geopackage

A Rust implementation of the OGC GeoPackage 1.4 format: vector features, attribute tables, spatial indexing, and columnar I/O through Apache Arrow. Pre-1.0: the API will change without notice.

Install

[dependencies]
geopackage = "0.1"
geo-types = "0.7"  # any geo-traits implementation works; this is the common one

Columnar read and write through Apache Arrow is behind an off-by-default feature, which adds the arrow-array and arrow-schema dependencies:

geopackage = { version = "0.1", features = ["arrow"] }

SQLite is bundled and built from source, so a C compiler is required and there is no system SQLite dependency. The minimum supported Rust version is 1.95.

Example

Create a file, declare a point layer, write features, and query by bounding box.

use geo_types::Point;
use geopackage::core::types::{ColumnType, GeometryType};
use geopackage::{
    BoundingBox, ColumnSpec, GeoPackage, GeometrySpec, NewFeature, TableSchemaBuilder, Value,
};

let gpkg = GeoPackage::create("cities.gpkg")?;

gpkg.create_layer(
    &TableSchemaBuilder::new("cities")
        .column(ColumnSpec::new("name", ColumnType::Text(None)))
        .geometry(GeometrySpec::new(GeometryType::Point, 4326)),
)?;

// `create_layer` builds a spatial index; decline it with
// `.spatial_index(false)` on the builder.
let layer = gpkg.layer("cities")?;

layer.write_all(
    vec![
        NewFeature::new(Point::new(-6.26, 53.35), vec![Value::Text("Dublin".into())]),
        NewFeature::new(Point::new(-0.13, 51.51), vec![Value::Text("London".into())]),
    ],
    1000,
)?;

// Uses the RTree index when one is present, a full scan otherwise.
for feature in layer.features_in(BoundingBox::new(-7.0, 53.0, -6.0, 54.0))? {
    println!("{:?}", feature?.value("name"));
}

Columnar read and write

With the arrow feature, Layer::read_arrow reads a layer as Arrow record batches, on min(4, available parallelism) threads by default, and Layer::write_arrow writes batches back through the same path as write_all. Geometry is a GeoArrow WKB column whose metadata stores the CRS as PROJJSON. Layer::arrow_schema and TableSchemaBuilder::from_arrow_schema give the two directions of the type mapping, so a layer can be copied without restating its schema.

More examples

Runnable programs in geopackage/examples:

cargo run --release --example bulk_load -- 200000 out.gpkg
cargo run --example inspect -- out.gpkg
cargo run --example bbox_query -- out.gpkg points -10 -5 10 5

Configuration

The defaults produce a single-file GeoPackage with an indexed feature layer. What is settable:

Each setting is documented in full on its type in the crate documentation. Anything not covered is reachable as SQL through GeoPackage::connection().

Workspace

Design

• Synchronous API over rusqlite. GeoPackage::connection() and from_connection() expose the underlying connection; SQLite is the query engine, so anything the API does not cover is a query away.
• WAL is opt-in, and a handle that opted into it checkpoints and resets the file to DELETE on close, so a handed-over .gpkg is a single file with no sidecars.
• New indexes get the GeoPackage 1.4 trigger set. Older and mixed trigger generations, a known source of file corruption, are detected and repaired with repair_spatial_index rather than silently mixed.
• CRS definitions are stored, never transformed: there is no PROJ dependency and no coordinate transformation. add_epsg_srs writes WKT1 where the code has one, and WKT2 through the gpkg_crs_wkt_1_1 extension otherwise (for codes with no WKT1 form, such as the geographic 3D EPSG:4979), matching GDAL.

Conformance

Files written by this crate are checked against OGC ets-gpkg12: 40 passed, 1 failed, where the failing test's regex hard-codes the GeoPackage 1.2 trigger set and rejects a correct 1.4 one (no 1.3/1.4 ETS exists). The PDOK validator reports two advisory findings, both on deliberate test-file choices. A GDAL round-trip byte-compares geometry WKB and attribute values after an ogr2ogr copy, and the test corpus includes GDAL-written, QGIS-written and raw-SQLite files.

Known limitations

• Untrusted files can trigger a large allocation. The wkb reader pre-allocates from element counts read out of the geometry blob without bounding them against the buffer, so a malformed 17-byte GPB blob declaring a 0xFFFFFFFF-member collection drives a multi-gigabyte allocation. The fix belongs upstream in georust/wkb; do not parse untrusted GeoPackage files. Tracked in #3.
• Non-linear curve types (CIRCULARSTRING, COMPOUNDCURVE, ...) cannot have their envelopes computed and so cannot be inserted into an indexed table. Tracked in #5.
• Tile pyramids are not implemented. A file's tile tables are visible through contents() but there is no read or write path for them.

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option.