path: root/contrib/gnu-sort/NEWS

Changes in release 2.1
[2.0.21]
* split accepts new option -a or --suffix-length.
* split no longer generates longer suffixes than requested; instead, it reports
  an error when suffixes are exhausted.  POSIX requires this behavior.
* The _POSIX2_VERSION environment variable lets you select which version
  of POSIX the utilities should conform to.  Its default value is system
  dependent.  Set _POSIX2_VERSION=199209 to cause the utilities to support
  obsolete usage like "sort +1".
* The following obsolete usages are no longer supported when conforming
  to POSIX 1003.1-2001, which prohibits most digit-string options:
    expand -N	(instead, use expand -t N)
    head -N	(instead, use head -c N or head -n N)
    fold -N	(instead, use fold -w N)
    split -N	(instead, use split -l N)
    tail -N	(instead, use tail -c N or tail -n N)
    unexpand -N	(instead, use unexpand --first-only -t N)
    uniq -N	(instead, use uniq -f N)
  The following obsolete usages (options without arguments) are no
  longer supported when conforming to POSIX 1003.1-2001, which
  prohibits most options with optional arguments:
    od -s	(instead, use od --strings)
    od -w	(instead, use od --width)
    pr -S	(instead, use pr --sep-string)
[2.0.20]
* tr no longer gets failed a assertion for [==] or [::]
* The following obsolete usages are no longer supported when conforming
  to POSIX 1003.1-2001, which prohibits most options with leading "+":
    sort +POS1 -POS2	(instead, use sort -k)
    tail +N		(instead, use tail -c +N or tail -n +N)
    uniq +N		(instead, use uniq -s N)
* Warnings are issued for obsolete usages on older hosts,
  unless POSIXLY_CORRECT is set in the environment.
* sort -m no longer segfaults when given an empty file
* sort -S now accepts 'K' as a synonym for 'k'.
* wc recognizes all locale-defined white-space characters, not just those
    in the "C" locale.
[2.0.19]
* portability tweak to make lib/regex.c compile
* split translatable strings only in the middle of sentences
[2.0.18]
* sort could segfault on systems without a working mkstemp function and
    with a gettimeofday function that clobbers the static buffer that
    localtime uses for it's return value -- introduced in 2.0.17
[2.0.17]
* csplit no longer gets a failed assertion for this:
    printf 'a\n\n'|csplit - '/^$/' 2
* sort detects physical memory attributes more portably
* tail no longer gets a segfault on Linux's /proc/ksyms
* sum -s produces the proper 16-bit checksum for large files
    (this fixes a bug that was introduced in 2.0f)
* uniq is now about 3 times faster than the version from 2.0 on Linux systems;
    the code uses lock-avoiding variants of common I/O functions
[2.0.16]
* tail -F no longer segfaults
[2.0.15]
* `head -c N' and `od -N N' now read no more than N bytes of input
* tail accepts new option: -F, equivalent to `--follow=name --retry',
    for compatibility with the FreeBSD and NetBSD versions of tail.
* fmt no longer segfaults when using a maximum line width larger than 32767
* uniq's --all-repeated option has new modes to delimit groups
    of duplicate lines: --all-repeated={precede,separate,none(default)}
[2.0.14]
* sort now accepts long options like "--reverse" and "--".
* sort now checks option syntax as POSIX requires, except that (as usual
  for GNU) options can follow file names unless POSIXLY_CORRECT is set.
  For example, invalid positional combinations like "sort +1 -r -2" are
  now rejected as per POSIX.
* The next POSIX standard will require that obsolescent 'sort'
  positional options like +1 be treated as file names, not options.
  Please use 'sort -k' instead.
[2.0.13]
* pr accepts new -D or --date option, to specify date format.
