panelui-native 0.49.0 → 0.53.0

package/src/components/context-menu/index.tsx

@@ -0,0 +1,658 @@
+/**
+ * ContextMenu — the actions that belong to a piece of content, opened on it.
+ *
+ * A `Menu` hangs off a control that exists to be pressed: a ⋯ button, a toolbar
+ * item, something whose whole job is to open the menu. A context menu has no
+ * such control. The target is the content itself — a message, a note, a photo,
+ * a row — and the actions are reached by pressing and holding it.
+ *
+ * ```tsx
+ * <ContextMenu>
+ *   <ContextMenu.Trigger>
+ *     <Message>Would you like an interactive todo list?</Message>
+ *   </ContextMenu.Trigger>
+ *   <ContextMenu.Content>
+ *     <ContextMenu.Item icon={<Share2 size={16} />}>Share</ContextMenu.Item>
+ *     <ContextMenu.Item icon={<Copy size={16} />}>Copy</ContextMenu.Item>
+ *     <ContextMenu.Separator />
+ *     <ContextMenu.Item variant="destructive" icon={<Flag size={16} />}>
+ *       Report
+ *     </ContextMenu.Item>
+ *   </ContextMenu.Content>
+ * </ContextMenu>
+ * ```
+ *
+ * ## It is a Menu, and deliberately so
+ *
+ * The rows here *are* `Menu`'s rows — the same components, not a second set
+ * styled to match. `ContextMenu.Item` and `Menu.Item` are one implementation,
+ * so the destructive colour, the press-in scale, the indicator column and the
+ * dismiss-on-select rule cannot drift apart between the two ways of reaching
+ * them. The panel is `Menu`'s panel, which is `Popover`'s, so `presentation`,
+ * submenus and edge-flipping all arrive already working.
+ *
+ * What this component owns is the two things a menu opened on content needs and
+ * a menu opened from a button does not: the long press, and where the panel
+ * goes.
+ *
+ * ## Anchored to the finger, not to the target
+ *
+ * A toolbar menu is placed against its trigger, because the trigger is small
+ * and its position is the only sensible answer. A context menu's target is
+ * often most of the screen — a whole message, a whole card — and the middle of
+ * it is not where the finger was. So the anchor is the press point by default,
+ * and the panel unfolds from it the way a popover unfolds from a button.
+ *
+ * `anchor="target"` places it against the target's bounds instead, which is the
+ * better answer for something small and list-shaped, where the panel lining up
+ * with the row reads as belonging to it.
+ *
+ * ## Why the gesture is not a Pressable
+ *
+ * The target usually has a press of its own — open the thread, play the video,
+ * follow the link — and the two must not both fire. React Native's `Pressable`
+ * decides between them after the fact, and the tap can still get through on the
+ * way to a long press; the recogniser here is asked for the arbitration up
+ * front instead, so a hold that opens the menu never also counts as a press.
+ *
+ * It is also what lets the target be anything at all. A cloned `onLongPress`
+ * needs a child that takes one, which rules out exactly the plain views —
+ * bubbles, cards, images — that content-native actions are usually attached to.
+ */
+import {
+  Children,
+  cloneElement,
+  createContext,
+  isValidElement,
+  useCallback,
+  useContext,
+  useEffect,
+  useMemo,
+  useRef,
+  useState,
+  type ReactElement,
+  type ReactNode,
+} from 'react';
+import { View, type ViewProps } from 'react-native';
+import { Gesture, GestureDetector } from 'react-native-gesture-handler';
+import Animated, {
+  FadeOut,
+  runOnJS,
+  useAnimatedStyle,
+  useReducedMotion,
+  useSharedValue,
+  withSpring,
+} from 'react-native-reanimated';
+import { useCSSVariable } from 'uniwind';
+import { Portal } from '../../primitives/portal';
+import {
+  Menu,
+  type MenuCheckboxItemProps,
+  type MenuContentProps,
+  type MenuItemProps,
+  type MenuProps,
+} from '../menu';
+import { usePopoverAnchor } from '../popover';
+import { cn } from '../../utils/cn';
+import { selectionTick } from '../../utils/haptics';
+
+/**
+ * How long the target is held before the menu opens.
