@CGGXANNX Both good, and they land differently: you are right about the switch chips, and the manual disagrees with you about DHCP.
On hardware offload you are correct, and it was worse than a missing feature. From the Switch Chip Features page:
"For 88E6393X, 88E6191X, 88E6190, MT7621, MT7531, EN7523, RTL8367 switch chips, the VLAN table is configured through bridge HW vlan-filtering using the /interface/bridge menu. For other switches with VLAN table support, see VLAN Table and configure the VLAN table using the /interface/ethernet/switch/vlan menu."
RB3011 and hAP ac²/ac³ are in that "other" group, so what we generate takes the VLAN traffic off the switch chip and into the CPU. On an RB3011 that is the difference between wire speed and software forwarding, and we said nothing about it.
The generator now says so. Pick one of those boards, turn VLANs on, and it tells you the VLAN table on this chip lives in /interface ethernet switch, that bridge filtering works but disables hardware offload, and that the switch-chip route is the one for full throughput. Ten of the models we carry are marked: hAP, hAP ac, hAP ac lite, hAP ac², hAP ac³, hEX PoE, hEX PoE lite, PowerBox Pro and OmniTIK 5 PoE ac.
Two things I found while doing it that are worth knowing.
First, I could not derive this from our own data. Our catalogue stores the chip from the vendor spec table, and for the hAP ac² that says IPQ-4018, which is the SoC. The switch is an Atheros8327. So the list is curated by hand from the manual's device table rather than computed, which means it is only as complete as that table.
Second, and it cuts against generating the switch-menu configuration everywhere: on IPQ-PPE boards such as the hAP ax³ and cAP ax, and on the QCA8386 in the hAP be³ Media, the capability table shows no VLAN table in the switch chip at all. On those, VLANs are CPU-bound no matter which menu you use. So this is not a simple two-way split, and generating /interface ethernet switch vlan for everything would be wrong.
As for producing that configuration rather than only warning about it: you predicted it adds complexity to the code base and you are right, it is a second VLAN emission model living beside the current one. It is worth doing for that family, but I would rather ship an honest warning today than a half-tested second path, so that is the order.
On the DHCP raw rule, the manual answers it directly. The line immediately above that rule on the Building Advanced Firewall page:
"accept DHCP discovery - most of the DHCP packets are not seen by an IP firewall, but some of them are, so make sure that they are accepted."
So your premise is mostly right and the conclusion does not follow: most DHCP bypasses the IP firewall, but not all of it, which is exactly why MikroTik put the rule in. We copied it from there verbatim.
That said, I did go and trace our own raw chain rather than just quote the manual at you, and I could not construct a case where removing it breaks DHCP on the LAN. A discover has source 0.0.0.0 and destination 255.255.255.255, and none of our bogon drops match that on a LAN interface: the destination lists are WAN-scoped since the multicast fix, and 0.0.0.0 is not in any source list. So in our specific rule set it may well be doing nothing. We are keeping it, because MikroTik say some DHCP packets do reach the firewall and one accept rule is a cheap way to not find out which ones the hard way, but you were not wrong that it earns its place less than it looks.