tree 3054c7ed427d74e065d1e82c30431da2cb3bc247
parent 0945ce8e5b5b506356e0f0cb941b51557f0b8b56
author Ming-Shin Lu <lumark@google.com> 1620364418 +0800
committer Ming-Shin Lu <lumark@google.com> 1620895468 +0800

Fix some IME transition issues

- Add InsetsSource#getVisibleFrame to check if the launching task
  needs to disable fixed rotation:

  As CL[1] disables fixed rotation when IME surface exists for
  fixing IME janky when IME receiving different orientations.

  Even the CL used StartingData#hasImeSurface to forecast if needs to
  disable fixed rotation, it may hit some timing cases that StartingData
  not yet be prepared when launching the task next time quickly.

  Add InsetsSource#getVisibleFrame to prevent this edge case happen
  since this value will be updated when app client requests relayout.

- Consolide DC#isImeAttachedToApp() check:

  When recents animation started, will check whether
  the task has to hide soft-keyboard when it's not able to attach IME
  surface through DC#isImeAttachedToApp.

  Since this method only checks the layering target window states
  but not check if the IME surface has actually attached to the app.
  it might have potential issue that IME may state
  persistents when switching between tasks if we don't check if the
  actual IME container surface.

  Consolide DC#isImeSurfaceAttachedOnApp check to ensure hiding keyboard
  in RecentsAnimationController as the above potential case.
  (Also renamed the original DC#isImeAttachedToApp as
   DC#shouldImeAttachedToApp to make it easier understand the difference
   within the new isImeAttachedToApp.)

[1]: Idfdf129adbfbee6634d8a27aa78da1f631bd213b

Fix: 160451808
Bug: 184020899
Test: manual as below steps
1) launching the app with focusing IME in portrait mode.
2) swiping out to launcher
3) rotate the device in landscape
4) tapping the shortcut to launch the app again
5) repeat 1-4, see if IME surface will mess up to see both portrait
   and landcape surface.

Change-Id: I966ab69f260f828b6e96b3479a36467181288504