Now that we have the display's context to hook into, we can use it to
construct other GL contexts and don't need a GdkSurface vfunc anymore.
This has the added benefit that backends can have different GdkGLContext
classes on the display and get new GLContexts generated from them, so
we get multiple GL backend support per GDK backend for free.
I originally wanted to make this a vfunc on GdkGLContextClass, but
it turns out all the abckends would just call g_object_new() anyway.
Instead of
Display::make_gl_context_current()
we now have
GLContext::clear_current()
GLContext::make_current()
This fits better with the backends (we can actually implement
clearCurrent on macOS now) and makes it easier to implement different GL
backends for backends (like EGL/GLX on X11).
We also pass a surfaceless boolean to make_current() so the calling code
can decide if a surface needs to be bound or not, because the backends
were all doing whatever, which was very counterproductive.
The code to create and manage a fake egl surface to bind to is
complex and completely untested because everyone seems to support this
extension.
nvidia and Mesa do support it and according to Mesa devs, adding support
in a new driver is rather simple and Mesa drivers gain that feature
automatically, so all future drivers shoould have it.
... or more exactly: Only use paint contexts with
gdk_cairo_draw_from_gl().
Instead of paint contexts being the only contexts who call swapBuffer(),
any context can be used for this, when it's used with
begin_frame()/end_frame().
This removes 2 features:
1. We no longer need a big sharing hierarchy. All contexts are now
shared with gdk_display_get_gl_context().
2. There is no longer a difference between attached and non-attached
contexts. All contexts work the same way.
Do not treat the context as already current when the value
of context::in-frame changes.
This is so we can bind to EGL_NO_SURFACE if context::in-frame == false
and to context::surface if context::in-frame == true.
This allows getting rid of the attached property in future commits.
The vfunc is called to initialize GL and it returns a "base" context
that GDK then uses as the context all others are shared with. So the GL
context share tree now looks like:
+ context from init_gl
- context1
- context2
...
So this is a flat tree now, the complexity is gone.
The only caveat is that backends now need to create a GL context when
initializing GL so some refactoring was needed.
Two new functions have been added:
* gdk_display_prepare_gl()
This is public API and can be used to ensure that GL has been
initialized or if not, retrieve an error to display (or debug-print).
* gdk_display_get_gl_context()
This is a private function to retrieve the base context from
init_gl(). It replaces gdk_surface_get_shared_data_context().
Create it during init and then reuse it for all contexts.
While doing that, also improve error reporting - that's not used yet but
will in later commits.
This is not used yet, but it allows surfaceless GL contexts.
For that purpose, we need to make the display a construct-only property,
so that it can be set when the surface isn't.
This adds a bunch of very picky checks in the constructor so nothing bad
can happen.
... and move some members from the GdkDisplay struct.
We've always wanted to add one to isolate the display from the backends
a bit more, but so far it's never happened.
Now that I'm about to add more data to GdkDisplay, it's a good excuse to
start.
We try EGL first, but are very picky about what we accept.
If that fails, we try to go with GLX instead.
And if that also fails, we try EGL again, but this time accept anything.
The idea here is that EGL is the preferred method going forward, but GLX is
the tried and tested method that we know works. So if we detect issues with
EGL, we want to avoid using it in favor of GLX.
Also add a GDK_DEBUG=gl-egl option to force EGL at all costs and not try
GLX.
That way, we can give a useful error message when things break down for
users.
These error messages could still be improved in places (like looking at
the actual EGL error codes), but that seemed overkill.
Query the EGL_VISUAL_ID from the egl Config and select a config with the
matching Visual.
This is currently broken on Mesa because it does not expose any RGBA
X Visuals in any EGL config, so we always end up with opaque Windows.
https://gitlab.freedesktop.org/mesa/mesa/-/issues/149
This reverts commit c35a6725b9.
This approach doesn't work because if NVIDIA doesn't work for EGL, the
EGL implementation won't be provided by NVIDIA, so checking the vendor
doesn't work.
