Comment by spijdar
2 days ago
Learn about what stuff? Linux? System V UNIX?
I haven't done LFS since my tweens (and I'm almost 30 now), but I remember the sysvinit portion amounted to, past building and installing the init binary, downloading and extracting a bunch of shell scripts into the target directory and following some instructions for creating the right symlinks.
Obviously, you can go and check out the init scripts (or any other individual part of LFS) as closely as you wish, and it is easier to "see" than systemd. But I strongly protest that sysvinit is either "Linux" (in that it constitutes a critical part of "understanding Linux" nor that it's really that understandable.
But setting aside all of that, and even setting aside the practical reasons given (maintenance burden), when the majority of "Linux" in the wild is based on systemd, if one wanted to do "Linux From Scratch" and get an idea of how an OS like Debian or Fedora works, you would want to build and install systemd from source.
For me, Linux From Scratch is not about compiling linux from scratch, but on building up an entire Linux distro from the ground up, understanding how every piece fits together.
Doing it via systemd is like drawing a big black box, writing LINUX on the side, and calling it a day.
You are necessarily working with very big blocks when you're doing this, anyway. You don't do a deep dive on a whole bunch of other topics in LFS, because otherwise the scope would become too big.
That's what I was trying to get at -- yes, you can say that sysvinit is easier to understand than systemd, and less of a black box. But, even still, a "real Linux distribution" is full of these black boxes, especially the closer you get to being able to run "real applications". I'd argue that once you get into full desktop seat management, you add so much complexity on top of sysvinit that the difference narrows...
Which is why I asked "learn about what stuff". I think if the goal is to learn about "Unix" or OS design/ideas, you're better off with a leaner, "pedagogical" OS, like xv6. If the goal is to piece together an OS and really understand each piece, I don't think you really want sysvinit. You want something closer to an /etc/rc.local that just kicks off a few daemons and hopes for the best.
You can argue that sysvinit makes a better "compromise" between usability and clarity, and I'd entertain that idea, but then I think dinit is far easier to understand than sysvinit. And of course, at that point you can shave yaks till you fill the bike shed with wool.
Realistically, as much as people may hate it, if you have to pick a single init to standardize on for clarity and "building an entire Linux distro from the ground up, understanding how every piece fits together", systemd is the most rational choice. It's the most representative of the ecosystem, and requires the least "extra layers" to make the "desktop layer" work.