Gestures

libinput supports Pinch gestures and Swipe gestures on most modern touchpads and other indirect touch devices. Note that libinput does not support gestures on touchscreens, see Touchscreen gestures.

Lifetime of a gesture

A gesture starts when the finger position and/or finger motion is unambiguous as to what gesture to trigger and continues until the first finger belonging to this gesture is lifted.

A single gesture cannot change the finger count. For example, if a user puts down a fourth finger during a three-finger swipe gesture, libinput will end (cancel) the three-finger gesture and, if applicable, start a four-finger swipe gesture. A caller may however decide that those gestures are semantically identical and continue the two gestures as one single gesture.

Pinch gestures

Pinch gestures are executed when two or more fingers are located on the touchpad and are either changing the relative distance to each other (pinching) or are changing the relative angle (rotate). Pinch gestures may change both rotation and distance at the same time. For such gestures, libinput calculates a logical center for the gestures and provides the caller with the delta x/y coordinates of that center, the relative angle of the fingers compared to the previous event, and the absolute scale compared to the initial finger position.

_images/pinch-gestures.svg

The pinch and rotate gestures

The illustration above shows a basic pinch in the left image and a rotate in the right angle. Not shown is a movement of the logical center if the fingers move unevenly. Such a movement is supported by libinput, it is merely left out of the illustration.

Note that while position and angle is relative to the previous event, the scale is always absolute and a multiplier of the initial finger position’s scale.

Swipe gestures

Swipe gestures are executed when three or more fingers are moved synchronously in the same direction. libinput provides x and y coordinates in the gesture and thus allows swipe gestures in any direction, including the tracing of complex paths. It is up to the caller to interpret the gesture into an action or limit a gesture to specific directions only.

_images/swipe-gestures.svg

The swipe gestures

The illustration above shows a vertical three-finger swipe. The coordinates provided during the gesture are the movements of the logical center.

Hold gestures

A hold gesture is one where the user places one or more fingers on the device without significant movement. The exact conditions when a hold gesture transitions to pointer motion, scrolling or other gestures are implementation-defined.

The hold gesture is intended to allow for the implementation of two specific features:

  • where a two-finger scrolling starts kinetic scrolling in the caller, a subsequent hold gesture can be used to stop that kinetic scroll motion, and

  • hold-to-trigger interactions where the interaction could be a click, a context menu, or some other context-specific interaction.

Hold gestures have three potential logical states:

  • begin: one or more fingers are placed on the device at the same time

  • end: all fingers are removed and the device enters a neutral logical state

  • end(cancelled): all fingers are part of a known interaction and the currenthold gesture is no longer active. This may also occurs when switching between hold gestures with different finger counts.

Note

By definition, a hold gesture does not move and thus no coordinate updates are available.

For example, a user that puts one finger, then a second finger down and releases them later may trigger the following event sequence:

Action

Event

Finger count

Finger 1 down

<no event>

Finger 2 down

begin

2

Finger 2 up

end

2

Finger 1 up

<no event>

A hold gesture may by be cancelled. This occurs when the hold gesture changes into some other interaction and should no longer be considered the current hold gesture. A end(cancelled) event applies to the whole gesture (all fingers). For example, a pointer motion on a touchpad may trigger this sequence:

Action

Event

Finger 1 down
hold begin
Finger 1 motion

hold cancel
pointer motion
Finger 1 motion
pointer motion
Finger 1 up
no event

Note

Many interactions with a touchpad will start with a hold gesture that is then cancelled as that gesture turns into e.g. pointer motion. A caller must handle hold gesture cancellations correctly.

A two-finger scroll motion on a touchpad may trigger this sequence:

Action

Event

Finger count

Finger 1 down
hold begin
1
Finger 2 down

hold cancel
hold begin
1
2
Finger 1+2 motion

hold cancel
pointer axis
2

Finger 1+2 motion
pointer axis
Finger 1 up
Finger 2 up
pointer axis
(scroll stop)

A three-finger-swipe on a touchpad may trigger this sequence:

Action

Event

Finger count

Finger 1 down
hold begin
1
Finger 2 down

hold cancel
hold begin
1
2
Finger 3 down

hold cancel
hold begin
2
3
Finger motion

hold cancel
swipe begin
3
3
Finger motion
swipe update
3
Finger 1 up
Finger 2 up
Finger 3 up
swipe end


3


Single-finger hold gestures

libinput uses implementation-defined timeouts based on other interactions to determine whether a single-finger hold gestures should start. In other words, a caller must not rely on a hold gesture always being triggered as soon as a single finger is placed on the touchpad. This is true for any hold gesture but especially so for single-finger hold gestures.

Hold gestures with a single finger are prone to being extremely short-lived. On many devices it is impossible to hold a finger still enough for there to be no pointer motion events, even if those deltas are miniscule. Changing movement thresholds to rely on hold gestures would reduce device responsiveness.

