The Debian init system general resolution returns

One can certainly consider improving upon sysvinit, but a rat's nest of processes and D-Bus interfaces is not an improvement.

Great. Feel free to design, implement, debug, and document something that is better than systemd. If systemd is really so terrible, people will be celebrating you for the next few decades.

In reality, though, actually coming up with a software system that does what systemd does but better is probably not all that easy. Talk is cheap, but actually doing the work is hard – which is why out of all the people who are moaning and griping about how systemd is so bad, nobody so far has actually got their gear together enough to come up with anything at all to compete with it. They tell us, for example, that sysvinit would be perfectly equal to systemd if only someone added component X, feature Y, or approach Z that is superior to what systemd does, but nobody develops the code or writes the documentation. Many of the missing bits and pieces purportedly exist and are working great, but nobody even comes up with a proof-of-concept derivative of a popular Linux distribution that integrates them by default. My personal guess is that most of these people simply lack the vision, technical competence, and will of someone like Lennart Poettering, but they still think they're entitled to tell everyone what they should or should not be using. I don't think these people are really worth listening to.

A good modular design would have been technically acceptable to everybody

Have you actually looked at systemd? The design is pretty modular as these things go. And the interfaces between the modules are way better documented than anything in the sysvinit universe.

Systemd is about seizing control of Linux.

Exactly who is supposed to be “seizing control of Linux” through systemd? The systemd developer community includes people from all major Linux distributions. If getting rid of every distribution's /etc/hostname, /etc/HOSTNAME, and /etc/sysconfig/whatever_the_file_is_called_this_week in favour of one standardised configuration file is “seizing control of Linux“ then I say bring it on.

And the extraordinarily serious downside of systemd's awful design is that the lack of modularity makes it hard to improve Linux in future.

Systemd consists of a bunch of modules with documented interfaces and a stability promise. It would certainly be possible to replace individual modules with different implementations that adhere to the same interfaces. It may be often be an easier route, however, to simply improve systemd's components instead. These could be forked if required, and/or improvements could be fed back into the systemd development process.