test_handle_interrupted? may expose improper handle_interrupt logic

[headius]

The test test_handle_interrupted? may fail at test/mri/ruby/test_thread.rb:862 due to flaws in Thread#handle_interrupt logic. If the handling is allowing exceptions to be thrown that should be masked, it could trigger push of :ng1 earlier in the test, causing a sporadic failure as seen in this job: https://travis-ci.org/jruby/jruby/jobs/430135641

I am quarantining this test until we can determine if it is a flawed test or a bug in handle_interrupt.

FWIW I have stepped through handle_interrupt and been unable to reproduce the issue locally, so it may involve more complicated timing than can be simulated in a debug session.

[headius]

Other tests have been masked for this reason; it seems there's a bug in handle_interrupt that should be fixed before 9.2.1.

[larskanis]

The reason for ng1 in the test is probably that the interrupt mask is currently not inherited to new threads in JRuby and Truffleruby. In MRI it's done here.

Here is a minimal test case. In JRuby and Truffleruby this aborts the thread, in MRI this print :ok after one second:

Thread.handle_interrupt(RuntimeError=>:never){ Thread.new{ sleep 1; p :ok } }.raise

Hope this can be fixed in 9.2.1.

[headius]

@larskanis Ah, thanks for the pointer, that makes sense. Should be an easy one-liner...I'll give it a shot.

[headius]

I fixed the case that @larskanis explained, and test_handle_interrupted? does appear to pass consistently locally. The other three excluded tests still don't pass, though, and need further investigation.