On Tuesday 19 December 2006 23:07, Sven Luther wrote: > Err, can you tell me the risk of regression. Please speak the issue > with someone else, like Colin and Joey, or Bastian, and weigth the > actual risk.
OK. So let me tell you why this patch will BREAK some installations.
finish-install.d/90console has:
<snip>
inst_pid=$(pidof debian-installer | sed "s/ /\n/g" | sort -n | head -n 1)
rawconsole=$(readlink /proc/${inst_pid}/fd/0)
</snip>
Why does it take the 'head'?
Reason is that during installs using network console there are _two_
debian-installer processes: the original one (possibly started over
sercon), and the second one started over ssh.
The second one will be the one finish-install is run from. This instance
will have TERM_TYPE=network, and not TERM_TYPE=serial like the original
one. So, in that case, even though the install was started over serial
console, the installed system would not be set up to use serial console
(while this works correctly with the current logic).
/me is so glad he is conservative with patches like this and takes the
time to think them through and really test them instead of applying them
blindly.
Hopefully this will teach you to not dismiss my concerns out of hand in
the future, though I won't hold my breath.
Cheers,
FJP
pgpSQb4YKxLcB.pgp
Description: PGP signature
