lsm / neokai / 25634109409

82%

push

github

fix: remove double-compensation causing mobile Safari chat composer jump (#1845)

* fix: remove double-compensation causing mobile Safari chat composer jump

PR #1837 introduced three simultaneous mechanisms to handle the iOS Safari
keyboard. Two of them conflicted, causing the composer to jump to the top
or hide behind the address bar:

1. Container shrinking via --safe-height (useViewportSafety.ts) already
   shrinks the app container to visualViewport.height when the keyboard
   opens. An absolute; bottom:0 child naturally sits at the bottom of the
   visible area within that shrunk container.

2. The CSS rule `html.keyboard-open .chat-footer { bottom: var(--keyboard-height) }`
   pushed the footer UP by the keyboard height within an already-shrunk
   container → double compensation → composer flew to the top.

3. The 350ms delayed scrollIntoView in InputTextarea.tsx triggers a known
   Safari scroll-to-top bug (WebKit #297779) when the viewport is being
   resized. With the container already shrunk correctly via --safe-height,
   the textarea is visible without scrolling.

Changes:
- Remove the @media (pointer: coarse) block with .chat-footer keyboard offset
  and its transition from styles.css
- Remove the handleFocus/handleBlur scrollIntoView workaround, the
  focusScrollTimerRef, and the isMobileDevice ref from InputTextarea.tsx
- The viewport meta tag already has interactive-widget=resizes-content,
  which tells iOS Safari 16+ to resize the layout viewport when the
  keyboard opens, making absolute bottom:0 work correctly without JS

* docs: update stale comment about --keyboard-height consumer

The previous comment referenced the chat composer as a consumer of
--keyboard-height, but after removing the double-compensation CSS rule
nothing in the codebase reads this variable. Updated the comment to
clarify it is set for potential future use.

8234 of 11506 branches covered (71.56%)

Branch coverage included in aggregate %.

64950 of 77297 relevant lines covered (84.03%)

289.69 hits per line