Can we make istanbul the default coverage provider over v8?

[ahayes91]

Reason I ask is that I came across this issue today in our codebase, where code coverage wasn't accurate for a new piece of code with an implicit else condition:
#2239

Nestled deep inside the description of that ticket, this is actually a known issue with v8, but it was fixed for users of istanbul:

The coverage report using @vitest/coverage-c8 reports this as 100% branch coverage, due to a known issue with v8 (bcoe/c8#227 and this v8 ticket).

This is a pretty glaring bug in terms of coverage reporting IMO - for new (or lazy 😉 ) engineers, having automated checks and thresholds on code coverage is the best reminder to make sure you test all the branching logic in your code properly. Or at least - you're made aware of the ones you're not testing and you have to qualify that for some reason.

https://vitest.dev/guide/coverage#coverage-providers could be updated to mention the bug at least (maybe a pros/cons of why you should use a certain provider).

[AriPerkkio]

We are not planning to change the default value. Users can still change the provider when needed.

There's on-going work to make the output of V8 coverage match Istanbul's accuracy in Vitest.

[ahayes91]

Fair enough! In the meantime though, it would be great to bring this issue (and any other known gotchas between the two) to the forefronts of the documentation to help folks out. It took me a long time to figure out that the issue was only fixed in istanbul and not in v8 from looking through that thread 😅

[AriPerkkio]

Reason I ask is that I came across this issue today in our codebase, where code coverage wasn't accurate for a new piece of code with an implicit else condition:

This will be fixed by:

• feat(coverage): v8 experimental AST-aware remapping #7736

Here's comparison of V8 coverage with AST based remapping vs @vitest/coverage-istanbul: