Dear Emacs, please make this -*-Text-*- mode!

This file covers NEWS up to the release of R-2.0.0.

See 'ONEWS' for subsequent changes.

**************************************************

* *

* 2.0 SERIES NEWS *

* *

**************************************************

CHANGES IN R VERSION 2.0.0

USER-VISIBLE CHANGES

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

DESCRIPTION file.

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.

NEW FEATURES

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

preserved.

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

.Deprecated().

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

DESCRIPTION file.

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

getOption("na.action").

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

specifications.

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

function.

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

components.

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

?Comparison.

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

sizes.

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>

upViewport(depth)

- 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

pushViewport(viewport(gp=gpar(alpha=0.5)))

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

pushViewport(viewport(gp=gpar(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.lines(name="line")

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

pushViewport(viewport(gp=gpar(cex=0.5)))

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

pushViewport(viewport(gp=gpar(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"

package.

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.

UTILITIES

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

help(citEntry).

- sessionInfo() can be used to include version information about

R and R packages in text or LaTeX documents.

DOCUMENTATION

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

example.

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'.

C-LEVEL FACILITIES

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

used.

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.

DEPRECATED & DEFUNCT

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.

INSTALLATION CHANGES

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.

PACKAGE INSTALLATION CHANGES

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

'Depends:'.

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

environment.

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).

BUG FIXES

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

installed.

o apply() failed on an array of dimension >=3 when for each

iteration the function returns a named vector of length >=2.

(PR#7205)

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@.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

connection.

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

environment.

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].

(PR#7245)

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

(PR#7255)

**************************************************

* *

* 1.9 SERIES NEWS *

* *

**************************************************

CHANGES IN R VERSION 1.9.1 Patched

INSTALLATION ISSUES

o Installation will now work even in Norwegian and Danish locales

which sort AA at the end (for package stats4 which has AAA.R).

BUG FIXES

o Various memory leaks have been plugged and uses of strcpy()

with overlapping src and dest corrected.

o R CMD INSTALL now also works for /bin/sh's such as the one from

Solaris 8 which fail when a function has the same name as a variable.

o The Date method for trunc() failed.

o window() failed if both start and end were outside the time

range of the original series (possible if extend = TRUE).

o coplot(..) doesn't give an extraneous warning anymore when called

on a fresh device.

o hasArg() used wrong logic to get the parent function. (sys.function()

behaves differently from what is documented.)

o prompt(f) now gives proper \usage{..} for

f <- function(x, g = function(u) { v <- u^2 ; sin(v)/v }) { g(x) }

o package.skeleton() now uses the supplied name in the DESCRIPTION

file.

o options(list('digits', 'scipen')) no longer seg.faults, the

problem being the misuse of a list (PR#7078).

o summary.Date() now has a more sensible default for 'digits'.

o list.files(all.files = TRUE, recursive = TRUE) died on infinite

recursion. (PR#7100)

o cor(as.array(c(a=1,b=2)), cbind(1:2)) no longer seg.faults (PR#7116).

cor(), cov() and var() no longer accidentally work with list()

arguments as if they were unlist()ed.

o as.matrix(data.frame(d=as.POSIXct("2004-07-20"))) doesn't give a

wrong warning anymore.

o gsub(perl=TRUE) code got the R length of return strings wrong

in some circumstances (when the string was shortened).

(Fixed also PR#7108)

o summaryRprof() was ignoring functions whose name begins with dot,

e.g. .C, .Call, .Fortran. (PR#7137)

o loglin() could segfault if 'start' was of the wrong length. (PR#7123)

o model.tables(type="means") could fail in a design where a

projection gave all zeros. (PR#7132)

o Applying attributes() to a pairlist, e.g. .Options, could segfault.

o The checking of R versions incorrectly assumed 1.9.1 >= 1.50.

o str(Surv(..)) failed for type = "counting" Surv objects and for

promises.

o approx(c(1,2),c(NA,NA),1.5,rule=2) does not segfault anymore

(PR#7177), but gives an error.

o nls(model = TRUE) was broken.

o Subsetted assignments of the form A[i1, i2, i3] <- B stopped as

soon as an NA was encountered in an index so subsequent non-NA

indices were ignored. (PR#7210)

o Fixed bug in handling of lwd=NA in contour().

o is.na() was returning undefined results on nested lists.

CHANGES IN R VERSION 1.9.1

NEW FEATURES

o as.Date() now has a method for "POSIXlt" objects.

o mean() has a method for "difftime" objects and so summary()

works for such objects.

o legend() has a new argument 'pt.cex'.

o plot.ts() has more arguments, particularly 'yax.flip'.

o heatmap() has a new 'keep.dendro' argument.

o The default barplot method now handles vectors and 1-d arrays

(e.g., obtained by table()) the same, and uses grey instead of

heat color palettes in these cases. (Also fixes PR#6776.)

o nls() now looks for variables and functions in its formula in

the environment of the formula before the search path, in the

same way lm() etc look for variables in their formulae.

INSTALLATION ISSUES

o src/modules/X11/dataentry.c would not build on some XFree

4.4.0 systems. (This is a bug in their header files but we have

added a workaround.)

o Building with gcc/g77 3.4.0 on ix86 platforms failed to produce

a working build: the critical LAPACK routines are now compiled

with -ffloat-store.

o Added patches to enable 64-bit builds on AIX 5.1: see the R-admin

manual for details.

o Added some patches to allow non-IEEE-754 installations to work

reasonably well. (Infs and NAs are still not handled properly

in complex arithmetic and functions such as sin(). See also

Deprecated, as support for non-IEEE-754 installations is about

to be removed.)

o Installation will now work in Estonian (et_EE*) locales, which

sort z before u. (PR#6958)

DEPRECATED & DEFUNCT

o Support for non-IEEE-754 arithmetic (which has been untested

for some time) will be removed in the next full release.

o Direct use of R INSTALL|REMOVE|BATCH|COMPILE|SHLIB is

deprecated: use R CMD instead.

o The gnome/GNOME graphics device is deprecated and will be

removed in the next full release.

BUG FIXES

o pbinom(q, N, prob) is now more accurate when prob is close to 0.

(PR#6757)

o pcauchy(x, .., log.p) is now more accurate for large x,

particularly when log.p = TRUE. (PR#6756)