From: Bil Kleb <W.L.Kleb@...>
Date: 2002-02-06T07:40:58+09:00
Subject: Re: smallest non-zero number (and other limits)

Yukihiro Matsumoto wrote:
> 
> Hi,

Hello.

>  (a) <float.h> is already checked by AC_HEADER_STDC in configure, so
>      that you don't have to add AC_CHECK_HEADERS.  You just have to
>      use STDC_HEADERS instead of HAVE_FLOAT_H.

Roger.

>  (b) Ruby only has double precision float (Float class), why do we
>      need these FLT_* constants?

Excellent point; no need for the LDBL's either...

So this narrows it down to ROUNDS, RADIX, MANT_DIG, DIG, MIN_EXP, MAX_EXP,
MIN_10_EXP, MAX_10_EXP, MIN, MAX, and EPSILON.  Although it would seem
that folks would only typically want access to MIN, MAX, EPSILON,
DIG, MANT_DIG, and perhaps a spacing function?

(The spacing function would be a generalization of EPSILON which takes an
argument for the origin as opposed to EPSILON's "1".  In other words, what is
the "epsilon" about "3.45" or "1.45e+34".)

>  (c) I still think we should do something other than just ommiting on
>      non ANSI platforms if we do check.  Or just abandon non ANSI C
>      platforms.

"[D]o something" as in compute them in place, search for another header
which might have them, or give the user some sort of a failure message
when trying to access these when Ruby is compiled with a non-ANSI compiler?

Computing them for the general case would require computing the radix
of the exponent representation. Does anyone have a bullet-proof radix
algorithm lying around?

Regards,
--
Bil