Re: So many packages. So few that I need started at every boot.
On Sat, Dec 27, 2003 at 05:36:40PM -0800, Vincent Bernat wrote:
> OoO Vers la fin de l'après-midi du samedi 27 décembre 2003, vers
> 16:47, Colin Watson <firstname.lastname@example.org> disait:
> > Having said that, there seems to be a patch in the BTS at the moment
> > which I ought to look at.
> Since the original one was quite old, I have sent two new.
That's what I was referring to.
> The original one get rid of /etc/ssh/sshd_not_to_be_run.
As I say, that should happen by splitting the package, post-sarge.
> If the user does not want the server, the symlinks are removed, it
> can still launch it via /etc/init.d.
That doesn't really work. Your original patch meant that the user's
choice was *only* remembered in the debconf cache, which might be
removed at any time. At the moment /etc/ssh/sshd_not_to_be_run is simply
a record of the answer in a validly persistent location.
> My first patch looks at the name of the script. If it is invoked as S*
> or K*, /etc/ssh/sshd_not_to_be_run is respected, otherwise the action
> is taken whatever this file exists or not. It is almost the same as
> the original approach.
> The second patch adds some targets : force-start, force-stop and
> force-reload. With start, stop and reload, the script behaves the same
> way as the original script, so we don't need to modify other scripts
> like the postinst one (and maybe some others). The force-* versions
> will do the appropriate actions even if /etc/ssh/sshd_not_to_be_run
> Just handling symlinks in /etc/rc.d is not sufficient enough since
> many postinst scripts should then be modified to avoid the server to
> start against the user will. In contrast, adding "force-*" targets
> would force the user to modify its own scripts (for example, when he
> could have written a script that launches Apache+MySQL).
I prefer the second option. ssh's postinst can't invoke the init script
via the S* and K* names because the user might wish to change the
ordering of their init scripts. Also, the user might not be using the S*
and K* scheme at all, but might be using file-rc.
In addition, the second option nicely avoids being invasive, since it
doesn't change the meaning of existing targets.
Colin Watson [email@example.com]