Date: Fri, 9 Jun 1995 16:25:34 +1000 From: Bruce Evans <bde@zeta.org.au> To: bugs@FreeBSD.org, mal@algonet.se Subject: Re: Bogus (?) SIGFPE's from divide by zero Message-ID: <199506090625.QAA20634@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>I don't know what the standards say, but other systems (at least SunOS)
>do not issue floating point exceptions on divide by zero. Bug in FreeBSD?
Standard C: result is undefined. Sometimes the result is to send mail
to bugs@freebsd.org :-)
POSIX: same as Standard C
IEEE: default is to record the exception but not trap on it
IBCS2: SIGFPE (at least according to (a comment in) 386BSD-0.0)
FreeBSD: IBCS2
Right: IEEE
FreeBSD conforms to Standard C here. However, libm divides by zero on
purpose to generate IEEE exceptions. This and other generation of IEEE
exceptions in libm makes libm nonconformant because the IEEE exceptions
generate SIGFPE's.
>(The constant divisions are replaced by constants by gcc, I suppose)
This is another bug. It is OK for Standard C (the behaviour is undefined)
but not for IEEE. In particular, it breaks the generation of IEEE
exceptions in libm. The side effects are lost when the calculations
are done at compile time. I believe Sun's compilers are more careful
than gcc here.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199506090625.QAA20634>