R 2.0.0 is released

Peter Dalgaard p.dalgaard at biostat.ku.dk
Mon Oct 4 17:29:39 CEST 2004

I've rolled up R-2.0.0.tar.gz a short while ago. This is a new version
with a number of new features. See below for the details.

As was the case with R 1.0.0, this new version represents a coming of
age more than a radical change to R. We do plan to celebrate the new
major version with press releases and such.

The release will be available from 


or you might wait for it to be mirrored at a CRAN site nearer to you.
Binaries for various platforms will appear in due course.
There is also a version split for floppies.

        For the R Core Team
        Peter Dalgaard

These are the md5sums for the freshly created files, in case you wish
to check that they are uncorrupted:

3900bca37cabb4b76b8d736d51cc9251  R-2.0.0.tar.gz
4a3c7595b112d879997f7455fa8c1c0d  R-2.0.0.tar.gz-split.aa
45d376c7c533c62c657ef0fafac8a784  R-2.0.0.tar.gz-split.ab
7dbdb241e1fb7263701719ea856ebe41  R-2.0.0.tar.gz-split.ac
1e1077f7593778b79e9b0d5c676af63b  R-2.0.0.tar.gz-split.ad
6f4ffaa48a54e002f586f80fd4f2461c  R-2.0.0.tar.gz-split.ae
34b62bb31f6ecf84da329d21c2a21561  R-2.0.0.tar.gz-split.af
c880f8f06ca3fe367bba6757e9cfbf32  R-2.0.0.tar.gz-split.ag
b1a2d17d3ae523d04ffc8d3c6db4b67b  R-2.0.0.tar.gz-split.ah

Here is the relevant part of the NEWS file



    o	The stub packages from 1.9.x have been removed: the library()
	function selects the new home for their code.

    o	`Lazy loading' of R code has been implemented, and is used for
	the standard and recommended packages by default.  Rather than
	keep R objects in memory, they are kept in a database on disc
	and only loaded on first use.  This accelerates startup (down
	to 40% of the time for 1.9.x) and reduces memory usage -- the
	latter is probably unimportant of itself, but reduces
	commensurately the time spent in garbage collection.

	Packages are by default installed using lazy loading if they
	have more than 25Kb of R code and did not use a saved image.
	This can be overridden by INSTALL --[no-]lazy or via a field
	in the DESCRIPTION file.  Note that as with --save, any other
	packages which are required must be already installed.

	As the lazy-loading databases will be consulted often, R
	will be slower if run from a slow network-mounted disc.

    o	All the datasets formerly in packages 'base' and 'stats' have
	been moved to a new package 'datasets'.	 data() does the
	appropriate substitution, with a warning.  However, calls to
	data() are not normally needed as the data objects are visible
	in the 'datasets' package.

	Packages can be installed to make their data objects visible
	via R CMD INSTALL --lazy-data or via a field in the

    o	Package 'graphics' has been split into 'grDevices' (the graphics
	devices shared between base and grid graphics) and 'graphics'
	(base graphics).  Each of the 'graphics' and 'grid' packages
	load 'grDevices' when they are attached.  Note that
	ps.options() has been moved to grDevices and user hooks may
	need to be updated.

    o	The semantics of data() have changed (and were incorrectly
	documented in recent releases) and the function has been moved
	to package 'utils'.  Please read the help page carefully if
	you use the 'package' or 'lib.loc' arguments.

	data() now lists datasets, and not just names which data() accepts.

    o	Dataset 'phones' has been renamed to 'WorldPhones'.

    o	Datasets 'sunspot.month' and 'sunspot.year' are available
	separately but not via data(sunspot) (which was used by package
	lattice to retrieve a dataset 'sunspot').

    o	Packages must have been re-installed for this version, and
	library() will enforce this.

    o	Package names must now be given exactly in library() and
	require(), regardless of whether the underlying file system is
	case-sensitive or not.	So 'library(mass)' will not work, even
	on Windows.

    o	R no longer accepts associative use of relational operators.
	That is, 3 < 2 < 1 (which used to evalute as TRUE!) now causes
	a syntax error.	 If this breaks existing code, just add
	parentheses -- or braces in the case of plotmath.

    o	The R parser now allows multiline strings, without escaping
	the newlines with backslashes (the old method still works).
	Patch by Mark Bravington.


