feat(c/sedona-proj): Implement item crs support for ST_Transform

[paleolimbot]

This PR integrates Item Crs inputs and outputs into ST_Transform, which now has a rather complicated signature with respect to what can happen:

• (Geometry or ItemCrs) + (String or Integer constant to) -> Geometery (type-level CRS)
• (Geometry or ItemCrs) + (String or integer from) + (String or Integer constant to) -> Geometery (type-level CRS)
• (Geometry or ItemCrs) + (String or Integer array/column to) -> Item CRS
• (Geometry or ItemCrs) + (String or integer from) + (String or Integer array/column to) -> Item CRS

There's probably no great way to implement all of that cleanly but I did try to reduce repetition as much as possible.

import pandas as pd
import sedona.db

sd = sedona.db.connect()

df = pd.DataFrame({"wkt": ["POINT (0 1)", "POINT (2 3)"], "srid_from": [4326, 3857], "srid_to": [3857, 4326]})
sd.create_data_frame(df).to_view("foofy")

# Array -> Array
sd.sql("SELECT ST_Transform(ST_GeomFromWKT(wkt), srid_from, srid_to) as g FROM foofy").show()
#> ┌───────────────────────────────────────────────────────────────────────────────┐
#> │                                       g                                       │
#> │                                     struct                                    │
#> ╞═══════════════════════════════════════════════════════════════════════════════╡
#> │ {item: POINT(0 111325.1428663851), crs: EPSG:3857}                            │
#> ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
#> │ {item: POINT(0.000017966305682390428 0.00002694945852358465), crs: EPSG:4326} │
#> └───────────────────────────────────────────────────────────────────────────────┘

# Array -> Scalar
sd.sql("SELECT ST_Transform(ST_GeomFromWKT(wkt), srid_from, 4326) as g FROM foofy").show()
#> ┌───────────────────────────────────────────────────────┐
#> │                           g                           │
#> │                        geometry                       │
#> ╞═══════════════════════════════════════════════════════╡
#> │ POINT(0 1)                                            │
#> ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
#> │ POINT(0.000017966305682390428 0.00002694945852358465) │
#> └───────────────────────────────────────────────────────┘

# Scalar -> Array
sd.sql("SELECT ST_Transform(ST_GeomFromWKT(wkt), 4326, srid_to) as g FROM foofy").show()
#> ┌────────────────────────────────────────────────────┐
#> │                          g                         │
#> │                       struct                       │
#> ╞════════════════════════════════════════════════════╡
#> │ {item: POINT(0 111325.1428663851), crs: EPSG:3857} │
#> ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
#> │ {item: POINT(2 3), crs: EPSG:4326}                 │
#> └────────────────────────────────────────────────────┘

# ItemCrs -> Array
sd.sql("SELECT ST_Transform(ST_SetSRID(ST_GeomFromWKT(wkt), srid_from), srid_to) as g FROM foofy").show()
#> ┌───────────────────────────────────────────────────────────────────────────────┐
#> │                                       g                                       │
#> │                                     struct                                    │
#> ╞═══════════════════════════════════════════════════════════════════════════════╡
#> │ {item: POINT(0 111325.1428663851), crs: EPSG:3857}                            │
#> ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
#> │ {item: POINT(0.000017966305682390428 0.00002694945852358465), crs: EPSG:4326} │
#> └───────────────────────────────────────────────────────────────────────────────┘

# ItemCrs -> Scalar
sd.sql("SELECT ST_Transform(ST_SetSRID(ST_GeomFromWKT(wkt), srid_from), 4326) as g FROM foofy").show()
#> ┌───────────────────────────────────────────────────────┐
#> │                           g                           │
#> │                        geometry                       │
#> ╞═══════════════════════════════════════════════════════╡
#> │ POINT(0 1)                                            │
#> ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
#> │ POINT(0.000017966305682390428 0.00002694945852358465) │
#> └───────────────────────────────────────────────────────┘

[Copilot]

Pull request overview

This PR implements item-level CRS (Coordinate Reference System) support for the ST_Transform function, enabling transformations where the target CRS varies per row rather than being fixed at the type level.

Changes:

• Adds support for item-level CRS input/output in ST_Transform, allowing per-row CRS specifications
• Refactors argument handling to support both scalar and array CRS inputs with appropriate return type determination
• Optimizes transformation by caching when both source and target CRS are constant across all rows

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 7 comments.

File	Description
rust/sedona-testing/src/testers.rs	Updates invoke method to compute return types with scalar arguments
rust/sedona-functions/src/executor.rs	Adds support for iterating over item-crs struct arrays by extracting geometry from the "item" field
c/sedona-proj/src/st_transform.rs	Major refactor to support item-crs inputs/outputs with new argument pattern matching and transformation logic
c/sedona-proj/Cargo.toml	Adds sedona-common dependency

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[zhangfengcdt]

Just wanted to confirm the builder.append_value([]); is expected to be called after the invoke_scalar call.

[paleolimbot]

Yes, that is expected (invoke_scalar writes into the builder directly but the append_value is needed to force a new element from the builder).

[zhangfengcdt]

Maybe caching could help here since this is per-row calculation?

[paleolimbot]

We have a thread local cache behind deserialize_crs() for this reason!

...there are probably more ways we could speed up per-row CRS operations once they actually work, though.