Re: [sysvinit-devel] Reading runlevel without utmp

[Jesse Smith]

> Personally, I believe it to be a really bad idea to duplicate the
> runlevel information in several locations, and would have instead
> choosen to simply state that systems without utmp(x) suppport are broken
> and need to get utmp support before they will work properly with
> sysvinit.  utmpx is part of POSIX.

utmpx may be part of POSIX, but that doesn't mean it is supported
everywhere. Refusing to implement a work around for systems without utmp
just leaves us in a stalemate where musl c won't implement utmp because
of the various issues with the standard. And, on the other side, us not
providing a workaround, leaving halt/reboot broken on those systems. In
that scenario nobody wins.

Providing a work around was fairly simple and, so far as I can tell, has
no negative side-effect.

What is your reasoning for believing providing the work around is a
"really bad idea"? What drawback do you see from having it?

> How do you handle systems where /var/run/utmp and /var/run/runlevel
> differ?  Who to believe?

If both exist, then init assumes utmp is correct. The /var/run/runlevel
file is only used as a fallback if utmp doesn't exist or is corrupted.

> Btw, isn't /run/ the proposed location these days?  Why /var/run/?
>

The /run directory doesn't exist on a lot of classic/legacy/UNIX
systems. In fact, I don't think it's used anywhere outside of modern,
mainstream Linux distros. This patch was intended to improve
cross-platform support. Relying on /run would have been a case of one
step forward, one step back.