Drop hover:on-demand support & let touch-screens report "none" instead.

ui/base/touch/touch_device_android.cc

Index: ui/base/touch/touch_device_android.cc
diff --git a/ui/base/touch/touch_device_android.cc b/ui/base/touch/touch_device_android.cc
index b8a458e8300467d6cd711fc4745f9cb7cf19f578..e46de92442e5d68ef33c7498ba84d3c59291c3f2 100644
--- a/ui/base/touch/touch_device_android.cc
+++ b/ui/base/touch/touch_device_android.cc
@@ -45,8 +45,6 @@ PointerType GetPrimaryPointerType(int available_pointer_types) {
 }
 
 HoverType GetPrimaryHoverType(int available_hover_types) {
-  if (available_hover_types & HOVER_TYPE_ON_DEMAND)
-    return HOVER_TYPE_ON_DEMAND;

[Rick Byers, 2017/03/21 20:47:17]

There's a subtle change in semantics here.  Note that unlike on Windows, on
Android we were always considering touch "primary" over "mouse".  Previously
when both touch and mouse were present we'd have pointer:coarse and
hover:on-demand.  Now in that case it seems we'd have pointer:coarse and
hover:hover, which seems inconsistent (is the primary pointer a hovering coarse
pointer?).

Assuming we want to stick with always considering touch "primary" on android,
then I think you want this here:
if (available_hover_types & HOVER_TYPE_NONE)	
      return HOVER_TYPE_NONE;

Then the only way you ever get hover:hover on Android is when there's no
touchscreen at all and only a mouse.

[mustaq, 2017/03/22 14:31:29]

Yes, I noted this subtle change in #msg25 above but I had ruled it out as
uncommon. Your suggestion seems better. Done.

   if (available_hover_types & HOVER_TYPE_HOVER)
     return HOVER_TYPE_HOVER;
   DCHECK_EQ(available_hover_types, HOVER_TYPE_NONE);

[Rick Byers, 2017/03/21 20:47:17]

This isn't a new problem, but I believe this DCHECK isn't necessarily true. 
Looking at the Java code, an Android device with neither a mouse or touchscreen
will have available_hover_times==0 (which is not HOVER_TYPE_NONE).  Perhaps just
remove the DCHECK?  I think returning HOVER_TYPE_NONE in both the no device and
touch-only cases is what we want here anyway.

[mustaq, 2017/03/22 14:31:29]

Did you mean some Java code other than TouchDevice.java:89? I have changed the
DCHECK to match your comment above but this code shouldn't see any unset
hover-types I believe.

[Rick Byers, 2017/03/23 19:27:32]

Ah, sorry - I missed the code that was overwriting the '0' value with the NONE
one.  Seems a little weird, but OK that means I was wrong and the DHECK is
guaranteed to be true.  Thanks!