Nightly Dotty workflow of 2023-12-18 failed

[github-actions]

See https://github.com/lampepfl/dotty/actions/runs/7243042276

[nicolasstucki]

Error:  Test dotty.tools.dotc.CompilationTests.parallelBackend failed: java.util.concurrent.TimeoutException: Compiling targets timed out, remaining targets: out/parallelBackend/reference, took 1200.005 sec
Error:      at dotty.tools.vulpix.ParallelTesting$Test.executeTestSuite(ParallelTesting.scala:707)

[nicolasstucki]

This was caused by 10f2c10.

[nicolasstucki]

I seems that 10f2c10 was not the cause after all. See #19289 (comment).

[nicolasstucki]

This issue started with 1f7aa24. We need to clean the project to reproduce the issue.

[nicolasstucki]

This issue is spurious. I am not sure if 1f7aa24 started it anymore. I will need to test this in depth.

[nicolasstucki]

Executing git bisect using

sbt> clean; testOnly dotty.tools.dotc.CompilationTests -- *parallelBackend; testOnly dotty.tools.dotc.CompilationTests -- *parallelBackend; testOnly dotty.tools.dotc.CompilationTests -- *parallelBackend

showed with more confidence that the issue comes from 2dcd64a

[odersky]

I have spurious parallel backend failures in my local runs all the time (maybe one in three runs fails). I suspect a race condition in parallel backend. What is the status of it? Do we run that in production?

[nicolasstucki]

I have spurious parallel backend failures in my local runs all the time (maybe one in three runs fails). I suspect a race condition in parallel backend. What is the status of it? Do we run that in production?

#19290 fixes the issue on the CI. However, we should try to figure out why accessing the symbol of an enum there fails with a race condition.

[odersky]

I can't see how such a commit could make any difference if parallel backend would not have some underlying race problems.

[nicolasstucki]

I assume it is only triggering the issue much more often.