RIDDL 1.26.0 Released

RIDDL 1.26.0 is now available. This release of the RIDDL compiler and language tooling includes the following changes:

What’s New

root2Json — AST → JSON serializer (the inverse of parseJson) — RiddlLib.root2Json(root, pretty) (and RiddlAPI.root2Json on the JavaScript side) serializes a RIDDL AST Root back into the JSON wire schema that parseJson consumes. For any model in the supported subset the round-trip is faithful: parseJson(root2Json(root)) re-validates identically, and the serialization is stable under a second round-trip. It is lossless for the documented JSON subset and best-effort (non-crashing) beyond it. Combined with bast2FlatAST, existing compiled models can now be turned into JSON — e.g. to mint training data for AI generation. Implemented as a plain recursive serializer (JsonSerializer), Native-safe, and additive (no existing signatures change). Documented in index.d.ts and JSON_INPUT.md.

parseJson tolerates the two most common AI-generated JSON mistakes — Real language models emitting model JSON reliably make two schema-usage slips that previously failed to parse; both are now accepted (strictly more permissively — every previously-valid document parses identically):
- An unknown type-expression kind (e.g. {"kind":"Username"} where Username is a declared type) is read as a reference to a declared type of that name instead of raising an error. Genuinely undefined names still surface as normal validation (resolution) errors when the model is validated — never as an exception.
- Inline cardinality, {"kind":"Date","cardinality":"optional"}, is now accepted and parses identically to the canonical wrapper form {"cardinality":"optional","of":{"kind":"Date"}}.
This makes parseJson a far more forgiving input path for programmatic and AI-assisted generation, without loosening what a valid document means.