type state doesn't handle closures properly

[nikomatsakis]

In the following test case:

fn test(-x: uint) {}

fn main() {
    let i = 3u;
    uint::range(0u, 10u) {|_x| test(i)}
}

the variable i is de-initialized multiple times.

[catamorphism]

@nikomatsakis : I can see that the current behavior is wrong, but I'm not sure what the correct behavior is. Typestate was supposed to be intraprocedural, and the change to allow closures to deinitialize upvars got made while I was away, IIRC, so I don't know the rationale for it. It seems like handling this test case requires interprocedural reasoning. We have to take into account that the closure that's an argument to range deinitializes i, which is in the outer scope, and that's not really something that typestate does right now.

Or is the point simply that a closure shouldn't be allowed to call a function that deinitializes any of its arguments?

[nikomatsakis]

actually I think you are right the bug is the move from an upvaris not reported

[catamorphism]

All right, that'll be a lot easier to fix :-)

[catamorphism]

I think I've fixed the bug to forbid callee-moves from upvars. However, now iter::foldl and iter::foldr get rejected. In the first case, iter calls blk, which deinitializes its first argument, on b, which is defined in the outer scope.

What was the intent here?

[catamorphism]

ping @nikomatsakis -- I still have code sitting around to fix this, but foldl and foldr are still inconsistent with it :-)

[nikomatsakis]

I'll have to check. I have a rewrite of foldl (and the other iter functions) in a branch but I can't commit it yet because of some bug in the snapshot compiler. :( As far as I can tell, my rewrite doesn't fall afoul of these rules.

[nikomatsakis]

ping @catamorphism I think I fixed the relevant functions now.

[catamorphism]

@nikomatsakis Great, thanks -- I'll try to dust off my patch that fixes this.