Build by PackageIdSpec, not name, to avoid ambiguity

[tedinski]

What does this PR try to resolve?

Fixes #11999

I previously added this code to ensure cargo install will build the intended package, but it turns out to cause ambiguity in the case where a package depends on prior versions of itself.

This ambiguity can be resolved by identifying the package to build by its pkg-spec, not name.

How should we test and review this PR?

A test case is included.

I have additionally manually tested that, with this patch, the issue in #11999 is fixed and now installs correctly.

[rustbot]

r? @ehuss

(rustbot has picked a reviewer for you, use r? to override)

[tedinski]

Ah, drat, I neglected to retry the complete test suite after this change. It would seem the full pkg spec does not work in some cases, for some reason.

[tedinski]

Ok, it broke for --git and this make a bit more sense a few lines down:

cargo/src/cargo/ops/cargo_install.rs

Lines 187 to 193 in de80432

	let pkg = if source_id.is_git() {
	// Don't use ws.current() in order to keep the package source as a git source so that
	// install tracking uses the correct source.
	pkg
	} else {
	ws.current()?.clone()
	};

So cargo install tracking is trying to retain the source git url, but for the build spec we need to unconditionally use ws.current() to correctly address the checked-out url (path) of the package we're trying to build within the workspace.

So just a little code juggling to fix.

[weihanglo]

Thanks for the fix. Could you add a test like path_install_workspace_root_despite_default_members but for git dependency? I feel like we missed that one.

[ehuss]

r? @weihanglo

[weihanglo]

@rustbot author

[tedinski]

I've added two more test cases: the one requested, and another slightly less recursive one, where the workspace contains both a package foo and a dependency on a registry package foo (which may just have colliding names, and not really be "recursive".)

[tedinski]

@rustbot ready

[weihanglo]

Thanks for the update! Just one question. Other parts look pretty great!

[weihanglo]

Pardon my correcting the term in a test name. And thank you so much for taking care of this issue!

@bors r+

[bors]

📌 Commit 7fb35c9 has been approved by weihanglo

It is now in the queue for this repository.

[bors]

⌛ Testing commit 7fb35c9 with merge a9465fe...

[bors]

☀️ Test successful - checks-actions
Approved by: weihanglo
Pushing a9465fe to master...