Variables bound with different modes in patterns

[Stebalien]

The following doesn't compile:

enum Test<'a> {
    A(&'a u64),
    B(u64),
}
fn foo(test: Test) {
    match test {
        Test::A(r) | Test::B(ref r) => println!("{}", r)
    }
}
fn main() {
    foo(Test::A(&0));
    foo(Test::B(1));
}

failing with the following error:

test.rs:7:30: 7:35 error: variable `r` is bound with different mode in pattern #2 than in pattern #1
test.rs:7         Test::A(r) | Test::B(ref r) => println!("{}", r)
                                       ^~~~~
error: aborting due to previous error

However, the following does work:

enum Test<'a> {
    A(&'a u64),
    B(u64),
}
fn foo(test: Test) {
    match test {
        Test::A(&ref r) | Test::B(ref r) => println!("{}", r)
    }
}
fn main() {
    foo(Test::A(&0));
    foo(Test::B(1));
}

Is there any reason rust can't just perform this reborrow automatically? &ref is a very cryptic pattern.

[ghost]

Is there any reason we put in this restriction in the first place in a6a5c48? Seems like the type equality restriction on bindings should be enough.

cc @nikomatsakis

[kdeeee]

this error still occur in rustc 1.11.0 (9b21dcd 2016-08-15)

[Mark-Simulacrum]

@nikomatsakis Is there a chance the &ref pattern is actually bypassing the match checking that is intended to not allow binding both by-move and by-reference in the same arm? It seems to be, so... perhaps this is a problem in a wider sense?

[nikomatsakis]

@Mark-Simulacrum

Is there a chance the &ref pattern is actually bypassing the match checking that is intended to not allow binding both by-move and by-reference in the same arm?

I don't understand what you mean here -- we have two bindings, both are "by reference", so how is the current code mixing by-move and by-ref in the same pattern?

[nikomatsakis]

I think that in principle we do not need to require all bindings to be in the same mode; in that sense I agree with @ghost. However, I'm not sure how hard it would be to keep current code "up and going" without that simplifying restriction. I'd have to dig in in more detail. I think that once we are doing borrowck on MIR, things ought to be simpler, since most of the rules we have to keep borrowck + match sound could go away.

[Mark-Simulacrum]

Oh, I think I misread about the two bindings being by-move and by-reference -- never mind.

I agree that we should probably leave this for after MIR borrowck (which might in its implementation resolve these sorts of issues on its own, I don't know).

[steveklabnik]

Triage: MIR borrowcheck has now landed, but this still gives this error.

[Dylan-DPC]

Current error:

error[E0409]: variable `r` is bound inconsistently across alternatives separated by `|`
 --> src/main.rs:7:34
  |
7 |         Test::A(r) | Test::B(ref r) => println!("{}", r)
  |                 - first binding  ^ bound in different ways

For more information about this error, try `rustc --explain E0409`