Gunnar> Raymond Toy <toy(a)rtp.ericsson.se&gt; writes: Gunnar> In my experience you can't trust the values of variables reported by Gunnar> gdb (incl. backtraces) unless you compiled with -g and without Gunnar> optimisation. Agreed. However when I say it's 8388xxxx, that's the value of the parameter in the routine make_lisp_buffer_stream which was initialized from start1, and that should be valid, right? Gunnar> It's supposed to be. Ray, could you try compiling with gcc but with Gunnar> less aggressive optimisation (say -O1 instead of -O3)? I tried recompiling fileio (I think) without -O. I still got crashes, but I'm not 100% sure if I was running the new version or the old. I am pretty sure that I'm running the Sun CC version now, and that seems stable. I do know for a fact that 2.95.x, x < 2, was not capable of building a temacs that could create an xemacs, so I stayed with 2.7.2.2 for a long time. Ray