Gamma-correctness pushed into Skia, top-down.

include/core/SkCanvas.h

Index: include/core/SkCanvas.h
diff --git a/include/core/SkCanvas.h b/include/core/SkCanvas.h
index 91952c3050002d8d4ee35d59546786a8452fe223..f0781069a39ca464a0d96e22c26b280ffbcb93cc 100644
--- a/include/core/SkCanvas.h
+++ b/include/core/SkCanvas.h
@@ -122,6 +122,12 @@ public:
      */
     SkImageInfo imageInfo() const;
 
+    /**
+    *  Return SurfaceProps for this canvas. If the canvas is not backed by pixels
+    *  (cpu or gpu), then this will return null.

[reed1, 2016/04/04 21:12:29]

1. I think recording canvases have a device, so they will return true? Would it
be more correct to say the canvas has to be backed by a surface?

2. How do you feel about returning a copy of the props instead of a bare ptr?

e.g.

bool getProps(SurfaceProps*) const;

[Brian Osman, 2016/04/04 21:42:35]

My understanding (and Brian's claim) was that the recording canvases didn't have
a device. If they do, then the comment on imageInfo() is also wrong - that
function is also implemented to check for the device being null.
I'd be okay with that. It will make a bit more typing in some of the GMs that
I'm fixing up, but not enough to matter. This version of props() is already
strange, because every other class with an SkSurfaceProps accessor returns by
const-ref, so changing the name/signature more to call out the potential for
failure is probably good.

+    */
+    const SkSurfaceProps* props() const;
+
     ///////////////////////////////////////////////////////////////////////////
 
     /**