test-only: replay #1806 (smooth drag-and-drop reordering of relation chips) on master

[jwalkingjew]

Replay of #1806 on current master, opened so a Vercel preview builds and the drag behavior can be tested. #1806 is 378 commits behind master and conflicting, so its preview is unusable. Not for merge — test target only. Review and merge the original.

What

Makes drag-and-drop reordering of relation chips feel smooth, including wrapped multi-row chip lists.

How

Migrates this one component to the next-gen dnd-kit:

• @dnd-kit/react + @dnd-kit/helpers (0.5.0) — measures the real DOM and FLIP-animates optimistic reordering, so variable-width wrapped chips reflow smoothly and land correctly. Reordering is optimistic during the drag; positions are committed on drop by reassigning the existing position set (same scheme as before).
• chip.tsx: LinkableRelationChip gains an optional sortableDragHandleRef (the new API marks drag handles via callback ref instead of spread listeners). The dots button remains the handle; grab UX unchanged.
• editable-entity-page.tsx: the add-entity button moves into the chip list via the new afterChips slot; IMAGE/VIDEO branching normalized to renderableTypeStrict (behavior-equivalent).

The classic @dnd-kit/core/@dnd-kit/sortable packages remain — other surfaces (table rows, tabs, etc.) still use them; the libraries coexist.

Differences from #1806

Content is identical apart from one conflict that master's movement forced:

• vitest.setup.ts did not exist when fix: smooth drag-and-drop reordering of relation chips #1806 branched. Master has since added it (the unmocked-network guard) and wired setupFiles in vite.config.js. So the ResizeObserver stub — needed because @dnd-kit/dom references ResizeObserver at module scope and jsdom does not implement it — is appended to master's setup file instead of creating one, and fix: smooth drag-and-drop reordering of relation chips #1806's duplicate setupFiles key in vite.config.js is dropped (vite.config.js is now unchanged from master).

reorderable-relation-chips-dnd.tsx is byte-identical to #1806. Original authorship preserved on the commit.

Verification on this branch

• bun install leaves bun.lock unchanged — lockfile is consistent with master's dependency set
• tsc --noEmit: only the 3 pre-existing failures already on master; none in the touched files
• bun run build: passes
• Full vitest suite: 462 files / 5122 tests passing

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
geogenesis	Ready	Preview	Sep 20, 2026 2:29am UTC

[ohohoreilly]

Checked this while auditing the PR queue, and it is worth un-parking.

The bug it replays is still live on master. reorderable-relation-chips-dnd.tsx there still imports @dnd-kit/core / @dnd-kit/sortable with CSS.Transform, and @dnd-kit/react is not in package.json. Nothing else has landed the migration since May, so wrapped multi-row chip lists still reorder the way #1806 set out to fix.

I briefly closed #1806 in a sweep of stale conflicting PRs and have reopened it — that was my error, and this PR is the evidence I should have looked at first.

This looks like the landable artifact, not #1806. That branch is 378 commits behind for the sake of one component; this one is on master, reorderable-relation-chips-dnd.tsx is byte-identical to it, original authorship is preserved on the commit, and it reports tsc clean, build passing, bun.lock unchanged, and 462 files / 5122 tests green.

So "not for merge — review and merge the original" is the more expensive path by a wide margin. Un-drafting this and landing it looks strictly better, unless preserving #1806 as the merged artifact matters for reasons outside the code.

One thing to confirm before it lands, since it is the part most likely to bite: the ResizeObserver stub appended to vitest.setup.ts. @dnd-kit/dom touches ResizeObserver at module scope and jsdom does not implement it, so that stub is load-bearing for every test file the setup touches — not just these. Worth a look that it cannot mask a real missing-observer failure elsewhere.