Match arm trailing whitespace triggering error; match arm lone comma before comment unformattable

[RuijieYu]

Hello, I am encountering a combination of two issues.

$ rustc --version
rustc 1.63.0 (4b91a6ea7 2022-08-08)
$ cargo fmt --version
rustfmt 1.5.1-stable (4b91a6e 2022-08-08)

To my understanding there is no config file in effect.

After a bit of research, I think this issue is tangentially related toe #5535, as both are somewhat related to unformattable code due to potential loss of comments.

---

The issues are demonstrated in the following snippet. Note that the first match arm has trailing comma on a separate line, and that the Err(()) expression on the second match arm has a few trailing spaces.

Snippet 1

enum A {
    A1,
    A2,
}

struct B { field : bool, unused: () }

fn main() {}

fn f(a: A, b: B) -> Result<(), ()> {
    match (a,b ) {
        // arm 1
        (A::A1, B { field: true, ..}) =>
            Ok(println!("A1 + B=T"))
,
        // arm 2
        _ => Err(()),    
    }
}

In this form, cargo fmt --check generates the following error regarding the space after Err(()), causing my IDE to not format the code at all:

$ cargo fmt --check
Diff in /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs at line 3:
     A2,
 }

-struct B { field : bool, unused: () }
+struct B {
+    field: bool,
+    unused: (),
+}

 fn main() {}

error[internal]: left behind trailing whitespace
  --> /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs:20:20:22
   |
20 |         _ => Err(()),
   |                      ^^^^^^^^
   |

warning: rustfmt has failed to format. See previous 1 errors.

As suggested by #5535, but I don't quite understand which comment would be lost and why.

$ cargo fmt --check -- --config=error_on_unformatted=true
Diff in /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs at line 3:
     A2,
 }

-struct B { field : bool, unused: () }
+struct B {
+    field: bool,
+    unused: (),
+}

 fn main() {}

error[internal]: not formatted because a comment would be lost
  --> /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs:11
   |
11 |     match (a,b ) {
   |
   = note: set `error_on_unformatted = false` to suppress the warning against comments or string literals

error[internal]: left behind trailing whitespace
  --> /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs:20:20:22
   |
20 |         _ => Err(()),
   |                      ^^^^
   |

warning: rustfmt has failed to format. See previous 2 errors.

---

After I remove the trailing spaces after Err(()) from snippet 1, the following is the formatted snippet.

Snippet 2

enum A {
    A1,
    A2,
}

struct B {
    field: bool,
    unused: (),
}

fn main() {}

fn f(a: A, b: B) -> Result<(), ()> {
    match (a,b ) {
        // arm 1
        (A::A1, B { field: true, ..}) =>
            Ok(println!("A1 + B=T"))
,
        // arm 2
        _ => Err(()),
    }
}

Note that the comma stays unchanged, whereas I expect it to be at the right of Ok(..). Even if I add spaces before the comma, it is never formatted to join the Ok(..) line. Also note that (a,b ) is not formatted.

Now as I format with error_on_unformatted:

$ cargo fmt --check -- --config=error_on_unformatted=true
error[internal]: not formatted because a comment would be lost
  --> /home/user/.cache/cargo-temp/tmp-cM6RJ2/src/main.rs:14
   |
14 |     match (a,b ) {
   |
   = note: set `error_on_unformatted = false` to suppress the warning against comments or string literals

warning: rustfmt has failed to format. See previous 1 errors.

---

When I join the comma with the previous line from snippet 2, the following is the formatted snippet.

Snippet 3

enum A {
    A1,
    A2,
}

struct B {
    field: bool,
    unused: (),
}

fn main() {}

fn f(a: A, b: B) -> Result<(), ()> {
    match (a, b) {
        // arm 1
        (A::A1, B { field: true, .. }) => Ok(println!("A1 + B=T")),
        // arm 2
        _ => Err(()),
    }
}

Notice that now the comma and Ok(..) are now joined together.

---

When I instead remove the "arm 2" comment from snippet 2, the following is the formatted snippet.

Snippet 4

enum A {
    A1,
    A2,
}

struct B {
    field: bool,
    unused: (),
}

fn main() {}

fn f(a: A, b: B) -> Result<(), ()> {
    match (a, b) {
        // arm 1
        (A::A1, B { field: true, .. }) => Ok(println!("A1 + B=T")),
        _ => Err(()),
    }
}

In this case we see again that Ok(..) are joined with the comma.

---

Edit: after a bit more of digging around the issues, I think #5521 might be related to the first issue I have, the trailing whitespaces.

[ytmimi]

Thanks for the report.

I also appreciate you doing some digging to try and find related issues! I made a quick edit to add some trailing spaces to example 1, since those spaces weren't there for some reason.

Also, confirming that I'm able to reproduce the issue locally using snippet 1 and rustfmt 1.5.1-nightly (38659ec6 2022-08-18)

I have a feeling that both #5535 and #5521 are unrelated. You're experiencing similar problems, but they're happening in different contexts. In this case we're in the context of a match block, while those other issues dealt with let expressions and impl blocks respectively. It's not totally impossible for them to be related, but I think it's more likely this is related to #4119. Possibly a duplicate, but I'd need to do a little more digging to confirm that.