </p>
 </li>
 <li> <p><code>length()&lt;-</code> on a pairlist succeeds.
 (<a href="https://stat.ethz.ch/pipermail/r-devel/2017-July/074680.html">https://stat.ethz.ch/pipermail/r-devel/2017-July/074680.html</a>)
 </p>
 </li>
 <li><p> Language objects such as <code>quote(("\n"))</code> or <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 functions are correctly printed again, where <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.4.1 accidentally
 duplicated the backslashes.
 </p>
 </li>
 <li><p> Construction of <code>names()</code> for very large objects in
 <code>c()</code> and <code>unlist()</code> now works, thanks to Suharto
 Anggono's patch proposals in <a href="https://bugs.R-project.org/show_bug.cgi?id=17292">PR#17292</a>.
 </p>
 </li>
 <li><p> Resource leaks (and similar) reported by Steve Grubb fixed.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17314">PR#17314</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=17316">PR#17316</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=17317">PR#17317</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=17318">PR#17318</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=17319">PR#17319</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=17320">PR#17320</a>)
 </p>
 </li>
 <li> <p><code>model.matrix(~1, mf)</code> now gets the row names from
 <code>mf</code> also when they differ from <code>1:nrow(mf)</code>, fixing
 <a href="https://bugs.R-project.org/show_bug.cgi?id=14992">PR#14992</a> thanks to the suggestion by Sebastian Meyer.
 </p>
 </li>
 <li> <p><code>sigma(fm)</code> now takes the correct denominator degrees
 of freedom for a fitted model with <code>NA</code> coefficients.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17313">PR#17313</a>)
 </p>
 </li>
 <li> <p><code>hist(x, "FD")</code> no longer &ldquo;dies&rdquo; with a somewhat
 cryptic error message when <code>x</code> has extreme outliers or
 <code>IQR()</code> zero:  <code>nclass.FD(x)</code> tries harder to find a
 robust bin width <i>h</i> in the latter case, and
 <code>hist.default(*, breaks)</code> now checks and corrects a too large
 <code>breaks</code> number.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17274">PR#17274</a>)
 </p>
 </li>
 <li> <p><code>callNextMethod()</code> works for <code>...</code> methods.
 </p>
 </li>
 <li> <p><code>qr.coef(qd, y)</code> now has correct names also when
 <code>qd</code> is a complex QR or stems from <code>qr(*, LAPACK=TRUE)</code>.
 </p>
 </li>
 <li><p> Setting <code>options(device = *)</code> to an invalid function no
 longer segfaults when plotting is initiated.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15883">PR#15883</a>)
 </p>
 </li>
 <li> <p><code>encodeString(&lt;very large string&gt;)</code> no longer
 segfaults.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15885">PR#15885</a>)
 </p>
 </li>
 <li><p> It is again possible to use <code>configure
 	--enable-maintainer-mode</code> without having installed
 <code>notangle</code> (it was required in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.4.[01]).
 </p>
 </li>
 <li><p> S4 method dispatch on <code>...</code> calls the method by name
 instead of <code>.Method</code> (for consistency with default dispatch),
 and only attempts to pass non-missing arguments from the generic.
 </p>
 </li>
 <li> <p><code>readRDS(textConnection(.))</code> works again.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17325">PR#17325</a>)
 </p>
 </li>
 <li> <p><code>(1:n)[-n]</code> no longer segfaults for <code>n &lt;- 2.2e9</code>
 (on a platform with enough RAM).
 </p>
 </li>
 <li> <p><code>x &lt;- 1:2; tapply(x, list(x, x), function(x) "")[1,2]</code>
 now correctly returns <code>NA</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17333">PR#17333</a>)
 </p>
 </li>
 <li><p> Running of finalizers after explicit GC request moved from
 the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> interface <code>do_gc</code> to the <code>C</code> interface
 <code>R_gc</code>.  This helps with reclaiming inaccessible connections.
 </p>
 </li>
 <li> <p><code>help.search(topic)</code> and <code>??topic</code> matching topics
 in vignettes with multiple file name extensions (e.g.,
 &lsquo;<span class="file">*.md.rsp</span>&rsquo; but not &lsquo;<span class="file">*.Rmd</span>&rsquo;) failed with an error
 when using <code>options(help_type = "html")</code>.
 </p>
 </li>
 <li><p> The X11 device no longer uses the Xlib backing store (<a href="https://bugs.R-project.org/show_bug.cgi?id=16497">PR#16497</a>).
 </p>
 </li>
 <li> <p><code>array(character(), 1)</code> now gives (a 1D array with)
 <code>NA</code> as has been documented for a long time as in the other
 cases of zero-length array initialization and also compatibly with
 <code>matrix(character(), *)</code>.  As mentioned there, this also
 fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=17333">PR#17333</a>.
 </p>
 </li>
 <li> <p><code>splineDesign(.., derivs = 4)</code> no longer segfaults.
 </p>
 </li>
 <li> <p><code>fisher.test(*, hybrid=TRUE)</code> now (again) will use the
 hybrid method when Cochran's conditions are met, fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=16654">PR#16654</a>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.4.1</h3>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> The deprecated support for PCRE versions older than 8.20
 has been removed.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>getParseData()</code> gave incorrect column information when
 code contained multi-byte characters.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17254">PR#17254</a>)
 </p>
 </li>
 <li><p> Asking for help using expressions like <code>?stats::cor()</code>
 did not work.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17250">PR#17250</a>)
 </p>
 </li>
 <li> <p><code>readRDS(url(....))</code> now works.
 </p>
 </li>
 <li> <p><code>R CMD Sweave</code> again returns <span class="samp">status = 0</span> on
 successful completion.
 </p>
 </li>
 <li><p> Vignettes listed in &lsquo;<span class="file">.Rbuildignore</span>&rsquo; were not being
 ignored properly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17246">PR#17246</a>)
 </p>
 </li>
 <li> <p><code>file.mtime()</code> no longer returns NA on Windows when the
 file or directory is being used by another process.  This affected
 <code>installed.packages()</code>, which is now protected against this.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> Windows .zip file obeys <code>--lock</code>
 and <code>--pkglock</code> flags.
 </p>
 </li>
 <li><p> (Windows only)  The <code>choose.files()</code> function could
 return incorrect results when called with <code>multi = FALSE</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17270">PR#17270</a>)
 </p>
 </li>
 <li> <p><code>aggregate(&lt;data.frame&gt;, drop = FALSE)</code> now also works in
 case of near-equal numbers in <code>by</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16918">PR#16918</a>)
 </p>
 </li>
 <li> <p><code>fourfoldplot()</code> could encounter integer overflow when
 calculating the odds ratio. (<a href="https://bugs.R-project.org/show_bug.cgi?id=17286">PR#17286</a>)
 </p>
 </li>
 <li> <p><code>parse()</code> no longer gives spurious warnings when
 extracting srcrefs from a file not encoded in the current locale.
 </p>
 <p>This was seen from <code>R CMD check</code> with &lsquo;<span class="file">inst/doc/*.R</span>&rsquo;
 files, and <code>check</code> has some additional protection for such files.
 </p>
 </li>
 <li> <p><code>print.noquote(x)</code> now always returns its argument
 <code>x</code> (invisibly).
 </p>
 </li>
 <li><p> Non-UTF-8 multibyte character sets were not handled properly
 in source references.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16732">PR#16732</a>)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.4.0</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li><p> (Unix-alike)
 The default methods for <code>download.file()</code> and <code>url()</code>
 now choose <code>"libcurl"</code> except for <span class="samp">file://</span>
 URLs.  There will be small changes in the format and wording of
 messages, including in rare cases if an issue is a warning or an
 error.  For example, when HTTP re-direction occurs, some messages
 refer to the final URL rather than the specified one.
 </p>
 <p>Those who use proxies should check that their settings are
 compatible (see <code>?download.file</code>: the most commonly used
 forms work for both <code>"internal"</code> and <code>"libcurl"</code>).
 </p>
 </li>
 <li> <p><code>table()</code> has been amended to be more internally
 consistent and become back compatible to <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> <i>&lt;=</i> 2.7.2
 again.  Consequently, <code>table(1:2, exclude = NULL)</code> no longer
 contains a zero count for <code>&lt;NA&gt;</code>, but <code>useNA = "always"</code>
 continues to do so.
 </p>
 </li>
 <li> <p><code>summary.default()</code> no longer rounds, but its print
 method does resulting in less extraneous rounding, notably of
 numbers in the ten thousands.
 </p>
 </li>
 <li> <p><code>factor(x, exclude = L)</code> behaves more rationally when
 <code>x</code> or <code>L</code> are character vectors.  Further,
 <code>exclude = &lt;factor&gt;</code> now behaves as documented for long.
 </p>
 </li>
 <li><p> Arithmetic, logic (<code>&amp;</code>, <code>|</code>) and comparison (aka
 &lsquo;relational&rsquo;, e.g., <code>&lt;</code>, <code>==</code>) operations with
 arrays now behave consistently, notably for arrays of length zero.
 </p>
 <p>Arithmetic between length-1 arrays and longer non-arrays had
 silently dropped the array attributes and recycled.  This now
 gives a warning and will signal an error in the future, as it has
 always for logic and comparison operations in these cases (e.g.,
 compare <code>matrix(1,1) + 2:3</code> and <code>matrix(1,1) &lt; 2:3</code>).
 </p>
 </li>
 <li><p> The JIT (&lsquo;Just In Time&rsquo;) byte-code compiler is now
 enabled by default at its level 3. This means functions will be
 compiled on first or second use and top-level loops will be
 compiled and then run.  (Thanks to Tomas Kalibera for extensive
 work to make this possible.)
 </p>
 <p>For now, the compiler will not compile code containing explicit
 calls to <code>browser()</code>: this is to support single stepping from
 the <code>browser()</code> call.
 </p>
 <p>JIT compilation can be disabled for the rest of the session using
 <code>compiler::enableJIT(0)</code> or by setting environment variable
 <span class="env">R_ENABLE_JIT</span> to <code>0</code>.
 </p>
 </li>
 <li> <p><code>xtabs()</code> works more consistently with <code>NA</code>s, also
 in its result no longer setting them to <code>0</code>.  Further, a new
 logical option <code>addNA</code> allows to count <code>NA</code>s where
 appropriate.  Additionally, for the case <code>sparse = TRUE</code>, the
 result's <code>dimnames</code> are identical to the default case's.
 </p>
 </li>
 <li><p> Matrix products now consistently bypass BLAS when the inputs
 have <code>NaN</code>/<code>Inf</code> values. Performance of the check of
 inputs has been improved. Performance when BLAS is used is
 improved for matrix/vector and vector/matrix multiplication (DGEMV
 is now used instead of DGEMM).
 </p>
 <p>One can now choose from alternative matrix product implementations
 <em>via</em> <code>options(matprod = )</code>.  The <code>"internal"</code>
 implementation is not optimized for speed but consistent in
 precision with other summations in R (using <code>long double</code>
 accumulators where available).  <code>"blas"</code> calls BLAS directly
 for best speed, but usually with undefined behavior for inputs
 with <code>NaN</code>/<code>Inf</code>.
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> User errors such as <code>integrate(f, 0:1, 2)</code> are now caught.
 </p>
 </li>
 <li><p> Add <code>signature</code> argument to <code>debug()</code>,
 <code>debugonce()</code>, <code>undebug()</code> and <code>isdebugged()</code> for
 more conveniently debugging S3 and S4 methods.  (Based on a patch by
 Gabe Becker.)
 </p>
 </li>
 <li><p> Add <code>utils::debugcall()</code> and
 <code>utils::undebugcall()</code> for debugging the function that would
 be called by evaluating the given expression. When the call is to
 an S4 generic or standard S3 generic, <code>debugcall()</code> debugs the
 method that would be dispatched. A number of internal utilities
 were added to support this, most notably
 <code>utils::isS3stdGeneric()</code>.  (Based on a patch by Gabe Becker.)
 </p>
 </li>
 <li><p> Add <code>utils::strcapture()</code>. Given a character vector and
 a regular expression containing capture expressions,
 <code>strcapture()</code> will extract the captured tokens into a
 tabular data structure, typically a <code>data.frame</code>.
 </p>
 </li>
 <li> <p><code>str()</code> and <code>strOptions()</code> get a new option
 <code>drop.deparse.attr</code> with improved but <em>changed</em> default
 behaviour for expressions.  For <code>expression</code> objects
 <code>x</code>, <code>str(x)</code> now may remove extraneous white space and
 truncate long lines.
 </p>
 </li>
 <li> <p><code>str(&lt;looooooooong_string&gt;)</code> is no longer very slow;
 inspired by Mikko Korpela's proposal in <a href="https://bugs.R-project.org/show_bug.cgi?id=16527">PR#16527</a>.
 </p>
 </li>
 <li> <p><code>str(x)</code>'s default method is more &ldquo;accurate&rdquo; and
 hence somewhat more generous in displaying character vectors; this
 will occasionally change <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> outputs (and need changes to some
 &lsquo;<span class="file">*.Rout(.save)</span>&rsquo; files).
 <br />
 For a classed integer vector such as <code>x &lt;- xtabs(~ c(1,9,9,9))</code>,
 <code>str(x)</code> now shows both the class and <code>"int"</code>, instead
 of only the latter.
 </p>
 </li>
 <li> <p><code>isSymmetric(m)</code> is much faster for large asymmetric
 matrices <code>m</code> <em>via</em> pre-tests and a new option <code>tol1</code>
 (with which strict back compatibility is possible but not the
 default).
 </p>
 </li>
 <li><p> The result of <code>eigen()</code> now is of class <code>"eigen"</code>
 in the default case when eigenvectors are computed.
 </p>
 </li>
 <li><p> Zero-length date and date-time objects (of classes
 <code>"POSIX[cl]?t"</code>) now <code>print()</code> &ldquo;recognizably&rdquo;.
 </p>
 </li>
 <li> <p><code>xy.coords()</code> and <code>xyz.coords()</code> get a new
 <code>setLab</code> option.
 </p>
 </li>
 <li><p> The <code>method</code> argument of <code>sort.list()</code>,
 <code>order()</code> and <code>sort.int()</code> gains an <code>"auto"</code> option
 (the default) which should behave the same as before when
 <code>method</code> was not supplied.
 </p>
 </li>
 <li> <p><code>stopifnot(E, ..)</code> now reports differences when
 <code>E</code> is a call to <code>all.equal()</code> and that is not true.
 </p>
 </li>
 <li> <p><code>boxplot(&lt;formula&gt;, *)</code> gain optional arguments
 <code>drop</code>, <code>sep</code>, and <code>lex.order</code> to pass to
 <code>split.default()</code> which itself gains an argument
 <code>lex.order</code> to pass to <code>interaction()</code> for more
 flexibility.
 </p>
 </li>
 <li><p> The <code>plot()</code> method for <code>ppr()</code> has enhanced
 default labels (<code>xmin</code> and <code>main</code>).
 </p>
 </li>
 <li> <p><code>sample.int()</code> gains an explicit <code>useHash</code> option
 (with a back compatible default).
 </p>
 </li>
 <li> <p><code>identical()</code> gains an <code>ignore.srcref</code> option
 which drops <code>"srcref"</code> and similar attributes when true (as
 by default).
 </p>
 </li>
 <li> <p><code>diag(x, nrow = n)</code> now preserves <code>typeof(x)</code>, also
 for logical, integer and raw <code>x</code> (and as previously for
 complex and numeric).
 </p>
 </li>
 <li> <p><code>smooth.spline()</code> now allows direct specification of
 <code>lambda</code>, gets a <code>hatvalues()</code> method and keeps
 <code>tol</code> in the result, and optionally parts of the internal
 matrix computations.
 </p>
 </li>
 <li> <p><code>addNA()</code> is faster now, e.g. when applied twice.
 (Part of <a href="https://bugs.R-project.org/show_bug.cgi?id=16895">PR#16895</a>.)
 </p>
 </li>
 <li><p> New option <code>rstandard(&lt;lm&gt;, type = "predicted")</code>
 provides the &ldquo;PRESS&rdquo;&ndash;related leave-one-out
 cross-validation errors for linear models.
 </p>
 </li>
 <li><p> After seven years of deprecation, duplicated factor levels
 now produce a warning when printed and an error in <code>levels&lt;-</code>
 instead of a warning.
 </p>
 </li>
 <li><p> Invalid factors, e.g., with duplicated levels (invalid but
 constructable) now give a warning when printed, <em>via</em> new
 function <code>.valid.factor()</code>.
 </p>
 </li>
 <li> <p><code>sessionInfo()</code> has been updated for Apple's change in
 OS naming as from &lsquo;10.12&rsquo; (&lsquo;macOS Sierra&rsquo; <em>vs</em>
 &lsquo;OS X El Capitan&rsquo;).
 </p>
 <p>Its <code>toLatex()</code> method now includes the <code>running</code>
 component.
 </p>
 </li>
 <li> <p><code>options(interrupt=)</code> can be used to specify a default
 action for user interrupts.  For now, if this option is not set
 and the <code>error</code> option is set, then an unhandled user
 interrupt invokes the <code>error</code> option.  (This may be dropped in
 the future as <code>interrupt</code> conditions are not <code>error</code>
 conditions.)
 </p>
 </li>
 <li><p> In most cases user interrupt handlers will be called with a
 <code>"resume"</code> restart available.  Handlers can invoke this
 restart to resume computation. At the browser prompt the <code>r</code>
 command will invoke a <code>"resume"</code> restart if one is
 available. Some read operations cannot be resumed properly when
 interrupted and do not provide a <code>"resume"</code> restart.
 </p>
 </li>
 <li><p> Radix sort is now chosen by <code>method = "auto"</code> for
 <code>sort.int()</code> for double vectors (and hence used for
 <code>sort()</code> for unclassed double vectors), excluding
 &lsquo;long&rsquo; vectors.
 </p>
 <p><code>sort.int(method = "radix")</code> no longer rounds double vectors.
 </p>
 </li>
 <li><p> The <code>default</code> and <code>data.frame</code> methods for
 <code>stack()</code> preserve the names of empty elements in the levels
 of the <code>ind</code> column of the return value.  Set the new
 <code>drop</code> argument to <code>TRUE</code> for the previous behavior.
 </p>
 </li>
 <li><p> Speedup in <code>simplify2array()</code> and hence <code>sapply()</code> and
 <code>mapply()</code> (for the case of names and common length &gt; 1),
 thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17118">PR#17118</a>.
 </p>
 </li>
 <li> <p><code>table(x, exclude = NULL)</code> now sets
 <code>useNA = "ifany"</code> (instead of <code>"always"</code>).
 Together with the bug fixes for this case, this recovers more
 consistent behaviour  compatible to older versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  As
 a consequence, <code>summary()</code> for a logical vector no longer
 reports (zero) counts for <code>NA</code> when there are no <code>NA</code>s.
 </p>
 </li>
 <li> <p><code>dump.frames()</code> gets a new option
 <code>include.GlobalEnv</code> which allows to also dump the global
 environment, thanks to Andreas Kersting's proposal in <a href="https://bugs.R-project.org/show_bug.cgi?id=17116">PR#17116</a>.
 </p>
 </li>
 <li> <p><code>system.time()</code> now uses <code>message()</code> instead of
 <code>cat()</code> when terminated early, such that
 <code>suppressMessages()</code> has an effect; suggested by Ben Bolker.
 </p>
 </li>
 <li> <p><code>citation()</code> supports &lsquo;<span class="file">inst/CITATION</span>&rsquo; files from
 package source trees, with <code>lib.loc</code> pointing to the
 directory containing the package.
 </p>


 </li>
 <li> <p><code>try()</code> gains a new argument <code>outFile</code> with a
 default that can be modified <em>via</em>
 <code>options(try.outFile = .)</code>, useful notably for <code>Sweave</code>.
 </p>
 </li>
 <li><p> The unexported low-level functions in package <span class="pkg">parallel</span>
 for passing serialized <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> objects to and from forked children now
 support long vectors on 64-bit platforms.  This removes some
 limits on higher-level functions such as <code>mclapply()</code> (but
 returning gigabyte results from forked processes <em>via</em>
 serialization should be avoided if at all possible).
 </p>
 </li>
 <li><p> Connections now <code>print()</code> without error even if
 invalid, e.g. after having been destroyed.
 </p>
 </li>
 <li> <p><code>apropos()</code> and <code>find(simple.words = FALSE)</code> no
 longer match object names starting with <span class="samp">.</span> which are known
 to be internal objects (such as <code>.__S3MethodsTable__.</code>).
 </p>
 </li>
 <li><p> Convenience function <code>hasName()</code> has been added; it
 is intended to replace the common idiom <code>!is.null(x$name)</code>
 without the usually unintended partial name matching.
 </p>
 </li>
 <li> <p><code>strcapture()</code> no longer fixes column names nor coerces
 strings to factors (suggested by Bill Dunlap).
 </p>
 </li>
 <li> <p><code>strcapture()</code> returns <code>NA</code> for non-matching
 values in <code>x</code> (suggested by Bill Dunlap).
 </p>
 </li>
 <li> <p><code>source()</code> gets new optional arguments, notably
 <code>exprs</code>; this is made use of in the new utility function
 <code>withAutoprint()</code>.
 </p>
 </li>
 <li> <p><code>sys.source()</code> gets a new <code>toplevel.env</code> argument.
 This argument is useful for frameworks running package tests;
 contributed by Tomas Kalibera.
 </p>
 </li>
 <li> <p><code>Sys.setFileTime()</code> and
 <code>file.copy(copy.date = TRUE)</code> will set timestamps with
 fractions of seconds on platforms/filesystems which support this.
 </p>
 </li>
 <li><p> (Windows only.) <code>file.info()</code> now returns file
 timestamps including fractions of seconds; it has done so on other
 platforms since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.14.0.  (NB: some filesystems do not
 record modification and access timestamps to sub-second
 resolution.)
 </p>
 </li>
 <li><p> The license check enabled by
 <code>options(checkPackageLicense = TRUE)</code> is now done when the
 package's namespace is first loaded.
 </p>
 </li>
 <li> <p><code>ppr()</code> and <code>supsmu()</code> get an optional
 <code>trace</code> argument, and <code>ppr(.., sm.method = ..spline)</code> is
 no longer limited to sample size <i>n &lt;= 2500</i>.
 </p>
 </li>
 <li><p> The <code>POSIXct</code> method for <code>print()</code> gets optional
 <code>tz</code> and <code>usetz</code> arguments, thanks to a report from
 Jennifer S. Lyon.
 </p>
 </li>
 <li><p> New function <code>check_packages_in_dir_details()</code> in
 package <span class="pkg">tools</span> for analyzing package-check log files to
 obtain check details.
 </p>
 </li>
 <li><p> Package <span class="pkg">tools</span> now exports function
 <code>CRAN_package_db()</code> for obtaining information about current
 packages in the <acronym><span class="acronym">CRAN</span></acronym> package repository, and several
 functions for obtaining the check status of these packages.
 </p>
 </li>
 <li><p> The (default) Stangle driver <code>Rtangle</code> allows
 <code>annotate</code> to be a function and gets a new
 <code>drop.evalFALSE</code> option.
 </p>
 </li>
 <li><p> The default method for <code>quantile(x, prob)</code> should now
 be monotone in <code>prob</code>, even in border cases, see <a href="https://bugs.R-project.org/show_bug.cgi?id=16672">PR#16672</a>.
 </p>
 </li>
 <li> <p><code>bug.report()</code> now tries to extract an email address
 from a <span class="samp">BugReports</span> field, and if there is none, from a
 <span class="samp">Contacts</span> field.
 </p>
 </li>
 <li><p> The <code>format()</code> and <code>print()</code> methods for
 <code>object.size()</code> results get new options <code>standard</code> and
 <code>digits</code>; notably, <code>standard = "IEC"</code> and
 <code>standard = "SI"</code> allow more standard (but less common)
 abbreviations than the default ones, e.g. for kilobytes.
 (From contributions by Henrik Bengtsson.)
 </p>
 </li>
 <li><p> If a reference class has a validity method,
 <code>validObject</code> will be called automatically from the
 default initialization method for reference classes.
 </p>
 </li>
 <li> <p><code>tapply()</code> gets new option <code>default = NA</code>
 allowing to change the previously hardcoded value.
 </p>
 </li>
 <li> <p><code>read.dcf()</code> now consistently interprets any
 &lsquo;whitespace&rsquo; to be stripped to include newlines.
 </p>
 </li>
 <li><p> The maximum number of DLLs that can be loaded into R
 e.g. <em>via</em> <code>dyn.load()</code> can now be increased by
 setting the environment variable <code>R_MAX_NUM_DLLS</code> before
 starting R.
 </p>
 </li>
 <li><p> Assigning to an element of a vector beyond the current
 length now over-allocates by a small fraction. The new vector is
 marked internally as growable, and the true length of the new
 vector is stored in the <code>truelength</code> field. This makes
 building up a vector result by assigning to the next element
 beyond the current length more efficient, though pre-allocating is
 still preferred.  The implementation is subject to change and not
 intended to be used in packages at this time.
 </p>
 </li>
 <li><p> Loading the <span class="pkg">parallel</span> package namespace no longer sets
 or changes the <code>.Random.seed</code>, even if <span class="env">R_PARALLEL_PORT</span>
 is unset.
 </p>
 <p>NB: This can break reproducibility of output, and did for a
 <acronym><span class="acronym">CRAN</span></acronym> package.
 </p>
 </li>
 <li><p> Methods <code>"wget"</code> and <code>"curl"</code> for
 <code>download.file()</code> now give an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> error rather than a non-zero
 return value when the external command has a non-zero status.
 </p>
 </li>
 <li><p> Encoding name <code>"utf8"</code> is mapped to <code>"UTF-8"</code>.
 Many implementations of <code>iconv</code> accept <code>"utf8"</code>, but not
 GNU <span class="pkg">libiconv</span> (including the late 2016 version 1.15).
 </p>
 </li>
 <li> <p><code>sessionInfo()</code> shows the full paths to the library or
 executable files providing the BLAS/LAPACK implementations
 currently in use (not available on Windows).
 </p>
 </li>
 <li><p> The binning algorithm used by bandwidth selectors
 <code>bw.ucv()</code>, <code>bw.bcv()</code> and <code>bw.SJ()</code> switches to a
 version linear in the input size <code>n</code> for <code>n &gt; nb/2</code>.
 (The calculations are the same, but for larger <code>n/nb</code> it is
 worth doing the binning in advance.)
 </p>
 </li>
 <li><p> There is a new option <code>PCRE_study</code> which controls when
 <code>grep(perl = TRUE)</code> and friends &lsquo;study&rsquo; the compiled
 pattern.  Previously this was done for 11 or more input strings:
 it now defaults to 10 or more (but most examples need many more
 for the difference from studying to be noticeable).
 </p>
 </li>
 <li> <p><code>grep(perl = TRUE)</code> and friends can now make use of
 PCRE's Just-In-Time mechanism, for PCRE <i>&gt;=</i> 8.20 on
 platforms where JIT is supported.  It is used by default whenever
 the <code>pattern</code> is studied (see the previous item).  (Based on
 a patch from Mikko Korpela.)
 </p>
 <p>This is controlled by a new option <code>PCRE_use_JIT</code>.
 </p>
 <p>Note that in general this makes little difference to the speed,
 and may take a little longer: its benefits are most evident on
 strings of thousands of characters.  As a side effect it reduces
 the chances of C stack overflow in the PCRE library on very long
 strings (millions of characters, but see next item).
 </p>
 <p>Warning: segfaults were seen using PCRE with JIT enabled on 64-bit
 Sparc builds.
 </p>
 </li>
 <li><p> There is a new option <code>PCRE_limit_recursion</code> for
 <code>grep(perl = TRUE)</code> and friends to set a recursion limit
 taking into account <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>'s estimate of the remaining C stack space
 (or 10000 if that is not available).  This reduces the chance of C
 stack overflow, but because it is conservative may report a
 non-match (with a warning) in examples that matched before.  By
 default it is enabled if any input string has 1000 or more bytes.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16757">PR#16757</a>)
 </p>
 </li>
 <li> <p><code>getGraphicsEvent()</code> now works on
 <code>X11(type = "cairo")</code> devices.  Thanks to Frederick Eaton (for
 reviving an earlier patch).
 </p>
 </li>
 <li><p> There is a new argument <code>onIdle</code> for
 <code>getGraphicsEvent()</code>, which allows an R function to be run
 whenever there are no pending graphics events.  This is currently
 only supported on X11 devices.  Thanks to Frederick Eaton.
 </p>
 </li>
 <li><p> The <code>deriv()</code> and similar functions now can compute
 derivatives of <code>log1p()</code>, <code>sinpi()</code> and similar
 one-argument functions, thanks to a contribution by Jerry Lewis.
 </p>
 </li>
 <li> <p><code>median()</code> gains a formal <code>...</code> argument, so
 methods with extra arguments can be provided.
 </p>
 </li>
 <li> <p><code>strwrap()</code> reduces <code>indent</code> if it is more than
 half <code>width</code> rather than giving an error.  (Suggested by Bill
 Dunlap.)
 </p>
 </li>
 <li><p> When the condition <code>code</code> in <code>if(.)</code> or
 <code>while(.)</code> is not of length one, an error instead of a
 warning may be triggered by setting an environment variable, see
 the help page.
 </p>
 </li>
 <li><p> Formatting and printing of bibliography entries
 (<code>bibentry</code>) is more flexible and better documented.  Apart
 from setting <code>options(citation.bibtex.max = 99)</code> you can also
 use <code>print(&lt;citation&gt;, bibtex=TRUE)</code> (or <code>format(..)</code>)
 to get the BibTeX entries in the case of more than one entry.
 This also affects <code>citation()</code>.
 Contributions to enable <code>style = "html+bibtex"</code> are welcome.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> Entry points <code>R_MakeExternalPtrFn</code> and
 <code>R_ExternalPtrFn</code> are now declared in header
 &lsquo;<span class="file">Rinternals.h</span>&rsquo; to facilitate creating and retrieving an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 external pointer from a C function pointer without ISO C warnings
 about the conversion of function pointers.
 </p>
 </li>
 <li><p> There was an exception for the native Solaris C++ compiler
 to the dropping (in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0) of legacy C++ headers from headers
 such as &lsquo;<span class="file">R.h</span>&rsquo; and &lsquo;<span class="file">Rmath.h</span>&rsquo; &mdash; this has now been
 removed.  That compiler has strict C++98 compliance hence does not
 include extensions in its (non-legacy) C++ headers: some
 packages will need to request C++11 or replace non-C++98 calls
 such as <code>lgamma</code>: see §1.6.4 of &lsquo;Writing R Extensions&rsquo;.
 </p>
 <p>Because it is needed by about 70 CRAN packages, headers &lsquo;<span class="file">R.h</span>&rsquo;
 and &lsquo;<span class="file">Rmath.h</span>&rsquo; still declare </p>
 <pre>use namespace std;</pre>
 <p>when included on Solaris.
 </p>
 </li>
 <li><p> When included from C++, the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> headers now use forms such as
 <code>std::FILE</code> directly rather than including the line </p>
 <pre>using std::FILE;</pre>
 <p>C++ code including these headers might be relying on the latter.
 </p>
 </li>
 <li><p> Headers &lsquo;<span class="file">R_ext/BLAS.h</span>&rsquo; and &lsquo;<span class="file">R_ext/Lapack.h</span>&rsquo; have
 many improved declarations including <code>const</code> for
 double-precision complex routines. <em>Inter alia</em> this avoids
 warnings when passing &lsquo;string literal&rsquo; arguments from C++11
 code.
 </p>
 </li>
 <li><p> Headers for Unix-only facilities &lsquo;<span class="file">R_ext/GetX11Image.h</span>&rsquo;,
 &lsquo;<span class="file">R_ext/QuartzDevice.h</span>&rsquo; and &lsquo;<span class="file">R_ext/eventloop.h</span>&rsquo; are no
 longer installed on Windows.
 </p>
 </li>
 <li><p> No-longer-installed headers &lsquo;<span class="file">GraphicsBase.h</span>&rsquo;,
 &lsquo;<span class="file">RGraphics.h</span>&rsquo;, &lsquo;<span class="file">Rmodules/RX11.h</span>&rsquo; and
 &lsquo;<span class="file">Rmodules/Rlapack.h</span>&rsquo; which had a LGPL license no longer do so.
 </p>
 </li>
 <li> <p><code>HAVE_UINTPTR_T</code> is now defined where appropriate by
 <code>Rconfig.h</code> so that it can be included before
 <code>Rinterface.h</code> when <code>CSTACK_DEFNS</code> is defined and a C
 compiler (not C++) is in use.  <code>Rinterface.h</code> now includes C
 header &lsquo;<span class="file">stdint.h</span>&rsquo; or C++11 header &lsquo;<span class="file">cstdint</span>&rsquo; where
 needed.
 </p>
 </li>
 <li><p> Package <span class="pkg">tools</span> has a new function
 <code>package_native_routine_registration_skeleton()</code> to assist
 adding native-symbol registration to a package.  See its help and
 §5.4.1 of &lsquo;Writing R Extensions&rsquo; for how to use it.  (At
 the time it was added it successfully automated adding
 registration to over 90% of <acronym><span class="acronym">CRAN</span></acronym> packages which lacked
 it.  Many of the failures were newly-detected bugs in the
 packages, e.g. 50 packages called entry points with varying
 numbers of arguments and 65 packages called entry points not in
 the package.)


 </p>
 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li> <p><code>readline</code> headers (and not just the library) are
 required unless configuring with <span class="option">--with-readline=no</span>.
 </p>
 </li>
 <li> <p><code>configure</code> now adds a compiler switch for C++11
 code, even if the compiler supports C++11 by default.  (This
 ensures that <code>g++</code> 6.x uses C++11 mode and not its default
 mode of C++14 with &lsquo;GNU extensions&rsquo;.)
 </p>
 <p>The tests for C++11 compliance are now much more comprehensive.
 For gcc &lt; 4.8, the tests from R 3.3.0 are used in order to maintain
 the same behaviour on Linux distributions with long-term support.
 </p>
 </li>
 <li><p> An alternative compiler for C++11 is now specified with
 <span class="samp">CXX11</span>, not <span class="samp">CXX1X</span>. Likewise C++11 flags are specified
 with <span class="samp">CXX11FLAGS</span> and the standard (e.g., <span class="samp">-std=gnu++11</span>)
 is specified with <span class="samp">CXX11STD</span>.
 </p>
 </li>
 <li> <p><code>configure</code> now tests for a C++14-compliant compiler
 by testing some basic features.  This by default tries flags for
 the compiler specified by <span class="samp">CXX11</span>, but an alternative compiler,
 options and standard can be specified by variables <span class="samp">CXX14</span>,
 <span class="samp">CXX14FLAGS</span> and <span class="samp">CXX14STD</span> (e.g., <span class="option">-std=gnu++14</span>).
 </p>
 </li>
 <li><p> There is a new macro <code>CXXSTD</code> to help specify the
 standard for C++ code, e.g. <span class="option">-std=c++98</span>.  This
 makes it easier to work with compilers which default to a later
 standard: for example, with <code>CXX=g++6 CXXSTD=-std=c++98</code>
 <code>configure</code> will select commands for <code>g++</code> 6.x
 which conform to C++11 and C++14 where specified but otherwise use
 C++98.
 </p>
 </li>
 <li><p> Support for the defunct IRIX and OSF/1 OSes and Alpha CPU
 has been removed.
 </p>
 </li>
 <li> <p><code>configure</code> checks that the compiler specified by
 <span class="samp">$CXX $CXXFLAGS</span> is able to compile C++ code.
 </p>
 </li>
 <li> <p><code>configure</code> checks for the required header
 &lsquo;<span class="file">sys/select.h</span>&rsquo; (or &lsquo;<span class="file">sys/time.h</span>&rsquo; on legacy systems) and
 system call <code>select</code> and aborts if they are not found.
 </p>
 </li>
 <li><p> If available, the POSIX 2008 system call <code>utimensat</code>
 will be used by <code>Sys.setFileTime()</code> and
 <code>file.copy(copy.date = TRUE)</code>.  This may result in slightly
 more accurate file times.  (It is available on Linux and FreeBSD
 but not macOS.)

 </p>
 </li>
 <li><p> The minimum version requirement for <code>libcurl</code> has been
 reduced to 7.22.0, although at least 7.28.0 is preferred and
 earlier versions are little tested.  (This is to support Debian 7
 &lsquo;Wheezy&rsquo; LTS and Ubuntu &lsquo;Precise&rsquo; 12.04 LTS,
 although the latter is close to end-of-life.)

 </p>
 </li>
 <li> <p><code>configure</code> tests for a C++17-compliant compiler.
 The tests are experimental and subject to change in the future.
 </p>
 </li></ul>


 <h4>INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> (Windows only) Tcl/Tk version 8.6.4 is now included in the
 binary builds.  The &lsquo;<span class="file">tcltk*.chm</span>&rsquo; help file is no longer
 included; please consult the online help at
 <a href="http://www.tcl.tk/man/">http://www.tcl.tk/man/</a> instead.
 </p>
 </li>
 <li><p> The version of LAPACK included in the sources has been
 updated to 3.7.0: no new routines have been added to <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> There is support for compiling C++14 or C++17 code in
 packages on suitable platforms: see &lsquo;Writing R Extensions&rsquo;
 for how to request this.
 </p>
 </li>
 <li><p> The order of flags when <span class="samp">LinkingTo</span> other packages has
 been changed so their include directories come earlier, before
 those specified in <code>CPPFLAGS</code>.  This will only have an effect
 if non-system include directories are included with <span class="option">-I</span>
 flags in <code>CPPFLAGS</code> (and so not the default
 <code>-I/usr/local/include</code> which is treated as a system include
 directory on most platforms).