+ *
+ * Long enough not to fire while a finger is on its way to a scroll, short
+ * enough that nobody wonders whether the hold is working. The platforms sit
+ * either side of this; it is the value the rest of this library holds at.
+ */
+const DEFAULT_DELAY = 350;
+
+/**
+ * How far the finger may travel during the hold before it stops counting.
+ *
+ * Generous rather than tight, and the reason is what the target usually sits
+ * in: a scroller. A threshold small enough to feel precise cancels the menu for
+ * anyone whose thumb drifts while holding still, and the gesture it is being
+ * told apart from — a scroll — has moved a great deal further than this by the
+ * time it matters.
+ */
+const DEFAULT_SLOP = 12;
+
+/**
+ * Extra height on every row.
+ *
+ * A context menu is opened by a hold and read with the hand still over it, at
+ * whatever angle the phone happened to be held at. A menu dropped from a button
+ * is aimed at deliberately; this one is landed on. The rows are taller than
+ * `Menu`'s for that reason alone, and it is the only measurement that differs.
+ */
+const ROW_CLASS = 'py-4 ps-4 pe-3.5';
+
+/**
+ * Floor for the panel's width.
+ *
+ * A context menu has no trigger to take a width from, and a column of one-word
+ * verbs left to size itself lands somewhere around a thumb's width — too narrow
+ * to aim at, and too narrow to read as a panel belonging to the whole piece of
+ * content it was opened on.
+ */
+const DEFAULT_MIN_WIDTH = 280;
+
+/**
+ * The colour an icon falls back to before the stylesheet has been read.
+ *
+ * A general-purpose icon set defaults an unset colour to `currentColor`, which
+ * React Native cannot resolve and refuses to paint. A neutral mid grey is
+ * legible on either a light or a dark panel for the frame or two it lasts.
+ */
+const ICON_FALLBACK = '#737373';
+
+function useTint(variable: string, fallback: string): string {
+  const raw = useCSSVariable(variable);
+  return typeof raw === 'string' ? raw : fallback;
+}
+
+/**
+ * Paints a row's glyph to match its label, without the caller saying so twice.
+ *
+ * The icons on these rows come from whatever set the app already uses, and a
+ * general-purpose one has no idea what an overlay's foreground is — left alone
+ * it paints `currentColor`, which React Native will not draw at all. Setting it
+ * here means a destructive row's icon turns red along with its label, which is
+ * the one place the two disagreeing would matter.
+ *
+ * An explicit colour on the element still wins: a brand mark that carries its
+ * own colours is not something to overrule.
+ */
+function useGlyph(variant: 'default' | 'destructive' | undefined) {
+  const foreground = useTint('--color-overlay-foreground', ICON_FALLBACK);
+  const destructive = useTint('--color-destructive', ICON_FALLBACK);
+  const tint = variant === 'destructive' ? destructive : foreground;
+
+  return useCallback(
+    (icon: ReactNode): ReactNode => {
+      if (!isValidElement(icon)) return icon;
+      const element = icon as ReactElement<{ color?: string }>;
+      if (element.props.color !== undefined) return element;
+      return cloneElement(element, { color: tint });
+    },
+    [tint]
+  );
+}
+
+/** The target's frame in window coordinates, taken as the menu opens. */
+interface TargetRect {
+  x: number;
+  y: number;
+  width: number;
+  height: number;
+}
+
+interface ContextMenuContextValue {
+  /** Where the target was when it was held, or `null` before anything has. */
+  target: TargetRect | null;
+  setTarget: (rect: TargetRect | null) => void;
+  /** The trigger's children, so the preview can draw the same thing again. */
+  content: ReactNode;
+  setContent: (node: ReactNode) => void;
+  /** Whether a `ContextMenu.Preview` was declared inside the panel. */
+  hasPreview: boolean;
+}
+
+const ContextMenuContext = createContext<ContextMenuContextValue | null>(null);
+
+function useContextMenu(part: string): ContextMenuContextValue {
+  const context = useContext(ContextMenuContext);
+  if (!context) throw new Error(`${part} must be used inside <ContextMenu>.`);
+  return context;
+}
+
+/**
+ * Whether a `ContextMenu.Preview` appears anywhere in the declared tree.
