Clarify rules around Self and .Self

[josh11b]

A number of smaller changes grouped together in one proposal:

• Make Self a keyword.
• Clarify that Self refers to the current type in a base class and in impl declarations.
• Clarify when .Self is legal, and what type it has.
• Also specify that where is not an associative operator.

[zygoloid]

Looks good to me.

[zygoloid]

Getting a bit into the fine details here, what happens for:

fn F[T:! InterfaceA where .Self is 
             (InterfaceB where .B == .Self)](x: T);

? The second .Self arguably only means one thing here -- it resolves to T -- but that fact is somewhat obscured via the indirection through the first .Self.

[josh11b]

As written, I agree that would be allowed. Do you think that should change? We could instead have a rule about is expression in where clauses. What happens when we greatly expand the kinds of expressions in where clauses for templates?

I'm mostly looking for a rule with a concise and easy-to-remember statement.

[zygoloid]

I think we also need to think about implementability, for example for:

fn F[T:! Identity(InterfaceA where .T == .Self)]();

I'd prefer a rule that can be checked immediately and locally. I think the current rule would require an instance of .Self to track all the possible results until we reach some enclosing context where we can resolve them and see if they're the same. Maybe we can special-case a where expression as the top-level expression on the right of a :! and treat all other cases as being immediately ambiguous?

[chandlerc]

LG for me as well, so I think we have consensus among leads. (Kate is out sick at the moment.)