Skip to content

Instantly share code, notes, and snippets.

@daeken
Created April 1, 2017 22:37
Show Gist options
  • Save daeken/ec6b84ab73b14ce2793d231ad69501f8 to your computer and use it in GitHub Desktop.
Save daeken/ec6b84ab73b14ce2793d231ad69501f8 to your computer and use it in GitHub Desktop.
------------------------------------------------------------------------
r212097 | carlosgc@webkit.org | 2017-02-10 08:26:45 +0000 (Fri, 10 Feb 2017) | 5 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.cpp
Unreviewed. Fix GTK+ build with threaded compositor disabled.
* WebProcess/WebPage/AcceleratedDrawingArea.cpp:
(WebKit::AcceleratedDrawingArea::enterAcceleratedCompositingMode):
(WebKit::AcceleratedDrawingArea::exitAcceleratedCompositingModeNow):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211967 | magomez@igalia.com | 2017-02-09 18:11:22 +0000 (Thu, 09 Feb 2017) | 23 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/cairo/CairoUtilities.cpp
[GTK] scroll with transparent background not repainted after scrollY >= 32768
https://bugs.webkit.org/show_bug.cgi?id=154283
Reviewed by Carlos Garcia Campos.
Due to a limitation of the pixman backend, which uses 16 bits to hold signed integers, cairo is
not able to draw anything when using transformation matrices with values bigger than 32768. When
drawing patterns into large pages, the matrices values can overflow those 16 bits, so cairo doesn't
draw anything in, which causes the reported transparent backgrounds.
The patch modifies the transformation matrices both from the current context and the pattern we
are painting, to avoid them to hold values that cannot stored in 16 bits.
There's still the possibility that this happens, but it would require using a pattern with a size
bigger than 32768.
Based on a previous patch by Gwang Yoon Hwang <yoon@igalia.com>.
No new tests.
* platform/graphics/cairo/CairoUtilities.cpp:
(WebCore::drawPatternToCairoContext):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211867 | magomez@igalia.com | 2017-02-08 12:19:05 +0000 (Wed, 08 Feb 2017) | 15 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/texmap/coordinated/TiledBackingStore.cpp
M /trunk/Source/WebCore/platform/graphics/texmap/coordinated/TiledBackingStore.h
[GTK] Reduce TiledBackingStore tile coverage when on memory pressure state
https://bugs.webkit.org/show_bug.cgi?id=167980
Reviewed by Carlos Garcia Campos.
Use a smaller coverAreaMultiplier in the TiledBackingStore when in memory pressure situation, to reduce
the amount of tiles created.
No new tests.
* platform/graphics/texmap/coordinated/TiledBackingStore.cpp:
(WebCore::TiledBackingStore::createTilesIfNeeded):
(WebCore::TiledBackingStore::createTiles):
* platform/graphics/texmap/coordinated/TiledBackingStore.h:
------------------------------------------------------------------------
------------------------------------------------------------------------
r211721 | magomez@igalia.com | 2017-02-06 12:22:04 +0000 (Mon, 06 Feb 2017) | 13 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/CompositingCoordinator.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/CompositingCoordinator.h
[GTK][EFL] Release unused UpdateAtlas when in memory pressure situation
https://bugs.webkit.org/show_bug.cgi?id=167872
Reviewed by Carlos Garcia Campos.
In a memory pressure situation, release all the unused UpdateAtlas as soon as possible, instead
of waiting 3 seconds before releasing them.
* WebProcess/WebPage/CoordinatedGraphics/CompositingCoordinator.cpp:
(WebKit::CompositingCoordinator::releaseInactiveAtlasesTimerFired):
(WebKit::CompositingCoordinator::releaseAtlases):
* WebProcess/WebPage/CoordinatedGraphics/CompositingCoordinator.h:
------------------------------------------------------------------------
------------------------------------------------------------------------
r212024 | bfulgham@apple.com | 2017-02-10 02:11:28 +0000 (Fri, 10 Feb 2017) | 23 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/shadow-dom/attach-shadow-teardown-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/attach-shadow-teardown.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/dom/Element.cpp
Tear down existing renderers when adding a shadow root.
https://bugs.webkit.org/show_bug.cgi?id=167117
Patch by Antti Koivisto <antti@apple.com> on 2017-02-09
Reviewed by Andreas Kling.
Source/WebCore:
Adding a shadow root may cause children no longer be part of the flat tree. If they had renderers
tear down code would no longer reach them.
Test: fast/shadow-dom/attach-shadow-teardown.html
* dom/Element.cpp:
(WebCore::Element::addShadowRoot):
Tear down existing render tree when adding a shadow root.
LayoutTests:
* fast/shadow-dom/attach-shadow-teardown-expected.txt: Added.
* fast/shadow-dom/attach-shadow-teardown.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r212023 | bfulgham@apple.com | 2017-02-10 02:09:25 +0000 (Fri, 10 Feb 2017) | 23 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/forms/input-type-change-during-selection-expected.txt
A /trunk/LayoutTests/fast/forms/input-type-change-during-selection.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/html/HTMLTextFormControlElement.cpp
Handle synchronous layout when setting a selection range
https://bugs.webkit.org/show_bug.cgi?id=167092
<rdar://problem/30041640>
Reviewed by Antti Koivisto.
Source/WebCore:
The 'innerTextElement' of a form control can change during layout due
to arbitrary JavaScript executing. Handle the case where the inner text
element has changed so that current render box height is while setting
a selection range.
Test: fast/forms/input-type-change-during-selection.html
* html/HTMLTextFormControlElement.cpp:
(WebCore::HTMLTextFormControlElement::setSelectionRange):
LayoutTests:
* fast/forms/input-type-change-during-selection-expected.txt: Added.
* fast/forms/input-type-change-during-selection.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211999 | bfulgham@apple.com | 2017-02-10 00:10:14 +0000 (Fri, 10 Feb 2017) | 22 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/shadow-dom/resources/start.html
A /trunk/LayoutTests/fast/shadow-dom/shadow-at-root-during-disconnect-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/shadow-at-root-during-disconnect.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/dom/ContainerNodeAlgorithms.cpp
Disconnect shadow children of root when detaching a frame
https://bugs.webkit.org/show_bug.cgi?id=166851
<rdar://problem/29930443>
Reviewed by Andy Estes.
Source/WebCore:
If the root of the tree we are disconnecting has a shadow element, include it in the set of
things to disconnect.
Tests: fast/shadow-dom/shadow-at-root-during-disconnect.html
* dom/ContainerNodeAlgorithms.cpp:
(WebCore::disconnectSubframes):
LayoutTests:
* fast/shadow-dom/resources/start.html: Added.
* fast/shadow-dom/shadow-at-root-during-disconnect-expected.txt: Added.
* fast/shadow-dom/shadow-at-root-during-disconnect.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r209145 | bfulgham@apple.com | 2016-11-30 20:20:32 +0000 (Wed, 30 Nov 2016) | 62 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/shadow-dom/color-input-element-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/color-input-element-shadow-manipulation.html
A /trunk/LayoutTests/fast/shadow-dom/file-input-element-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/file-input-element-shadow-manipulation.html
A /trunk/LayoutTests/fast/shadow-dom/keygen-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/keygen-shadow-manipulation.html
A /trunk/LayoutTests/fast/shadow-dom/media-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/media-shadow-manipulation.html
A /trunk/LayoutTests/fast/shadow-dom/range-input-element-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/range-input-element-shadow-manipulation.html
A /trunk/LayoutTests/fast/shadow-dom/textarea-shadow-manipulation-expected.txt
A /trunk/LayoutTests/fast/shadow-dom/textarea-shadow-manipulation.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/dom/Element.h
M /trunk/Source/WebCore/html/ColorInputType.cpp
M /trunk/Source/WebCore/html/FileInputType.cpp
M /trunk/Source/WebCore/html/HTMLKeygenElement.cpp
M /trunk/Source/WebCore/html/HTMLMediaElement.cpp
M /trunk/Source/WebCore/html/HTMLTextAreaElement.cpp
M /trunk/Source/WebCore/html/RangeInputType.cpp
M /trunk/Source/WebCore/html/shadow/SliderThumbElement.h
M /trunk/Source/WebCore/svg/SVGUseElement.cpp
Use 'childOfType' template when retrieving Shadow DOM elements
https://bugs.webkit.org/show_bug.cgi?id=165145
<rdar://problem/29331830>
Reviewed by Antti Koivisto.
Source/WebCore:
Tests: fast/shadow-dom/color-input-element-shadow-manipulation.html
fast/shadow-dom/file-input-element-shadow-manipulation.html
fast/shadow-dom/keygen-shadow-manipulation.html
fast/shadow-dom/media-shadow-manipulation.html
fast/shadow-dom/range-input-element-shadow-manipulation.html
fast/shadow-dom/textarea-shadow-manipulation.html
Switch to using 'childOfType' when retrieving Shadow DOM elements, rather
than relying on expected element positions, as these can be changed by
JavaScript.
Drive by fix: Make more use of is<> and downcast<> templates rather than blindly casting.
* dom/Element.h:
(WebCore::Element::isUploadButton): Added.
(WebCore::Element::isSliderContainerElement): Added.
* html/ColorInputType.cpp:
(WebCore::ColorInputType::shadowColorSwatch): Use 'childOfType' rather than assuming
the first child is the one we want.
* html/FileInputType.cpp:
(isType): Added.
(WebCore::FileInputType::disabledAttributeChanged): Use 'childOfType' rather than assuming
the first child is the one we want.
(WebCore::FileInputType::multipleAttributeChanged): Ditto.
* html/HTMLKeygenElement.cpp:
(WebCore::HTMLKeygenElement::shadowSelect): Ditto.
* html/HTMLMediaElement.cpp:
(WebCore::HTMLMediaElement::mediaControls): Ditto.
(WebCore::HTMLMediaElement::hasMediaControls): Ditto.
* html/HTMLTextAreaElement.cpp:
(WebCore::HTMLTextAreaElement::innerTextElement): Ditto.
* html/RangeInputType.cpp:
(WebCore::RangeInputType::sliderTrackElement): Ditto.
* html/shadow/SliderThumbElement.h:
(isType): Added.
* svg/SVGUseElement.cpp:
(WebCore::SVGUseElement::targetClone): Use 'childOfType' rather than assuming
the first child is the one we want.
LayoutTests:
* fast/shadow-dom/color-input-element-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/color-input-element-shadow-manipulation.html: Added.
* fast/shadow-dom/file-input-element-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/file-input-element-shadow-manipulation.html: Added.
* fast/shadow-dom/keygen-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/keygen-shadow-manipulation.html: Added.
* fast/shadow-dom/media-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/media-shadow-manipulation.html: Added.
* fast/shadow-dom/range-input-element-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/range-input-element-shadow-manipulation.html: Added.
* fast/shadow-dom/textarea-shadow-manipulation-expected.txt: Added.
* fast/shadow-dom/textarea-shadow-manipulation.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r210122 | bfulgham@apple.com | 2016-12-23 00:20:27 +0000 (Fri, 23 Dec 2016) | 25 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/loader/nested-document-handling-expected.txt
A /trunk/LayoutTests/fast/loader/nested-document-handling.html
A /trunk/LayoutTests/fast/loader/resources/subframe-success.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/loader/DocumentWriter.cpp
M /trunk/Source/WebCore/page/Frame.cpp
M /trunk/Source/WebCore/page/Frame.h
Nested calls to setDocument can omit firing 'unload' events
https://bugs.webkit.org/show_bug.cgi?id=166422
<rdar://problem/29763012>
Reviewed by Alex Christensen.
Source/WebCore:
Test: fast/loader/nested-document-handling.html
Only allow a single document change to be taking place during a given runloop cycle.
* bindings/js/ScriptController.cpp:
(WebCore::ScriptController::executeIfJavaScriptURL): Block script changing the document
when we are in the middle of changing the document.
* page/Frame.cpp:
(WebCore::Frame::setDocument): Keep track of document change state.
* page/Frame.h:
LayoutTests:
* fast/loader/nested-document-handling-expected.txt: Added.
* fast/loader/nested-document-handling.html: Added.
* fast/loader/resources/subframe-success.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211254 | cdumez@apple.com | 2017-01-27 05:36:19 +0000 (Fri, 27 Jan 2017) | 121 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
M /trunk/LayoutTests/editing/mac/input/unconfirmed-text-navigation-with-page-cache.html
M /trunk/LayoutTests/fast/harness/page-cache-crash-on-data-urls.html
M /trunk/LayoutTests/fast/harness/use-page-cache.html
A /trunk/LayoutTests/fast/history/page-cache-after-window-open-expected.txt
A /trunk/LayoutTests/fast/history/page-cache-after-window-open.html
A /trunk/LayoutTests/fast/history/page-cache-back-navigation-crash-expected.txt
A /trunk/LayoutTests/fast/history/page-cache-back-navigation-crash.html
A /trunk/LayoutTests/fast/history/page-cache-with-opener-expected.txt
A /trunk/LayoutTests/fast/history/page-cache-with-opener.html
A /trunk/LayoutTests/fast/history/resources/page-cache-window-with-iframe.html
A /trunk/LayoutTests/fast/history/resources/page-cache-window-with-opener.html
M /trunk/LayoutTests/fast/loader/stateobjects/no-popstate-when-back-to-stateless-entry-with-page-cache.html
M /trunk/LayoutTests/fast/loader/stateobjects/popstate-fires-with-page-cache.html
M /trunk/LayoutTests/tiled-drawing/tiled-drawing-scroll-position-page-cache-restoration.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/dom/Document.cpp
M /trunk/Source/WebCore/dom/Document.h
M /trunk/Source/WebCore/history/PageCache.cpp
M /trunk/Source/WebCore/page/DOMWindow.cpp
M /trunk/Source/WebCore/page/DiagnosticLoggingKeys.cpp
M /trunk/Source/WebCore/page/DiagnosticLoggingKeys.h
M /trunk/Source/WebCore/page/Page.cpp
M /trunk/Source/WebCore/page/Page.h
M /trunk/Source/WebCore/page/Settings.in
M /trunk/Source/WebKit/mac/ChangeLog
M /trunk/Source/WebKit/mac/WebView/WebPreferenceKeysPrivate.h
M /trunk/Source/WebKit/mac/WebView/WebPreferences.mm
M /trunk/Source/WebKit/mac/WebView/WebPreferencesPrivate.h
M /trunk/Source/WebKit/mac/WebView/WebView.mm
M /trunk/Source/WebKit/win/ChangeLog
M /trunk/Source/WebKit/win/Interfaces/IWebPreferencesPrivate.idl
M /trunk/Source/WebKit/win/WebPreferenceKeysPrivate.h
M /trunk/Source/WebKit/win/WebPreferences.cpp
M /trunk/Source/WebKit/win/WebPreferences.h
M /trunk/Source/WebKit/win/WebView.cpp
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/WebPreferencesDefinitions.h
M /trunk/Source/WebKit2/UIProcess/API/C/WKPreferences.cpp
M /trunk/Source/WebKit2/UIProcess/API/C/WKPreferencesRefPrivate.h
M /trunk/Source/WebKit2/WebProcess/InjectedBundle/InjectedBundle.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/WebPage.cpp
M /trunk/Tools/ChangeLog
M /trunk/Tools/DumpRenderTree/mac/DumpRenderTree.mm
M /trunk/Tools/DumpRenderTree/win/DumpRenderTree.cpp
M /trunk/Tools/WebKitTestRunner/TestController.cpp
Crash when navigating back to a page in PacheCache when one of its frames has been removed
https://bugs.webkit.org/show_bug.cgi?id=167421
<rdar://problem/30188490>
Reviewed by Darin Adler.
Source/WebCore:
Disallow page caching of a page if:
1. The main window has an opener (i.e. it was opened via window.open)
2. It has ever used window.open()
This is because allowing page caching in this case would allow such
windows to script each other even after one of them entered Page
Cache. Allowing this is dangerous and easily causes crashes.
This is a short term workaround until we find a better solution to
the problem. One issue is this workaround is that navigating back
to a page that has an opener or used window.open() will not longer
get the page from PageCache. As a result, state may be lost upon
navigating back. However, we never guarantee that pages get page
cached, and Chrome does not have a PageCache.
Tests: fast/history/page-cache-after-window-open.html
fast/history/page-cache-back-navigation-crash.html
fast/history/page-cache-with-opener.html
* dom/Document.cpp:
(WebCore::Document::hasEverCalledWindowOpen):
(WebCore::Document::markHasCalledWindowOpen):
* dom/Document.h:
* history/PageCache.cpp:
(WebCore::canCachePage):
* page/DOMWindow.cpp:
(WebCore::DOMWindow::createWindow):
* page/DiagnosticLoggingKeys.cpp:
(WebCore::DiagnosticLoggingKeys::hasCalledWindowOpenKey):
(WebCore::DiagnosticLoggingKeys::hasOpenerKey):
* page/DiagnosticLoggingKeys.h:
* page/Page.cpp:
(WebCore::Page::openedByWindowOpen):
* page/Page.h:
* page/Settings.in:
Source/WebKit/mac:
Add a new setting allowing layout tests to enable PageCache in a window
that has an opener, for convenience.
* WebView/WebPreferenceKeysPrivate.h:
* WebView/WebPreferences.mm:
(+[WebPreferences initialize]):
(-[WebPreferences allowsPageCacheWithWindowOpener]):
(-[WebPreferences setAllowsPageCacheWithWindowOpener:]):
* WebView/WebPreferencesPrivate.h:
* WebView/WebView.mm:
(-[WebView _preferencesChanged:]):
Source/WebKit/win:
Add a new setting allowing layout tests to enable PageCache in a window
that has an opener, for convenience.
* WebPreferenceKeysPrivate.h:
* WebPreferences.cpp:
(WebPreferences::initializeDefaultSettings):
* WebPreferences.h:
Source/WebKit2:
Add a new setting allowing layout tests to enable PageCache in a window
that has an opener, for convenience.
* Shared/WebPreferencesDefinitions.h:
* UIProcess/API/C/WKPreferences.cpp:
(WKPreferencesSetAllowsPageCacheWithWindowOpener):
(WKPreferencesGetAllowsPageCacheWithWindowOpener):
* UIProcess/API/C/WKPreferencesRefPrivate.h:
* WebProcess/InjectedBundle/InjectedBundle.cpp:
(WebKit::InjectedBundle::overrideBoolPreferenceForTestRunner):
* WebProcess/WebPage/WebPage.cpp:
(WebKit::WebPage::updatePreferences):
Tools:
Add a new setting allowing layout tests to enable PageCache in a window
that has an opener, for convenience.
