I would say that the changes fix a bug. According to the documentation, an event should have been emitted, but it was not.
If I remember correctly, an invite was necessary for my tests before issuing the cancel request, and then issuing the RPC was another way of triggering it that did not generate the event as per the reporter.
Probably other ways of generating a locally `CANCEL` did not produce the event, but I did not verify this before.