NEWS

GNU grep NEWS                                    -*- outline -*-

* Noteworthy changes in release ?.? (????-??-??) [?]

** Bug fixes

  Preprocessing N patterns would take at least O(N^2) time when too many
  patterns hashed to too few buckets. This now takes seconds, not days:
  : | grep -Ff <(seq 6400000 | tr 0-9 A-J)
  [Bug#44754 introduced in grep 3.5]


* Noteworthy changes in release 3.6 (2020-11-08) [stable]

** Changes in behavior

  The GREP_OPTIONS environment variable no longer affects grep's behavior.
  The variable was declared obsolescent in grep 2.21 (2014), and since
  then any use had caused grep to issue a diagnostic.

** Bug fixes

  grep's DFA matcher performed an invalid regex transformation
  that would convert an ERE like a+a+a+ to a+a+, which would make
  grep a+a+a+ mistakenly match "aa".
  [Bug#44351 introduced in grep 3.2]

  grep -P now reports the troublesome input filename upon PCRE execution
  failure.  Before, searching many files for something rare might fail with
  just "exceeded PCRE's backtracking limit".  Now, it also reports which file
  triggered the failure.


* Noteworthy changes in release 3.5 (2020-09-27) [stable]

** Changes in behavior

  The message that a binary file matches is now sent to standard error
  and the message has been reworded from "Binary file FOO matches" to
  "grep: FOO: binary file matches", to avoid confusion with ordinary
  output or when file names contain spaces and the like, and to be
  more consistent with other diagnostics.  For example, commands
  like 'grep PATTERN FILE | wc' no longer add 1 to the count of
  matching text lines due to the presence of the message.  Like other
  stderr messages, the message is now omitted if the --no-messages
  (-s) option is given.

  Two other stderr messages now use the typical form too.  They are
  now "grep: FOO: warning: recursive directory loop" and "grep: FOO:
  input file is also the output".

  The --files-without-match (-L) option has reverted to its behavior
  in grep 3.1 and earlier.  That is, grep -L again succeeds when a
  line is selected, not when a file is listed.  The behavior in grep
  3.2 through 3.4 was causing compatibility problems.

** Bug fixes

  grep -I no longer issues a spurious "Binary file FOO matches" line.
  [Bug#33552 introduced in grep 2.23]

  In UTF-8 locales, grep -w no longer ignores a multibyte word
  constituent just before what would otherwise be a word match.
  [Bug#43225 introduced in grep 2.28]

  grep -i no longer mishandles ASCII characters that match multibyte
  characters.  For example, 'LC_ALL=tr_TR.utf8 grep -i i' no longer
  dumps core merely because 'i' matches 'İ' (U+0130 LATIN CAPITAL
  LETTER I WITH DOT ABOVE) in Turkish when ignoring case.
  [Bug#43577 introduced partly in grep 2.28 and partly in grep 3.4]

  A performance regression with -E and many patterns has been mostly fixed.
  "Mostly" as there is a performance tradeoff between Bug#22357 and Bug#40634.
  [Bug#40634 introduced in grep 2.28]

  A performance regression with many duplicate patterns has been fixed.
  [Bug#43040 introduced in grep 3.4]

  An N^2 RSS performance regression with many patterns has been fixed
  in common cases (no backref, and no use of -o or --color).
  With only 80,000 lines of /usr/share/dict/linux.words, the following
  would use 100GB of RSS and take 3 minutes. With the fix, it used less
  than 400MB and took less than one second:
    head -80000 /usr/share/dict/linux.words > w; grep -vf w w
  [Bug#43527 introduced in grep 3.4]

** Build-related

  "make dist" builds .tar.gz files again, as they are still used in
  some barebones builds.


* Noteworthy changes in release 3.4 (2020-01-02) [stable]

** New features

  The new --no-ignore-case option causes grep to observe case
  distinctions, overriding any previous -i (--ignore-case) option.

** Bug fixes

  '.' no longer matches some invalid byte sequences in UTF-8 locales.
  [bug introduced in grep 2.7]

  grep -Fw can no longer false match in non-UTF-8 multibyte locales
  For example, this command would erroneously print its input line:
    echo ab | LC_CTYPE=ja_JP.eucjp grep -Fw b
  [Bug#38223 introduced in grep 2.28]

  The exit status of 'grep -L' is no longer incorrect when standard
  output is /dev/null.
  [Bug#37716 introduced in grep 3.2]

  A performance bug has been fixed when grep is given many patterns,
  each with no back-reference.
  [Bug#33249 introduced in grep 2.5]

  A performance bug has been fixed for patterns like '01.2' that
  cause grep to reorder tokens internally.
  [Bug#34951 introduced in grep 3.2]

** Build-related

  The build procedure no longer relies on any already-built src/grep
  that might be absent or broken.  Instead, it uses the system 'grep'
  to bootstrap, and uses src/grep only to test the build.  On Solaris
  /usr/bin/grep is broken, but you can install GNU or XPG4 'grep' from
  the standard Solaris distribution before building GNU Grep yourself.
  [bug introduced in grep 2.8]


* Noteworthy changes in release 3.3 (2018-12-20) [stable]

** Bug fixes

  Some uses of \b in the C locale and with the DFA matcher would fail, e.g.,
  the following would print nothing (it should print the input line):
    echo 123-x|LC_ALL=C grep '.\bx'
  Using a multibyte locale, using certain regexp constructs (some ranges,
  back-references), or forcing use of the PCRE matcher via --perl-regexp (-P)
  would avoid the bug.
  [bug introduced in grep 3.2]


* Noteworthy changes in release 3.2 (2018-12-20) [stable]

** Changes in behavior

  The --files-without-match (-L) option now causes grep to succeed
  when a file is listed, instead of when a line is selected.  This
  resembles what git-grep does.

