R 3.6.0 is released

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

R 3.6.0 is released

R devel mailing list
The build system rolled up R-3.6.0.tar.gz (codename "Planting of a Tree") this morning.

The list below details the changes in this release.

You can get the source code from

http://cran.r-project.org/src/base/R-3/R-3.6.0.tar.gz

or wait for it to be mirrored at a CRAN site nearer to you.

Binaries for various platforms will appear in due course.


For the R Core Team,

Peter Dalgaard

These are the checksums (md5 and SHA-256) for the freshly created files, in case you wish
to check that they are uncorrupted:

MD5 (AUTHORS) = b9c44f9f78cab3184ad9898bebc854b4
MD5 (COPYING) = eb723b61539feef013de476e68b5c50a
MD5 (COPYING.LIB) = a6f89e2100d9b6cdffcea4f398e37343
MD5 (FAQ) = 28a3942a7129877e9af1d5ea16202052
MD5 (INSTALL) = 7893f754308ca31f1ccf62055090ad7b
MD5 (NEWS) = 6d227865440cc1ece3d97bdf4a8ee41e
MD5 (NEWS.0) = bfcd7c147251b5474d96848c6f57e5a8
MD5 (NEWS.1) = eb78c4d053ec9c32b815cf0c2ebea801
MD5 (NEWS.2) = 591dcf615162127f904e4e461f330ce9
MD5 (R-latest.tar.gz) = 65601eac6d353f7efb5b48c29097c2fb
MD5 (README) = f468f281c919665e276a1b691decbbe6
MD5 (RESOURCES) = 529223fd3ffef95731d0a87353108435
MD5 (THANKS) = 08158353102084599797db8c9ccf8e2a
MD5 (VERSION-INFO.dcf) = 97d5e3df5e5ac56750695e4b49145fae
MD5 (R-3/R-3.6.0.tar.gz) = 65601eac6d353f7efb5b48c29097c2fb

2cde824a7b18958e5f06b391c801c8288be0f84fa8934b7ddefef23c67e60c09  AUTHORS
e6d6a009505e345fe949e1310334fcb0747f28dae2856759de102ab66b722cb4  COPYING
6095e9ffa777dd22839f7801aa845b31c9ed07f3d6bf8a26dc5d2dec8ccc0ef3  COPYING.LIB
38219d9c6221ccfbf075ef03711b420a1aa8731f890c8f2337148b602a217c2d  FAQ
f87461be6cbaecc4dce44ac58e5bd52364b0491ccdadaf846cb9b452e9550f31  INSTALL
184dfa18e3069782d1092b289a6fe6ef85feb951cd4d1566b225c746d29a5420  NEWS
4e21b62f515b749f80997063fceab626d7258c7d650e81a662ba8e0640f12f62  NEWS.0
12b30c724117b1b2b11484673906a6dcd48a361f69fc420b36194f9218692d01  NEWS.1
ca04f78ffe54afa326fe3ed40e7e1411aca0000ed2fa5ead97ddf51c6aa5b7bc  NEWS.2
36fcac3e452666158e62459c6fc810adc247c7109ed71c5b6c3ad5fc2bf57509  R-latest.tar.gz
2fdd3e90f23f32692d4b3a0c0452f2c219a10882033d1774f8cadf25886c3ddc  README
408737572ecc6e1135fdb2cf7a9dbb1a6cb27967c757f1771b8c39d1fd2f1ab9  RESOURCES
2d2e85e85574c4430951f6b070c08cd5aff1602abfd1bb162bed6d89c436b11f  THANKS
05bd1202b132c8e91a8887d923fcc525cfb3bf81c67847d15876b88bf0c68a71  VERSION-INFO.dcf
36fcac3e452666158e62459c6fc810adc247c7109ed71c5b6c3ad5fc2bf57509  R-3/R-3.6.0.tar.gz

This is the relevant part of the NEWS file

CHANGES IN R 3.6.0:

  SIGNIFICANT USER-VISIBLE CHANGES:

    * Serialization format version 3 becomes the default for
      serialization and saving of the workspace (save(), serialize(),
      saveRDS(), compiler::cmpfile()).  Serialized data in format 3
      cannot be read by versions of R prior to version 3.5.0.
      Serialization format version 2 is still supported and can be
      selected by version = 2 in the save/serialization functions.  The
      default can be changed back for the whole R session by setting
      environment variables R_DEFAULT_SAVE_VERSION and
      R_DEFAULT_SERIALIZE_VERSION to 2.  For maximal
      back-compatibility, files vignette.rds and partial.rdb generated
      by R CMD build 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).

