Re: i386 and the COMPAT_50 option

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Mon, 26 Jan 2009 11:02:40 +0200
    From:        Antti Kantee <pooka%cs.hut.fi@localhost>
    Message-ID:  <20090126090240.GC28742%cs.hut.fi@localhost>

  | Always include non-modularized compat by default, but flip the defs and
  | provide NO_COMPAT for the people who must compile their own kernels and
  | just can't live with compat compiled in.

That would work, though positive options are generally easier to understand
and use than negative ones.

  | That way the people who want
  | to ultratweak can continue to do so, but we give a strong statement of
  | what is considered default

That's a reasonably strong point.

  | and don't have to maintain an evergrowing
  | list of compat options in hundreds of config files.

but you don't really win there, the long list doesn't really need to
be maintained anyway, COMPAT_14 implies COMPAT_!5 which implies
COMPAT_16 ... up to COMPAT_40 implies COMPAT_50.   The config files
only really need the oldest desired.   The rest are just there as more
accurate documentation/information of what options exist and their
status in the generated kernel.

So, if you flipped them, the likely consequence would be to replace
the (not all that long, but growing) sequence
        option  COMPAT_14
        option  COMPAT_15
 ...
        option  COMPAT_50
with
        #option NO_COMPAT_14
        #option NO_COMPAT_15
 ...

which isn't much of an advantage really (it will save a few bytes of
the compiled in kernel config in the kernel binary).

The big negative of this is that you'd be making it even harder than
it is now to maintain a single kernel config file that is able to build
lots of different netbsd versions, which is what I try do do, and mostly
works with the aid of a little sed magic to work around devices (etc) that
change their attachment logic from version to version...

kre