Decrement CreateDoubleValue count

components/policy/core/common/registry_dict_win.cc

Index: components/policy/core/common/registry_dict_win.cc
diff --git a/components/policy/core/common/registry_dict_win.cc b/components/policy/core/common/registry_dict_win.cc
index 8e03fbc4f722f136739d4a4220f0746103c42914..d2478f760a64f602b3fccd7641f0e0e50042afc3 100644
--- a/components/policy/core/common/registry_dict_win.cc
+++ b/components/policy/core/common/registry_dict_win.cc
@@ -89,11 +89,11 @@ scoped_ptr<base::Value> ConvertValue(const base::Value& value,
       double double_value = 0;
       if (value.GetAsInteger(&int_value)) {
         return scoped_ptr<base::Value>(
-            base::Value::CreateDoubleValue(int_value));
+            new base::FundamentalValue(int_value));

[battre, 2014/07/16 07:37:12]

Do you want to cast to double here? Otherwise you change the type of the result.
I did not check whether this has an impact anywhere downstream.

[bartfab (slow), 2014/07/16 09:07:55]

Yes, you definitely want to cast. Otherwise, the policy will have an incorrect
type and there will be a CHECK or DCHECK failure later on.

[battre, 2014/07/16 09:53:20]

Maybe that is an indication that going for an overloaded constructor is a bad
move.

[Evan Stade, 2014/07/16 22:15:25]

Equally, I think it's an indication that this code should not have been relying
on an implicit cast. Luckily, several tests caught this as well. Going forward,
I don't think the overloaded constructor is more likely to cause bugs (although
I will grant it is a potential source of regressions in the switch over).

       } else if (value.GetAsString(&string_value) &&
                  base::StringToDouble(string_value, &double_value)) {
         return scoped_ptr<base::Value>(
-            base::Value::CreateDoubleValue(double_value));
+            new base::FundamentalValue(double_value));
       }
       break;
     }