@doom369 sounds like something we can change .. Do you want to do a PR ?

I think it should be message.payload().retainedDuplicate()

@normanmaurer I was thinking about:

message.payload().duplicate() -> message.payload()

In a typical use case you would do:

        ByteBuf buffer = ctx.alloc().buffer(body.length());
        buffer.writeCharSequence(body, StandardCharsets.UTF_8);

        MqttPublishMessage message = MqttMessageBuilders.publish()
                .payload(buffer)
                .qos(MqttQoS.AT_LEAST_ONCE)
                .messageId(packetId)
                .build();
       ctx.write(message);

Do you think this duplicate() has some purpose?

I would allocate an ioBuffer and perform the copy into it, which will still happen later in the transport layers...
I am not a big fan of copies eh...we could have a builder which allow to pass some arbitrary buffer too, and will allow users to allocate it themselves, so we can enable some zero-copy semantic, if is a direct one.

we could have a builder which allow to pass some arbitrary buffer too, and will allow users to allocate it themselves

That's what the current MqttMessageBuilders has already. The question - why the encoder does duplicate(), doesn't seem necessary for me.

Also, from the netty-end-user point of view - buffers are not necessary, I just want to send a string message back into mqtt channel (most common use case). So for me, it would be more convenient to send String instead of ByteBuf. But that's another issue.