closure type fails to be inferred properly for vec foldl method

[thestinger]

This snippet works:

fn combine_flags(flags: &[MagicFlag]) -> c_int {
   vec::foldl(0, flags, |a: c_int, b: &MagicFlag| a | (*b as c_int))
}

Changing to the following doesn't compile because it gets inferred as an @ closure:

fn combine_flags(flags: &[MagicFlag]) -> c_int {
    flags.foldl(0, |a: c_int, b: &MagicFlag| a | (*b as c_int))
}

Error:

magic.rs:62:15: 62:61 error: mismatched types: expected `&fn(&core::libc::types::os::arch::c95::c_int, &MagicFlag) -> core::libc::types::os::arch::c95::c_int` but found `&fn(core::libc::types::os::arch::c95::c_int, &MagicFlag) -> core::libc::types::os::arch::c95::c_int` (expected &-ptr but found core::libc::types::os::arch::c95::c_int)
magic.rs:62   flags.foldl(0, |a: c_int, b: &MagicFlag| a | (*b as c_int))
                           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
error: aborting due to previous error
zsh: exit 101   rustc magic.rs --test -O

The full code is in this repo.

[huonw]

It looks like the problem is actually the first argument of the closure: it should be of type &c_int in the second example.

Confusingly, the foldl method (from iter) and the function in vec have subtly different types for the closure (the standalone function in iter has the same type as the method).

[nikomatsakis]

Not critical for 0.6; removing milestone

[nikomatsakis]

The vec function has the correct signature. The iter signature is weaker
(it does not pass the accumulator by value but rather by reference). This
is partly because, in the generic iter case, you cannot safely write the
vec signature. The compiler isn't smart enough to permit the moves (it fears
recursion). One option might be to pass in &mut, or else have iter::foldl use
a bit of unsafe code. But anyway this sounds like a separate issue.

[catamorphism]

It doesn't look like there's a bug here. Reopen if I'm wrong.