CI is running with threads?

[vtjnash]

Taking a look at a CI crash, I noticed it was initializing an extra thread, which isn't really supported by our testsuite, and so may be causing the recent crashes. I don't know how this computation suddenly got corrupted:

$ ./julia -E 'Threads.maxthreadid()'
2

[oscardssmith]

Is this that we are now defaulting to 1 regular and 1 interactive thread?

[vtjnash]

Are we? That sounds like a bug

[DilumAluthge]

I believe it is intentional behavior: #57087

[vtjnash]

Do we need to revert that entirely or just some part of it?

[DilumAluthge]

Can we instead fix our test suite to work correctly in the case of 1 regular thread, 1 interactive thread?

[oscardssmith]

Agreed with Dilum. Julia has multithreading as a core feature. If it's broken, we need to fix the bugs.

[IanButterworth]

and so may be causing the recent crashes

Specifics would help here.

[IanButterworth]

Also, the approach in #57087 was to not special-case tests, but to make them robust to threading state.

I'm aware of one open flaky issue #57198

[DilumAluthge]

Also, for the test suite, we probably don't need to support the full general case. It might be sufficient to only support the following two special cases:

• 1 regular thread, 0 interactive threads.
• 1 regular thread, 1 interactive thread.