In 2.x local replication endpoints only seem to work like in 1.x but they do something quite confusing: instead of replicating a clustered database they replicate a node local database on the node where the replication job happens to run. The node were replication runs is essentially random as far as users are concerned. So "local" replications in 2.x almost never do what users expect.
Over the years this confusion led to numerous question and issues raised in the IRC channel. So the proposal is to remove support for "local" replications from versions 3.x and up, and only support "remote" replication endpoints. That will reduce the support and issue load as well as replicator code size. What does everyone think?