attempt to fix 32 bit architectures #54
Conversation
| finally: | ||
| sys.setswitchinterval(original_switch) |
There was a problem hiding this comment.
Will this just skip parallel execution if the switch interval value cannot be defined?
There was a problem hiding this comment.
The finally block is just to make sure we always restore the original switch interval no matter what happens.
The change as a whole makes sure the binding for the original switch interval is always valid. I'm not totally sure why it's sometimes invalid on 32 bit Python. I think the code as you had it originally should have worked.
There was a problem hiding this comment.
But if the setswitchinterval call emits a ValueError, wouldn't the parallel execution will be skipped? Shouldn't we also enclose that instruction in a try/except statement?
There was a problem hiding this comment.
I don't think a ValueError is possible now, since we're always getting a valid orig_interval before calling setswitchinterval.
If it somehow does happen there's another bug we're not considering and I'd rather fix that instead of papering over it by ignoring a valid exception.
There was a problem hiding this comment.
Sorry! I misread the issue completely, I thought the issue was with sys.setswitchinterval(0.000001)
2 participants
Fixes #53.
For reasons I don't understand, binding
orig_intervalin the outer function scope doesn't always work on 32 bit python.I also took the opportunity to make sure calling
sys.setswitchinterval(orig_interval)always happens after we set it to 10 microseconds in the pytest plugin using atry/finallyblock.I manually verified this works on a 32 bit linux docker image from manylinux. I don't particularly want to fight with github actions to figure out how to add 32 bit CI. If anyone wants to take that on, please feel free.