Devuan Jessie beta released

Posted May 2, 2016 14:50 UTC (Mon) by mgb (guest, #3226)
In reply to: Devuan Jessie beta released by Cyberax
Parent article: Devuan Jessie beta released

> Technical arguments? Who cares about technical arguments?! SystemD is the Great Satan of software and must be destroyed.

Great Satan? No. Harmful monolithic hack? Yes.

Hacking is creative and wonderful. Much that is good in computer science started as hacks. But init systems are not new. Dependency-based init systems are not new. Gnu/Linux needs an engineered init solution, not a hack backed by a tawdry political movement. I don't know who V.R. is but this is half of the sort of document that the Debian Tech Committee should have produced instead of the hand-waving we saw from both sides: http://blog.darknedgy.net/technology/2015/10/11/0/ Anyone who has to use systemd would do well to read V.R.'s document for a much better understanding than systemd's own documentation offers.

Nevertheless, despite ignoring decades of software engineering, the systemd hack would still be a valuable experiment if it were truly modular. It is not. It has a vast and untidy surface over a convoluted and unstable interior. There are arguments to the contrary by those who look at the nitty gritty of makefiles and build options while ignoring the big picture but the fact is that distros are splitting into with-systemd and without-systemd flavors because by intent and by poor design systemD is in practice monolithic.

Systemd succeeded because Gnu/Linux is modular - truly modular in practice. But SystemdD itself is monolithic. It blocks the way forward for Gnu/Linux. This is why I have always opposed systemd and why the route forward from Wheezy is Devuan. Not because systemd was an undesigned hack that just grew and metastasized, not because systemd is a bad implementation of a good idea, but because systemd seriously impedes the future development of F/LOSS.

to post comments

Devuan Jessie beta released

Posted May 2, 2016 15:16 UTC (Mon) by peter-b (guest, #66996) [Link]

I'm looking forward to your forthcoming good implementation of systemd's "good idea". If it successfully addresses the problems that systemd solves, then I'm sure it will be adopted very enthusiastically.

I don't mean this sarcastically or dismissively -- I genuinely believe that there is space for multiple implementations of a good dependency-based, race-free init system for Linux. Even better if it exposes an implementation of the public systemd API so that people can try it out with their existing tools more easily.

Devuan Jessie beta released

Posted May 2, 2016 15:21 UTC (Mon) by anselm (subscriber, #2796) [Link] (1 responses)

… but because systemd seriously impedes the future development of F/LOSS.

Considering that the cobbled-together suckage that is the traditional setup (System-V init, init scripts, inetd, cron, at, syslogd, …) apparently hasn't managed to seriously impede the future development of F/LOSS for the last 25 years or so, I wouldn't worry unduly about systemd. People have obviously dealt with much worse stuff than systemd (and by all indications don't mind planning on still dealing with it going forward, see Devuan), or we wouldn't be here now.

Devuan Jessie beta released

Posted May 2, 2016 15:59 UTC (Mon) by johannbg (guest, #65743) [Link]

Well there are cases where you still "cobbled-together" components with systemd like for example it's a common misconception that type timer units replace a cron based solution that stems from people filled with the bright idea of migrate "everything" to native systemd type units which ends up like a work of bunch of idiots re-implementing cron in the form of timer units but now with the double the administrative overhead to it's end users and no technical benefits.

When I went through the whole components that ship cron scripts in Fedora ( ca 100 components ) only half of that was applicable to be migrated to timer units and once I had gone through those ca 50 components that ship those cron scripts and filtered out the unused, obsoleted scripted trash which seem to be shipped and exist only to waste peoples space on hardrives or serve as an example to the terminal world how not to write a script, I ended up with roughly 30 components which would have been migrated.

In the above example these two solution complement each other short comings and should be implemented as such where applicable in the distribution and elsewhere not re-implemented in the form of type timer units instead of cron scripts because they "can" or systemd is "hot" for the moment.

Devuan Jessie beta released

Posted May 2, 2016 18:19 UTC (Mon) by Cyberax (✭ supporter ✭, #52523) [Link]

> Gnu/Linux needs an engineered init solution
Then provide one.

> I don't know who V.R. is but this is half of the sort of document that the Debian Tech Committee should have produced instead of the hand-waving we saw from both sides: http://blog.darknedgy.net/technology/2015/10/11/0/ Anyone who has to use systemd would do well to read V.R.'s document for a much better understanding than systemd's own documentation offers.
I read this document. I don't see any real technical points against systemd design in principle. And no, "inelegant" does not count. Real-life systems are almost always "inelegant" from theoretical standpoint.

There are also lots of inaccuracies in details (like the statement that cgroups v2 requires the single writer).

And really, statements like:
> It should be noted that dynamic tracing is another potential way to achieve deferred execution conditional upon resource availability.
Are a _huge_ warning sign for anybody maintaining real-life systems.