SRFI 144: fl-least should not be negative

[jpellegrini]

Hello,

Chibi implements fl-least as -DBL_MAX (lib/srfi/144/math.stub, line 46), but the SRFI says it should be the smallest positive finite flonum (so it's not -DBL_MAX, but rather DBL_TRUE_MIN).
However: this breaks compatibility with OpenBSD, since DBL_TRUE_MIN isn't there yet.
Using DBL_MIN would discard subnormal numbers which are less than DBL_MIN (and the SRFI recommends DBL_TRUE_MIN).

Maybe relevant:

• I see Gauche forces the encoding of a flonum with an ad-hoc procedure,

(define-constant fl-least     (encode-float '#(1 -1074 1)))

(It's in ext/scheme/flonum.scm, line 123)

• I have used DBL_TRUE_MIN in STklos, but it won't compile on OpenBSD.

• The Chicken egg, srfi-144 seems to use DBL_TRUE_MIN also.

[jpellegrini]

If you trust that all supported platforms will always use IEEE 754-encoded floating numbers, one way to get DBL_TRUE_MIN is to make a properly-sized (32 or 64-bit) unsigned integer equal to one. That would give you a zero exponent, zero sign bit and a one in the mantissa (which is exactly what DBL_TRUE_MIN should be).

Or you could take DBL_MIN and keep dividing it by 2.0 until it becomes zero. The last non-zero number should be DBL_TRUE_MIN (this is what I think Sagittarius does, and what I have included in a PR for STklos).