Research and implement tarpaulin instead of grcov

[greg-szabo]

Not a done deal, but according to Tony, tarpaulin has less false negatives.
https://github.com/xd009642/tarpaulin

[liamsi]

related: #250

[thanethomson]

I'm assuming this will be superseded by #811?

[thanethomson]

Since our code coverage seems broken (and seems like it has been broken for some time), we should consider fixing it or integrating tarpaulin-based coverage calculation sooner rather than later.

@romac are you planning on taking #811 further? What are the pros/cons at this point of going with cargo-llvm-cov? I see it's still labelled as being "experimental" right now?

[romac]

are you planning on taking #811 further?

I wish I could but I don't have the bandwidth at the moment.

What are the pros/cons at this point of going with cargo-llvm-cov?

Pros:

• the coverage should be much more precise
  Cons:
• at the moment I was not able to get codecov.io to understand it's output on anything other than toy crates
• -Zinstrument-coverage is still pretty new so there might be issues

I see it's still labelled as being "experimental" right now?

The wrapper itself is marked as experimental, but it basically only wraps cargo build (adding nightly-only options), llvm-profdata, and llvm-cov.

[romac]

By the way, we use tarpaulin on ibc-rs but have seen no real improvements (eg. the coverage still jumps around by 30% sometimes, etc) hence why I was looking into source-based code coverage.

[thanethomson]

Currently superseded by #840.

We can reopen this issue again in future if we want to reconsider using tarpaulin.