Interoperability with trybuild

[jhpratt]

Currently, the trybuild crate does not contribute anything to the coverage report. I'm not sure whether this should be fixed in cargo-llvm-cov or trybuild. compiletest_rs, on the other hand, does contribute to the coverage report (but is far more finicky).

[taiki-e]

I'm not sure what we can do about this on our side...

• IIRC, the code that trybuild generates for tests is located in "target/tests".
• The path to the object file will be collected in the object_files function.
• The object_files function does not currently search "target/tests/target".
• However, I tried a change a while ago to search for "target/tests/target", but the coverage results did not change.

cargo-llvm-cov/src/main.rs

Lines 142 to 170 in a04fcab

	fn object_files(cx: &Context) -> Result<Vec<OsString>> {
	let mut files = vec![];
	// To support testing binary crate like tests that use the CARGO_BIN_EXE
	// environment variable, pass all compiled executables.
	// This is not the ideal way, but the way unstable book says it is cannot support them.
	// https://doc.rust-lang.org/nightly/unstable-book/compiler-flags/instrument-coverage.html#tips-for-listing-the-binaries-automatically
	let mut build_dir = cx.target_dir.join(if cx.release { "release" } else { "debug" });
	if let Some(target) = &cx.target {
	build_dir.push(target);
	}
	fs::remove_dir_all(build_dir.join("incremental"))?;
	for f in WalkDir::new(build_dir.as_str()).into_iter().filter_map(Result::ok) {
	let f = f.path();
	if is_executable::is_executable(&f) {
	files.push(f.to_owned().into_os_string());
	}
	}
	if cx.doctests {
	for f in glob::glob(cx.doctests_dir.join("*/rust_out").as_str())?.filter_map(Result::ok) {
	if is_executable::is_executable(&f) {
	files.push(f.into_os_string());
	}
	}
	}
	// This sort is necessary to make the result of `llvm-cov show` match between macos and linux.
	files.sort_unstable();
	trace!(object_files = ?files);
	Ok(files)
	}

[jhpratt]

Hmm…that's odd. What does compiletest_rs do differently that it is covered?

[jhpratt]

Just coming back to this. I just looked through trybuild's code and see nothing that could conceivably be changed to help this. It (unfortunately) looks like this has to be handled on llvm-cov's end if my understanding is correct. As far as I can tell the bullet points you've listed are correct as they relate to trybuild.

[taiki-e]

I just found a way to fix this.

[taiki-e]

This will be fixed by #44 and dtolnay/trybuild#121.

[jhpratt]

You're a wizard, what can I say. You always find a way to do things, and generally quite quickly at that.

[davidhewitt]

🚀 thank you, really excited to see this!

[taiki-e]

trybuild 1.0.44 yanked due to regression: dtolnay/trybuild#122

[taiki-e]

FYI: 1.0.34 or older versions of trybuild provide similar behavior to 1.0.44, so you can work around this issue by pinning trybuild to those versions for now.

trybuild = "=1.0.34"