Destroy the first levels of a value by recursion

The pointer-reversing walk in destroy_container() avoids the call stack but
moves a value around for every container it visits. Recurse for the first
recursion_depth_limit() levels instead - no allocation, no thread-local, a
bounded frame - and hand anything deeper to the walk. Destroy twitter
0.49 -> 0.36 ms, citm 0.80 -> 0.74 ms (3.12.0: 0.47 / 1.00).

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

docs/mkdocs/docs/api/basic_json/array_t.md

@@ -27,7 +27,7 @@ To store objects in C++, a type is defined by the template parameters explained
 
 ### Default type
 
-With the default values for `ArrayType` (`std::vector`) and `AllocatorType` (`std::allocator`), the default value for
+With the default values for `ArrayType` (`std::vector`) and `AllocatorType` (`std::allocator`), the default type for
 `array_t` is:
 
 ```cpp

docs/mkdocs/docs/api/basic_json/basic_json.md

@@ -68,6 +68,14 @@ basic_json(basic_json&& other) noexcept;
      `std::unordered_map`, `std::multimap`, and `std::unordered_multimap` with a `key_type` compatible to `string_t`
      and a `value_type` from which a `basic_json` value can be constructed.
     - **strings**: `string_t`, string literals, and all compatible string containers can be used.
+
+    !!! note "Rvalue containers are moved from"
+
+        If `val` is an rvalue of a compatible array-like or object-like container other than `array_t`/`object_t`
+        (e.g., `json j = std::move(vec);` or `json j(std::vector<std::string>{...});`), the elements are moved into
+        the JSON value instead of copied. Afterwards, the elements of the source container are in a valid but
+        unspecified state (**moved-from**). Keys of associative containers are `const` and therefore still copied.
+        Lvalues and `const` rvalues are copied, and the source is left unchanged.
     - **numbers**: [`number_integer_t`](number_integer_t.md), [`number_unsigned_t`](number_unsigned_t.md),
      [`number_float_t`](number_float_t.md), and all convertible number types such as `int`, `size_t`, `int64_t`, `float`
      or `double` can be used.
@@ -236,7 +244,9 @@ basic_json(basic_json&& other) noexcept;
 2. No-throw guarantee: this constructor never throws exceptions.
 3. Depends on the called constructor. For types directly supported by the library (i.e., all types for which no
    `to_json()` function was provided), a strong guarantee holds: if an exception is thrown, there are no changes to any
-   JSON value.
+   JSON value. **Exception:** when `val` is an rvalue array-like or object-like container, its elements are moved one
+   by one; if an exception is thrown in the middle of the conversion (e.g., an allocation failure), the JSON value is
+   unchanged, but the source container may be left partially moved-from.
 4. Depends on the called constructor. For types directly supported by the library (i.e., all types for which no
    `to_json()` function was provided), a strong guarantee holds: if an exception is thrown, there are no changes to any
    JSON value.

docs/mkdocs/docs/api/basic_json/boolean_t.md

@@ -23,7 +23,7 @@ To store boolean values in C++, a type is defined by the template parameter  `Bo
 
 ### Default type
 
-With the default values for `BooleanType` (`#!cpp bool`), the default value for `boolean_t` is `#!cpp bool`.
+With the default values for `BooleanType` (`#!cpp bool`), the default type for `boolean_t` is `#!cpp bool`.
 
 ### Storage
 

docs/mkdocs/docs/api/basic_json/number_float_t.md

@@ -35,7 +35,7 @@ type to use.
 
 ### Default type
 
-With the default values for `NumberFloatType` (`double`), the default value for `number_float_t` is `#!cpp double`.
+With the default values for `NumberFloatType` (`double`), the default type for `number_float_t` is `#!cpp double`.
 
 ### Default behavior
 
@@ -52,8 +52,9 @@ With the default values for `NumberFloatType` (`double`), the default value for
 > interoperability can be achieved by implementations that expect no more precision or range than these provide, in the
 > sense that implementations will approximate JSON numbers within the expected precision.
 
-This implementation does exactly follow this approach, as it uses double precision floating-point numbers. Note that
-numbers smaller than `-1.79769313486232e+308` or greater than `1.79769313486232e+308` are rejected during parsing with
+This implementation does exactly follow this approach, as it uses double precision floating-point numbers by default.
+Note that with the default `double`, numbers smaller than `-1.79769313486232e+308` or greater than
+`1.79769313486232e+308` are rejected during parsing with
 [`out_of_range.406`](../../home/exceptions.md#jsonexceptionout_of_range406). Non-finite values (NaN and infinity) can
 only be stored by assigning them directly and are serialized to `null`.
 

docs/mkdocs/docs/api/basic_json/number_integer_t.md

@@ -31,7 +31,7 @@ to use.
 
 ### Default type
 
-With the default values for `NumberIntegerType` (`std::int64_t`), the default value for `number_integer_t` is
+With the default values for `NumberIntegerType` (`std::int64_t`), the default type for `number_integer_t` is
 `#!cpp std::int64_t`.
 
 ### Default behavior

docs/mkdocs/docs/api/basic_json/number_unsigned_t.md

@@ -32,7 +32,7 @@ the type to use.
 
 ### Default type
 
-With the default values for `NumberUnsignedType` (`std::uint64_t`), the default value for `number_unsigned_t` is
+With the default values for `NumberUnsignedType` (`std::uint64_t`), the default type for `number_unsigned_t` is
 `#!cpp std::uint64_t`.
 
 ### Default behavior

docs/mkdocs/docs/api/basic_json/object_t.md

@@ -36,7 +36,7 @@ To store objects in C++, a type is defined by the template parameters described
 ### Default type
 
 With the default values for `ObjectType` (`std::map`), `StringType` (`std::string`), and `AllocatorType`
-(`std::allocator`), the default value for `object_t` is:
+(`std::allocator`), the default type for `object_t` is:
 
