Hi Patrick,
On Tue, 3 Jun 2025, Patrick Steinhardt wrote:
> On Mon, Jun 02, 2025 at 10:59:53AM -0500, Justin Tobler wrote:
> > On 25/06/02 05:01PM, Johannes Schindelin wrote:
> > > Hi Justin,
> > >
> > > On Tue, 20 May 2025, Justin Tobler wrote:
> > >
> > > > diff --git a/t/t5410-receive-pack.sh b/t/t5410-receive-pack.sh
> > > > index 9afea54a26..f76a22943e 100755
> > > > --- a/t/t5410-receive-pack.sh
> > > > +++ b/t/t5410-receive-pack.sh
> > > > @@ -62,4 +62,26 @@ test_expect_success 'receive-pack missing objects fails connectivity check' '
> > > > test_must_fail git -C remote.git cat-file -e $(git -C repo rev-parse HEAD)
> > > > '
> > > >
> > > > +test_expect_success 'receive-pack missing objects bypasses connectivity check' '
> > > > + test_when_finished rm -rf repo remote.git setup.git &&
> > > > +
> > > > + git init repo &&
> > > > + git -C repo commit --allow-empty -m 1 &&
> > > > + git clone --bare repo setup.git &&
> > > > + git -C repo commit --allow-empty -m 2 &&
> > > > +
> > > > + # Capture git-send-pack(1) output sent to git-receive-pack(1).
> > > > + git -C repo send-pack ../setup.git --all \
> > > > + --receive-pack="tee ${SQ}$(pwd)/out${SQ} | git-receive-pack" &&
> > > > +
> > > > + # Replay captured git-send-pack(1) output on new empty repository.
> > > > + git init --bare remote.git &&
> > > > + git receive-pack --skip-connectivity-check remote.git <out >actual 2>err &&
> > > > +
> > > > + test_grep ! "missing necessary objects" actual &&
> > > > + test_must_be_empty err &&
> > > > + git -C remote.git cat-file -e $(git -C repo rev-parse HEAD) &&
> > > > + test_must_fail git -C remote.git rev-list $(git -C repo rev-parse HEAD)
> > > > +'
> > > > +
> > > > test_done
> > >
> > > This test case seems to hang occasionally in the "win+Meson test" jobs on
> > > GitHub (I tried to find the same failure at
> > > https://gitlab.com/gitlab-org/git/-/pipelines but couldn't find any). See
> > > for example
> > > https://github.com/gitgitgadget/git/actions/runs/15383915635/job/43279134837#step:6:627
> > >
> > > Note that this problem afflicts only the "win+Meson test" jobs; The
> > > corresponding "win test" job seems not to hang.
> > >
> > > Even in the Git for Windows project, where the `win+VS test` jobs are run,
> > > the t5410 test passes within a dozen seconds or so, see e.g.
> > > https://github.com/git-for-windows/git/actions/runs/15383945895/job/43279689086#step:5:143
> > > (confusingly, the subset of tests run in the matrix jobs differs between
> > > the `win+Meson test` jobs and the `win+VS test` jobs, but if you click
> > > through all of the `win+Meson test` jobs, expand the `test` step,
> > > patiently wait a few seconds for the log to be lazy loaded "enough" for
> > > the search to work, you will notice that t5410 is not mentioned in any of
> > > them, and the only one that times out after 4h37m11s is
> > > https://github.com/git-for-windows/git/actions/runs/15383945895/job/43279753911,
> > > likely while running 5410, too).
> > >
> > > Do you have any idea why this particular test case, in conjunction with
> > > Windows and Meson (and only on GitHub) acts up like this?
> >
> > Thanks Johannes for the report. I'm not quite sure yet what is going on
> > here, but I'll dig into this a bit and see what I can figure out. :)
>
> I've been banging my head against this issue for a bit today. A couple
> of findings:
>
> - The issue is specific to Git for Windows, I could only reproduce it
> when working with aa550efd0bb (fixup??? survey: add command line
> opts to select references, 2025-05-08).
I can reproduce it consistently with Git's `master`, see e.g.
https://github.com/git/git/actions/runs/15454602308/job/43504424816#step:6:627
> - When working on top of the above commit the bug is consistent. It
> doesn't only happen in GitHub, but also happens in GitLab CI [1].
>
> - That being said, I still can't reproduce it locally?! This one is
> quite puzzling to me. I have tried to get my environment as close as
> possible to the environment we have in the CI systems.
I, too, was unable to reproduce locally (probably because of the way the
runners start the processes, without an initial Win32 Console and all). So
I took to mxschmitt/action-tmate to debug on the runner itself. It is a
bit tricky to do, as MSVC's debugger runs in a graphical IDE and gdb is
unable to find the symbols.
> - I have a fix, see the patch further down. But I don't understand
> that fix just yet.
I would like to propose an alternative:
https://lore.kernel.org/git/pull.1932.git.1749118606047.gitgitgadget@xxxxxxxxx
The reason why I prefer that solution is that I suspect the extra script
to make the conditions only less likely, but not impossible, for the bug
to rear its ugly head.
Ciao,
Johannes
>
> I saw that all other sites where inject a custom receive-pack command
> also use a wrapper script, so it's not the worst thing to do. But it
> would be great to understand why this issue exists in the first place.
>
> Patrick
>
> [1]: https://github.com/pks-t/git/actions/runs/15416185892/job/43379399861
>
> diff --git a/t/t5410-receive-pack.sh b/t/t5410-receive-pack.sh
> index f76a22943ef..112da408d45 100755
> --- a/t/t5410-receive-pack.sh
> +++ b/t/t5410-receive-pack.sh
> @@ -49,9 +49,13 @@ test_expect_success 'receive-pack missing objects fails connectivity check' '
> git clone --bare repo setup.git &&
> git -C repo commit --allow-empty -m 2 &&
>
> + write_script receive-pack-wrapper <<-EOF &&
> + tee "$(pwd)/out" | git-receive-pack "\$@"
> + EOF
> +
> # Capture git-send-pack(1) output sent to git-receive-pack(1).
> git -C repo send-pack ../setup.git --all \
> - --receive-pack="tee ${SQ}$(pwd)/out${SQ} | git-receive-pack" &&
> + --receive-pack="${SQ}$(pwd)${SQ}/receive-pack-wrapper" &&
>
> # Replay captured git-send-pack(1) output on new empty repository.
> git init --bare remote.git &&
> @@ -70,9 +74,13 @@ test_expect_success 'receive-pack missing objects bypasses connectivity check' '
> git clone --bare repo setup.git &&
> git -C repo commit --allow-empty -m 2 &&
>
> + write_script receive-pack-wrapper <<-EOF &&
> + tee "$(pwd)/out" | git-receive-pack "\$@"
> + EOF
> +
> # Capture git-send-pack(1) output sent to git-receive-pack(1).
> git -C repo send-pack ../setup.git --all \
> - --receive-pack="tee ${SQ}$(pwd)/out${SQ} | git-receive-pack" &&
> + --receive-pack="${SQ}$(pwd)${SQ}/receive-pack-wrapper" &&
>
> # Replay captured git-send-pack(1) output on new empty repository.
> git init --bare remote.git &&
>
