Slate v2 New Tests Architecture Cleanup Ralplan

status: done score: 0.94 date: 2026-05-08 target: .tmp/slate-v2 source plan: docs/editor-test-harvester/lexical/report.md skill: slate-ralplan

Current Verdict

Do not rewrite Slate v2 for this. The new Lexical-derived tests are a useful pressure pass, and the selected lane is green, but the cleanup target is small:

1. Keep external HTML paste policy app-owned and move the growing parser out of the example component body.
2. Split boolean mark controls from style-only leaf attributes so examples stop repeating casts around state.marks.get().
3. Keep table tests honest: current Slate v2 proves containment and cell selection, not Lexical's whole-table selection model.
4. Keep browser transport claims explicit: native clipboard, synthetic ClipboardEvent, semantic handles, and raw device proof are different claims.

The architecture is basically right. The current rough edge is example-local typing and organization, not core/runtime design.

Execution Update

Implemented the approved cleanup in .tmp/slate-v2:

• added site/examples/ts/mark-utils.ts;
• changed site/examples/ts/custom-types.d.ts so CustomTextKey is derived from boolean leaf fields and excludes style-only attrs like fontSize;
• removed repeated state.marks.get() casts from richtext.tsx, iframe.tsx, and hovering-toolbar.tsx;
• extracted paste HTML import/parser policy into site/examples/ts/paste-html-import.ts;
• kept site/examples/ts/paste-html.tsx focused on rendering and editor composition.

No core, runtime, public API, issue claim, or table-selection model changed.

Intent Boundary Record

• intent: clean up the architecture exposed by the new harvested tests without losing the green proof.
• desired outcome: all harvested test rows stay green, the example code is easier to maintain, and Slate core remains unopinionated.
• in scope:
  • .tmp/slate-v2/site/examples/ts/paste-html.tsx
  • .tmp/slate-v2/site/examples/ts/custom-types.d.ts
  • repeated mark helpers in richtext.tsx, iframe.tsx, and hovering-toolbar.tsx
  • proof naming and verification commands around the new Playwright rows
• non-goals:
  • no core rich-HTML import API
  • no whole-table selection feature without a first-class table-selection model
  • no raw mobile claim without Appium/device artifacts
  • no ProseMirror/Lexical API cloning
• decision boundary: this plan may require example-local helper extraction and type cleanup; it must not move source-app paste policy into slate, slate-dom, or slate-react.
• unresolved user decision points: none for this cleanup. A future table selection model is a separate product/API decision.

Pressure test: if the fix starts adding Google Docs, Google Sheets, Word, or table-selection policy to Slate core, it is going the wrong way.

Decision Brief

Principles:

• Slate core owns model, operations, DOM/runtime boundaries, and extension points.
• Apps/examples own source-specific import policy.
• Browser proof must name the real transport.
• Tests should describe behavior, not Lexical internals.
• Small helper extraction beats a grand rewrite here.

Top drivers:

• The new tests are already green in the selected lane.
• paste-html.tsx now mixes UI, parser policy, transport handling, and leaf rendering.
• CustomTextKey now excludes fontSize, but callers still cast marks because boolean marks and style attributes share one broad leaf type.

Viable options:

Consequences:

• The next implementation should be boring: helper extraction plus type cleanup.
• No behavior should change.
• If any harvested test fails after the refactor, revert the refactor shape before touching runtime code.

Live Source Evidence

Ecosystem Strategy Synthesis

Public API Target

No public Slate API change for this cleanup.

Accepted target:

• dom.clipboard.insertData remains the app-owned rich paste extension point.
• fontSize stays a leaf attribute in the paste-html example.
• Boolean toolbar marks use an explicit example-local boolean mark type/helper.
• No whole-table selection public API is introduced by this lane.

Rejected target:

• editor.clipboard as a new public namespace.
• core parseExternalHtmlFromGoogleDocsOrSheets.
• Lexical-style TableSelection without a real Slate table-selection design.

Internal Runtime Target

No runtime rewrite. The existing runtime tests are good because they isolate transport classes:

• IME composition: stress rows plus slate-browser helper.
• Browser transport: plaintext and editable-void Playwright rows.
• Clipboard model boundary: slate-dom unit tests.
• App rich HTML policy: paste-html example Playwright rows.

The next implementation should avoid touching slate-react runtime unless a test fails after pure helper extraction.

Hook, Component, And Render DX Target

Implement a tiny example-local helper layer:

• site/examples/ts/mark-utils.ts
  • type BooleanCustomTextKey = ...
  • toggleBooleanMark(editor, key)
  • isBooleanMarkActive(editor, key)
• Replace repeated casts in richtext, iframe, and hovering-toolbar.
• Keep fontSize out of toolbar mark controls.

