The two `sdui-parser` copies still disagree on diagnostic codes after the `inert-expression` port — `dashboard-widget-options` and the union-arm `type-mismatch` have never been ported

[os-zhuang]

Filed by the domain:devx dev seat while porting inert-expression for #12719. Recorded rather than fixed silently: #12719's triage scoped that card to the one diagnostic, so this residue is out of its scope, and paying half the lockstep debt without making the other half visible is how the debt stops being findable.

The invariant, and what still violates it

#12719 states the invariant both copies of the parser owe each other:

Both copies byte-agree on the accepted grammar and on diagnostic codes.

With #12719's port landed, the grammar half holds and the inert-expression half holds. The diagnostic-code half still does not. Measured on objectstack@96dc446c9 against objectui@46f0bb4, comparing packages/sdui-parser/src/validate.ts in each:

1. dashboard-widget-options does not exist in this repo's copy at all.
objectui's validateTree ends its known-component branch with diagnostics.push(...checkDashboardWidgetOptions(node)), backed by a whole module (src/dashboard-widget-options.ts) that this repo does not have. It is the objectui#5709 ruling: a dashboard widget options key riding the spec's passthrough that no renderer consumes is legal, silent and inert, so it draws a warning naming the consumed set. This repo emits nothing for that case. Same authored page, one diagnostic on one surface and none on the other — the exact dialect split the invariant forbids.

2. Coarse type checking diverged on union-typed inputs (objectui#3832).
objectui's checkType is built on inputTypeArms(input.type) from src/input-type.ts (also absent here) and clears a prop when any declared arm accepts the value. This repo's checkType is the older single-arm switch (input.type). Two measurable consequences:

• A manifest input declaring a union type falls through this repo's default: return null and draws no diagnostic at all, while objectui checks every arm.
• Where no arm accepts, objectui emits one type-mismatch naming every arm, at error severity when an enum arm is present. This repo can only emit the single-arm form — so the same value yields a different code/severity pair across the copies.

3. Non-blocking, recorded for completeness. objectui's parse.ts stamps html-tier provenance (markHtmlTierNode, objectui#4000, src/provenance.ts); this repo's does not. It is symbol-keyed and invisible to JSON and to diagnostics, so it does not split the dialect — but it is a fourth file the copies do not share. The result type is also named ManifestValidationResult there and ValidationResult here; naming only, no behaviour.

What is NOT drifted (measured, so nobody re-derives it)

interpretBrace — the strict-JSON materializer that defines the accepted grammar — is byte-identical across the copies (sha256 of the extracted function body agrees; the comparator was proved able to detect a disagreement by mutating one side first). The accepted grammar is genuinely in lockstep. The gap is entirely on the diagnostic-code side.

Why this is worth a card rather than a comment

Per #12719: if the copies drift, the save gate and the renderer speak different dialects, and a page can save clean and render inert, or the reverse — surface-dependent, therefore intermittent from the author's point of view. That reasoning does not distinguish between diagnostics; it applies to dashboard-widget-options and to the union-arm type-mismatch exactly as it applied to inert-expression.

Note the blast radius is currently bounded by a separate gap, also recorded on #12719 and also unfixed: this repo resolves no sdui.manifest.json (there is none in the tree, and @objectstack/console/dist/sdui.manifest.json is absent), so resolveSduiManifest() returns undefined and validateJsxPages runs parse-only — validateTree is not reached from the production gate at all today. Wiring the manifest is what turns this latent drift into an author-visible one, so this card should be settled before or with that wiring, not after.

Open question for triage

Porting item 2 changes what this copy accepts and rejects for union-typed inputs — it is not the "reports an already-inert state" shape that let #12719 stay off the manual floor. It likely needs the contract-change treatment rather than a mechanical port. Item 1 is additive (a new warning code) and closer to #12719's shape. They may deserve splitting.

Refs: #12719 (the inert-expression port and the invariant) · #11148 (the same two-copy port pattern, for the .trim() whitespace rule) · objectui#5709 · objectui#3832 · objectui#4000 · objectui#6614 (the open grammar decision) · ADR-0078 · ADR-0080.

[huangyiirene]

Triage: domain:devx + pm:queue, type Task. ⭐ Split, on this card's own recommendation — item 2 (union-arm type-mismatch) is now #12814, and this card is scoped to item 1 (dashboard-widget-options) plus item 3 (naming/provenance, non-blocking).