** Bug fixes

  The --recursive (-r) option no longer fails on MS-Windows.
  [bug introduced in grep 2.11]

** Improvements

  An over-30x performance improvement when many 'or'd expressions
  share a common prefix, thanks to improvements in gnulib's dfa.c,
  by Norihiro Tanaka.  See gnulib commits v0.1-2110-ge648401be,
  v0.1-2111-g4299106ce, v0.1-2117-g617a60974

  An additional 3-23% speed-up when searching large files, via
  increased initial buffer size.

  grep now diagnoses stack overflow.  Before grep-2.6, the included
  regexp code would detect it.  Since 2.6, grep defaulted to using
  glibc's regexp, which lost that capability.


* Noteworthy changes in release 3.1 (2017-07-02) [stable]

** Improvements

  grep '[0-9]' is now just as fast as grep '[[:digit:]]' when run
  in a multi-byte locale.  Before, it was several times slower.

** Changes in behavior

  Context no longer excludes selected lines omitted because of -m.
  For example, 'grep "^" -m1 -A1' now outputs the first two input
  lines, not just the first line.  This fixes a glitch that has been
  present since -m was added in grep 2.5.

  The following changes affect only MS-Windows platforms.  First, the
  --binary (-U) option now governs whether binary I/O is used, instead
  of a heuristic that was sometimes incorrect.  Second, the
  --unix-byte-offsets (-u) option now has no effect on MS-Windows too.


* Noteworthy changes in release 3.0 (2017-02-09) [stable]

** Bug fixes

  grep without -F no longer goes awry when given two or more patterns
  that contain no special characters other than '\' and also contain a
  subpattern like '\.' that escapes a character to make it ordinary.
  [bug introduced in grep 2.28]

  grep no longer fails to build on PCRE versions before 8.20.
  [bug introduced in grep 2.28]


* Noteworthy changes in release 2.28 (2017-02-06) [stable]

** Bug fixes

  When grep -Fo finds matches of differing length, it could
  mistakenly print a shorter one.  Now it prints a longest one.
  [bug introduced in grep-2.26]

  When standard output is /dev/null, grep no longer fails when
  standard input is a file in the Linux /proc file system, or when
  standard input is a pipe and standard output is in append mode.
  [bugs introduced in grep-2.27]

  Fix performance regression with multiple patterns, e.g., for -Fi in
  a multi-byte locale, or for -Fw in a single-byte locale.
  [bugs introduced in grep-2.19, grep-2.22 and grep-2.26]

** Improvements

  Improve performance for -E or -G pattern lists that are easily
  converted to -F format.


* Noteworthy changes in release 2.27 (2016-12-06) [stable]

** Bug fixes

  grep no longer reports a false match in a multibyte, non-UTF8 locale
  like zh_CN.gb18030, with a regular expression like ".*7" that just
  happens to match the 4-byte representation of gb18030's \uC9, the
  final byte of which is the digit "7".
  [bug introduced in grep-2.19]

  Unless an early-exit option like -q, -l, -L, -m, or -f /dev/null is
  specified, grep now reads all of a non-seekable standard input,
  even if this cannot affect grep's output or exit status.  This works
  better with nonportable scripts that run "PROGRAM | grep PATTERN
  >/dev/null" where PROGRAM dies when writing into a broken pipe.
  [bug introduced in grep-2.26]

  grep no longer mishandles ranges in nontrivial unibyte locales.
  [bug introduced in grep-2.26]

  grep -P no longer attempts multiline matches.  This works more
  intuitively with unusual patterns, and means that grep -Pz no longer
  rejects patterns containing ^ and $ and works when combined with -x.
  [bugs introduced in grep-2.23] A downside is that grep -P is now
  significantly slower, albeit typically still faster than pcregrep.

  grep -m0 -L PAT FILE now outputs "FILE".  [bug introduced in grep-2.5]

  To output ':' and tab-align the following character C, grep -T no
  longer outputs tab-backspace-':'-C, an approach that has problems if
  run inside an Emacs shell window.  [bug introduced in grep-2.5.2]

  grep -T now uses worst-case widths of line numbers and byte offsets
  instead of guessing widths that might not work with larger files.
  [bug introduced in grep-2.5.2]

  grep's use of getprogname no longer causes a build failure on HP-UX.

** Improvements

  grep no longer reads the input in a few more cases when it is easy
  to see that matching cannot succeed, e.g., 'grep -f /dev/null'.


* Noteworthy changes in release 2.26 (2016-10-02) [stable]

** Bug fixes

  Grep no longer omits output merely because it follows an output line
  suppressed due to encoding errors.  [bug introduced in grep-2.21]

  In the Shift_JIS locale, grep no longer mistakenly matches in the
  middle of a multibyte character. [bug present since "the beginning"]

** Improvements

  grep can be much faster now when standard output is /dev/null.

  grep -F is now typically much faster when many patterns are given,
  as it now uses the Aho-Corasick algorithm instead of the
  Commentz-Walter algorithm in that case.

  grep -iF is typically much faster in a multibyte locale, if the
  pattern and its case counterparts contain only single byte characters.

  grep with complicated expressions (e.g., back-references) and without
  -i now uses the regex fastmap for better performance.

  In multibyte locales, grep now handles leading "." in patterns more
  efficiently.

  grep now prints a "FILENAME:LINENO: " prefix when diagnosing an
  invalid regular expression that was read from an '-f'-specified file.


* Noteworthy changes in release 2.25 (2016-04-21) [stable]

** Bug fixes

  In the C or POSIX locale, grep now treats all bytes as valid
  characters even if the C runtime library says otherwise.  The
  revised behavior is more compatible with the original intent of
  POSIX, and the next release of POSIX will likely make this official.
  [bug introduced in grep-2.23]

  grep -Pz no longer mistakenly diagnoses patterns like [^a] that use
  negated character classes. [bug introduced in grep-2.24]

  grep -oz now uses null bytes, not newlines, to terminate output lines.
  [bug introduced in grep-2.5]

** Improvements

  grep now outputs details more consistently when reporting a write error.
  E.g., "grep: write error: No space left on device" rather than just
  "grep: write error".


