Fix handling of modules in non_modrs_mods style

[flodiebold]

We need to keep track of the module relative to which we're resolving paths,
instead of always passing None.

Fixes #2673.

WIP because I still need to add a test; I'm making the PR already because I wanted to make sure this is the right approach.

[flodiebold]

Added a test; I couldn't fit it into the existing test patterns because if I put the files e.g. into target, it will try to format foo.rs alone and fail to parse it (which I think is legitimate, but maybe not?)

[nrc]

Added a test; I couldn't fit it into the existing test patterns because if I put the files e.g. into target, it will try to format foo.rs alone and fail to parse it (which I think is legitimate, but maybe not?)

I think that should work, maybe we need to update rustc_ap_syntax?

[nrc]

maybe we need to update rustc_ap_syntax?

Just did this, could you try a regular test now please?

[nrc]

Thanks for the PR! Code looks good, it would be nice to have a regular test, rather than a special case one, or at least properly understand why the regular test doesn't work.

[flodiebold]

I rebased, but it still doesn't work. I get

error[E0583]: file not found for module `bar`
 --> tests/target/issue-2673-nonmodrs-mods/foo.rs:1:5
  |
1 | mod bar;
  |     ^^^
  |
  = help: name the file either bar.rs or bar/mod.rs inside the directory "tests/target/issue-2673-nonmodrs-mods"

when it tries to format foo.rs by itself, because the parser doesn't know to look for bar in foo/bar.rs. I've tried guessing the directory ownership from the file name (so when parsing a file foo.rs as the root, it treats it as a module foo), but that results in lots of other tests failing, because e.g. in tests/target/configs/reorder_modules/false.rs, it looks for the module lorem in false/lorem.rs instead of lorem/mod.rs.

I could make this a normal test if there were some way of telling the test runner not to format foo.rs by itself.

[topecongiro]

@flodiebold There is skip_children config option which lets rustfmt ignore any submodules. I hope this could help to solve the issue.

[flodiebold]

@topecongiro Ah, of course, why didn't I think of that :)

@nrc I've turned the test into an idempotence test now.

[nrc]

Looks good, thanks!