Re: [Sbcl-devel] Problem with one patch from October

[Martin C. <cra...@co...>]

Gbor Melis wrote on Mon, Feb 02, 2009 at 08:43:30PM +0100: 
> On Lunes 02 Febrero 2009, Martin Cracauer wrote:
> > Gbor Melis wrote on Mon, Feb 02, 2009 at 06:53:25PM +0100:
> > > On Lunes 02 Febrero 2009, Martin Cracauer wrote:
> > > > I apologize for giving feedback so late, there's a bit of a
> > > > backlog here.
> > > >
> > > > I didn't find any discussion of the following patch.  It hurts
> > > > performance badly, a straight 4% for my employer's overall mix.
> > > > Given that we only do a certain amount of foreign calls this
> > > > seems to slow down actual calls by an order of magnitude.  On top
> > > > of that, it conses.  That call to cons in
> > > > invoke-with-saved-fp-and-pc is a real killer.
> > > >
> > > > I hope I didn't overlook a previous discussion, but can we
> > > > re-hash why this is required? Since it is labeled as a hack, I
> > > > assume that a better solution is possible but hasn't been
> > > > implemented yet?
> > > >
> > > > Maybe we can introduce a switch so that the extended marking is
> > > > only done when debug == 3?
> > >
> > > It's actually not the consing but accessing and binding that
> > > special what
> > > hurts.
> >
> > But that would break if a piece of C code would ever call back into
> > Lisp and the Lisp code spawns a new thread, right?
> 
> New threads don't inherit specials, if that's what you mean.

Yes, so if you create threads (in Lisp or C) with the binding in
effect it'll be lost.

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cra...@co...>   http://www.cons.org/cracauer/
FreeBSD - where you want to go, today.      http://www.freebsd.org/