* The following changes are required by POSIX:
  - If POSIXLY_CORRECT is set, dates in pr headers now look something like
    'Dec  4 23:59 2001', with the exact appearance affected by LC_TIME.
  - pr -h now affects only the center header string, not the entire header.
  - pr no longer truncates headers.
* Spacing in pr headers has been adjusted slightly.
* `fmt --prefix=S' now works when S contains a byte with the high bit set
[2.0.12]
* sort has improved performance when using very little main memory
* sort has improved memory management
* sort is no longer susceptible to certain denial of service attacks
* sort no longer suffers from a race condition whereby an interrupt received
    during cleanup could cause it to fail to remove temporary files.
    This problem could arise only on hosts without sigaction.
[2.0.11]
* sort accepts new -S SIZE option, to specify main-memory usage.
[2.0.10]
* od is faster and more portable than it was in 2.0.9
* tail avoids an uninitialized memory reference
[2.0.9]
* od now prints valid addresses for offsets of 2^32 and larger, and allows
    the byte offset (-j) and byte count (-N) arguments to be 2^32 and larger.
* tail now works with line and byte counts of 2^32 and larger, on systems
    with large file support
* join now works with an 8-bit delimiter
* fix a compilation failure on some Solaris systems with wc.c
[2.0.8]
* od now supports 8-byte integers, assuming they're printable with e.g., %lld
* new program: sha1sum
* wc accepts new -m option: count (potentially multi-byte) characters
* wc's `--chars' option is now equivalent to -m, not --bytes as it used to be
* `cat -n' works properly when processing 2^31 or more lines
[2.0g]
* sort's --help output now warns that it is locale-aware
* tail: fix a buffer underrun error that occurred on an empty pipe,
  also thanks to bounded pointers
* pr: fix a bounds violation found by Greg McGary's bounded-pointers-enabled gcc
  It could have caused (with low probability) the columns on the last page of
  output *not* to be `balanced' when they should have been.
* sort: if the -T tmpdir option is given multiple times, all the given
  directories are used; this can improve performance for huge sort/merges.
[2.0f]
* all programs fail when printing --help or --version output to a full device
* cut no longer gets a segfault under some circumstances
* unexpand accepts new option: --first-only
[2.0e]
* `tail -f directory' no longer gets a failed assertion
* sort: big performance improvement when sorting many small files;
  from Charles Randall
* configure and portability changes in m4/ and lib/
[2.0d]
* preliminary sort performance improvements
* tsort now works more like the traditional UNIX tsort.  Before it would
  exit when it found a loop.  Now it continues and outputs all items.
* unexpand no longer infloops on certain sequences of white space
* unified lib/: now that directory and most of the configuration framework
  is common between fileutils, textutils, and sh-utils
[2.0c]
* include lib/nanosleep.h.
[2.0b]
* portability tweaks for error.c vs. systems with deficient strerror_r
[2.0a]
* `tail --follow=name' no longer gets a failed assertion for a
  dev,inode-reusing race condition
* sort and comm no longer consider newlines to be part of the line,
  as this requirement will likely be removed from POSIX.2.
  This undoes some changes made for textutils 1.22m and 1.22n.
* tail's (short only) -f option no longer accepts an optional argument,
  so e.g., `tail -fn 2 file' works again.
* tail no longer refuses to operate on certain types of files
* fixed bug in tsort's handling of cycles
Changes in release 2.0
[1.22q]
* HPUX portability fix: md5sum would dump core due to use of libc's getline