    o	There is a new atomic vector type, class "raw".	 See ?raw for
	full details including the operators and utility functions provided.

    o	The default barplot() method by default uses a
	gamma-corrected grey palette (rather than the heat color
	palette) for coloring its output when given a matrix.

    o	The 'formula' method for boxplot() has a 'na.action' argument,
	defaulting to NULL.  This is mainly useful if the response
	is a matrix when the previous default of 'na.omit' would omit
	entire rows.  (Related to PR#6846.)

	boxplot() and bxp() now obey global 'par' settings and also
	allow the specification of graphical options in more detail,
	compatibly with S-PLUS (fulfilling wishlist entry PR#6832)
	thanks to contributions from Arni Magnusson.  For consistency,
	'boxwex' is not an explicit argument anymore.

    o	chull() has been moved to package graphics (as it uses xy.coords).

    o	There is now a coef() method for summaries of "nls" objects.

    o	compareVersion(), packageDescription() and read.00Index()
	have been moved to package 'utils'.

    o	convolve(), fft(), mvfft() and nextn() have been moved to
	package stats.

    o	coplot() now makes use of cex.lab and font.lab par() settings.

    o	cumsum/prod/max/min() now preserve names.

    o	data(), .path.packages() and .find.packages() now interpret
	package = NULL to mean all loaded packages.

    o	data.frame() and its replacement methods remove the names from
	vector columns.	 Using I() will ensure that names are

    o	data.frame(check.names = TRUE) (the default) enforces unique
	names, as S does.

    o	.Defunct() now has 'new' and 'package' arguments like those of

    o	The plot() method for "dendrogram" objects now respects many more
	nodePar and edgePar settings and for edge labeling computes the
	extents of the diamond more correctly.

    o	deparse(), dput() and dump() have a new 'control' argument to
	control the level of detail when deparsing.  dump() defaults to
	the most detail, the others default to less.  See ?.deparseOpts
	for the details.

	They now evaluate promises by default: see ?dump for details.

    o	dir.create() now expands '~' in filenames.

    o	download.file() has a new progress meter (under Unix) if the
	length of the file is known -- it uses 50 equals signs.

    o	dyn.load() and library.dynam() return an object describing the
	DLL that was loaded.  For packages with namespaces, the DLL
	objects are stored in a list within the namespace.

    o	New function eapply() - apply for environments.	 The supplied
	function is applied to each element of the environment; the order
	of application is not specified.

    o	edit() and fix() use the object name in the window caption on
	some platforms (e.g. Windows).

    o	Function file.edit() function added: like file.show(), but
	allows editing.

    o	Function file.info() can return file sizes > 2G if the
	underlying OS supports such.

    o	fisher.test(*, conf.int=FALSE) allows the confidence interval
	computation to be skipped.

    o	formula() methods for classes "lm" and "glm" used the expanded
	formula (with '.' expanded) from the terms component.

    o	The `formula' method for ftable() now looks for variables in the
	environment of the formula before the usual search path.

    o	A new function getDLLRegisteredRoutines() returns information
	about the routines available from a DLL that were explicitly
	registered with R's dynamic loading facilities.

    o	A new function getLoadedDLLs() returns information about the
	DLLs that are currently loaded within this session.

    o	The package element returned by getNativeSymbolInfo() contains
	reference to both the internal object used to resolve symbols
	with the DLL, and the internal DllInfo structure used to
	represent the DLL within R.

    o	help() now returns information about available documentation for
	a given topic, and notifies about multiple matches.  It has a
	separate print() method.

	If the latex help files were not installed, help() will offer
	to create a latex file on-the-fly from the installed .Rd file.

    o	heatmap() has a new argument 'reorderfun'.

    o	Most versions of install.packages() have an new optional
	argument 'dependencies = TRUE' which will not only fetch the
	packages but also their uninstalled dependencies and their
	dependencies ....

