Rustfmt: cargo fmt makes correct code incorrect when using pub with auto

[droundy]

This bug can be seen when formatting the dimensioned crate. Here is a screenshot:

cargo build
   Compiling dimensioned v0.6.0 (file:///home/droundy/src/dimensioned)
    Finished dev [unoptimized + debuginfo] target(s) in 6.06s
bennet:dimensioned$ cargo fmt
bennet:dimensioned$ cargo fmt
error: expected one of `!` or `::`, found `pub`
   --> /home/droundy/src/dimensioned/src/traits.rs:123:6
    |
123 | auto pub trait NotDim {}
    |      ^^^ expected one of `!` or `::` here

bennet:dimensioned$ git diff /home/droundy/src/dimensioned/src/traits.rs
diff --git a/src/traits.rs b/src/traits.rs
index 1464479..d5e94bb 100644
--- a/src/traits.rs
+++ b/src/traits.rs
@@ -120,7 +120,7 @@ pub trait Map<ValueOut>: Dimensionless {

 #[cfg(feature = "oibit")]
 /// Everything that is not a quantity implements this trait
-pub auto trait NotDim {}
+auto pub trait NotDim {}

 macro_rules! impl_unary {
     ($Type:ty, $Trait:ident, $fun:ident) => {

It looks like all it's doing to break the code is swapping the order of auto and pub. I have no knowledge of what auto does, as this isn't my code. But clearly formatting the code should not make it fail to compile.

[isaacazuelos]

I’d like to tackle this.

[nrc]

@isaacazuelos great, thanks! The best way to start is to look at items.rs. There you'll find functions for reformatting module-level items, including traits. If you look at how the 'signature' of a trait is formatted you should be able to see where the pub and auto come from and why the order is wrong. I think the fix should be easy. You'll also need to add a test, testing is described in Contributing.md. You'll need to find an appropriate source and target test file and add a trait like the one in the OP, then run cargo test to ensure it passes.

[nrc]

Let me know if anything is unclear

[isaacazuelos]

I'm not able to reproduce the issue. It looks like this is already tested too. I think this issue would happen at about items.rs:990.

Is it possible the issue is coming from an older version of rustfmt, prior to this commit?

[nrc]

Ah, that looks likely. Thanks for investigating @isaacazuelos

If you want another issue to look into, https://github.com/rust-lang-nursery/rustfmt/issues/2848 might be good if it still reproduces. If you're interested in performance, then I'm sure there are lots of possible optimisation wins possible and that would be awesome!

[max-sixty]

I confirmed by running cargo fmt on https://github.com/paholg/dimensioned. A couple of changes, but no crash. Can this be closed?