Fix nested workspace resolution

[Jake-Shadle]

This fixes a bug that was introduced in #10776 with nested workspaces.

As an example, say we have two workspaces:

/code/example/Cargo.toml and /code/example/sub/Cargo.toml, and a crate within the sub workspace /code/example/sub/test-crate/Cargo.toml.

Since the ws_roots is a HashMap with randomized ordering, this code will sometimes cause the workspace at /code/example/Cargo.toml to be discovered and used before /code/example/sub/Cargo.toml,

cargo/src/cargo/core/workspace.rs

Lines 1704 to 1710 in b1dd22e

	// Check if there are any workspace roots that have already been found that would work
	for (ws_root, ws_root_config) in config.ws_roots.borrow().iter() {
	if manifest_path.starts_with(ws_root) && !ws_root_config.is_excluded(manifest_path) {
	// Add `Cargo.toml` since ws_root is the root and not the file
	return Ok(Some(ws_root.join("Cargo.toml").clone()));
	}
	}

This will then cause the validate_members method to fail as the member thinks it is a member of a different workspace than it should be.

cargo/src/cargo/core/workspace.rs

Lines 874 to 891 in b1dd22e

	fn validate_members(&mut self) -> CargoResult<()> {
	for member in self.members.clone() {
	let root = self.find_root(&member)?;
	if root == self.root_manifest {
	continue;
	}
	match root {
	Some(root) => {
	bail!(
	"package `{}` is a member of the wrong workspace\n\
	expected: {}\n\
	actual: {}",
	member.display(),
	self.root_manifest.as_ref().unwrap().display(),
	root.display()
	);
	}

This change just makes it so that the input manifest path is walked up to find the (presumably) most appropriate workspace so that the ordering of the HashMap doesn't matter.

If you encounter this bug by running cargo nightly, you can workaround it by adding the crate(s) to the excluded field in the workspace they don't belong to.

[rust-highfive]

r? @ehuss

(rust-highfive has picked a reviewer for you, use r? to override)

[Muscraft]

I was worried about problems with nested workspaces when working on this, so thank you for finding the problem and posting a solution!

It looks like the performance is the same with these changes so all is good on that front!

[epage]

@bors r+

[bors]

📌 Commit fb56406 has been approved by epage

It is now in the queue for this repository.

[bors]

⌛ Testing commit fb56406 with merge 92ff479...

[bors]

☀️ Test successful - checks-actions
Approved by: epage
Pushing 92ff479 to master...

[ehuss]

Can someone please add a test for this regression? I assume all it needs is two workspaces, with the inner one having a single member? The test should likely go in https://github.com/rust-lang/cargo/blob/master/tests/testsuite/workspaces.rs. I would just run cargo tree in the inner member and make sure it returns the correct output without an error.