namespaces -- time to do something?

["Marc W. Mengel" <mengel at fnal dot gov>]

Well, now that egcs-1.0 is out, it seems to me it's time to do something about
namespaces for egcs-1.1.  Towards that end, I have a simple idea that I think 
will work which I'd like to toss out for folks to knock around.

Premise:
--------
Unless I am missing something, it looks like egcs *declares* namespaces 
okay, creates symbols in namespaces properly, and does fully-qualified names 
correctly, it just doesn't do "using" correctly.   

Proposal:
---------
If so, I propose we fix this by adding the following tokens to the grammar:

%token NAMESPACE_ID_ALIAS 
%token NAMESPACE_TYPE_ALIAS 
%token NAMESPACE_NAMESPACE_ALIAS

similar to TYPENAME, which carry around a tree-node for 
"namespace::name" instead of a type.  We then expand the symbol
to the fully qualified name when needed, like:

qualified_id : NAMESPACE_ID_ALIAS
	{ $$ = EXPAND_NAMESPACE_ALIAS($1); }

qualified_type : NAMESPACE_TYPE_ALIAS
	{ $$ = EXPAND_NAMESPACE_ALIAS($1); }

identifier : NAMESPACE_NAMESPACE_ALIAS
	{ $$ = EXPAND_NAMESPACE_ALIAS($1); }

Then "using" statements simply define the appropriate NAMESPACE_xxx_ALIAS
symbols in the current scope, similiar to how typedefs define a TYPENAME.
(By the way, how *do* typedefs define a TYPENAME?  I've tried to follow
through the code by hand to figure this out and keep getting lost...)

The only slightly painful part is when we see a "using xxx;" statement to
import a whole namespace, we have to make aliases for the whole list of 
symbols in the namespace...

This has one possibly confusing side-effect; if we have:

namespace fred {
	int a;
};

namespace jane {
	int b;
	using fred::a;
};

namespace paul {
	int c;
	using jane::a;
}

this could well succeed in making the "a" seen in namespace paul be actually 
fred::a, or fail, depending on how we define EXPAND_NAMESPACE_ALIAS()...
It's not clear to me which way this is supposed to work.

Marc