Re: [C++ Patch] PR 34892

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Hi again,

On 10/24/2012 09:53 PM, Jason Merrill wrote:
> The parm shouldn't be a parameter pack in that case; the ... is part 
> of the default argument, not the parameter.
I tracked down where this goes wrong, in the parser: the below 
recognizes when cp_parser_parameter_declaration doesn't parse a valid 
default argument and in that case "*is_parameter_pack = true;" doesn't 
happen. This means we have a very good diagnostics, only a single error 
message and nothing else, not even the maybe_warn_variadic_templates 
message in C++98 mode because I don't think the code can make any sense 
in any C++ dialect whatsoever (the latter would be easy to change of course)

Tested x86_64-linux, as usual.

Thanks,
Paolo.

PS: interestingly, detecting error_mark_node toward the end of 
cp_parser_parameter_declaration and turning it into a NULL_TREE leads to 
a single regression, at the end of pr39060.C, because NULL_TREE implies 
that we don't do anymore the diagnostics about default argument part of 
duplicate_decls (which ICC also gives). Besides that, for 34892 we give 
like 6 different error messages ;)

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

Attachment: CL_34892_2
Description: Text document

Attachment: patch_34892_2
Description: Text document