Return of #2985? Segv in protobuf but with pytorch

[aifartist]

Problems in the coexistence of TF 1.13.1 with the ProtoBuf 3.12.3 #2985
I was replacing my Torch 1.13.1 with Torch 2.0.0 locally built and get a SEGV at
google::protobuf::internal::ReflectionOps::FindInitializationErrors(google::protobuf::Message const&, std::string const&, std::vector<std::string, std::allocatorstd::string >*) ()
just like the old bug. Looks like a buffer overrun. The SEGV is a deference of $rax=0x656c676f6f67200a
at call *0x88(%rax)
This is obviously a char string which has overwritten the memory rax was fetched from .
The char string is "\n google" and interestingly the message being passed in is:
\n google/protobuf/descriptor.proto\022\017google.protobuf...
which is a 4372 byte long string.
The context I saw this was using AUTOMATIC1111 which involves Stable Diffusion.
I was trying the latest Torch.