Re: v2.25.0-rc0 test failure on cygwin

From: Ramsay Jones <ramsay@ramsayjones.plus.com>
Date: Jun 3, 2025, 19:03 UTC
Message-ID: <780f09cb-5399-4256-91ca-d93a553540c0@ramsayjones.plus.com>
In-Reply-To: <xmqq5xhcpzyp.fsf@gitster.g>

On 03/06/2025 17:11, Junio C Hamano wrote:

> Ramsay Jones <ramsay@ramsayjones.plus.com> writes:
>

[snip]

Show 7 quoted lines

>> A quick squint at the failing tests made it clear that the failure was
>> caused by the cygwin build treating a quoted glob character sequence
>> (e.g. '\*') as a directory separator char followed by a glob character.
> 
> Should we revert ec727e18 (dir.c: literal match with wildcard in
> pathspec should still glob, 2025-05-03) before deciding to how to
> proceed, as we will be deep in prerelease freeze?

Heh, that decision is above my pay grade! :)

BTW, I realized earlier today why we had not seen any CI failures on windows (yes I'm a bit slow on the uptake!); the test has this conditional at the top:

  test_have_prereq FUNNYNAMES || {
         skip_all='skipping: needs FUNNYNAMES (non-Windows only)'
         test_done
  }

And FUNNYNAMES (as indicated) is set on MINGW/Gfw, but not cygwin. (cygwin can cope with many 'funny' filenames, but it can't change the use of '\' as a win32 directory separator).

> It's not a "fix" for something that is gravely wrong but a glitch
> people have lived with almost forever, so it can be reattempted in
> the next cycle without hurting much.

The POSIX/win32 pathname syntax 'issue' is not going away anytime soon, so I suspect that just skipping these tests (or the entire test file ala Gfw) is probably the best it gets. ;)

[see Patrick's commit 5f8af25ff9 (t5500, t5601: skip tests which exercise paths with '[::1]' on Cygwin, 2024-10-16), which had been failing from v2.25.0-rc0 (25 Dec 2019) to v2.48.0-rc0 (16 Dec 2024), which was actually caused by a commit which started allowing a win32 drive 'letter' to be an multi-byte UTF8 Unicode char! :) ]

Thanks.

ATB, Ramsay Jones