Multi-touch

One finger is unambiguous: it presses, it moves, it lifts. A second contact changes the question being asked. A surface with two fingers on it has to work out whether they are travelling together, spreading apart, turning about a shared midpoint, or whether one of them is simply resting while the other does the work. That decision is what multi-touch means, and it makes gesture recognition mostly bookkeeping: every contact arrives with an identity of its own, and the gesture lives in the relationship between them rather than in any single one.

Two fingers carry most of the vocabulary. Pinch open and pinch closed scale content by the distance between the contacts, which is the mechanism behind pinch to zoom; the rotate gesture turns content by the angle of the line the two contacts span; and a two-finger drag scrolls or pans without being mistaken for a selection or a drag of the content itself. Above two, the gestures usually belong to the system rather than to the application: three-finger swipes move between spaces or undo, and four or five finger pinches on a tablet go home. The finger count is part of the name, so an instruction that says swipe when it means two-finger swipe will be followed wrongly by nearly everyone.

On the web the contacts arrive as pointer events, each carrying a pointerId that stays stable from pointerdown through to pointerup, so tracking a gesture comes down to keeping a small map of live pointers and recomputing the distance and angle whenever one of them moves. Whether those events arrive at all is decided by touch-action. The browser’s own panning and zooming are claimed before your handler runs and cannot be taken back afterwards, so a surface that wants a two-finger gesture of its own has to declare touch-action: none, or pan-y to keep vertical scrolling and claim the rest, in the stylesheet rather than in the handler.

The conflict with the platform is unavoidable, so it is worth designing rather than discovering. A map inside a scrolling page that swallows one-finger drags strands the reader halfway down the article, which is why the usual settlement gives the page one finger and the map two, and says so on the surface. Anything the operating system has claimed, an edge swipe or a system-wide three-finger gesture, is taken before your code sees it at all. And multi-touch is a capability rather than a guarantee: a mouse has one pointer, a trackpad synthesizes some gestures and not others, and someone using a switch, a head pointer, or one hand has a single contact at most. Every pinch needs zoom controls behind it, and every two-finger scroll needs a scrollbar.