advanced_slice_patterns has unreachable expression

[adjivas]

Hello,

According to this chapiter, I have writted this example:

#![feature(advanced_slice_patterns, slice_patterns)]


fn check(list: &[u8]) {
    match list {
        &[] => {},
        &[u1 @ b'\xF0' ... b'\xF4', u2 @ b'\x8F' ... b'\x90',
          u3 @ b'\x80' ... b'\xBF', u4 @ b'\x80' ... b'\xBF', ref next..] => {
            println!("{} {} {} {}", u1, u2, u3, u4);
            check(next)
        },
        &[u1, ref next..] => {
            println!("{}", u1);
            check(next)
        },
    }
}

fn main() {
    let buf = &[b'0', b'1', b'4'];
    
    check(&buf[..]);
}

But after the rustc 1.14.0-nightly version from 2016-10-25 -aef18be1b-,
I expected to see this happen:

error[E0001]: unreachable pattern
  --> <anon>:12:9
   |
12 |         &[u1 @ b'\xF0' ... b'\xF4',
   |         ^ this is an unreachable pattern

error: aborting due to previous error

[djzin]

Minimal example

#![feature(slice_patterns)]

fn check(list: &[u8]) {
    match list {
        &[] => {},
        &[u1, u2, ref next..] => {},
        &[u1] => {},
    }
}

Fails with

rustc 1.14.0-nightly (7c69b0d5a 2016-11-01)
error[E0001]: unreachable pattern
 --> <anon>:6:9
  |
6 |         &[u1, u2, ref next..] => {},
  |         ^^^^^^^^^^^^^^^^^^^^^ this is an unreachable pattern

error: aborting due to previous error

[arielb1]

regression introduced in #36695.

[TimNN]

@arielb1: How can this be a stable-nightly regression when this affects a gated feature?

[arielb1]

Not sure what's the policy about it. It's a regression.

[TimNN]

Yeah, I totally agree that this is a regression, although the label is inaccurate. I have wanted a regression-from-nightly-to-nightly label in the past (for eg. #36954), and think that once upon a time there even was such a label.