[New git version] v2.33.1

[git-for-windows-ci]

https://github.com/git/git/releases/tag/v2.33.1

[dscho]

Right. This was unexpected, and I had counted on a couple of weeks to be able to stabilize a couple of things:

• We recently replaced FSMonitor with v4, and wanted to get v5 integrated, too: FSMonitor: deepening a directory causes confusing events #3456
• We had to rebuild the MSYS2 runtime to suppress an annoying warning when running on Windows/ARM64 (which was not without problems)
• A new GCM Core version was published as prerelease and is hopefully promoted to a full release soon: [New gcm-core version] GCM 2.0.567 #3457
• I had hoped to merge init: respect core.symlinks before copying the templates #3417 (using it as an excuse to have a snapshot that integrated the new GCM Core version)
• We also upgraded ca-certificates and less for i686 (because the MSYS2 project stopped building i686 MSYS packages) and wanted to figure out what other packages need to be upgraded for i686.
• Assuming that I had a couple of weeks until the next full Git for Windows release, I finished rotating my GPG key. This already led to a couple of hiccups in the automation and I fully expect more hiccups down the road.
• We picked up an installer PR that really could have used some testing in the wild: Use code.cmd instead of Code.exe build-extra#384
• Fully expecting no .1 release, I merged the PR to drop GCM for Windows (which is an archived project, no longer maintained): Drop Git Credential Manager for Windows build-extra#377. I had hoped to introduce this change in a .0 release.

I probably forgot more things that pose a particular challenge for upgrading Git for Windows to this Git version. With all that in mind, it might take me a while to get this release out.

[dscho]

• I finished rotating my GPG key. This already led to a couple of hiccups in the automation and I fully expect more hiccups down the road.

To illustrate what I mean, have a look at this failure:

gpg: assuming signed data in '/tmp/tmp.iiV7YLCmLW/i686/git-for-windows.db.tar.xz'
gpg: Signature made Tue, Oct 12, 2021  6:28:16 PM GMT
gpg:                using RSA key 91883E11E83DC29D14104DB4EDD44359093056EE
gpg: checking the trustdb
gpg: marginals needed: 3  completes needed: 1  trust model: pgp
gpg: depth: 0  valid:   1  signed:   0  trust: 0-, 0q, 0n, 0m, 0f, 1u
gpg: next trustdb check due at 2026-03-22
gpg: Good signature from "Johannes Schindelin <[email protected]>" [ultimate]
Downloading current i686/git-for-windows.files.tar.xz...
gpg: assuming signed data in '/tmp/tmp.iiV7YLCmLW/i686/git-for-windows.files.tar.xz'
gpg: Signature made Tue, Oct 12, 2021  6:28:17 PM GMT
gpg:                using RSA key 91883E11E83DC29D14104DB4EDD44359093056EE
gpg: Good signature from "Johannes Schindelin <[email protected]>" [ultimate]
Downloading current i686/git-for-windows-mingw32.db.tar.xz...
gpg: assuming signed data in '/tmp/tmp.iiV7YLCmLW/i686/git-for-windows-mingw32.db.tar.xz'
gpg: Signature made Mon, Oct 11, 2021  3:52:54 PM GMT
gpg:                using RSA key 57759F748D223F034D8BE870BB3AA74136C569BB
gpg: Can't check signature: No public key
Could not verify GPG signature: /tmp/tmp.iiV7YLCmLW/i686/git-for-windows-mingw32.db.tar.xz
##[error]PowerShell exited with code '1'.

The problem here is that simply rotating the GPG key causes a failure while validatingof the current signature (which was done using the previous GPG key).

[dscho]

• We had to rebuild the MSYS2 runtime

And of course this also had to cause issues: the git-sdk-64-minimal artifact (which we use to build Git and run its test suite in CI) now has an msys-2.0.dll that is incompatible with the one in C:\Program Files\Git\usr\bin, but the latter is picked up when running gpg.exe (due to the unfortunate setup of the build agents whereby all of Git for Windows' Unix tools are exposed via PATH). As a result, when trying to run the GPG-related tests, GPG would fail, but not when determining the prereq. So now I will have to fix that, too.

[dscho]

My current plan is to wait for the Git artifacts run to pass, then to merge #3417.

After that, I hope that the final version of #3456 is ready to be merged. At that stage, I can start the final merging-rebase onto v2.33.1.

As mentioned before, all of this might spill over into tomorrow.

[dscho]

Here is the PR: #3465

[dscho]

And the Git artifacts run is here.

[dscho]

And my testing showed that we need one more thing. See #3465 (comment) for details.

[calle2010]

Thanks for all the efforts you put into Git for Windows!

I just wanted to leave here that the update from 2.33.0(2) to 2.33.1 contains two potentially breaking changes in dependencies:

• OpenSSH 8.8 disables usage of SHA-1 signatures: https://www.openssh.com/txt/release-8.8
• LFS 3.0.1 doesn't work with Bitbucket Data Center/Server before release 7.16.1: https://community.atlassian.com/t5/Bitbucket-articles/Git-LFS-v3-0-1-includes-backwards-incompatible-changes-with/ba-p/1819819

The OpenSSH incompatibility can easily be worked around (see release notes). But for the LFS change I must wait for an upgrade of our Bitbucket server.

[dscho]

@calle2010 good point. Could you put up two discussions for those two issues? I would like to pin them, for visibility.