* DumpRenderTree/mac/DumpRenderTree.mm:
(resetWebPreferencesToConsistentValues):
* DumpRenderTree/win/DumpRenderTree.cpp:
(resetWebPreferencesToConsistentValues):
* WebKitTestRunner/TestController.cpp:
(WTR::TestController::resetPreferencesToConsistentValues):
LayoutTests:
* fast/history/page-cache-after-window-open-expected.txt: Added.
* fast/history/page-cache-after-window-open.html: Added.
* fast/history/page-cache-back-navigation-crash-expected.txt: Added.
* fast/history/page-cache-back-navigation-crash.html: Added.
* fast/history/page-cache-with-opener-expected.txt: Added.
* fast/history/page-cache-with-opener.html: Added.
* fast/history/resources/page-cache-window-with-iframe.html: Added.
* fast/history/resources/page-cache-window-with-opener.html: Added.
Add layout test coverage.
* editing/mac/input/unconfirmed-text-navigation-with-page-cache.html:
* fast/harness/page-cache-crash-on-data-urls.html:
* fast/harness/use-page-cache.html:
* fast/history/page-cache-after-window-open-expected.txt: Added.
* fast/history/page-cache-after-window-open.html: Added.
* fast/history/page-cache-with-opener-expected.txt: Added.
* fast/history/page-cache-with-opener.html: Added.
* fast/history/resources/page-cache-window-with-opener.html: Added.
* fast/loader/stateobjects/no-popstate-when-back-to-stateless-entry-with-page-cache.html:
* fast/loader/stateobjects/popstate-fires-with-page-cache.html:
* tiled-drawing/tiled-drawing-scroll-position-page-cache-restoration.html:
These tests relied on using window.open() to test PageCache for convenience. They now
need to override a setting in order to be allowed to do so.
------------------------------------------------------------------------
------------------------------------------------------------------------
r210120 | zalan@apple.com | 2016-12-22 23:48:53 +0000 (Thu, 22 Dec 2016) | 21 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/regions/flow-fragment-as-anonymous-block-crash-expected.txt
A /trunk/LayoutTests/fast/regions/flow-fragment-as-anonymous-block-crash.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/rendering/RenderObject.h
Do not destroy the RenderNamedFlowFragment as leftover anonymous block.
https://bugs.webkit.org/show_bug.cgi?id=166436
rdar://problem/29772233
Reviewed by Simon Fraser.
Source/WebCore:
When as the result of certain style change, the generated anonymous block is not needed anymore, we
move its descendants up to the parent and destroy the generated box. While RenderNamedFlowFragment is a generated
block, the cleanup code should just ignore it the same way we ignore boxes like multicolumn, mathml etc.
Test: fast/regions/flow-fragment-as-anonymous-block-crash.html
* rendering/RenderObject.h:
(WebCore::RenderObject::isAnonymousBlock):
LayoutTests:
* fast/regions/flow-fragment-as-anonymous-block-crash-expected.txt: Added.
* fast/regions/flow-fragment-as-anonymous-block-crash.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r209926 | zalan@apple.com | 2016-12-16 18:48:31 +0000 (Fri, 16 Dec 2016) | 29 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/accessibility/accessibility-crash-with-dynamic-inline-content-expected.txt
A /trunk/LayoutTests/accessibility/accessibility-crash-with-dynamic-inline-content.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/accessibility/AXObjectCache.cpp
M /trunk/Source/WebCore/accessibility/AXObjectCache.h
M /trunk/Source/WebCore/page/FrameView.cpp
M /trunk/Source/WebCore/rendering/RenderBlock.cpp
M /trunk/Source/WebCore/rendering/RenderBlockLineLayout.cpp
Defer certain accessibility callbacks until after layout is finished.
https://bugs.webkit.org/show_bug.cgi?id=165861
rdar://problem/29646301
Reviewed by Chris Fleizach.
Source/WebCore:
Currently with certain AXObjectCache callbacks, we can end up in a layout while the render tree is being mutated.
This patch ensures that such callbacks are deferred until after tree mutation/layout is finished.
Test: accessibility/accessibility-crash-with-dynamic-inline-content.html
* accessibility/AXObjectCache.cpp:
(WebCore::AXObjectCache::remove):
(WebCore::AXObjectCache::performDeferredIsIgnoredChange):
(WebCore::AXObjectCache::insertDeferredIsIgnoredChange):
* accessibility/AXObjectCache.h:
* page/FrameView.cpp:
(WebCore::FrameView::performPostLayoutTasks):
* rendering/RenderBlock.cpp:
(WebCore::RenderBlock::deleteLines):
* rendering/RenderBlockLineLayout.cpp:
(WebCore::RenderBlockFlow::createAndAppendRootInlineBox):
LayoutTests:
* accessibility/accessibility-crash-with-dynamic-inline-content-expected.txt: Added.
* accessibility/accessibility-crash-with-dynamic-inline-content.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r210288 | bfulgham@apple.com | 2017-01-04 22:35:44 +0000 (Wed, 04 Jan 2017) | 26 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/bindings/js/ScriptController.cpp
M /trunk/Source/WebCore/bindings/js/ScriptController.h
M /trunk/Source/WebCore/loader/FrameLoader.cpp
Correct DOMWindow handling during FrameLoader::clear
https://bugs.webkit.org/show_bug.cgi?id=166357
<rdar://problem/29741862>
Reviewed by Andy Estes.
Make sure that we always clean up the DOM window when clearing Window properties, even if the document will
remain in the page cache. Since 'clearWindowShell' is only used in FrameLoader, divide it's beahvior into
two steps:
1. Rename 'clearWindowShell' to 'clearWIndowShellsNotMatchingDOMWindow' to better describe its function.
Switch to a modern C++ loop. Do not switch to the new DOMWindow here, but detach and clear existing
DOMWindow connections.
2. Add a new method 'setDOMWindowForWindowShell'. Complete switch to the new DOMWindow.
This change allows us to disconnect the old DOMWindow, perform the 'setDocument(nullptr)' operation, and then
connect to the new Window without leaving the loader in an inconsistent state.
* loader/bindings/js/ScriptController.cpp:
(WebCore::clearWindowShellsNotMatchingDOMWindow): Renamed from 'clearWindowShell'
(WebCore::setDOMWindowForWindowShell): Added.
* loader/bindings/js/ScriptController.h:
* loader/FrameLoader.cpp:
(WebCore::FrameLoader::clear): Revise to use the new two-step DOMWindow switch logic.
------------------------------------------------------------------------
------------------------------------------------------------------------
r209990 | bfulgham@apple.com | 2016-12-19 19:05:36 +0000 (Mon, 19 Dec 2016) | 28 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/html/form-mutate-expected.txt
A /trunk/LayoutTests/fast/html/form-mutate.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/html/HTMLFormElement.cpp
M /trunk/Source/WebCore/html/HTMLFormElement.h
Side effects while restting form elements
https://bugs.webkit.org/show_bug.cgi?id=165959
<rdar://problem/29705967>
Reviewed by Anders Carlsson.
Source/WebCore:
JavaScript logic can run while resetting FormElement objects. This can
lead to unintended side-effets and other unwanted behavior. We should
protect these elements during the reset.
Test: fast/html/form-mutate.html
* html/HTMLFormElement.cpp:
(WebCore::HTMLFormElement::HTMLFormElement): Switch to C++11 initialization.
(WebCore::HTMLFormElement::reset): Protect elements until the reset
operation is finished.
(WebCore::HTMLFormElement::resetAssociatedFormControlElements): Added to share
code with 'resumeFromDocument'.
(WebCore::HTMLFormElement::resumeFromDocument): Protect elements until the
reset operation is finished.
LayoutTests:
* fast/html/form-mutate-expected.txt: Added.
* fast/html/form-mutate.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r209424 | mark.lam@apple.com | 2016-12-06 22:43:16 +0000 (Tue, 06 Dec 2016) | 107 lines
Changed paths:
M /trunk/JSTests/ChangeLog
M /trunk/JSTests/stress/get-from-scope-dynamic-onto-proxy.js
D /trunk/JSTests/stress/proxy-dont-infinite-loop.js
D /trunk/JSTests/stress/proxy-json-path.js
D /trunk/JSTests/stress/rest-parameter-allocation-elimination-watchpoints-6.js
M /trunk/LayoutTests/ChangeLog
M /trunk/LayoutTests/TestExpectations
M /trunk/LayoutTests/http/tests/security/window-named-valueOf-expected.txt
M /trunk/LayoutTests/js/dom/proxy-is-not-allowed-in-global-prototype-chain-expected.txt
M /trunk/LayoutTests/js/dom/proxy-is-not-allowed-in-global-prototype-chain.html
A /trunk/LayoutTests/js/prototype-assignment-expected.txt
A /trunk/LayoutTests/js/prototype-assignment.html
A /trunk/LayoutTests/js/script-tests/prototype-assignment.js
M /trunk/LayoutTests/js/setPrototypeOf-expected.txt
M /trunk/Source/JavaScriptCore/ChangeLog
M /trunk/Source/JavaScriptCore/runtime/JSObject.cpp
M /trunk/Source/JavaScriptCore/runtime/JSTypeInfo.h
M /trunk/Source/JavaScriptCore/runtime/ObjectConstructor.cpp
M /trunk/Source/JavaScriptCore/runtime/ObjectPrototype.h
M /trunk/Source/JavaScriptCore/runtime/Structure.h
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/bindings/js/JSDOMWindowProperties.h
M /trunk/Source/WebCore/bindings/scripts/CodeGeneratorJS.pm
M /trunk/Source/WebCore/bindings/scripts/IDLAttributes.txt
M /trunk/Source/WebCore/dom/EventTarget.idl
M /trunk/Source/WebCore/page/DOMWindow.idl
Introduce the concept of Immutable Prototype Exotic Objects to comply with the spec.
https://bugs.webkit.org/show_bug.cgi?id=165227
<rdar://problem/29442665>
Reviewed by Saam Barati.
JSTests:
* stress/get-from-scope-dynamic-onto-proxy.js:
- Updated error message.
* stress/proxy-dont-infinite-loop.js: Removed.
* stress/proxy-json-path.js: Removed.
* stress/rest-parameter-allocation-elimination-watchpoints-6.js: Removed.
- Removed these tests because the issue they are testing relies on being able to
set Object.prototype.__proto__ to something else (which is now not possible).
Source/JavaScriptCore:
* runtime/JSObject.cpp:
(JSC::JSObject::setPrototypeWithCycleCheck):
- This is where we check for immutable prototype exotic objects and refuse to set
the prototype if needed.
See https://tc39.github.io/ecma262/#sec-immutable-prototype-exotic-objects.
* runtime/JSTypeInfo.h:
(JSC::TypeInfo::isImmutablePrototypeExoticObject):
* runtime/Structure.h:
- Add flag for declaring immutable prototype exotic objects.
* runtime/ObjectPrototype.h:
- Declare that Object.prototype is an immutable prototype exotic object.
See https://tc39.github.io/ecma262/#sec-properties-of-the-object-prototype-object.
* runtime/ObjectConstructor.cpp:
(JSC::objectConstructorSetPrototypeOf):
- Use better error messages.
Source/WebCore:
Make all objects in window.__proto__'s prototype chain immutable prototype exotic
objects. This gives us roughly equivalent behavior to other browsers.
Firefox's behavior differ slightly in that Firefox will fail any attempted
assignment their __proto__, while the immutable prototype exotic objects will
only fail if the assignment is of a different value. See
https://tc39.github.io/ecma262/#sec-immutable-prototype-exotic-objects.
Chrome differs in that assignment to window.__proto__ is also handled like an
immutable prototype exotic object. Instead we adhere to the current HTML spec
that says that the assignment should fail unconditionally. See
https://html.spec.whatwg.org/#the-windowproxy-exotic-object and
https://html.spec.whatwg.org/#windowproxy-setprototypeof.
If the HTML spec is changed to make the WindowProxy and Location objects into
immutable prototype exotic objects later, we can update to match the spec then.
Test: js/prototype-assignment.html
* bindings/js/JSDOMWindowProperties.h:
* bindings/scripts/CodeGeneratorJS.pm:
(GenerateHeader):
(GeneratePrototypeDeclaration):
* bindings/scripts/IDLAttributes.txt:
* dom/EventTarget.idl:
* page/DOMWindow.idl:
LayoutTests:
The new prototype-assignment.js test is currently only enabled for LLInt only
run in the JSC tests until webkit.org/b/165401 is fixed.
* TestExpectations:
- Skip js/prototype-assignment.html for now until webkit.org/b/165401 is fixed.
* http/tests/security/window-named-valueOf-expected.txt:
* js/dom/proxy-is-not-allowed-in-global-prototype-chain-expected.txt:
* js/dom/proxy-is-not-allowed-in-global-prototype-chain.html:
- Updated error messages.
* js/prototype-assignment-expected.txt: Added.
* js/prototype-assignment.html: Added.
* js/script-tests/prototype-assignment.js: Added.
(else):
(reportError):
(shouldEqual):
(shouldThrow):
(stringify):
(makeTestID):
(doInternalSetPrototypeOf):
(ordinarySetPrototypeOf):
(setImmutablePrototype):
(windowProxySetPrototypeOf):
(initSetterExpectation):
(throwIfNoExceptionPending):
(objectSetPrototypeOf):
(setUnderscoreProto):
(reflectSetPrototypeOf):
(newObjectProto.toString):
(this.testObject.targets.push.value):
(this.testProxy.targets.push.setPrototypeOf):
(Symbol):
(test):
(runTests):
* js/setPrototypeOf-expected.txt:
------------------------------------------------------------------------
------------------------------------------------------------------------
r209149 | mark.lam@apple.com | 2016-11-30 21:13:42 +0000 (Wed, 30 Nov 2016) | 17 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/js/dom/proxy-is-not-allowed-in-global-prototype-chain-expected.txt
A /trunk/LayoutTests/js/dom/proxy-is-not-allowed-in-global-prototype-chain.html
M /trunk/Source/JavaScriptCore/ChangeLog
M /trunk/Source/JavaScriptCore/runtime/ProgramExecutable.cpp
Proxy is not allowed in the global prototype chain.
https://bugs.webkit.org/show_bug.cgi?id=165205
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
* runtime/ProgramExecutable.cpp:
(JSC::ProgramExecutable::initializeGlobalProperties):
- We'll now throw a TypeError if we detect a Proxy in the global prototype chain.
LayoutTests:
* js/dom/proxy-is-not-allowed-in-global-prototype-chain-expected.txt: Added.
* js/dom/proxy-is-not-allowed-in-global-prototype-chain.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r208745 | bfulgham@apple.com | 2016-11-15 20:14:02 +0000 (Tue, 15 Nov 2016) | 27 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/forms/search-cancel-button-change-input-expected.txt
A /trunk/LayoutTests/fast/forms/search-cancel-button-change-input.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/WebCore.order
M /trunk/Source/WebCore/html/HTMLInputElement.cpp
M /trunk/Source/WebCore/html/InputType.h
M /trunk/Source/WebCore/html/SearchInputType.h
Correct handling of changing input type
https://bugs.webkit.org/show_bug.cgi?id=164759
<rdar://problem/29211174>
Reviewed by Darin Adler.
Source/WebCore:
Test: fast/forms/search-cancel-button-change-input.html
It is possible for JavaScript to change the type property of an input field. WebKit
needs to gracefully handle this case.
Add a type traits specialization so we can properly downcast InputType elements.
Use this to only call search functions on actual search input types.
* html/HTMLInputElement.cpp:
(WebCore::HTMLInputElement::onSearch): Only perform search functions if the
input type is actually a search field.
* html/InputType.h: Add type traits specialization for 'downcast' template.
* html/SearchInputType.h: Ditto.
LayoutTests:
* fast/forms/search-cancel-button-change-input-expected.txt: Added.
* fast/forms/search-cancel-button-change-input.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r208628 | bfulgham@apple.com | 2016-11-12 00:32:59 +0000 (Sat, 12 Nov 2016) | 20 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/canvas/neutered-imagedata-expected.txt
A /trunk/LayoutTests/fast/canvas/neutered-imagedata.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/bindings/js/SerializedScriptValue.cpp
Neutered ArrayBuffers are not properly serialized
https://bugs.webkit.org/show_bug.cgi?id=164647
<rdar://problem/29213490>
Reviewed by David Kilzer.
Source/WebCore:
Correct binding logic to handle ImageBuffers being deserialized from neutered ArrayBuffers.
Test: fast/canvas/neutered-imagedata.html
* bindings/js/SerializedScriptValue.cpp:
(WebCore::CloneDeserializer::readTerminal):
LayoutTests:
* fast/canvas/neutered-imagedata-expected.txt: Added.
* fast/canvas/neutered-imagedata.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r208825 | bfulgham@apple.com | 2016-11-17 00:35:32 +0000 (Thu, 17 Nov 2016) | 24 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/media/track/audio-track-add-remove-expected.txt
A /trunk/LayoutTests/media/track/audio-track-add-remove.html
A /trunk/LayoutTests/media/track/video-track-add-remove-expected.txt
A /trunk/LayoutTests/media/track/video-track-add-remove.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/html/HTMLMediaElement.cpp
Clear track client when removing a track
https://bugs.webkit.org/show_bug.cgi?id=164842
<rdar://problem/29213621>
Reviewed by Eric Carlson.
Source/WebCore:
Call 'clearClient' when removing a track from an HTMLMediaElement.
Test: media/track/audio-track-add-remove.html
media/track/video-track-add-remove.html
* html/HTMLMediaElement.cpp:
(WebCore::HTMLMediaElement::removeAudioTrack): Call 'clearClient'
(WebCore::HTMLMediaElement::removeVideoTrack): Ditto.
LayoutTests:
* media/track/audio-track-add-remove-expected.txt: Added.
* media/track/audio-track-add-remove.html: Added.
* media/track/video-track-add-remove-expected.txt: Added.
* media/track/video-track-add-remove.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r210112 | dbates@webkit.org | 2016-12-22 22:20:25 +0000 (Thu, 22 Dec 2016) | 24 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame-expected.txt
A /trunk/LayoutTests/fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame.html
A /trunk/LayoutTests/fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame2-expected.txt
A /trunk/LayoutTests/fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame2.html
A /trunk/LayoutTests/fast/events/popup-blocked-from-unique-frame-via-window-open-named-sibling-frame-expected.txt
A /trunk/LayoutTests/fast/events/popup-blocked-from-unique-frame-via-window-open-named-sibling-frame.html
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/page/DOMWindow.cpp
Bypass pop-up blocker from cross-origin or sandboxed frame
https://bugs.webkit.org/show_bug.cgi?id=166290
<rdar://problem/29742039>
Reviewed by Darin Adler.