 ```cpp
 // until C++14

docs/mkdocs/docs/api/basic_json/string_t.md

@@ -32,7 +32,7 @@ JSON class into byte-sized characters during deserialization.
 
 ### Default type
 
-With the default values for `StringType` (`std::string`), the default value for `string_t` is `#!cpp std::string`.
+With the default values for `StringType` (`std::string`), the default type for `string_t` is `#!cpp std::string`.
 
 ### Encoding
 

docs/mkdocs/docs/features/types/index.md

@@ -120,7 +120,7 @@ requirements are collected in [Template Parameter Requirements](template_paramet
 
 ### Default type
 
-With the default values for *ObjectType* (`std::map`), *StringType* (`std::string`), and *AllocatorType* (`std::allocator`), the default value for `object_t` is:
+With the default values for *ObjectType* (`std::map`), *StringType* (`std::string`), and *AllocatorType* (`std::allocator`), the default type for `object_t` is:
 
 ```cpp
 std::map<
@@ -177,7 +177,7 @@ json j = m;
 
 ### Default type
 
-With the default values for *ArrayType* (`std::vector`) and *AllocatorType* (`std::allocator`), the default value for `array_t` is:
+With the default values for *ArrayType* (`std::vector`) and *AllocatorType* (`std::allocator`), the default type for `array_t` is:
 
 ```cpp
 std::vector<
@@ -209,7 +209,7 @@ Unicode values are split by the JSON class into byte-sized characters during des
 
 ### Default type
 
-With the default values for *StringType* (`std::string`), the default value for `string_t` is `#!cpp std::string`.
+With the default values for *StringType* (`std::string`), the default type for `string_t` is `#!cpp std::string`.
 
 ### Encoding
 
@@ -234,7 +234,7 @@ String values are stored as pointers in a `basic_json` type. That is, for any ac
 
 ### Default type
 
-With the default values for *BooleanType* (`#!cpp bool`), the default value for `boolean_t` is `#!cpp bool`.
+With the default values for *BooleanType* (`#!cpp bool`), the default type for `boolean_t` is `#!cpp bool`.
 
 ### Storage
 
@@ -252,9 +252,9 @@ This description includes both integer and floating-point numbers. However, C++
 
 ### Default types
 
-With the default values for *NumberIntegerType* (`std::int64_t`), the default value for `number_integer_t` is `std::int64_t`.
-With the default values for *NumberUnsignedType* (`std::uint64_t`), the default value for `number_unsigned_t` is `std::uint64_t`.
-With the default values for *NumberFloatType* (`#!cpp double`), the default value for `number_float_t` is `#!cpp double`.
+With the default values for *NumberIntegerType* (`std::int64_t`), the default type for `number_integer_t` is `std::int64_t`.
+With the default values for *NumberUnsignedType* (`std::uint64_t`), the default type for `number_unsigned_t` is `std::uint64_t`.
+With the default values for *NumberFloatType* (`#!cpp double`), the default type for `number_float_t` is `#!cpp double`.
 
