Make the value lifecycle inline-friendly (#5840)

* Make the value lifecycle inline-friendly

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

* Compare standard containers without a thread-local depth count

operator== on objects and arrays recursed through the containers' own
operator== and counted the depth in a thread-local variable. Where reading a
thread-local costs a call (macOS), that added about 10% to comparing documents
with many small containers. For the standard containers, walk objects and
arrays directly, compare scalars and strings inline, and pass the depth as an
argument, falling back to compare_iteratively beyond the depth limit. Other
container types keep the previous path.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

* Address review: owns_memory() helper, leaf compare naming, astyle

- Add detail::owns_memory(value_t), a switch listing every enumerator so
  -Wswitch-enum flags it when a type is added; destroy() and the copy
  constructor use it instead of hand-written comparisons. Clang emits the
  same bitmask test as before.
- Drop the parentheses around the scalar operands in
  JSON_IMPLEMENT_SCALAR_CASES; they are only needed for the container
  operands (#1530), which the comment now says.
- Build equal_bounded_containers from detail::conjunction so astyle no
  longer pads the template brackets.
- Rename equal_scalar() to equal_leaf() and document why it also compares
  strings, unlike JSON_IMPLEMENT_SCALAR_FAST.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

---------

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

include/nlohmann/detail/value_t.hpp

@@ -117,6 +117,34 @@ inline bool operator<(const value_t lhs, const value_t rhs) noexcept
 }
 #endif
 
+/*!
+@brief whether a value of the given type owns heap memory
+
+Strings, binaries, objects, and arrays are stored as pointers to allocated
+values; all other types are stored directly in the union. Every enumerator is
+listed so that -Wswitch-enum flags this function when a new type is added.
+*/
+inline bool owns_memory(const value_t t) noexcept
+{
+    switch (t)
+    {
+        case value_t::object:
+        case value_t::array:
+        case value_t::string:
+        case value_t::binary:
+            return true;
+
+        case value_t::null:
+        case value_t::boolean:
+        case value_t::number_integer:
+        case value_t::number_unsigned:
+        case value_t::number_float:
+        case value_t::discarded:
+        default:
+            return false;
+    }
+}
+
 
 /*!
 @brief compare an integer with a floating point number without precision loss

include/nlohmann/json.hpp

@@ -585,26 +585,12 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             switch (t)
             {
                 case value_t::object:
-                {
-                    object = create<object_t>();
-                    break;
-                }
-
                 case value_t::array:
-                {
-                    array = create<array_t>();
-                    break;
-                }
-
                 case value_t::string:
-                {
-                    string = create<string_t>("");
-                    break;
-                }
-
                 case value_t::binary:
                 {
-                    binary = create<binary_t>();
+                    // allocating: kept out of line, so the scalar cases below inline
+                    init_owned(t);
                     break;
                 }
 
@@ -651,6 +637,34 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             }
         }
 
+        JSON_HEDLEY_NEVER_INLINE
+        void init_owned(value_t t)
+        {
+            switch (t)
+            {
+                case value_t::object:
+                    object = create<object_t>();
+                    break;
+                case value_t::array:
+                    array = create<array_t>();
+                    break;
+                case value_t::string:
+                    string = create<string_t>("");
+                    break;
+                case value_t::binary:
+                    binary = create<binary_t>();
+                    break;
+                case value_t::null:
+                case value_t::boolean:
+                case value_t::number_integer:
+                case value_t::number_unsigned:
+                case value_t::number_float:
+                case value_t::discarded:
+                default:
+                    break;
+            }
+        }
+
         /// constructor for strings
         json_value(const string_t& value) : string(create<string_t>(value)) {}
 
@@ -911,7 +925,18 @@ public:
             }
         }
 
+        // keep the check whether t owns memory inline so destroying a scalar
+        // is a compare and a branch, and leave the actual work to an
+        // out-of-line function
         void destroy(value_t t)
+        {
+            if (JSON_HEDLEY_UNLIKELY(detail::owns_memory(t)))
+            {
+                destroy_owned(t);
+            }
+        }
+
+        JSON_HEDLEY_NEVER_INLINE void destroy_owned(value_t t)
         {
             switch (t)
             {
@@ -2559,15 +2584,14 @@ public:
         // check of passed value is valid
         other.assert_invariant();
 
-        if (m_data.m_type == value_t::object || m_data.m_type == value_t::array)
+        if (JSON_HEDLEY_LIKELY(!detail::owns_memory(m_data.m_type)))
         {
-            // copying the container directly would call this constructor again
-            // for every element, once per nesting level
-            copy_structured(other);
+            // null, boolean, numbers, discarded: plain copy of the union
+            m_data.m_value = other.m_data.m_value;
         }
         else
         {
-            copy_leaf_value(other, *this);
+            copy_owning(other);
         }
 
         assert_invariant();
@@ -2799,6 +2823,22 @@ public:
     // value access //
     //////////////////
 
+    /// out-of-line part of the copy constructor for owning types
+    JSON_HEDLEY_NEVER_INLINE
+    void copy_owning(const basic_json& other)
+    {
+        if (m_data.m_type == value_t::object || m_data.m_type == value_t::array)
+        {
+            // copying the container directly would call this constructor again
+            // for every element, once per nesting level
+            copy_structured(other);
+        }
+        else
+        {
+            copy_leaf_value(other, *this);
+        }
+    }
+
     /// get a pointer to the value (object)
     object_t* get_impl_ptr(object_t* /*unused*/) noexcept
     {
@@ -5173,7 +5213,22 @@ public:
     /// @name lexicographical comparison operators
     /// @{
 
