accessBitmap is no longer called/needed

skia/ext/platform_canvas_unittest.cc

Index: skia/ext/platform_canvas_unittest.cc
diff --git a/skia/ext/platform_canvas_unittest.cc b/skia/ext/platform_canvas_unittest.cc
index 8f30a5c2b071098f93e9a07298d602c2aeb31069..eb3f7ef2b7f2ec2e7a1fe5a8b68f96c0ea7ceee8 100644
--- a/skia/ext/platform_canvas_unittest.cc
+++ b/skia/ext/platform_canvas_unittest.cc
@@ -102,9 +102,11 @@ bool VerifyRoundedRect(const SkCanvas& canvas,
                        int y,
                        int w,
                        int h) {
-  SkBaseDevice* device = canvas.getTopDevice(true);
-  const SkBitmap& bitmap = device->accessBitmap(false);
-  SkAutoLockPixels lock(bitmap);
+  SkBitmap bitmap;
+  bitmap.allocN32Pixels(w, h);
+  if (!canvas.readPixels(&bitmap, x, y)) {

[f(malita), 2016/09/17 14:21:22]

nit: can readPixels ever fail for this unit test?  If not, maybe
ASSERT_TRUE(canvas.readPixels(...))?

[reed1, 2016/09/19 00:11:52]

This is a general pixel checker. I'm not sure that all callers ensure that read
should always work.

+    return false;
+  }
 
   // Check corner points first. They should be of canvas_color.
   if (!IsOfColor(bitmap, x, y, canvas_color)) return false;

[f(malita), 2016/09/17 14:21:22]

Previously we were reading the full backing store, but now we're only extracting
[x y w h].  I think the coords below need to be adjusted relative to the new
origin (x,y).  Might be easier to just read the full canvas, like before.

(kinda scary that it passes as-is; looks like there's only one client which uses
smallish insets so that may explain it)

[reed1, 2016/09/19 00:11:52]

Good point about needing the entire canvas read. Will fix.