ContextQMD
Back to Plate

Goal

docs/plans/2026-04-15-slate-v2-perf-architecture-research.md

53.0.67.8 KB
Original Source

Goal

Run a full research-wiki pass on the editor-architecture candidate lane for the specific question:

  • can Slate v2 honestly claim same-or-better perf architecture than the serious candidates on theory alone?
  • what does React 19.2 actually buy us?
  • if we want the field-best version of the decoration stack, what reshape is still justified?

Scope

  • direct compare corpora:
    • local Slate v2
    • ProseMirror
    • Lexical
    • VS Code
    • React 19.2 official docs
  • contextual compiled corpora from the existing landscape:
    • Tiptap
    • Premirror / Pretext
    • edix / use-editable / rich-textarea
    • TanStack DB
    • EditContext

Loaded Skills

  • research-wiki
  • major-task
  • goal workflow
  • learnings-researcher
  • repo-research-analyst

Phases

  • Read research layer entrypoints and full-pipeline workflow
  • Scope direct and contextual corpora
  • Read strongest direct evidence and classify each corpus
  • Update compiled source/system/decision pages
  • Update research index and log
  • Summarize the honest verdict and reshape pressure

Per-Corpus Evidence Ledger

Direct corpora

Slate v2 local proof substrate

  • compiled pages inspected:
    • docs/research/sources/editor-architecture/slate-v2-local-proof-substrate.md
    • docs/research/systems/slate-v2-overlay-architecture.md
    • docs/research/decisions/slate-v2-overlay-superiority-vs-legacy-and-field.md
  • raw paths inspected:
    • .tmp/slate-v2/packages/slate-react/src
    • .tmp/slate-v2/packages/slate-react/test
    • .tmp/slate-v2/docs/walkthroughs/09-performance.md
  • direct raw files read:
    • .tmp/slate-v2/packages/slate-react/src/projection-store.ts
    • .tmp/slate-v2/packages/slate-react/src/hooks/use-slate-selector.tsx
    • .tmp/slate-v2/packages/slate-react/src/hooks/use-slate-annotations.tsx
    • .tmp/slate-v2/packages/slate-react/src/hooks/use-slate-widgets.tsx
    • .tmp/slate-v2/docs/walkthroughs/09-performance.md
  • official source entrypoints checked:
    • local repo source
  • strongest evidence found:
    • runtime-id keyed local subscription is real
    • useSyncExternalStore-backed selectors and stores are real
    • current projection recompute still starts from source(Editor.getSnapshot(editor))
  • disposition:
    • evidenced
  • next action:
    • compile the local limit honestly: same class on UI subscription, not yet clearly field-best on core invalidation

ProseMirror

  • compiled pages inspected:
    • docs/research/entities/prosemirror.md
    • docs/research/sources/editor-architecture/prosemirror-mapped-overlays-and-bookmarks.md
    • docs/research/systems/editor-architecture-landscape.md
  • raw paths inspected:
    • ../prosemirror
    • ../prosemirror/view/src
    • ../prosemirror/state/src
  • direct raw files read:
    • ../prosemirror/README.md
    • ../prosemirror/view/src/decoration.ts
    • ../prosemirror/view/src/viewdesc.ts
    • ../prosemirror/state/src/selection.ts
  • official source entrypoints checked:
    • ../prosemirror/README.md
    • https://prosemirror.net
  • strongest evidence found:
    • child-scoped mapped overlay propagation
    • durable selection bookmarks
    • incremental view/update discipline
  • disposition:
    • evidenced
  • next action:
    • update compiled source/system pages with the stronger perf-architecture read