Source/WebCore:
Tests: fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame.html
fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame2.html
fast/events/popup-blocked-from-unique-frame-via-window-open-named-sibling-frame.html
* page/DOMWindow.cpp:
(WebCore::DOMWindow::open): Use FrameLoader::findFrameForNavigation() to find the
target frame to navigate with respect to the active document just as we do in WebCore::createWindow().
LayoutTests:
* fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame-expected.txt: Added.
* fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame.html: Added.
* fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame2-expected.txt: Added.
* fast/events/popup-blocked-from-sandboxed-frame-via-window-open-named-sibling-frame2.html: Added.
* fast/events/popup-blocked-from-unique-frame-via-window-open-named-sibling-frame-expected.txt: Added.
* fast/events/popup-blocked-from-unique-frame-via-window-open-named-sibling-frame.html: Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211040 | gns@gnome.org | 2017-01-23 18:20:10 +0000 (Mon, 23 Jan 2017) | 12 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/text/hyphen/HyphenationLibHyphen.cpp
[GTK] asserting on unknown locale for hyphenation is wrong
https://bugs.webkit.org/show_bug.cgi?id=167312
Reviewed by Carlos Garcia Campos.
The fact that we hit the assert on a test called 'hyphenation-unknown-locale' is already
a strong indication we should not have it. In addition to that, Carlos Lopez pointed out
a similar assert was removed from the Mac codepaths when the test was introduced.
* platform/text/hyphen/HyphenationLibHyphen.cpp:
(WebCore::lastHyphenLocation): early return when a locale that is not available is provided
for hyphenation, instead of asserting.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211627 | eocanha@igalia.com | 2017-02-03 12:03:10 +0000 (Fri, 03 Feb 2017) | 24 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamer.cpp
M /trunk/Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamer.h
[GStreamer] Store preloaded media in webkit's cache
https://bugs.webkit.org/show_bug.cgi?id=119477
Reviewed by Xabier Rodriguez-Calvar.
Files cached on disk by MediaPlayerPrivateGStreamer are deleted only when the player is closed. If the
WebProcess crashed, they're just left there in the cache directory. This patch changes the location
of those temporary files to a proper temporary directory (/var/tmp, as those files aren't actually
reusable, so they don't belong to a cache directory, and /tmp is a bad place because it's RAM-based on
some distros), unlinks (deletes) them right after creation and also deletes any other stalled temporary
file on the old legacy cache directory.
There's no API in GstPlaybin to control the temporary file location, so we do it manually by locating
the GstDownloadBuffer element in the pipeline as soon as it's created, reconfiguring it with the right
temporary file path and deleting the file as soon as it's created.
* platform/graphics/gstreamer/MediaPlayerPrivateGStreamer.cpp:
(WebCore::MediaPlayerPrivateGStreamer::~MediaPlayerPrivateGStreamer): Stop listening to element-added.
(WebCore::MediaPlayerPrivateGStreamer::uriDecodeBinElementAddedCallback): Look for GstDownloadBuffer.
(WebCore::MediaPlayerPrivateGStreamer::downloadBufferFileCreatedCallback): Remove the file after creation.
(WebCore::MediaPlayerPrivateGStreamer::purgeOldDownloadFiles): Delete legacy files.
(WebCore::MediaPlayerPrivateGStreamer::sourceChanged): Listen to element-added signals on GstUriDecodeBin.
* platform/graphics/gstreamer/MediaPlayerPrivateGStreamer.h: New reference to GstDownloadBuffer.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211225 | eocanha@igalia.com | 2017-01-26 20:39:57 +0000 (Thu, 26 Jan 2017) | 10 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/gstreamer/GRefPtrGStreamer.cpp
[GStreamer] ASSERTION FAILED: !g_object_is_floating(ptr) in adoptGRef(GstContext* ptr)
https://bugs.webkit.org/show_bug.cgi?id=167458
Reviewed by Xabier Rodriguez-Calvar.
GstContext is a mini_object, not a GObject. Therefore it shouldn't (and can't) be checked for floatingness.
* platform/graphics/gstreamer/GRefPtrGStreamer.cpp:
(WTF::adoptGRef): Removed incorrect assertion.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211815 | carlosgc@webkit.org | 2017-02-07 18:04:50 +0000 (Tue, 07 Feb 2017) | 9 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp
[GStreamer] Deadlock when media player is destroyed
https://bugs.webkit.org/show_bug.cgi?id=167861
Reviewed by Michael Catanzaro.
The problem is that we are calling notifyOne() for the draw mutex without taking the lock.
* platform/graphics/gstreamer/MediaPlayerPrivateGStreamerBase.cpp:
(WebCore::MediaPlayerPrivateGStreamerBase::repaint):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211816 | carlosgc@webkit.org | 2017-02-07 18:08:56 +0000 (Tue, 07 Feb 2017) | 10 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/soup/WebKitSoupRequestInputStream.cpp
[Soup] WebKitSoupRequestInputStream can still leave requests unfinished after r211773
https://bugs.webkit.org/show_bug.cgi?id=167929
Reviewed by Michael Catanzaro.
The while loop to process the pending data requests was not a good idea, because having a new pending request
doesn't mean we have more data available, but that more data was requested.
* WebProcess/soup/WebKitSoupRequestInputStream.cpp:
(webkitSoupRequestInputStreamPendingReadAsyncComplete): Process only one request if needed.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211773 | carlosgc@webkit.org | 2017-02-07 06:12:53 +0000 (Tue, 07 Feb 2017) | 19 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/soup/WebKitSoupRequestInputStream.cpp
[Soup] Long resources loaded by custom protocols sometimes never finish loading
https://bugs.webkit.org/show_bug.cgi?id=167890
Reviewed by Michael Catanzaro.
It's another bug that has appeared in WebKitSoupRequestInputStream after moving the custom protocols handling to
the main thread. The problem is that webkitSoupRequestInputStreamPendingReadAsyncComplete invalidates
pendingAsyncRead after calling webkitSoupRequestInputStreamReadAsyncResultComplete, but in some cases
webkitSoupRequestInputStreamReadAsyncResultComplete completes the task in the same run loop iteration. In that
case webkitSoupRequestInputStreamReadAsync is called again creating a new AsyncReadData that is destroyed right
after webkitSoupRequestInputStreamReadAsyncResultComplete returns.
* WebProcess/soup/WebKitSoupRequestInputStream.cpp:
(AsyncReadData::AsyncReadData): Use an rvalue reference for the task.
(webkitSoupRequestInputStreamPendingReadAsyncComplete): Use WTFMove to ensure pendingAsyncRead is cleared before
webkitSoupRequestInputStreamReadAsyncResultComplete is called, and continue processing pending requests if there
are new ones after webkitSoupRequestInputStreamReadAsyncResultComplete.
(webkitSoupRequestInputStreamReadAsync): Use WTFMove to transfer the task to AsyncReadData.
(webkitSoupRequestInputStreamDidFailWithError): Use WTFMove to ensure pendingAsyncRead is cleared.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211734 | carlosgc@webkit.org | 2017-02-06 18:00:52 +0000 (Mon, 06 Feb 2017) | 14 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/soup/WebKitSoupRequestInputStream.cpp
[Soup] Deadlock in NetworkProcess
https://bugs.webkit.org/show_bug.cgi?id=167876
Reviewed by Michael Catanzaro.
WebKitSoupRequestInputStream uses a read lock. What is happening is that webkitSoupRequestInputStreamAddData
takes the lock, and it calls webkitSoupRequestInputStreamPendingReadAsyncComplete with the lock help. That
causes webkitSoupRequestInputStreamReadAsync to be called again to read the next chunk, but in the same run loop
operation. We don't really need the read lock because both webkitSoupRequestInputStreamAddData and
webkitSoupRequestInputStreamReadAsync shoudl always be called from the main thread.
* WebProcess/soup/WebKitSoupRequestInputStream.cpp:
(webkitSoupRequestInputStreamReadAsync): Remove the read lock and assert if called from a secondary thread.
(webkitSoupRequestInputStreamAddData): Ditto.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211775 | carlosgc@webkit.org | 2017-02-07 06:45:43 +0000 (Tue, 07 Feb 2017) | 36 lines
Changed paths:
M /trunk/LayoutTests/ChangeLog
A /trunk/LayoutTests/fast/scrolling/scroll-animator-overlay-scrollbars-clicked-expected.txt
A /trunk/LayoutTests/fast/scrolling/scroll-animator-overlay-scrollbars-clicked.html
M /trunk/LayoutTests/platform/ios-simulator/TestExpectations
M /trunk/LayoutTests/platform/mac-wk1/TestExpectations
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/page/EventHandler.cpp
M /trunk/Source/WebCore/page/EventHandler.h
Overlay scrolling with iframe-s broken
https://bugs.webkit.org/show_bug.cgi?id=165056
Reviewed by Antonio Gomes.
