Regression in 1.9.11 -> 1.10.1 : line coverage stats >100%

Dependabot wants to update my project from 1.9.11 to 1.10.1, but the PR is failing as line coverage on the first subproject has dropped from 100% to 98% according to the build output.

When I try to recreate the failure locally the generated report shows an overall line coverage stat of 106%, and even 133% in one class. Predictably enough, the >100% coverage class reports show no uncovered lines and no obvious weirdness. I think my tests are pretty good, but they’re not that good.

Hopefully others can recreate this result by:

1. git clone https://github.com/Mastercard/flow.git && cd flow
2. mkdir -p api/target && touch api/target/mutate
3. mvn test-compile

Cheers!

[hcoles]

Fortunately looks to be a simple issue. Lines are being double counted when generating the threshold if they’re included in the bytecode more than once. The compiler would do this for finally blocks, and perhaps in some other scenarios.

It also seems that the old html report code is calculating this stuff seperately, which seems less than ideal.

Will poke around for a bit to see if anything else looks broken, then produce a release shortly.

[hcoles]

@therealryan Ok, looks like something was badly messed up with the thresholds during the 1.9 -> 1.10 refactor, which is a little embarassing. Thanks again for reporting, hopefully this was stupidity confined to a single place rather than something wide spread.