release 2020-03 follow-up

release date: Q3 2020 (end-july, early august)

Expected:

unexpected addition:

no more expected ready for this release:

Focus of the release:

internals simplification:
- remove old and hacky legacy internals by switching to msvc_runtime,
- Tornado hack is removed but is in cgohlke wheels
adapt to coming pip new strict dependancy resolver:
- build something that is "pip check" happy,
- ping upstream to avoid recompiling too much package
python-vscode issues

Exploratory work:

looking non-MKL to benefit of AMD Zen2+: BLIS / OpenBLAS / improved MKL, ... pushing latest numpy is best action.
Spyder-5.1 compatible with PySide2-5.15 this fall ... pushing PyQt-5.15 / PySide2-5.15 on conda is best action
include virtualenv to give a way out of pip strict resolver impact
SQLite progress/pushing on math (iee option added in future SQLite-3.33.0, hopefully a hint of progress)
check Tornado progress on SelectorEventLoop
Mxnet https://github.com/StacyYang/MXNet-Gluon-Style-Transfer

Hopes/Wishes/Forecast/Strategy for coming 2021-2022: (to look back in 6 month)

25% of DataScientist on PC laptop moves from Intel to AMD in 2021 ==> favorize latest numpy usage
Python-3.10 simplifies multi-thread (hpy) and async (TaskGroups): Q3 2021 ==> include Trio as a training set
Python GIL-less is Q3 2022 : python-3.11 or Python-4 ==> provide a "python-next-3.10"
Python-vscode becomes DataScientist super-friendly (invading azure/devops/o365) Q4 2020 => keep/refine VSCode branch
PyQt5 becomes less Popular than PySide2: Q2 2021 (currently we are 25/75)==> accelerate Spyder-5.1 arrival
still to find ways to builld on "conpty", "onnx", "winrt" (integrated ML inside Windows)
Python hits 33% on PYPL end 2020 (covid wave 1&2 boost) ==> keep Winpython packages set well fresh
Python hits 34% on PYPL end 2021 (azure/o365/devops boost) ==> Speed-up on Micro-Services/Pypa-Forge/Qt-6
Python hits 35% on PYPL end 2022 (wasm/GIL-less/pypy boost) ==> a pypy branch ?
end 2022 on PYPL, TypeScript overtakes PHP, Rust overtakes Go