import_granularity: Don't normalize imports with comments

[davidlattimore]

Issue #4991

[ytmimi]

Thank for the PR!

Overall the code changes look good to me.

Would you mind adding some additional test cases with nested block comments. For example:

use a::{
    b,
    c,
    // Comment before a::d
    d,
    e, // Comment after a::e
};
use a::{/* before a::f::g? */ f::g, h::{i, /* after a::h::i? */ j}};
use a::{l::{self, m, n::o, p::*} /* after a::l::p::*? */};
use a::q::{self};
use a::q::{/* before */ self};
use a::q::{self /* after */};
// Comment before a::r
use a::r;

[davidlattimore]

I tried adding some more tests as described and ran into a few issues. In particular, some of the other import granularities, flatten then regroup imports. This ends up being lossy, especially if there are comments before / after a group. I'm starting to wonder if having any comments within a use statement should result in that use statement being left alone.

There's already code in normalize_use_trees_with_granularity that checks if a use-statement has a comment, and if it does, leaves it as-is, but it only checks for top-level comments, it doesn't check for comments within the use-statement. Perhaps I should just expand that. Thoughts?

[ytmimi]

@davidlattimore Thanks for continuing to look into this. I think we should prefer not to flatten use statements if it would result in rustfmt removing comments. I think airing on the side of caution is probably better here. It would be great if you could introduce that logic and add the relevant test cases 😁.

I wonder if following an approach outlined in #3999 would be useful. This was a PR that was merged into an unreleased branch, and also deals with dropping comments from use statements.

[davidlattimore]

I've updated so that use statements that contain comments won't be normalized.

[calebcartwright]

Just to summarize and ensure I understand correctly, is the consensus that trying to flatten/merge in the presence of comments has far too many complexities and scenarios for rustfmt to be able to format them well, and we're instead just going to bail in the presence of comments?

[davidlattimore]

Yep, that was my thinking. The one possible exception might be comments that were after an individual import and on the same line. e.g.

use a::b; // A comment
use a::c;

could be transformed into:

use a::{
   b, // A comment
   c,
};

And vice-versa.

I'm not sure though if it's worth that special case though. If an import is special enough that it deserves a comment, then perhaps it should be left separate from other imports.

All the other cases, e.g. comments before / after a group get really messy.

[calebcartwright]

I'm not sure though if it's worth that special case though. If an import is special enough that it deserves a comment, then perhaps it should be left separate from other imports.

All the other cases, e.g. comments before / after a group get really messy.

Makes sense, and agreed. Maintaining comments on flat reorders is challenging enough technically but straightforward enough logically. Whereas when changing the granularity rustfmt would be making under informed guesses in even many of the better cases, so this is best left to the developer writing the code since they can temporarily remove/reshuffle such comments and put them back wherever it makes sense in the new granularity post formatting

[calebcartwright]

lgtm, thank you!