3. Floating-Point Arithmetic

< Day Day Up >

3. Floating-Point Arithmetic This section concerns arithmetic on the floating-point types: float and double. 3.1. Floating-point arithmetic is inexact Prescription: Don't use floating-point where exact results are required; instead, use an integral type or BigDecimal. Avoid floating-point loop indices. Avoid using the ++ and -- operators on floating-point variables, as these operators have no effect on most floating-point values. Avoid testing floating-point values for equality. Prefer double to float. References: Puzzles 2, 28, and 34; [JLS3 4.2.3], [EJ Item 31], and [IEEE-754]. 3.2. NaN is not equal to any floating-point value, including itself Prescription: Avoid testing floating-point values for equality. This is not always sufficient to avoid problems, but it's a good start. References: Puzzle 29; [JLS 15.21.1] and [IEEE-754]. 3.3. Conversions from int to float, long to float, and long to double are lossy Prescription: Avoid computations that mix integral and floating-point types. Prefer integral arithmetic to floating-point. References: Puzzles 34 and 87; [JLS 5.1.2]. 3.4. The BigDecimal(double) constructor returns the exact value of its floating-point argument Prescription: Always use the BigDecimal(String) constructor; never use BigDecimal(double). References: Puzzle 2.

< Day Day Up >