Constant expression evaluation in statics is broken

[mahkoh]

static X: u64 = -1 as u16 as u64;
static Y: u64 = -1 as u32 as u64;

fn main() {
    match 1 {
        X => { },
        Y => { },
        _ => { },
    }
}

<anon>:7:9: 7:10 error: unreachable pattern [E0001] (pass `--explain E0001` to see a detailed explanation)
<anon>:7         Y => { },

   eddyb_ | http://is.gd/skhSd8
   eddyb_ | hmm
   eddyb_ | I wonder
  Gliptic | well, they have the same value in this case
arrrrrrr1 | Gliptic: do they?
   eddyb_ | ah, yes
   eddyb_ | arrrrrrr1: it's abug
   eddyb_ | probably filed a dozen times already
arrrrrrr1 | don't tell me it matches against -1
   eddyb_ | under different and seemingly unrelated descriptions
   eddyb_ | arrrrrrr1: yes it does
   eddyb_ | and it's very complicated to even get there
   eddyb_ | arrrrrrr1: jakub- has worked on that code ~recently
   eddyb_ | I don't know anyknow who knows everything about pattern handling
   eddyb_ | maybe nmatsakis, if you give him a few hours to read it all :P
   eddyb_ | http://is.gd/HDrH3f this is worse
   eddyb_ | and this works http://is.gd/aOHVkr
   eddyb_ | hmm
   eddyb_ | arrrrrrr1: I wonder if they are computed wrong
   eddyb_ | how can we check...
   eddyb_ | enum discriminants :D
   eddyb_ | "error: discriminant value already exists"
   eddyb_ | arrrrrrr1: okay, this might be simpler
   eddyb_ | const_eval is simply wrong
   eddyb_ | eugh
   eddyb_ | so what is this cast... u8 as u64?
   eddyb_ | arrrrrrr1: it ignores the size, lol
   eddyb_ | but, hmm
   eddyb_ | what is the order of operations
   eddyb_ | (-(1u8)) as u64
   eddyb_ | arrrrrrr1: lol it's so wrong
   eddyb_ | arrrrrrr1: it doesn't normalize the expression to its type
   eddyb_ | arrrrrrr1: so -(1u8) produces const_uint(-1u64)
   eddyb_ | same for !0u8
   eddyb_ | arrrrrrr1: and... casts ignore bit sizes so there's no truncate or zero/sign-extend
   eddyb_ | arrrrrrr1: rustc::middle::const_eval::eval_const_expr_partial is just wrong. if you want the location of the wrong code :P