Fix: Make path dependencies with the same name stays locked

[linyihai]

What does this PR try to resolve?

Fixes: #13405

This is a workround based on #13405 (comment)

How should we test and review this PR?

first commit will pass, second commit fixed it and update test.

Additional information

[rustbot]

r? @ehuss

rustbot has assigned @ehuss.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[linyihai]

@rustbot ready

[linyihai]

@ehuss any chance for some feedback? I'm not sure I'm doing the right thing.

[ehuss]

I'm still a little concerned about the behavior here when there are multiple versions, but none of them match. There are two concerns:

• This randomly chooses one of the packages, which is not great for consistency.
• Depending on what changed and which random package it picks, you can end up with two packages with the same SourceId.

I'm still wondering if it is appropriate to return None here if there are multiple versions, but none of them match. For example, something like:

Suggested change

	}
	}
	return None;

Does that make sense?

[ehuss]

Here is a test that illustrates the problem with having multiple versions:

#[cargo_test]
fn same_name_version_changed() {
    // Illustrates having two path packages with the same name, but different versions.
    // Verifies it works correctly when one of the versions is changed.
    let p = project()
        .file(
            "Cargo.toml",
            r#"
                [package]
                name = "foo"
                version = "1.0.0"
                edition = "2021"

                [dependencies]
                foo2 = { path = "foo2", package = "foo" }
            "#,
        )
        .file("src/lib.rs", "")
        .file(
            "foo2/Cargo.toml",
            r#"
                [package]
                name = "foo"
                version = "2.0.0"
                edition = "2021"
            "#,
        )
        .file("foo2/src/lib.rs", "")
        .build();

    p.cargo("tree")
        .with_stderr("[LOCKING] 2 packages to latest compatible versions")
        .with_stdout(
            "\
foo v1.0.0 ([ROOT]/foo)
└── foo v2.0.0 ([ROOT]/foo/foo2)
",
        )
        .run();

    p.change_file(
        "foo2/Cargo.toml",
        r#"
            [package]
            name = "foo"
            version = "2.0.1"
            edition = "2021"
        "#,
    );
    p.cargo("tree")
        .with_stderr(
            "\
[LOCKING] 1 package to latest compatible version
[ADDING] foo v2.0.1 ([ROOT]/foo/foo2)
",
        )
        .with_stdout(
            "\
foo v1.0.0 ([ROOT]/foo)
└── foo v2.0.1 ([ROOT]/foo/foo2)
",
        )
        .run();
}

With this PR as-is, this test will randomly fail.

I'd recommend adding this test to the path.rs testsuite module.

[linyihai]

Oh on, the ci faild due to

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    bench::many_similar_names

test result: FAILED. 3326 passed; 1 failed; 7 ignored; 0 measured; 0 filtered out; finished in 390.01s

error: test failed, to rerun pass `-p cargo --test testsuite`

I guess this pr merge can solve this

[ehuss]

Thanks! I moved the suggested test to path.rs since it is unrelated to patching. I'm still not 100% confident with that final None return, but I can't think of any real alternatives.

@bors r+

[bors]

📌 Commit ab92717 has been approved by ehuss

It is now in the queue for this repository.

[bors]

⌛ Testing commit ab92717 with merge 62f1a51...

[bors]

💔 Test failed - checks-actions

[ehuss]

@bors retry

[bors]

⌛ Testing commit ab92717 with merge 198ba31...

[bors]

☀️ Test successful - checks-actions
Approved by: ehuss
Pushing 198ba31 to master...