Update mKeyIsWaitingForEventsTimeout separately from the prune check
Before this CL, the variable mKeyIsWaitingForEventsTimeout was getting
updated inside 'shouldPruneInboundQueueLocked', but that is not the
correct place to do this update. In this CL, we move this update to the
enqueue function, and this allows us to make
'shouldPruneInboundQueueLocked' const.
The current behaviour of:
```
When there are unprocessed motion events, we wait up to 500 ms to send
any key events, in anticipation that focus might change. However, if
there are subsequent "ACTION_DOWN" events, we don't want to be adding
this 500ms wait, so we short-circuit the timeout and process the pending
keys immediately (send them to the currently focused window).
```
is not changed in this CL.
However, we are updating the test for this behaviour to:
- Be correctly named. The pending key is not dropped, but it's processed
immediately instead.
- Actually test the behaviour in question. Previously, the test passed
even with the 'mKeyIsWaitingForEventsTimeout' not getting updated
during the queueing stage. This is because the test used a large
(1000ms) timeout for event consumption. So in practice, the test was
still waiting for 500 ms and only then the consumption passed.
In this CL, the test is changed so that it consumes within the allotted
timeframe. If this becomes problematic (flaky), we would need to change
the KEY_WAITING_FOR_EVENTS_TIMEOUT and possibly provide that value from
the policy, thus allowing the test to override it. Or, another way to
fix flakes here would be to switch to an injectable clock, although in
that case we would still ideally open up KEY_WAITING_FOR_EVENTS_TIMEOUT
for tests, or maybe just hardcode the value with a comment, similarly to
how it's currently done.
Bug: 294553757
Test: TEST=inputflinger_tests; m $TEST && $ANDROID_HOST_OUT/nativetest64/$TEST/$TEST
Change-Id: Ife71634960b365df1d106fd28f0dc0047ac2ed2b
diff --git a/services/inputflinger/dispatcher/include/InputDispatcherPolicyInterface.h b/services/inputflinger/dispatcher/include/InputDispatcherPolicyInterface.h
index 9e6209b..62c2b02 100644
--- a/services/inputflinger/dispatcher/include/InputDispatcherPolicyInterface.h
+++ b/services/inputflinger/dispatcher/include/InputDispatcherPolicyInterface.h
@@ -131,6 +131,18 @@
return std::chrono::nanoseconds(currentTime - eventTime) >= STALE_EVENT_TIMEOUT;
}
+ /**
+ * Get the additional latency to add while waiting for other input events to process before
+ * dispatching the pending key.
+ * If there are unprocessed events, the pending key will not be dispatched immediately. Instead,
+ * the dispatcher will wait for this timeout, to account for the possibility that the focus
+ * might change due to touch or other events (such as another app getting launched by keys).
+ * This would give the pending key the opportunity to go to a newly focused window instead.
+ */
+ virtual std::chrono::nanoseconds getKeyWaitingForEventsTimeout() {
+ return KEY_WAITING_FOR_EVENTS_TIMEOUT;
+ }
+
/* Notifies the policy that a pointer down event has occurred outside the current focused
* window.
*
@@ -150,6 +162,13 @@
/* Notifies the policy that there was an input device interaction with apps. */
virtual void notifyDeviceInteraction(DeviceId deviceId, nsecs_t timestamp,
const std::set<gui::Uid>& uids) = 0;
+
+private:
+ // Additional key latency in case a connection is still processing some motion events.
+ // This will help with the case when a user touched a button that opens a new window,
+ // and gives us the chance to dispatch the key to this new window.
+ static constexpr std::chrono::nanoseconds KEY_WAITING_FOR_EVENTS_TIMEOUT =
+ std::chrono::milliseconds(500);
};
} // namespace android