f2cl, cernlib, and maxima (Re: [Maxima] Release plans)

>>>>> "Richard" == Richard Fateman <fateman@cs.berkeley.edu> writes:

    Richard> The ACM software copyright is on this page:
    Richard> http://www.acm.org/pubs/copyright_policy/softwareCRnotice.html
    Richard> and contacting them by permissions@acm.org is suggested
    Richard> for commercial use.

I'll contact them.

    Richard> There is a key difference in the numerical software (claims)
    Richard> of Mathematica vs this other stuff.  Mathematica numerics
    Richard>   are all variable-precision, and so one can crank up the
    Richard> accuracy of any of their functions. You should be aware
    Richard> of this.   I'm not sure what Maple does for special functions.

My only intent was to replace the existing routines with something
that had better than the single precision accuracy we have today.
Variable precision would be the long-term goal, I think.

    Richard> Macsyma has bigfloats, but only for rational operations plus
    Richard> sin/cos/exp/log/atan.  We could get more by importing mpfun
    Richard> bigfloats.

I don't think mpfun goes far enough.  You get a few more, but nothing
like gamma, Bessel, etc.  And if we have to do it from scratch, maxima
bigfloats or mpfun would be equally easy or difficult.  Are there
libraries for mpfun for these other functions?  That would help a
lot. 

    Richard> There is also a problem in mapping notions like
    Richard> IEEE-infinity and IEEE-not-a-number into the symbolic
    Richard> part of Maxima.

Wouldn't you map IEEE infinity to INF and MINF?  It's not the really
the same, but if you're converting a floating-point number into some
kind of symbolic thing, I think that's pretty close.  NaN should be an
error of some sort, I guess.

Ray