Intermittent Failure in Queue Test

[insertinterestingnamehere]

As noted in #323, there is an extremely intermittent failure in the queue test. Thus far this has only been observed in the sherwood/binders config. The assertion at

qthreads/test/basics/queue.c

Line 105 in 7f2f008

test_check(qthread_readstate(NODE_BUSYNESS) == 1);

fails. When testing locally this only happens once every few thousand runs of that test. Thus far it's been observed on x86 OSX and x86 Linux. In both cases with the clang compiler.

While switching to the CMake build in CI is what unearthed this, I think that's likely not the actual cause. It seems like we're hitting a case where there's some undefined behavior giving the optimizer more latitude than it ought to have somewhere. Given that this is in a rarely used configuration and that the test failure is for a minor consistency check in a test for a portion of our API that's unused downstream I'm marking this as low priority and moving on for now.

[insertinterestingnamehere]

Just saw this fail on an ARM OSX instance using gcc as well, so it's not specific to any architecture or compiler. OSX seems to be much more susceptible to it though. Monitoring.

[insertinterestingnamehere]

Just saw this in an nvc/arm/sherwood/no topology config too. Increasing the priority since it's clearly happening far more frequently in CI than it was locally.

[insertinterestingnamehere]

Looks like the OSX/gcc falure was actually apple clang in disguise, so this may still be restricted to LLVM-based compilers.

[insertinterestingnamehere]

Fixed in #332