Re: [Patch] PR c++/42824

[Jason Merrill <jason at redhat dot com>]

On 01/31/2010 04:20 PM, Dodji Seketeli wrote:
> On Sun, Jan 31, 2010 at 03:29:52PM -0500, Jason Merrill wrote:
> > On 01/31/2010 02:40 PM, Dodji Seketeli wrote:
> > > Ah. I think as Wrap<Quan<V2>   >   is a partial specialization
> >
> > It is not.
>
> Sorry, I meant partial instantiation as in the comment in
> lookup_template_class.

Ah, I see.

Eval<Mrtl>::Wrap is a partial instantiation of Eval<T>::Wrap.

Eval<Mrtl>::Wrap<Quan<0> > is a full instantiation of that partial 
instantiation.

Eval<Mrtl>::Wrap<Quan<V2> > is also a full instantiation of 
Eval<Mrtl>::Wrap, but it appears in the signature of the partial 
instantiation Eval<Mrtl>::value #0, and is still a dependent type after 
the partial instantiation of that function signature.

I guess the bug is that lookup_template_class fails to make this 
distinction, and thinks that it too is a partial instantiation.  That 
is, this code:

>       /* This type is a "partial instantiation" if any of the template
>          arguments still involve template parameters.  Note that we set
>          IS_PARTIAL_INSTANTIATION for partial specializations as
>          well.  */
>       is_partial_instantiation = uses_template_parms (arglist);

Is wrong.  I guess it ought to compare the innermost args with the 
innermost args of the template, instead.

Jason