If I am reading the situation correctly based on my admittedly incomplete information, TensorFlow is falling behind on build system updates. We have the Bzlmod support] “Modified by moderator” issue hanging without any concrete answers from active maintainers for over nine months at this point. Issues like outdated [protobuf]“Modified by moderator” and gRPC (link omitted doe to forum rules for new users) are holding back Bzlmod support and introducing more recent and human-friendly versions of Bazel.
This is critical for any client/dependent projects, such as Google’s own MediaPipe, who suffer the same dependency hell issues by being forced to stick to older versions of Bazel. In many places, we at my company have to do three/four-level-deep patches to Mediapipe(Tensorflow(gRPC/etc/(something even deeper))) in order to get things to just build for various edge platforms (e.g. on arm64 Macs, for example).
The only thing I can do from my side at the moment is add posts burried in TensorFlow issues about how to resolve this or that bug by patching.
Sorry for the long-winded introduction to the issue, I hope I was able to stress the importance of this matter.
The question:
what are the plans/timeline re:
- Protobuf upgrade?
- gRPC upgrade?
- Bazel upgrade?
- Bzlmod support?