Previously, the input dispatcher assumed that the input channel's
receive pipe file descriptor was a sufficiently unique identifier for
looking up input channels in its various tables. However, it can happen
that an input channel is disposed and then a new input channel is
immediately created that reuses the same file descriptor. Ordinarily
this is not a problem, however there is a small opportunity for a race
to arise in InputQueue.
When InputQueue receives an input event from the dispatcher, it
generates a finishedToken that encodes the channel's receive pipe fd,
and a sequence number. The finishedToken is used by the ViewRoot
as a handle for the event so that it can tell the InputQueue when
the event has finished being processed.
Here is the race:
1. InputQueue receives an input event, assigns a new finishedToken.
2. ViewRoot begins processing the input event.
3. During processing, ViewRoot unregisters the InputChannel.
4. A new InputChannel is created and is registered with the Input Queue.
This InputChannel happens to have the same receive pipe fd as
the one previously registered.
5. ViewRoot tells the InputQueue that it has finished processing the
input event, passing along the original finishedToken.
6. InputQueue throws an exception because the finishedToken's receive
pipe fd is registered but the sequence number is incorrect so it
assumes that the client has called finish spuriously.
The fix is to include a unique connection id within the finishedToken so
that the InputQueue can accurately confirm that the token belongs to
the currently registered InputChannel rather than to an old one that
happened to have the same receive pipe fd. When it notices this, it
ignores the spurious finish.
I've also made a couple of other small changes to avoid similar races
elsewhere.
This patch set also includes a fix to synthesize a finished signal
when the input channel is unregistered on the client side to
help keep the server and client in sync.
Bug: 2834068
Change-Id: I1de34a36249ab74c359c2c67a57e333543400f7b
Added a new asynchronous injection mode and made the existing
synchronization mechanism more robust.
Change-Id: Ia4aa04fd9b75ea2461a844c5b7933c831c1027e6
Added several new coordinate values to MotionEvents to capture
touch major/minor area, tool major/minor area and orientation.
Renamed NDK input constants per convention.
Added InputDevice class in Java which will eventually provide
useful information about available input devices.
Added APIs for manufacturing new MotionEvent objects with multiple
pointers and all necessary coordinate data.
Fixed a bug in the input dispatcher where it could get stuck with
a pointer down forever.
Fixed a bug in the WindowManager where the input window list could
end up containing stale removed windows.
Fixed a bug in the WindowManager where the input channel was being
removed only after the final animation transition had taken place
which caused spurious WINDOW DIED log messages to be printed.
Change-Id: Ie55084da319b20aad29b28a0499b8dd98bb5da68
This factors out the boiler-plate code from the sample
app to a common glue code that can be used for everyone
writing this style of app: a dedicated app thread that
takes care of waiting for events and processing them.
As part of doing this, ALooper has a new facility to allow
registration of fds that cause ALooper_pollOnce() to return
the fd that has data, allowing the app to drive the loop
without callbacks. Hopefully this makes some people feel better. :)
Also do some other cleanup of the ALooper API, plus some
actual documentation.
Change-Id: Ic53bd56bdf627e3ba28a3c093faa06a92be522b8
Removed old input dispatch code.
Refactored the policy callbacks.
Pushed a tiny bit of the power manager state down to native.
Fixed long press on MENU.
Made the virtual key detection and cancelation a bit more precise.
Change-Id: I5d8c1062f7ea0ab3b54c6fadb058c4d5f5a9e02e
Set a default orientation of ROTATION_0.
Added some more careful checks based on whether we have valid
absolute axis information from the driver.
Reset key repeating during configuration changes since the keyboard
device may have been removed.
Change-Id: I685960828acffcb17595fc5683309e8064a76714
Target identification is now fully native.
Fixed a couple of minor issues related to input injection.
Native input enabled by default, can be disabled by setting
WindowManagerPolicy.ENABLE_NATIVE_INPUT_DISPATCH to false.
Change-Id: I7edf66ed3e987cc9306ad4743ac57a116af452ff
Added ANRs handling.
Added event injection.
Fixed a NPE ActivityManagerServer writing ANRs to the drop box.
Fixed HOME key interception.
Fixed trackball reporting.
Fixed pointer rotation in landscape mode.
Change-Id: I50340f559f22899ab924e220a78119ffc79469b7
Added more tests.
Fixed a regression in Vector.
Fixed bugs in pointer tracking.
Fixed a starvation issue in PollLoop when setting or removing callbacks.
Fixed a couple of policy nits.
Modified the internal representation of MotionEvent to be more
efficient and more consistent.
Added code to skip/cancel virtual key processing when there are multiple
pointers down. This helps to better disambiguate virtual key presses
from stray touches (such as cheek presses).
Change-Id: I2a7d2cce0195afb9125b23378baa94fd2fc6671c
Refactored the code to eliminate potential deadlocks due to re-entrant
calls from the policy into the dispatcher. Also added some plumbing
that will be used to notify the framework about ANRs.
Change-Id: Iba7a10de0cb3c56cd7520d6ce716db52fdcc94ff
The old dispatch mechanism has been left in place and continues to
be used by default for now. To enable native input dispatch,
edit the ENABLE_NATIVE_DISPATCH constant in WindowManagerPolicy.
Includes part of the new input event NDK API. Some details TBD.
To wire up input dispatch, as the ViewRoot adds a window to the
window session it receives an InputChannel object as an output
argument. The InputChannel encapsulates the file descriptors for a
shared memory region and two pipe end-points. The ViewRoot then
provides the InputChannel to the InputQueue. Behind the
scenes, InputQueue simply attaches handlers to the native PollLoop object
that underlies the MessageQueue. This way MessageQueue doesn't need
to know anything about input dispatch per-se, it just exposes (in native
code) a PollLoop that other components can use to monitor file descriptor
state changes.
There can be zero or more targets for any given input event. Each
input target is specified by its input channel and some parameters
including flags, an X/Y coordinate offset, and the dispatch timeout.
An input target can request either synchronous dispatch (for foreground apps)
or asynchronous dispatch (fire-and-forget for wallpapers and "outside"
targets). Currently, finding the appropriate input targets for an event
requires a call back into the WindowManagerServer from native code.
In the future this will be refactored to avoid most of these callbacks
except as required to handle pending focus transitions.
End-to-end event dispatch mostly works!
To do: event injection, rate limiting, ANRs, testing, optimization, etc.
Change-Id: I8c36b2b9e0a2d27392040ecda0f51b636456de25