* Noteworthy changes in release 2.24 (2016-03-10) [stable]

** Bug fixes

  grep -z would match strings it should not.  To trigger the bug, you'd
  have to use a regular expression including an anchor (^ or $) and a
  feature like a range or a back-reference, causing grep to forego its DFA
  matcher and resort to using re_search.  With a multibyte locale, that
  matcher could mistakenly match a string containing a newline.
  For example, this command:
    printf 'a\nb\0' | LC_ALL=en_US.utf-8 grep -z '^[a-b]*b'
  would mistakenly match and print all four input bytes.  After the fix,
  there is no match, as expected.
  [bug introduced in grep-2.7]

  grep -Pz now diagnoses attempts to use patterns containing ^ and $,
  instead of mishandling these patterns.  This problem seems to be
  inherent to the PCRE API; removing this limitation is on PCRE's
  maint/README wish list.  Patterns can continue to match literal ^
  and $ by escaping them with \ (now needed even inside [...]).
  [bug introduced in grep-2.5]


* Noteworthy changes in release 2.23 (2016-02-04) [stable]

** Bug fixes

  Binary files are now less likely to generate diagnostics and more
  likely to yield text matches.  grep now reports "Binary file FOO
  matches" and suppresses further output instead of outputting a line
  containing an encoding error; hence grep can now report matching text
  before a later binary match.  Formerly, grep reported FOO to be
  binary when it found an encoding error in FOO before generating
  output for FOO, which meant it never reported both matching text and
  matching binary data; this was less useful for searching text
  containing encoding errors in non-matching lines.
  [bug introduced in grep-2.21]

  grep -c no longer stops counting when finding binary data.
  [bug introduced in grep-2.21]

  grep no longer outputs encoding errors in unibyte locales.
  For example, if the byte '\x81' is not a valid character in a
  unibyte locale, grep treats the byte as binary data.
  [bug introduced in grep-2.21]

  grep -oP is no longer susceptible to an infinite loop when processing
  invalid UTF8 just before a match.
  [bug introduced in grep-2.22]

  --exclude and related options are now matched against trailing
  parts of command-line arguments, not against the entire arguments.
  This partly reverts the --exclude-related change in 2.22.
  [bug introduced in grep-2.22]

  --line-buffer is no longer ineffective when combined with -l.
  [bug introduced in grep-2.5]

  -xw is now equivalent to -x more consistently, with -P and with backrefs.
  [bug only partially fixed in grep-2.19]


* Noteworthy changes in release 2.22 (2015-11-01) [stable]

** Improvements

  Performance has improved for patterns containing very long strings,
  reducing preprocessing time for an N-byte regexp from O(N^2) to
  only slightly superlinear for most patterns.  Before, a command like
  the following would take over a minute, but now, it takes less than
  a second:
  : | grep -f <(seq -s '' 99999)

  When building grep, 'configure' now uses PCRE's pkg-config module for
  configuration information, rather than attempting to guess it by hand.

** Bug fixes

  A DFA matcher bug made this command mistakenly print its input line:
    echo axb | grep -E '^x|x$'
  Likewise for this equivalent command:
    echo axb | grep -e '^x' -e 'x$'
  [bug introduced in grep-2.19 ]

  grep no longer reads from uninitialized memory or from beyond the end
  of the heap-allocated input buffer.  This fix addressed CVE-2015-1345.
  [bug introduced in grep-2.19 ]

  With -z, '.' and '[^x]' in a pattern now consistently match newline.
  Previously, they sometimes matched newline, and sometimes did not.
  [bug introduced in grep-2.4]

  When the JIT stack is exhausted, grep -P now grows the stack rather
  than reporting an internal PCRE error.

  'grep -D skip PATTERN FILE' no longer hangs if FILE is a fifo.
  [bug introduced in grep-2.12]

  --exclude and related options are now matched against entire
  command-line arguments, not against command-line components.
  [bug introduced in grep-2.6]

  Fix performance degradation of grep -Fw in unibyte locales.
  [bug introduced in grep-2.19 ]


* Noteworthy changes in release 2.21 (2014-11-23) [stable]

** Improvements

  Performance has been greatly improved for searching files containing
  holes, on platforms where lseek's SEEK_DATA flag works efficiently.

  Performance has improved for rejecting data that cannot match even
  the first part of a nontrivial pattern.

  Performance has improved for very long strings in patterns.

  If a file contains data improperly encoded for the current locale,
  and this is discovered before any of the file's contents are output,
  grep now treats the file as binary.

  grep -P no longer reports an error and exits when given invalid UTF-8 data.
  Instead, it considers the data to be non-matching.

** Bug fixes

  grep no longer mishandles patterns that contain \w or \W in multibyte
  locales.

  grep would fail to count newlines internally when operating in non-UTF8
  multibyte locales, leading it to print potentially many lines that did
  not match.  E.g., the command, "seq 10 | env LC_ALL=zh_CN src/grep -n .."
  would print this:
  1:1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  implying that the match, "10" was on line 1.
  [bug introduced in grep-2.19]

  grep -F -x -o no longer prints an extra newline for each match.
  [bug introduced in grep-2.19]

  grep in a non-UTF8 multibyte locale could mistakenly match in the middle
  of a multibyte character when using a '^'-anchored alternate in a pattern,
  leading it to print non-matching lines.  [bug present since "the beginning"]

  grep -F Y no longer fails to match in non-UTF8 multibyte locales like
  Shift-JIS, when the input contains a 2-byte character, XY, followed by
  the single-byte search pattern, Y.  grep would find the first, middle-
  of-multibyte matching "Y", and then mistakenly advance an internal
  pointer one byte too far, skipping over the target "Y" just after that.
  [bug introduced in grep-2.19]

  grep -E rejected unmatched ')', instead of treating it like '\)'.
  [bug present since "the beginning"]

  On NetBSD, grep -r no longer reports "Inappropriate file type or format"
  when refusing to follow a symbolic link.
  [bug introduced in grep-2.12]

** Changes in behavior

  The GREP_OPTIONS environment variable is now obsolescent, and grep
  now warns if it is used.  Please use an alias or script instead.

