On Sat, Jun 19, 2010 at 09:10:55AM +0200, Petter Reinholdtsen wrote:
> Currently, virtual facilities are defined either in /etc/insserv.conf
> or in /etc/insserv.conf.d/, and in Debian we expect packages providing
> virtual facilities like $named to add a file in /etc/insserv.conf.d/.
>
> Such package would have a init.d script header like this:
>
> ## BEGIN INIT INFO
> # Provides: pdns
> # ...
> ## END INIT INFO
>
> and a insserv.conf.d entry like this:
>
> $named pdns
>
> Instead, what about teaching insserv to understand notation like this,
> and skip the insserv specific file in /etc/:
>
> ## BEGIN INIT INFO
> # Provides: pdns $named
> # ...
> ## END INIT INFO
>
> For this to work, insserv will have to accept and handle several
> scripts providing the same virtual facility, allowing setup like this:
>
> ## BEGIN INIT INFO
> # Provides: pdns $named
> # ...
> ## END INIT INFO
>
> ## BEGIN INIT INFO
> # Provides: bind $named
> # ...
> ## END INIT INFO
>
> ## BEGIN INIT INFO
> # Provides: dnsmasq $named
> # ...
> ## END INIT INFO
>
> Werner, any opinion on this?
Hmmm ... the problem with this approach is e.g. if network script
is not installed no $network will be provided even if other
scripts refer to $network. Also it will move the responsibility
from the insserv maintainer forward to the individual package
maintainers including foreign package maintainers which may
(in my experience;) increase the numbers of potential problems.
> I suspect the LSB should be updated to document that this is an
> allowed setup, if we go this way.
For optional contributions for a system facility we could do
somewthing like
# Provides: dnsmasq
# X-Contribute: $named
that could be a way to do this, and it would allow to use
nick names for a Provide.
Werner
--
"Having a smoking section in a restaurant is like having
a peeing section in a swimming pool." -- Edward Burr
