Don't infer 100% coverage on partial results

Code coverage is incorrectly assumed to be 100% on partial builds.

We have a mono repo with multiple subdirectories as projects which we do not build and test if there are no changes to save CI time. We upload coverage using the -F <topic> parameter to segment out the various subdirectories.

On pull requests where a <topic> is not built the Github coverage numbers show as 100%. While it would be great if it showed the percentage from the branch base for this <topic>, 100% is certainly not correct.

I have uploaded a screen shot of the report on a build that only touched the “goservices” topic. You can see that the “android” topic is reported as 100% even though no coverage results were pushed for this topic.


@nickpalmer saw this expressed in “Feature Ideas” too, definitely a feature for us / how we will expect product work in future.

This is certainly related to the partial coverage case, but it is weird that it reports 100% instead of not reporting at all so I filed it as a bug instead. Looking forward to “carry forward” resolving this either way.