[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Proposal: switch default desktop to xfce



On Fri, Oct 25, 2013 at 06:15:27PM +0100, Neil Williams wrote:
> > On Fri, Oct 25, 2013 at 12:41:26PM +0200, Josselin Mouette wrote:
> > > Le jeudi 24 octobre 2013 à 16:40 +0100, Steve McIntyre a écrit : 
> > > > This goes back to during the wheezy release cycle. There was a
> > > > little discussion around a change in tasksel [1], but rather too
> > > > late in the day for the change to make sense. Now we have rather
> > > > more time, I feel. Let's change the default desktop for
> > > > installation to xfce.

> > > What are the reasons exactly for deliberately depriving the default
> > > installation’s users of a more complete and featureful desktop?

> > > So far, in this discussion, I only read “I don’t like
> > > systemd” (which is irrelevant)

> > This is a strawman.  The real objection is "tight coupling of a
> > desktop to an init system and kernel is at best lazy engineering, and
> > at worst a massive power grab by an upstream, and is contrary to
> > Debian's values."

> +1

> The option existed to make the desired features optional and that
> option was deliberately written out in an effort to extend GNOME beyond
> a desktop. A land grab.

> Other desktop environments have similar features without requiring a
> change of init system. It was a choice by GNOME upstream and a choice
> that could have had a much more friendly and supportable alternative.
> The choice has been made and it seems unlikely that GNOME upstream will
> now accept requests to change direction, leaving distributions to sort
> out the mess.

I'd like to be clear here that I do not fault GNOME upstream for making the
decision to leverage the systemd dbus APIs.  The session manager is a very
important component of the desktop, and as Olav has pointed out, the choices
in the field are logind or consolekit.  Of the two, logind is a much better
code base, with a better architecture that learns from the mistakes of
consolekit and takes advantages of Linux features (i.e., cgroups) that were
not available at the time consolekit was written.

The problem comes when logind in turn dictates the init system, which is
what happens with systemd 205.  Tollef, one of the systemd maintainers in
Debian, has said here that he "[does] explicitly not guarantee [running
logind without systemd init] will work at all."  As a systemd maintainer,
that's not an unreasonable position for him to hold; providing such a
guarantee would imply a lot of work that he doesn't want to commit to, work
which doesn't align with his and upstream's goal of seeing systemd more
widely adopted.

But each of these individual decisions, which are made in good faith and
represent a local optimum, result in a terrible outcome for Debian.

My assertion is that it's not acceptable to Debian for:

  GNOME to be the default desktop, and
  GNOME to depend on logind, and
  logind to require systemd init.

But there are multiple ways to solve this problem.  Steve McIntyre has
already made it clear that he thinks we should negate the first of these.  I
have no strong opinion on that, but do feel we should address the third of
these, because the logind service is actually quite good and desktop
environments *should* prefer that service over consolekit... they just
shouldn't be forced to adopt systemd init as baggage along with it.  And I
think Debian, as a project, should push back on this tight coupling, and
require an implementation of the logind interface independent from systemd
init.

In the short term, this could be a committment from the systemd maintainers
to hold the package at version 204 until the dust settles around cgroup
manager interfaces[1].  Or it could be an agreement to provide a 'logind'
virtual package for GNOME to depend on, and someone could repackage systemd
204 to provide the dbus services separately, so they remain independently
available even if the systemd package moves on in unstable.  This is
something I can imagine being willing to help with.  (It would probably be
less of a time committment than trying to keep up with this thread!)

In the long term, we certainly need a decision for the default init system
in Debian.  If Debian decides to adopt systemd as the default (via its
constitutional processes, not because a GNOME dependency forces the issue),
then it's not worth anyone's effort to maintain a fork of logind.  If the
decision is for something other than systemd, then we need to address the
maintenance of such a logind fork to make sure it remains viable, so that it
doesn't wind up bit rotting like consolekit.  That requires someone to put
their money where their mouth is, and be willing to maintain the code.

-- 
Steve Langasek                   Give me a lever long enough and a Free OS
Debian Developer                   to set it on, and I can move the world.
Ubuntu Developer                                    http://www.debian.org/
slangasek@ubuntu.com                                     vorlon@debian.org

[1] Those interested in the cgroup management might also want to be aware of
ongoing discussions in Ubuntu about this, including a session at the
upcoming UDS next month:
https://blueprints.launchpad.net/ubuntu/+spec/core-1311-cgroup-manager

Attachment: signature.asc
Description: Digital signature


Reply to: