Re: [ANNOUNCE] Git v2.50.0-rc1 - Test Failed

From: Johannes Sixt <j6t@kdbg.org>
Date: Jun 4, 2025, 19:25 UTC
Message-ID: <44fe8627-5680-443d-bf02-a6e85afd46b4@kdbg.org>
In-Reply-To: <xmqqjz5rcz90.fsf@gitster.g>

Am 04.06.25 um 17:17 schrieb Junio C Hamano:

Show 5 quoted lines

> So the build procedure for git-gui (but not gitk) has changed rather
> extensively after we tagged the preview before -rc1?  Honestly, I
> would have preferred to see a change with this impact go through the
> regular 'seen' to 'next' to 'master' way before -rc0, but that is
> water under the bridge.

I don't think we ever had such a cycle for gitk and git-gui. I carry inofficial branches 'j6t-testing' in my repositories that interested parties could track instead of 'master'. I would be happy to hear that people actually do use them.

Show 16 quoted lines

> I do not spot anything obviously wrong (and it is not expected that
> I would---we wouldn't have this code sent to me in the first place
> if this is something I can immediately notice).  git-gui/Makefile
> sets ALL_LIBFILES to $(wildcard lib/*.tcl) and then does
> 
>     $(SHELL_PATH) generate-tclindex.sh . ./GIT-GUI-BUILD-OPTIONS $(ALL_LIBFILES)
> 
> So the error message in Becker's message, i.e.
> 
>> /usr/coreutils/bin/bash generate-tclindex.sh . ./GIT-GUI-BUILD-OPTIONS 
>> usage: generate-tclindex.sh <BUILD_DIR> <BUILD_OPTIONS> <LIBFILE> [<LIBFILE>...]
>> Makefile:200: recipe for target 'lib/tclIndex' failed
> 
> suggests that $(wildcard lib/*tcl) expanded to *nothing*, which
> sounds horribly wrong.  They are source material and should exist in
> an unmodified checkout or a tarball extract.

I don't see anything wrong, either. I can easily verify your theory that the $(wildcard) produces an empty list by modifying the pattern.

Randall, would it be possible for you to find out why $(wildcard lib/*tcl) produces an empty list in your case?

-- Hannes