They were ultimately caused by killing Launcher process from tests.
Now having a test info handler request to clear db.
Bug: 152629799
Change-Id: Ia81ddc3e338718c4cff08c7396b9fda1b7091024
-> Synchronize calls in LoaderTask.java
issue 156041043
-> Remove non system users on setup
If a work test crashes before getting to run its teardown - we might end up with a user profile that could throw off subsequent tests.
issue 156022161
Test: Manual
Change-Id: Ife708a3de01572f7cb2187078d592d8d570dd951
This is tightening the makeshift strictmode criteria.
Starting with this moment, we will know that there is
no memory growth during tests execution, which is a big deal.
Big: 139137636
Change-Id: I5edc84524463bd1736d727496ad0fc031bb9624c
Strictmode leak detector is still a goal, but we might not be able to
achieve it in R. Strictmode has several framework-side bugs that perhaps
hide Launcher-side strictmode violations, while the time to fix
everything is limited, and new leaks get introduced all the time.
For now, implementing a check that is slightly more relaxed than
Strictmode, but still ensures the absence of leaks. I’ll keep
eliminating Strictmode violations as well as keep strengthening the
makeshift checker conditions until we’ll be able to enable Strictmode in
continuous testing.
I’m disabling Strictmode checks for now so that they don’t generate
unnecessary hprof dumps, but leaving the code dealing with strictmode.
Bug: 139137636
Change-Id: Ib10136b0d4e9892f70a19cd052ae5a54cf0a4efb
Now doing this before branching points, thus avoiding flakes when the
execution can go to an unexpected branch and not produce an event.
Bug: 153824894
Change-Id: If117da0498eaf2d94c9610552724981be34c6569
Slight revert of ag/10668129 with adjustment
of disabling it for tests.
Fixes: 151456795
Test: Ran the labtest command for OOP
tests for crosshatch (where this issue
was first detected)
Change-Id: I315d138c2e4a6d4068304e9b5fb2e1b7feb34e63
It also turned out that Pilfer event seems to come in a
non-deterministic order relative to the events from the Main and TIS
sequences. So I moved it to its own sequence.
Change-Id: Ie4ea5865afd900bebbd8287dad2372c94dce8ad5
Unlocking the device at the beginning
of the test does help, but perhaps we should prevent
locking during the test execution.
Some failures still remain.
Bug: 151613234
This is a workaround for the infra not unlocking
the device properly.
I don't have access to may machine, so
I've created this CL in Gerrit editor, didn't test,
but it should work.
Bug: 151613234
- hide overlay icon in landscape mode
- don't show edu if user has already seen legacy work profile edu
- make sure personal tab is highlighted when work profile is reinstalled
- always go home after a work profile is added or removed
- add tests for work edu flow
Bug: 150122946
Test: Manual
Change-Id: I8f80ac763acf03ca31a534464f4ddfd84528d329
This replaces PendingAppWidgetHostView with configured widget view when
OPTION_APPWIDGET_RESTORE_COMPLETED is set to True by a widget provider.
Bug: 63667276
Test: Manual
Change-Id: Ide21f5e9a7dac7e3d6a745660a38ad0b951b47d3
Can do this via API with the exact heap state that triggered the
callback because using API allows to avoid deadlock that we'd have in
main thread if we used shell command.
checkDetectedLeaks() is still needed, as I'll add throwing an
exception there soon.
Bug: 139137636
Change-Id: I3c8670071b058c5c146ac58bb16152c980553065
This is perhaps not a leak, but it conflicts with how StrictMode
activity leak detection works.
Bug: 139137636
Change-Id: I1608e348d8b238142c42ed81ae125fd07ef6b673
WM is making changes which allows apps to maintain
their orientation independent of the orientation of
the foreground app. This allows recents to always start
in portrait even when the app currently running is in
landscape. This means we have to give the illusion of
a landscape oriented overview when user swipes up in
gesterual nav when launcher is started in portrait
configuration.
PagedOrientationHandler abstracts all coordinate specific
logic from Paged/RecentsView primarily, but also all
other dynamic calculations throughout launcher.
PagedViewOrientationState is the single point of exposure
to other classes that depend on those changes. The goal
is to also minimize holding state to allow for default
implementations of PagedOrientationHandler for all the
3p/Fallback classes. PagedViewOrientationState also
holds other data around rotation that isn't
specifically tied to view logic.
The fake landscape overview can be toggled with:
adb shell settings put global forced_rotation [0/1]
Fixes: 146176182
Change-Id: I65d8d4e9f92b93931cbe0053ccaf0cda8d2ffd6c