	The Unix version of install.packages() attempts to install
	packages in an order that reflects their dependencies.	(This
	is not needed for binary installs as used under Windows.)

    o	interaction() has new argument 'sep'.

    o	interaction.plot() allows 'type = "b"' and doesn't give spurious
	warnings when passed a matplot()-only argument such as 'main'.

    o	is.integer() and is.numeric() always return FALSE for a
	factor.	 (Previously they were true and false respectively for
	well-formed factors, but it is possible to create factors
	with non-integer codes by underhand means.)

    o	New functions is.leaf(), dendrapply() and a labels() method for
	dendrogram objects.

    o	legend() has an argument 'pt.lwd' and setting 'density' now works
	because 'angle' now defaults to 45 (mostly contributed by Uwe Ligges).

    o	library() now checks the version dependence (if any) of
	required packages mentioned in the Depends: field of the

    o	load() now detects and gives a warning (rather than an error)
	for empty input, and tries to detect (but not correct) files
	which have had LF replaced by CR.

    o	ls.str() and lsf.str() now return an object of class "ls_str" which
	has a print method.

    o	make.names() has a new argument allow_, which if false allows
	its behaviour in R 1.8.1 to be reproduced.

    o	The 'formula' method for mosaicplot() has a 'na.action' argument
	defaulting to 'na.omit'.

    o	model.frame() now warns if it is given data = newdata and it
	creates a model frame with a different number of rows from
	that implied by the size of 'newdata'.

	Time series attributes are never copied to variables in the
	model frame unless na.action = NULL.  (This was always the
	intention, but they sometimes were as the result of an earlier
	bug fix.)

    o	There is a new 'padj' argument to mtext() and axis().
	Code patch provided by Uwe Ligges (fixes PR#1659 and PR#7188).

    o	Function package.dependencies() has been moved to package 'tools'.

    o	The 'formula' method for pairs() has a 'na.action' argument,
	defaulting to 'na.pass', rather than the value of

    o	There are five new par() settings:

	'family' can be used to specify a font family for graphics
	text.  This is a device-independent family specification
	which gets mapped by the graphics device to a device-specific
	font specification (see, for example, postscriptFonts()).
	Currently, only PostScript, PDF, X11, Quartz, and Windows
	respond to this setting.

	'lend', 'ljoin', and 'lmitre' control the cap style and
	join style for drawing lines (only noticeable on thick lines
	or borders).  Currently, only PostScript, PDF, X11, and Quartz
	respond to these settings.

	'lheight' is a multiplier used in determining the vertical
	spacing of multi-line text.

	All of these settings are currently only available via par()
	(i.e., not in-line as arguments to plot(), lines(), ...)

    o	PCRE (as used by grep etc) has been updated to version 5.0.

    o	A 'version' argument has been added to pdf() device.  If this is
	set to "1.4", the device will support transparent colours.

    o	plot.xy(), the workhorse function of points(), lines() and
	plot.default() now has 'lwd' as explicit argument instead of
	implicitly in '...', and now recycles lwd where it makes
	sense, i.e. for line-based plot symbols.

    o	The png() and jpeg() devices (and the bmp() device under Windows)
	now allow a nominal resolution to be recorded in the file.

    o   New functions to control mapping from device-independent
	graphics font family to device-specific family:
	postscriptFont() and postscriptFonts() (for both postscript()
	and pdf()); X11Font() and X11Fonts(); windowsFont() and
	windowsFonts(); quartzFont() and quartzFonts().

    o	power (x^y) has optimised code for y == 2.

    o	prcomp() is now generic, with a formula method (based on an
	idea of Jari Oksanen).

	prcomp() now has a simple predict() method.

    o	printCoefmat() has a new logical argument 'signif.legend'.

    o	quantile() has the option of several methods described in
	Hyndman & Fan (1996). (Contributed by Rob Hyndman.)

    o	rank() has two new 'ties.method's, "min" and "max".

    o	New function read.fortran() reads Fortran-style fixed-format

    o	read.fwf() reads multiline records, is faster for large files.

    o	read.table() now accepts "NULL", "factor", "Date" and
	"POSIXct" as possible values of colClasses, and colClasses can
	be a named character vector.

    o	readChar() can now read strings with embedded nuls.

    o	The "dendrogram" method for reorder() now has a 'agglo.FUN'
	argument for specification of a weights agglomeration

    o	New reorder() method for factors, slightly extending that in
	lattice.  Contributed by Deepayan Sarkar.

    o	Replaying a plot (with replayPlot() or via autoprinting) now
	automagically opens a device if none is open.

    o	replayPlot() issues a warning if an attempt is made to replay
	a plot that was recorded using a different R version (the
	format for recorded plots is not guaranteed to be stable
	across different R versions).  The Windows-menu equivalent
	(History...Get from variable) issues a similar warning.

    o	reshape() can handle multiple 'id' variables.

    o	It is now possible to specify colours with a full alpha
	transparency channel via the new 'alpha' argument to the
	rgb() and hsv() functions, or as a string of the form "#RRGGBBAA".

