feat(rust/sedona-functions): Implement ST_GeomFromEWKT

[yutannihilation]

Now that we have item-level CRS, I think ST_GeomFromEWKT can be implemented. Unlike ST_GeomFromEWKB (#490), if I understand correctly, EWKT doesn't need a dedicated parser except for spliting SRID=...; and pass the rest to the existing WKT parser.

It seems to work. I think I still don't understand how the item-level CRS should be handled, so this pull request takes some more time to finish.

> select st_geomfromewkt('SRID=3857;POINT (1 1)');
┌────────────────────────────────────────────────┐
│ st_geomfromewkt(Utf8("SRID=3857;POINT (1 1)")) │
│                     struct                     │
╞════════════════════════════════════════════════╡
│ {item: POINT(1 1), crs: EPSG:3857}             │
└────────────────────────────────────────────────┘

> select st_geomfromewkt('POINT (1 1)');
┌──────────────────────────────────────┐
│ st_geomfromewkt(Utf8("POINT (1 1)")) │
│                struct                │
╞══════════════════════════════════════╡
│ {item: POINT(1 1), crs: }            │
└──────────────────────────────────────┘

[paleolimbot]

Only minor nits from me...this looks great!

I think I still don't understand how the item-level CRS should be handled

Me too! Supporting both type-level and item-level CRS is new territory and feel free to flag anything that doesn't make sense.

[paleolimbot]

Feel free to leave this out for now...the CRS engine variants of some of the other functions aren't currently used with an engine and we probably need a more reliable way to inject support for validation.

[yutannihilation]

Oh, thanks. Then, I remove this. I noticed engine is currently unused, but wondered if this is somehow useful for testing.

[paleolimbot]

Can this be a U32?

[paleolimbot]

I don't think you need the repeated case here (I'm pretty sure tester.invoke_array() will do that cast for you).

[paleolimbot]

Thank you!

I opened #501 to track CRS validation. There are a number of places where the situation can and should be improved 🙂

[yutannihilation]

I kept this pull request draft as I was wondering if I should add Python tests. However, since the output is different from PostGIS, probably it's not worth...? If I should add some, I can do.

[paleolimbot]

I kept this pull request draft as I was wondering if I should add Python tests.

In this case you probably could check ST_SRID(ST_GeomFromEWKT(...)) and ST_AsText(ST_GeomFromEWKT(...)), which I think should be the same in both PostGIS and SedonaDB now, although I wasn't personally expecting anybody to try this yet (there's no helpers for this in Python land at the moment!)

[yutannihilation]

Thanks, I added some tests, but I found ST_AsText() doesn't work on per-item-CRS geometry yet. So, I think testing only the SRID is enough for now.

> SELECT ST_AsText(ST_GeomFromEWKT('POINT(0 0)'));
This feature is not implemented: st_astext([Arrow(Struct([Field { data_type: Binary, nullable: true, metadata: {"ARROW:extension:name": "geoarrow.wkb", "ARROW:extension:metadata": "{}"} }, Field { name: "crs", data_type: Utf8View, nullable: true }]))]): No kernel matching arguments

[paleolimbot]

If you rebase or merge in the main branch I think this will work:

Suggested change

	# TODO: currently, a per-item-CRS geometry cannot be cast to string
	# eng.assert_query_result(
	# f"SELECT ST_GeomFromEWKT({val_or_null(ewkt)})",
	# expected,
	# )
	eng.assert_query_result(
	f"SELECT ST_AsWKT(ST_GeomFromEWKT({val_or_null(ewkt)}))",
	expected,
	)

sedona-db/rust/sedona-functions/src/st_astext.rs

Line 38 in af5e947

ItemCrsKernel::wrap_impl(vec![Arc::new(STAsText {})]),

Also feel free to remove this bit and I can follow up when I do ST_GeomFromEKWB() (FWIW I prefer not to merge TODOs that don't have an tracker link associated with them, since TODOs in the code base are much harder to prioritize than issues.)

[yutannihilation]

Fair enough! Then, I remove this for now as ST_AsWKT() doesn't work either.

[yutannihilation]

Sorry, I just didn't notice ST_AsText() needs #492 to work. Added the test again.

[paleolimbot]

Thank you!