[mattijn]

Summary

Follows #275, the other way: Topology.to_polars() gives a Polars DataFrame, the id and the properties as columns and the geometry as WKB in the column geometry, as polars-st stores it.

import polars_st as st

gdf = st.GeoDataFrame(topo.to_polars())

• With a CRS that has an EPSG code, of the input (such as the SRID of polars-st, or the CRS of a GeoDataFrame) or given with crs=, the WKB is EWKB with this code as SRID, so polars-st knows the CRS. Without one it is plain WKB.
• The options are those of to_gdf(): crs, validate, winding_order and object_name; the geometries are the same as those of to_gdf().
• The ids that go in come out: the column id holds the ids of the features, from the index of a GeoDataFrame, the ids of GeoJSON features, or a column id of a Polars DataFrame. That last one is new on the input side: a column id of a Polars DataFrame now gives the ids of the features, as the index of a GeoDataFrame, instead of being a property (so to_json and to_gdf keep them too). Duplicate ids raise as in a FeatureCollection; with ignore_index=True the features are numbered.
• A feature whose geometry collapsed on the grid (Drop rings that collapse on the grid and keep the feature #257) has null as geometry; a property missing on some features is null there.
• Neither Polars nor polars-st is imported by topojson until to_polars() is called; the tests build and read the WKB with shapely, so they run without polars-st.
• A section .to_polars() on Retrieval data types, whose example runs while the site is built.

🤖 Generated with Claude Code

[github-actions]

Benchmark result

These PR changes are benchmarked against five files (see tests/benchmark_compute.py for what each one tests), compared with PyPi's latest release and the repository's main branch.

[github-actions]

Benchmark result

These PR changes are benchmarked against five files (see tests/benchmark_compute.py for what each one tests), compared with PyPi's latest release and the repository's main branch.