Merge branch 'drawing-model-refresh' into 'master'

doc: Rewrite the drawing model overview

See merge request GNOME/gtk!603
This commit is contained in:
Matthias Clasen 2019-02-23 21:05:35 +00:00
commit fad9468e77

View File

@ -26,56 +26,42 @@
widgets and windows, you should read this chapter; this will be
useful to know if you decide to implement your own widgets. This
chapter will also clarify the reasons behind the ways certain
things are done in GTK; for example, why you cannot change the
background color of all widgets with the same method.
things are done in GTK.
</para>
<refsect2 id="drawing model windows">
<title>Windows and events</title>
<para>
Programs that run in a windowing system generally create
rectangular regions in the screen called
<firstterm>windows</firstterm>. Traditional windowing systems
do not automatically save the graphical content of windows, and
instead ask client programs to repaint those windows whenever it
is needed. For example, if a window that is stacked below other
windows gets raised to the top, then a client program has to
repaint the area that was previously obscured. When the
windowing system asks a client program to redraw part of a
window, it sends an <firstterm>exposure event</firstterm> to the
program for that window.
Applications that use a windowing system generally create
rectangular regions in the screen called <firstterm>surfaces</firstterm>
(GTK is following the Wayland terminology, other windowing systems
such as X11 may call these <firstterm>windows</firstterm>).
Traditional windowing systems do not automatically save the
graphical content of surfaces, and instead ask applications to
provide new content whenever it is needed.
For example, if a window that is stacked below other
windows gets raised to the top, then the application has to
repaint it, so the previously obscured area can be shown.
When the windowing system asks an application to redraw
a window, it sends an <firstterm>frame event</firstterm>
(<firstterm>expose event</firstterm> in X11 terminology)
for that window.
</para>
<para>
Each GTK toplevel window or dialog is associated with a
windowing system surface. Child widgets such as buttons or
entries don't have their own surface; they use the surface
of their toplevel.
</para>
<para>
Here, "windows" means "rectangular regions with automatic
clipping", instead of "toplevel application windows". Most
windowing systems support nested windows, where the contents of
child windows get clipped by the boundaries of their parents.
Although GTK and GDK in particular may run on a windowing
system with no such notion of nested windows, GDK presents the
illusion of being under such a system. A toplevel window may
contain many subwindows and sub-subwindows, for example, one for
the menu bar, one for the document area, one for each scrollbar,
and one for the status bar. In addition, controls that receive
user input, such as clickable buttons, are likely to have their
own subwindows as well.
</para>
<para>
In practice, most windows in modern GTK application are client-side
constructs. Only few windows (in particular toplevel windows) are
<emphasis>native</emphasis>, which means that they represent a
window from the underlying windowing system on which GTK is running.
For example, on X11 it corresponds to a <type>Window</type>; on Win32,
it corresponds to a <type>HANDLE</type>.
</para>
<para>
Generally, the drawing cycle begins when GTK receives an
exposure event from the underlying windowing system: if the
Generally, the drawing cycle begins when GTK receives
a frame event from the underlying windowing system: if the
user drags a window over another one, the windowing system will
tell the underlying window that it needs to repaint itself. The
tell the underlying surface that it needs to repaint itself. The
drawing cycle can also be initiated when a widget itself decides
that it needs to update its display. For example, when the user
types a character in a <link
@ -85,13 +71,10 @@
</para>
<para>
The windowing system generates events for native windows. The GDK
interface to the windowing system translates such native events into
<link linkend="GdkEvent"><structname>GdkEvent</structname></link>
structures and sends them on to the GTK layer. In turn, the GTK layer
finds the widget that corresponds to a particular
<classname>GdkSurface</classname> and emits the corresponding event
signals on that widget.
The windowing system generates frame events for surfaces. The GDK
interface to the windowing system translates such events into
emissions of the #GtkSurface::render signal on the affected surfaces.
The GTK toplevel window connects to that signal, and reacts appropriately.
