Some initial diagnosis of memtemp89.C

[Paul Burchard]

I have started looking into the crash test memtemp89.C (reproduced
below), and have discovered at least two distinct problems with the
handling of template template parameters.

class Q {
	template<class T>
	class X {
	};
};
template<template<class> class XX>
class Y {
	XX<int> x_;
};
Y<Q::X> y;


1. The template template arg `Q::X' of `Y<Q::X>' is mis-parsed.  Tthe
parser erroneously accepts `Q::X' as an expression template arg
(specifically, as the overqualified_id of a fictitious data member `X'
of class `Q').  I'm not yet sure why the template_arg => type_id
production fails (thus falling through to the undesired template_arg =>
expr_no_commas rule), since type_id => nested_type is a valid
production.

2. The function tsubst() seems to make incorrect assumptions about
template template args (the comment before the call to
lookup_template_class() is wrong, I think).  More specifically, when
calling lookup_template_class() on the template template arg `Q::X', the
`context' argument should be NULL; no function context is involved. 
Indeed, if I make `context' NULL, the crash goes away and gcc gives the
message "In instantiation of `Y<Q::X>': `X' is not a template" (which is
probably correct, given the erroneous parsing of this arg).

-- 
----------------------------------------------------------------------
Paul Burchard   <burchard@pobox.com>   http://www.pobox.com/~burchard/
----------------------------------------------------------------------