Unified traversal #CT-997

[ubik2]

Changes the way we do traversal so the same system can be used by both the cell.get() implementation and the query implementation that determines which linked docs we need.

The cell.get() system, or validateAndTransform doesn't follow asCell links, just creating the cell objects that will let us access those in the future.

The query system does follow those links to determine what we need to have available.

To unify tracking, the query system was re-written to use the transaction interface we use for reactivity instead of directly accessing storage.

There's a lot more link magic that needed to go on for the validateAndTransform, and to make this all work, I split the code into a SchemaObjectTraverser portion that knows how to walk objects with rules like the schema, and an IObjectCreator portion that knows how to create the objects.

• For the validateAndTransform, that IObjectCreator does all the magic of creating new cells.
• For the querySchema, that IObjectCreator does very little, mostly just passing along information about whether the data matched the schema pattern.

Things I intend to address later:

• Our typing is unsound. The IObjectCreator returns either the complex mix of proxy and cell objects, or simple JSONValue | undefined depending on the mode.
• We're not trying to be perfect with schemas, but many anyOf/allOf that expect to combine object properties may break.
• SchemaObjectTraverser should probably switch to a link interface instead of IAttribution.
• We really need some documentation on how these cell.get calls work, and what's going on behind the scenes. A lot of the rules when we combine schemas from links that we're following should have unit tests.

---

Summary by cubic

Unifies traversal and validation to run on storage transactions and normalizes selectors to start at "value", improving consistency, default handling, and link behavior. Addresses CT-997 by removing tight coupling to manager classes, stabilizing array/link semantics, and preventing actions from running with invalid arguments.

• Refactors

  • Traversal/validation run on IExtendedStorageTransaction (via ManagedStorageTransaction); removed direct BaseObjectManager usage.
  • Selectors and addresses are now rooted at ["value", …]; tests and callers updated.
  • Improved schema handling: merges schemas across links, adds limited anyOf merging, applies defaults to missing/undefined values (including top-level), preserves asCell/asStream, follows first array element link when appropriate, unifies additionalProperties behavior, and treats broken links as undefined.
  • ContextualFlowControl now supports prefixItems and false schemas, reuses a single instance per traversal for better performance, and tightens legacy cell link detection.
  • validateAndTransform delegates to SchemaObjectTraverser and drops the unused sync flag; lowered /memory/transact logs to debug.
  • Actions only run when arguments validate; still run if argumentSchema is false.
  • Shared schema tracker maps space+doc to schema selectors, acting as a watch list and cache to avoid redundant re-traversals.
  • Event streams with false/undefined schemas now use a permissive schema so handlers still run.

• Migration

  • Update SchemaPathSelector.path to start with "value".
  • Use an ExtendedStorageTransaction for traversal instead of passing a manager.
  • Note: charm list schema now defaults to [] (no action needed unless relying on undefined).

^{Written for commit b951078. Summary will update automatically on new commits.}

[seefeldb]

do we actually see this at this point or would have this been collapsed to a true schema earlier?

[ubik2]

I think we still see this. I don't think @mathpirate 's transformer code ever generates this style of schema, but you can specify it manually still.