It would allow the configuation of asymmetric VLANs as described in Annex F of IEEE Std 802.1Q-2014.
Now you got me intrigued. I can't find document containing the Annex you mentioned on line, so can you briefly describe how it's supposed to work?
A simple example to demonstrate the concept. (This is an example only. It's
not what I'm trying to achieve, so please, no suggestions of other ways in which it might be done.)
Start with a 5-port router: ether1 is the uplink; ether2–ether5 are connected to bridge1, which has a PVID of 1. I want to share a single subnet across 2 VLANs (for no particular reason). Ports ether2 and ether3 are both on VLAN 2; ports ether4 and ether5 are both on VLAN 3.
/interface bridge port
set 0 pvid=2
set 1 pvid=2
set 2 pvid=3
set 3 pvid=3
/interface bridge vlan
add bridge=bridge vlan-id=1 untagged=ether2,ether3,ether4,ether5
add bridge=bridge vlan-id=2 untagged=bridge
add bridge=bridge vlan-id=3 untagged=bridge
/interface bridge
set 0 vlan-filtering=yes
Ports 2 and 3 are isolated from ports 4 and 5, but they all share the subnet configured on the bridge (e.g. 192.168.88.0/24).
This works, but at the expense of flooding. For example, the bridge (on VLAN 1) can't learn the MAC addresses of the hosts connected to ports 2 and 3 (VLAN 2) and to ports 4 and 5 (VLAN 3), because they are on different VLANs, so it floods traffic to all 4 ports. Shared VLAN learning would fix this.