The GTK Input and Event Handling Model

Overview of GTK input and event handling

This chapter describes in detail how GTK handles input. If you are interested in what happens to translate a key press or mouse motion of the users into a change of a GTK widget, you should read this chapter. This knowledge will also be useful if you decide to implement your own widgets.

Devices and events

The most basic input devices that every computer user has interacted with are keyboards and mice; beyond these, GTK supports touchpads, touchscreens and more exotic input devices such as graphics tablets. Inside GTK, every such input device is represented by a GdkDevice object.

To simplify dealing with the variability between these input devices, GTK has a concept of logical and physical devices. The concrete physical devices that have many different characteristics (mice may have 2 or 3 or 8 buttons, keyboards have different layouts and may or may not have a separate number block, etc) are represented as “slave” devices. Each physical device is associated with a virtual logical device. Logical devices always come in pointer/keyboard pairs - you can think of such a pair as a ‘seat’.

GTK widgets generally deal with the logical devices, and thus can be used with any pointing device or keyboard.

When a user interacts with an input device (e.g. moves a mouse or presses a key on the keyboard), GTK receives events from the windowing system. These are typically directed at a specific window - for pointer events, the window under the pointer (grabs complicate this), for keyboard events, the window with the keyboard focus.

GDK translates these raw windowing system events into GdkEvents. Typical input events are:

Additionally, GDK/GTK synthesizes other signals to let know whether grabs (system-wide or in-app) are taking input away:

When GTK is initialized, it sets up an event handler function with gdk_event_handler_set(), which receives all of these input events (as well as others, for instance window management related events).

Event propagation

For widgets which have a GdkWindow set, events are received from the windowing system and passed to gtk_main_do_event(). See its documentation for details of what it does: compression of enter/leave events, identification of the widget receiving the event, pushing the event onto a stack for gtk_get_current_event(), and propagating the event to the widget.

When a GDK backend produces an input event, it is tied to a GdkDevice and a GdkWindow, which in turn represents a windowing system surface in the backend. If a widget has grabbed the current input device, or all input devices, the event is propagated to that GtkWidget. Otherwise, it is propagated to the the GtkWidget which called gtk_widget_register_window() on the GdkWindow receiving the event.

Grabs are implemented for each input device, and globally. A grab for a specific input device (gtk_device_grab_add()), is sent events in preference to a global grab (gtk_grab_add()). Input grabs only have effect within the GtkWindowGroup containing the GtkWidget which registered the event’s GdkWindow. If this GtkWidget is a child of the grab widget, the event is propagated to the child — this is the basis for propagating events within modal dialogs.

An event is propagated to a widget using gtk_propagate_event(). Propagation differs between event types: key events (GDK_KEY_PRESS, GDK_KEY_RELEASE) are delivered to the top-level GtkWindow; other events are propagated down and up the widget hierarchy in three phases (see GtkPropagationPhase).

For key events, the top-level window’s default GtkWidget::key-press-event and GtkWidget::key-release-event signal handlers handle mnemonics and accelerators first. Other key presses are then passed to gtk_window_propagate_key_event() which propagates the event upwards from the window’s current focus widget (gtk_window_get_focus()) to the top-level.

For other events, in the first phase (the “capture” phase) the event is delivered to each widget from the top-most (for example, the top-level GtkWindow or grab widget) down to the target GtkWidget. Gestures that are attached with GTK_PHASE_CAPTURE get a chance to react to the event.

After the “capture” phase, the widget that was intended to be the destination of the event will run gestures attached to it with GTK_PHASE_TARGET. This is known as the “target” phase, and only happens on that widget.

Next, the “event” signal is emitted, then the appropriate signal for the event in question, for example GtkWidget::motion-notify-event. Handling these signals was the primary way to handle input in GTK widgets before gestures were introduced. If the widget is realized, the GtkWidget::event-after signal is emitted. The signals are emitted from the target widget up to the top-level, as part of the “bubble” phase.

The default handlers for the event signals send the event to gestures that are attached with GTK_PHASE_BUBBLE. Therefore, gestures in the “bubble” phase are only used if the widget does not have its own event handlers, or takes care to chain up to the default GtkWidget handlers.

Events are not delivered to a widget which is insensitive or unmapped.

Any time during the propagation phase, a widget may indicate that a received event was consumed and propagation should therefore be stopped. In traditional event handlers, this is hinted by returning GDK_EVENT_STOP. If gestures are used, this may happen when the widget tells the gesture to claim the event touch sequence (or the pointer events) for its own. See the “gesture states” section below to know more of the latter.

Event masks

Each widget instance has a basic event mask and another per input device, which determine the types of input event it receives. Each event mask set on a widget is added to the corresponding (basic or per-device) event mask for the widget’s GdkWindow, and all child GdkWindows.

If a widget is windowless (gtk_widget_get_has_window() returns FALSE) and an application wants to receive custom events on it, it must be placed inside a GtkEventBox to receive the events, and an appropriate event mask must be set on the box. When implementing a widget, use a GDK_INPUT_ONLY GdkWindow to receive the events instead.

