Index: third_party/base/numerics/safe_math.h |
diff --git a/third_party/base/numerics/safe_math.h b/third_party/base/numerics/safe_math.h |
index 013af1eb60972fa7ebed4ba8cf004cdda5eb1c1a..a0c41a467b904fb742bfc7f0fe0a098112a3378f 100644 |
--- a/third_party/base/numerics/safe_math.h |
+++ b/third_party/base/numerics/safe_math.h |
@@ -2,140 +2,268 @@ |
// Use of this source code is governed by a BSD-style license that can be |
// found in the LICENSE file. |
-#ifndef PDFIUM_THIRD_PARTY_BASE_SAFE_MATH_H_ |
-#define PDFIUM_THIRD_PARTY_BASE_SAFE_MATH_H_ |
+#ifndef PDFIUM_THIRD_PARTY_BASE_NUMERICS_SAFE_MATH_H_ |
+#define PDFIUM_THIRD_PARTY_BASE_NUMERICS_SAFE_MATH_H_ |
-#include "safe_math_impl.h" |
+#include <stddef.h> |
+ |
+#include <limits> |
+#include <type_traits> |
+ |
+#include "third_party/base/numerics/safe_math_impl.h" |
namespace pdfium { |
namespace base { |
namespace internal { |
-// CheckedNumeric implements all the logic and operators for detecting integer |
+// CheckedNumeric<> implements all the logic and operators for detecting integer |
// boundary conditions such as overflow, underflow, and invalid conversions. |
// The CheckedNumeric type implicitly converts from floating point and integer |
// data types, and contains overloads for basic arithmetic operations (i.e.: +, |
-// -, *, /, %). |
+// -, *, / for all types and %, <<, >>, &, |, ^ for integers). Type promotions |
+// are a slightly modified version of the standard C arithmetic rules with the |
+// two differences being that there is no default promotion to int and bitwise |
+// logical operations always return an unsigned of the wider type. |
+// |
+// You may also use one of the variadic convenience functions, which accept |
+// standard arithmetic or CheckedNumeric types, perform arithmetic operations, |
+// and return a CheckedNumeric result. The supported functions are: |
+// CheckAdd() - Addition. |
+// CheckSub() - Subtraction. |
+// CheckMul() - Multiplication. |
+// CheckDiv() - Division. |
+// CheckMod() - Modulous (integer only). |
+// CheckLsh() - Left integer shift (integer only). |
+// CheckRsh() - Right integer shift (integer only). |
+// CheckAnd() - Bitwise AND (integer only with unsigned result). |
+// CheckOr() - Bitwise OR (integer only with unsigned result). |
+// CheckXor() - Bitwise XOR (integer only with unsigned result). |
+// CheckMax() - Maximum of supplied arguments. |
+// CheckMin() - Minimum of supplied arguments. |
+// |
+// The unary negation, increment, and decrement operators are supported, along |
+// with the following unary arithmetic methods, which return a new |
+// CheckedNumeric as a result of the operation: |
+// Abs() - Absolute value. |
+// UnsignedAbs() - Absolute value as an equal-width unsigned underlying type |
+// (valid for only integral types). |
+// Max() - Returns whichever is greater of the current instance or argument. |
+// The underlying return type is whichever has the greatest magnitude. |
+// Min() - Returns whichever is lowest of the current instance or argument. |
+// The underlying return type is whichever has can represent the lowest |
+// number in the smallest width (e.g. int8_t over unsigned, int over |
+// int8_t, and float over int). |
// |
// The following methods convert from CheckedNumeric to standard numeric values: |
-// IsValid() - Returns true if the underlying numeric value is valid (i.e. has |
-// has not wrapped and is not the result of an invalid conversion). |
-// ValueOrDie() - Returns the underlying value. If the state is not valid this |
-// call will crash on a CHECK. |
-// ValueOrDefault() - Returns the current value, or the supplied default if the |
-// state is not valid. |
-// ValueFloating() - Returns the underlying floating point value (valid only |
-// only for floating point CheckedNumeric types). |
+// AssignIfValid() - Assigns the underlying value to the supplied destination |
+// pointer if the value is currently valid and within the range |
+// supported by the destination type. Returns true on success. |
+// **************************************************************************** |
+// * WARNING: All of the following functions return a StrictNumeric, which * |
+// * is valid for comparison and assignment operations, but will trigger a * |
+// * compile failure on attempts to assign to a type of insufficient range. * |
+// **************************************************************************** |
+// IsValid() - Returns true if the underlying numeric value is valid (i.e. has |
+// has not wrapped and is not the result of an invalid conversion). |
+// ValueOrDie() - Returns the underlying value. If the state is not valid this |
+// call will crash on a CHECK. |
+// ValueOrDefault() - Returns the current value, or the supplied default if the |
+// state is not valid (will not trigger a CHECK). |
// |
-// Bitwise operations are explicitly not supported, because correct |
-// handling of some cases (e.g. sign manipulation) is ambiguous. Comparison |
-// operations are explicitly not supported because they could result in a crash |
-// on a CHECK condition. You should use patterns like the following for these |
-// operations: |
-// Bitwise operation: |
-// CheckedNumeric<int> checked_int = untrusted_input_value; |
-// int x = checked_int.ValueOrDefault(0) | kFlagValues; |
-// Comparison: |
-// CheckedNumeric<size_t> checked_size; |
-// CheckedNumeric<int> checked_size = untrusted_input_value; |
-// checked_size = checked_size + HEADER LENGTH; |
+// The following wrapper functions can be used to avoid the template |
+// disambiguator syntax when converting a destination type. |
+// IsValidForType<>() in place of: a.template IsValid<Dst>() |
+// ValueOrDieForType<>() in place of: a.template ValueOrDie() |
+// ValueOrDefaultForType<>() in place of: a.template ValueOrDefault(default) |
+// |
+// The following are general utility methods that are useful for converting |
+// between arithmetic types and CheckedNumeric types: |
+// CheckedNumeric::Cast<Dst>() - Instance method returning a CheckedNumeric |
+// derived from casting the current instance to a CheckedNumeric of |
+// the supplied destination type. |
+// MakeCheckedNum() - Creates a new CheckedNumeric from the underlying type of |
+// the supplied arithmetic, CheckedNumeric, or StrictNumeric type. |
+// |
+// Comparison operations are explicitly not supported because they could result |
+// in a crash on an unexpected CHECK condition. You should use patterns like the |
+// following for comparisons: |
+// CheckedNumeric<size_t> checked_size = untrusted_input_value; |
+// checked_size += HEADER LENGTH; |
// if (checked_size.IsValid() && checked_size.ValueOrDie() < buffer_size) |
// Do stuff... |
+ |
template <typename T> |
class CheckedNumeric { |
+ static_assert(std::is_arithmetic<T>::value, |
+ "CheckedNumeric<T>: T must be a numeric type."); |
+ |
public: |
- typedef T type; |
+ using type = T; |
- CheckedNumeric() {} |
+ constexpr CheckedNumeric() {} |
// Copy constructor. |
template <typename Src> |
- CheckedNumeric(const CheckedNumeric<Src>& rhs) |
- : state_(rhs.ValueUnsafe(), rhs.validity()) {} |
+ constexpr CheckedNumeric(const CheckedNumeric<Src>& rhs) |
+ : state_(rhs.state_.value(), rhs.IsValid()) {} |
template <typename Src> |
- CheckedNumeric(Src value, RangeConstraint validity) |
- : state_(value, validity) {} |
+ friend class CheckedNumeric; |
// This is not an explicit constructor because we implicitly upgrade regular |
