Bug#1079983: src:lua-luv: autopkgtest regression on arm64: Segmentation fault

[Paul Gevers]

Hi James,

Nothing changed in the package itself since the last time it was
consistently succeeding, so that implies something else broke it and
wasn't caught before it migrated to testing.

Correct. This can happen if the breakage is caused by something lower in
the stack. For migration testing we only look at direct reverse (test)
dependencies. The system isn't perfect.

I wasn't able to reproduce this on amdahl. I attempted to reproduce it
locally via qemu and also couldn't reproduce it there.

If you want we can arrange access to a testbed where the test just ran
and failed for you to inspect and "play with". Unfortunately I need to
be on-line in the current scheme, so we'd need to align our availability.

Paul

[James McCoy]

Hi,

I wasn't able to reproduce this on amdahl. I attempted to reproduce it
locally via qemu and also couldn't reproduce it there.

It seems it's flaky, so might be a race condition. On 2024-08-29 I scheduled
10 runs on our infra (5 in unstable and 5 in testing) and they all failed.
Today I did the same, and 1 passed. So that's 19/20 failures.

Indeed, I've hit it locally now and I see that it's just waiting an
arbitrary time for an asynchronous callback to get called. I've filed an
issue [upstream] to make this more deterministic, but it looks like
increasing the timeout will reduce the flakiness.

[upstream]: https://github.com/luvit/luv/issues/715

Cheers,