 </p>
 </li>
 <li><p> Packages which register native routines for <code>.C</code> or
 <code>.Fortran</code> need to be re-installed for this version (unless
 installed with R-devel SVN revision r72375 or later).
 </p>
 </li>
 <li><p> Make variables with names containing <code>CXX1X</code> are
 deprecated in favour of those using <code>CXX11</code>, but for the time
 being are still made available <em>via</em> file
 &lsquo;<span class="file">etc/Makeconf</span>&rsquo;.  Packages using them should be converted to
 the new forms and made dependent on <span class="samp">R (&gt;= 3.4.0)</span>.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li><p> Running <code>R CMD check --as-cran</code> with
 <span class="env">_R_CHECK_CRAN_INCOMING_REMOTE_</span> false now skips tests that
 require remote access.  The remaining (local) tests typically
 run quickly compared to the remote tests.
 </p>
 </li>
 <li> <p><code>R CMD build</code> will now give priority to vignettes
 produced from files in the &lsquo;<span class="file">vignettes</span>&rsquo; directory over those
 in the &lsquo;<span class="file">inst/doc</span>&rsquo; directory, with a warning that the latter
 are being ignored.
 </p>
 </li>
 <li> <p><code>R CMD config</code> gains a <span class="option">--all</span> option for
 printing names and values of all basic configure variables.
 </p>
 <p>It now knows about all the variables used for the C++98, C++11 and
 C++14 standards.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now checks that output files in
 &lsquo;<span class="file">inst/doc</span>&rsquo; are newer than the source files in &lsquo;<span class="file">vignettes</span>&rsquo;.
 </p>
 </li>
 <li><p> For consistency with other package subdirectories, files
 named &lsquo;<span class="file">*.r</span>&rsquo; in the &lsquo;<span class="file">tests</span>&rsquo; directory are now recognized
 as tests by <code>R CMD check</code>. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=17143">PR#17143</a>.)
 </p>
 </li>
 <li> <p><code>R CMD build</code> and <code>R CMD check</code> now use the
 <em>union</em> of <span class="env">R_LIBS</span> and <code>.libPaths()</code>.  They may not
 be equivalent, e.g., when the latter is determined by
 <span class="env">R_PROFILE</span>.
 </p>
 </li>
 <li> <p><code>R CMD build</code> now preserves dates when it copies files
 in preparing the tarball.  (Previously on Windows it changed the
 dates on all files; on Unix, it changed some dates when installing
 vignettes.)
 </p>
 </li>
 <li><p> The new option <code>R CMD check --no-stop-on-test-error</code>
 allows running the remaining tests (under &lsquo;<span class="file">tests/</span>&rsquo;) even if
 one gave an error.
 </p>
 </li>
 <li><p> Check customization <em>via</em> environment variables to
 detect side effects of <code>.Call()</code> and <code>.External()</code> calls
 which alter their arguments is described in §8 of the &lsquo;R
 Internals&rsquo; manual.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now checks any <span class="samp">BugReports</span> field
 to be non-empty and a suitable single URL.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> now NOTEs if the package
 does not register its native routines or does not declare its
 intentions on (native) symbol search.  (This will become a WARNING
 in due course.)
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> (Windows only) Function <code>setInternet2()</code> is defunct.
 </p>
 </li>
 <li><p> Installation support for <code>readline</code> emulations based on
 <code>editline</code> (aka <code>libedit</code>) is deprecated.
 </p>
 </li>
 <li><p> Use of the C/C++ macro <span class="samp">NO_C_HEADERS</span> is defunct and
 silently ignored.
 </p>
 </li>
 <li> <p><code>unix.time()</code>, a traditional synonym for
 <code>system.time()</code>, has been deprecated.
 </p>
 </li>
 <li> <p><code>structure(NULL, ..)</code> is now deprecated as you cannot
 set attributes on <code>NULL</code>.
 </p>
 </li>
 <li><p> Header &lsquo;<span class="file">Rconfig.h</span>&rsquo; no longer defines
 <span class="samp">SUPPORT_OPENMP</span>; instead use <span class="samp">_OPENMP</span> (as documented
 for a long time).
 </p>
 </li>
 <li><p> (C-level Native routine registration.)  The deprecated
 <code>styles</code> member of the <code>R_CMethodDef</code> and
 <code>R_FortranMethodDef</code> structures has been removed.
 Packages using these will need to be re-installed for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.4.0.
 </p>
 </li>
 <li><p> The deprecated support for PCRE versions older than 8.20
 will be removed in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.4.1. (Versions 8.20&ndash;8.31 will still be
 accepted but remain deprecated.)
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> Getting or setting <code>body()</code> or <code>formals()</code> on
 non-functions for now signals a warning and may become an error
 for setting.
 </p>
 </li>
 <li> <p><code>match(x, t)</code>, <code>duplicated(x)</code> and
 <code>unique(x)</code> work as documented for complex numbers with <code>NA</code>s
 or <code>NaN</code>s, where all those containing <code>NA</code> do match,
 whereas in the case of <code>NaN</code>'s both real and imaginary parts
 must match, compatibly with how <code>print()</code> and <code>format()</code>
 work for complex numbers.
 </p>
 </li>
 <li> <p><code>deparse(&lt;complex&gt;, options = "digits17")</code> prints more
 nicely now, mostly thanks to a suggestion by Richie Cotton.
 </p>
 </li>
 <li><p> Rotated symbols in plotmath expressions are now positioned
 correctly on <code>x11(type = "Xlib")</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16948">PR#16948</a>)
 </p>
 </li>
 <li> <p><code>as&lt;-()</code> avoids an infinite loop when a virtual class
 is interposed between a subclass and an actual superclass.
 </p>
 </li>
 <li><p> Fix level propagation in <code>unlist()</code> when the list
 contains zero-length lists or factors.
 </p>
 </li>
 <li><p> Fix S3 dispatch on S4 objects when the <span class="pkg">methods</span> package is
 not attached.
 </p>
 </li>
 <li><p> Internal S4 dispatch sets <code>.Generic</code> in the method
 frame for consistency with <code>standardGeneric()</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16929">PR#16929</a>)
 </p>
 </li>
 <li><p> Fix <code>order(x, decreasing = TRUE)</code> when <code>x</code> is an
 integer vector containing <code>MAX_INT</code>.  Ported from a fix Matt
 Dowle made to <a href="https://CRAN.R-project.org/package=data.table"><span class="pkg">data.table</span></a>.
 </p>
 </li>
 <li><p> Fix caching by <code>callNextMethod()</code>, resolves <a href="https://bugs.R-project.org/show_bug.cgi?id=16973">PR#16973</a>
 and <a href="https://bugs.R-project.org/show_bug.cgi?id=16974">PR#16974</a>.
 </p>
 </li>
 <li> <p><code>grouping()</code> puts NAs last, to be consistent with the
 default behavior of <code>order()</code>.
 </p>
 </li>
 <li><p> Point mass limit cases: <code>qpois(-2, 0)</code> now gives
 <code>NaN</code> with a warning and <code>qgeom(1, 1)</code> is <code>0</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16972">PR#16972</a>)
 </p>
 </li>
 <li> <p><code>table()</code> no longer drops an <code>"NaN"</code> factor level,
 and better obeys <code>exclude = &lt;chr&gt;</code>, thanks to Suharto Anggono's
 patch for <a href="https://bugs.R-project.org/show_bug.cgi?id=16936">PR#16936</a>.  Also, in the case of <code>exclude = NULL</code>
 and <code>NA</code>s, these are tabulated correctly (again).
 </p>
 <p>Further, <code>table(1:2, exclude = 1, useNA = "ifany")</code> no longer
 erroneously reports <code>&lt;NA&gt;</code> counts.
 </p>
 <p>Additionally, all cases of empty <code>exclude</code> are equivalent,
 and <code>useNA</code> is not overwritten when specified (as it was by
 <code>exclude = NULL</code>).
 </p>
 </li>
 <li> <p><code>wilcox.test(x, conf.int=TRUE)</code> no longer errors out in
 cases where the confidence interval is not available, such as for
 <code>x = 0:2</code>.
 </p>
 </li>
 <li> <p><code>droplevels(f)</code> now keeps &lt;NA&gt; levels when present.
 </p>
 </li>
 <li><p> In integer arithmetic, <code>NULL</code> is now treated as
 <code>integer(0)</code> whereas it was previously treated as <code>double(0)</code>.
 </p>
 </li>
 <li><p> The radix sort considers <code>NA_real_</code> and <code>NaN</code> to
 be equivalent in rank (like the other sort algorithms).
 </p>
 </li>
 <li><p> When <code>index.return=TRUE</code> is passed to <code>sort.int()</code>,
 the radix sort treats <code>NA</code>s like <code>sort.list()</code> does
 (like the other sort algorithms).
 </p>
 </li>
 <li><p> When in <code>tabulate(bin, nbin)</code> <code>length(bin)</code> is
 larger than the maximal integer, the result is now of type
 <code>double</code> and hence no longer silently overflows to wrong
 values.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17140">PR#17140</a>)
 </p>
 </li>
 <li> <p><code>as.character.factor()</code> respects S4 inheritance when
 checking the type of its argument. (<a href="https://bugs.R-project.org/show_bug.cgi?id=17141">PR#17141</a>)
 </p>
 </li>
 <li><p> The <code>factor</code> method for <code>print()</code> no longer sets
 the class of the factor to <code>NULL</code>, which would violate a
 basic constraint of an S4 object.
 </p>
 </li>
 <li> <p><code>formatC(x, flag = f)</code> allows two new flags, and
 signals an error for invalid flags also in the case of character
 formatting.
 </p>
 </li>
 <li><p> Reading from <code>file("stdin")</code> now also closes the
 connection and hence no longer leaks memory when reading from a
 full pipe, thanks to Gábor Csárdi, see thread starting at
 <a href="https://stat.ethz.ch/pipermail/r-devel/2016-November/073360.html">https://stat.ethz.ch/pipermail/r-devel/2016-November/073360.html</a>.
 </p>
 </li>
 <li><p> Failure to create file in <code>tempdir()</code> for compressed
 <code>pdf()</code> graphics device no longer errors (then later
 segfaults).  There is now a warning instead of error and
 compression is turned off for the device.  Thanks to Alec
 Wysoker (<a href="https://bugs.R-project.org/show_bug.cgi?id=17191">PR#17191</a>).
 </p>
 </li>
 <li><p> Asking for <code>methods()</code> on <code>"|"</code> returns only S3
 methods. See
 <a href="https://stat.ethz.ch/pipermail/r-devel/2016-December/073476.html">https://stat.ethz.ch/pipermail/r-devel/2016-December/073476.html</a>.
 </p>
 </li>
 <li> <p><code>dev.capture()</code> using Quartz Cocoa device (macOS) returned
 invalid components if the back-end chose to use ARGB instead of RGBA
 image format. (Reported by Noam Ross.)
 </p>
 </li>
 <li> <p><code>seq("2", "5")</code> now works too, equivalently to
 <code>"2":"5"</code> and <code>seq.int()</code>.
 </p>
 </li>
 <li> <p><code>seq.int(to = 1, by = 1)</code> is now correct, other cases are
 integer (instead of <code>double</code>) when <code>seq()</code> is integer
 too, and the &quot;non-finite&quot; error messages are consistent between
 <code>seq.default()</code> and <code>seq.int()</code>, no longer mentioning
 <code>NaN</code> etc.
 </p>
 </li>
 <li> <p><code>rep(x, times)</code> and <code>rep.int(x, times)</code> now work
 when <code>times</code> is larger than the largest value representable
 in an integer vector.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16932">PR#16932</a>)
 </p>
 </li>
 <li> <p><code>download.file(method = "libcurl")</code> does not check for
 URL existence before attempting downloads; this is more robust to
 servers that do not support HEAD or range-based retrieval, but may
 create empty or incomplete files for aborted download requests.
 </p>
 </li>
 <li><p> Bandwidth selectors <code>bw.ucv()</code>, <code>bw.bcv()</code> and
 <code>bw.SJ()</code> now avoid integer overflow for large sample sizes.
 </p>
 </li>
 <li> <p><code>str()</code> no longer shows <code>"list output truncated"</code>,
 in cases that list was not shown at all.  Thanks to Neal Fultz
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17219">PR#17219</a>)
 </p>
 </li>
 <li><p> Fix for <code>cairo_pdf()</code> (and <code>svg()</code> and
 <code>cairo_ps()</code>) when replaying a saved display list that contains
 a mix of <span class="pkg">grid</span> and <span class="pkg">graphics</span> output.
 (Report by Yihui Xie.)
 </p>


 </li>
 <li><p> The <code>str()</code> and <code>as.hclust()</code> methods for
 <code>"dendrogram"</code> now also work for deeply nested dendrograms
 thanks to non-recursive implementations by Bradley Broom.
 </p>
 </li>
 <li> <p><code>sample()</code> now uses two uniforms for added precision
 when the uniform generator is <code>Knuth-TAOCP</code>,
 <code>Knuth-TAOCP-2002</code>, or a user-defined generator and the
 population size is <i>2^25</i> or greater.
 </p>
 </li>
 <li><p> If a vignette in the &lsquo;<span class="file">vignettes</span>&rsquo; directory is listed in
 &lsquo;<span class="file">.Rbuildignore</span>&rsquo;, <code>R CMD build</code> would not include it
 in the tarball, but would include it in the vignette database,
 leading to a check warning.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17246">PR#17246</a>)
 </p>
 </li>
 <li> <p><code>tools::latexToUtf8()</code> infinite looped on certain
 inputs.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17138">PR#17138</a>)
 </p>
 </li>
 <li> <p><code>terms.formula()</code> ignored argument names when determining
 whether two terms were identical.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17235">PR#17235</a>)
 </p>
 </li>
 <li> <p><code>callNextMethod()</code> was broken when called from a method
 that augments the formal arguments of a primitive generic.
 </p>
 </li>
 <li><p> Coercion of an S4 object to a vector during sub-assignment
 into a vector failed to dispatch through the <code>as.vector()</code>
 generic (often leading to a segfault).
 </p>
 </li>
 <li><p> Fix problems in command completion:  Crash (<a href="https://bugs.R-project.org/show_bug.cgi?id=17222">PR#17222</a>)
 and junk display in Windows, handling special characters in
 filenames on all systems.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.3.3</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> Changes when redirection of a <span class="samp">http://</span> URL to a
 <span class="samp">https://</span> URL is encountered:
 </p>

 <ul>
 <li><p> The internal methods of <code>download.file()</code> and
 <code>url()</code> now report that they cannot follow this (rather
 than failing silently).
 </p>
 </li>
 <li><p> (Unix-alike) <code>download.file(method = "auto")</code> (the
 default) re-tries with <code>method = "libcurl"</code>.
 </p>
 </li>
 <li><p> (Unix-alike) <code>url(method = "default")</code> with an
 explicit <code>open</code> argument re-tries with
 <code>method = "libcurl"</code>.  This covers many of the usages,
 e.g. <code>readLines()</code> with a URL argument.
 </p>
 </li></ul>

 </li></ul>


 <h4>INSTALLATION on a UNIX-ALIKE</h4>


 <ul>
 <li><p> The <code>configure</code> check for the <code>zlib</code> version is
 now robust to versions longer than 5 characters, including
 <code>1.2.11</code>.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li><p> Environmental variable <span class="env">_R_CHECK_TESTS_NLINES_</span> controls
 how <code>R CMD check</code> reports failing tests (see §8 of the
 &lsquo;R Internals&rsquo; manual).
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> (C-level Native routine registration.)  The undocumented
 <code>styles</code> field of the components of <code>R_CMethodDef</code> and
 <code>R_FortranMethodDef</code> is deprecated.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>vapply(x, *)</code> now works with long vectors <code>x</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=17174">PR#17174</a>)
 </p>
 </li>
 <li> <p><code>isS3method("is.na.data.frame")</code> and similar are
 correct now.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17171">PR#17171</a>)
 </p>
 </li>
 <li> <p><code>grepRaw(&lt;long&gt;, &lt;short&gt;, fixed = TRUE)</code> now works,
 thanks to a patch by Mikko Korpela.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17132">PR#17132</a>)
 </p>
 </li>
 <li><p> Package installation into a library where the package exists
 <em>via</em> symbolic link now should work wherever
 <code>Sys.readlink()</code> works, resolving <a href="https://bugs.R-project.org/show_bug.cgi?id=16725">PR#16725</a>.
 </p>
 </li>
 <li> <p><code>"Cincinnati"</code> was missing an <code>"n"</code> in the
 <code>precip</code> dataset.
 </p>
 </li>
 <li><p> Fix buffer overflow vulnerability in <code>pdf()</code> when
 loading an encoding file.  Reported by Talos (TALOS-2016-0227).
 </p>
 </li>
 <li> <p><code>getDLLRegisteredRoutines()</code> now produces its warning
 correctly when multiple DLLs match, thanks to Matt Dowle's <a href="https://bugs.R-project.org/show_bug.cgi?id=17184">PR#17184</a>.
 </p>
 </li>
 <li> <p><code>Sys.timezone()</code> now returns non-NA also on platforms
 such as <span class="samp">Ubuntu 14.04.5 LTS</span>, thanks to Mikko Korpela's
 <a href="https://bugs.R-project.org/show_bug.cgi?id=17186">PR#17186</a>.
 </p>
 </li>
 <li> <p><code>format(x)</code> for an illegal <code>"POSIXlt"</code> object
 <code>x</code> no longer segfaults.
 </p>
 </li>
 <li> <p><code>methods(f)</code> now also works for <code>f</code> <code>"("</code>
 or <code>"{"</code>.
 </p>
 </li>
 <li><p> (Windows only)  <code>dir.create()</code> did not check the length
 of the path to create, and so could overflow a buffer and crash
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17206">PR#17206</a>)
 </p>
 </li>
 <li><p> On some systems, very small hexadecimal numbers in hex notation
 would underflow to zero.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17199">PR#17199</a>)
 </p>
 </li>
 <li> <p><code>pmin()</code> and <code>pmax()</code> now work again for
 <code>ordered</code> factors and 0-length S3 classed objects, thanks to
 Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17195">PR#17195</a> and <a href="https://bugs.R-project.org/show_bug.cgi?id=17200">PR#17200</a>.
 </p>
 </li>
 <li> <p><code>bug.report()</code> did not do any validity checking on a
 package's <span class="samp">BugReports</span> field.  It now ignores an empty field,
 removes leading whitespace and only attempts to open
 <span class="samp">http://</span> and <span class="samp">https://</span> URLs, falling back to emailing
 the maintainer.
 </p>
 </li>
 <li><p> Bandwidth selectors <code>bw.ucv()</code> and
 <code>bw.SJ()</code> gave incorrect answers or incorrectly reported an
 error (because of integer overflow) for inputs longer than
 46341.  Similarly for <code>bw.bcv()</code> at length 5793.
 </p>
 <p>Another possible integer overflow is checked and may result in an
 error report (rather than an incorrect result) for much longer
 inputs (millions for a smooth distribution).
 </p>
 </li>
 <li> <p><code>findMethod()</code> failed if the active signature had
 expanded beyond what a particular package used. (Example with
 packages <a href="https://CRAN.R-project.org/package=XR"><span class="pkg">XR</span></a> and <a href="https://CRAN.R-project.org/package=XRJulia"><span class="pkg">XRJulia</span></a> on <acronym><span class="acronym">CRAN</span></acronym>.)
 </p>
 </li>
 <li> <p><code>qbeta()</code> underflowed too early in some very asymmetric
 cases.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17178">PR#17178</a>)
 </p>
 </li>
 <li> <p><code>R CMD Rd2pdf</code> had problems with packages with
 non-ASCII titles in &lsquo;<span class="file">.Rd</span>&rsquo; files (usually the titles were
 omitted).
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.3.2</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>extSoftVersion()</code> now reports the version (if any) of
 the <code>readline</code> library in use.
 </p>
 </li>
 <li><p> The version of LAPACK included in the sources has been
 updated to 3.6.1, a bug-fix release including a speedup for the
 non-symmetric case of <code>eigen()</code>.
 </p>
 </li>
 <li><p> Use <code>options(deparse.max.lines=)</code> to limit the number
 of lines recorded in <code>.Traceback</code> and other deparsing
 activities.
 </p>
 </li>
 <li> <p><code>format(&lt;AsIs&gt;)</code> looks more regular, also for
 non-character atomic matrices.
 </p>
 </li>
 <li> <p><code>abbreviate()</code> gains an option <code>named = TRUE</code>.
 </p>
 </li>
 <li><p> The online documentation for package <span class="pkg">methods</span> is
 extensively rewritten.  The goals are to simplify
 documentation for basic use, to note old features not
 recommended and to correct out-of-date information.
 </p>
 </li>
 <li><p> Calls to <code>setMethod()</code> no longer print a message when
 creating a generic function in those cases where that is
 natural: S3 generics and primitives.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> Versions of the <code>readline</code> library &gt;= 6.3 had been
 changed so that terminal window resizes were not signalled to
 <code>readline</code>: code has been added using a explicit signal
 handler to work around that (when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is compiled against readline
 &gt;= 6.3).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16604">PR#16604</a>)
 </p>
 </li>
 <li> <p><code>configure</code> works better with Oracle Developer
 Studio 12.5.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> reports more dubious flags in files
 &lsquo;<span class="file">src/Makevars[.in]</span>&rsquo;, including <span class="option">-w</span> and <span class="option">-g</span>.
 </p>
 </li>
 <li> <p><code>R CMD check</code> has been set up to filter important
 warnings from recent versions of <code>gfortran</code> with
 <span class="option">-Wall -pedantic</span>: this now reports non-portable GNU
 extensions such as out-of-order declarations.
 </p>
 </li>
 <li> <p><code>R CMD config</code> works better with paths containing
 spaces, even those of home directories (as reported by Ken Beath).
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> Use of the C/C++ macro <span class="samp">NO_C_HEADERS</span> is deprecated
 (no C headers are included by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> headers from C++ as from <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.3.0, so it should no longer be needed).
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> The check for non-portable flags in <code>R CMD check</code>
 could be stymied by &lsquo;<span class="file">src/Makevars</span>&rsquo; files which contained
 targets.
 </p>
 </li>
 <li><p> (Windows only) When using certain desktop themes in Windows
 7 or higher, <kbd>Alt-Tab</kbd> could cause <code>Rterm</code> to stop
 accepting input.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=14406">PR#14406</a>; patch submitted by Jan Gleixner.)
 </p>
 </li>
 <li> <p><code>pretty(d, ..)</code> behaves better for date-time <code>d</code>
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16923">PR#16923</a>).
 </p>
 </li>
 <li><p> When an S4 class name matches multiple classes in the S4
 cache, perform a dynamic search in order to obey namespace
 imports.  This should eliminate annoying messages about multiple
 hits in the class cache.  Also, pass along the package from the
 <code>ClassExtends</code> object when looking up superclasses in the
 cache.
 </p>
 </li>
 <li> <p><code>sample(NA_real_)</code> now works.
 </p>
 </li>
 <li><p> Packages using non-ASCII encodings in their code did not
 install data properly on systems using different encodings.
 </p>
 </li>
 <li> <p><code>merge(df1, df2)</code> now also works for data frames with
 column names <code>"na.last"</code>, <code>"decreasing"</code>, or
 <code>"method"</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17119">PR#17119</a>)
 </p>
 </li>
 <li> <p><code>contour()</code> caused a segfault if the <code>labels</code>
 argument had length zero.  (Reported by Bill Dunlap.)
 </p>
 </li>
 <li> <p><code>unique(warnings())</code> works more correctly, thanks to a
 new <code>duplicated.warnings()</code> method.
 </p>
 </li>
 <li> <p><code>findInterval(x, vec = numeric(), all.inside = TRUE)</code>
 now returns <code>0</code>s as documented.  (Reported by Bill Dunlap.)
 </p>
 </li>
 <li><p> (Windows only) <code>R CMD SHLIB</code> failed when a symbol
 in the resulting library had the same name as a keyword in
 the &lsquo;<span class="file">.def</span>&rsquo; file.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17130">PR#17130</a>)
 </p>
 </li>
 <li> <p><code>pmax()</code> and <code>pmin()</code> now work with (more ?)
 classed objects, such as <code>"Matrix"</code> from the <a href="https://CRAN.R-project.org/package=Matrix"><span class="pkg">Matrix</span></a>
 package, as documented for a long time.
 </p>
 </li>
 <li> <p><code>axis(side, x = D)</code> and hence <code>Axis()</code> and
 <code>plot()</code> now work correctly for <code>"Date"</code> and time
 objects <code>D</code>, even when &ldquo;time goes backward&rdquo;, e.g.,
 with decreasing <code>xlim</code>.  (Reported by William May.)
 </p>
 </li>
 <li> <p><code>str(I(matrix(..)))</code> now looks as always intended.
 </p>
 </li>
 <li> <p><code>plot.ts()</code>, the <code>plot()</code> method for time series,
 now respects <code>cex</code>, <code>lwd</code> and <code>lty</code>.  (Reported by
 Greg Werbin.)
 </p>
 </li>
 <li> <p><code>parallel::mccollect()</code> now returns a named list (as
 documented) when called with <code>wait = FALSE</code>.  (Reported by
 Michel Lang.)
 </p>
 </li>
 <li><p> If a package added a class to a class union in another
 package, loading the first package gave erroneous warnings
 about &ldquo;undefined subclass&rdquo;.
 </p>
 </li>
 <li> <p><code>c()</code>'s argument <code>use.names</code> is documented now, as
 belonging to the (C internal) default method.   In
 &ldquo;parallel&rdquo;, argument <code>recursive</code> is also moved from
 the generic to the default method, such that the formal argument list
 of <span class="pkg">base</span> generic <code>c()</code> is just <code>(...)</code>.
 </p>
 </li>
 <li> <p><code>rbeta(4, NA)</code> and similarly <code>rgamma()</code> and
 <code>rnbinom()</code> now return <code>NaN</code>'s with a warning, as other
 r&lt;dist&gt;(), and as documented.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17155">PR#17155</a>)
 </p>
 </li>
 <li><p> Using <code>options(checkPackageLicense = TRUE)</code> no longer
 requires acceptance of the licence for non-default standard
 packages such as <span class="pkg">compiler</span>.  (Reported by Mikko Korpela.)
 </p>
 </li>
 <li> <p><code>split(&lt;very_long&gt;, *)</code> now works even when the split
 off parts are long. (<a href="https://bugs.R-project.org/show_bug.cgi?id=17139">PR#17139</a>)
 </p>
 </li>
 <li> <p><code>min()</code> and <code>max()</code> now also work correctly when
 the argument list starts with <code>character(0)</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17160">PR#17160</a>)
 </p>
 </li>
 <li><p> Subsetting very large matrices (<code>prod(dim(.)) &gt;= 2^31</code>)
 now works thanks to Michael Schubmehl's <a href="https://bugs.R-project.org/show_bug.cgi?id=17158">PR#17158</a>.
 </p>
 </li>
 <li> <p><code>bartlett.test()</code> used residual sums of squares
 instead of variances, when the argument was a list of <code>lm</code> objects.
 (Reported by Jens Ledet Jensen).
 </p>
 </li>
 <li> <p><code>plot(&lt;lm&gt;, which = *)</code> now correctly labels the
 contour lines for the standardized residuals for <code>which = 6</code>.
 It also takes the correct <i>p</i> in case of singularities (also
 for <code>which = 5</code>).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=17161">PR#17161</a>)
 </p>
 </li>
 <li> <p><code>xtabs(~ exclude)</code> no longer fails from wrong scope,
 thanks to Suharto Anggono's <a href="https://bugs.R-project.org/show_bug.cgi?id=17147">PR#17147</a>.
 </p>
 </li>
 <li><p> Reference class calls to <code>methods()</code> did not
 re-analyse previously defined methods, meaning that calls to
 methods defined later would fail. (Reported by Charles Tilford).
 </p>
 </li>
 <li> <p><code>findInterval(x, vec, left.open = TRUE)</code> misbehaved in some
 cases. (Reported by Dmitriy Chernykh.)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.3.1</h3>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>R CMD INSTALL</code> and hence <code>install.packages()</code>
 gave an internal error installing a package called
 <span class="pkg">description</span> from a tarball on a case-insensitive file system.
 </p>
 </li>
 <li> <p><code>match(x, t)</code> (and hence <code>x %in% t</code>) failed
 when <code>x</code> was of length one, and either <code>character</code> and <code>x</code>
 and <code>t</code> only differed in their <code>Encoding</code> or when
 <code>x</code> and <code>t</code> where <code>complex</code> with <code>NA</code>s or <code>NaN</code>s.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16885">PR#16885</a>.)
 </p>
 </li>
 <li> <p><code>unloadNamespace(ns)</code> also works again when <code>ns</code> is a
 &lsquo;namespace&rsquo;, as from <code>getNamespace()</code>.
 </p>
 </li>
 <li> <p><code>rgamma(1, Inf)</code> or <code>rgamma(1, 0, 0)</code> no longer give
 <code>NaN</code> but the correct limit.
 </p>
 </li>
 <li> <p><code>length(baseenv())</code> is correct now.
 </p>
 </li>
 <li> <p><code>pretty(d, ..)</code> for date-time <code>d</code> rarely failed
 when <code>"halfmonth"</code> time steps were tried (<a href="https://bugs.R-project.org/show_bug.cgi?id=16923">PR#16923</a>) and
 on &lsquo;inaccurate&rsquo; platforms such as 32-bit Windows or
 a configuration with <code>--disable-long-double</code>; see comment
 #15 of <a href="https://bugs.R-project.org/show_bug.cgi?id=16761">PR#16761</a>.
 </p>
 </li>
 <li><p> In <code>text.default(x, y, labels)</code>, the rarely(?) used
 default for <code>labels</code> is now correct also for the case of a
 2-column matrix <code>x</code> and missing <code>y</code>.
 </p>
 </li>
 <li> <p><code>as.factor(c(a = 1L))</code> preserves <code>names()</code> again
 as in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> &lt; 3.1.0.
 </p>
 </li>
 <li> <p><code>strtrim(""[0], 0[0])</code> now works.
 </p>
 </li>
 <li><p> Use of <code>Ctrl-C</code> to terminate a reverse incremental
 search started by <code>Ctrl-R</code> in the <code>readline</code>-based Unix
 terminal interface is now supported when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> was compiled against
 <code>readline</code> &gt;= 6.0 (<code>Ctrl-G</code> always worked).
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16603">PR#16603</a>)
 </p>
 </li>
 <li> <p><code>diff(&lt;difftime&gt;)</code> now keeps the <code>"units"</code>
 attribute, as subtraction already did, <a href="https://bugs.R-project.org/show_bug.cgi?id=16940">PR#16940</a>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.3.0</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li> <p><code>nchar(x, *)</code>'s argument <code>keepNA</code> governing how
 the result for <code>NA</code>s in <code>x</code> is determined, gets a new
 default <code>keepNA = NA</code> which returns <code>NA</code> where
 <code>x</code> is <code>NA</code>, except for <code>type = "width"</code> which still
 returns <code>2</code>, the formatting / printing width of <code>NA</code>.
 </p>
 </li>
 <li><p> All builds have support for <span class="samp">https:</span> URLs in the
 default methods for <code>download.file()</code>, <code>url()</code> and code
 making use of them.
 </p>
 <p>Unfortunately that cannot guarantee that any particular
 <span class="samp">https:</span> URL can be accessed.  For example, server and client
 have to successfully negotiate a cryptographic protocol (TLS/SSL,
 ...) and the server's identity has to be verifiable <em>via</em>
 the available certificates.  Different access methods may allow
 different protocols or use private certificate bundles: we
 encountered a <span class="samp">https:</span> CRAN mirror which could be accessed by
 one browser but not by another nor by <code>download.file()</code> on
 the same Linux machine.
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> The <code>print</code> method for <code>methods()</code> gains a
 <code>byclass</code> argument.
 </p>
 </li>
 <li><p> New functions <code>validEnc()</code> and <code>validUTF8()</code> to
 give access to the validity checks for inputs used by
 <code>grep()</code> and friends.
 </p>
 </li>
 <li><p> Experimental new functionality for S3 method checking, notably
 <code>isS3method()</code>.
 </p>
 <p>Also, the names of the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> &lsquo;language elements&rsquo; are exported
 as character vector <code>tools::langElts</code>.
 </p>
 </li>
 <li> <p><code>str(x)</code> now displays <code>"Time-Series"</code> also for
 matrix (multivariate) time-series, i.e. when <code>is.ts(x)</code> is true.
 </p>
 </li>
 <li><p> (Windows only)  The GUI menu item to install local packages
 now accepts &lsquo;<span class="file">*.tar.gz</span>&rsquo; files as well as &lsquo;<span class="file">*.zip</span>&rsquo; files
 (but defaults to the latter).
 </p>
 </li>
 <li><p> New programmeR's utility function <code>chkDots()</code>.
 </p>
 </li>
 <li> <p><code>D()</code> now signals an error when given invalid input,
 rather than silently returning <code>NA</code>.  (Request of John Nash.)
 </p>
 </li>
 <li> <p><code>formula</code> objects are slightly more &ldquo;first class&rdquo;:
 e.g., <code>formula()</code> or <code>new("formula", y ~ x)</code> are now
 valid.  Similarly, for <code>"table"</code>, <code>"ordered"</code> and
 <code>"summary.table"</code>.  Packages defining S4 classes with the above
 S3/S4 classes as slots should be reinstalled.


 </p>
 </li>
 <li><p> New function <code>strrep()</code> for repeating the elements of a
 character vector.
 </p>
 </li>
 <li> <p><code>rapply()</code> preserves attributes on the list when
 <code>how = "replace"</code>.
 </p>
 </li>
 <li><p> New S3 generic function <code>sigma()</code> with methods for
 extracting the estimated standard deviation aka &ldquo;residual
 standard deviation&rdquo; from a fitted model.
 </p>
 </li>
 <li> <p><code>news()</code> now displays <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> and package news files within
 the HTML help system if it is available.  If no news file is found,
 a visible <code>NULL</code> is returned to the console.
 </p>
 </li>
 <li> <p><code>as.raster(x)</code> now also accepts <code>raw</code> arrays
 <code>x</code> assuming values in <code>0:255</code>.
 </p>
 </li>
 <li><p> Subscripting of matrix/array objects of type
 <code>"expression"</code> is now supported.
 </p>
 </li>
 <li> <p><code>type.convert("i")</code> now returns a factor instead of a
 complex value with zero real part and missing imaginary part.
 </p>
 </li>
 <li><p> Graphics devices <code>cairo_pdf()</code> and <code>cairo_ps()</code>
 now allow non-default values of the cairographics &lsquo;fallback
 resolution&rsquo; to be set.
 </p>
 <p>This now defaults to 300 on all platforms: that is the default
 documented by cairographics, but apparently was not used by all
 system installations.
 </p>
 </li>
 <li> <p><code>file()</code> gains an explicit <code>method</code> argument
 rather than implicitly using
 <code>getOption("url.method", "default")</code>.
 </p>
 </li>
 <li><p> Thanks to a patch from Tomas Kalibera, <code>x[x != 0]</code> is
 now typically faster than <code>x[which(x != 0)]</code> (in the case
 where <code>x</code> has no NAs, the two are equivalent).
 </p>
 </li>
 <li> <p><code>read.table()</code> now always uses the names for a named
 <code>colClasses</code> argument (previously names were only used when
 <code>colClasses</code> was too short). (In part, wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16478">PR#16478</a>.)
 </p>
 </li>
 <li><p> (Windows only) <code>download.file()</code> with default
 <code>method = "auto"</code> and a <span class="samp">ftps://</span> URL chooses
 <code>"libcurl"</code> if that is available.
 </p>
 </li>
 <li><p> The out-of-the box Bioconductor mirror has been changed to
 one using <span class="samp">https://</span>: use <code>chooseBioCmirror()</code> to choose
 a <span class="samp">http://</span> mirror if required.
 </p>
 </li>
 <li><p> The data frame and formula methods for <code>aggregate()</code>
 gain a <code>drop</code> argument.
 </p>
 </li>
 <li> <p><code>available.packages()</code> gains a <code>repos</code> argument.
 </p>
 </li>
 <li><p> The undocumented switching of methods for <code>url()</code> on
 <span class="samp">https:</span> and <span class="samp">ftps:</span> URLs is confined to
 <code>method = "default"</code> (and documented).
 </p>
 </li>
 <li> <p><code>smoothScatter()</code> gains a <code>ret.selection</code> argument.
 </p>
 </li>
 <li> <p><code>qr()</code> no longer has a <code>...</code> argument to pass
 additional arguments to methods.
 </p>
 </li>
 <li> <p><code>[</code> has a method for class <code>"table"</code>.
 </p>
 </li>
 <li><p> It is now possible (again) to <code>replayPlot()</code> a
 display list snapshot that was created by <code>recordPlot()</code>
 in a different <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session.
 </p>
 <p>It is still not a good idea to use snapshots as a persistent
 storage format for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> plots, but it is now not completely silly to
 use a snapshot as a format for transferring an R plot between two
 R sessions.
 </p>
 <p>The underlying changes mean that packages providing graphics
 devices (e.g., <a href="https://CRAN.R-project.org/package=Cairo"><span class="pkg">Cairo</span></a>, <a href="https://CRAN.R-project.org/package=RSvgDevice"><span class="pkg">RSvgDevice</span></a>,
 <a href="https://CRAN.R-project.org/package=cairoDevice"><span class="pkg">cairoDevice</span></a>, <a href="https://CRAN.R-project.org/package=tikzDevice"><span class="pkg">tikzDevice</span></a>) will need to be
 reinstalled.
 </p>
 <p>Code for restoring snapshots was contributed by Jeroen Ooms
 and JJ Allaire.
 </p>
 <p>Some testing code is available at
 <a href="https://github.com/pmur002/R-display-list">https://github.com/pmur002/R-display-list</a>.
 </p>
 </li>
 <li> <p><code>tools::undoc(dir = D)</code> and <code>codoc(dir = D)</code> now
 also work when <code>D</code> is a directory whose <code>normalizePath()</code>ed
 version does not end in the package name, e.g. from a symlink.
 </p>
 </li>
 <li> <p><code>abbreviate()</code> has more support for multi-byte
 character sets &ndash; it no longer removes bytes within characters and
 knows about Latin vowels with accents.  It is still only
 really suitable for (most) European languages, and still warns on
 non-ASCII input.
 </p>
 <p><code>abbreviate(use.classes = FALSE)</code> is now implemented, and
 that is more suitable for non-European languages.
 </p>
 </li>
 <li> <p><code>match(x, table)</code> is faster (sometimes by an order of
 magnitude) when <code>x</code> is of length one and <code>incomparables</code>
 is unchanged, thanks to Peter Haverty (<a href="https://bugs.R-project.org/show_bug.cgi?id=16491">PR#16491</a>).
 </p>
 </li>
 <li><p> More consistent, partly not back-compatible behavior of
 <code>NA</code> and <code>NaN</code> coercion to complex numbers, operations
 less often resulting in complex <code>NA</code> (<code>NA_complex_</code>).
 </p>
 </li>
 <li> <p><code>lengths()</code> considers methods for <code>length</code> and
 <code>[[</code> on <code>x</code>, so it should work automatically on any
 objects for which appropriate methods on those generics are
 defined.
 </p>
 </li>
 <li><p> The logic for selecting the default screen device on OS X
 has been simplified: it is now <code>quartz()</code> if that is
 available even if environment variable <span class="env">DISPLAY</span> has been set
 by the user.
 </p>
 <p>The choice can easily be overridden <em>via</em> environment
 variable <span class="env">R_INTERACTIVE_DEVICE</span>.
 </p>
 </li>
 <li><p> On Unix-like platforms which support the <code>getline</code> C
 library function, <code>system(*, intern = TRUE)</code> no longer
 truncates (output) lines longer than 8192 characters, thanks to
 Karl Millar. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16544">PR#16544</a>)
 </p>
 </li>
 <li> <p><code>rank()</code> gains a <code>ties.method = "last"</code> option,
 for convenience (and symmetry).
 </p>