	NOTE: most devices draw nothing if a colour is not opaque,
	but PDF and Quartz devices will render semitransparent colours.

	A new argument 'alpha' to the function col2rgb()
	provides the ability to return the alpha component of
	colours (as well as the red, green, and blue components).

    o	save() now checks that a binary connection is used.

    o	seek() on connections now accepts and returns a double for the
	file position.	This allows >2Gb files to be handled on a
	64-bit platform (and some 32-bit platforms).

    o	source() with 'echo = TRUE' uses the function source attribute
	when displaying commands as they are parsed.

    o	setClass() and its utilities now warn if either superclasses
	or classes for slots are undefined.  (Use setOldClass to
	register S3 classes for use as slots)

    o	str(obj) now displays more reasonably the STRucture of S4 objects.

	It is also improved for language objects and lists with promise

	The method for class "dendrogram" has a new argument 'stem' and
	indicates when it's not printing all levels (as typically when
	e.g., 'max.level = 2').

	Specifying 'max.level = 0' now allows to suppress all but the top
	level for hierarchical objects such as lists. This is different
	to previous behavior which was the default behavior of giving all
	levels is unchanged.  The default behavior is unchanged but now
	specified by 'max.level = NA'.

    o	system.time() has a new argument 'gcFirst' which, when TRUE,
	forces a garbage collection before timing begins.

    o	tail() of a matrix now displays the original row numbers.

    o	The default method for text() now coerces a factor to character
	and not to its internal codes.	This is incompatible with S
	but seems what users would expect.

	It now also recycles (x,y) to the length of 'labels' if that
	is longer.  This is now compatible with grid.text() and
	S. (See also PR#7084.)

    o	TukeyHSD() now labels comparisons when applied to an
	interaction in an aov() fit.  It detects non-factor terms in
	'which' and drops them if sensible to do so.

    o	There is now a replacement method for window(), to allow a
	range of values of time series to be replaced by specifying the
	start and end times (and optionally a frequency).

    o	If writeLines() is given a connection that is not open, it now
	attempts to open it in mode = "wt" rather than the default
	mode specified when creating the connection.

    o	The screen devices x11(), windows() and quartz() have a new
	argument 'bg' to set the default background colour.

    o	Subassignments involving NAs and with a replacement value of
	length > 1 are now disallowed.	(They were handled
	inconsistently in R < 2.0.0, see PR#7210.)  For data frames
	they are disallowed altogether, even for logical matrix indices
	(the only case which used to work).

    o	The way the comparison operators handle a list argument has
	been rationalized so a few more cases will now work -- see

    o	Indexing a vector by a character vector was slow if both the
	vector and index were long (say 10,000).  Now hashing is used
	and the time should be linear in the longer of the lengths
	(but more memory is used).

    o	Printing a character string with embedded nuls now prints the
	whole string, and non-printable characters are represented by
	octal escape sequences.

    o	Objects created from a formally defined class now include the
	name of the corresponding package as an attribute in the
	object's class.	 This allows packages with namespaces to have
	private (non-exported) classes.

    o	Changes to package 'grid':

	- Calculation of number of circles to draw in circleGrob now
	  looks at length of y and r as well as length of x.

	- Calculation of number of rectangles to draw in rectGrob now
	  looks at length of y, w, and h as well as length of x.

	- All primitives (rectangles, lines, text, ...) now handle
	  non-finite values (NA, Inf, -Inf, NaN) for locations and

	  Non-finite values for locations, sizes, and scales of
	  viewports result in error messages.

	  There is a new vignette ("nonfinite") which describes this
	  new behaviour.

	- Fixed (unreported) bug in drawing circles.  Now checks that
	  radius is non-negative.

	- downViewport() now reports the depth it went down to find a
	  viewport.  Handy for "going back" to where you started, e.g., ...

	    depth <- downViewport("vpname")
	    <draw stuff>

	- The "alpha" gpar() is now combined with the alpha channel of
	  colours when creating a gcontext as follows: (internal C code)

	    finalAlpha = gpar("alpha")*(R_ALPHA(col)/255)

	  This means that gpar(alpha=) settings now affect internal
	  colours so grid alpha transparency settings now are sent to
	  graphics devices.

