fix(InputNumber): work uncontrolled with only a default value (nuxt/ui@2d4782b)

.sync/dep-parity.json

@@ -1,6 +1,6 @@
 {
   "$note": "Upstream's pinned versions for every dependency both trees declare in the SAME section, snapshotted at `cursor`. The sync ports deltas, which is correct per commit and lets a one-time divergence become permanent: once a version is off upstream's line, every later `chore(deps)` batch skips it, because those ports bump only where this fork already matched upstream's pre-image. `prettier` sat at ^3.8.4 against upstream's ^3.9.6 for that reason, through four ported batches, until this file was written. Section-aware on purpose: 18 packages are declared on both sides but in different sections — the whole `@tiptap/*` family is a peer `^3` upstream and a dependency `^3.29.2` here, and `ai` is a peer there and a devDependency here. Those are structural divergences, not drift, and comparing a peer range against a dependency range says nothing. They are absent from this file by construction rather than by omission. Guarded by `test/utils/dep-parity.spec.ts`. Refresh with `node .sync/dep-parity.mjs <path-to-nuxt-ui-mirror> [cursor]`, which preserves `exceptions`.",
-  "cursor": "ebd4adfccc9e983ccd65d8d2bbbbaaa4e8e7d9e6",
+  "cursor": "2d4782ba926ff911a2a7178372c062a3e7150fff",
   "manifests": {
     "package.json": {
       "dependencies": {

.sync/log/2d4782ba926ff911a2a7178372c062a3e7150fff.md

@@ -0,0 +1,59 @@
+# port — nuxt/ui@2d4782ba926ff911a2a7178372c062a3e7150fff
+
+**Upstream:** fix(InputNumber): work uncontrolled with only a default value (#6859)
+
+**Decision:** port, verbatim. Third of ten commits in this run.
+
+## The defect
+
+`useVModel(props, 'modelValue', emits, { defaultValue })` kept its **own** copy
+of the value and that copy was bound to `NumberFieldRoot`'s `model-value`. With
+only a `defaultValue` and no `v-model`, reka has internal state of its own, and
+the two disagreed: the field stepped off the wrapper's copy instead of off reka's,
+so an uncontrolled `InputNumber` did not increment.
+
+Removing the wrapper alone would trade that for a different bug, which is why
+the fix has a second half. Reka writes `update:modelValue` on **every** write —
+blur, Enter, stepping at a bound — whether or not the value actually moved. With
+`useVModel` gone there is nothing absorbing those, so `onUpdate` gains an
+equality guard and emits by hand:
+
+```ts
+if (value === props.modelValue || (value == null && props.modelValue == null)) {
+  return
+}
+
+emits('update:modelValue', value as ApplyModifiers<T, Mod>)
+```
+
+The second clause is not redundant. `undefined` and `null` are different values
+but the same emptiness, and the `.optional` modifier produces the first while a
+caller supplies the second.
+
+`useForwardProps(…, emits)` loses its second argument for the same reason: the
+root must not re-emit what `onUpdate` now owns.
+
+## b24ui port
+
+Applies verbatim. Our pre-image is upstream's at all five sites, `eslint-disable`
+comment included, and `useVModel` appears nowhere else in the file. Selectors and
+imports the new tests need — `data-slot="increment"`, `flushPromises`, `test` —
+were already present.
+
+## Tests
+
+Upstream's four, taken as written. Verified against the pre-image rather than
+assumed: with the component reverted and the tests kept, **all four fail**; with
+the port, all pass. So this fixes a defect that was live in this fork, not a
+theoretical one.
+
+Mutation-checked, and the counts separate the two halves of the fix:
+
+| mutation | result |
+| --- | --- |
+| drop the `emits('update:modelValue', …)` call | **14 red** |
+| drop the equality guard entirely | **6 red** |
+| keep only `value === props.modelValue`, drop the `== null` clause | **2 red** |
+
+The last row is the one worth having: it shows the `== null` clause is
+load-bearing and covered, rather than defensive padding that no test would miss.