 </li>
 <li> <p><code>regmatches(invert = NA)</code> can now be used to extract
 both non-matched and matched substrings.
 </p>
 </li>
 <li> <p><code>data.frame()</code> gains argument <code>fix.empty.names</code>;
 <code>as.data.frame.list()</code> gets new <code>cut.names</code>,
 <code>col.names</code> and <code>fix.empty.names</code>.
 </p>
 </li>
 <li> <p><code>plot(x ~ x, *)</code> now warns that it is the same as
 <code>plot(x ~ 1, *)</code>.
 </p>
 </li>
 <li> <p><code>recordPlot()</code> has new arguments <code>load</code> and
 <code>attach</code> to allow package names to be stored as part of a
 recorded plot.  <code>replayPlot()</code> has new argument
 <code>reloadPkgs</code> to load/attach any package names that were
 stored as part of a recorded plot.
 </p>
 </li>
 <li><p> S4 dispatch works within calls to <code>.Internal()</code>. This
 means explicit S4 generics are no longer needed for
 <code>unlist()</code> and <code>as.vector()</code>.
 </p>
 </li>
 <li><p> Only font family names starting with <span class="samp">"Hershey"</span> (and
 not <span class="samp">"Her"</span> as before) are given special treatment by the
 graphics engine.
 </p>
 </li>
 <li><p> S4 values are automatically coerced to vector (<em>via</em>
 <code>as.vector</code>) when subassigned into atomic vectors.
 </p>
 </li>
 <li> <p><code>findInterval()</code> gets a <code>left.open</code> option.
 </p>
 </li>
 <li><p> The version of LAPACK included in the sources has been
 updated to 3.6.0, including those &lsquo;deprecated&rsquo; routines
 which were previously included.  <em>Ca</em> 40 double-complex
 routines have been added at the request of a package maintainer.
 </p>
 <p>As before, the details of what is included are in
 &lsquo;<span class="file">src/modules/lapack/README</span>&rsquo; and this now gives information on
 earlier additions.
 </p>
 </li>
 <li> <p><code>tapply()</code> has been made considerably more efficient
 without changing functionality, thanks to proposals from Peter
 Haverty and Suharto Anggono.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16640">PR#16640</a>)
 </p>
 </li>
 <li> <p><code>match.arg(arg)</code> (the one-argument case) is faster; so
 is <code>sort.int()</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16652">PR#16652</a>)
 </p>
 </li>
 <li><p> The <code>format</code> method for <code>object_size</code> objects now
 also accepts &ldquo;binary&rdquo; units such as <code>"KiB"</code> and e.g.,
 <code>"Tb"</code>.  (Partly from <a href="https://bugs.R-project.org/show_bug.cgi?id=16649">PR#16649</a>.)
 </p>
 </li>
 <li><p> Profiling now records calls of the form <code>foo::bar</code> and
 some similar cases directly rather than as calls to
 <code>&lt;Anonymous&gt;</code>.  Contributed by Winston Chang.
 </p>
 </li>
 <li><p> New string utilities <code>startsWith(x, prefix)</code> and
 <code>endsWith(x, suffix)</code>.  Also provide speedups for some
 <code>grepl("^...", *)</code> uses (related to proposals in <a href="https://bugs.R-project.org/show_bug.cgi?id=16490">PR#16490</a>).
 </p>
 </li>
 <li><p> Reference class finalizers run at exit, as well as on
 garbage collection.
 </p>
 </li>
 <li><p> Avoid <span class="pkg">parallel</span> dependency on <span class="pkg">stats</span> for port
 choice and random number seeds.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16668">PR#16668</a>)
 </p>
 </li>
 <li><p> The radix sort algorithm and implementation from
 <a href="https://CRAN.R-project.org/package=data.table"><span class="pkg">data.table</span></a> (<code>forder</code>) replaces the previous radix
 (counting) sort and adds a new method for <code>order()</code>.
 Contributed by Matt Dowle and Arun Srinivasan, the new algorithm
 supports logical, integer (even with large values), real, and
 character vectors.  It outperforms all other methods, but there
 are some caveats (see <code>?sort</code>).
 </p>
 </li>
 <li><p> The <code>order()</code> function gains a <code>method</code> argument
 for choosing between <code>"shell"</code> and <code>"radix"</code>.
 </p>
 </li>
 <li><p> New function <code>grouping()</code> returns a permutation that
 stably rearranges data so that identical values are adjacent.  The
 return value includes extra partitioning information on the
 groups.  The implementation came included with the new radix sort.
 </p>
 </li>
 <li> <p><code>rhyper(nn, m, n, k)</code> no longer returns <code>NA</code> when
 one of the three parameters exceeds the maximal integer.
 </p>
 </li>
 <li> <p><code>switch()</code> now warns when no alternatives are provided.
 </p>
 </li>
 <li> <p><code>parallel::detectCores()</code> now has default <code>logical
 	= TRUE</code> on all platforms &ndash; as this was the default on Windows,
 this change only affects Sparc Solaris.
 </p>
 <p>Option <code>logical = FALSE</code> is now supported on Linux and recent
 versions of OS X (for the latter, thanks to a suggestion of Kyaw
 Sint).
 </p>
 </li>
 <li> <p><code>hist()</code> for <code>"Date"</code> or <code>"POSIXt"</code> objects
 would sometimes give misleading labels on the breaks, as they
 were set to the day before the start of the period being
 displayed.  The display format has been changed, and the shift of the
 start day has been made conditional on <code>right = TRUE</code> (the
 default).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16679">PR#16679</a>)
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> now uses a new version of the logo (donated to the R
 Foundation by RStudio).  It is defined in &lsquo;<span class="file">.svg</span>&rsquo; format, so
 will resize without unnecessary degradation when displayed on
 HTML pages&mdash;there is also a vector PDF version.  Thanks to
 Dirk Eddelbuettel for producing the corresponding X11 icon.
 </p>
 </li>
 <li><p> New function <code>.traceback()</code> returns the stack trace
 which <code>traceback()</code> prints.
 </p>
 </li>
 <li> <p><code>lengths()</code> dispatches internally.
 </p>
 </li>
 <li> <p><code>dotchart()</code> gains a <code>pt.cex</code> argument to control
 the size of points separately from the size of plot labels.
 Thanks to Michael Friendly and Milan Bouchet-Valat for ideas
 and patches.
 </p>
 </li>
 <li> <p><code>as.roman(ch)</code> now correctly deals with more diverse
 character vectors <code>ch</code>; also arithmetic with the resulting
 roman numbers works in more cases.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16779">PR#16779</a>)
 </p>
 </li>
 <li> <p><code>prcomp()</code> gains a new option <code>rank.</code> allowing to
 directly aim for less than <code>min(n,p)</code> PC's.  The
 <code>summary()</code> and its <code>print()</code> method have been amended,
 notably for this case.
 </p>
 </li>
 <li> <p><code>gzcon()</code> gains a new option <code>text</code>, which marks
 the connection as text-oriented (so e.g. <code>pushBack()</code>
 works).  It is still always opened in binary mode.
 </p>
 </li>
 <li><p> The <code>import()</code> namespace directive now accepts an
 argument <code>except</code> which names symbols to exclude from the
 imports. The <code>except</code> expression should evaluate to a
 character vector (after substituting symbols for strings). See
 Writing R Extensions.
 </p>
 </li>
 <li><p> New convenience function <code>Rcmd()</code> in package
 <span class="pkg">tools</span> for invoking <code>R CMD</code> tools from within <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li><p> New functions <code>makevars_user()</code> and
 <code>makevars_site()</code> in package <span class="pkg">tools</span> to determine the
 location of the user and site specific &lsquo;<span class="file">Makevars</span>&rsquo; files for
 customizing package compilation.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> has a new option
 <span class="option">--ignore-vignettes</span> for use with non-Sweave vignettes
 whose <span class="samp">VignetteBuilder</span> package is not available.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now by default checks code usage
 (<em>via</em> <a href="https://CRAN.R-project.org/package=codetools"><span class="pkg">codetools</span></a>) with only the base package
 attached.  Functions from default packages other than <span class="pkg">base</span>
 which are used in the package code but not imported are reported
 as undefined globals, with a suggested addition to the
 <code>NAMESPACE</code> file.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> now also checks DOIs in
 package &lsquo;<span class="file">CITATION</span>&rsquo; and Rd files.
 </p>
 </li>
 <li> <p><code>R CMD Rdconv</code> and <code>R CMD Rd2pdf</code> each have
 a new option <span class="option">--RdMacros=pkglist</span> which allows Rd macros
 to be specified before processing.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> The previously included versions of <code>zlib</code>,
 <code>bzip2</code>, <code>xz</code> and PCRE have been removed, so suitable
 external (usually system) versions are required (see the
 &lsquo;R Installation and Administration&rsquo; manual).
 </p>
 </li>
 <li><p> The unexported and undocumented Windows-only devices
 <code>cairo_bmp()</code>, <code>cairo_png()</code> and <code>cairo_tiff()</code>
 have been removed.  (These devices should be used as
 e.g. <code>bmp(type = "cairo")</code>.)
 </p>
 </li>
 <li><p> (Windows only)  Function <code>setInternet2()</code> has no effect
 and will be removed in due course.  The choice between methods
 <code>"internal"</code> and <code>"wininet"</code> is now made by the
 <code>method</code> arguments of <code>url()</code> and <code>download.file()</code>
 and their defaults can be set <em>via</em> options.  The
 out-of-the-box default remains <code>"wininet"</code> (as it has been
 since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.2).
 </p>
 </li>
 <li> <p><code>[&lt;-</code> with an S4 value into a list currently embeds the
 S4 object into its own list such that the end result is roughly
 equivalent to using <code>[[&lt;-</code>.  That behavior is deprecated.  In
 the future, the S4 value will be coerced to a list with
 <code>as.list()</code>.
 </p>
 </li>
 <li><p> Package <span class="pkg">tools</span>' functions
 <code>package.dependencies()</code>, <code>pkgDepends()</code>, etc are
 deprecated now, mostly in favor of <code>package_dependencies()</code>
 which is both more flexible and efficient.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> Support for very old versions of <code>valgrind</code>
 (e.g., 3.3.0) has been removed.
 </p>
 </li>
 <li><p> The included <code>libtool</code> script (generated by
 <code>configure</code>) has been updated to version 2.4.6 (from 2.2.6a).
 </p>
 </li>
 <li> <p><code>libcurl</code> version 7.28.0 or later with support for the
 <code>https</code> protocol is required for installation (except on
 Windows).
 </p>
 </li>
 <li><p> BSD networking is now required (except on Windows) and so
 <code>capabilities("http/ftp")</code> is always true.
 </p>
 </li>
 <li> <p><code>configure</code> uses <code>pkg-config</code> for PNG, TIFF
 and JPEG where this is available.  This should work better with
 multiple installs and with those using static libraries.
 </p>
 </li>
 <li><p> The minimum supported version of OS X is 10.6 (&lsquo;Snow
 Leopard&rsquo;): even that has been unsupported by Apple since 2012.
 </p>
 </li>
 <li><p> The <code>configure</code> default on OS X is
 <span class="option">--disable-R-framework</span>: enable this if you intend to
 install under &lsquo;<span class="file">/Library/Frameworks</span>&rsquo; and use with <code>R.app</code>.
 </p>
 </li>
 <li><p> The minimum preferred version of PCRE has since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0
 been 8.32 (released in Nov 2012).  Versions 8.10 to 8.31 are now
 deprecated (with warnings from <code>configure</code>), but will
 still be accepted until <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.4.0.
 </p>
 </li>
 <li> <p><code>configure</code> looks for C functions <code>__cospi</code>,
 <code>__sinpi</code> and <code>__tanpi</code> and uses these if <code>cospi</code>
 <em>etc</em> are not found.  (OS X is the main instance.)
 </p>
 </li>
 <li><p> (Windows) R is now built using <code>gcc</code>
 4.9.3.  This build will require recompilation of at least those packages
 that include C++ code, and possibly others.  A build of R-devel using
 the older toolchain will be temporarily available for comparison
 purposes.
 </p>
 <p>During the transition, the environment variable
 <span class="env">R_COMPILED_BY</span> has been defined to indicate which toolchain
 was used to compile R (and hence, which should be used to compile
 code in packages).  The <code>COMPILED_BY</code> variable described below
 will be a permanent replacement for this.
 </p>
 </li>
 <li><p> (Windows) A <code>make</code> and <code>R CMD config</code> variable
 named <code>COMPILED_BY</code> has been added.  This indicates
 which toolchain was used to compile R (and hence, which should be
 used to compile code in packages).
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> The <code>make</code> macro <code>AWK</code> which used to be made
 available to files such as &lsquo;<span class="file">src/Makefile</span>&rsquo; is no longer set.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> The API call <code>logspace_sum</code> introduced in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.0 is
 now remapped as an entry point to <code>Rf_logspace_sum</code>, and its
 first argument has gained a <code>const</code> qualifier.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16470">PR#16470</a>)
 </p>
 <p>Code using it will need to be reinstalled.
 </p>
 <p>Similarly, entry point <code>log1pexp</code> also defined in
 &lsquo;<span class="file">Rmath.h</span>&rsquo; is remapped there to <code>Rf_log1pexp</code>
 </p>
 </li>
 <li> <p><code>R_GE_version</code> has been increased to <code>11</code>.
 </p>
 </li>
 <li><p> New API call <code>R_orderVector1</code>, a faster
 one-argument version of <code>R_orderVector</code>.
 </p>
 </li>
 <li><p> When <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> headers such as &lsquo;<span class="file">R.h</span>&rsquo; and &lsquo;<span class="file">Rmath.h</span>&rsquo; are
 called from C++ code in packages they include the C++ versions of
 system headers such as &lsquo;<span class="file">&lt;cmath&gt;</span>&rsquo; rather than the legacy
 headers such as &lsquo;<span class="file">&lt;math.h&gt;</span>&rsquo;.  (Headers &lsquo;<span class="file">Rinternals.h</span>&rsquo; and
 &lsquo;<span class="file">Rinterface.h</span>&rsquo; already did, and inclusion of system headers
 can still be circumvented by defining <code>NO_C_HEADERS</code>,
 including as from this version for those two headers.)
 </p>
 <p>The manual has long said that <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> headers should <b>not</b> be
 included within an <code>extern "C"</code> block, and almost all the
 packages affected by this change were doing so.
 </p>
 </li>
 <li><p> Including header &lsquo;<span class="file">S.h</span>&rsquo; from C++ code would fail on some
 platforms, and so gives a compilation error on all.
 </p>
 </li>
 <li><p> The deprecated header &lsquo;<span class="file">Rdefines.h</span>&rsquo; is now compatible
 with defining <code>R_NO_REMAP</code>.
 </p>
 </li>
 <li><p> The connections interface now includes a function
 <code>R_GetConnection()</code> which allows packages implementing
 connections to convert R <code>connection</code> objects to
 <code>Rconnection</code> handles. Code which previously
 used the low-level R-internal <code>getConnection()</code> entry point
 should switch.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> C-level <code>asChar(x)</code> is fixed for when <code>x</code> is not a
 vector, and it returns <code>"TRUE"</code>/<code>"FALSE"</code> instead of
 <code>"T"</code>/<code>"F"</code> for logical vectors.
 </p>
 </li>
 <li><p> The first arguments of <code>.colSums()</code> etc (with an
 initial dot) are now named <code>x</code> rather than <code>X</code> (matching
 <code>colSums()</code>): thus error messages are corrected.
 </p>
 </li>
 <li><p> A <code>coef()</code> method for class <code>"maov"</code> has been
 added to allow <code>vcov()</code> to work with multivariate
 results. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16380">PR#16380</a>)
 </p>
 </li>
 <li> <p><code>method = "libcurl"</code> connections signal errors rather
 than retrieving HTTP error pages (where the ISP reports the
 error).
 </p>
 </li>
 <li> <p><code>xpdrows.data.frame()</code> was not checking for unique
 row names; in particular, this affected assignment to non-existing
 rows <em>via</em> numerical indexing. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16570">PR#16570</a>)
 </p>
 </li>
 <li> <p><code>tail.matrix()</code> did not work for zero rows matrices,
 and could produce row &ldquo;labels&rdquo; such as <code>"[1e+05,]"</code>.
 </p>
 </li>
 <li><p> Data frames with a column named <code>"stringsAsFactors"</code>
 now format and print correctly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16580">PR#16580</a>)
 </p>
 </li>
 <li> <p><code>cor()</code> is now guaranteed to return a value with
 absolute value less than or equal to 1. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16638">PR#16638</a>)
 </p>
 </li>
 <li><p> Array subsetting now keeps <code>names(dim(.))</code>.
 </p>
 </li>
 <li><p> Blocking socket connection selection recovers more
 gracefully on signal interrupts.
 </p>
 </li>
 <li><p> The <code>data.frame</code> method of <code>rbind()</code> construction
 <code>row.names</code> works better in borderline integer cases, but
 may change the names assigned.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16666">PR#16666</a>)
 </p>
 </li>
 <li><p> (X11 only)  <code>getGraphicsEvent()</code> miscoded buttons and
 missed mouse motion events.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16700">PR#16700</a>)
 </p>
 </li>
 <li> <p><code>methods(round)</code> now also lists <code>round.POSIXt</code>.
 </p>
 </li>
 <li> <p><code>tar()</code> now works with the default <code>files = NULL</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16716">PR#16716</a>)
 </p>
 </li>
 <li><p> Jumps to outer contexts, for example in error recovery, now
 make intermediate jumps to contexts where <code>on.exit()</code> actions
 are established instead of trying to run all <code>on.exit()</code>
 actions before jumping to the final target. This unwinds the stack
 gradually, releases resources held on the stack, and significantly
 reduces the chance of a segfault when running out of C stack
 space. Error handlers established using
 <code>withCallingHandlers()</code> and <code>options("error")</code>
 specifications are ignored when handling a C stack overflow error
 as attempting one of these would trigger a cascade of C stack
 overflow errors.  (These changes resolve <a href="https://bugs.R-project.org/show_bug.cgi?id=16753">PR#16753</a>.)
 </p>
 </li>
 <li><p> The spacing could be wrong when printing a complex array.
 (Report and patch by Lukas Stadler.)
 </p>
 </li>
 <li> <p><code>pretty(d, n, min.n, *)</code> for date-time objects <code>d</code>
 works again in border cases with large <code>min.n</code>, returns a
 <code>labels</code> attribute also for small-range dates and in such cases
 its returned length is closer to the desired <code>n</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16761">PR#16761</a>)
 Additionally, it finally does cover the range of <code>d</code>, as it
 always claimed.
 </p>
 </li>
 <li> <p><code>tsp(x) &lt;- NULL</code> did not handle correctly objects
 inheriting from both <code>"ts"</code> and <code>"mts"</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16769">PR#16769</a>)
 </p>
 </li>
 <li> <p><code>install.packages()</code> could give false errors when
 <code>options("pkgType")</code> was <code>"binary"</code>.  (Reported by
 Jose Claudio Faria.)
 </p>
 </li>
 <li><p> A bug fix in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.2 fixed problems with <code>locator()</code>
 in X11, but introduced problems in Windows.  Now both should be
 fixed.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15700">PR#15700</a>)
 </p>
 </li>
 <li> <p><code>download.file()</code> with <code>method = "wininet"</code>
 incorrectly warned of download file length difference when
 reported length was unknown. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16805">PR#16805</a>)
 </p>
 </li>
 <li> <p><code>diag(NULL, 1)</code> crashed because of missed type
 checking.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16853">PR#16853</a>)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.5</h3>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>format.POSIXlt()</code> behaved incorrectly in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.4.
 E.g. the output of
 <code>format(as.POSIXlt(paste0(1940:2000, "-01-01"), tz = "CET"),
 	usetz = TRUE)</code> ended in two <code>"CEST"</code> time formats.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.4</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>install.packages()</code> and related functions now give a
 more informative warning when an attempt is made to install a base
 package.
 </p>
 </li>
 <li> <p><code>summary(x)</code> now prints with less rounding when
 <code>x</code> contains infinite values. (Request of <a href="https://bugs.R-project.org/show_bug.cgi?id=16620">PR#16620</a>.)
 </p>
 </li>
 <li> <p><code>provideDimnames()</code> gets an optional <code>unique</code> argument.
 </p>
 </li>
 <li> <p><code>shQuote()</code> gains <code>type = "cmd2"</code> for quoting
 in <code>cmd.exe</code> in Windows.  (Response to <a href="https://bugs.R-project.org/show_bug.cgi?id=16636">PR#16636</a>.)
 </p>
 </li>
 <li><p> The <code>data.frame</code> method of <code>rbind()</code> gains an
 optional argument <code>stringsAsFactors</code> (instead of only
 depending on <code>getOption("stringsAsFactors")</code>).
 </p>
 </li>
 <li> <p><code>smooth(x, *)</code> now also works for long vectors.
 </p>
 </li>
 <li> <p><code>tools::texi2dvi()</code> has a workaround for problems with
 the <code>texi2dvi</code> script supplied by <span class="pkg">texinfo 6.1</span>.
 </p>
 <p>It extracts more error messages from the LaTeX logs when in
 emulation mode.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> will leave a log file
 &lsquo;<span class="file">build_vignettes.log</span>&rsquo; from the re-building of vignettes in
 the &lsquo;<span class="file">.Rcheck</span>&rsquo; directory if there is a problem, and always if
 environment variable <span class="env">_R_CHECK_ALWAYS_LOG_VIGNETTE_OUTPUT_</span> is
 set to a true value.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> Use of <span class="samp">SUPPORT_OPENMP</span> from header &lsquo;<span class="file">Rconfig.h</span>&rsquo; is
 deprecated in favour of the standard OpenMP define <span class="samp">_OPENMP</span>.
 </p>
 <p>(This has been the recommendation in the manual for a while now.)
 </p>
 </li>
 <li><p> The <code>make</code> macro <code>AWK</code> which is long unused by
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> itself but recorded in file &lsquo;<span class="file">etc/Makeconf</span>&rsquo; is deprecated
 and will be removed in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0.
 </p>
 </li>
 <li><p> The C header file &lsquo;<span class="file">S.h</span>&rsquo; is no longer documented: its
 use should be replaced by &lsquo;<span class="file">R.h</span>&rsquo;.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>kmeans(x, centers = &lt;1-row&gt;)</code> now works. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16623">PR#16623</a>)
 </p>
 </li>
 <li> <p><code>Vectorize()</code> now checks for clashes in argument names.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16577">PR#16577</a>)
 </p>
 </li>
 <li> <p><code>file.copy(overwrite = FALSE)</code> would signal a successful
 copy when none had taken place.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16576">PR#16576</a>)
 </p>
 </li>
 <li> <p><code>ngettext()</code> now uses the same default domain as
 <code>gettext()</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=14605">PR#14605</a>)
 </p>
 </li>
 <li> <p><code>array(.., dimnames = *)</code> now warns about
 non-<code>list</code> dimnames and, from <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0, will signal the same
 error for invalid dimnames as <code>matrix()</code> has always done.
 </p>
 </li>
 <li> <p><code>addmargins()</code> now adds dimnames for the extended
 margins in all cases, as always documented.
 </p>
 </li>
 <li> <p><code>heatmap()</code> evaluated its <code>add.expr</code> argument
 in the wrong environment.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16583">PR#16583</a>)
 </p>
 </li>
 <li> <p><code>require()</code> etc now give the correct entry of
 <code>lib.loc</code> in the warning about an old version of a package
 masking a newer required one.
 </p>
 </li>
 <li><p> The internal deparser did not add parentheses when
 necessary,  e.g. before <code>[]</code> or <code>[[]]</code>.  (Reported by
 Lukas Stadler; additional fixes included as well).
 </p>
 </li>
 <li> <p><code>as.data.frame.vector(*, row.names=*)</code> no longer
 produces &lsquo;corrupted&rsquo; data frames from row names of incorrect
 length, but rather warns about them.  This will become an error.
 </p>
 </li>
 <li> <p><code>url</code> connections with <code>method = "libcurl"</code> are
 destroyed properly. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16681">PR#16681</a>)
 </p>
 </li>
 <li> <p><code>withCallingHandler()</code> now (again) handles warnings
 even during S4 generic's argument evaluation.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16111">PR#16111</a>)
 </p>
 </li>
 <li> <p><code>deparse(..., control = "quoteExpressions")</code>
 incorrectly quoted empty expressions.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16686">PR#16686</a>)
 </p>
 </li>
 <li> <p><code>format()</code>ting datetime objects (<code>"POSIX[cl]?t"</code>)
 could segfault or recycle wrongly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16685">PR#16685</a>)
 </p>
 </li>
 <li> <p><code>plot.ts(&lt;matrix&gt;, las = 1)</code> now does use <code>las</code>.
 </p>
 </li>
 <li> <p><code>saveRDS(*, compress = "gzip")</code> now works as
 documented.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16653">PR#16653</a>)
 </p>
 </li>
 <li><p> (Windows only) The <code>Rgui</code> front end did not
 always initialize the console properly, and could cause
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> to crash.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16698">PR#16698</a>)
 </p>
 </li>
 <li> <p><code>dummy.coef.lm()</code> now works in more cases, thanks to a
 proposal by Werner Stahel (<a href="https://bugs.R-project.org/show_bug.cgi?id=16665">PR#16665</a>).  In addition, it now works
 for multivariate linear models (<code>"mlm"</code>, <code>manova</code>)
 thanks to a proposal by Daniel Wollschlaeger.
 </p>
 </li>
 <li><p> The <code>as.hclust()</code> method for <code>"dendrogram"</code>s
 failed often when there were ties in the heights.
 </p>
 </li>
 <li> <p><code>reorder()</code> and <code>midcache.dendrogram()</code> now are
 non-recursive and hence applicable to somewhat deeply nested
 dendrograms, thanks to a proposal by Suharto Anggono in <a href="https://bugs.R-project.org/show_bug.cgi?id=16424">PR#16424</a>.
 </p>
 </li>
 <li> <p><code>cor.test()</code> now calculates very small p values
 more accurately (affecting the result only in extreme not
 statistically relevant cases).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16704">PR#16704</a>)
 </p>
 </li>
 <li> <p><code>smooth(*, do.ends=TRUE)</code> did not always work correctly
 in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> versions between 3.0.0 and 3.2.3.
 </p>
 </li>
 <li> <p><code>pretty(D)</code> for date-time objects <code>D</code> now also
 works well if <code>range(D)</code> is (much) smaller than a second.  In
 the case of only one unique value in <code>D</code>, the pretty range
 now is more symmetric around that value than previously.
 <br />
 Similarly, <code>pretty(dt)</code> no longer returns a length 5 vector
 with duplicated entries for <code>Date</code> objects <code>dt</code> which
 span only a few days.
 </p>
 </li>
 <li><p> The figures in help pages such as <code>?points</code> were
 accidentally damaged, and did not appear in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.3.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16708">PR#16708</a>)
 </p>
 </li>
 <li> <p><code>available.packages()</code> sometimes deleted the wrong
 file when cleaning up temporary files.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16712">PR#16712</a>)
 </p>
 </li>
 <li><p> The <code>X11()</code> device sometimes froze on Red Hat
 Enterprise Linux 6. It now waits for <code>MapNotify</code> events
 instead of <code>Expose</code> events, thanks to Siteshwar
 Vashisht. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16497">PR#16497</a>)
 </p>
 </li>
 <li> <p><code>[dpqr]nbinom(*, size=Inf, mu=.)</code> now works as limit
 case, for &lsquo;dpq&rsquo; as the Poisson.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16727">PR#16727</a>)<br />
 <code>pnbinom()</code> no longer loops infinitely in border cases.
 </p>
 </li>
 <li> <p><code>approxfun(*, method="constant")</code> and hence <code>ecdf()</code>
 which calls the former now correctly &ldquo;predict&rdquo; <code>NaN</code>
 values as <code>NaN</code>.
 </p>
 </li>
 <li> <p><code>summary.data.frame()</code> now displays <code>NA</code>s in
 <code>Date</code> columns in all cases.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16709">PR#16709</a>)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.3</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> Some recently-added Windows time zone names have been added
 to the conversion table used to convert these to Olson names.
 (Including those relating to changes for Russia in Oct 2014, as in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=16503">PR#16503</a>.)
 </p>
 </li>
 <li><p> (Windows) Compatibility information has been added to the
 manifests for &lsquo;<span class="file">Rgui.exe</span>&rsquo;, &lsquo;<span class="file">Rterm.exe</span>&rsquo; and
 &lsquo;<span class="file">Rscript.exe</span>&rsquo;.  This should allow <code>win.version()</code> and
 <code>Sys.info()</code> to report the actual Windows version up to
 Windows 10.
 </p>
 </li>
 <li><p> Windows <code>"wininet"</code> FTP first tries EPSV / PASV mode
 rather than only using active mode (reported by Dan Tenenbaum).

 </p>
 </li>
 <li> <p><code>which.min(x)</code> and <code>which.max(x)</code> may be much
 faster for logical and integer <code>x</code> and now also work for long
 vectors.
 </p>
 </li>
 <li><p> The &lsquo;emulation&rsquo; part of <code>tools::texi2dvi()</code> has
 been somewhat enhanced, including supporting <code>quiet = TRUE</code>.
 It can be selected by <code>texi2dvi = "emulation"</code>.
 </p>
 <p>(Windows) MiKTeX removed its <code>texi2dvi.exe</code> command in Sept
 2015: <code>tools::texi2dvi()</code> tries <code>texify.exe</code> if it is
 not found.
 </p>
 </li>
 <li><p> (Windows only) Shortcuts for printing and saving have
 been added to menus in <code>Rgui.exe</code>.  (Request of <a href="https://bugs.R-project.org/show_bug.cgi?id=16572">PR#16572</a>.)
 </p>
 </li>
 <li> <p><code>loess(..., iterTrace=TRUE)</code> now provides diagnostics
 for robustness iterations, and the <code>print()</code> method for
 <code>summary(&lt;loess&gt;)</code> shows slightly more.
 </p>
 </li>
 <li><p> The included version of PCRE has been updated to 8.38, a
 bug-fix release.
 </p>
 </li>
 <li> <p><code>View()</code> now displays nested data frames in a more
 friendly way.  (Request with patch in <a href="https://bugs.R-project.org/show_bug.cgi?id=15915">PR#15915</a>.)
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The included configuration code for <code>libintl</code> has been
 updated to that from <code>gettext</code> version 0.19.5.1 &mdash; this
 should only affect how an external library is detected (and the
 only known instance is under OpenBSD).  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16464">PR#16464</a>.)
 </p>
 </li>
 <li> <p><code>configure</code> has a new argument
 <span class="option">--disable-java</span> to disable the checks for Java.
 </p>
 </li>
 <li><p> The <code>configure</code> default for <code>MAIN_LDFLAGS</code> has
 been changed for the FreeBSD, NetBSD and Hurd OSes to one more
 likely to work with compilers other than <code>gcc</code> (FreeBSD 10
 defaults to <code>clang</code>).
 </p>
 </li>
 <li> <p><code>configure</code> now supports the OpenMP flags
 <span class="option">-fopenmp=libomp</span> (clang) and <span class="option">-qopenmp</span> (Intel C).
 </p>
 </li>
 <li><p> Various macros can be set to override the default
 behaviour of <code>configure</code> when detecting OpenMP: see file
 &lsquo;<span class="file">config.site</span>&rsquo;.
 </p>
 </li>
 <li><p> Source installation on Windows has been modified to allow
 for MiKTeX installations without <code>texi2dvi.exe</code>.  See file
 &lsquo;<span class="file">MkRules.dist</span>&rsquo;.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>regexpr(pat, x, perl = TRUE)</code> with Python-style named
 capture did not work correctly when <code>x</code> contained <code>NA</code>
 strings.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16484">PR#16484</a>)
 </p>
 </li>
 <li><p> The description of dataset <code>ToothGrowth</code> has been
 improved/corrected.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15953">PR#15953</a>)
 </p>
 </li>
 <li> <p><code>model.tables(type = "means")</code> and hence
 <code>TukeyHSD()</code> now support <code>"aov"</code> fits without an
 intercept term.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16437">PR#16437</a>)
 </p>
 </li>
 <li> <p><code>close()</code> now reports the status of a <code>pipe()</code>
 connection opened with an explicit <code>open</code> argument.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16481">PR#16481</a>)
 </p>
 </li>
 <li><p> Coercing a list without names to a data frame is faster if
 the elements are very long. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16467">PR#16467</a>)
 </p>
 </li>
 <li><p> (Unix-only) Under some rare circumstances piping the output
 from <code>Rscript</code> or <code>R -f</code> could result in
 attempting to close the input file twice, possibly crashing the
 process.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16500">PR#16500</a>)
 </p>
 </li>
 <li><p> (Windows) <code>Sys.info()</code> was out of step with
 <code>win.version()</code> and did not report Windows 8.
 </p>
 </li>
 <li> <p><code>topenv(baseenv())</code> returns <code>baseenv()</code> again as
 in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.0 and earlier.  This also fixes <code>compilerJIT(3)</code>
 when used in &lsquo;<span class="file">.Rprofile</span>&rsquo;.
 </p>
 </li>
 <li> <p><code>detach()</code>ing the <span class="pkg">methods</span> package keeps
 <code>.isMethodsDispatchOn()</code> true, as long as the methods
 namespace is not unloaded.
 </p>
 </li>
 <li><p> Removed some spurious warnings from <code>configure</code> about the
 preprocessor not finding header files.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15989">PR#15989</a>)
 </p>
 </li>
 <li> <p><code>rchisq(*, df=0, ncp=0)</code> now returns <code>0</code> instead
 of <code>NaN</code>, and <code>dchisq(*, df=0, ncp=*)</code> also no longer
 returns <code>NaN</code> in limit cases (where the limit is unique).
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16521">PR#16521</a>)
 </p>
 </li>
 <li> <p><code>pchisq(*, df=0, ncp &gt; 0, log.p=TRUE)</code> no longer
 underflows (for ncp &gt; ~60).
 </p>
 </li>
 <li> <p><code>nchar(x, "w")</code> returned -1 for characters it did not
 know about (e.g. zero-width spaces): it now assumes 1.
 It now knows about most zero-width characters and a few more
 double-width characters.
 </p>
 </li>
 <li><p> Help for <code>which.min()</code> is now more precise about
 behavior with logical arguments.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16532">PR#16532</a>)
 </p>
 </li>
 <li><p> The print width of character strings marked as
 <code>"latin1"</code> or <code>"bytes"</code> was in some cases computed
 incorrectly.
 </p>
 </li>
 <li> <p><code>abbreviate()</code> did not give names to the return value
 if <code>minlength</code> was zero, unlike when it was positive.
 </p>
 </li>
 <li><p> (Windows only) <code>dir.create()</code> did not always warn
 when it failed to create a directory.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16537">PR#16537</a>)
 </p>
 </li>
 <li><p> When operating in a non-UTF-8 multibyte locale
 (e.g. an East Asian locale on Windows), <code>grep()</code> and
 related functions did not handle UTF-8 strings properly.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16264">PR#16264</a>)
 </p>
 </li>
 <li> <p><code>read.dcf()</code> sometimes misread lines longer than 8191
 characters.  (Reported by Hervé Pagès with a patch.)
 </p>
 </li>
 <li> <p><code>within(df, ..)</code> no longer drops columns whose name
 start with a <code>"."</code>.
 </p>
 </li>
 <li><p> The built-in <code>HTTP</code> server converted entire
 <code>Content-Type</code> to lowercase including parameters which
 can cause issues for multi-part form boundaries (<a href="https://bugs.R-project.org/show_bug.cgi?id=16541">PR#16541</a>).
 </p>
 </li>
 <li><p> Modifying slots of S4 objects could fail when the
 <span class="pkg">methods</span> package was not attached. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16545">PR#16545</a>)
 </p>
 </li>
 <li> <p><code>splineDesign(*, outer.ok=TRUE)</code> (<span class="pkg">splines</span>) is better now
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16549">PR#16549</a>), and <code>interpSpline()</code> now allows <code>sparse=TRUE</code>
 for speedup with non-small sizes.
 </p>
 </li>
 <li><p> If the expression in the traceback was too long,
 <code>traceback()</code> did not report the source line number.  (Patch
 by Kirill Müller.)
 </p>
 </li>
 <li><p> The browser did not truncate the display of the function when
 exiting with <code>options("deparse.max.lines")</code> set.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16581">PR#16581</a>)
 </p>
 </li>
 <li><p> When <code>bs(*, Boundary.knots=)</code> had boundary knots inside
 the data range, extrapolation was somewhat off.  (Patch by Trevor
 Hastie.)
 </p>
 </li>
 <li> <p><code>var()</code> and hence <code>sd()</code> warn about
 <code>factor</code> arguments which are deprecated now. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16564">PR#16564</a>)
 </p>
 </li>
 <li> <p><code>loess(*, weights = *)</code> stored wrong weights and hence
 gave slightly wrong predictions for <code>newdata</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16587">PR#16587</a>)
 </p>
 </li>
 <li> <p><code>aperm(a, *)</code> now preserves <code>names(dim(a))</code>.
 </p>
 </li>
 <li> <p><code>poly(x, ..)</code> now works when either <code>raw=TRUE</code> or
 <code>coef</code> is specified.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16597">PR#16597</a>)
 </p>
 </li>
 <li> <p><code>data(package=*)</code> is more careful in determining the path.
 </p>
 </li>
 <li> <p><code>prettyNum(*, decimal.mark, big.mark)</code>: fixed bug
 introduced when fixing <a href="https://bugs.R-project.org/show_bug.cgi?id=16411">PR#16411</a>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.2</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li><p> It is now easier to use secure downloads from
 <span class="samp">https://</span> URLs on builds which support them: no longer do
 non-default options need to be selected to do so.  In particular,
 packages can be installed from repositories which offer
 <span class="samp">https://</span> URLs, and those listed by <code>setRepositories()</code>
 now do so (for some of their mirrors).
 </p>
 <p>Support for <span class="samp">https://</span> URLs is available on Windows, and on
 other platforms if support for <code>libcurl</code> was compiled in and
 if that supports the <code>https</code> protocol (system installations
 can be expected to do).  So <span class="samp">https://</span> support can be
 expected except on rather old OSes (an example being OS X
 &lsquo;Snow Leopard&rsquo;, where a non-system version of
 <code>libcurl</code> can be used).
 </p>
 <p>(Windows only) The default method for accessing URLs <em>via</em>
 <code>download.file()</code> and <code>url()</code> has been changed to be
 <code>"wininet"</code> using Windows API calls.  This changes the way
 proxies need to be set and security settings made: there have been
 some reports of <span class="samp">ftp:</span> sites being inaccessible under the new default
 method (but the previous methods remain available).
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>cmdscale()</code> gets new option <code>list.</code> for increased
 flexibility when a list should be returned.
 </p>
 </li>
 <li> <p><code>configure</code> now supports <code>texinfo</code> version 6.0,
 which (unlike the change from 4.x to 5.0) is a minor update.
 (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16456">PR#16456</a>.)
 </p>
 </li>
 <li><p> (Non-Windows only) <code>download.file()</code> with default
 <code>method = "auto"</code> now chooses <code>"libcurl"</code> if that is
 available and a <span class="samp">https://</span> or <span class="samp">ftps://</span> URL is used.
 </p>
 </li>
 <li><p> (Windows only) <code>setInternet2(TRUE)</code> is now the default.
 The command-line option <code>--internet2</code> and environment
 variable <span class="env">R_WIN_INTERNET2</span> are now ignored.
 </p>
 <p>Thus by default the <code>"internal"</code> method for
 <code>download.file()</code> and <code>url()</code> uses the <code>"wininet"</code>
 method: to revert to the previous default use
 <code>setInternet2(FALSE)</code>.
 </p>
 <p>This means that <span class="samp">https://</span> URLs can be read by default by
 <code>download.file()</code> (they have been readable by <code>file()</code>
 and <code>url()</code> since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.0).
 </p>
 <p>There are implications for how proxies need to be set (see
 <code>?download.file</code>).
 </p>
 </li>
 <li> <p><code>chooseCRANmirror()</code> and <code>chooseBioCmirror()</code> now
 offer HTTPS mirrors in preference to HTTP mirrors.  This changes
 the interpretation of their <code>ind</code> arguments: see their help
 pages.
 </p>
 </li>
 <li> <p><code>capture.output()</code> gets optional arguments <code>type</code>
 and <code>split</code> to pass to <code>sink()</code>, and hence can be used to
 capture messages.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> Header &lsquo;<span class="file">Rconfig.h</span>&rsquo; now defines <code>HAVE_ALLOCA_H</code> if
 the platform has the &lsquo;<span class="file">alloca.h</span>&rsquo; header (it is needed to
 define <code>alloca</code> on Solaris and AIX, at least: see
 &lsquo;Writing R Extensions&rsquo; for how to use it).
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The <code>libtool</code> script generated by
 <code>configure</code> has been modified to support FreeBSD &gt;= 10
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16410">PR#16410</a>).
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> The HTML help page links to demo code failed due to a change
 in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.0.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16432">PR#16432</a>)
 </p>
 </li>
 <li><p> If the <code>na.action</code> argument was used in
 <code>model.frame()</code>, the original data could be
 modified. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16436">PR#16436</a>)
 </p>
 </li>
 <li> <p><code>getGraphicsEvent()</code> could cause a crash if a graphics
 window was closed while it was in use. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16438">PR#16438</a>)
 </p>
 </li>
 <li> <p><code>matrix(x, nr, nc, byrow = TRUE)</code> failed if <code>x</code>
 was an object of type <code>"expression"</code>.
 </p>
 </li>
 <li> <p><code>strptime()</code> could overflow the allocated storage on
 the C stack when the timezone had a non-standard format much
 longer than the standard formats. (Part of <a href="https://bugs.R-project.org/show_bug.cgi?id=16328">PR#16328</a>.)
 </p>
 </li>
 <li> <p><code>options(OutDec = s)</code> now signals a warning (which will
 become an error in the future) when <code>s</code> is not a string with
 exactly one character, as that has been a documented requirement.
 </p>
 </li>
 <li> <p><code>prettyNum()</code> gains a new option <code>input.d.mark</code>
 which together with other changes, e.g., the default for
 <code>decimal.mark</code>, fixes some <code>format()</code>ting variants with
 non-default <code>getOption("OutDec")</code> such as in <a href="https://bugs.R-project.org/show_bug.cgi?id=16411">PR#16411</a>.
 </p>
 </li>
 <li> <p><code>download.packages()</code> failed for <code>type</code> equal to
 either <code>"both"</code> or <code>"binary"</code>.  (Reported by Dan
 Tenenbaum.)
 </p>
 </li>
 <li><p> The <code>dendrogram</code> method of <code>labels()</code> is much more
 efficient for large dendrograms, now using <code>rapply()</code>.
 (Comment #15 of <a href="https://bugs.R-project.org/show_bug.cgi?id=15215">PR#15215</a>)
 </p>
 </li>
 <li><p> The <code>"port"</code> algorithm of <code>nls()</code> could give
 spurious errors.  (Reported by Radford Neal.)
 </p>
 </li>
 <li><p> Reference classes that inherited from reference classes in
 another package could invalidate methods of the inherited
 class. Fixing this requires adding the ability for methods to be
 &ldquo;external&rdquo;, with the object supplied explicitly as the first
 argument, named <code>.self</code>. See &quot;Inter-Package Superclasses&quot;
 in the documentation.
 </p>
 </li>
 <li> <p><code>readBin()</code> could fail on the SPARC architecture
 due to alignment issues.  (Reported by Radford Neal.)
 </p>
 </li>
 <li> <p><code>qt(*, df=Inf, ncp=.)</code> now uses the natural
 <code>qnorm()</code> limit instead of returning <code>NaN</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16475">PR#16475</a>)
 </p>
 </li>
 <li><p> Auto-printing of S3 and S4 values now searches for
 <code>print()</code> in the base namespace and <code>show()</code> in the
 <span class="pkg">methods</span> namespace instead of searching the global
 environment.
 </p>
 </li>
 <li> <p><code>polym()</code> gains a <code>coefs = NULL</code> argument and
 returns class <code>"poly"</code> just like <code>poly()</code> which gets a
 new <code>simple=FALSE</code> option.  They now lead to correct
 <code>predict()</code>ions, e.g., on subsets of the original data.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16239">PR#16239</a>)
 </p>
 </li>
 <li> <p><code>rhyper(nn, &lt;large&gt;)</code> now works correctly. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16489">PR#16489</a>)
 </p>
 </li>
 <li> <p><code>ttkimage()</code> did not (and could not) work so was
 removed. Ditto for <code>tkimage.cget()</code> and
 <code>tkimage.configure()</code>. Added two Ttk widgets and missing
 subcommands for Tk's <code>image</code> command:
 <code>ttkscale()</code>,
 <code>ttkspinbox()</code>,
 <code>tkimage.delete()</code>,
 <code>tkimage.height()</code>,
 <code>tkimage.inuse()</code>,
 <code>tkimage.type()</code>,
 <code>tkimage.types()</code>,
 <code>tkimage.width()</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15372">PR#15372</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=16450">PR#16450</a>)
 </p>
 </li>
 <li> <p><code>getClass("foo")</code> now also returns a class definition when it is
 found in the cache more than once.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.1</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>utf8ToInt()</code> now checks that its input is valid UTF-8
 and returns <code>NA</code> if it is not.
 </p>
 </li>
 <li> <p><code>install.packages()</code> now allows <code>type = "both"</code>
 with <code>repos = NULL</code> if it can infer the type of file.
 </p>
 </li>
 <li> <p><code>nchar(x, *)</code> and <code>nzchar(x)</code> gain a new argument
 <code>keepNA</code> which governs how the result for <code>NA</code>s in
 <code>x</code> is determined.  For <code>nzchar()</code> in general and
 <code>nchar()</code> in the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.x series, the default
 remains <code>FALSE</code> which is fully back compatible.
 From <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0, <code>nchar()</code>'s default will change to
 <code>keepNA = NA</code> and you are advised to consider this for code
 portability.
 </p>
 </li>
 <li> <p><code>news()</code> more flexibly extracts dates from package
 &lsquo;<span class="file">NEWS.Rd</span>&rsquo; files.
 </p>
 </li>
 <li> <p><code>lengths(x)</code> now also works (trivially) for atomic
 <code>x</code> and hence can be used more generally as an efficient
 replacement of <code>sapply(x, length)</code> and similar.
 </p>
 </li>
 <li><p> The included version of PCRE has been updated to 8.37, a
 bug-fix release.
 </p>
 </li>
 <li> <p><code>diag()</code> no longer duplicates a matrix when extracting
 its diagonal.
 </p>
 </li>
 <li> <p><code>as.character.srcref()</code> gains an argument to allow
 characters corresponding to a range of source references to be
 extracted.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>acf()</code> and <code>ccf()</code> now guarantee values strictly
 in <i>[-1,1]</i> (instead of sometimes very slightly outside). <a href="https://bugs.R-project.org/show_bug.cgi?id=15832">PR#15832</a>.
 </p>
 </li>
 <li> <p><code>as.integer("111111111111")</code> now gives NA (with a warning) as
 it does for the corresponding numeric or negative number
 coercions.  Further, <code>as.integer(M + 0.1)</code> now gives <code>M</code>
 (instead of NA) when M is the maximal representable integer.
 </p>
 </li>
 <li><p> On some platforms <code>nchar(x, "c")</code> and <code>nchar(x, "w")</code>
 would return values (possibly <code>NA</code>) for inputs which were
 declared to be UTF-8 but were not, or for invalid strings without a
 marked encoding in a multi-byte locale, rather than give an
 error.  Additional checks have been added to mitigate this.
 </p>
 </li>
 <li> <p><code>apply(a, M, function(u) c(X = ., Y = .))</code> again
 has dimnames containing &quot;X&quot; and &quot;Y&quot; (as in R &lt; 3.2.0).
 </p>
 </li>
 <li><p> (Windows only) In some cases, the <code>--clean</code> option
 to <code>R CMD INSTALL</code> could fail.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16178">PR#16178</a>)
 </p>
 </li>
 <li><p> (Windows only) <code>choose.files()</code> would occasionally
 include characters from the result of an earlier call in
 the result of a later one.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16270">PR#16270</a>)
 </p>
 </li>
 <li><p> A change in <code>RSiteSearch()</code> in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.0 caused it to
 submit invalid URLs.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16329">PR#16329</a>)
 </p>
 </li>
 <li> <p><code>Rscript</code> and command line <code>R</code> silently
 ignored incomplete statements at the end of a script; now they
 are reported as parse errors.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16350">PR#16350</a>)
 </p>
 </li>
 <li><p> Parse data for very long strings was not stored.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16354">PR#16354</a>)
 </p>
 </li>
 <li> <p><code>plotNode()</code>, the workhorse of the <code>plot</code> method
 for <code>"dendrogram"</code>s is no longer recursive, thanks to Suharto
 Anggono, and hence also works for deeply nested dendrograms.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15215">PR#15215</a>)
 </p>
 </li>
 <li><p> The parser could overflow internally when given numbers
 in scientific format with extremely large exponents.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16358">PR#16358</a>)
 </p>
 </li>
 <li><p> If the CRAN mirror was not set, <code>install.packages(type = "both")</code>
 and related functions could repeatedly query the user for it.
 (Part of <a href="https://bugs.R-project.org/show_bug.cgi?id=16362">PR#16362</a>)
 </p>
 </li>
 <li><p> The low-level functions <code>.rowSums()</code> etc. did not
 check the length of their argument, so could segfault. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16367">PR#16367</a>)
 </p>
 </li>
 <li><p> The <code>quietly</code> argument of <code>library()</code> is now
 correctly propagated from <code>.getRequiredPackages2()</code>.
 </p>
 </li>
 <li><p> Under some circumstances using the internal PCRE when
 building <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> from source would cause external libs such as
 <code>-llzma</code> to be omitted from the main link.
 </p>
 </li>
 <li><p> The .Primitive default methods of the logic operators, i.e.,
 <code>!</code>, <code>&amp;</code> and <code>|</code>, now give correct error messages
 when appropriate, e.g., for <code>`&amp;`(TRUE)</code> or <code>`!`()</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16385">PR#16385</a>)
 </p>
 </li>
 <li> <p><code>cummax(x)</code> now correctly propagates <code>NA</code>s also
 when <code>x</code> is of type <code>integer</code> and begins with an <code>NA</code>.
 </p>
 </li>
 <li> <p><code>summaryRprof()</code> could fail when the profile contained
 only two records.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16395">PR#16395</a>)
 </p>
 </li>
 <li><p> HTML vignettes opened using <code>vignette()</code> did not support
 links into the rest of the HTML help system.  (Links worked properly
 when the vignette was opened using <code>browseVignettes()</code> or from
 within the help system.)
 </p>
 </li>
 <li> <p><code>arima(*, xreg = .)</code> (for <i>d &gt;= 1</i>)
 computes estimated variances based on a the number of effective
 observations as in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> version 3.0.1 and earlier.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16278">PR#16278</a>)
 </p>
 </li>
 <li> <p><code>slotNames(.)</code> is now correct for <code>"signature"</code>
 objects (mostly used internally in <span class="pkg">methods</span>).
 </p>
 </li>
 <li><p> On some systems, the first string comparison after
 a locale change would result in <code>NA</code>.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.2.0</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>anyNA()</code> gains a <code>recursive</code> argument.
 </p>
 </li>
 <li><p> When <code>x</code> is missing and <code>names</code> is not false
 (including the default value), <code>Sys.getenv(x, names)</code> returns
 an object of class <code>"Dlist"</code> and hence prints tidily.
 </p>
 </li>
 <li><p> (Windows.)  <code>shell()</code> no longer consults the
 environment variable <span class="env">SHELL</span>: too many systems have been
 encountered where it was set incorrectly (usually to a path where
 software was compiled, not where it was installed).
 <span class="env">R_SHELL</span>, the preferred way to select a non-default shell,
 can be used instead.
 </p>
 </li>
 <li><p> Some unusual arguments to <code>embedFonts()</code> can now be
 specified as character vectors, and the defaults have been changed
 accordingly.
 </p>
 </li>
 <li><p> Functions in the <code>Summary</code> group duplicate less.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15798">PR#15798</a>)
 </p>
 </li>
 <li><p> (Unix-alikes.) <code>system(cmd, input = )</code> now uses
 &lsquo;shell-execution-environment&rsquo; redirection, which will be
 more natural if <code>cmd</code> is not a single command (but requires a
 POSIX-compliant shell). (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15508">PR#15508</a>)
 </p>
 </li>
 <li> <p><code>read.fwf()</code> and <code>read.DIF()</code> gain a
 <code>fileEncoding</code> argument, for convenience.
 </p>
 </li>
 <li><p> Graphics devices can add attributes to their description in
 <code>.Device</code> and <code>.Devices</code>.  Several of those included with
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> use a <code>"filepath"</code> attribute.
 </p>
 </li>
 <li> <p><code>pmatch()</code> uses hashing in more cases and so is faster
 at the expense of using more memory. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15697">PR#15697</a>)
 </p>
 </li>
 <li> <p><code>pairs()</code> gains new arguments to select sets of
 variables to be plotted against each other.
 </p>
 </li>
 <li> <p><code>file.info(, extra_cols = FALSE)</code> allows a minimal set
 of columns to be computed on Unix-alikes: on some systems without
 properly-configured caching this can be significantly faster with
 large file lists.
 </p>
 </li>
 <li><p> New function <code>dir.exists()</code> in package <span class="pkg">base</span> to
 test efficiently whether one or more paths exist and are
 directories.
 </p>
 </li>
 <li> <p><code>dput()</code> and friends gain new controls
 <span class="samp">hexNumeric</span> and <span class="samp">digits17</span> which output double and
 complex quantities as, respectively, binary fractions (exactly,
 see <code>sprintf("%a")</code>) and as decimals with up to 17
 significant digits.
 </p>
 </li>
 <li> <p><code>save()</code>, <code>saveRDS()</code> and <code>serialize()</code> now
 support <code>ascii = NA</code> which writes ASCII files using
 <code>sprintf("%a")</code> for double/complex quantities.  This is
 read-compatible with <code>ascii = TRUE</code> but avoids
 binary-&gt;decimal-&gt;binary conversions with potential loss of
 precision.  Unfortunately the Windows C runtime's lack of C99
 compliance means that the format cannot be read correctly there in
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> before 3.1.2.
 </p>
 </li>
 <li><p> The default for <code>formatC(decimal.mark =)</code> has been
 changed to be <code>getOption("OutDec")</code>; this makes it more
 consistent with <code>format()</code> and suitable for use in print
 methods, e.g. those for classes <code>"density"</code>, <code>"ecdf"</code>,
 <code>"stepfun"</code> and <code>"summary.lm"</code>.
 </p>
 <p><code>getOption("OutDec")</code> is now consulted by the print method
 for class <code>"kmeans"</code>, by <code>cut()</code>, <code>dendrogram()</code>,
 <code>plot.ts()</code> and <code>quantile()</code> when constructing labels
 and for the report from <code>legend(trace = TRUE)</code>.
 </p>
 <p>(In part, wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15819">PR#15819</a>.)
 </p>
 </li>
 <li> <p><code>printNum()</code> and hence <code>format()</code> and
 <code>formatC()</code> give a warning if <code>big.mark</code> and
 <code>decimal.mark</code> are set to the same value (period and comma
 are not uncommonly used for each, and this is a check that
 conventions have not got mixed).
 </p>
 </li>
 <li> <p><code>merge()</code> can create a result which uses long vectors
 on 64-bit platforms.
 </p>
 </li>
 <li> <p><code>dget()</code> gains a new argument <code>keep.source</code> which
 defaults to <code>FALSE</code> for speed (<code>dput()</code> and
 <code>dget()</code> are most often used for data objects where this can
 make <code>dget()</code> many times faster).
 </p>
 </li>
 <li><p> Packages may now use a file of common macro definitions in
 their help files, and may import definitions from other packages.
 </p>
 </li>
 <li><p> A number of macros have been added in the new
 &lsquo;<span class="file">share/Rd</span>&rsquo; directory for use in package overview help pages,
 and <code>promptPackage()</code> now makes use of them.
 </p>
 </li>
 <li> <p><code>tools::parse_Rd()</code> gains a new <code>permissive</code>
 argument which converts unrecognized macros into text.  This is
 used by <code>utils:::format.bibentry</code> to allow LaTeX markup to be
 ignored.
 </p>
 </li>
 <li> <p><code>options(OutDec =)</code> can now specify a multi-byte
 character, e.g., <code>options(OutDec = "\u00b7")</code> in a UTF-8
 locale.
 </p>
 </li>
 <li> <p><code>is.recursive(x)</code> is no longer true when <code>x</code> is an
 external pointer, a weak reference or byte code; the first enables
 <code>all.equal(x, x)</code> when <code>x &lt;- getClass(.)</code>.
 </p>
 </li>
 <li> <p><code>ls()</code> (aka <code>objects()</code>) and
 <code>as.list.environment()</code> gain a new argument <code>sorted</code>.
 </p>
 </li>
 <li><p> The <code>"source"</code> attribute (which has not been added to
 functions by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> since before <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> version 2.14.0) is no longer
 treated as special.
 </p>
 </li>
 <li><p> Function <code>returnValue()</code> has been added to give
 <code>on.exit()</code> code access to a function's return value for
 debugging purposes.
 </p>
 </li>
 <li> <p><code>crossprod(x, y)</code> allows more matrix coercions when
 <code>x</code> or <code>y</code> are vectors, now equalling
 <code>t(x) %*% y</code> in these cases (also reported by Radford Neal).
 Similarly, <code>tcrossprod(x,y)</code> and <code>%*%</code> work in more
 cases with vector arguments.
 </p>
 </li>
 <li><p> Utility function <code>dynGet()</code> useful for detecting
 cycles, aka infinite recursions.
 </p>
 </li>
 <li><p> The byte-code compiler and interpreter include new
 instructions that allow many scalar subsetting and assignment and
 scalar arithmetic operations to be handled more efficiently. This
 can result in significant performance improvements in scalar
 numerical code.
 </p>
 </li>
 <li> <p><code>apply(m, 2, identity)</code> is now the same as the matrix
 <code>m</code> when it has <em>named</em> row names.
 </p>
 </li>
 <li><p> A new function <code>debuggingState()</code> has been added,
 allowing to temporarily turn off debugging.
 </p>
 </li>
 <li> <p><code>example()</code> gets a new optional argument
 <code>run.donttest</code> and <code>tools::Rd2ex()</code> a corresponding
 <code>commentDonttest</code>, with a default such that
 <code>example(..)</code> in help examples will run <code>\donttest</code>
 code only if used interactively (a change in behaviour).
 </p>
 </li>
 <li> <p><code>rbind.data.frame()</code> gains an optional argument
 <code>make.row.names</code>, for potential speedup.
 </p>
 </li>
 <li><p> New function <code>extSoftVersion()</code> to report on the
 versions of third-party software in use in this session.
 Currently reports versions of <code>zlib</code>, <code>bzlib</code>, the
 <code>liblzma</code> from <code>xz</code>, PCRE, ICU, TRE and the <code>iconv</code>
 implementation.
 </p>
 <p>A similar function <code>grSoftVersion()</code> in package <span class="pkg">grDevices</span>
 reports on third-party graphics software.
 </p>
 <p>Function <code>tcltk::tclVersion()</code> reports the Tcl/Tk version.
 </p>
 </li>
 <li><p> Calling <code>callGeneric()</code> without arguments now works
 with primitive generics to some extent.
 </p>
 </li>
 <li> <p><code>vapply(x, FUN, FUN.VALUE)</code> is more efficient notably
 for large <code>length(FUN.VALUE)</code>; as extension of <a href="https://bugs.R-project.org/show_bug.cgi?id=16061">PR#16061</a>.
 </p>
 </li>
 <li> <p><code>as.table()</code> now allows tables with one or more
 dimensions of length 0 (such as <code>as.table(integer())</code>).
 </p>
 </li>
 <li> <p><code>names(x) &lt;- NULL</code> now clears the names of call
 and <code>...</code> objects.
 </p>
 </li>
 <li> <p><code>library()</code> will report a warning when an insufficient
 dependency version is masking a sufficient one later on the
 library search path.
 </p>
 </li>
 <li><p> A new <code>plot()</code> method for class <code>"raster"</code> has
 been added.
 </p>
 </li>
 <li><p> New <code>check_packages_in_dir_changes()</code> function in
 package <span class="pkg">tools</span> for conveniently analyzing how changing
 sources impacts the check results of their reverse dependencies.
 </p>
 </li>
 <li><p> Speed-up from Peter Haverty for <code>ls()</code> and
 <code>methods:::.requirePackage()</code> speeding up package loading.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16133">PR#16133</a>)
 </p>
 </li>
 <li><p> New <code>get0()</code> function, combining <code>exists()</code> and
 <code>get()</code> in one call, for efficiency.
 </p>
 </li>
 <li> <p><code>match.call()</code> gains an <code>envir</code> argument for
 specifying the environment from which to retrieve
 the <code>...</code> in the call, if any; this environment was wrong
 (or at least undesirable) when the <code>definition</code> argument was
 a function.
 </p>
 </li>
 <li> <p><code>topenv()</code> has been made <code>.Internal()</code> for
 speedup, based on Peter Haverty's proposal in <a href="https://bugs.R-project.org/show_bug.cgi?id=16140">PR#16140</a>.
 </p>
 </li>
 <li> <p><code>getOption()</code> no longer calls <code>options()</code> in the main
 case.
 </p>
 </li>
 <li><p> Optional use of <code>libcurl</code> (version 7.28.0 from Oct 2012
 or later) for Internet access:
 </p>

 <ul>
 <li> <p><code>capabilities("libcurl")</code> reports if this is
 available.
 </p>
 </li>
 <li> <p><code>libcurlVersion()</code> reports the version in use, and
 other details of the <code>"libcurl"</code> build including which URL
 schemes it supports.
 </p>
 </li>
 <li> <p><code>curlGetHeaders()</code> retrieves the headers for
 <span class="samp">http://</span>, <span class="samp">https://</span>, <span class="samp">ftp://</span> and
 <span class="samp">ftps://</span> URLs: analysis of these headers can provide
 insights into the &lsquo;existence&rsquo; of a URL (it might for example be
 permanently redirected) and is so used in <code>R CMD
 	  check --as-cran</code>.
 </p>
 </li>
 <li> <p><code>download.file()</code> has a new optional method
 <code>"libcurl"</code> which will handle more URL schemes, follow
 redirections, and allows simultaneous downloads of multiple
 URLs.
 </p>
 </li>
 <li> <p><code>url()</code> has a new method <code>"libcurl"</code> which
 handles more URL schemes and follows redirections.  The default
 method is controlled by a new option <code>url.method</code>, which
 applies also to the opening of URLs <em>via</em> <code>file()</code>
 (which happens implicitly in functions such as
 <code>read.table</code>.)
 </p>
 </li>
 <li><p> When <code>file()</code> or <code>url()</code> is invoked with a
 <code>https://</code> or <code>ftps://</code> URL which the current method
 cannot handle, it switches to a suitable method if one is
 available.
 </p>
 </li></ul>

 </li>
 <li><p> (Windows.) The DLLs &lsquo;<span class="file">internet.dll</span>&rsquo; and
 &lsquo;<span class="file">internet2.dll</span>&rsquo; have been merged.  In this version it is safe
 to switch (repeatedly) between the internal and Windows internet
 functions within an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session.
 </p>
 <p>The Windows internet functions are still selected by flag
 <span class="option">--internet2</span> or <code>setInternet2()</code>.  This can be
 overridden for an <code>url()</code> connection <em>via</em> its new
 <code>method</code> argument.
 </p>
 <p><code>download.file()</code> has new method <code>"wininet"</code>, selected
 as the default by <span class="option">--internet2</span> or <code>setInternet2()</code>.
 </p>
 </li>
 <li> <p><code>parent.env&lt;-</code> can no longer modify the parent of a
 locked namespace or namespace imports environment.  Contributed by
 Karl Millar.
 </p>
 </li>
 <li><p> New function <code>isNamespaceLoaded()</code> for readability and speed.
 </p>
 </li>
 <li> <p><code>names(env)</code> now returns all the object names of an
 <code>environment</code> <code>env</code>, equivalently to
 <code>ls(env, all.names = TRUE, sorted = FALSE)</code> and also to
 the names of the corresponding list,
 <code>names(as.list(env, all.names = TRUE))</code>.
 Note that although <code>names()</code> returns a character vector, the
 names have no particular ordering.
 </p>
 </li>
 <li><p> The memory manager now grows the heap more aggressively. This
 reduces the number of garbage collections, in particular while
 data or code are loaded, at the expense of slightly increasing the
 memory footprint.
 </p>
 </li>
 <li><p> New function <code>trimws()</code> for removing leading/trailing
 whitespace.
 </p>
 </li>
 <li> <p><code>cbind()</code> and <code>rbind()</code> now consider S4 inheritance
 during S3 dispatch and also obey <code>deparse.level</code>.
 </p>
 </li>
 <li> <p><code>cbind()</code> and <code>rbind()</code> will delegate recursively
 to <code>methods::cbind2</code> (<code>methods::rbind2</code>) when at least
 one argument is an S4 object and S3 dispatch fails (due to
 ambiguity).
 </p>
 </li>
 <li><p> (Windows.)  <code>download.file(quiet = FALSE)</code> now uses
 text rather than Windows progress bars in non-interactive use.
 </p>
 </li>
 <li><p> New function <code>hsearch_db()</code> in package <span class="pkg">utils</span> for
 building and retrieving the help search database used by
 <code>help.search()</code>, along with functions for inspecting the
 concepts and keywords in the help search database.
 </p>
 </li>
 <li><p> New function <code>.getNamespaceInfo()</code>, a no-check version
 of <code>getNamespaceInfo()</code> mostly for internal speedups.
 </p>
 </li>
 <li><p> The help search system now takes <span class="samp">\keyword</span> entries in
 Rd files which are not standard keywords (as given in
 &lsquo;<span class="file">KEYWORDS</span>&rsquo; in the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> documentation directory) as concepts.
 For standard keyword entries the corresponding descriptions are
 additionally taken as concepts.
 </p>
 </li>
 <li><p> New <code>lengths()</code> function for getting the lengths of all
 elements in a list.
 </p>
 </li>
 <li><p> New function <code>toTitleCase()</code> in package <span class="pkg">tools</span>,
 tailored to package titles.
 </p>
 </li>
 <li><p> The matrix methods of <code>cbind()</code> and <code>rbind()</code>
 allow matrices as inputs which have <i>2^31</i> or more
 elements.  (For <code>cbind()</code>, wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16198">PR#16198</a>.)
 </p>
 </li>
 <li><p> The default method of <code>image()</code> has an explicit check
 for a numeric or logical matrix (which was always required).
 </p>
 </li>
 <li> <p><code>URLencode()</code> will not by default encode further URLs
 which appear to be already encoded.
 </p>
 </li>
 <li> <p><code>BIC(mod)</code> and <code>BIC(mod, mod2)</code> now give non-NA
 numbers for <code>arima()</code> fitted models, as <code>nobs(mod)</code> now
 gives the number of &ldquo;used&rdquo; observations for such models.
 This fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=16198">PR#16198</a>, quite differently than proposed there.
 </p>
 </li>
 <li><p> The <code>print()</code> methods for <code>"htest"</code>,
 <code>"pairwise.htest"</code> and <code>"power.htest"</code> objects now have
 a <code>digits</code> argument defaulting to (a function of)
 <code>getOption("digits")</code>, and influencing all printed numbers
 coherently.  Unavoidably, this changes the display of such test
 results in some cases.
 </p>
 </li>
 <li><p> Code completion for namespaces now recognizes all loaded
 namespaces, rather than only the ones that are also attached.
 </p>
 </li>
 <li><p> The code completion mechanism can now be replaced by a
 user-specified completer function, for (temporary) situations
 where the usual code completion is inappropriate.
 </p>
 </li>
 <li> <p><code>unzip()</code> will now warn if it is able to detect truncation
 when unpacking a file of 4GB or more (related to <a href="https://bugs.R-project.org/show_bug.cgi?id=16243">PR#16243</a>).
 </p>
 </li>
 <li> <p><code>methods()</code> reports S4 in addition to S3 methods;
 output is simplified when the <code>class</code> argument is
 used.  <code>.S3methods()</code> and <code>methods::.S4methods()</code> report
 S3 and S4 methods separately.
 </p>
 </li>
 <li><p> Higher order functions such as the <code>apply</code> functions
 and <code>Reduce()</code> now force arguments to the functions they apply
 in order to eliminate undesirable interactions between lazy
 evaluation and variable capture in closures.  This resolves
 <a href="https://bugs.R-project.org/show_bug.cgi?id=16093">PR#16093</a>.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The <code>\donttest</code> sections of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>'s help files can be
 tested by<br />
 <code>make check TEST_DONTTEST=TRUE</code> .
 </p>
 </li>
 <li><p> It is possible to request the use of system
 <code>valgrind</code> headers <em>via</em> <code>configure</code> option
 <span class="option">--with-system-valgrind-headers</span>: note the possible future
 incompatibility of such headers discussed in the 'R Installation
 and Administration' manual. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16068">PR#16068</a>.)
 </p>
 </li>
 <li><p> The included version of <code>liblzma</code> has been updated to
 <code>xz-utils</code> 5.0.7 (minor bug fixes from 5.0.5).
 </p>
 </li>
 <li> <p><code>configure</code> options <span class="option">--with-system-zlib</span>,
 <span class="option">--with-system-bzlib</span> and <span class="option">--with-system-pcre</span> are
 now the default.  For the time being there is fallback to the
 versions included in the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> sources if no system versions are
 found or (unlikely) if they are too old.
 </p>
 <p>Linux users should check that the <code>-devel</code> or <code>-dev</code>
 versions of packages <span class="pkg">zlib</span>, <span class="pkg">bzip2</span>/<span class="pkg">libbz2</span> and
 <span class="pkg">pcre</span> as well as <span class="pkg">xz-devel</span>/<span class="pkg">liblzma-dev</span> (or
 similar names) are installed.
 </p>
 </li>
 <li> <p><code>configure</code> by default looks for the
 <code>texi2any</code> script from <span class="pkg">texinfo</span> 5.1 or later, rather
 than the <code>makeinfo</code> program.  (<code>makeinfo</code> is a
 link to the Perl script <code>texi2any</code> in <span class="pkg">texinfo</span> 5.x.)
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> gains an  option
 <span class="option">--built-timestamp=STAMP</span> allowing 100% reproducible
 package building, thanks to Dirk Eddelbuettel.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li><p> There is support for testing the <code>\dontrun</code> and
 <code>\donttest</code> parts of examples in packages.
 </p>
 <p><code>tools::testInstalledPackage()</code> accepts new arguments
 <code>commentDontrun = FALSE</code> and <code>commentDonttest = FALSE</code>.
 </p>
 <p><code>R CMD check</code> gains options <span class="option">--run-dontrun</span> and
 <span class="option">--run-donttest</span>.
 </p>
 </li>
 <li><p> The <acronym><span class="acronym">HTML</span></acronym> generated by <code>tools::Rd2HTML()</code> and
 <code>tools::toHTML()</code> methods is now
 &lsquo;<acronym><span class="acronym">XHTML</span></acronym> 1.0 Strict&rsquo;.
 </p>
 </li>
 <li><p> The <span class="pkg">compiler</span> package's utility function
 <code>setCompilerOptions()</code> now returns the old values
 invisibly. The initial optimization level can also be set with the
 environment variable <span class="env">R_COMPILER_OPTIMIZE</span>.
 </p>
 </li>
 <li> <p><code>R CMD build</code> adds a <span class="samp">NeedsCompilation</span> field
 if one is not already present in the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file.
 </p>
 </li>
 <li> <p><code>R CMD check</code> gains option <span class="option">--test-dir</span> to
 specify an alternative set of tests to run.
 </p>
 </li>
 <li> <p><code>R CMD check</code> will now by default continue with
 testing after many types of errors, and will output a summary
 count of errors at the end if any have occurred.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now checks that the <span class="samp">Title</span> and
 <span class="samp">Description</span> fields are correctly terminated.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> now:
 </p>

 <ul>
 <li><p> checks a &lsquo;<span class="file">README.md</span>&rsquo; file can be processed: this
 needs <code>pandoc</code> installed.
 </p>
 </li>
 <li><p> checks the existence and accessibility of URLs in the
 &lsquo;<span class="file">DESCRIPTION</span>&rsquo;, &lsquo;<span class="file">CITATION</span>&rsquo;, &lsquo;<span class="file">NEWS.Rd</span>&rsquo; and
 &lsquo;<span class="file">README.md</span>&rsquo; files and in the help files (provided the build
 has <code>libcurl</code> support).
 </p>
 </li>
 <li><p> reports non-ASCII characters in R source files when there
 is no package encoding declared in the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file.
 </p>
 </li>
 <li><p> reports (apparent) S3 methods exported but not registered.
 </p>
 </li>
 <li><p> reports overwriting registered S3 methods from
 base/recommended packages.  (Such methods are replaced in the
 affected package for the rest of the session, even if the
 replacing namespace is unloaded.)
 </p>
 </li>
 <li><p> reports if the <code>Title</code> field does not appear to be in
 title case (see &lsquo;Writing R Extensions&rsquo;: there may be
 false positives, but note that technical words should be
 single-quoted and will then be accepted).
 </p>
 </li></ul>

 <p>Most of these checks can also be selected by environment
 variables: see the &lsquo;R Internals&rsquo; manual.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> New C API utility <code>logspace_sum(logx[], n)</code>.
 </p>
 </li>
 <li><p> Entry points <code>rbinom_mu</code>, <code>rnbinom_mu</code> and
 <code>rmultinom</code> are remapped (by default) to <code>Rf_rbinom_mu</code>
 etc.  This requires packages using them to be re-installed.
 </p>
 </li>
 <li> <p><code>.C(DUP = FALSE)</code> and <code>.Fortran(DUP = FALSE)</code> are
 now ignored, so arguments are duplicated if <code>DUP = TRUE</code>
 would do so.  As their help has long said, <code>.Call()</code> is much
 preferred.
 </p>
 </li>
 <li><p> New entry point <code>R_allocLD</code>, like <code>R_alloc</code> but
 guaranteed to have sufficient alignment for <code>long double</code>
 pointers.
 </p>
 </li>
 <li> <p><code>isPairList()</code> now returns <code>TRUE</code> for DOTSXP.
 </p>
 </li></ul>


 <h4>WINDOWS BUILD CHANGES</h4>

 <p>A number of changes to the Windows build system are in development.
 The following are currently in place.
 </p>

