- If the task is previously not visible or has no visible children at
the point when we start controlling it in the task org, hide the task
until we send taskAppeared to ensure that the task org can reparent
and show it otherwise we could see a flash of the task.
This happens mainly from two cases:
- when starting a new task with a given win mode, we show it and wait
for first draw before notifying the task org
- when transitioning into pip from swipe up, the activity is hidden
and when it requests to enter pip is made visible again
Since we are hiding the task w/ the pending transaction, we also need
to defer all task org callbacks until that's applied to ensure proper
lifecycle of the calls.
- Also skip app transitions for task org tasks for now
Bug: 152809695
Bug: 152134460
Test: Open a bubble, ensure that we don't see the task in fullscreen
first. Enter pip, ensure that we don't see flash of the task
before SysUI can fade it in.
Test: atest PipAnimationControllerTest
Test: atest TaskOrganizerTests
Test: atest SplitScreenTests
Change-Id: If51e98cd007faef35e99acd31b27b20eebbea010
am skip reason: Change-Id I22182d90b0057df67fbd6d20785da23d6c17dcf3 with SHA-1 c8048daed2 is in history
Change-Id: I0d5375b98bac66de51cfe1a5398ca591f3a31562
Once a new Request is in flight, we probably don't want to display the
old responses.
This prevents a VIEW_ENTERED flow from displaying the suggestions
(either as a dropdown or as inline chips) if there is a pending request
for the same partition. Thus fixing a bug where the suggestions may be
rendered twice - once from the VIEW_ENTERED update and again from the
Fill Request succeeding (see http://b/152620157#comment7). An example of
how this may occur:
1. User taps on a View that previously showed suggestions.
2. App manually triggers autofill, which issues a new fill request.
3. The Session is updated with VIEW_ENTERED (due to (1)).
Session#requestShowInlineSuggestionsLocked returns false because of the
pending request. The system fallsback to drawing the dropdown UI.
4. FillRequest from (2) completes. This hides the dropdown and shows
inline chips.
An alternative is to exit the VIEW_ENTERED flow when there's a pending
request for the same view. This should also fix the bug, but is no
longer necessary with the current fix. The advantage of the current
approach is that the VIEW_ENTERED update logic and the request logic are
less coupled.
Fix: 152620157
Test: manual
Test: atest android.autofillservice.cts
Change-Id: Id15887ffdf28d0a3ea32e1f68ffd81f994f93187
clearResponse=true doesn't clear the response if the `response` arg is non null.
The new behavior matches the corresponding logic for authenticated
responses above (lines 3197-3200).
This affects only 1 invoking method, replaceResponseLocked, which sets
the responses again soon after, so this change shouldn't affect any
behavior.
This change also fixes the javadoc for the method.
Bug: 152620157
Test: atest android.autofillservice.cts
Test: manual
Change-Id: I78262baa1c6c11ef7e4b3c12c2ecd3d9dd28bfd7
Also test MidiFramer
Add end-to-end test through encoder and decoder.
Add test for reserved bit in header.
Bug: 35669198
Bug: 140638458
Bug: 149927520
Test: atest BluetoothMidiTests
Change-Id: I767deab6847d2b45e01bed29f5eed3df0f0e46c0
Merged-In: I767deab6847d2b45e01bed29f5eed3df0f0e46c0