-    // note parentheses around operands are necessary; see
+    // the same-type cases for non-owning scalars, shared by JSON_IMPLEMENT_OPERATOR
+    // and JSON_IMPLEMENT_SCALAR_FAST
+#define JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                     \
+  case value_t::null:                                                                                  \
+        return (null_result);                                                                            \
+  case value_t::boolean:                                                                               \
+        return lhs.m_data.m_value.boolean op rhs.m_data.m_value.boolean;                                 \
+  case value_t::number_integer:                                                                        \
+        return lhs.m_data.m_value.number_integer op rhs.m_data.m_value.number_integer;                   \
+  case value_t::number_unsigned:                                                                       \
+        return lhs.m_data.m_value.number_unsigned op rhs.m_data.m_value.number_unsigned;                 \
+  case value_t::number_float:                                                                          \
+        return lhs.m_data.m_value.number_float op rhs.m_data.m_value.number_float;
+
+    // note parentheses around the container operands are necessary, because
+    // GCC otherwise parses `array <` as std::array; see
     // https://github.com/nlohmann/json/issues/1530
     // Mixed signed/unsigned integer comparisons check whether the signed value
     // is negative before casting. If it is, the comparison is performed with
@@ -5209,24 +5264,11 @@ public:
                 return (*lhs.m_data.m_value.object) op (*rhs.m_data.m_value.object);                                   \
             }                                                                                            \
             \
-            case value_t::null:                                                                          \
-                return (null_result);                                                                    \
-                \
+            JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                 \
+            \
             case value_t::string:                                                                        \
                 return (*lhs.m_data.m_value.string) op (*rhs.m_data.m_value.string);                                   \
                 \
-            case value_t::boolean:                                                                       \
-                return (lhs.m_data.m_value.boolean) op (rhs.m_data.m_value.boolean);                                   \
-                \
-            case value_t::number_integer:                                                                \
-                return (lhs.m_data.m_value.number_integer) op (rhs.m_data.m_value.number_integer);                     \
-                \
-            case value_t::number_unsigned:                                                               \
-                return (lhs.m_data.m_value.number_unsigned) op (rhs.m_data.m_value.number_unsigned);                   \
-                \
-            case value_t::number_float:                                                                  \
-                return (lhs.m_data.m_value.number_float) op (rhs.m_data.m_value.number_float);                         \
-                \
             case value_t::binary:                                                                        \
                 return (*lhs.m_data.m_value.binary) op (*rhs.m_data.m_value.binary);                                   \
                 \
@@ -5270,6 +5312,24 @@ public:
     \
     return (default_result);
 
+    // inline fast path for two values of the same non-owning scalar type; anything
+    // else falls through to the out-of-line general comparison
+#define JSON_IMPLEMENT_SCALAR_FAST(op, null_result)                                                      \
+    if (lhs.m_data.m_type == rhs.m_data.m_type)                                                          \
+    {                                                                                                    \
+        switch (lhs.m_data.m_type)                                                                       \
+        {                                                                                                \
+                JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                 \
+            case value_t::object:                                                                       \
+            case value_t::array:                                                                        \
+            case value_t::string:                                                                       \
+            case value_t::binary:                                                                       \
+            case value_t::discarded:                                                                    \
+            default:                                                                                     \
+                break;                                                                                   \
+        }                                                                                                \
+    }
+
   JSON_PRIVATE_UNLESS_TESTED:
     // returns true if:
     // - any operand is NaN and the other operand is of number type
@@ -5297,6 +5357,173 @@ public:
         return compares_unordered(*this, rhs, inverse);
     }
 
