Unconstrained region variables when making str-ptr-ptr hash and eq

[pcwalton]

Test case:

pure fn view_hash(s: & &str) -> uint {
    hash::hash_str(*s) as uint
}

pure fn view_eq(a: & &str, b: & &str) -> bool {
    str::eq_slice(*a, *b)
}

fn main() {
    let map = send_map::linear::linear_map(view_hash, view_eq);
    map.insert("foo", ());
}

Errors:

/home/pcwalton/Source/rust/build/test.rs:11:58: 11:65 error: Unconstrained region variable #3
/home/pcwalton/Source/rust/build/test.rs:11         let map = send_map::linear::linear_map(view_hash, view_eq);
                                                                                                      ^~~~~~~
/home/pcwalton/Source/rust/build/test.rs:11:58: 11:65 error: Unconstrained region variable #4
/home/pcwalton/Source/rust/build/test.rs:11         let map = send_map::linear::linear_map(view_hash, view_eq);

If the functions become inline lambdas, the problem goes away.

@nikomatsakis, any ideas?

[bblum]

I was fiddling around with regioned stack closures, trying to see if it'd ever mean something to return one, and ran into the same error for the first time:

fn foo(+blk: fn(+p: &a/fn() -> &a/fn())) {
    let mut state = 0;
    let statep = &mut state;
    do blk {
        || { *statep = 1; }
    }
}
fn main() {
    do foo |p| { p()() }
}


frames.rs:1:0: 1:0 error: Unconstrained region variable #8
frames.rs:1 fn foo(+blk: fn(+p: &a/fn() -> &a/fn())) {
            ^
frames.rs:1:0: 1:0 error: Unconstrained region variable #10
frames.rs:1 fn foo(+blk: fn(+p: &a/fn() -> &a/fn())) {
            ^

[catamorphism]

@pcwalton 's example no longer works since the linear_map function being used is now trait-based, rather than taking functions explicitly. I get a different error with @bblum's example:

/Users/tchevalier/rust2/src/test/run-pass/issue-3283.rs:4:11: 6:5 error: mismatched types: expected `&fn()` but found `@fn()` (expected & closure, found @ closure)
/Users/tchevalier/rust2/src/test/run-pass/issue-3283.rs:4     do blk {
/Users/tchevalier/rust2/src/test/run-pass/issue-3283.rs:5         || { *statep = 1; }
/Users/tchevalier/rust2/src/test/run-pass/issue-3283.rs:6     }

I don't know if that's right or not.

[catamorphism]

bblum's example compiles now (if you add a & before the fn type of blk), and pcwalton's example won't compile now for other reasons. So I'm considering this done. Reopen if you have a repro case

[catamorphism]

bblum's example compiled and ran just fine on my Mac, but segfaulted on the Linux and Windows bots. I don't have time right now to figure out why, so I'm checking it in xfailed.

[nikomatsakis]

I think this system should not type check. This is probably a bug in how we are handling region closures (I know there is work to be done there, not sure what bug #, let me fish it up).

[nikomatsakis]

Dup of issue #3696, I think.