For paste HTML:

• Extract parser/policy helpers from paste-html.tsx into an example-local file, likely site/examples/ts/paste-html-import.ts.
• Keep rendering components in paste-html.tsx.
• Keep the parser narrow: safe schemes, basic tags, inline styles already covered by tests. No generic sanitizer engine.

Plate And slate-yjs Migration Backbone

No new adapter work. The cleanup preserves the important migration shape:

• raw Slate remains unopinionated;
• app/plugin layers own rich import policy;
• history composition proof stays operation/transaction-level;
• transport proof remains browser/runtime-level.

This is friendly to Plate and slate-yjs because it does not bake product import rules into the core substrate.

Issue Ledger Accounting

ClawSweeper related-issue pass: skipped.

Reason: the implementation is a behavior-neutral helper extraction/type cleanup. It does not change public API, runtime behavior, browser behavior, examples' observable output, issue claims, or PR narrative. Existing issue-ledger rows and PR references stay unchanged.

Known nearby issue refs from existing ledgers:

Legacy Regression Proof Matrix

Required after implementation:

cd /Users/zbeyens/git/slate-v2
bunx playwright test playwright/integration/examples/paste-html.test.ts --project=chromium
bunx playwright test playwright/integration/examples/plaintext.test.ts --project=chromium
bunx playwright test playwright/integration/examples/plaintext.test.ts --project=firefox
bunx playwright test playwright/integration/examples/editable-voids.test.ts --project=chromium -g "keeps native paste inside editable void input"
bunx playwright test playwright/integration/examples/editable-voids.test.ts --project=firefox -g "keeps native paste inside editable void input"
PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/tables.test.ts --project=chromium
STRESS_FAMILIES=ime-composition-inline-void-boundary,ime-composition-undo,paste-html-image-void PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/stress/generated-editing.test.ts --project=chromium
bun check

Run Playwright commands sequentially. The Next webserver build can collide when parallelized.

Fresh planning-pass verification, run from /Users/zbeyens/git/slate-v2 on 2026-05-08:

These proved the current new-test lane was green before cleanup.

Fresh implementation verification, run from /Users/zbeyens/git/slate-v2 after cleanup:

Implementation Skill Review Matrix

High-Risk Pre-Mortem

1. Parser extraction changes behavior by accident.
   • proof: full paste-html Chromium file plus stress image paste row.
2. Mark helper hides fontSize and breaks formatted paste.
   • proof: paste-html font-size row plus site typecheck.
3. Table tests get "improved" into false whole-table claims.
   • proof: keep table assertions as containment/cell-selection only.

Rollback answer: revert helper extraction first; do not patch runtime unless a focused test proves runtime drift.

Hard Cuts And Rejected Alternatives

• Drop: core rich HTML parser.
• Drop: generic sanitizer package in Slate core.
• Drop: whole-table selection in this lane.
• Drop: raw mobile closure from Playwright mobile viewport.
• Keep: current dom.clipboard.insertData capability shape.
• Keep: current slate-browser transport distinction.

Maintainer Objection Ledger

Pass-State Ledger

Plan Deltas From Review

• Added a cleanup plan on top of the completed Lexical apply lane.
• Downgraded "improve Slate v2 architecture" from rewrite to two example-local refactors.
• Kept full-table selection out of scope.
• Kept raw mobile proof deferred.

Open Questions

None for this cleanup. Future table selection needs a separate API/model plan.

Implementation Phases

1. Mark typing cleanup.
   • Add example-local boolean mark type/helper.
   • Remove repeated state.marks.get() casts.
   • Verify site typecheck and richtext/iframe/hovering-toolbar behavior.
2. Paste HTML helper extraction.
   • Move parser/policy helpers out of the component file.
   • Keep insertHtmlData wired through dom.clipboard.insertData.
   • Keep fontSize rendering behavior unchanged.
3. Sequential proof.
   • Run focused Playwright rows.
   • Run bun check.
4. Planning closure.
   • Update this plan with changed files and command results.
   • Update completion state only after the closure pass.

Final Completion Gates

• All touched source has live source pointers.
• No runtime/core API changes unless a focused failing test forces them.
• Focused harvested tests pass from /Users/zbeyens/git/slate-v2.
• bun check passes from /Users/zbeyens/git/slate-v2.
• bun run completion-check passes from /Users/zbeyens/git/plate-2.

Current state is done because the approved cleanup was implemented under an execution lane after the Ralplan pass, with no claim-changing issue surface and fresh .tmp/slate-v2 verification.