Source/WebCore:
Mouse press events for overlay scrollbars are ignored if there's a subframe under the scrollbar. This doesn't
happen with normal scrollbars, because the subframe is not really under the scrollbar, so events are always
correctly passed to the scrollbar. With overlay scrollbars, the hit test detects the scrollbar, but events are
always passed first to the subframe. Scrollbars are correctly updated on hover though, because
handleMouseMoveEvent checks the presence of scrollbars before checking for subframes and move events are
actually passed to both, the scrollbar and the subframe. Overlay scrollbars should always take precedence over
subframes to handle mouse press events, so we should check first if mouse is over a scrollbar and never pass the
event to a subframe in that case. Another problem is that the cursor is not updated either when the overlay
scrollbar is hovered and there's a subframe. This is because in handleMouseMoveEvent we pass the event to both
the scrollbar and subframe but we never update the cursor when a suframe was found. So, here again we need to make
an exception for scrollbars and upate the cursor when mouse is over the scrollbar even if a subframe was found.
Test: fast/scrolling/scroll-animator-overlay-scrollbars-clicked.html
* page/EventHandler.cpp:
(WebCore::EventHandler::handleMousePressEvent): Move the scrollbar check before the subframe check.
(WebCore::EventHandler::handleMouseMoveEvent): Update the cursor when hovering a scrollbar even if a subframe
was found.
(WebCore::EventHandler::updateLastScrollbarUnderMouse): Use an enum instead of a boolean for setLast parameter.
* page/EventHandler.h:
LayoutTests:
Add a new test to check that clicking on an overlay scrollbar works even it's over a subframe.
* fast/scrolling/scroll-animator-overlay-scrollbars-clicked-expected.txt: Added.
* fast/scrolling/scroll-animator-overlay-scrollbars-clicked.html: Added.
* platform/ios-simulator/TestExpectations:
* platform/mac-wk1/TestExpectations:
------------------------------------------------------------------------
------------------------------------------------------------------------
r211366 | carlosgc@webkit.org | 2017-01-30 16:47:15 +0000 (Mon, 30 Jan 2017) | 6 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.cpp
Unreviewed. Fix GTK+ debug build after r211365.
Remove invalid assert.
* WebProcess/WebPage/AcceleratedDrawingArea.cpp:
(WebKit::AcceleratedDrawingArea::exitAcceleratedCompositingModeNow):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211206 | zandobersek@gmail.com | 2017-01-26 09:21:46 +0000 (Thu, 26 Jan 2017) | 22 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/cairo/ImageBufferCairo.cpp
ImageBufferCairo: cairo_image_surface should use bmalloc-allocated memory
https://bugs.webkit.org/show_bug.cgi?id=165751
Reviewed by Carlos Garcia Campos.
Allocate the underlying memory for cairo_image_surface objects through FastMalloc.
This way we can steer such large allocations away from the default libc allocator.
Objects of this class can create Cairo surfaces that need as much as 4MB of memory
for the underlying pixel buffer. Allocating such objects through the default
libc allocator can lead to increased memory usage because of non-optimal allocation
strategy in libc. In contrast, bmalloc performs large allocations by directly using
mmap() to reserve the necessary memory.
The improvements can be significant. On nytimes.com, with the threaded version of
the CoordinatedGraphics system, the memory consumption can drop by roughly 20%.
* platform/graphics/cairo/ImageBufferCairo.cpp:
(WebCore::ImageBuffer::ImageBuffer): Zero-allocate the necessary memory via FastMalloc.
Tie that memory lifetime to the lifetime of the surface by using
cairo_surface_set_user_data() with the specific user data key.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211277 | carlosgc@webkit.org | 2017-01-27 08:54:23 +0000 (Fri, 27 Jan 2017) | 15 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/UIProcess/gtk/WebPreferencesGtk.cpp
[GTK] Stop forcing accelerated compositing by default
https://bugs.webkit.org/show_bug.cgi?id=167492
Reviewed by Žan Doberšek.
We are now ready to bring back the on demand accelerated compositing mode with the threaded
compositor. Unfortunately, having AC always enabled brought a lot of issues for many people. Not only it
requires a lot more memory but also caused rendering issues (or even nothing rendered at all) with some graphics
drivers. People started to use WEBKIT_DISABLE_COMPOSITING_MODE as a workaround, but that was not expected to be
used by users. So, entering/leaving AC when required by web contents is not actually the solution but a huge
improvement for many people. If we are eventually ready to force AC mode again, we'll change this again. Note
that we still have WEBKIT_DISABLE_COMPOSITING_MODE, but now also WEBKIT_FORCE_COMPOSITING_MODE for testing and debugging.
* UIProcess/gtk/WebPreferencesGtk.cpp:
(WebKit::WebPreferences::platformInitializeStore):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211365 | carlosgc@webkit.org | 2017-01-30 16:42:12 +0000 (Mon, 30 Jan 2017) | 44 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.h
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.h
M /trunk/Source/WebKit2/WebProcess/WebPage/DrawingAreaImpl.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/LayerTreeHost.h
[GTK] Do not release OpenGL resource immediately when leaving accelerated compositing mode
https://bugs.webkit.org/show_bug.cgi?id=167544
Reviewed by Michael Catanzaro.
Sometimes the conditions to be in AC mode or not change quickly, and then we leave AC mode just enter it again
after a very short period of time. In those cases we are dropping all the GL resources and the compositor
thread, and creating it again. We could keep the layer tree host alive for a while when exiting AC mode, and
reuse it if we enter AC mode before the previous one has been discarded. While the previous layer tree host is
alive we still need to keep it up to date, for example if the web view is resized or contents size change, and
synchronize with the threaded compositor when it becomes the layer tree host again.
* WebProcess/WebPage/AcceleratedDrawingArea.cpp:
(WebKit::AcceleratedDrawingArea::~AcceleratedDrawingArea): Discard the previous layer tree host.
(WebKit::AcceleratedDrawingArea::AcceleratedDrawingArea): Initialize the timer to discard the previous layer
tree host.
(WebKit::AcceleratedDrawingArea::pageBackgroundTransparencyChanged): Notify the previous layer tree host if needed.
(WebKit::AcceleratedDrawingArea::mainFrameContentSizeChanged): Ditto.
(WebKit::AcceleratedDrawingArea::updateBackingStoreState): Ditto.
(WebKit::AcceleratedDrawingArea::enterAcceleratedCompositingMode): Reuse the previous layer tree host if possible.
(WebKit::AcceleratedDrawingArea::exitAcceleratedCompositingModeNow): Exit AC mode and save the layer tree host
starting a timer of 5 seconds to discard it if not reused.
(WebKit::AcceleratedDrawingArea::discardPreviousLayerTreeHost): Invalidate and destroy the previous layer tree host.
(WebKit::AcceleratedDrawingArea::didChangeViewportAttributes): Notify the previous layer tree host if needed.
(WebKit::AcceleratedDrawingArea::deviceOrPageScaleFactorChanged): Ditto.
* WebProcess/WebPage/AcceleratedDrawingArea.h:
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp:
(WebKit::ThreadedCoordinatedLayerTreeHost::scrollNonCompositedContents): If it's discardable add the action to
be synchronized instead.
(WebKit::ThreadedCoordinatedLayerTreeHost::contentsSizeChanged): Ditto.
(WebKit::ThreadedCoordinatedLayerTreeHost::deviceOrPageScaleFactorChanged): Ditto.
(WebKit::ThreadedCoordinatedLayerTreeHost::pageBackgroundTransparencyChanged): Ditto.
(WebKit::ThreadedCoordinatedLayerTreeHost::sizeDidChange): Ditto.
(WebKit::ThreadedCoordinatedLayerTreeHost::didChangeViewportAttributes): Ditto.
(WebKit::ThreadedCoordinatedLayerTreeHost::setIsDiscardable): When the layer tree host becomes discardable,
reset the sync actions and return. When it becomes the real layer tree host again, apply all pending actions to
synchronize with the threaded compositor.
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.h:
* WebProcess/WebPage/DrawingAreaImpl.cpp:
(WebKit::DrawingAreaImpl::scroll): Notify the previous layer tree host if needed.
(WebKit::DrawingAreaImpl::mainFrameContentSizeChanged): Ditto.
(WebKit::DrawingAreaImpl::exitAcceleratedCompositingMode): Use AcceleratedDrawingArea::exitAcceleratedCompositingModeNow().
* WebProcess/WebPage/LayerTreeHost.h:
(WebKit::LayerTreeHost::setIsDiscardable): Added.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211350 | carlosgc@webkit.org | 2017-01-29 10:06:30 +0000 (Sun, 29 Jan 2017) | 30 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/SimpleViewportController.cpp
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/SimpleViewportController.h
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.h
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.h
M /trunk/Source/WebKit2/WebProcess/WebPage/DrawingArea.h
M /trunk/Source/WebKit2/WebProcess/WebPage/LayerTreeHost.h
M /trunk/Source/WebKit2/WebProcess/WebPage/WebPage.cpp
[Coordinated Graphics] WebPage shouldn't use the layerTreeHost directly
https://bugs.webkit.org/show_bug.cgi?id=167494
Reviewed by Michael Catanzaro.
In Coordinated Graphics we have a couple of methods that the WebPage uses directly from the layer tree host,
instead of using the drawing area interface. This patch adds DrawingArea::didChangeViewportAttributes and
DrawingArea::deviceOrPageScaleFactorChanged and renames LayerTreeHost::didChangeViewportProperties as
LayerTreeHost::didChangeViewportAttributes for consistency.
* Shared/CoordinatedGraphics/SimpleViewportController.cpp:
(WebKit::SimpleViewportController::didChangeViewportAttributes): Receive an rvalue reference to avoid copies.
