1 回答
It sounds like you've basically figured it out.
You're getting deep into APIs inside the protocol compiler which were not really designed for public consumption. It gets complicated because no one has written a helper layer to simplify things, because not many people use these features.
I am not sure what you mean about Message being a "God class". Message is merely the abstract interface for a protobuf instance. Descriptors describe types of protobuf instances. Message::getDescriptor() returns the type of a message, but beyond that there isn't much direct connection between these APIs...
Isn't there a way to include the source information without using a FileDescriptorSet?
The comments are intentionally stripped from the descriptors embedded into the generated code, so you need to run the parser separately, generate a descriptor set, and consume it dynamically.
Is it possible to "connect"/set the FileDescriptorSet with a concrete Message class/instance, since that would drastically simplify things?
Do you mean that you want Message::getDescriptor() to return a descriptor that includes the comment data from the source file? That would require that the comment data be embedded into generated code, which would be trivial for protoc to implement (it currently intentionally strips them out, so it would just have to not do that) but potentially bloat-y and dangerous (could reveal secrets for people shipping closed-source binaries built with protobufs).