 <ul>
 <li><p> Installation using external binary distributions
 of <span class="pkg">zlib</span>, <span class="pkg">bzip2</span>, <span class="pkg">liblzma</span>, <span class="pkg">pcre</span>,
 <span class="pkg">libpng</span>, <span class="pkg">jpeglib</span> and <span class="pkg">libtiff</span> is now required,
 and the build instructions have been revised.
 </p>
 </li>
 <li><p> A new <code>make</code> target <code>rsync-extsoft</code> has been
 added to obtain copies of the external libraries from
 <acronym><span class="acronym">CRAN</span></acronym>.
 </p>
 </li>
 <li><p> Building the manuals now requires <code>texi2any</code> from
 <span class="pkg">texinfo</span> 5.1 or later.  <acronym><span class="acronym">CRAN</span></acronym> binary builds include
 the manuals, but by default builds from source will not, and they
 will be accessed from <acronym><span class="acronym">CRAN</span></acronym>.  See the comments in
 &lsquo;<span class="file">src/gnuwin32/MkRules.dist</span>&rsquo; for how to specify the location
 of <code>texi2any</code>.
 </p>
 </li>
 <li><p> (Windows) Changes have been made to support an experimental
 Windows toolchain based on GCC 4.9.2.  The default toolchain
 continues to be based on GCC 4.6.3, as the new toolchain is not
 yet stable enough.  A change to a new toolchain is expected during
 the R 3.2.x lifetime.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p>  (Windows) The use of macro <code>ZLIB_LIBS</code> in file
 &lsquo;<span class="file">src/Makevars.win</span>&rsquo; (which has not been documented for a long
 time) now requires an external &lsquo;<span class="file">libz.a</span>&rsquo; to be available (it is
 part of the &lsquo;goodies&rsquo; used to compile Windows binary
 packages).  It would be simpler to use <code>-lz</code> instead.
 </p>
 </li>
 <li><p> The default for option <code>pkgType</code> on platforms using
 binary packages is now <code>"both"</code>, so source packages will be
 tried if binary versions are not available or not up to date.
 </p>
 <p>There are options for what <code>install.packages(type = "both")</code>
 (possibly called <em>via</em> <code>update.packages()</code>) will do if
 compilation of a source package is desirable: see <code>?options</code>
 (under <span class="pkg">utils</span>).
 </p>
 <p>If you intend not to accept updates as source packages, you should
 use <code>update.packages(type = "binary")</code>.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li> <p><code>download.file(method = "lynx")</code> is defunct.
 </p>
 </li>
 <li><p> Building <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> using the included versions of <code>zlib</code>,
 <code>bzip2</code>, <code>xz</code> and PCRE is deprecated: these are frozen
 (bar essential bug-fixes) and will be removed for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.3.0.
 </p>
 </li>
 <li><p> The <code>configure</code> option
 <span class="option">--with-valgrind-instrumentation=3</span> has been withdrawn, as
 it did not work with recent <code>valgrind</code> headers: it is now
 treated as level <code>2</code>.
 </p>
 </li>
 <li><p> The <code>MethodsList</code> class in package <span class="pkg">methods</span> had
 been deprecated in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.11.0 and is defunct now.  Functions using
 it are defunct if they had been deprecated in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.11.0, and are
 deprecated now, otherwise.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> Fixed two obscure bugs in pairlist subassignment, reported by
 Radford Neal as part of pqR issue 16.
 </p>
 </li>
 <li><p> Fixes for bugs in handling empty arguments and argument
 matching by name in <code>log()</code>.
 </p>
 </li>
 <li> <p><code>all.equal()</code> gains methods for <code>environment</code>s and
 <code>refClass</code>es.
 </p>
 </li>
 <li> <p><code>[&lt;-</code> and <code>[[&lt;-</code> gain S4 <code>data.frame</code> methods
 to avoid corruption of S4 class information by the S3 methods.
 </p>
 </li>
 <li> <p><code>callNextMethod()</code> should now work within a <code>.local</code>
 call when <code>...</code> is absent from <code>formals(.local)</code>.
 </p>
 </li>
 <li> <p><code>dput(pairlist(x))</code> generates a call to the
 <code>pairlist</code> constructor instead of the <code>list</code>
 constructor.
 </p>
 </li>
 <li><p> Fix <code>missing()</code> when arguments are propagated through
 <code>...</code> . (<a href="https://bugs.R-project.org/show_bug.cgi?id=15707">PR#15707</a>)
 </p>
 </li>
 <li> <p><code>eigen(m)</code> now defaults to <code>symmetric = TRUE</code> even
 when the dimnames are asymmetric if the matrix is otherwise
 symmetric.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16151">PR#16151</a>)
 </p>
 </li>
 <li><p> Fix issues with forwarding <code>...</code> through
 <code>callGeneric()</code> and <code>callNextMethod()</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16141">PR#16141</a>)
 </p>
 </li>
 <li> <p><code>callGeneric()</code> now works after a <code>callNextMethod()</code>.
 </p>
 </li>
 <li><p> Subclass information is kept consistent when replacing an
 ordinary S4 class with an &ldquo;old class&rdquo; <em>via</em> the
 <code>S4Class</code> argument to <code>setOldClass()</code>. Thus, for
 example, a <code>data.frame</code> is valid for a <code>list</code> argument
 in the signature, and a <code>factor</code> is valid for <code>vector</code>
 arguments.
 </p>
 </li>
 <li><p> In <code>qbeta()</code> the inversion of <code>pbeta()</code> is much
 more sophisticated.  This works better in corner cases some of
 which failed completely previously (<a href="https://bugs.R-project.org/show_bug.cgi?id=15755">PR#15755</a>), or were using
 too many iterations.
 </p>
 </li>
 <li><p> Auto-printing no longer duplicates objects when printing is
 dispatched to a method.
 </p>
 </li>
 <li> <p><code>kmeans(x, k)</code> would fail when <code>nrow(x) &gt;= 42949673</code>.
 (Comment 6 of <a href="https://bugs.R-project.org/show_bug.cgi?id=15364">PR#15364</a>)
 </p>
 </li>
 <li> <p>&lsquo;Abbreviated&rsquo; locale-specific day and month names
 could have been truncated in those rare locales where there are
 the same as the full names.
 </p>
 </li>
 <li><p> An irrelevant warning message from updating subclass
 information was silenced (the namespace would not be writable
 in this case).
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.1.3</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> The internal method of <code>download.file()</code> can now handle
 files larger than 2GB on 32-bit builds which support such files
 (tested on 32-bit <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> running on 64-bit Windows).
 </p>
 </li>
 <li> <p><code>kruskal.test()</code> warns on more types of suspicious input.
 </p>
 </li>
 <li><p> The <code>as.dendrogram()</code> method for <code>"hclust"</code>
 objects gains a <code>check</code> argument protecting against memory
 explosion for invalid inputs.
 </p>
 </li>
 <li> <p><code>capabilities()</code> has a new item <code>long.double</code>
 which indicates if the build uses a <code>long double</code> type which
 is longer than <code>double</code>.
 </p>
 </li>
 <li> <p><code>nlm()</code> no longer modifies the callback argument in
 place (a new vector is allocated for each invocation, which mimics
 the implicit duplication that occurred in R &lt; 3.1.0); note that
 this is a change from the previously documented
 behavior. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15958">PR#15958</a>)
 </p>
 </li>
 <li> <p><code>icuSetCollate()</code> now accepts <code>locale = "ASCII"</code>
 which uses the basic C function <code>strcmp</code> and so collates
 strings byte-by-byte in numerical order.
 </p>
 </li>
 <li> <p><code>sessionInfo()</code> tries to report the OS version in use
 (not just that compiled under, and including details of Linux
 distributions).
 </p>
 </li>
 <li> <p><code>model.frame()</code> (used by <code>lm()</code> and many other
 modelling functions) now warns when it drops contrasts
 from factors.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=16119">PR#16119</a>)
 </p>
 </li>
 <li> <p><code>install.packages()</code> and friends now accept the value
 <code>type = "binary"</code> as a synonym for the native binary type on
 the platform (if it has one).
 </p>
 </li>
 <li><p> Single source or binary files can be supplied for
 <code>install.packages(type = "both")</code> and the appropriate type
 and <code>repos = NULL</code> will be inferred.
 </p>
 </li>
 <li><p> New function <code>pcre_config()</code> to report on some of the
 configuration options of the version of PCRE in use.  In
 particular, this reports if regular expressions using
 <span class="samp">\p{xx}</span> are supported.
 </p>
 </li>
 <li><p> (Windows.) <code>download.file(cacheOK = FALSE)</code> is now
 supported when &lsquo;<span class="file">internet2.dll</span>&rsquo; is used.
 </p>
 </li>
 <li> <p><code>browseURL()</code> has been updated to work with Firefox
 36.0 which has dropped support for the <span class="option">-remote</span> interface.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The included version of PCRE has been updated to 8.36.
 </p>
 </li>
 <li> <p><code>configure</code> accepts <span class="samp">MAKEINFO=texi2any</span> as
 another way to ensure <span class="pkg">texinfo</span> 5.x is used when both 5.x and
 4.x are installed.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> now checks the packages used in
 <code>\donttest</code> sections of the examples are specified in the
 &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file.
 (These are needed to run the examples interactively.)
 </p>
 </li>
 <li> <p><code>R CMD check</code> checks for the undeclared use of GNU
 extensions in Makefiles, and for Makefiles with a missing final
 linefeed.
 </p>
 <p><code>R CMD build</code> will correct line endings in all Makefiles,
 not just those in the &lsquo;<span class="file">src</span>&rsquo; directory.
 </p>
 </li>
 <li> <p><code>R CMD check</code> notes uses of <code>library()</code> and
 <code>require()</code> in package code: see the section
 &lsquo;Suggested packages&rsquo; of &lsquo;Writing R Extensions&rsquo; for
 good practice.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> The <code>configure</code> option
 <span class="option">--with-valgrind-instrumentation=3</span> is deprecated and will
 be removed in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.2.0.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> (Windows.) <code>Rscript.exe</code> was missing a
 manifest specifying the modern style for common controls (e.g.,
 the download progress bar).
 </p>
 </li>
 <li><p> If a package had extra documentation files but no vignette,
 the HTML help system produced an empty index page.
 </p>
 </li>
 <li><p> The parser now gives an error if a null character is included
 in a string using Unicode escapes. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16046">PR#16046</a>)
 </p>
 </li>
 <li> <p><code>qr.Q()</code> failed on complex arguments due to
 pre-3.0(!) typo. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16054">PR#16054</a>)
 </p>
 </li>
 <li> <p><code>abs()</code> failed with named arguments when the argument
 was complex.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16047">PR#16047</a>)
 </p>
 </li>
 <li> <p><code>"noquote"</code> objects may now be used as columns in
 data frames.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15997">PR#15997</a>)
 </p>
 </li>
 <li><p> Some values with extremely long names were printed
 incorrectly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15999">PR#15999</a>)
 </p>
 </li>
 <li><p> Extremely large exponents on zero expressed in scientific
 notation (e.g. <code>0.0e50000</code>) could give <code>NaN</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15976">PR#15976</a>)
 </p>
 </li>
 <li> <p><code>download.file()</code> reported downloaded sizes as 0KB if
 less than 1MB, only for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.2 and only on big-endian platforms.
 </p>
 </li>
 <li> <p><code>prompt()</code> did not escape percent signs in the
 automatically generated usage section of help files.
 </p>
 </li>
 <li> <p><code>drop.terms()</code> dropped some of the attributes of the
 object it was working with.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16029">PR#16029</a>)
 </p>
 </li>
 <li><p> (Windows.) The command completion in <code>Rgui.exe</code>
 messed up the console.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15791">PR#15791</a>)
 </p>
 </li>
 <li><p> (Windows.) The <code>choose.files()</code> command returned a
 blank string when the user asked for a single file but cancelled
 the request.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16074">PR#16074</a>)
 </p>
 </li>
 <li> <p><code>Math2</code> S4 group generics failed to correctly dispatch
 <code>"structure"</code>- and <code>"nonStructure"</code>-derived classes.
 </p>
 </li>
 <li> <p><code>loadNamespace()</code> imposed undocumented restrictions
 on the <code>versionCheck</code> parameter.  (Reported by Geoff Lee.)
 </p>
 </li>
 <li><p> Rare over-runs detected by AddressSanitizer in
 <code>substr()</code> and its replacement version have been avoided.
 </p>
 <p><em>Inter alia</em> that fix gives the documented behaviour for
 <code>substr(x, 1, 2) &lt;- ""</code> (subsequently reported as
 <a href="https://bugs.R-project.org/show_bug.cgi?id=16214">PR#16214</a>).
 </p>
 </li>
 <li><p> Loading packages incorrectly defining an S4 generic followed
 by a function of the same name caused an erroneous cyclic
 namespace dependency error.
 </p>
 </li>
 <li><p> Declared vignette encodings are now always passed to the
 vignette engine.
 </p>
 </li>
 <li><p> Port Tomas Kalibera's fix from R-devel that restores the
 <code>loadMethod()</code> fast path, effectively doubling the speed of
 S4 dispatch.
 </p>
 </li>
 <li> <p><code>power.t.test()</code> and <code>power.prop.test()</code> now make
 use of the <code>extendInt</code> option of <code>uniroot()</code> and hence
 work in more extreme cases.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15792">PR#15792</a>)
 </p>
 </li>
 <li><p> If a package was updated and attached when its namespace was
 already loaded, it could end up with parts from one version and
 parts from the other.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16120">PR#16120</a>)
 </p>
 </li>
 <li> <p><code>tools:::.Rdconv()</code> didn't accept <code>--encoding=</code> due
 to a typo.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16121">PR#16121</a>)
 </p>
 </li>
 <li><p> Unix-alike builds without a suitable <code>makeinfo</code> were
 documented to link the missing HTML manuals to CRAN, but did not.
 </p>
 </li>
 <li> <p><code>save(*, ascii=TRUE)</code> and <code>load()</code> now correctly
 deal with <code>NaN</code>'s.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16137">PR#16137</a>)
 </p>
 </li>
 <li> <p><code>split.Date()</code> retains fractional representations while
 avoiding incomplete class propagation.
 </p>
 </li>
 <li> <p>&lsquo;<span class="file">R_ext/Lapack.h</span>&rsquo; had not been updated for changes made
 by LAPACK to the argument lists of its (largely internal)
 functions <code>dlaed2</code> and <code>dlaed3</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16157">PR#16157</a>)
 </p>
 </li>
 <li> <p><code>RShowDoc("NEWS", "txt")</code> had not been updated for the
 layout changes of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.0.
 </p>
 </li>
 <li><p> The <code>xtfrm()</code> method for class <code>"Surv"</code> has been
 corrected and its description expanded.
 </p>
 </li>
 <li> <p><code>mode(x) &lt;- y</code> would incorrectly evaluate <code>x</code> before
 changing its mode. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16215">PR#16215</a>)
 </p>
 </li>
 <li> <p><code>besselJ(1, 2^64)</code> and <code>besselY(..)</code> now signal a
 warning, returning <code>NaN</code> instead of typically
 segfaulting. (Issue 3 of <a href="https://bugs.R-project.org/show_bug.cgi?id=15554">PR#15554</a>)
 </p>
 </li>
 <li><p> HTML conversion of <span class="samp">\href</span> markup in &lsquo;<span class="file">.Rd</span>&rsquo; files
 did not remove the backslash from <span class="samp">\%</span> and so gave an
 invalid URL.  In a related change, the <span class="samp">\</span> escape is now
 required in such URLs.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.1.2</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>embedFonts()</code> now defaults to
 <code>format = "ps2write"</code> for &lsquo;<span class="file">.ps</span>&rsquo; and &lsquo;<span class="file">.eps</span>&rsquo; files.
 This is available in Ghostscript 9.x (since 2010) whereas the
 previous default, <code>format = "pswrite"</code>, was removed in
 Ghostscript 9.10.
 </p>
 </li>
 <li><p> For consistency with <code>[dpqr]norm()</code>,
 <code>[dp]lnorm(sdlog = 0)</code> model a point mass at
 <code>exp(mulog)</code> rather than return <code>NaN</code> (for an error).
 </p>
 </li>
 <li> <p><code>capabilities()</code> now reports if ICU is compiled in for
 use for collation (it is only actually used if a suitable locale
 is set for collation, and never for a <code>C</code> locale).
 </p>
 </li>
 <li><p> (OS X only.) Package <span class="pkg">tcltk</span> checks when loaded if it is
 linked against the CRAN X11-based Tcl/Tk and if so that the Tcl/Tk
 component and the X11 libraries are installed.  This allows more
 informative error messages to be given advising the installation
 of the missing component or of XQuartz.
 </p>
 <p>The <code>X11()</code> device and X11-based versions of the data editor
 and viewer (invoked by <code>edit()</code> and <code>View()</code> for data
 frames and matrices from command-line <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>) check that the X11
 libraries are installed and if not advises installing XQuartz.
 </p>
 </li>
 <li> <p><code>icuSetCollate()</code> allows <code>locale = "default"</code>, and
 <code>locale = "none"</code> to use OS services rather than ICU for
 collation.
 </p>
 <p>Environment variable <span class="env">R_ICU_LOCALE</span> can be used to set the
 default ICU locale, in case the one derived from the OS locale is
 inappropriate (this is currently necessary on Windows).
 </p>
 </li>
 <li><p> New function <code>icuGetCollate()</code> to report on the ICU
 collation locale in use (if any).
 </p>
 </li>
 <li> <p><code>utils::URLencode()</code> was updated to use unreserved and
 reserved characters from RFC 3986
 (<a href="http://tools.ietf.org/html/rfc3986">http://tools.ietf.org/html/rfc3986</a>) instead of RFC 1738.
 </p>
 </li>
 <li> <p><code>unique(warnings())</code> and <code>c(warnings())</code> are now
 supported.
 </p>
 </li>
 <li><p> The Bioconductor &lsquo;version&rsquo; used by
 <code>setRepositories()</code> now defaults to <code>3.0</code>. (It can be
 set at runtime <em>via</em> environment variable
 <span class="env">R_BIOC_VERSION</span>.)
 </p>
 </li>
 <li><p> Omegahat is no longer listed as providing Windows binary
 packages, e.g. by <code>setRepositories()</code>.  It has no binary
 packages available for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.x and those for earlier versions
 were 32-bit only.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The <code>configure</code> script reports on the more important
 capabilities/options which will not be compiled in.
 </p>
 <p>More types of external BLAS are recognized by name in that report.
 </p>
 </li>
 <li><p> When building <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> as a shared library, the
 <span class="option">-L${R_HOME}/lib${R_ARCH}</span> flag is placed earlier in the
 link commands used during installation and when packages are
 installed: this helps ensure that the current build has priority
 if an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> shared library has already been installed by
 e.g. <code>install-libR</code> in a library mentioned in
 <span class="env">LDFLAGS</span> (and not in &lsquo;your system's library directory&rsquo;
 as documented). (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15790">PR#15790</a>.)
 </p>
 </li>
 <li><p> LaTeX package <span class="pkg">upquote</span> is no longer required for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>'s
 use of <span class="pkg">inconsolata</span>.
 </p>
 </li>
 <li><p> (Windows only) If both 32- and 64-bit versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> are
 installed, the &lsquo;<span class="file">bin/R.exe</span>&rsquo; and &lsquo;<span class="file">bin/Rscript.exe</span>&rsquo; executables
 now run 64-bit <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  (To run 32-bit <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>, overwrite these files
 with copies of &lsquo;<span class="file">bin/i386/Rfe.exe</span>&rsquo;.)
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li><p> Running <code>R CMD check</code> with
 <span class="env">_R_CHECK_DEPENDS_ONLY_</span> true now makes the
 <span class="samp">VignetteBuilder</span> packages available even if they are
 listed in <span class="samp">Suggests</span>, since they are needed to recognise and
 process non-Sweave vignettes.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now reports empty <code>importFrom</code>
 declarations in a &lsquo;<span class="file">NAMESPACE</span>&rsquo; file, as these are common
 errors (writing <code>importFrom(<var>Pkg</var>)</code> where
 <code>import(<var>Pkg</var>)</code> was intended).
 </p>
 </li>
 <li> <p><code>R CMD check</code> now by default checks code usage
 directly on the package namespace without loading and attaching
 the package and its suggests and enhances.  For good practice with
 packages in the <span class="samp">Suggests</span> field, see §1.1.3.1 of
 &lsquo;Writing R Extensions&rsquo;.  For use of lazy-data objects in
 the package's own code, see <code>?data</code>.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>dmultinom()</code> did not handle non-finite probabilities
 correctly.
 </p>
 </li>
 <li> <p><code>prettyNum(x, zero.print=*)</code> now also works when
 <code>x</code> contains <code>NA</code>s.
 </p>
 </li>
 <li><p> A longstanding bug exhibited by <code>nlminb()</code> on Windows
 was traced to a compiler bug in gcc 4.6.3; a workaround has been
 put in place.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15244">PR#15244</a> and <a href="https://bugs.R-project.org/show_bug.cgi?id=15914">PR#15914</a>).
 </p>
 </li>
 <li><p> Rendering of <code>\command</code> in HTML versions of help pages
 has been improved: this is particularly evident on the help page
 for <code>INSTALL</code>.
 </p>
 </li>
 <li> <p><code>as.hexmode(x)</code> and <code>as.octmode(x)</code> now behave
 correctly for some numeric <code>x</code>, e.g., <code>c(NA, 1)</code> or
 <code>c(1, pi)</code>.
 </p>
 </li>
 <li> <p><code>drop1()</code> failed if the <code>scope</code> argument had
 no variables to drop.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15935">PR#15935</a>)
 </p>
 </li>
 <li> <p><code>edit()</code> (and hence <code>fix()</code>) failed if
 an object had a non-character attribute named <code>"source"</code>
 (an attribute that had been used in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> prior to version 2.14.0).
 </p>
 </li>
 <li> <p><code>callGeneric()</code> could fail if the generic had
 <code>...</code> as a formal argument. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15937">PR#15937</a>).
 </p>
 </li>
 <li><p> Forking in package <span class="pkg">parallel</span> called C entry point
 <code>exit</code> in the child.  This was unsafe (<code>_exit</code> should
 have been called), and could flush <code>stdin</code> of the main <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 process (seen most often on Solaris).
 </p>
 <p>As good practice, <code>stdout</code> is now flushed before forking a child.
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> objects such as <code>list(`a\b` = 1)</code> now print correctly.
 </p>
 </li>
 <li> <p><code>getAnywhere("C_pbinom")</code> now returns correctly a
 single object (rather than unlisting it).
 </p>
 </li>
 <li><p> The <code>confint()</code> method for <code>nls()</code> fits failed it
 these has specified parameter limits despite using an algorithm
 other than <code>"port"</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15960">PR#15960</a>)
 </p>
 </li>
 <li><p> Subclassing an S4 class failed if the class required
 arguments to the generator, through its <code>initialize()</code> method.
 </p>
 </li>
 <li> <p><code>removeSource()</code> did not properly handle expressions
 containing arguments that were supplied as missing, e.g.
 <code>x[i,]</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15957">PR#15957</a>)
 </p>
 </li>
 <li> <p><code>as.environment(list())</code> now works, and
 <code>as.list()</code> of such an environment is now the same as
 <code>list()</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15926">PR#15926</a>)
 </p>
 </li>
 <li><p> Several <span class="pkg">tcltk</span> functions failed when run in unusual
 environments. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15970">PR#15970</a>)
 </p>
 </li>
 <li> <p><code>options(list())</code> now works (trivially). (<a href="https://bugs.R-project.org/show_bug.cgi?id=15979">PR#15979</a>)
 </p>
 </li>
 <li> <p><code>merge(&lt;dendrogram&gt;, ..)</code> now works correctly for two
 &lsquo;independent&rsquo; dendrograms (<a href="https://bugs.R-project.org/show_bug.cgi?id=15648">PR#15648</a>), and still compatibly
 <em>via</em> <code>adjust = "auto"</code> e.g. for two branches of an
 existing dendrogram.
 </p>
 </li>
 <li><p> The <code>plot</code> method for <code>"hclust"</code> objects gets an
 optional argument <code>check</code>; when that is true (the default) it
 checks more carefully for valid input.
 </p>
 </li>
 <li><p> (Windows only) If a user chose to install 64 bit <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> but not
 32 bit <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>, the &lsquo;<span class="file">bin/R</span>&rsquo; and &lsquo;<span class="file">bin/Rscript</span>&rsquo; executables
 failed to run.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15981">PR#15981</a>)
 </p>
 </li>
 <li><p> Various possible buffer overruns have been prevented,
 and missed memory protection added. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15990">PR#15990</a>)
 </p>
 </li>
 <li> <p><code>Rscript</code> no longer passes <code>--args</code> to
 <code>R</code> when there are no extra (&ldquo;user&rdquo;) arguments.
 </p>
 </li>
 <li><p> objects like <code>getClass("refClass")@prototype</code> now
 <code>print()</code> and <code>str()</code> without error.
 </p>
 </li>
 <li> <p><code>identical()</code> now also looks at the S4 bit.
 </p>
 </li>
 <li> <p><code>hist(x, breaks)</code> is more robust in adding a small fuzz
 to few breaks when some are very large. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15988">PR#15988</a>)
 </p>
 </li>
 <li> <p><code>sub()</code> and <code>gsub()</code> did not handle regular
 expressions like <code>"\s{2,}"</code> properly if the text
 contained <code>NA</code> or non-ASCII elements in a UTF-8
 locale.  Part of this was due to a bug in the TRE library.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16009">PR#16009</a>)
 </p>
 </li>
 <li> <p><code>RShowDoc("NEWS")</code> now displays the PDF version.
 </p>
 </li>
 <li><p> Matrices and arrays with last dimension zero did not print
 at all or incompletely.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=16012">PR#16012</a>)
 </p>
 </li>
 <li> <p><code>plot.histogram()</code> and hence <code>hist()</code> now respect
 the <code>xaxs</code>, <code>yaxs</code> and <code>lab</code> graphics parameters.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=16021">PR#16021</a>)
 </p>
 </li>
 <li> <p><code>bw.SJ(x)</code> and other bw.*() no longer segfault when
 <code>x</code> contains non-finite values. (<a href="https://bugs.R-project.org/show_bug.cgi?id=16024">PR#16024</a>)
 </p>
 </li>
 <li> <p><code>R CMD Rd2pdf</code> unintentionally ignored its
 <span class="option">--os</span> option.
 </p>
 </li>
 <li><p> The internal method of <code>download.file()</code> was not
 reporting file sizes and progress correctly on files larger than
 2GB (inherited from <code>libxml2</code>).  This is corrected for 64-bit
 builds (32-bit platforms may not support such files, but where
 possible will be supported in future versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>).
 </p>
 </li>
 <li><p> Work around a bug in OS X Yosemite where key environment
 variables may be duplicated causing issues in subprocesses. The
 duplicates are now removed on <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> startup (<em>via</em>
 Rprofile). (<a href="https://bugs.R-project.org/show_bug.cgi?id=16042">PR#16042</a>)
 </p>
 </li>
 <li><p> Adjust X11 auto-launch detection in DISPLAY on OS X to
 recognize latest XQuartz.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.1.1</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> When <code>attach()</code> reports conflicts, it does so
 compatibly with <code>library()</code> by using <code>message()</code>.
 </p>
 </li>
 <li> <p><code>R CMD Sweave</code> no longer cleans any files by
 default, compatibly with versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> prior to 3.1.0.  There are
 new options <span class="option">--clean</span>, <span class="option">--clean=default</span> and
 <span class="option">--clean=keepOuts</span>.
 </p>
 </li>
 <li> <p><code>tools::buildVignette()</code> and <code>tools::buildVignettes()</code>
 with <code>clean = FALSE</code> no longer remove any created files.
 <code>buildvignette()</code> gains a <code>keep</code> argument for more
 cleaning customization.
 </p>
 </li>
 <li><p> The Bioconductor &lsquo;version&rsquo; used by
 <code>setRepositories()</code> can now be set by environment variable
 <span class="env">R_BIOC_VERSION</span> at runtime, not just when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is installed.
 (It has been stated that Bioconductor will switch from
 &lsquo;version&rsquo; 2.14 to &lsquo;version&rsquo; 3.0 during the lifetime
 of the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1 series.)
 </p>
 </li>
 <li><p> Error messages from bugs in embedded <span class="samp">Sexpr</span> code
 in Sweave documents now report the source location.
 </p>
 </li>
 <li> <p><code>type.convert()</code>, <code>read.table()</code> and similar
 <code>read.*()</code> functions get a new <code>numerals</code> argument,
 specifying how numeric input is converted when its conversion to
 double precision loses accuracy.  The default value,
 <code>"allow.loss"</code> allows accuracy loss, as in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> versions before
 3.1.0.
 </p>
 </li>
 <li><p> For some compilers, integer addition could overflow without
 a warning.  <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>'s internal code for both integer addition and
 subtraction is more robust now.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15774">PR#15774</a>)
 </p>
 </li>
 <li><p> The function determining the default number of knots for
 <code>smooth.spline()</code> is now exported, as <code>.nknots.smspl()</code>.
 </p>
 </li>
 <li> <p><code>dbeta(, a,b)</code>, <code>pbeta()</code>, <code>qbeta()</code> and
 <code>rbeta()</code> are now defined also for <i>a = 0</i>, <i>b = 0</i>,
 or infinite <i>a</i> and <i>b</i> (where they typically returned
 <code>NaN</code> before).
 </p>
 </li>
 <li><p> Many package authors report that the RStudio graphics device
 does not work correctly with their package's use of
 <code>dev.new()</code>.  The new option <code>dev.new(noRStudioGD =
       TRUE)</code> replaces the RStudio override by the default device as
 selected by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> itself, still respecting environment variables
 <span class="env">R_INTERACTIVE_DEVICE</span> and <span class="env">R_DEFAULT_DEVICE</span>.
 </p>
 </li>
 <li> <p><code>readRDS()</code> now returns visibly.
 </p>
 </li>
 <li><p> Modifying internal logical scalar constants now results in
 an error instead of a warning.
 </p>
 </li>
 <li> <p><code>install.packages(repos = NULL)</code> now accepts
 <span class="samp">http://</span> or <span class="samp">ftp://</span> URLs of package archives as well
 as file paths, and will download as required.  In most cases
 <code>repos = NULL</code> can be deduced from the extension of the URL.
 </p>
 </li>
 <li><p> The warning when using partial matching with the <code>$</code>
 operator on data frames is now only given when
 <code>options("warnPartialMatchDollar")</code> is <code>TRUE</code>.
 </p>
 </li>
 <li><p> Package help requests like <code>package?foo</code> now
 try the package <code>foo</code> whether loaded or not.
 </p>
 </li>
 <li><p> General help requests now default to trying all loaded
 packages, not just those on the search path.
 </p>
 </li>
 <li><p> Added a new function <code>promptImport()</code>, to generate a
 help page for a function that was imported from another package
 (and presumably re-exported, or help would not be needed).
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li> <p><code>configure</code> option <span class="option">--with-internal-tzcode</span>
 can now be used with variable <code>rsharedir</code>.
 </p>
 </li>
 <li><p> The included version of PCRE has been updated to 8.35.
 </p>
 </li>
 <li><p> There is a new target <code>make uninstall-libR</code> to
 remove an installed shared/static &lsquo;<span class="file">libR</span>&rsquo;.
 </p>
 <p><code>make install-libR</code> now works if a sub-architecture is
 used, although the user will need to specify <code>libdir</code>
 differently for different sub-architectures.
 </p>
 </li>
 <li><p> There is more extensive advice on which LaTeX packages are
 required to install <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> or to make package manuals (as done by
 <code>R CMD check</code>) in the &lsquo;Writing R Extensions&rsquo;
 manual.
 </p>
 </li>
 <li><p> Compilers/linkers were handling the visibility control in
 &lsquo;<span class="file">src/extra/xz</span>&rsquo; inconsistently (and apparently in some cases
 incorrectly), so it has been simplified.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15327">PR#15327</a>)
 </p>
 </li>
 <li><p> (Windows)  There is updated support for the use of ICU for
 collation: see the &lsquo;R Installation and Administration Manual&rsquo;.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>dbinom(x, n)</code>, <code>pbinom()</code>, <code>dpois()</code>, etc,
 are slightly less restrictive in checking if <code>n</code> is
 integer-valued.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15734">PR#15734</a>.)
 </p>
 </li>
 <li> <p><code>pchisq(x, df, ncp, log.p = TRUE)</code> is more accurate and
 no longer underflows for small <code>x</code> and <code>ncp &lt; 80</code>, e.g,
 for <code>pchisq(1e-5, df = 100, ncp = 1, log = TRUE)</code>.
 (Based on <a href="https://bugs.R-project.org/show_bug.cgi?id=15635">PR#15635</a> and a suggestion by Roby Joehanes.)
 </p>
 </li>
 <li><p> The <code>s</code> (&ldquo;step into&rdquo;) command in the debugger
 would cause <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> to step into expressions evaluated there,
 not just into functions being debugged.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15770">PR#15770</a>)
 </p>
 </li>
 <li><p> The C code used by <code>strptime()</code> rejected time-zone
 offsets of more than <code>+1200</code> (<code>+1245</code>, <code>+1300</code> and
 <code>+1400</code> can occur).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15768">PR#15768</a>)
 </p>
 </li>
 <li><p> (Windows only.)
 <code>png(type = "cairo", antialias = "gray")</code>
 was not accepted.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15760">PR#15760</a>)
 </p>
 </li>
 <li><p> Use of <code>save(..., envir=)</code> with named objects could
 fail.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15758">PR#15758</a>)
 </p>
 </li>
 <li> <p><code>Sweave()</code> mis-parsed <span class="samp">Sexpr</span> expressions that
 contained backslashes.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15779">PR#15779</a>)
 </p>
 </li>
 <li><p> The return value from <code>options(foo = NULL)</code> was not
 the previous value of the option.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15781">PR#15781</a>)
 </p>
 </li>
 <li> <p><code>enc2utf8()</code> and <code>enc2native()</code> did not always
 mark the encoding of the return values when it was known.
 </p>
 </li>
 <li> <p><code>dnbinom(x, size = &lt;large&gt;, mu, log = TRUE)</code> no longer
 underflows to -Inf for large <code>mu</code>, thanks to a suggestion
 from Alessandro Mammana (MPI MolGen, Berlin).
 </p>
 </li>
 <li> <p><code>pbeta(x, a, b, log = TRUE)</code> no longer behaves
 discontinuously (in a small x-region) because of denormalized
 numbers.  Also, <code>pbeta(1-1e-12, 1e30, 1.001, log=TRUE)</code> now
 terminates &ldquo;in real time&rdquo;.
 </p>
 </li>
 <li><p> The <code>"CRAN"</code> filter (see <code>available.packages()</code>)
 no longer removes duplicates other than of packages on CRAN, and
 does not fail if there is no CRAN repository in
 <code>getOption("repos")</code>.
 </p>
 </li>
 <li><p> The device listing from <code>dev2bitmap()</code> and
 <code>bitmap()</code> was truncated to 1000 characters: modern versions
 of GhostScript on most platforms have many more devices.
 </p>
 </li>
 <li><p> (Windows.)  Commands such as <code>Sys.which()</code> and
 <code>pipe()</code> which needed to find the full path to a command
 could segfault if the &lsquo;long&rsquo; path name was much longer than
 the &lsquo;short&rsquo; path name (which <code>Sys.which()</code> returns),
 as the behaviour of the Windows API call had changed.
 </p>
 </li>
 <li> <p><code>R CMD build</code> will fail with an error if one of the
 packages specified in the <span class="samp">VignetteBuilder</span> field is not
 installed.  (Without loading those packages it cannot be
 ascertained which files are intended to be vignettes. This means
 that the <span class="samp">VignetteBuilder</span> packages have to be installed for
 package checking too.)
 (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15775">PR#15775</a>.)
 </p>
 </li>
 <li><p> Misguided attempts to use <code>chull()</code> with non-finite
 points now give an error (related to <a href="https://bugs.R-project.org/show_bug.cgi?id=15777">PR#15777</a>).
 </p>
 </li>
 <li><p> For a formula with exactly 32 variables the 32nd variable
 was aliased to the intercept in some C-level computations of
 terms, so that for example attempting to remove it would remove
 the intercept instead (and leave a corrupt internal structure).
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15735">PR#15735</a>)
 </p>
 </li>
 <li> <p><code>anyDuplicated()</code> silently returned wrong values when
 the first duplicate was at an index which was too large to be
 stored in an integer vector (although a lot of RAM and patience
 would have been needed to encounter this).
 </p>
 </li>
 <li> <p><code>tools::Rd2ex(commentDontrun = FALSE)</code> failed if the
 block had only one line.
 </p>
 </li>
 <li><p> Hexadecimal constants such as <code>0x110p-5L</code> which were
 incorrectly qualified by <code>L</code> were parsed incorrectly since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.0.0, with a slightly garbled warning. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15753">PR#15753</a>)
 </p>
 </li>
 <li> <p><code>system()</code> returned success on some platforms even if
 the system was unable to launch a process. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15796">PR#15796</a>)
 </p>
 </li>
 <li><p> (Windows <code>Rgui</code> console.)  Unbuffered output was
 sometimes not output immediately if the prompt was not on the last
 line of the console.
 </p>
 </li>
 <li><p> The built-in help server did not declare the encoding for
 the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; or other text files to be the package
 encoding, so non-ASCII characters could be displayed incorrectly.
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is now trying harder to not cleanup child processes
 that were not spawned by <code>mcparallel()</code> on platforms that
 provide information about the source process of the <code>SIGCHLD</code>
 signal. This allows 3rd party libraries to manage the exit status
 of children that they spawn without <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> interfering.
 </p>
 </li>
 <li> <p><code>mcmapply()</code> was only parallelizing if the number of
 jobs was bigger than the number of cores. It now parallelizes if
 the number of jobs is more than one.
 </p>
 </li>
 <li><p> Auto-printing would re-evaluate its argument when trying to
 dispatch to a print method.  This is now avoided when possible.
 </p>
 </li>
 <li><p> Unserializing (including <code>load()</code> and <code>readRDS()</code>)
 could silently return incorrect numeric values from ASCII saves if
 there was a read error.
 </p>
 </li>
 <li> <p><code>getParseData()</code> could return incorrect values for
 the parents of some elements. (Reported by Andrew Redd.)
 </p>
 </li>
 <li><p> Attempting to use data frames of 2^31 or more rows with
 <code>merge()</code> or to create a merged data frame of that size now
 gives a clearer error message.
 </p>
 </li>
 <li> <p><code>parse()</code> did not check its <code>file</code> argument was a
 connection if it was not a character string, so
 e.g. <code>parse(FALSE)</code> attempted to read from <code>stdin</code>.
 </p>
 <p>Nor did <code>dump()</code> and <code>dput()</code>.
 </p>
 </li>
 <li><p> The <code>"help.try.all.packages"</code> option was ignored when
 the shortcut syntax for help was used, e.g. <code>?foo</code>.
 </p>
 </li>
 <li><p> A potential segfault in string allocation has been fixed.
 (Found by Radford Neal.)
 </p>
 </li>
 <li><p> Potential memory protection errors in <code>sort()</code> and
 <code>D()</code> have been fixed. (Found by Radford Neal.)
 </p>
 </li>
 <li><p> Fixed a lack of error checking in graphics event functions.
 (Found by Radford Neal; a different patch used here than the
 one in pqR.)
 </p>
 </li>
 <li> <p><code>numericDeriv()</code> sometimes miscalculated the gradient.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15849">PR#15849</a>, reported originally by Radford Neal)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.1.0</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>type.convert()</code> (and hence by default
 <code>read.table()</code>) returns a character vector or factor when
 representing a numeric input as a double would lose accuracy.
 Similarly for complex inputs.
 </p>
 <p>If a file contains numeric data with unrepresentable numbers of
 decimal places that are intended to be read as numeric, specify
 <code>colClasses</code> in <code>read.table()</code> to be <code>"numeric"</code>.
 </p>
 </li>
 <li> <p><code>tools::Rdiff(useDiff = FALSE)</code> is closer to the POSIX
 definition of <code>diff -b</code> (as distinct from the description
 in the <code>man</code> pages of most systems).
 </p>
 </li>
 <li><p> New function <code>anyNA()</code>, a version of
 <code>any(is.na(.))</code> which is fast for atomic vectors, based on a
 proposal by Tim Hesterberg. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15239">PR#15239</a>.)
 </p>
 </li>
 <li> <p><code>arrayInd(*, useNames = TRUE)</code> and, analogously,
 <code>which(*, arr.ind = TRUE)</code> now make use of
 <code>names(.dimnames)</code> when available.
 </p>
 </li>
 <li> <p><code>is.unsorted()</code> now also works for <code>raw</code> vectors.
 </p>
 </li>
 <li><p> The <code>"table"</code> method for <code>as.data.frame()</code> (also
 useful as <code>as.data.frame.table()</code>) now passes <code>sep</code> and
 <code>base</code> arguments to <code>provideDimnames()</code>.
 </p>
 </li>
 <li> <p><code>uniroot()</code> gets new optional arguments, notably
 <code>extendInt</code>, allowing to auto-extend the search interval when
 needed.  The return value has an extra component, <code>init.it</code>.
 </p>


 </li>
 <li> <p><code>switch(f, ...)</code> now warns when <code>f</code> is a factor,
 as this typically happens accidentally where the useR meant to
 pass a character string, but <code>f</code> is treated as integer (as
 always documented).
 </p>
 </li>
 <li><p> The parser has been modified to use less memory.
 </p>
 </li>
 <li><p> The way the unary operators (<code>+ - !</code>) handle attributes
 is now more consistent.  If there is no coercion, all attributes
 (including class) are copied from the input to the result:
 otherwise only names, dims and dimnames are.
 </p>
 </li>
 <li> <p><code>colorRamp()</code> and <code>colorRampPalette()</code> now allow
 non-opaque colours and a ramp in opacity <em>via</em> the new argument
 <code>alpha = TRUE</code>.  (Suggested by Alberto Krone-Martins, but
 optionally as there are existing uses which expect only RGB values.)

