This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: C++0x issues
- From: Benjamin Kosnik <bkoz at redhat dot com>
- To: Christopher Jefferson <chris at bubblescope dot net>
- Cc: libstdc++ <libstdc++ at gcc dot gnu dot org>
- Date: Mon, 12 Oct 2009 17:26:58 -0700
- Subject: Re: C++0x issues
- References: <FB5C4A22-9CE8-4CAD-A414-5EFD17E92DE5@bubblescope.net>
Thanks for doing this.
> On a related note, we should probably start running the testsuite in
> C+ +0x mode more regularly, and somehow mark tests which won't pass
> in C+ +0x. I'm not sure how hard that is to do.
Yes, we should be thinking about ways to improve C++0x coverage.
As far as marking the tests that fail in C++0x mode, feel free to add a
common comment string to at least mark this for developers if not for
the test harness.
Like so:
// XXX Known fail with __GXX_EXPERIMENTAL_CXX0X__
As of 2009-08-11 there is a test file that includes all the std headers
and compiles with -std=gnu++0x.
testsuite/17_intro/headers/c++200x/stdc++.cc
I'm working on some testsuite additions that will compile/instantiate
all the C++0x container types.
And then C++0x-only features get tested, like mutex etc.
However, there are some parts of the library are lacking C++0x test
coverage pretty much entirely and include io, algorithm, c++2003
numerics like complex and valarray, and some of the extensions. If we
can figure out how to do simple checks for these areas in the C++0x
dialect it might make sense to add it, or else just wait to see what
breaks and add on a per-case basis...
best,
benjamin