Skip to content

python: add version 3.10.7 #212

Merged
merged 1 commit into from
Sep 30, 2022
Merged

python: add version 3.10.7 #212

merged 1 commit into from
Sep 30, 2022

Conversation

thomas
Copy link
Contributor

@thomas thomas commented Sep 12, 2022

Compared to python-3.9.12-0 the script is restructured to
a larger extend:

  • packages are taken from a list, no more 'pip install ...'
    spaghetti.

  • package evolution is tracked with a small git repo inside
    the build dir. With 'git log --patch' downgrades are easily
    spotted.

Major changes:

  • Cuda version 11.5.2
  • Include (aka dev) files for dbus-python, pycairo, and PyGObject.
    Making it possible to build applications that not only import
    the packages.
  • Qt-bindings from Riverbank computing up to date
  • snakemake version 7.14.0
  • Tensorflow version 2.10.0
  • Napari added (a tool to work with Zeiss images)
  • Git-Cola added to the package, obsoleting the separate package

Removals:

  • JaxLib (requires too many old packages)
  • thonny, a tiny IDE (uses too much obscure packages)

Trivia:

  • The annoying 'pip version check' can be disabled by
    changing a single line.

What else:

  • some minor formatting errors remain.

Compared to python-3.9.12-0 the script is restructured to
a larger extend:

  - packages are taken from a list, no more 'pip install ...'
    spaghetti.

  - package evolution is tracked with a small git repo inside
    the build dir. With 'git log --patch' downgrades are easily
    spotted.

Major changes:

  - Cuda version 11.5.2
  - Include (aka dev) files for dbus-python, pycairo, and PyGObject.
    Making it possible to build applications that not only import
    the packages.
  - Qt-bindings from Riverbank computing up to date
  - snakemake version 7.14.0
  - Tensorflow version 2.10.0
  - Napari added (a tool to work with Zeiss images)
  - Git-Cola added to the package, obsoleting the separate package

Removals:
  - JaxLib (requires too many old packages)
  - thonny, a tiny IDE (uses too much obscure packages)

Trivia:

  - The annoying 'pip version check' can be disabled by
    changing a single line.

What else:

  - some minor formatting errors remain.
@donald
Copy link
Contributor

donald commented Sep 12, 2022

Is

PACKAGES=$(sed -e s/#.*$// <<- __PKGLIST__
        protobuf==3.19.4    # tensorflow likes to have an slightly older one [< 3.20], no conflict with this
        ipython[all]

        SciPy Matplotlib pandas SymPy nose statsmodels
        [...]
        __PKGLIST__
)

for PKG in $PACKAGES; do
  install $PKG
done

really better than

install protobuf==3.19.4    # tensorflow likes to have an slightly older one [< 3.20], no conflict with this
install ipython[all]
install SciPy Matplotlib pandas SymPy nose statsmodels
[...]

would be?

python-3.10.7-0.build.sh Show resolved Hide resolved
python-3.10.7-0.build.sh Show resolved Hide resolved
@niclas niclas linked an issue Sep 16, 2022 that may be closed by this pull request
@thomas
Copy link
Contributor Author

thomas commented Sep 30, 2022

Results for the 'CUDA-enhanced' packages included:

             works in general | works on A-100
tensorflow:        Yes             Yes
pytorch:           Yes             No [1]
pycuda:            Yes             Yes
numba [2]:         No              No

[1] pytorch comes with an embedded CUDA, and this version doesn't support recent GPUs.
[2] numba: the provided pypi-package fails completely and switches back to CPU mode.

In both cases own source builds might solve the issues present. OTOH: in case of numba it is also OK to have it as a 'NO-OP'-Module since the depending applications are used on non-GPU machines anyway (Also, IMHO numba isn't worth the hassle).


As for the launch-date. Next Tuesday would have been nice, but since we are enjoying ourselves on Wednesdays lunch-date, I'll opt for releasing it on Wednesday evening.

@thomas thomas merged commit 64a2877 into master Sep 30, 2022
Sign in to join this conversation on GitHub.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Python: Missing module *perf*
2 participants