 ### Default behavior
 
@@ -281,7 +281,7 @@ As this range is a subrange of the exactly supported range [`INT64_MIN`, `INT64_
 
 > This specification allows implementations to set limits on the range and precision of numbers accepted. Since software that implements IEEE 754-2008 binary64 (double precision) numbers is generally available and widely used, good interoperability can be achieved by implementations that expect no more precision or range than these provide, in the sense that implementations will approximate JSON numbers within the expected precision.
 
-This implementation does exactly follow this approach, as it uses double precision floating-point numbers. Note that numbers smaller than `#!c -1.79769313486232e+308` or greater than `#!c 1.79769313486232e+308` are rejected during parsing with [`out_of_range.406`](../../home/exceptions.md#jsonexceptionout_of_range406). Non-finite values (NaN and infinity) can only be stored by assigning them directly and are serialized to `#!json null`.
+This implementation does exactly follow this approach, as it uses double precision floating-point numbers by default (see *NumberFloatType* above). Note that with the default `#!cpp double`, numbers smaller than `#!c -1.79769313486232e+308` or greater than `#!c 1.79769313486232e+308` are rejected during parsing with [`out_of_range.406`](../../home/exceptions.md#jsonexceptionout_of_range406). Non-finite values (NaN and infinity) can only be stored by assigning them directly and are serialized to `#!json null`.
 
 ### Storage
 

docs/mkdocs/docs/home/releases.md

@@ -9,6 +9,15 @@ published on the [GitHub releases page](https://github.com/nlohmann/json/release
     This documentation is built from the `develop` branch and may describe changes that are not part of a release
     yet. Their version numbers are followed by an <span class="unreleased-version">unreleased</span> badge.
 
+## Unreleased
+
+- Converting an rvalue container to JSON (`json j = std::move(vec);`, `json j = std::map<std::string, std::string>{...};`)
+  now **moves** the elements instead of copying them (e.g., `std::vector<std::string>` with 100,000 strings: 4.7 ms to
+  2.3 ms). Two consequences: the source container is left in a valid but unspecified (moved-from) state, and if an
+  exception is thrown during the conversion, the source may be partially moved-from, so the strong exception guarantee
+  no longer holds for the source of an rvalue conversion (the JSON value itself is unaffected). Lvalues and `const`
+  objects are still copied. See [`basic_json`](../api/basic_json/basic_json.md).
+
 ## v3.12.0 (2025-04-11)
 
 Fixes bugs found in 3.11.3 and adds several features. All changes are backward-compatible.

include/nlohmann/detail/conversions/from_json.hpp

@@ -266,6 +266,22 @@ template<typename ConstructibleArrayType>
 void from_json_array_reserve(ConstructibleArrayType& /*arr*/, std::size_t /*size*/, priority_tag<0> /*unused*/)
 {}
 
+// append through push_back() where available; containers without it (such as
+// std::set) fall back to insert(end(), ...), which is what std::inserter did
+template<typename ConstructibleArrayType, typename ValueType>
+auto from_json_array_append(ConstructibleArrayType& arr, ValueType&& value, priority_tag<1> /*unused*/)
+-> decltype(arr.push_back(std::forward<ValueType>(value)), void())
+{
+    arr.push_back(std::forward<ValueType>(value));
+}
+
+template<typename ConstructibleArrayType, typename ValueType>
+void from_json_array_append(ConstructibleArrayType& arr, ValueType&& value, priority_tag<0> /*unused*/)
+{
+    using std::end;
+    arr.insert(end(arr), std::forward<ValueType>(value));
+}
+
 template<typename BasicJsonType, typename ConstructibleArrayType,
          enable_if_t<
              std::is_assignable<ConstructibleArrayType&, ConstructibleArrayType>::value,
@@ -275,17 +291,15 @@ auto from_json_array_impl(const BasicJsonType& j, ConstructibleArrayType& arr, p
     j.template get<typename ConstructibleArrayType::value_type>(),
     void())
 {
-    using std::end;
-
     ConstructibleArrayType ret;
-    from_json_array_reserve(ret, j.size(), priority_tag<1> {});
-    std::transform(j.begin(), j.end(),
-                   std::inserter(ret, end(ret)), [](const BasicJsonType & i)
+    const auto* inner_array = j.template get_ptr<const typename BasicJsonType::array_t*>();
+    from_json_array_reserve(ret, inner_array->size(), priority_tag<1> {});
+    for (const auto& elem : *inner_array)
     {
         // get<BasicJsonType>() returns *this, this won't call a from_json
         // method when value_type is BasicJsonType
-        return i.template get<typename ConstructibleArrayType::value_type>();
-    });
+        from_json_array_append(ret, elem.template get<typename ConstructibleArrayType::value_type>(), priority_tag<1> {});
+    }
     arr = std::move(ret);
 }
 
