field: Type /* Comment */ , /* Comment */ causes rustfmt to introduce breaking comma

[hlbarber]

While strange,

struct Struct {
    field: String /* Comment */ , /* Comment */
}

compiles.

When hit with cargo fmt it's transformed into

struct Struct {
    field: String, /* Comment */ , /* Comment */
}

which fails to compile.

Observed on rustfmt 1.5.1-nightly (2022-06-30 7425fb2) and rustfmt 1.4.38-stable (db9d1b2 2022-01-20).

[ytmimi]

Thanks for the report.

Confirming that I can reproduce this with the latest master rustfmt 1.5.1-nightly (2ae63f00 2022-06-30).

[ytmimi]

The following test case illustrates that this is only an issue with the last item in a list.

input

struct Struct {
    field1: String /* Comment */ , /* Comment */
    filed2: String /* Comment */ , /* Comment */
}

output

struct Struct {
    field1: String, /* Comment */
    /* Comment */
    filed2: String, /* Comment */ , /* Comment */
}

Also running the original input with trailing_comma=Never also prevents the issue since the struct is returned as is. You can run rustfmt --config=trailing_comma=Never to verify.

struct Struct {
    field: String /* Comment */ , /* Comment */
}

For reference, this is where we handle post comments in lists:

rustfmt/src/lists.rs

Lines 404 to 504 in 2ae63f0

	// Post-comments
	if tactic == DefinitiveListTactic::Horizontal && item.post_comment.is_some() {
	let comment = item.post_comment.as_ref().unwrap();
	let formatted_comment = rewrite_comment(
	comment,
	true,
	Shape::legacy(formatting.shape.width, Indent::empty()),
	formatting.config,
	)?;
	result.push(' ');
	result.push_str(&formatted_comment);
	}
	if separate && sep_place.is_back() {
	result.push_str(formatting.separator);
	}
	if tactic != DefinitiveListTactic::Horizontal && item.post_comment.is_some() {
	let comment = item.post_comment.as_ref().unwrap();
	let overhead = last_line_width(&result) + first_line_width(comment.trim());
	let rewrite_post_comment = |item_max_width: &mut Option<usize>| {
	if item_max_width.is_none() && !last && !inner_item.contains('\n') {
	*item_max_width = Some(max_width_of_item_with_post_comment(
	&cloned_items,
	i,
	overhead,
	formatting.config.max_width(),
	));
	}
	let overhead = if starts_with_newline(comment) {
	0
	} else if let Some(max_width) = *item_max_width {
	max_width + 2
	} else {
	// 1 = space between item and comment.
	item_last_line_width + 1
	};
	let width = formatting.shape.width.checked_sub(overhead).unwrap_or(1);
	let offset = formatting.shape.indent + overhead;
	let comment_shape = Shape::legacy(width, offset);
	let block_style = if !formatting.ends_with_newline && last {
	true
	} else if starts_with_newline(comment) {
	false
	} else {
	comment.trim().contains('\n') || comment.trim().len() > width
	};
	rewrite_comment(
	comment.trim_start(),
	block_style,
	comment_shape,
	formatting.config,
	)
	};
	let mut formatted_comment = rewrite_post_comment(&mut item_max_width)?;
	if !starts_with_newline(comment) {
	if formatting.align_comments {
	let mut comment_alignment =
	post_comment_alignment(item_max_width, inner_item.len());
	if first_line_width(&formatted_comment)
	+ last_line_width(&result)
	+ comment_alignment
	+ 1
	> formatting.config.max_width()
	{
	item_max_width = None;
	formatted_comment = rewrite_post_comment(&mut item_max_width)?;
	comment_alignment =
	post_comment_alignment(item_max_width, inner_item.len());
	}
	for _ in 0..=comment_alignment {
	result.push(' ');
	}
	}
	// An additional space for the missing trailing separator (or
	// if we skipped alignment above).
	if !formatting.align_comments
	|| (last
	&& item_max_width.is_some()
	&& !separate
	&& !formatting.separator.is_empty())
	{
	result.push(' ');
	}
	} else {
	result.push('\n');
	result.push_str(indent_str);
	}
	if formatted_comment.contains('\n') {
	item_max_width = None;
	}
	result.push_str(&formatted_comment);
	} else {
	item_max_width = None;
	}

The issue can be found in lists::write_list. Specifically on lines 418-420. The condition evaluates to true and we end up writing a trailing separator. Then we enter the block starting on line 423 where we format and write the post comment, which in this case contains an uncommented ,. This leads to parse errors on future runs because the compiler gets two , tokens back to back.

[hlbarber]

@ytmimi Great info! Thank you