The 18/09/11, Duncan wrote:

> > I don't see any added benefit from using DBUS on my servers.

Insterstingly, Duncan just answered your question...

> Interesting question.  I hadn't seen the suggestion until this thread, 
> either, and it bothered me too.

From here:

> With a moment's thought, I decided I could probably return to a semi-
> static dev setup reasonably easily.  I'd potentially turn on the early-dev 
> option in the kernel that I still have off, ATM, which presumably would 
> mount a tmpfs on dev and populate it with the earliest devices.  After 
> that, if necessary, I'd copy the existing udev-created nodes out to a 
> persistent state dir, and copy them back in with a little init-time 
> script of my own.  As long as the device ordering remains stable, this 
> could include by-label, etc, symlinks, or I could simply kill the by-
> label, by-uid stuff in fstab, and go back to traditional devices there, 
> too.
> 
> Either that, or simply go back to a static /dev entirely.
> 
> People with dynamic ordered devices may have to devise their own scripts, 
> tho, or perhaps more likely, fork off udev from the pre-union state.

...to here.

> But it's also possible that's far enough in the future that we can't 
> really answer the question now, since technology will have changed enough 
> to make an answer now look senseless, then.  Consider trying to answer 
> the question in terms of the kernel devfs back before udev.  The tech 
> simply changed and those answers wouldn't really work, today.

Upstream changes the init process is done. So, you're free to either:

 stick to upstream (with best long term support);

or

 fork off upstream (requires knowledges, manpower and time);

or

 go back to 1960 with a full/partial static /dev (asking to manually
 maintain the crap).

See the benenfit, now?

-- 
Nicolas Sebrecht