chore(sync): resolve forward-merge conflict (main into next)

[mfal]

Resolves the forward-merge conflict between main and next (ADR 0004 §4), resolved locally with the merge drivers active via pnpm sync:resolve.

Merge this as a true merge commit — a squash or rebase merge would break the superset invariant (ADR 0004 §1) and is blocked by branch protection on next.

This PR is not only conflict resolution. Beyond the three conflicts it carries a generator fix, a new test, and one regenerated file — two cross-side interactions that only break once both branches are combined. Those are the parts worth reviewing.

The three conflicts

apps/docs/src/lib/PropertiesTables/lib/loadProperties.ts — a real semantic conflict. next moved the properties table onto the generated component index (#2897); main taught it to hide @deprecatedValues (#2960, on top of #2961). Kept next's data source and ported main's value hiding onto the new prop shape: name comes from the map key, type is a plain string, description is optional.

Dropped the top-level undefined/null filter from the docs side — the index generator already strips them in dev/component-index/filterProps.ts, so keeping it here would duplicate the rule. The docs side now only removes deprecated values.

docs/adr/0004-forward-merge-main-into-next.md (§3) — took main's passage verbatim. It already contains both changes: the publish-next.yml → publish.yml rename (#2968) and the version-consistency-guard paragraph (#2958).

pnpm-lock.yaml — kept next's, verified rather than assumed. Zero entries exist only on main (next already carried the same Dependabot bumps), and pnpm install reports it up to date against every merged manifest.

Two cross-side interactions

Neither is in a conflicted file. Both are silent — nothing in either branch's own CI run would have caught them.

1. Button's color came out of the merge flagged deprecated as a whole. main's new @deprecatedValues tag contains the substring next's index generator matched to set deprecated: true:

if (prop.description.includes("@deprecated")) {   // matches "@deprecatedValues accent"

dev/component-index/buildComponentIndex.ts now uses main's /@deprecated(?!\w)/, which was introduced in #2960 for exactly this distinction. Without it, #2960's "nothing about the prop itself changes" stops holding — color keeps working and still warns via useWarnDeprecation, but the docs would have advertised the whole prop as deprecated. A regression test covers it (verified to fail on the old code).

2. RemoteRatingSegmentElement carried a render remote property. main put react-aria's render on the generator's global ignore list (#2966) after next had generated RatingSegment (#2863), so the merged generator output differs from what is committed. Regenerated.

This is not cosmetic: render returns rendered output the host would have to call, and one such property drops the entire mutation batch over postMessage — the bug #2966 fixed for Table. It would also have failed CI's "Check all generated code is committed" gate.

Verification

Check	Result
pnpm lint (eslint + stylelint + format:check)	exit 0 — 0 errors, warnings all pre-existing
node .github/scripts/version-consistency-guard.mjs	16 managed packages at 1.1.0-next.0
node --test .github/scripts/*.test.mjs	38/38
unit tests (components, docs, remote-*)	445 passed
test:compile components, tsc --noEmit apps/docs	clean
pnpm build + git diff	clean after the one regenerated file

End-to-end against the real generated index, Button's color reads … | "secondary" with accent gone, deprecated: false, and the tag stripped from the description. ColumnLayout's s still reads (number | null)[], so #2961's nested-null fix survives.

Carries the verify-only run-visual-tests label: this is the first time main's Table and FileField fixes render alongside next's new RatingSegment, and no prior PR's visual run covered that combination.

Escalation: #2963 — closed automatically once next contains main again.

The red run-visual-tests check is inherited from main

Do not add update-screenshots to this PR. All four required checks (main, Conventional PR title, Routing, Version contract) pass; run-visual-tests is not required.

Two Remote/firefox cases in List.browser.test.tsx fail. Both already fail on main's tip — verified by running the same test on each parent commit in a clean worktree:

Test (Remote, firefox)	main @ 83d6086	next @ b171f12	this PR
List items	✗ 782 px (0.01)	✗ 782 px (0.01)	✗ 782 px (0.01)
List date range filter	✗ 529 px (0.01)	✓	✗ (Linux only)
List table	✓	✗ 95158 px (0.11)	✓ fixed here

The merge does not cause either failure, and it repairs List table (Remote), which main's #2966 fixes and next was still failing.

The List items diff is the focus ring missing on the checked filter item in the dropdown — 782 px is the ring, not noise. It comes with an unhandled error that reproduces on both parents:

TypeError: listener is not a function
  at eventListenerCallbackWrapper (@mittwald/remote-dom-react/host/hooks/props-for-element.mjs:54)
  at onChange (packages/components/src/lib/remote/useControlledHostValueProps.ts:34)

The host hands down an onChange that is truthy but not a function, so the onChangeFromProps?.(v) guard does not catch it. That is a real bug on both release lines and wants its own fix — baking the current rendering into the baselines would entrench it.

Nothing here blocks this PR; the visual suite only ran because the label was applied deliberately, and the workflow removed it again after the run.

[github-actions]

🚀 Preview Deployment

Preview environments are ready:

Type	URL
docs	pr-2981.docs.review.flow-components.de
storybook	pr-2981.storybook.review.flow-components.de

Images:

• docs: ghcr.io/mittwald/flow/docs:pr-2981
• storybook: ghcr.io/mittwald/flow/storybook:pr-2981

[github-actions]

❌ Visual Regression Tests Failed

The visual test step did not pass. If snapshots differ from the committed baselines, download the visual-diffs artifact from this run to inspect the actual/diff images (the artifact is absent if the run failed before comparison, e.g. during install or browser setup).

If the differences are intentional, update the baselines by adding the update-screenshots label to the PR.

Run details

[mfal]

⚠️ Please do not follow the update-screenshots suggestion above on this PR. It would entrench a real regression.

The run-visual-tests failure is inherited from main, not caused by this merge. Verified by running the same test on each parent commit in a clean worktree:

Test (Remote, firefox)	main @ 83d6086	next @ b171f12	this PR
List items	✗ 782 px (0.01)	✗ 782 px (0.01)	✗ 782 px (0.01)
List date range filter	✗ 529 px (0.01)	✓	✗ (Linux only)
List table	✓	✗ 95158 px (0.11)	✓ fixed here

Neither branch touched anything that could affect it — next only changed Rating, main only Button/FileField/List/Table. The merge in fact repairs List table (Remote), which next was still failing until #2966 arrived.

The 1% diff is not noise: the checked filter item in the open dropdown loses its focus ring, and those 782 px are the ring. It comes with an unhandled error that reproduces on both parents:

TypeError: listener is not a function
  at eventListenerCallbackWrapper (@mittwald/remote-dom-react/…/host/hooks/props-for-element.mjs:54)
  at onChange (packages/components/src/lib/remote/useControlledHostValueProps.ts:34)

Line 34 is onChangeFromProps?.(v) — the host passes an onChange that is truthy but not a function, so the optional-call guard does not catch it. That wants its own fix: on the main line so it forward-merges up.

It went unnoticed because run-visual-tests is an opt-in label, not a required check. All four required checks here (main, Conventional PR title, Routing, Version contract) pass.

🤖 Addressed by Claude Code