+ *
+ * Sniffed from the elements rather than reported by the preview when it mounts,
+ * because the answer is needed *before* it does: the trigger has to know at the
+ * moment of the hold whether the panel will be sharing the screen with a lifted
+ * copy of the target, and by the time a child of the panel exists the anchor has
+ * already been set.
+ */
+function declaresPreview(children: ReactNode): boolean {
+  let found = false;
+  Children.forEach(children, (child) => {
+    if (found || !isValidElement(child)) return;
+    if (child.type === ContextMenuPreview) {
+      found = true;
+      return;
+    }
+    const inner = (child.props as { children?: ReactNode }).children;
+    if (inner && declaresPreview(inner)) found = true;
+  });
+  return found;
+}
+
+/**
+ * The root takes exactly what `Menu`'s root takes — `open`, `onOpenChange`,
+ * `defaultOpen`, `presentation` and `haptics` — because it *is* that root.
+ */
+export type ContextMenuProps = MenuProps;
+
+/**
+ * The root. Provides the menu's own context and the popover underneath it.
+ *
+ * It renders a `Menu`, which is not a shortcut — it is the point. Everything a
+ * menu is, this is, and the parts below are the only difference.
+ */
+function ContextMenuRoot({ children, ...props }: ContextMenuProps) {
+  const [target, setTarget] = useState<TargetRect | null>(null);
+  const [content, setContent] = useState<ReactNode>(null);
+  const hasPreview = useMemo(() => declaresPreview(children), [children]);
+
+  const context = useMemo<ContextMenuContextValue>(
+    () => ({ target, setTarget, content, setContent, hasPreview }),
+    [target, content, hasPreview]
+  );
+
+  return (
+    <ContextMenuContext.Provider value={context}>
+      <Menu {...props}>{children}</Menu>
+    </ContextMenuContext.Provider>
+  );
+}
+
+/** Which rectangle the panel is placed against. */
+export type ContextMenuAnchor = 'point' | 'target';
+
+export interface ContextMenuTriggerProps extends Omit<ViewProps, 'children'> {
+  /**
+   * Classes on the wrapper the content sits in, which lays out like any other
+   * view — it does not shrink to its child, because the things held are usually
+   * meant to fill their place in the layout. It is also the rect
+   * `anchor="target"` measures.
+   */
+  className?: string;
+  /**
+   * The content the actions belong to. Anything at all — it is not required to
+   * be pressable, and is not cloned or altered.
+   */
+  children: ReactNode;
+  /**
+   * `point` anchors the panel where the finger landed, `target` against the
+   * bounds of the whole trigger.
+   *
+   * Point is the default because a context menu's target is usually large, and
+   * the middle of a whole message is not where the press was. Reach for
+   * `target` when the target is small and list-shaped and the panel should read
+   * as lining up with it.
+   */
+  anchor?: ContextMenuAnchor;
+  /** How long the hold has to last, in milliseconds. 350 by default. */
+  delay?: number;
+  /**
+   * How far the finger may move during the hold before it stops being one, in
+   * points. 12 by default.
+   *
+   * Loose rather than tight, because the target is usually inside a scroller: a
+   * threshold small enough to feel precise cancels the menu for anyone whose
+   * thumb drifts while holding still, and a scroll has travelled much further
+   * than this by the time the two need telling apart. Tighten it only for a
+   * target that cannot be scrolled.
+   */
+  slop?: number;
+  /** A short press on the target, which the hold never also counts as. */
+  onPress?: () => void;
+  /**
+   * Tick the haptic engine as the menu opens. Needs the optional
+   * `expo-haptics`, and is silent without it.
+   *
+   * Worth setting more often than not. A hold has no edge to it the way a press
+   * does — nothing moves under the finger at the moment it takes — so the tick
+   * is what tells someone the hold has been long enough, before the panel has
+   * had time to say so.
+   */
+  haptics?: boolean;
+  /** Nothing opens the menu, and the short press stops firing too. */
+  disabled?: boolean;
+}
+
+/**
+ * Wraps the content and opens the menu when it is held.
+ *
+ * The wrapper is a plain view and lays out like one, stretching as a view does
+ * rather than shrinking to its child. That is the opposite of what a tooltip's
+ * trigger wants, and for the opposite reason: a tooltip names a control and
+ * belongs over it, while the things held here — a bubble, a card, a row — are
+ * usually meant to fill their place in the layout, and a wrapper that collapsed
+ * around them would change it.
