Re: BUG: git grep behave oddly with alternatives

Am 03.01.23 um 19:13 schrieb Marco Nenciarini:

Meaning you used Apple's version of git before?

   $ echo abd > test.file
   $ /usr/bin/git grep --untracked 'b\|c' test.file
   test.file:abd
   $ /usr/bin/git version
   git version 2.37.1 (Apple Git-137.1)

   $ git grep --untracked 'b\|c' test.file
   $ git version
   git version 2.39.0

So I can reproduce your findings on macOS Ventura.  Same with grep:

   $ grep 'b\|c' test.file
   abd
   $ grep --version
   grep (BSD grep, GNU compatible) 2.6.0-FreeBSD

re_format(7) says:

   "Obsolete (“basic”) regular expressions differ in several respects.
    ‘|’ is an ordinary character and there is no equivalent for its
    functionality.".

Under the headline "ENHANCED FEATURES" it continues, however:

   "When the REG_ENHANCED flag is passed to one of the regcomp()
    variants, additional features are activated."

And:

   "For enhanced basic REs, ‘+’, ‘?’ and ‘|’ remain regular characters,
    but ‘\+’, ‘\?’ and ‘\|’ have the same special meaning as the
    unescaped characters do for extended REs, i.e., one or more
    matches, zero or one matches and alteration, respectively."

So apparently Apple chose a middle ground between basic and extended regular expressions for its grep and git grep.

Under "IMPLEMENTATION CHOICES" it says:

   "The regex implementation in Mac OS X 10.8 and later is based on a
    heavily modified subset of TRE (http://laurikari.net/tre/)."

The manpage of GNU grep says:

   "grep understands three different versions of regular expression
    syntax: “basic” (BRE), “extended” (ERE) and “perl” (PCRE).  In
    GNU grep there is no difference in available functionality
    between basic and extended syntax.  In other implementations,
    basic regular expressions are less powerful."

And under the headline "Basic vs Extended Regular Expressions":

   "In basic regular expressions the meta-characters ?, +, {, |, (,
    and ) lose their special meaning; instead use the backslashed
    versions \?, \+, \{, \|, \(, and \)."

So GNU grep apparently made the same choice as Apple, probably far earlier.

When I compile git with NO_REGEX and thus with the fallback in compat/regex/, the result supports alternation as well:

   $ ./git grep --untracked 'b\|c' test.file
   test.file:abd
   $ nm ./git | grep regcomp
   0000000100255978 T _git_regcomp

Based on that I suggest:

--- >8 ---
Subject: grep: use REG_ENHANCED on macOS

GNU grep(1) and regcomp(3) use enhanced basic regular expressions by default, which means that it e.g. supports alternation, e.g. "a\|b" matches both "a" and "b". The same is true for our compat/regex/ implementation.

On macOS Ventura (and possibly earlier) grep(1) also uses enhanced BREs, but regcomp(3) requires the flag REG_ENHANCED to turn them on. Do that for git grep if possible, for consistency with the platform's grep(1) and our fallback library.

It would be simpler to use REG_ENHANCED everywhere it is defined instead of adding a Makefile setting, but it's a non-standard extension and might mean something else on other platforms.

Reported-by: Marco Nenciarini <marco.nenciarini@enterprisedb.com>
Signed-off-by: René Scharfe <l.s.r@web.de>
---
 Makefile         | 8 ++++++++
 config.mak.uname | 1 +
 grep.c           | 4 ++++
 3 files changed, 13 insertions(+)