PR c++/42758

[Dodji Seketeli]

On Tue, Jan 19, 2010 at 03:31:10PM -0500, Jason Merrill wrote:
> Perhaps rather than try to recreate the information about which
> arguments were defaulted at formatting time we should just remember
> it, either by hanging the count off the TREE_VEC somewhere or
> marking/wrapping the arguments themselves.

The patch below follows the first choice of storing the count in the
arguments TREE_VEC.

I think it fixes a couple of other defaulted template arguments count 
related PRs and mitigates a little bit PR c++/42336. So I have added 
testcases of those tangent bugs I noticed in the patch.

Tested on x86_64-unknown-linux-gnu against trunk.

Thanks.

        Dodji