On Mon, Jan 06, 2025 at 12:16:51PM +0100, Patrick Steinhardt wrote:
> Fix the issue by making the writer fill the pipe buffer before we
> execute git-submodule(1). Ideally, it would be git-submodule(1) itself
> that does produce all that data, but it would require us to create a
> large amount of submodules, which is inefficient. Instead, we use Perl
> to print gibberish until the buffer is filled.
>
> To verify that this works as expected one can apply the following patch
> to the preimage of this commit, which used to reliably trigger the race:
>
> diff --git a/t/t7422-submodule-output.sh b/t/t7422-submodule-output.sh
> index f21e920367..9338c75626 100755
> --- a/t/t7422-submodule-output.sh
> +++ b/t/t7422-submodule-output.sh
> @@ -168,7 +168,7 @@ done
>
> test_expect_success !MINGW 'git submodule status --recursive propagates SIGPIPE' '
> { git submodule status --recursive 2>err; echo $?>status; } |
> - grep -q X/S &&
> + { sleep 1 && grep -q X/S; } &&
> test_must_be_empty err &&
> test_match_signal 13 "$(cat status)"
> '
>
> With the pipe-stuffing workaround the test runs successfully.
Sadly this isn't enough. The pipe-stuffing solves the race with grep
_starting_ (and thus the extra "sleep"), but the fundamental race we've
seen in practice still remains. See my reply the v1 thread for details.
-Peff
