🚀 go-pugleaf

> Path:
newsbf01.news.aol.com!newstf01.news.aol.com!uunet!in1.uu.net!news.bluesky.
net!news.mathworks.com!newshost.marcam.com!usc!howland.reston.ans.net!news.
nic.surfnet.nl!sun4nl!cwi.nl!dik
> From: dik@cwi.nl (Dik T. Winter)
> 
> ...
> Stringized versions of comparison are much better.  But there is again
> one caveat; I do not know the exact requirements on the behaviour of
> sprintf; are there any?  Is it allowable that sprintf displays identical
> strings for f-p numbers that are different?  (I think it is BTW.)  In
> that case it is not a proper substitute.
> 
> The portion of C code I saw (using frexp) is appears to be correct for
base
> 2 and 16.  But for other bases there is something uncertain left (the
> f-p calculations in there worry me).  But when converting from float to
> int, using frexp is correct.

Since almost all floating point implementations of C's "float", "double",
and "long double" use base 2 or 16, the example would avoid floating point
most of the time.  I guess I could have added case statements for base 3,
4, 5, ... up to say, 256?  But what about bases greater that 256 -- use
the "log" functions.  Even so, you only need to call the function *once*
per program -- only marginally longer on machines with oddball bases.

Your solution requires floating point operations as you convert the
floating point to a string.  Furthermore, you need to convert to a
string for *each* comparison.  Comparing your solution to mine:

	Mine: Calling "log" one per program (the "log(2)" can be replaced
	by a constant) and *only* for machines with odd bases.

	Yours: Calling "sprintf" for each floating and integer conversion
	and comparison.

I don't think your solution offers better performance.

-FF
-------------------------------------------------------------------
(``I only use AOL for reading netnews.'')
Frank Farance, Farance Inc.
E-mail: frank@farance.com, Telephone: +1 212 486 4700
ISO JTC1/SC22/WG14 & ANSI X3J11 (C Programming Language) Project Editor