 </p>
 </li>
 <li> <p><code>grid.show.layout()</code> and <code>grid.show.viewport()</code>
 get an optional <code>vp.ex</code> argument.
 </p>
 </li>
 <li><p> There is a new function <code>find_gs_cmd()</code> in the
 <span class="pkg">tools</span> package to locate a GhostScript executable.  (This is
 an enhanced version of a previously internal function there.)
 </p>
 </li>
 <li> <p><code>object.size()</code> gains a <code>format()</code> method.
 </p>
 </li>
 <li><p> There is a new family, <code>"ArialMT"</code>, for the
 <code>pdf()</code> and <code>postscript()</code> devices.  This will only be
 rendered correctly on viewers which have access to Monotype
 TrueType fonts (which are sometimes requested by journals).
 </p>
 </li>
 <li><p> The text and PDF news files, including &lsquo;<span class="file">NEWS</span>&rsquo; and
 &lsquo;<span class="file">NEWS.2</span>&rsquo;, have been moved to the &lsquo;<span class="file">doc</span>&rsquo; directory.
 </p>
 </li>
 <li> <p><code>combn(x, simplify = TRUE)</code> now gives a factor result
 for factor input <code>x</code> (previously user error).
 (Related to <a href="https://bugs.R-project.org/show_bug.cgi?id=15442">PR#15442</a>.)
 </p>
 </li>
 <li><p> Added <code>utils::fileSnapshot()</code> and
 <code>utils::changedFiles()</code> functions to allow snapshots and
 comparison of directories of files.
 </p>
 </li>
 <li> <p><code>make.names(names, unique=TRUE)</code> now tries to preserve
 existing names. (Suggestion of <a href="https://bugs.R-project.org/show_bug.cgi?id=15452">PR#15452</a>.)
 </p>
 </li>
 <li><p> New functions <code>cospi(x)</code>, <code>sinpi(x)</code>, and
 <code>tanpi(x)</code>, for more accurate computation of
 <code>cos(pi*x)</code>, etc, both in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> and the C API.  Using these
 gains accuracy in some cases, e.g., inside <code>lgamma()</code> or
 <code>besselI()</code>.  (Suggested by Morten Welinder in <a href="https://bugs.R-project.org/show_bug.cgi?id=15529">PR#15529</a>.)
 </p>
 </li>
 <li> <p><code>print.table(x, zero.print = ".")</code> now also has an
 effect when <code>x</code> is not integer-valued.
 </p>
 </li>
 <li><p> There is more support to explore the system's idea of
 time-zone names.  <code>Sys.timezone()</code> tries to give the current
 system setting by name (and succeeds at least on Linux, OS X,
 Solaris and Windows), and <code>OlsonNames()</code> lists the names in
 the system's Olson database. <code>Sys.timezone(location = FALSE)</code>
 gives the previous behaviour.
 </p>
 </li>
 <li><p> Platforms with a 64-bit <code>time_t</code> type are allowed to
 handle conversions between the <code>"POSIXct"</code> and
 <code>"POSIXlt"</code> classes for date-times outside the 32-bit range
 (before 1902 or after 2037): the existing workarounds are used on
 other platforms.  (Note that time-zone information for post-2037
 is speculative at best, and the OS services are tested for known
 errors and so not used on OS X.)
 </p>
 <p>Currently <code>time_t</code> is usually <code>long</code> and hence 64-bit on
 Unix-alike 64-bit platforms: however in several cases the
 time-zone database is 32-bit.  For <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> for Windows it is 64-bit
 (for both architectures as from this version).
 </p>
 </li>
 <li><p> The <code>"save.defaults"</code> option can include a value for
 <code>compression_level</code>.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15579">PR#15579</a>.)
 </p>
 </li>
 <li> <p><code>colSums()</code> and friends now have support for arrays and
 data-frame columns with <i>2^31</i> or more elements.
 </p>
 </li>
 <li> <p><code>as.factor()</code> is faster when <code>f</code> is an unclassed
 integer vector (for example, when called from <code>tapply()</code>).
 </p>
 </li>
 <li> <p><code>fft()</code> now works with longer inputs, from the 12
 million previously supported up to 2 billion.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15593">PR#15593</a>)
 </p>
 </li>
 <li><p> Complex <code>svd()</code> now uses LAPACK subroutine
 <code>ZGESDD</code>, the complex analogue of the routine used for the
 real case.
 </p>
 </li>
 <li><p> Sweave now outputs &lsquo;<span class="file">.tex</span>&rsquo; files in UTF-8 if the
 input encoding is declared to be UTF-8, regardless of the
 local encoding.  The UTF-8 encoding may now be declared using
 a LaTeX comment containing the string <code>%\SweaveUTF8</code>
 on a line by itself.
 </p>
 </li>
 <li> <p><code>file.copy()</code> gains a <code>copy.date</code> argument.
 </p>
 </li>
 <li><p> Printing of date-times will make use of the time-zone
 abbreviation in use at the time, if known.  For example, for Paris
 pre-1940 this could be <span class="samp">LMT</span>, <span class="samp">PMT</span>, <span class="samp">WET</span> or
 <span class="samp">WEST</span>.  To enable this, the <code>"POSIXlt"</code> class has an
 optional component <code>"zone"</code> recording the abbreviation for
 each element.
 </p>
 <p>For platforms which support it, there is also a component
 <code>"gmtoff"</code> recording the offset from GMT where known.
 </p>
 </li>
 <li><p> (On Windows, by default on OS X and optionally elsewhere.)
 The system C function <code>strftime</code> has been replaced by a more
 comprehensive version with closer conformance to the POSIX 2008
 standard.
 </p>
 </li>
 <li> <p><code>dnorm(x, log = FALSE)</code> is more accurate (but somewhat
 slower) for |x| &gt; 5; as suggested in <a href="https://bugs.R-project.org/show_bug.cgi?id=15620">PR#15620</a>.
 </p>
 </li>
 <li><p> Some versions of the <code>tiff()</code> device have further
 compression options.
 </p>
 </li>
 <li> <p><code>read.table()</code>, <code>readLines()</code> and <code>scan()</code>
 have a new argument to influence the treatment of embedded nuls.
 </p>
 </li>
 <li><p> Avoid duplicating the right hand side values in complex
 assignments when possible.  This reduces copying of replacement
 values in expressions such as <code>Z$a &lt;- a0</code> and
 <code>ans[[i]] &lt;- tmp</code>: some package code has relied on there
 being copies.
 </p>
 <p>Also, a number of other changes to reduce copying of objects; all
 contributed by or based on suggestions by Michael Lawrence.
 </p>
 </li>
 <li><p> The <code>fast</code> argument of <code>KalmanLike()</code>,
 <code>KalmanRun()</code> and <code>KalmanForecast()</code> has been replaced
 by <code>update</code>, which instead of updating <code>mod</code> in place,
 optionally returns the updated model in an attribute <code>"mod"</code>
 of the return value.
 </p>
 </li>
 <li> <p><code>arima()</code> and <code>makeARIMA()</code> get a new optional
 argument <code>SSinit</code>, allowing the choice of a different
 <b>s</b>tate <b>s</b>pace initialization which has been observed
 to be more reliable close to non-stationarity: see <a href="https://bugs.R-project.org/show_bug.cgi?id=14682">PR#14682</a>.
 </p>
 </li>
 <li> <p><code>warning()</code> has a new argument <code>noBreaks.</code>, to
 simplify post-processing of output with <code>options(warn = 1)</code>.
 </p>
 </li>
 <li> <p><code>pushBack()</code> gains an argument <code>encoding</code>, to
 support reading of UTF-8 characters using <code>scan()</code>,
 <code>read.table()</code> and related functions in a non-UTF-8 locale.
 </p>
 </li>
 <li> <p><code>all.equal.list()</code> gets a new argument <code>use.names</code>
 which by default labels differing components by names (if they
 match) rather than by integer index.  Saved <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> output in packages may
 need to be updated.
 </p>
 </li>
 <li><p> The methods for <code>all.equal()</code> and
 <code>attr.all.equal()</code> now have argument <code>check.attributes</code>
 after <code>...</code> so it cannot be partially nor positionally
 matched (as it has been, unintentionally).
 </p>
 <p>A side effect is that some previously undetected errors of passing
 empty arguments (no object between commas) to <code>all.equal()</code>
 are detected and reported.
 </p>
 <p>There are explicit checks that <code>check.attributes</code> is logical,
 <code>tolerance</code> is numeric and <code>scale</code> is <code>NULL</code> or
 numeric.  This catches some unintended positional matching.
 </p>
 <p>The message for <code>all.equal.numeric()</code> reports a
 <code>"scaled difference"</code> only for <code>scale != 1</code>.
 </p>
 </li>
 <li> <p><code>all.equal()</code> now has a <code>"POSIXt"</code> method
 replacing the <code>"POSIXct"</code> method.
 </p>
 </li>
 <li><p> The <code>"Date"</code> and <code>"POSIXt"</code> methods of
 <code>seq()</code> allows <code>by = "quarter"</code> for completeness
 (<code>by = "3 months"</code> always worked).
 </p>
 </li>
 <li> <p><code>file.path()</code> removes any trailing separator on
 Windows, where they are invalid (although sometimes accepted).
 This is intended to enhance the portability of code written by
 those using POSIX file systems (where a trailing <code>/</code> can be
 used to confine path matching to directories).
 </p>
 </li>
 <li><p> New function <code>agrepl()</code> which like <code>grepl()</code>
 returns a logical vector.
 </p>
 </li>
 <li> <p><code>fifo()</code> is now supported on Windows. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15600">PR#15600</a>)
 </p>
 </li>
 <li> <p><code>sort.list(method = "radix")</code> now allows negative
 integers (wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15644">PR#15644</a>).
 </p>
 </li>
 <li><p> Some functionality of <code>print.ts()</code> is now available in
 <code>.preformat.ts()</code> for more modularity.
 </p>
 </li>
 <li> <p><code>mcparallel()</code> gains an option <code>detach = TRUE</code>
 which allows execution of code independently of the current
 session.  It is based on a new <code>estranged = TRUE</code> argument to
 <code>mcfork()</code> which forks child processes such that they become
 independent of the parent process.
 </p>
 </li>
 <li><p> The <code>pdf()</code> device omits circles and text at extremely
 small sizes, since some viewers were failing on such files.
 </p>
 </li>
 <li><p> The rightmost break for the <code>"months"</code>,
 <code>"quarters"</code> and <code>"years"</code> cases of
 <code>hist.POSIXlt()</code> has been increased by a day.  (Inter alia,
 fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=15717">PR#15717</a>.)
 </p>
 </li>
 <li><p> The handling of <code>DF[i,] &lt;- a</code> where <code>i</code> is of
 length 0 is improved.  (Inter alia, fixes <a href="https://bugs.R-project.org/show_bug.cgi?id=15718">PR#15718</a>.)
 </p>
 </li>
 <li> <p><code>hclust()</code> gains a new method <code>"ward.D2"</code> which
 implements Ward's method correctly.  The previous <code>"ward"</code>
 method is <code>"ward.D"</code> now, with the old name still working.
 Thanks to research and proposals by Pierre Legendre.
 </p>
 </li>
 <li><p> The <code>sunspot.month</code> dataset has been amended and
 updated from the official source, whereas the <code>sunspots</code> and
 <code>sunspot.year</code> datasets will remain immutable. The
 documentation and source links have been updated correspondingly.
 </p>
 </li>
 <li><p> The <code>summary()</code> method for <code>"lm"</code> fits warns if
 the fit is essentially perfect, as most of the summary may be
 computed inaccurately (and with platform-dependent values).
 </p>
 <p>Programmers who use <code>summary()</code> in order to extract just
 a component which will be reliable (e.g., <code>$cov.unscaled</code>)
 should wrap their calls in <code>suppressWarnings()</code>.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The included version of LAPACK has been updated to 3.5.0.
 </p>
 </li>
 <li><p> There is some support for parallel testing of an
 installation, by setting <span class="env">TEST_MC_CORES</span> to an integer
 greater than one to indicate the maximum number of cores to be
 used in parallel. (It is worth specifying at least 8 cores if
 available.)  Most of these require a <code>make</code> program (such
 as GNU <code>make</code> and <code>dmake</code>) which supports the
 <code>$MAKE -j nproc</code> syntax.
 </p>
 <p>Except on Windows: the tests of standard package examples in
 <code>make check</code> are done in parallel.  This also applies to
 running <code>tools::testInstalledPackages()</code>.
 </p>
 <p>The more time-consuming regression tests are done in parallel.
 </p>
 <p>The package checks in <code>make check-devel</code> and <code>make
 	check-recommended</code> are done in parallel.
 </p>
 </li>
 <li><p> More of <code>make check</code> will work if recommended packages
 are not installed: but recommended packages remain needed for
 thorough checking of an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> build.
 </p>
 </li>
 <li><p> The version of <span class="samp">tzcode</span> included in
 &lsquo;<span class="file">src/extra/tzone</span>&rsquo; has been updated.  (Formerly used only on
 Windows.)
 </p>
 </li>
 <li><p> The included (64-bit) time-zone conversion code and Olson
 time-zone database can be used instead of the system version: use
 <code>configure</code> option <span class="option">--with-internal-tzcode</span>.  This
 is the default on Windows and OS X.  (Note that this does not
 currently work if a non-default <code>rsharedir</code>
 <code>configure</code> variable is used.)
 </p>
 <p>(It might be necessary to set environment variable <span class="env">TZ</span> on
 OSes where this is not already set, although the system timezone is
 deduced correctly on at least Linux, OS X and Windows.)
 </p>
 <p>This option also switches to the version of <code>strftime</code>
 included in directory &lsquo;<span class="file">src/extra/tzone</span>&rsquo;.
 </p>
 </li>
 <li> <p><code>configure</code> now tests for a C++11-compliant compiler
 by testing some basic features.  This by default tries flags for
 the compiler specified by <span class="samp">CXX</span>, but an alternative compiler,
 options and standard can be specified by variables <span class="samp">CXX1X</span>,
 <span class="samp">CXX1XFLAGS</span> and <span class="samp">CXX1XSTD</span> (e.g., <span class="samp">-std=gnu++11</span>).
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> can now optionally be compiled to use reference
 counting instead of the <code>NAMED</code> mechanism by defining
 <code>SWITCH_TO_REFCNT</code> in &lsquo;<span class="file">Rinternals.h</span>&rsquo;. This may become
 the default in the future.
 </p>
 </li>
 <li><p> There is a new option <span class="option">--use-system-tre</span> to use a
 suitable system <span class="pkg">tre</span> library: at present this means a version
 from their <code>git</code> repository, after corrections.
 (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15660">PR#15660</a>.)
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> The <code>CRANextra</code> repository is no longer a default
 repository on Windows: all the binary versions of packages from
 CRAN are now on CRAN, although <code>CRANextra</code> contains packages
 from Omegahat and elsewhere used by CRAN packages.
 </p>
 </li>
 <li><p> Only vignettes sources in directory &lsquo;<span class="file">vignettes</span>&rsquo; are
 considered to be vignettes and hence indexed as such.
 </p>
 </li>
 <li><p> In the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file, </p>
 <pre>    License: X11</pre>
 <p>is no longer recognized as valid.  Use <span class="samp">MIT</span> or
 <span class="samp">BSD_2_clause</span> instead, both of which need <span class="samp">+ file LICENSE</span>.
 </p>
 </li>
 <li><p> For consistency, entries in &lsquo;<span class="file">.Rinstignore</span>&rsquo; are now matched
 case-insensitively on all platforms.
 </p>
 </li>
 <li><p> Help for S4 methods with very long signatures now tries
 harder to split the description in the <span class="samp">Usage</span> field to no
 more than 80 characters per line (some packages had over 120
 characters).
 </p>
 </li>
 <li> <p><code>R CMD INSTALL --build</code> (not Windows) now defaults to
 the internal <code>tar()</code> unless <span class="env">R_INSTALL_TAR</span> is set.
 </p>
 </li>
 <li><p> There is support for compiling C++11 code in packages on
 suitable platforms: see &lsquo;Writing R Extensions&rsquo;.
 </p>
 </li>
 <li><p> Fake installs now install the contents of directory
 &lsquo;<span class="file">inst</span>&rsquo;: some packages use this to install e.g. C++ headers
 for use by other packages that are independent of the package
 itself.  Option <span class="option">--no-inst</span> can be used to get the previous
 behaviour.
 </p>
 </li></ul>


 <h4>DEBUGGING</h4>


 <ul>
 <li><p> The behaviour of the code browser has been made more
 consistent, in part following the suggestions in <a href="https://bugs.R-project.org/show_bug.cgi?id=14985">PR#14985</a>.
 </p>
 </li>
 <li><p> Calls to <code>browser()</code> are now consistent with calls
 to the browser triggered by <code>debug()</code>, in that <kbd>Enter</kbd>
 will default to <code>n</code> rather than <code>c</code>.
 </p>
 </li>
 <li><p> A new browser command <code>s</code> has been added, to
 &ldquo;step into&rdquo; function calls.
 </p>
 </li>
 <li><p> A new browser command <code>f</code> has been added, to
 &ldquo;finish&rdquo; the current loop or function.
 </p>
 </li>
 <li><p> Within the browser, the command <code>help</code> will
 display a short list of available commands.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li><p> Only vignettes sources in directory &lsquo;<span class="file">vignettes</span>&rsquo; are
 considered to be vignettes by <code>R CMD check</code>.  That has
 been the preferred location since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.14.0 and is now obligatory.
 </p>
 </li>
 <li><p> For consistency, <code>R CMD build</code> now matches entries
 in &lsquo;<span class="file">.Rbuildignore</span>&rsquo;  and &lsquo;<span class="file">vignettes/.install_extras</span>&rsquo;
 case-insensitively on all platforms (not just on Windows).
 </p>
 </li>
 <li> <p><code>checkFF()</code> (called by <code>R CMD check</code> by
 default) can optionally check foreign function calls for
 consistency with the registered type and argument count.  This is
 the default for <code>R CMD check --as-cran</code> or can be enabled
 by setting environment variable <span class="env">_R_CHECK_FF_CALLS_</span> to
 <span class="samp">registration</span> (but is in any case suppressed by
 <span class="option">--install=no</span>).   Because this checks calls in which
 <code>.NAME</code> is an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> object and not just a literal character
 string, some other problems are detected for such calls.
 </p>
 <p>Functions <code>suppressForeignCheck()</code> and <code>dontCheck()</code>
 have been added to allow package authors to suppress false
 positive reports.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> warns about a false value of
 the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; field <span class="samp">BuildVignettes</span> for Open Source
 packages, and ignores it.  (An Open Source package needs to have
 complete sources for its vignettes which should be usable on a
 suitably well-equipped system).
 </p>
 </li>
 <li> <p><code>R CMD check --no-rebuild-vignettes</code> is defunct:<br />
 <code>R CMD check --no-build-vignettes</code> has been preferred since
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0.
 </p>
 </li>
 <li> <p><code>R CMD build --no-vignettes</code> is defunct:<br />
 <code>R CMD build --no-build-vignettes</code> has been preferred since
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0.
 </p>
 </li>
 <li> <p><code>R CMD Sweave</code> and <code>R CMD Stangle</code> now
 process both Sweave and non-Sweave vignettes.  The
 <code>tools::buildVignette()</code> function has been added to do the
 same tasks from within <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li><p> The flags returned by <code>R CMD config --ldflags</code> and
 (where installed) <code>pkg-config --libs libR</code> are now those
 needed to link a front-end against the (shared or static) <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 library.
 </p>
 </li>
 <li> <p>&lsquo;<span class="file">Sweave.sty</span>&rsquo; has a new option <span class="samp">[inconsolata]</span>.
 </p>
 </li>
 <li> <p><code>R CMD check</code> customizations such as
 <span class="env">_R_CHECK_DEPENDS_ONLY_</span> make available packages only in
 <span class="samp">LinkingTo</span> only for installation, and not for
 loading/runtime tests.
 </p>
 </li>
 <li> <p><code>tools::checkFF()</code> reports on <code>.C</code> and
 <code>.Fortran</code> calls with <code>DUP = FALSE</code> if argument
 <code>check_DUP</code> is true.  This is selected by
 <code>R CMD check</code> by default.
 </p>
 </li>
 <li> <p><code>R CMD check --use-gct</code> can be tuned to
 garbage-collect less frequently using <code>gctorture2()</code>
 <em>via</em> the setting of environment variable
 <span class="env">_R_CHECK_GCT_N_</span>.
 </p>
 </li>
 <li><p> Where supported, <code>tools::texi2dvi()</code> limits the number
 of passes tried to 20.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> (Windows only) A function <code>R_WaitEvent()</code> has been
 added (with declaration in header&lsquo;<span class="file">R.h</span>&rsquo;) to block execution
 until the next event is received by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li><p> Remapping in the &lsquo;<span class="file">Rmath.h</span>&rsquo; header can be suppressed by
 defining <span class="samp">R_NO_REMAP_RMATH</span>.
 </p>
 </li>
 <li><p> The remapping of <code>rround()</code> in header &lsquo;<span class="file">Rmath.h</span>&rsquo;
 has been removed: use <code>fround()</code> instead.
 </p>
 </li>
 <li> <p><code>ftrunc()</code> in header &lsquo;<span class="file">Rmath.h</span>&rsquo; is now a wrapper
 for the C99 function <code>trunc()</code>, which might as well be used
 in C code: <code>ftrunc()</code> is still needed for portable C++ code.
 </p>
 </li>
 <li><p> The never-documented remapping of <code>prec()</code> to
 <code>fprec()</code> in header &lsquo;<span class="file">Rmath.h</span>&rsquo; has been removed.
 </p>
 </li>
 <li><p> The included LAPACK subset now contains <code>ZGESDD</code> and
 <code>ZGELSD</code>.
 </p>
 </li>
 <li><p> The function <code>LENGTH()</code> now checks that it is only
 applied to vector arguments.  However, in packages <code>length()</code>
 should be used. (In <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> itself <code>LENGTH()</code> is a macro without
 the function overhead of <code>length()</code>.)
 </p>
 </li>
 <li><p> Calls to <code>SET_VECTOR_ELT()</code> and <code>SET_STRING_ELT()</code>
 are now checked for indices which are in-range: several packages
 were writing one element beyond the allocated length.
 </p>
 </li>
 <li> <p><code>allocVector3</code> has been added which allows custom
 allocators to be used for individual vector allocations.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li> <p><code>chol(pivot = TRUE, LINPACK = TRUE)</code> is defunct.
 </p>
 <p>Arguments <code>EISPACK</code> for <code>eigen()</code> and <code>LINPACK</code> for
 <code>chol()</code>, <code>chol2inv()</code>, <code>solve()</code> and <code>svd()</code>
 are ignored: LAPACK is always used.
 </p>
 </li>
 <li> <p><code>.find.package()</code> and <code>.path.package()</code> are
 defunct: only the versions without the initial dot introduced in
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.13.0 have ever been in the API.
 </p>
 </li>
 <li><p> Partial matching when using the <code>$</code> operator <em>on
 data frames</em> now throws a warning and may become defunct in the
 future. If partial matching is intended, replace <code>foo$bar</code>
 by <code>foo[["bar", exact = FALSE]]</code>.
 </p>
 </li>
 <li><p> The long-deprecated use of <code>\synopsis</code> in the
 <span class="samp">Usage</span> section of &lsquo;<span class="file">.Rd</span>&rsquo; files has been removed: such
 sections are now ignored (with a warning).
 </p>
 </li>
 <li> <p><code>package.skeleton()</code>'s deprecated argument
 <code>namespace</code> has been removed.
 </p>
 </li>
 <li><p> Many methods are no longer exported by package <span class="pkg">stats</span>.
 They are all registered on their generic, which should be called
 rather than calling a method directly.
 </p>
 </li>
 <li><p> Functions <code>readNEWS()</code> and <code>checkNEWS()</code> in
 package <span class="pkg">tools</span> are defunct.
 </p>
 </li>
 <li> <p><code>download.file(method = "lynx")</code> is deprecated.
 </p>
 </li>
 <li> <p><code>.C(DUP = FALSE)</code> and <code>.Fortran(DUP = FALSE)</code> are
 now deprecated, and may be disabled in future versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  As
 their help has long said, <code>.Call()</code> is much preferred.
 </p>
 <p><code>R CMD check</code> notes such usages (by default).
 </p>
 </li>
 <li><p> The workaround of setting <span class="env">R_OSX_VALGRIND</span> has been
 removed: it is not needed in current valgrind.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> Calling <code>lm.wfit()</code> with no non-zero weights gave an
 array-overrun in the Fortran code and a not very sensible answer.
 It is now special-cased with a simpler answer (no <code>qr</code>
 component).
 </p>
 </li>
 <li><p> Error messages involving non-syntactic names (e.g., as produced
 by <code>`\r`</code> when that object does not exist) now encode
 the control characters.  (Reported by Hadley Wickham.)
 </p>
 </li>
 <li> <p><code>getGraphicsEvent()</code> caused 100% usage of one CPU in
 Windows. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15500">PR#15500</a>)
 </p>
 </li>
 <li> <p><code>nls()</code> with no <code>start</code> argument may now work
 inside another function (scoping issue).
 </p>
 </li>
 <li> <p><code>pbeta()</code> and similar work better for very large
 (billions) <code>ncp</code>.
 </p>
 </li>
 <li><p> Where time zones have changed abbreviations over the years,
 the software tries to more consistently use the abbreviation
 appropriate to the time or if that is unknown, the current
 abbreviation.  On some platforms where the C function
 <code>localtime</code> changed the <code>tzname</code> variables the reported
 abbreviation could have been that of the last time converted.
 </p>
 </li>
 <li> <p><code>all.equal(list(1), identity)</code> now works.
 </p>
 </li>
 <li><p> Bug fix for pushing viewports in <span class="pkg">grid</span> (reported by
 JJ Allaire and Kevin Ushey).
 </p>
 <p>NOTE for anyone poking around within the graphics engine display
 list (despite the warnings not to) that this changes what
 is recorded by <span class="pkg">grid</span> on the graphics engine display list.
 </p>
 </li>
 <li><p> Extra checks have been added for unit resolution and
 conversion in <span class="pkg">grid</span>, to catch instances of division-by-zero.
 This may introduce error messages in existing code and/or produce
 a different result in existing code (but only where a non-finite
 location or dimension may now become zero).
 </p>
 </li>
 <li><p> Some bugs in TRE have been corrected by updating from the
 <code>git</code> repository.  This allows <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> to be installed on some
 platforms for which this was a blocker (<a href="https://bugs.R-project.org/show_bug.cgi?id=15087">PR#15087</a> suggests Linux
 on ARM and HP-UX).
 </p>
 </li>
 <li> <p><code>?</code> applied to a call to an S4 generic failed in
 several cases. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15680">PR#15680</a>)
 </p>
 </li>
 <li><p> The implicit S4 generics for primitives with <code>...</code> in
 their argument list were incorrect. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15690">PR#15690</a>)
 </p>
 </li>
 <li><p> Bug fixes to <code>methods::callGeneric()</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15691">PR#15691</a>)
 </p>
 </li>
 <li><p> The bug fix to <code>aggregrate()</code> in <a href="https://bugs.R-project.org/show_bug.cgi?id=15004">PR#15004</a> introduced
 a new bug in the case of no grouping variables. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15699">PR#15699</a>)
 </p>
 </li>
 <li><p> In rare cases printing deeply nested lists overran a buffer
 by one byte and on a few platforms segfaulted. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15679">PR#15679</a>)
 </p>
 </li>
 <li><p> The dendrogram method of <code>as.dendrogram()</code> was hidden
 accidentally, (<a href="https://bugs.R-project.org/show_bug.cgi?id=15703">PR#15703</a>), and <code>order.dendrogram(d)</code> gave
 too much for a leaf <code>d</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15702">PR#15702</a>)
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> would try to kill processes on exit that have pids ever
 used by a child process spawned by <code>mcparallel</code> even though
 the current process with that pid was not actually its child.
 </p>
 </li>
 <li> <p><code>cophenetic()</code> applied to a <code>"dendrogram"</code> object
 sometimes incorrectly returned a <code>"Labels"</code> attribute with
 dimensions.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15706">PR#15706</a>)
 </p>
 </li>
 <li> <p><code>printCoefmat()</code> called from quite a few <code>print()</code>
 methods now obeys small <code>getOption("width")</code> settings,
 line wrapping the <span class="samp">"signif. codes"</span> legend appropriately.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15708">PR#15708</a>)
 </p>
 </li>
 <li> <p><code>model.matrix()</code> assumed that the stored dimnames for a
 matrix was <code>NULL</code> or length 2, but length 1 occurred.
 </p>
 </li>
 <li><p> The clipping region for a device was sometimes used in base
 graphics before it was set.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.0.3</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> On Windows there is support for making &lsquo;<span class="file">.texi</span>&rsquo; manuals
 using <code>texinfo</code> 5.0 or later: the setting is in file
 &lsquo;<span class="file">src/gnuwin32/MkRules.dist</span>&rsquo;.
 </p>
 <p>A packaging of the Perl script and modules for <code>texinfo</code>
 5.2 has been made available at
 <a href="http://www.stats.ox.ac.uk/pub/Rtools/">http://www.stats.ox.ac.uk/pub/Rtools/</a>.
 </p>
 </li>
 <li> <p><code>write.table()</code> now handles matrices of
 <i>2^31</i> or more elements, for those with large amounts
 of patience and disc space.
 </p>
 </li>
 <li><p> There is a new function, <code>La_version()</code>, to report the
 version of LAPACK in use.
 </p>
 </li>
 <li><p> The HTML version of &lsquo;An Introduction to R&rsquo; now has
 links to PNG versions of the figures.
 </p>
 </li>
 <li><p> There is some support to produce manuals in ebook
 formats. (See &lsquo;<span class="file">doc/manual/Makefile</span>&rsquo;.  Suggested by Mauro
 Cavalcanti.)
 </p>
 </li>
 <li><p> On a Unix-alike <code>Sys.timezone()</code> returns <code>NA</code> if
 the environment variable <span class="env">TZ</span> is unset, to distinguish it from
 an empty string which on some OSes means the <span class="samp">UTC</span> time zone.
 </p>
 </li>
 <li><p> The backtick may now be escaped in strings, to allow names
 containing them to be constructed, e.g. <code>`\``</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15621">PR#15621</a>)
 </p>
 </li>
 <li> <p><code>read.table()</code>, <code>readLines()</code> and <code>scan()</code>
 now warn when an embedded nul is found in the input.  (Related to
 <a href="https://bugs.R-project.org/show_bug.cgi?id=15625">PR#15625</a> which was puzzled by the behaviour in this unsupported
 case.)
 </p>
 </li>
 <li><p> (Windows only.)  <code>file.symlink()</code> works around the
 undocumented restriction of the Windows system call to
 backslashes.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15631">PR#15631</a>.)
 </p>
 </li>
 <li> <p><code>KalmanForecast(fast = FALSE)</code> is now the default, and
 the help contains an example of how <code>fast = TRUE</code> can be used
 in this version.  (The usage will change in 3.1.0.)
 </p>
 </li>
 <li> <p><code>strptime()</code> now checks the locale only when
 locale-specific formats are used and caches the locale in use:
 this can halve the time taken on OSes with slow system
 functions (e.g., OS X).
 </p>
 </li>
 <li> <p><code>strptime()</code> and the <code>format()</code> methods for
 classes <code>"POSIXct"</code>, <code>"POSIXlt"</code> and <code>"Date"</code>
 recognize strings with marked encodings: this allows, for example,
 UTF-8 French month names to be read on (French) Windows.
 </p>
 </li>
 <li> <p><code>iconv(to = "utf8")</code> is now accepted on all platforms
 (some implementations did already, but GNU <span class="pkg">libiconv</span> did not:
 however converted strings were not marked as being in UTF-8).  The
 official name, <code>"UTF-8"</code> is still preferred.
 </p>
 </li>
 <li> <p><code>available.packages()</code> is better protected against
 corrupt metadata files.  (A recurring problem with Debian package
 <span class="pkg">shogun-r</span>: <a href="https://bugs.R-project.org/show_bug.cgi?id=14713">PR#14713</a>.)
 </p>
 </li>
 <li><p> Finalizers are marked to be run at garbage collection, but
 run only at a somewhat safer later time (when interrupts are
 checked).  This circumvents some problems with finalizers running
 arbitrary code during garbage collection (the known instances being
 running <code>options()</code> and (C-level) <code>path.expand()</code>
 re-entrantly).
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The included version of PCRE has been updated to 8.34.  This
 fixes bugs and makes the behaviour closer to Perl 5.18.  In
 particular, the concept of &lsquo;space&rsquo; includes <span class="samp">VT</span> and
 hence agrees with POSIX's.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> The new field <span class="samp">SysDataCompression</span> in the
 &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file allows user control over the compression
 used for &lsquo;<span class="file">sysdata.rda</span>&rsquo; objects in the lazy-load database.
 </p>
 </li>
 <li> <p><code>install.packages(dependencies = value)</code> for <code>value =
 	NA</code> (the default) or <code>value = TRUE</code> omits packages only in
 <code>LinkingTo</code> for binary package installs.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> The long undocumented remapping of <code>rround()</code> to
 <code>Rf_fround()</code> in header &lsquo;<span class="file">Rmath.h</span>&rsquo; is now formally
 deprecated: use <code>fround()</code> directly.
 </p>
 </li>
 <li><p> Remapping of <code>prec()</code> and <code>trunc()</code> in the
 &lsquo;<span class="file">Rmath.h</span>&rsquo; header has been disabled in C++ code (it has caused
 breakage with <code>libc++</code> headers).
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>getParseData()</code> truncated the imaginary part of
 complex number constants.  (Reported by Yihui Xie.)
 </p>
 </li>
 <li> <p><code>dbeta(x, a, b)</code> with <code>a</code> or <code>b</code> within a
 factor of 2 of the largest representable number could
 infinite-loop.  (Reported by Ioannis Kosmidis.)
 </p>
 </li>
 <li> <p><code>provideDimnames()</code> failed for arrays with a 0
 dimension. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15465">PR#15465</a>)
 </p>
 </li>
 <li> <p><code>rbind()</code> and <code>cbind()</code> did not handle
 list objects correctly. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15468">PR#15468</a>)
 </p>
 </li>
 <li> <p><code>replayPlot()</code> now checks if it is replaying a plot
 from the same session.
 </p>
 </li>
 <li> <p><code>rasterImage()</code> and <code>grid.raster()</code> now give
 error on an empty (zero-length) raster.  (Reported by Ben North.)
 </p>
 </li>
 <li> <p><code>plot.lm()</code> would sometimes scramble the labels
 in plot type 5.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15458">PR#15458</a> and <a href="https://bugs.R-project.org/show_bug.cgi?id=14837">PR#14837</a>)
 </p>
 </li>
 <li> <p><code>min()</code> did not handle <code>NA_character_</code> values
 properly.  (Reported by Magnus Thor Torfason.)
 </p>
 </li>
 <li><p> (Windows only.)  <code>readRegistry()</code> would duplicate
 default values for keys.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15455">PR#15455</a>)
 </p>
 </li>
 <li> <p><code>str(..., strict.width = "cut")</code> did not handle
 it properly when more than one line needed to be cut.  (Reported
 by Gerrit Eichner.)
 </p>
 </li>
 <li><p> Removing subclass back-references when S4 classes were
 removed or their namespace unloaded had several bugs (e.g., <a href="https://bugs.R-project.org/show_bug.cgi?id=15481">PR#15481</a>).
 </p>
 </li>
 <li> <p><code>aggregate()</code> could fail when there were too many
 levels present in the <code>by</code> argument.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15004">PR#15004</a>)
 </p>
 </li>
 <li> <p><code>namespaceImportFrom()</code> needed to detect primitive
 functions when checking for duplicated imports (reported by
 Karl Forner).
 </p>
 </li>
 <li> <p><code>getGraphicsEvent()</code> did not exit when a user closed
 the graphics window.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15208">PR#15208</a>)
 </p>
 </li>
 <li><p> Errors in vignettes were not always captured and displayed
 properly.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15495">PR#15495</a>)
 </p>
 </li>
 <li> <p><code>contour()</code> could fail when dealing with extremely
 small z values.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15454">PR#15454</a>)
 </p>
 </li>
 <li><p> Several functions did not handle zero-length vectors properly,
 including <code>browseEnv()</code>, <code>format()</code>, <code>gl()</code>,
 <code>relist()</code> and <code>summary.data.frame()</code>.  (E.g., <a href="https://bugs.R-project.org/show_bug.cgi?id=15499">PR#15499</a>)
 </p>
 </li>
 <li> <p><code>Sweave()</code> did not restore the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> output to the
 console if it was interrupted by a user in the middle of evaluating
 a code chunk.  (Reported by Michael Sumner.)
 </p>
 </li>
 <li><p> Fake installs of packages with vignettes work again.
 </p>
 </li>
 <li><p> Illegal characters in the input caused <code>parse()</code>
 (and thus <code>source()</code>) to segfault.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15518">PR#15518</a>)
 </p>
 </li>
 <li><p> The nonsensical use of <code>nmax = 1</code> in
 <code>duplicated()</code> or <code>unique()</code> is now silently ignored.
 </p>
 </li>
 <li> <p><code>qcauchy(p, *)</code> is now fully accurate even when p is
 very close to 1. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15521">PR#15521</a>)
 </p>
 </li>
 <li><p> The <code>validmu()</code> and <code>valideta()</code> functions in the
 standard <code>glm()</code> families now also report non-finite values,
 rather than failing.
 </p>
 </li>
 <li><p> Saved vignette results (in a &lsquo;<span class="file">.Rout.save</span>&rsquo; file) were
 not being compared to the new ones during <code>R CMD check</code>.
 </p>
 </li>
 <li><p> Double-clicking outside of the list box (e.g., on the scrollbar)
 of a Tk listbox widget generated by <code>tk_select.list()</code> no
 longer causes the window to close. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15407">PR#15407</a>)
 </p>
 </li>
 <li><p> Improved handling of edge cases in
 <code>parallel::splitindices()</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15552">PR#15552</a>)
 </p>
 </li>
 <li><p> HTML display of results from <code>help.search()</code> and
 <code>??</code> sometimes contained badly constructed links.
 </p>
 </li>
 <li> <p><code>c()</code> and related functions such as <code>unlist()</code>
 converted raw vectors to invalid logical vectors. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15535">PR#15535</a>)
 </p>
 </li>
 <li><p> (Windows only) When a call to <code>system2()</code> specified
 one of <code>stdin</code>, <code>stdout</code> or <code>stderr</code> to be a file,
 but the command was not found (e.g., it contained its arguments,
 or the program was not on the <span class="env">PATH</span>), it left the file open
 and unusable until <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> terminated.  (Reported by Mathew McLean.)
 </p>
 </li>
 <li><p> The <code>bmp()</code> device was not recording <code>res = NA</code>
 correctly: it is now recorded as 72 ppi.
 </p>
 </li>
 <li><p> Several potential problems with compiler-specific behaviour
 have been identified using the &lsquo;Undefined Behaviour
 Sanitizer&rsquo; in conjunction with the <code>clang</code> compiler.
 </p>
 </li>
 <li> <p><code>hcl()</code> now honours <code>NA</code> inputs (previously they
 were mapped to black).
 </p>
 </li>
 <li><p> Some translations in base packages were being looked up in
 the main catalog rather than that for the package.
 </p>
 </li>
 <li><p> As a result of the 3.0.2 change about &lsquo;the last
 second before the epoch&rsquo;, most conversions which should have given
 <code>NA</code> returned that time.  (The platforms affected include
 Linux and OS X, but not Windows nor Solaris.)
 </p>
 </li>
 <li> <p><code>rowsum()</code> has more support for matrices and data frames
 with <i>2^31</i> or more elements. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15587">PR#15587</a>)
 </p>
 </li>
 <li> <p><code>predict(&lt;lm object&gt;, interval = "confidence", scale =
 	&lt;something&gt;)</code> now works.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15564">PR#15564</a>)
 </p>
 </li>
 <li><p> The bug fix in 3.0.2 for <a href="https://bugs.R-project.org/show_bug.cgi?id=15411">PR#15411</a> was too aggressive,
 and sometimes removed spaces that should not have been removed.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15583">PR#15583</a>)
 </p>
 </li>
 <li><p> Running <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> code in a <span class="pkg">tcltk</span> callback failed to set the
 busy flag, which will be needed to tell OS X not to &lsquo;App Nap&rsquo;.
 </p>
 </li>
 <li><p> The code for date-times before 1902 assumed that the offset
 from GMT in 1902 was a whole number of minutes: that was not true
 of Paris (as recorded on some platforms).
 </p>
 </li>
 <li><p> Using <code>Sys.setlocale</code> to set <code>LC_NUMERIC</code> to
 <code>"C"</code> (to restore the sane behavior) no longer gives a
 warning.
 </p>
 </li>
 <li> <p><code>deparse()</code> now deparses complex vectors in a way that
 re-parses to the original values.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15534">PR#15534</a>, patch based on code
 submitted by Alex Bertram.)
 </p>
 </li>
 <li><p> In some extreme cases (more than <i>10^15</i>)
 integer inputs to <code>dpqrxxx()</code> functions might have been
 rounded up by one (with a warning about being non-integer).
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15624">PR#15624</a>)
 </p>
 </li>
 <li><p> Plotting symbol <code>pch = 14</code> had the triangle upside down
 on some devices (typically screen devices). The triangle is
 supposed to be point up.  (Reported by Bill Venables.)
 </p>
 </li>
 <li> <p><code>getSrcref()</code> did not work on method definitions if
 <code>rematchDefinition()</code> had been used.
 </p>
 </li>
 <li> <p><code>KalmanForecast(fast = FALSE)</code> reported a (harmless)
 stack imbalance.
 </p>
 </li>
 <li><p> The count of observations used by <code>KalmanRun()</code> did not
 take missing values into account.
 </p>
 </li>
 <li><p> In locales where the abbreviated name of one month is a
 partial match for the full name of a later one, the <code>%B</code>
 format in <code>strptime()</code> could fail.  An example was French on
 OS X, where <span class="samp">juin</span> is abbreviated to <span class="samp">jui</span> and partially
 matches <code>juillet</code>.  Similarly for weekday names.
 </p>
 </li>
 <li> <p><code>pbeta(x, a, b, log.p = TRUE)</code> sometimes underflowed to
 zero for very small and very differently sized <code>a</code>, <code>b</code>.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15641">PR#15641</a>)
 </p>
 </li>
 <li> <p><code>approx()</code> and <code>approxfun()</code> now handle infinite
 values with the <code>"constant"</code> method. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15655">PR#15655</a>)
 </p>
 </li>
 <li> <p><code>stripchart()</code> again respects reversed limits in
 <code>xlim</code> and <code>ylim</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15664">PR#15664</a>)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.0.2</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li><p> The &lsquo;<span class="file">NEWS</span>&rsquo; files have been re-organized.
 </p>
 <p>This file contains news for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> &gt;= 3.0.0: news for the 0.x.y, 1.x.y
 and 2.x.y releases is in files &lsquo;<span class="file">NEWS.0</span>&rsquo;, &lsquo;<span class="file">NEWS.1</span>&rsquo; and
 &lsquo;<span class="file">NEWS.2</span>&rsquo;.  The latter files are now installed when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is
 installed.  An HTML version of news from 2.10.0 to 2.15.3 is
 available as &lsquo;<span class="file">doc/html/NEWS.2.html</span>&rsquo;.
 </p>
 </li>
 <li> <p><code>sum()</code> for integer arguments now uses an integer
 accumulator of at least 64 bits and so will be more accurate in
 the very rare case that a cumulative sum exceeds
 <i>2^53</i> (necessarily summing more than 4 million
 elements).
 </p>
 </li>
 <li><p> The <code>example()</code> and <code>tools::Rd2ex()</code> functions now
 have parameters to allow them to ignore <code>\dontrun</code> markup in
 examples.  (Suggested by Peter Solymos.)
 </p>
 </li>
 <li> <p><code>str(x)</code> is considerably faster for very large lists,
 or factors with 100,000 levels, the latter as in <a href="https://bugs.R-project.org/show_bug.cgi?id=15337">PR#15337</a>.
 </p>
 </li>
 <li> <p><code>col2rgb()</code> now converts factors to character strings
 not integer codes (suggested by Bryan Hanson).
 </p>
 </li>
 <li> <p><code>tail(warnings())</code> now works, <em>via</em> the new
 <code>`[`</code> method.
 </p>
 </li>
 <li><p> There is now support for the LaTeX style file &lsquo;<span class="file">zi4.sty</span>&rsquo;
 which has in some distributions replaced &lsquo;<span class="file">inconsolata.sty</span>&rsquo;.
 </p>
 </li>
 <li> <p><code>unlist(x)</code> now typically returns all non-list
 <code>x</code>s unchanged, not just the &ldquo;vector&rdquo; ones.
 Consequently, <code>format(lst)</code> now also works when the list
 <code>lst</code> has non-vector elements.
 </p>
 </li>
 <li><p> The <code>tools::getVignetteInfo()</code> function has been added
 to give information about installed vignettes.
 </p>
 </li>
 <li><p> New <code>assertCondition()</code>, etc. utilities in <span class="pkg">tools</span>, useful
 for testing.
 </p>
 </li>
 <li><p> Profiling now records non-inlined calls from byte-compiled
 code to <code>BUILTIN</code> functions.
 </p>
 </li>
 <li><p> Various functions in <span class="pkg">stats</span> and elsewhere that use
 non-standard evaluation are now more careful to follow the
 namespace scoping rules.  E.g., <code>stats::lm()</code> can now find
 <code>stats::model.frame()</code> even if <span class="pkg">stats</span> is not on the
 search path or if some package defines a function of that name.
 </p>
 </li>
 <li><p> If an invalid/corrupt <code>.Random.seed</code> object is
 encountered in the workspace it is ignored with a warning rather
 than giving an error.  (This allows <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> itself to rely on a working
 RNG, e.g. to choose a random port.)
 </p>
 </li>
 <li> <p><code>seq()</code> and <code>seq.int()</code> give more explicit error
 messages if called with invalid (e.g., <code>NaN</code>) inputs.
 </p>
 </li>
 <li><p> When <code>parse()</code> finds a syntax error, it now makes
 partial parse information available up to the location of the
 error.  (Request of Reijo Sund.)
 </p>
 </li>
 <li><p> Methods invoked by <code>NextMethod()</code> had a different
 dynamic parent to the generic. This was causing trouble where S3
 methods invoked <em>via</em> lazy evaluation could lose track of
 their generic.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15267">PR#15267</a>)
 </p>
 </li>
 <li><p> Code for the negative binomial distribution now treats the case
 <code>size == 0</code> as a one-point distribution at zero.
 </p>
 </li>
 <li> <p><code>abbreviate()</code> handles without warning non-ASCII input
 strings which require no abbreviation.
 </p>
 </li>
 <li> <p><code>read.dcf()</code> no longer has a limit of 8191 bytes per
 line. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15250">PR#15250</a>.)
 </p>
 </li>
 <li> <p><code>formatC(x)</code> no longer copies the class of <code>x</code> to
 the result, to avoid misuse creating invalid objects as in
 <a href="https://bugs.R-project.org/show_bug.cgi?id=15303">PR#15303</a>.  A warning is given if a class is discarded.
 </p>
 </li>
 <li><p> Dataset <code>npk</code> has been copied from <a href="https://CRAN.R-project.org/package=MASS"><span class="pkg">MASS</span></a> to
 allow more tests to be run without recommended packages being
 installed.
 </p>
 </li>
 <li><p> The initialization of the regression coefficients for
 non-degenerate differenced models in <code>arima()</code> has been
 changed and in some examples avoids a local maximum.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15396">PR#15396</a>)
 </p>
 </li>
 <li> <p><code>termplot()</code> now has an argument <code>transform.x</code>
 to control the display of individual terms in the plot.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15329">PR#15329</a>)
 </p>
 </li>
 <li> <p><code>format()</code> now supports <code>digits = 0</code>, to
 display <code>nsmall</code> decimal places.
 </p>
 </li>
 <li><p> There is a new read-only <code>par()</code> parameter called
 <code>"page"</code>, which returns a logical value indicating whether
 the next <code>plot.new()</code> call will start a new page.
 </p>
 </li>
 <li><p> Processing Sweave and Rd documents to PDF now renders
 backticks and single quotes better in several instances, including
 in <span class="samp">\code</span> and <span class="samp">\samp</span> expressions.
 </p>
 </li>
 <li> <p><code>utils::modifyList()</code> gets a new argument <code>keep.null</code>
 allowing <code>NULL</code> components in the replacement to be retained,
 instead of causing corresponding components to be deleted.
 </p>
 </li>
 <li> <p><code>tools::pkgVignettes()</code> gains argument <code>check</code>;
 if set to <code>TRUE</code>, it will warn when it appears a vignette requests
 a non-existent vignette engine.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check --as-cran</code> checks the line widths in
 usage and examples sections of the package Rd files.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> now implies <span class="option">--timings</span>.
 </p>
 </li>
 <li> <p><code>R CMD check</code> looks for command <code>gfile</code> if a
 suitable <code>file</code> is not found.  (Although <code>file</code> is
 not from GNU, OpenCSW on Solaris installs it as <code>gfile</code>.)
 </p>
 </li>
 <li> <p><code>R CMD build</code> (with the internal <code>tar</code>) checks
 the permissions of &lsquo;<span class="file">configure</span>&rsquo; and &lsquo;<span class="file">cleanup</span>&rsquo; files and
 adds execute permission to the recorded permissions for these
 files if needed, with a warning.  This is useful on OSes and file
 systems which do not support execute permissions (notably, on
 Windows).
 </p>
 </li>
 <li> <p><code>R CMD build</code> now weaves and tangles all vignettes,
 so suggested packages are not required during package installation
 if the source tarball was prepared with current
 <code>R CMD build</code>.
 </p>
 </li>
 <li> <p><code>checkFF()</code> (used by <code>R CMD check</code>) does a
 better job of detecting calls from other packages, including not
 reporting those where a function has been copied from another
 namespace (e.g., as a default method).  It now reports calls where
 <code>.NAME</code> is a symbol registered in another package.
 </p>
 </li>
 <li><p> On Unix-alike systems, <code>R CMD INSTALL</code> now installs packages
 group writably whenever the library (<code>lib.loc</code>) is group
 writable.  Hence, <code>update.packages()</code> works for other group
 members (suggested originally and from a patch by Dirk Eddelbuettel).
 </p>
 </li>
 <li> <p><code>R CMD javareconf</code> now supports the use of symbolic
 links for <span class="env">JAVA_HOME</span> on platforms which have
 <code>realpath</code>.  So it is now possible to
 use </p>
 <pre>R CMD javareconf JAVA_HOME=/usr/lib/jvm/java-1.7.0</pre>
 <p>on a Linux system and record that value rather than the
 frequently-changing full path such as
 &lsquo;<span class="file">/usr/lib/jvm/java-1.7.0-openjdk-1.7.0.25.x86_64</span>&rsquo;.
 </p>
 </li>
 <li><p> (Windows only.) <code>Rscript -e</code> requires a non-empty
 argument for consistency with Unix versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  (Also
 <code>Rterm -e</code> and <code>R -e</code>.)
 </p>
 </li>
 <li> <p><code>R CMD check</code> does more thorough checking of
 declared packages and namespaces.  It reports
 </p>

 <ul>
 <li><p> packages declared in more than one of the
 <span class="samp">Depends</span>, <span class="samp">Imports</span>, <span class="samp">Suggests</span> and
 <span class="samp">Enhances</span> fields of the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file.
 </p>
 </li>
 <li><p> namespaces declared in <span class="samp">Imports</span> but not
 imported from, neither in the &lsquo;<span class="file">NAMESPACE</span>&rsquo; file nor using
 the <code>::</code> nor <code>:::</code> operators.
 </p>
 </li>
 <li><p> packages which are used in <code>library()</code> or
 <code>requires()</code> calls in the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> code but were already put on
 the search path <em>via</em> <span class="samp">Depends</span>.
 </p>
 </li>
 <li><p> packages declared in <span class="samp">Depends</span> not imported
 <em>via</em> the &lsquo;<span class="file">NAMESPACE</span>&rsquo; file (except the standard
 packages).  Objects used from <span class="samp">Depends</span> packages should be
 imported to avoid conflicts and to allow correct operation when
 the namespace is loaded but not attached.
 </p>
 </li>
 <li><p> objects imported <em>via</em> <code>:::</code> calls where
 <code>::</code> would do.
 </p>
 </li>
 <li><p> objects imported by <code>::</code> which are not exported.
 </p>
 </li>
 <li><p> objects imported by <code>:::</code> calls which do not exist.
 </p>
 </li></ul>

 <p>See &lsquo;Writing R Extensions&rsquo; for good practice.
 </p>
 </li>
 <li> <p><code>R CMD check</code> optionally checks for non-standard
 top-level files and directories (which are often mistakes): this
 is enabled for <span class="option">--as-cran</span>.
 </p>
 </li>
 <li><p> LaTeX style file <code>upquote.sty</code> is no longer included
 (the version was several years old): it is no longer used in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 A much later version is commonly included in LaTeX distributions
 but does not play well with the <code>ae</code> fonts which are the
 default for Sweave vignettes.
 </p>
 </li>
 <li> <p><code>R CMD build</code> makes more use of the &lsquo;<span class="file">build</span>&rsquo;
 sub-directory of package sources, for example to record
 information about the vignettes.
 </p>
 </li>
 <li> <p><code>R CMD check</code> analyses <code>:::</code> calls.
 </p>
 </li></ul>


 <h4>INSTALLATION and INCLUDED SOFTWARE</h4>


 <ul>
 <li><p> The macros used for the texinfo manuals have been changed to
 work better with the incompatible changes made in <code>texinfo
       5.x</code>.
 </p>
 </li>
 <li><p> The minimum version for a system <code>xz</code> library is now
 5.0.3 (was 4.999).  This is in part to avoid 5.0.2, which can
 compress in ways other versions cannot decompress.
 </p>
 </li>
 <li><p> The included version of PCRE has been updated to 8.33.
 </p>
 </li>
 <li><p> The included version of <code>zlib</code> has been updated to
 1.2.8, a bug-fix release.
 </p>
 </li>
 <li><p> The included version of xz utils's <code>liblzma</code> has been
 updated to 5.0.5.
 </p>
 </li>
 <li><p> Since <code>javareconf</code> (see above) is used when <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is
 installed, a stable link for <span class="env">JAVA_HOME</span> can be supplied then.
 </p>
 </li>
 <li><p> Configuring with <span class="option">--disable-byte-compilation</span> will
 override the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; files of recommended packages,
 which typically require byte-compilation.
 </p>
 </li>
 <li><p> More of the installation and checking process will work even
 when <span class="env">TMPDIR</span> is set to a path containing spaces, but this is
 not recommended and external software (such as
 <code>texi2dvi</code>) may fail.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> Installation is aborted immediately if a <code>LinkingTo</code>
 package is not installed.
 </p>
 </li>
 <li> <p><code>R CMD INSTALL</code> has a new option
 <code>--no-byte-compile</code> which will override a <span class="samp">ByteCompile</span>
 field in the package's &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file.
 </p>
 </li>
 <li><p> License <span class="samp">BSD</span> is deprecated: use <span class="samp">BSD_3_clause</span>
 or <span class="samp">BSD_2_clause</span> instead.
 </p>
 <p>License <span class="samp">X11</span> is deprecated: use <span class="samp">MIT</span> or
 <span class="samp">BSD_2_clause</span> instead.
 </p>
 </li>
 <li><p> Version requirements for <code>LinkingTo</code> packages are now
 recognized: they are checked at installation.  (Fields with
 version requirements were previously silently ignored.)
 </p>
 </li>
 <li><p> The limit of 500 <code>S3method</code> entries in a
 <code>NAMESPACE</code> file has been removed.
 </p>
 </li>
 <li><p> The default &lsquo;version&rsquo; of Bioconductor for its
 packages has been changed to the upcoming <span class="samp">2.13</span>, but this
 can be set by the environment variable <span class="env">R_BIOC_VERSION</span> when
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is installed.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li> <p>&lsquo;<span class="file">Rdefines.h</span>&rsquo; has been tweaked so it can be included in
 C++ code after &lsquo;<span class="file">R_ext/Boolean.h</span>&rsquo; (which is included by
 &lsquo;<span class="file">R.h</span>&rsquo;).
 </p>
 <p>Note that &lsquo;<span class="file">Rdefines.h</span>&rsquo; is not kept up-to-date, and
 &lsquo;<span class="file">Rinternals.h</span>&rsquo; is preferred for new code.
 </p>
 </li>
 <li> <p><code>eval</code> and <code>applyClosure</code> are now protected
 against package code supplying an invalid <code>rho</code>.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> The unused <code>namespace</code> argument to
 <code>package.skeleton()</code> is now formally deprecated and will be
 removed in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.1.0.
 </p>
 </li>
 <li> <p><code>plclust()</code> is deprecated: use the <code>plot()</code> method
 for class <code>"hclust"</code> instead.
 </p>
 </li>
 <li><p> Functions <code>readNEWS()</code> and <code>checkNEWS()</code> in
 package <span class="pkg">tools</span> are deprecated (and they have not worked with
 current &lsquo;<span class="file">NEWS</span>&rsquo; files for a long time).
 </p>
 </li></ul>


 <h4>DOCUMENTATION</h4>


 <ul>
 <li> <p>&lsquo;An Introduction to R&rsquo; has a new chapter on using <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 as a scripting language including interacting with the OS.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>help.request()</code> could not determine the current version
 of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> on CRAN.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15241">PR#15241</a>)
 </p>
 </li>
 <li><p> On Windows, <code>file.info()</code> failed on root directories unless
 the path was terminated with an explicit <code>"."</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15302">PR#15302</a>)
 </p>
 </li>
 <li><p> The <code>regmatches&lt;-()</code> replacement function mishandled
 results coming from <code>regexpr()</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15311">PR#15311</a>)
 </p>
 </li>
 <li><p> The help for <code>setClass()</code> and
 <code>representation()</code> still suggested the deprecated argument
 <code>representation=</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15312">PR#15312</a>)
 </p>
 </li>
 <li> <p><code>R CMD config</code> failed in an installed build of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.0.1 (only) when a sub-architecture was used.  (Reported by
 Berwin Turlach.)
 </p>
 </li>
 <li><p> On Windows, the installer modified the &lsquo;<span class="file">etc/Rconsole</span>&rsquo;
 and &lsquo;<span class="file">etc/Rprofile.site</span>&rsquo; files even when default options were
 chosen, so the MD5 sums did not refer to the installed versions.
 (Reported by Tal Galili.)
 </p>
 </li>
 <li> <p><code>plot(hclust(), cex =)</code> respects <code>cex</code> again
 (and possibly others similarly).  (Reported by Peter Langfelder.)
 </p>
 </li>
 <li><p> If multiple packages were checked by <code>R CMD check</code>,
 and one was written for a different OS, it would set
 <code>--no-install</code> for all following packages as well as
 itself.
 </p>
 </li>
 <li> <p><code>qr.coef()</code> and related functions did not properly coerce
 real vectors to complex when necessary.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15332">PR#15332</a>)
 </p>
 </li>
 <li> <p><code>ftable(a)</code> now fixes up empty <code>dimnames</code> such
 that the result is printable.
 </p>
 </li>
 <li> <p><code>package.skeleton()</code> was not starting its search for
 function objects in the correct place if <code>environment</code> was
 supplied.  (Reported by Karl Forner.)
 </p>
 </li>
 <li><p> Parsing code was changing the length field of vectors and
 confusing the memory manager.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15345">PR#15345</a>)
 </p>
 </li>
 <li><p> The Fortran routine <code>ZHER2K</code> in the reference BLAS
 had a comment-out bug in two places.
 This caused trouble with <code>eigen()</code> for Hermitian matrices.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15345">PR#15345</a> and report from Robin Hankin)
 </p>
 </li>
 <li> <p><code>vignette()</code> and <code>browseVignettes()</code> did not
 display non-Sweave vignettes properly.
 </p>
 </li>
 <li><p> Two warning/error messages have been corrected:
 the (optional) warning produced by a partial name match
 with a pairlist, the error message from a zero-length argument to
 the <code>:</code> operator.  (Found by Radford Neal; <a href="https://bugs.R-project.org/show_bug.cgi?id=15358">PR#15358</a>,
 <a href="https://bugs.R-project.org/show_bug.cgi?id=15356">PR#15356</a>)
 </p>
 </li>
 <li> <p><code>svd()</code> returned <code>NULL</code> rather than omitting
 components as documented.  (Found by Radford Neal; <a href="https://bugs.R-project.org/show_bug.cgi?id=15360">PR#15360</a>)
 </p>
 </li>
 <li> <p><code>mclapply()</code> and <code>mcparallel()</code> with
 <code>silent = TRUE</code> could break a process that uses
 <code>stdout</code> output
 unguarded against broken pipes (e.g., <code>zip</code> will fail
 silently). To work around such issues, they now replace
 <code>stdout</code> with a descriptor pointed to &lsquo;<span class="file">/dev/null</span>&rsquo;
 instead. For this purpose, internal <code>closeStdout</code> and
 <code>closeStderr</code> functions have gained the <code>to.null</code> flag.
 </p>
 </li>
 <li> <p><code>log()</code>, <code>signif()</code> and <code>round()</code> now raise an
 error if a single named argument is not named <code>x</code>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15361">PR#15361</a>)
 </p>
 </li>
 <li> <p><code>deparse()</code> now deparses raw vectors in a form that
 is syntactically correct. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15369">PR#15369</a>)
 </p>
 </li>
 <li><p> The <code>jpeg</code> driver in Sweave created a JPEG file, but gave
 it a &lsquo;<span class="file">.png</span>&rsquo; extension. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15370">PR#15370</a>)
 </p>
 </li>
 <li><p> Deparsing of infix operators with named arguments is
 improved. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15350">PR#15350</a>)
 </p>
 </li>
 <li> <p><code>mget()</code>, <code>seq.int()</code> and <code>numericDeriv()</code>
 did not duplicate arguments properly. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15352">PR#15352</a>, <a href="https://bugs.R-project.org/show_bug.cgi?id=15353">PR#15353</a>,
 <a href="https://bugs.R-project.org/show_bug.cgi?id=15354">PR#15354</a>)
 </p>
 </li>
 <li> <p><code>kmeans(algorithm = "Hartigan-Wong")</code> now always stops
 iterating in the QTran stage. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15364">PR#15364</a>).
 </p>
 </li>
 <li> <p><code>read.dcf()</code> re-allocated incorrectly and so could
 segfault when called on a file with lines of more than 100 bytes.
 </p>
 </li>
 <li><p> On systems where <code>mktime()</code> does not set <code>errno</code>,
 the last second before the epoch could not be converted from
 <code>POSIXlt</code> to <code>POSIXct</code>. (Reported by Bill Dunlap.)
 </p>
 </li>
 <li> <p><code>add1.glm()</code> miscalculated F-statistics when df &gt; 1. (Bill
 Dunlap, <a href="https://bugs.R-project.org/show_bug.cgi?id=15386">PR#15386</a>).
 </p>
 </li>
 <li> <p><code>stem()</code> now discards infinite inputs rather than
 hanging.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15376">PR#15376</a>)
 </p>
 </li>
 <li><p> The parser now enforces C99 syntax for floating point
 hexadecimal constants (e.g., <code>0x1.1p0</code>), rather than returning
 unintended values for malformed constants. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15234">PR#15234</a>)
 </p>
 </li>
 <li> <p><code>model.matrix()</code> now works with very long LHS
 names (more than 500 bytes).  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15377">PR#15377</a>)
 </p>
 </li>
 <li> <p><code>integrate()</code> reverts to the pre-2.12.0 behaviour: from
 2.12.0 to 3.0.1 it sometimes failed to achieve the requested
 tolerance and reported error estimates that were exceeded.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15219">PR#15219</a>)
 </p>
 </li>
 <li> <p><code>strptime()</code> now handles <span class="samp">%W</span> fields with value
 0. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15915">PR#15915</a>)
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> is now better protected against people trying to interact
 with the console in startup code.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15325">PR#15325</a>)
 </p>
 </li>
 <li><p> Subsetting 1D arrays often lost dimnames (<a href="https://bugs.R-project.org/show_bug.cgi?id=15301">PR#15301</a>).
 </p>
 </li>
 <li><p> Unary <code>+</code> on a logical vector did not coerce to
 integer, although unary <code>-</code> did.
 </p>
 </li>
 <li> <p><code>na.omit()</code> and <code>na.exclude()</code> added a row
 to a zero-row data frame.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15399">PR#15399</a>)
 </p>
 </li>
 <li><p> All the (where necessary cut-down) vignettes are installed
 if <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> was configured with <span class="option">--without-recommended-packages</span>.
 </p>
 </li>
 <li> <p><code>source()</code> did not display filenames when reporting
 syntax errors.
 </p>
 </li>
 <li><p> Syntax error reports misplaced the caret pointing out the bad
 token.
 </p>
 </li>
 <li><p> (Windows only) Starting <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> with <code>R</code> (instead of
 <code>Rterm</code> or <code>Rgui</code>) would lose any
 zero-length strings from the command line arguments. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15406">PR#15406</a>)
 </p>
 </li>
 <li><p> Errors in the encoding specified on the command line
 <em>via</em> <code>--encoding=foo</code> were not handled properly.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15405">PR#15405</a>)
 </p>
 </li>
 <li><p> If <code>x</code> is a symbol, <code>is.vector(x, "name")</code> now
 returns <code>TRUE</code>, since <code>"name"</code> and <code>"symbol"</code>
 should be synonyms.  (Reported by Hervé Pagès.)
 </p>
 </li>
 <li> <p><code>R CMD rtags</code> works on platforms (such as OS X)
 with a XSI-conformant shell command <code>echo</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15231">PR#15231</a>)
 </p>
 </li>
 <li> <p><code>is.unsorted(NA)</code> returns false as documented (rather than
 <code>NA</code>).
 </p>
 </li>
 <li> <p><code>R CMD LINK</code> did not know about sub-architectures.
 </p>
 </li>
 <li> <p><code>system()</code> and <code>system2()</code> are better protected
 against users who misguidedly have spaces in the temporary
 directory path.
 </p>
 </li>
 <li> <p><code>file.show()</code> and <code>edit()</code> are now more likely to
 work on file paths containing spaces.  (Where external utilities
 are used, not the norm on Windows nor in <code>R.app</code> which
 should previously have worked.)
 </p>
 </li>
 <li><p> Packages using the <span class="pkg">methods</span> package are more likely to
 work when they import it but it is not attached.  (Several parts
 of its C code were looking for its <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> functions on the search path
 rather than in its namespace.)
 </p>
 </li>
 <li> <p><code>lgamma(-x)</code> is no longer <code>NaN</code> for very small x.
 </p>
 </li>
 <li><p> (Windows) <code>system2()</code> now respects specifying
 <code>stdout</code> and <code>stderr</code> as files if called from
 <code>Rgui</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15393">PR#15393</a>)
 </p>
 </li>
 <li><p> Closing an <code>x11()</code> device whilst <code>locator()</code> or
 <code>identify()</code> is in progress no longer hangs <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15253">PR#15253</a>)
 </p>
 </li>
 <li> <p><code>list.dirs(full.names = FALSE)</code> was not implemented.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15170">PR#15170</a>)
 </p>
 </li>
 <li> <p><code>format()</code> sometimes added unnecessary spaces.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15411">PR#15411</a>)
 </p>
 </li>
 <li> <p><code>all.equal(check.names = FALSE)</code> would ignore the request
 to ignore the names and would check them as attributes.
 </p>
 </li>
 <li><p> The symbol set by <code>tools::Rd2txt_options(itemBullet=)</code>
 was not respected in some locales. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15435">PR#15435</a>)
 </p>
 </li>
 <li> <p><code>mcMap()</code> was not exported by package
 <span class="pkg">parallel</span>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15439">PR#15439</a>)
 </p>
 </li>
 <li> <p><code>plot()</code> for <code>TukeyHSD</code> objects did not balance
 <code>dev.hold()</code> and <code>dev.flush()</code> calls on multi-page plots.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15449">PR#15449</a>)
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.0.1</h3>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>chooseCRANmirror()</code> and <code>chooseBioCmirror()</code> gain
 an <code>ind</code> argument (like <code>setRepositories()</code>).
 </p>
 </li>
 <li> <p><code>mcparallel</code> has a new argument <code>mc.interactive</code>
 which can modify the interactive flag in the child process. The
 new default is <code>FALSE</code> which makes child processes
 non-interactive by default (this prevents lock-ups due to children
 waiting for interactive input).
 </p>
 </li>
 <li> <p><code>scan()</code> now warns when end-of-file occurs within
 a quoted string.
 </p>
 </li>
 <li> <p><code>count.fields()</code> is now consistent with <code>scan()</code>
 in its handling of newlines in quoted strings.  Instead of
 triggering an error, this results in the current line receiving
 <code>NA</code> as the field count, with the next line getting the total
 count of the two lines.
 </p>
 </li>
 <li><p> The default method of <code>image()</code> will plot axes of the
 class of <code>xlim</code> and <code>ylim</code> (and hence of <code>x</code> and
 <code>y</code> if there is a suitable <code>range()</code> method).  Based on
 a suggestion of Michael Sumner.
 </p>
 </li>
 <li> <p><code>load()</code> now has a <code>verbose</code> argument for
 debugging support, to print the names of objects just before
 loading them.
 </p>
 </li>
 <li><p> When loading a serialized object encounters a reference to a
 namespace which cannot be loaded, this is replaced by a reference
 to the global environment, with a warning.
 </p>
 </li>
 <li> <p><code>pairs()</code> gains a <code>line.main</code> option for title placement.
 </p>
 </li>
 <li><p> The remaining instances in which serialization to a raw
 vector was limited to 2GB have been unlimited on a 64-bit
 platform, and in most cases serialization to a vector of more than
 1GB will be substantially faster.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD config</code> now make use of personal
 &lsquo;<span class="file">Makevars</span>&rsquo; files under &lsquo;<span class="file">~/.R</span>&rsquo; and a site file
 &lsquo;<span class="file">Makevars.site</span>&rsquo;, in the same way as <code>R CMD SHLIB</code> and
 <code>R CMD INSTALL</code>.  This makes the utility more useful in
 package <code>configure</code> scripts.
 </p>
 <p>On Windows finding the personal files may require the environment
 variable <span class="env">HOME</span> set.
 </p>
 <p>The old behaviour can be obtained with the new options
 <span class="option">--no-user-files</span> and <span class="option">--no-site-files</span>.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> Alternatives to the site and user customization files
 &lsquo;<span class="file">Makevars.site</span>&rsquo; and &lsquo;<span class="file">~/.R/Makevars</span>&rsquo; can be specified
 <em>via</em> the environment variables <span class="env">R_MAKEVARS_SITE</span> and
 <span class="env">R_MAKEVARS_USER</span> respectively.  These can be used to suppress
 the use of the default files by setting an empty value (where
 possible) or a non-existent path.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li> <p><code>sys.source()</code> did not report error locations when
 <code>keep.source = TRUE</code>.
 </p>
 </li>
 <li> <p><code>as.POSIXct.numeric</code> was coercing <code>origin</code> using
 the <code>tz</code> argument and not <code>"GMT"</code> as documented
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=14973">PR#14973</a>).
 </p>
 </li>
 <li><p> The active binding to assign fields in reference classes
 has been cleaned up to reduce dependence on the class' package
 environment, also fixing bug in initializing read-only fields
 (inspired by a report from Hadley Wickham).
 </p>
 </li>
 <li> <p><code>str(d)</code> no longer gives an error when <code>names(d)</code>
 contain illegal multibyte strings (<a href="https://bugs.R-project.org/show_bug.cgi?id=15247">PR#15247</a>).
 </p>
 </li>
 <li><p> Profiling of built-in functions with <code>line.profiling=
       TRUE</code> did not record the line from which they were called.
 </p>
 </li>
 <li> <p><code>citation(pkg)</code> dropped the header and footer specified
 in the &lsquo;<span class="file">CITATION</span>&rsquo; file (<a href="https://bugs.R-project.org/show_bug.cgi?id=15257">PR#15257</a>).
 </p>
 </li>
 <li><p> Quotes were handled differently when reading the first line
 and reading the rest, so <code>read.table()</code> misread some files
 that contained quote characters (<a href="https://bugs.R-project.org/show_bug.cgi?id=15245">PR#15245</a>).
 </p>
 </li>
 <li> <p><code>cat()</code> with <code>sep</code> a character vector of length
 greater than one and more than one argument was using separators
 inconsistently (<a href="https://bugs.R-project.org/show_bug.cgi?id=15261">PR#15261</a>).
 </p>
 </li>
 <li><p> On Windows in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0, <code>savePlot()</code> failed because of
 an incorrect check on the argument count.
 </p>
 </li>
 <li> <p><code>unzip(list = TRUE)</code> returned <code>Names</code> as a factor
 and not a character vector (as documented) for the internal method.
 (Noticed by Sean O'Riordain.)
 </p>
 </li>
 <li> <p><code>contourLines()</code> now checks more comprehensively for
 conformance of its <code>x</code>, <code>y</code> and <code>z</code> arguments (it
 was used incorrectly in package <a href="https://CRAN.R-project.org/package=R2G2"><span class="pkg">R2G2</span></a>).
 </p>
 </li>
 <li><p> Saved graphics display lists are <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> version-specific.
 Attempting to load workspaces containing them (or some other
 version-specific objects) aborted the load in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0 and
 earlier; now it does a partial load and generates a warning
 instead.
 </p>
 </li>
 <li><p> In <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0, <code>identify()</code> and <code>locator()</code> did
 not record information correctly, so replaying a graph (e.g., by
 copying it to another device) would fail. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15271">PR#15271</a>)
 </p>
 </li>
 <li><p> Calling <code>file.copy()</code> or <code>dirname()</code> with the
 invalid input <code>""</code> (which was being used in packages, despite
 not being a file path) could have caused a segfault.
 </p>
 <p><code>dirname("")</code> is now <code>""</code> rather than <code>"."</code> (unless
 it segfaulted).
 </p>
 </li>
 <li> <p><code>supsmu()</code> could read/write outside its input vectors
 for very short inputs (seen in package <a href="https://CRAN.R-project.org/package=rms"><span class="pkg">rms</span></a> for <code>n = 4</code>).
 </p>
 </li>
 <li> <p><code>as.dendrogram()</code>'s <code>hclust</code> method uses less
 memory and hence gets considerably faster for large (n ~ 1000)
 clusterings, thanks to Daniel Müllner. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15174">PR#15174</a>)
 </p>
 </li>
 <li><p> The return value when all workers failed from
 <code>parallel::mclapply(mc.preschedule = TRUE)</code> was a list of
 strings and not of error objects.  (Spotted by Karl Forner and
 Bernd Bischl.)
 </p>
 </li>
 <li><p> In <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0, when <code>help()</code> found multiple pages with
 the same alias, the HTML display of all the selections was not
 produced.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15282">PR#15282</a>)
 </p>
 </li>
 <li> <p><code>splinefun(method="monoH.FC")</code> now produces a
 function with first argument named <code>x</code> and allows
 <code>deriv=3</code>, as documented. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15273">PR#15273</a>)
 </p>
 </li>
 <li> <p><code>summaryRprof()</code> would only read the first
 <code>chunksize</code> lines of an <code>Rprof</code> file produced with
 <code>line.profiling=TRUE</code>. By default, this is the first 100
 seconds.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=15288">PR#15288</a>)
 </p>
 </li>
 <li> <p><code>lsfit()</code> produced an incorrect error message when
 argument <code>x</code> had more columns than rows or <code>x</code> had a
 different number of rows than <code>y</code>. (Spotted by Renaud Gaujoux.)
 </p>
 </li>
 <li><p> Binary operations on equal length vectors copied the
 class name from the second operand when the first had no
 class name, but did not set the object bit. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15299">PR#15299</a>)
 </p>
 </li>
 <li><p> The <code>trace()</code> method for reference generator objects
 failed after those objects became function definitions.
 </p>
 </li>
 <li> <p><code>write.table()</code> did not check that factors were
 constructed correctly, and so caused a segment fault when
 writing bad ones. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15300">PR#15300</a>)
 </p>
 </li>
 <li><p> The internal HTTP server no longer chokes on POST requests
 without body. It will also pass-through other request types for
 custom handlers (with the method stored in Request-Method header)
 instead of failing.
 </p>
 </li></ul>


 <h3><img src="../help/figures/../../html/Rlogo.svg" class="toplogo" alt="[R logo]" /> CHANGES IN R 3.0.0</h3>


 <h4>SIGNIFICANT USER-VISIBLE CHANGES</h4>


 <ul>
 <li><p> Packages need to be (re-)installed under this version
 (3.0.0) of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>.
 </p>
 </li>
 <li><p> There is a subtle change in behaviour for numeric index
 values <i>2^31</i> and larger.  These never used to be
 legitimate and so were treated as <code>NA</code>, sometimes with a
 warning.  They are now legal for long vectors so there is no
 longer a warning, and <code>x[2^31] &lt;- y</code> will now extend the
 vector on a 64-bit platform and give an error on a 32-bit one.
 </p>
 </li>
 <li><p> It is now possible for 64-bit builds to allocate amounts of
 memory limited only by the OS.  It may be wise to use OS
 facilities (e.g., <code>ulimit</code> in a <code>bash</code> shell,
 <code>limit</code> in <code>csh</code>), to set limits on overall memory
 consumption of an <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> process, particularly in a multi-user
 environment.  A number of packages need a limit of at least 4GB of
 virtual memory to load.
 </p>
 <p>64-bit Windows builds of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> are by default limited in memory usage
 to the amount of RAM installed: this limit can be changed by
 command-line option <span class="option">--max-mem-size</span> or setting environment
 variable <span class="env">R_MAX_MEM_SIZE</span>.
 </p>
 </li>
 <li><p> Negative numbers for colours are consistently an error:
 previously they were sometimes taken as transparent, sometimes
 mapped into the current palette and sometimes an error.
 </p>
 </li></ul>


