Re: __FreeBSD_version
On Tuesday, November 29, 2011 1:49:24 pm Robert Millan wrote:
> 2011/11/29 John Baldwin <jhb@freebsd.org>:
> > Is this code in a Debian-specific package or from another party,
>
> Christoph probably knows.
>
> > and is this
> > dependent on something that is unique to FreeBSD but not GNU/kFreeBSD? (That
> > is, which syscall ABI does GNU/kFreeBSD use and is this change something that
> > is specific to the "native" ABI?)
>
> I don't know the details to this particular case (it's Christoph's
> package), but in case it helps: we use the same syscall ABI as FreeBSD
> [1]. Our pthread library however, is based on rfork() (move to thr()
> is planned). It is very different from the two implementations of
> userland-side threading in FreeBSD (KSE is one of them, right?).
>
> [1] barring a few cases in which we were forced to diverge, but has
> gotten a lot better lately since some of those patches were merged in
> 9-CURRENT.
Ahh, ok. If you are using the "native" signal trampolines, etc. then you
probably want the #ifdef Christoph posted enabled as needed.
FWIW, FreeBSD has dropped KSE at this point as the complexity simply wasn't
worth the gain, so FreeBSD is down to only one real user thread library now
(thr() based). One could argue that scheduler activations is a hack in the
threading implementation to workaround ineffecient uses of threads in
userland, but that is an entirely different can of worms. :)
--
John Baldwin
Reply to: