matches! problem with edition 2021

[leonardo-m]

If I compile this:

fn main() {
    let _b = matches!(b'3', b'0' ..= b'9');
}

With:

rustc 1.53.0-nightly (b84932674 2021-04-21)
binary: rustc
commit-hash: b849326744a8eec939e592f0ab13bff85cc865d3
commit-date: 2021-04-21
host: x86_64-pc-windows-gnu
release: 1.53.0-nightly
LLVM version: 12.0.0

If I use the command:

...>rustc --edition 2018 -Z unstable-options test.rs

The compilation seems OK. If I compile with:

...>rustc --edition 2021 -Z unstable-options test.rs

rustc gives an error without line numbers:

error: `$pattern:pat` may be followed by `|`, which is not allowed for `pat` fragments
    |
    = note: allowed there are: `=>`, `,`, `=`, `if` or `in`

error: aborting due to previous error

[ehuss]

I believe this was caused by #83468 cc @hi-rustin

There seems to be some edition hygiene problem here. The matches macro in std should be 2018, and thus use the pat2015 behavior.

This makes it impossible to test 2021 on any project that uses the matches! macro.

[Rustin170506]

There seems to be some edition hygiene problem here. The matches macro in std should be 2018, and thus use the pat2015 behavior.

Yes, The pat2021 matcher will match new patterns, which include the | character. @nikomatsakis How do you think this can be fixed? Do we need to adjust matches! Or is there another way?

Do we need to adjust matches!

I suppose this should be fixed in the compiler by correcting the edition hygiene? I don't think it should be fixed by adjusting matches!, otherwise all macros from pre-2021-edition crates that use pat in this way will also need to be adjusted to be used in a 2021-edition crate.

@rustbot claim

[nikomatsakis]

Interesting! Let me poke a second into this.

[nikomatsakis]

@hyd-dev do you need any help or do you think you know what to do? Your diagnosis sounds correct to me.

Thanks @nikomatsakis! I don't know what to do yet and am digging into it now.

[Aaron1011]

This is caused by the fact that we currently map ExpnId::root() from a foreign crate to ExpnId::root() frrom the current crate during hygiene decoding. However the two root ExpnIds may have a different edition, so we end up using the current crate's edition instead of the foreign crate's edition.