InnerSpace::normalize produces a Vector of NaNs when called on a zeroed Vector

[mitchmindtree]

This makes sense of course as we divide by the magnitude! However it can be a little tricky for newer users to hunt down the source of these NaNs.

Processing solves this by first checking if the magnitude is zero and if so returns [0, 0]. This is a little more expensive but perhaps also a little more user-friendly?

Perhaps we could leave the normalize method as is and instead add a new method which does this check? Currently it looks like the only solution for users is to 1. check that v.magnitude() > 0.0 before calling normalize() (this means that magnitude ends up gettinng calculated twice) or 2. implement their own normalize function.

cc @JoshuaBatty

[brendanzab]

Might be interesting to compare to what other maths libraries do with this. Perhaps some documentation on this behaviour could also help. 🤔

[richard-uk1]

If you use magnitude2 you avoid the sqrt, otherwise yeah could add a try_normalize

Or alternatively you could add an example to the docs using is_finite.

[thefakeplace]

this got me while I was coding

I picked the cgmath because it’s marketed as “for games/CG” and this behavior is totally useless in games/CG

adding checked_normalize is one option, but IMO, should just change magnitude to return 0 in this case. it aligns better with the project goals, and I seriously doubt anybody is relying on this behavior for anything— pragmatism is cool in “games/CG”.