    * The default method for generating from a discrete uniform
      distribution (used in sample(), for instance) has been changed.
      This addresses the fact, pointed out by Ottoboni and Stark, that
      the previous method made sample() noticeably non-uniform on large
      populations.  See PR#17494 for a discussion.  The previous method
      can be requested using RNGkind() or RNGversion() if necessary for
      reproduction of old results.  Thanks to Duncan Murdoch for
      contributing the patch and Gabe Becker for further assistance.

      The output of RNGkind() has been changed to also return the
      'kind' used by sample().

  NEW FEATURES:

    * Sys.setFileTime() has been vectorized so arguments path and time
      of length greater than one are now supported.

    * axis() gets new option gap.axis = NA for specifying a
      multiplication factor for the minimal "gap" (distance) between
      axis labels drawn.  Its default is 1 for labels _parallel_ to the
      axis, and 0.25 for perpendicular ones.

      Perpendicular labels no longer overlap, fixing bug PR#17384.

    * The default method of plot() gains new arguments xgap.axis = NA
      and ygap.axis = NA to be passed to the x- and y- axis(..,
      gap.axis=*) calls.

    * removeSource() now works not only for functions but also for some
      language objects.

    * as.call(), rep.int(), rep_len() and nchar() dispatch internally.

    * is(object, class2) looks for class2 in the calling namespace
      after looking in the namespace of class(object).

    * extendrange(.., f) with a length-2 f now extends separately to
      the left and the right.

    * lengths() dispatches internally to S4 methods.

    * download.file() on Windows now uses URLdecode() to determine the
      file extension, and uses binary transfer (mode = "wb") also for
      file extension .rds.

      The help page for download.file() now contains the same
      information on all platforms.

    * Setting C locale for collation _via_ environment variables LC_ALL
      and LC_COLLATE and _via_ a call to Sys.setlocale() now takes
      precedence over environment variable R_ICU_LOCALE.

    * There is a new function, nullfile(), to give the file name of the
      null system device (e.g., /dev/null) on the current platform.

    * There are two new options, keep.parse.data and
      keep.parse.data.pkgs, which control whether parse data are
      included into sources when keep.source or keep.source.pkgs is
      TRUE.  By default, keep.parse.data.pkgs is now FALSE, which
      changes previous behavior and significantly reduces space and
      time overhead when sources are kept when installing packages.

    * In rapply(x, ..), x can also be "list-like" and of length >=
      2^{31}.

    * trimws() gets new optional whitespace argument, allowing more
      extensive definitions of "space", such as including Unicode
      spaces (as wished in PR#17431).

    * weighted.mean() no longer coerces the weights to a double/numeric
      vector, since sum() now handles integer overflow. This makes
      weighted.mean() more polymorphic and endomorphic, but be aware
      that the results are no longer guaranteed to be a vector of type
      double.

    * When loading namespaces, S3 method registrations which overwrite
      previous registrations are now noted by default (using
      packageStartupMessage()).

    * compiler::cmpfile() gains a version argument, for use when the
      output file should be saved in serialization format 2.

    * The axis labeling in the default method of pairs() may now be
      toggled by new options horOdd and verOdd.

    * (Not Windows nor macOS.) Package tcltk now supports an
      environment variable R_DONT_USE_TK 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.

    * The numeric method of all.equal() gets optional arguments countEQ
      and formatFUN.  If countEQ is true, the mean error is more
      sensible when many entries are *eq*ual.

    * outer(x,y, FUN = "*") is more efficient using tcrossprod(u,v)
      instead of u %*% t(v).

    * vcov(<mlm>) is more efficient via new optional arguments in
      summary.mlm().

    * The default method of summary() gets an option to choose the
      _kind_ of quantile()s to use; wish of PR#17438.

    * Fitting multiple linear models _via_ lm() does work with _matrix_
      offsets, as suggested in PR#17407.

    * The new functions mem.maxVSize() and mem.maxMSize() allow the
      maximal size of the vector heap and the maximal number of nodes
      allowed in the current R process to be queried and set.

    * news() gains support for NEWS.md files.

    * 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 X11() / x11()
      got identical formal arguments to the Unix one.

    * sessionInfo()$running has been factored out in a new variable
      osVersion.

    * slice.index() now also works for multi-dimensional margins.

    * untar() used with an external tar command assumes this supports
      decompression including xz and automagically detecting the
      compression type.  This has been true of all mainstream
      implementations since 2009 (for GNU tar, since version 1.22):
      older implementations are still supported _via_ the new argument
      support_old_tars whose default is controlled by environment
      variable R_SUPPORT_OLD_TARS.  (It looks like NetBSD and OpenBSD
      have 'older' tar commands for this purpose.)

    * The new function asplit() allow splitting an array or matrix by
      its margins.

    * New functions errorCondition() and warningCondition() provide a
      convenient way to create structured error and warning objects.

      .Deprecated() now signals a warning of class "deprecatedWarning",
      and .Defunct() now signals an error of class "defunctError".

    * Many 'package not found' errors are now signaled as errors of
      class "packageNotFoundError".

    * As an experimental feature, when loadNamespace() fails because
      the requested package is not available the error is initially
      signaled with a retry_loadNamespace restart available.  This
      allows a calling handler to try to install the package and
      continue.

    * S3method() directives in NAMESPACE can now also be used to
      perform _delayed_ S3 method registration.

    * Experimentally, setting environment variable
      _R_CHECK_LENGTH_1_LOGIC2_ will lead to warnings (or errors if the
      variable is set to a 'true' value) when && or || encounter and
      use arguments of length more than one.

    * Added "lines" and "chars" coordinate systems to grconvertX() and
      grconvertY().

    * getOption() is more efficient notably for the rare case when
      called with two arguments, from several contributors in PR#17394.

    * In .col(dim) and .row(dim), dim now may also be an integer-valued
      "double".

    * sQuote() and dQuote() get an explicit q argument with obvious
      default instead of using getOption("fancyQuotes") implicitly and
      unconditionally.

    * unzip() can list archives with comments and with spaces in file
      names even using an external unzip command.

    * Command line completion has a new setting rc.settings(dots =
      FALSE) to remove ... from the list of possible function
      arguments.

    * library() no longer checks packages with compiled code match
      R.version$platform.  loadNamespace() never has, and increasingly
      the 'canonical name' does not reflect the important
      characteristics of compiled code.

    * The primitive functions drop() and unclass() now avoid
      duplicating their data for atomic vectors that are large enough,
      by returning ALTREP wrapper objects with adjusted attributes.
      R-level assignments to change attributes will also use wrapper
      objects to avoid duplicating data for larger atomic vectors. R
      functions like structure() and unname() will therefore not
      duplicate data in these settings.  Generic vectors as produced by
      list() are not yet covered by this optimization but may be in due
      course.

    * In formals(), envir becomes an optional argument instead of being
      hardwired.

    * Instead of signalling an error for an invalid S4 object x, str(x)
      now gives a warning and subsequently still shows most parts of x,
      e.g., when slots are missing.

    * gamma(x) and lgamma(x) no longer warn when correctly returning
      Inf or underflowing to zero.  This helps maximum likelihood and
      similar computations.

    * convertColor() is now vectorized, so a lot faster for converting
      many colours at once.  The new argument vectorized to
      colorConverter() ensures that non-vectorized colour converters
      still work.  (Thanks to Brodie Gaslam.)

    * download.file() and url() get new argument headers for custom
      HTTP headers, e.g., allowing to perform basic http
      authentication, thanks to a patch contributed by G'abor Cs'ardi.

    * File-based connection functions file(), gzfile(), bzfile() and
      xzfile() now signal an error when used on a directory.

    * For approx(), splinefun() _etc_, a new setting ties =
      c("ordered", <fun>) allows skipping the sorting and still treat
      ties.

    * format(x) gives a more user friendly error message in the case
      where no method is defined.  A minimal method is provided in
      format.default(x) when isS4(x) is true.

    * which(x) now also works when x is a long vector, thanks to
      Suharto Anggono's PR#17201.  *NB*: this may return a double
      result, breaking the previous guarantee of an integer result.

    * seq.default() is more careful to return an integer (as opposed to
      double) result when its arguments are large and/or classed
      objects; see comment #9 of Suharto Anggono's PR#17497.

    * The plot() method for lm and glm fits, plot.lm(), gains a new
      option iter.smooth with a default of 0 for binomial fits, no
      longer down-weighting when smoothing the residuals.

    * zip() passes its list of files _via_ standard input to the
      external command when too long for the command line (on some
      platforms).

    * data() gains an overwrite argument.

    * t.test() now also returns the standard error (in list component
      stderr).

    * model.matrix(*, contrasts.arg = CC) now warns about invalid
      contrasts.args.

    * Performance of substr() and substring() has been improved.

    * stopifnot() has been simplified thanks to Suharto Anggono's
      proposals to become considerably faster for cheap expressions.

    * The default 'user agent' has been changed when accessing http://
      and https:// sites using libcurl.  (A site was found which caused
      libcurl to infinite-loop with the previous default.)

    * sessionInfo() now also contains RNGkind() and prints it when it
      differs from the default; based on a proposal and patch by Gabe
      Becker in PR#17535.  Also, RNGversion(getRversion()) works
      directly.

    * library() and require() now allow more control over handling
      search path conflicts when packages are attached. The policy is
      controlled by the new conflicts.policy option.

    * barplot() gets a formula method, thanks to a patch proposal by
      Arni Magnusson in PR#17521.

    * pmax() and pmin(x) now also work for long vectors, thanks to
      Suharto Anggono's PR#17533.

    * bxp() now warns when omitting duplicated arguments.

    * New hcl.colors() function to provide wide range of HCL-based
      colour palettes with much better perceptual properties than the
      existing RGB/HSV-based palettes like rainbow().

      Also a new hcl.pals() function to list available palette names
      for hcl.colors().

      Contributed by Achim Zeileis.

    * The default colours for image() and filled.contour() are now
      based on hcl.colors().

    * The palette-generating functions rainbow(), gray.colors(), etc.
      get a new rev argument to facilitate reversing the order of
      colors.

    * New str2lang() and str2expression() as streamlined versions of
      parse(text=., keep.source=FALSE) allow to abstract typical call
      constructions, e.g., in formula manipulations.  (Somewhat
      experimental)

    * Add update_PACKAGES() for incrementally updating a package
      repository index, instead of rebuilding the index from scratch.
      Thanks to Gabe Becker in PR#17544 for the patch, based on part of
      his switchr package.

  INSTALLATION on a UNIX-ALIKE:

    * The options selected for the C++ compiler default to the C++11
      standard if supported, otherwise to the C++98 standard.

    * Visibility macros such as C_VISIBILITY can now be user-set
      (including to empty), e.g. in config.site.

    * Macro FCLIBS, which has sometimes been needed on Solaris, has
      been renamed to FCLIBS_XTRA.

    * Macro F77 is always set to the value of FC, so the latter should
      be set to user-select the Fortran compiler for both fixed-form
      and free-form Fortran.  In particular, gfortran is now the first
      choice for F77, not f95.

      Macros FFLAGS and FCFLAGS 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: .f is fixed-form and
      .f90 and .f95 are free-form).

      If only one of them is set, its value is used for both.

    * The special-casing of CFLAGS, CXXFLAGS and FFLAGS 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 x86_64.

      If necessary, override the defaults on the configure command line
      or in file config.site.

    * Long-untested configure support for HP-UX and very old versions
      of Linux has been removed.

    * configure --with-blas (without specifying a value) includes
      OpenBLAS in its search (before ATLAS and a generic BLAS).  This
      follows recent versions of the ax_blas autoconf macro.

    * The configure macro MAKEINFO has been updated to TEXI2ANY.

    * Support for make install-strip has been enhanced.

  PACKAGE INSTALLATION:

    * Source package installation is by default 'staged': 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 R sessions.

      The overall default is set by environment variable
      R_INSTALL_STAGED.  R CMD INSTALL has new options --staged-install
      and --no-staged-install, and packages can use the StagedInstall
      field in their DESCRIPTION file to opt out.  (That opt-out is a
      temporary measure which may be withdrawn in future.)

      Staged installation requires either --pkglock or --lock, one of
      which is used by default.

    * The interpretation of source code with extension .f is changing.
      Previously this denoted FORTRAN 77 code, but current compilers no
      longer have a FORTRAN 77 mode and interpret it as 'fixed-form'
      Fortran 90 (or later where supported) code.  Extensions .f90 and
      .f95 continue to indicate 'free-form' Fortran code.