Filtering events against event masks happens inside GdkWindow, which exposes event masks to the windowing system to reduce the number of events GDK receives from it. On receiving an event, it is filtered against the GdkWindow’s mask for the input device, if set. Otherwise, it is filtered against the GdkWindow’s basic event mask.

This means that widgets must add to the event mask for each event type they expect to receive, using gtk_widget_set_events() or gtk_widget_add_events() to preserve the existing mask. Widgets which are aware of floating devices should use gtk_widget_set_device_events() or gtk_widget_add_device_events(), and must explicitly enable the device using gtk_widget_set_device_enabled(). See the GdkDeviceManager documentation for more information.

All standard widgets set the event mask for all events they expect to receive, and it is not necessary to modify this. Masks should be set when implementing a new widget.

Touch events

Touch events are emitted as events of type GDK_TOUCH_BEGIN, GDK_TOUCH_UPDATE or GDK_TOUCH_END, those events contain an “event sequence” that univocally identifies the physical touch until it is lifted from the device.

On some windowing platforms, multitouch devices perform pointer emulation, this works by granting a “pointer emulating” hint to one of the currently interacting touch sequences, which will be reported on every GdkEventTouch event from that sequence. By default, if a widget didn’t request touch events by setting GDK_TOUCH_MASK on its event mask and didn’t override GtkWidget::touch-event, GTK will transform these “pointer emulating” events into semantically similar GdkEventButton and GdkEventMotion events. Depending on GDK_TOUCH_MASK being in the event mask or not, non-pointer-emulating sequences could still trigger gestures or just get filtered out, regardless of the widget not handling those directly.

If the widget sets GDK_TOUCH_MASK on its event mask and doesn’t chain up on GtkWidget::touch-event, only touch events will be received, and no pointer emulation will be performed.

Grabs

Grabs are a method to claim all input events from a device, they happen either implicitly on pointer and touch devices, or explicitly. Implicit grabs happen on user interaction, when a GDK_BUTTON_PRESS event type happens, all events from then on, until after the corresponding GDK_BUTTON_RELEASE event type, will be reported to the widget that got the first event. Likewise, on touch events, every GdkEventSequence will deliver only events to the widget that received its GDK_TOUCH_BEGIN event.

Explicit grabs happen programatically (both activation and deactivation), and can be either system-wide (GDK grabs) or application-wide (GTK grabs). On the windowing platforms that support it, GDK grabs will prevent any interaction with any other application/window/widget than the grabbing one, whereas GTK grabs will be effective only within the application (across all its windows), still allowing for interaction with other applications.

But one important aspect of grabs is that they may potentially happen at any point somewhere else, even while the pointer/touch device is already grabbed. This makes it necessary for widgets to handle the cancellation of any ongoing interaction. Depending on whether a GTK or GDK grab is causing this, the widget will respectively receive a GtkWidget::grab-notify signal, or a GdkEventGrabBroken event.

On gestures, these signals are handled automatically, causing the gesture to cancel all tracked pointer/touch events, and signal the end of recognition.

Keyboard input

Event controllers and gestures

Event controllers are standalone objects that can perform specific actions upon received GdkEvents. These are tied to a GtkWidget, and can be told of the event propagation phase at which they will manage the events.

Gestures are a set of specific controllers that are prepared to handle pointer and/or touch events, each gestures implementation attempts to recognize specific actions out the received events, notifying of the state/progress accordingly to let the widget react to those. On multi-touch gestures, every interacting touch sequence will be tracked independently.

Being gestures “simple” units, it is not uncommon to tie several together to perform higher level actions, grouped gestures handle the same event sequences simultaneously, and those sequences share a same state across all grouped gestures. Some examples of grouping may be:

  • A “drag” and a “swipe” gestures may want grouping. The former will report events as the dragging happens, the latter will tell the swipe X/Y velocities only after gesture has finished.
  • Grouping a “drag” gesture with a “pan” gesture will only effectively allow dragging in the panning orientation, as both gestures share state.

If “press” and “long press” are wanted simultaneously, those would need grouping.

Gesture states

Gestures have a notion of “state” for each individual touch sequence. When events from a touch sequence are first received, the touch sequence will have “none” state, this means the touch sequence is being handled by the gesture to possibly trigger actions, but the event propagation will not be stopped.

When the gesture enters recognition, or at a later point in time, the widget may choose to claim the touch sequences (individually or as a group), hence stopping event propagation after the event is run through every gesture in that widget and propagation phase. Anytime this happens, the touch sequences are cancelled downwards the propagation chain, to let these know that no further events will be sent.

Alternatively, or at a later point in time, the widget may choose to deny the touch sequences, thus letting those go through again in event propagation. When this happens in the capture phase, and if there are no other claiming gestures in the widget, a GDK_TOUCH_BEGIN/GDK_BUTTON_PRESS event will be emulated and propagated downwards, in order to preserve consistency.

Grouped gestures always share the same state for a given touch sequence, so setting the state on one does transfer the state to the others. They also are mutually exclusive, within a widget there may be only one gesture group claiming a given sequence. If another gesture group claims later that same sequence, the first group will deny the sequence.