* Shared/CoordinatedGraphics/SimpleViewportController.h:
* WebProcess/WebPage/AcceleratedDrawingArea.cpp:
(WebKit::AcceleratedDrawingArea::didChangeViewportAttributes): Forward it to the layer tree host if any.
(WebKit::AcceleratedDrawingArea::deviceOrPageScaleFactorChanged): Ditto.
* WebProcess/WebPage/AcceleratedDrawingArea.h:
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp:
(WebKit::ThreadedCoordinatedLayerTreeHost::didChangeViewportAttributes): Renamed and updated to pass an rvalue reference.
(WebKit::ThreadedCoordinatedLayerTreeHost::didChangeViewportProperties): Deleted.
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.h:
* WebProcess/WebPage/DrawingArea.h:
* WebProcess/WebPage/LayerTreeHost.h:
(WebKit::LayerTreeHost::didChangeViewportAttributes): Renamed and updated to pass an rvalue reference.
(WebKit::LayerTreeHost::didChangeViewportProperties): Deleted.
* WebProcess/WebPage/WebPage.cpp:
(WebKit::WebPage::sendViewportAttributesChanged): Use the drawing area.
(WebKit::WebPage::scalePage): Ditto
(WebKit::WebPage::setDeviceScaleFactor): Ditto.
(WebKit::WebPage::viewportPropertiesDidChange): Ditto.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211358 | carlosgc@webkit.org | 2017-01-30 08:47:30 +0000 (Mon, 30 Jan 2017) | 11 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperPlatformLayerProxy.cpp
[Threaded Compositor] Crash in GraphicsContext3D::deleteTexture when destroying TextureMapperPlatformLayerProxy
https://bugs.webkit.org/show_bug.cgi?id=167575
Reviewed by Žan Doberšek.
We should clear all the buffers on invalidate to ensure we don't have textures alive after CoordinatedGraphicsScene::purgeGLResources().
Fix crash in media/video-poster-background.html.
* platform/graphics/texmap/TextureMapperPlatformLayerProxy.cpp:
(WebCore::TextureMapperPlatformLayerProxy::invalidate): Clear current, pending and all used buffers.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211357 | carlosgc@webkit.org | 2017-01-30 05:55:34 +0000 (Mon, 30 Jan 2017) | 20 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperPlatformLayerProxy.cpp
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/CoordinatedGraphicsScene.cpp
[Threaded Compositor] Crash on WebCore::GLContext::version()
https://bugs.webkit.org/show_bug.cgi?id=167559
Reviewed by Michael Catanzaro.
Source/WebCore:
Fixes crashes in several media tests.
* platform/graphics/texmap/TextureMapperPlatformLayerProxy.cpp:
(WebCore::TextureMapperPlatformLayerProxy::invalidate): Clear m_compositorThreadUpdateTimer and call the update function.
Source/WebKit2:
This is happening because TextureMapperPlatformLayerProxy::compositorThreadUpdateTimerFired() is fired after the
threaded compositor is deleted. CoordinatedGraphicsScene::purgeGLResources() should invalidate the proxies
before clearing the map.
* Shared/CoordinatedGraphics/CoordinatedGraphicsScene.cpp:
(WebKit::CoordinatedGraphicsScene::purgeGLResources):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211367 | magomez@igalia.com | 2017-01-30 16:53:09 +0000 (Mon, 30 Jan 2017) | 12 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebCoreSupport/WebChromeClient.cpp
[GTK] Scrolling iframes, doesn't redraw their content
https://bugs.webkit.org/show_bug.cgi?id=167581
Reviewed by Carlos Garcia Campos.
Take into account whether we are using AC or not in order to repaint an area after scrolling.
No behaviour change, no new tests.
* WebProcess/WebCoreSupport/WebChromeClient.cpp:
(WebKit::WebChromeClient::invalidateContentsForSlowScroll):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211348 | carlosgc@webkit.org | 2017-01-29 07:16:51 +0000 (Sun, 29 Jan 2017) | 10 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/CoordinatedGraphicsScene.cpp
[Threaded Compositor] Crash when detaching the CoordinatedGraphicsScene
https://bugs.webkit.org/show_bug.cgi?id=167547
Reviewed by Michael Catanzaro.
It seems that commitSceneState() can be called after the CoordinatedGraphicsScene has been detached.
* Shared/CoordinatedGraphics/CoordinatedGraphicsScene.cpp:
(WebKit::CoordinatedGraphicsScene::commitSceneState): Return early if scene has been detached.
(WebKit::CoordinatedGraphicsScene::detach): Take the render queue lock before clearing the render queue.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211347 | carlosgc@webkit.org | 2017-01-29 07:15:33 +0000 (Sun, 29 Jan 2017) | 11 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/threadedcompositor/CompositingRunLoop.cpp
[Threaded Compositor] Crash when deleting the compositor run loop
https://bugs.webkit.org/show_bug.cgi?id=167545
Reviewed by Michael Catanzaro.
The problem is that we are releasing the WorkQueue before the update timer that keeps a reference to the run
loop, destroyed by the WorkQueue. So, invalidate the WorkQueue in the next run loop iteration to ensure it
happens after the CompositingRunLoop destructor.
* Shared/CoordinatedGraphics/threadedcompositor/CompositingRunLoop.cpp:
(WebKit::CompositingRunLoop::~CompositingRunLoop):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211346 | carlosgc@webkit.org | 2017-01-29 07:14:12 +0000 (Sun, 29 Jan 2017) | 13 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/DrawingAreaImpl.cpp
[GTK] ASSERTION FAILED: !m_layerTreeHost in DrawingAreaImpl::display()
https://bugs.webkit.org/show_bug.cgi?id=167548
Reviewed by Michael Catanzaro.
The problem is that non accelerated compositing forceRepaint implementation is doing a layout and then calling
display. The layout makes the drawing area enter in AC mode and display asserts that we have a layer tree
host. forceRepaint shouldn't do the layout because display already does that and it correctly handles the case
of entering AC mode during the layout. It shouldn't call setNeedsDisplay either, because that schedules a
display, but we are going to display synchronously.
* WebProcess/WebPage/DrawingAreaImpl.cpp:
(WebKit::DrawingAreaImpl::forceRepaint):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211281 | carlosgc@webkit.org | 2017-01-27 12:10:56 +0000 (Fri, 27 Jan 2017) | 22 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.cpp
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.h
[Threaded Compositor] Stop creating the GLContext on demand the first time makeContextCurrent is called
https://bugs.webkit.org/show_bug.cgi?id=167496
Reviewed by Žan Doberšek.
This is causing problems with animations when entering AC mode on demand. What happens is that the threaded
compositor is created, then the animation is scheduled and during the first animation iteration the GLContext is
created, making the first frame of the animation quite slow. In my computer creating the GLContext takes 0.8
seconds. If the animation duration is less than the time it takes to create the GLContext, the animation ends
without iterating. This causing timeouts in the bots in tests like
animations/animation-iteration-event-destroy-renderer.html that expect webkitAnimationIteration events that
never fire.
* Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.cpp:
(WebKit::ThreadedCompositor::ThreadedCompositor): Create the GLContext right after the compositing thread is
created if we already have a native surface handle.
(WebKit::ThreadedCompositor::createGLContext): Helper to create the GLContext.
(WebKit::ThreadedCompositor::setNativeSurfaceHandleForCompositing): Create the GLContext when a native surface
handle is given.
(WebKit::ThreadedCompositor::makeContextCurrent): Deleted.
(WebKit::ThreadedCompositor::renderLayerTree): Make the context cunrrent directly here.
* Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.h:
------------------------------------------------------------------------
------------------------------------------------------------------------
r211205 | carlosgc@webkit.org | 2017-01-26 09:19:48 +0000 (Thu, 26 Jan 2017) | 12 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp
[Threaded Compositor] Update also the contents size when creating the threaded compositor
https://bugs.webkit.org/show_bug.cgi?id=167452
Reviewed by Žan Doberšek.
In r210954 we ensured that the threaded compositor and the viewport controller were created with the initial web
page size. If we don't update the contents size, the visible rectangle will be empty and tiles won't be created
until contentsSizeDidChange is called. It's even possible, when entering AC mode on demand, that we create the
layer tree host after the contents size changed, in which case nothing is rendered unless we resize the window.
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp:
(WebKit::ThreadedCoordinatedLayerTreeHost::ThreadedCoordinatedLayerTreeHost):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211204 | magomez@igalia.com | 2017-01-26 09:05:01 +0000 (Thu, 26 Jan 2017) | 13 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperGL.cpp
[GTK] WebProcess from WebKitGtk+ 2.15.3 SIGSEVs in WebCore::GraphicsContext3D::drawArrays(unsigned int, int, int) at Source/WebCore/platform/graphics/opengl/GraphicsContext3DOpenGLCommon.cpp:740
https://bugs.webkit.org/show_bug.cgi?id=167296
Reviewed by Sergio Villar Senin.
Use a VBO to pass the vertex data when using the stencil for clipping. Passing a custom array without
using an VBO is deprecated when using OpenGL >= 3.0 and a Core profile.
No behaviour change, no new tests.
