[TriLUG] Recommendations for a systemd-less Linux distribution

[Gregory Woodbury via TriLUG]

When SysVInit was written (for UNIX Sys IV) there was a deliberate
choice to punt the order of execution to humans for a few reasons;
we couldn't decide on a reasonable dependency syntax, and solving the
dependencies was a computationally hard problem at
the time (hardware was not as advanced.)

There is no real reason for some of those init scripts to be so
bizarre. As packages we added to distros, the authors tried to
have the init scripts handle all sorts of conditions about the system
environment at runtime, rather than examining it at
install and putting a reasonable configuration file in an appropriate
place.  In System IV all sort of new stuff had been added
to System IIIs base and the init system had to be re-written, with a
requirement that most of the SysIII stuff could be still used.
We wanted to do away with the inittab, but that was forbidden by mgmt.
We did at least get the Murray Hill group to make
a few changes to the shell to make some magic easier (using exec to
change file connections.)  I was working at the Holmdel
Bell Labs site on one of the first internal projects to use System IV
(never made it out of internal use) so our sysadmin group
was doing a lot of the nitpicking fixes that Murray Hill was to
disconnected to see or care about.

On Thu, Jul 16, 2015 at 2:24 PM, Steve Litt via TriLUG
<trilug at trilug.org> wrote:
> Yes. Sysvinit and OpenRC have init scripts from hell. But runit, s6,
> Epoch, and Suckless Init + Daemontools-encore + LittKit all would have
> cleaned up those init scripts.
>
> In discussing systemd, it's important to avoid the false choice of
> systemd vs upstart vs sysvinit, because there are *a lot* more init
> systems out there, and in my opinion, systemd, upstart and sysvinit are
> three of the worst.
> Steve Litt
-- 
G.Wolfe Woodbury
redwolfe at gmail.com