// numerics to CheckedNumerics to make them easier to use. |
template <typename Src> |
- CheckedNumeric(Src value) |
+ constexpr CheckedNumeric(Src value) // NOLINT(runtime/explicit) |
: state_(value) { |
- COMPILE_ASSERT(std::numeric_limits<Src>::is_specialized, |
- argument_must_be_numeric); |
+ static_assert(std::is_arithmetic<Src>::value, "Argument must be numeric."); |
+ } |
+ |
+ // This is not an explicit constructor because we want a seamless conversion |
+ // from StrictNumeric types. |
+ template <typename Src> |
+ constexpr CheckedNumeric( |
+ StrictNumeric<Src> value) // NOLINT(runtime/explicit) |
+ : state_(static_cast<Src>(value)) {} |
+ |
+ // IsValid() - The public API to test if a CheckedNumeric is currently valid. |
+ // A range checked destination type can be supplied using the Dst template |
+ // parameter. |
+ template <typename Dst = T> |
+ constexpr bool IsValid() const { |
+ return state_.is_valid() && |
+ IsValueInRangeForNumericType<Dst>(state_.value()); |
} |
- // IsValid() is the public API to test if a CheckedNumeric is currently valid. |
- bool IsValid() const { return validity() == RANGE_VALID; } |
+ // AssignIfValid(Dst) - Assigns the underlying value if it is currently valid |
+ // and is within the range supported by the destination type. Returns true if |
+ // successful and false otherwise. |
+ template <typename Dst> |
+ constexpr bool AssignIfValid(Dst* result) const { |
+ return IsValid<Dst>() ? ((*result = static_cast<Dst>(state_.value())), true) |
+ : false; |
+ } |
- // ValueOrDie() The primary accessor for the underlying value. If the current |
- // state is not valid it will CHECK and crash. |
- T ValueOrDie() const { |
- CHECK(IsValid()); |
- return state_.value(); |
+ // ValueOrDie() - The primary accessor for the underlying value. If the |
+ // current state is not valid it will CHECK and crash. |
+ // A range checked destination type can be supplied using the Dst template |
+ // parameter, which will trigger a CHECK if the value is not in bounds for |
+ // the destination. |
+ // The CHECK behavior can be overridden by supplying a handler as a |
+ // template parameter, for test code, etc. However, the handler cannot access |
+ // the underlying value, and it is not available through other means. |
+ template <typename Dst = T, class CheckHandler = CheckOnFailure> |
+ constexpr StrictNumeric<Dst> ValueOrDie() const { |
+ return IsValid<Dst>() ? static_cast<Dst>(state_.value()) |
+ : CheckHandler::template HandleFailure<Dst>(); |
} |
- // ValueOrDefault(T default_value) A convenience method that returns the |
+ // ValueOrDefault(T default_value) - A convenience method that returns the |
// current value if the state is valid, and the supplied default_value for |
// any other state. |
- T ValueOrDefault(T default_value) const { |
- return IsValid() ? state_.value() : default_value; |
+ // A range checked destination type can be supplied using the Dst template |
+ // parameter. WARNING: This function may fail to compile or CHECK at runtime |
+ // if the supplied default_value is not within range of the destination type. |
+ template <typename Dst = T, typename Src> |
+ constexpr StrictNumeric<Dst> ValueOrDefault(const Src default_value) const { |
+ return IsValid<Dst>() ? static_cast<Dst>(state_.value()) |
+ : checked_cast<Dst>(default_value); |
} |
- // ValueFloating() - Since floating point values include their validity state, |
- // we provide an easy method for extracting them directly, without a risk of |
- // crashing on a CHECK. |
- T ValueFloating() const { |
- COMPILE_ASSERT(std::numeric_limits<T>::is_iec559, argument_must_be_float); |
- return CheckedNumeric<T>::cast(*this).ValueUnsafe(); |