 <h4>NEW FEATURES</h4>


 <ul>
 <li> <p><code>identical()</code> has a new argument,
 <code>ignore.environment</code>, used when comparing functions (with
 default <code>FALSE</code> as before).
 </p>
 </li>
 <li><p> There is a new option, <code>options(CBoundsCheck=)</code>, which
 controls how <code>.C()</code> and <code>.Fortran()</code> pass arguments to
 compiled code.  If true (which can be enabled by setting the
 environment variable <span class="env">R_C_BOUNDS_CHECK</span> to <span class="samp">yes</span>), raw,
 integer, double and complex arguments are always copied, and
 checked for writing off either end of the array on return from the
 compiled code (when a second copy is made).  This also checks
 individual elements of character vectors passed to <code>.C()</code>.
 </p>
 <p>This is not intended for routine use, but can be very helpful in
 finding segfaults in package code.
 </p>
 </li>
 <li><p> In <code>layout()</code>, the limits on the grid size have been
 raised (again).
 </p>
 </li>
 <li><p> New simple <code>provideDimnames()</code> utility function.
 </p>
 </li>
 <li><p> Where methods for <code>length()</code> return a double value
 which is representable as an integer (as often happens for package
 <a href="https://CRAN.R-project.org/package=Matrix"><span class="pkg">Matrix</span></a>), this is converted to an integer.
 </p>
 </li>
 <li><p> Matrix indexing of data frames by two-column numeric indices
 is now supported for replacement as well as extraction.
 </p>
 </li>
 <li> <p><code>setNames()</code> now has a default for its <code>object</code>
 argument, useful for a character result.
 </p>
 </li>
 <li> <p><code>StructTS()</code> has a revised additive constant in the
 <code>loglik</code> component of the result: the previous definition is
 returned as the <code>loglik0</code> component.  However, the help page has
 always warned of a lack of comparability of log-likelihoods for
 non-stationary models.  (Suggested by Jouni Helske.)
 </p>
 </li>
 <li><p> The logic in <code>aggregate.formula()</code> has been revised.
 It is now possible to use a formula stored in a variable;
 previously, it had to be given explicitly in the function call.
 </p>
 </li>
 <li> <p><code>install.packages()</code> has a new argument <code>quiet</code> to
 reduce the amount of output shown.
 </p>
 </li>
 <li><p> Setting an element of the graphics argument <code>lwd</code> to a
 negative or infinite value is now an error.  Lines corresponding
 to elements with values <code>NA</code> or <code>NaN</code> are silently
 omitted.
 </p>
 <p>Previously the behaviour was device-dependent.
 </p>
 </li>
 <li><p> Setting graphical parameters <code>cex</code>, <code>col</code>,
 <code>lty</code>, <code>lwd</code> and <code>pch</code> in <code>par()</code> now requires a
 length-one argument.   Previously some silently took the first
 element of a longer vector, but not always when documented to do so.
 </p>
 </li>
 <li> <p><code>Sys.which()</code> when used with inputs which would be
 unsafe in a shell (e.g., absolute paths containing spaces) now uses
 appropriate quoting.
 </p>
 </li>
 <li> <p><code>as.tclObj()</code> has been extended to handle raw vectors.
 Previously, it only worked in the other direction.
 (Contributed by Charlie Friedemann, <a href="https://bugs.R-project.org/show_bug.cgi?id=14939">PR#14939</a>.)
 </p>
 </li>
 <li><p> New functions <code>cite()</code> and <code>citeNatbib()</code> have
 been added, to allow generation of in-text citations from
 <code>"bibentry"</code> objects.  A <code>cite()</code> function may be added
 to <code>bibstyle()</code> environments.
 </p>
 </li>
 <li><p> A <code>sort()</code> method has been added for <code>"bibentry"</code>
 objects.
 </p>
 </li>
 <li><p> The <code>bibstyle()</code> function now defaults to setting the
 default bibliography style. The <code>getBibstyle()</code> function
 has been added to report the name of the current default style.
 </p>
 </li>
 <li> <p><code>scatter.smooth()</code> now has an argument <code>lpars</code> to
 pass arguments to <code>lines()</code>.
 </p>
 </li>
 <li> <p><code>pairs()</code> has a new <code>log</code> argument, to allow some
 or all variables to be plotted on logarithmic scale.
 (In part, wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=14919">PR#14919</a>.)
 </p>
 </li>
 <li> <p><code>split()</code> gains a <code>sep</code> argument.
 </p>
 </li>
 <li> <p><code>termplot()</code> does a better job when given a model with
 interactions (and no longer attempts to plot interaction terms).
 </p>
 </li>
 <li><p> The parser now incorporates code from Romain Francois'
 <a href="https://CRAN.R-project.org/package=parser"><span class="pkg">parser</span></a> package, to support more detailed computation on
 the code, such as syntax highlighting, comment-based
 documentation, etc.  Functions <code>getParseData()</code> and
 <code>getParseText()</code> access the data.
 </p>
 </li>
 <li><p> There is a new function <code>rep_len()</code> analogous to
 <code>rep.int()</code> for when speed is required (and names are not).
 </p>
 </li>
 <li><p> The undocumented use <code>rep(NULL, length.out = n)</code> for
 <code>n &gt; 0</code> (which returns <code>NULL</code>) now gives a warning.
 </p>
 </li>
 <li> <p><code>demo()</code> gains an <code>encoding</code> argument for those
 packages with non-ASCII demos: it defaults to the package encoding
 where there is one.
 </p>
 </li>
 <li> <p><code>strwrap()</code> converts inputs with a marked encoding to
 the current locale: previously it made some attempt to pass
 through as bytes inputs invalid in the current locale.
 </p>
 </li>
 <li><p> Specifying both <code>rate</code> and <code>scale</code> to
 <code>[dpqr]gamma</code> is a warning (if they are essentially the same
 value) or an error.
 </p>
 </li>
 <li> <p><code>merge()</code> works in more cases where the data frames
 include matrices.  (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=14974">PR#14974</a>.)
 </p>
 </li>
 <li> <p><code>optimize()</code> and <code>uniroot()</code> no longer use a
 shared parameter object across calls.  (<code>nlm()</code>,
 <code>nlminb()</code> and <code>optim()</code> with numerical derivatives
 still do, as documented.)
 </p>
 </li>
 <li><p> The <code>all.equal()</code> method for date-times is now
 documented: times are regarded as equal (by default) if they
 differ by up to 1 msec.
 </p>
 </li>
 <li> <p><code>duplicated()</code> and <code>unique()</code> gain a <code>nmax</code>
 argument which can be used to make them much more efficient when
 it is known that there are only a small number of unique entries.
 This is done automatically for factors.
 </p>
 </li>
 <li><p> Functions <code>rbinom()</code>, <code>rgeom()</code>, <code>rhyper()</code>,
 <code>rpois()</code>, <code>rnbinom(),</code> <code>rsignrank()</code> and
 <code>rwilcox()</code> now return integer (not double) vectors.  This
 halves the storage requirements for large simulations.
 </p>
 </li>
 <li> <p><code>sort()</code>, <code>sort.int()</code> and <code>sort.list()</code> now
 use radix sorting for factors of less than 100,000 levels when
 <code>method</code> is not supplied.  So does <code>order()</code> if called
 with a single factor, unless <code>na.last = NA</code>.
 </p>
 </li>
 <li> <p><code>diag()</code> as used to generate a diagonal matrix has been
 re-written in C for speed and less memory usage.  It now forces
 the result to be numeric in the case <code>diag(x)</code> since it is
 said to have &lsquo;zero off-diagonal entries&rsquo;.
 </p>
 </li>
 <li> <p><code>backsolve()</code> (and <code>forwardsolve()</code>) are now
 internal functions, for speed and support for large matrices.
 </p>
 </li>
 <li><p> More matrix algebra functions (e.g., <code>chol()</code> and
 <code>solve()</code>) accept logical matrices (and coerce to numeric).
 </p>
 </li>
 <li> <p><code>sample.int()</code> has some support for  <i>n &gt;= 2^31</i>: see its help for the limitations.
 </p>
 <p>A different algorithm is used for <code>(n, size, replace = FALSE,
 	prob = NULL)</code> for <code>n &gt; 1e7</code> and <code>size &lt;= n/2</code>.  This
 is much faster and uses less memory, but does give different results.
 </p>
 </li>
 <li> <p><code>approxfun()</code> and <code>splinefun()</code> now return a
 wrapper to an internal function in the <span class="pkg">stats</span> namespace
 rather than a <code>.C()</code> or <code>.Call()</code> call.  This is more
 likely to work if the function is saved and used in a different
 session.
 </p>
 </li>
 <li><p> The functions <code>.C()</code>, <code>.Call()</code>,
 <code>.External()</code> and <code>.Fortran()</code> now give an error (rather
 than a warning) if called with a named first argument.
 </p>
 </li>
 <li> <p><code>Sweave()</code> by default now reports the locations in
 the source file(s) of each chunk.
 </p>
 </li>
 <li> <p><code>clearPushBack()</code> is now a documented interface to a
 long-existing internal call.
 </p>
 </li>
 <li> <p><code>aspell()</code> gains filters for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> code, Debian Control
 Format and message catalog files, and support for <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> level
 dictionaries.  In addition, package <span class="pkg">utils</span> now provides
 functions <code>aspell_package_R_files()</code> and
 <code>aspell_package_C_files()</code> for spell checking <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> and C level
 message strings in packages.
 </p>
 </li>
 <li> <p><code>bibentry()</code> gains some support for &ldquo;incomplete&rdquo;
 entries with a <span class="samp">crossref</span> field.
 </p>
 </li>
 <li> <p><code>gray()</code> and <code>gray.colors()</code> finally allow
 <code>alpha</code> to be specified.
 </p>
 </li>
 <li> <p><code>monthplot()</code> gains parameters to control the look of
 the reference lines.  (Suggestion of Ian McLeod.)
 </p>
 </li>
 <li><p> Added support for new <code>%~%</code> relation
 (&ldquo;is distributed as&rdquo;) in plotmath.
 </p>
 </li>
 <li> <p><code>domain = NA</code> is accepted by <code>gettext()</code> and
 <code>ngettext()</code>, analogously to <code>stop()</code> etc.
 </p>
 </li>
 <li> <p><code>termplot()</code> gains a new argument <code>plot = FALSE</code>
 which returns information to allow the plots to be modified for
 use as part of other plots, but does not plot them.
 (Contributed by Terry Therneau, <a href="https://bugs.R-project.org/show_bug.cgi?id=15076">PR#15076</a>.)
 </p>
 </li>
 <li> <p><code>quartz.save()</code>, formerly an undocumented part of
 <code>R.app</code>, is now available to copy a device to a
 <code>quartz()</code> device.  <code>dev.copy2pdf()</code> optionally does
 this for PDF output: <code>quartz.save()</code> defaults to PNG.
 </p>
 </li>
 <li><p> The default method of <code>pairs()</code> now allows
 <code>text.panel = NULL</code> and the use of <code>&lt;foo&gt;.panel = NULL</code>
 is now documented.
 </p>
 </li>
 <li> <p><code>setRefClass()</code> and <code>getRefClass()</code> now return
 class generator functions, similar to <code>setClass()</code>, but
 still with the reference fields and methods as before
 (suggestion of Romain Francois).
 </p>
 </li>
 <li><p> New functions <code>bitwNot()</code>, <code>bitwAnd()</code>,
 <code>bitwOr()</code> and <code>bitwXor()</code>, using the internal
 interfaces previously used for classes <code>"octmode"</code> and
 <code>"hexmode"</code>.
 </p>
 <p>Also <code>bitwShiftL()</code> and <code>bitwShiftR()</code> for shifting bits
 in elements of integer vectors.
 </p>
 </li>
 <li><p> New option <code>"deparse.cutoff"</code> to control the deparsing
 of language objects such as calls and formulae when printing.
 (Suggested by a comment of Sarah Goslee.)
 </p>
 </li>
 <li> <p><code>colors()</code> gains an argument <code>distinct</code>.
 </p>
 </li>
 <li><p> New <code>demo(colors)</code> and <code>demo(hclColors)</code>, with
 utility functions.
 </p>
 </li>
 <li> <p><code>list.files()</code> (aka <code>dir()</code>) gains a new optional
 argument <code>no..</code> which allows to exclude <code>"."</code> and
 <code>".."</code> from listings.
 </p>
 </li>
 <li><p> Multiple time series are also of class <code>"matrix"</code>;
 consequently, <code>head()</code>, e.g., is more useful.
 </p>
 </li>
 <li> <p><code>encodeString()</code> preserves UTF-8 marked encodings.
 Thus if factor levels are marked as UTF-8 an attempt is made to
 print them in UTF-8 in <code>RGui</code> on Windows.
 </p>
 </li>
 <li> <p><code>readLines()</code> and <code>scan()</code> (and hence
 <code>read.table()</code>) in a UTF-8 locale now discard a UTF-8
 byte-order-mark (BOM).  Such BOMs are allowed but not recommended
 by the Unicode Standard: however Microsoft applications can
 produce them and so they are sometimes found on websites.
 </p>
 <p>The encoding name <code>"UTF-8-BOM"</code> for a connection will
 ensure that a UTF-8 BOM is discarded.
 </p>
 </li>
 <li> <p><code>mapply(FUN, a1, ..)</code> now also works when <code>a1</code> (or
 a further such argument) needs a <code>length()</code> method (which the
 documented arguments never do).  (Requested by Hervé Pagès; with a
 patch.)
 </p>
 </li>
 <li> <p><code>.onDetach()</code> is supported as an alternative to
 <code>.Last.lib</code>.  Unlike <code>.Last.lib</code>, this does not need to
 be exported from the package's namespace.
 </p>
 </li>
 <li><p> The <code>srcfile</code> argument to <code>parse()</code> may now be a
 character string, to be used in error messages.
 </p>
 </li>
 <li><p> The <code>format()</code> method for <code>ftable</code> objects gains
 a <code>method</code> argument, propagated to <code>write.ftable()</code> and
 <code>print()</code>, allowing more compact output, notably for LaTeX
 formatting, thanks to Marius Hofert.
 </p>
 </li>
 <li><p> The <code>utils::process.events()</code> function has been added
 to trigger immediate event handling.
 </p>
 </li>
 <li> <p><code>Sys.which()</code> now returns <code>NA</code> (not <code>""</code>) for
 <code>NA</code> inputs (related to <a href="https://bugs.R-project.org/show_bug.cgi?id=15147">PR#15147</a>).
 </p>
 </li>
 <li><p> The <code>print()</code> method for class <code>"htest"</code> gives
 fewer trailing spaces (wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15124">PR#15124</a>).
 </p>
 <p>Also print output from <code>HoltWinters()</code>, <code>nls()</code> and others.
 </p>
 </li>
 <li> <p><code>loadNamespace()</code> allows a version specification to be
 given, and this is used to check version specifications given in
 the <span class="samp">Imports</span> field when a namespace is loaded.
 </p>
 </li>
 <li> <p><code>setClass()</code> has a new argument, <code>slots</code>, clearer
 and less ambiguous than <code>representation</code>.  It is recommended
 for future code, but should be back-compatible.  At the same time,
 the allowed slot specification is slightly more general.  See the
 documentation for details.
 </p>
 </li>
 <li> <p><code>mget()</code> now has a default for <code>envir</code> (the frame
 from which it is called), for consistency with <code>get()</code> and
 <code>assign()</code>.
 </p>
 </li>
 <li> <p><code>close()</code> now returns an integer status where available,
 invisibly. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15088">PR#15088</a>.)
 </p>
 </li>
 <li><p> The internal method of <code>tar()</code> can now store paths too
 long for the <span class="samp">ustar</span> format, using the (widely supported) GNU
 extension.  It can also store long link names, but these are much
 less widely supported.  There is support for larger files, up to
 the <span class="samp">ustar</span> limit of 8GB.
 </p>
 </li>
 <li><p> Local reference classes have been added to package
 <span class="pkg">methods</span>.  These are a technique for avoiding unneeded
 copying of large components of objects while retaining standard <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 functional behavior.  See <code>?LocalReferenceClasses</code>.
 </p>
 </li>
 <li> <p><code>untar()</code> has a new argument <code>restore_times</code> which
 if false (not the default) discards the times in the tarball.
 This is useful if they are incorrect (some tarballs submitted to
 <acronym><span class="acronym">CRAN</span></acronym> have times in a local time zone or many years in the
 past even though the standard required them to be in UTC).
 </p>
 </li>
 <li> <p><code>replayplot()</code> cannot (and will not attempt to) replay
 plots recorded under <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> &lt; 3.0.0.  It may crash the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session if
 an attempt is made to replay plots created in a different build of
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> &gt;= 3.0.0.
 </p>
 </li>
 <li><p> Palette changes get recorded on the display list, so
 replaying plots (including when resizing screen devices and using
 <code>dev.copy()</code>) will work better when the palette is changed
 during a plot.
 </p>
 </li>
 <li> <p><code>chol(pivot = TRUE)</code> now defaults to LAPACK, not LINPACK.
 </p>
 </li>
 <li><p> The <code>parse()</code> function has a new parameter
 <code>keep.source</code>, which defaults to <code>options("keep.source")</code>.
 </p>
 </li>
 <li><p> Profiling <em>via</em> <code>Rprof()</code> now optionally records
 information at the statement level, not just the function level.
 </p>
 </li>
 <li><p> The <code>Rprof()</code> function now quotes function names in
 in its output file on Windows, to be consistent with the quoting
 in Unix.
 </p>
 </li>
 <li><p> Profiling <em>via</em> <code>Rprof()</code> now optionally records
 information about time spent in GC.
 </p>
 </li>
 <li><p> The HTML help page for a package now displays non-vignette
 documentation files in a more accessible format.
 </p>
 </li>
 <li><p> To support <code>options(stringsAsFactors = FALSE)</code>,
 <code>model.frame()</code>, <code>model.matrix()</code> and
 <code>replications()</code> now automatically convert character
 vectors to factors without a warning.
 </p>
 </li>
 <li><p> The <code>print</code> method for objects of class <code>"table"</code>
 now detects tables with 0-extents and prints the results as, e.g.,
 <span class="samp">&lt; table of extent 0 x 1 x 2 &gt;</span>. (Wish of <a href="https://bugs.R-project.org/show_bug.cgi?id=15198">PR#15198</a>.)
 </p>
 </li>
 <li><p> Deparsing involving calls to anonymous functions has been
 made closer to reversible by the addition of extra parentheses.
 </p>
 </li>
 <li><p> The function <code>utils::packageName()</code> has been added as
 a lightweight version of <code>methods::getPackageName()</code>.
 </p>
 </li>
 <li> <p><code>find.package(lib.loc = NULL)</code> now treats loaded
 namespaces preferentially in the same way as attached packages
 have been for a long time.
 </p>
 </li>
 <li><p> In Windows, the Change Directory dialog now defaults to
 the current working directory, rather than to the last directory
 chosen in that dialog.
 </p>
 </li>
 <li> <p><code>available.packages()</code> gains a
 <code>"license/restricts_use"</code> filter which retains only packages
 for which installation can proceed solely based on packages which
 are guaranteed not to restrict use.
 </p>
 </li>
 <li><p> New <code>check_packages_in_dir()</code> function in package
 <span class="pkg">tools</span> for conveniently checking source packages along with
 their reverse dependencies.
 </p>
 </li>
 <li><p> R's completion mechanism has been improved to handle help
 requests (starting with a question mark).  In particular, help
 prefixes are now supported, as well as quoted help topics.  To
 support this, completion inside quotes are now handled by R by
 default on all platforms.
 </p>
 </li>
 <li><p> The memory manager now allows the strategy used to balance
 garbage collection and memory growth to be controlled by setting
 the environment variable <span class="env">R_GC_MEM_GROW</span>. See <code>?Memory</code>
 for more details.
 </p>
 </li>
 <li><p> (&lsquo;For experts only&rsquo;, as the introductory manual
 says.)  The use of environment variables <span class="env">R_NSIZE</span> and
 <span class="env">R_VSIZE</span> to control the initial (= minimum) garbage
 collection trigger for number of cons cels and size of heap has
 been restored: they can be overridden by the command-line options
 <code>--min-nsize</code> and <code>--min-vsize</code>; see <code>?Memory</code>.
 </p>
 </li>
 <li><p> On Windows, the device name for bitmap devices as reported
 by <code>.Device</code> and <code>.Devices</code> no longer includes the file
 name.   This is for consistency with other platforms and was
 requested by the <a href="https://CRAN.R-project.org/package=lattice"><span class="pkg">lattice</span></a> maintainer.
 </p>
 <p><code>win.metafile()</code> still uses the file name: the exact form is
 used by package <a href="https://CRAN.R-project.org/package=tkrplot"><span class="pkg">tkrplot</span></a>.
 </p>
 </li>
 <li> <p><code>set.seed(NULL)</code> re-initializes <code>.Random.seed</code> as
 done at the beginning of the session if not already set.
 (Suggestion of Bill Dunlap.)
 </p>
 </li>
 <li><p> The <code>breaks</code> argument in <code>hist.default()</code> can now be
 a function that returns the breakpoints to be used (previously it
 could only return the suggested number of breakpoints).
 </p>
 </li>
 <li><p> File &lsquo;<span class="file">share/licenses/licenses.db</span>&rsquo; has some
 clarifications, especially as to which variants of &lsquo;BSD&rsquo;
 and &lsquo;MIT&rsquo; is intended and how to apply them to packages.
 The problematic licence &lsquo;Artistic-1.0&rsquo; has been removed.
 </p>
 </li></ul>


 <h4>LONG VECTORS</h4>

 <p>This section applies only to 64-bit platforms.
 </p>

 <ul>
 <li><p> There is support for vectors longer than <i>2^31
       - 1</i> elements.  This applies to raw, logical, integer, double,
 complex and character vectors, as well as lists.  (Elements of
 character vectors remain limited to <i>2^31 - 1</i>
 bytes.)
 </p>
 </li>
 <li><p> Most operations which can sensibly be done with long vectors
 work: others may return the error &lsquo;long vectors not
 supported yet&rsquo;.  Most of these are because they explicitly work
 with integer indices (e.g., <code>anyDuplicated()</code> and
 <code>match()</code>) or because other limits (e.g., of character strings
 or matrix dimensions) would be exceeded or the operations would be
 extremely slow.
 </p>
 </li>
 <li> <p><code>length()</code> returns a double for long vectors, and
 lengths can be set to <i>2^31</i> or more by the
 replacement function with a double value.
 </p>
 </li>
 <li><p> Most aspects of indexing are available.  Generally
 double-valued indices can be used to access elements beyond
 <i>2^31 - 1</i>.
 </p>
 </li>
 <li><p> There is some support for matrices and arrays with each
 dimension less than <i>2^31</i> but total number of
 elements more than that.  Only some aspects of matrix algebra work
 for such matrices, often taking a very long time.  In other cases
 the underlying Fortran code has an unstated restriction (as was
 found for complex <code>svd()</code>).
 </p>
 </li>
 <li> <p><code>dist()</code> can produce dissimilarity objects for more
 than 65536 rows (but for example <code>hclust()</code> cannot process
 such objects).
 </p>
 </li>
 <li> <p><code>serialize()</code> to a raw vector is unlimited in size
 (except by resources).
 </p>
 </li>
 <li><p> The C-level function <code>R_alloc</code> can now allocate
 <i>2^35</i> or more bytes.
 </p>
 </li>
 <li> <p><code>agrep()</code> and <code>grep()</code> will return double vectors
 of indices for long vector inputs.
 </p>
 </li>
 <li><p> Many calls to <code>.C()</code> have been replaced by
 <code>.Call()</code> to allow long vectors to be supported (now or in
 the future).  Regrettably several packages had copied the non-API
 <code>.C()</code> calls and so failed.
 </p>
 </li>
 <li> <p><code>.C()</code> and <code>.Fortran()</code> do not accept long vector
 inputs.  This is a precaution as it is very unlikely that existing
 code will have been written to handle long vectors (and the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 wrappers often assume that <code>length(x)</code> is an integer).
 </p>
 </li>
 <li><p> Most of the methods for <code>sort()</code> work for long vectors.
 </p>
 <p><code>rank()</code>, <code>sort.list()</code> and <code>order()</code> support
 long vectors (slowly except for radix sorting).
 </p>
 </li>
 <li> <p><code>sample()</code> can do uniform sampling from a long vector.
 </p>
 </li></ul>


 <h4>PERFORMANCE IMPROVEMENTS</h4>


 <ul>
 <li><p> More use has been made of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> objects representing registered
 entry points, which is more efficient as the address is provided
 by the loader once only when the package is loaded.
 </p>
 <p>This has been done for packages <code>base</code>, <code>methods</code>,
 <code>splines</code> and <code>tcltk</code>: it was already in place for the
 other standard packages.
 </p>
 <p>Since these entry points are always accessed by the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> entry
 points they do not need to be in the load table which can be
 substantially smaller and hence searched faster.  This does mean
 that <code>.C</code> / <code>.Fortran</code> / <code>.Call</code> calls copied from
 earlier versions of <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> may no longer work &ndash; but they were never
 part of the API.
 </p>
 </li>
 <li><p> Many <code>.Call()</code> calls in package <span class="pkg">base</span> have been
 migrated to <code>.Internal()</code> calls.
 </p>
 </li>
 <li> <p><code>solve()</code> makes fewer copies, especially when <code>b</code>
 is a vector rather than a matrix.
 </p>
 </li>
 <li> <p><code>eigen()</code> makes fewer copies if the input has dimnames.
 </p>
 </li>
 <li><p> Most of the linear algebra functions make fewer copies when
 the input(s) are not double (e.g., integer or logical).
 </p>
 </li>
 <li><p> A foreign function call (<code>.C()</code> etc) in a package
 without a <code>PACKAGE</code> argument will only look in the first DLL
 specified in the &lsquo;<span class="file">NAMESPACE</span>&rsquo; file of the package rather than
 searching all loaded DLLs.  A few packages needed <code>PACKAGE</code>
 arguments added.
 </p>
 </li>
 <li><p> The <code>@&lt;-</code> operator is now implemented as a primitive,
 which should reduce some copying of objects when used.  Note that
 the operator object must now be in package <span class="pkg">base</span>: do not try
 to import it explicitly from package <span class="pkg">methods</span>.
 </p>
 </li></ul>


 <h4>PACKAGE INSTALLATION</h4>


 <ul>
 <li><p> The transitional support for installing packages without
 namespaces (required since <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.14.0) has been removed.
 <code>R CMD build</code> will still add a namespace, but a
 <code>.First.lib()</code> function will need to be converted.
 </p>
 <p><code>R CMD INSTALL</code> no longer adds a namespace (so
 installation will fail), and a <code>.First.lib()</code> function in a
 package will be ignored (with an installation warning for now).
 </p>
 <p>As an exception, packages without a &lsquo;<span class="file">R</span>&rsquo; directory and no
 &lsquo;<span class="file">NAMESPACE</span>&rsquo; file can still be installed.
 </p>
 </li>
 <li><p> Packages can specify in their &lsquo;<span class="file">DESCRIPTION file</span>&rsquo; a line
 like </p>
 <pre>     Biarch: yes
 </pre>
 <p>to be installed on Windows with <span class="option">--force-biarch</span>.
 </p>
 </li>
 <li><p> Package vignettes can now be processed by other engines
 besides <code>Sweave</code>; see &lsquo;Writing R Extensions&rsquo; and the
 <code>tools::vignetteEngine</code> help topic for details.
 </p>
 </li>
 <li><p> The &lsquo;<span class="file">*.R</span>&rsquo; tangled source code for vignettes is now
 included in tarballs when <code>R CMD build</code> is used to produce
 them.  In <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 3.0.0, &lsquo;<span class="file">*.R</span>&rsquo; files not in the sources will be
 produced at install time, but eventually this will be dropped.
 </p>
 </li>
 <li><p> The package type <code>"mac.binary"</code> now looks in a path in
 the repository without any Mac subtype (which used to be
 <span class="samp">universal</span> or <span class="samp">leopard</span>): it looks in
 &lsquo;<span class="file">bin/macosx/contrib/3.0</span>&rsquo; rather than
 &lsquo;<span class="file">bin/macosx/leopard/contrib/2.15</span>&rsquo;).  This is the type used
 for the <acronym><span class="acronym">CRAN</span></acronym> binary distribution for OS X as from <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.0.0.
 </p>
 </li>
 <li><p> File &lsquo;<span class="file">etc/Makeconf</span>&rsquo; makes more use of the macros
 <code>$(CC)</code>, <code>$(CXX)</code>, <code>$(F77)</code> and <code>$(FC)</code>, so
 the compiler in use can be changed by setting just these (and if
 necessary the corresponding flags and <code>FLIBS</code>) in file
 &lsquo;<span class="file">~/.R/Makevars</span>&rsquo;.
 </p>
 <p>This is convenient for those working with binary distributions of
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>, e.g. on OS X.
 </p>
 </li></ul>


 <h4>UTILITIES</h4>


 <ul>
 <li> <p><code>R CMD check</code> now gives a warning rather than a
 note if it finds calls to <code>abort</code>, <code>assert</code> or
 <code>exit</code> in compiled code, and has been able to find the
 &lsquo;<span class="file">.o</span>&rsquo; file in which the calls occur.
 </p>
 <p>Such calls can terminate the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> process which loads the package.
 </p>
 </li>
 <li><p> The location of the build and check environment files can
 now be specified by the environment variables
 <span class="env">R_BUILD_ENVIRON</span> and <span class="env">R_CHECK_ENVIRON</span>, respectively.
 </p>
 </li>
 <li> <p><code>R CMD Sweave</code> gains a <span class="option">--compact</span> option
 to control possibly reducing the size of the PDF file it creates
 when <span class="option">--pdf</span> is given.
 </p>
 </li>
 <li> <p><code>R CMD build</code> now omits Eclipse's &lsquo;<span class="file">.metadata</span>&rsquo;
 directories, and <code>R CMD check</code> warns if it finds them.
 </p>
 </li>
 <li> <p><code>R CMD check</code> now does some checks on functions
 defined within reference classes, including of <code>.Call()</code> etc
 calls.
 </p>
 </li>
 <li> <p><code>R CMD check --as-cran</code> notes assignments to the
 global environment, calls to <code>data()</code> which load into the
 global environment, and calls to <code>attach()</code>.
 </p>
 </li>
 <li> <p><code>R CMD build</code> by default uses the internal method of
 <code>tar()</code> to prepare the tarball.  This is more likely to
 produce a tarball compatible with <code>R CMD INSTALL</code> and
 <code>R CMD check</code>: an external <code>tar</code> program,
 including options, can be specified <em>via</em> the environment
 variable <span class="env">R_BUILD_TAR</span>.
 </p>
 </li>
 <li> <p><code>tools::massageExamples()</code> is better protected against
 packages which re-define base functions such as <code>cat()</code> and
 <code>get()</code> and so can cause <code>R CMD check</code> to fail when
 checking examples.
 </p>
 </li>
 <li> <p><code>R CMD javareconf</code> has been enhanced to be more
 similar to the code used by <code>configure</code>.
 </p>
 <p>There is now a test that a JNI program can be compiled (like
 <code>configure</code> did) and only working settings are used.
 </p>
 <p>It makes use of custom settings from configuration recorded in
 &lsquo;<span class="file">etc/javaconf</span>&rsquo;.
 </p>
 </li>
 <li><p> The <span class="option">--no-vignettes</span> argument of <code>R CMD
 	build</code> has been renamed to the more accurate
 <span class="option">--no-build-vignettes</span>: its action has always been to
 (re)build vignettes and never omitted them.
 </p>
 <p><code>R CMD check</code> accepts <span class="option">--no-build-vignettes</span> as a
 preferred synonym for <span class="option">--no-rebuild-vignettes</span>.
 </p>
 </li></ul>


 <h4>DEPRECATED AND DEFUNCT</h4>


 <ul>
 <li><p> The <code>ENCODING</code> argument to <code>.C()</code> is defunct.
 Use <code>iconv()</code> instead.
 </p>
 </li>
 <li><p> The <code>.Internal(eval.with.vis)</code> non-API function has
 been removed.
 </p>
 </li>
 <li><p> Support for the converters for use with <code>.C()</code> has been
 removed, including the oft misused non-API header
 &lsquo;<span class="file">R_ext/RConverters.h</span>&rsquo;.
 </p>
 </li>
 <li><p> The previously deprecated uses of <code>array()</code> with a
 0-length <code>dim</code> argument and <code>tapply()</code> with a 0-length
 <code>INDEX</code> list are now errors.
 </p>
 </li>
 <li> <p><span class="samp">Translation</span> packages are defunct.
 </p>
 </li>
 <li><p> Calling <code>rep()</code> or <code>rep.int()</code> on a pairlist or
 other non-vector object is now an error.
 </p>
 </li>
 <li><p> Several non-API entry points have been transferred to
 packages (e.g., <code>R_zeroin2</code>) or replaced by different non-API
 entry points (e.g., <code>R_tabulate</code>).
 </p>
 </li>
 <li><p> The &lsquo;internal&rsquo; graphics device invoked by
 <code>.Call("R_GD_nullDevice", package = "grDevices")</code> has
 been removed: use <code>pdf(file = NULL)</code> instead.
 </p>
 </li>
 <li><p> The <code>.Fortran()</code> entry point <code>"dqrls"</code> which has
 not been used by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> since version 2.15.1 is no longer available.
 </p>
 </li>
 <li><p> Functions <code>traceOn()</code> and <code>traceOff()</code> in package
 <span class="pkg">methods</span> are now defunct.
 </p>
 </li>
 <li><p> Function <code>CRAN.packages()</code> is finally defunct.
 </p>
 </li>
 <li><p> Use of <code>col2rgb(0)</code> is defunct: use <code>par("bg")</code> or
 <code>NA</code> instead.
 </p>
 </li>
 <li><p> The long-defunct functions <code>Rd_parse()</code>,
 <code>anovalist.lm()</code>, <code>categpry()</code>, <code>clearNames()</code>,
 <code>gammaCody()</code>, <code>glm.fit.null()</code>, <code>lm.fit.null()</code>,
 <code>lm.wfit.null()</code>, <code>manglePackageNames()</code>,
 <code>mauchley.test()</code>, <code>package.contents()</code>,
 <code>print.coefmat()</code>, <code>reshapeLong()</code>,
 <code>reshapeWide()</code>, <code>tkclose()</code>, <code>tkcmd()</code>,
 <code>tkfile.dir()</code>, <code>tkfile.tail()</code>, <code>tkopen()</code>,
 <code>tkputs()</code>, <code>tkread()</code>, <code>trySilent()</code> and
 <code>zip.file.extract()</code> have been removed entirely (but are
 still documented in the help system).
 </p>
 </li>
 <li><p> The unused <code>dataPath</code> argument to
 <code>attachNamespace()</code> has been removed.
 </p>
 </li>
 <li> <p><code>grid.prompt()</code> has been removed: use
 <code>devAskNewPage()</code> instead.
 </p>
 </li>
 <li><p> The long-deprecated <code>intensities</code> component is no
 longer returned by <code>hist()</code>.
 </p>
 </li>
 <li> <p><code>mean()</code> for data frames and <code>sd()</code> for data
 frames and matrices are defunct.
 </p>
 </li>
 <li> <p><code>chol(pivot = FALSE, LINPACK = TRUE)</code>,
 <code>ch2inv(LINPACK = TRUE)</code>, <code>eigen(EISPACK = TRUE)</code>,
 <code>solve(LINPACK = TRUE)</code> and <code>svd(LINPACK = TRUE)</code> are
 defunct: LAPACK will be used, with a warning.
 </p>
 </li>
 <li><p> The <code>keep.source</code> argument to <code>library()</code> and
 <code>require()</code> is defunct.  This option needs to be set
 at install time.
 </p>
 </li>
 <li><p> Documentation for <code>real()</code>, <code>as.real()</code> and
 <code>is.real()</code> has been moved to &lsquo;defunct&rsquo; and the
 functions removed.
 </p>
 </li>
 <li><p> The <code>maxRasters</code> argument of <code>pdf()</code> (unused since
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> 2.14.0) has been removed.
 </p>
 </li>
 <li><p> The unused <code>fontsmooth</code> argument has been removed from
 the <code>quartz()</code> device.
 </p>
 </li>
 <li><p> All the (non-API) EISPACK entry points in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> have been removed.
 </p>
 </li>
 <li> <p><code>chol(pivot = TRUE, LINPACK = TRUE)</code> is deprecated.
 </p>
 </li>
 <li><p> The long-deprecated use of <code>\synopsis</code> in the
 <span class="samp">Usage</span> section of &lsquo;<span class="file">.Rd</span>&rsquo; files will be removed in <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 3.1.0.
 </p>
 </li>
 <li> <p><code>.find.package()</code> and <code>.path.package()</code> are
 deprecated: only the public versions without the dot have ever
 been in the API.
 </p>
 </li>
 <li><p> In a package's &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file, </p>
 <pre>    License: X11</pre>
 <p>is deprecated, since it includes
 &lsquo;Copyright (C) 1996 X Consortium&rsquo; which cannot be
 appropriate for a current <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> package.  Use &lsquo;MIT&rsquo; or
 &lsquo;BSD_2_clause&rsquo; instead.
 </p>
 </li></ul>


 <h4>CODE MIGRATION</h4>


 <ul>
 <li><p> The C code underlying base graphics has been migrated to the
 <span class="pkg">graphics</span> package (and hence no longer uses
 <code>.Internal()</code> calls).
 </p>
 </li>
 <li><p> Most of the <code>.Internal()</code> calls used in the <span class="pkg">stats</span>
 package have been migrated to C code in that package.
 </p>
 <p>This means that a number of <code>.Internal()</code> calls which have
 been used by packages no longer exist, including
 <code>.Internal(cor)</code> <code>.Internal(cov)</code>,
 <code>.Internal(optimhess)</code> and <code>.Internal(update.formula)</code>.
 </p>
 </li>
 <li><p> Some <code>.External()</code> calls to the <code>base</code> package
 (really to the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> executable or shared library) have been moved to
 more appropriate packages.  Packages should not have been using
 such calls, but some did (mainly those used by <code>integrate()</code>).
 </p>
 </li></ul>


 <h4>PACKAGE parallel</h4>


 <ul>
 <li><p> There is a new function <code>mcaffinity()</code> which allows
 getting or setting the CPU affinity mask for the current <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> process on
 systems that supports this (currently only Linux has been tested
 successfully).  It has no effect on systems which do not support
 process affinity.  Users are not expected to use this function
 directly (with the exception of fixing libraries that break
 affinity settings like OpenBLAS) &ndash; the function is rather
 intended to support affinity control in high-level parallel
 functions.  In the future, <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> may supplement lack of affinity
 control in the OS by its own bookkeeping <em>via</em> <code>mcaffinity()</code>
 related to processes and threads it spawns.
 </p>
 </li>
 <li> <p><code>mcparallel()</code> has a new argument <code>mc.affinity</code>
 which attempts to set the affinity of the child process according
 to the specification contained therein.
 </p>
 </li>
 <li><p> The port used by socket clusters is chosen randomly: this
 should help to avoid clashes observed when two users of a
 multi-user machine try to create a cluster at the same time.  To
 reproduce the previous behaviour set environment variable
 <span class="env">R_PARALLEL_PORT</span> to <code>10187</code>.
 </p>
 </li></ul>


 <h4>C-LEVEL FACILITIES</h4>


 <ul>
 <li><p> There has been some minor re-organization of the non-API
 header files.  In particular, &lsquo;<span class="file">Rinternals.h</span>&rsquo; no longer
 includes the non-API header &lsquo;<span class="file">R_exts/PrtUtil.h</span>&rsquo;, and that no
 longer includes &lsquo;<span class="file">R_exts/Print.h</span>&rsquo;.
 </p>
 </li>
 <li><p> Passing <code>NULL</code> to <code>.C()</code> is now an error.
 </p>
 </li>
 <li> <p><code>.C()</code> and <code>.Fortran()</code> now warn if
 <code>"single"</code> arguments are used with <code>DUP = FALSE</code>, as
 changes to such arguments are not returned to the caller.
 </p>
 </li>
 <li><p> C entry points <code>R_qsort</code> and <code>R_qsort_I</code> now have
 <code>start</code> and <code>end</code> as <code>size_t</code> to allow them to work
 with longer vectors on 64-bit platforms.  Code using them should
 be recompiled.
 </p>
 </li>
 <li><p> A few recently added C entry points were missing the
 remapping to <code>Rf_</code>, notably <code>[dpq]nbinom_mu</code>.
 </p>
 </li>
 <li><p> Some of the interface pointers formerly available only to
 <code>R.app</code> are now available to front-ends on all
 Unix-alikes: one has been added for the interface to
 <code>View()</code>.
 </p>
 </li>
 <li> <p><code>PACKAGE = ""</code> is now an error in <code>.C()</code> etc calls:
 it was always contrary to the documentation.
 </p>
 </li>
 <li><p> Entry point <code>rcont2</code> has been migrated to package
 <span class="pkg">stats</span> and so is no longer available.
 </p>
 </li>
 <li> <p><code>R_SVN_REVISION</code> in &lsquo;<span class="file">Rversion.h</span>&rsquo; is now an integer
 (rather than a string) and hence usable as e.g.
 <code>#if R_SVN_REVISION &lt; 70000</code>.
 </p>
 </li>
 <li><p> The entry points <code>rgb2hsv</code> and <code>hsv2rgb</code> have been
 migrated to package <span class="pkg">grDevices</span> and so are no longer available.
 </p>
 </li>
 <li> <p><code>R_GE_version</code> has been increased to <code>10</code> and
 <code>name2col</code> removed (use <code>R_GE_str2col</code> instead).  <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 internal colour codes are now defined using the typedef
 <code>rcolor</code>.
 </p>
 </li>
 <li><p> The <code>REPROTECT</code> macro now checks that the protect index
 is valid.
 </p>
 </li>
 <li><p> Several non-API entry points no longer used by <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> have been
 removed, including the Fortran entry points <code>chol</code>,
 <code>chol2inv</code>, <code>cg</code>, <code>ch</code> and <code>rg</code>, and the C
 entry points <code>Brent_fmin</code>, <code>fft_factor</code> and <code>fft_work</code>.
 </p>
 </li>
 <li><p> If a <code>.External</code> call is registered with a number of
 arguments (other than <code>-1</code>), the number of arguments passed
 is checked for each call (as for other foreign function calls).
 </p>
 </li>
 <li><p> It is now possible to write custom connection
 implementations outside core R using &lsquo;<span class="file">R_ext/Connections.h</span>&rsquo;.
 Please note that the implementation of connections is still
 considered internal and may change in the future (see the above
 file for details).
 </p>
 </li></ul>


 <h4>INTERNATIONALIZATION</h4>


 <ul>
 <li><p> The management of translations has been converted to <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span>
 code: see <code>?tools::update_pkg_po</code>.
 </p>
 </li>
 <li><p> The translations for the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> interpreter and
 <code>RGui.exe</code> are now part of the <span class="pkg">base</span> package (rather than
 having sources in directory &lsquo;<span class="file">po</span>&rsquo; and being installed to
 &lsquo;<span class="file">share/locale</span>&rsquo;).  Thus the <span class="pkg">base</span> package supports three
 translation domains, <code>R-base</code>, <code>R</code> and <code>RGui</code>.
 </p>
 </li>
 <li><p> The compiled translations which ship with <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> are all
 installed to the new package <span class="pkg">translations</span> for easier
 updating.  The first package of that name found on
 <code>.libPaths()</code> at the start of the <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> session will be used.
 (It is possible messages will be used before <code>.libPaths()</code> is
 set up in which case the default translations will be used: set
 environment variable <span class="env">R_TRANSLATIONS</span> to point to the location
 of the intended <span class="pkg">translations</span> package to use this right from
 the start.)
 </p>
 </li>
 <li><p> The translations form a separate group in the Windows
 installer, so can be omitted if desired.
 </p>
 </li>
 <li><p> The markup for many messages has been changed to make them
 easier to translate, incorporating suggestions from Łukasz Daniel.
 </p>
 </li></ul>


 <h4>INSTALLATION</h4>


 <ul>
 <li><p> There is again support for building without using the C
 &lsquo;long double&rsquo; type.  This is required by C99, but system
 implementations can be slow or flawed.  Use <code>configure</code>
 option <span class="option">--disable-long-double</span>.
 </p>
 </li>
 <li> <p><code>make pdf</code> and <code>make install-pdf</code> now make
 and install the full reference index (including all base and
 recommended packages).
 </p>
 </li>
 <li><p> The 'reference manual' on the Windows GUI menu and included
 in the installer is now the full reference index, including all
 base and recommended packages.
 </p>
 </li>
 <li> <p><span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> help pages and manuals have no ISBNs because ISBN
 rules no longer allow constantly changing content to be assigned
 an ISBN.
 </p>
 </li>
 <li><p> The Windows installer no longer installs a Start Menu
 link to the static help pages; as most pages are generated
 dynamically, this led to a lot of broken links.
 </p>
 </li>
 <li><p> Any custom settings for Java configuration are recorded in
 file &lsquo;<span class="file">etc/javaconf</span>&rsquo; for subsequent use by <code>R CMD
 	javareconf</code>.
 </p>
 </li>
 <li><p> There is now support for <code>makeinfo</code> version 5.0
 (which requires a slightly different &lsquo;<span class="file">.texi</span>&rsquo; syntax).
 </p>
 </li>
 <li><p> The minimum versions for <span class="option">--use-system-zlib</span> and
 <code>--use-system-pcre</code> are now tested as 1.2.5 and 8.10
 respectively.
 </p>
 </li>
 <li><p> On Windows, the stack size is reduced to 16MB on 32-bit
 systems: misguided users were launching many threads without
 controlling the stack size.
 </p>
 </li>
 <li> <p><code>configure</code> no longer looks for file
 &lsquo;<span class="file">~/.Rconfig</span>&rsquo;: &lsquo;<span class="file">~/.R/config</span>&rsquo; has long been preferred.
 </p>
 </li></ul>


 <h4>BUG FIXES</h4>


 <ul>
 <li><p> When <code>R CMD build</code> is run in an encoding other than
 the one specified in the package's &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file it
 tries harder to expand the <code>authors@R</code> field in the specified
 encoding.  (<a href="https://bugs.R-project.org/show_bug.cgi?id=14958">PR#14958</a>)
 </p>
 </li>
 <li><p> If <code>R CMD INSTALL</code> is required to expand the
 <code>authors@R</code> field of the &lsquo;<span class="file">DESCRIPTION</span>&rsquo; file, it tries
 harder to do so in the encoding specified for the package (rather
 than using ASCII escapes).
 </p>
 </li>
 <li><p> Fix in package <span class="pkg">grid</span> for pushing a viewport into a
 layout cell, where the layout is within a viewport that has zero
 physical width OR where the layout has zero total relative width
 (likewise for height).  The layout column widths (or row heights)
 in this case were being calculated with non-finite values.
 (Reported by Winston Chang.)
 </p>
 </li>
 <li> <p><code>solve(A, b)</code> for a vector <code>b</code> gave the answer
 names from <code>colnames(A)</code> for <code>LINPACK = TRUE</code> but not in
 the default case.
 </p>
 </li>
 <li> <p><code>La.svd()</code> accepts logical matrices (as documented, and
 as <code>svd()</code> did).
 </p>
 </li>
 <li> <p><code>legend()</code> now accepts negative <code>pch</code> values, in
 the same way <code>points()</code> long has.
 </p>
 </li>
 <li><p> Parse errors when installing files now correctly display
 the name of the file containing the bad code.
 </p>
 </li>
 <li><p> In Windows, tcltk windows were not always properly constructed.
 (<a href="https://bugs.R-project.org/show_bug.cgi?id=15150">PR#15150</a>)
 </p>
 </li>
 <li><p> The internal functions implementing <code>parse()</code>,
 <code>tools::parseLatex()</code> and <code>tools::parse_Rd()</code> were not
 reentrant, leading to errors in rare circumstances such as a
 garbage collection triggering a recursive call.
 </p>
 </li>
 <li><p> Field assignments in reference class objects <em>via</em>
 <code>$&lt;-</code> were not being checked
 because the magic incantation to turn methods on for that
 primitive operator had been inadvertently omitted.
 </p>
 </li>
 <li> <p><code>setHook(hookname, value, action="replace")</code> set the
 hook to be the value, rather than a list containing the value as
 documented. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15167">PR#15167</a>)
 </p>
 </li>
 <li><p>  If a package used a &lsquo;<span class="file">NEWS.Rd</span>&rsquo; file, the main HTML
 package index page did not link to it.  (Reported by Dirk
 Eddelbuettel.)
 </p>
 </li>
 <li><p> The primitive implementation of <code>@&lt;-</code> was not
 checking the class of the replacement.  It now does a check,
 quicker but less general than <code>slot&lt;-</code>. See the help.
 </p>
 </li>
 <li> <p><code>split(x, f)</code> now recycles classed objects <code>x</code> in
 the same way as vectors.  (Reported by Martin Morgan.)
 </p>
 </li>
 <li> <p><code>pbeta(.28, 1/2, 2200, lower.tail=FALSE, log.p=TRUE)</code>
 is no longer <code>-Inf</code>; ditto for corresponding <code>pt()</code> and
 <code>pf()</code> calls, such as <code>pt(45, df=5000, lower.tail=FALSE,
 	log.p=TRUE)</code>. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15162">PR#15162</a>)
 </p>
 </li>
 <li><p> The Windows graphics device would crash <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> if a user
 attempted to load the graphics history from a variable that was
 not a saved history. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15230">PR#15230</a>)
 </p>
 </li>
 <li><p> The workspace size for the <code>predict()</code>
 method for <code>loess()</code> could exceed the maximum integer size.
 (Reported by Hiroyuki Kawakatsu.)
 </p>
 </li>
 <li> <p><code>ftable(x, row.vars, col.vars)</code> now also works when the
 <code>*.vars</code> arguments are (integer or character vectors) of
 length zero.
 </p>
 </li>
 <li><p> Calling <code>cat()</code> on a malformed UTF-8 string could cause
 the Windows GUI to lock up. (<a href="https://bugs.R-project.org/show_bug.cgi?id=15227">PR#15227</a>)
 </p>
 </li>
 <li> <p><code>removeClass(cc)</code> gave &quot;node stack overflow&quot; for some
 class definitions containing <code>"array"</code> or <code>"matrix"</code>.
 </p>
 </li></ul>


 <h3>CHANGES in previous versions</h3>


 <ul>
 <li><p> Older news can be found in text format in files
 <a href="../NEWS.0">NEWS.0</a>, <a href="../NEWS.1">NEWS.1</a>
 and <a href="../NEWS.2">NEWS.2</a>
 in the &lsquo;<span class="file">doc</span>&rsquo; directory.  News in HTML format for
 <span style="font-family: Courier New, Courier; color: #666666;"><b>R</b></span> versions from 2.10.0 to 2.15.3 is in
 <a href="NEWS.2.html">NEWS.2.html</a>.
 </p>
 </li></ul>


 </div></body></html>