How these libraries work

Every library here exists in several languages and behaves the same way in each of them, down to its error codes.

One spec, many ports

Each library has a spec repository (<library>-spec). It describes the behavior once, in a language-neutral way, and holds the conformance cases: inputs with expected outputs, including expected errors.

Each language gets its own port repository, laid out the way that language expects. A port passes every conformance case before it claims a level (core, io or full). Where a reference implementation exists, the cases are generated from it; where it can’t produce a case, the case is written from the spec.

Same shape everywhere

• Same operations, named in each language’s own style.
• Same errors: a kind (invalid_input, limit_exceeded, …) and a stable code such as polyline.invalid_char, plus a byte offset where it applies.
• Same limits on untrusted input, with documented defaults.
• Same three examples, which is what the tabs on each library page show.

Tested and measured

• Every decoder of untrusted input is fuzzed.
• Every port is benchmarked against a mature reference implementation on the same input, with the same method. Each library page shows the ratio.

License

Everything is MIT OR Apache-2.0.