+ // Returns a checked numeric of the specified type, cast from the current |
+ // CheckedNumeric. If the current state is invalid or the destination cannot |
+ // represent the result then the returned CheckedNumeric will be invalid. |
+ template <typename Dst> |
+ constexpr CheckedNumeric<typename UnderlyingType<Dst>::type> Cast() const { |
+ return *this; |
} |
- // validity() - DO NOT USE THIS IN EXTERNAL CODE - It is public right now for |
- // tests and to avoid a big matrix of friend operator overloads. But the |
- // values it returns are likely to change in the future. |
- // Returns: current validity state (i.e. valid, overflow, underflow, nan). |
- // TODO(jschuh): crbug.com/332611 Figure out and implement semantics for |
- // saturation/wrapping so we can expose this state consistently and implement |
- // saturated arithmetic. |
- RangeConstraint validity() const { return state_.validity(); } |
- |
- // ValueUnsafe() - DO NOT USE THIS IN EXTERNAL CODE - It is public right now |
- // for tests and to avoid a big matrix of friend operator overloads. But the |
- // values it returns are likely to change in the future. |
- // Returns: the raw numeric value, regardless of the current state. |
- // TODO(jschuh): crbug.com/332611 Figure out and implement semantics for |
- // saturation/wrapping so we can expose this state consistently and implement |
- // saturated arithmetic. |
- T ValueUnsafe() const { return state_.value(); } |
+ // This friend method is available solely for providing more detailed logging |
+ // in the the tests. Do not implement it in production code, because the |
+ // underlying values may change at any time. |
+ template <typename U> |
+ friend U GetNumericValueForTest(const CheckedNumeric<U>& src); |
// Prototypes for the supported arithmetic operator overloads. |
- template <typename Src> CheckedNumeric& operator+=(Src rhs); |
- template <typename Src> CheckedNumeric& operator-=(Src rhs); |
- template <typename Src> CheckedNumeric& operator*=(Src rhs); |
- template <typename Src> CheckedNumeric& operator/=(Src rhs); |
- template <typename Src> CheckedNumeric& operator%=(Src rhs); |
- |
- CheckedNumeric operator-() const { |
- RangeConstraint validity; |
- T value = CheckedNeg(state_.value(), &validity); |
- // Negation is always valid for floating point. |
- if (std::numeric_limits<T>::is_iec559) |
- return CheckedNumeric<T>(value); |
- |
- validity = GetRangeConstraint(state_.validity() | validity); |
- return CheckedNumeric<T>(value, validity); |
+ template <typename Src> |
+ CheckedNumeric& operator+=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator-=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator*=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator/=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator%=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator<<=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator>>=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator&=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator|=(const Src rhs); |
+ template <typename Src> |
+ CheckedNumeric& operator^=(const Src rhs); |
+ |
+ constexpr CheckedNumeric operator-() const { |
+ return CheckedNumeric<T>( |
+ NegateWrapper(state_.value()), |
+ IsValid() && |
+ (!std::is_signed<T>::value || std::is_floating_point<T>::value || |
+ NegateWrapper(state_.value()) != |
+ std::numeric_limits<T>::lowest())); |
+ } |
+ |
+ constexpr CheckedNumeric operator~() const { |
+ return CheckedNumeric<decltype(InvertWrapper(T()))>( |
+ InvertWrapper(state_.value()), IsValid()); |
} |
- CheckedNumeric Abs() const { |
- RangeConstraint validity; |
- T value = CheckedAbs(state_.value(), &validity); |
