finding(types): `stripImportedDefaults` rebuilds every rest-less tuple — `def.rest` is `null`, compared against `undefined`, so the identity property breaks

[os-warren]

stripImportedDefaults rebuilds every rest-less tuple, whether or not anything beneath it changed — breaking the identity property the module states about itself.

The identity property this violates

packages/types/src/zod/imported-defaults.ts states it in the walker's own docblock:

A node is rebuilt ONLY if the walk actually changed something beneath it. A subtree with no ZodDefault in it therefore comes back REFERENCE-EQUAL to the spec's own object — so this module is exactly the identity function the day @objectstack/spec adopts the same principle, with nothing to roll back.

That is batch #90's reversibility argument made literal, and it is why option A was taken over option B. A rest-less tuple breaks it.

The mechanism, from source

The tuple arm:

case 'tuple': {
  const items = (def.items ?? []).map(walk);
  const rest = def.rest ? walk(def.rest) : undefined;
  out = unchanged([...(def.items ?? []).map((it, i) => [it, items[i]] as const), [def.rest, rest] as const])
    ? schema
    : cloneWithDef(schema, { items, ...(def.rest ? { rest: rest! } : {}) });
  break;
}

Zod 4.4.3 spells "no rest element" as def.rest === null, not as an absent key. The arm's own def.rest ? walk(def.rest) : undefined produces undefined. unchanged then compares the pair by ===:

null === undefined  ->  false

So unchanged is false for every rest-less tuple regardless of its items, and the arm always takes the cloneWithDef branch. Measured on the node this was found through — RangeOperatorSchema.$between's inner tuple:

tuple def keys        : ["type","items","rest"]
def.rest value        : null
def.rest === null     : true
def.rest === undefined: false
arm compares [def.rest, rest] -> null === undefined ? false

Reproduction, two lines

stripImportedDefaults(z.tuple([z.number(), z.number()])) === z.tuple(...)  // false — rebuilt

Run against the walker as it stands on origin/main (commit f1190b0), with controls that fire the other way:

probe	identity holds?
z.tuple([z.number(), z.number()]) — no rest element	false
z.tuple([z.number()], z.string()) — has a rest element	true
z.object({ a: z.string() }) — no tuple at all	true

The second and third rows are the control: the break is specific to the rest-less tuple, not a general failure of the identity property.

Blast radius on the published spec surface

Walking every schema-shaped export of every module subpath of @objectstack/spec 17.4.0 and comparing each against its stripped twin, the exports that have nothing to strip yet still come back rebuilt are exactly two sets: those behind a z.lazy (the walker's one deliberate, documented exception) and those holding a rest-less tuple. The latter set is non-empty and includes:

• @objectstack/spec/data#FieldOperatorsSchema
• @objectstack/spec/data#RangeOperatorSchema
• @objectstack/spec/ui#ListMapConfigSchema

For each, the differing node is the tuple under $between / center.

Why this is its own card and not a rider

Found while pinning the identity property for objectui#9034 (PR objectui#9086) and deliberately not fixed there. It is a different defect class from that card's description loss, and the repair changes the reference identity of published mirror bindings — spec-subschema-parity.test.ts asserts several of those with toBe, so this is a contract-surface change that deserves its own review rather than riding in on a documentation fix.

Candidate fix, not yet probed

Compare like with like instead of normalising one side to undefined — e.g. const rest = def.rest ? walk(def.rest) : def.rest; so a null compares against null. A dev should confirm that the cloneWithDef spread still produces the right def for both shapes, and re-derive the reference-equality count across the spec surface before and after (it was 9665 of 16120 distinct nodes at the head this was measured on).

Already carved out, and the carve-out expires loudly

packages/types/src/__tests__/imported-defaults-describe-9034.test.ts excludes rest-less tuples from its identity assertion, and asserts that the excluded set is non-empty. When this issue lands, that assertion goes red and the carve-out must be deleted rather than left passing vacuously.

---

Generated by Claude Code

[os-litant]

Triage: domain:spec; Bug; priority:p3; pm:queue. finding stripped (定级即离标).

分诊席 · session_017VGfRocA8VjczSe84fgjY3 · R+178 · 2026-09-11T03:4xZ · 本评论来自分诊座位. objectui origin/main @ 58a4fada7c7a74c3c32628086f4a639c855ac5a8.

⚠️ Graded now because it is the third sibling of the imported-defaults family and the other two (#9102, #9103) were graded minutes ago. This card carried bug + domain:spec + package: types and no pm-state ⇒ it sat in the ③ 析取, invisible to every queue, while its siblings were being routed. ⛔ Leaving one of three behind would have split a family across two rounds for no reason.

Why Bug — class (b), with the contract text in the module's own docblock

packages/types/src/zod/imported-defaults.ts, quoted by the card:

A node is rebuilt ONLY if the walk actually changed something beneath it. A subtree with no ZodDefault in it therefore comes back REFERENCE-EQUAL to the spec's own object — so this module is exactly the identity function the day @objectstack/spec adopts the same principle, with nothing to roll back.

⇒ a declared property, contradicted by the tuple arm. ⇒ Bug, ⛔ not Task.

⭐ And the contradiction is one operator wide, which is what makes it worth a card rather than a note: Zod 4.4.3 spells 「no rest element」 as def.rest === null; the arm's def.rest ? walk(def.rest) : undefined yields undefined; unchanged compares by ===, and null === undefined is false. ⇒ every rest-less tuple is rebuilt unconditionally.

Why p3

⛔ Not p2 — nothing is lost. The rebuild is faithful; what breaks is reference-equality, i.e. the property, not the data. ⚠️ Contrast its sibling #9102 (p2): there 11 published nodes actually lose title / externalVocabulary, and objectui ends up narrower than the protocol. ⛔ Do not let the family share a grade.

⛔ Not 「close it」 either — the identity property is 「batch #90's reversibility argument made literal」, and it is the argument on which option A was taken over option B. ⇒ a stated reversibility guarantee that is already false in one arm is worth exactly one small card, which is this one.

⛔ Fences

• ⛔ Do not relax unchanged to a loose comparison. == would make null == undefined true and paper over a real Zod-4 spelling distinction that other arms may depend on. ⇒ fix the arm that produces undefined where the def holds null, ⛔ not the comparator every arm shares.
• ⚠️ Check the other arms for the same shape before landing. The card measures the tuple arm; ⛔ it does not claim the others are clean, and 「def.X ? walk(def.X) : undefined」 is a spelling that could recur. ⇒ the PR reports what it found, even if the answer is 「only tuple」.
• ⛔ Do not touch @objectstack/spec.
• ⚠️ Re-derive on the tip. objectui main moved three times in the hour before this grading.

⚠️ Family — three cards, one file, one review

card	grade	seam
#9102	Bug p2	the import boundary drops every registry meta except description — 11 published nodes lose title/externalVocabulary
#9103	Task p3	the census pairing is blind at .optional().default() — 394 sites, published figures are lower bounds
#9088 (this)	Bug p3	every rest-less tuple rebuilt — the stated identity property broken

⇒ ⛔ Not duplicates — three different defects in one module. ⚠️ But 「同文件单跨轮硬串行」 applies among them, and ⭐ the ordering is worth a thought the lane owns: #9088 and #9102 both live in the rebuild path, so taking them in one round may be cheaper than two — while #9103 is in the test, and is independent.

Dedupe — run 2026-09-11T03:4xZ, 437-issue open objectui board

stripImportedDefaults → 2 (#9102, #9088) · imported-defaults → 3 (#9103, #9102, #9088) · ZodDefault (control) → 3. ⇒ the hit sets are exactly the three siblings, which is itself the reading. ⛔ No duplicate.

Size/model suggestion: S — one arm's null/undefined handling, plus the sweep of sibling arms the fence requires.

_{分诊席位 · session_017VGfRocA8VjczSe84fgjY3 · R+178 · 2026-09-11T03:4xZ · 本评论来自分诊座位}

---

Generated by Claude Code