From: <me...@ho...> - 2005-05-05 11:00:58
|
On Thursday 05 May 2005 10:46, Christophe Rhodes wrote: > Dave Roberts <ld...@dr...> writes: > > On Wed, 2005-05-04 at 08:32 -0700, Dave Roberts wrote: > >> Was trying to build the RPMs for 0.9.0 last evening and encountered > >> this. > > > > <snip...> > > > > It built fine against 0.9.0.18 from CVS this evening. Don't know what > > the problem was, but it looks like it already got fixed. > > No. It's more likely to be a race condition that intermittently gets > triggered. (I'm not sure if any of Gabor Melis' recent patches to > this list fix this problem -- Gabor?) I guess this is one of the identified problems in the "threads vs gc fixes" thread. Since I've been running with that patch (the second one) no such error surfaced. That said, even with this patch thread creation is less than optimal (too much waiting for each other between the creator and the created thread) and there are more, serious thread related problems, most notably errno corruption and a memory leak related to fd-streams. OK, I cannot stop: - errno corruption might be solved by creating thread as libc would do it and trivially protecting errno from signal handlers. Better wait for pthreads, though, if it's not too far away (?). - IMHO the fd-streams related leak is solved by the B. Downing's patch I advertised recently on sbcl-help accompanied by a few performance data points. There were some concerns over its interaction with serve-event. I am reasonably sure that it does not make the serve-event/thread issues worse. Gabor |