[shu-chen-dot-ca]

Fixes #1510.

Adds an enabled/disabled state to SQL notebook cells. A disabled cell keeps its source, configuration and last output, but nothing in the notebook sends it to the database until it is enabled again. That covers a direct run, Run All and automatic prerequisite resolution.

• types/notebook.ts
  • Optional isDisabled on SQL cells and in the file format. Absent means enabled, so legacy cells and files load as enabled.
  • RunAllResult gains blocked and blockedCells. Each blocked cell lists the disabled cells it waits on.
• utils/notebookRunAll.ts
  • getExecutableCells leaves disabled cells out; getDisabledCells lists them, so Run All counts them as skipped.
  • findDisabledPrerequisites(sql, cells, prepare) returns the disabled cells that stop a query: referenced directly, or through referenced cells that would have to run first because they have no current result. It resolves notebook parameters (so @source = {{cell_1}} counts) and handles reference cycles.
  • addBlocked records a blocked cell; formatRunAllSummary reports blocked cells.
• utils/notebookVariables.ts: a disabled cell is never scheduled as a dependency, and its cached output is never substituted into a consumer's CTE (it stays an unresolved reference).
• NotebookView.tsx
  • The run path returns at once for a disabled cell, whatever triggered it: the header button, the editor shortcut or a dependency.
  • It checks again after each lazy dependency run and on both sides of the query guard, which can wait on the dangerous-query confirmation. So a cell disabled while it waits is never sent: it stops with no error and Run All counts it as skipped.
  • A consumer of a disabled prerequisite is marked "Blocked: prerequisite {{cell_N}} is disabled" before any guard or execute_query call.
  • Run All re-reads each cell right before starting it. Blocked cells are counted separately: they are neither failures nor a reason to stop on error.
  • The in-flight result update only patches runtime fields, so a reply to a query sent before disabling cannot clear isDisabled. Disabling does not claim the query was cancelled.
  • No query plan is prepared for a disabled cell.
• NotebookCellHeader.tsx / NotebookCellWrapper.tsx
  • A power toggle with aria-pressed, a translated aria-label/title ("Disable cell" / "Enable cell"), visible focus and theme tokens.
  • A "Disabled" badge. While disabled, the run button is disabled and labelled "This cell is disabled. Enable it to run it."
  • The query plan button stays in place but is disabled, so the header buttons don't shift under the pointer.
• SqlCell.tsx
  • Keeps the plan section closed on disabled cells, since an open plan sends EXPLAIN (or EXPLAIN ANALYZE) as soon as it mounts.
  • Shows a status line on disabled cells. When an earlier result or error is still shown, it adds that the output is from an earlier run and is not current.
• RunAllSummary.tsx: "N blocked", and for each blocked cell a link to it and to the disabled cell it waits on.
• notebookFile.ts / notebookUndo.ts
  • isDisabled is written only when true, and read only from a literal true on SQL cells.
  • It is part of the undo signature, with "Disabled cell N" / "Enabled cell N" in the edit history.
  • Re-enabling is only a document change and executes nothing.
• New strings in all 13 locales.

Out of scope, as the issue says: cancellation, scheduling and read-only enforcement. The SQL source export still includes disabled cells' SQL, since the issue only covers execution. Happy to leave them out or comment them out there if preferred.

Tests:

• tests/utils/notebookRunAll.test.ts
  • Disabled and empty cells left out of Run All; legacy cells count as enabled.
  • Direct and transitive disabled prerequisites, with and without cached output; a chain stops at an enabled cell that already has a result.
  • Several disabled prerequisites reported once each, sorted; reference cycles terminate.
  • Markdown, missing and enabled cells ignored; parameter-made references counted.
  • Blocked counting and the summary text.
• tests/utils/notebookFile.test.ts: round trip, enabled cells not written, legacy file loads as enabled, malformed values ("true", 1, null, {}) ignored, never set on Markdown.
• tests/utils/notebookUndo.test.ts: signature change, describe disable/enable, undo/redo of a disable.
• tests/utils/notebookVariables.test.ts: a disabled cell's cached output is not substituted, and a disabled cell is not scheduled as a dependency.
• tests/components/notebook/NotebookView.test.tsx
  • Direct run: a disabled cell sends nothing and keeps its output.
  • Consumers of a disabled cell, with and without cached output, are blocked with no guard or query. So are the end of a dependency chain and a consumer whose reference comes from a parameter.
  • Disabled while waiting: a consumer disabled while its lazy dependency runs, and a cell disabled while the guard waits for confirmation, send nothing.
  • Run All with disabled, blocked, plain and chained cells sends only the plain query; the summary shows 1 succeeded, 1 skipped, 2 blocked with links to the disabled cell.
  • Run All edge cases:
    • Stop on error does not stop on a blocked cell.
    • Parallel cells are handled.
    • A cell disabled while earlier cells or the parallel group run is skipped.
    • A sequential consumer disabled during its own lazy dependency re-run is skipped, not failed or sent.
  • Re-enabling runs nothing, and undo/redo restore the flag.
  • A reply arriving after disabling keeps the cell disabled.
  • No plan is prepared for a disabled cell.
• tests/components/notebook/NotebookCellWrapper.test.tsx, SqlCell.test.tsx and the new RunAllSummary.test.tsx: toggle name, state and handler; disabled run button with its reason; badge; disabled plan button with no EXPLAIN request; no toggle on Markdown; the notices; the summary links.

