Re: [PATCH] wifi: move action code from per-type frame structs

[Johannes Berg <johannes@sipsolutions.net>]

On Wed, 2026-02-25 at 13:18 -0800, Jeff Johnson wrote:
> On 2/25/2026 9:08 AM, Johannes Berg wrote:
> > 
> > And a couple of other notes:
> > 
> > >  				struct{
> > > -					u8 action_code;
> > > +					u8 no_fixed_fields[0];
> > >  					u8 variable[];
> > >  				} __packed chan_switch;
> > 
> > This thing in a couple of places isn't ideal, but I think it doesn't
> > outweigh the benefits of this structural approach.
> 
> My fear is that "helpful" members of the community will refer to the
> following, believing this is an old style variable array definition:
> 
> https://www.kernel.org/doc/html/latest/process/deprecated.html#zero-length-and-one-element-arrays

Yeah ... We'll have helpful AIs to tell the "helpful" people off, right?
Right? ;-)

> So I personally feel that a /* no fixed fields */ comment would provide the
> same benefit without bringing unwanted noise (unless the code just doesn't
> build without the member being present).

Ah, but that's exactly the thing, it isn't valid to have structs
completely without members (there's one case without "u8 variable[];"),
nor structs with only a variable member. So indeed it doesn't build
without it.

So ... yeah, not ideal, but I think the structural benefit outweighs the
disadvantage here.

Unless folks disagree with that and think that the action_code really
_should_ be part of the per-action-frame sub-struct? Personally, I think
the code looks better this way, and it has always slightly bothered me,
but I could be convinced otherwise :)

johannes