Why:

To build TensorFlow from source, we need to be able to run Bazel from Fromager, in our AIPCC builder, it's the only way to build the wheel.

What:

We need to find a way to execute Bazel in our ecosystem among the dependencies to build TensorFlow. Both Bazel and the dependencies will be tracked via another EPIC.

How:

A prototype was proposed here: https://gitlab.com/redhat/rhel-ai/wheels/builder/-/merge_requests/1711

Note: there is a remaining issue where we'd want to drop the "dev0" from the wheel name. Something was tried in the MR but it didn't work yet.

Scope of this EPIC:

This will track the work done in a Tensorflow package plugin and the addition of this package into the right collections for the right variants. Bazel and the dependencies will be figured out in different EPICs.