Closed ericstj closed 6 months ago
I tested this fix in https://github.com/dotnet/machinelearning/pull/6922 as well
Attention: 1 lines
in your changes are missing coverage. Please review.
Comparison is base (
efab011
) 68.79% compared to head (3ca267a
) 68.79%. Report is 3 commits behind head on main.
This seems to have fixed things, but now our failures are all timeouts. Hitting here and in https://github.com/dotnet/machinelearning/pull/6923.
@michaelgsharp didn't you mention wanting to address the timeouts by splitting up the test assemblies?
@ericstj Yeah, as soon as I finish the NER issue I'll test splitting the assemblies, so I'll have a test out for that later today.
Fixes #6921
Asserting on
_rowCount < Utils.Size(_valueBoundaries)
was catching a case where_rowCount
's update was reordered before_valueBoundaries
This was unnecessary, since this method doesn't need to use
_rowCount
.Instead, make the asserts use only
idx
which will be maintained consistent with the waiter logic in this cache. This will lock and synchronize with the main thread to ensureidx
is always within bounds.Ensure we only ever use
_rowCount
from the caching thread, so write reordering won't matter.