Validation:

• pnpm typecheck && pnpm lint && pnpm format:check && pnpm web:boundaries && pnpm web:packaging && pnpm web:docs && pnpm web:manual-parity && pnpm test:contrast && pnpm build && pnpm test run --maxWorkers=4: all passed (565 files, 9125 tests). No Rust changes.
• GitNexus upstream impact:
  • serializeNotebook HIGH: it is the save path of every notebook (create, rename, import); the change there is one optional field written only when true.
  • NotebookCellHeader, documentSignature, deserializeNotebook and runCellInner LOW.
  • getExecutableCells, findUnresolvedDependencies and describeChange have no resolved callers in the index; text search shows they are used only in NotebookView and NotebookHistoryPanel.
• detect-changes before each commit: notebook symbols only. Risk is high because NotebookView is in many flows; the export handlers it lists only moved down the file.
• Manual check in Tabularis Web (nightly binary serving this branch's build) with a SQLite connection and a two-cell notebook where cell 2 reads {{cell_1}}:
  • A notebook saved before the change opened with both cells enabled.
  • After disabling cell 1, Run All reported "1 skipped, 1 blocked" with links, and cell 2 showed the blocked message.
  • The state survived a reload.
  • The server's audit log showed no execute_query after disabling, including after re-enabling.

Before	After

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

I found no defects in the new commit 355fab19, which re-checks a cell's disabled state before its query is sent. I read its diff and the surrounding code in NotebookView.tsx. I did not run the tests.

• The disabled-state re-checks sit in the right places: after dependency runs and on both sides of the query guard's confirmation.
• A cell that is disabled mid-run is recorded as skipped in run-all.
• A rejected sequential run is now recorded as a failure instead of aborting the whole run.
• The other files in the incremental range come from the base branch, not from this PR's changes.

Files Reviewed (2 files)

• packages/web-ui/src/components/notebook/NotebookView.tsx
• packages/web-ui/tests/components/notebook/NotebookView.test.tsx

Previous Review Summary (commit a3b1b9a)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit a3b1b9a)

Status: No Issues Found | Recommendation: Merge

I found no defects in the changed code. I read the diff and the saved patch output for the source files. I did not run the tests, and I did not read the changes under packages/web-ui/src/i18n/locales/ or packages/web-ui/src/types/notebook.ts in detail.

Files Reviewed (31 files)

• packages/web-ui/src/components/notebook/NotebookCellHeader.tsx
• packages/web-ui/src/components/notebook/NotebookCellWrapper.tsx
• packages/web-ui/src/components/notebook/NotebookView.tsx
• packages/web-ui/src/components/notebook/RunAllSummary.tsx
• packages/web-ui/src/components/notebook/SqlCell.tsx
• packages/web-ui/src/i18n/locales/*.json (13 locale files)
• packages/web-ui/src/types/notebook.ts
• packages/web-ui/src/utils/notebookFile.ts
• packages/web-ui/src/utils/notebookRunAll.ts
• packages/web-ui/src/utils/notebookUndo.ts
• packages/web-ui/src/utils/notebookVariables.ts
• packages/web-ui/tests/** (8 test files)

---

_{Reviewed by claude-sonnet-5.5 · Input: 248 · Output: 1.1K · Cached: 194.9K}

_{Review guidance: REVIEW.md from base branch main}

[shu-chen-dot-ca]

Rebased onto main (5725dad) to resolve the conflict. The only conflict was in utils/notebookUndo.ts, which git treats as binary because of the raw NUL separator in describeChange. It is now a text merge that keeps both sides: main's kpiConfig in the undo signature and the kpiCell change kind, plus this PR's isDisabled / disableCell / enableCell. The NUL bytes are unchanged. Everything else applied cleanly. Full frontend checks pass again (578 files, 9492 tests).