int vs long vs double vs BigDecimal for Money

£19.99 has a decimal point, so double looks right, and double is the type that makes invoices a penny out.

Money needs a type that stores amounts exactly. double does not, so the choice is between a whole number of pence in an int or a long, and BigDecimal, a class that stores decimal numbers exactly and carries its own number of decimal places.

An invoice line says £19.99. It has a decimal point in it, so double looks like the answer, and double is the answer that produces invoices a penny out.

Four ways to hold an amount of money, on exactness, range and effort
Type	Exact?	Range	Costs you
`double`	No: binary fractions cannot hold 0.1	Enormous	Nothing, until the totals disagree
`int` of pence	Yes	Up to about £21 million	Dividing by 100 everywhere you display it
`long` of pence	Yes	More than any real ledger needs	The same division, and a wider field
`BigDecimal`	Yes, and it carries its scale	Unbounded	An object per value, and `compareTo` instead of `==`

The decision rule. If the number is money, never use double or float. Use long of the smallest unit when the arithmetic is addition and subtraction and you control both ends. Use BigDecimal when there is division, a tax rate, a currency with a different number of decimal places, or an accountant who will check.

java

BigDecimal price = new BigDecimal("19.99");
BigDecimal vat = price.multiply(new BigDecimal("0.20"));

System.out.println(vat.setScale(2, RoundingMode.HALF_UP));

the string constructor, not the double one: new BigDecimal(19.99) brings the problem back

text

4.00

For anything that is not money the rule relaxes: double is the right type for a temperature, a distance or a physics simulation, where being within a fifteenth decimal place is not a defect.

← → move through sections and cards