Implement vertex buffer caching in the tessellated path renderer.

include/core/SkPath.h

Index: include/core/SkPath.h
diff --git a/include/core/SkPath.h b/include/core/SkPath.h
index dc50ae76c317b1ad487da5470ecb6f73c7698bf6..f3aa1eab5bec075df92ec21dd15d44b6d9aae07e 100644
--- a/include/core/SkPath.h
+++ b/include/core/SkPath.h
@@ -1001,11 +1001,14 @@ private:
 
     void setPt(int index, SkScalar x, SkScalar y);
 
+    SkPathRef* pathRef() const { return fPathRef; }
+
     friend class SkAutoPathBoundsUpdate;
     friend class SkAutoDisableOvalCheck;
     friend class SkAutoDisableDirectionCheck;
     friend class SkBench_AddPathTest; // perf test reversePathTo
     friend class PathTest_Private; // unit test reversePathTo
+    friend class TessellatingPathBatch; // for pathRef()

[bsalomon, 2015/08/03 18:08:26]

Why not just pass the listeners through the skpath to pathref? Or was this done
to avoid having listeners in the public API? Maybe the right place to add this
interface is src/core/SkPathPriv.h? That way we don't need to add a friend and
also don't need to worry about clients using the mechanism.

[Stephen White, 2015/08/03 18:54:41]

I didn't want to put it on the SkPath, partly for the reason you noted, and
partly because the listener really is associated with the SkPathRef, not the
SkPath: if you add it to an empty SkPath, it'll fail (since it'll be trying to
add it to the canonical empty SkPathRef). At least by requiring to go through
the SkPathRef, this is a little more clear. Not sure that's a strong argument,
though. I'll try putting it in SkPathPriv.h to see how that looks.

 };
 
 #endif