I recently installed  .tgz files for the 1.6U snapshot from ftp.netbsd.org.
After the update I've seen two instances of what looks like a compiler bug.

First problme is that my local build of racoon coredumps.  Diagnosing
with gdb, next inspecting cc -S showed that two file-scope read/write
variables (which just happened to come after a lex/yacc-generated
readonly table) were, in fact, being outputted into the .rodata
section of the assembly file.

I've also had strange errors when testing snapshots of my fast-ipsec
port.  (e.g., newly-built kernels fail to configure a cardbus tlp
device).  Older kernels work fine.

It sounds very like hardware error, maybe bad memory. But I tried
moving the disk to a second machine, and after recompiling a GENERIC
kernel (no fast-ipsec), that box misbehaves in the same way.

Has anyone seen anything even vaguely similar, or should I continue to
look for local hardware errors?