Instead, use the display's "leader surface" when no surface is required,
because we have it lying around.
Really, we want to use EGL_NO_SURFACE, but if that's not supported...
Instead of going via GdkVisual, doing a preselection and letting the GL
initialization improve it, let the GL initialization pick an X Visual
directly using X Visual code directly.
The code should select the same visuals as before as it tries to apply
the same logic, but it's a rewrite, so I expect I messed something up.
1. We're using EGL most of the time anyway, so if we wanted to cache
things, we'd need to port it there.
2. Our GL handling is massively configurable, so determining when to use
the cache and when not is a challenge.
3. It makes startup nondeterministic and depend on whether a GTK4 app
has previously been started on this display and nobody thinks about
that when debugging.
4. The only benefit of the caching is delaying GL initialization - which
made sense in GTK3 where almost no app used GL but doesn't make sense
in GTK4 where almost every app uses GL.
So unless I find a big benefit to reintroducing it, this cache will be
gone for good.
Avoids having to use private data, though the benefit is somewhat
limited as we still have to put the destructor in the egl code and can't
just put it in gdk_surface_x11_finalize().
We only have one config, because we use the same Visual everywhere.
Store this config in the GdkDisplayX11 struct for easy access.
Also do this on initialize, because if creating the config fails, we
want to switch to GLX instead of failing to do GL at all.
This also simplifies a lot of code as we can share Visual, Colormap, etc
across surfaces.
There's no need to use g_object_set_data() for it.
We can also stop caching it elsewhere because we know the display has
it.
And finally, we can remove the display->have_egl boolean and use
display->egl_display != NULL instead. We initialize the display at
startup, so that variable is the perfect indicator.
We need to initialize GL to select the Visual we are going to use for
all our Windows.
As the Visual needs to be known before we know if we are even gonna use
GL later, we can't avoid initializing it.
Note that this previously happened, too. It was just hidden behind the
GdkScreen initialization.
We don't want to bind ourselves to GTK3 - both because we don't want to
accidentally cause bugs in a different codebase and because we want to
deviate from it.
While doing so, also store visuals as visuals and not as integers.
And only store one Visual because GTK4 only uses one visual.
And then remove the code that is leftover for dealing with the
compatibility Visual for GTK3.
PS: I'm kinda proud of my STRINGIFY_WITHOUT_BRACKETS hack.
The old code was ordering visuals by depth, but considering that these
days we either use the default visual or a 32bit RGBA visual, that
reordering does not have an effect anymore.
In theory, the only effect is that the GLX Visual selection might select
a different replacement Visual when it checks for improved GL Visuals, but
even there I can't come up with a case where that matters, because
again, the visuals are only reordered by depth and we want to keep the
depth.
In any case, make this a separate commit so bisecting can find this
problem if it ever shows up.
Instead of the display telling the screen to tell the visuals to tell
the display to initialize itself, just init the display directly.
What a concept.
That's a sneaky trick so my edit/compile/test cycle goes faster:
I usually use demos for testing so the tools don't have to be linked for
me to start testing.
If the pointer capability is added, pointer swipe and pinch gestures
will be created. However, if the pointer capability is removed, the
gesture objects won't be destroyed.
If the pointer capability is removed and added several times in a row,
for example due to plugging and unplugging physical mouse, this can lead
to leaking the old gesture objects.
In order to prevent that, this change makes the seat destroy swipe and
pinch gestures when the pointer capability is withdrawn.
It's only used during DND to allow use of the root window's cow window
as a DND target, because apparently gnome-shell used to think that was a
great idea to DND to the overview.
Somebody complain to gnome-shell devs about it not being a good idea if
they want it fixed.
Potentially using Wayland is a better idea though.
This reverts 85ae875dcb
Related: https://bugzilla.gnome.org/show_bug.cgi?id=601731
It's not 2011 anymore, and we shouldn't randomly build one of 10.000
different combinations of X11 backends (I counted the possibilities) but
exactly the one we expect people to use.
