Strangely low coverage for the browser `unittest`

[Snuffleupagus]

Looking at the coverage data sorted by test, see https://app.codecov.io/gh/mozilla/pdf.js/flags, it seems that the unittest percentage is too low compared to what you'd expect:

• unittestcli runs in Node.js and the latest GitHub workflow reports 1291 specs, 0 failures, 35 pending specs, which gives a coverage percentage of 64.87%.

• unittest runs in browsers and the latest GitHub workflow runs 1336 tests per browser, since testing is done using both Firefox and Chrome, which gives a coverage percentage of only 38.91%.

Given that unittest includes more test-cases you'd expect it to have a higher coverage percentage, and not significantly lower than unittestcli.

---

Is it possible that the coverage reporting for unittest doesn't map the results correctly to the source files, since that may explain the low percentage?

/cc @calixteman, @timvandermeij

[calixteman]

Interesting, thanks for flagging this.

[timvandermeij]

I think, but I'm not 100% sure, it's because unittest also pulled the web/ folder in scope of the coverage report, a lot of which is uncovered because we don't have the integration/browser tests reporting coverage yet. The addition of that new folder probably pulls down the overall coverage for the target because effectively a larger chunk of code is uncovered. It could be that unittestcli skips the tests that actually use the web/ parts, and that's probably why web/ hasn't been visible in Codecov before but is now.

[calixteman]

First, I think we don't collect correctly the worker ccov.
Second, maybe there's something else which could cause a diff: in unittestcli case we use a build when in the unittest case we use src/.
Let me try to fix that.

[Snuffleupagus]

I think, but I'm not 100% sure, it's because unittest also pulled the web/ folder in scope of the coverage report, a lot of which is uncovered because we don't have the integration/browser tests reporting coverage yet.

After PR #21168 unittestcli also includes the web/ folder in the report, so that's probably not the explanation.

Additionally, if you compare https://app.codecov.io/gh/mozilla/pdf.js?flags%5B0%5D=unittestcli and https://app.codecov.io/gh/mozilla/pdf.js?flags%5B0%5D=unittest you can see that even the src/ folder coverage is significantly lower.