Require canonical type comparison for typedefs again.

[Jason Merrill]

On 10/23/2010 05:32 PM, Dodji Seketeli wrote:
> If I change the declaration of T to make it correct, we don't have that
> error message.

Ah, I see.

> It is because T is actually not properly declared that we
> get the message. So should I just go back to using the tf_none argument
> when tsubsting into the template parms/args altogether?

I suppose so.

Jason