Handling of "bbox" and "crs" in geo metadata for "proj:geometry"

[scottyhq]

There's currently a difference in how rustac-py versus stac-geoparquet write geo-metadata. Specifically, rustac-py writes out a bbox into geo metadata and does not set "crs":null, which can confuse downstream tools like GDAL

stac-geoparquet:

        "proj:geometry": {
            "encoding": "WKB",
            "geometry_types": [],
            "crs": null
        }

see also: https://github.com/stac-utils/stac-geoparquet/blob/8dca071c146effd5d71392ddb5e6292321d4bb55/stac_geoparquet/arrow/_to_parquet.py#L167-L179

rustac-py:

        "proj:geometry": {
            "encoding": "WKB",
            "geometry_types": [
                "Polygon"
            ],
            "bbox": [
                239993.99966284665,
                4149993.9999659467,
                780005.999984643,
                4240006.000035326
            ]
        }

[gadomski]

@scottyhq I just chatted with Kyle, and we agree that the idea of a null crs doesn't make a ton of sense b/c you can't do any operations on the columns w/o a crs. I think I'd prefer to just not declare proj:geometry as a "official" geoparquet geometry column if crs isn't set. Does that make sense to you?

[scottyhq]

Sounds good to me. I first noticed this comparing unfamiliar behavior with GDAL where it was recognizing both geometry and proj:geometry even with "primary_column": "geometry"... maybe that's actually a GDAL issue.

In any case, I agree "official" geometry columns should have an unambiguous single CRS.

[gadomski]

Releasing is blocked temporarily on what I think is a compiler regression (https://github.com/stac-utils/rustac-py/actions/runs/17974805205/job/51125701324#step:5:962) but if you wanted to try out stac-utils/rustac#808 to see if it fixes things, uv pip install git+https://github.com/stac-utils/rustac-py

[gadomski]

Ok, released in v0.8.3, let me know if that helps: https://pypi.org/project/rustac/0.8.3/

[gadomski]

Going to close as fixed, please re-open if that isn't the case!