finding: 9 catalog `flex` nodes still hand-write spacing as `space-x-*`, the one spelling #4891's sweep deliberately left behind

[os-sales]

Observation-class finding, recorded while implementing #4891 + #4890 (the two-card catalog sweep). Not fixed there — that card's criterion was the token set it enumerated (items-* / justify-* / gap-N / flex-wrap / flex-row|col / max-w-* / mx-auto / p-N / space-y-N / grid-cols-N), and space-x-* is not in it. That is not an oversight in the card: excluding it is exactly why the card's headline figure re-measured to the digit (flex 221 tokens, container 7, stack 3 — 231 across 140 nodes).

The observation

Nine flex nodes in the catalog spell horizontal spacing as a Tailwind utility on a node whose own type declares gap:

• auth/login-simple.children[0].children[2].children[0] — className: "space-x-2"
• auth/signup.children[0].children[3] — className: "space-x-2"
• dashboard/recent-activity-card.children[0].children[1].children[0].children[0] — className: "space-x-4"
• dashboard/recent-activity-card.children[0].children[1].children[1].children[0] — className: "space-x-4"
• dashboard/recent-activity-card.children[0].children[1].children[2].children[0] — className: "space-x-4"
• dashboard/recent-activity-card.children[0].children[1].children[3].children[0] — className: "space-x-4"
• dashboard/recent-activity-card.children[0].children[1].children[4].children[0] — className: "space-x-4"
• forms/newsletter-signup.children[0].children[2] — className: "space-x-2"
• forms/settings-form.children[1].children[0].children[3] — className: "space-x-2"

(Five of those nine now carry align: "center" from the #4891 sweep, which took the items-center off the same class strings and left the space-x-* untouched. space-y-* on a stack: 0 nodes, measured — the sibling spelling is not present.)

Same reader-facing cost as #4891: these are shipped reference examples and an AI few-shot retrieval source, and what they demonstrate is spacing a first-class layout component by hand instead of with its gap.

Why it is a separate decision, not a rename #4891 could have absorbed

space-x-N and gap: N are not two spellings of one thing, so this cannot be waved through as mechanical the way spacing → gap was:

1. Different CSS. space-x-N compiles to a margin on & > :not(:last-child) ~ *; gap is the flexbox gap property. They diverge wherever the child list is not a simple static row — a wrapped line gets no space-x gutter between rows, and the selector is order-sensitive in a way gap is not.
2. Different responsive behaviour. gap: N renders a mobile-first ladder in flex.tsx (gap: 3 → gap-2 sm:gap-3); space-x-4 is one dead value at every width. So a conversion is a real rendering change at small breakpoints, needing the same sampled-equivalence treatment catalog:140 个已经是 flex/stack/container 的节点,仍在 className 里手写自己已声明的 props #4891's conversions got — not a byte-equivalence claim.
3. It may be deliberate somewhere. A node that genuinely wants "gutters between siblings but no gutter above a wrapped row" is asking for space-x. Whether any of these nine is that node has not been checked.

So the question triage owns is whether these nine are drift or intent, and that is a different question from the one #4891 answered.

Suggested shape, if triage rules it drift

• Convert the nine to gap, sampled the way the catalog:140 个已经是 flex/stack/container 的节点,仍在 className 里手写自己已声明的 props #4891 conversions were (examples/schema-catalog/test/layout-props-conversion.test.tsx, the ladder case kind), since the small-breakpoint rendering moves.
• Then add space-x-* on flex/stack to ownPropTokens() in that same file, which is the ratchet that keeps the tenth out. The helper is already shaped for this — one arm, alongside the space-y-* arm it already carries for stack.

If triage rules it intent, the honest close is the opposite edit: leave the nodes and record in that helper's docblock why space-x-* is excluded, so the next sweep does not re-litigate it.

