gtk_snapshot_pop() => removed
gtk_snapshot_pop_and_append() => gtk_snapshot_pop()
So now there is no way to get a rendernode out of the snapshotting API
until you gtk_snapshot_finish().
We already take ints when setting the translation, so it can't
currently take any other values. Additionally, I was seeing large
costs in int -> double -> int for the rects in
gtk_snapshot_clips_rect(), as all callers really are ints (widget
allocations) and the clip region is int-based.
This change completely cleared a 2% rectangle_init_from_graphene from
the profile and is likely to have nice performance effects elsewhere
too.
So we can set the css name of a widget to something that's not related
to the class name. If the css-name property is set to NULL, we will
still fall back to the one set using gtk_widget_class_set_css_name which
is alwasys non-NULL since GtkWidget itself sets it to "widget".
This causes the snapshotting algorithm to dump all widget nodes into
their own container node. We then name that group accordingly (ie
"GtkSwitch<0xdeadbeef>") so you can easily see which node belongs where.
The feature is toggleable in the inspector's visual tab.
There's a few problems with it, becuse GtkSnapshot optimized container
nodes away if they are not needed, so we are losing some widgets...
It turns out, some simple getters - such as
gdk_drawing_context_get_clip() - love copying things before returning
them.
I guess somebody has to burn cycles...
That way we can capture both the actual changes (clip region) and the
area that was redrawn (render region), which in OpenGL might not be
identical.
Nothing shows the render region yet though...
Just to avoid having to do NULL checks when calling
widget_class->snapshot. We were crashing with drawing areas who don't
have a draw or a snapshot vfunc (woot!).
We need so subtract the allocation from the clip to get the clip offset,
not the other way around.
This was screwing in particular with marks on GtkScale, because GtkScale
mark clip computation is broken and always returns (0,0) which makes
scales have a waaaaay too large clip.
But that's another bug.
And use this to cull widgets and gadgets that are completely outside the
clip region.
A potential optimization is to apply this clip region to cairo contexts
created with gtk_snapshot_append_cairo_node(), but for that we'd need to
apply the inverse matrix to the clip region, and that causes rounding
errors.
Plus, I hope that cairo drawing becomes exceedingly rare so it won't be
used for the whole widget factory like today (which might also explain
why no culling happens in the widget factory outside the header bar.
We now look at which of get_render_mode, draw or snapshot vfuncs is the
latest to have been overwritten in the class tree and then use that one.
This allows GtkContainerClass and GtkBinClass to override all of them
for without screwing things up.
We now try to emulate cairo_t:
We keep a stack of nodes via push/pop and a transform matrix.
So whenever a new node is added to the snapshot, we transform it
by the current transform matrix and append it to the current node.
This one introduces the Recording object which is essentially a single
instance of something that happened.
The RenderRecording is an instance of an actual rendering operation.
Switch code to use gdk_display_is_composited() instead.
The new code also doesn't use a vfunc to query the property but rather
requires the backend to call set_composited()/set_rgba() to change the
value.
Before, we would immediately invalidate the GdkWindow of the widget, now
we call the parent's GtkWidgetClass.queue_draw_child() function.
This allows the parent to track redraw queueing of children.
By default GtkWidgetClass.queue_draw_child() will again chain up to its
parent while respecting the GdkWindow hierarchy for clipping.
GtkWindow is then the only widget actually invalidating the GdkWindow.
This essentially moves redraw queueing from GDK to GTK.
These complicate a lot of GdkWindow internals to implement features
that not a lot of apps use, and will be better achieved using gsk.
So, we just drop it all.
Add a new ::measure vfunc similar to GtkCssGadget's that widget
implementations have to override instead of the old get_preferred_width,
get_preferred_height, get_preferred_width_for_height,
get_preferred_height_for_width and
get_preferred_height_and_baseline_for_width.
While porting GTK to GskRenderer we noticed that the current fallback
code for widgets using Cairo to draw is not enough to cover all the
possible cases.
For instance, if a container widget still uses GtkWidget::draw to render
its children, and at least one of them has been ported to using render
nodes instead, the container won't know how to draw it.
For this reason we want to provide to layers above GSK the ability to
create a "fallback" renderer instance, created using a "parent"
GskRenderer instance, but using a Cairo context as the rendering target
instead of a GdkDrawingContext.
GTK will use this inside the gtk_widget_draw() implementation, if a
widget implements GtkWidgetClass.get_render_node().
Change get_render_node to return nodes that are sized to the clip
area and expect to be placed at the clip position; change
gtk_container_propagate_render_node to place child render nodes
accordingly, and change gtk_css_gadget_get_render_node to return
nodes that are sized accordingly as well.
GtkWidget.create_render_node() sets up a GskRenderNode appropriate for
rendering the contents of a widget, including its bounds,
transformation, and anchor point.
The naming is consistent with other scene graph libraries, as it
represents an additional translation transformation applied on top of
the provided transformation matrices.
We can also simplify the implementation by applying the translation when
we compute the world matrix.
We were allocating a surface thats big enough for the clip, and
we were setting the transform for that, but then GtkContainer
was overriding the transform with the one for the allocation.
