What do the non-JS libraries encode? It seems strange to encode a one-of with the default since you can still decode the default. I've never tested this.

We should encode default value to know what one-of field was set. Official python and js protobuf library encode default value of one-of too:

message Id {
  oneof type {
    uint64 numId = 1;
    double doubleId = 2;
    string strId = 3;
  }
}

js

const id = new google_pb.Id();
id.setNumid(0);
const s = id.serializeBinary();
console.log(Buffer.from(s))

<Buffer 08 00>

python

import p_pb2

id = p_pb2.Id()
id.numId = 0
data = id.SerializeToString()
print(data)

b'\x08\x00'

Of course it will be nice to add check user set only one one-of field but logic will be more complex. For now, let's assume the user is doing the right thing.

Ah, okay. I think I misunderstood. I thought you were saying that when the oneof isn't set it should still encode the oneof. That's not what the C++ implementation does. If you don't set any fields, it should be 0 bytes.

It sounds like this is a different issue from that, if I understand correctly.

Yes, the issue is not about whole oneof field but is about oneof subfield with default value. We should encode it anyway.

Is it any change this PR will be merged? Behavior is still incorrect.