Make plane generic over pixel component type

[rom1v]

In order to support both u8 and u16 for plane components, make the Plane structure generic over the component type. As a consequence, many other structures and functions also become generic.

Some functions are not u8-compatible yet, although they have been make generic over the component type to make the compilation work. They assert that the size of the generic parameter is 16 bits wide.

For this reason, the root context structure is unconditionally created as Context<u16> for now.

[coveralls]

Coverage decreased (-0.3%) to 79.812% when pulling 32c19dd on rom1v:generic into baec5d3 on xiph:master.

[tdaede]

FrameInvariants becoming generic really adds a lot of noise :/ this is because of the reference frame structure correct?

[rom1v]

this is because of the reference frame structure correct?

Yes, just because of its field:

pub rec_buffer: ReferenceFramesSet<T>,

[tdaede]

Do you think it would make sense to do some dynamic dispatch over that field? Or does that make it even more complicated?

[rom1v]

Do you think it would make sense to do some dynamic dispatch over that field?

That would mean erasing its concrete type behind a trait:

pub trait RFS {}

impl<T: Pixel> RFS for ReferenceFramesSet<T> {}

pub struct FrameInvariants {
    // …
    pub rec_buffer: Box<dyn RFS>,
    // …
}

But then, from a FrameInvariants, we could not expose the Frames of the rec_buffer (because its concrete type would be lost), so every action on the ReferenceFrameSet<T> would need to be implemented through methods on the RFS trait.

It would be impractical IMO.

In fact, ReferenceFrameSet needs to updated with a FrameState<T>, which is not possible if concrete type of the ReferenceFrameSet is not known (unless we also erase the FrameState parameter…).

[tdaede]

I see. I guess we could also split out ReferenceFrameSet to be its own parameter. But that doesn't make many things better.

[tdaede]

lu_zero can you check how the api.rs changes interact with crav1e?

[lu-zero]

They would break everything, but I can fix after it lands.

[lu-zero]

I need to check what works better compared to dyn dispatch, but we could also at a certain point of the chain use just a proxy like this

enum Something {
   Something16(SomethingInner<u16>),
   Something8(SomethingInner<u8>
}

With From and Deref made to unwrap to the actual type.

[tdaede]

I still wish we could have FI not be per-depth but I think this is currently the best option.