</para>
<para>
@ -112,8 +95,13 @@
it does. On top of this GTK has a frame clock that gives a
“pulse” to the application. This clock beats at a steady rate,
which is tied to the framerate of the output (this is synced to
the monitor via the window manager/compositor). The clock has
several phases:
the monitor via the window manager/compositor). A typical
refresh rate is 60 frames per second, so a new “pulse” happens
roughly every 16 milliseconds.
</para>
<para>
The clock has several phases:
<itemizedlist>
<listitem><para>Events</para></listitem>
<listitem><para>Update</para></listitem>
@ -125,24 +113,24 @@
</para>
<para>
The Events phase is a long stretch of time between each
redraw where we get input events from the user and other events
The Events phase is a stretch of time between each redraw where
GTK processes input events from the user and other events
(like e.g. network I/O). Some events, like mouse motion are
compressed so that we only get a single mouse motion event per
clock cycle.
compressed so that only a single mouse motion event per clock
cycle needs to be handled.
</para>
<para>
Once the Events phase is over we pause all external events and
run the redraw loop. First is the Update phase, where all
Once the Events phase is over, external events are paused and
the redraw loop is run. First is the Update phase, where all
animations are run to calculate the new state based on the
estimated time the next frame will be visible (available via
the frame clock). This often involves geometry changes which
drives the next phase, Layout. If there are any changes in
widget size requirements we calculate a new layout for the
widget hierarchy (i.e. we assign sizes and positions). Then
we go to the Paint phase where we redraw the regions of the
window that need redrawing.
drive the next phase, Layout. If there are any changes in
widget size requirements the new layout is calculated for the
widget hierarchy (i.e. sizes and positions for all widgets are
determined). Then comes the Paint phase, where we redraw the
regions of the window that need redrawing.
</para>
<para>
@ -184,162 +172,57 @@
</para>
</refsect2>
<refsect2 id="scene-graph">
<title>The scene graph</title>
<para>
The first step in “drawing” a window is that GTK creates
<firstterm>render nodes</firstterm> for all the widgets
in the window. The render nodes are combined into a tree
that you can think of as a <firstterm>scene graph</firstterm>
describing your window contents.
</para>
<para>
Render nodes belong to the GSK layer, and there are various kinds
of them, for the various kinds of drawing primitives you are likely
to need when translating widget content and CSS styling. Typical
examples are text nodes, gradient nodes, texture nodes or clip nodes.
<para>
<para>
In the past, all drawing in GTK happened via cairo. It is still possible
to use cairo for drawing your custom widget contents, by using a cairo
render node.
</para>
</para>
A GSK <firstterm>renderer</firstterm> takes these render nodes, transforms
them into rendering commands for the drawing API it targets, and arranges
for the resulting drawing to be associated with the right surface. GSK has
renderers for OpenGL, Vulkan and cairo.
</para>
</refsect2>
<refsect2 id="hierarchical-drawing">
<title>Hierarchical drawing</title>
<para>
During the Paint phase we will send a single expose event to
the toplevel window. The event handler will create a cairo
context for the window and emit a GtkWidget::draw() signal
on it, which will propagate down the entire widget hierarchy
in back-to-front order, using the clipping and transform of
the cairo context. This lets each widget draw its content at
the right place and time, correctly handling things like
partial transparencies and overlapping widgets.
During the Paint phase we will send a single ::render signal the toplevel
window. The signal handler will create a snapshot object (which is a
helper for creating a scene graph) and emit a GtkWidget::snapshot() signal,
which will propagate down the entire widget hierarchy. This lets each widget
snapshot its content at the right place and time, correctly handling things
like partial transparencies and overlapping widgets.
</para>
<para>
When generating the event, GDK also sets up double buffering to
avoid the flickering that would result from each widget drawing
itself in turn. <xref linkend="double-buffering"/> describes
the double buffering mechanism in detail.
</para>
<para>
Normally, there is only a single cairo context which is used in
the entire repaint, rather than one per GdkSurface. This means you
have to respect (and not reset) existing clip and transformations
set on it.
</para>
<para>
Most widgets, including those that create their own GdkSurfaces have
a transparent background, so they draw on top of whatever widgets
are below them. This was not the case in GTK 2 where the theme set
the background of most widgets to the default background color. (In
fact, transparent GdkSurfaces used to be impossible.)
</para>
<para>
The whole rendering hierarchy is captured in the call stack, rather
than having multiple separate draw emissions, so you can use effects
like e.g. cairo_push/pop_group() which will affect all the widgets
below you in the hierarchy. This makes it possible to have e.g.
partially transparent containers.
To avoid excessive work when generating scene graphs, GTK caches render nodes.
GtkWidget keeps a reference to its render node (which in turn, will refer to
the render nodes of children, and grandchildren, and so on), and will reuse
that node during the Paint phase. Invalidating a widget (e.g. by calling
gtk_widget_queue_draw) discards the cached render node, forcing GTK to
regenerate it the next time it needs to snapshot the widget.
</para>
</refsect2>
<refsect2 id="scrolling drawing model">
<title>Scrolling</title>
<para>
Traditionally, GTK has used self-copy operations to implement
scrolling with native windows. With transparent backgrounds, this
no longer works. Instead, we just mark the entire affected area for
repainting when these operations are used. This allows (partially)
transparent backgrounds, and it also more closely models modern
hardware where self-copy operations are problematic (they break the
rendering pipeline).
</para>
</refsect2>
</refsect1>
<refsect1 id="double-buffering">
<title>Double buffering</title>
<para>
If each of the drawing calls made by each subwidget's
<literal>draw</literal> handler were sent directly to the
windowing system, flicker could result. This is because areas may get
redrawn repeatedly: the background, then decorative frames, then text
labels, etc. To avoid flicker, GTK employs a <firstterm>double
buffering</firstterm> system at the GDK level. Widgets normally don't
know that they are drawing to an off-screen buffer; they just issue their
normal drawing commands, and the buffer gets sent to the windowing system
when all drawing operations are done.
</para>
<para>
Two basic functions in GDK form the core of the double-buffering
mechanism: <link
linkend="gdk_surface_begin_paint_region"><function>gdk_surface_begin_paint_region()</function></link>
and <link
linkend="gdk_surface_end_paint"><function>gdk_surface_end_paint()</function></link>.
The first function tells a <classname>GdkSurface</classname> to
create a temporary off-screen buffer for drawing. All
subsequent drawing operations to this window get automatically
redirected to that buffer. The second function actually paints
the buffer onto the on-screen window, and frees the buffer.
</para>
<refsect2 id="automatic-double-buffering">
<title>Automatic double buffering</title>
<para>
It would be inconvenient for all widgets to call
<function>gdk_surface_begin_paint_region()</function> and
<function>gdk_surface_end_paint()</function> at the beginning
and end of their draw handlers.
</para>
<para>
To make this easier, GTK normally calls
<function>gdk_surface_begin_paint_region()</function>
before emitting the #GtkWidget::draw signal, and
then it calls <function>gdk_surface_end_paint()</function>
after the signal has been emitted. This is convenient for
most widgets, as they do not need to worry about creating
their own temporary drawing buffers or about calling those
functions.
</para>
<para>
However, some widgets may prefer to disable this kind of
automatic double buffering and do things on their own.
To do this, call the
<function>gtk_widget_set_double_buffered()</function>
function in your widget's constructor. Double buffering
can only be turned off for widgets that have a native
window.
</para>
<example id="disabling-double-buffering">
<title>Disabling automatic double buffering</title>
<programlisting>
static void
my_widget_init (MyWidget *widget)
{
...
gtk_widget_set_double_buffered (widget, FALSE);
...
}
</programlisting>
</example>
<para>
When is it convenient to disable double buffering? Generally,
this is the case only if your widget gets drawn in such a way
that the different drawing operations do not overlap each
other. For example, this may be the case for a simple image
viewer: it can just draw the image in a single operation.
This would <emphasis>not</emphasis> be the case with a word
processor, since it will need to draw and over-draw the page's
background, then the background for highlighted text, and then
the text itself.
</para>
<para>
Even if you turn off double buffering on a widget, you
can still call
<function>gdk_surface_begin_paint_region()</function> and
<function>gdk_surface_end_paint()</function> by hand to use
temporary drawing buffers.
</para>
</refsect2>
</refsect1>
</refentry>