Module Resolution Fixes

[ytmimi]

Fixes #7038

This mostly addresses #7038, but I've added a mod resolution test to more exhaustively check resolution with path attributes as mentioned in the reference, and I've updated the doc comments on SubModKind.

[purplesyringa]

Just to make sure, does this construct paths with the cartesian product? i.e.

#[cfg_attr(unix, path = "a")]
#[cfg_attr(not(unix), path = "b")]
mod outer1 {
    #[cfg_attr(unix, path = "c")]
    #[cfg_attr(not(unix), path = "d")]
    mod outer2 {
        #[cfg_attr(unix, path = "e.rs")]
        #[cfg_attr(not(unix), path = "f.rs")]
        mod inner;
    }
}

should format each of {a,b}/{c,d}/{e,f}.rs.

[ytmimi]

Just to make sure, does this construct paths with the cartesian product? i.e.

#[cfg_attr(unix, path = "a")]
#[cfg_attr(not(unix), path = "b")]
mod outer1 {
    #[cfg_attr(unix, path = "c")]
    #[cfg_attr(not(unix), path = "d")]
    mod outer2 {
        #[cfg_attr(unix, path = "e.rs")]
        #[cfg_attr(not(unix), path = "f.rs")]
        mod inner;
    }
}

should format each of {a,b}/{c,d}/{e,f}.rs.

@purplesyringa yes, as currently implemented this would construct paths for each of these:

• a/c/e.rs (all unix)
• a/c/f.rs
• a/d/e/rs
• a/d/f.rs
• b/c/e.rs
• b/c/f.rs
• b/d/e.rs
• b/df.rs (all not(unix))

Having written that all out it feels like only a/c/e.rs (all unix) and b/d/f.rs (all not(unix)) are valid since any other path can't be constructed because you can't have #[cfg_attr(unix)] and #[cfg_attr(not(unix))] at the same time.

[matthewhughes934]

The comments make this enum much clearer!

View changes since the review

[ytmimi]

Just to make sure, does this construct paths with the cartesian product? i.e.

#[cfg_attr(unix, path = "a")]

#[cfg_attr(not(unix), path = "b")]

mod outer1 {

#[cfg_attr(unix, path = "c")]

#[cfg_attr(not(unix), path = "d")]

mod outer2 {

    #[cfg_attr(unix, path = "e.rs")]

    #[cfg_attr(not(unix), path = "f.rs")]

    mod inner;

}

}

should format each of {a,b}/{c,d}/{e,f}.rs.

@purplesyringa yes, as currently implemented this would construct paths for each of these:

• a/c/e.rs (all unix)

• a/c/f.rs

• a/d/e/rs

• a/d/f.rs

• b/c/e.rs

• b/c/f.rs

• b/d/e.rs

• b/df.rs (all not(unix))

Having written that all out it feels like only a/c/e.rs (all unix) and b/d/f.rs (all not(unix)) are valid since any other path can't be constructed because you can't have #[cfg_attr(unix)] and #[cfg_attr(not(unix))] at the same time.

We should probably add a test case for this before merging.