Document SkSurface::MakeRaster's memory initialization

include/core/SkSurface.h

Index: include/core/SkSurface.h
diff --git a/include/core/SkSurface.h b/include/core/SkSurface.h
index a092a81ca592745fc599804952f680194f29f437..e93413d3afea400c27fdab065513ac42de596345 100644
--- a/include/core/SkSurface.h
+++ b/include/core/SkSurface.h
@@ -49,10 +49,10 @@ public:
                                                  void* context, const SkSurfaceProps* = nullptr);
 
     /**
-     *  Return a new surface, with the memory for the pixels automatically allocated, but respecting
-     *  the specified rowBytes. If rowBytes==0, then a default value will be chosen. If a non-zero
-     *  rowBytes is specified, then any images snapped off of this surface (via newImageSnapshot())
-     *  are guaranteed to have the same rowBytes.
+     *  Return a new surface, with the memory for the pixels automatically allocated and
+     *  zero-initialized, but respecting the specified rowBytes. If rowBytes==0, then a default

[reed1, 2016/06/15 15:30:35]

Do we consistently make this promise for surfaces (gpu backed as well)? That if
Skia allocates the buffer, it will *always* be initialized? If so, then lgtm

[f(malita), 2016/06/15 15:36:37]

Even if we only do this for raster, it's still worth documenting, no?

(the CL only makes this promise for Make*Raster* currently)

[bsalomon, 2016/06/15 16:18:23]

I believe we do this for GPU as well.

+     *  value will be chosen. If a non-zero rowBytes is specified, then any images snapped off of
+     *  this surface (via makeImageSnapshot()) are guaranteed to have the same rowBytes.
      *
      *  If the requested surface cannot be created, or the request is not a
      *  supported configuration, NULL will be returned.