- // Absolute value is always valid for floating point. |
- if (std::numeric_limits<T>::is_iec559) |
- return CheckedNumeric<T>(value); |
+ constexpr CheckedNumeric Abs() const { |
+ return CheckedNumeric<T>( |
+ AbsWrapper(state_.value()), |
+ IsValid() && |
+ (!std::is_signed<T>::value || std::is_floating_point<T>::value || |
+ AbsWrapper(state_.value()) != std::numeric_limits<T>::lowest())); |
+ } |
+ |
+ template <typename U> |
+ constexpr CheckedNumeric<typename MathWrapper<CheckedMaxOp, T, U>::type> Max( |
+ const U rhs) const { |
+ using R = typename UnderlyingType<U>::type; |
+ using result_type = typename MathWrapper<CheckedMaxOp, T, U>::type; |
+ // TODO(jschuh): This can be converted to the MathOp version and remain |
+ // constexpr once we have C++14 support. |
+ return CheckedNumeric<result_type>( |
+ static_cast<result_type>( |
+ IsGreater<T, R>::Test(state_.value(), Wrapper<U>::value(rhs)) |
+ ? state_.value() |
+ : Wrapper<U>::value(rhs)), |
+ state_.is_valid() && Wrapper<U>::is_valid(rhs)); |
+ } |
- validity = GetRangeConstraint(state_.validity() | validity); |
- return CheckedNumeric<T>(value, validity); |
+ template <typename U> |
+ constexpr CheckedNumeric<typename MathWrapper<CheckedMinOp, T, U>::type> Min( |
+ const U rhs) const { |
+ using R = typename UnderlyingType<U>::type; |
+ using result_type = typename MathWrapper<CheckedMinOp, T, U>::type; |
+ // TODO(jschuh): This can be converted to the MathOp version and remain |
+ // constexpr once we have C++14 support. |
+ return CheckedNumeric<result_type>( |
+ static_cast<result_type>( |
+ IsLess<T, R>::Test(state_.value(), Wrapper<U>::value(rhs)) |
+ ? state_.value() |
+ : Wrapper<U>::value(rhs)), |
+ state_.is_valid() && Wrapper<U>::is_valid(rhs)); |
+ } |
+ |
+ // This function is available only for integral types. It returns an unsigned |
+ // integer of the same width as the source type, containing the absolute value |
+ // of the source, and properly handling signed min. |
+ constexpr CheckedNumeric<typename UnsignedOrFloatForSize<T>::type> |
+ UnsignedAbs() const { |
+ return CheckedNumeric<typename UnsignedOrFloatForSize<T>::type>( |
+ SafeUnsignedAbs(state_.value()), state_.is_valid()); |
} |
CheckedNumeric& operator++() { |
@@ -160,113 +288,223 @@ class CheckedNumeric { |
return value; |
} |
- // These static methods behave like a convenience cast operator targeting |
- // the desired CheckedNumeric type. As an optimization, a reference is |
- // returned when Src is the same type as T. |
+ // These perform the actual math operations on the CheckedNumerics. |
+ // Binary arithmetic operations. |
+ template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R> |
+ static CheckedNumeric MathOp(const L lhs, const R rhs) { |
+ using Math = typename MathWrapper<M, L, R>::math; |
+ T result = 0; |
+ bool is_valid = |
+ Wrapper<L>::is_valid(lhs) && Wrapper<R>::is_valid(rhs) && |
+ Math::Do(Wrapper<L>::value(lhs), Wrapper<R>::value(rhs), &result); |
+ return CheckedNumeric<T>(result, is_valid); |
+ }; |
+ |
+ // Assignment arithmetic operations. |
+ template <template <typename, typename, typename> class M, typename R> |
+ CheckedNumeric& MathOp(const R rhs) { |
+ using Math = typename MathWrapper<M, T, R>::math; |
+ T result = 0; // Using T as the destination saves a range check. |
+ bool is_valid = state_.is_valid() && Wrapper<R>::is_valid(rhs) && |
+ Math::Do(state_.value(), Wrapper<R>::value(rhs), &result); |
+ *this = CheckedNumeric<T>(result, is_valid); |
+ return *this; |
+ }; |
+ |
+ private: |
+ CheckedNumericState<T> state_; |
+ |
template <typename Src> |
- static CheckedNumeric<T> cast( |
- Src u, |
- typename std::enable_if<std::numeric_limits<Src>::is_specialized, |
- int>::type = 0) { |
- return u; |
