Floating-point numbers represent a wide range of magnitudes with limited precision. Go provides float32 and float64 under IEEE-754: float32 uses 4 bytes and about six guaranteed decimal digits, while float64 uses 8 bytes and 15. section 6.2.1
Treat floating-point results as approximations:
- Most decimal fractions cannot be represented exactly; 0.625 is an exception because it combines inverses of powers of 2. section 6.2.1
- float32(123_456_789) prints as 123456792, and float32(0.0123456789) prints as 0.0123456791. section 6.2.1
- Adding or subtracting values with very different magnitudes can lose the smaller value; subtracting 1 from float32(100_000_000) leaves 100 million. section 6.2.3
- Avoid == for approximate results; compare the difference against an appropriate precision range. section 6.2.3
Comparing a computed zero
math.Sin(math.Pi) returns 1.2246467991473515e-16 rather than exactly zero. A tolerance check such as math.Abs(math.Sin(math.Pi)-0) < math.Pow(10, -15) treats the result as indistinguishable from zero. section 6.2.3
fmt.Printf("%.8f", float32(1)/float32(41)*float32(41))
Run the expression to observe how intermediate float32 rounding can affect a result. section 6.2.1