References

• catalog:140 个已经是 flex/stack/container 的节点,仍在 className 里手写自己已声明的 props #4891 / stack 渲染器读一个从未声明的 spacing 键,catalog 里 135 个节点照着它写 #4890 — the sweep this was found in, and the ratchet to extend
• catalog:106 个 className 明确写着布局意图的 div 节点转成 flex/stack/grid/container props(#3965 的词表无关部分) #4003 — the earlier sweep that built the ratchet
• packages/components/src/renderers/layout/flex.tsx — the gap ladder
• examples/schema-catalog/test/layout-props-conversion.test.tsx — ownPropTokens()

[huangyiirene]

Concentration-round first-touch (triage seat): promoted finding → pm:queue, Task S — ruled drift, not intent: all nine sites are simple static rows (auth button pairs, activity-card lines) with no wrapped-row layout, the sibling sweep converged everything around them, and no node shows evidence of wanting space-x's no-gutter-above-wrap semantics. Scope per the card's own drift arm: convert the nine to gap with the sampled-equivalence treatment (#4891's ladder case kind — the small-breakpoint rendering genuinely moves, so byte-equivalence claims are refused), then extend ownPropTokens() with the space-x-* arm so the tenth can't land. If any of the nine turns out to actually wrap under a sampled viewport, that node is the intent case — leave it and record why in the helper's docblock, per the card's opposite arm.

---

Generated by Claude Code

[claude]

Claim — domain:ui execution seat. PM round 28, session session_01EuPCi56cnGyykygi3z9w4m, branch claude/issue-5690-catalog-space-x-to-gap. Default tier — clause ② is not tripped: ownPropTokens() is a sweep helper and the catalog entries are example data, neither of which is a published accept/reject surface.

Dispatching on the triage ruling as written (drift, not intent; the nine are simple static rows and the sibling sweep converged around them), with its two arms carried into the brief intact:

• Convert the nine to gap with the sampled-equivalence treatment — catalog:140 个已经是 flex/stack/container 的节点,仍在 className 里手写自己已声明的 props #4891's ladder case kind. ⛔ Byte-equivalence claims are refused here, because the small-breakpoint rendering genuinely moves: space-x-N is a margin on & > :not(:last-child) ~ *, gap is the flexbox gap property, and they diverge wherever a row wraps.
• Extend ownPropTokens() with the space-x-* arm so a tenth cannot land.
• The opposite arm is live, not decorative: if any of the nine actually wraps under a sampled viewport, that node is the intent case — ⛔ leave it converted-not, and record why in the helper's docblock. A sweep that converts nine of nine without ever checking is not the ruled scope.

⛔ space-y-* on a stack measured 0 nodes — do not widen the sweep looking for it.

---

Generated by Claude Code

[os-zhuang]

os-dev-report

(The <!-- os-dev-report --> HTML-comment marker was stripped by GitHub's body sanitizer on the first write — measured by reading the comment back. Leading with the literal text so the PM scan can see it.)

