Despite SQLite documentation saying these clauses follow the "expr"
syntax, queries will fail with obscure "row value misused" errors,
so remove the strict enforcement for now.
Test: manual
Bug: 111480817
Change-Id: Ib8cdcb4e6456110589203189fcfa14e75cd68718
Previously battery history is kept in the in-memory mHistoryBuffer
with size of 512KB (96KB on low memory device). When the buffer is
close to full we drop certain types of history. When the buffer is
full the whole buffer is reset and we lost elder history. On a
device with long battery life this problem is more frequent.
This CL changes mHistoryBuffer to 128KB (64KB on low memory device).
When the buffer is full, it is saved to history file on file system.
By default we allow 32 history files (64 history files on low-memory
device) which gives us 4MB history (compare to 512KB today).
The MAX_HISTORY_BUFFER and MAX_HISTORY_FILES can be remote configured
through GServices or P/H.
In case of history exceeding 4MB, the oldest history file is deleted
and new history file is open.
This change increases battery history by using disk file and the
chance of losing history is greatly reduced.
Bug: 67297625
Test: adb shell dumpsys batterystats --history
Change-Id: Id9aafea761649d7323b97d1e44135f7880a95414
Developers often accept selection clauses from untrusted code, and
SQLiteQueryBuilder already supports a "strict" mode to help catch
SQL injection attacks. This change extends the builder to support
update() and delete() calls, so that we can help secure those
selection clauses too.
Extend it to support selection arguments being provided when
appending appendWhere() clauses, meaning developers no longer need
to manually track their local selection arguments along with
remote arguments.
Extend it to support newer ContentProvider.query() variant that
accepts "Bundle queryArgs", and have all query() callers flow
through that common code path. (This paves the way for a future
CL that will offer to gracefully extract non-WHERE clauses that
callers have tried smashing into their selections.)
Updates ContentValues to internally use more efficient ArrayMap.
Bug: 111268862
Test: atest frameworks/base/core/tests/utiltests/src/com/android/internal/util/ArrayUtilsTest.java
Test: atest cts/tests/tests/database/src/android/database/sqlite/cts/SQLiteQueryBuilderTest.java
Change-Id: I60b6f69045766bb28d2f21a32c120ec8c383b917
DateUtilsBridge is only used in one place in framework.
The usage is avoidable by sticking to java.util. This
decreases the number of non-public libcore classes in
use by framework.
Test: build
Bug: 111055375
Change-Id: Ifd6576b5b6df1e154f16ec201a4575d0b95de66d
The updates to the HWUI renderer now require that setOverlappingRendering be
set when shadows are used. Without this change modifying the alpha of the
TextView will produce incorrect blending results.
Bug: 111381525
Test: Test app in bug
Change-Id: I523c5ef698e8db912dd2ef848eeeffb3465b990a
Keeping the code in memory of the currently set home app is
important for latency as we don't have any kind of starting
window/splash screen when pressing the home app to hide any latency.
Memory impact:
Pinning dex/vdex:
In practical scenarios, this should be < 500kb.
The home app is usually profile-speed compiled, for which the
resulting dex/vdex files are about 2 mb. However, during regular
use, at least 1.5 MB of it is referenced in memory. This makes
sense: By definition profile-speed only compiles the things that
is usually frequently executed during regular execution.
Pinning apk:
With Launcher 3 in practical scenarios this should be about 3.7 MB,
as the APK is about 5.7 MB but 2 MB are usually referenced in any
case.
Bug: 111132016
Bug: 78585335
Test: Inspect "adb shell dumpsys pinner" after boot.
Test: Check for pinned files after updating camera/home.
Test: Check for pinned files after user switch with different
default apps.
Test: Check for pinned files after bg-dexopt.
Test: Check for pinned files after bg-dexopt + kill pid.
Change-Id: I6cdbc06d089efeb1c72a51216879ba0573502009
Merged-In: I6cdbc06d089efeb1c72a51216879ba0573502009
Keeping the code in memory of the currently set home app is
important for latency as we don't have any kind of starting
window/splash screen when pressing the home app to hide any latency.
Memory impact:
Pinning dex/vdex:
In practical scenarios, this should be < 500kb.
The home app is usually profile-speed compiled, for which the
resulting dex/vdex files are about 2 mb. However, during regular
use, at least 1.5 MB of it is referenced in memory. This makes
sense: By definition profile-speed only compiles the things that
is usually frequently executed during regular execution.
Pinning apk:
With Launcher 3 in practical scenarios this should be about 3.7 MB,
as the APK is about 5.7 MB but 2 MB are usually referenced in any
case.
Bug: 111132016
Bug: 78585335
Test: Inspect "adb shell dumpsys pinner" after boot.
Test: Check for pinned files after updating camera/home.
Test: Check for pinned files after user switch with different
default apps.
Test: Check for pinned files after bg-dexopt.
Test: Check for pinned files after bg-dexopt + kill pid.
Change-Id: I6cdbc06d089efeb1c72a51216879ba0573502009
Settings need LocaleInfo can be serializabled.
So, it can be passed by Intent.
Bug: 111373939
Test: manual test
Change-Id: Ie42c99583688be6fdcec5dd12b105bee90b2577e
- Use the correct state when comparing the association
with the process. (Was causing warning spam as well.)
- Don't wtf if we call startAssociationIfNeeded() no a
ServiceRecord that doesn't have a process... that is normal
when first bringing up a service.
Also add a switch to turn this all off, which for now is
leaving it on.
Test: manual
Bug: 111391911
Bug: 111302807
Bug: 111131998
Change-Id: I2d3f43fd99c7ade24c867ad80a728d898ab0e6c5
The critical methods on this interface - like updateSession() - were already
void, so all we had to do were to "onewaywize" the other methods. We could
either refactor them to be truly async, or implement a blocking mechanism that
let them still be sync *and* oneway - because these methods are not in the
critical path, we opted for the latter, which is simpler and less risky.
Fixes: 73536867
Test: mmma -j ./frameworks/base/apct-tests/perftests/autofill/ && \
adb install -r $OUT/data/app/AutofillPerfTests/AutofillPerfTests.apk && \
adb shell am instrument -w -e class android.view.autofill.LoginTest \
com.android.perftests.autofill/android.support.test.runner.AndroidJUnitRunner
Test: CtsAutoFillServiceTestCases
Change-Id: I380430aa2a7805aed6f629afb360566fc5402abb
Root cause has been identified, but fix is too risky. Instead, we
remove the WTF for now and readd the fix as well as the WTF
in master.
Note that due to defensive programming, in case we land in the WTF
case, it doesn't cause any real bug.
Test: boots
Bug: 110834518
Change-Id: I0da1e48e420c3fcde0e818b7fe0527da9155a159
An Activity may not yet create on client side, there is another
launch request with flags to clear task, then a destroy transaction
is scheduled. If client side keeps blocking until destroy timeout,
the token on server side will be removed. When client begins to
handle the first creation, it will report its activity token to
server that causes IllegalArgumentException because there is no
matched ActivityRecord.
Bug: 32375307
Test: atest FrameworksCoreTests:TransactionExecutorTests
Change-Id: I1b7e0c2863b13091c3fd50df602ff31ae02ff38d