[1.22p]
* portability fixes from Paul Eggert based largely on tar-1.13 reports
* `tail --pid=PID' now works even when PID belongs to some other user
[1.22o]
* tail accepts new option: --pid=PID
[1.22n]
* tail accepts the following new options (some of which were added in 1.22g):
  --retry
  --follow[={name|descriptor}]
  --max-unchanged-stats=N
  --max-consecutive-size-changes=N
  --sleep-interval=S
* wc uses the POSIX-mandated output format when POSIXLY_CORRECT is set
* To maintain compatibility with sort, comm and join now obey the LC_COLLATE
  locale, and comm now considers newlines to be part of the lines.
* use lib/memchr.c only if it's not provided by the system -- this means
  that on systems with a fast library memchr function you may notice an
  improvement.  If you use a system with a buggy or signifcantly slower
  memchr, please report it.
[1.22m]
* sort now considers newlines to be part of the line, as required by POSIX.2.
  E.g. a line starting with a tab now sorts before an empty line,
  since tab precedes newline in the ASCII collating sequence.
* sort handles NUL bytes correctly when configured/compiled with --enable-nls
* fix typos in my version of AC_SEARCH_LIBS.
* fix dates on config files so builders don't need autoconf/automake
[1.22l]
* sort no longer autodetects the locale of numbers and months,
   as that conflicts with POSIX.2
* `join -tC' now works when input contains trailing spaces
* portability tweaks for Irix's cc
[1.22k]
* `sort -n' works with negative numbers when configured/compiled
   with --enable-nls
* head accepts byte and line counts of type uintmax_t (so up to 2^64 - 1)
[1.22j]
* tail: fix bug introduced in 1.22i
[1.22i]
* tail now terminates in `yes > k & sleep 1; tail -2c k'
* `tail -f' now ensures that stdout is unbuffered
* fix a bug in cut to allow use of 8-bit delimiters
* pr accepts POSIX compliant options -s and -w,
  the new capital letter options -J, -S and _W turn off the
  unexpected interferences of the small letter options -s and -w
  if used together with the column options.
* pr output has been adapted to other UNIXes in some cases.
[1.22h]
* portability tweaks
* Window/NT/DOS support
[1.22g]
* uniq accepts new option: --all-repeated (-D).
* Windows/DOS portability fixes
* new program: tsort
* tail has several new options
* md5sum can handle file names with embedded backslash characters
* pr accepts long option names (see `pr --help')
* new program: ptx (moved to this package from being its own distribution)
[1.22f]
* cut accepts new --output-delimiter=STR option
* `sort -o no-such-file no-such-file' now fails, as it should
* fix pr bug: pr -td didn't double space
* fix tac bug when using -b, -r, and -s SEPARATOR
* fix sort bug whereby using key-local `d' option would cause following
  key specs to be ignored when any two keys (in the `d'-modified test)
  compared equal.
[1.22e]
* remove maintainer mode
[1.22d]
* wc accepts new option: --max-line-length (-L)
* sort can sort according to your locale if your C library supports that
[1.22c]
[1.22b]
* od supports a new trailing `z' character in a type specification:
    $ od -tx1z .
    0000000 be ef c6 0f fd f9 d7 e0 ec cb f3 c6 00 db e8 00  >................<
    0000020 00 00 d2 00 00 00 00 00 00 00 00 00 00 00 00 00  >................<
    0000040 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  >................<
    *
    0000600 00 00 00 00 00 00 00 00 00 00 00 00 00 00 35 cc  >..............5.<
    0000620 05 63 76 74 2e 6f 00 00 29 ac 08 70 72 6f 6a 65  >.cvt.o..)..proje<
    0000640 63 74 73 00 00 00 18 9a 05 63 76 74 2e 63 00 00  >cts......cvt.c..<
    0000660 18 d9 03 52 43 53 00 00 18 c0 05 78 2e 64 61 74  >...RCS.....x.dat<