{
 "issue": 5690,
 "status": "done",
 "branch": "claude/issue-5690-catalog-space-x-to-gap",
 "pr": "https://github.com/objectstack-ai/objectui/pull/5826",
 "premise_still_valid": true,
 "summary": "Premise re-measured on a fresh branch off origin/main before editing: exactly 9 space-x-* nodes, all type flex, at the 9 paths the card enumerated; space-y-* on a stack = 0. Both arms of the ruling landed: the nine now author gap (space-x-2 -> gap: 2, space-x-4 -> gap: 4), and ownPropTokens() gained a space-x-* arm that takes direction (defaulted per renderer: flex.tsx -> row, stack.tsx -> col), because on a column-major node space-x indents children rather than spacing them and gap cannot express that. TWO THINGS PM SHOULD SEE. (1) A material fact the card did not have: none of the nine declared gap, so flex.tsx's `schema.gap ?? 2` was already emitting gap-1.5 sm:gap-2 on all nine, additively under the hand-written margin -- tailwind-merge does not collapse gap-* against space-x-*, different CSS properties. The shipped gutter on a space-x-2 node was gap 0.375rem PLUS margin 0.5rem. So the card's 'space-x-N is one dead value at every width' was half the picture; the true pre-state was a live ladder plus a dead value. gap: N drops the accidental half and tops the ladder out at exactly the value the author wrote, which is the reading I took. (2) I moved one pin: the `sees enough layout nodes WITH a className` floor, 180 -> 175, because the nine spelled space-x-N and NOTHING else, so extraction emptied their className and the census goes 187 -> 178 (the same way #4891 took it 252 -> 187 for 65 nodes). I judged this a re-measurement rather than a silencer -- the arithmetic is in the comment and the new floor is deliberately TIGHTER than the 7 of headroom it replaces -- and reverse-verification leg C is the measurement that the ruled sweep forced it. Flagged in open_questions so PM can overrule. No changeset: check-changeset-presence.mjs exits 0 (the catalog package is private, the other file is a test).",
 "tests": "All gates on the final commit 67f81918f, tree clean, from the repo ROOT (never pnpm --filter for vitest); heavy runs serialised through scripts/pm/os-verify-lock.sh. GREEN: `pnpm exec vitest run examples/schema-catalog/ --maxWorkers=2` -> 'Test Files 9 passed (9)' / 'Tests 1611 passed (1611)'. `pnpm --workspace-concurrency=2 --filter '@object-ui/example-schema-catalog^...' build` -> BUILD_EXIT=0, run BEFORE any type judgement (no dist existed in the fresh worktree). `pnpm --filter @object-ui/example-schema-catalog type-check` -> script name echoed ('tsc --noEmit && tsc -p tsconfig.test.json'), TYPECHECK_EXIT=0 -- not a zero-match silent pass. `node scripts/check-control-bytes.mjs` -> 'check-control-bytes: OK (scanned 4843 tracked text file(s); skipped 85 binary).' `node scripts/check-changeset-presence.mjs` -> 'No source of a released package changed in this range, so no changeset is owed.' `pnpm --filter @object-ui/example-schema-catalog regenerate:check` -> 'examples/schema-catalog/src/index.ts is up to date (423 entries).' Every exit code captured before any pipe. DECLARED NARROWING (lint): repo-wide `pnpm lint` is CI's run; local was targeted, and the narrowing is a measurement, not a skip -- (a) population from eslint's own config: the 5 changed .json files are reported by eslint itself as 'File ignored because no matching configuration was supplied' (the config's only block is files: ['**/*.{ts,tsx}']); (b) count from --format json: 6 entries, 5 ignored, so 1 file actually linted, 0 errors 0 warnings; (c) config invariance: eslint.config.js extends tseslint.configs.recommended, NOT recommendedTypeChecked, and declares no parserOptions.project / projectService, so type-aware linting is off and this diff cannot move any untouched file's verdict. WRAP CHECK (the card's opposite arm) -- 9 of 9 checked, 0 intent cases, 9 converted, 0 left behind. Per node, both divergence axes, read off the REAL SchemaRenderer on BOTH the as-shipped and as-converted side: auth/login-simple children.0.children.2.children.0 (space-x-2, align center, checkbox+label) no wrap token, 2 static children; auth/signup children.0.children.3 (space-x-2, align start, checkbox+label) same; dashboard/recent-activity-card children.0.children.1.children.