On Wed, Sep 01, 2004 at 09:27:12AM -0500, Stephen R Marenka wrote:
> On Wed, Sep 01, 2004 at 01:36:24AM +0200, Wouter Verhelst wrote:
> > [...serial check thinks real consoles are serial as well...]
> >
> > Hm. I just referenced the code which was already in rootskel, supposing
> > it was working OK, but that doesn't seem to be the case. Checking on my
> > VME box, however, the /proc/self/fd/* files also link to /dev/console,
> > so there is no way this problem could be fixed by slightly modifying the
> > check.
>
> The difference is that the original code was called by debian-installer
> and the new code is called in sbin/init, which has a different
> context.
Ah, that explains. Darn.
I'll try and see whether I can find a difference from that context,
then, but I'm afraid not.
> Can the following sbin/init code be moved to
> lib/debian-installer.d/S30term?
>
> if [ $TERM_TYPE = "serial" ]; then
> for i in 2 3 4 5; do
> rm -f /dev/vc/$i
> done
> fi
No, I tried that; by that time, it's too late.
> This seems to solve the detection problem for me.
>
> Do you really need to delete the extra vc devices?
Yes; because busybox init currently checks whether the devices are there
(whether the files exist), and if they are, tries to open them.
Repeatedly. Issuing an error every time it fails (which, of course,
happens every time). Removing the files once those error messages start
to appear only changes the error that appears; and with those messages
going over the screen, the usefulness of d-i is quite, uh, low.
This obviously really is a bug in busybox init, hence my previous
suggestion to try and fix that one.
--
EARTH
smog | bricks
AIR -- mud -- FIRE
soda water | tequila
WATER
-- with thanks to fortune
Attachment:
signature.asc
Description: Digital signature