Recompiles workspace crates every time

[Eugeny]

Currently, running cargo llvm-cov run recompiles the workspace crates every time, making merging multiple runs with --no-report and --no-run --html unfeasible.

[taiki-e]

Could you provide an example to reproduce this?

Also, old results will be removed unless you pass --no-report to cargo llvm-cov run.

cargo llvm-cov run --no-report -- a
cargo llvm-cov run --no-report -- b
cargo llvm-cov --no-run --html

[Eugeny]

I've got the repro ready and then tried putting --no-report after run, and it worked! Shame on me.

Might still be unexpected behaviour however, that:

• --no-report before run is silently ignored and
• --no-report affects whether a rebuild happens

[taiki-e]

• --no-report before run is silently ignored and

This is #151, and will be fixed by #197

• --no-report affects whether a rebuild happens

readme example describes cargo llvm-cov clean as a command to remove artifacts.

[Eugeny]

This is #151, and will be fixed by #197

👍

readme example describes cargo llvm-cov clean as a command to remove artifacts.

Sorry if I wasn't clear - I've already figured out how to do / not do a rebuild, but it's still unexpected for a new user that adding --no-report will also change the build behaviour (and not just whether a report is generated)

[taiki-e]

I've already figured out how to do / not do a rebuild, but it's still unexpected for a new user that adding --no-report will also change the build behaviour (and not just whether a report is generated)

Indeed. Adding a document to mention it seems like a good idea.

[fenollp]

I see that the workspace's bin crates are recompiled everytime as well. I can reproduce with the provided repro repo.
My usage:

while :; do cargo llvm-cov --all-targets --all-features --locked --frozen; done

# Note: plain cargo test call does not recompile things:
while :; do cargo test --all-targets --all-features --locked --frozen; done

Please advise.

[Eugeny]

@fenollp adding --no-report should help. In my cases, I've ended up running it once and then running the compiled binary multiple times directly during my integration tests.

[taiki-e]

Added --no-clean flag as a workaround for this. See #214 for more.