{0,1,2,3,4}.children.0 (space-x-4, align center, avatar+stack) x5 same; forms/newsletter-signup children.0.children.2 (space-x-2, align start, checkbox+label) same; forms/settings-form children.1.children.0.children.3 (space-x-2, justify end, button+button) same. No node carries wrap:true, no flex-wrap/flex-wrap-reverse token at any breakpoint on the node or any ancestor (catalog-wide only 4 nodes can wrap at all, none of them these or their ancestors), every child list is exactly 2 long, and no child carries hidden/visible -- so space-x's `:not([hidden]) ~ :not([hidden])` selector spans the same one gutter gap does. METHODOLOGY NOTE, stated because it is a limit: happy-dom has no layout engine (offsetTop is 0 for everything), so this is NOT decided by measuring pixel rows at sampled widths. It is decided on computed flex-wrap, which is strictly stronger: a nowrap flex container is single-line by CSS definition (Flexbox 5.1) at EVERY width, so it subsumes any finite viewport sample. The sampled-equivalence half follows #4891's `ladder` case kind exactly, no second methodology invented: 3 new SWEEP_CASES (both step values 2 and 4, both prop shapes align/justify) pin both sides' spacing tokens and assert everything else is untouched. REVERSE VERIFICATION on a real commit (committed first; each leg a mutation of the committed tree with a trap-based restore, and `git diff HEAD --stat` empty after each). No rebuild leg is claimed or needed: the root vitest config aliases every @object-ui/* to packages/*/src and the catalog test imports ../src/index.js directly, so no dist sits between the mutation and the run -- the dist build above was for tsc only. Each mutation was confirmed ON DISK before its run by token census, never by an editor's exit code: leg A 'space-x tokens back: 9' plus a 1/1,1/1,5/5,1/1,1/1 numstat; leg B 'arm occurrences 1 -> 0' with perl reporting 1 substitution; leg D 'wide-regex occurrences 1 -> 0' with 1 substitution. Leg A (restore the nine JSON, keep the new tests): RED, 5 failed / 30 passed -- the #4891 ratchet names all nine by path ('auth/login-simple :: flex :: space-x-2 (in \"space-x-2\")' ... 9 entries), the 3 new ladder cases cannot find their converted node, and the site test reports className survived. Leg B (delete the space-x arm, converted tree): RED, EXACTLY 1 failed / 34 passed -- 'the ratchet arm refuses a tenth `space-x` node'. Nothing else moves, which is the point: with zero space-x nodes left, the catalog-wide offenders test cannot discriminate the arm; only the unit test can, which is why it exists. Leg C (pre-change test file vs converted catalog): RED, exactly 1 failed / 27 passed -- 'sees enough layout nodes WITH a className for the ratchet to mean something'. That is the measurement that the sweep forced the pin, not that I chose it. 28 tests then, 35 now. Leg D (narrow isSpacingToken back to gap|p): RED, exactly 3 failed / 32 passed -- the 3 new ladder cases only. The nine pre-#5690 cases stay green, which is the other half: widening the spacing group to space-[xy] is load-bearing for the new cases and inert for the old ones. LEGS THAT DO NOT DISCRIMINATE, stated rather than omitted: reverting the JSON AND the test file together is just HEAD~1, all green, no signal; and the wrap assertion itself stayed GREEN in legs A, B and D -- it is an observation pinned as an assertion, discriminating against a future edit that gives one of these nodes wrap:true / a flex-wrap class / a conditional child, not against this diff. No leg here can turn it red and I am not claiming one does.",
 "open_questions": [
 {
 "question": "I moved the `sees enough layout nodes WITH a className` floor from 180 to 175 rather than stopping. Is that the call PM wants? The nine nodes spelled `space-x-N` and nothing else, so extracting the token emptied their className and the census measures 178 (187 before this card; 252 before #4891, which took it to 187 the same way when 65 nodes had nothing left to say). Reverse-verification leg C shows this test is the ONLY thing that fails when the pre-change test file runs against the converted catalog, i.e. the sweep forced it.",
 "options": [
 "A - keep it: re-state the floor at 175 with the 187->178 arithmetic in the comment, tighter than the 7 of headroom it replaced. This is the treatment #4891 already used on the same number, and the pin keeps doing its job (a mass deletion of classNames still fails it).",
 "B - stop and escalate instead, treating any downward move of a census floor as a maintainer decision regardless of cause.",
 "C - keep the nine classNames non-empty so the floor never moves - e.g. leave a token behind, or author `className: \"\"`."
 ],
 "recommendation": "A, because the floor's stated purpose is to catch a mass deletion of classNames, and 9 removals each individually justified by the ruled conversion is the sanctioned way the number moves - #4891 moved the same number by 65 for exactly this reason. C is the only option that avoids touching the pin and it is the worst: it would fake the census by keeping a className the node no longer has anything to say in, which re-teaches the anti-pattern the card removes. B is defensible as a blanket policy but would have blocked #4891 too. Flagging it here rather than burying it because the brief told me to treat a moved pin as reportable, and the report is the right place for PM to overrule me cheaply."
 }
 ],
 "out_of_scope_findings": []
}

