https://pantsbuild.org/ logo
#development
Title
# development
h

hundreds-breakfast-49010

08/17/2020, 9:07 AM
I've been looking at https://github.com/pantsbuild/pants/issues/10129 for a while now trying to debug this issue so we can un-revert that commit.
I've replicated the issue stu saw by adding a deliberate source of nondeterminstic failure to a pants test, namely adding
from random import randint; assert 0 = randint(1, 2000)
so that the test will always fail but in a different way each time. I can see that every time I change the exact number in the randint line, I see one additional instance of the MultiPlatformProcess corresponding to the pytest runner execute
i.e. I see
[INFO] Completed: - Run Pytest for src/python/pants/build_graph/address_test.py:tests
in an effort to debug this I added a debug print
log::warn!("Calling complete(), node: {:?}", entry.node());
to the
complete()
method in
graph/src/lib.rs