Function getParsed

Parse and cache an OpenAPI document. Returns the cached parse for the same object identity.

Before parsing, the document is run through the version-aware normalisation pipeline (normaliseOpenApiSchemas) so OpenAPI 3.0.x keywords (nullable, discriminator, example), OpenAPI 3.1.x discriminator, and Swagger 2.0 documents are all converted to canonical Draft 2020-12 form. The parser and downstream extractors (extractRequestBody, extractResponses, etc.) then observe schemas in the same form <SchemaComponent> does, keeping the OpenAPI components on the same pipeline as the top-level adapter.

Caching and diagnostics

Normalisation runs at most once per document identity. The full set of doc-level diagnostics emitted during that single run is captured into the cache alongside the parsed result. Each caller-supplied sink receives the captured diagnostics exactly once per cached entry, no matter how many times getParsed is called with that (doc, sink) pair.

The previous implementation bypassed the cache whenever diagnostics was supplied and re-ran the entire normalisation pipeline against the new sink. That fired every doc-level diagnostic once per call, so a parent like ApiWebhooks that renders ApiWebhook per webhook entry caused N-fold emission of a single real cause. With the new strategy, cardinality stays at one per real cause regardless of how many child renders share the sink.

Strict mode is treated as a per-call invariant — see the internal replayCapturedDiagnostics helper below for the rationale.