How many rows are sampled when using DESCRIBE with st_read ?

[mhkeller]

Looking at the docs for read_json and read_csv, a user can pass in the sample_size option to determine how many rows get included for things like DESCRIBE. I don't see an equivalent option for st_read but I can still do DESCRIBE SELECT * FROM st_read('my-map.geojson').

Does that statement read through every row by default or a subset? Would it be possible to include the same sample_size option to be able to match the behavior of the other read functions?

[Maxxen]

Hello!

I don't think there is a definite answer to this, DuckDB just asks GDAL for the schema, and how that schema is derived is driver-dependent. It doesn't seem like the GeoJSON driver has an option to specify this, so we can't expose a sample_size. My guess would be that it probably parses the whole input to detect the types.

[mhkeller]

Makes sense! Thanks!