	  The alpha setting is also cumulative.	 For example, ...

	    grid.rect(width=0.5, height=0.5,
		      gp=gpar(fill="blue"))		 # alpha = 1
	    grid.rect(height=0.25, gp=gpar(fill="red"))	 # alpha = 0.5
	    grid.rect(width=0.25, gp=gpar(fill="red"))	 # alpha = 0.25 !

	- Editing a gp slot in a grob is now incremental.  For example ...

	    grid.edit("line", gp=gpar(col="red")) # line turns red
	    grid.edit("line", gp=gpar(lwd=3)) # line becomes thick
					      # AND STAYS red

	- The "cex" gpar is now cumulative.  For example ...

	    grid.rect(height=unit(4, "char")) # cex = 1
	    grid.rect(height=unit(4, "char")) # cex = 0.5
	    grid.rect(height=unit(4, "char")) # cex = 0.125 !!!

	- New childNames() function to list the names of children
	  of a gTree.

	- The "grep" and "global" arguments have been implemented for
	  grid.[add|edit|get|remove]Grob() functions.

	  The "grep" argument has also been implemented for the
	  grid.set() and setGrob().

	- New function grid.grab() which creates a gTree from the
	  current display list (i.e., the current page of output can
	  be converted into a single gTree object with all grobs
	  on the current page as children of the gTree and all the
	  viewports used in drawing the current page in the childrenvp
	  slot of the gTree).

	- New "lineend", "linejoin", and "linemitre" gpar()s:

	  line end can be "round", "butt", or "square".
	  line join can be "round", "mitre", or "bevel".
	  line mitre can be any number larger than 1
	    (controls when a mitre join gets turned into a bevel join;
	     proportional to angle between lines at join;
	     very big number means that conversion only happens for lines
	     that are almost parallel at join).

	- New grid.prompt() function for controlling whether the user is
	  prompted before starting a new page of output.

	  Grid no longer responds to the par(ask) setting in the "graphics"

    o	The tcltk package has had the tkcmd() function renamed as
	tcl() since it could be used to invoke commands that had
	nothing to do with Tk. The old name is retained, but will be
	deprecated in a future release. Similarly, we now have
	tclopen(), tclclose(), tclread(), tclputs(), tclfile.tail(),
	and tclfile.dir() replacing counterparts starting with "tk",
	with old names retained for now.


    o	R CMD check now checks for file names in a directory that
	differ only by case.

    o	R CMD check now checks Rd files using R code from package tools,
	and gives refined diagnostics about "likely" Rd problems (stray
	top-level text which is silently discarded by Rdconv).

    o	R CMD INSTALL now fails for packages with incomplete/invalid
	DESCRIPTION metadata, using new code from package tools which is
	also used by R CMD check.

    o	list_files_with_exts (package tools) now handles zipped directories.

    o	Package 'tools' now provides Rd_parse(), a simple top-level
	parser/analyzer for R documentation format.

    o	tools::codoc() (and hence R CMD check) now checks any documentation
	for registered S3 methods and unexported objects in packages
	with namespaces.

    o	Package 'utils' contains several new functions:

	- Generics toBibtex() and toLatex() for converting
	  R objects to BibTeX and LaTeX (but almost no methods yet).

	- A much improved citation() function which also has a package
	  argument.  By default the citation is auto-generated from
	  the package DESCRIPTION, the file 'inst/CITATION' can be
	  used to override this, see help(citation) and

	- sessionInfo() can be used to include version information about
	  R and R packages in text or LaTeX documents.


    o	The DVI and PDF manuals are now all made on the paper specified
	by R_PAPERSIZE (default 'a4'), even the .texi manuals which
	were made on US letter paper in previous versions.

    o	The reference manual now omits 'internal' help pages.

    o	There is a new help page shown by help("Memory-limits") which
	documents the current design limitations on large objects.

    o	The format of the LaTeX version of the documentation has
	changed.  The old format is still accepted, but only the new
	resolves cross-references to object names containing _, for

    o	HTML help pages now contain a reference to the package and
	version in the footer, and HTML package index pages give their
	name and version at the top.

