Rafael Schouten

The type restriction means this is a breaking change, so we will have to bump a breaking version. Wondering if there is a better way to get stability.

Sure, but GPT5 is not a very good programmer ;) We have to follow semantic versioning and make breaking releases if we make breaking changes. But we try our hardest...

I think we just need a PR

I guess we can have a `geometrycolum` keyword with `:geometry` as the default?

~~How does `materialize_traced_array` work? is the idea to get the wrapped `TracedArray` out of `x`?~~ ~~Can that work through multiple layers of wrappers?~~ Ok think I get it. `ancestor` will...

Doesn't that risk losing information from the wrapper? E.g. would `SubArray` be a wrapper?

I meant this kind of method generally, but yes, I assume getting the ancestor through a SubArray would be broken in your example.

It could also do `ancestors(TracedArray, A)` to be more general. That would allow getting closer ancestors like Diagonal with `ancestors(Diagonal, A)` rather than the vector it wraps.

Nothing direct, I'm in a workshop ecologists talking about using it as a simple terrain classifier. It seems like an interesting and alternative to the methods already here, and I...

Yeah, this is stuff that Raster.jl does already with RasterStack/Raster, mostly defined on AbstractDimStack/AbstractDimArray. A DimStack has its own metadata and a NamedTuple of metadata for all the layers, and...