Regression on array pattern-matching

[renato-zannon]

Until friday (commit 3047614) the following (very contrived) code compiled:

fn main() {
    println!("{}", count_members(&[1, 2, 3, 4]));
}

fn count_members(v: &[uint]) -> uint {
    match v {
        []         => 0,
        [_]        => 1,
        [_x, ..xs] => 1 + count_members(xs)
    }
}

With ca3e557, it fails with:

array_match.rs:9:5: 9:15 error: unreachable pattern
array_match.rs:9     [_x, ..xs] => 1 + count_members(xs)
                     ^~~~~~~~~~
error: aborting due to previous error

I'll try to bisect and get to a single commit

[alexcrichton]

cc @jakub-, I think one of your recent patches fixed this, right?

@alexcrichton Yeah but this looks different... looking!

[renato-zannon]

My bisect ended up pointing f5e513b as the first bad commit. This should help narrowing down the problem, as it has a relatively small diff

[bstrie]

I just noticed this as well. A more minimal test case:

fn main() {
    match &[1,2,3] {
        [] => (),
        [_] => (),
        [_, ..] => ()  // error: unreachable pattern
    }
}

Notably, the error vanishes if you remove the one-element pattern:

fn main() {
    match &[1,2,3] {
        [] => (),
        [_, ..] => ()  // works fine
    }
}

This is fixed now.

[alexcrichton]

Thanks @jakub-!