      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 'obsolescent' in Fortran 90
      and 'deleted' in Fortran 2018 which gfortran 8.x and later warn
      about.

    * Packages containing files in the src directory with extensions
      .f90 or .f95 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.

      Consequentially, a package which includes free-form Fortran 9x
      code which uses OpenMP should include SHLIB_OPENMP_CFLAGS (or the
      CXXFLAGS version if they also include C++ code) in PKG_LIBS
      rather than SHLIB_OPENMP_FCFLAGS - fortunately on almost all
      current platforms they are the same flag.

    * Macro PKG_FFLAGS will be used for the compilation of both
      fixed-form and free-form Fortran code unless PKG_FCFLAGS is also
      set (in src/Makevars or src/Makevars.win).

    * The make macro F_VISIBILITY is now preferred for both fixed-form
      and free-form Fortran, for use in src/Makevars and similar.

    * R CMD INSTALL gains a new option --strip which (where supported)
      strips installed shared object(s): this can also be achieved by
      setting the environment variable _R_SHLIB_STRIP_ to a true value.

      The new option --strip-lib attempts stripping of static and
      shared libraries installed under lib.

      These are most useful on platforms using GNU binutils (such as
      Linux) and compiling with -g flags.

    * 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 R code (and NAMESPACE files) are worked
      around better.

  UTILITIES:

    * R CMD check now optionally checks makefiles for correct and
      portable use of the SHLIB_OPENMP_*FLAGS macros.

    * R CMD check now evaluates \Sexpr{} expressions (including those
      in macros) before checking the contents of Rd files and so
      detects issues both in evaluating the expressions and in the
      expanded contents.

    * R CMD check now lists missing packages separated by commas and
      with regular quotes such as to be useful as argument in calling
      install.packages(c(..)); from a suggestion by Marcel Ramos.

    * tools::Rd2latex() now uses UTF-8 as its default output encoding.

    * R CMD check now checks line endings of files with extension .hpp
      and those under inst/include.  The check now includes that a
      non-empty file is terminated with a newline.

      R CMD build will correct line endings in such files.

    * R CMD check now tries re-building all vignettes rather than
      stopping at the first error: whilst doing so it adds 'bookmarks'
      to the log.  By default (see the 'R Internals' manual) it
      re-builds each vignette in a separate process.

      It now checks for duplicated vignette titles (also known as
      'index entries'): they are used as hyperlinks on CRAN package
      pages and so do need to be unique.

    * R CMD check has more comprehensive checks on the data directory
      and the functioning of data() in a package.

    * R CMD check now checks autoconf-generated configure files have
      their corresponding source files, including optionally attempting
      to regenerate them on platforms with autoreconf.

    * R CMD build has a new option --compression to select the
      compression used for the tarball.

    * R CMD build now removes src/*.mod files on all platforms.

  C-LEVEL FACILITIES:

    * New pointer protection C functions R_PreserveInMSet and
      R_ReleaseFromMSet have been introduced to replace UNPROTECT_PTR,
      which is not safe to mix with UNPROTECT (and with
      PROTECT_WITH_INDEX). Intended for use in parsers only.

    * NAMEDMAX has been raised to 7 to allow further protection of
      intermediate results from (usually ill-advised) assignments in
      arguments to BUILTIN functions. Properly written package code
      should not be affected.

    * R_unif_index is now considered to be part of the C API.

    * R_GetCurrentEnv() allows C code to retrieve the current
      environment.

  DEPRECATED AND DEFUNCT:

    * Argument compressed of untar() is deprecated - it is only used
      for external tar commands which increasingly for extraction
      auto-detect compression and ignore their zjJ flags.

    * var(f) and hence sd(f) now give an error for factor arguments;
      they gave a deprecation warning since R 3.2.3, PR#16564.

    * Package tools' vignetteDepends() has been deprecated (it called a
      function deprecated since Feb 2016), being partly replaced by
      newly exported vignetteInfo().

    * The f77_f2c script has been removed: it no longer sufficed to
      compile the .f files in R.

    * The deprecated legacy support of make macros such as CXX1X has
      been removed: use the CXX11 forms instead.

    * Make macro F77_VISIBILITY is deprecated in favour of
      F_VISIBILITY.

