CollectionMetadata -> CubeMetadata?

[soxofaan]

openeo-python-client/openeo/metadata.py

Lines 198 to 200 in cbd4053

	# TODO: "CollectionMetadata" is also used as "cube metadata" where the link to original collection
	# might be lost (if any). Better separation between rich EO raster collection metadata and
	# essential cube metadata? E.g.: also thing of vector cubes.

"CollectionMetadata" got its name when data cubes where still called "ImageCollection" (instead of "DataCube"). But now that is getting a bit confusing: in openEO, "collections" is the source data one loads with load_collection, while "cubes" are the intermediate data structure one works with. So "CollectionMetadata" should better be called "CubeMetadata".

Also, CollectionMetadata contains some specific dimension metadata parsing logic (cube:dimensions/eo:bands), which is closely tied to the STAC description of collections. I think it would be better to separate this from the essential cube metadata logic.

[soxofaan]

I think we should change it to:

• CubeMetadata: just the bare metadata essentials for a data cube: interface to interact with the cube dimensions and their labels (if available), maybe some resolution info too. Methods allow to construct a new CubeMetadata instance from an existing one (e.g. reduce_dimension eliminates a dimension)
• CollectionMetadata (subclass of CubeMetadata): interface to interact with all original collection (STAC) metadata. Calling a "process" like reduce_dimension produces a CubeMetadata

By making CollectionMetadata a subclass of CubeMetadata we can stay backward compatible I think, because the interface of CollectionMetadata would not really change

[jdries]

ok, makes sense!

[soxofaan]

Another reason for better abstraction: load_stac should produce a data cube with metadata (#425) , but that is not the same as collection metadata.

[VictorVerhaert]

openeo-python-client/openeo/metadata.py

Lines 231 to 242 in 498c4c7

	def _clone_and_update(
	self, metadata: dict = None, dimensions: List[Dimension] = None, **kwargs
	) -> CubeMetadata: # python >= 3.11: -> Self to be more correct for subclasses
	"""Create a new instance (of same class) with copied/updated fields."""
	# TODO: do we want to keep the type the same or force it to be CubeMetadata?
	# this method is e.g. used by reduce_dimension, which should return a CubeMetadata
	# If adjusted, name should be changed to e.g. _create_updated
	# Alternative is to use an optional argument to specify the class to use
	cls = type(self)
	if dimensions == None:
	dimensions = self._dimensions
	return cls(metadata=metadata or self._orig_metadata, dimensions=dimensions, **kwargs)

Currently functionality does not change due to this function.
As this function checks the type and returns an object of the same class, functions like reduce_dimension (who use this function to update a metadata object) will still return a CollectionMetadata object if that was the original object.

Changing this to return a CubeMetadata changes functionality and fails some existing tests.

An alternative is to provide an optional argument to set the expected returned class, however this feels more like a quick fix and might be confusing in the future.