RFC: enforce module directory structure more strictly

[chris-morgan]

No description provided.

[pnkfelix]

@nikomatsakis has a more general idea on how to fix the common problems here. He should probably weigh in (or Niko: if you prefer, I can write an RFC for your proposal).

[steveklabnik]

ahhahahahah :D

[nikomatsakis]

@pnkfelix this is basically the "opinionated" version of what I wanted to do.

The less opinionated version would support either the mod.rs or else a pattern where the main module is "foo.rs" and submodules are "foo/bar.rs"

foo.rs
foo/bar.rs

The lenient and strict version are basically the same, except for the behaviour when the current file name is not mod.rs. In that case, the lenient version would, given a declaration mod foo, we would look for current/foo.rs (where current is the name of the current module).

[o11c]

More choices in things like this are bad.

[pcwalton]

This seems reasonable to me at first glance.

[alexcrichton]

I also think this seems like a reasonable solution to a common problem.

[pnkfelix]

i prefer @nikomatsakis 's generalized form, but I would not object to accepting this RFC, especially since I believe it to be forward-compatible with niko's generalization.

[nikomatsakis]

@chris-morgan I took the liberty of updating the title to be more precise and formal. Feel free to update it again if you don't think it correctly reflects your proposal.

[chris-morgan]

@nikomatsakis the rename is fine. I wasn't enthralled with the name it had!

As for the alternative of allowing foo.rs to load foo/bar.rs, that would indeed be backwards compatible, and I could live with such a thing. I do believe that enforcing more strict conventions is a good thing, however, so I would prefer this to not head in that direction.

It's much easier for a tool to correct style if there is one, and only one, way of doing something.

[bstrie]

+1 to more restrictions to make the module system clearer.

[emberian]

Strong +1

[alexcrichton]

We discussed this at today's meeting, and have decided to merge this.

[pnkfelix]

Just a quick note of something I've noticed since attempting to adopt the stricter style imposed here: it can yield many open buffers that are all named mod.rs.

A good editor/IDE will distinguish them in some way, e.g by using the directory that they fall in (so that the buffers are named things like mod.rs<foo> and mod.rs<bar> for foo/mod.rs and bar/mod.rs).

Nonetheless, the slightly looser semantic that Niko was proposing (which we could still add on top of this, I believe) would enable me to avoid ever using the same filename, which can ease switching between open buffers in some editors.

[nikomatsakis]

@pnkfelix yeah, this has been mentioned a few times. I can't believe that there are editors out there that fail to make the distinction between files with the same name in distinct directories, but apparently it is true. (Hello, people, index.html?) Anyway, that is certainly one legit reason that someone might prefer foo.rs and foo/bar.rs vs foo/mod.rs and foo/bar.rs. It is interesting that node.js supports both of these styles (they call it index.js, not mod.js), but apparently foo.rs and foo/bar.rs is considered preferred? I'm not sure if that's actually true, since I'm not a big node user. Certainly, for me at least, foo/mod.rs has proven to be much more compatible with how I think mentally: the foo directory has everything related to foo.

[chris-morgan]

Another possibility is foo/foo.rs rather than foo/mod.rs.