Patch c++/40155

[Jason Merrill]

On 01/05/2010 07:36 AM, Dodji Seketeli wrote:
> This patch makes sure that we end up in the last "else" branch of the
> unify_pack_expansion to re-use the argument deduced earlier, when we are
> in a non-deduced context.

I think it would be clearer to move the template_parm_level_and_index 
call out from the various if/elses into the main body of the loop and 
then just copy the TMPL_ARG = from the final else.

> This bug does not seem to be a regression, but as it's an
> 'accept-invalid' on c++0x (should I say c++1x now?), maybe the final fix
> could go into 4.5?

Rather, it's a rejects-valid.

We're going to keep calling it c++0x until we have an actual date.

Yes, the fix can go into 4.5 since it only affects C++0x.

Jason