  In locales with multibyte character encodings other than UTF-8,
  grep -P now reports an error and exits instead of misbehaving.

  When searching binary data, grep now may treat non-text bytes as
  line terminators.  This can boost performance significantly.

  grep -z no longer automatically treats the byte '\200' as binary data.

* Noteworthy changes in release 2.20 (2014-06-03) [stable]

** Bug fixes

  grep --max-count=N FILE would no longer stop reading after the Nth match.
  I.e., while grep would still print the correct output, it would continue
  reading until end of input, and hence, potentially forever.
  [bug introduced in grep-2.19]

  A command like echo aa|grep -E 'a(b$|c$)' would mistakenly
  report the input as a matched line.
  [bug introduced in grep-2.19]

** Changes in behavior

  grep --exclude-dir='FOO/' now excludes the directory FOO.
  Previously, the trailing slash meant the option was ineffective.


* Noteworthy changes in release 2.19 (2014-05-22) [stable]

** Improvements

  Performance has improved, typically by 10% and in some cases by a
  factor of 200.  However, performance of grep -P in UTF-8 locales has
  gotten worse as part of the fix for the crashes mentioned below.

** Bug fixes

  grep no longer mishandles patterns like [a-[.z.]], and no longer
  mishandles patterns like [^a] in locales that have multicharacter
  collating sequences so that [^a] can match a string of two characters.

  grep no longer mishandles an empty pattern at the end of a pattern list.
  [bug introduced in grep-2.5]

  grep -C NUM now outputs separators consistently even when NUM is zero,
  and similarly for grep -A NUM and grep -B NUM.
  [bug present since "the beginning"]

  grep -f no longer mishandles patterns containing NUL bytes.
  [bug introduced in grep-2.11]

  Plain grep, grep -E, and grep -F now treat encoding errors in patterns
  the same way the GNU regular expression matcher treats them, with respect
  to whether the errors can match parts of multibyte characters in data.
  [bug present since "the beginning"]

  grep -w no longer mishandles a potential match adjacent to a letter that
  takes up two or more bytes in a multibyte encoding.
  Similarly, the patterns '\<', '\>', '\b', and '\B' no longer
  mishandle word-boundary matches in multibyte locales.
  [bug present since "the beginning"]

  grep -P now reports an error and exits when given invalid UTF-8 data.
  Previously it was unreliable, and sometimes crashed or looped.
  [bug introduced in grep-2.16]

  grep -P now works with -w and -x and back-references. Before,
  echo aa|grep -Pw '(.)\1' would fail to match, yet
  echo aa|grep -Pw '(.)\2' would match.

  grep -Pw now works like grep -w in that the matched string has to be
  preceded and followed by non-word components or the beginning and end
  of the line (as opposed to word boundaries before).  Before, this
  echo a@@a| grep -Pw @@ would match, yet this
  echo a@@a| grep -w @@ would not.  Now, they both fail to match,
  per the documentation on how grep's -w works.

  grep -i no longer mishandles patterns containing titlecase characters.
  For example, in a locale containing the titlecase character
  'Lj' (U+01C8 LATIN CAPITAL LETTER L WITH SMALL LETTER J),
  'grep -i Lj' now matches both 'LJ' (U+01C7 LATIN CAPITAL LETTER LJ)
  and 'lj' (U+01C9 LATIN SMALL LETTER LJ).


* Noteworthy changes in release 2.18 (2014-02-20) [stable]

** Bug fixes

  grep no longer mishandles patterns like [^^-~] in unibyte locales.
  [bug introduced in grep-2.8]

  grep -i in a multibyte, non-UTF8 locale could be up to 200 times slower
  than in 2.16.  [bug introduced in grep-2.17]


* Noteworthy changes in release 2.17 (2014-02-17) [stable]

** Improvements

  grep -i in a multibyte locale is now typically 10 times faster
  for patterns that do not contain \ or [.

  grep (without -i) in a multibyte locale is now up to 7 times faster
  when processing many matched lines.

** Maintenance

  grep's --mmap option was disabled in March of 2010, and began to
  elicit a warning in January of 2012.  Now it is completely gone.


* Noteworthy changes in release 2.16 (2014-01-01) [stable]

** Bug fixes

  Fix gnulib-provided maint.mk so that the release procedure described
  in README-release actually does what we want.  Before that fix, that
  procedure resulted in a grep-2.15 tarball that would lead to a grep
  binary whose --version-reported version number was 2.14.51...

  The fix to make \s and \S work with multi-byte white space broke
  the use of each shortcut whenever followed by a repetition operator.
  For example, \s*, \s+, \s? and \s{3} would all malfunction in a
  multi-byte locale.  [bug introduced in grep-2.15]

  The fix to make grep -P work better with UTF-8 made it possible for
  grep to evoke a larger set of PCRE errors, some of which could trigger
  an abort.  E.g., this would abort:
    printf '\x82'|LC_ALL=en_US.UTF-8 grep -P y
  Now grep handles arbitrary PCRE errors.  [bug introduced in grep-2.15]

  Handle very long lines (2GiB and longer) on systems with a deficient
  read system call.

* Noteworthy changes in release 2.15 (2013-10-26) [stable]

** Bug fixes

  grep's \s and \S failed to work with multi-byte white space characters.
  For example, \s would fail to match a non-breaking space, and this
  would print nothing: printf '\xc2\xa0' | LC_ALL=en_US.UTF-8 grep '\s'
  A related bug is that \S would mistakenly match an invalid multibyte
  character.  For example, the following would match:
    printf '\x82\n' | LC_ALL=en_US.UTF-8 grep '^\S$'
  [bug present since grep-2.6]

  grep -i would segfault on systems using UTF-16-based wchar_t (Cygwin)
  when converting an input string containing certain 4-byte UTF-8
  sequences to lower case.  The conversions to wchar_t and back to
  a UTF-8 multibyte string did not take surrogate pairs into account.
  [bug present since at least grep-2.6, though the segfault is new with 2.13]

  grep -E would segfault when given a regexp like '([^.]*[M]){1,2}'
  for any multibyte character M. [bug introduced in grep-2.6, which would
  segfault, but 2.7 and 2.8 had no problem, and 2.9 through 2.14 would
  hit a failed assertion. ]

  grep -F would get stuck in an infinite loop when given a search string
  that is an invalid byte sequence in the current locale and that matches
  the bytes of the input twice on a line.  Now grep fails with exit status 1.

  grep -P could misbehave.  While multi-byte mode is only supported by PCRE
  with UTF-8 locales, grep did not activate it.  This would cause failures
  to match multibyte characters against some regular expressions, especially
  those including the '.' or '\p' metacharacters.

** New features

  grep -P can now use a just-in-time compiler to greatly speed up matches,
  This feature is transparent to the user; no flag is required to enable
  it.  It is only available if the corresponding support in the PCRE
  library is detected when grep is compiled.


* Noteworthy changes in release 2.14 (2012-08-20) [stable]

** Bug fixes

  grep -i '^$' could exit 0 (i.e., report a match) in a multi-byte locale,
  even though there was no match, and the command generated no output.
  E.g., seq 2 | LC_ALL=en_US.utf8 grep -il '^$' would mistakenly print
  "(standard input)".  Related, seq 9 | LC_ALL=en_US.utf8 grep -in '^$'
  would print "2:4:6:8:10:12:14:16" and exit 0.  Now it prints nothing
  and exits with status of 1.  [bug introduced in grep-2.6]

  'grep' no longer falsely reports text files as being binary on file
  systems that compress contents or that store tiny contents in metadata.


* Noteworthy changes in release 2.13 (2012-07-04) [stable]

** Bug fixes

  grep -i, in a multi-byte locale, when matching a line containing a character
  like the UTF-8 Turkish I-with-dot (U+0130) (whose lower-case representation
  occupies fewer bytes), would print an incomplete output line.
  Similarly, with a matched line containing a character (e.g., the Latin
  capital I in a Turkish UTF-8 locale), where the lower-case representation
  occupies more bytes, grep could print garbage.
  [bug introduced in grep-2.6]

  --include and --exclude can again be combined, and again apply to
  the command line, e.g., "grep --include='*.[ch]' --exclude='system.h'
  PATTERN *" again reads all *.c and *.h files except for system.h.
  [bug introduced in grep-2.6]

** New features

  'grep' without -z now treats a sparse file as binary, if it can
  easily determine that the file is sparse.

** Dropped features

  Bootstrapping with Makefile.boot has been broken since grep 2.6,
  and was removed.


* Noteworthy changes in release 2.12 (2012-04-23) [stable]

** Bug fixes

  "echo P|grep --devices=skip P" once again prints P, as it did in 2.10
  [bug introduced in grep-2.11]

  grep no longer segfaults with -r --exclude-dir and no file operand.
  I.e., ":|grep -r --exclude-dir=D PAT" would segfault.
  [bug introduced in grep-2.11]

  Recursive grep now uses fts for directory traversal, so it can
  handle much-larger directories without reporting things like "File
  name too long", and it can run much faster when dealing with large
  directory hierarchies. [bug present since the beginning]

  grep -E 'a{1000000000}' now reports an overflow error rather than
  silently acting like grep -E 'a\{1000000000}'.

  grep -E 'a{,10}' was not treated equivalently to grep -E 'a{0,10}'.

** New features

  The -R option now has a long-option alias --dereference-recursive.

** Changes in behavior

  The -r (--recursive) option now follows only command-line symlinks.
  Also, by default -r now reads a device only if it is named on the command
  line; this can be overridden with --devices.  -R acts as before, so
  use -R if you prefer the old behavior of following all symlinks and
  defaulting to reading all devices.


* Noteworthy changes in release 2.11 (2012-03-02) [stable]

** Bug fixes

  grep no longer dumps core on lines whose lengths do not fit in 'int'.
  (e.g., lines longer than 2 GiB on a typical 64-bit host).
  Instead, grep either works as expected, or reports an error.
  An error can occur if not enough main memory is available, or if the
  GNU C library's regular expression functions cannot handle such long lines.
  [bug present since "the beginning"]

  The -m, -A, -B, and -C options no longer mishandle context line
  counts that do not fit in 'int'.  Also, grep -c's counts are now
  limited by the type 'intmax_t' (typically less than 2**63) rather
  than 'int' (typically less than 2**31).

  grep no longer silently suppresses errors when reading a directory
  as if it were a text file.  For example, "grep x ." now reports a
  read error on most systems; formerly, it ignored the error.
  [bug introduced in grep-2.5]

  grep now exits with status 2 if a directory loop is found,
  instead of possibly exiting with status 0 or 1.
  [bug introduced in grep-2.3]

  The -s option now suppresses certain input error diagnostics that it
  formerly failed to suppress.  These include errors when closing the
  input, when lseeking the input, and when the input is also the output.
  [bug introduced in grep-2.4]

  On POSIX systems, commands like "grep PAT < FILE >> FILE"
  now report an error instead of looping.
  [bug present since "the beginning"]

  The --include, --exclude, and --exclude-dir options now handle
  command-line arguments more consistently.  --include and --exclude
  apply only to non-directories and --exclude-dir applies only to
  directories.  "-" (standard input) is never excluded, since it is
  not a file name.
  [bug introduced in grep-2.5]

  grep no longer rejects "grep -qr . > out", i.e., when run with -q
  and an input file is the same as the output file, since with -q
  grep generates no output, so there is no risk of infinite loop or
  of an output-affecting race condition.  Thus, the use of the following
  options also disables the input-equals-output failure:
    --max-count=N (-m) (for N >= 2)
    --files-with-matches (-l)
    --files-without-match (-L)
  [bug introduced in grep-2.10]

  grep no longer emits an error message and quits on MS-Windows when
  invoked with the -r option.

  grep no longer misinterprets some alternations involving anchors
  (^, $, \<  \>  \B, \b).  For example, grep -E "(^|\B)a" no
  longer reports a match for the string "x a".
  [bug present since "the beginning"]

** New features

  If no file operand is given, and a command-line -r or equivalent
  option is given, grep now searches the working directory.  Formerly
  grep ignored the -r and searched standard input nonrecursively.
  An -r found in GREP_OPTIONS does not have this new effect.

  grep now supports color highlighting of matches on MS-Windows.

** Changes in behavior

  Use of the --mmap option now elicits a warning.  It has been a no-op
  since March of 2010.

  grep no longer diagnoses write errors repeatedly; it exits after
  diagnosing the first write error.  This is better behavior when
  writing to a dangling pipe.

  Syntax errors in GREP_COLORS are now ignored, instead of sometimes
  eliciting warnings.  This is more consistent with programs that
  (e.g.) ignore errors in termcap entries.

* Noteworthy changes in release 2.10 (2011-11-16) [stable]

** Bug fixes

  grep no longer mishandles high-bit-set pattern bytes on systems
  where "char" is a signed type. [bug appears to affect only MS-Windows]

  On POSIX systems, grep now rejects a command like "grep -r pattern . > out",
  in which the output file is also one of the inputs,
  because it can result in an "infinite" disk-filling loop.
  [bug present since "the beginning"]

** Build-related

  "make dist" no longer builds .tar.gz files.
  xz is portable enough and in wide-enough use that distributing
  only .tar.xz files is enough.


* Noteworthy changes in release 2.9 (2011-06-21) [stable]

** Bug fixes

  grep no longer clobbers heap for an ERE like '(^| )*( |$)'
  [bug introduced in grep-2.6]

  grep is faster on regular expressions that match multibyte characters
  in brackets (such as '[áéíóú]').

  echo c|grep '[c]' would fail for any c in 0x80..0xff, with a uni-byte
  encoding for which the byte-to-wide-char mapping is nontrivial.  For
  example, the ISO-88591 locales are not affected, but ru_RU.KOI8-R is.
  [bug introduced in grep-2.6]

  grep -P no longer aborts when PCRE's backtracking limit is exceeded
  Before, echo aaaaaaaaaaaaaab |grep -P '((a+)*)+$' would abort.  Now,
  it diagnoses the problem and exits with status 2.


* Noteworthy changes in release 2.8 (2011-05-13) [stable]

** Bug fixes

  echo c|grep '[c]' would fail for any c in 0x80..0xff, and in many locales.
  E.g., printf '\xff\n'|grep "$(printf '[\xff]')" || echo FAIL
  would print FAIL rather than the required matching line.
  [bug introduced in grep-2.6]

  grep's interpretation of range expression is now more consistent with
  that of other tools.  [bug present since multi-byte character set
  support was introduced in 2.5.2, though the steps needed to reproduce
  it changed in grep-2.6]

  grep erroneously returned with exit status 1 on some memory allocation
  failure. [bug present since "the beginning"]


* Noteworthy changes in release 2.7 (2010-09-16) [stable]

** Bug fixes

  grep --include=FILE works once again, rather than working like --exclude=FILE
  [bug introduced in grep-2.6]

  Searching with grep -Fw for an empty string would not match an
  empty line. [bug present since "the beginning"]

  X{0,0} is implemented correctly.  It used to be a synonym of X{0,1}.
  [bug present since "the beginning"]

  In multibyte locales, regular expressions including back-references
  no longer exhibit quadratic complexity (i.e., they are orders
  of magnitude faster). [bug present since multi-byte character set
  support was introduced in 2.5.2]

  In UTF-8 locales, regular expressions including "." can be orders
  of magnitude faster.  For example, "grep ." is now twice as fast
  as "grep -v ^$", instead of being immensely slower.  It remains
  slow in other multibyte locales. [bug present since multi-byte
  character set support was introduced in 2.5.2]

  --mmap was meant to be ignored in 2.6.x, but it was instead
  removed by mistake.  [bug introduced in 2.6]

** New features

  grep now diagnoses (and fails with exit status 2) commonly mistyped
  regular expression like [:space:], [:digit:], etc.  Before, those were
  silently interpreted as [ac:eps] and [dgit:] respectively.  Virtually
  all who make that class of mistake should have used [[:space:]] or
  [[:digit:]].  This new behavior is disabled when the POSIXLY_CORRECT
  environment variable is set.

  On systems using glibc, grep can support equivalence classes.  However,
  whether they actually work depends on glibc's locale definitions.

* Noteworthy changes in release 2.6.3 (2010-04-02) [stable]

** Bug fixes

  Searching with grep -F for an empty string in a multibyte locale
  would hang grep. [bug introduced in 2.6.2]

  PCRE support is once again detected on systems with <pcre/pcre.h>
  [bug introduced in 2.6.2]


* Noteworthy changes in release 2.6.2 (2010-03-29) [stable]

** Bug fixes

  grep -F no longer mistakenly reports a match when searching
  for an incomplete prefix of a multibyte character.
  [bug present since "the beginning"]

  grep -F no longer goes into an infinite loop when it finds a match for an
  incomplete (non-prefix of a) multibyte character.  [bug introduced in 2.6]

  Using any of the --include or --exclude* options would cause a NULL
  dereference.  [bugs introduced in 2.6]

** Build-related

  configure no longer relies on pkg-config to detect PCRE support.


* Noteworthy changes in release 2.6.1 (2010-03-25) [stable]

** Bug fixes

  Character classes could cause a segmentation fault if they included a
  multibyte character.  [bug introduced in 2.6]

  Character ranges would not work in single-byte character sets other
  than C (for example, ISO-8859-1 or KOI8-R) and some multi-byte locales.
  For example, this should print "1", but would find no match:
    $ echo 1 | env -i LC_COLLATE=en_US.UTF-8 grep '[0-9]'
  [bug introduced in 2.6]

  The output of grep was incorrect for whole-word (-w) matches if the
  patterns included a back-reference. [bug introduced in grep-2.5.2]

** Portability

  Avoid a link failure on Solaris 8.


* Noteworthy changes in release 2.6 (2010-03-23) [stable]

** Speed improvements

  grep is much faster on multibyte character sets, especially (but not
  limited to) UTF-8 character sets.  The speed improvement is also very
  pronounced with case-insensitive matches.

** Bug fixes

  Character classes would malfunction in multi-byte locales when using grep -i.
  Examples which would print nothing for LC_ALL=en_US.UTF-8 include:
  - for ranges, echo Z | grep -i '[a-z]'
  - for single characters, echo Y | grep -i '[y]'
  - for character types, echo Y | grep -i '[[:lower:]]'

  grep -i -o would fail to report some matches; grep -i --color, while not
  missing any line containing a match, would fail to color some matches.

  grep would fail to report a match in a multibyte character set other than
  UTF-8, if another match occurred earlier in the line but started in the
  middle of a multibyte character.

  Various bugs in grep -P, caused by expressions such as [^b] or \S matching
  newlines, were fixed.  grep -P also supports the special sequences \Z and
  \z, and can be combined with the command-line option -z to perform searches
  on NUL-separated records.

  grep would mistakenly exit with status 1 upon error, rather than 2,
  as it is documented to do.

  Using options like -1 -2 or -1 -v -2 results in two lines of
  context (the last value that appears on the command line) instead
  twelve (the concatenation of all the values).  This is consistent
  with the behavior of options -A/-B/-C.

  Two new command-line options, --group-separator=ARGUMENT and
  --no-group-separator, enable further customization of the output
  when -A, -B or -C is being used.

** Other changes

  egrep accepts the -E option and fgrep accepts the -F option.  If egrep
  and fgrep are given another of the -E/-F/-G options, they print a more
  meaningful error message.

* Noteworthy changes in release 2.5.4 (2009-02-10) [stable]

  - This is a bugfix release. No new features.

Version 2.5.3
  - The new option --exclude-dir allows to specify a directory pattern that
    will be excluded from recursive grep.
  - Numerous bug fixes

Version 2.5.1
  - This is a bugfix release. No new features.

Version 2.5
  - The new option --label allows to specify a different name for input
    from stdin. See the man or info pages for details.

  - The internal lib/getopt* files are no longer used on systems providing
    getopt functionality in their libc (e.g. glibc 2.2.x).
    If you need the old getopt files, use --with-included-getopt.

  - The new option --only-matching (-o) will print only the part of matching
    lines that matches the pattern. This is useful, for example, to extract
    IP addresses from log files.

  - i18n bug fixed ([A-Z0-9] wouldn't match A in locales other than C on
    systems using recent glibc builds

  - GNU grep can now be built with autoconf 2.52.

  - The new option --devices controls how grep handles device files. Its usage
    is analogous to --directories.

  - The new option --line-buffered fflush on everyline.  There is a noticeable
    slow down when forcing line buffering.

  - Back-references are now local to the regex.
    grep -e '\(a\)\1' -e '\(b\)\1'
    The last backref \1 in the second expression refer to \(b\)

  - The new option --include=PATTERN will search only matching files
    when recursing in directories

  - The new option --exclude=PATTERN will skip matching files when
    recursing in directories.

  - The new option --color will use the environment variable GREP_COLOR
    (default is red) to highlight the matching string.
    --color takes an optional argument specifying when to colorize a line:
      --color=always, --color=tty, --color=never

  - The following changes are for POSIX conformance:

    . The -q or --quiet or --silent option now causes grep to exit
      with zero status when a input line is selected, even if an error
      also occurs.

    . The -s or --no-messages option no longer affects the exit status.

    . Bracket regular expressions like [a-z] are now locale-dependent.
      For example, many locales sort characters in dictionary order,
      and in these locales the regular expression [a-d] is not
      equivalent to [abcd]; it might be equivalent to [aBbCcDd], for
      example.  To obtain the traditional interpretation of bracket
      expressions, you can use the C locale by setting the LC_ALL
      environment variable to the value "C".

  - The -C or --context option now requires an argument, partly for
    consistency, and partly because POSIX recommends against
    optional arguments.

  - The new -P or --perl-regexp option tells grep to interpret the pattern as
    a Perl regular expression.

  - The new option --max-count=num makes grep stop reading a file after num
    matching lines.
    New option -m; equivalent to --max-count.

  - Translations for bg, ca, da, nb and tr have been added.

Version 2.4.2

  - Added more check in configure to default the grep-${version}/src/regex.c
    instead of the one in GNU Lib C.

Version 2.4.1

  - If the final byte of an input file is not a newline, grep now silently
    supplies one.

  - The new option --binary-files=TYPE makes grep assume that a binary input
    file is of type TYPE.
    --binary-files='binary' (the default) outputs a 1-line summary of matches.
    --binary-files='without-match' assumes binary files do not match.
    --binary-files='text' treats binary files as text
        (equivalent to the -a or --text option).

  - New option -I; equivalent to --binary-files='without-match'.

Version 2.4:

  - egrep is now equivalent to 'grep -E' as required by POSIX,
    removing a longstanding source of confusion and incompatibility.
    'grep' is now more forgiving about stray '{'s, for backward
    compatibility with traditional egrep.

  - The lower bound of an interval is not optional.
    You must use an explicit zero, e.g. 'x{0,10}' instead of 'x{,10}'.
    (The old documentation incorrectly claimed that it was optional.)

  - The --revert-match option has been renamed to --invert-match.

  - The --fixed-regexp option has been renamed to --fixed-strings.

  - New option -H or --with-filename.

  - New option --mmap.  By default, GNU grep now uses read instead of mmap.
    This is faster on some hosts, and is safer on all.

  - The new option -z or --null-data causes 'grep' to treat a zero byte
    (the ASCII NUL character) as a line terminator in input data, and
    to treat newlines as ordinary data.

  - The new option -Z or --null causes 'grep' to output a zero byte
    instead of the normal separator after a file name.

  - These two options can be used with commands like 'find -print0',
    'perl -0', 'sort -z', and 'xargs -0' to process arbitrary file names,
    even those that contain newlines.

  - The environment variable GREP_OPTIONS specifies default options;
    e.g. GREP_OPTIONS='--directories=skip' reestablishes grep 2.1's
    behavior of silently skipping directories.

  - You can specify a matcher multiple times without error, e.g.
    'grep -E -E' or 'fgrep -F'.  It is still an error to specify
    conflicting matchers.

  - -u and -U are now allowed on non-DOS hosts, and have no effect.

  - Modifications of the tests scripts to go around the "Broken Pipe"
    errors from bash. See Bash FAQ.

  - New option -r or --recursive or --directories=recurse.
    (This option was also in grep 2.3, but wasn't announced here.)

  - --without-included-regex disable, was causing bogus reports .i.e
    doing more harm then good.

Version 2.3:

  - When searching a binary file FOO, grep now just reports
    "Binary file FOO matches" instead of outputting binary data.
    This is typically more useful than the old behavior,
    and it is also more consistent with other utilities like 'diff'.
    A file is considered to be binary if it contains a NUL (i.e. zero) byte.

    The new -a or --text option causes 'grep' to assume that all
    input is text.  (This option has the same meaning as with 'diff'.)
    Use it if you want binary data in your output.

  - 'grep' now searches directories just like ordinary files; it no longer
    silently skips directories.  This is the traditional behavior of
    Unix text utilities (in particular, of traditional 'grep').
    Hence 'grep PATTERN DIRECTORY' should report
    "grep: DIRECTORY: Is a directory" on hosts where the operating system
    does not permit programs to read directories directly, and
    "grep: DIRECTORY: Binary file matches" (or nothing) otherwise.

    The new -d ACTION or --directories=ACTION option affects directory handling.
    '-d skip' causes 'grep' to silently skip directories, as in grep 2.1;
    '-d read' (the default) causes 'grep' to read directories if possible,
    as in earlier versions of grep.

  - The MS-DOS and Microsoft Windows ports now behave identically to the
    GNU and Unix ports with respect to binary files and directories.

Version 2.2:

Bug fix release.

  - Status error number fix.
  - Skipping directories removed.
  - Many typos fix.
  - -f /dev/null fix(not to consider as an empty pattern).
  - Checks for wctype/wchar.
  - -E was using the wrong matcher fix.
  - bug in regex char class fix
  - Fixes for DJGPP

Version 2.1:

This is a bug fix release(see Changelog) i.e. no new features.

  - More compliance to GNU standard.
  - Long options.
  - Internationalization.
  - Use automake/autoconf.
  - Directory hierarchy change.
  - Sigvec with -e on Linux corrected.
  - Sigvec with -f on Linux corrected.
  - Sigvec with the mmap() corrected.
  - Bug in kwset corrected.
  - -q, -L and -l stop on first match.
  - New and improve regex.[ch] from Ulrich Drepper.
  - New and improve dfa.[ch] from Arnold Robbins.
  - Prototypes for over zealous C compiler.
  - Not scanning a file, if it's a directory
    (cause problems on Sun).
  - Ported to MS-DOS/MS-Windows with DJGPP tools.

See Changelog for the full story and proper credits.

Version 2.0:

The most important user visible change is that egrep and fgrep have
disappeared as separate programs into the single grep program mandated
by POSIX 1003.2.  New options -G, -E, and -F have been added,
selecting grep, egrep, and fgrep behavior respectively.  For
compatibility with historical practice, hard links named egrep and
fgrep are also provided.  See the manual page for details.

In addition, the regular expression facilities described in Posix
draft 11.2 are now supported, except for internationalization features
related to locale-dependent collating sequence information.

There is a new option, -L, which is like -l except it lists
files which don't contain matches.  The reason this option was
added is because '-l -v' doesn't do what you expect.

Performance has been improved; the amount of improvement is platform
dependent, but (for example) grep 2.0 typically runs at least 30% faster
than grep 1.6 on a DECstation using the MIPS compiler.  Where possible,
grep now uses mmap() for file input; on a Sun 4 running SunOS 4.1 this
may cut system time by as much as half, for a total reduction in running
time by nearly 50%.  On machines that don't use mmap(), the buffering
code has been rewritten to choose more favorable alignments and buffer
sizes for read().

Portability has been substantially cleaned up, and an automatic
configure script is now provided.

The internals have changed in ways too numerous to mention.
People brave enough to reuse the DFA matcher in other programs
will now have their bravery amply "rewarded", for the interface
to that file has been completely changed.  Some changes were
necessary to track the evolution of the regex package, and since
I was changing it anyway I decided to do a general cleanup.

========================================================================
Copyright (C) 1992, 1997-2002, 2004-2020 Free Software Foundation, Inc.

  Copying and distribution of this file, with or without modification,
  are permitted in any medium without royalty provided the copyright
  notice and this notice are preserved.

Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with no
Invariant Sections, with no Front-Cover Texts, and with no Back-Cover
Texts.  A copy of the license is included in the "GNU Free
Documentation License" file as part of this distribution.