search

StanfordAHA / garnet

Next generation CGRA generator
BSD 3-Clause "New" or "Revised" License
106 stars 11 forks source link

Figure out why GC coverage decreases #102

Open rsetaluri opened 6 years ago

rsetaluri commented 6 years ago

See: https://coveralls.io/builds/18924947

The file where decreased coverage is reported is untouched. Unsure why it is reporting decreased coverage. Maybe some thing is stale.

leonardt commented 6 years ago

So looking at https://github.com/StanfordAHA/garnet/pull/101, I see 

Change from base Build 305: | -0.1%

For Build 305, if we look at https://coveralls.io/builds/18922701/source?filename=global_controller/global_controller.py#L124 and Build 310 (the one referenced in the top comment), https://coveralls.io/builds/18924947/source?filename=global_controller/global_controller.py#L124

We see that line 124 isn't covered. Is there some non-determinism in the test case that could cause this line to be occasionally uncovered? E.g. maybe there's a case when no mask bit is set.

@alexcarsello do you know how that mask is generated in the test case?

rsetaluri commented 6 years ago

I notice a lot of random's in that the GC tests. @alexcarsello maybe you can do a pass and make it pseudo-constrained-random (maybe just hard code some special cases) to ensure coverage? Not very urgent. In the meantime we can add #nocover to them.

alexcarsello commented 6 years ago

Yes there is some non-determinism there, so it's possible that line 124 could occasionally be uncovered. I'll fix this.