Re: Reformulation of criticisms about console-setup switch (was: Re: Review of console-setup wrt D-I [very long])

[Anton Zinoviev <anton@lml.bas.bg>]

On Mon, Jun 29, 2009 at 07:32:39PM +0200, Frans Pop wrote:
> >
> > Should I add one additional question or preseeding is enough?
> 
> No, you need to implement automatic detection for this case.
> 
> Please see kbd-chooser.c. It's not all that hard to find the code, nor 
> should it be all that hard to implement it in C. See also the 90console 
> script in finish-install.

I didn't known that such detection is possible.  No C is required in 
order to detect the presence of a physical keyboard - the following 
shell code does the trick:

tests=......
if grep "$tests" /proc/bus/input/devices /proc/bus/usb/devices >/dev/null; then
   # there is a keyboard
else
   # there is no keyboard
fi

The detection of the serial console requires an ioctl but I think 
console-setup-udeb doesn't need to know whether the console is serial or 
not.

I have some questions:

1. Can console-setup-udeb expect that /proc is mounted? Is there some 
dependency about this?

2. Can console-setup-udeb expect that /proc/bus/usb is mounted? Is there 
some dependency about this?

3. If the answer of 1 or 2 is no, then what can be done?

4. Are there architectures where kernel <=2.4 is used?  The detection 
doesn't work for such kernels.

5. What about the following scenario: the system is installed without a 
keyboard, console-setup-udeb leaves the keyboard unconfigured and then 
the user attaches a keyboard.

Anton Zinoviev