    * Make macros F77, FCPIFCPLAGS and SHLIB_OPENMP_FCFLAGS are
      deprecated in favour of FC, FPICFLAGS and SHLIB_OPENMP_FFLAGS
      respectively.

    * $.data.frame 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.)

  BUG FIXES:

    * replayPlot(r) now also works in the same R session when r has
      been "reproduced" from serialization, typically after saving to
      and reading from an RDS file.

    * substr() and substring() now signal an error when the input is
      invalid UTF-8.

    * file.copy() now works also when its argument to is of length
      greater than one.

    * mantelhaen.test() no longer suffers from integer overflow in
      largish cases, thanks to Ben Bolker's PR#17383.

    * Calling setGeneric("foo") in a package no longer fails when the
      enclosing environment of the implicit generic foo() is
      .GlobalEnv.

    * untar(file("<some>.tar.gz"), *) now gives a better error message,
      suggesting to use gzfile() instead.

    * Method dispatch uses more relevant environments when looking up
      class definitions.

    * The documentation for identify() incorrectly claimed that the
      indices of identified points were returned in the order that the
      points were selected.  identify() now has a new argument order 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.

    * order(...., decreasing=c(TRUE, FALSE)) could fail in some cases.
      Reported from StackOverflow via Karl Nordstr"om.

    * User macros in Rd files now accept empty and multi-line
      arguments.

    * Changes in print.*(), thanks to Lionel Henry's patches in
      PR#17398:

        * Printing lists, pairlists or attributes containing calls with
          S3 class no longer evaluate those.

        * Printing S4 objects within lists and pairlists dispatches
          with show() rather than print(), as with auto-printing.

        * The indexing tags (names or [[<n>]]) of recursive data
          structures are now printed correctly in complex cases.

        * Arguments supplied to print() are now properly forwarded to
          methods when printing lists, pairlists or attributes
          containing S3 objects.

        * 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.

        * Printing lists, pairlists or attributes containing functions
          now uses srcref attributes if present.

    * Calling install.packages() with a length zero pkgs argument now
      is a no-op (PR#17422).

    * unlist(x) now returns a correct factor when x is a nested list
      with factor leaves, fixing PR#12572 and PR#17419.

    * The documentation help(family) gives more details about the aic
      component, thanks to Ben Bolker's prompting.

    * The documentation for attributes and `attributes<-` now gives x
      as name of the first and main argument which the implementation
      has been requiring, fixing PR#17434.  For consistency, the first
      argument name is also changed from obj to x for
      `mostattributes<-`.

    * strwidth() now uses par("font") as default font face (PR#17352).

    * plot(<table>, log="x") no longer warns about log.

    * The print() method for "htest" objects now formats the test
      statistic and parameter directly and hence no longer rounds to
      units _before_ the decimal point.  Consequently, printing of
      t.test() results with a small number of digits now shows
      non-large df's to the full precision (PR#17444).

    * kruskal.test() and fligner.test() no longer erroneously insist on
      numeric g group arguments (PR#16719).

    * Printing a news db via the browser now does a much better job
      (PR#17433).

    * print.aov() missed column names in the multivariate case due to
      misspelling (reported by Chris Andrews).

    * axis() now creates valid at locations also for small subnormal
      number ranges in log scale plots.

    * format.POSIXlt() now also recycles the zone and gmtoff list
      components to full length when needed, and its internal C code
      detects have_zone in more cases.  In some cases, this changes its
      output to become compatible with format.POSIXct().

    * On Windows, detectCores() in package parallel now detects
      processors in all processor groups, not just the group R is
      running in (impacts particularly systems with more than 64
      logical processors).  Reported by Arunkumar Srinivasan.

    * On Windows, socketSelect() would hang with more than 64 sockets,
      and hence parallel::clusterApplyLB() would hang with more than 64
      workers.  Reported by Arunkumar Srinivasan.

    * as(1L, "double") now does coerce (PR#17457).

    * lm.influence(), influence.measures(), rstudent() etc now work
      (more) correctly for multivariate models ("mlm"), thanks to
      (anonymous) stackoverflow remarks.

    * sample.int(2.9, *, replace=TRUE) again behaves as documented and
      as in R < 3.0.0, namely identically to sample.int(2, ..).

    * Fixes to convertColor() for chromatic adaptation; thanks to
      Brodie Gaslam PR#17473.

