The DIRTY flag is used to track which elements of the view hierarchy need
to be redrawn on the next drawing operation. This flag is set on the parent
hierarchy of a view when that view is invalidated. There is an optimization for
opaque views that tells the parent that it is dirty, but that it need not
redraw its own content because the view will cover it (since it is opaque).
This dirty-opaque logic breaks down in the current code because we only set
these dirty flags on the parent hierarchy, not on the view itself. In the situation
raised by this bug, we would invalidate the parent container directly (which does
not case the dirty flag to be set), then we would invalidate a child of that view.
Because the child is opaque, the DIRTY_OPAQUE flag would be set on the parent
container. This would cause us, in the later rendering process, to skip the
drawing on that parent container, assuming that it was only asked to be redrawn
because of its opaque child's invalidation.
The fix is to now set the DIRTY flag on an invalidated view, not just on its parent
hierarchy. The DIRTY_OPAQUE logic will avoid setting the opaque flag on views/parents
that are already marked DIRTY, thus an invalidated parent will be correctly drawn
during the drawing process.
Change-Id: Ib5e014a125a4f5168f6a5a3166e941474659aca5
1. There was a bug that was not handling correctly the
case for which the interrogator requests an accessibility
node info and the message describing how to fetch the
latter for the same process case was delivered after the
code that checks whether the message is there in order to
dispatch it is executed. Now the message handling is done
correctly - the caller checks if the message is present and
if so processes it, otherwise the caller sleeps and is
interrupted if such a message arrives.
bug:5138933
Change-Id: I4c2940b46c9a52a51c5ee48b83ca6811489765d6
External storage volumes that were emulated+encrypted needed to have
their encryption mapping removed so that it doesn't try to encrypt the
volume after formatting them.
This just wires through an argument through vold, and assumes that vold
will do the right thing even if there is no encryption mapping set.
Bug: 5017638
Change-Id: I858fae3d12cb415bc34637f520f71220ad9daaad
Use the URL host and path rather than the complete url to store
form autocomplete data. This helps in the situation that a site
uses some dynamic query string on the page that contains the form.
Also set the autoocmplete threshold to 1 so that we don't flick the
autocomplete options up and down as you type the first few characters.
Bug: 5265606
Change-Id: I7b372400062ae34f70a78b786007910dc179b101
* commit '2295cc77c72d451e7a1c9ec48f120a8e73651b6d':
Fix bug #2246815 Error popups on EditText should span the width of the screen instead of the width of the EditText field
* commit '16e1678df48f6de4eb602f9cd18baa1e40b1181f':
Make the SyncManager have wakelocks per account, not per account type, otherwise the parallel sync behavior of gmail can cause the first account to finish to release the wakelock that is used for the other account, which will result in the device going to sleep before the sync is complete.
