[1.38] Regression: proc-macro derive panicked

[petrochenkov]

#63628 (comment)

proc-macro derive panicked

[INFO] [stderr] error: proc-macro derive panicked
[INFO] [stderr]   --> src/lib.rs:11:1
[INFO] [stderr]    |
[INFO] [stderr] 11 | / proc_macro_item_decl! {
[INFO] [stderr] 12 | |     /// Generates a default blanket impl for a generic trait that gets passed into
[INFO] [stderr] 13 | |     /// `autoimpl!`, with the same type parameters and type bounds as its trait.
[INFO] [stderr] 14 | |     ///
[INFO] [stderr] ...  |
[INFO] [stderr] 44 | |     autoimpl! => generate_auto_impl_impl
[INFO] [stderr] 45 | | }
[INFO] [stderr]    | |_^
[INFO] [stderr]    |
[INFO] [stderr]    = help: message: assertion failed: `(left == right)`
[INFO] [stderr]              left: `Some("#[allow(unused,")`,
[INFO] [stderr]             right: `Some("#[allow(unused")`

All of the 62 regressions are due to old versions of proc-macro-hack having an assert expecting #[allow(unused, instead of #[allow(unused ,.
So this is again a consequence of whitespace changes in pretty-printing (#62667).

I think we should be able to fix this by special casing the comma token during token stream pretty-printing and not printing the space before it.

[petrochenkov]

Fixed in #63897.

[nikomatsakis]

Just FYI, we discussed this in the Sep 05 @rust-lang/lang meeting. We felt like the action here was appropriate -- while in general we do not guarantee stability of pretty printing, it makes sense to be judicious, and in this particular case not having a leading space before , just looks better anyway. 😛