Reading geoparquet but not using geometry returns ValueError about dask-geopandas HOT 1 CLOSED

Hmm, good question!

So in geopandas, we discussed this on the PR and in the end decided (for now) to error when you specify a column selection without the geometry column (so eg geopandas.read_parquet(.., columns=['pop_est']), which would otherwise result in a normal pandas dataframe and not a geopandas GeoDataFrame).

So under the hood, that is happening here. The

ddf = dask_geopandas.read_parquet("test/test_*.pq")
ddf['pop_est']

gets translated under the hood to

ddf = dask_geopandas.read_parquet("test/test_*.pq", columns=['pop_est'])

to make use of the fact that we can read columns selectively with the parquet file format.

Of course, the user should not directly be aware of this. So indeed, with dask_geopandas, we should somehow be able to fall back to the plain dask reader when the "pushed down" column selection doesn't include a geometry column.

from dask-geopandas.