☰ 🦠 Datalayer Cells

A spreadsheet drawn on canvas and exposed as a React component, over a headless model that also runs in Node and in workers. Cells is built on Reactor: every extension point (formula functions, cell types, commands, importers and exporters, data sources, the collaboration provider) is a Reactor contribution point, and the built-in features are Reactor plugins contributing through the same points.

Work in progress.

Formulas: 117 functions, each checked against what LibreOffice computes; the list is in docs/functions.md.

Charts: column, bar, line, area, pie, scatter and combo charts anchored on a sheet, fed by ranges that follow recalculation, carried through XLSX, and drawn with Apache ECharts from @datalayer/cells/charts, a lazy plugin the grid loads only when a chart is drawn. See them live on the charts example: twelve months of figures with a column, a line, a doughnut and a combo chart beside them, in light and dark.

Packages

npm install @datalayer/cells @primer/react @primer/octicons-react @datalayer/primer-addons

@primer/react, @primer/octicons-react and @datalayer/primer-addons are required peers: the grid's context menu, the chrome and the named themes are Primer. loro-crdt (./collab), @datalayer/lexical-loro (its transport), echarts and echarts-for-react (./charts) are optional peers, installed by the hosts that use those entries.

Cells is built on Reactor: every extension point is a datalayer.cells.* contribution point, and the built-in features are plugins contributing through the same points. docs/plugins.md walks through a third-party plugin that reaches every one of them, running on the plugin example.

Documentation

• API: every public export of every entry point, linked to its source.
• Keyboard: moving, selecting, editing, and the commands' keybindings.
• Functions: the built-in formula functions.
• Plugins: extending Cells through its Reactor contribution points.
• Code cells: Python in a cell, run on a Jupyter kernel, its result in the grid and its outputs beside it; the security notes.
• File format: CellsWorkbook JSON, its validation, and its version.
• Docusaurus: live sheets in a documentation site.

Python

datalayer_cells reads and writes the same workbooks, with the same validation, for agents and scripts, and adds the datalayer cells command (also datalayer-cells on its own):

from datalayer_cells import Workbook

wb = Workbook.load("budget.cells.json")
wb.read("Summary!A1:C4")            # rows of values
wb.write("B2", [[12, "=B1*2"]])     # formulas are kept as typed
wb.save("budget.cells.json")

datalayer cells show budget.cells.json
datalayer cells get budget.cells.json 'Summary!A1:C4'
datalayer cells set budget.cells.json B2 12 '=B1*2'
datalayer cells export budget.cells.json --format csv

Examples

make start          # or make examples: http://127.0.0.1:3077
# The dev server loads loro-crdt's WebAssembly through vite-plugin-wasm (a
# devDependency) and a modern build target; a production build needs neither.

A page with a menu of examples, from a few values to collaboration, each with its source. What is not built yet says so, and names the milestone that brings it.

Published

The examples are online at cells.datalayer.tech. To publish a new version of them:

make build-examples   # builds the site into examples/dist
make deploy-examples  # builds, replaces the datalayer-cells S3 bucket, and refreshes the CDN

deploy-examples needs the AWS CLI with access to the datalayer-cells bucket and the CloudFront distribution EJMQSIG0ZTQVR, from the datalayer conda environment.

Development

@datalayer/reactor resolves as a package: from npm, or from the workspace in the Datalayer monorepo (install at its root).

npm install
npm run lint        # Prettier and the licence header on every file
npm run typecheck   # everything, tests and examples included
npm test            # unit tests (vitest, two workers), the React tests under React 19 and 18.3
npm run build       # lib/
npm run bench       # Node benches into bench/results/
python -m pytest    # the Python package (pip install -e ".[test]" first)
npm run size        # bundle sizes against their budgets

Browser benches and end-to-end tests run on the system Chrome, headless:

env -u WAYLAND_DISPLAY -u DISPLAY node scripts/bench.mjs --browser

License

MIT, see LICENSE.