Lexical

  • compiled pages inspected:
    • docs/research/entities/lexical.md
    • docs/research/sources/editor-architecture/lexical-mark-store-and-decorator-split.md
    • docs/research/systems/editor-architecture-landscape.md
  • raw paths inspected:
    • ../lexical
    • ../lexical/packages/lexical/src
    • ../lexical/packages/lexical-react/src
    • ../lexical/packages/lexical-mark/src
    • ../lexical/packages/lexical-playground/src
  • direct raw files read:
    • ../lexical/README.md
    • ../lexical/packages/lexical/src/LexicalUpdates.ts
    • ../lexical/packages/lexical-react/src/useLexicalSubscription.tsx
    • ../lexical/packages/lexical-mark/src/MarkNode.ts
    • ../lexical/packages/lexical-playground/src/commenting/index.ts
    • ../lexical/packages/lexical-playground/src/plugins/CommentPlugin/index.tsx
  • official source entrypoints checked:
    • ../lexical/README.md
    • https://lexical.dev/docs/intro
  • strongest evidence found:
    • dirty leaf/element transform heuristic
    • explicit React subscription helper
    • mark ids, comment store, and decorator UI are split
  • disposition:
    • evidenced
  • next action:
    • update compiled source/system pages with the stronger runtime verdict

VS Code

  • compiled pages inspected:
    • docs/research/entities/vscode.md
    • docs/research/sources/editor-architecture/service-channels-and-live-stores.md
    • docs/research/systems/editor-architecture-landscape.md
  • raw paths inspected:
    • ../vscode
    • ../vscode/src/vscode-dts
    • ../vscode/src/vs/editor
  • direct raw files read:
    • ../vscode/README.md
    • ../vscode/src/vscode-dts/vscode.d.ts
    • ../vscode/src/vs/editor/common/services/markerDecorationsService.ts
    • ../vscode/src/vs/editor/browser/widget/codeEditor/codeEditorWidget.ts
    • ../vscode/src/vs/editor/common/viewModel/viewModelImpl.ts
  • official source entrypoints checked:
    • ../vscode/README.md
    • https://code.visualstudio.com
  • strongest evidence found:
    • typed decoration handles
    • independent comment controller surface
    • view-model separation from widget/editor shell
  • disposition:
    • evidenced
  • next action:
    • promote VS Code from “typed channels” only to a real perf-architecture reference

React 19.2

  • compiled pages inspected:
    • none; current lane has no dedicated React 19.2 compiled page
  • raw paths inspected:
    • official docs only
  • direct raw files read:
    • https://react.dev/reference/react/useTransition
    • https://react.dev/reference/react/useDeferredValue
    • https://react.dev/reference/react/useSyncExternalStore
    • https://react.dev/reference/react/Activity
  • official source entrypoints checked:
    • https://react.dev
  • strongest evidence found:
    • external store subscriptions are a first-class React primitive
    • transitions are non-blocking and interruptible
    • hidden Activity preserves state/DOM for later restore
  • disposition:
    • compile gap
  • next action:
    • create a dedicated compiled React 19.2 source page for this lane

Context corpora

Existing context-only candidates

  • compiled pages inspected:
    • docs/research/systems/editor-architecture-landscape.md
    • docs/analysis/editor-architecture-candidates.md
  • strongest evidence found:
    • Tiptap, Premirror / Pretext, TanStack DB, EditContext, and the lightweight surfaces still matter as context, but they do not materially change the direct perf-architecture verdict against the named serious compare set
  • disposition:
    • synthesis gap
  • next action:
    • keep them in the landscape/system page as context, not fake direct proof

Progress

2026-04-15

  • read docs/research/README.md, index.md, log.md, and docs/research/commands/full-pipeline.md
  • inspected the current compiled editor-architecture lane
  • verified direct candidate repos and revisions:
    • ProseMirror c7f2f1d7bde70728dfedaa68ca8f5fc3dffa17cc
    • Lexical d52f66e250e031a6c6fd8836d160373b0df557c7
    • VS Code bd5165ae94b03646adfba2fd80095d7d26c226b1
  • updated the compiled research layer with:
    • a dedicated React 19.2 source page
    • stronger ProseMirror / Lexical / VS Code source summaries
    • a sharper landscape/system verdict
    • an explicit decision rejecting blanket theory-only superiority claims while naming the best next reshape
  • updated docs/research/index.md, docs/research/log.md, and the editor architecture source README

Errors

  • there is still no normalized ../raw/<corpus> family for these editor architecture candidates; the lane remains partially backed by official local clones under ../