Well, after a little work I've come to the conclusion that I think
the runtime library needs to catch a FPE unless you want a core
dump on overflow (or you don't write your own handler in your app).

I wrote several programs, compiled them to asm, and hacked the asm
to come to that conclusion.  So the next question is "Should a
floating point overflow cause a SIGFPE by default, or should it
not?  If not, how should IEEE results be returned?  I think Linux
requires a -lieee flag on the compiler to return INFs.  Is that
proper for NetBSD as well, or should we make IEEE mode default,
and allow a flag to speed things up (and SIGFPE if you're not
careful)?  Should it be a seperate library, or some other device?

I guess what I'm really asking is for advice, direction, worth of
effort, and is this already being worked on?

Jon