When reporting views to accessibility services we are ordering the
children in a parent based on their location on the screen. The
initial implementation worked pretty well in practive but violated
the transitivity property leading to rare crashes in apps. The current
implementation does not violate transitivity but does not produce
good ordering.
Given the lack of time and to minumize risk this change uses the
old strategy which works most of the time and if that fails we
fall back to the current strategy. Coming up with a correct strategy
that produces good results requires more time.
bug:17887986
Change-Id: I1c233ecdf318befc315e793696ac48bd6c652ab6
Negative margins can cause all sorts of other problematic assumptions,
subtle errors and unexpected behavior. While they technically work
developers should steer clear of them.
Change-Id: Ia1f7d9a33e25444ebdc3c32679ae1179210558e1
Help developers during debugging bad calls to addView by throwing an
IllegalArgumentException earlier rather than implicitly tossing NPE
the first time the child view is accessed.
Change-Id: I4a554635815a5d0b415f637b93592fb45973f26d
Bug: 17765082
DeferredLayerUpdater had fallen behind RT updates. Re-snap to
latest expectations, ensuring to call requireGlContext() prior
to detachSurfaceTexture to avoid leaking SurfaceTextures
Change-Id: Ic65fb9831e5284f658866da8da9ad5af1d227699
Bug: 17578553
Games seem to be doing this to prevent destruction of their
GL contexts, and they assume it works even if it doesn't. However,
GLSurfaceView is clunky here, so while the app is doing something
questionable we don't really offer a better way. For now revert
back to kitkat behavior.
Change-Id: Icfa9e496279b9cfa47f9bc7f6848d9313caed0d5
The motivation is an API change: FloatMath is going to be
deprecated and/or removed. Performance is not the goal of
this change.
That said...
Math is faster than FloatMath with AOT compilation.
While making the change, occurances of:
{Float}Math.sqrt(x * x + y * y) and
{Float}Math.sqrt({Float}Math.pow(x, 2) + {Float}Math.pow(y, 2))
have been replaced with:
{(float)} Math.hypot(x, y)
Right now there is no runtime intrinsic for hypot so is not faster
in all cases for AOT compilation:
Math.sqrt(x * x + y * y) is faster than Math.hypot(x, y) with
AOT, but all other combinations of FloatMath, use of pow() etc.
are slower than hypot().
hypot() has the advantage of being self documenting and
could be optimized in future. None of the behavior differences
around NaN and rounding appear to be important for the cases
looked at: they all assume results and arguments are in range
and usually the results are cast to float.
Different implementations measured on hammerhead / L:
AOT compiled:
[FloatMath.hypot(x, y)]
benchmark=Hypot_FloatMathHypot} 633.85 ns; σ=0.32 ns @ 3 trials
[FloatMath.sqrt(x*x + y*y)]
benchmark=Hypot_FloatMathSqrtMult} 684.17 ns; σ=4.83 ns @ 3 trials
[FloatMath.sqrt(FloatMath.pow(x, 2) + FloatMath.pow(y, 2))]
benchmark=Hypot_FloatMathSqrtPow} 1270.65 ns; σ=12.20 ns @ 6 trials
[(float) Math.hypot(x, y)]
benchmark=Hypot_MathHypot} 96.80 ns; σ=0.05 ns @ 3 trials
[(float) Math.sqrt(x*x + y*y)]
benchmark=Hypot_MathSqrtMult} 23.97 ns; σ=0.01 ns @ 3 trials
[(float) Math.sqrt(Math.pow(x, 2) + Math.pow(y, 2))]
benchmark=Hypot_MathSqrtPow} 156.19 ns; σ=0.12 ns @ 3 trials
Interpreter:
benchmark=Hypot_FloatMathHypot} 1180.54 ns; σ=5.13 ns @ 3 trials
benchmark=Hypot_FloatMathSqrtMult} 1121.05 ns; σ=3.80 ns @ 3 trials
benchmark=Hypot_FloatMathSqrtPow} 3327.14 ns; σ=7.33 ns @ 3 trials
benchmark=Hypot_MathHypot} 856.57 ns; σ=1.41 ns @ 3 trials
benchmark=Hypot_MathSqrtMult} 1028.92 ns; σ=9.11 ns @ 3 trials
benchmark=Hypot_MathSqrtPow} 2539.47 ns; σ=24.44 ns @ 3 trials
Bug: https://code.google.com/p/android/issues/detail?id=36199
Change-Id: I06c91f682095e627cb547d60d936ef87941be692
When the status guard is not available, (e.g. when the window is not
drawing system bars) do not consume the insets when an action mode is
active.
