add a msys2-runtime-3.3 package

[rimrul]

The first step of phase 2 of the current timeline(git-for-windows/git#4279 (comment))

[dscho]

These changes look pretty good so far, thank you so much for working on this @rimrul!

I allowed myself to add a commit to trigger i686 builds for PRs containing the needle msys2-runtime-3.3 in their title, as this here PR does. I hope to get an artifact that I can then test.

[dscho]

@rimrul from my side, I consider this one done (modulo the final interactive rebase). Anything else you want to see before flipping this to "ready-for review"?

[dscho]

Yeah!

[dscho]

/deploy msys2-runtime-3.3

The x86_64 and the i686 workflow runs were started.

[dscho]

I guess technically we do not need the x86_64 builds, but what's the harm, eh? 😁

[rimrul]

Yes, that should be fine.

[dscho]

I think the next step should be a PR in git-sdk-32 that teaches the sync job (or more correctly: update-via-pacman.ps1 to replace the msys2-runtime (if installed) package by the msys2-runtime-3.3.

[dscho]

I guess technically we do not need the x86_64 builds, but what's the harm, eh? 😁

So now I know what the harm is: https://github.com/git-for-windows/git-sdk-64/actions/runs/4921830149/jobs/8792046619 says this:

[...]
:: Starting core system upgrade...
warning: terminate other MSYS2 programs before proceeding
resolving dependencies...
looking for conflicting packages...
error: failed to prepare transaction (could not satisfy dependencies)
:: installing msys2-runtime (3.4.6-2) breaks dependency 'msys2-runtime=3.3.6' required by msys2-runtime-devel
[...]

Same problem in the sync job of git-sdk-arm64.

This suggests to me that the -3.3- in msys2-runtime-3.3-3.3.6-1 is causing problems.

I fear that we'll have to repo-remove it ASAP, and hope that simply renaming the package to msys2-runtime3.3 will work...

[dscho]

Hmm. MSYS2 does not seem to have any problems with this package name! See https://packages.msys2.org/package/msys2-runtime-3.3-devel?repo=msys&variant=x86_64

Also:

$ tar tvf /var/lib/pacman/sync/msys.db | grep msys2-runtime
drwxr-xr-x repo/repo         0 2023-04-02 10:39 msys2-runtime-3.3-3.3.6-5/
-rw-r--r-- repo/repo      1514 2023-04-02 10:39 msys2-runtime-3.3-3.3.6-5/desc
drwxr-xr-x repo/repo         0 2023-04-02 10:39 msys2-runtime-3.3-devel-3.3.6-5/
-rw-r--r-- repo/repo      1562 2023-04-02 10:39 msys2-runtime-3.3-devel-3.3.6-5/desc
drwxr-xr-x repo/repo         0 2023-04-02 10:39 msys2-runtime-3.4.6-2/
-rw-r--r-- repo/repo      1491 2023-04-02 10:39 msys2-runtime-3.4.6-2/desc
drwxr-xr-x repo/repo         0 2023-04-02 10:39 msys2-runtime-devel-3.4.6-2/
-rw-r--r-- repo/repo      1534 2023-04-02 10:39 msys2-runtime-devel-3.4.6-2/desc

So the problem must lie with Git for Windows' tooling... I suspect pacman-helper.sh...

[dscho]

So the problem must lie with Git for Windows' tooling... I suspect pacman-helper.sh...

Indeed, it looks that way. I consistently used the pattern ${dirname%%-[0-9]*} (e.g. here) in Git for Windows' tooling to deduce the package name from the directory name in the tar files /var/lib/pacman/sync/*.db. But Pacman actually uses ${dirname%-*-*}, see e.g. here.

Gotta fix that in All The Places. And then fix Git for Windows' Pacman databases...

[dscho]

Gotta fix that in All The Places.

See git-for-windows/build-extra#499 for that fix.

And then fix Git for Windows' Pacman databases...

Done in https://github.com/git-for-windows/git-for-windows-automation/actions/runs/4930413416/jobs/8811298893. The git-sdk-arm64 sync worked and I started the re-run of the git-sdk-64 sync, too.