[clang] [llvm] [clang][coverage] Add opt-in call continuation counters for coverage (PR #201079)

[Alan Phipps via cfe-commits]

evodius96 wrote:

Sorry for the silence -- the summertime is usually busy, we we just had July 4th holiday in the US.

> Thanks for the suggestion. My impression was that this change is relatively self-contained with opt-in option and doesn't rise to the level of an RFC, but I'm happy to follow the maintainers' preference here. I'll defer to @evodius96 and @chapuni on whether they'd like to see an RFC first.

I agree in general that this change works best with an opt-in option, and so functionally nothing is imposed on folks who don't want to incur the overhead.  You've done a good job of characterizing that.  However, I'm concerned about the challenges posed above to extending clang's baked-in region-based counter infrastructure (via `RegionCounterMap`) and would like to open this up more to encourage broader buy-in and also to solicit suggestions.  So to that end, I agree with @justincady and @AaronBallman.  I'm fairly confident we can arrive at a workable solution.  I'd also like to get some suggestions from @chapuni since this impacts a lot of the work he extended.

It would be good if you could take your [existing discussion and proposal](https://discourse.llvm.org/t/is-source-based-coverage-on-llvm-clang-being-actively-maintained/90918) and port that into an RFC and invite some comment.  Referencing the existing PR, please summarizing clearly what the challenges are, and we'll go from there.

I apologize that this process can be tedious, but I think it would be worth it in the long-run.

https://github.com/llvm/llvm-project/pull/201079