It is thus the responsibility of the caller to determine where hold gestures transition in and out of other interactions. For example, a two-finger hold may produce a cancelled single-finger hold gesture first:

Action

Event

Finger count

Notes

Finger 1 down
hold begin
1
Finger 1 motion

hold cancel
pointer motion
1

tiny deltas

Finger 2 down
hold begin
2
Finger 1 up
Finger 2 up
hold end



Note how the second hold gesture started with a finger count of 2 - without the user ever lifting the first finger. Cancellation of hold gesture does not imply the user has lifted a finger.

A hold gesture may start after a previous gesture completed. For example, a single finger move-and-hold may trigger different sequences for the same user interaction:

Action

Device 1

Device 2

Notes

Finger 1 down
hold begin
hold begin
Finger 1 motion
hold cancel
pointer motion
tiny deltas

hold begin
Finger 1 up
hold end
hold end

A caller that wants to use hold gestures must thus be able to infer the same interaction based on a stream of pointer motion events with small deltas.

libinput may start a new hold begin gesture once the pointer stops moving. The time between the last pointer motion event and the hold begin event is implementation-defined.

Hold gestures and thumb/palm detection

Thumb and palm detection effectively remove touches from being counted towards an interaction, see Thumb detection and Palm detection for details.

In the context of hold gestures, thumbs and palms are treated by libinput as if the finger was removed from the device. Where other non-thumb/non-palm fingers remain on the device, an hold update event is sent. Otherwise, the hold gesture terminates with a hold cancel event.

Notably, libinput’s thumb and palm detection is not a simple boolean per touch but specific to the state of that touch in the overall context. For example, a touch may be a thumb for tapping but not for clickfinger interactions. A caller must not infer the number of physical fingers from the hold gesture.

Likewise, libinput may classify a finger as thumb in the same hardware event as a new finger is placed on the touchpad. In that case, the hold gesture may continue as one-finger gesture despite there being two physical touch points.

Information to determine whether a touch is a thumb or a palm may not be available until some time into an interaction. Thus very short brushes of the touchpad by a palm may trigger a hold begin followed by an immediate hold end as libinput lacks sufficient information to identify the touch as thumb/palm and send the corresponding hold cancel event. A caller must not assume that a hold gesture always represents a valid finger down.

Hold gestures and tap-to-click

Tap-to-click behaviour is the feature that enables short-lived touches to trigger button presses.

Warning

Summary: do not use hold gestures to do your own tap-to-click implementation

In the context of hold gestures, tap-to-click cancels current hold gestures and a finger dragging (see Tap-and-drag) does not begin a hold gesture. Where tap-to-click is disabled a tap-like gesture may create hold begin followed by a hold end event. Callers must not use hold gestures for their own tap-to-click implementation as the data is not reliable enough. libinput may change internal timeouts and thresholds depending on whether tap-to-click is enabled and the hold gesture event may not match touch sequences that a user would expect to be a tap-to-click interaction.

Touchscreen gestures

Touchscreen gestures are not interpreted by libinput. Rather, any touch point is passed to the caller and any interpretation of gestures is up to the caller or, eventually, the X or Wayland client.

Interpreting gestures on a touchscreen requires context that libinput does not have, such as the location of windows and other virtual objects on the screen as well as the context of those virtual objects:

_images/touchscreen-gestures.svg

Context-sensitivity of touchscreen gestures

In the above example, the finger movements are identical but in the left case both fingers are located within the same window, thus suggesting an attempt to zoom. In the right case both fingers are located on a window border, thus suggesting a window movement. libinput has no knowledge of the window coordinates and thus cannot differentiate the two.

Gestures with enabled software buttons

If the touchpad device is a Clickpad, it is recommended that a caller switches to Clickfinger behavior. Usually fingers placed in a software button area are not considered for gestures, resulting in some gestures to be interpreted as pointer motion or two-finger scroll events.

_images/pinch-gestures-softbuttons.svg

Interference of software buttons and pinch gestures

In the example above, the software button area is highlighted in red. The user executes a three-finger pinch gesture, with the thumb remaining in the software button area. libinput ignores fingers within the software button areas, the movement of the remaining fingers is thus interpreted as a two-finger scroll motion.

Gestures on two-finger touchpads

As of kernel 4.2, many Partial multi-touch touchpads provide only two slots. This affects how gestures can be interpreted. Touchpads with only two slots can identify two touches by position but can usually tell that there is a third (or fourth) finger down on the touchpad - without providing positional information for that finger.

Touchpoints are assigned in sequential order and only the first two touch points are trackable. For libinput this produces an ambiguity where it is impossible to detect whether a gesture is a pinch gesture or a swipe gesture whenever a user puts the index and middle finger down first. Since the third finger does not have positional information, it’s location cannot be determined.

_images/gesture-2fg-ambiguity.svg

Ambiguity of three-finger gestures on two-finger touchpads

The image above illustrates this ambiguity. The index and middle finger are set down first, the data stream from both finger positions looks identical. In this case, libinput assumes the fingers are in a horizontal arrangement (the right image above) and use a swipe gesture.