Dev.to WebDev 🛠 Dev 👁 0 📖 4 min read

touch-action Is an Intersection: Un-trapping Page Scroll on 35 Mobile Games

I'm building a small site of free browser minigames. On a phone, if you put a finger on a game board and drag vertically, the page would not scroll. The board takes up 60-70% of the screen, so once you were in a game, ge

I'm building a small site of free browser minigames. On a phone, if you put a finger on a game board and drag vertically, the page would not scroll. The board takes up 60-70% of the screen, so once you were in a game, getting down to the guide and leaderboard below it was awkward.

I caused this myself. To make touch controls work, I had put touch-action: none on the whole board. Counting afterwards, 37 components did this. This post covers the fix, how I measured it, and the parts I have not verified.

https://raw.githubusercontent.com/Piousangel/minigames-hub/main/docs/marketing/blog/2026-11-18-touch-action-scroll-trap/01-2048.png

The rule that bit me: touch-action is an intersection

My first idea was to set pan-y on the inner canvas. That does nothing. The browser computes touch-action as the intersection of the values on the touched element and its ancestors, so a child cannot re-enable something a parent disabled with none.

The structure has to change instead:

  • The board container does not restrict touch-action.
  • Only a central "play zone" overlay gets touch-action: none.
  • The top and bottom edges are left alone, so a finger resting there and dragging scrolls the page.

I extracted this into a shared PlayZone component:

export function PlayZone({ top = 0.12, bottom = 0.12, left = 0, right = 0, axis = 'none', className, style, ...rest }: PlayZoneProps) {
  const pct = (n: number) => `${Math.max(0, Math.min(0.45, n)) * 100}%`
  const s: CSSProperties = { position: 'absolute', top: pct(top), bottom: pct(bottom), left: pct(left), right: pct(right), touchAction: axis, zIndex: 3, ...style }
  return <div data-play-zone="" aria-hidden="true" className={className} style={s} {...rest} />
}

export function inPlayZone(e: { target: EventTarget | null }): boolean {
  const t = e.target
  return t instanceof Element && t.closest('[data-play-zone]') !== null
}

The default scroll bands are 12% at the top and bottom. Games that only use horizontal input pass axis="pan-y". In 2048 the change is one <PlayZone /> inside the board plus a guard at the start of the touch handler:

const onTouchStart = (e: React.TouchEvent) => {
  if (!inPlayZone(e)) {
    touchStart.current = null // a touch that starts in the edge scroll band is not game input
    return
  }
  touchStart.current = { x: e.touches[0].clientX, y: e.touches[0].clientY }
}

Classify before you change anything

Before touching 36 games I sorted them by input style. The most important finding: when you relax touch-action, the browser fires pointercancel once a scroll gesture begins, but pointerdown has already fired by then. A tap game that acts on pointerdown (jump, shoot, whack a mole) registers one input the moment you put a finger down to scroll. So touches that start in the edge bands are ignored via inPlayZone.

Drag games must treat pointercancel as a cancel, not a commit. While classifying I found WordSearch ran its answer check on cancel, and Unblock committed the block move on cancel. Both become bugs once scrolling can cancel a gesture.

Some games needed no change: Hogline already only disables touch on its handle, Shadowfit is a 176px panel with scrollable UI around it, and FizzBeat, MorseRelay, Slosh and Tetris are small button-style controls. Backspace uses a two-finger rewind gesture, so the classification doc recommends none in the central zone only, with taps in the top and bottom bands ignored.

Measuring with CDP touch events

I did not want to stop at "seems to work", so I sent touches through CDP Input.dispatchTouchEvent from Playwright and measured how far scrollY moved. On the first game I fixed (Plop), a swipe in the center scrolled 0px, the top band +65px and the bottom band +112px, and tap-to-jump still worked. Across the 35 games in total: 0px in the play zone and roughly +75 to +128px from the edges. I also confirmed 2048 moves, Flappy starts, the Breakout paddle follows, AcornHop ignores taps in the bands but fires from the center, and GemMatch swipes still work.

It shipped as four PRs (Plop #75, first batch of 9 #79, then 11 in #83 and 10 in #84).

https://raw.githubusercontent.com/Piousangel/minigames-hub/main/docs/marketing/blog/2026-11-18-touch-action-scroll-trap/02-snake.png

What I haven't verified, and known limits

  • No real-device testing yet. All measurements above are emulated CDP touch. I haven't checked iOS Safari or Android Chrome on hardware. Unblock's horizontal pan-y / vertical pan-x blocks and Backspace's two-finger rewind near the bands are the cases I most want to see on a real phone.
  • Drags that start on the first or last row get cancelled. In WordSearch, GemMatch and Nonogram, a vertical or diagonal drag starting at an edge row turns into a scroll and is cancelled. Starting from the other end, or using tap-then-tap in GemMatch, works around it. It is a compromise, not a clean fix.
  • Polarity's "tap to start" card doesn't start if you press it in a band.
  • BubbleShooter's mouse aim line doesn't update inside the 12% top and bottom bands.

The band sizes (12% by default, 14%/16% on Plop) are design choices, not derived from user data.

Takeaways

  • If an ancestor says none, a child can't undo it. Leave the container unrestricted and put the restriction on a smaller zone.
  • Tap-on-pointerdown games get the input before the scroll starts, so exclude edge touches.
  • Treat pointercancel in drag handlers as cancel, never commit.
  • Even emulated scrollY numbers beat "seems fine", but real devices still need their own pass.

If you want to see the behavior, open 2048 on a phone and drag from the top or bottom edge of the board.

https://raw.githubusercontent.com/Piousangel/minigames-hub/main/docs/marketing/blog/2026-11-18-touch-action-scroll-trap/03-flappy.png

Drafted with AI assistance.

📰 Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.