diff options
| -rw-r--r-- | services/core/java/com/android/server/InputMethodManagerService.java | 17 |
1 files changed, 16 insertions, 1 deletions
diff --git a/services/core/java/com/android/server/InputMethodManagerService.java b/services/core/java/com/android/server/InputMethodManagerService.java index a5d536e..2f153a0 100644 --- a/services/core/java/com/android/server/InputMethodManagerService.java +++ b/services/core/java/com/android/server/InputMethodManagerService.java @@ -2112,8 +2112,23 @@ public class InputMethodManagerService extends IInputMethodManager.Stub if (DEBUG) Slog.v(TAG, "Not hiding: forced show not cancelled by not-always hide"); return false; } + + // There is a chance that IMM#hideSoftInput() is called in a transient state where + // IMMS#InputShown is already updated to be true whereas IMMS#mImeWindowVis is still waiting + // to be updated with the new value sent from IME process. Even in such a transient state + // historically we have accepted an incoming call of IMM#hideSoftInput() from the + // application process as a valid request, and have even promised such a behavior with CTS + // since Android Eclair. That's why we need to accept IMM#hideSoftInput() even when only + // IMMS#InputShown indicates that the software keyboard is shown. + // TODO: Clean up, IMMS#mInputShown, IMMS#mImeWindowVis and mShowRequested. + final boolean shouldHideSoftInput = (mCurMethod != null) && (mInputShown || + (mImeWindowVis & InputMethodService.IME_ACTIVE) != 0); boolean res; - if ((mImeWindowVis & InputMethodService.IME_ACTIVE) != 0 && mCurMethod != null) { + if (shouldHideSoftInput) { + // The IME will report its visible state again after the following message finally + // delivered to the IME process as an IPC. Hence the inconsistency between + // IMMS#mInputShown and IMMS#mImeWindowVis should be resolved spontaneously in + // the final state. executeOrSendMessage(mCurMethod, mCaller.obtainMessageOO( MSG_HIDE_SOFT_INPUT, mCurMethod, resultReceiver)); res = true; |