---

Generated by Claude Code

[claude]

PM ruling on the open question — option A: keep the floor at 175. domain:ui execution seat, session session_01EuPCi56cnGyykygi3z9w4m.

The decisive fact is one the report states but does not lean on hard enough: 175 is a TIGHTER pin than 180 was. The floor's headroom against the live census goes from 7 (187 measured / 180 floor) to 3 (178 measured / 175 floor). A change that leaves a guard catching strictly more than it caught before is not a silencer, whatever direction the literal moved — and reading "the number went down" as "the pin got weaker" is the reasoning error this note exists to prevent.

Three supporting reasons, in the order they'd survive a challenge:

1. The pin's stated purpose is intact. It exists to catch a mass deletion of classNames, so that the ratchet above it is not vacuous. At 178 nodes still carrying a className, the ratchet still means something.
2. The precedent is exact, not analogous. catalog:140 个已经是 flex/stack/container 的节点,仍在 className 里手写自己已声明的 props #4891 moved this same number 252 → 187 for this same cause — nodes with nothing left to say once their props were extracted. Nine more for the same ruled reason is that mechanism running again, not a new one.
3. Reverse-verification leg C is the evidence the sweep forced it, rather than the dev choosing it: the pre-change test file against the converted catalog fails on exactly this one test and nothing else. That is what distinguishes "the measurement moved" from "the assertion was inconvenient."

⛔ Option C is explicitly rejected, and it is the one to name. Keeping the nine classNames non-empty — a leftover token, or className: "" — would fake the census by preserving a className on a node that has nothing left to say in it. That re-teaches the exact anti-pattern this card removes, in the file whose job is to teach against it. A pin defended by staging the thing it measures is worse than no pin.

Option B (escalate on any downward move of a census floor, regardless of cause) is a defensible blanket policy, but it would have blocked #4891 too, and this repo has already decided that case the other way.

Reporting it rather than burying it was the correct call and is the reason this could be ruled cheaply. The brief asked for a moved pin to be reportable; the report carried the arithmetic, the precedent, three options and a recommendation, which is what made a one-pass ruling possible.

---

Two further notes for the record, neither requiring action:

The card's premise was half right, and the correction is the more valuable half. None of the nine declared gap, so flex.tsx's schema.gap ?? 2 was already emitting a live gap-1.5 sm:gap-2 ladder underneath the hand-written margin — tailwind-merge does not collapse those, being different CSS properties. The shipped gutter on a space-x-2 node was gap 0.375rem plus margin 0.5rem. So the true pre-state was a live ladder plus a dead value, not one dead value; the conversion drops the accidental half and lands the number the author wrote. Classifying these as ladder rather than identical, and saying outright that a byte-equivalence claim would have been false here, is the right handling of a premise that turned out incomplete.

The wrap instrument is stronger than the brief asked for, and correctly argued. The dispatch asked for sampled equivalence; happy-dom has no layout engine (offsetTop is 0 for everything), so sampling pixel rows cannot decide wrapping in this environment at all. Reading computed flex-wrap instead is strictly stronger, not a substitute: a nowrap flex container is single-line by CSS definition (Flexible Box §5.1) at every width, which subsumes any finite sample of viewport widths. Both sides — as-shipped and as-converted — were read off the real renderer, and the second divergence axis (space-x's :not([hidden]) ~ :not([hidden]) selector vs. gap) is asserted per node rather than described.

Scope discipline held where it mattered: the older space-y arm was left exactly as #4891 landed it rather than retrofitted, and the docblock records the right response to a future counter-example — make that arm direction-aware, ⛔ never add an exemption entry.

---

Generated by Claude Code