implement item macros

[paulstansifer]

It's already possible to invoke macros in item position, but macro_rules currently only ever produces macros that mr_expr(...). It should be pretty easy to make it produce both kinds of macros. Perhaps (...) => (...) defines expr macros and (...) => {...} defines item macros?

Ideally, item macros should be able to produce multiple items. This would require changing fold.rs to handle ~[@item] instead of option<@item>, as it does now. This would require a bit more hacking; perhaps changing the AST to allow sequences of items in more places is an option (it should always be semantically valid, I believe).

[eholk]

What about having different macro defining forms for expression macros and item macros? I don't know that want the behavior to be different based on which delimiter I use. I could also see something like macro_rules! foo:expr { } and macro_rules! foo:items { }, although allowing those makes the macro syntax more flexible/subtle...

[eholk]

It'd be nice if an item macro could expand into multiple (including zero) items.

[nikomatsakis]

In particular, #[auto_serialize] as currently implemented requires
this. However, auto_serialize is likely being converted to use the
visitor when possible, so maybe it's not an important case. Also,
auto_serialize is this sort of weird decorator form that may never
become something more standard.

[graydon]

I agree, expanding to >1 item is probably useful; at the moment you can probably fake it by expanding to a module-full-of-items.

[nikomatsakis]

Two interesting use cases are ignore! and include!; both of which seem equally applicable to item or expr pos.

[paulstansifer]

I've been alerted to the fact that X macro-like techniques (see http://stackoverflow.com/questions/264269/what-is-a-good-reference-documenting-patterns-of-use-of-x-macros-in-c-or-possib) may require that we allow macros that can expand either to exprs or to items depending on where they are invoked. This should be possible. I previously didn't think there was any value to it, but I guess I was wrong.

[catamorphism]

This is done; open more specific issues if there are bugs.