float and double concepts

I'm reading the JLS and dont unerstand some of the concepts related to floating point numbers. What is normalized and denormalized? What is FP-strict? What is value set conversion?
If anyone can help explain this stuff in english I would appreciate it.

Hi Randall,
I had trouble with this as well, from what I understand (and I'm not great at math); there is a standard, IEEE 754, which defines the rules for handling floating point numbers particularly how infinities are to be treated and how high-precision numbers are rounded during calculations.
Java basically guarantees that

The rules for operations on infinities match normal mathematical expectations.
The Java Programming Language: Second Edition
Ken Arnold and James Gosling

The same book gives the rules for inifinities as:
<pre>

x	y	x/y	x%y
finite	0.0	infinity	NaN
finite	infinity	0.0	x
0.0	0.0	NaN	NaN
infinity	finite	infinity	NaN
infinity	infinity	NaN	NaN

</pre>
Where NaN is Not a number and infinity can be POSITIVE_INFINITY or NEGATIVE_INFINITY.
What this boils down to is: floating point arithmetic will never throw an <code>ArithmeticException</code>. So a divide by zero will not cause a Runtime error and underflow/overflow conditions (results that won't fit within 32-bit or 64-bit memory) don't cause runtime errors.
The value set is the range a number can fall within. When a number with many decimals is divided by another number large decimals values which exceed what can be held by 32-bit or 64-bit memory storage result. How the runtime rounds the result to fit within the available memory is also defined in the standard. For the most part, Java rounds toward zero or chops off the portion that won't fit into storage. There are other extended value sets that allow for different rounding mechanisms.
When strictfp is used, Java guarantees that any arithmetic operation will strictly conform to the standard; producing the same result any other system conforming to the standard would produce. If strictfp is not used there are cases when an extended or different value set is allowed whose results may or may not match the result arrived at by strictly following the standard.
Bottom line, unless your doing really intensive and precise mathematical calculations you don't really need to be concerned with infinitesimal differences in calculations.
As for normalized and denormalized from what I can make out a number is normalized if it fits within the standard value set; otherwise it's denormalized (although I'm really iffy on this).
Whew! Hope that makes sense and if I've got anything wrong someone with a better understanding will jump in.
------------------
Jane
The cure for boredom is curiosity.
There is no cure for curiousity.
-- Dorothy Parker
[This message has been edited by Jane Griscti (edited November 01, 2000).]

thanks that helped a lot. also i love the quote!

Your welcome

I swiped the quote from "The Java Programming Language" ... they've got some really good ones.
------------------
Jane
The cure for boredom is curiosity.
There is no cure for curiousity.
-- Dorothy Parker