+    /// @brief whether objects and arrays are the standard containers, whose
+    ///        elements equal_bounded() may therefore compare itself
+    using standard_object_t = std::map<StringType, basic_json, default_object_comparator_t,
+          AllocatorType<std::pair<const StringType, basic_json>>>;
+    using standard_array_t = std::vector<basic_json, AllocatorType<basic_json>>;
+    using equal_bounded_containers = detail::conjunction<std::is_same<object_t, standard_object_t>, std::is_same<array_t, standard_array_t>>;
+
+    // out-of-line general comparisons; the operators below handle same-type
+    // scalars inline and fall through to these
+    JSON_HEDLEY_NEVER_INLINE
+    static bool equal_general(const_reference lhs, const_reference rhs) noexcept
+    {
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_PUSH
+        JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
+#endif
+        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
+                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_POP
+#endif
+    }
+
+    JSON_HEDLEY_NEVER_INLINE
+    static bool less_general(const_reference lhs, const_reference rhs) noexcept
+    {
+        // default_result is used if we cannot compare values. In that case,
+        // we compare types. Note we have to call the operator explicitly,
+        // because MSVC has problems otherwise.
+        JSON_IMPLEMENT_OPERATOR( <, false, false, operator<(lhs_type, rhs_type),
+                                 compare_iteratively<true>(lhs, rhs, true) == compare_result::less, false)
+    }
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+    JSON_HEDLEY_NEVER_INLINE
+    static std::partial_ordering three_way_general(const_reference lhs, const_reference rhs) noexcept // *NOPAD*
+    {
+        // default_result is used if we cannot compare values. In that case,
+        // we compare types.
+        JSON_IMPLEMENT_OPERATOR(<=>, // *NOPAD*
+                                std::partial_ordering::equivalent,
+                                std::partial_ordering::unordered,
+                                lhs_type <=> rhs_type, // *NOPAD*
+                                to_partial_ordering(compare_iteratively<true>(lhs, rhs, false)), true)
+    }
+#endif
+
+    /// @brief compare two leaves of the same type inside equal_structured()
+    /// @return 1 if they are equal, 0 if they are not, and -1 if the values are
+    ///         of different types or are not handled here, which this leaves to
+    ///         equal_structured()
+    /// @note Unlike JSON_IMPLEMENT_SCALAR_FAST, this also compares strings:
+    ///       they are the most common leaves in a document, so comparing them
+    ///       inline pays off in the container walk, while the operators keep
+    ///       their inlined code at each call site small. Binary values are
+    ///       rare and left to equal_general().
+    JSON_HEDLEY_ALWAYS_INLINE
+    static int equal_leaf(const_reference lhs, const_reference rhs) noexcept
+    {
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_PUSH
+        JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
+#endif
+        if (lhs.m_data.m_type == rhs.m_data.m_type)
+        {
+            switch (lhs.m_data.m_type)
+            {
+                case value_t::null:
+                    return 1;
+                case value_t::boolean:
+                    return lhs.m_data.m_value.boolean == rhs.m_data.m_value.boolean;
+                case value_t::number_integer:
+                    return lhs.m_data.m_value.number_integer == rhs.m_data.m_value.number_integer;
+                case value_t::number_unsigned:
+                    return lhs.m_data.m_value.number_unsigned == rhs.m_data.m_value.number_unsigned;
+                case value_t::number_float:
+                    return lhs.m_data.m_value.number_float == rhs.m_data.m_value.number_float;
+                case value_t::string:
+                    return *lhs.m_data.m_value.string == *rhs.m_data.m_value.string;
+                case value_t::object:
+                case value_t::array:
+                case value_t::binary:
+                case value_t::discarded:
+                default:
+                    break;
+            }
+        }
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_POP
+#endif
+        return -1;
+    }
+
+    /// @brief compare for equality, walking objects and arrays here instead of
+    ///        in their containers so that the depth is a parameter and needs no
+    ///        thread-local count; only used for the standard containers
+    static bool equal_bounded(const_reference lhs, const_reference rhs, std::size_t depth) noexcept
+    {
+        const int leaf = equal_leaf(lhs, rhs);
+        return leaf < 0 ? equal_structured(lhs, rhs, depth) : leaf != 0;
+    }
+
+    static bool equal_structured(const_reference lhs, const_reference rhs, std::size_t depth) noexcept
+    {
+        const auto type = lhs.m_data.m_type;
+        if (type != rhs.m_data.m_type || (type != value_t::array && type != value_t::object))
+        {
+            return equal_general(lhs, rhs);
+        }
+
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return compare_iteratively<false>(lhs, rhs, false) == compare_result::equal;
+        }
+
+        if (type == value_t::array)
+        {
+            const auto& a = *lhs.m_data.m_value.array;
+            const auto& b = *rhs.m_data.m_value.array;
+            if (a.size() != b.size())
+            {
+                return false;
+            }
+            auto it_b = b.cbegin();
+            for (auto it_a = a.cbegin(); it_a != a.cend(); ++it_a, ++it_b)
+            {
+                const int leaf = equal_leaf(*it_a, *it_b);
+                if (leaf < 0 ? !equal_structured(*it_a, *it_b, depth + 1) : leaf == 0)
+                {
+                    return false;
+                }
+            }
+            return true;
+        }
+
+        const auto& a = *lhs.m_data.m_value.object;
+        const auto& b = *rhs.m_data.m_value.object;
+        if (a.size() != b.size())
+        {
+            return false;
+        }
+        auto it_b = b.cbegin();
+        for (auto it_a = a.cbegin(); it_a != a.cend(); ++it_a, ++it_b)
+        {
+            if (!(it_a->first == it_b->first))
+            {
+                return false;
+            }
+            const int leaf = equal_leaf(it_a->second, it_b->second);
+            if (leaf < 0 ? !equal_structured(it_a->second, it_b->second, depth + 1) : leaf == 0)
+            {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    static bool equal_dispatch(const_reference lhs, const_reference rhs, std::true_type /*standard containers*/) noexcept
+    {
+        return equal_bounded(lhs, rhs, 0);
+    }
+
+    static bool equal_dispatch(const_reference lhs, const_reference rhs, std::false_type /*other containers*/) noexcept
+    {
+        return equal_general(lhs, rhs);
+    }
+
   public:
 #if JSON_HAS_THREE_WAY_COMPARISON
     /// @brief comparison: equal
@@ -5308,11 +5535,11 @@ public:
         JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
 #endif
         const_reference lhs = *this;
-        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
-                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+        JSON_IMPLEMENT_SCALAR_FAST( ==, true)
 #ifdef __GNUC__
         JSON_HEDLEY_DIAGNOSTIC_POP
 #endif
+        return equal_dispatch(lhs, rhs, equal_bounded_containers {});
     }
 
     /// @brief comparison: equal
@@ -5329,13 +5556,8 @@ public:
     std::partial_ordering operator<=>(const_reference rhs) const noexcept // *NOPAD*
     {
         const_reference lhs = *this;
-        // default_result is used if we cannot compare values. In that case,
-        // we compare types.
-        JSON_IMPLEMENT_OPERATOR(<=>, // *NOPAD*
-                                std::partial_ordering::equivalent,
-                                std::partial_ordering::unordered,
-                                lhs_type <=> rhs_type, // *NOPAD*
-                                to_partial_ordering(compare_iteratively<true>(lhs, rhs, false)), true)
+        JSON_IMPLEMENT_SCALAR_FAST(<=>, std::partial_ordering::equivalent) // *NOPAD*
+        return three_way_general(lhs, rhs);
     }
 
     /// @brief comparison: 3-way
@@ -5423,11 +5645,11 @@ public:
         JSON_HEDLEY_DIAGNOSTIC_PUSH
         JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
 #endif
-        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
-                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+        JSON_IMPLEMENT_SCALAR_FAST( ==, true)
 #ifdef __GNUC__
         JSON_HEDLEY_DIAGNOSTIC_POP
 #endif
+        return equal_dispatch(lhs, rhs, equal_bounded_containers {});
     }
 
     // The friend comparisons with a scalar name the JSON type via decltype of
@@ -5482,11 +5704,8 @@ public:
     /// @sa https://json.nlohmann.me/api/basic_json/operator_lt/
     friend bool operator<(const_reference lhs, const_reference rhs) noexcept
     {
-        // default_result is used if we cannot compare values. In that case,
-        // we compare types. Note we have to call the operator explicitly,
-        // because MSVC has problems otherwise.
-        JSON_IMPLEMENT_OPERATOR( <, false, false, operator<(lhs_type, rhs_type),
-                                 compare_iteratively<true>(lhs, rhs, true) == compare_result::less, false)
+        JSON_IMPLEMENT_SCALAR_FAST( <, false)
+        return less_general(lhs, rhs);
     }
 
