WIP: GeoParquet reader - Githubissues

kylebarron commented 1 year ago

I started to take a stab at this to resolve https://github.com/georust/geozero/issues/37 but ran into a few issues/questions:

I put process_geoarrow_feature_chunk in the geoarrow mod, because the parquet reader's result is automatically geoarrow (of the wkb variant). In theory we could implement GeozeroDatasource for geoarrow directly as well to convert from an arrow table, but that might be less useful.
It seemed not possible to write an impl for GeozeroGeometry in addition to GeozeroDatasource, because GeozeroGeometry's process_geom only provides a non-mutable reader, and the parquet FileReader needed a mutable reference. Or maybe I'm missing something there.
I'm not sure the best way of doing arrow -> geozero feature objects. Arrow2's "chunk" object is a vec of Array trait objects, so we need to do downcasting to get values out. Is it possible to do that downcasting outside of the loop so that we don't have to downcast on every iteration?
I assume geozero has a record-only API? I.e. I can't process all the data column by column, it needs to be row-by-row?

pka commented 1 year ago

Rebasing should solve the failed CI test.

nyurik commented 1 year ago

I did the merge, but i have no domain knowledge about this PR

nyurik commented 1 year ago

@kylebarron & @pka please update on this PR's path, if it needs more changes, or should be closed or merged

kylebarron commented 6 months ago

Closing this because it's been (mostly) implemented in https://github.com/geoarrow/geoarrow-rs, with transitive bindings into geozero.

georust / geozero