| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
| |
The hit test on the nav cache returns a node/frame pair.
It looks for the closest hit, but gives priority to a direct
hit. Track the frame associated with the direct hit
separately, so that the correct node/frame pair is returned.
Change-Id: Icb1e3de4a0aad3c6dd9b2b81669f9c7bbb260282
http://b/2316138
|
| |
|
|
|
|
|
|
|
|
|
|
| |
A fix for http://b/2319610 incorrectly retried finding
nodes if the node was clipped, possibly causing infinite
recursion. Leave all of the prior fix but only try once.
I tested the prior test case and ensured that it still
works.
Change-Id: Ib229b6fc0ba57c131a8c1f33350982ac22a445c0
http://b/2501914
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A layer may be inside or outside of an anchor. If it is inside,
the corresponding CachedNode has already been created, but is
being tracked. These tracked nodes need their layer and unclipped
bits set. For now, node in layers are assumed to draw last in
their layer, and aren't obscured by other drawing.
We may allow nodes in layers to be obscured one day. Preparing
for that, translate the layer picture by its global position
when testing to see if the node is hidden.
Fix debugging by allowing CachedLayer to see inside LayerAndroid.
http://b/2453974
|
| |
|
|
|
|
|
|
|
|
|
| |
- get rid of the FloatPoint interface in LayerAndroid;
use (x, y) instead
- make CachedFrame a friend of CachedRoot and CacheBuilder
a friend of CachedNode so they alone can access private fields.
- assume the LayerAndroid picture can sometimes be null.
If it is, use the main page's picture instead.
http://b/2369549
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
companion fix is in framework/base
With fixed layers, parts of the web page are now
in motion relative to the document when the page
scrolls. Many routines that formerly read static
coordinates need to compute locations. In some cases,
new computations are cached for speed -- for instance,
the current cursor position is cached when it is
frequently compared.
The cursor rings and other drawing elements like
finding text on the page now to be drawn in the correct
order so that they appear both under and over layers.
There's quite a bit more work to be done. Major
pieces are drawing the text selection in the correct
order, and computing locations based on nest layers.
With this checkin, only the position of the child-
most layer is considered when computing bounds.
http://b/2369549
JavaScriptCore/wtf/Platform.h
- Turn compositing on. All routines that
reference LayerAndroid are bracketed by this
condition.
WebCore/platform/graphics/android/LayerAndroid.h
WebCore/platform/graphics/android/LayerAndroid.cpp
- Add a unique id to each layer. The unique id is
used to associate a layer created when the DOM is
parsed in the webkit thread with its copy in the
UI thread.
- Add: draw the text found on the page, as a call
out in the primary draw. The call out must follow
the drawing the layers' contents to show the found
text correctly. Note that this adds a new slot with
identical contents in every child LayerAndroid. In
a future optimization, a RootLayerAndroid could hold
data common for all child layers.
- Add: clipArea(), which returns an array of rectangles
describing the clip for this LayerAndroid and its
children. Generally, this is the part of the webpage
which is covered by one or more fixed layers.
- Add: find(FloatPoint) that returns the deepest layer
that contains this point. This is used to match taps
to the layer that is tapped on.
- Add: draw all layer pictures and identify
which layer is being drawn. This is used to analyze
the picture contents for finding and selecting text.
- Add: find the layer that matches a given id; this is
used to map cached DOM node data back to the layer
that contains it.
- Fix up includes, delete unused interfaces
WebKit/android/jni/WebViewCore.h
WebKit/android/jni/WebViewCore.cpp
- Remove local mRootLayer; use the one in WebView.cpp
instead (which is in sync with WebView.java)
WebKit/Android.mk
WebKit/android/nav/CachedLayer.h
WebKit/android/nav/CachedLayer.cpp
- CacheLayer associates the cached node with the
LayerAndroid instance. It contains the index to the
node in the cached frame, the LayerAndroid's unique id,
and the spacial offset of the node within the layer
when the DOM information was captured. It also caches
a pointer to the LayerAndroid instance.
CacheLayer computes the node's location each time
it is called, since the fixed layer may be constantly
moving relative to the document's coordinates.
WebKit/android/nav/CacheBuilder.h
WebKit/android/nav/CacheBuilder.cpp
- Track the active layer while building the nav cache.
The 'Tracker' structs were refactored to share common
code, and a new 'LayerTracker' struct identifies when
the node walker is inside a layer.
- Added code to dump layer information for debugging.
