I agree that it's kind of ugly to have lots of scripts with boilerplate code in them as a configuration management system -- but: it's simple, and it works. System startup is a complex task, and the init scripts manage to do that with code that a competent sysadmin could understand. It is extremely transparent, with a minimum of black box code. So while it is ugly, it is also a very elegant solution in its simplicity. It let sysadmins and power users understand and tweak anything they needed.
I disagree that it's elegant. In fact I disagree that it's even a solution. I'm a power user/admin of my own system and many systemd's additions proved very handy for me. I always had problems reading complex bash scripts, a script based solution for system startup is practically undebuggable for me. I'm better at C than at shell. And unit files are easier for me to deal with than shell scripts. Ability to display system log only for current boot is also extremely handy.
In general this is true, but in this case the dependencies are too heavy. There is a clear hierarchy of importance: the init process is much more important than DBus. Specifically, you don't want the system to be unbootable if dbus fails. It's sort of like if the kernel relied on Avahi to boot. Maybe it's better to have a lightweight custom IPC code instead. Also, there's a big difference between library dependencies and service dependencies. In principle they are the similar, you either pass messages via function calls or via sockets or TCP. In practice there is so much more that could go wrong with a service like dbus than a library. And with non-service libraries you can even link statically, which would be useful to ensure that the init didn't fail after a botched upgrade. Also, having a lot of state outside the filesystem makes it difficult to create chroot environment and to isolate services.
DBus seems to be all about boot speed. In the process they made all kinds of troubleshooting (except performance) significantly harder.
It's not so clear to me. I don't see how DBus is too heavy while custom IPC solution isn't. I don't feel comfortable evaluating design of boot systems. I'm not interested in developing them, only using them. I'd rather let people who actually develop them do so.