* platform/graphics/texmap/TextureMapperGL.cpp:
(WebCore::TextureMapperGL::beginClip):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211145 | carlosgc@webkit.org | 2017-01-25 16:54:28 +0000 (Wed, 25 Jan 2017) | 25 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/x11/PlatformDisplayX11.cpp
M /trunk/Source/WebCore/platform/graphics/x11/PlatformDisplayX11.h
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/UIProcess/gtk/AcceleratedBackingStoreX11.cpp
M /trunk/Source/WebKit2/UIProcess/gtk/WebPreferencesGtk.cpp
[GTK] UIProcess from WebKitGtk+ 2.15.x SIGSEGVs because of X Error BadDamage in WebKit::AcceleratedBackingStoreX11::update(WebKit::LayerTreeContext const&) () at Source/WebKit2/UIProcess/gtk/AcceleratedBackingStoreX11.cpp:145
https://bugs.webkit.org/show_bug.cgi?id=165656
Reviewed by Michael Catanzaro.
Source/WebCore:
Also return the base error code from PlatformDisplayX11::supportsXDamage().
* platform/graphics/x11/PlatformDisplayX11.cpp:
(WebCore::PlatformDisplayX11::supportsXDamage):
* platform/graphics/x11/PlatformDisplayX11.h:
Source/WebKit2:
We are incorrectly handling BadDamage errors because the BadDamage value we pass to the XErrorTrapper is not
the actual error code used by X11. Since XDamage is an extension, it has its own errors and a base error
code. We need to use the base error code we get when calling XDamageQueryExtension to pass the right error code
to the XErrorTrapper.
* UIProcess/gtk/AcceleratedBackingStoreX11.cpp:
(WebKit::AcceleratedBackingStoreX11::create): Get also the damage base error.
(WebKit::xDamageErrorCode): Helper to get the actual error code.
(WebKit::AcceleratedBackingStoreX11::~AcceleratedBackingStoreX11): Use xDamageErrorCode().
(WebKit::AcceleratedBackingStoreX11::update): Ditto.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211141 | magomez@igalia.com | 2017-01-25 13:26:01 +0000 (Wed, 25 Jan 2017) | 31 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/page/PageOverlayController.cpp
M /trunk/Source/WebCore/page/PageOverlayController.h
M /trunk/Source/WebCore/rendering/RenderLayerCompositor.cpp
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebCoreSupport/WebInspectorClient.cpp
[GTK] The inspector is broken when AC support is disabled
https://bugs.webkit.org/show_bug.cgi?id=165237
Reviewed by Carlos Garcia Campos.
Source/WebCore:
Ensure that PageOverlayController destroys its root GraphicsLayers when leaving AC. They
will be recreated if we enter AC again, avoiding a crash because the layers don't belong to
the appropriate compositor. Also, don't append the document overlay layers if we are not
in compositing mode.
No new tests added.
* page/PageOverlayController.cpp:
(WebCore::PageOverlayController::willDetachRootLayer):
* page/PageOverlayController.h:
* rendering/RenderLayerCompositor.cpp:
(WebCore::RenderLayerCompositor::appendDocumentOverlayLayers):
(WebCore::RenderLayerCompositor::detachRootLayer):
Source/WebKit2:
WebInspector must not use PageOverlays or GraphicsLayers when AC is not avaialable.
No new tests added.
* WebProcess/WebCoreSupport/WebInspectorClient.cpp:
(WebKit::WebInspectorClient::highlight):
(WebKit::WebInspectorClient::showPaintRect):
------------------------------------------------------------------------
------------------------------------------------------------------------
r211084 | magomez@igalia.com | 2017-01-24 16:25:31 +0000 (Tue, 24 Jan 2017) | 13 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/rendering/RenderLayerBacking.cpp
[GTK] Do not paint non composited content into the window when using the threaded compositor
https://bugs.webkit.org/show_bug.cgi?id=167367
Reviewed by Carlos Garcia Campos.
When using the threaded compositor we need to send the non composited content for compositing as well,
not painting it directly into the window.
No new tests.
* rendering/RenderLayerBacking.cpp:
(WebCore::RenderLayerBacking::paintsIntoWindow):
------------------------------------------------------------------------
------------------------------------------------------------------------
r210954 | carlosgc@webkit.org | 2017-01-20 06:55:46 +0000 (Fri, 20 Jan 2017) | 22 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/SimpleViewportController.cpp
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/SimpleViewportController.h
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.cpp
M /trunk/Source/WebKit2/Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.h
M /trunk/Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp
[Threaded Compositor] Initialize the threaded compositor with the current size
https://bugs.webkit.org/show_bug.cgi?id=167196
Reviewed by Žan Doberšek.
We are always creating the threaded compositor with an empty size and then a sizeDidChange always happen when
the backing store state changes. This is always happening because the threaded compositor is created before the
first backing store state, but if we wanted to create it later, for example to enter/leave AC mode on demand,
the threaded compositor will not have the viewport size unless the window is resized, or sizeDidChange is called
manually when entering AC mode. Creating the threaded compositor is sync and changing the size too, so it's
better to do both things at the same time using the same sync operation.
* Shared/CoordinatedGraphics/SimpleViewportController.cpp:
(WebKit::SimpleViewportController::SimpleViewportController): Pass an initial size to the constructor.
* Shared/CoordinatedGraphics/SimpleViewportController.h:
* Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.cpp:
(WebKit::ThreadedCompositor::create): Add viewportSize and scaleFactor construction parameters,
(WebKit::ThreadedCompositor::ThreadedCompositor): Ditto. Also mark as needs resize if the given size is not empty.
* Shared/CoordinatedGraphics/threadedcompositor/ThreadedCompositor.h:
* WebProcess/WebPage/CoordinatedGraphics/ThreadedCoordinatedLayerTreeHost.cpp:
(WebKit::ThreadedCoordinatedLayerTreeHost::ThreadedCoordinatedLayerTreeHost): Initialize the threaded compositor
with an initial viewport size and scale factor.
------------------------------------------------------------------------
------------------------------------------------------------------------
r211083 | magomez@igalia.com | 2017-01-24 14:10:43 +0000 (Tue, 24 Jan 2017) | 14 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/WebProcess/WebPage/AcceleratedDrawingArea.cpp
[Coordinated Graphics] Ensure that we're in AC mode before trying to create GraphicsLayers
https://bugs.webkit.org/show_bug.cgi?id=167365
Reviewed by Carlos Garcia Campos.
When entering AC on demand, the RenderLayerCompositor wants to create the root GraphicsLayer before the
AcceleratedDrawingArea is in AC mode. This means there's no LayerTreeHost to request the GraphicsLayerFactory,
so no layers are created. Ensure that we are in AC mode before requesting the GraphicsLayerFactory.
No behaviour change, no new tests.
* WebProcess/WebPage/AcceleratedDrawingArea.cpp:
(WebKit::AcceleratedDrawingArea::graphicsLayerFactory):
------------------------------------------------------------------------
------------------------------------------------------------------------
r210920 | carlosgc@webkit.org | 2017-01-19 08:59:42 +0000 (Thu, 19 Jan 2017) | 17 lines
Changed paths:
M /trunk/Source/WebKit2/ChangeLog
M /trunk/Source/WebKit2/UIProcess/DrawingAreaProxyImpl.cpp
[GTK] Do not update the backing store state unnecessarily when page visibility changes
https://bugs.webkit.org/show_bug.cgi?id=167195
Reviewed by Sergio Villar Senin.
I've noticed that we are receiving UpdateBackingStoreState messages in the web process even when size and scale
factor didn't change. That's quite unfortunate because we do a lot of work unnecessarily and we tell the
threaded compositor that the size changed. This is not only a problem because it's not true, but also because
changing the size is a sync operation in the threaded compositor, so we block the compositing thread for
nothing. This is happening because the WebPageProxy notifies the drawing area that the backing store is
discardable when the page is not visible, and that always produces a new backing store state. In accelerating
compositing mode we don't even have a backing store to discard, so we should check we have something to discard
and also make sure we only generate a new backing store state if we really discarded the backing store.
* UIProcess/DrawingAreaProxyImpl.cpp:
(WebKit::DrawingAreaProxyImpl::discardBackingStoreSoon):
(WebKit::DrawingAreaProxyImpl::discardBackingStore):
------------------------------------------------------------------------
------------------------------------------------------------------------
r210847 | magomez@igalia.com | 2017-01-18 08:50:16 +0000 (Wed, 18 Jan 2017) | 19 lines
Changed paths:
M /trunk/Source/WebCore/ChangeLog
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperGL.cpp
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperShaderProgram.cpp
M /trunk/Source/WebCore/platform/graphics/texmap/TextureMapperShaderProgram.h
[GTK] [TextureMapper] [GLES2] Draw repeated patterns for NPOT textures manually
https://bugs.webkit.org/show_bug.cgi?id=167118
Reviewed by Žan Doberšek.
When using GLES2 without the GL_OES_texture_npot extension, some implementations fail to properly
draw repeated patters from a texture (using GL_REPEAT). For those cases, perform the repetition
manually using the shader.
Covered by existent tests.
* platform/graphics/texmap/TextureMapperGL.cpp:
(WebCore::driverSupportsNPOTTextures):
(WebCore::TextureMapperGL::drawTexture):
(WebCore::TextureMapperGL::drawTexturedQuadWithProgram):
* platform/graphics/texmap/TextureMapperShaderProgram.cpp:
(WebCore::TextureMapperShaderProgram::create):
* platform/graphics/texmap/TextureMapperShaderProgram.h:
------------------------------------------------------------------------
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment