Oh, mpfr supports signed zeros, since it mimics the IEEE float
design.
There are other issues that come up. For example, do you
want to do error handling by exceptions or just returning NaNs?
What should the various functions return for NaN arguments?
What is NaN* %i ?
Kahan answers some of these questions, but often the answers
are his bestconsidered tradeoffs, and ultimately some lead to
contradictory results.
RJF
----- Original Message -----
From: "Raymond Toy" <raymond.toy at ericsson.com>
To: <macrakis at alum.mit.edu>
Cc: "Maxima list mailing" <maxima at math.utexas.edu>
Sent: Monday, February 27, 2006 7:51 AM
Subject: Re: [Maxima] branch cut fixup for atanh(bfloat(x))
>>>>>> "Stavros" == Stavros Macrakis <macrakis at gmail.com> writes:
>
> Stavros> 1. (*) text/plain ( ) text/html
> >>
> >> Maxima's bigfloats don't support signed-zeroes.
>
>
> Stavros> 1) They could be implemented.
>
> Stavros> 2) They probably don't *need* to be implemented, because
> Maxima bigfloats
> Stavros> never underflow (try 2.0b0^-(2^1000) for example).
>
> Yes, but it's useful for things like log(1-z) where z = 2+0*%i. If
> you don't have signed zeroes, you'll get a different branch cut.
>
> I think that's something maxima needs to decide. Should signed zeroes
> be supported?
>
> Ray
>
> _______________________________________________
> Maxima mailing list
> Maxima at math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
>