Re: Patch ping

[Benjamin Kosnik <bkoz at redhat dot com>]

> > I believe in this *specific* case we can at least use mutexes without
> > penalizing performances.
> 
> I'm testing this patch.

Thanks for your patience with this.

Your patch looks acceptable to me, modulo some versioning tweaks. Since
gcc-4.0.2 was released with a defined set of GLIBCXX_3.4.6 exports, to
add new symbols we have to use GLIBCXX_3.4.7 and bump versions
accordingly. We are trying to keep mainline and gcc-4_0-branch in sync.

Using concurrence.h is an interesting exercise: we are not exporting
mutex objects, and using in .h files is out due to the multiple-copies
thing. It's helped us write exception-safe locks via the scoped lock
idiom but there is obvious room for improvement.

The long term plan (as far as I am concerned) is to look into using
Han's atomic_ops package, and building higher-level locking mechanisms
into concurrence.h that don't have such picky linkage issues. We
won't be able to use them in all situations, but hopefully more options
will mean more (and better) solutions.

Anyway.

This is ok for mainline/gcc-4_0-branch.

-benjamin

Attachment: p.20051010
Description: Text document