The GTK Input Model 3 GTK Library The GTK Input Model input and event handling in detail 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. 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 master and slave 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 slave device is associated with a virtual master device. Master devices always come in pointer/keyboard pairs - you can think of such a pair as a 'seat'. GTK widgets generally deal with the master 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 surface - for pointer events, the surface under the pointer (grabs complicate this), for keyboard events, the surface with the keyboard focus. GDK translates these raw windowing system events into #GdkEvents. Typical input events are: #GdkEventButton #GdkEventMotion #GdkEventCrossing #GdkEventKey #GdkEventFocus #GdkEventTouch Additionally, GDK/GTK synthesizes other signals to let know whether grabs (system-wide or in-app) are taking input away: #GdkEventGrabBroken #GtkWidget::grab-notify When GTK creates a GdkSurface, it connects to the ::event signal on it, which receives all of these input events. Surfaces have have signals and properties, e.g. to deal with window management related events. The function which initially receives input events on the GTK side is 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 #GdkSurface, 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 #GtkRoot which owns the #GdkSurface 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 #GdkSurface. 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 goes down and up the widget hierarchy in three phases (see #GtkPropagationPhase) towards a target widget. For key events, the top-level window gets a first shot at activating mnemonics and accelerators. If that does not consume the events, the target widget for event propagation is window's current focus widget (see gtk_window_get_focus()). For pointer events, the target widget is determined by picking the widget at the events coordinates (see gtk_window_pick()). In the first phase (the “capture” phase) the event is delivered to each widget from the top-most (the top-level #GtkWindow or grab widget) down to the target #GtkWidget. Event controllers 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 event controllers attached to it with %GTK_PHASE_TARGET. This is known as the “target” phase, and only happens on that widget. In the last phase (the “bubble” phase), the event is delivered to each widget from the target to the top-most, and event controllers attached with %GTK_PHASE_BUBBLE are run. Events are not delivered to a widget which is insensitive or unmapped. Any time during the propagation phase, a controller may indicate that a received event was consumed and propagation should therefore be stopped. If gestures are used, this may happen when the gesture claims the event touch sequence (or the pointer events) for its own. See the “gesture states” section below to learn more about gestures and sequences. 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. 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 #GdkEventButtonPress happens, all events from then on, until after the corresponding #GdkEventButtonRelease, 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. Every #GtkWindow maintains a single focus location (in the ::focus-widget property). The focus widget is the target widget for key events sent to the window. Only widgets which have ::can-focus set to %TRUE can become the focus. Typically these are input controls such as entries or text fields, but e.g. buttons can take the focus too. Input widgets can be given the focus by clicking on them, but focus can also be moved around with certain key events (this is known as “keyboard navigation”). GTK reserves the Tab key to move the focus to the next location, and Shift-Tab to move it back to the previous one. In addition many containers allow “directional navigation” with the arrow keys. 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 gesture 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. Since gestures are “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 recognition 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. 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.