On Sun, Sep 28, 2003 at 03:46:45PM +0100, Daniel Barlow wrote:
> Unfortunately, not for the threading problem, which seems to grow the
> more I look at is (can anyone else give me a convincing explanation
> that SIGSEGV is not in sigaddset_blockable()?)
> There's a problem with GC hooks that causes SBCL to die ("Killed")
> during GC on a couple of platforms (some BSD that I'm not sure about,
> and, we recently discovered also, Linux 2.6). It appears to be
> related to hooks that cons. The attached patch disables pre-gc hooks
> altogether, and moves the post-gc hook running outside of the various
> GC locking stuff. (I'd like to retain post-hooks if possible because
> finalization depends on them). I view the pre-gc disabling as
> temporary and the post-gc move as permanent
> With this patch linux-2.6 can again build SBCL.
It seems reasonable to me to apply such a patch and then to delay
release while we encourage people to test again.
I'm not so enthusiastic about releasing SBCL unbuildable on several
platforms, so a patch sounds like a good idea.
The patch sounds about as conservative as can be, which is nice, but
it's still necessarily adventurous enough that I'm not enthusiastic
about the idea of applying it, crossing my fingers, and releasing
0.8.4 a few hours later. Thus, I propose that you commit the patch or
something similar, and then we plan to play with it for a few days,
more or less starting over on code freeze. If it looks good on
Thursday or so, then I can release the new version.
William Harold Newman <william.newman@...>
"There's nothing an agnostic can't do as long as he doesn't know
whether he believes in anything or not." -- Monty Python.
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C B9 25 FB EE E0 C3 E5 7C