ProseMirror transaction/view DOM runtime

Purpose

Compile the ProseMirror evidence that matters to Slate v2's transaction and DOM selection discipline.

Strongest Evidence

• editor state is updated by applying transactions.
• transactions subclass transforms and track document changes, selection changes, stored marks, scroll intent, and metadata.
• transaction selection is mapped through step mappings as document changes are accumulated.
• selections expose bookmarks that can map without the current document and resolve later.
• commands receive state, optional dispatch, and sometimes view.
• view/input owns DOM event handling, DOM observer flushing, composition flags, DOM selection import, and DOM selection export.
• decorations are a view data channel, not document content.

What To Steal

1. Transaction owns composite local mutation

Slate already has operations and transactions. ProseMirror reinforces that the transaction should own:

• document changes
• selection before/after
• mark state
• metadata
• scroll / UI event policy

For Slate v2, editor.update should create the transaction, and EditorCommit should be the durable local record.

2. Selection mapping discipline

ProseMirror maps selection through transaction steps. Slate v2 should map selection through operations and runtime ids instead of relying on stale public fields or full snapshots.

3. Durable bookmarks

SelectionBookmark is the clean durable anchor model:

• map without needing the current document
• resolve later against current document

Slate v2 should preserve this as the public durable selection/anchor story for history, collaboration, comments, and review systems.

4. One DOM bridge owner

ProseMirror's view code has explicit selectionFromDOM, selectionToDOM, DOM observer flushing, and composition handling. Slate v2 should keep the same discipline:

• DOM import has one owner
• DOM export/repair has one owner
• composition is an explicit mode
• app commands do not read DOM selection directly

5. Decorations as view data

ProseMirror decorations prove overlays should be mapped view data. Slate v2 should keep projection sources and dirty-source invalidation instead of reviving render-time decorate as the primary API.

What Not To Steal

• Do not copy integer document positions.
• Do not make schema-first content matching the Slate v2 core identity.
• Do not expose ProseMirror-style plugin complexity as the normal Plate/Slate extension API.
• Do not make React a wrapper around a ProseMirror-like view tree.

Take For Slate v2

The ProseMirror lesson is not "be ProseMirror".

It is:

transactions own state changes
selection maps through mutations
DOM selection import/export is centralized
view overlays are data

Slate v2 should implement those disciplines with paths, runtime ids, operations, commits, and a React-optimized renderer.