Hacker News Reader

Comment by BeetleB | Hacker News Reader

If I read you right, you are saying that if you compute a+b, you'll get a different result than if you compute b+a, if a is tiny and b is huge.

I believe this is incorrect. Feel free to show me a counterexample.

View on HN

Not quite a+b, but a+b+c reordered may give different results[1]. Example from Python:

  >>> 0.1+0.2+0.3
  0.6000000000000001
  >>> 0.2+0.3+0.1
  0.6

It's an associativity problem and not a commutative one, however, that is relevant to the original point about folding floating point operations over lists.

[1] https://www.quora.com/Is-floating-point-addition-commutative...

a + b commutes all right, but some-start-value +a +b ... +x does not necessarily.

As rhimenoceros points out, that is a failure of associativity, not commutativity.

Reply on news.ycombinator.com