ST_Distance and ST_DWithin based on georust/geo

[Kontinuation]

This patch implements ST_Distance and ST_DWithin using georust/geo's Euclidean distance function.

Performance comparison with GEOS-based implementation:

GEO:

geo-st_distance-ArrayScalar(Point, Polygon(10))
                        time:   [19.120 ms 19.171 ms 19.227 ms]
geo-st_distance-ArrayScalar(Point, Polygon(500))
                        time:   [446.14 ms 448.10 ms 450.64 ms]

GEOS:

geos-st_distance-ArrayScalar(Polygon(10), Polygon(10))
                        time:   [187.51 ms 188.38 ms 189.72 ms]
geos-st_distance-ArrayScalar(Polygon(10), Polygon(500))
                        time:   [2.8651 s 2.8735 s 2.8822 s]

Benchmarking ternary functions such as ST_DWithin takes forever, it could be a problem with our benchmarking framework.

[Copilot]

Pull Request Overview

This PR implements ST_Distance and ST_DWithin spatial functions using the georust/geo library's Euclidean distance calculations, providing a significant performance improvement over the GEOS-based implementation.

• Adds ST_Distance function for calculating Euclidean distance between geometries
• Adds ST_DWithin function for checking if geometries are within a specified distance
• Includes comprehensive test coverage and benchmarks for both functions

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
rust/sedona-geo/src/st_distance.rs	Implements ST_Distance function with Euclidean distance calculation
rust/sedona-geo/src/st_dwithin.rs	Implements ST_DWithin function for distance-based geometric queries
rust/sedona-geo/src/register.rs	Registers the new functions in the scalar kernels list
rust/sedona-geo/src/lib.rs	Adds module declarations for the new functions
rust/sedona-geo/benches/geo-functions.rs	Adds benchmarks for performance testing of the new functions

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[jesspav]

Looks great!
Do these functions not apply to geographies?

[Kontinuation]

Looks great! Do these functions not apply to geographies?

Geographies has its own implementation of ST_Distance:

sedona-db/c/sedona-s2geography/src/scalar_kernel.rs

Lines 85 to 91 in 3d01227

	pub fn st_distance_impl() -> ScalarKernelRef {
	S2ScalarKernel::new_ref(
	S2ScalarUDF::Distance,
	vec![ArgMatcher::is_geography(), ArgMatcher::is_geography()],
	SedonaType::Arrow(DataType::Float64),
	)
	}

[paleolimbot]

Thank you! I think the GeoTypesExecutor is worth addressing now (array distance inputs could be a follow-up although I think it is not too hard to handle).

[paleolimbot]

This will error when distance isn't a constant? I think you could probably handle this case with:

let arg2_array = arg2.to_array(executor.num_iterations())?;
let arg2_f64_array = as_float64_array(&arg2_array)?;
let arg2_iter = arg2_f64_array.iter();

// In the loop
let distance = arg2_iter.next().unwrap()

That is possibly slower for the scalar case and could be kept separate if it ends up mattering.

[Kontinuation]

I am simply replicating what st_dwithin for GEOS does:

sedona-db/c/sedona-geos/src/st_dwithin.rs

Lines 56 to 70 in aa4f80a

	// Extract the constant scalar value before looping over the input geometries
	let distance: Option<f64>;
	let arg2 = args[2].cast_to(&DataType::Float64, None)?;
	if let ColumnarValue::Scalar(scalar_arg) = &arg2 {
	if scalar_arg.is_null() {
	distance = None;
	} else {
	distance = Some(f64::try_from(scalar_arg.clone())?);
	}
	} else {
	return Err(DataFusionError::Execution(format!(
	"Invalid distance: {:?}",
	args[2]
	)));
	}

I'll address the same problem in GEOS UDF code in this patch.

[paleolimbot]

Just double-checking: do we need these conversions, or is the distance metric implemented for the generic case?

Also, it should be more efficient to use the GeoTypesExecutor, which should ensure that scalar inputs on either the right or left are only converted once.

[Kontinuation]

The generic distance PR was merged 1 hour ago. I'll remove the conversion and call distance_ext directly on WKB values.