- Note that CachedNode::cursorRingsPtr can only be called
during nav data construction
- The cache builder can limit or exclude nodes that
are clipped out -- but until I have more understanding
of layer clipping, treat contained nodes as unclipped.
WebKit/android/nav/CachedDebug.h
- Add a variant that can dump either to a log file
or the console including the function it was dumped from.
WebKit/android/nav/CachedFrame.h
WebKit/android/nav/CachedFrame.cpp
- Add an array of CacheLayer instances.
- Protect bounds from direct access since they must
always be computed.
- Remove misnamed focus parameter from many routines
since the cursor node can be read from the root frame.
- Add: adjustBounds(), which computes the bounds as the
layer moves.
- Add: checkRings(), which gets the appropriate picture
for the node.
- Remove disabled code
- Find the layer list for the matching node by
using a binary search
- Add: resetLayers() to reset the LayerAndroid pointer
in CachedLayers when the layer world changes.
WebKit/android/nav/CachedHistory.h
WebKit/android/nav/CachedHistory.cpp
- Update history data to have matching frame and node
WebKit/android/nav/CachedNode.h
WebKit/android/nav/CachedNode.cpp
- Refactor functions that directly read coordinates
to compute them. In some cases, pass the frame in so
that the layer coordinates can be found.
- Add a bit to note that the node belongs to a layer.
- Remove duplicate bounds interfaces.
- Add methods to get cursor ring data at runtime.
- Update debugging info.
WebKit/android/nav/CachedRoot.h
WebKit/android/nav/CachedRoot.cpp
- Isolate direct picture access so that the layer
picture can be returned.
- Add knowledge of how the base is covered by layers.
- Add a pointer to the root LayerAndroid.
- delete disabled code.
- Move the cursor ring into view if it is obscured
by a layer (this isn't totally working)
- Before finding the next node to move to, set
up 'cursor cache' data, including the visible picture.
WebKit/android/nav/FindCanvas.h
WebKit/android/nav/FindCanvas.cpp
- Move find code here so that it can be called from
layers.
WebKit/android/nav/WebView.cpp
- Add java interface to get viewport metrics on demand.
- Pass frame with the node.
- Remove the find on page code (now in FindCanvas).
- Compute focus rings instead of reading them directly.
- Transfer layer id when getting new nav cache.
- Set up root LayerAndroid.
- Add utility to track if cursor is in a layer.
- Simplify drawLayers() to use common view metrics.
|
| |
|
|
|
|
| |
This code was written because we previously only had one action
button on the IME for textfields. Now that textfields can always
have a next button, we no longer need it.
|
| |
|
|
|
|
|
| |
Also, if a smaller object, like text, is drawn over a occluding rectangle,
ignore it, and don't reset the occluded state.
fixes http://b/2319610
|
| |
|
|
|
|
| |
like one.
Fix for http://b/issue?id=2368868
|
| |
|
|
| |
Fixes http://b/issue?id=2201866
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fix for http://b/issue?id=2219166
Requires a change to frameworks/base
Remove the old change to update the WebTextView when a key is pressed,
since the IME does not generate key events. Instead, when the focus
changes, and the IME is serving the WebTextView, immediately clear
the cursor and update the WebTextView, so the user can continue
typing. Also, allow "Next" to work on the currently focused
textfield, even if it's not the cursor. Further, check for
a new action if there is a focus but not a cursor.
|
| |
|
|
| |
Fixes http://b/issue?id=2210152
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Help to fix http://b/issue?id=1890360 and http://b/issue?id=2150538
CacheBuilder.cpp:
Explicitly set isTextField to false for textareas.
CachedRoot:
Remove the code which checks to see if the textfield is a search,
since if it is, we can avoid this path altogether.
WebView:
Return a single integer which tells what type the current text
input field is.
Requires a change to frameworks/base.
|
| |
|
|
|
|
|
| |
SEARCH <input>
Fixes http://b/issue?id=2299660 and http://b/issue?id=2299650
Also remove isPassword, which is redundant.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Some of the fields in CachedNode are relevant only to input
fields and text areas. Move these into their own vector
so that we can add more data without making all CacheNodes
bigger.
Remove CacheNode entries that are no longer used, or can
be consolidated into the node type. Alphabetize some interfaces
and implementations. Update the debugging output.
part of http://b/2299660
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
These files have not yet been upstreamed to webkit.org.
WebKit requires either a BSD-style or LGPL 2.1 license for all code.
We use a BSD-style 2-clause license for Android-specific files that will be upstreamed to webkit.org.
This change adds licenses where absent or simply fixes the names of copyright holders in the license
text to 'THE COPYRIGHT OWNER' and cleans up formatting.
