Title: [139479] trunk
- Revision
- 139479
- Author
- le...@chromium.org
- Date
- 2013-01-11 11:59:48 -0800 (Fri, 11 Jan 2013)
Log Message
RenderGeometryMap and TransformState disagree with sub-pixel layout and translations
https://bugs.webkit.org/show_bug.cgi?id=106047
Reviewed by Simon Fraser.
Source/WebCore:
Mirror RenderGeometryMap's optimization for integer-translated transforms in TransformState.
This avoids the current behavior where the two can disagree on mappings, since RenderGeometryMap
pixel-snapped later when a translation occurred between two sub-pixel containers.
Test: fast/layers/geometry-map-transform-state-translation-mismatch.html
* platform/graphics/transforms/TransformState.h:
(WebCore::TransformState::setQuad): Clear accumulatedOffset when setting a new quad. Note: this
implementation only works properly when only tracking a quad.
* platform/graphics/transforms/TransformState.cpp:
(WebCore::TransformState::applyTransform): apply integral translations to the accumulatedOffset
for performance and consistency with RenderGeometryMap.
LayoutTests:
* fast/layers/geometry-map-transform-state-translation-mismatch-expected.txt: Added.
* fast/layers/geometry-map-transform-state-translation-mismatch.html: Added.
Modified Paths
Added Paths
Diff
Modified: trunk/LayoutTests/ChangeLog (139478 => 139479)
--- trunk/LayoutTests/ChangeLog 2013-01-11 19:58:00 UTC (rev 139478)
+++ trunk/LayoutTests/ChangeLog 2013-01-11 19:59:48 UTC (rev 139479)
@@ -1,3 +1,13 @@
+2013-01-11 Levi Weintraub <le...@chromium.org>
+
+ RenderGeometryMap and TransformState disagree with sub-pixel layout and translations
+ https://bugs.webkit.org/show_bug.cgi?id=106047
+
+ Reviewed by Simon Fraser.
+
+ * fast/layers/geometry-map-transform-state-translation-mismatch-expected.txt: Added.
+ * fast/layers/geometry-map-transform-state-translation-mismatch.html: Added.
+
2013-01-11 Stephen Chenney <schen...@chromium.org>
[Chromium] Mac editing test rebaseline
Added: trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch-expected.txt (0 => 139479)
--- trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch-expected.txt (rev 0)
+++ trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch-expected.txt 2013-01-11 19:59:48 UTC (rev 139479)
@@ -0,0 +1,2 @@
+This tests that RenderGeometryMap and TransformState both avoid pixel snapping prior to applying translation-only transforms. The test passes if it doesn't assert on Debug.
+
Added: trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch.html (0 => 139479)
--- trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch.html (rev 0)
+++ trunk/LayoutTests/fast/layers/geometry-map-transform-state-translation-mismatch.html 2013-01-11 19:59:48 UTC (rev 139479)
@@ -0,0 +1,33 @@
+<!DOCTYPE html>
+<html>
+<style>
+.wrapper {
+ width: 200px;
+ margin-left: 101.7px;
+}
+
+.content {
+ background: rgba(255, 0, 0, 0.2);
+ -webkit-transform: translateZ(0);
+ width: 100px;
+}
+
+.inner {
+ background: rgba(0, 0, 255, 0.2);
+ position: relative;
+ left: -6.25px;
+}
+</style>
+<script>
+if (window.testRunner)
+ testRunner.dumpAsText();
+</script>
+<body>
+<div id="description">This tests that RenderGeometryMap and TransformState both avoid pixel snapping prior to applying translation-only transforms.
+The test passes if it doesn't assert on Debug.</div>
+<div class="wrapper">
+ <div class="content">
+ <div class="inner"></div>
+ </div>
+</div>
+</body>
Modified: trunk/Source/WebCore/ChangeLog (139478 => 139479)
--- trunk/Source/WebCore/ChangeLog 2013-01-11 19:58:00 UTC (rev 139478)
+++ trunk/Source/WebCore/ChangeLog 2013-01-11 19:59:48 UTC (rev 139479)
@@ -1,3 +1,23 @@
+2013-01-11 Levi Weintraub <le...@chromium.org>
+
+ RenderGeometryMap and TransformState disagree with sub-pixel layout and translations
+ https://bugs.webkit.org/show_bug.cgi?id=106047
+
+ Reviewed by Simon Fraser.
+
+ Mirror RenderGeometryMap's optimization for integer-translated transforms in TransformState.
+ This avoids the current behavior where the two can disagree on mappings, since RenderGeometryMap
+ pixel-snapped later when a translation occurred between two sub-pixel containers.
+
+ Test: fast/layers/geometry-map-transform-state-translation-mismatch.html
+
+ * platform/graphics/transforms/TransformState.h:
+ (WebCore::TransformState::setQuad): Clear accumulatedOffset when setting a new quad. Note: this
+ implementation only works properly when only tracking a quad.
+ * platform/graphics/transforms/TransformState.cpp:
+ (WebCore::TransformState::applyTransform): apply integral translations to the accumulatedOffset
+ for performance and consistency with RenderGeometryMap.
+
2013-01-11 Abhishek Arya <infe...@chromium.org>
Heap-use-after-free in WebCore::RenderText::computePreferredLogicalWidths
Modified: trunk/Source/WebCore/platform/graphics/transforms/TransformState.cpp (139478 => 139479)
--- trunk/Source/WebCore/platform/graphics/transforms/TransformState.cpp 2013-01-11 19:58:00 UTC (rev 139478)
+++ trunk/Source/WebCore/platform/graphics/transforms/TransformState.cpp 2013-01-11 19:59:48 UTC (rev 139479)
@@ -111,6 +111,11 @@
if (wasClamped)
*wasClamped = false;
+ if (transformFromContainer.isIntegerTranslation()) {
+ move(LayoutSize(transformFromContainer.e(), transformFromContainer.f()), accumulate);
+ return;
+ }
+
applyAccumulatedOffset();
// If we have an accumulated transform from last time, multiply in this transform
Modified: trunk/Source/WebCore/platform/graphics/transforms/TransformState.h (139478 => 139479)
--- trunk/Source/WebCore/platform/graphics/transforms/TransformState.h 2013-01-11 19:58:00 UTC (rev 139478)
+++ trunk/Source/WebCore/platform/graphics/transforms/TransformState.h 2013-01-11 19:59:48 UTC (rev 139479)
@@ -73,7 +73,14 @@
TransformState& operator=(const TransformState&);
- void setQuad(const FloatQuad& quad) { m_lastPlanarQuad = quad; }
+ void setQuad(const FloatQuad& quad)
+ {
+ // FIXME: this assumes that the quad being added is in the coordinate system of the current state.
+ // This breaks if we're simultaneously mapping a point. https://bugs.webkit.org/show_bug.cgi?id=106680
+ ASSERT(!m_mapPoint);
+ m_accumulatedOffset = LayoutSize();
+ m_lastPlanarQuad = quad;
+ }
void move(LayoutUnit x, LayoutUnit y, TransformAccumulation accumulate = FlattenTransform)
{
_______________________________________________
webkit-changes mailing list
webkit-changes@lists.webkit.org
http://lists.webkit.org/mailman/listinfo/webkit-changes
