Running UDF on Vector Cube

[soxofaan]

reported by @mbuchhorn :

• loading vector cube with connection.vectorcube_from_paths (which uses load_uploaded_files process)
• run UDF on that cube

fails with

  File ".../openeo_driver/ProcessGraphDeserializer.py", line 1491, in run_udf
    data = data.to_legacy_save_result()
  File ".../openeo_driver/datacube.py", line 366, in to_legacy_save_result
    if cube.dims == (self.DIM_GEOMETRIES, "t"):
AttributeError: 'NoneType' object has no attribute 'dims'

[soxofaan]

apparently, the VITO backend now has 3 non-official processes to load geojson data from url:

• read_vector -> DelayedVector
• get_geometries -> DelayedVector
• load_uploaded_files -> DriverVectorCube

running run_udf on DriverVectorCube is not properly supported yet, so it might be better to use one of the others for now.

also related: Open-EO/openeo-processes#322 which recommends to use load_geojson to load geojson from URL

[soxofaan]

hmm, I tried to set up a POC based on read_vector/DelayedVector, but that doesn't seem to be working either:
DelayedVector strips out all properties along the way (keeping only the geometries):

• before passing the vector data to run_udf: so you can not access properties from your intput data
• after run_udf, e.g. in save_result: so you can not access properties that were added in the UDF

[soxofaan]

I iterated on PR Open-EO/openeo-api#200 and merged that in f83d1cf which adds initial support for doing apply_dimension with a UDF on a vector cube

[soxofaan]

Some related API/process questions/discussions:

• load_geojson issues openeo-processes#448
• How to run a UDF on vector cube to access geometry? openeo-processes#449
• Support load_geojson and load_url #211
• Vector cube design: property preservation when using aggregate spatial openeo-processes#466