Fix ASSERT failure in RenderGeomtryMap for SVGs

Source/core/rendering/svg/SVGRenderSupport.cpp

 /*
  * Copyright (C) 2007, 2008 Rob Buis <buis@kde.org>
  * Copyright (C) 2007 Nikolas Zimmermann <zimmermann@kde.org>
  * Copyright (C) 2007 Eric Seidel <eric@webkit.org>
  * Copyright (C) 2009 Google, Inc.  All rights reserved.
  * Copyright (C) 2009 Dirk Schulze <krit@webkit.org>
  * Copyright (C) Research In Motion Limited 2009-2010. All rights reserved.
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Library General Public
@@ skipping 49 matching lines @@
 {
     repaintRect.inflate(object->style()->outlineWidth());
 
     // Translate to coords in our parent renderer, and then call computeFloatRectForRepaint() on our parent.
     repaintRect = object->localToParentTransform().mapRect(repaintRect);
     object->parent()->computeFloatRectForRepaint(repaintContainer, repaintRect, fixed);
 }
 
 void SVGRenderSupport::mapLocalToContainer(const RenderObject* object, const RenderLayerModelObject* repaintContainer, TransformState& transformState, bool* wasFixed)
 {
-    transformState.applyTransform(object->localToParentTransform());
-
     RenderObject* parent = object->parent();
 
     // At the SVG/HTML boundary (aka RenderSVGRoot), we apply the localToBorderBoxTransform
     // to map an element from SVG viewport coordinates to CSS box coordinates.
     // RenderSVGRoot's mapLocalToContainer method expects CSS box coordinates.
     if (parent->isSVGRoot())
         transformState.applyTransform(toRenderSVGRoot(parent)->localToBorderBoxTransform());
 
+    transformState.applyTransform(object->localToParentTransform());

[fs, 2014/01/21 09:07:57]

This doesn't look right to me - I take it that in the relevant case, |object| is
the <g transform="scale(3)">, and hence that |parent| is the <svg ...>? If so,
then the currently local coordinate space is the coordinate space of the <g> -
from which we get to the local coordinate space of the <svg> by applying the
scale(3) - yes? Then to get to the "CSS (border) box" space by applying the
transform produced by the viewBox (mapping the content-box width/height to a
specified local coordinate space). And then RenderBox adds in any extras. Or
maybe it's me missing something? =)

[Stephen Chennney, 2014/01/21 22:12:11]

I think that, like me, you are confused by the ambiguity of "applyTransform",
which apparently multiplies it to the current transform on the right. In other
words, the last transform to be applied in code will actually be the first one
applied mathematically. I don't know who wrote this, but it's as confusing as
can be.

+
     MapCoordinatesFlags mode = UseTransforms;
     parent->mapLocalToContainer(repaintContainer, transformState, mode, wasFixed);
 }
 
 const RenderObject* SVGRenderSupport::pushMappingToContainer(const RenderObject* object, const RenderLayerModelObject* ancestorToStopAt, RenderGeometryMap& geometryMap)
 {
     ASSERT_UNUSED(ancestorToStopAt, ancestorToStopAt != object);
 
     RenderObject* parent = object->parent();
 
@@ skipping 316 matching lines @@
 
 bool SVGRenderSupport::isEmptySVGInlineText(const RenderObject* object)
 {
     // RenderSVGInlineText performs whitespace filtering in order to support xml:space
     // (http://www.w3.org/TR/SVG/struct.html#LangSpaceAttrs), and can end up with an empty string
     // even when its original constructor argument is non-empty.
     return object->isSVGInlineText() && toRenderSVGInlineText(object)->hasEmptyText();
 }
 
 }