Re: Renaming IS_AGGR_TYPE & co

[Mark Mitchell <mark at codesourcery dot com>]

Paolo Carlini wrote:
> Hi Mark,
> > In all three uses of that function, we want a class type, AFAICT.  So,
> > I would change the function to be is_class_type and use CLASS_TYPE_P
> > as the predicate in the test.
> I tried that, and unfortunately the build fails in libstdc++-v3 with the
> below. I'm pretty sure (if you want I can double check) that
> build_offset_ref (in init.c) is the caller passing sometimes a
> non-CLASS_TYPE_P, it asserts only TYPE_P. I was about to submit a
> version just using maybe_class_type_p, but was unsure about the error
> message...

What is the "type" in build_offset_ref at this point?  I'm guessing it's 
a TYPENAME_TYPE.  That's a little surprising, since we check for 
dependent types right above the call to is_aggr_type.  What's the call 
stack?

If it is a TYPENAME_TYPE, the fix might be to do:

  if (TREE_CODE (type) == TYPENAME_TYPE
      || class_type_p (...))
    ...

We still shouldn't need to allow things like TEMPLATE_TYPE_PARM; those 
are *always* dependent.

Thanks,

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713