Justin Chu

Is there a way to either disable pyi gen or fall back to the old method when building with an older protobuf version? Whichever is easier.

Generated ```python from google.protobuf.internal import containers as _containers from google.protobuf.internal import enum_type_wrapper as _enum_type_wrapper from google.protobuf import descriptor as _descriptor from google.protobuf import message as _message from typing import ClassVar...

vs before ```python # @generated by protoc-gen-mypy.py. Do not edit! # mypy: disable-error-code=override from typing import ( Iterable, List, Optional as OptionalType, Tuple, cast, ) from google.protobuf.message import ( #...

> We can also specify/install the correct version on the Linux pipeline -- practically I don't think it really makes a difference. I'm afraid we can't. `pip install` still needs...

It seems possible to check the version here: https://cmake.org/cmake/help/latest/module/FindProtobuf.html using `Protobuf_VERSION`

@kumar-ish Did you get a chance to update this?

The issue predates protobuf-24.4, so I am not sure?

I think it is due to another package linking to onnx at the same time: https://groups.google.com/g/protobuf/c/RZRCnOywdSk/m/HsOaheyFCQAJ

PyTorch builds with ONNX. I wonder why it would fail with torch_geometric? Does setting `BUILD_SHARED_LIBS` help?

For float8 usage, we may be better of using https://github.com/jax-ml/ml_dtypes?