fix(validator): keep an inactive boolean branch out of schemas

[David Chicaiza (david0723)]

Follow-up to #65.

What breaks

A single-branch boolean (nested: [...], applies when true) with the toggle at false is still walked since 1.18.0 so stale values keep their type rules. That walk also pushed the branch's fields into schemas, flattened, with required: true intact.

Consumers persist schemas as the module's resolved form and validate runtime bundles against it. Those validators never see the toggle, so a required field from the inactive branch gets demanded at run time, e.g. Missing value of required parameter 'fallbackConnectionId' on a module whose fallback toggle is false.

Why it surfaced now

Before #65 an absent toggle failed as mandatory and never reached the walk, so the leak only fired when the caller sent false explicitly. With fillDefaults: 'requiredOnly' the toggle fills to false and the inactive walk runs on every module that predates the toggle, which is most of them. The fill was correct; what it exposed was this.

The fix

suppressRequired has exactly one setter, the inactive single-branch walk, so both schemaFields.push sites now skip when it is set: the nested site (same-domain branches) and the domain-root site (a single-branch nested: { domain, store } pointing at another domain). The two-branch form already stayed out of schemas via registerOnly; a test now pins that.

Validation itself is unchanged: stale values in the inactive branch are still type-checked, strict mode still knows their names, and an active branch still lands in schemas exactly as before.

Tests

Five cases in boolean-nested.spec.ts: explicit false leaves the branch out, true keeps it in, a false that fillDefaults filled leaves it out, a cross-domain inactive branch stays out of the other domain's schemas, and the two-branch form reports only its active side (the last one is a pin, it passed before). The first four go red on main. 508/508 green, tsc and build clean.

Not in this PR

• states for the inactive branch is untouched. It carries labels for stale values, nothing a runtime validator reads. Happy to align it in a follow-up if you'd rather the inactive branch contribute nothing at all.
• schemas spreads an rpc:// string sitting inside a collection spec into an indexed-character object ({"0":"r","1":"p",...}). Separate defect in the same output, I'll file it rather than widen this one.

[Copilot]

🟡 Changes recommended

The behavioral change for two-branch boolean nested (registerOnly) affecting schemas isn’t covered by a regression test, and one newly added test name is unclear/grammatically incorrect.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Fixes validateForman(..., { schemas: true }) so fields from an inactive boolean nested branch are no longer emitted into the flattened schemas output, preventing downstream validators (that don’t see the toggle value) from incorrectly requiring fields from an inactive branch.

Changes:

• Skip pushing fields into roots[domain].schemaFields when walking under suppressRequired (inactive single-branch boolean) and registerOnly (inactive two-branch boolean) contexts.
• Add regression tests for single-branch boolean nested to ensure inactive branches don’t leak into schemas, including when fillDefaults fills the toggle to false.

File summaries

File	Description
src/validator.ts	Prevents inactive boolean nested branches from contributing fields to schemas by gating schema-field collection on registerOnly / suppressRequired.
test/boolean-nested.spec.ts	Adds regression coverage for schemas output under single-branch boolean nested, including a fillDefaults scenario.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.