jupyterlab-cell-enhancements

Enhancements for JupyterLab 4 notebook cells:

• Cell titles — give any cell an editable title shown above the input. Titles are stored in the cell's metadata (cell_title) so they travel with the notebook. Hover a cell and click "Add a cell title…", or click an existing title to rename it. Press Enter to save, Esc to cancel.

• Focus mode — a corner-brackets button in the cell toolbar expands a single cell to fill the notebook, hiding everything else. Toggle the button again or press Esc to exit. Keyboard shortcut: Ctrl/Cmd + Shift + Enter while the notebook is focused.

• Floating notes — attach a note to a whole cell, or to a specific span of text. Notes float above everything (including the sidebars), follow their anchor as you scroll, hide once it leaves the viewport, and are draggable — dragging sets an offset that stays glued to the anchor.

  Notes render Markdown and LaTeX through JupyterLab's own renderer, so **bold**, lists, code, links and $x^2$ math all work.

• Side-by-side input/output — put a code cell's output beside its input instead of below it, per cell, saved in the notebook.

Adding a note

Anchored notes highlight their text and draw a leader line to the card.

Code anchors track your edits: they're held in a CodeMirror StateField that maps every range through each document change, so a highlight stays on its text as you type around it. Across reloads, anchors are re-verified against the quoted text and re-found if they've moved.

Markdown and output anchors are matched by quoted text instead, which is weaker — and note that outputs are regenerated on every run, so re-running a cell will usually orphan its output notes. Anything that can't be re-anchored is kept and flagged outdated (amber, with the original quote struck through) rather than silently pointing at the wrong place.

Resolving

Click ✓ on a note to resolve (archive) it. Resolved notes are hidden; enable Show resolved notes to bring them back, each with a ↺ to reopen.

Identity and colour

Notes are attributed using JupyterLab's own user identity, so they work without setup. Each author gets a colour generated from their name in OKLCH — any hue on the circle, with lightness and chroma constrained so the avatar text always keeps sufficient contrast. The same name yields the same colour for everyone opening the notebook.

Double-click a note's name to change it (you'll be offered the chance to update existing notes), or its avatar to pick any colour. The Notes toolbar button and the command palette expose the same options.

Resizing

Notes carry a native resize grip at their bottom-right corner — drag it to size the card in both directions. Text wraps as you narrow it (the note reflows and scrolls rather than clipping), and the note stays pinned to its anchor as it grows. The size is saved per note, alongside its position.

Storage

Everything lives in cell metadata under cell_comments — text, author, anchor, position, size, and resolved state — so notes travel with the notebook and are saved automatically shortly after any change. Notebooks written by earlier versions (cell_comment) are read transparently and migrated on first write.

Saving is routed through a single shared coordinator so that features writing the same notebook never issue overlapping saves — which, on a slow or network-backed filesystem, could otherwise race into a spurious "file changed on disk" conflict prompt.

Side-by-side input and output

Split a code cell so its output sits beside its input. Use the split button in the cell toolbar, right-click the cell, or the command palette.

Drag the divider between the panes to resize them; double-click it to go back to the default, and it's focusable with ←/→ (hold Shift for larger steps).

A few deliberate behaviours:

• The split only engages once the cell has output, so an un-run cell keeps the full width and then splits by itself when it executes.
• When the notebook gets too narrow for two useful panes, they wrap back to stacked rather than overflowing.
• Wide output (data frames, long lines) scrolls inside its own pane instead of forcing the pane past its share of the width.

State is stored per cell in metadata, so it travels with the notebook:

"metadata": { "cell_split": true, "cell_split_ratio": 0.45 }

Cells with no cell_split of their own follow the Side-by-side input/output setting, and metadata is written only when a cell differs from that default — so toggling cells doesn't sprinkle a redundant key across the notebook.

Cell appearance

Under the Notes toolbar button → Cell Appearance:

Colour options left blank fall through to JupyterLab's own styling rather than overriding it with a default, so a fresh install looks untouched.

Requirements

• JupyterLab >= 4.0.0

Install

pip install jupyterlab-cell-enhancements

Uninstall

pip uninstall jupyterlab-cell-enhancements

Development install

# Clone, then from the repo root:
pip install -e .
jupyter labextension develop . --overwrite
jlpm build

Rebuild after source changes with jlpm build, or run jlpm watch in one terminal and jupyter lab in another.

Building on a network / mapped drive (important)

This project's canonical location is on a mapped network drive (Z:), which Node resolves to a UNC path (\\server\share\...). webpack's resolver cannot handle UNC absolute paths, so jlpm build:prod / python -m build must be run from a copy of the repo on a local disk (e.g. C:\...). Editing and version control can stay on Z:; only the build/package step needs local disk. Copy the resulting dist/*.whl (and jupyterlab_cell_enhancements/labextension/) back if you want them alongside the source.

license-webpack-plugin patch

webpack >= 5.107 changed the ProvideSharedModule identifier separator from = to |, which crashes license-webpack-plugin@4.0.2 during the production build. A Yarn patch in .yarn/patches/ fixes this (wired via the resolutions field in package.json); it is applied automatically on jlpm install.