It's OK to compare floating-points for equality

4 days ago

Floating-point epsilon comparisons are often problematic because they are arbitrary, lack transitivity, and can lead to inconsistent program behavior.
Epsilon comparisons are usually unnecessary; better solutions exist, such as using fixed-point numbers, integer arithmetic, or exact comparisons (x == y).
Examples demonstrate that epsilons are not the best solution in cases like game unit movement, slerp, vector length, solving linear systems, ray-box intersection, and geometric predicates.
There are specific cases where epsilon comparisons are acceptable, such as in visualization for filtering close points or in testing mathematical libraries.
The key takeaway is to think critically about why floating-point comparisons are needed and to choose appropriate solutions based on the problem context, rather than defaulting to epsilons.

Hasty Briefsbeta