Generic parameter for unit's underlying float

[tower120]

Is it possible to make something like Mass<T=f64> instead of just Mass?

For example, I would like to work with f32 precision mass, volume, etc...

[sunsided]

I tried to give that a spin here using num-traits constrained generics but it ended up in one difficult spot: When T can be anything, the automatically implemented commutative implementations for Mul and Add fail, particularly with Duration:

impl<T: num_trait::Float> Mul<T> for Duration and impl<T: num_trait::Float> Mul<Duration> for T cannot coexist because a user of the create may implement num_trait::Float for Duration, leading to a conflicting trait implementation. That's blocked by RFC 1210.

Resolving this by making operations non-commutative could work, but then the specializations for f64 would have to be implemented manually.

I also thought of feature-gating the implementation, defaulting to <f64> specializations if generic types are not wished for - that's a bridge to be crossed once I'm there.