.sync/nuxt-ui.json

@@ -1,7 +1,7 @@
 {
   "upstream": "nuxt/ui",
   "branch": "v4",
-  "cursor": "ebd4adfccc9e983ccd65d8d2bbbbaaa4e8e7d9e6",
+  "cursor": "2d4782ba926ff911a2a7178372c062a3e7150fff",
   "_cursor_note": "cursor = last upstream commit ported into b24ui. The sync is manual by decision: one commit at a time, oldest-first, each with a `.sync/log/<sha>.md` journal and an entry in `processed`. There is no dispatcher, no porter workflow and no kill-switch — `.sync/PORTING.md` is the whole procedure. `processed` is maintained per port (backfilled #68-#72 on 2026-06-09).",
   "processed": {
     "2799fa6f2b25ce3eb15e050f3ef7c57d0d9a2fdb": {
@@ -1777,10 +1777,16 @@
       "summary": "fix(Table): exclude hidden columns from colspan (nuxt/ui #6894) — PORT, verbatim. First of ten commits in this run, taken alone; the next is a chore(deps) batch with nothing in common. Five colspan expressions counted ALL columns while the body rendered only the visible ones: row.getAllCells() -> getVisibleCells() for the expanded row, and tableApi.getAllLeafColumns() -> getVisibleLeafColumns() for empty, loading and the two virtualizer padding rows. Hide a column and every full-width cell was one too wide, so the <td> overhung the row it was meant to span. Applies verbatim: our pre-image has the same five sites and the same two method names, and Table.vue:569 already rendered row.getVisibleCells() — the row body and the colspan were reading different column sets, which is the whole defect. Coverage measured PER SITE rather than assumed, and the measurement changed what got written: reverting expanded alone turns 2 red and empty alone turns 2 red, but loading alone leaves 78 green and the two virtualizer rows alone leave 78 green — so three of the five were riding on a test that cannot see them. Added one assertion for loading, which reads the count from the same object as empty but through a different branch, so a fix applied to one and not the other would still have passed; it now turns 2 red when reverted alone. The two virtualizer padding rows stay UNCOVERED and this is stated rather than glossed: they are aria-hidden spacers existing only when the virtualizer is active, so reaching them means standing up a virtualizer harness, out of proportion to a one-word change, and upstream has no such test either. The added assertion nearly went in wrong — it first passed ...props, which carries data, but the loading row is a sibling v-else-if of the rows and renders only when there are none, so the rows mounted and the query came back empty; it failed on correct code, which is the good failure mode."
     },
     "ebd4adfccc9e983ccd65d8d2bbbbaaa4e8e7d9e6": {
+      "pr": 551,
+      "b24ui_sha": "d57db68231238c2a2e06e2e96eea1dcf3be622dc",
+      "decision": "port",
+      "summary": "chore(deps): update all non-major dependencies (nuxt/ui #6925) — PORT, minus one package, plus one fork-only companion change. Second of ten commits in this run. Every shared package sat at upstream's exact pre-image, checked package by package rather than inferred from dep-parity being green: root takes @internationalized/date ^3.12.4, @internationalized/number ^3.6.8, motion-v ^2.4.2, ai ^7.0.93, eslint ^10.10.0; docs takes @ai-sdk/vue ^4.0.93, ai, joi ^18.2.8, maska ^3.2.1, motion-v; playgrounds/nuxt takes @ai-sdk/vue, @internationalized/date, ai; playgrounds/vue takes vue-router ^5.3.1. Six upstream names are absent here and the @internationalized/* PEER ranges are ^3.0.0 on both sides, untouched by upstream either. Two upstream changes n/a: the zod entry in playgrounds/nuxt/nuxt.config.ts paths (no typescript block here at all) and the ContentToc snapshot, which moves only under happy-dom 20.14 — a version we do not take. playgrounds/demo is ours alone and declares four of the bumped packages: two at upstream's pre-image take the bump mechanically, and TWO HAD DRIFTED BEHIND OUR OWN ROOT (@ai-sdk/vue ^4.0.77 vs our ^4.0.85, ai ^7.0.77 vs ^7.0.85), brought up with the rest — exactly the 'one-time divergence becomes permanent' failure dep-parity's note describes, and demo is invisible to that file because upstream has no counterpart. maska and @internationalized/date in the nuxt and vue playgrounds stay put: upstream does not declare them there, so there is no pre-image to move from — structural, not drift. THE ONE PACKAGE NOT TAKEN IS happy-dom. Upstream goes ^20.12.0 -> ^20.14.0; taking it turns 1450 tests and 136 snapshots red across 64 files with TypeError: 'defineProperty' on proxy: trap returned falsish for property '__v_skip' — Vue's markRaw against a happy-dom proxy, raised from reka's Presence watcher. Isolated rather than guessed (every other bump applied, happy-dom alone moved back => green) then bisected: 20.12.0 passes, 20.12.1 / 20.12.2 / 20.13.0 / 20.13.1 / 20.14.0 all fail. The finding is LARGER than the port: the declared range was ^20.12.0, which admits every broken version, and it worked only because the lockfile happened to hold 20.12.0 — any fresh resolution or Dependabot bump would have reddened the suite with no source change to point at, and ~20.12.0 is not enough either since 20.12.1 already breaks. So the declaration becomes an exact pin, 20.12.0, with the reason in dep-parity.json exceptions where the next chore(deps) port will read it. That closes a landmine rather than adding a restriction. Upstream ships ^20.14.0 and is presumably green, so whatever differs is on our side — reka's version or the vitest environment — and finding it is not something a dependency-bump port can do; recorded for a follow-up. Verified: lint and typecheck green, test 7857 passed / 6 skipped / 0 failed, resolved versions read from pnpm-lock.yaml rather than node_modules."
