On: How to average floating point numbers (2015)

[johnbcoughlin]

> the only real question is how much error you are willing to tolerate

So, you are, in fact, gonna need an algorithm for summation that controls that error, if you're not willing to tolerate very much.

[jfrunyon → johnbcoughlin]

Yes. Similarly, if you design something with total disregard to optimization, it will likely need optimization.

YAGNI is a prediction, not a rule. Just like premature optimization, alternatives to floating point - or using longer floating point types - have real costs, which must be balanced against the needed precision.