Advice on BigDecimal vs. double

[Dave Van Even]

howdy
I'm in the progress of modeling a system for a financial company and I was wondering wether I should use BigDecimal or double for financial data.
From what I've read by now you can't really trust double to store financial numbers but BigDecimal has some downsided too:
- it's slow (MUCH slower)
- can't use normal operators on them; have to resort to method calls
- only basic operations on BigDecimals are available

One possible improvement over BigDecimal would be to use IBM's BigDecimal class ( http://www2.hursley.ibm.com/decimalj/ ). It's 'compatible' with java.math.BigDecimal but IBM has added aditional methods and functionality + they say it's 23 times faster. (btw: JSR13 )
So; should I use double ? or BigDecimal ? or IBM's BigDecimal ? what would YOU do ?

any advice on when to use BigDecimal would be welcome too
thanks

[Snigdha Solanki]

Calculations with BigDecimal are slower as compared to those with double but if accuracy and precision are important than you should go for BigDecimal rather than double.

[Thomas Paul]

Why not just use ints? Multiply all your numbers by 100 so that you carry cents. Then just worry about converting them to a decimal format when you need to store or display them.

[Ilja Preuss]

Originally posted by Thomas Paul:
Why not just use ints? Multiply all your numbers by 100 so that you carry cents. Then just worry about converting them to a decimal format when you need to store or display them.

Well, afaik most financial applications need to work on the thousandth part of a cent or something, but you should be able to use int or long as long as you don't need *arbitrary* precision. (Don't forget to encapsulate it in its own Money class!)

[Dave Van Even]

thanks so far everybody
I'm not so keen on implementing my own 'Money' class The thing is I'm using these things with EJB and I'm wondering wether it'll all work...
Perhaps the BigDecimal isn't so bad after all. I know that works for shure...
Pitty java doesn't have operator overloading I remember stuff back from the C++ days, it can really help you out in some situations. I do understand SUNs decision not to implement operator overloading though..

[Dirk Schreckmann]

What do you think could be accomplished with operator overloading that cannot be otherwise accomplished?

[Jim Yingst]

Well, you could write things like

    x = (a + b)/(c - d);

where all variables are BigDecimal. Instead we have to use things like

    x = a.add(b).divide(c.subtract(d), BigDecimal.ROUND_HALF_UP);

If you're using lots of formulas, operator overloading can give you much better readability. Of course it can also obfuscate the heck out of other examples.

[Thomas Paul]

C# has operator overloading.