Precision measures how much significant detail a numerical representation preserves. Floating-point types store approximations because infinitely many real values must fit into a finite number of bits. section 3.3
The surprising multiplication result
A developer may expect float32 multiplication of 1.0001 by itself to print 1.00020001. Instead, it commonly prints 1.0002 because the stored value is already an approximation. Increasing the representation’s available detail can improve accuracy, but cannot make floating-point arithmetic exact. section 3.3
Go floating-point values follow IEEE-754: sign, exponent, and mantissa bits determine the stored approximation. float32 allocates fewer mantissa bits than float64, so its representation can lose more detail. section 3.3
Precision affects more than the initial conversion:
- Avoid treating == as a universally reliable floating-point comparison; compare the difference instead. section 3.3
