The callbacks for the notification listener and notification
ranker were delivered on binder threads which is problematic
becuase: 1) permission checks and app ops checks would fail
unless the app developer knows to clear binder calling id and
restore it after that; 2) developers need to synchronize their
implementation as they get callbacks on different threads (
arguably callbacks should not be concurrent); 3) this doesn't
follow the pattern in the platform;
Also the code delivering callbacks was catching Throwable which
we shouldn't do in general and also masks bugs in the listener
or ranker implementation. Now that the callbacks are offloaded
to the main listener/ranker thread system code should not be
guarding against Throwable to handle exceptions propagated
over binder calls.
bug:26704777
Change-Id: I171fb41bbe25e6105dd05e4166193dbcec594f82
Unbundled apps should not be allowed to share
linker-namespaces with default namespace; it implies
that they will not have unlimited access to private
native libs once updated.
Bug: http://b/27189432
Change-Id: I60295b9b31d700016add93537418fdcce5a77719
This reverts commit 90bd36363c.
Seems that the semantics of InputConnectionWrapper#setTarget() is more
complicated than I thought. At least the following cases have worked
fine.
case 1:
InputConnectionWrapper wrapper =
new InputConnectionWrapper(null, false);
wrapper.SetTarget(ic);
...
case 2:
InputConnectionWrapper wrapper =
new InputConnectionWrapper(null, true);
wrapper.SetTarget(ic);
...
case 3:
InputConnectionWrapper wrapper =
new InputConnectionWrapper(ic, true);
wrapper.SetTarget(null);
wrapper.SetTarget(ic2);
...
The previous code did not intended to break existing code. Let's revert
it we decide how to deal with above cases.
Bug: 27407697
Change-Id: I8bc84d484ab0b27a02e74f11110430f70646e69a
This CL makes it clear that InputConnectionWrapper does not support null
target. In other words, the semantics of null InputConnection can never
be emulated by a non-null InputConnectionWrapper.
This is particularly problematic when app developers are just forwarding
the return value of super.onCreateInputConnection() to
InputConnectionWrapper or its subclass, because there are many chance
that super.onCreateInputConnection() starts returning null, e.g. when:
A. the application is extending a Framework class, and the Framework
class is updated by OTA.
B. the application is extending system WebView, and the WebView is
updated.
C. the application is extending a 3rd party library, and the app
developer creates a new build with a new version of the 3rd party
library.
To make it easy to catch these kind of bugs, this CL lets the
constructor of InputMethodWrapper throw NullPointerException when target
is null. Bugs like crbug.com/571229 should be caught by developers
more easily.
Bug: 27407697
Change-Id: I83875bea886d4784f9507c930050efc29708d9db
...isUserAMonkey for testing purpose
Add an argument for the caller to specify if they are a poo flinging
monkey.
Change-Id: I0e149a8d78776abaf07517bd4ae886047b7f4252
Sometimes pointer change request is delivered after view is detached from its
ViewRootImpl. E.g. when popup is present click outside to close it.
Bug: 27292939
Change-Id: I925728af334a1e1ae53f7e530d639e50b0c37f2b
When decoding from a file descriptor, if the descriptor is seekable and
has no offset, use SkFILEStream directly instead of wrapping in an
SkFrontBufferedStream.
This will let the SkRawCodec take advantage of a seekable stream.
One can see 2x speed up for RAW cases.
Bug: 27097104
Change-Id: I369fbb3af1170c94ec7cd9bb35e8e0007dde9fd5
Ensure we're using file paths from the correct package. Also cleans up
synchronization to avoid synchronizing on local variables which, while
not an issue in the current implementation, is generally unsafe.
Bug: 27313689
Change-Id: Ib8550909e7b02ea88d19ce2fc43756a16f9664ab
Similar to first patch, but now using new "rethrowFromSystemServer()"
method which internally translates DeadObjectException into
DeadSystemException. New logic over in Log.printlns() now
suppresses the DeadSystemException stack traces, since they're
misleading and just added pressure to the precious log buffer space.
Add some extra RuntimeInit checks to suppress logging-about-logging
when the system server is dead.
Bug: 27364859
Change-Id: I05316b3e8e42416b30a56a76c09cd3113a018123
* Don't race with construct/init; we can't abide partial state here
* Defense in depth: null check the contained native state to catch
double invocations.
Bug 27377823
Change-Id: I38f91668d36404c2a7b93d5f7c6616d9eb3f0a30
Clones notifications before adding them to remote views so
the original stays parcelable on its own.
Also prevents the compatibility inflation from triggering when
an app uses a decorating style to prevent recursive wrapping.
Those styles only exist on N and later anyways.
Also fixes the compatibility inflation in listeners.
Bug: 27368615
Change-Id: Iedf3036bf315dd9c7b476c7e8bcce57de5b5c9c8
Bug: 27286867
If the system/app is slow, it might take too long to
stop drawing. Switch the ordering of destroying stuff so
that we switch to the pbuffer surface first, then do
cleanup
Change-Id: If64a3dbb71bb9fd53567231590436a89b2f1a09e
I forgot to include Parcelable.Creator when I added SearchView
iconified state saving.
Bug: 27357167
Change-Id: Ia96490aa9c12c4042f50ba4b1aeb7422b37a1d4c
Remove an unused field in BoringLayout. This has also the positive
side effect of allowing static initialization in the boot image.
Bug: 27248115
Change-Id: I9d6beec5f5e045ef899e8979ece96e0239f82076
The registered shortcut will be called from PhoneWindowManager,
before dispatching
Change-Id: If26128939b45a639c8895719a7a23ca433f39fd9
(cherry picked from commit 4da863c5a8872dcabb179a978a2b2157d9081679)
