This patch allows parsing of the syntax init [ v1, v2, v3, .. ] and certain level of type checking on HIR level.

See Draft RFC, under tracking rust-lang/lang-team#336

Dragons ahead

This is a draft PR that is very experimental and it is posted only for very early feedback and it is not immediately mergeable.

• This patch introduces changes to the parser to accept init # ... AST fragment as a start.
  The reason that this deviates from the original proposal, with the additional #, is to help with syntax ambiguity between init[0] as indexing into a value init, versus init [ 0 ] as in-place initialisation.
  However, I would propose to switch to a macro based approach to the syntax to avoid such ambiguity.
• This patch introduces a new type category Init. It has the same generics structure as a Closure and it is almost type-checked in the same way as Closure, except that its return type expectation also contains the error type, which is to be unified with those from the nested in-place initialisation.
  This is still not functional, because for instance I still have to fill out some tables such as late_bound_vars.

This patch does not include any THIR or MIR building.

The final state at which I would see this patch can be merged, is to at least we can type check the following smoke test.

#![feature(in_place_initialization)]
use std::init::Init;

fn smoke() -> impl Init<[i32; 1]> {
    init # [0]
}

fn main() {}

Areas to improve

As of July 7th, I see that I need to work these out

• The AST node InitTail is not necessary, I should just drop it
• The HIR node InitKind::Array could be squashed so that it is just a list of InitTail nodes
• The HIR node names. I don't find them descriptive. InitTail sounds terrible and I regret the choice.

r? @compiler-errors