Fix bubble client area clipping regression.

ui/views/view.cc

Index: ui/views/view.cc
diff --git a/ui/views/view.cc b/ui/views/view.cc
index beedbb1ff261813e9add59d03cbd91e8fe2efdeb..d733034153bb82110ef12b88c6d642c7f74c73d2 100644
--- a/ui/views/view.cc
+++ b/ui/views/view.cc
@@ -813,7 +813,7 @@ void View::Paint(const ui::PaintContext& parent_context) {
     } else {
       gfx::Path clip_path_in_parent = clip_path_;
       clip_path_in_parent.offset(GetMirroredX(), y());
-      clip_recorder.ClipPath(clip_path_in_parent);
+      clip_recorder.ClipPathWithAntiAliasing(clip_path_in_parent);

[Evan Stade, 2016/03/29 01:27:58]

I tried to avoid this before and instead inset by 0.5 for perf reasons, but as
the bug(s) point out, this doesn't work.

[msw, 2016/03/29 23:08:27]

Maybe add a comment about why the aa is needed here?

[Evan Stade, 2016/03/29 23:11:52]

Do you have a suggestion for a good comment? It seems to boil down to "the clip
path could have partially transparent pixels, so use AA" but that seems sort of
self evident (as in, that's the only reason you'd use AA to begin with).

[msw, 2016/03/29 23:19:13]

Shrug, I don't understand why it would have transparent pixels now, but not
before this change... shouldn't the round rect corners yield some mid-pixel
paths regardless of your old 0.5f extra inset? I dunno, I suppose it might be
self-evident to people more familiar with clip paths than myself.

     }
   }