Back then we added VLAN_ARRAY not to save memory (which sounds ridiculous today, but also was questionable in the 32-bit world), but to avoid cache misses in case if VLAN ids are selected with large gaps in their numbers.

Haven't yet looked in detail, but have a quick question. Did you try VLAN_ARRAY option? Many years ago we left it in assumption that in some cases a user may want a full array of VLANs. I wonder if it is still functioning, since it was never compiled by default or advertised.

In D48902#1115700, @ngie wrote:

Sidenote: you can require the program in the Kyuafile as well. That's generally the best way/least amount of code to ensure that the requirement is actually being met.

In D48864#1114767, @kib wrote:

What can be done with POLLINIGNEOF is to hide it from userspace. Brace the bit definition with #ifdef _KERNEL, so no new usage could appear.

Thanks for working on this! Once you come to an agreement on how to properly handle that, can you please also look at other tests that use scapy, outside of pf? There are quite a lot in netinet, netinet6 and net tests.

In D48864#1114692, @kib wrote:

This is what called ABI stability, which is provided by any OS that tries to be useful. And this is the big reason why an idea needs very good reason before being put into the ABI visible space.

In D48864#1114651, @kib wrote:

How do you know that no software uses the flag? Exp run only shows that the current ports do not.
What troubles does the flag cause? Referenced bug only talked about incorrect use of the flag to implement some other features. As far as that attempts were eliminated, what are the kernel problems?

If we move the fib to pcb would we be able to preserve this API? Of course it would start fail on sockets where FIB doesn't make any sense. But would we be able to preserve for meaningful sockets?

Maybe commit changes to ip_output.c and ip_fastfwd.c separately? Cause they are a generic change, not something related to ICMP.

I will let Kirk to decide what solution is better. I just had fixed the build.