Support currentTransform in 2D Canvas.

Source/core/html/canvas/CanvasRenderingContext2D.cpp

Index: Source/core/html/canvas/CanvasRenderingContext2D.cpp
diff --git a/Source/core/html/canvas/CanvasRenderingContext2D.cpp b/Source/core/html/canvas/CanvasRenderingContext2D.cpp
index 4bc56b201c406dea598180b30114f4f276b1512c..dc6a715568c025621e8982e815ee922366330fca 100644
--- a/Source/core/html/canvas/CanvasRenderingContext2D.cpp
+++ b/Source/core/html/canvas/CanvasRenderingContext2D.cpp
@@ -596,6 +596,11 @@ void CanvasRenderingContext2D::setGlobalCompositeOperation(const String& operati
     c->setCompositeOperation(op, blendMode);
 }
 
+void CanvasRenderingContext2D::setCurrentTransform(const SVGMatrix& matrix)
+{
+    setTransform(matrix.a(), matrix.b(), matrix.c(), matrix.d(), matrix.e(), matrix.f());
+}
+
 void CanvasRenderingContext2D::scale(float sx, float sy)
 {
     GraphicsContext* c = drawingContext();
@@ -698,12 +703,12 @@ void CanvasRenderingContext2D::transform(float m11, float m12, float m21, float
 
     realizeSaves();
 
+    modifiableState().m_transform = newTransform;

[Justin Novosad, 2013/10/09 17:17:04]

Interesting bug you fixed here, but is this the right way to fix it?  I am
guessing this is to fix the case of a setCurrentTransform/currentTransform
round-trip with a non-invertible matrix? With this change, in the case where the
transform is non-invertible, the transform will now be retained in the canvas
state, but won't be applied to the context because the call to c->concatCTM
below will be skipped. I think that is wrong. The canvas state and the graphics
context state should remain in sync. Probably not a big deal given Blink's
implementation of 2D canvas does not currently handle non-invertible transforms
correctly, but it is something we should eventually fix, and we can start right
here.

[dshwang, 2013/10/09 17:48:20]

exactly. This change is only for currentTransform with a non-invertible matrix.
We already have some inconsistent states between canvas and GraphicsContext
For example,
1. canvas calls save() lazily
2. When Canvas.currentTransform is identity, GraphicsContext.CTM is
HTMLCanvasElement::baseTransform()

In addtion, we never use non-invertible modifiableState().m_transform because we
always check modifiableState().m_invertibleCTM before using
modifiableState().m_transform.
IMO there will be nothing that we should eventually fix.


Nevertheless, if we are afraid this approach, there would be two options.
1. Add new variable to trace currentTransform although non-invertible.
  modifiableState().m_transform will be consistent with js world
  modifiableState().m_effectiveTransform will be consistent with GraphicsContext

2. Make GraphicsContext handle non-invertible matrix.
GraphicsContext can not handle non-invertible matrix. GraphicsContext just calls
SkCanvas::concat().
Introduce m_invertibleCTM in GraphicsContext and check m_invertibleCTM in every
methods before using SkCanvas like CanvasRenderingContext2D.

WDYT?

     if (!newTransform.isInvertible()) {
         modifiableState().m_invertibleCTM = false;
         return;
     }
 
-    modifiableState().m_transform = newTransform;
     c->concatCTM(transform);
     m_path.transform(transform.inverse());
 }