Don't set the STATE_SYSTEM_HOTTRACKED state in tests.

content/browser/accessibility/browser_accessibility_state_impl.h

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef CONTENT_BROWSER_ACCESSIBILITY_BROWSER_ACCESSIBILITY_STATE_IMPL_H_
 #define CONTENT_BROWSER_ACCESSIBILITY_BROWSER_ACCESSIBILITY_STATE_IMPL_H_
 
 #include <vector>
 
 #include "base/basictypes.h"
@@ skipping 42 matching lines @@
 
   // Adds the given accessibility mode to the current accessibility mode bitmap.
   void AddAccessibilityMode(AccessibilityMode mode);
 
   // Removes the given accessibility mode from the current accessibility mode
   // bitmap, managing the bits that are shared with other modes such that a
   // bit will only be turned off when all modes that depend on it have been
   // removed.
   void RemoveAccessibilityMode(AccessibilityMode mode);
 
+  // Accessibility objects can have the "hot tracked" state set when

[David Tseng, 2015/02/03 02:19:42]

I wonder if an easier fix to this is to just move the mouse to some known
position programmatically as part of a test's setup?

[dmazzoni, 2015/02/03 06:54:11]

I think that'd be annoying for anyone who runs content_browsertests locally.
Also I think it sets a bad precedent because tests run in parallel but there's
only one mouse. If I set it to 0,0 and another test set it to 9999, 9999 they
could mess with each other and cause more flakiness. I think it's better to not
have tests depend on the local machine.

The exception is interactive_ui_tests - those tests are allowed to do things
like move the mouse, but I don't see any reason to migrate these tests there -
they'd run much more slowly since those tests can't be parallelized.

Other options would include:
* Setting the flag as a static bool in BrowserAccessibility or
BrowserAccessibilityWin instead
* Ignoring STATE_SYSTEM_HOTTRACKED when checking states in test code, rather
than changing the code being tested - this has the advantage of not messing with
the real code, but it requires remembering to do that in every test.

+  // the mouse is hovering over them, but this makes tests flaky because
+  // the test behaves differently when the mouse happens to be over an
+  // element.  This is a global switch to not use the "hot tracked" state
+  // in a test.
+  void set_disable_hot_tracking_for_testing(bool disable_hot_tracking) {
+    disable_hot_tracking_ = disable_hot_tracking;
+  }
+  bool disable_hot_tracking_for_testing() const {
+    return disable_hot_tracking_;
+  }
+
  private:
   friend class base::RefCountedThreadSafe<BrowserAccessibilityStateImpl>;
   friend struct DefaultSingletonTraits<BrowserAccessibilityStateImpl>;
 
   // Resets accessibility_mode_ to the default value.
   void ResetAccessibilityModeValue();
 
   // Called a short while after startup to allow time for the accessibility
   // state to be determined. Updates histograms with the current state.
   void UpdateHistograms();
 
   // Leaky singleton, destructor generally won't be called.
   ~BrowserAccessibilityStateImpl() override;
 
   void UpdatePlatformSpecificHistograms();
 
   // Updates the accessibility mode of all web contents, including swapped out
   // ones. |add| specifies whether the mode should be added or removed.
   void AddOrRemoveFromAllWebContents(AccessibilityMode mode, bool add);
 
   AccessibilityMode accessibility_mode_;
 
   std::vector<base::Closure> histogram_callbacks_;
 
+  bool disable_hot_tracking_;
+
   DISALLOW_COPY_AND_ASSIGN(BrowserAccessibilityStateImpl);
 };
 
 }  // namespace content
 
 #endif  // CONTENT_BROWSER_ACCESSIBILITY_BROWSER_ACCESSIBILITY_STATE_IMPL_H_