fix the visual difference of raster vs gpu -- inverse_paths

src/core/SkCanvas.cpp

Index: src/core/SkCanvas.cpp
diff --git a/src/core/SkCanvas.cpp b/src/core/SkCanvas.cpp
index 93dc14bd1e5e6700c7b2bef784255dc86b4fff90..5a0821144af66047a6f17e4435526432077e0983 100644
--- a/src/core/SkCanvas.cpp
+++ b/src/core/SkCanvas.cpp
@@ -1746,7 +1746,9 @@ void SkCanvas::drawPath(const SkPath& path, const SkPaint& paint) {
             return;
         }
     }
-    if (path.isEmpty()) {
+
+    const SkRect& r = path.getBounds();
+    if (r.width() <= 0 && r.height() <= 0) {

[reed2, 2014/02/21 10:48:49]

Do we have unit-tests added to exercise this?
Does this catch a width > 0 but height < 0 filled path (e.g. horizontal line as
a path)?
Do we perform the same checks on drawRect/drawOval?

[yunchao, 2014/02/24 03:07:20]

Hi Mike,
Do you think we need to add a unit-tests like tests/GpuDrawPathTest.cpp to show
that inverse-filling an empty path will not crash? The previous tests didn't
test this in unit-tests too. If we want to test the images generated by drawPath
for inverse-filled empty path, gm/inversepaths have done this.

a width > 0 but height < 0 filled path is not included in this fix, I think.
because It is not an empty path. It is a hairline, right? Skia can draw it
correctly.

Rects and Ovals have no FillType if they are not added to a path, so they didn't
need to check this. If they are empty, just draw nothing. GM cases have tested
empty rects/ovals too.

         if (path.isInverseFillType()) {
             this->internalDrawPaint(paint);
         }