sysvinit should depend on initscripts for a functional /lib/sysvinit/init

[Josh Triplett]

On Sun, Jul 17, 2016 at 04:37:10PM +0200, Michael Biebl wrote:
> Control: retitle -1 Drop transitional sysvinit package in stretch
> 
> Am 05.05.2015 um 17:57 schrieb Josh Triplett:
> > On Tue, May 05, 2015 at 02:08:56PM +0200, Michael Biebl wrote:
> >> Hi,
> >>
> >> On Thu, 02 Apr 2015 11:24:53 -0700 Josh Triplett <josh at joshtriplett.org>
> >> wrote:
> >>> sysvinit-core depends on initscripts, but sysvinit does not.  However,
> >>> sysvinit ships /lib/sysvinit/init, and it should be possible to use
> >>> sysvinit by booting with init=/lib/sysvinit/init without having
> >>> sysvinit-core installed.  Thus, sysvinit needs to have dependencies on
> >>> any packages needed for a functional sysvinit init system, including
> >>> initscripts (and potentially other dependencies of sysvinit-core).
> >>
> >> The sysvinit package (as shipped in jessie) was mostly intended as a
> >> transitional measure when upgrading from wheezy to jessie.
> >> I tried to make that clear also in the package description.
> > 
> > I had assumed that the sysvinit package would stick around as long as
> > sysvinit does.  What is your plan post-jessie?
> > 
> >> Keep in mind, that when installing jessie from scratch, there will be no
> >> /etc/inittab. This means, installing the sysvinit package (in parallel
> >> to systemd-sysv) will not result in a bootable system via
> >> init=/lib/sysvinit/init.
> >>
> >> Now, if there is desire to make the sysvinit package useful beyond the
> >> wheezy -> jessie upgrade, someone would have to deal with this inittab
> >> problem first, I think.
> > 
> > That's a good point.  I would have assumed that one of the sysvinit
> > packages would create /etc/inittab on installation.  I suspect that some
> > people installing jessie from scratch will find it rather surprising
> > when they're unable to switch to sysvinit simply by installing the
> > appropriate package.
> 
> So, we've been discussing the role of the "sysvinit" binary package
> within the pkg-systemd team just recently. We've come to the conclusion
> that keeping the sysvinit transitional package is actively harmful, as
> it's confusing for users. They need to pick between sysvinit-core and
> sysvinit, and intalling sysvinit won't actually lead to a bootable
> system if there is no pre-existing /etc/inittab.

This change would break the mechanism that allowed switching between
init systems via init= (or the generated options in GRUB).  That makes
it more difficult for someone who needs to test sysvinit for whatever
reason.  (The reverse works fine: having sysvinit-core installed but
booting via systemd.)

I realize that the missing /etc/inittab and missing initscripts
dependency already broke that for anyone who didn't start out with
sysvinit.  And I also realize that having a transitional package named
"sysvinit" that doesn't actually install sysvinit (unless you had it
already) could lead to confusion.  I don't have any objection to
removing the transitional package named "sysvinit".  However, it seems
unfortunate to completely lose the "dual-boot" functionality.

- Josh Triplett