> > softdeps definitely needs to be made more stable.
>
> 	On what system/os rev, and how did you make it fall over?

I've killed softdeps on NetBSD 1.5 on m68k Mac with SCSI, m68k Amiga with
IDE, SCSI, and wide ultra SCSI, and on i386 with IDE and SCSI. On the i386
this only seems to happen with slower machines.

After 1.5 was officially released, I did installs on a number of machines,
and I turned on softdeps. During post-install, I have the machines do a
bit of stuff that thrashes (building packages while supping can last till
3:00am, when the locate database updates; this has killed many an
install).

> 	We've been running softdeps on our internal server (mixed IDE
> 	SCSI Dell box, nfs server, samba, netatalk, apache, tomcat and
> 	frequently ends up paging furiously while building from pkgsrc
> 	and running a good sized (400 million entries) analog run),
> 	since 1.4X occasionally updating through to its current 1.5.1_BETA
> 	I also run it on a couple of other boxes without problem.

It would be great if the problem could be isolated; I'm not very familiar
with gdb, nor have I been able to find a good source of information about
it (yet), but if someone wants to help me debug a crashed softdep machine,
I'm sure I can duplicate the problem.

> 	I fully believe there are problems left in softdep, I'm just
> 	curious to try to find out what other people are doing to tickle
> 	them, to make sure its PRed (and that I know what not to try :)

The speed thing makes me curious. What happens when the softdep code has
more work lined up for it than it has CPU to take care of?

Thanks,
John
-- 
The proof of a system's value is its existence.