Reduce_temporal throws MetadataException after load_stac

[VictorVerhaert]

When using load_stac, the python client tries to read metadata from the provided stac. The only metadata currently being read is bands, so no spatial or temporal metadata.

When the python client fails to read metadata from the stac, metadata is simply set to None.
These lines check is metadata is present and have default behaviour otherwise:

openeo-python-client/openeo/rest/datacube.py

Lines 1350 to 1359 in 36ecc2b

	def reduce_temporal(self, reducer: Union[str, PGNode, typing.Callable, UDF]) -> DataCube:
	"""
	Shortcut for :py:meth:`reduce_dimension` along the temporal dimension
	:param reducer: "child callback" function, see :ref:`callbackfunctions`
	"""
	return self.reduce_dimension(
	dimension=self.metadata.temporal_dimension.name if self.metadata else "t",
	reducer=reducer,
	)

and

openeo-python-client/openeo/rest/datacube.py

Line 1216 in 36ecc2b

metadata=self.metadata.reduce_dimension(dimension_name=dimension) if self.metadata else None,

The problem arises however when metadata IS being read from the STAC. As only bands are read cube.metadata is not None, but it does not contain a temporal dimension resulting in a MetadataException originating for example from:

openeo-python-client/openeo/metadata.py

Lines 271 to 275 in 36ecc2b

	@property
	def temporal_dimension(self) -> TemporalDimension:
	if not self.has_temporal_dimension():
	raise MetadataException("No temporal dimension")
	return self._temporal_dimension

The solution is threefold I think:

1. Not only check if the metadata is not None but also whether it contains a temporal dimension
2. Implement reading a temporal dimension (and spatial) from stac metadata
3. Catch metadata related errors and throw them as warnings (Discussion) as this blocks the execution of a batch job while it would run fine.

[soxofaan]

as discussed, a workaround would be to keep the openeo package below 0.29 for now

[VictorVerhaert]

Another workaround is to add the following:
cube.metadata = cube.metadata.add_dimension("t", label=None, type="temporal")

[soxofaan]

Started with a PR at #568

[soxofaan]

I already merged an initial fix that avoids throwing the MetadataException, so at least that can already be included in the next release

next steps are properly detecting real properties of the temporal dimension

[clausmichele]

@soxofaan @VictorVerhaert this method parses already correctly the dimensions, why not re using that? It seems we are trying to re-implement something that already exists:

openeo-python-client/openeo/metadata.py

Line 413 in 3c6045b

def _parse_dimensions(cls, spec: dict, complain: Callable[[str], None] = warnings.warn) -> List[Dimension]:

[clausmichele]

@soxofaan I created a PR here: #591 . Currently it solves the issue I have, but there are probably other scenarios that could be covered (currently it makes only a difference for STAC Collections having cube:dimensions available.

[soxofaan]

I think this can now be closed with #591 being merged