[bbingz]

Summary

Fixes #8879. Since 22d5989, dragging a project header or Project Group header only produced a drop indicator while the pointer was inside another header's own ~28px strip; over expanded section bodies — most of the sidebar — the preview returned null, the indicator unmounted, and releasing was a silent no-op. Precise reordering with expanded sections was nearly impossible.

computeWorktreeSidebarHeaderDropPreview now maps a pointer in a section body or interior gap to the nearest boundary slot (previous section's bottom vs next header's top, midpoint rule; ties resolve toward the next header). The indicator follows the pointer through the whole list, and drops land where the indicator shows — dropIndex is derived from the same slot that positions the line, for both header consumers.

Guard against fabricated drops: the measured content bound (container.scrollHeight, same coordinate space as the virtualizer-measured header tops) is applied up front to every branch, so a pointer below the actually-rendered list still cancels exactly like before. This also tightens the pre-existing edge-zone/final-boundary path — an overshoot final drop that lands below real content now cancels too (regression test added); that path's internal slot logic is otherwise unchanged. #6609's reorder-index fix (sibling reindexing) is untouched; that issue never asked for a dead zone — the same commit even accepts sectionBottom + gap for the final boundary slot, so the interior dead zone was accidental scope.

Out of scope (documented in #8879): the Math.max(scrollTop, …) indicator clamp divergence, and replacing the estimator-based per-section boundaries with measured offsets (pre-existing, shared with the final-boundary path — see Notes).

Screenshots

UI behavior change (drop indicator tracking). No recording attached: the semantics are pinned by the unit suites (including the short-list differential regression added in 0788d5ad8) and the rewritten e2e spec in CI.

Testing

• pnpm lint — oxlint clean on all 8 touched files (one hook refactored to stay under the max-lines cap rather than bumping it); full lint chain deferred to CI
• pnpm typecheck — node, cli, and web tsconfigs all pass
• pnpm test — all 17 sidebar drag/drop suites: 118 passed (focused trio: 49); full suite deferred to CI
• pnpm build / e2e — deferred to CI (no quick single-spec e2e path; the rewritten spec's expected order is derived from commitProjectHeaderDragDrop's ungrouped path and is boundary-choice-independent at the drag target used)
• Added or updated high-quality tests that would catch regressions

Test coverage: the three previous expect-null pins are flipped to nearest-boundary assertions; new cases pin both boundary choices across an interior gap and the midpoint tie; a content-bound group pins both drift directions of the estimated section end (overshoot: pointer below measured content → null; within measured content → still snaps), for both the project-header and project-group-header consumers; the pinned final-boundary tests and the in-band header behavior stay byte-identical.

AI Review Report

Two adversarial Codex rounds (gpt-5.6-sol, read-only sandbox). Round 1 surfaced a concern about the estimated sectionBottom; adjudicated with a full evidence chain: the per-section boundary estimator (worktree-header-section-boundaries.ts) omits the virtualizer's 6px inter-row gap and measured sizes, but it is pre-existing shared infrastructure — the new snap consumes it byte-identically to two shipped paths, committed dropIndex is a pure row-model integer classified against measured header tops (drift-immune), and near boundaries the nearest-boundary choice self-corrects to measured geometry. Round 2 found one high-severity real defect in the first draft: with estimate overshoot, a pointer below the actually-rendered list could be classified "inside" and commit a reorder where the parent returned null. Fixed exactly per the reviewer's recommendation — an authoritative measured content bound rejects such points before snapping — with tests for both drift directions and both consumers. Cross-platform: pure renderer pointer-geometry; no platform APIs, shortcuts, or paths touched.

Security Audit

Renderer-only pointer-geometry logic. No input handling, IPC, command execution, path handling, auth, secrets, or dependency changes. The one security-adjacent failure mode — converting an outside-the-list release (cancel intent) into a persistent reorder — was caught in review and is now bounded by measured content height with regression tests.

Notes

• Cosmetic residual (pre-existing): for drops deep inside a very tall last section, the indicator line can render a few px off the true boundary because the shared per-section estimator undershoots by ~6px per row; the committed slot is unaffected. A proper fix should replace the estimator with measured offsets across all three consumers at once (preview snap, in-band bottom-half, final-boundary path) — happy to file that as a follow-up if maintainers want it.
• The rewritten e2e ("dropping a project over another project body snaps to that section boundary", expected order [bravo, alpha, charlie]) encodes the new WYSIWYG semantics; the old test pinned the accidental no-op.

[coderabbitai]

📝 Walkthrough

Walkthrough

Sidebar drag previews now receive the measured scroll-content bottom and snap drops in expanded or interior section areas to the nearest header boundary. The shared preview logic handles midpoint tie-breaking and rejects positions below the rendered content end. Unit tests cover boundary selection and content limits, while the end-to-end manual-sort test verifies the resulting persisted project order.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The changes implement #8879's nearest-boundary snapping for project and Project Group headers and preserve the existing reorder-index fix.
Out of Scope Changes check	✅ Passed	The touched files and tests stay within the sidebar header drag-drop fix; no unrelated code paths or features were introduced.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Title check	✅ Passed	The title clearly summarizes the main change: sidebar header drags snap to the nearest boundary slot instead of a dead zone.
Description check	✅ Passed	All required sections are present and substantive, with testing, review, security, and notes covered; screenshots are deferred to a follow-up comment.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 400c3872-a860-4aa7-a4eb-a9b87027b309

📥 Commits

Reviewing files that changed from the base of the PR and between 190ed7d and 6b69cd2.

📒 Files selected for processing (8)

• src/renderer/src/components/sidebar/project-group-header-drag.ts
• src/renderer/src/components/sidebar/project-group-header-drop.test.ts
• src/renderer/src/components/sidebar/project-group-header-drop.ts
• src/renderer/src/components/sidebar/project-header-drag.ts
• src/renderer/src/components/sidebar/project-header-drop.test.ts
• src/renderer/src/components/sidebar/project-header-drop.ts
• src/renderer/src/components/sidebar/worktree-sidebar-header-drop-preview.ts
• tests/e2e/project-group-manual-sort.spec.ts

[AmethystLiang]

reviewing now