python-3.11.7: build a new 'package python'

[thomas]

Some more relevant version numbers (apps and libs):

meson: 1.3.0
snakemake: 7.32.4
spyder: 5.5.0

numpy: 1.25.2
pandas: 2.1.4
matplotlib: 3.8.2
scipy: 1.11.4
scikit-image: 0.22.0
jupyterlab: 4.0.9
torch: 2.1.2+cu118
tensorflow: 2.14.1

Tensorflow 2.14.1 was build against /pkg/cuda-11.8.0-0. It requires now bazel 6.1 (or higher), which in turn is only happy with a more recent JRE, so java21 was choosen.

Tensorflow's CPU architecture is no longer 'nehalem', but 'broadwell'. Most of our cluster machines should be able to handle this.

A more general change is the deprecation of source builds via:

'python setup.py build; python setup.py install'

Now replaced with the recommended way:

'pip install .', or
'pip install git+https://github.com/xyz/abc.git@v1.0'

As for the QT5 bindings, here nothing was changed. For QT5 I expect no dramatic changes anymore. At some point QT6 will join in or replace QT5.

Other changes and removals:

'helen' does not build anymore, removed
Older 'gmpy' was replaced with 'gmpy2'

Note on the strategy to avoid downgrades:

If a package requires 'pkgXY' it installs the
most recent versison. A follow up might want a slightly
older version, so before 'pkgXY' is installed, just
install this older requirement with oldie==1.2.3
(see urllib3 as example)

[thomas]

A final note about CUDA. To make full use of the GPU aware python packages it is required to use a nvidia driver that supports CUDA 11.8. Current default kernel is mariux-5.15.131-457 with Nvidia driver version: 510.108.03 (release date: Nov/22/2022) supporting only CUDA 11.6 and below. Right now only mariux-6.1.53-458 uses the Nvidia driver version: 535.104.05 (release date: Aug./29/2023) supporting CUDA up to 12.2.