Call EnableNonClientDpiScaling

ui/gfx/win/window_impl.cc

Index: ui/gfx/win/window_impl.cc
diff --git a/ui/gfx/win/window_impl.cc b/ui/gfx/win/window_impl.cc
index f7744687c97b396501d9cd4fe8ecb0258d637954..c1cefa80887e9d20ac28b050c571788fce902292 100644
--- a/ui/gfx/win/window_impl.cc
+++ b/ui/gfx/win/window_impl.cc
@@ -292,7 +292,6 @@ LRESULT CALLBACK WindowImpl::WndProc(HWND hwnd,
     window->got_create_ = true;
     if (hwnd)
       window->got_valid_hwnd_ = true;
-    return TRUE;
   }
 
   WindowImpl* window = reinterpret_cast<WindowImpl*>(GetWindowUserData(hwnd));

[sky, 2016/12/14 02:42:41]

This isn't necessary for the if case, maybe move declaration above if and move
this to else case?

[robliao, 2016/12/14 18:50:19]

While that is true, this seems more readable and GetWindowUserData is pretty
cheap (it's just grabbing a pointer off of the underlying HWND and only for
extra for WM_NCCREATE which is sent only once in an HWND's lifetime).

Happy to change it if you feel strongly about it, but that would mean we would
have to leave window uninitialized before the if-else check:

WindowImpl* window; // Uninitialized
if (message == WM_NCCREATE) {
  ...
  window = reinterpret_cast<>(cs->lpCreateParams)
  ...
} else {
  window = reinterpret_cast<>(GetWindowUserData(hwnd))
}

or check for WM_NCCREATE twice

const WindowImpl* window = (message == WM_NCCREATE) ? (Get WindowImpl from
l_param) : GetWindowUserData(...)
if (message == WM_NCCREATE) {
  SetWindowLongData(hwnd, window);
  ...
}

[sky, 2016/12/14 18:54:24]

I personally think it's clearer overall to not do the lookup twice.
How about setting it to null?

[robliao, 2016/12/14 19:03:18]

Done.

The main reason I don't like setting it to null is we never read (and shouldn't
read) from the value before it is set below.