This reverts commit 5073145f85c8197ec4387d5e2599442a7291ca9c.
I spoke to Dianne and she said that if this is making a problem go away then it is only hiding it and there is a deeper problem we need to fix.
I think if killedByApp is true then we should set app and app.thread to null. I'll look at it further when I have more time.
Change-Id: I44c29e15ca654666440472eab2d2af1c5af80b9a
Symptom: No activity was started for startActivity, low repro rate.
Root Cause: Starting activity on a process which was killed by ActivityManagerService.killUnneededProcessLocked() and before receiving AppDeathRecipient of it.
Solution: Choose the flow of starting a new process if ProcessRecord.killedByAm is true.
Change-Id: Ida2639b0fb4631222fc92d65aadc9fd2da637b45
This patch simplifies the RenderScript JNI layer by replacing six functions
with a single funciton. This new function now handles all previous cases.
Functions in android.renderscript.script have been updated to use this new JNI
function.
Change-Id: I77e4b155cc7ca1581b05bf901c70ae53a9ff0b12
This complements https://android-review.googlesource.com/#/c/103231/
where the instruction set was mapped just for comptilation. The same
should have been done for move,remove and getSize.
The list of dex code ISAs is alo de-duped.
Bug: 16185267
Change-Id: I8fe453a800812e382e8f41b5f7922997aa9c18a9
In the presence of a native bridge it is more efficient to compile the
dex directly to the native ISA than to use the shared library ISA as a
reference.
This can be achieve by configuring the readonly system properties to map
between the .so ISA and the desired dex code .ISA (e.g.
ro.dalvik.vm.isa.ISA1=ISA2).
Bug: 16185267
Change-Id: I50baa7b37e1465b9adf72d6f6b96f526a08d59c7
There was a JNI error where when you got an OOM and called
report_exception, it would call two NewStringUTF in a row without
checking the return values. This could mean that the first one
threw a new OOME and the second one would cause a JNI error when
it also attempted to throw an OOME with a pending OOME.
Bug: 16843627
Change-Id: Ie4f9f9a5f8b7993cd3655d42a6718c0a5e1199f8
This will allow us to conditionally change the compiler-filter based on other properties.
Bug: 15165413
Change-Id: I1676987c69a3cad4217dc4b56465ef3ae886122a
When switching to a newly created user, the user may face this bug when he opens the RecentTask screen.
A possible bug scenario is described as follows:
A user id of a removed user may be recycled when created a new user.
However, mRecentTasks is not correctly controlled so that old information may still remain and be possibly mapped to wrong user.
This patch prevents this bug by explicitly removing old information in mRecentTasks when removing existing user.
Change-Id: I1874dbd604598a5d740ae1e034981e21214c15c6
Signed-off-by: Eunae Kim <eunae.kim@lge.com>
Sympton:
During testing, skip kill native crash process manually because it will continue to die by default.
Root Cause:
Large process may take some time to do coredump.In auto test, crash process will be killed immediately that results incomplete coredump file.
Solution:
If the tester (IActivityController) will handle app crash event,
Do not kill native crashed process if the rom is debuggable.
Change-Id: Ia360af147d694125d440e5ba2f958c4759a50494
Sympton:
Next activity only adds to history but does not launch/resume then results ANR.
Root Cause:
In a rare timing, some windows are switched at the same time,
it will cause some finishing records on the top temporarily,
then set startIt to false that skip to resume the real top activty.
Solution:
If all activities in a task are finishing, do not use it to check.
The behavior/checking is the same concept as in JellyBean:
// If starting in an existing task, find where that is...
boolean startIt = true;
for (int i = NH-1; i >= 0; i--) {
ActivityRecord p = mHistory.get(i);
if (p.finishing) { // <--
continue;
}
Change-Id: I9d81a7b5182400c52e173da23eee61c74692beee
Also, note that this method never worked. Locale
settings were stored with underscores (like Locale.toString)
but matched against AssetManager.getLocales() which
returned language-tag like output.
bug: 10090157
Change-Id: I36ffea6e39ff7d1907b2787f8fd4545c24d326a8
Symptom: Sometimes HardwareLayer display is corrupted.
Root Cause: In some cases, ViewRootImpl destroys HardwareRenderer without
clearing HardwareLayers in the view hirarchy. If the system trims memory fully
and kill the glContext at the same time, HardwareLayer may corrupt.
Solution: Clear Hardware-Layers before system destroys HardwareRenderer.
Reproduce steps: Modify the cache's trim memory mechanism to make it easier to
trim fully, HardwareLayers implemented in app will easily be corrupted.
Modification of the cache is not required but make it easier to reproduce.
Change-Id: If57c802cdacf8dc37cbbcc199e74f9482834f981
Fixes an issue where the stencil buffer is corrupted
(used in overdraw debugging, path clipping) when used in
conjunction with a HW accelerated WebView.
bug:15023700
bug:16465551
Change-Id: If201b503931380f7107a34e9a9a2f877a50f72b1
If an application that has a broadcast receiver is killed
during broadcasting, thread variable of ProcessRecord becomes null
so that IIntentReceiver#performReceive() is called
in BroadcastQueue#performReceiveLocked(). But if binder driver has not
noticed the death of the application yet, it can't throw
DeadObjectException. After that, binder driver notices. But it can't
notify DeadObjectException to the caller because performReceive() is
async call. So broadcasting keeps on waiting for finishing
performReceive() until timeout.
This change checks the death of the application before calling
performReceive() and skips broadcasting to the receiver
if the application has already died.
Change-Id: Ifa02b8b1a7e7b6fd314de90fedff5b7a5326825d