Also, we were drawing at the clip position, not the allocation
position.
Render nodes need access to rendering information like scaling factors.
If we keep render nodes separate from renderers until we submit a nodes
tree for rendering we're going to have to duplicate all that information
in a way that makes the API more complicated and fuzzier on its
semantics.
By having GskRenderer create GskRenderNode instances we can tie nodes
and renderers together; since higher layers will also have access to
the renderer instance, this does not add any burden to callers.
Additionally, if memory measurements indicate that we are spending too
much time in the allocation of new render nodes, we can now easily
implement a free-list or a renderer-specific allocator without breaking
the API.
The clip rectangle may have non-zero offsets, so we need to ensure that
the GskRenderNode associated to the rendered area is translated by those
same offsets.
We need a virtual function to retrieve the GskRenderNode for each
widget, which is supposed to attach its own children's GskRenderNodes.
Additionally, we want to maintain the existing GtkWidget::draw mechanism
for widgets that do not implement get_render_node() — as well as widgets
that have handlers connected to the ::draw signal.
And replaces its usages in GtkTextView/GtkStyleContext with a hard-coded
0.04 which was the default value for cursor-aspect-ratio. Also remove
the public gtk_draw_insertion_cursor which used draw_insertion_cursor
which in turn looked up cursor-aspect-ratio
And with it, gtk_widget_get_visual() and gtk_widget_set_visual() are
gone.
We now always use the RGBA visual (if available) and otherwise fall back
to the system visual.
This was meant to be silenced unless expicitly requested but
G_ENABLE_DEBUG is defined by default unless --disable-debug is passed to
configure, so use G_ENABLE_CONSISTENCY_CHECKS instead which is only
defined if --enable-debug is explicitly passed.
If somebody decides to use gtk_widget_set_double_buffered() in the
middle of a draw() then there's the risk of calling end_draw_frame()
with an invalid pointer.
Some overeager compilers may warn about the double_buffered bit field
changing values and leading to a potentially uninitialized variable.
In order to avoid compiler warnings or crashes, we can simply store the
value of the double_buffered bit field at the beginning of the rendering
and use that instead of the actual bit field.
https://bugzilla.gnome.org/show_bug.cgi?id=771463
Not all occurrences of this warning can be fixed today, so put it behind
a G_ENABLE_DEBUG flag since it still shows legitimate problems even if
some of them are false positives.
It is important to know whether the returned object can or cannot
change, for a certain widget. For example to connect to the
GtkStyleContext::changed signal.
https://bugzilla.gnome.org/show_bug.cgi?id=769047
Firefox does a bunch of interesting things with GTK.
If the top-level GtkWindow does not have a "csd" style class associated,
Firefox will happily draw the contents of the container used to render
HTML and XUL directly on the top level's GdkWindow; on the other hand,
if a "csd" style class is found, the MozContainer will create a new
child window, and draw on it.
Then, Firefox will proceed to disable double buffering on both the
top-level window and the MozContainer (unless they are backed by the
same GdkWindow, in which case only the top-level will be
single-buffered) *and* it will add a GDK_EXPOSURE_MASK flag to the
MozContainer events for good measure (even if this is only needed for
GTK+ 2.x).
After landing the GdkDrawingContext API in GdkWindow, GTK enabled
automatic double buffering on all top-level windows backed by a native
surface, ad most users of single buffering rely on child widgets instead
of top-levels, and we'd still like to have the same double buffering
behaviour for all top-levels on all backends. Obviously, with Firefox
disabling double buffering on the top-level window, the change broke
their drawing mechanism.
Ideally, Firefox could be fixed to not disable double buffering on the
top-level window when MozContainer has a separate GdkWindow — i.e. the
CSD case — but since we did introduce a slight change of behaviour in
fringe users of the GTK+ API, let's keep backwards compatibility with
the old code for a little while longer, and create an intermediate Cairo
context unbound from the GdkDrawingContext, like we used to do until
GTK+ 3.20.
Instead of associating the GdkWindow that created the GdkDrawingContext
we can directly bind the Cairo context to the GDK drawing context.
Cairo contexts created via gdk_cairo_create() go back to not having a
GdkWindow associated to them, like they did before we introduced the
gdk_window_begin_draw_frame() API.
https://bugzilla.gnome.org/show_bug.cgi?id=766675
Instead of giving out Cairo contexts, GdkWindow should provide a
"drawing context", which can then create Cairo contexts on demand; this
allows us to future proof the API for when we're going to use a
different rendering pipeline, like OpenGL.
https://bugzilla.gnome.org/show_bug.cgi?id=766675
This removes leftover code from when classes where added to the style
context.
Now that they get added directly to css nodes, the classes can exist
without a style context.
https://bugzilla.gnome.org/show_bug.cgi?id=767312
When we emit ::draw, the widget should not have alloc_needed set
anymore. If this happens, it indicates a broken situation. Add a
warning to help tracking down why this might occur.