     /// @brief comparison: less than
@@ -5597,6 +5816,8 @@ public:
 #endif
 
 #undef JSON_IMPLEMENT_OPERATOR
+#undef JSON_IMPLEMENT_SCALAR_FAST
+#undef JSON_IMPLEMENT_SCALAR_CASES
 
     /// @}
 

single_include/nlohmann/json.hpp

@@ -3338,6 +3338,34 @@ inline bool operator<(const value_t lhs, const value_t rhs) noexcept
 }
 #endif
 
+/*!
+@brief whether a value of the given type owns heap memory
+
+Strings, binaries, objects, and arrays are stored as pointers to allocated
+values; all other types are stored directly in the union. Every enumerator is
+listed so that -Wswitch-enum flags this function when a new type is added.
+*/
+inline bool owns_memory(const value_t t) noexcept
+{
+    switch (t)
+    {
+        case value_t::object:
+        case value_t::array:
+        case value_t::string:
+        case value_t::binary:
+            return true;
+
+        case value_t::null:
+        case value_t::boolean:
+        case value_t::number_integer:
+        case value_t::number_unsigned:
+        case value_t::number_float:
+        case value_t::discarded:
+        default:
+            return false;
+    }
+}
+
 
 /*!
 @brief compare an integer with a floating point number without precision loss
@@ -30366,26 +30394,12 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             switch (t)
             {
                 case value_t::object:
-                {
-                    object = create<object_t>();
-                    break;
-                }
-
                 case value_t::array:
-                {
-                    array = create<array_t>();
-                    break;
-                }
-
                 case value_t::string:
-                {
-                    string = create<string_t>("");
-                    break;
-                }
-
                 case value_t::binary:
                 {
-                    binary = create<binary_t>();
+                    // allocating: kept out of line, so the scalar cases below inline
+                    init_owned(t);
                     break;
                 }
 
@@ -30432,6 +30446,34 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             }
         }
 
+        JSON_HEDLEY_NEVER_INLINE
+        void init_owned(value_t t)
+        {
+            switch (t)
+            {
+                case value_t::object:
+                    object = create<object_t>();
+                    break;
+                case value_t::array:
+                    array = create<array_t>();
+                    break;
+                case value_t::string:
+                    string = create<string_t>("");
+                    break;
+                case value_t::binary:
+                    binary = create<binary_t>();
+                    break;
+                case value_t::null:
+                case value_t::boolean:
+                case value_t::number_integer:
+                case value_t::number_unsigned:
+                case value_t::number_float:
+                case value_t::discarded:
+                default:
+                    break;
+            }
+        }
+
         /// constructor for strings
         json_value(const string_t& value) : string(create<string_t>(value)) {}
 
@@ -30692,7 +30734,18 @@ public:
             }
         }
 
+        // keep the check whether t owns memory inline so destroying a scalar
+        // is a compare and a branch, and leave the actual work to an
+        // out-of-line function
         void destroy(value_t t)
+        {
+            if (JSON_HEDLEY_UNLIKELY(detail::owns_memory(t)))
+            {
+                destroy_owned(t);
+            }
+        }
+
+        JSON_HEDLEY_NEVER_INLINE void destroy_owned(value_t t)
         {
             switch (t)
             {
@@ -32340,15 +32393,14 @@ public:
         // check of passed value is valid
         other.assert_invariant();
 
-        if (m_data.m_type == value_t::object || m_data.m_type == value_t::array)
+        if (JSON_HEDLEY_LIKELY(!detail::owns_memory(m_data.m_type)))
         {
-            // copying the container directly would call this constructor again
-            // for every element, once per nesting level
-            copy_structured(other);
+            // null, boolean, numbers, discarded: plain copy of the union
+            m_data.m_value = other.m_data.m_value;
         }
         else
         {
-            copy_leaf_value(other, *this);
+            copy_owning(other);
         }
 
         assert_invariant();
@@ -32580,6 +32632,22 @@ public:
     // value access //
     //////////////////
 
+    /// out-of-line part of the copy constructor for owning types
+    JSON_HEDLEY_NEVER_INLINE
+    void copy_owning(const basic_json& other)
+    {
+        if (m_data.m_type == value_t::object || m_data.m_type == value_t::array)
+        {
+            // copying the container directly would call this constructor again
+            // for every element, once per nesting level
+            copy_structured(other);
+        }
+        else
+        {
+            copy_leaf_value(other, *this);
+        }
+    }
+
     /// get a pointer to the value (object)
     object_t* get_impl_ptr(object_t* /*unused*/) noexcept
     {
@@ -34954,7 +35022,22 @@ public:
     /// @name lexicographical comparison operators
     /// @{
 
-    // note parentheses around operands are necessary; see
+    // the same-type cases for non-owning scalars, shared by JSON_IMPLEMENT_OPERATOR
+    // and JSON_IMPLEMENT_SCALAR_FAST
+#define JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                     \
+  case value_t::null:                                                                                  \
+        return (null_result);                                                                            \
+  case value_t::boolean:                                                                               \
+        return lhs.m_data.m_value.boolean op rhs.m_data.m_value.boolean;                                 \
+  case value_t::number_integer:                                                                        \
+        return lhs.m_data.m_value.number_integer op rhs.m_data.m_value.number_integer;                   \
+  case value_t::number_unsigned:                                                                       \
+        return lhs.m_data.m_value.number_unsigned op rhs.m_data.m_value.number_unsigned;                 \
+  case value_t::number_float:                                                                          \
+        return lhs.m_data.m_value.number_float op rhs.m_data.m_value.number_float;
+
+    // note parentheses around the container operands are necessary, because
+    // GCC otherwise parses `array <` as std::array; see
     // https://github.com/nlohmann/json/issues/1530
     // Mixed signed/unsigned integer comparisons check whether the signed value
     // is negative before casting. If it is, the comparison is performed with
@@ -34990,24 +35073,11 @@ public:
                 return (*lhs.m_data.m_value.object) op (*rhs.m_data.m_value.object);                                   \
             }                                                                                            \
             \
-            case value_t::null:                                                                          \
-                return (null_result);                                                                    \
-                \
+            JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                 \
+            \
             case value_t::string:                                                                        \
                 return (*lhs.m_data.m_value.string) op (*rhs.m_data.m_value.string);                                   \
                 \
-            case value_t::boolean:                                                                       \
-                return (lhs.m_data.m_value.boolean) op (rhs.m_data.m_value.boolean);                                   \
-                \
-            case value_t::number_integer:                                                                \
-                return (lhs.m_data.m_value.number_integer) op (rhs.m_data.m_value.number_integer);                     \
-                \
-            case value_t::number_unsigned:                                                               \
-                return (lhs.m_data.m_value.number_unsigned) op (rhs.m_data.m_value.number_unsigned);                   \
-                \
-            case value_t::number_float:                                                                  \
-                return (lhs.m_data.m_value.number_float) op (rhs.m_data.m_value.number_float);                         \
-                \
             case value_t::binary:                                                                        \
                 return (*lhs.m_data.m_value.binary) op (*rhs.m_data.m_value.binary);                                   \
                 \
@@ -35051,6 +35121,24 @@ public:
     \
     return (default_result);
 
+    // inline fast path for two values of the same non-owning scalar type; anything
+    // else falls through to the out-of-line general comparison
+#define JSON_IMPLEMENT_SCALAR_FAST(op, null_result)                                                      \
+    if (lhs.m_data.m_type == rhs.m_data.m_type)                                                          \
+    {                                                                                                    \
+        switch (lhs.m_data.m_type)                                                                       \
+        {                                                                                                \
+                JSON_IMPLEMENT_SCALAR_CASES(op, null_result)                                                 \
+            case value_t::object:                                                                       \
+            case value_t::array:                                                                        \
+            case value_t::string:                                                                       \
+            case value_t::binary:                                                                       \
+            case value_t::discarded:                                                                    \
+            default:                                                                                     \
+                break;                                                                                   \
+        }                                                                                                \
+    }
+
   JSON_PRIVATE_UNLESS_TESTED:
     // returns true if:
     // - any operand is NaN and the other operand is of number type
@@ -35078,6 +35166,173 @@ public:
         return compares_unordered(*this, rhs, inverse);
     }
 