- } |
+ constexpr CheckedNumeric(Src value, bool is_valid) |
+ : state_(value, is_valid) {} |
+ // These wrappers allow us to handle state the same way for both |
+ // CheckedNumeric and POD arithmetic types. |
template <typename Src> |
- static CheckedNumeric<T> cast( |
- const CheckedNumeric<Src>& u, |
- typename std::enable_if<!std::is_same<Src, T>::value, int>::type = 0) { |
- return u; |
- } |
+ struct Wrapper { |
+ static constexpr bool is_valid(Src) { return true; } |
+ static constexpr Src value(Src value) { return value; } |
+ }; |
- static const CheckedNumeric<T>& cast(const CheckedNumeric<T>& u) { return u; } |
+ template <typename Src> |
+ struct Wrapper<CheckedNumeric<Src>> { |
+ static constexpr bool is_valid(const CheckedNumeric<Src> v) { |
+ return v.IsValid(); |
+ } |
+ static constexpr Src value(const CheckedNumeric<Src> v) { |
+ return v.state_.value(); |
+ } |
+ }; |
- private: |
- CheckedNumericState<T> state_; |
+ template <typename Src> |
+ struct Wrapper<StrictNumeric<Src>> { |
+ static constexpr bool is_valid(const StrictNumeric<Src>) { return true; } |
+ static constexpr Src value(const StrictNumeric<Src> v) { |
+ return static_cast<Src>(v); |
+ } |
+ }; |
}; |
-// This is the boilerplate for the standard arithmetic operator overloads. A |
-// macro isn't the prettiest solution, but it beats rewriting these five times. |
-// Some details worth noting are: |
-// * We apply the standard arithmetic promotions. |
-// * We skip range checks for floating points. |
-// * We skip range checks for destination integers with sufficient range. |
-// TODO(jschuh): extract these out into templates. |
-#define BASE_NUMERIC_ARITHMETIC_OPERATORS(NAME, OP, COMPOUND_OP) \ |
- /* Binary arithmetic operator for CheckedNumerics of the same type. */ \ |
- template <typename T> \ |
- CheckedNumeric<typename ArithmeticPromotion<T>::type> operator OP( \ |
- const CheckedNumeric<T>& lhs, const CheckedNumeric<T>& rhs) { \ |
- typedef typename ArithmeticPromotion<T>::type Promotion; \ |
- /* Floating point always takes the fast path */ \ |
- if (std::numeric_limits<T>::is_iec559) \ |
- return CheckedNumeric<T>(lhs.ValueUnsafe() OP rhs.ValueUnsafe()); \ |
- if (IsIntegerArithmeticSafe<Promotion, T, T>::value) \ |
- return CheckedNumeric<Promotion>( \ |
- lhs.ValueUnsafe() OP rhs.ValueUnsafe(), \ |
- GetRangeConstraint(rhs.validity() | lhs.validity())); \ |
- RangeConstraint validity = RANGE_VALID; \ |
- T result = Checked##NAME(static_cast<Promotion>(lhs.ValueUnsafe()), \ |
- static_cast<Promotion>(rhs.ValueUnsafe()), \ |
- &validity); \ |
- return CheckedNumeric<Promotion>( \ |
- result, \ |
- GetRangeConstraint(validity | lhs.validity() | rhs.validity())); \ |
- } \ |
- /* Assignment arithmetic operator implementation from CheckedNumeric. */ \ |
- template <typename T> \ |
- template <typename Src> \ |
- CheckedNumeric<T>& CheckedNumeric<T>::operator COMPOUND_OP(Src rhs) { \ |
- *this = CheckedNumeric<T>::cast(*this) OP CheckedNumeric<Src>::cast(rhs); \ |
- return *this; \ |
- } \ |
- /* Binary arithmetic operator for CheckedNumeric of different type. */ \ |
- template <typename T, typename Src> \ |
- CheckedNumeric<typename ArithmeticPromotion<T, Src>::type> operator OP( \ |
- const CheckedNumeric<Src>& lhs, const CheckedNumeric<T>& rhs) { \ |
- typedef typename ArithmeticPromotion<T, Src>::type Promotion; \ |
- if (IsIntegerArithmeticSafe<Promotion, T, Src>::value) \ |
- return CheckedNumeric<Promotion>( \ |
- lhs.ValueUnsafe() OP rhs.ValueUnsafe(), \ |
- GetRangeConstraint(rhs.validity() | lhs.validity())); \ |
- return CheckedNumeric<Promotion>::cast(lhs) \ |
- OP CheckedNumeric<Promotion>::cast(rhs); \ |
- } \ |
- /* Binary arithmetic operator for left CheckedNumeric and right numeric. */ \ |
