GitLab

CachedUAStyles are large (160b) and not cheap to create because they
include several Length objects which have non-trivial destructors,
making CachedUAStyles expensive to destruct. This patch reduces the
number of CachedUAStyles by only creating them when needed.

There are two changes in this patch:
1) CachedUAStyle& is no longer a member on StyleAdjuster but is
instead passed in through StyleAdjuster::adjustRenderStyle.

2) The cached UA style is now constructed as-needed in
StyleResolverState::cacheUserAgentBorderAndBackground.
StyleResolverState always owned the CachedUAStyle but this patch
makes it explicit by changing the type to OwnPtr<CachedUAStyle>. A
null check was needed in RenderTheme::adjustStyle but otherwise
CachedUAStyle was not accessed without being set.

Before patch:
  facebook.com - 19,850 CachedUAStyles constructed
  wikipedia cat - 22,686 CachedUAStyles constructed

With patch:
  facebook.com - 538 CachedUAStyles constructed
  wikipedia cat - 16 CachedUAStyles constructed

Review URL: https://codereview.chromium.org/338673002

git-svn-id: svn://svn.chromium.org/blink/trunk@176163 bbb929c8-8fbe-4397-9dbb-9b2b20218538

With OOPI, the main frame can be either local or remote. Features that depend
on the main frame being local will have to be updated to account for this fact.

This patch takes an aggressive approach--most locations that depend on main
frame being local have been updated to directly attempt to downcast to
LocalFrame, with no attempt to check that the main frame is local first. The
only places that check are those on the critical path towards getting
rendering/input events working with the updated OOPI infrastructure.

BUG=346764

Review URL: https://codereview.chromium.org/336553003

git-svn-id: svn://svn.chromium.org/blink/trunk@176162 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Also fixes a real bug where we reversed the order of the
type/release/debug columns for the list of failing builders,
but we didn't adjust cellByBuildType or the tests to account
for it. Instead, re-reverse the order. Tests pass, bug fixed.
I'm not sure when this bug got introduced, but its at least
from before the Blink fork, as are most of these other test
failures.

NOTRY=true

Review URL: https://codereview.chromium.org/332073002

git-svn-id: svn://svn.chromium.org/blink/trunk@176161 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Traditionally, frame tree mutation has been driven from inside the renderer
itself. However, for out-of-process iframes, the embedder needs to mirror the
same frame tree structure between several different renderers.

BUG=346764

Review URL: https://codereview.chromium.org/334483002

git-svn-id: svn://svn.chromium.org/blink/trunk@176159 bbb929c8-8fbe-4397-9dbb-9b2b20218538

This CL takes care the fetch, fileapi, frame and part of html subdirs under core.

BUG=381876

Review URL: https://codereview.chromium.org/336893002

git-svn-id: svn://svn.chromium.org/blink/trunk@176158 bbb929c8-8fbe-4397-9dbb-9b2b20218538

BUG=379012

Review URL: https://codereview.chromium.org/324893002

git-svn-id: svn://svn.chromium.org/blink/trunk@176157 bbb929c8-8fbe-4397-9dbb-9b2b20218538

The biggest change is removing RelativeTime. We were creating these,
but not doing anything with them. These were also the only
things that ever called setIndex on Notification, so we can remove
index/setIndex from Notification.

The rest of the patch is just removing dead code.

NOTRY=true

Review URL: https://codereview.chromium.org/336033002

git-svn-id: svn://svn.chromium.org/blink/trunk@176156 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Remove the very complicated logic from computeLogicalHeight() and
computeLogicalWidth() whose sole purpose was to use the writing mode of
the child vs the containing block when appropriate. The complexity was
unnecessary and the same result can be achieved by ensuring we use the
child's writing mode when computing the size of the box and the parent's
writing mode when computing the margins for position (e.g. for margin
collapsing).

Review URL: https://codereview.chromium.org/298563002

git-svn-id: svn://svn.chromium.org/blink/trunk@176154 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Depends on the yasm stuff landing and sticking.

Review URL: https://codereview.chromium.org/336633003

git-svn-id: svn://svn.chromium.org/blink/trunk@176153 bbb929c8-8fbe-4397-9dbb-9b2b20218538

We focus the plugin on mousedown/touchstart, so similarly we should
focus it on gesturetapdown.

This fixes IME input on plugins, if a plugin gets gesturetapdown
without any mousedown/touchstart, the plugin will be focused correctly
and input fed to the plugin would work properly.
This could happen if the plugin doesn't have any touch handlers, then
tapping sends gesture event(s) to plugin, without any touchstart.

BUG=383484
Test=Using chrome app webview, which is a plugin.
Enable virutal keyboard in chrome://flags if it's not enabled.
In chromebook pixel: load the sample
browser app
https://github.com/GoogleChrome/chrome-app-samples/tree/master/webview-samples/browser
Navigate to a simple page that has <input> element, e.g.
http://jsbin.com/togif/1/
Now without generating any mouse/touchpad inside the content area, tap on the screen on the input box in the jsbin page.
Notice virtual keyboard opening and it is possible to type stuffs into the input box.

Review URL: https://codereview.chromium.org/333813002

git-svn-id: svn://svn.chromium.org/blink/trunk@176152 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Depending on the inlineBox bidiLevel use the right or left offset.
Do not use bidiLevel of the previous box when attribute is auto.

R=eseidel, leviw
BUG=296847
TEST=TextArea with strong RTL word, when on next line LTR word is
entered, caret position is changed.

Review URL: https://codereview.chromium.org/302433015

git-svn-id: svn://svn.chromium.org/blink/trunk@176151 bbb929c8-8fbe-4397-9dbb-9b2b20218538

In the process, inline RenderLayerRepainter::setBackingNeedsRepaint's one call site
and remove the method.

BUG=384520

Review URL: https://codereview.chromium.org/337793004

git-svn-id: svn://svn.chromium.org/blink/trunk@176150 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Request.method should be ByteString (now supported)
Response.status should be assignable
Response.statusText should be assignable
Response.statusText should be ByteString (now supported)

And added tests for assignable properties/invalid ByteStrings.

R=falken@chromium.org,horo@chromium.org

Review URL: https://codereview.chromium.org/337793003

git-svn-id: svn://svn.chromium.org/blink/trunk@176148 bbb929c8-8fbe-4397-9dbb-9b2b20218538

The json mirror is apparently better able to scale to load
and is less dependent on the buildbot master being in
good state.

As part of this, we how have a way of querying the step
information for the last build on a builder, so we can
skip the request and code related to finding out the
build number of the latest run.

Change the masters data structure to store a url name
instead of the whole URL. The url name is the master name
to use for constructing URLs. This way buildershandler
can use the mirror, but builders.js can continue pointing
to the actual master for it's non-json queries.

Also, get rid of the force_update code path in buildershandler.
This seemed like a good idea at the time, but in practice it
just masks errors by making it seem like things are working.
The downside of deleting this is that persistent errors loading
the JSON won't be noticed until the memcache is flushed for
some reason. We should add monitoring of 500s to catch these
things earlier.

Review URL: https://codereview.chromium.org/337433003

git-svn-id: svn://svn.chromium.org/blink/trunk@176147 bbb929c8-8fbe-4397-9dbb-9b2b20218538

readline blocks, so reading stdout, then stderr doesn't make any
sense because we'll block on each on. So if we print only to one
of them, we'll just wait until the process completes.

Also, simplify the readline loop in RebaselineOMatic to just
use readline instead of poll. That way it avoids missing any
output between the last poll and when the process finishes.

Review URL: https://codereview.chromium.org/329873005

git-svn-id: svn://svn.chromium.org/blink/trunk@176146 bbb929c8-8fbe-4397-9dbb-9b2b20218538

BUG=384565,245025

Review URL: https://codereview.chromium.org/333063004

git-svn-id: svn://svn.chromium.org/blink/trunk@176145 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Revert of Disable the HTMLVideoElement-specific prefixed fullscreen API (2nd attempt) (https://codereview.chromium.org/251873002/)

Reason for revert:
Wait with removal until the unprefixed Fullscreen API has been shipped.

Original issue's description:
> Disable the HTMLVideoElement-specific prefixed fullscreen API (2nd attempt)
> 
> BUG=346236
> 
> Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=172693

Review URL: https://codereview.chromium.org/328303003

git-svn-id: svn://svn.chromium.org/blink/trunk@176144 bbb929c8-8fbe-4397-9dbb-9b2b20218538

This CL implements frame scrollbar update for the overflow recalc fast path
without invoking layout. This fast path first determines whether scrollbars
need to be created or removed (in that case, a layout will be needed), and
update scrollbars without layout if possible.

TEST=fast/transforms/transform-update-frame-overflow.html
BUG=354167

Review URL: https://codereview.chromium.org/328023002

git-svn-id: svn://svn.chromium.org/blink/trunk@176143 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Add APIs
- WebFrame::brandColor() to get the brand color.
- WebFrameClinet::didChangeBrandColor to notify the brand color change.