Files in stl/ currently use licenses other than BSD-style and will require more careful treatment.
Change-Id: I67ad4b8932e432d3eaaeecdfeb0d09418496228d
|
| |
|
|
|
|
|
|
|
|
|
| |
Remove some code that simulates a mouse click at the beginning of
a textarea and the end of a textfield. The original goal was to
make the click change the selection to be at the beginning or end
of the field, respectively. However, we actually make another call
which prevents this click from the selection. Further, the
selection actually gets changed elsewhere.
Fixes http://b/issue?id=2219233
|
| |
|
|
|
|
|
|
|
| |
Most of the time, the simulated mouse position can be
computed from the current cursor. But when the cursor
is changed, the current cursor info could be out of
date and generate a bus error.
fixes http://b/issue?id=2061211
|
| |
|
|
|
|
| |
Fix for http://b/issue?id=2136097
Change-Id: I8ca00498dd5a8d611bc0de5a941faaf596d9f76d
|
| |
|
|
|
|
|
|
|
|
|
| |
When a focused node is found while walking the DOM, the
CachedFrame tree doesn't have its parent pointer set.
Fix the focus index in the parent when fixing the parent pointer.
Also fix a debug statement so that it doesn't depend on
the focus parent.
Fixes http://b/issue?id=2048186
|
| |
|
|
| |
fixes http://b/issue?id=2048180
|
| |
|
|
|
|
|
|
|
|
| |
Zooming in to yahoo.com fails sometimes if the block of text has
a large bitmap on the left. This change prevents the bitmap from
being seen as a part of the text paragraph.
Also, cleaned up the debugging statements.
Fixes http://b/issue?id=2066236
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
While the page is loading, the nav cache is not rebuilt.
Double-click zooms out the web page by using the nav cache to
find the left edge of the column -- but fails to work during
page load. This change rebuilds the nav cache (if the page is
loading) each time a mouse click is sent to webkit.
This doesn't fix the bug where the first double click doesn't
align the column correctly, but helps with subsequent clicks.
Also, pass scale information to getBlockLeftEdge so it can
restrict its search to the area that will be zoomed to. Default
to the point clicked if no alignment info can be found.
|
| |
|
|
|
|
|
|
| |
In CachedFrame, add methods to find the next textfield and to
determine which ImeAction should be associated with a given textfield.
In WebView, uses these apis to determine the ImeAction and jump
to the next textfield and scroll it on screen. Requires a change to
frameworks/base.
|
| |
|
|
|
| |
The function is currently unused, but will be used by
Grace's double-tap experiment.
|
| |
|
|
|
|
|
|
| |
Add clicking, key debug statements in WebViewCore.cpp.
Fix dumping nav tree to use NamedNodeMap.
Set click point to be start of text field, end of text area, middle
of others.
Make CachedRoot::getSimulatedMousePosition, callers const
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
WebViewCore.cpp:
Add parameter to CachedRoot::findAt to suppress check for
hidden nodes. This helps find a best match for newly built
nav caches.
Require that the new cursor node closely match the bounds
of the prior cursor node. This may need tuning.
CacheBuilder.cpp:
Remove isInput flag on cached nodes (no longer used)
CachedFrame.cpp:
Add hideCursor to complement clearCursor. Hide prevents
the cursor from drawing but does not move it. Clear removes
it altogether so that the next movement starts from the viewPort
edge.
CachedHistory.cpp:
Don't special case text fields when doing history navigation.
This special casing in part allowed setting the focus to
the homepage input, but since focus is no longer set by nav,
it is not required.
CachedNode.cpp:
Add hideCursor; clean up debugging
CachedRoot.cpp:
Use navBounds from history instead of cursor bounds to
determine next move. Clean up some obsolete code.
WebView.cpp:
Add hideCursor; call it when appropriate.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Older code treated the trackball as a four way dpad with
equivalents to moving up, down, left and right by generating
arrow key events. This change makes the trackball solely
generate mousemove events.
The old arrow keys in turn were mapped to be as close as
possible to tab-key events that moved the focus. The new
model leaves focus-changes to the DOM.
Clicking the dpad is distinguished from pressing the enter
key to be more compatible with desktop-authored web pages.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Check to see if the potentially larger hit-test bounds
can be used in place of the normal bounds, or if the
normal bounds can be used in place of the individual
text bounds.
Construct a region out of the individual focus ring
rectangles, then see if any text is drawn inside the
bounds but outside of the focus ring. If not, use one
rectangle instead of the rings.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
