Currently Magpylib converts everything into float64.

At the bottom levels tests with dtype=int inputs fail because of typecasting. (specifically casting a float 0<f<1 into an dtype int array results in a lot of zeros.

however, with numpy py float32, float64 and float128 available one should consider to allow for different input types, and adjust the output types accordingly.

Reason for using float128: While all expressions are explicit, near undetermined forms they have limited precision
Reason for using float32: speed