As well as being able to hold decimal fractions exactly, BigDecimal has a number of capabilities that are absent from the primitive binary floating-point types float and double. First, a BigDecimal can hold very big values, both in terms of exponent range and in terms of digits; in contrast, the primitive floating-point types have a fixed precision and limited exponent range; see Table 2.
|float||2-149 to 2 127 ~= 10-45 to 10 38||24 bits ~= 6 to 9 decimal digits|
|double||2-1074 to 21023 ~= 10-324 to 10308||53 bits ~= 15 to 17 decimal digits|
|BigDecimal||10-2147486647 to 102147483648||1 to billions of decimal digits|
Table 2: Properties of numeric types.
BigDecimal arithmetic operations allow full control over both the rounding mode and precision of the result. Operations on float and double always use the round to the nearest even rounding mode and the precision of the result is fixed. The string representation of a BigDecimal transparently denotes the number’s value. Printing out float and double numbers is more subtle. The double floating-point value that prints out as “0.1” is not actually equal to 1/10; it is just the double value closest to 1/10.The primitive floating-point types have their own advantages. Expressions on those types can be written with infix operations such as “+”, “*”, and “/”. Today, binary floating-point enjoys near-ubiquitous hardware support, so operations should be quite fast. The primitive types also take up less space than BigDecimal objects.
In summary, if raw performance and space are the most important factors, primitive floating-point types are appropriate. If decimal values need to be represented exactly, high-precision computation is needed, or fine control of rounding is desired, only BigDecimal has the needed capabilities.
(from Dr. Dobb’s Journal #362)