Re: [C++ PATCH for 3.3] Fix PR10849

[Kriang Lerdsuwanakij <lerdsuwa at users dot sourceforge dot net>]

Mark Mitchell wrote

> Thanks for working on this.  Dealing with the old parser is so
> unappetizing...
>
> I'm not sure why you're setting *new_type_p the way you do; that's
> supposed to be set if a new type is being declared (i.e., if a
> class-head, rather than an elaborated-type-specifier, is being
> processed).  Why is the logic you used right?
>
> Also, where is the pop_scope() corresponding to the push_scope() you
> added?  Is there any way that you could cue the disabling off of
> maybe_process_partial_specialization?
This particular use of 'new_type_flag' is for case like

 class A::B {
   ...
 };

where 'A' is a class or namespace.  So inside the definition of this 
class, name lookup is performed in the scope of 'A'.  Also this
affects access checking of the class head 'A::B' itself.  The
'new_type_flag' is checked later in 'finish_class_definition'
called by 'structsp' non-terminal.  So the 'pop_scope' is actually
inside that function.

About the 'maybe_process_partial_specialization' solution, deferred 
access for both class head and base classes are already inserted
in the 'type_lookups' list.  We could walk the list and get rid of
the class head there but I think it is ugly.

--Kriang