[WIP] Allow macros in identifier position

[bgamari]

This allows macros to be used in identifier position of items, making concat_idents! far more useful. This is very much a work in progress.

[eddyb]

cc me @sfackler

[huonw]

This probably needs an RFC.

[bgamari]

@huonw I can try putting together an RFC if necessary. Is it the grammar changes that are concerning?

[huonw]

And the interactions with hygiene.

[alexcrichton]

Closing, this change is definitely one that warrants an RFC. This looks pretty awesome, however, and I would love to see this implemented!

Some thoughts that may affect the writeup of the RFC:

• How does this interact with hygiene? What scope are the identifiers created bound to, can they reference local variables for example?
• How does this interact with our grammar? Are there ambiguities allowing macros as idents? If we one day allow for invoke macros via paths, this may require infinite lookahead to determine if an ident is a macro invocation (when used as an expression). This may not be super-relevant, however (just something to think about).

There may be other concerns, but I'm sure an RFC would flesh them out!

[vadimcn]

@bgamari, @alexcrichton: So why was this PR abandoned? Were there any technical problems with it, or just the absence of an RFC?

[alexcrichton]

Ah sorry, my memory is a little hazy on this, but I believe it would definitely need to start out with an RFC today.

[eddyb]

There was an RFC and it was rejected. Though it doesn't point to this PR specifically (which is technically superior in that it doesn't increase the size of Ident or make it non-Copy).

It seems a bit odd to me that it was, because we have since added pattern macros and trait/impl interior macros (i.e. macros that can expand to methods or associated items).

Making ident macros unstable to use should avoid any de facto problematic design stabilization.
I proposed foo.(concat_idents!(bar, baz)) for method calls, which is far from optimal, but we have UFCS now, so maybe we don't need this to work with method call syntax.

[SiegeLord]

I don't quite understand how UFCS resolves the desired syntax sugar from my comment in that RFC (if it's meant to). I specifically want a sugaring that looks like this: foo.bar!(a = 5) and desugars into foo.baz(5) (or the UFCS equivalent). Deviations into bar!(foo, a = 5) are unacceptable, and foo.(bar!(a = 5)) are unfortunate.

[eddyb]

@SiegeLord I don't see how that's relevant, you seem to want something which doesn't expand to a single identifier. I am talking about identifier macros in method name position, which caused some issues during the implementation of this, though I don't see them mentioned anywhere now.

[eddyb]

Oh, I think this was is about conflicts between ident macros and hypothetical method macros (which you seem to be referring to). In which case, what I meant about UFCS is that it removes the requirement for the former (you wouldn't use ident macros for methods outside of macros yourself, I wouldn't think), leaving the obvious syntax for the latter.

[vadimcn]

So shall we reopen the RFC (and this PR)? Not being able to create new idents in macros is annoying.