- template <typename T, typename Src> \ |
- CheckedNumeric<typename ArithmeticPromotion<T, Src>::type> operator OP( \ |
- const CheckedNumeric<T>& lhs, Src rhs) { \ |
- typedef typename ArithmeticPromotion<T, Src>::type Promotion; \ |
- if (IsIntegerArithmeticSafe<Promotion, T, Src>::value) \ |
- return CheckedNumeric<Promotion>(lhs.ValueUnsafe() OP rhs, \ |
- lhs.validity()); \ |
- return CheckedNumeric<Promotion>::cast(lhs) \ |
- OP CheckedNumeric<Promotion>::cast(rhs); \ |
- } \ |
- /* Binary arithmetic operator for right numeric and left CheckedNumeric. */ \ |
- template <typename T, typename Src> \ |
- CheckedNumeric<typename ArithmeticPromotion<T, Src>::type> operator OP( \ |
- Src lhs, const CheckedNumeric<T>& rhs) { \ |
- typedef typename ArithmeticPromotion<T, Src>::type Promotion; \ |
- if (IsIntegerArithmeticSafe<Promotion, T, Src>::value) \ |
- return CheckedNumeric<Promotion>(lhs OP rhs.ValueUnsafe(), \ |
- rhs.validity()); \ |
- return CheckedNumeric<Promotion>::cast(lhs) \ |
- OP CheckedNumeric<Promotion>::cast(rhs); \ |
- } |
+// Convenience functions to avoid the ugly template disambiguator syntax. |
+template <typename Dst, typename Src> |
+constexpr bool IsValidForType(const CheckedNumeric<Src> value) { |
+ return value.template IsValid<Dst>(); |
+} |
+ |
+template <typename Dst, typename Src> |
+constexpr StrictNumeric<Dst> ValueOrDieForType( |
+ const CheckedNumeric<Src> value) { |
+ return value.template ValueOrDie<Dst>(); |
+} |
+ |
+template <typename Dst, typename Src, typename Default> |
+constexpr StrictNumeric<Dst> ValueOrDefaultForType( |
+ const CheckedNumeric<Src> value, |
+ const Default default_value) { |
+ return value.template ValueOrDefault<Dst>(default_value); |
+} |
+ |
+// These variadic templates work out the return types. |
+// TODO(jschuh): Rip all this out once we have C++14 non-trailing auto support. |
+template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R, |
+ typename... Args> |
+struct ResultType; |
+ |
+template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R> |
+struct ResultType<M, L, R> { |
+ using type = typename MathWrapper<M, L, R>::type; |
+}; |
+ |
+template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R, |
+ typename... Args> |
+struct ResultType { |
+ using type = |
+ typename ResultType<M, typename ResultType<M, L, R>::type, Args...>::type; |
+}; |
+ |
+// Convience wrapper to return a new CheckedNumeric from the provided arithmetic |
+// or CheckedNumericType. |
+template <typename T> |
+constexpr CheckedNumeric<typename UnderlyingType<T>::type> MakeCheckedNum( |
+ const T value) { |
+ return value; |
+} |
+ |
+// These implement the variadic wrapper for the math operations. |
+template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R> |
+CheckedNumeric<typename MathWrapper<M, L, R>::type> ChkMathOp(const L lhs, |
+ const R rhs) { |
+ using Math = typename MathWrapper<M, L, R>::math; |
+ return CheckedNumeric<typename Math::result_type>::template MathOp<M>(lhs, |
+ rhs); |
+} |
+ |
+// General purpose wrapper template for arithmetic operations. |
+template <template <typename, typename, typename> class M, |
+ typename L, |
+ typename R, |
+ typename... Args> |
+CheckedNumeric<typename ResultType<M, L, R, Args...>::type> |
+ChkMathOp(const L lhs, const R rhs, const Args... args) { |
+ auto tmp = ChkMathOp<M>(lhs, rhs); |
+ return tmp.IsValid() ? ChkMathOp<M>(tmp, args...) |
+ : decltype(ChkMathOp<M>(tmp, args...))(tmp); |
+}; |
-BASE_NUMERIC_ARITHMETIC_OPERATORS(Add, +, += ) |
-BASE_NUMERIC_ARITHMETIC_OPERATORS(Sub, -, -= ) |
-BASE_NUMERIC_ARITHMETIC_OPERATORS(Mul, *, *= ) |
-BASE_NUMERIC_ARITHMETIC_OPERATORS(Div, /, /= ) |
-BASE_NUMERIC_ARITHMETIC_OPERATORS(Mod, %, %= ) |
