fmt gives up on some pub const fields in types

[fulara]

using rustfmt 1.4.38-nightly (777bb86 2022-01-20)

Example code:

use std::mem;

struct X;

fn foo() {}

impl X {
    pub const SOMETHING: usize = mem::size_of::<i64>() // field A
        + mem::size_of::<i64>() // field B
        + mem::size_of::<i64>() // field C
+ mem::size_of::<i64>() // field D
        + mem::size_of::<i64>() // field E
        + mem::size_of::<i64>() // field F
        + mem::size_of::<i64>() // field G
        + mem::size_of::<i64>(); // field H
}

fn main() {
    println!("Hello, world! X {}", X::SOMETHING);
}

rustfmt fails to format above for field D.
However if I try to put the: mem::size_of::<i64>() // field A on the next line rustfmt does put it back in there.
it also formats foo() in this function as expected so I am invoking it correctly.

In my other tests it sort of put the comment for field H in a totally odd location I had rustfmt give me something like this:
(unable to produce MVE for this, but I did not try hard )

...
-        + mem::size_of::<i64>(); // some_field_c
+        + mem::size_of::<i64>();
+
+    // some_field_d

[ytmimi]

Thanks for submitting the report!

I think the trailing comment not sticking with the impl item is partially related to #4027, #5168, #3720. Those linked issues deal with mod and use statements, and go down a slightly different code path, but the general issue is the same. rustfmt tries to reorder these items while retaining the trailing comments, but it's a little harder to get the comment for the last item in the group.

When you get a chance, could you please provide an example without comments to better highlight the issue you're describing.

[calebcartwright]

This is a duplicate of #5099 so going to close accordingly

[fulara]

hey @calebcartwright well if you think this is a duplicate of that one it probably is.
@ytmimi here is a complete example showcasing seemingly both issue.

1. rustfmt gives up on formatting the things
2. the comments after the comma sticks to the next element.

It seems to me that comments here are the causing factor, when I was testing it without comments it seems to work.

use std::mem;

struct X;

impl X {
    pub const SOMETHING: usize = mem::size_of::<i64>() // field A
        + mem::size_of::<i64>() // field B
        + mem::size_of::<i64>() // field C
        + mem::size_of::<i64>() // field D
+ mem::size_of::<i64>() // field E        
        + mem::size_of::<i64>() // field F
        + mem::size_of::<i64>() // field G
        + mem::size_of::<i64>(); // field H

    pub const OTHER: usize = 3;
}

fn main() {
    println!("Hello, world! X {}", X::SOMETHING);
}

Results in...

use std::mem;

struct X;

impl X {
    // field H

    pub const OTHER: usize = 3;
    pub const SOMETHING: usize = mem::size_of::<i64>() // field A
        + mem::size_of::<i64>() // field B
        + mem::size_of::<i64>() // field C
        + mem::size_of::<i64>() // field D
+ mem::size_of::<i64>() // field E
        + mem::size_of::<i64>() // field F
        + mem::size_of::<i64>() // field G
        + mem::size_of::<i64>();
}

fn main() {
    println!("Hello, world! X {}", X::SOMETHING);
}

I was not able reproduce the comment behavior earlier because it turns out it's related to my config:

reorder_impl_items = true

It's the reorder_impl_items that causes this odd stickyness of the comments.