Add support for new canvas ellipse method.

Source/core/html/canvas/CanvasPathMethods.cpp

Index: Source/core/html/canvas/CanvasPathMethods.cpp
diff --git a/Source/core/html/canvas/CanvasPathMethods.cpp b/Source/core/html/canvas/CanvasPathMethods.cpp
index 5053696562cb38d7d19c84453fc190fb0964628e..5ea30de9a4d2158e44de10ef4aa9dfa25e885762 100644
--- a/Source/core/html/canvas/CanvasPathMethods.cpp
+++ b/Source/core/html/canvas/CanvasPathMethods.cpp
@@ -130,6 +130,16 @@ void CanvasPathMethods::arcTo(float x1, float y1, float x2, float y2, float r, E
         m_path.addArcTo(p1, p2, r);
 }
 
+static float adjustEndAngle(float sa, float ea, bool anticlockwise)
+{
+    // If 'sa' and 'ea' differ by more than 2Pi, just add a circle starting/ending at 'sa'.
+    if (anticlockwise && sa - ea >= 2 * piFloat)
+        return sa - 2 * piFloat;

[Stephen Chennney, 2013/04/26 14:51:03]

This is not correct. If they differ by more than 4 pi you will fail. The right
way to do this is with a while loop.

And what about cases where you are told to go anticlockwise and the angle is
less than -2 pi (or negative at all?).

[Stephen Chennney, 2013/04/26 14:57:56]

I see, it is correct according to the spec. But then you know you have a
complete ellipse, so it should be optimizable somehow.

+    if (!anticlockwise && ea - sa >= 2 * piFloat)
+        return sa + 2 * piFloat;
+    return ea;
+}
+
 void CanvasPathMethods::arc(float x, float y, float r, float sa, float ea, bool anticlockwise, ExceptionCode& ec)
 {
     ec = 0;
@@ -150,19 +160,36 @@ void CanvasPathMethods::arc(float x, float y, float r, float sa, float ea, bool
     if (!isTransformInvertible())
         return;
 
-    // If 'sa' and 'ea' differ by more than 2Pi, just add a circle starting/ending at 'sa'.
-    if (anticlockwise && sa - ea >= 2 * piFloat) {
-        m_path.addArc(FloatPoint(x, y), r, sa, sa - 2 * piFloat, anticlockwise);
+    float adjustedEndAngle = adjustEndAngle(sa, ea, anticlockwise);
+    m_path.addArc(FloatPoint(x, y), r, sa, adjustedEndAngle, anticlockwise);
+}
+
+void CanvasPathMethods::ellipse(float x, float y, float rx, float ry, float rot, float sa, float ea, bool anticlockwise, ExceptionCode& ec)
+{
+    ec = 0;
+    if (!std::isfinite(x) || !std::isfinite(y) || !std::isfinite(rx) || !std::isfinite(ry) || !std::isfinite(rot) || !std::isfinite(sa) || !std::isfinite(ea))
+        return;
+
+    if (rx < 0 || ry < 0) {
+        ec = INDEX_SIZE_ERR;
         return;
     }
-    if (!anticlockwise && ea - sa >= 2 * piFloat) {
-        m_path.addArc(FloatPoint(x, y), r, sa, sa + 2 * piFloat, anticlockwise);
+
+    if (!rx || !ry || sa == ea) {
+        // The ellipse is empty but we still need to draw the connecting line to start point.
+        lineTo(x + rx * cosf(sa + rot), y + rx * sinf(sa + rot));
         return;
     }
 
-    m_path.addArc(FloatPoint(x, y), r, sa, ea, anticlockwise);
+    if (!isTransformInvertible())
+        return;
+
+    float adjustedEndAngle = adjustEndAngle(sa, ea, anticlockwise);
+    m_path.addEllipse(FloatPoint(x, y), rx, ry, rot, sa, adjustedEndAngle, anticlockwise);

[Stephen Chennney, 2013/04/26 14:51:03]

You should catch the common case of an entire ellipse, and use the optimized
path for that.

[dshwang, 2013/04/26 15:23:33]

Thank you for review, Stephen. Yes, your opinion is great.
However, without CanvasPathMethods::ellipse, adjustEndAngle() is just extracted
from CanvasPathMethods::arc(). I mean that CanvasPathMethods::arc() behavior is
not changed. And your opinion is possible to apply arc() also. How about
separating optimization patch for both arc() and ellipse()?

If you say ok, I'll file it.

[Stephen Chennney, 2013/04/26 16:02:04]

You can certainly change both ellipse and arc. I'm not certain that it will
behave correctly due to possible differences in Skia behavior, but I would
certainly like you to try. We are actively attempting to use Skia optimized
primitives whenever we can.

 }
 
+
+
 void CanvasPathMethods::rect(float x, float y, float width, float height)
 {
     if (!isTransformInvertible())