+ *
+ * It is also the rect measured under `anchor="target"`, which is why that
+ * anchoring lines the panel up with the row rather than with the text in it.
+ */
+function ContextMenuTrigger({
+  className,
+  children,
+  anchor = 'point',
+  delay = DEFAULT_DELAY,
+  slop = DEFAULT_SLOP,
+  onPress,
+  haptics = false,
+  disabled = false,
+  ...props
+}: ContextMenuTriggerProps) {
+  const { setOpen, anchorTo } = usePopoverAnchor('ContextMenu.Trigger');
+  const { setTarget, setContent, hasPreview } = useContextMenu('ContextMenu.Trigger');
+  const ref = useRef<View>(null);
+
+  /*
+   * A preview overrules `anchor`, and has to.
+   *
+   * The panel is placed outside whatever rectangle it is given, so anchoring to
+   * the target is what keeps it clear of the lifted copy of that target. Anchor
+   * to the press instead and the panel opens over the very thing the preview
+   * exists to show.
+   */
+  const against: ContextMenuAnchor = hasPreview ? 'target' : anchor;
+
+  /*
+   * Both branches end up doing the same thing — set an anchor, then open — and
+   * differ only in which rectangle they set. A point is a zero-sized rect,
+   * which the popover places a panel against exactly as it does a trigger's
+   * bounds; there is no separate code path for it downstream.
+   *
+   * The target is measured either way, because the preview draws at that rect
+   * whatever the panel is placed against.
+   */
+  const openAt = useCallback(
+    (x: number, y: number) => {
+      // Ticked here rather than in the gesture callback so it fires once the
+      // hold has been accepted, which is the moment there is something to
+      // confirm — and on the same side of the bridge as the opening.
+      if (haptics) selectionTick();
+
+      // Measured on opening rather than on layout: the target may have
+      // scrolled since, and a stale rect anchors the panel to where it was.
+      ref.current?.measureInWindow((mx, my, width, height) => {
+        setTarget({ x: mx, y: my, width, height });
+        setContent(children);
+
+        /*
+         * The panel is placed outside the rectangle it is given, so under a
+         * preview that rectangle has to be the target's *lifted* bounds rather
+         * than its resting ones. The lift grows the target about its middle,
+         * and anchoring to the smaller rect left the panel overlapping the
+         * last few points of it — where the lifted copy, drawn over the panel,
+         * simply hid the row underneath.
+         */
+        const grown = hasPreview ? (PREVIEW_SCALE - 1) / 2 : 0;
+        anchorTo(
+          against === 'point'
+            ? { x, y, width: 0, height: 0 }
+            : {
+                x: mx - width * grown,
+                y: my - height * grown,
+                width: width * (1 + grown * 2),
+                height: height * (1 + grown * 2),
+              }
+        );
+        setOpen(true);
+      });
+    },
+    [against, anchorTo, setOpen, haptics, setTarget, setContent, children, hasPreview]
+  );
+
+  const gesture = useMemo(() => {
+    /*
+     * The hold and the tap are given to the recogniser as alternatives, so it
+     * decides between them rather than both firing. That is the whole reason
+     * this is a gesture and not a `Pressable`: the target below usually has a
+     * press of its own, and a hold that opened the menu must not also count as
+     * one.
+     *
+     * `absoluteX`/`absoluteY` are window coordinates, which is the space the
+     * popover places panels in — so the press point needs no conversion.
+     */
+    const hold = Gesture.LongPress()
+      .minDuration(delay)
+      .maxDistance(slop)
+      .enabled(!disabled)
+      .onStart((event) => {
+        runOnJS(openAt)(event.absoluteX, event.absoluteY);
+      });
+
+    const tap = Gesture.Tap()
+      .enabled(!disabled && !!onPress)
+      .onEnd((_event, success) => {
+        if (success && onPress) runOnJS(onPress)();
+      });
+
+    return Gesture.Exclusive(hold, tap);
+  }, [delay, slop, disabled, onPress, openAt]);
+
+  return (
+    <GestureDetector gesture={gesture}>
+      {/*
+        `collapsable={false}` keeps the wrapper as a real view on Android, where
+        a view that only groups children is otherwise flattened away — and a
+        flattened view cannot be measured, which `anchor="target"` needs.
+      */}
+      <View ref={ref} collapsable={false} className={className} {...props}>
+        {children}
+      </View>
+    </GestureDetector>
+  );
+}
+
+/** How far the lifted target grows. Enough to read as off the page. */
+const PREVIEW_SCALE = 1.04;
+
+/** Coming off the page. Soft, because the target is large and close to the eye. */
+const PREVIEW_SPRING = { damping: 20, stiffness: 220, mass: 0.7 } as const;
+
+export interface ContextMenuPreviewProps {
+  /**
+   * Drawn instead of the target itself. For a target that would be wrong to
+   * repeat — one carrying a video, a live map, a text field with a cursor in
+   * it — or one that should show more of itself once it has the screen.
+   *
+   * Left out, the target is drawn again as it stands, which is what makes the
+   * lift read as the content coming forward rather than as a picture of it
+   * appearing.
+   */
+  children?: ReactNode;
+  /** Extra classes on the lifted copy. */
+  className?: string;
+}
+
+/**
+ * The target, lifted off the page while its actions are up.
+ *
+ * Declared inside `ContextMenu.Content`, but not drawn there — it floats over
+ * the dimmed screen at the place the target was measured, and the panel is
+ * anchored to that same rectangle so the two never overlap. Its presence is
+ * what switches the anchor: a panel placed at the press point would open across
+ * the very content the preview exists to hold up.
+ *
+ * What it draws is the trigger's own children, rendered a second time. That
+ * keeps the lift honest — it is the content itself coming forward, at the size
+ * and in the place it already occupied — and it is why a target that should not
+ * simply be repeated can pass its own `children` instead.
+ *
+ * It takes no touches. The actions are in the panel; the lifted content is
+ * there to say what they are about, and a second live copy of a pressable card
+ * would be a second place to press.
+ */
+function ContextMenuPreview({ children, className }: ContextMenuPreviewProps) {
+  const { target, content } = useContextMenu('ContextMenu.Preview');
+  const lift = useSharedValue(0);
+  const reduced = useReducedMotion();
+
+  useEffect(() => {
+    lift.value = reduced ? 1 : withSpring(1, PREVIEW_SPRING);
+  }, [lift, reduced]);
+
+  const style = useAnimatedStyle(() => ({
+    opacity: lift.value,
+    transform: [{ scale: 1 + lift.value * (PREVIEW_SCALE - 1) }],
+  }));
+
+  if (!target) return null;
+
+  return (
+    <Portal>
+      <Animated.View
+        pointerEvents="none"
+        exiting={reduced ? undefined : FadeOut.duration(140)}
+        style={[
+          style,
+          {
+            position: 'absolute',
+            left: target.x,
+            top: target.y,
+            width: target.width,
+          },
+        ]}
+        className={className}
+      >
+        {children ?? content}
+      </Animated.View>
+    </Portal>
+  );
+}
+
+ContextMenuPreview.displayName = 'ContextMenu.Preview';
+
+export interface ContextMenuItemProps extends MenuItemProps {
+  /**
+   * The row's glyph, drawn at the trailing edge rather than in front of the
+   * label. Painted to match the label unless it carries a colour of its own.
+   */
+  icon?: ReactNode;
+}
+
+/**
+ * One row: the verb at the leading edge, its glyph at the trailing one.
+ *
+ * The other way round is right for a menu dropped from a button, where the
+ * glyphs form a column the eye runs down to find the row it wants. A context
+ * menu is not read that way. It appears under the hand that opened it, already
+ * over the content, and what is being scanned is the *words* — so the words
+ * start at the edge, flush with one another, and the glyph sits at the far side
+ * confirming the row rather than introducing it.
+ *
+ * It is still `Menu.Item` underneath, handed the glyph as its trailing slot. So
+ * the press-in fill, the destructive colour and the dismiss-on-select rule are
+ * one implementation shared with `Menu`, and only the arrangement differs.
