Here, our precomposition layer renders the precomposition myprecomp at a size of 256x256. This means we need to encode the precomposition, then scale it down to a size of 256x256. This proves difficult because we scale by multipliers, not absolute sizes, and we don't know the size of the precomposition, neither before nor after encoding it as a scene. It is not clear if we need the bounding box of the control points (which would be easy), or the bounding box of the rendered scene, which would be difficult.
I was mistaken, the width and height of the precomp layer aren't for scaling, but for clipping. This is more evident with an ellipse than a rectangle...
Consider
Here, our precomposition layer renders the precomposition
myprecomp
at a size of 256x256. This means we need to encode the precomposition, then scale it down to a size of 256x256. This proves difficult because we scale by multipliers, not absolute sizes, and we don't know the size of the precomposition, neither before nor after encoding it as a scene. It is not clear if we need the bounding box of the control points (which would be easy), or the bounding box of the rendered scene, which would be difficult.