ICE: rust: task failed at 'Assertion fcx.locals.contains_key(nid) failed',

[dgryski]

This is with a build of master. I'm testing with now a build of incoming (but my poor laptop is taking a while to get through the build of llvm...)

MacBook-Pro-de-Damian-Gryski-2:siphash dgryski$ cat siphash-ice.rs
use std;

iface siphash {
    fn result() -> u64;
    fn reset();
}

fn siphash(k0 : u64, k1 : u64) -> siphash {
    type sipstate = {
        mut v0 : u64,
        mut v1 : u64,
    };

    fn mk_result(st : sipstate) -> u64 {

        let v0 = st.v0,
            v1 = st.v1;
        ret v0 ^ v1;
    }

   impl of siphash for sipstate {
        fn reset() {
            self.v0 = k0 ^ 0x736f6d6570736575;
            self.v1 = k1 ^ 0x646f72616e646f6d;
        }
        fn result() -> u64 { ret mk_result(self); }
    }
}
MacBook-Pro-de-Damian-Gryski-2:siphash dgryski$ RUST_LOG=rustc=0,::rt::backtrace rustc --test siphash-ice.rs
rust: task failed at 'Assertion fcx.locals.contains_key(nid) failed', /Users/dgryski/work/src/cvs/rust/src/rustc/middle/typeck/check.rs:1952
error: internal compiler error: unexpected failure
note: the compiler hit an unexpected failure path. this is a bug
note: try running with RUST_LOG=rustc=0,::rt::backtrace to get further details and report the results to github.com/mozilla/rust/issues
rust: task failed at 'explicit failure', /Users/dgryski/work/src/cvs/rust/src/rustc/driver/rustc.rs:256
rust: domain main @0x103800010 root task failed
rust: task failed at 'killed', /Users/dgryski/work/src/cvs/rust/src/libcore/task.rs:523

[catamorphism]

I'll take a look at this.

[catamorphism]

FYI, I reproduced this with incoming. I'll investigate further.

[catamorphism]

A more minimal example that pokes the same bug:

use std;

iface siphash {
    fn reset();
}

fn siphash(k0 : u64) -> siphash {
    type sipstate = {
        mut v0 : u64,
    };

   impl of siphash for sipstate {
        fn reset() {
           self.v0 = k0 ^ 0x736f6d6570736575;
        }
    }
    fail;
}

fn main() {}

If I comment out the body of the reset method, it works. It seems that methods in nested impls can't refer to upvars, but I'll keep looking.

[nikomatsakis]

impls, like other items, are certainly not supposed to refer to upvars. I think this constraint is enforced in resolve for nested fns---seems like resolve ought to enforce a similar constraint for impls etc.

[catamorphism]

Ok, I'll look at why resolve isn't doing that, then.

[catamorphism]

I understand what the bug is: resolve was resolving impl methods with a NormalRibKind rather than an OpaqueFunctionRibKind. Fixing that fixed the test case, but then I realized the behavior should be the same for classes and traits (which is currently isn't). That's a problem because the obvious fix stops class methods from being able to refer to their class-bound ty params.

Resolve is doing something weird with def_ids for def_ty_params... so I'm still working on this.

[graydon]

Fwiw siphash itself arrived in libcore yesterday.

[catamorphism]

Fixed in a0d0584