Feature Request: C++ logging from non-TensorFlow frameworks

TensorBoard logging can be done from a variety of ML frameworks, including PyTorch with torch.utils.tensorboard. The FileWriter and protos exposed from TensorBoard provide an API contract that can be easily accessed in Python code to log everything from scalars and histograms to model graphs.

However, things get complicated with C++. Any proto specifications or code would need to be built as a static or dynamic library and referenced by frameworks at build as well as runtime.

A couple of questions:

1. How much interest is there in the community for TensorBoard logging from non-Python languages and non-TensorFlow libraries?
2. Should proto files exist in a separate TensorFlow/TensorBoard repository so they can be pinned, updated, etc. across logging implementations or just be copied as needed into other code bases?
3. Is there value in a shared C++ logging implementation and a precompiled library to match?

We are exploring C++ logging for PyTorch with libtorch and so wanted to get community involvement on this topic. Comments welcome!

cc @nfelt @manivaradarajan @GalOshri @natalialunova @lanpa @sanekmelnikov @caraya10 @jspisak

No files, tests, or entry points are identified. Start by reviewing the questions about proto ownership, cross-language logging, and a shared C++ implementation, then establish a concrete scope and acceptance criteria. Done would require an agreed direction and an implementation plan for C++ logging outside TensorFlow.