Re: GUADEC 2008 GTK+ Meeting Minutes

["Colin Walters" <walters verbum org>]

On Wed, Jul 23, 2008 at 10:59 PM, Paul Davis <paul linuxaudiosystems com> wrote:

> Here's an example of real world program almost certainly on your
> desktop that needs it:
> http://blogs.gnome.org/dcbw/2008/06/11/great-taste-less-filling/
> That mockup really needs a canvas that has features like painless
> GtkWidget embedding, powerful
> styling and layout.

That mockup absolutely does NOT need a canvas.

Ok; how would you do it?  If you're thinking WebKit or GtkMozembed or something - while I agree that's a possibility for this, it's a very heavyweight solution for a lot of things that just need a flexible layout system without dragging in HTTP and _javascript_ for example - other reasons are outlined by Owen at the bottom of this mail:
http://groups.google.com/group/mugshot/browse_thread/thread/73ba19bdd8f2f993#

Not sure I agree with that. The core team seem to have a pretty great
grasp of what a canvas *should* do, but they also have a huge reluctance
to commit to anything that doesn't get really close to their ideal
vision. Out in app-devel land, most of us have used canvases (Gnome,
Goo, Hippo, ccc, Dia and more) that are not perfect but enable us to get
stuff done reasonably, even if not ideally.

Sure; but we should be keeping in mind that in the end we need to choose.
 

> One possible approach with the canvas would be to have parts of GTK+
> that are not API/ABI stable - perhaps they require a Havoc-style
> "#define GTK_I_WANT_A_CANVAS 1".  Once the canvas is stable we declare
> it GTK+ 3.0 from a marketing point of view (but don't change the
> soname), like what GNOME is doing with GNOME 3.

Not a bad idea, except that when an app is distributed that WANTS-A
canvas to a system where GTK was built with FORGET-A canvas, what
happens?

The idea is that the canvas is not ABI stable for a while - applications depending on it that are shipped by distribution vendors would need to be carefully managed (possibly go the parallel install route).  But it would still get the canvas feedback from some real world appplications.   I admit I'm just thinking out loud here.