Make types generic

[bojohnson5]

I’m wondering what it would look like to make the types generic for any floating point number rather than having f64 hardcoded. Would implementing generics using num-traits be viable? That’s a pull request I could work on if it makes sense.

[ethanbarry]

Well, I'm not sure whether algorithms which are stable for 64-bit floats are necessarily stable for 32-bit floats. It seems like stepping down the precision shouldn't cause an issue, but floating point errors can sometimes be counter-intuitive. I guess we'd have to test it!

Unfortunately, I've been quite busy in the last couple of weeks with a new job, and I know @Axect has had a lot on his plate for some time now. :( It sounds like a great feature to have, though. 32-bit arithmetic would allow this library to support hardware without 64-bit FPUs or software emulation.

[bojohnson5]

Cool, I'll see about implementing this and testing it out when I have time, too! Thanks!

[Axect]

Hello @bojohnson5 ,
Thank you for your interest in Puruspe! I'd like to share some context about this.

When I initially developed Puruspe, the primary focus was on implementing special functions with sufficient precision, which is why I chose to work exclusively with f64. However, I completely agree that extending support to f32 through generic implementation would be a valuable addition to the library, particularly for use cases involving hardware without 64-bit FPUs or where software emulation is needed.

As @ethanbarry correctly pointed out, we need to be careful about numerical stability when working with different floating-point precisions. While implementing generics using num-traits is technically feasible, we would need to develop tailored test suites specifically for f32 to ensure the algorithms maintain their stability and accuracy at lower precision.

This would be a significant and welcome contribution to Puruspe. If you're interested in working on this feature, I'd be happy to provide guidance and review the implementation. We should pay particular attention to:

1. Implementing comprehensive test cases for f32
2. Verifying numerical stability across different precision levels
3. Documenting any precision-specific considerations or limitations

Let me know if you'd like to proceed with this enhancement, and we can discuss the implementation details further.