4 ms·

Part of the reason for writing the slow version is that you need to profile it to find out where the problems actually are, because intuition is not a very good

Part of the reason for writing the slow version is that you need to profile it to find out where the problems actually are, because intuition is not a very good guide. Some of the worst bottlenecks seem tiny but happen so frequently they add up, while a lot of things that sounded slow don’t actually matter.

carlmr 6y ago
That's also the thing with premature. I would agree with you that premature is before you know what's slow. But if you have a CI Profiling run showing a few hotspots it's not premature anymore. Still people quote this adage. Further there are things you should know are usually performance issues, that I would expect you to avoid entirely, except with good reason. Using O(N²) or worse algorithms on big collections when there are much faster alternatives in the standard library of your language for example. I don't think these good habits are premature optimization.