Re: /tset/ANSI.os/maths/pow/T.pow 4

[Christopher Yeoh <cyeoh@samba.org>]

At 2002/2/25 16:31-0500  Dave Prosser writes:
> Matt Wilson wrote:
> > From lsb-distribution-test-1.1.3-1
> > 
> > ************************************************************************
> > /tset/ANSI.os/maths/pow/T.pow 4 Warning
> > 
> >         Test Information:
> > pow(0.0, -1.0) gave
> > RETURN VALUES: expected: -inf, observed: inf
> >     Bit Representation: expected value: \000\000\000\000\000\000\360\377
> >     Bit Representation: observed value: \000\000\000\000\000\000\360\177
> > 
> > ************************************************************************
> > 
> > Isn't pow(0.0, -1.0) undefined?  From the looks of C99, pow(long
> > double x, long double y) may cause a domain error if x is zero and y
> > is less than or equal to zero.
> 
> For IEEE F.P. systems, C99 says (F.9.4.4):
> 
>   pow(+-0, y) returns +-infinity and raises the "divide-by-zero"
>   floating point exception for y an odd integer < 0.
> 
> So the problem with the test is that it expects the wrong sign
> for the infinity returned given the information in your message.

That tests suite was written for POSIX.1-1990. I don't have a copy of
that, but SUSv2 states:

If x is 0.0 and y is negative, -HUGE_VAL is returned and errno may be
set to [EDOM] or [ERANGE].

So I'm guessing the specification changed since the test was written.
I think we classify Warnings as Pass anyway (Andrew please correct me
if I'm wrong) so we might not have to worry about this one. If not, it
appears to be following future behaviour so it should be waived
regardless.

Regards,

Chris
-- 
cyeoh@au.ibm.com
IBM OzLabs Linux Development Group
Canberra, Australia