Properly cancel timeout on compilation error in ParallelCompiler #7428

[allansson]

The :timed_out message was not cancelled properly when compilation failed in ParallelCompiler. This meant the timeout message would be delivered even after compilation completed. This could be problematic when invoking compilation programmatically via e.g. Mix.Task.run("compile", ["--return-errors"]) since it means that the calling process, depending on the implemenation, might crash unexpectedly or get its inbox filled with messages.

The fix simply calls the cancel_waiting_timer when a compilation error has ocurred in the same way it is called in other similar cases.

This fixes issue #7428

[josevalim]

@allansson can you see if it is possible to check for this behaviour in a test? The test suite is here and we already have failing tests for the parallel compiler, maybe one of those have a leaked message that will be fixed with your test? It should be a matter of calling refute_received {:timed_out, _}. Thanks!

[allansson]

Of course! I did look at the tests and considered implementing one, but decided not to because there can be a considerable delay between compilation finishing and the timeout message being received. I didn't want to create a test which would have to run for a at least 10 seconds or so.

I was not aware of the refute_received assertion. It will make things really easy.

I'll try to fix at soon as possible. While I am at it, I suppose could write some tests for when compilation is successful as well. ^.^

[josevalim]

I see your point about the timeout. We could set it to one second but then we can have races. With all that said, I think this is good to merge as is. -- *José Valimwww.plataformatec.com.br <http://www.plataformatec.com.br/>Founder and Director of R&D*

[allansson]

Ok. I have almost completed writing two tests.

I can get around the long wait by setting the long_compilation_threshold option to something low. But that means that if compilation was to stall for too long for whatever reason, the tests would fail.

If it was possible to "spy" on calls to send_after and cancel_timer I could of course verify that all references are cleared up after execution has finished, but I am not aware of any way of doing so?

[josevalim]

❤️ 💚 💙 💛 💜