fix(ui): NcButton native-type is inert on @nextcloud/vue 9 — six forms never submitted

[rubenvdlinde]

The defect

@nextcloud/vue 9 renamed NcButton's visual prop type → variant and repurposed type as the native button type, with default: "button". nativeType was removed outright — it was the v8 name.

So native-type="submit" on v9 is an undeclared prop: it falls through to the DOM as an inert attribute while the button keeps type="button". A type="button" inside a <form> raises no submit event, so @submit.prevent never runs.

No console warning, no lint error, no failed request. User-facing: someone fills in a purchase order, clicks Create purchase order, sees no error, and nothing was ever sent.

This repo is on @nextcloud/vue ^9.9.0, lockfile-resolved to 9.9.0 — checked in the lockfile, not the caret.

The six sites

Every one already used variant= (the v9 visual prop), so these were half-finished v8→v9 migrations where only the button type was missed. Each verified per-site to sit inside a form with a submit handler — not blanket-rewritten:

site	enclosing form
src/components/goods-receipt-note/GoodsReceiptNoteForm.vue:152	<form @submit.prevent="onSubmit">
src/components/purchase-order/PurchaseOrderForm.vue:160	<form @submit.prevent="onSubmit">
src/views/DeadlineCalendarSettings.vue:57	<form @submit.prevent="save">
src/views/bookings/BookingForm.vue:63	<form @submit.prevent="submit">
src/views/settings/PipelinqIntegration.vue:38	<form @submit.prevent="save">
src/views/settings/Settings.vue:21	<form @submit.prevent="save">

Each is native-type="submit" → type="submit"; the diff is exactly 6 changed lines in 6 files. git grep native-type -- src now returns nothing.

Evidence

Behaviour is proven in ConductionNL/openconnector#1140, which mounts the real NcButton from the installed 9.9.0 and asserts DOM behaviour, with negative controls:

spelling	renders	submit handler
type="submit" (the fix)	type=submit	fires
native-type="submit" (the bug)	type=button	never fires

That test is mutation-checked: flipping the fix assertion back to the broken spelling makes it fail, so the assertion is shown capable of failing rather than merely passing.

The component-level mechanism is shared by all six sites here. What is verified by reading rather than execution is that each of these specific forms wraps its button — the table above — which is why every site was confirmed individually.

Coordination note

Peer agents were reported to have fixed some of these files earlier. At the time this branch was cut, all six were still present on origin/development and no remote branch had them fixed, so all six are included here. If a peer PR lands first, this rebases down to whatever remains rather than conflicting on content.

🤖 Generated with Claude Code

[github-actions]

Quality Report — ConductionNL/shillinq @ 1d1b275

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
composer			✅	✅ 100/100
npm			✅	✅ 547/547
PHPUnit					✅
Newman					❌
Playwright					⏭️

---

Quality workflow — 2026-08-04 10:18 UTC

Download the full PDF report from the workflow artifacts.

[rubenvdlinde]

Merging with --admin. Failure set is identical to the base, compared against the last completed run rather than an in-flight one:

	run	failing
this PR	latest	Integration Tests (Newman), Quality Report
development	30896861325 (31 jobs)	same two

Quality Report is red only because Newman is. Newman is pre-existing test debt and is being worked separately; this PR adds no new failure.

⚠️ Method note, since it bit me twice today: a run with a blank conclusion is still running, not passing. Reading an in-flight base as "no failures" would have made this PR look like it introduced two — I compared against the last completed verdict instead.