On Sat, 14 Apr 2007, Johnny Billquist wrote:

> We still have the problem with gcc with any optimization causing a crash.

   I have gotten around this by building cc1 and cc1plus with -O1.  I've 
got these changes on my build system, but I've not completely verified 
that it works completely.

Index: gnu/usr.bin/gcc4/cc1/Makefile
===================================================================
RCS file: /cvsroot/src/gnu/usr.bin/gcc4/cc1/Makefile,v
retrieving revision 1.3
diff -u -r1.3 Makefile
--- gnu/usr.bin/gcc4/cc1/Makefile	12 Nov 2006 01:24:44 -0000	1.3
+++ gnu/usr.bin/gcc4/cc1/Makefile	14 Apr 2007 16:49:23 -0000
@@ -9,4 +9,8 @@

  COPTS+=	-Wno-stack-protector

+.if ${MACHINE} == "vax"
+CFLAGS+=	-O1
+.endif
+
  .PATH: ${DIST}/gcc ${G_out_file:H}
Index: gnu/usr.bin/gcc4/cc1plus/Makefile
===================================================================
RCS file: /cvsroot/src/gnu/usr.bin/gcc4/cc1plus/Makefile,v
retrieving revision 1.2
diff -u -r1.2 Makefile
--- gnu/usr.bin/gcc4/cc1plus/Makefile	1 Jul 2006 19:06:51 -0000	1.2
+++ gnu/usr.bin/gcc4/cc1plus/Makefile	14 Apr 2007 16:49:23 -0000
@@ -12,4 +12,8 @@
  # YUCK.  but see timevar.h:POP_TIMEVAR_AND_RETURN
  CFLAGS+=	-Wno-error # not good enough: -Wno-return-type

+.if ${MACHINE} == "vax"
+CFLAGS+=	-O1
+.endif
+
  .PATH: ${DIST}/gcc/cp ${DIST}/gcc ${G_out_file:H}

> I was reminded of it today when I thought I'd try to build gcc natively on 
> the VAX, by passing an empty string as CFLAGS, so that gcc would have been 
> built without any optimizations. That should be interesting, I thought.

   I've been working on building the tools on the VAX, and I've been using
HOST_CFLAGS=-O1.  I've not actually completed a build yet, but have gotten 
a couple of more fixes into the tree.  -O0 would be needed if gcc had been 
built with -O2.

   My first attempt used -O2, and the resulting gcc toolchain failed when 
trying to build a kernel on the VAX.  I then did a GENERIC kernel without 
useing the toolchain, and have been running that kernel for 25 hours now - 
trying to build the toolchain with the -O1 flag.

> However, the build stops before doing any real work. autoconf is running 
> through for libiberty, and when it's about time to build, the config.result 
> file is read in for parsing, and then it crashes. Looking in the 
> config.result file shows sections of NUL characters some way in.
>
> So. Anyone else seen something like this?

   I have not seen this on my system.  I've had lots of other problems, but 
not this one.

--
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA