Struct constness/mutability needs to be represented in metadata

[bblum]

Whether or not a struct is const is forgotten across crate boundaries. The compiler should reject this program:

const.rs:

use nonconst;
import nonconst::nonconst;

fn x<T: const>(_x: T) { }

fn main() { x(nonconst()) }

nonconst.rc:

#[crate_type = "lib"];

nonconst.rs:

struct nonconst { mut z: (); }
fn nonconst() -> nonconst { nonconst { mut z: () } }

If you change the type nonconst to be e.g. an old-style record type, the program is correctly rejected.

[catamorphism]

Seems like this bug even manifests with non-cross-crate structs:

struct S<T: Const> {
    s: T,
    mut cant_nest: ()
}

fn main() {
    let a1  = ~S{ s: true, cant_nest: () };
    let _a2 = ~S{ s: move a1, cant_nest: () };
}

This seems like it shouldn't compile, as ~S doesn't have kind Const (it contains a mutable field). Yet it does compile (as of 59a034a ).

[emberian]

@bblum is this still an issue?

[bblum]

No, it isn't. The test has bitrotted slightly and should now be:

extern mod extra;
use extra::arc::RWARC;

fn main() {
    let arc1  = ~RWARC(true);
    let _arc2 = ~RWARC(arc1);
}

Anyone who updates it should close.

[bblum]

Actually, yes it is. The cant-nest-rw-arcs test seems to work fine, but the OP test case is still broken, not even cross-crate. It correctly falis to compile if the struct has a mutable thing inside (@mut), but not if the only thing making it nonconst is the #[mutable] tag. The problem is likely with ty::type_contents.

I don't know why the mutable tag works for RWARC but not for this.

#[mutable]
struct Nonconst { x: () }
pub fn nonconst() -> Nonconst { Nonconst { x: () } } 

fn x<T: Freeze>(_x: T) { }

fn main() { x((nonconst())) }

[thestinger]

The #[mutable] attribute got renamed to #[no_freeze] so that's the issue. It would be really nice if we only accepted valid attributes...

[bblum]

DAMMIT