This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: [Patch] libstdc++/26970
- From: Jakub Jelinek <jakub at redhat dot com>
- To: Gabriel Dos Reis <gdr at cs dot tamu dot edu>
- Cc: Paolo Carlini <pcarlini at suse dot de>, libstdc++ <libstdc++ at gcc dot gnu dot org>
- Date: Mon, 12 Jun 2006 16:11:06 -0400
- Subject: Re: [Patch] libstdc++/26970
- References: <448C71C2.4050801@suse.de> <87bqszii90.fsf@soliton.cs.tamu.edu>
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
On Sun, Jun 11, 2006 at 02:50:51PM -0500, Gabriel Dos Reis wrote:
> | the below started noticing that __convert_from_v was not exported and
> | that -Wformat-nonliteral is happy with variadic functions. Overall, in
> | my opinion the code is not less clean than before...
> |
> | I would ask Benjamin and Jakub to help me double checking that
> | everything is still fine wrt long double on powerpc... I will wait
> | until tomorrow, in any case.
With the latest patch Paolo sent I don't see why it couldn't just work
on the long double {64,128} arches, though I haven't tested it.
> Variadic functions are usually slower than non-variadic ones. Any
> reason why the issue could not be addressed differently?
But s*printf is already variadic and on most targets sprintf and snprintf
are just tiny wrappers around vsprintf and vsnprintf (the code duplication
would be otherwise huge). So it really doesn't matter if you do
va_start/va_end in the vsnprintf caller or call snprintf which in turn
does va_start/va_end.
Jakub