+    /// @brief whether objects and arrays are the standard containers, whose
+    ///        elements equal_bounded() may therefore compare itself
+    using standard_object_t = std::map<StringType, basic_json, default_object_comparator_t,
+          AllocatorType<std::pair<const StringType, basic_json>>>;
+    using standard_array_t = std::vector<basic_json, AllocatorType<basic_json>>;
+    using equal_bounded_containers = detail::conjunction<std::is_same<object_t, standard_object_t>, std::is_same<array_t, standard_array_t>>;
+
+    // out-of-line general comparisons; the operators below handle same-type
+    // scalars inline and fall through to these
+    JSON_HEDLEY_NEVER_INLINE
+    static bool equal_general(const_reference lhs, const_reference rhs) noexcept
+    {
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_PUSH
+        JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
+#endif
+        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
+                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_POP
+#endif
+    }
+
+    JSON_HEDLEY_NEVER_INLINE
+    static bool less_general(const_reference lhs, const_reference rhs) noexcept
+    {
+        // default_result is used if we cannot compare values. In that case,
+        // we compare types. Note we have to call the operator explicitly,
+        // because MSVC has problems otherwise.
+        JSON_IMPLEMENT_OPERATOR( <, false, false, operator<(lhs_type, rhs_type),
+                                 compare_iteratively<true>(lhs, rhs, true) == compare_result::less, false)
+    }
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+    JSON_HEDLEY_NEVER_INLINE
+    static std::partial_ordering three_way_general(const_reference lhs, const_reference rhs) noexcept // *NOPAD*
+    {
+        // default_result is used if we cannot compare values. In that case,
+        // we compare types.
+        JSON_IMPLEMENT_OPERATOR(<=>, // *NOPAD*
+                                std::partial_ordering::equivalent,
+                                std::partial_ordering::unordered,
+                                lhs_type <=> rhs_type, // *NOPAD*
+                                to_partial_ordering(compare_iteratively<true>(lhs, rhs, false)), true)
+    }
+#endif
+
+    /// @brief compare two leaves of the same type inside equal_structured()
+    /// @return 1 if they are equal, 0 if they are not, and -1 if the values are
+    ///         of different types or are not handled here, which this leaves to
+    ///         equal_structured()
+    /// @note Unlike JSON_IMPLEMENT_SCALAR_FAST, this also compares strings:
+    ///       they are the most common leaves in a document, so comparing them
+    ///       inline pays off in the container walk, while the operators keep
+    ///       their inlined code at each call site small. Binary values are
+    ///       rare and left to equal_general().
+    JSON_HEDLEY_ALWAYS_INLINE
+    static int equal_leaf(const_reference lhs, const_reference rhs) noexcept
+    {
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_PUSH
+        JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
+#endif
+        if (lhs.m_data.m_type == rhs.m_data.m_type)
+        {
+            switch (lhs.m_data.m_type)
+            {
+                case value_t::null:
+                    return 1;
+                case value_t::boolean:
+                    return lhs.m_data.m_value.boolean == rhs.m_data.m_value.boolean;
+                case value_t::number_integer:
+                    return lhs.m_data.m_value.number_integer == rhs.m_data.m_value.number_integer;
+                case value_t::number_unsigned:
+                    return lhs.m_data.m_value.number_unsigned == rhs.m_data.m_value.number_unsigned;
+                case value_t::number_float:
+                    return lhs.m_data.m_value.number_float == rhs.m_data.m_value.number_float;
+                case value_t::string:
+                    return *lhs.m_data.m_value.string == *rhs.m_data.m_value.string;
+                case value_t::object:
+                case value_t::array:
+                case value_t::binary:
+                case value_t::discarded:
+                default:
+                    break;
+            }
+        }
+#ifdef __GNUC__
+        JSON_HEDLEY_DIAGNOSTIC_POP
+#endif
+        return -1;
+    }
+
+    /// @brief compare for equality, walking objects and arrays here instead of
+    ///        in their containers so that the depth is a parameter and needs no
+    ///        thread-local count; only used for the standard containers
+    static bool equal_bounded(const_reference lhs, const_reference rhs, std::size_t depth) noexcept
+    {
+        const int leaf = equal_leaf(lhs, rhs);
+        return leaf < 0 ? equal_structured(lhs, rhs, depth) : leaf != 0;
+    }
+
+    static bool equal_structured(const_reference lhs, const_reference rhs, std::size_t depth) noexcept
+    {
+        const auto type = lhs.m_data.m_type;
+        if (type != rhs.m_data.m_type || (type != value_t::array && type != value_t::object))
+        {
+            return equal_general(lhs, rhs);
+        }
+
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return compare_iteratively<false>(lhs, rhs, false) == compare_result::equal;
+        }
+
+        if (type == value_t::array)
+        {
+            const auto& a = *lhs.m_data.m_value.array;
+            const auto& b = *rhs.m_data.m_value.array;
+            if (a.size() != b.size())
+            {
+                return false;
+            }
+            auto it_b = b.cbegin();
+            for (auto it_a = a.cbegin(); it_a != a.cend(); ++it_a, ++it_b)
+            {
+                const int leaf = equal_leaf(*it_a, *it_b);
+                if (leaf < 0 ? !equal_structured(*it_a, *it_b, depth + 1) : leaf == 0)
+                {
+                    return false;
+                }
+            }
+            return true;
+        }
+
+        const auto& a = *lhs.m_data.m_value.object;
+        const auto& b = *rhs.m_data.m_value.object;
+        if (a.size() != b.size())
+        {
+            return false;
+        }
+        auto it_b = b.cbegin();
+        for (auto it_a = a.cbegin(); it_a != a.cend(); ++it_a, ++it_b)
+        {
+            if (!(it_a->first == it_b->first))
+            {
+                return false;
+            }
+            const int leaf = equal_leaf(it_a->second, it_b->second);
+            if (leaf < 0 ? !equal_structured(it_a->second, it_b->second, depth + 1) : leaf == 0)
+            {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    static bool equal_dispatch(const_reference lhs, const_reference rhs, std::true_type /*standard containers*/) noexcept
+    {
+        return equal_bounded(lhs, rhs, 0);
+    }
+
+    static bool equal_dispatch(const_reference lhs, const_reference rhs, std::false_type /*other containers*/) noexcept
+    {
+        return equal_general(lhs, rhs);
+    }
+
   public:
 #if JSON_HAS_THREE_WAY_COMPARISON
     /// @brief comparison: equal
@@ -35089,11 +35344,11 @@ public:
         JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
 #endif
         const_reference lhs = *this;
-        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
-                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+        JSON_IMPLEMENT_SCALAR_FAST( ==, true)
 #ifdef __GNUC__
         JSON_HEDLEY_DIAGNOSTIC_POP
 #endif
+        return equal_dispatch(lhs, rhs, equal_bounded_containers {});
     }
 
     /// @brief comparison: equal
@@ -35110,13 +35365,8 @@ public:
     std::partial_ordering operator<=>(const_reference rhs) const noexcept // *NOPAD*
     {
         const_reference lhs = *this;
-        // default_result is used if we cannot compare values. In that case,
-        // we compare types.
-        JSON_IMPLEMENT_OPERATOR(<=>, // *NOPAD*
-                                std::partial_ordering::equivalent,
-                                std::partial_ordering::unordered,
-                                lhs_type <=> rhs_type, // *NOPAD*
-                                to_partial_ordering(compare_iteratively<true>(lhs, rhs, false)), true)
+        JSON_IMPLEMENT_SCALAR_FAST(<=>, std::partial_ordering::equivalent) // *NOPAD*
+        return three_way_general(lhs, rhs);
     }
 
     /// @brief comparison: 3-way
@@ -35204,11 +35454,11 @@ public:
         JSON_HEDLEY_DIAGNOSTIC_PUSH
         JSON_HEDLEY_PRAGMA(GCC diagnostic ignored "-Wfloat-equal")
 #endif
-        JSON_IMPLEMENT_OPERATOR( ==, true, false, false,
-                                 compare_iteratively<false>(lhs, rhs, false) == compare_result::equal, true)
+        JSON_IMPLEMENT_SCALAR_FAST( ==, true)
 #ifdef __GNUC__
         JSON_HEDLEY_DIAGNOSTIC_POP
 #endif
+        return equal_dispatch(lhs, rhs, equal_bounded_containers {});
     }
 
     // The friend comparisons with a scalar name the JSON type via decltype of
@@ -35263,11 +35513,8 @@ public:
     /// @sa https://json.nlohmann.me/api/basic_json/operator_lt/
     friend bool operator<(const_reference lhs, const_reference rhs) noexcept
     {
-        // default_result is used if we cannot compare values. In that case,
-        // we compare types. Note we have to call the operator explicitly,
-        // because MSVC has problems otherwise.
-        JSON_IMPLEMENT_OPERATOR( <, false, false, operator<(lhs_type, rhs_type),
-                                 compare_iteratively<true>(lhs, rhs, true) == compare_result::less, false)
+        JSON_IMPLEMENT_SCALAR_FAST( <, false)
+        return less_general(lhs, rhs);
     }
 
