Release 2024-03 follow-up #1309

release date: mid-june

wanted:

• Jupyterlab-4.2.1+ (seems it will be on time nearly on time)
• a free-threading [experimental] WinPython:
  • goal is keep developers moving forward also on windows,
  • usability hope is 25% slower only single-thread, 3x faster on a 4 cpu
• simplify, clean-up, and debug WinPython internals
• numba-0.60.0
• preparing for "the great Scientific Python Apocalypse" (next cycle):
  • very late & problematic eco-system: numpy-2.0.1 (see Ecosystem compatibility with numpy 2.0 numpy/numpy#26191)
  • very late: pandas-3.0.1 (pyarrow as fast internal machinery)
  • packages that will still be not ready for Numpy-2.0: GDAL, h5py, imageio, Keras, opencv, torch
• DELAYED:
  • python-3.13.0b3 (b2 only), about 7% faster single-thread, JIT & free-threding remain experimental
  • full cleanups to welcome numpy-2

focus:

• simplification of build and maintenance:
  • cleanup in the pre-pip aera code, now outdated
  • faster building: reduce package list, final size + pip progressing in speed New resolver takes 1-2 hours to install a large requirements file pypa/pip#12314
  • remove qtconsole icon
  • re-add powershell icon
  • less code to carry-on: drop no more used Qt code (since the drop of Qt ControlPanel), cx_freeze outdated helpers, ...
  • Python needs for a build is down to WinPythondot + flit + packaging (but requires WinPython-3.8+)
• follow packaging summit, consider hash checking
• free-threading test:
  • scientific builds tests there https://github.com/Quansight-Labs/free-threaded-compatibility
  • known-failure: IDLE
• update examples on import _xxsubinterpreters as _interpreters, and free_threading
• llm: AI agents and the 7b models, whispers made easy coming to town ?
• fighting the fat: polars for example is growing in size by 50% per year (now twice bigger than pandas-2)

exploration:

• jupyterlite more
• llm and generative AI: whisper.cpp
• replacing torch per jax[cpu] would shrink WinPython by 100Mo (or not increase per 200 More) ?

Direction:

• speed-debt reduction:
  • latest python 3.13 facts, 3.14 seems the place where jit/free-threads/multi-interpreters will be available in full glory ...
  • easier multi-thread python 3.13 'free-threading' progress
  • faster data analysis: numpy-2, pandas-3, scikit-learn integrating polars
• llvm local small equivalent to ggerganov/whisper.cpp
• ubiquity:
  • in browser (and WebApp): Jupyterlite prototype (https://stonebig.github.io/my_demo_4/lab/index.html),
  • WASI: a local jupyterlite ?
  • in Excel: asks Operational Research packages
  • in Windows Store: a requirement file