recent 3.0.3 regression for g++.pt/memtemp100.C

[Bob Wilson <bwilson at tensilica dot com>]

I built from yesterday's 3.0 CVS files and saw a regression from 3.0.2. 
  The g++.pt/memtemp100.C test fails with:

.../g++.old-deja/g++.pt/memtemp100.C:18: parse error before `>' token
.../g++.old-deja/g++.pt/memtemp100.C:25: confused by earlier errors, 
bailing out
compiler exited with status 1

The only recent changes that look even remotely related are:

2001-11-20  Mark Mitchell  <mark@codesourcery.com>

         PR c++/3637
         * call.c (non_reference): Add documentation.
         (convert_class_to_reference): Do not strip reference types
         from conversion operators.
         (maybe_handle_ref_bind): Simplify.
         (compare_ics): Correct handling of references.

2001-11-19  Mark Mitchell  <mark@codesourcery.com>

         * cp/decl2.c (grok_alignof): Make sure that expression created
         while processing a template do not have a type.
         * cp/typeck.c (c_sizeof): Likewise.
         (expr_sizeof): Likewise.
         (c_alignof): Likewise.

Unfortunately, it is hard for me to tell you exactly how to reproduce 
this, because I'm using the Xtensa port for which you do not yet have 
the sources (see my as-yet-unanswered question from yesterday on the 
gcc-patches list about when and where I can contribute this port). 
However, I'm pretty sure this problem is not specific to the Xtensa 
port, since we're doing nothing special with regard to C++ and 
templates.  I'd be glad to provide whatever information you want and 
even help debug it, but I'm not at all familiar with the C++ frontend.

Thanks!