How much importance is given to code coverage?

[cousteaulecommandant]

I've noticed that you have a codecov bot adding code coverage reports on PRs, and also that there is a make run-gcov utility in the repo source. However:

• The report shows that only about 85% of the code is covered, which isn't much.
• The codecov bot's report seems to skip certain code blocks. For example, in this report it doesn't highlight the body of isIdentifier() at all, despite of it being actually executed. If I run make run-gcov locally it does get covered.
• Conversely, some lines of the report I get locally are not covered, but are on codecov's report (for example, this entire block appears marked as ##### in my local report).
• Some lines that "should never happen" (Assert, VERBOSE...) are marked as uncovered (obviously). Maybe there should be an exception for those.
• Finally, the unit tests could be improved to cover more code. For example, this part about '{' is marked because there isn't any line in the Verilog unit tests that does something like {c, s} = x + y; (or simply {z} = x + y;).

What exactly is the purpose of this codecov bot and code coverage effort? Is it there only to ensure that new commits don't introduce untested features (which I suppose is useful on its own)? Or does the project plan to increase code coverage? If the latter, I could think of a few examples to improve the unit tests.

[masatake]

What exactly is the purpose of this codecov bot and code coverage effort? Is it there only to ensure that new commits don't introduce untested features (which I suppose is useful on its own)?

Yes. I, the head maintainer of the project, don't set a target percentage. If I can remember correctly, it was introduced to help contributors know what they did when making a pull request.
We don't have a plan. To be honest, I have my hands full, trying not to dead the project.

[masatake]

I don't mean to reject your contribution to improving coverage, but I prioritize updating parsers, especially the languages (and frameworks) I must read daily.

[cousteaulecommandant]

Thanks! It was just to know if the aim was to have everything covered or just detect potential regressions early (which seems to be the case). Just to see if it made sense to do a little "coverage hunting".