JAXB, BigDecimal or double?

Question

I working on different web-services, and I always use WSDL First.

JAXB generates for a Type like:

<xsd:simpleType name="CurrencyFormatTyp">
    <xsd:restriction base="xsd:decimal">
        <xsd:totalDigits value="13"/>
        <xsd:fractionDigits value="2"/>
        <xsd:minInclusive value="0.01"/>
    </xsd:restriction>
</xsd:simpleType>

a Java binding type BigDecimal (as it's mentioned in JAXB specification).

When I then do some simple arithmetic operation with values of the type double (which are stored in a database and mapped via hibernate to the type double) I run into trouble.

<ns5:charge>0.200000000000000011102230246251565404236316680908203125</ns5:charge>        
<ns5:addcharge>0.0360000000000000042188474935755948536098003387451171875</ns5:addcharge>
<ns5:tax>0.047199999999999998900879205621095024980604648590087890625</ns5:tax>
<ns5:totalextax>0.2360000000000000153210777398271602578461170196533203125</ns5:totalextax>

What would be the right way?

1. Convert all my values into double (JAXB binding from BigDecimal to double)
2. Hibernate mapping double to Bigdecimal

and do all my arithmetic operations in one object type.

Answer 1

You don't ever want to use floating point formats (such as double and float in Java) for currency operations, because they have limited precision and have been designed to hold values that are somehow derived from measuring (in which case they aren't absolutely precise to begin with, in which case the limited precision is less of a problem).

What Every Computer Scientist Should Know About Floating-Point Arithmetic is the article on the topic. It's a bit heavy on math, but it really helps to understand that (Alternatively, the article linked to by Michael Borgwardt is much easier to understand and still demonstrates/explains the problem).

To avoid this kind of problems, make sure that you use BigDecimal exclusively in your code and that all external storage/transfer points use fixed-point/arbitrary-precision values as well (i.e. your database shouldn't store floating point numbers either).

Answer 2

• Read the Floating-Point Guide
• Never use double or float for money amounts
• Use BigDecimal instead, that's exactly what it's for