[Openmcl-devel] linux process hang

[Larry Pells]

I just built the bleeding edge version of openmcl and found that if I 
recompile uffi, then the first 2 test cases don't hang.  Interesting.

Larry

On Saturday 05 February 2005 07:11, Larry Pells wrote:
> On Friday 04 February 2005 14:13, Gary Byers wrote:
> > Larry sent me the code that had been hanging.  I wasn't able to reproduce
> > the problem, but something about this looked kind of familiar.
> >
> > Some months ago I'd found a long-standing problem: the GC sometimes
> > frees certain unreferenced foreign pointers; when the GC is entered
> > in some thread, other threads are suspended and resumed when it's
> > finished.  If one of those suspended threads owns a malloc/free-related
> > lock, the GC will deadlock when calling free().
> >
> > That's pretty nasty (and pretty stupid); to make matters even more
> > stupid (and nasty, in their own way) I promptly checked in a fix
> > to the bleeding-edge CVS tree and completely forgot to check that
> > fix into the main tree.
> >
> > I think that I did do a moment ago.  I can't be certain that this
> > fixes the problem that Larry reported (since I couldn't reproduce
> > that in either the "main" or "bleeding-edge" lisps), but the conditions
> > under which this bug could occur and the symptoms were similar to
> > what he described.
> >
> > Larry: if you do a "cvs update" and recompile the lisp kernel now, does
> > the problem persist ?  (I found that the version of UFFI I have -
> > which may be fairly old - complained that WITH-CAST-POINTER was
> > not yet implemented; that would seem to be completely unrelated
> > to the hanging/deadlock that you saw.)
>
> Since I have rebuilt the lisp kernel (I also rebuilt the lisp image as I
> have a Makefile which does both), I removed the uffi compiled files and
> regenerated them (just doing some testing).  Now when I add the callback
> in, it always hangs, regardless if I put the dribble in or not.
>
> Larry
> _______________________________________________
> Openmcl-devel mailing list
> Openmcl-devel at clozure.com
> http://clozure.com/mailman/listinfo/openmcl-devel