[Maxima-discuss] SIGN1 strangeness, was: Problem with bfloat

[Robert D. <rob...@gm...>]

On 2014-08-30, couraud <l.c...@gm...> wrote:

> solve(eq, r6);
>
> 1 Enter bfloat [0.0 + 1..#INF00e+000E+2678440]
>
>  2 Enter bfloat [0.0]
>
>  2 Exit  bfloat 0.0b0
>
>  2 Enter bfloat [1..#INF00e+000E+69]
>
> bfloat: attempted conversion of floating-point infinity.

I think what's going on here is that Maxima is attempting to determine
the sign of some constant expression (involving ratios of enormous
integers), and tries to compute a floating point approximation.
I think there is some provision in the code for failure to compute a
float, and then it goes to a bigfloat approximation. That scheme
works OK for Lisp implementations (Clisp, SBCL, Clozure CL) which can't
compute floating point infinity by default, but it fails for those
which do (GCL, ECL) -- after computing floating point infinity, Maxima
tries to convert it to bigfloat (I don't know what's the purpose of
that) and fails.

The code of interest is in SIGN1 (src/compar.lisp) at lines 1243--1259.
I guess one way to fix it is to account for non-finite floats.
Can someone fix it?

best

Robert Dodier