CHANGELOG.md

Changelog

[0.5.2] - 2026-04-15

Changed (JS)

• Observer Schema Context — ObjectTree._observer hook signature extended with 5th parameter: fn(op, path, key, val, schema). The schema parameter provides the schema node for the accessed property, enabling:
  • Filter properties by x-observable or other extensions
  • Format-aware serialization based on schema.format
  • Type-aware logging with schema.type and schema.description
  • Reactive bindings with full schema metadata in event payload
• Backward compatible — existing observer hooks with 4-parameter signature continue to work (5th parameter ignored if unused)

Added (JS)

• Observer examples — examples/observer_schema_context.mjs demonstrates 4 use cases: filter by extension, format-aware serialization, type-aware logging, schema-less property handling

Tests (JS)

• All existing tests pass: 922/922 Draft-07 + 19/19 validate tests

[0.5.1] - 2026-04-15

Added (JS)

• Dot Key Support — nested property paths via dot notation syntax. obj["a.b.c"] = value automatically creates intermediate objects and validates against nested schema. Simplifies deep property updates without manual object traversal.
  • Auto-creates intermediate objects when undefined/null
  • Full schema validation on final property
  • Works with existing observer hooks
• ObjectTree._observer — static hook point for property access observation. Set to fn(op, path, key, val) to observe all get/set across every ObjectTree instance. null disables (zero cost). Structural equivalent of Objective-C KVO — every schema-driven object is natively observable without wrapper or Proxy layer.
  • get emits after value resolution (primitive, object wrap, array cursor)
  • set emits after validation and write
  • Cached object/array re-access does not re-emit (cache hit returns before hook)

Tests (JS)

• 5 new dot key support tests in test_dotkey.mjs: basic nested set, multiple paths, overwrite, type validation, auto-create intermediate
• 6 observer tests in validate.mjs: get primitive, get object, get array, set, null-disable, cached-no-re-emit
• Total: 922/922 Draft-07 (100%)

[0.5.0] - 2026-03-30

Changed (JS)

• Lazy defaults — schema default values are the initial value of any field. Reading a field missing from raw data returns its schema default transparently. Object.keys(tree) and in include default-only fields. Applies at all nesting levels.
• Array cursor — array properties wrap as a lazy Proxy cursor instead of eager .map(). Per-index ObjectTree instances are created on first access and cached for stable identity (arr[0] === arr[0]). Mutations invalidate only the affected index. for...of, spread, and destructuring work correctly.
• $withDefaults() semantics — materializes defaults into raw data. Recursively fills nested objects and array items.
• $value includes defaults — includes schema-default fields and expands array item defaults.

Internal (JS)

• ctx pipeline — internal rewrite: C-style ctx struct + pure fn pipeline replaces class methods. External API unchanged.
• Removed graph scheduler — simplified execution model.

Fixed (JS)

• Nested remote $ref — correctly resolves nested $ref in remote schemas.

Note

• Python and Rust implementations have not been updated for 0.5.0 changes (lazy defaults, array cursor, ctx pipeline). They remain at 0.4.x semantics.

[0.4.1] - 2026-03-28

Breaking Changes (JS)

• Filesystem resolver no longer accepts HTTP URIs — $ref with http:// scheme now throws TypeError when resolver is a directory path string. Use function or object-map resolver for HTTP schemas.
• Resolver no longer retries with original $ref — only the normalized URI is passed to resolver. Resolver must handle the canonical form.

Fixed

• Loader #idMap lookup (JS) — removed incorrect fallback that searched fragment-bearing URIs against $id map
• Test runner — switched from filesystem resolver to function resolver for Draft-07 remote $ref tests, correctly exercising the resolver callback interface

Removed

• #fileExists helper, .json extension auto-append, path traversal guard — these were runtime concerns that don't belong in schema tooling

[0.4.0] - 2026-03-21

Breaking Changes (JS)

• Unified #value storage — removed #isObjectNode() and dual #value/#data fields. Every schema node is an object; type/properties are just attributes, not node classification. Schemas without type or properties (e.g. pure anyOf/oneOf) now correctly build ObjectTree.
• Removed file path as schema parameter — schema must be a plain object. Read and parse files yourself; the lib doesn't do I/O.

Fixed

• Nested wrapping for schemeless nodes (JS) — schemas without type: "object" or properties now correctly wrap nested data as ObjectTree
• Primitive data storage (JS) — ObjectTree("hello", { type: "string" }) now correctly stores and returns the value

Changed

• #defineProperties guard (JS) — handles null/boolean schemas without crashing
• README — rewritten constructor docs, added "Loading Schema from File" section

[0.3.0] - 2026-03-17

Breaking Changes (JS only)

• $ prefix on all ObjectTree API — value → $value, toDict() → $toDict(), oneOf() → $oneOf(), etc. Data properties keep no prefix (tree.name). Prevents collision when schema property names match API method names.

Added

• toJSON() protocol (JS) — JSON.stringify(tree) now works correctly, recursively unwraps nested ObjectTree instances
• Invalid schema guard (JS) — validateType rejects null/non-object schemas with clear error message

Fixed

• Eager nested wrapping (JS) — nested objects eagerly wrapped as ObjectTree at construction, aligned with Python/Rust. Mutation at any depth validates against sub-schema.
• Default preservation (JS) — constructor overlays data onto defaults instead of replacing; set $value also preserves defaults
• anyOf/oneOf sub-schema wrapping (JS) — removed isObj guard; any object value is wrapped regardless of schema shape
• $project() (JS) — uses constructor instead of direct #data assignment, ensuring nested wrapping

Changed

• configurable: false (JS) — property descriptors block delete to prevent validation bypass

[0.2.0] - 2026-03-14

Added

• Rust codegen (rust/codegen/) — generates static .rs file: runtime source + typed structs from JSON Schema
  • python rust/codegen/codegen.py schema.json --out src/generated.rs
  • Full draft-07 suite (922/922) + API tests (44)
• docs/ — architecture docs: schema-as-object.md, schema2object-api.json, structure.md, types.md, etc.
• JS validate() — standalone validation export without constructing ObjectTree
• Rust schema()/get_schema() — now use schema_to_dict, aligned with JS/Python

Fixed

• Internal re-bind Loader propagation (JS + Python + Rust) — oneOf/anyOf/allOf/ifThen/project/withDefaults now pass Loader instance internally; sub-schemas no longer lose $definitions context
• Rust get_extensions() — empty path now calls self.schema() via schema_to_dict, consistent with JS/Python
• multipleOf precision — relative tolerance aligned across all three implementations
• Resolver propagation — ifThen/project/withDefaults correctly propagate resolver (JS + Python)
• required ordering — cross-language alignment

Changed

• Rust rewritten as single file (schema2object.rs), aligned with JS/Python structure
• Python rewritten as single file (schema2object.py), full draft-07 suite
• JS full Draft-07 $ref resolution with flexible resolver API

[0.1.0] - initial

• Three-language implementation: JS, Python, Rust
• JSON Schema Draft-07 as object class definition
• ObjectTree construction validates; mutation validates; to_dict() projects schema fields only
• Draft-07 combinators as methods: oneOf, anyOf, allOf, notOf, ifThen, project, contains, withDefaults