Re: BigBinary class

From:

Patricia Shanahan <pats@acm.org>

Newsgroups:

comp.lang.java.help

Date:

Sat, 25 Jul 2009 09:00:52 -0700

Message-ID:

<I-GdnQPHRNSgsfbXnZ2dnUVZ_tqdnZ2d@earthlink.com>

Jeff Higgins wrote:

Patricia Shanahan wrote:

Jeff Higgins wrote:

Ken T. wrote:

On Fri, 24 Jul 2009 11:29:19 -0400, Jeff Higgins wrote:

Is anyone aware of a BigBinary class? It would be like BigDecimal, but
for base-2 numbers.

Depending on what you want this for, you might also consider
BitSet. We can't really tell what features of BigDecimal you want
from your post. Could you provide more information?

Hi,
yup thanks. I've been reading about fixed-point arithmetic.

Among many others:
Fixed-Point Arithmetic: An Introduction
<http://www.digitalsignallabs.com/fp.pdf>
Optimizing Fixed Point (FP) Math with J2ME
<http://www.devx.com/Java/Article/21850>
IAppli Fixed Point Math Library
<http://www.beartronics.com/imode/fplib/index.html>

As suggested by Patricia Shanahan, this would be more
a convenience class.
As suggested by rossum, this is mostly a learning exercise.
So far no one has responded, "Sure use woodie's BB class...,
so Eric Sosman's suggestion seems on target. I'm not sure about
"without much trouble" knowing myself as I do.

[Maybe there is something I'm missing,
(not understanding)with BigDecimal/Integer]

Anyway, I think what I'd like to see is something like:

enum BRep{ONES,TWOS,NATURAL;}
maybe BinaryMathContext extends MathContext

BigBinary(byte[] value, int width, int radix, boolean signed, BRep r)
BigBinary(int value, int width, int radix, boolean signed, BRep r)
BigBinary(BitSet value, int width, int radix, boolean signed, BRep r)
etc. ...

arithmetic methods
conversion methods
properties methods
maybe scaling, bit manipulation methods?

Unless you have major scaling requirements, why not just use BigDecimal,
with a wrapper class that provides additional construction methods?

Well, mainly because I've not figured out how to use BigDecimal to:

byte[4] (or int for that matter) is to represent, let's say;
a signed, fixed-point number
with (1 sign bit + 15 integer bits + 16 fraction bits)

BigBinary would be overkill for that. BigDecimal does a form of decimal
floating point arithmetic. Why not use an int, and write some methods to
do the fixed point arithmetic? Most methods will just use Java's normal
2's complement arithmetic.

Also, consider whether double meets your needs. It can exactly represent
any number you could represent in a fixed point binary type with no more
than 53 bits total for the "integer" and "fraction" bits.

do arithmetic with these numbers...
store result back in 4 bytes.

For example only.

What are major scaling requirements v minor scaling requirements?

The distinction I should have made is between rules that need to be
enforced during arithmetic vs. rules that only need to be enforced
during construction and conversion.

Patricia