    * Using \Sexpr[stage=install]{..} to create an Rd section no longer
      gives a warning in R CMD check; problem originally posted by
      G'abor Cs'ardi, then reported as PR#17479 with a partial patch by
      Duncan Murdoch.

    * Parse data now include a special node for equal assignment.

    * split.default() no longer relies on [[<-(), 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.

    * Line numbers in messages about .Rd files are now more reliable,
      thanks to a patch from Duncan Murdoch.

    * In the numeric method for all.equal(), a numeric scale argument
      is now checked to be positive and allowed to be of length > 1.
      (The latter worked originally and with a warning in recent
      years).

    * Deferred string conversions now record the OutDec option setting
      when not equal to the default.  Reported by Michael Sannella.

    * When y is numeric and f a factor, plot(y ~ f) nicely uses "y" and
      "f" as y- and x-labels.  The more direct boxplot(y ~ f) now does
      too.  The new argument ann = FALSE may be used to suppress these.

    * 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 PR#17483.

    * Calls like formatC(*, zero.print = "< 0.001") no longer give an
      error and are further improved via new optional argument
      replace.zero.  Reported by David Hugh-Jones.

    * methods::formalArgs("<fn>") now finds the same function as
      formals("<fn>"), fixing Emil Bode's PR#17499.

    * The methods package better handles duplicated class names across
      packages.

    * The default method of seq() now avoids integer overflow, thanks
      to the report and "cumsum" patch of Suharto Anggono's PR#17497.

    * sub() no longer loses encodings for non-ASCII replacements
      (PR#17509).

    * Fix for rotated raster image on X11 device.  (Partial fix for
      PR#17148; thanks to Mikko Korpela).

    * formula(model.frame(frml, ..)) now returns frml in all cases,
      thanks to Bill Dunlap.  The previous behavior is available as
      DF2formula(<model.frame>).

    * ar.ols() also returns scalar var.pred in univariate case
      (PR#17517).

    * normalizePath() now treats NA path as non-existent and normalizes
      it to NA.  file.access() treats NA file name as non-existent.
      file.edit() and connection functions such as file() now treat NA
      file names as errors.

    * The internal regularize.values() auxiliary of approx(),
      splinefun() etc now warns again when there are ties and the
      caller did not specify ties.  Further, it no longer duplicates x
      and y unnecessarily when x is already sorted (PR#17515).

    * strtoi("", base) now gives NA on all platforms, following its
      documentation.  Reported by Michael Chirico.

    * 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.)

    * From setClassUnion(), if environment variable
      _R_METHODS_SHOW_CHECKSUBCLASSES is set to true, the internal
      .checkSubclasses() utility prints debugging info to see where it
      is used.

    * max.col(m) with an m of zero columns now returns integer NA
      (instead of 1).

    * axTicks() no longer returns small "almost zero" numbers (in
      exponential format) instead of zero, fixing Ilario Gelmetti's
      PR#17534.

    * isSymmetric(matrix(0, dimnames=list("A","b"))) is FALSE again, as
      always documented.

    * The cairo_pdf graphics device (and other Cairo-based devices) now
      clip correctly to the right and bottom border.

      There was an off-by-one-pixel bug, reported by Lee Kelvin.

    * as.roman(3) <= 2:4 and all other comparisons now work, as do
      group "Summary" function calls such as max(as.roman(sample(20)))
      and as.roman(NA).  (Partly reported by Bill Dunlap in PR#17542.)

    * reformulate("x", response = "sin(y)") no longer produces extra
      back quotes, PR#17359, and gains new optional argument env.

    * When reading console input from stdin with re-encoding (R
      --encoding=enc < input) the code on a Unix-alike now ensures that
      each converted input line is terminated with a newline even if
      re-encoding fails.

    * as.matrix.data.frame() now produces better strings from logicals,
      thanks to PR#17548 from Gabe Becker.

    * The S4 generic signature of rowSums(), rowMeans(), colSums() and
      colMeans() is restricted to "x".

    * match(x, tab) now works for long _character_ vectors x, thanks to
      PR#17552 by Andreas Kersting.

    * Class unions are unloaded when their namespace is unloaded
      (PR#17531, adapted from a patch by Brodie Gaslam).

    * selectMethod() is robust to ANY-truncation of method signatures
      (thanks to Herve Pages for the report).



--
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Office: A 4.23
Email: [hidden email]  Priv: [hidden email]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel