cvs commit: src/lib/libpthread Makefile src/lib/libpthread/test sigsuspend_d.c src/lib/libpthread/thread thr_cancel.c thr_concurrency.c thr_create.c thr_find_thread.c thr_init.c thr_kern.c thr_nanosleep.c thr_private.h thr_sig.c thr_sigmask.c...

David Schultz das at FreeBSD.ORG
Sat Jul 19 02:06:33 PDT 2003


On Fri, Jul 18, 2003, David O'Brien wrote:
> On Fri, Jul 18, 2003 at 05:45:46PM -0700, Marcel Moolenaar wrote:
> > On Fri, Jul 18, 2003 at 05:15:24PM -0700, David O'Brien wrote:
> > > 
> > > Who are "the arches"?  KSE is a new feature, so the responsibility is on
> > > the KSE developers to complete their implementation on all Tier-1
> > > platforms.
> > 
> > No. If there's not enough support from <arch>-gurus, then <arch>
> > should be bumped down to tier 2.
> 
> In general I agree.
> 
> > We cannot realisticly require MI
> > developers to know the MD details of all tier 1 platforms. We
> > need to have active maintainers for the MD bits. If they don't
> > exist, then we cannot really support the platform anyway.
> 
> HOWEVER, in this case, you can only take that line if you are paying for
> the labor.  All our platforms have MD maintainers today -- but that does
> not mean they are interested in doing KSE work.  KSE is the new comer,
> not the existing Tier-1 platforms; thus the is the KSE developers that
> need to have a complete implementation or not commit the work at all.

I think it's pretty clear that the KSE folks are ultimately
responsible for making sure that the work gets done.  On the other
hand, it's also clear that writing the MD glue is an order of
magnitude harder for someone unfamiliar with the architecture.
The KSE folks have done 95% of the work already, so they deserve
some assistance here.  There's no need to treat a request for help
as an imposition.

FWIW, the same is true of libthr, which, last I checked, still
doesn't work on Alpha.  libc_r is being phased out, and without
it, Alpha has no threading implementation.  Marcel has a point
here; it's hard to say that FreeBSD supports a platform if nobody
is willing and able to keep it up to date.  And it isn't as though
there's a whole lot of work involved.  Either one of libthr or
libkse would be perfectly adequate, but someone needs to do it, or
at the very least offer to explain how to do it.  I have nothing
against Alpha and I'm not trying to be critical of it, but someone
needs to step up to the plate.


More information about the cvs-src mailing list