Allow `foo.rs` to be parent to `foo/bar.rs`

[withoutboats]

Prior to this commit, in order for a module to have submodules, it had
to be at the /foo/mod.rs filepath. After this commit, modules at the
/foo.rs filepath have submodules located in the /foo/ directory.

[rust-highfive]

r? @pnkfelix

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

[withoutboats]

Not sure if this needs an RFC or not.

As written this doesn't work (and doesn't even get past stage 1), because #[path] attributes need to search paths from the directory the file is in, but this causes them to search in a subdirectory.

For example, in src/librustc_borrowck/borrowck/move_data.rs:

#[path="fragments.rs"]
mod fragments;

Needs to search for that file at src/librustc_borrowck/borrowck/fragments.rs, but instead on this branch src/librustc_borrowck/borrowck/mock_data/fragments.rs is searched.

I'll keep iterating on it though.

EDIT: Fixed this.

[withoutboats]

Not sure if this needs an RFC or not.

Talked with the lang team, we agreed that this needs an RFC. I'll write one.

[withoutboats]

Question for the RFC - Consider this code:

// src/lib.rs
#[path="bar.rs"]
mod foo;

// src/bar.rs (the `foo` module)
mod baz;

Should the baz module be searched for at src/foo/baz.rs or src/bar/baz.rs?

My inclination is to search at src/bar/baz.rs because if you chose #[path="bar/mod.rs"] that is where it would search.

[jmesmon]

On the paths: unless there is an existing example otherwise, it probably makes sense to keep them as being relative to the directory containing the file which contains the #[path="..."] attribute. Otherwise thinking about path gets harder than it really needs to be.

[KalitaAlexey]

I'd like to say against this feature because it adds complexity without any benefits.

[jmesmon]

I don't mind it so much: it could make moving from a single-file module (src/foo.rs) to a multi-file module (src/foo.rs + src/foo/blah.rs) not require a renaming of a file, which makes moving to multi-file modules easier (smaller diff & better vcs logs).

[withoutboats]

On the paths: unless there is an existing example otherwise, it probably makes sense to keep them as being relative to the directory containing the file which contains the #[path="..."] attribute. Otherwise thinking about path gets harder than it really needs to be.

The way the path attribute is parsed can't change. The question is whether path'd modules have submodules at the subdirectory matching their module name or their path name.

[jmesmon]

Ah yes, I see what you were getting at. In that case, your suggestion sounds reasonable.

[withoutboats]

After considering all the cases, what this implements (and the RFC will propose) is that a path attribute will contain submodules using these rules:

• foo/mod.rs will have submodules in foo/
• foo/bar.rs will have submodules in bar if "bar.rs" is valid unicode matching the regex .+\.rs
• Any other file name (not ending in .rs, not valid unicode, just .rs with no prefix) cannot have submodules, just as non-mod.rs modules cannot have today.

Again this only applies to modules at path attribute locations; I would be surprised if any Rust code in the wild has a module falling under the third category.

[nagisa]

Question for the RFC - Consider this code:

Take arbitrary stance in the RFC and pose this question in the RFC’s unresolved questions section.

[bors]

☔️ The latest upstream changes (presumably #39765) made this pull request unmergeable. Please resolve the merge conflicts.

[pnkfelix]

@withoutboats any update on the status of the RFC being drafted? Should I close this PR in the meantime?

[withoutboats]

Yea I'll close this for now. Hope to get back to the RFC in the next few weeks.