+ */
+function ContextMenuItem({ className, icon, variant, ...props }: ContextMenuItemProps) {
+  const glyph = useGlyph(variant);
+
+  return (
+    <Menu.Item
+      variant={variant}
+      className={cn(ROW_CLASS, className)}
+      trailing={icon ? glyph(icon) : undefined}
+      {...props}
+    />
+  );
+}
+
+ContextMenuItem.displayName = 'ContextMenu.Item';
+
+/**
+ * A row carrying a state. Its tick stays at the leading edge, where `Menu` puts
+ * it, because a tick is not a glyph naming the row — it is the answer to it,
+ * and a column of them is what makes a set of choices readable as one.
+ */
+function ContextMenuCheckboxItem({ className, ...props }: MenuCheckboxItemProps) {
+  return <Menu.CheckboxItem className={cn(ROW_CLASS, className)} {...props} />;
+}
+
+ContextMenuCheckboxItem.displayName = 'ContextMenu.CheckboxItem';
+
+/**
+ * Everything `Menu.Content` takes. The four listed here are the ones whose
+ * defaults differ, and they are listed so the difference is visible.
+ */
+export interface ContextMenuContentProps extends MenuContentProps {
+  /** Which side of the anchor the panel opens on. Down from the press, flipping
+   * above it near the bottom of the screen. */
+  placement?: MenuContentProps['placement'];
+  /** Where it sits along the other axis. From the press, not centred on it. */
+  align?: MenuContentProps['align'];
+  /** Gap between the anchor and the panel. Small, so it reads as coming out of
+   * the press rather than floating near it. */
+  offset?: number;
+  /**
+   * Floor for the panel's width. A context menu has no trigger to take its
+   * width from, and a column of one-word verbs is too narrow to aim at.
+   */
+  minWidth?: number;
+  /** Dim the screen behind the panel. On here, unlike a plain popover. */
+  scrim?: boolean;
+}
+
+/**
+ * The panel, with the defaults a context menu wants rather than a popover's.
+ *
+ * It unfolds down and from the press rather than being centred on it —
+ * centring is right for a panel under a button, and wrong for one at a
+ * fingertip, where it would put half the panel back under the hand that opened
+ * it. Near the bottom of the screen `Popover` flips it above the press and
+ * clamps it into the safe area, so the one case where down does not work
+ * answers itself.
+ *
+ * The gap to the anchor is small, so the panel reads as coming out of the press
+ * rather than floating near it.
+ *
+ * The screen dims behind it, which a popover does not do. A context menu is
+ * modal in practice — the content underneath is what the actions are *about*,
+ * so nothing else on the screen is available while it is up, and the dim is
+ * what says so.
+ */
+function ContextMenuContent({
+  placement = 'bottom',
+  align = 'start',
+  offset = 8,
+  minWidth = DEFAULT_MIN_WIDTH,
+  scrim = true,
+  children,
+  ...props
+}: ContextMenuContentProps) {
+  // The panel is portalled, so it mounts outside this provider's subtree.
+  // Re-provided here so `ContextMenu.Preview`, which is declared among these
+  // rows, can still reach the rectangle the trigger measured.
+  const context = useContextMenu('ContextMenu.Content');
+
+  return (
+    <Menu.Content
+      placement={placement}
+      align={align}
+      offset={offset}
+      minWidth={minWidth}
+      scrim={scrim}
+      {...props}
+    >
+      <ContextMenuContext.Provider value={context}>{children}</ContextMenuContext.Provider>
+    </Menu.Content>
+  );
+}
+
+/*
+ * Everything not listed here is Menu's, passed straight through. The three that
+ * are listed are wrappers rather than second implementations — they set a class
+ * or move a slot and hand the row back to `Menu.Item`, so the press-in fill,
+ * the destructive colour and the dismiss-on-select rule stay in one place and
+ * cannot drift between the two ways of reaching a list of verbs.
+ */
+export const ContextMenu = Object.assign(ContextMenuRoot, {
+  Trigger: ContextMenuTrigger,
+  Content: ContextMenuContent,
+  Preview: ContextMenuPreview,
+  Background: Menu.Background,
+  Label: Menu.Label,
+  Item: ContextMenuItem,
+  CheckboxItem: ContextMenuCheckboxItem,
+  RadioGroup: Menu.RadioGroup,
+  RadioItem: Menu.RadioItem,
+  Separator: Menu.Separator,
+  Sub: Menu.Sub,
+  SubTrigger: Menu.SubTrigger,
+  SubContent: Menu.SubContent,
+});