release date: 2019-09

Expected:

• Jupyterlab-1.1.1,
• dashboarding:
  • Anaconda Panel,
  • Quantstack Voila, (in 64 bit only)
  • not sure for Plotly Dash (but AJ Pryor is a fan),
• deep learning:
  • WinML / ONNX, that is in Windows10-1809 32/64bit,
  • PyTorch-1.2.0 with improved ONNX support
  • Tensorflow2 (if on time, rc announced around August 20th).
• vscode integration :
  • vscode launcher that should work for system vscode or "portable vscode"
  • a procedure to setup a "portable vscode inside WinPython"
  • an experimental "cod" version in Winpython-3.7, with "portable vscode" included.

Focus of the release:

• vscode integration effort:

  • make vscode works and portable : integrate a "VS Code" launcher #770
  • add a vscode launcher
  • do a notebook to precise how to install / connect,
  • may do a vscode branch for users, would replace 32 bit that starts to fall appart

• julia (re-)integration effort:

  • rework installation procedure in julia notebook,(but not portable at the moment):
    • critical is to put in env.bat "set PYTHON=%winpydir%\python.exe"
    • install via "import julia;julia.install()"
    • then via %julia magic, install Julia Optimization packages: https://github.com/JuliaOpt
    • simpler then previous hacks, but not portable apparently, packages goes in users home,
  • a second wave of effort would be needed to try reach portability.

• Spyder-4(beta4+) for PySide2 (the end game)

• tune Jupyterlab stack,

• simplify further the build process, to be able to move to the cloud:

  • simplify python binary directory-based installation process in make.py
  • split in steps to remove need for an initial Python environnment.
  • simplify winpython wheel build process (since a long time, python version independant),

non-Focus of the release:

• pandas:-1.0 (3 month lag recently), so patching pandas-0.25.1 for python-3.8
• Tensorflow-2.0: no roadmap dates (smells november)

Windows 32 bit E.O.L:

• after Deep Learning packages, Jupyterlab is now starting to fail to build on 32 bit. This is:

  • the natural bloatness of Programs (you always need a bigger machine),
  • this won't be fixed / taken care of per maintainers,
  • droping Jupyterlab "rebuild" from 32bit builds: only default extensions in it

• this will be the last release with some 32bit distro, if ever.

• hopes for coming 2020:
  - Spyder-5 compatible with PySide2 Q1 2020
  - nodejs no longer necessary to update Jupyterlab-2 Q2 2020,
  - Tensorflow-2 and Pytorch get smaller, using windows ONNX infrastructure Q3 2020,
  - Python-3.9 hugely simplify Asyncio and multi-core Q4 2020