See https://bugzilla.gnome.org/show_bug.cgi?id=765410
Fix testsuite/a11y/about.ui GtkAboutDialog :
"CRITICAL **: atk_hyperlink_get_start_index: assertion 'ATK_IS_HYPERLINK (link)' failed"
That is set widget->priv->accessible as soon as accessible object is generated.
When accessible object is created accessible->priv->widget is set,
if widget->priv->accessible is not , then _gtk_label_accessible_update_links
exits early, thus without creating the links on the accessible side.
(This as it checks for the widget to have the accessible set before proceeding).
https://bugzilla.gnome.org/show_bug.cgi?id=766458
GTK used to not emit GtkWidget::style-updated on widgets that weren't
realized. This sped up construction of complex widgetry in the early
days of GTK3 where we instantly invalidated on every change.
We don't do that anymore, so in theory (and in my limited testing with
widget-factory) this shouldn't be a prolem anymore.
What is a problem though is that postponing style-updated leads to 2
problems:
(1) Unrealized widgets will not emit style-updated which may cause them
to not properly update their state and return wrong values from
get_preferred_width/height() etc
(2) Emitting style-updated during realize can happen too late.
When a widget is not made child-visible by its parent (common
examples: notebook, paned) it will also not be realized when the
parent is initially shown. However, when they get realized later
(after a resize of the parent), they will emit style-updated (and
potentially queue a resize) during size-allocate.
https://bugzilla.gnome.org/show_bug.cgi?id=765700
This will almost certainly overwritten before the widget gets
to the screen, but while we are doing this, we might as well
use the same state that we initialize the widgets state to.
e8aa9b0440 introduced a new debug mode
that highlights resizes. Unfortunately it has the side effect of
always queueing redraws even when the debug mode is not enabled.
Make the redraw conditional.
When a gtk_widget_queue_allocate() on some widget increases the clip,
widget->parent's clip was not updated. This appraoch naively just
unions widget's new clip with widget->parent's clip.
This of course only works if widget and parent share the same GDK
window. In the cases where they don't we can't do anything and need a
better fix.
Fixes label-text-shadow-changes-modify-clip.ui reftest.
The g_print documentation explicitly says not to do this, since
g_print is meant to be redirected by applications. Instead use
g_message for logging that can be triggered via GTK_DEBUG.
GtkWidget uses gtk_container_foreach() to iterate over children and
check whether they need their allocation reset.
However, that leaves out internal children, such as scrollbars of a
GtkScrolledWindow. Use gtk_container_forall() instead.
Now selecting a widget by class name no longer works.
This is probably most relevant for users outside of GTK that want to
style their own widgets. Those widgets should now either add their own
style classes (if they want to adjust existing CSS) or use
gtk_widget_class_set_css_name() themselves (if they want to get rid of
all "upstream" styling).
Check that non-native window are indeed children of the event window and
only then confirm that they should be drawn.
Fixes Glade thinking that it's okay to have the draw function do
different things depending on what window to draw. (This should really
be fixed in Glade.)
... and remove the also forgotten void function that lingered around
with it.
Fixes opacity=0 parts like inactive spinners or sort indicators in
treeview headers being drawn since last commit.
Oops.
Previously, we had a special cae to draw subwindows of widgets.
This is not necessary as conformant widgets should be able to properly
render themselves when all windows need to be painted.
From now on assume that is the case.
We therefore paint nonnative GDK windows "inline" by just returning TRUE
for gtk_cairo_should_draw_window() for those windows.
This speeds up hilighting different rows in the listbox gtk-demo example
tremendously (by a factor of 10 or more) as the previous code was
O(<number of non-window subwidgets> *
<number of subwindows>) which in the listbox example were ~15,000 and
~2,000 respectively.
If a GtkGestureSingle is set as touch-only, pointer events would be
discarded without giving an opportunity to the regular GtkGesture
handler to manage those.
Because the pointer events weren't actually managed by the gesture,
gtk_gesture_get_sequence_state() (rather unhelpfully here) will resort
to returning GTK_EVENT_SEQUENCE_NONE, which is in turn interpreted
by _gtk_widget_consumes_motion() as "may be handling the events for
this sequence", because gestures in this state presumably handle
the events, just that it's not "claimed" yet.
Instead, use gtk_gesture_handles_sequence(), which will perform the
expected check on the event sequence being managed, as we expect
here.
In https://bugzilla.gnome.org/show_bug.cgi?id=601425 the annotations
were changed to int as they not only take the predefined enum values
but also user defined values registered through gtk_icon_size_register()
As a result the typelib doesn't contain any information about
GtkIconSize for those arguments and the Python docstring only
shows the corresponding Python type "int".
This changes the argument docs to mention the type explicitly
so the Python doc generator can add a link to Gtk.IconSize
which contains the most useful predefined values.
https://bugzilla.gnome.org/show_bug.cgi?id=757411
Previously, the ID was only set on the CSS node as a side-effect
of calling gtk_widget_get_style_context. This was showing up
in CSS style tests as nodes lacking their IDs.
gdk_widget_get_frame_clock can return NULL. In particular,
this can happen when the drag window is destroyed at the end
of a DND operation. Handle this gracefully when it happens.