+// The following macros are just boilerplate for the standard arithmetic |
+// operator overloads and variadic function templates. A macro isn't the nicest |
+// solution, but it beats rewriting these over and over again. |
+#define BASE_NUMERIC_ARITHMETIC_VARIADIC(NAME) \ |
+ template <typename L, typename R, typename... Args> \ |
+ CheckedNumeric<typename ResultType<Checked##NAME##Op, L, R, Args...>::type> \ |
+ Check##NAME(const L lhs, const R rhs, const Args... args) { \ |
+ return ChkMathOp<Checked##NAME##Op, L, R, Args...>(lhs, rhs, args...); \ |
+ } |
+#define BASE_NUMERIC_ARITHMETIC_OPERATORS(NAME, OP, COMPOUND_OP) \ |
+ /* Binary arithmetic operator for all CheckedNumeric operations. */ \ |
+ template <typename L, typename R, \ |
+ typename std::enable_if<IsCheckedOp<L, R>::value>::type* = \ |
+ nullptr> \ |
+ CheckedNumeric<typename MathWrapper<Checked##NAME##Op, L, R>::type> \ |
+ operator OP(const L lhs, const R rhs) { \ |
+ return decltype(lhs OP rhs)::template MathOp<Checked##NAME##Op>(lhs, rhs); \ |
+ } \ |
+ /* Assignment arithmetic operator implementation from CheckedNumeric. */ \ |
+ template <typename L> \ |
+ template <typename R> \ |
+ CheckedNumeric<L>& CheckedNumeric<L>::operator COMPOUND_OP(const R rhs) { \ |
+ return MathOp<Checked##NAME##Op>(rhs); \ |
+ } \ |
+ /* Variadic arithmetic functions that return CheckedNumeric. */ \ |
+ BASE_NUMERIC_ARITHMETIC_VARIADIC(NAME) |
+ |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Add, +, +=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Sub, -, -=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Mul, *, *=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Div, /, /=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Mod, %, %=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Lsh, <<, <<=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Rsh, >>, >>=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(And, &, &=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Or, |, |=) |
+BASE_NUMERIC_ARITHMETIC_OPERATORS(Xor, ^, ^=) |
+BASE_NUMERIC_ARITHMETIC_VARIADIC(Max) |
+BASE_NUMERIC_ARITHMETIC_VARIADIC(Min) |
+ |
+#undef BASE_NUMERIC_ARITHMETIC_VARIADIC |
#undef BASE_NUMERIC_ARITHMETIC_OPERATORS |
+// These are some extra StrictNumeric operators to support simple pointer |
+// arithmetic with our result types. Since wrapping on a pointer is always |
+// bad, we trigger the CHECK condition here. |
+template <typename L, typename R> |
+L* operator+(L* lhs, const StrictNumeric<R> rhs) { |
+ uintptr_t result = CheckAdd(reinterpret_cast<uintptr_t>(lhs), |
+ CheckMul(sizeof(L), static_cast<R>(rhs))) |
+ .template ValueOrDie<uintptr_t>(); |
+ return reinterpret_cast<L*>(result); |
+} |
+ |
+template <typename L, typename R> |
+L* operator-(L* lhs, const StrictNumeric<R> rhs) { |
+ uintptr_t result = CheckSub(reinterpret_cast<uintptr_t>(lhs), |
+ CheckMul(sizeof(L), static_cast<R>(rhs))) |
+ .template ValueOrDie<uintptr_t>(); |
+ return reinterpret_cast<L*>(result); |
+} |
+ |
} // namespace internal |
using internal::CheckedNumeric; |
+using internal::IsValidForType; |
+using internal::ValueOrDieForType; |
+using internal::ValueOrDefaultForType; |
+using internal::MakeCheckedNum; |
+using internal::CheckMax; |
+using internal::CheckMin; |
+using internal::CheckAdd; |
+using internal::CheckSub; |
+using internal::CheckMul; |
+using internal::CheckDiv; |
+using internal::CheckMod; |
+using internal::CheckLsh; |
+using internal::CheckRsh; |
+using internal::CheckAnd; |
+using internal::CheckOr; |
+using internal::CheckXor; |
} // namespace base |
} // namespace pdfium |
-#endif // PDFIUM_THIRD_PARTY_BASE_SAFE_MATH_H_ |
+#endif // PDFIUM_THIRD_PARTY_BASE_NUMERICS_SAFE_MATH_H_ |