     /// @brief comparison: less than
@@ -35378,6 +35625,8 @@ public:
 #endif
 
 #undef JSON_IMPLEMENT_OPERATOR
+#undef JSON_IMPLEMENT_SCALAR_FAST
+#undef JSON_IMPLEMENT_SCALAR_CASES
 
     /// @}
 

tests/src/unit-comparison.cpp

@@ -19,6 +19,7 @@
 
 #include <cctype>
 #include <cstdint>
+#include <limits>
 #include <map>
 #include <string>
 #include <utility>
@@ -1149,3 +1150,157 @@ TEST_CASE("operator<=> of binary values with a different subtype does not depend
     }
 }
 #endif
+
+TEST_CASE("comparison of scalars (inline fast path)")
+{
+    const json::number_integer_t neg = -1;
+    const json::number_integer_t pos = 5;
+    const json::number_unsigned_t big = (std::numeric_limits<json::number_unsigned_t>::max)();
+
+    SECTION("same type")
+    {
+        CHECK(json(pos) == json(pos));
+        CHECK(json(pos) != json(neg));
+        CHECK(json(neg) < json(pos));
+        CHECK_FALSE(json(pos) < json(neg));
+        CHECK_FALSE(json(pos) < json(pos));
+        CHECK(json(pos) <= json(pos));
+
+        CHECK(json(1u) == json(1u));
+        CHECK(json(1u) < json(2u));
+        CHECK_FALSE(json(2u) < json(1u));
+
+        CHECK(json(1.5) == json(1.5));
+        CHECK(json(1.5) < json(2.5));
+        CHECK_FALSE(json(2.5) < json(1.5));
+        CHECK(json(-0.0) == json(0.0));
+    }
+
+    SECTION("mixed number types")
+    {
+        // negative signed vs unsigned must not wrap around
+        CHECK(json(neg) < json(1u));
+        CHECK(json(1u) > json(neg));
+        CHECK(json(neg) != json(big));
+        CHECK(json(neg) < json(big));
+        CHECK_FALSE(json(big) < json(neg));
+
+        CHECK(json(pos) == json(5u));
+        CHECK(json(5u) == json(pos));
+        CHECK(json(pos) < json(6u));
+        CHECK(json(pos) == json(5.0));
+        CHECK(json(5.0) == json(pos));
+        CHECK(json(5u) == json(5.0));
+        CHECK(json(neg) < json(0.5));
+        CHECK(json(0.5) < json(1u));
+    }
+
+    SECTION("boolean and null")
+    {
+        CHECK(json(true) == json(true));
+        CHECK(json(true) != json(false));
+        CHECK(json(false) < json(true));
+        CHECK_FALSE(json(true) < json(false));
+        CHECK_FALSE(json(true) < json(true));
+
+        CHECK(json(nullptr) == json(nullptr));
+        CHECK_FALSE(json(nullptr) < json(nullptr));
+        CHECK(json(nullptr) <= json(nullptr));
+        CHECK(json(nullptr) < json(false));
+        CHECK(json(nullptr) != json(false));
+        CHECK(json(nullptr) != json(0));
+    }
+
+    SECTION("string vs number")
+    {
+        CHECK(json("5") != json(5));
+        CHECK(json(5) != json("5"));
+        CHECK(json(5) < json("5")); // number < string by type order
+        CHECK_FALSE(json("5") < json(5));
+        CHECK(json(1.5) < json(""));
+        CHECK(json("a") == json("a"));
+        CHECK(json("a") < json("b"));
+    }
+
+    SECTION("NaN")
+    {
+        const json nan = std::numeric_limits<json::number_float_t>::quiet_NaN();
+        const json one = 1.0;
+
+        CHECK_FALSE(nan == nan);
+        CHECK(nan != nan);
+        CHECK_FALSE(nan < nan);
+        CHECK_FALSE(nan <= nan);
+        CHECK_FALSE(nan >= nan);
+        CHECK_FALSE(nan > nan);
+
+        CHECK_FALSE(nan == one);
+        CHECK_FALSE(one == nan);
+        CHECK_FALSE(nan < one);
+        CHECK_FALSE(one < nan);
+        CHECK_FALSE(nan <= one);
+        CHECK_FALSE(one <= nan);
+        CHECK_FALSE(nan >= one);
+        CHECK_FALSE(one >= nan);
+
+        // NaN against other number types
+        CHECK_FALSE(nan == json(1));
+        CHECK_FALSE(json(1) == nan);
+        CHECK_FALSE(nan < json(1u));
+        CHECK_FALSE(json(1u) < nan);
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+        CHECK((nan <=> nan) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((nan <=> one) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((one <=> nan) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((one <=> json(2.0)) == std::partial_ordering::less); // *NOPAD*
+        CHECK((json(2.0) <=> one) == std::partial_ordering::greater); // *NOPAD*
+        CHECK((one <=> one) == std::partial_ordering::equivalent); // *NOPAD*
+        CHECK((json(neg) <=> json(1u)) == std::partial_ordering::less); // *NOPAD*
+        CHECK((json(nullptr) <=> json(nullptr)) == std::partial_ordering::equivalent); // *NOPAD*
+        CHECK((json(true) <=> json(false)) == std::partial_ordering::greater); // *NOPAD*
+#endif
+    }
+
+    SECTION("discarded")
+    {
+        const json d(json::value_t::discarded);
+
+        CHECK_FALSE(d == d);
+        CHECK_FALSE(d < d);
+        CHECK_FALSE(d == json(1));
+        CHECK_FALSE(json(1) == d);
+        CHECK_FALSE(d < json(1));
+        CHECK_FALSE(json(1) < d);
+        CHECK_FALSE(d == json(nullptr));
+        CHECK_FALSE(json(nullptr) == d);
+
+#if JSON_HAS_THREE_WAY_COMPARISON
+        CHECK((d <=> d) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((d <=> json(1)) == std::partial_ordering::unordered); // *NOPAD*
+        CHECK((json(1) <=> d) == std::partial_ordering::unordered); // *NOPAD*
+#endif
+    }
+
+    SECTION("scalar overloads")
+    {
+        CHECK(json(5) == 5);
+        CHECK(5 == json(5));
+        CHECK(json(5) < 6);
+        CHECK(4 < json(5));
+        CHECK(json(1.5) == 1.5);
+        CHECK(json(true) == true);
+        CHECK(json(5) != 6);
+        CHECK(json(1u) == 1.0);
+    }
+
+    SECTION("structured values still compare")
+    {
+        CHECK(json({1, 2}) == json({1, 2}));
+        CHECK(json({1, 2}) < json({1, 3}));
+        CHECK(json({{"a", 1}}) == json({{"a", 1}}));
+        CHECK(json::binary({1, 2}) == json::binary({1, 2}));
+        CHECK(json(1) < json::array());
+        CHECK(json::object() < json::array());
+    }
+}

