Will Tebbutt

The best first contribution would be to pick one of your implementations (maybe the `qr` decomposition, pretty sure we don't have that yet) and make a PR to `ChainRules`. Probably...

A version that requires a signature would be preferable imho :) Is this something you've got time to implement?

Hi @FerreolS . Thanks for opening this. I think you should probably consider using `ParameterHandling.fixed` -- this lets you use arbitrary types when you don't want their data to be...

Ahh I see. Unfortunately I'm not sure that ParameterHandling.jl is really the tool for this kind of thing. It's intended use is largely not that you would build an arbitrary...

This would be nice, but I'm not sure how you could achieve this in practice. Do you have any thoughts?

(The reason that we only support open intervals at the minute is because the bijections that we use map the bounds to -Inf and Inf )

I agree with getting rid of the supertyping -- it feels redundant in that I think we really just want a trait-based system. However, I'm not of the opinion that...

> Hmm I'm not sure I see the downside yet. Wouldn't it just be nice if we could just do positive(2.0) * SEKernel() and nobody would bat an eye? This...

Good point. There are definitely presumably more complicated parameter tying settings that you might consider though. e.g. I don't know why you would want this, but something like ```julia kernel(a)...

Oooo I like this `ParametrizedKernel` idea -- feels like it would be very straightforward to implement, and potentially really quite clear to the user.