Add support for CheckedNumeric Max and Min constexpr functions

base/numerics/safe_conversions_impl.h

Index: base/numerics/safe_conversions_impl.h
diff --git a/base/numerics/safe_conversions_impl.h b/base/numerics/safe_conversions_impl.h
index c52b8f25c77f10bd4d9babd7560c329eed52d0bd..be9eaa16ba6008a58c6fa33e5a5ad83b37385c8c 100644
--- a/base/numerics/safe_conversions_impl.h
+++ b/base/numerics/safe_conversions_impl.h
@@ -335,11 +335,7 @@ enum ArithmeticPromotionCategory {
   RIGHT_PROMOTION  // Use the type of the right-hand argument.
 };
 
-template <ArithmeticPromotionCategory Promotion,
-          typename Lhs,
-          typename Rhs = Lhs>
-struct ArithmeticPromotion;
-
+// Determines the type that can represent the largest positive value.
 template <typename Lhs,
           typename Rhs,
           ArithmeticPromotionCategory Promotion =
@@ -358,6 +354,34 @@ struct MaxExponentPromotion<Lhs, Rhs, RIGHT_PROMOTION> {
   using type = Rhs;
 };
 
+// Determines the type that can represent the lowest arithmetic value.

[dcheng, 2016/12/03 21:34:29]

Nit: it kind of feels like we should be more consistent about terminology here
and just call this minimum instead of lowest.

[jschuh, 2016/12/03 22:29:12]

I tried to be consistent with std::numeric_limits<> (where the right thing is
max and lowest, not min). I'm not wedded to that approach, but I was going to
replace all the occurrences of -numeric_limits<FLOAT>::max() and
numeric_limits<INTEGER>::min() with numeric_limits<>::lowest(). So, this just
felt more consistent in that world.

+template <typename Lhs,
+          typename Rhs,
+          ArithmeticPromotionCategory Promotion =
+              std::is_signed<Lhs>::value
+                  ? (std::is_signed<Rhs>::value
+                         ? (MaxExponent<Lhs>::value > MaxExponent<Rhs>::value
+                                ? LEFT_PROMOTION
+                                : RIGHT_PROMOTION)
+                         : LEFT_PROMOTION)
+                  : (std::is_signed<Rhs>::value
+                         ? RIGHT_PROMOTION
+                         : (MaxExponent<Lhs>::value < MaxExponent<Rhs>::value
+                                ? LEFT_PROMOTION
+                                : RIGHT_PROMOTION))>
+struct LowestValuePromotion;
+
+template <typename Lhs, typename Rhs>
+struct LowestValuePromotion<Lhs, Rhs, LEFT_PROMOTION> {
+  using type = Lhs;
+};
+
+template <typename Lhs, typename Rhs>
+struct LowestValuePromotion<Lhs, Rhs, RIGHT_PROMOTION> {
+  using type = Rhs;
+};
+
+// Determines the type that is best able to represent an arithmetic result.
 template <typename Lhs,
           typename Rhs = Lhs,
           bool is_intmax_type =