Ubuntu To Abandon Upstart, Switch To Systemd

[Delgarde]

Extremely surprised. Especially since Debian has only decided on systemd for Jessie and not necessarily for later releases.

They made the decision for Jessie only, because that's what was relevant. But that doesn't mean that they've decided to trial systemd for one release, then repeat this whole blood-on-the-floor decision process for the next one. It just means they're not committing themselves to stick with the same system forever, even if circumstances change.

Seriously - if they're going to the effort of switching init implementations, they're really not going to be treating it as a short-term thing, expecting to have to do all the work again in a few years time...

[Delgarde]

Upstart was a technically promising solution which the Debian Technical Committee (TC) praised for its code quality and technical merits.

Debian could have went with Upstart, but was held back by the CLA.
If Canonical didn't require CLA on Upstart, the Debian outcome could have had an other outcome.

No, I think it goes back much further than that. For all the arguing, it was fairly obvious from the start that the CTTE would decide for systemd - and that that decision was mostly based on technical merit, and on the fact that any other decision would be sweeping back the tide of what others were doing. The CLA was a factor for the pro-systemd voters, but not the deciding one.

To my mind, the real damage from the CLA was done long before, when it prevented widespread adoption of upstart early on. I think systemd was a superior design from the start, but not so decisively so - had upstart managed to establish itself more widely, systemd would have struggled to get a foot in the door. But as things stood, upstart received relatively little attention outside Ubuntu, leaving things wide open for systemd.

[Delgarde]

Next: Mir to Wayland?

Hell=Frozen. It will enter eternal ice age once they adopt Wayland.

Nah, it'll happen eventually. Canonical just doesn't have the resources to maintain *everything* themselves, and it's hard to justify something that's a lot of effort for little benefit compared to just going with what all your upstream projects are doing. That's why they're giving up on upstart in favour of systemd, and while they might persist for a year or two, I think the same will happen with Mir once Fedora and others put Wayland into production.

[Delgarde]

I didn't get that either. They decide to adopt systemd, but are still ensuring that logind will work without systemd. Why?

Because it's work that's mostly complete already, and something Debian do actually want (since they intend to support multiple inits), and something that's still useful to Ubuntu until they move onto systemd themselves.

[Delgarde]

Plus, Gnome still probably wants to work on things like Solaris and FreeBSD.

No, I don't think they do. 99.999% of Gnome developers are using Linux, and while they're not against portability, nor is it something they consider of great importance.

Back in the day, Sun put a lot of effort into supporting Solaris, as part of their SunRay thin-client platform - they employed several Gnome developers, and sponsored a lot of usability testing. But that's the past - Oracle still sell that stuff, but Solaris is effectively dead as a desktop platform, and I don't think they contribute much (if anything) to Gnome these days.

As for FreeBSD, they're just too small, lacking in influence compared to the 99.999% of developers working on Linux. Always have been - as Linux built more and more system infrastructure (stuff like HAL, DeviceKit/Udev, Udisk/Upower, NetworkManager, etc), BSD developers tried to port those things, but they were just too few - they'd spend effort getting something half working, just in time for the Linux developers to write that thing off as a failed experiment.

[erendorn]

But systemd starting everything willy nilly has never made sense to me. That's how Windows works. Windows takes forever to load the desktop. Ubuntu (upstart) is very fast boot with no delay.

Really? Fedora boots faster than ubuntu, and definitely less than 5 seconds to usable desktop on my NUC.

[amehaye]

No.
Mir and Wayland are different things.

Mir is a display server. Wayland is a display server protocol.

You can't abandon mir to use wayland. Wayland it not usable, only implementable.

What can happens in the future that a 2.0 version of mir can implement the wayland protocol. And according to some ubuntu devs, it not so hard to do this.

This. Just make Mir a Wayland compositor and be done with it. We already have Gnome/Mutter, KDE/KWin etc. all behaving as Wayland compositors. Why not Unity/Mir?

Canonical would get the control they desire, everybody can just program against the Wayland API, everybody wins.

[GreatEmerald]

This. Just make Mir a Wayland compositor and be done with it. We already have Gnome/Mutter, KDE/KWin etc. all behaving as Wayland compositors. Why not Unity/Mir?

Canonical would get the control they desire, everybody can just program against the Wayland API, everybody wins.

I'm pretty sure that's what people mean when they say that Ubuntu should adopt Wayland. Not that they should switch to Weston or anything like that.

[Rich Oliver]

Well this decision lays bare Canonical's game, I'm surprised its not been more remarked on. At any point Canonical could have abandoned the CLA for Upstart. There was a chance at least of swaying Debian. But Canonical had no interest in Upstart without the CLA. Canonical's plan no doubt was to do an Android at some point creating a semi open platform based on Linux that they could control and monetarise. Canonical's problem was that they were just far too small a player to be able to pull it off. Canonical might have been able to fool a lot of the linux user community, but there were far too many players in the industry who spotted what they were up to.

[Skrapion]

What can happens in the future that a 2.0 version of mir can implement the wayland protocol. And according to some ubuntu devs, it not so hard to do this.

Do you have a reference for this? I'd like to read it.

It doesn't really make sense to implement the Wayland protocol in Mir, unless they're planning on only implementing part of it, or adding proprietary extensions.

Wayland and Mir aim to do the same thing: to be the glue between the compositor and the clients. They both do it through a server library (libwayland-server/libmirserver) and a client library (libwayland-client/libmirclient) which communicate to each other through IPC.

The major differences?

1) Wayland has a well-defined protocol between libwayland-client and libwayland-server, so you could theoretically reimplement the protocol yourself instead of using the standard libraries. Mir makes no promises here, and warns that things will probably break if you try this.

2) The server libraries draw the line between library and compositor at a different point. A lot of stuff that Wayland expects to be done in the compositor is done in libmirserver instead of in Unity.

3) The two projects prioritize features differently.

That's pretty much it. Architecturally, they're more similar than dissimilar.