cityjson-rs implements the CityJSON 2.0 data model in Rust.
This crate provides types and accessor methods for working with a flattened, columnar representation of the CityJSON data model.
cityjson-rs is meant to be a core library for downstream specialized libraries that implement serialization, indexing, geometry processing, and other features.
Serialization is implemented by:
- json: cityjson-json
- arrow: cityjson-arrow
- parquet: cityjson-parquet
For a higher-level library that integrates serialization, implements geometry processing, and other features into a single crate, see cityjson-lib.
For generating fake, schema-valid data for any combination of the CityJSON specs, see cityjson-fake.
For efficient indexing and querying individual CityObjects across multiple files, see cityjson-index.
cargo add cityjsonuse cityjson::v2_0::*; // all CityJSON v2.0 types
use cityjson::prelude::*; // handles, storage strategies, error typesThe prelude re-exports crate-wide types (handles, errors, storage strategies) but not the cityjson-domain types from v2_0.
use cityjson::v2_0::{CityJSONVersion, CityModel, CityModelType};
fn main() {
let model = CityModel::<u32>::new(CityModelType::CityJSON);
assert_eq!(model.version(), Some(CityJSONVersion::V2_0));
assert!(model.cityobjects().is_empty());
assert_eq!(model.iter_geometries().count(), 0);
assert_eq!(model.iter_geometry_templates().count(), 0);
assert!(model.template_vertices().is_empty());
assert_eq!(model.iter_semantics().count(), 0);
assert_eq!(model.iter_materials().count(), 0);
assert_eq!(model.iter_textures().count(), 0);
assert!(model.vertices_texture().is_empty());
assert!(model.vertices().is_empty());
assert_eq!(model.transform(), None);
assert_eq!(model.metadata(), None);
assert_eq!(model.extra(), None);
assert_eq!(model.extensions(), None);
}cityjson-rs uses a flat, columnar internal representation that differs from the nested JSON structure of the CityJSON specification:
- Geometry boundaries: stored as sibling offset arrays (
surfaces,shells,solidswith corresponding vertex/ring/surface/shell offsets) instead of nested coordinate arrays - Resource pools: global pools for semantics, materials, textures, and UV coordinates; geometry-local maps store handle references into these pools instead of dense array indices
- Semantic and material assignments: flat primitive-assignment arrays (one per surface/point/linestring) instead of nested index structures
- Texture assignments: per-ring resource references with flat UV coordinate arrays, rather than nested per-ring texture entries
This representation is more efficient for traversal and serialization to columnar formats (Arrow, Parquet) while maintaining round-trip fidelity with the spec format. See Geometry Mappings for detailed layout rules and examples.
todo: link to docs.rs
| Module | Contents |
|---|---|
v2_0 |
Domain types: CityModel, CityObject, Geometry, GeometryDraft, Metadata, Transform, Semantic, Material, Texture, … |
resources |
Typed handles, resource pools, and string storage strategies |
raw |
Zero-copy read views for use in downstream serializers |
This crate follows semantic versioning (MAJOR.MINOR.PATCH):
MAJOR: incompatible API changesMINOR: backwards-compatible feature additionsPATCH: backwards-compatible fixes
The minimum supported rustc version is 1.93.0.
Contributions are welcome in all forms. Please open an issue to discuss any potential changes before working on a patch. You can submit LLM-generated PRs for bug fixes and documentation improvements. Regardless of handwritten or LLM-generated code, the PR should follow these guidelines:
- relatively small, focused changes, otherwise I won't be able to review it,
- follow the existing style and conventions,
- include unit tests and documentation for new features and bug fixes,
- the patched code should pass:
just check / lint / fmt / test / docs / miri / perf-check
- if you remove or merge tests or examples or benchmarks, please explain why and update the documentation accordingly.
Licensed under either:
- Apache License, Version 2.0 (
LICENSE-APACHE) - MIT license (
LICENSE-MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in cityjson-rs by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without additional terms or conditions.
This crate was originally developed without the use of AI.
Since then, it underwent multiple significant refactors and various LLM models (Claude, ChatGPT) were used for experimenting with alternative designs, in particular for the resource pool and attribute storage strategies.
LLM generated code is also used for improving the test coverage and documentation and mechanical improvements.
Code correctness and performance are verified by carefully curated test cases and benchmarks that cover the entire CityJSON 2.0 specification.
There are no major features planned for the near future, beyond bug fixes, test coverage, documentation improvements.