Re: Init system for non-Linux ports
On 12/02/14 10:45, Robert Millan wrote:
> I'm a bit afraid that even if sysvinit itself stays mostly fine, the scripts written for
> it could turn into a bunch of bitrot.
There are a few reasons to keep sysvinit scripts maintained for jessie:
1. for smoother upgrades from wheezy
2. in order to backport jessie packages to wheezy
3. for non-Linux (or non-systemd) ports
So ports are not the only reason. And yet all of the above points still
apply to ports; we'd have to support sysvinit even if we went with
something else.
I don't think it matters much what we choose, but seems we'd want to
make use of legacy sysvinit compatibility - write very few scripts in
specific formats (e.g. OpenRC runscripts / Upstart jobs). We probably
have right up until freeze to make a preference, and perhaps by then
there will be more than one fully-working init system.
GNU/Hurd porters already said they aim to maintain sysvinit scripts:
https://lists.debian.org/debian-hurd/2014/01/msg00051.html
And there are plenty of GNU/Linux users who will want to run systems
without systemd. (individuals, derivatives, quite possibly Google).
Thinking ahead, package maintainers won't have such need to support
sysvinit for jessie+1 so that's when we'll really have problems. Having
something like OpenRC or Upstart might allow to add/override broken init
scripts with native/declarative ones. Perhaps by then we'll have new
ways to convert init scripts or generate them from metalanguage; or
built-in support for reading each others' formats.
> And AFAICS we may be able to use Upstart some time in the future but this doesn't seem
> possible right now.
We could initially ask the maintainer to apply Dmitri's patches and try
to keep it building on kfreebsd. And just keep it around for people to
play around with and possibly get it fully working. IIRC it runs, but
will need early boot scripts re-writing, equivalent to /etc/rcS.d/
> What is the current status of OpenRC? Is it usable?
Pretty good as I recall; it was very easy to test using jails.
Regards,
--
Steven Chamberlain
steven@pyro.eu.org
Reply to: