The way of cgmath

[kvark]

I believe that cgmath went in the wrong direction lately. It got more traits to consider that are not as much graphics related as they are purely mathematical (EuclideanSpace, InnerSpace, etc). I don't think that's the way cgmath needs to evolve itself - we already got nalgebra as a highly abstracted (non graphics-focused) library. Instead, I'd like to see cgmath being simpler and having next to no traits.

cc @brendanzab

[msiglreith]

I actually like these traits as they allow to abstract over functionality and don't restrict yourself to either 2d or 3d. For real-time rendering isn't needed very often imo, but for e.g physically based animations, it's quite nice.

With 'simpler' it's meant 'simpler for users' or implementation side?

[kvark]

@msiglreith Good point. It appears that 2d/3d abstraction would be sacrificed if theses traits are removed. What do you mean by "physically based animations" btw?

With 'simpler' it's meant 'simpler for users' or implementation side?

Simpler for documentation, no need to include a magical prelude, and the implementation may become more readable, even if not with less code.

[msiglreith]

What do you mean by "physically based animations" btw?

Mostly rigid/soft-body physics and fluid dynamics (in the computer graphics domain).
E.g having to write a SPH implementation for 2d and 3d at the same time due to the trait abstraction saves a lot of time (mainly debugging).

[brendanzab]

Thanks for the issue, and good points raised! Sorry for the lack of response - I have been on holiday and just got back today!

[brendanzab]

Personally I'm more a fan of using mathematical structures as a basis for traits, hence why I pushed it in cgmath. It has been really nice to see @sebcrozet following this lead in nalgebra, and pushing it much further than I did. That said, it might be more helpful for the community to have a greater point of difference between the two libraries, so perhaps it might make sense to unwind some of these changes in cgmath. 🤔

I would like to hear from those who are in favour of the mathematical traits, and whether nalgebra might be a viable alternative for them. I would also have to consider switching to nalgebra, and transferring ownership of cgmath to another maintainer, or the community. I don't rush into anything just yet. I hope that makes sense!

[kvark]

@brendanzab it's almost like your paragraphs should be moved out to separate comments, given they induce entirely opposite emotions to many of us.

[LaylBongers]

I personally think cgmath strikes a good balance, where nalgebra goes to a far extreme that makes it harder to use in projects. I always feel like the best approach isn't in going to one of two extremes but finding a practical middleground.

[brendanzab]

Yeah, that's what I was kind of aiming for. Perhaps we could try implementing some method aliases that forward to the trait versions. Then people could use the direct methods initially, then level up to use the more generic traits if they need them.

[MaikKlein]

Something like this? I think that would be the best approach. (At least for the fundamental operations)

[brendanzab]

Yup - that's the one!