[nd 3/5] Frontend transaction sync by manzt · Pull Request #8844 · marimo-team/marimo

manzt · 2026-03-24T01:37:10Z

Prev #8842, #8843. This closes the frontend ↔ session loop in three steps:

Frontend → session. A reducer middleware intercepts cell actions (create, delete, move, rename, set-config) as the user performs them, batches them into transaction ops, debounces (400ms), and flushes to POST /api/document/transaction. A suppress guard prevents echo loops when applying server-originated ops.

Session → frontend. applyTransactionOps maps each op to the corresponding cell reducer action.

Queued-cell staleness. When the kernel queues a cell for execution, the frontend now snapshots lastCodeRun. Previously only the user-initiated "Run" button did this, so kernel-initiated runs (e.g. via code_mode) left cells visually stale even after successful execution.

vercel · 2026-03-24T01:37:15Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Mar 25, 2026 9:04pm

mscolnick · 2026-03-24T02:53:14Z

frontend/src/core/cells/apply-transaction.ts

+  NotificationMessageData<"notebook-document-transaction">["transaction"];
+type TransactionOp = Transaction["ops"][number];
+
+interface CellActions {


might be able to Pick<CellActions, ...> here.

I think i ran into some circular reference (because the type is inferred). We could probably have an explicit type somewhere that we reuse as a type-only import.

ah yes it is inferred. maybe we can make middleware additive

maybe creates other problems/smells, but something like this

manzt · 2026-03-24T02:55:57Z

frontend/src/core/cells/document-transaction-middleware.ts

+/**
+ * Actions the middleware intercepts. Payload types are hardcoded from
+ * the notebook reducer in cells.ts. A follow-up can infer these from
+ * the reducer directly once it's extracted to a named variable.
+ */
+type DocumentAction =
+  | { type: "createNewCell"; payload: { cellId: CellId } }
+  | { type: "deleteCell"; payload: { cellId: CellId } }
+  | { type: "moveCell"; payload: { cellId: CellId } }
+  | { type: "sendToTop"; payload: { cellId: CellId } }
+  | { type: "sendToBottom"; payload: { cellId: CellId } }
+  | { type: "dropCellOverCell"; payload: unknown }
+  | { type: "dropCellOverColumn"; payload: unknown }
+  | { type: "updateCellCode"; payload: { cellId: CellId; code: string } }
+  | { type: "updateCellName"; payload: { cellId: CellId; name: string } };


@mscolnick similar here... our middleware interface isn't strictly typed on the reducer. I hard-coded types to try to reduce the diff.

mscolnick · 2026-03-24T02:57:01Z

frontend/src/core/cells/apply-transaction.ts

+        let cellId: CellId | "__end__" = "__end__";
+        let before = false;
+        if (op.after) {
+          cellId = op.after as CellId;


i think these as could be cleaned up now from your other PRs

The document transaction middleware and apply-transaction logic were spread across two files with duplicated type definitions and several gaps in the op mapping. This PR consolidates them into `document-ops.ts` as two pure functions — `toDocumentOps` (action → ops) and `fromDocumentOps` (ops → actions) — with thin impure wrappers for debouncing and dispatch. The middleware was missing the `updateCellConfig` handler entirely (config changes were silently dropped), sent an empty config object for `createNewCell` instead of reading the actual `CellConfig` from post-reducer state, and produced `after: null` when a cell moved to the first position. Position derivation now uses an `anchorOf` helper that reads from `newState`, eliminating type casts and correctly handling `__end__`, column objects, and left/right moves without special-casing. Column structure actions (`addColumnBreakpoint`, `dropOverNewColumn`, etc.) previously only emitted `reorder-cells`, losing the per-cell column index. They now diff column assignments between prev and new state and emit `set-config` ops for cells whose column changed, keeping the Python `NotebookDocument` in sync. The middleware switch is exhaustive with `assertNever` — adding a new reducer action without handling it here is a compile error. A new round-trip test suite proves the two directions are correct inverses by performing actions on a primary notebook, applying the emitted ops to a replica, and asserting both converge to identical document state.

Prev #8842, #8843, #8844. This PR closes the kernel → session loop. **Transaction emission.** `_apply_ops` previously assembled three separate legacy notifications (`UpdateCellCodesNotification` grouped by stale status, plus `UpdateCellIdsNotification`). Now it builds typed `Op`s via `_plan_to_document_ops()` and broadcasts a **single** `NotebookDocumentTransactionNotification`. **Document context.** The execution endpoint snapshots `session.document.cells` into `ExecuteScratchpadCommand.notebook_cells`. The kernel handler sets a `ContextVar` from it so `AsyncCodeModeContext` can read cell state from the document.`_CellsView` now delegates entirely to the document and returns `NotebookCell` directly, removing the `NotebookCellData` wrapper and the module-level `_cell_names` dict.

