On Sun, 2024-12-15 at 22:34 +0100, Torsten Keßler wrote:
Dear all,
over the last weeks, Arch Linux Package Maintainers have been working on the Python rebuild for the new minor release 3.13. All but one package, TensorFlow, where successfully updated to work with Python 3.13. TensorFlow proved to be difficult to update [0]. As a large software stack with different python modules and low level C/C++ interfaces, we can't progress without upstream. Unfortunately, a new TensorFlow release with Python 3.13 is only expected for early next year [1]. We have evaluated different options. TensorFlow and related packages could be dropped from the official repositories. But we think that this step would cause too much confusion for users. Instead, we add a warning printed during the installation of a TensorFlow package that encourages users to use TensorFlow via pip or via the official docker images:
WARNING: This version of the TensorFlow package is not compatible with Python 3.13
If you want to use TensorFlow, consider installing an older version of Python and use a virtual environment or use the official Docker image, https://www.tensorflow.org/install
Best, Torsten
[0] https://github.com/tensorflow/tensorflow/issues/78774#issuecomment-243956835... [1] https://github.com/tensorflow/tensorflow/issues/78774#issuecomment-249814853... Hi Filipe, Just to clarify, you mean the that this warning will be added to the 'tensorflow' package, which only includes the base libraries, but the 'python-tensorflow' package will still be dropped, right? That's not possible with tensorflow's build system. The C++ library and
Am 15.12.24 um 22:56 schrieb Filipe Laíns: the python modules are built in the same bazel call. Thus, it's not possible to only partially build or partially drop tensorflow packages. The warning was added to all packages of the tensorflow split package.
Cheers, Filipe Laíns
Cheers, Torsten