Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.20%. Comparing base (0731edd) to head (0637dac).
⚠️ Report is 3 commits behind head on main.

@@            Coverage Diff             @@
##             main     #951      +/-   ##
==========================================
+ Coverage   92.19%   92.20%   +0.01%     
==========================================
  Files          49       49              
  Lines        7561     7572      +11     
==========================================
+ Hits         6971     6982      +11     
  Misses        590      590              

Hi @keller-mark, this is ready for review, correct?

Yes, thanks for the reminder, I have just updated the status

Thanks! Doing some light adjustments on the PR. I will push soon. Meanwhile you can find here a benchmark in Python for read-write operations with the new encoding. https://github.com/giovp/spatialdata-sandbox/blob/main/notebooks/benchmark_geoparquet_encoding.ipynb

Take home message: write operations are only slightly slower with geoarrow, but read operations are generally faster. The benchmark is done in pure geopandas and spatialdata. spatialdata has some overhead that disappears when the data is large.

Final changes are up. Key points:

• I added some global settings to allow reducing the number of arguments to pass to functions.
• I will keep the default as WKB. We can experiment and eventually change.
• I keep the shapes format unchanged since APIs to read back share the same syntax
• There is an edge case with geoarrow: mixed types force a coercion to the more general type. In practice this means that mixed columns polygons+multipolygon are written to disk as a column of multipolygon (see tests). Anyway, since downstream applications should expect the possibility of having multipolygons in the shapes layer, and in particular multipolygons containing a single polygon, this does not introduce a breaking change.