ignore timing out errors unless the timeout was specified to INFINITE #3
Conversation
The documentation states that for timeout failures: If a completion packet does not appear within the specified time, the function times out, returns FALSE, and sets *lpOverlapped to NULL. Unless the user specifiedwaiting for INFINITE, ignore timeout errors. https://docs.microsoft.com/en-us/windows/win32/api/ioapiset/nf-ioapiset-getqueuedcompletionstatus
|
Right, the failure alao happens on 1.0.5, and I'm Confused about it given it's saying the std (non group) implementation also fails. Which is new, it definitely passed when I published it. Perhaps DIE_WAIT needs incrementing for this test now for whatever reason. In any case: tests! I'm Also Confused because I figured that with my comparative tests of grouped/normal (non grouped) I was testing behaviour compatibility. So, uh, if I'm not in this case, I need a test to show this. |
|
Note, these unit tests pass when testing on machines with 32 cores as is. |
passcod
force-pushed
the
main
branch
5 times, most recently
from
382447b
to
d16879f
Compare
|
Alright, so I've fixed that test in CI. You can merge or rebase as you like. I still need a test written for this PR that fails without your changes and passes with. |
|
@passcod , I've merged main into my PR. |
|
Note, I believe reliably testing for this specific change will be difficult. It's highly dependent on Window's scheduler as to when an IO completion packet makes it to the queue. The previous behavior was more reliable on machines with large numbers of unused cores. |
|
Alright then, I'll trust you know what you're about. |
|
Released in 1.0.6 |
The documentation states that for timeout failures:
Unless the user specified waiting for INFINITE, ignore timeout errors.
https://docs.microsoft.com/en-us/windows/win32/api/ioapiset/nf-ioapiset-getqueuedcompletionstatus