This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: [v3] more consistent api testing
- From: Benjamin Kosnik <bkoz at redhat dot com>
- To: Paolo Carlini <pcarlini at suse dot de>
- Cc: libstdc++ <libstdc++ at gcc dot gnu dot org>
- Date: Tue, 20 Feb 2007 21:04:40 +0100
- Subject: Re: [v3] more consistent api testing
- References: <45DB3F0B.5090007@redhat.com> <45DB4F95.10609@suse.de>
I'm going to look a bit into the issue myself, but can you detail a bit
what's going wrong with <locale>??
Yes. Instead of required specializations for ctype_byname, we've just
specialized the ctors.
The compiler error makes little sense in this case: something about
specializing after instantiating.
But, the interface diergence seem clear. The actual inpact is minor, IMHO.
But, somewhat troubling. I am having problems testing for required
specializations, see:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30857
As regards the unordered containers, the issue coming from the
additional template parameter is in fact well know (are there any
other?): as apparent also from some of Matt' comments, he wanted to add
some additional flexibility, along the lines paved by Ami. Time ago, I
mentioned on the list that the additional parameter would cause strict
conformance problems (and we avoided such exact problems while recentlt
reworking shared_ptr), now probably it's time to rediscuss the thing:
shall we avoid it, by forwarding from conforming classes to the flexible
ones? The latter would be uglified but still available, similarly to
shared_ptr.
Yes, it's just the undocumented extra parameter. We should at least
document this....
Yes, is my vote. I think we should conform to the interfaces as
specified... or change the specified interfaces.... :)
This is a portability issue, as people using the GNU TR1 interfaces
could conceivably have specializations that don't compile on
TR1-conformant systems.
best,
benjamin
ps. Sorry about the churn