Hi,
Theo de Raadt wrote:
Review the lines that dmassage has commented-out. You can fairly
safely remove unused drivers for network/scsi/audio controllers/USB
devices, but other drivers/pseudo-devices are more likely to give
problems. Trimming out devices (especially some scsi and nic drivers)
will trim out a lot, and if you then find you need to go further,
you'll just need to take it step by step with educated guesses.
dmassage is about 12 years old, it is useful in some cases but the
generated config cannot be used directly.
And remember that if you use it, you are running a non-GENERIC kernel.
It isn't that we don't like people running non-GENERIC kernels. The
isue is that people who run custom kernels are often the type who
don't switch back to GENERIC kernels before telling us of a problem
they have encountered, and they have waste our time enough in the
past. So it isn't that we hate non-GENERIC kernels, it is that we
hate people who treat us so poorly.
I know, with both NetBSD and OpenBSD I always keep the original
reference kernel installed, to boot when a problem is suspected. On
machines where it is not a problem, I just keep the GENERIC and am happy
with that. Sometimes a non-generic kernel is needed for certain hardware
to work, but this luckily is becoming less and less the case.
In this case, we were speaking about tweaking on some stranger hardware
and squeezing where resources are more limited.
I think dmassage being unmaintained for 12 years, and this issue just
coming up now, probably says a lot about that "type" of person. It's
a type of person who can't fix dmassage, and then, sends us a mail.
Sorry, but it's the truth.
If it is not broken don't fix it: fine, but it still can be made
user-friendly, e.g. by marking certain commented out devices as "to
review" or by not commenting them out at all if they are known or
something like that.
Or perhaps a comment line in the kernel config like "do not disable this
ever or if you use X", something which config fails to catch, that's
all. One would think that config should catch dependencies.
I wasn't saying there is a bug or something is seriously broken, just
improvable.
It means "dmassage" produces an output which is not readily usable, but
needs to be massaged. Perhabs it could be noted in the manpage, instead
of the cryptic "POD ERRORS" section.
Anyway, thanks, I did not want to stir up the pot. I'll keep building
kernels on a faster machine and disable options piece after piece. I
just finished setting up a more modern laptop yesterday with OpenBSD!
Riccardo