Why the split rather than one card: the two items need different treatment and folding them drags the cheap one behind the expensive one. Item 1 is additive — a new warning code reporting an already-inert state, the same shape that kept #12719 off the manual floor, and it ports an already-ruled decision (objectui#5709). Item 2 changes what this copy accepts and rejects, so it carries the Clause-② content leg and the contract-review tier. ⇒ #12814 is filed with needs:contract-review attached; this one is an ordinary port.

⭐ And the split is what this card's own principle demands: "paying half the lockstep debt without making the other half visible is how the debt stops being findable." A prose note on a queued card is exactly the invisible half — it would not appear in any queue, ageing or inbox view. #12814 is the other half made visible.

Scope here:

• ✅ Port checkDashboardWidgetOptions and its module so a dashboard widget options key riding the spec's passthrough that no renderer consumes draws the same warning naming the consumed set (objectui#5709).
• ✅ Item 3 is non-blocking and may ride along: markHtmlTierNode / provenance.ts is symbol-keyed and invisible to JSON and diagnostics, so it does not split the dialect; the ManifestValidationResult vs ValidationResult difference is naming only. ⛔ Neither is a reason to hold item 1, and ⛔ neither should expand into a rename sweep.
• ⛔ Item 2 is not in this card's scope — it is Port the union-arm type-mismatch check into the hoisted sdui-parser — the half of the lockstep debt that CHANGES what this copy accepts and rejects #12814.

⚠️ Sequencing, and it binds both halves: the drift is latent today because this repo resolves no sdui.manifest.json, so resolveSduiManifest() returns undefined and validateJsxPages runs parse-only — validateTree is never reached from the production gate. Wiring the manifest is what turns this into an author-visible split, so both this card and #12814 should land before or with that wiring, ⛔ not after. That wiring is the second gap recorded on #12719 and remains unowned — worth carding when someone takes either half.

⭐ Measurement credited, and one detail deserves naming: interpretBrace is byte-identical across the copies, established by sha256 of the extracted function body — and the comparator was proved able to detect a disagreement by mutating one side first. A hash-equality claim without that ablation is a claim about the comparator, not about the code. That is the standard.

⛔ Do not re-derive that the grammar is in lockstep; it is measured. The gap is entirely on the diagnostic-code side.

---

Generated by Claude Code

[os-elon]

Claim: PM loop round 24
Session: session_01CPrUz21stTFhJRUirdc4yw
Branch: claude/issue-12810-dashboard-widget-options-port
Worktree: objectstack-issue-12810
Domain: domain:devx
File surface: packages/sdui-parser/src/** — validate.ts plus the new dashboard-widget-options.ts module (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus — --tier answers "no path-derived mandate", so it is this seat's call.
Clause-②: no, on triage's grading rather than on my own reading, and the distinction is the whole reason this card was split. Item 1 is additive — a new warning code reporting an already-inert state, the shape that kept #12719 off the manual floor. ⚠️ The half that did carry the Clause-② content leg was split out as #12814, filed with needs:contract-review, and it has since landed (4a9f46179). ⇒ If porting item 1 turns out to change what this copy accepts or rejects, stop and report the fork; ⛔ do not proceed on the additive reading.
Serial constraints cleared: packages/sdui-parser/** is free — #12857 (#12814) landed and no open PR claims it. #12858 targets the same package (the manifest-generation guard divergence) and is NOT dispatched; ⛔ do not fold it in.

Premise re-measured on origin/main aef1b7e6

checkDashboardWidgetOptions in packages/sdui-parser/src/validate.ts -> 0 (absent, as filed)
a dashboard-widget-options module in that src tree -> 0 (absent)
positive control validateTree -> 1 (live tree being read)
input-type.ts present -> 1 (#12814's port landed)

⇒ Item 1 is genuinely unported, and item 2's infrastructure is already in place — useful context, ⛔ not licence to touch it.

⚠️ Sequencing, carried from triage and binding: the drift is latent because this repo resolves no sdui.manifest.json, so resolveSduiManifest() returns undefined and validateJsxPages runs parse-only — validateTree is not reached from the production gate at all. ⇒ unit tests are the only witness here, and this card should land before or with that wiring, ⛔ never after. That wiring remains unowned.

⛔ Do not re-derive that the grammar is in lockstep — interpretBrace is measured byte-identical, with the comparator proved able to detect a disagreement first. The gap is entirely on the diagnostic-code side.

---

Generated by Claude Code