phil@eecs.nwu.edu (William LeFebvre) (08/30/89)
Vitals: Machine: Multimax model 320 OS: UMAX+NFS 4.2, version B3_3.21 named version: 4.8 #16 Our named will core dump occasionally, and I'm not sure why. Has anyone else out there seen something similar? I've narrowed it down to that part of ns_resp after the long block comment that starts "Penalize those who had earlier chances but failed..." The code is doing floating point computations involving ALPHA, BETA, and GAMMA. This makes me suspect that one of the computations is overflowing. Adb indicates that core was dumped while processing "mull $1.2000000000000000e+00,f0". Won't I get a SIGFPE if that multiply overflows? Sound familiar anyone? William LeFebvre Department of Electrical Engineering and Computer Science Northwestern University <phil@eecs.nwu.edu>
soper@maxzilla.Encore.COM (Pete Soper) (08/30/89)
The "$F" command in adb will show you the saved float registers
and MAXDOUBLE is about 1.798e308 so you can determine for sure
if an overflow happened. The next step would be to backtrack and
figure out what had to happen to bring this about. One suspect might
be a function returning MAXDOUBLE as the result of an out of range
argument.
----------------------------------------------------------------------
Pete Soper +1 919 481 3730
arpa: soper@encore.com uucp: {bu-cs,decvax,gould}!encore!soper
Encore Computer Corp, 901 Kildaire Farm Rd, bldg D, Cary, NC 27511 USAjb@CS.BROWN.EDU (08/30/89)
I've heard of the problem before. The complaint came from the person running the nameserver at Encore. The problem has never been located. I haven't run a nameserver on a multimax, so I haven't seen it personally. Jim