Mangling: initial attempt

[Daveed Vandevoorde]

Martin von Loewis wrote:
[...]
> It seems that would work. However, I feel it is inconsistent to use
> one way of qualifying for nested names, and another way for the
> toplevel name. Instead, I'd say that, instead of
> 
>   <prefix><length><name>
> 
> you have
> 
>   <prefix><identifier>
> 
> and <identifier> is either
> 
>   <length><name>
> or
>   Q{<length><name>}*E
> 
> So instead of _11fQ1N, you get _1Q1N1fE. So method name would be
> mangled just as a class name.

I like that.  It would remove a case where something that appears
early in the demangled representation (the qualifier) is encoded
later (here: at the end).

(I think the only other case of this is return types; should we
change that?  My inclination is "no".)

Unless there are objections, I'll go that way.  Actually, I think
we can use the N...E segment for that and do away with Q altogether:

	_1N1N1fE --> Ret? N::f()

	Daveed