Cisco’s converged access and (in)directly connected access-points

In every document you can find about Cisco’s converged access you will read that with 3650 and 3850 switches access-points need to be directly connected. I understand the reasons why Cisco requires that, but the nerd in me wanted to know how that it works and if the switch can be tricked in the process. So what happens if you connect an access-point on another layer 2 switch and build a dot1Q trunk between that switch and the 3650/3850?

Nov 18 00:58:18.893: %CAPWAP-3-AP_PORT_CFG: AP connected port Gi1/0/24 is not an access port.

Busted! But what happens when we make it an access port? Great success! Sadly this only works for just the first access-point. So to summarize the requirements if you really need/want to do this:
1. The access-point needs to be in the same VLAN as the wireless management interface of the 3650/3850 switch
2. The interface where the access-point is being located on/after needs to be in access mode
3. Only one access-point per interface.

Note: I tested this on an 3650 with IOS-XE 03.06.00E, but I do not believe that this is something what will change in newer software versions.