Add closures

[jackh726]

c.c. #368

This works but isn't rebased.

[jackh726]

Okay I just applied my changes in a new commit against master, since that was easier.

@nikomatsakis I would appreciate a review. The biggest "missing" part of this PR is tests. Not sure the best was to test closures currently.

[jackh726]

Downloaded lalrpop v0.19.0
error: failed to parse manifest at /home/runner/.cargo/registry/src/jackfan.us.kg-1ecc6299db9ec823/lalrpop-0.19.0/Cargo.toml

The heck. I'm just gonna restart CI.

[nikomatsakis]

This looks basically right. I left a few nits. I guess that for testing we'd have to have a way to declare closure types in the source, similar to our fn foo declarations. Not 100% sure what the obvious syntax for that is, maybe we should just do closure foo(...); declarations that are otherwise very much like fn declarations.

There was occasional talk of merging fns and closures in rustc, but I don't know of any concrete plans to do that. I guess that until we do so we can leave things as they are in chalk, though it does seem like you can imagine merging them and having a notion of some way to

(a) get the upvar type for a "callable" thing (for fn-def, returns (), for closures returns the tuple)
(b) get the fn-kind for a "callable" thing (for fn-def, returns Fn, for closures returned whatever is inferred)
(c) get the signature for the callable thing

Certainly would fit in with our general trend in chalk of trying to group together things that are "really quite similar", but I kind of think I'd rather land your PR as is (modulo changes I suggested) and then re-approach this later.

[jackh726]

So, testing closures with no upvars is fairly easy (like you said, a closure foo(...) syntax isn't terrible). The question is more of how would we test upvars?

We could do something like

closure foo[...]()

where the [...] would be the upvars. But even still, the question is how do we write the upvars? [Foo, &mut Bar]?

[nikomatsakis]

I was imagining

closure foo<...>(self|&self|&mut self, t0...tn) -> treturn {
   t_upvar1,
   t_upvar2,
   ...
}

where the self part captures fn/fn-mut/fn-once.

[nikomatsakis]

👍 so far

[jackh726]

This should be ready to review

[nikomatsakis]

This looks awesome! I left some comments, I'm a bit confused about the handling of generic type parameters.

[jackh726]

Ok @nikomatsakis I addressed your comments. I wasn't handling the generic types correctly. I also just moved from the single ClosureDatum struct to just functions on RustIrDatabase, since that seems to be the direction we want to go. Can you look over again? I wrote some doc comments, but they could use a second pair of eyes for clarity.

[jackh726]

Btw there might be a bug in here somewhere. Trying to update rustc to use this causes the closure tests to fail. So don't merge yet even if it looks good. Debugging...

[jackh726]

False alarm :) this is good for review