There seems to be some variability in the coverage results reported. See discussion in PR #324. A potential solution is changing the coverage testing to disallow lowering the coverage, not only to compare against a specified threshold.
Activity
mihaic commentedon Feb 2, 2018
The increase in coverage from PR #327 should get us over the threshold again, but it does not help with reliability.
Also, see this code comment in Coverage itself which indicates expected randomness:
https://bitbucket.org/ned/coveragepy/src/7b080c304b9219cdc33dd0b303db14d411c419fe/tests/test_concurrency.py?at=default&fileviewer=file-view-default#test_concurrency.py-362
dev: Add coverage variation to deal with flakiness
dev: Add coverage variation to deal with flakiness (#335)