This PR introduces the core data model, with wiring to session (#8843), frontend (#8844), and kernel (#8845). `NotebookDocument` is an ordered list of `NotebookCell` entries that applies `Transaction`s atomically. A transaction is a tuple of typed ops (`CreateCell`, `DeleteCell`, `MoveCell`, `ReorderCells`, `SetCode`, `SetName`, `SetConfig`) with a `source` tag identifying the writer (e.g. `"frontend"`, `"kernel"`, `"file-watch"`). The document validates ops for conflicts (e.g. delete + update on the same cell) before applying. Version is `None` on creation and stamped by `apply()`. ```python doc = NotebookDocument([NotebookCell(id="a", code="x = 1", ...)]) tx = Transaction(ops=(SetCode(cell_id="a", code="x = 2"),), source="kernel") applied = doc.apply(tx) # stamps version → Transaction(version=1) ``` --------- Co-authored-by: Shahmir Varqha <Sham9871@gmail.com> Co-authored-by: Myles Scolnick <myles@marimo.io> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

manzt requested a review from Light2Dark as a code owner March 24, 2026 01:37

manzt changed the title ~~[nd 3/5] Frontend transaction sync~~ [nd 3/5] Frontend transaction sync Mar 24, 2026

vercel bot deployed to Preview March 24, 2026 01:38 View deployment

manzt added the enhancement New feature or request label Mar 24, 2026

This was referenced Mar 24, 2026

[nd 4/5] Emit transactions from code_mode #8845

Merged

Emit document transactions from --watch file reload #8846

Merged

manzt force-pushed the manzt/nd-session branch from 510769c to 8f1aab9 Compare March 24, 2026 01:45

manzt force-pushed the manzt/nd-frontend branch from ee2fc9f to 552dbb6 Compare March 24, 2026 01:45

vercel bot deployed to Preview March 24, 2026 01:49 View deployment

manzt force-pushed the manzt/nd-frontend branch from 552dbb6 to 2bb42ff Compare March 24, 2026 02:08

github-actions bot added the bash-focus Area to focus on during release bug bash label Mar 24, 2026

vercel bot deployed to Preview March 24, 2026 02:11 View deployment

manzt force-pushed the manzt/nd-session branch from 8f1aab9 to 527fa05 Compare March 24, 2026 02:16

manzt force-pushed the manzt/nd-frontend branch from 2bb42ff to 468dd42 Compare March 24, 2026 02:16

vercel bot deployed to Preview March 24, 2026 02:20 View deployment

mscolnick reviewed Mar 24, 2026

View reviewed changes

manzt commented Mar 24, 2026

View reviewed changes

mscolnick reviewed Mar 24, 2026

View reviewed changes

manzt mentioned this pull request Mar 24, 2026

[nd 6/5] Back NotebookDocument cell text with Loro CRDT #8849

Draft

manzt force-pushed the manzt/nd-session branch from 527fa05 to 22a83e0 Compare March 24, 2026 04:20

manzt force-pushed the manzt/nd-frontend branch from 468dd42 to a766c32 Compare March 24, 2026 04:20

vercel bot deployed to Preview March 24, 2026 04:23 View deployment

vercel bot deployed to Preview March 24, 2026 18:53 View deployment

manzt force-pushed the manzt/nd-frontend branch from ef3c69f to 6626bda Compare March 24, 2026 18:53

vercel bot deployed to Preview March 24, 2026 18:58 View deployment

manzt force-pushed the manzt/nd-frontend branch from 6626bda to 8f2f8bb Compare March 24, 2026 19:41

manzt force-pushed the manzt/nd-session branch from daa7926 to 1be085c Compare March 24, 2026 19:41

vercel bot deployed to Preview March 24, 2026 19:42 View deployment

vercel bot deployed to Preview March 25, 2026 19:58 View deployment

manzt added 2 commits March 25, 2026 16:58

Remove suppressDepth in favor of non-middleware actions

5a47cc6

manzt force-pushed the manzt/nd-session branch from 1be085c to d0c7932 Compare March 25, 2026 20:59

manzt force-pushed the manzt/nd-frontend branch from 671e569 to 8002a4a Compare March 25, 2026 20:59

manzt requested a review from dmadisetti as a code owner March 25, 2026 20:59

manzt merged commit 5a3191b into manzt/nd-session Mar 25, 2026
9 of 11 checks passed

manzt deleted the manzt/nd-frontend branch March 25, 2026 21:00

vercel bot deployed to Preview March 25, 2026 21:04 View deployment

manzt mentioned this pull request Mar 25, 2026

Add new NotebookDocument model #8842

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[nd 3/5] Frontend transaction sync#8844

[nd 3/5] Frontend transaction sync#8844
manzt merged 3 commits intomanzt/nd-sessionfrom
manzt/nd-frontend

manzt commented Mar 24, 2026 •

edited

Loading

Uh oh!

vercel bot commented Mar 24, 2026 •

edited

Loading

Uh oh!

mscolnick Mar 24, 2026

Uh oh!

manzt Mar 24, 2026

Uh oh!

mscolnick Mar 24, 2026

Uh oh!

mscolnick Mar 24, 2026

Uh oh!

manzt Mar 24, 2026

Uh oh!

mscolnick Mar 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

manzt commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vercel bot commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mscolnick Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

manzt Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

mscolnick Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

mscolnick Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

manzt Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

mscolnick Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

manzt commented Mar 24, 2026 •

edited

Loading

vercel bot commented Mar 24, 2026 •

edited

Loading