BUG=383941

Intent to implement and ship link
https://groups.google.com/a/chromium.org/forum/#!topic/blink-dev/nzRY-h_-_ig

Review URL: https://codereview.chromium.org/329943004

git-svn-id: svn://svn.chromium.org/blink/trunk@176142 bbb929c8-8fbe-4397-9dbb-9b2b20218538

http://src.chromium.org/viewvc/blink?view=revision&revision=176121

BUG=383495
TBR=senorblanco@chromium.org

Review URL: https://codereview.chromium.org/333973002

git-svn-id: svn://svn.chromium.org/blink/trunk@176139 bbb929c8-8fbe-4397-9dbb-9b2b20218538

squashing layer state. This is no longer necessary after crrev.com/311813005.

BUG=

Review URL: https://codereview.chromium.org/331563005

git-svn-id: svn://svn.chromium.org/blink/trunk@176138 bbb929c8-8fbe-4397-9dbb-9b2b20218538

This is a speculative fix for the flakiness we are seeing on the
textPath-change-reference2.svg test[1]. The failure is not reproducable
locally but appears as if runTest() is never called. This patch
refactors the test to not depend on runTest().

In addition, the test ids have been refactored to be understandable.

[1] http://test-results.appspot.com/dashboards/flakiness_dashboard.html#group=%40ToT%20Blink&showAllRuns=true&showExpectations=true&tests=svg%2Fcustom%2FtextPath-change-reference2.svg

BUG=384566
TBR=dpranke

Review URL: https://codereview.chromium.org/333063003

git-svn-id: svn://svn.chromium.org/blink/trunk@176137 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Cleaned up style of translatez-removed.html while updating it.

Review URL: https://codereview.chromium.org/324363003

git-svn-id: svn://svn.chromium.org/blink/trunk@176136 bbb929c8-8fbe-4397-9dbb-9b2b20218538

We'd miss any output between the last poll and the end of
the process.

Also, print out the log lines so that you can see them when
looking at the local log.

BUG=382248

Review URL: https://codereview.chromium.org/327633003

git-svn-id: svn://svn.chromium.org/blink/trunk@176135 bbb929c8-8fbe-4397-9dbb-9b2b20218538

SystemParametersInfo will fail if we've disconnected win32k
via the sandbox on Win8+. So, just default to nicer font
settings for now, since we know the OS supports them.

BUG=365160

Review URL: https://codereview.chromium.org/331843004

git-svn-id: svn://svn.chromium.org/blink/trunk@176134 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Intent-to-ship:
https://groups.google.com/a/chromium.org/d/msg/blink-dev/Ae18uqxjzp8/C4K_J1xpnGwJ

BUG=140205

Review URL: https://codereview.chromium.org/330873002

git-svn-id: svn://svn.chromium.org/blink/trunk@176133 bbb929c8-8fbe-4397-9dbb-9b2b20218538

- mark crypto/rsassa-pkcs1-v1_5-generate-key.html as flaky timeout
- mark svg/custom/textPath-change-reference2.svg as flaky failure
- mark inspector/console/console-viewport-selection.html as flaky (Fix FlakyTest entry)
- virtual/gpu/fast/canvas/canvas-draw-canvas-on-canvas-shadow.html is no longer failing

TBR=schenney@chromium.org

Review URL: https://codereview.chromium.org/331843006

git-svn-id: svn://svn.chromium.org/blink/trunk@176132 bbb929c8-8fbe-4397-9dbb-9b2b20218538

The test now appears to time out periodically (i.e., it's not flaky).
Reverting.

Original description:
> Crash test case:
> 1. Open page with javascript
> 2. Add breakpoint
> 3. Stop on breakpoint
> 4. Use live edit and save script
> The test worked for Chromium revision until May 30.

TBR=kostyatinskiy@google.com
BUG=none

Review URL: https://codereview.chromium.org/331853007

git-svn-id: svn://svn.chromium.org/blink/trunk@176131 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Changes to the compositing path meant the toggle of an iframe to having
a self-painting layer wasn't getting updated.

R=jamesr@chromium.org

Review URL: https://codereview.chromium.org/330283008

git-svn-id: svn://svn.chromium.org/blink/trunk@176130 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Break Android build, clearly it was a typo.

BUG=357163

Review URL: https://codereview.chromium.org/333943002

git-svn-id: svn://svn.chromium.org/blink/trunk@176129 bbb929c8-8fbe-4397-9dbb-9b2b20218538

This makes RLC::applyUpdateLayerCompositingStateChickenEggHacks
issue a repaint when a RenderLayer gains or loses a CLM.

Review URL: https://codereview.chromium.org/336493004

git-svn-id: svn://svn.chromium.org/blink/trunk@176128 bbb929c8-8fbe-4397-9dbb-9b2b20218538

What this update type actually does is update the geometry of all the
composited layer mappings that have dirty geometry. Rather than naming the
value for its caller, we should name the value for its effect.

R=vollick@chromium.org

Review URL: https://codereview.chromium.org/334973002

git-svn-id: svn://svn.chromium.org/blink/trunk@176127 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Revert of Removing "using" declarations that import names in the C++ Standard library. (https://codereview.chromium.org/329183002/)

Reason for revert:
This patch is causing failures on the Chromium Linux Tsan v2 bot: http://build.chromium.org/p/chromium.memory.fyi/builders/Linux%20Builder%20(TSan%20v2)

The problem is that std::abs(int) is found in <cstdlib> but we apparently only include <cmath> everywhere. So the compiler only knows about std::abs(float) and variants.

You can probably fix this by including <cstdlib> in the places where you use std::abs.

Original issue's description:
> Removing "using" declarations that import names in the C++ Standard library.
> 
> This is as per the the coding guidelines for blink, which disallows the
> use of "using" declarations of any kind to import names in the standard
> template library.
> 
> BUG=NONE
> 
> Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=176032

TBR=tkent@chromium.org,ch.dumez@samsung.com,sunil.ratnu@samsung.com
NOTREECHECKS=true
NOTRY=true
BUG=NONE

Review URL: https://codereview.chromium.org/330933002

git-svn-id: svn://svn.chromium.org/blink/trunk@176125 bbb929c8-8fbe-4397-9dbb-9b2b20218538

BUG=None

Review URL: https://codereview.chromium.org/338523004

git-svn-id: svn://svn.chromium.org/blink/trunk@176124 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Add missing null check in PageSerializer::serializeCSSStyleSheet() for
importRule->styleSheet(). CSSImportRule::styleSheet() can return null but
the function it is being passed to expects a non-null pointer.

Also update PageSerializer::serializeCSSStyleSheet() to take its CSSStyleSheet
argument by reference, instead of pointer to make it obvious to the caller
that passing null is not OK.

R=morrita@chromium.org

Review URL: https://codereview.chromium.org/332573003

git-svn-id: svn://svn.chromium.org/blink/trunk@176123 bbb929c8-8fbe-4397-9dbb-9b2b20218538

The legacy implementation of FETurbulence actually never generated
tile-able noise. The algorithmic problem was fixed in Skia
(https://codereview.chromium.org/332523006/), but Blink was also not
passing the tile size (filter primitive subregion) to the turbulence
shader. This change fixes the latter (small) problem.

BUG=383495
R=fmalita

Review URL: https://codereview.chromium.org/333783002

git-svn-id: svn://svn.chromium.org/blink/trunk@176121 bbb929c8-8fbe-4397-9dbb-9b2b20218538

ASSERT_ENABLED doens't have more not double negatives.

Note the #ifndef ASSERT_DISABLED in DatabaseThread.cpp was always compiled out
because ASSERT_DISABLED was always defined to be either 0 or 1.

Review URL: https://codereview.chromium.org/331793004

git-svn-id: svn://svn.chromium.org/blink/trunk@176120 bbb929c8-8fbe-4397-9dbb-9b2b20218538

This function is iterating the stacking tree. It needs to ensure that the
stacking tree is up-to-date.

TBR=esprehn@chromium.org

Review URL: https://codereview.chromium.org/331923002

git-svn-id: svn://svn.chromium.org/blink/trunk@176119 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Styles are un-applied when an import with the style is removed.
The style however should be re-applied when the import is inserted back to the tree.

BUG=381052
R=dglazkov@chromium.org
TEST=import-readd*

Review URL: https://codereview.chromium.org/331763002

git-svn-id: svn://svn.chromium.org/blink/trunk@176117 bbb929c8-8fbe-4397-9dbb-9b2b20218538

Update SelectorQuery constructor to adopt CSSSelectorList object instead of
copying it. This CL adds a adopt(CSSSelectorList&) factory function to
SelectorQuery class to make this behavior obvious.

Copying a CSSSelectorList is relatively expensive and it should therefore be
avoided when possible.

R=esprehn@chromium.org

Review URL: https://codereview.chromium.org/330713003

git-svn-id: svn://svn.chromium.org/blink/trunk@176116 bbb929c8-8fbe-4397-9dbb-9b2b20218538