tests/src/unit-constructor2.cpp

@@ -8,6 +8,9 @@
 
 #include "doctest_compatibility.h"
 
+#include <string>
+#include <vector>
+
 #include <nlohmann/json.hpp>
 using nlohmann::json;
 
@@ -184,3 +187,139 @@ TEST_CASE("other constructors and destructor")
         }
     }
 }
+
+TEST_CASE("copy constructor and destructor across value types")
+{
+    SECTION("copy of a scalar is independent of the source")
+    {
+        const json src_null = nullptr;
+        const json src_bool = true;
+        const json src_int = -7;
+        const json src_uint = 7u;
+        const json src_float = 2.5;
+
+        json c_null(src_null);
+        json c_bool(src_bool);
+        json c_int(src_int);
+        json c_uint(src_uint);
+        json c_float(src_float);
+
+        CHECK(c_null.is_null());
+        CHECK(c_bool == true);
+        CHECK(c_int == -7);
+        CHECK(c_uint == 7u);
+        CHECK(c_float == 2.5);
+
+        c_int = 8;
+        c_float = 3.5;
+        CHECK(src_int == -7);
+        CHECK(src_float == 2.5);
+    }
+
+    SECTION("copy of an owning value, then mutate the copy")
+    {
+        const json src_string = "hello";
+        const json src_array = {1, 2, 3};
+        const json src_object = {{"a", 1}, {"b", {{"c", 2}}}};
+        const json src_binary = json::binary({1, 2, 3});
+
+        json c_string(src_string);
+        json c_array(src_array);
+        json c_object(src_object);
+        json c_binary(src_binary);
+
+        CHECK(c_string == src_string);
+        CHECK(c_array == src_array);
+        CHECK(c_object == src_object);
+        CHECK(c_binary == src_binary);
+
+        c_string.get_ref<json::string_t&>() += " world";
+        c_array.push_back(4);
+        c_object["b"]["c"] = 3;
+        c_object["d"] = 4;
+        c_binary.get_binary().push_back(4);
+
+        CHECK(src_string == "hello");
+        CHECK(src_array.size() == 3);
+        CHECK(src_object["b"]["c"] == 2);
+        CHECK_FALSE(src_object.contains("d"));
+        CHECK(src_binary.get_binary().size() == 3);
+
+        CHECK(c_string == "hello world");
+        CHECK(c_array.size() == 4);
+        CHECK(c_object["b"]["c"] == 3);
+        CHECK(c_binary.get_binary().size() == 4);
+    }
+
+    SECTION("copy keeps parent pointers valid (JSON_DIAGNOSTICS)")
+    {
+        const json src = {{"a", {1, 2}}, {"b", "x"}};
+        json c(src);
+        c["a"].push_back(3);
+        CHECK(c["a"].size() == 3);
+        CHECK(src["a"].size() == 2);
+    }
+
+    SECTION("vector of mixed types: build, copy, and destroy")
+    {
+        {
+            std::vector<json> v;
+            for (int i = 0; i < 70; ++i)
+            {
+                switch (i % 9)
+                {
+                    case 0:
+                        v.emplace_back(nullptr);
+                        break;
+                    case 1:
+                        v.emplace_back(i % 2 == 0);
+                        break;
+                    case 2:
+                        v.emplace_back(-i);
+                        break;
+                    case 3:
+                        v.emplace_back(static_cast<json::number_unsigned_t>(i));
+                        break;
+                    case 4:
+                        v.emplace_back(i * 0.5);
+                        break;
+                    case 5:
+                        v.emplace_back(std::string(static_cast<std::size_t>(i), 'x'));
+                        break;
+                    case 6:
+                        v.emplace_back(json::array({i, "s", nullptr}));
+                        break;
+                    case 7:
+                        v.emplace_back(json::object({{"k", i}}));
+                        break;
+                    default:
+                        v.emplace_back(json::binary({1, 2, 3}));
+                        break;
+                }
+            }
+            REQUIRE(v.size() == 70);
+
+            std::vector<json> copy(v);
+            CHECK(copy == v);
+
+            // mutating the copy leaves the original alone
+            copy[5] = "changed";
+            copy[6].push_back(1);
+            CHECK(copy != v);
+            CHECK(v[5] == std::string(5, 'x'));
+            CHECK(v[6].size() == 3);
+
+            v.erase(v.begin(), v.begin() + 10);
+            CHECK(v.size() == 60);
+        }
+
+        // destroying the discarded and default-constructed values
+        std::vector<json> w(4);
+        w.emplace_back(json::value_t::discarded);
+        w.emplace_back(json::value_t::object);
+        w.emplace_back(json::value_t::array);
+        w.emplace_back(json::value_t::string);
+        w.emplace_back(json::value_t::binary);
+        CHECK(w.size() == 9);
+    }
+}