Re: Module Proposal: GNOME Shell

[Owen Taylor <otaylor redhat com>]

On Sat, 2010-04-03 at 14:52 +0200, Josselin Mouette wrote:
> Le vendredi 02 avril 2010 à 08:34 -0400, Owen Taylor a écrit : 
> > We've always planned to require graphics acceleration. To review:
> > 
> >  * We can't take advantage of the capabilities of graphics
> >    acceleration in the user interface design unless we can count
> >    on it - otherwise the graphics acceleration is at best tack-on
> >    eye candy.
> > 
> >  * Developing two separate code paths for accelerated and
> >    non-accelerated graphics is also a large increase in
> >    development resources.
> 
> These are good arguments, but you are still being way too optimistic
> when it comes to the state of drivers. 
>       * Intel drivers are in a terrible shape, and it’s a nightmare to
>         get a kernel+X combination that works with all adapters. 

I don't think "in a terrible shape" is at all accurate. There are enough
different pieces that yes, it can be hard to assemble a working
solution. Assembling a working solution is not GNOME's problem, except
that we have to provide Intel and distros with the necessary pieces so
they can do regression testing.

>       * Radeon KMS drivers are very slow (too slow to run gnome-shell,
>         at least) and still stabilizing. 

Almost since the beginning 18 months ago, my personal gnome-shell
development has been done on an ancient Radeon rv350 with KMS drivers.
Not sure where your information is coming from.

>       * Nouveau is still experimental. 

Yes, but some encouraging progress recently.

>       * Non-free drivers are terribly lagging behind in terms of
>         non-core functionality (RandR? Ha, ha, ha.) 

Not GNOME's problem.

>       * Hardware from manufacturers other than the big 3 has zero
>         support.

Hardware from manufacturers other than the big 3 has close to zero
market presence too. If the documentation and motivation for developers
was there, the support would be there.

> > * There is zero reason that virtual machines can't also have 3D
> >    acceleration - and VirtualBox, VMWare, Parallels, etc, do this
> >    currently. KVM/qemu lags, that just needs to be fixed.
> 
> This is easier said than done.

"Easier said than done" also applies to providing a modern desktop
without using modern hardware. Of the two projects, virtualized 3D is
vastly more straightforward and tractable.

> And if VMware has 3D support, it is still
> very unstable. Which is expected when you try to access the same
> hardware with different drivers in the host and the guest, both as
> crappy.
> 
> You are also forgetting the trend towards thin clients. The protocols to
> display 3D remotely do not even exist. Everything has yet to be invented
> if you want to see 3D on what will be the standard desktop for an
> increasing part of our users.

I don't think there is a trend toward thin clients. Yes, there are
people running thin clients, as there have been for 20+ years. Protocols
to display GL remotely have existed certainly for over 10 years. There
are current deployements like the City of Largo using Compiz over remote
GLX. GNOME Shell will not perform as well as Compiz over remote GLX, but
much of that is fixable.  (More sophisticated protocols can also be
written and will be needed to run 3D desktops over low bandwidth links.)

Windows Vista and Windows 7 can be remoted with 3D acceleration. We
can't hide our head in the sand and claim that remoting 3D is too hard
and we must stick to 2D, because the end effect will simply be that
people won't use our software.

> > So, the official plan is basically that people can still use the 
> > GNOME 2 panel and window manager with GNOME 3 applications and
> > libraries, if necessary, but this is a transitional state, 
> > and to get the GNOME 3 experience, you need hardware acceleration.
> 
> As things are going, you are leaving us no choice but to keep shipping
> gnome-panel by default for a very long time, unless we want to provide
> two radically different user experiences.

How does "shipping gnome-panel by default" not equal providing two
radically different user experiences?

But anyways, if you are speaking for Debian, the GNOME Shell team's goal
is to make you ship GNOME Shell by default, by making it so cool and
obviously better that your users will run you out of town if you leave
them with old the GNOME 2 panel. :-)

- Owen