+    },
+    "2d4782ba926ff911a2a7178372c062a3e7150fff": {
       "pr": "pending-merge",
       "b24ui_sha": "pending-merge",
       "decision": "port",
-      "summary": "chore(deps): update all non-major dependencies (nuxt/ui #6925) — PORT, minus one package, plus one fork-only companion change. Second of ten commits in this run. Every shared package sat at upstream's exact pre-image, checked package by package rather than inferred from dep-parity being green: root takes @internationalized/date ^3.12.4, @internationalized/number ^3.6.8, motion-v ^2.4.2, ai ^7.0.93, eslint ^10.10.0; docs takes @ai-sdk/vue ^4.0.93, ai, joi ^18.2.8, maska ^3.2.1, motion-v; playgrounds/nuxt takes @ai-sdk/vue, @internationalized/date, ai; playgrounds/vue takes vue-router ^5.3.1. Six upstream names are absent here and the @internationalized/* PEER ranges are ^3.0.0 on both sides, untouched by upstream either. Two upstream changes n/a: the zod entry in playgrounds/nuxt/nuxt.config.ts paths (no typescript block here at all) and the ContentToc snapshot, which moves only under happy-dom 20.14 — a version we do not take. playgrounds/demo is ours alone and declares four of the bumped packages: two at upstream's pre-image take the bump mechanically, and TWO HAD DRIFTED BEHIND OUR OWN ROOT (@ai-sdk/vue ^4.0.77 vs our ^4.0.85, ai ^7.0.77 vs ^7.0.85), brought up with the rest — exactly the 'one-time divergence becomes permanent' failure dep-parity's note describes, and demo is invisible to that file because upstream has no counterpart. maska and @internationalized/date in the nuxt and vue playgrounds stay put: upstream does not declare them there, so there is no pre-image to move from — structural, not drift. THE ONE PACKAGE NOT TAKEN IS happy-dom. Upstream goes ^20.12.0 -> ^20.14.0; taking it turns 1450 tests and 136 snapshots red across 64 files with TypeError: 'defineProperty' on proxy: trap returned falsish for property '__v_skip' — Vue's markRaw against a happy-dom proxy, raised from reka's Presence watcher. Isolated rather than guessed (every other bump applied, happy-dom alone moved back => green) then bisected: 20.12.0 passes, 20.12.1 / 20.12.2 / 20.13.0 / 20.13.1 / 20.14.0 all fail. The finding is LARGER than the port: the declared range was ^20.12.0, which admits every broken version, and it worked only because the lockfile happened to hold 20.12.0 — any fresh resolution or Dependabot bump would have reddened the suite with no source change to point at, and ~20.12.0 is not enough either since 20.12.1 already breaks. So the declaration becomes an exact pin, 20.12.0, with the reason in dep-parity.json exceptions where the next chore(deps) port will read it. That closes a landmine rather than adding a restriction. Upstream ships ^20.14.0 and is presumably green, so whatever differs is on our side — reka's version or the vitest environment — and finding it is not something a dependency-bump port can do; recorded for a follow-up. Verified: lint and typecheck green, test 7857 passed / 6 skipped / 0 failed, resolved versions read from pnpm-lock.yaml rather than node_modules."
+      "summary": "fix(InputNumber): work uncontrolled with only a default value (nuxt/ui #6859) — PORT, verbatim. Third of ten commits in this run. The defect: useVModel(props,'modelValue',emits,{defaultValue}) kept its OWN copy of the value and that copy was bound to NumberFieldRoot's model-value, so with only a defaultValue and no v-model the field stepped off the wrapper's copy instead of off reka's internal state and did not increment at all. Removing the wrapper alone trades that for a different bug, which is why the fix has a second half: reka writes update:modelValue on EVERY write — blur, Enter, stepping at a bound — whether or not the value moved, and with useVModel gone nothing absorbs those, so onUpdate gains an equality guard and emits by hand. The guard's second clause is not redundant: undefined and null are different values but the same emptiness, and the .optional modifier produces the first while a caller supplies the second. useForwardProps loses its emits argument for the same reason — the root must not re-emit what onUpdate now owns. Applies verbatim: our pre-image is upstream's at all five sites, eslint-disable comment included, useVModel appears nowhere else in the file, and the selectors and imports the new tests need (data-slot=increment, flushPromises, test) were already present. Upstream's four tests taken as written and VERIFIED AGAINST THE PRE-IMAGE rather than assumed: with the component reverted and the tests kept, all four fail; with the port, all pass — so this fixes a defect that was live in this fork, not a theoretical one. Mutation-checked, and the counts separate the two halves of the fix: dropping the emits('update:modelValue') call turns 14 red, dropping the equality guard entirely turns 6 red, and keeping only value === props.modelValue while dropping the == null clause turns 2 red. That last row is the one worth having — it shows the == null clause is load-bearing and covered rather than defensive padding no test would miss."
     }
   }
 }

src/runtime/components/InputNumber.vue

@@ -108,7 +108,7 @@ export interface InputNumberSlots {
 <script setup lang="ts" generic="T extends InputNumberValue = InputNumberValue, Mod extends Pick<ModelModifiers, 'optional'> = Pick<ModelModifiers, 'optional'>">
 import { onMounted, computed, useTemplateRef, toRef } from 'vue'
 import { NumberFieldRoot, NumberFieldInput, NumberFieldDecrement, NumberFieldIncrement } from 'reka-ui'
-import { reactivePick, useVModel } from '@vueuse/core'
+import { reactivePick } from '@vueuse/core'
 import { useAppConfig } from '#imports'
 import { useComponentProps } from '../composables/useComponentProps'
 import { useForwardProps } from '../composables/useForwardProps'
@@ -136,13 +136,10 @@ defineSlots<InputNumberSlots>()
 
 const props = useComponentProps<InputNumberProps<T, Mod>>('inputNumber', _props)
 
-// eslint-disable-next-line vue/no-dupe-keys
-const modelValue = useVModel<InputNumberProps<T, Mod>, 'modelValue', 'update:modelValue'>(props, 'modelValue', emits, { defaultValue: props.defaultValue })
-
 const { t } = useLocale()
 const appConfig = useAppConfig() as InputNumber['AppConfig']
 
-const rootProps = useForwardProps(reactivePick(props, 'as', 'stepSnapping', 'formatOptions', 'disableWheelChange', 'invertWheelChange', 'required', 'readonly', 'focusOnChange', 'locale'), emits)
+const rootProps = useForwardProps(reactivePick(props, 'as', 'stepSnapping', 'formatOptions', 'disableWheelChange', 'invertWheelChange', 'required', 'readonly', 'focusOnChange', 'locale'))
 
 const { emitFormBlur, emitFormFocus, emitFormChange, emitFormInput, id, color: formFieldColor, size: formFieldSize, name, highlight: formFieldHighlight, disabled: formFieldDisabled, ariaAttrs } = useFormField<InputNumberProps<T, Mod>>(_props)
 const { orientation, size: fieldGroupSize } = useFieldGroup<InputNumberProps<T, Mod>>(_props)
@@ -184,9 +181,16 @@ const inputRef = useTemplateRef('inputRef')
 
 function onUpdate(value: ApplyModifiers<T, Mod> | undefined) {
   if (props.modelModifiers?.optional) {
-    modelValue.value = value = value ?? undefined
+    value = value ?? undefined
   }
 
+  // In controlled mode reka emits on every write, even when nothing changed (blur, Enter, stepping at a bound).
+  if (value === props.modelValue || (value == null && props.modelValue == null)) {
+    return
+  }
+
+  emits('update:modelValue', value as ApplyModifiers<T, Mod>)
+
   // @ts-expect-error - 'target' does not exist in type 'EventInit'
   const event = new Event('change', { target: { value } })
   emits('change', event)
@@ -222,7 +226,7 @@ defineExpose({
     v-bind="rootProps"
     :id="id"
     :default-value="props.defaultValue"
-    :model-value="modelValue"
+    :model-value="props.modelValue"
     :min="props.min"
     :max="props.max"
     :step="props.step"