> Yow!  That isn't going to work either.  I've now started it on two
   > occasions, and after running for about two hours it falls over whilst
   > running "make depend" in lib/librpcsvc.  The compiler and "make"
   > abort with Bus Errors and exec() causes more of 'em (so I can no
   > longer login to the machine).  This has now happened twice.

   This sounds like the "everything dumps core" bug.  It happens on some
   machines with depressing reproducibility; it doesn't happen on others.
   For example, I have a -3/150 that it happens on and a -3/260 that has
   been up for a couple of weeks (of software building, too) without a
   single strange coredump.  It's a hard bug and AFAIK nobody knows what's
   doing it, though I think gwr has some vague guesses.

   > ... from which I surmise that exec() has stopped working again :-)

   It's not really exec().  It appears to have something to do with shared
   libraries; an executable that uses no shared libraries will
   (invariably, in my experience) work, even when "everything dumps core".



Hello [another core dumps story]....


[given two 3/50s which can get into this state on demand]


The two Sun 3/50s were running diskless for quite a while.  Now one of
them has its own disk and, as far as I can tell, has *never* gotten
into the "all things dump core" state again.  The one which is still
diskless can still do this on demand, with an NFS copy of the same
binaries as the other one.

Further [even stranger], while moving the system from NFS to the disk,
I found that the 'tar' command would drop core...... However, 'tar' is
a static binary.  Compiling a *dynamic* version of 'tar' worked ok....

On the up-side, after I compiled everything, the 3/50 with the disk
has been very stable, the other one is lucky to make it a couple of
hours.

I have not had a whole lot of time to fiddle with this any more....





Brad Spencer - brad@anduin.eldar.org   http://anduin.eldar.org