Re: cpow

On 9/10/25 10:07, Steve Kargl wrote:
> On Wed, Sep 10, 2025 at 03:16:51PM +0000, Vincenzo Innocente wrote:
>>
>>> On Sep 10, 2025, at 5:13 PM, Steve Kargl <kargls@comcast.net> wrote:
>>>
>>> On 9/9/25 23:45, Paul Zimmermann wrote:
>>>>         Hi,
>>>> we noticed that in FreeBSD 14.3, cpow(x,y) for x = y = (+0,+0) yields (+0,+0),
>>>> whereas GNU MPC yields (1,+0), which matches the real case, where IEEE 754
>>>> requires x^+0 = 1 when x is not a signaling NaN.
>>>> Is there any reason for the choice (+0,+0)?
>>>
>>> It seems that FreeBSD's cpow was imported from OpenBSD.
>>> Looking at the Copyright, it was written by Stephen L.
>>> Moshier, which means the code ultimately comes from
>>> CEPHES.
>> WOW!
>> suspect last update was in the `90s
>>
> 
> If one looks at FreeBSD's pow(3) manual page, there
> is a rather long NOTES section about x**0 = 1 forall
> x.  It would seem logical to update cpow() to have
> the same semantics.  I won't do the change as I try
> to avoid looking at CEPHES derived codes in FreeBSD
> libm.
> 

Well, I couldn't help myself.  There's a patch attached to

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=289447

-- 
steve