This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: [v3] make check-abi
- From: Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: libstdc++ at gcc dot gnu dot org
- Date: Thu, 22 Aug 2002 22:36:52 -0500 (CDT)
- Subject: Re: [v3] make check-abi
- Organization: Networks and Infrastructure Lab (IL02/2240), Motorola Labs
In article <200208222001.g7MK1Od03578@fillmore.constant.com> Benjamin writes:
> At the moment, there is only a baseline file for x86/linux. Loren has
> indicated a BSD baseline exists, so I expect this will be checked in
> shortly.
Baseline installed for i386-unknown-freebsd4.6 as well, but I see I
only mailed the libstdc++ list with my actual patch (and I avoided
posting the large new file as well).
Shall we tweak the configure fragments as required so that if
config/abi/CPU-VENDOR-OSX.Y/baseline_symbols.txt is not found then
(at least) config/abi/CPU-VENDOR-OSX/baseline_symbols.txt and then (maybe)
config/abi/CPU-VENDOR-OS/baseline_symbols.txt are candidates for use.
gcc 3.2 on CPU-unknown-freebsd4 has one C++ ABI and gcc 3.2 on
CPU-unknown-freebsd5 may have another (although it may coincidentally
be identical). By default, the target triples constructed by
config.guess on FreeBSD, Solaris and elsewhere add the OS version
number.
> It is my intention to require library check-ins on the gcc-3_2-branch
> to pass 'make check-abi' before being put on the branch. I think this is sane.
Most sane. As well, check-ins on mainline should have explainable
changes only. For example, we have three symbol removals that are
explainable in that no user code should have called them directly or
indirectly.
Regards,
Loren