Slate v2 integration-local triage

Status: complete

Goal

Identify .tmp/slate-v2 Playwright integration rows that are redundant, low-value, or better covered by generated stress / package contracts, so bun test:integration-local can shrink without gutting browser-regression coverage.

Constraints

• Do not remove tests by vibe.
• Prefer deleting rows that duplicate a stronger generated scenario or assert only route smoke.
• Keep tests that cover user-reported regression classes, browser selection, DOM/model parity, void navigation, table navigation, IME, clipboard, and toolbar selection repair unless there is a strictly stronger row nearby.
• Do not require bun test:integration-local just to triage; use listing, static clustering, and targeted rows first.

Working Plan

1. Inventory Playwright projects and test rows.
2. Cluster by route and scenario family.
3. Compare duplicate rows against generated stress / slate-browser contracts.
4. Produce safe-cut candidates, risky candidates, and keepers.
5. If the user approves, delete a first batch and verify with bun check plus targeted Playwright rows.

Evidence Log

• Started from user request: identify useless / DRY removable rows from bun test:integration-local, which takes over 10 minutes.
• bunx playwright test --list before cleanup: 676 rows, 30 files, 169 rows per project across Chromium, Firefox, mobile, and WebKit.
• Biggest immediate waste: playwright/stress/** was included in test:integration-local, even though test:stress and test:stress:replay already own that lane.
• Updated .tmp/slate-v2/package.json so test:integration and test:integration-local run playwright/integration only.
• bunx playwright test playwright/integration --list after cleanup: 588 rows, 28 files, 147 rows per project.
• bun check passes after the script change.
• Fail-fast browser sweep after fixing current reds: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration --max-failures=1 --reporter=line passes 588/588 across Chromium, Firefox, mobile, and WebKit.
• Final fast gate: bun check passes in .tmp/slate-v2.

Safe Cuts Found

1. Exclude playwright/stress/** from test:integration*.

   • Already applied.
   • Removes 88 listed rows from integration-local: 84 generated stress rows plus 4 skipped replay rows.
   • Coverage remains available through explicit test:stress and test:stress:replay.

2. Delete route-smoke rows that are strictly weaker than existing rows in the same file or generated stress:

   • playwright/integration/examples/richtext.test.ts
     • renders rich text
     • inserts text through browser input
     • runs a traced slate-browser scenario
   • playwright/integration/examples/markdown-shortcuts.test.ts
     • contains quote
   • playwright/integration/examples/forced-layout.test.ts
     • checks for the elements
   • playwright/integration/examples/images.test.ts
     • contains image
   • playwright/integration/examples/hovering-toolbar.test.ts
     • hovering toolbar appears

3. Delete Playwright rows that are already better covered by package-level app-owned customization tests:

   • playwright/integration/examples/markdown-preview.test.ts
     • checks for markdown
     • stronger owner: packages/slate-react/test/app-owned-customization.tsx Editable supports app-owned markdown preview projections
   • playwright/integration/examples/forced-layout.test.ts
     • remaining forced-layout row can be considered after confirming the package-level normalization/app-owned customization tests cover the exact contract.

4. DRY or chromium-scope, not delete blindly:

   • playwright/integration/examples/styling.test.ts
     • both rows are pure prop/style rendering and should become React/package tests or Chromium-only integration rows.
   • richtext kernel trace metadata rows:
     • records kernel commands for structural browser edits
     • records kernel commands for proof-handle edits
     • records allowed kernel transitions for movement commands
     • records core command metadata for keydown movement
     • records kernel policies for browser command and repair traces
     • records core command metadata for text input and delete
     • records selectionchange and repair kernel results
     • These are important contracts, but running them across all projects is wasteful. Keep one browser-level row and move the rest to package/core tests or make them Chromium-only.

Keepers

• Void / inline / table navigation rows that match recent user regressions.
• Decoration focus and render-budget rows.
• IME/composition rows.
• Clipboard/paste rows that use real browser clipboard on desktop and semantic fallback on mobile.
• Shadow DOM and iframe rows.
• The generated stress lane itself, but only through test:stress.

Execution Pass: fail-fast reds

Status: complete

Trigger: user sees many reds and wants fixes to begin before the full integration sweep finishes.

Run policy:

• Run playwright/integration with --max-failures=1 and retries disabled.
• Fix the first current red from command output.
• Rerun the focused project/file/title before continuing the broader sweep.
• Keep status in_progress until the active red cluster is fixed or a real blocker prevents autonomous progress.

Current red cluster:

• Command: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration --max-failures=1 --reporter=line
• First failure: playwright/integration/examples/inlines.test.ts / wraps typed URL text as a link command on Chromium.
• Root cause: the example put URL wrapping in onDOMBeforeInput, while the slate-browser handle's insertText() path applies Editable inputRules. That made the app behavior split between native browser events and command style text insertion.
• Patch: move the URL wrapping rule to Editable inputRules and keep the test on editor.insertText(...), which is cross-browser and exercises the shared command/input-rule path.
• Follow-up build red: EditableInputRule was not exported and EditableTextBlocks did not expose/forward inputRules to EditableDOMRoot. Patch: export the input-rule types and forward inputRules through the public Editable wrapper.
• Focused verification: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/inlines.test.ts -g "wraps typed URL text as a link command" --reporter=line passes on Chromium, Firefox, mobile, and WebKit.

Second red cluster:

• Command: same fail-fast sweep.
• First failure: playwright/integration/examples/markdown-shortcuts.test.ts / can add list items on Chromium.
• Root cause: same app policy split as inlines. Markdown shortcut conversion lived only in onDOMBeforeInput, while the browser handle's command-style insertText() path only runs Editable inputRules.
• Patch: move markdown text shortcut conversion to Editable inputRules; keep onDOMBeforeInput only for the Android pending-diff microtask.
• Focused verification: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/markdown-shortcuts.test.ts --reporter=line passes 16/16 across Chromium, Firefox, mobile, and WebKit.

Third red cluster:

• Command: same fail-fast sweep.
• First failure: playwright/integration/examples/persistent-annotation-anchors.test.ts / keeps the annotation anchor attached across fragment insert, text insert, and clear on Chromium.
• Root cause: the example's "Insert fragment before anchor" button used raw tx.nodes.insert(...) at a text point even though the contract expects Slate fragment insertion semantics, where the trailing inserted block merges with the current text block.
• Patch: select the insertion point and call tx.fragment.insert(...).
• Focused verification: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/persistent-annotation-anchors.test.ts --reporter=line passes 4/4 across Chromium, Firefox, mobile, and WebKit.

Fourth red cluster:

• Command: same fail-fast sweep.
• First failure: playwright/integration/examples/large-document-runtime.test.ts / preserves app-owned rich HTML paste over shell-backed selection on Firefox.
• Root cause: test paste transport dispatched a synthetic paste event, then forced the browser-handle fallback for Firefox. The synthetic Firefox event could mutate the shell-backed selection before the fallback inserted data.
• Deeper root cause: the model selection was imported correctly, but the app-owned rich HTML insertData handler used raw node insertion for content that should follow fragment replacement semantics. That left browser/transport differences to decide whether replacement happened.
• Patch: use the browser-handle insertData fallback directly where needed, import DOM selection before fallback, and make the custom rich HTML paste handler call tx.fragment.insert(...) with the deserialized paragraph.
• Focused verification: PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/large-document-runtime.test.ts -g "preserves .*paste" --reporter=line passes 12/12 across Chromium, Firefox, mobile, and WebKit.