Hacker News Reader

But test against what? This only makes sense for the property based testing, but then random sampling should give the same result up to a statistical margin of error (unless there's some specific pathological case). I suppose this is good for extremely critical routines, kind of like the modified condition/decision coverage safe critical flight software has to undergo.

If you have an inverse function, use that to test all round tripa?

That quickly becomes unfeasible for things that are not single input (and still requires being able to identify a correct result).

ecef_to_lla requires two values, so even if one is willing to take the reduced geospatial resolution of float32, that's still 64-bits of parameter space.

if you do need 1 mm accuracy, including in the Pacific Ocean around 180 E/W then neither float32 nor scaled 32-bit ints are enough.