rustfmt makes it impossible to explain attributes with comments in the same line

[RalfJung]

Input

Sometimes an attribute needs a comment explaining why it exists:

#[cfg(not(miri))] // Miri does not deduplicate consts (https://github.com/rust-lang/miri/issues/131)

Output

Unfortunately rustfmt insists on putting the comment on a separate line (and it moves it after the attribute, which is even worse):

#[cfg(not(miri))]
// Miri does not deduplicate consts (https://github.com/rust-lang/miri/issues/131)

Expected output

Just like I can have comments to the right of code, explaining what the code does, I expect to have comments to the right of attributes, explaining what the attributes do:

#[cfg(not(miri))] // Miri does not deduplicate consts (https://github.com/rust-lang/miri/issues/131)

Meta

• rustfmt version: whatever ./x.py fmt does in rustc

[jyn514]

rustfmt version: whatever ./x.py fmt does in rustc

This is currently nightly-2020-11-19 (you can tell by looking at src/stage0.txt).

[RalfJung]

This looks very similar to #3255 -- in both cases rustfmt changes comment-after-code to move the comment onto its own line, thus changing the semantic meaning of the comment. IMO rustfmt should never do this.

[ytmimi]

I'm unable to reproduce this on rustfmt 1.4.38-nightly (2d9bc460 2022-04-07). Also It's not totally clear to me which PR resolved this. @calebcartwright any recollection of a PR that adjusted how missing comment spans were joined? without digging to deeply into it I'd assume that's where the fix was. I'll open a PR with tests cases for this.

[calebcartwright]

I don't recall anything offhand, though it is getting a bit late in the day. Could you make sure you're testing locally with the same configuration settings that are used in the rust-lang/rust repo?

[calebcartwright]

Closed as this does appear to be sorted, but please let us know if you see a recurrence @RalfJung