Different inference than rustc for integer

[jdupak]

fn test_cast() {
    let i = 1;
    let j = i as i64;
}

rustc says that i is i32 and we say it is i64.

[philberty]

Awh thats horrible .... :( i wonder can we get away with this one.

In theory it could be a simple fix though, by default we always append the reference chain for all unifications so in this case we are doing:

let i = 1; let j = i as i64; // > = <?I> as <?I64>

So i think the reference chain gets appended to the i64 and thus it gets infered that way... that's a nasty one