Hi,

I tracked down this regression to r225621, a clean up committed by Jason 
a while ago: unless we want to try something more aggressive, we can fix 
it the regression by simply restoring a few lines in 
cp_parser_template_parameter which consume the ellipsis. To clarify 
implementation-wise, the types affected are essentially all those for 
which cp_parser_simple_type_specifier (thus 
cp_parser_parameter_declaration) doesn't set the type in decl_specs:

       if (decl_specs
       && (token->keyword != RID_SIGNED
           && token->keyword != RID_UNSIGNED
           && token->keyword != RID_SHORT
           && token->keyword != RID_LONG))
     cp_parser_set_decl_spec_type (decl_specs,
                       type,
                       token,
                       /*type_definition_p=*/false);

Weird that we didn't have any testcase!

Tested x86_64-linux.

Thanks,
Paolo.

/////////////////////////