Where to put kwayland integration plugins

[David Faure]

On samedi 2 juillet 2016 15:25:39 CEST Martin Graesslin wrote:
> On Saturday, July 2, 2016 11:41:11 AM CEST David Faure wrote:
> > On lundi 6 juin 2016 11:26:58 CEST Aleix Pol wrote:
> > > On Mon, Jun 6, 2016 at 8:32 AM, Martin Graesslin <mgraesslin at kde.org>
> 
> wrote:
> > > > Hi framework-developers,
> > > > 
> > > > in Plasma we have a repository called kwayland-integration. It
> > > > provides
> > > > a
> > > > plugin for:
> > > > * KWindowSystem
> > > > * KIdleTime
> > > > 
> > > > using the KWayland client API. Thus it makes these frameworks
> > > > functional
> > > > on
> > > > windowing system Wayland.
> > > > 
> > > > I want to move the plugins into frameworks and are wondering how to do
> > > > that
> > > > and where to move them to.
> > > > 
> > > > I see the following options:
> > > > 1. Integrate directly into kwindowsystem and kidletime
> > > > 2. Merge them into frameworks-integration
> > > > 3. Create a new framework kwayland-integration
> > > > 4. create a new framework "kwindowsystem-wayland" and
> > > > "kidletime-wayland"
> > > > 
> > > > Option 1 is I think an absolute no-no as it would turn kwindowsystem
> > > > and
> > > > kidletime from tier 1 to tier 2 and that would affect a huge amount of
> > > > other frameworks. For everything which is not in tier 1, I think it's
> > > > the
> > > > way to go.
> > > > 
> > > > Option 2 is something I don't want to do as that would combine
> > > > windowing
> > > > system integration code with random other stuff and would result in
> > > > weird
> > > > dependencies. (To use KIdleTime on Wayland one needs X11 and Phonon?)
> > > > 
> > > > The same actually also applies to Option 3, though it is just two
> > > > frameworks and all dependencies would be tier 1.
> > > > 
> > > > So personally I think option 3 or option 4 are the way to go.
> > > > 
> > > > What do you think? Better ideas?
> > > 
> > > I agree ideally it's 3 or 4. Where do you have it now? Are they
> > > already separate repositories? If so, just moving them to frameworks
> > > could make sense.
> > > 
> > > In Purpose I have a similar problem (and I know the discussion is also
> > > relevant for other frameworks). We usually don't want plugins to raise
> > > the tier of your base framework, but you still need them to be
> > > distributed together and either way you need to make sure the plugins
> > > will be available when the system is deployed (which is actually much
> > > easier in option 1).
> > 
> > The alternative is to use option 1 with a cmake option to disable the
> > kwayland-based plugin. This offers benefits because
> > - on systems with wayland enabled, you are sure the plugin is present (no
> > bug report like "idle detection doesn't work" (because of a missing
> > package)) - on systems where dependencies should be kept to a minimum
> > (e.g.
> > an X11-based embedded system) one can easily compile without the kwayland
> > dependency
> > 
> > Application developers using KIdleTime or KWindowSystem do expect it to
> > work on all platforms where the application works, so surely a dependency
> > on kwayland can't be a problem on wayland.
> 
> That's actually not the case. Both KIdleTime and KWindowSystem (runtime)
> depend on additional Wayland protocols currently only provided in KWin/
> Wayland. That means your KIdleTime based application won't work in e.g.
> GNOME.

OK so things are less interoperable than I thought
  -- so much for "whatever the problem is, wayland will fix it!" ;-)

However that only makes my last paragraph moot, not the rest of the argument, 
AFAICS.

In fact it even leads to another possible solution : if these plugins are only 
useful in Plasma5, then there's no problem with putting them into plasma-
integration, is there?

-- 
David Faure, faure at kde.org, http://www.davidfaure.fr
Working on KDE Frameworks 5