On Sun, 2007-09-16 at 20:51 -0700, Ralf W. Grosse-Kunstleve wrote: > Under Linux, gcc/g++-compiled code happily continues running after > producing NAN and INF. Often it is time-consuming to backtrack to the > actual source of the numerical problems. In addition, such problems may > go undetected for some time, which can cause all kinds of confusion.
Yes, NaNs are quiet by default. > Other platforms stop after a floating-point exception (e.g. HP's Tru64 Unix > with cc/cxx), or support customizations of FPE behavior via environment > variables (e.g. SGI's old IRIX). Is something like this possible with > gcc/g++ under Linux? This really has very little to do with GCC -- this list is for GCC development. Please try asking on the gcc-help mailing list, or a C programming list. Cheers, Ben
