Re: [PATCH] PR c++/53609 - Wrong argument deduction for pack expansion in argument pack

[Jason Merrill <jason at redhat dot com>]

It seems like your new code is a generalization of the old code for 
handling substitution of a pack for itself (arg_from_parm_pack and such) 
and the code for handling other packs with a single pack expansion 
argument, and should replace those rather than adding on.

The solution that if at a certain index all the packs have expansion 
arguments then the substitution produces a pack expansion seems right to 
me, but if one pack has an expansion and another pack has a normal 
argument, we can't do the substitution and need to fall back on the 
PACK_EXPANSION_EXTRA_ARGS mechanism.

> +set_arg_pack_select_index_for_pack_expansion (tree aps,
> +					      int i,
> +					      tree arg_pack)
> +{
> +  if (any_non_real_argument_pack_element_p (arg_pack))

I don't think we care if *any* element is an expansion (and please talk 
about expansions rather than "non-real elements").  What we care about 
is whether the i'th element is an expansion.  And we need to compare all 
the pack elements, so I think this needs to be handled in the main 
function rather than encapsulated here.

> +	TREE_VEC_ELT (args_vec, i) =
> +	  TREE_VEC_ELT (ARGUMENT_PACK_ARGS (arg_pack), i);

Aren't the LHS and RHS the same location here?

Jason