[comp.unix.ultrix] cc core dumping

mf@ircam.ircam.fr (Michel Fingerhut) (10/16/90)

Setup: DECsystem 5820. Ultrix 4.0.  Compiling an X11 Xw file (Manager.c) with

   cc -c -O -Wf,-XNh2000 -Olimit 2000  -I/usr/local/include -I.. -I. Manager.c
	
It then says:

   Fatal error in: /usr/lib/cmplrs/cc/uopt Segmentation fault - core dumped

What should I do now?

mjr@hussar.dco.dec.com (Marcus J. Ranum) (10/18/90)

In article <1990Oct15.201820.7348@ircam.ircam.fr> mf@ircam.ircam.fr (Michel Fingerhut) writes:

>   Fatal error in: /usr/lib/cmplrs/cc/uopt Segmentation fault - core dumped

	Is uopt the optimizer ? I think so. Obviously, the internal error
is something that should be SPRred (is that a verb?) but you may be able to
temporarily work around it by turning off optimization for that one file.

	Ick.

mjr.

russ@silver.ucs.indiana.edu (Jeff Russ) (10/19/90)

Our 5840 running Ultrix 4.0 also dumps core occassionaly.  If I keep
repeating the cc command it will eventually compile okay.  It looks like
the logic for certain instructions in the machine needs to be warmed up for
a compile to succeed. (My old PDP8s at home sometimes have the same problem).

We currently see two different types of errors when it aborts:

	% cc -O frac.c
	Fatal error in: /usr/lib/cmplrs/cc/uopt Segmentation fault - core dumped

	% cc -O frac.c
	pid 11309 (uopt2.1) was killed on unaligned access, at pc 0x1a
	Fatal error in: /usr/lib/cmplrs/cc/uopt Bus error - core dumped

The last one occurrs less frequently.  The code compiles without any problems
on a 5200 running Ultrix 4.0.

mf@ircam.ircam.fr (Michel Fingerhut) (10/22/90)

Jeff Russ: the type of problem you report (with cc dumping core erratically)
appears to me to be different from the one I reported (which is repeatable).
It reminds me of what we had when we ran the two processors of the 5820:
the file system went bezerk: access to files sometimes worked sometimes didn't,
their type (as returned with "file") changed from command to command -- it
was really crazy.

Fortunately DEC found the right solution to this one: they recommended not
using the second CPU and run monoprocessor until further notice.  You can
imagine my relief and pleasure, right?