    o	All manuals in the 2.x series have new ISBN numbers.

    o	The 'R Data Import/Export' manual has been revised and has a
	new chapter on `Reading Excel spreadsheets'.


    o	The PACKAGE argument for .C/.Call/.Fortran/.External can be
	omitted if the call is within code within a package with a
	namespace.  This ensures that the native routine being called
	is found in the DLL of the correct version of the package if
	multiple versions of a package are loaded in the R session.
	Using a namespace and omitting the PACKAGE argument is
	currently the only way to ensure that the correct version is

    o	The header Rmath.h contains a definition for R_VERSION_STRING
	which can be used to track different versions of R and libRmath.

    o	The Makefile in src/nmath/standalone now has 'install' and
	'uninstall' targets -- see the README file in that directory.

    o	More of the header files, including Rinternals.h, Rdefines.h and
	Rversion.h, are now suitable for calling directly from C++.

    o   Configure looks to a suitable option for inlining C code which
	made available as macro R_INLINE: see `Writing R Extensions'
	for further details.


    o	Direct use of R INSTALL|REMOVE|BATCH|COMPILE|SHLIB has been
	removed: use R CMD instead.

    o	La.eigen(), tetragamma(), pentagamma(), package.contents() and
	package.description() are defunct.

    o	The undocumented function newestVersion() is no longer exported
	from package utils.  (Mainly because it was not completely general.)

    o	C-level entry point ptr_R_GetX11Image has been removed, as it
	was replaced by R_GetX11Image at 1.7.0.

    o	The undocumented C-level entry point R_IsNaNorNA has been
	removed.  It was used in a couple of packages, and should be
	replaced by a call to the documented macro ISNAN.

    o	The gnome/GNOME graphics device is now defunct.


    o	Arithmetic supporting +/-Inf, NaNs and the IEC 60559 (aka
	IEEE 754) standard is now required -- the partial and often
	untested support for more limited arithmetic has been removed.

	The C99 macro isfinite is used in preference to finite if available
	(and its correct functioning is checked at configure time).

	Where isfinite or finite is available and works, it is used as
	the substitution value for R_FINITE.  On some platforms this
	leads to a performance gain.  (This applies to compiled code
	in packages only for isfinite.)

    o	The dynamic libraries libR and libRlapack are now installed in
	R_HOME/lib rather than R_HOME/bin.

    o	When --enable-R-shlib is specified, the R executable is now a
	small executable linked against libR: see the R-admin manual
	for further discussion.	 The 'extra' libraries bzip2, pcre,
	xdr and zlib are now compiled in a way that allows the code to
	be included in a shared library only if this option is
	specified, which might improve performance when it is not.

    o	The main R executable is now R_HOME/exec/R not R_HOME/R.bin, to
	ease issues on MacOS X.	 (The location is needed when debugging
	core dumps, on other platforms.)

    o	Configure now tests for 'inline' and alternatives, and the
	src/extra/bzip2 code now (potentially) uses inlining where
	available and not just under gcc.

    o	The XPG4 sed is used on Solaris for forming dependencies,
	which should now be done correctly.

    o	Makeinfo 4.5 or later is now required for building the HTML and
	Info versions of the manuals.  However, binary distributions
	need to be made with 4.7 or later to ensure some of the
	links are correct.

    o	f2c is not allowed on 64-bit platforms, as it uses longs for
	Fortran integers.

    o	There are new options on how to make the PDF version of the
	reference manual -- see the 'R Administration and Installation
	Manual' section 2.2.

    o	The concatenated Rd files in the installed 'man' directory are
	now compressed and the R CMD check routines can read the
	compressed files.

    o   There is a new configure option --enable-linux-lfs that will
	build R with support for > 2Gb files on suitably recent 32-bit
	Linux systems.


    o	The DESCRIPTION file of packages may contain a 'Imports:'
	field for packages whose namespaces are used but do not need
	to be attached.	 Such packages should no longer be listed in

    o	There are new optional fields 'SaveImage', 'LazyLoad' and
	'LazyData' in the DESCRIPTION file.  Using 'SaveImage' is
	preferred to using an empty file 'install.R'.

    o	A package can contain a file 'R/sysdata.rda' to contain
	system datasets to be lazy-loaded into the namespace/package

    o	The packages listed in 'Depends' are now loaded before a package
	is loaded (or its image is saved or it is prepared for lazy
	loading).  This means that almost all uses of R_PROFILE.R and
	install.R are now unnecessary.

    o	If installation of any package in a bundle fails, R CMD
	INSTALL will back out the installation of all of the bundle,
	not just the failed package (on both Unix and Windows).


    o	Complex superassignments were wrong when a variable with the same
	name existed locally, and were not documented in R-lang.

    o	rbind.data.frame() dropped names/rownames from columns in all
	but the first data frame.

    o	The dimnames<- method for data.frames was not checking the
	validity of the row names.

    o	Various memory leaks reported by valgrind have been plugged.

    o	gzcon() connections would sometimes read the crc bytes from
	the wrong place, possibly uninitialized memory.

    o	Rd.sty contained a length \middle that was not needed after a
	revision in July 2000.	It caused problems with LaTeX systems
	based on e-TeX which are starting to appear.

    o	save() to a connection did not check that the connection was
	open for writing, nor that non-ascii saves cannot be made to a
	text-mode connection.

    o	phyper() uses a new algorithm based on Morten Welinder's bug
	report (PR#6772).  This leads to faster code for large arguments
	and more precise code, e.g. for phyper(59, 150,150, 60,	lower=FALSE).
	This also fixes bug (PR#7064) about fisher.test().

    o	print.default(*, gap = <n>) now in principle accepts all
	non-negative values <n>.

    o	smooth.spline(...)$pen.crit had a typo in its computation;
	note this was printed in print.smooth.spline(*) but not used in
	other "smooth.spline" methods.

    o	write.table() handles zero-row and zero-column inputs correctly.

    o	debug() works on trivial functions instead of crashing. (PR#6804)

    o	eval() could alter a data.frame/list second argument, so
	with(trees, Girth[1] <- NA) altered 'trees' (and any copy of
	'trees' too).

    o	cor() could corrupt memory when the standard deviation was
	zero. (PR#7037)

    o	inverse.gaussian() always printed 1/mu^2 as the link function.

    o	constrOptim() now passes ... arguments through optim to the
	objective function.

    o	object.size() now has a better estimate for character vectors:
	it was in general too low (but only significantly so for
	very short character strings) but over-estimated NA and
	duplicated elements.

    o	quantile() now interpolates correctly between finite and
	infinite values (giving +/-Inf rather than NaN).

    o	library() now gives more informative error messages mentioning
	the package being loaded.

    o	Building the reference manual no longer uses roman upright
	quotes in typewriter output.

    o	model.frame() no longer builds invalid data frames if the
	data contains time series and rows are omitted by na.action.

    o	write.table() did not escape quotes in column names.  (PR#7171)

    o	Range checks missing in recursive assignments using [[ ]].  (PR#7196)

    o	packageStatus() reported partially-installed bundles as

    o	apply() failed on an array of dimension >=3 when for each
	iteration the function returns a named vector of length >=2.

    o	The GNOME interface was in some circumstances failing if run
	from a menu -- it needed to always specify that R be interactive.

    o	depMtrxToStrings (part of pkgDepends) applied nrow() to a
	non-matrix and aborted on the result.

    o	Fix some issues with nonsyntactical names in modelling code
	(PR#7202), relating to backquoting.  There are likely more.

    o	Support for S4 classes that extend basic classes has been fixed
	in several ways.  as() methods and x at .Data should work better.

    o	hist() and pretty() accept (and ignore) infinite values.  (PR#7220)

    o	It is no longer possible to call gzcon() more than once on a

    o	t.test() now detects nearly-constant input data.  (PR#7225)

    o	mle() had problems if ndeps or parscale was supplied in the
	control arguments for optim().  Also, the profiler is now more
	careful to reevaluate modified mle() calls in its parent

    o	Fix to rendering of accented superscripts and subscripts e.g.,
	expression((b[dot(a)])).  (Patch from Uwe Ligges.)

    o	attach(*, pos=1) now gives a warning (and will give an error).

    o	power.*test() now gives an error when 'sig.level' is outside [0,1].

    o	Fitting a binomial glm with a matrix response lost the names of
	the response, which should have been transferred to the
	residuals and fitted values.

    o   print.ts() could get the year wrong because rounding issue

   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907

More information about the R-announce mailing list