Bug 17691453
Change-Id: I459622eaf161a96152fb9ac5f60bb9508aa4de89
When device is encrypted the user has to authenticate in order to decrypt
the data partition which is required for running accessibility services
and Text-To-Speech. In order to address this issue we are falling back
to use the default password if there is an enabled accessibility service
and the user has secure lock. This will enable the user to authenticate
when accessibility layer is completely functional.
bug:17671790
Change-Id: Iafffe7bcd234008cf91ffb5011b21b803dca227a
Bug 17671834
GhostViews can be added in arbitrary order since they are added
depending on how they are insterted into the TransitionValues
maps. This CL does two things: it ensures that GhostViews are always
pulled to the top of the Overlay and inserts GhostViews into
the Overlay in the order that their ghosted views are drawn.
Change-Id: I9f68105126834cc8f30a6cfe5d58fc3c51465afd
If the virtual child id is not specified we should ask the node
provider to create a node info for the hosting view bu passing
the special host node id.
bug:17664111
Change-Id: Iaf1387acdfe33b50b50729d293adcdf82340a648
If the virtual child id is not specified we should ask the node
provider to create a node info for the hosting view bu passing
the special host node id.
bug:17664111
Change-Id: I9c2d3195ad813eed7ea4108fe0d7ccb9cc32a1a0
When getting an accessibility node info we also prefetch its predecessors,
siblings and descendants up to fifty. If a view has a node provider, i.e.
virtual descendants, and the id of the virtual descendant is not specified
we use a speacial constant to refer that the provider should create a node
for its hosting view. The bug was that we were not using the correct constant
when calling into the provider.
bug:17664111
Change-Id: I4f6b2333e1e7e3e2bb6b164c0752613417c60d16
In accessibility mode we compute a location to where to send down and up
events to click a view (triggered by a double tap). This solves the problem
of interacting with views that are partially covered by other interactive
views. The logic that computes the click point was not ignoring siblings
that are not visible. As a result a blind user cannot interact with some
views.
bug:17632224
Change-Id: I93e637ee6702e4ba3e79faa736205430e2196d01
Once upon a time when the world was fresh and new, the heavens
had an easy rhythm. Day and night. Night and day. In the day,
the pixel fairies would cavort and play in the bright gardens
with narry a mark of shadow or gloom. In the night, they would
rest peacefully, dreaming no dreams and knowing no fear.
Then one night a fairy dreamed the first dream. At first
the dream was peaceful, full of colors and delight, hopes and
memories. Then all at once, jarringly, it awoke in bright
daylight. The pixel fairy knew fear, for the world had changed
and it was unprepared.
Time passed and the pixel fairies grew accustomed to their
fate, day and night, night and day, sometimes dreaming, until
there came a night when a fairy did not sleep. It roamed
the land in a dreamless doze, lost and afraid amid a grim haze
of grey and darkness. The fairy despaired. It wanted no
part of this place. It pretended for a time to be awake but
the bright daylight would not come. It pretended for a time to
be dreaming but the colors and memories would not come.
That is when the fairy wished for oblivion. Then just as
suddenly, it awoke in the daylight. It fell to the ground,
stunned as if it had forgotten how to walk in the too bright
daylight.
Though the world again grew softer and kinder in time, the pixel
fairies were never the same. For the night is dark and full
of terrors.
---
It used to be easy. Screen on and screen off could explain almost
everything about the state of the device but it's different now with
ambient display. We need to be able to wait for all windows to be
drawn even in the case where the device is still nominally asleep.
In truth, the window manager policy which drives a lot of these
interactions is a thicket of outdated assumptions.
Added a new method to tell the window manager policy when the screen
is being turned off so that it can correctly account for changes
to the interactive state (wakeUp and goingToSleep) and screen state
(screenTurningOn and screenTurnedOff). Now we can independently
poke keyguard during interactive state changes and we can apply
screen on blocking during screen state changes.
Moved the code which manages screen on blocking (which is what
ensures the UI has fully drawn before revealing screen contents)
from the power manager to the display manager since the display
manager is in a better position to accurately track the state of
the screen, particularly when the screen is being turned off.
Fixed a bunch of synchronization issues. Previously some work
had been moved to a handler without considering what might
happen if it became reordered relative to other work happening
elsewhere. Documented the desired behavior in the code to
prevent this from happening again.
There's still a bunch of stuff in here that isn't quite right,
particularly the assumption that there's only one screen, but
it's good enough for now. Hopefully there aren't too many bugs.
Bug: 17605802
Change-Id: Ic7319e09948c8a3cda014d7e169c964a3ad86f14
Bug: 16978006
Don't HWUI-accelerate KeyguardScrim
Aggressively trim memory as soon as a ViewRootImpl
dies or has its visibility changed.
Change-Id: Ie1b7c9d30653456bd2e9f309128174f972999368
