approx_eq_ulps() ? about nalgebra HOT 2 CLOSED

I think this is a very reasonable change for floating point numbers. I must admit, that I've been using some workaround to achieve something numerically similar in my projects using nalgebra (mostly multiplying epsilon by one of the numbers to compare, but the ULPs seem more efficient and and more adequate for floating point math).

As far as this library is concerned, whose main applications probably use some floating point type, I think we can do it like that. However, considering more abstract mathematical objects we should see how this relates to more general metric spaces.

from nalgebra.

That's an interesting approach. Though I am not sure we should do this by default (i-e. in .approx_eq) instead of the common epsilon-based solution as I think most people will have the latter in mind when using it.

from nalgebra.