[1.22a]
* sort -c reports both the number and the contents of the first out-of-order
  line, in addition to the file name.
* `head -c 4096m' is no longer treated just like `head -c 0'
  now it gets a diagnostic about 4096m being too large.
* pr: For compatibility (also more POSIX compliant): Include default
  separator `TAB' when merging lines of full length.
* When POSIXLY_CORRECT is not set, tail -N now accepts more than one file
  argument, to be consistent with the way head -N works.  If POSIXLY_CORRECT
  is set, using two or more file arguments with the obsolescent form (-N)
  evokes an error.  To avoid the warning or failure, use the POSIX -n N option
  or the GNU --lines=N option.

Changes in release 1.22
[1.21a]
* Fix a bug in tail when invoked with an argument like `+NUMBERc'
* Add test suite for tail

Changes in release 1.21
* Using --program-prefix no longer applies the prefix twice

Changes in release 1.20
* fix pr: -l now uses total number of lines per page also with -f
* fix pr: use left-hand-side truncation of header string to avoid line
  overflow
* fix pr: it now accepts `form feeds set in input files', also with -m
  and multiple form feeds at different pages in each file
* pr now accepts: -h "", print a blank line header
* pr: when skipping pages (+FIRST_PAGE option) line counting (-n option)
  starts with 1st line of input file (not of 1st page printed) by default
* pr accepts new option: -N, start printing with an optional line number
* pr -t retains `form feeds set in input files' (`don't destroy page layout')
* pr accepts new option: -T, equivalent to -t, but eliminate also form feeds
  (`clear file')
* pr accepts the extension: +FIRST_PAGE[:LAST_PAGE]
* pr -w and -s option disentangled (`use a separator' no longer destroys
  column alignment)
* pr accepts new option: -j, merge lines of full length
* pr accepts the extension: -s[STRING], use separator string instead of
  character only
* pr -b is no longer an independent option, balancing is always used
  with -COLUMN (a requirement of unrestricted use of form feeds)
* pr accepts new option: --test, to run the pr tests with a constant
  header string
* join passes all of its tests on Alpha OSF 4.0.
* sort no longer improperly ignores blanks in determining starting and ending
  positions for keys with explicit character offsets
* fix bug in csplit with regexp and negative offset that led to infinite loop
 Changes in test release 1.19q
* fix bug in sort -c that sometimes resulted in a segfault
 Changes in test release 1.19p
* md5sum's --string option is being deprecated and is no longer documented.
  It is still accepted, but will be removed altogether in 1.22.
* tr '[:lower:]' '[:upper:]' no longer fails when LC_CTYPE is set to
  iso_8859_1 on Solaris -- or any other character set with differing
  numbers of uppercase and lowercase characters
* split and tail diagnose unrecognized multiplier suffixes, in e.g.,
  `split --bytes=1M' (should be `-b 1m' or `--bytes=1m')
* fix bug in md5sum's handling of partial reads
* fix bug in treatment by sort -f of bytes with high-bit set
* update configuration system to use automake's aclocal program
* configure performs sanity check on CC and CFLAGS to avoid a misleading
  failure that suggested cross-compiling was the cause
* distribute test suites for cut, join, sort, and tr
* unexpand no longer gets in endless loop
* when verifying checksums, md5sum uses the binary mode flag from the
  input stream rather than the one from the command line

Changes in release 1.19
* md5sum can verify digests of files with names containing newline characters
* update from gettext-0.10.20.

Changes in release 1.18
* when building sort, link with -lm on systems that use the replacement strtod
* update from gettext-0.10.17.

Changes in release 1.17
* include texinfo.tex in the distribution

Changes in release 1.16
* sort is compatible with Unix sort when a key-end spec refers to the N'th
  character in a field that has fewer than N characters
* tail with old-style options like -20k and +31m operates on units of bytes,
  as the --help usage message says.  Before, it used units of lines.

Changes in release 1.15
* od gives better diagnostics for invalid format specs
* uses automake-generated Makefile templates
* configure takes a new option: --enable-maintainer-mode
* fix a bug in fmt when prefix has trailing white space
* internationalized diagnostic messages
* fix a couple bugs in tr involving use of -c and/or -d flags -- see ChangeLog
* diagnose some improper or questionable invocations of csplit
* properly handle `echo |csplit - 1 1', rather than aborting
* fix join: without -t it now ignores leading blanks
* sort accepts new option: -z for NUL terminated records
* join accepts new option: --ignore-case, -i
* uniq accepts new option: --ignore-case, -i

User-visible changes in release 1.14
* sort -i and sort -d properly order strings containing ignored characters
* nl: rename misleading --first-page=N option to --starting-line-number=N.
* sort diagnoses invalid arguments to -k, then fails
* sort -n properly orders invalid integers with respect to valid integers
* sorting works with character offsets larger than corresponding field width
* sort's -b option and `b' modifier work
* sort -k2,2 works.
* csplit detects integer overflow when converting command line arguments
* sort accepts new option/flag, -g, for sorting numbers in scientific notation
* join accepts POSIX `-o 0' field specifier.
* tr 'a[b*512]' '[a*]' < /dev/null terminates
* tr '[:*3][:digit:]' 'a-m' and tr 'a[=*2][=c=]' 'xyyz' no longer fail
* special characters in tr's string1 and string2 may be escaped with backslash

User-visible changes in release 1.13
* md5sum: with --check, distinguish between open/read failure and bad checksum
* md5sum: remove -h, -s, -v short options
* md5sum: rename --verbose to --warn, --quiet to --status
* md5sum --check fails if it finds no properly formatted checksum lines
* sort -c prints `disorder on...' message on standard error, not stdout
* sort -k works as described in the texinfo documentation
* tail works on NetBSD
* md5sum reads and writes (de facto) standard Plumb/Lankester format
* sort accepts -.1 +.2 options for compatibility
* od works properly when dump limit is specified and is a multiple of
  bytes_per_block (set by --width, 16 by default).

User-visible changes in release 1.12
* sort no longer reports spurious errors on Ultrix systems
* new program: md5sum
* all --help messages have been improved
* join's -a1 and -a2 options work
* tr '[:upper:]' '[:lower:]' no longer reads uninitialized memory
* sort properly handles command line arguments like `+7.2n'
* fmt properly formats paragraphs not terminated by a newline
* tail -f flushes stdout before sleeping so that it will output partial
  lines sooner
* sort properly orders fields where one field is a proper prefix of the other
* sort properly interprets field offsets specified via the -k option
* dd, od, and tail work on systems for which off_t is long long (e.g. BSD4.4)
* wc is faster when not counting words
* wc now works even when file pointer isn't at beginning of file
* expand no longer seg faults with very long tab lists

User-visible changes in release 1.11
* fmt is built

User-visible changes in release 1.10
* skeletal texinfo documentation (mainly just the `invoking' nodes)
* new program: fmt
* tail -f on multiple files reports file truncation
* tail -q has been fixed so it never prints headers
* wc -c is much faster when operating on non-regular files
* unexpand gives a diagnostic (rather than a segfault) when given a name of
  a nonexistent file.
* cat, csplit, head, split, sum, tac, tail, tr, and wc no longer fail
  gratuitously when continued after a suspended read or write system call.
* cut interprets -d '' to mean `use the NUL byte as the delimiter' rather
  than reporting that no delimiter was specified and failing.
* `echo a:b:c: | cut -d: -f3,4' prints `c:'.  Before it printed just `c'.
* cut has been rewritten, is markedly faster for large inputs, and passes a
  fairly large test suite.
* sort properly handles the argument to the -T option.

Major changes in release 1.9.1:
* cut no longer ignores the last line of input when that line lacks a
  trailing newline character

Major changes in release 1.9:
* `echo a:b:c: | cut -d: -f3-' prints `c:' and
  `echo a:b | cut -d: -f1' prints `a'.
* the command `printf '\t\n' |fold -w n' now terminates.
  Before, it wouldn't stop for n less than 8.
* sort accepts and ignores -y[string] options for compatibilty with Solaris.
* cat -v /dev/null works on more systems
* od's --compatible (-C) flag renamed to --traditional (no short option)
* --help and --version exit successfully
* --help gives a one-line description of each option and shows the
  correspondence between short and long-named options.
* fix bug in cut.  Now `echo 'a:b:c:' | cut -d: -f3-' works.
  Before it printed `c' instead of `c:'
* csplit allows repeat counts to be specified via `{*}'.
* csplit accepts a new option, --suffix=format that supercedes the
  --digits option.  The --digits option will continue to work.
* csplit accepts a new option, --elide-empty-files.
* configure uses config.h, so DEFS won't exceed preprocessor limits of
  some compilers on the number of symbols defined via -D.
* work around problem where $(srcdir)/config.h was used instead of
  ../config.h -- this happened only when building in a subdirectory
  and when config.h remained in $(srcdir) from a previous ./configure.

Major changes in release 1.8:
* added non-ANSIfied version of memchr.c from GNU libc.

Major changes in release 1.7:
* none
Major changes in release 1.6:
* with the --version option programs print the version and exit immediately
* pr -2a really terminates
* pr -n produces multi-column output

Major changes in release 1.5:
* sort is 8-bit clean
* sort's -n and -M options no longer imply -b
* several bugs in sort have been fixed
* all programs accept --help and --version options
* od --compatible accepts pre-POSIX arguments
* pr -2a terminates

Major changes in release 1.4:
* add od and cksum programs
* move cmp to GNU diff distribution
* tail -f works for multiple files
* pr prints the file name in error messages
* fix some off by 1 errors in pr and fold
* optimize wc -c on regular files
* sort handles `-' argument correctly
* sort supports -T option
* tr ranges like a-a work
* tr x '' fails gracefully
* default sum output format is BSD compatible
* paste -d '' works