Re: git bug: Perl compatible regular expressions do not work as expected

[Mario Grgic <mario_grgic@hotmail.com>]

Interesting. I could be wrong. Let me clarify what I did and what I observe. I edited the Makefile and put back the NO_REGEX = YesPlease line. Then I configured the build as 

./configure —with-libpcre2=/usr/local 

then built and installed git. 

otool -L git shows git binary is dynamically linking the libpcre in /usr/local

However, it looks like this third party library is also used for -G  searches (plain POSIX regex or PCRE), since it accepts and correctly finds things like ‘\btext\b’. That is 

git log —all -p -G ‘\main\b’

works

Furthermore, -G works with multibyte strings as well: 

git log --all -p  -G '顔🏁’

What doesn’t work is multibyte strings with PCRESs, so for example this does not work:

git log --all -p  -G ‘\b顔🏁\b’

This doesn’t seem to be consistent with the thesis that regex library from compat is used for -G searches and that it doesn’t support multibyte strings?


> On Mar 27, 2023, at 5:11 PM, Junio C Hamano <gitster@pobox.com> wrote:
> 
> Mario Grgic <mario_grgic@hotmail.com> writes:
> 
> [administrivia: do not top post]
> 
>>> Mario Grgic <mario_grgic@hotmail.com> writes:
>>> ...
>>>> +	NO_REGEX = YesPlease
>>>> 	PTHREAD_LIBS =
>>>> endif
>>> 
>>> It will unfortunately break multibyte support on macOS by reverting
>>> what 1819ad32 (grep: fix multibyte regex handling under macOS,
>>> 2022-08-26) did.
> 
>> In my case, I compiled git with pcre2 support, using third party
>> PCRE2 library: https://github.com/PCRE2Project/pcre2 and PCRE and
>> multibyte support in git works with it just fine.
> 
> Sorry, you misunderstood.  1819ad32 is about enabling multi-byte
> support for normal regexp types, and does not have anything to do
> with pcre.  By setting NO_REGEX, the build will not link with
> Apple's regex library but the one from compat/ and that version is
> what is used for -G and -E (not -P).  -G/-E patterns with multi-byte
> would not work with compat/ stuff, but they should work when linked
> with Apple's regex library.
>