TiltIndication

Metro's press feedback, in the slot Material fills with a ripple — and, since B-40, the keyboard focus ring that goes in the same slot.

Both live here for the reason the tilt does: LocalIndication is one place, and a component that has to remember to apply something is a component that will one day forget. Every clickable/toggleable/selectable in the library got the ring the moment this drew it, and nothing had to be edited to receive it. See drawKvadrantFocusRing for what is drawn and TiltNode.draw for when.

The plane leans towards the finger and sinks away from the eye; see tiltFor for the geometry. The press itself is not animated — the original sets the properties outright on every manipulation delta, and animates only the return.

The original pushed the plane away along Z. Compose has no translationZ, and its perspective term is applied only when there is a rotation — so a centre press, which is pure depression, would get no perspective at all. The sinking is therefore emulated with a uniform scale computed from the same camera the rotation uses, which is the one deviation from the original geometry.

cameraDistance is a Dp, and that is the whole of what B-25 turned out to be. Compose's own GraphicsLayerScope.cameraDistance lands the camera at a fixed number of pixels — measured at value * 72 px on both skiko and hwui, the two backends this library has — while every piece of geometry it looks at is in dp. On a 2.625x screen a 100 dp tile is 262 px and subtends 0.456 of a 576 px camera where the same tile subtends 0.174 on the desktop, so the denser the screen the more perspective a press gets and the deeper it sinks. Measured, before and after: a tile drew at 0.9225 of itself on a Pixel 6a against 0.9685 on the desktop.

Holding the camera at a distance in dp makes the whole projection density-independent, which is what "the same design on every screen" has to mean. The default reproduces Compose's own camera exactly at density 1, so nothing about the desktop moves.

The distance itself is still a decision rather than a transcription: the original's camera was global to the screen and Microsoft never published how far away it was.