☠️ EXPERIMENTAL ☠️ jupyterlab-offline

(Re)build JupyterLab and Extensions Faster With More Energy, but without an npm registry

☠️ NO REALLY ☠️ THIS WILL MESS UP YOUR JUPYTERLAB ☠️

Features

• store yarn.lock and dependencies
• build static assets from offline lockfiles and dependencies
• scaffold command for conda recipe
• better tests

Use cases

• npm is down
• you can't run your own npm repository

Alternative Approaches

• provide a local npm repository

  Could be couchapp or an alternative, lightweight, scope-enabled, npm-compatible registry

  • pro
    • doesn't need git (needed to merge lockfiles)
    • probably more robust to minor semver changes
  • meh
    • still can't account for "missed-chance" duplication
    • probably the same on disk
  • con
    • another port opened, probably...
    • more node

Failure modes

Missed-chance duplication

If...

• package A requires version 1.1 and less than 1.3 or greater of B
  • and bundles B@1.1
• package C requires version 1.2 or greater of B
  • and bundles B@1.3

Then the one valid resolution (B@1.2) would not be in the offline cache.

• Impact: build fails, requires manual intervention (e.g. conda remove last-thing)
• Likelihood: ??? (needs data)

Command-Line Usage

jupyter laboffline build

Attempt to build the current JupyterLab application using only offline assets. These are usually put in $APP_DIR/offline by installing packages.

jupyter laboffline install

Attempt to perform a clean install of the extension at that version, capturing its yarn.lock and npm dependencies.

jupyter laboffline install @jupyter-widgets/jupyterlab-manager 0.33.2

Known dependencies should already have been installed through your package manager.

jupyterlab-offline scaffold conda <@npm/name of extension> <simple semver>

Print out a scaffold conda-build meta.yaml for an extension

TODO: explore pip option

git clone https://github.com/conda-forge/staged-recipes
cd staged-recipes
mkdir -p recipes/bqplot
jupyterlab-offline scaffold conda bqplot 0.3.6 > recipes/bqplot/meta.yaml
# Add already-packged upstreams with good semantic versioning
# Make a PR!
# Add it to your release automation
# TODO: tie into bot

API

TBD

Installation

TBD

Dependencies

jupyterlab 0.31.12

jupyterlab-offline reuses much of the jupyterlab infrastructure, and its bundled package manager (yarn ne jlpm)

git

git is needed to merge all of the yarn lockfiles together

nodejs

nodejs is required to run jupyterlab build

Other JavaScript runtimes, e.g. PyMiniRacer, might be possible

Development

conda env update
source activate jupyterlab-offline-dev
anaconda-project run test # tbd