Support collecting coverage from end-to-end testing

[mathstuf]

Describe the feature
We have a test that is run manually with lots of setup (podman or docker, another executable, and the project under test). Is it possible to have cargo tarpaulin collect coverage from this run if I make an out-of-crate integration test that does the std::process::Command setup to run the test?

[xd009642]

In theory yes, it should be more robust with the llvm engine instead of ptrace just because of occasional flakiness following spawned executables. One of the key things is that the process you run needs to be built with the right rust flags. So you can use cargo tarpaulin --print-rust-flags and if that's set in the environment propagating it to the process when you build it and it might work.

The in-theory/might caveats are generally because it can be hard to say exactly without trying it as there might be other complexities with the setup I'm not aware of so 🤞

[mathstuf]

I've had other thoughts on how to manage complicated coverage data (including collecting it per-test) which will also be able to handle these things. I'll report back once I flesh out the idea some more and see whether it can be done with existing mechanisms rather than rewriting/updating instrumentation everywhere.