Distro design needed

Distro design needed

Posted Jan 21, 2014 19:07 UTC (Tue) by coling (guest, #95094)
In reply to: Distro design needed by johannbg
Parent article: A Debian init system update

I think you do us a bit of a disservice by saying we "just took the migration work... that had already taken place".

As I've known Lennart for a number of years prior to the birth of systemd, I was very much tracking the systemd project from it's early days and had started the job of converting the distro to be one of the early adopters. When we shipped Mageia 2 in May 2012, we were certainly among the first to use it by default having been involved with the upstream systemd project since Jan 2011 (second only to Fedora and perhaps OpenSuse, but not sure if they defaulted to it until 12.2 which was later... don't know and I'm not really trying to keep "score" anyway)

Ultimately, the "migration work" involved a lot more than just sysvinit scripts -> native units. While we do regularly "liberate" such units from Fedora and other places, this is not really where the bulk of the work actually came in with systemd adoption. Most of it was adapting tools and utilities built on top of sysvinit and debugging/fixing the new lower levels that were now introduced.

On several occasions I've also found, corrected and properly notified Fedora package maintainers of the bugs and problems in their units - the NFS units being probably the most fundamentally broken ones I've come across! This is part of the FOSS ecosystem I enjoy and collaborating with maintainers in this way is IMO very healthy, but blindly taking Fedora units and using them has led to several bugs at our side, so it's not something one can simply take with "little to no effort" - at least not if you have even a passing regard for QA!

So for you to say that "they decide to switch to systemd and migrated with little to no effort compared to us in Fedora" is actually really completely untrue and completely misrepresents the time, effort and dedication that myself another other Mageia contributors/users put into the project and continue to do so.

Now, I know it wasn't meant in that sense (despite it being very demoralizing to read!), but the time taken for conversion IMO really has very little to do with the units themselves - after all, one of the primary advantages of systemd is that writing units is simple! The time is accrued due to the fact that we actively want to avoid writing new units, we want them to be the same as in other distros and indeed try to push them upstream whenever feasible and, regardless of the source of the units themselves, things still need tested! It actually takes more time in many cases to check around the other downstream sources than it would to just bash out a brand new unit - another key benefit systemd brings is standardisation and it's a shame if people don't put in the extra effort to ensure consistency at this final step!

Forgetting the various bits of a distro that are relatively unique (tools, installers, service management GUIs etc), at a package conversion level the time (IMO) is really spent:

• checking around the systemd distros to see if they do things differently
• if there are differences which is better
• find out if there are upstream efforts and if not try and instigate them (and find out which version has gone upstream etc. - essentially try very hard to reduce any x-distro-variation)
• and of course actually test the package and make sure it works properly.

I'd say most of the package-level problems we've had have come from the lack of (or insufficient time spent on) that last step - either inheriting bugs from the Fedora conversion, not dealing with a variation/deviation in sysconfig file format from Fedora, or introducing our own shiny new problems accidentally!

That my general take on it anyway. YMMV :)

---

(Log in to post comments)