Due to code such as https://github.com/google/protobuf/blob/e3c807d4e752b477b707f5d021264faf9b127304/src/google/protobuf/map.h#L868, it's not in general safe to use a protobuf in one C++ module that is built with NDEBUG with code in another that is built without NDEBUG. E.g. passing an in-memory proto between the modules via a function call. This results in hard to discover corruptions of the mapped field hash table.

While this is acceptable in a monorepo, in general OSS where C++ dependencies are sourced from prebuilt libraries, this seems like a bad situation. Can we remove this in OSS protobuf?