RFC: Remove bind

[nikomatsakis]

I think it's time for bind to go. My reasons:

• The language will have that many fewer keywords and syntactic constructs.
• The semantics are incompatible with closures. Right now, if I write foo(_, f()) this evaluates f() right away and binds the result into the closure. This is quite different from {|x| foo(x, f())} which is surprising. This has led to bugs.
• The type system is complex enough. Bind is a weird corner case we tend to forget about. We already found one such problem (move mode arguments) and I am confident there are more lurking. Let's simplify our lives and only think about closures, which are hard enough.

As a side effect of this change, naming a method like a.b without calling it will presumably become an error.

[graydon]

Agree.

[brson]

I agree, though I do like the syntax of bind better. Maybe in the future we can add something syntactically similar as an extension.

[nikomatsakis]

FWIW, I still like the change I proposed a while back where expressions
with "_" were converted into closures. The main objection was that the
extent of the created closure could be confusing (with which I am
sympathetic). I'd be in favor of a similar extension that addressed
this issue, whatever it is.

[catamorphism]

I agree. I do think bind syntax is occasionally convenient -- like currying in Haskell, except the captured arguments don't have to be a prefix of the non-captured arguments, so it's more flexible -- but I don't think that necessarily justifies the complexity and potential confusion.

[catamorphism]

Do we have consensus? Any objections from @pcwalton or @marijnh ?

[marijnh]

I'd like to see classical bind removed and Niko's old proposal for desugaring calls containing _ holes to closures.

[brson]

@Lenny222. Not yet. The value of methods can still be taken, so there is some bind machinery still there.

[msullivan]

I have a patch to get rid of our code that depends on taking the values of methods and a patch that rips out the rest of the bind code. Still need to reject taking values of methods in the typechecker...