@@ -383,9 +397,13 @@ inline void from_json(const BasicJsonType& j, ConstructibleObjectType& obj)
     ConstructibleObjectType ret;
     const auto* inner_object = j.template get_ptr<const typename BasicJsonType::object_t*>();
     from_json_object_reserve(ret, inner_object->size(), priority_tag<1> {});
+    using value_type = typename ConstructibleObjectType::value_type;
     for (const auto& p : *inner_object)
     {
-        ret.emplace(p.first, p.second.template get<typename ConstructibleObjectType::mapped_type>());
+        // the keys of the source are sorted for the usual object types, so
+        // hinting the end makes every insertion constant-time (for containers
+        // that order their keys the same way)
+        ret.insert(ret.end(), value_type(p.first, p.second.template get<typename ConstructibleObjectType::mapped_type>()));
     }
     obj = std::move(ret);
 }

include/nlohmann/detail/conversions/to_json.hpp

@@ -167,6 +167,17 @@ struct external_constructor<value_t::array>
         j.replace_value(value_t::array, value);
     }
 
+    /// an rvalue container: move its elements
+    template<typename BasicJsonType, typename CompatibleArrayType>
+    static void construct_moving(BasicJsonType& j, CompatibleArrayType& arr)
+    {
+        using std::begin;
+        using std::end;
+
+        const typename BasicJsonType::json_value value(j.template create<typename BasicJsonType::array_t>(std::make_move_iterator(begin(arr)), std::make_move_iterator(end(arr))));
+        j.replace_value(value_t::array, value);
+    }
+
     template<typename BasicJsonType>
     static void construct(BasicJsonType& j, const std::vector<bool>& arr)
     {
@@ -217,6 +228,17 @@ struct external_constructor<value_t::object>
         j.replace_value(value_t::object, value);
     }
 
+    /// an rvalue container: move its values
+    template<typename BasicJsonType, typename CompatibleObjectType>
+    static void construct_moving(BasicJsonType& j, CompatibleObjectType& obj)
+    {
+        using std::begin;
+        using std::end;
+
+        const typename BasicJsonType::json_value value(j.template create<typename BasicJsonType::object_t>(std::make_move_iterator(begin(obj)), std::make_move_iterator(end(obj))));
+        j.replace_value(value_t::object, value);
+    }
+
     template < typename BasicJsonType, typename CompatibleObjectType,
                enable_if_t < !std::is_same<CompatibleObjectType, typename BasicJsonType::object_t>::value, int > = 0 >
     static void construct(BasicJsonType& j, const CompatibleObjectType& obj)
@@ -354,6 +376,66 @@ inline void to_json(BasicJsonType& j, const CompatibleArrayType& arr)
     external_constructor<value_t::array>::construct(j, arr);
 }
 
+// types that have a dedicated to_json overload and must not be treated as generic containers
+template<typename T>
+struct has_dedicated_to_json : std::false_type {};
+
+#if JSON_HAS_FILESYSTEM || JSON_HAS_EXPERIMENTAL_FILESYSTEM
+template<>
+struct has_dedicated_to_json<std_fs::path> : std::true_type {};
+#endif
+
+#if defined(__cpp_lib_char8_t)
+template<typename Tr, typename Allocator>
+struct has_dedicated_to_json<std::basic_string<char8_t, Tr, Allocator>> : std::true_type {};
+#endif
+
+#if JSON_HAS_RANGE_VIEW_CONVERSION
+template<typename T>
+struct is_range_view_type : is_compatible_range_view<T> {};
+#else
+template<typename T>
+struct is_range_view_type : std::false_type {};
+#endif
+
+#if JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS
+template<typename T>
+struct is_enum_keyed_map_type : is_enum_keyed_map<T> {};
+#else
+template<typename T>
+struct is_enum_keyed_map_type : std::false_type {};
+#endif
+
+// rvalue (non-const) container: the generic to_json takes its elements by move
+template<typename T>
+using is_movable_container_rvalue = conjunction <
+                                    negation<std::is_lvalue_reference<T>>,
+                                    negation<std::is_const<T>>,
+                                    negation<has_dedicated_to_json<T>>,
+                                    negation<is_basic_json<T>>,
+                                    negation<is_range_view_type<T>>,
+                                    negation<is_enum_keyed_map_type<T>>,
+                                    negation<std::is_same<T, std::vector<bool>>> >;
+
+// rvalue array-like containers
+template<typename BasicJsonType, typename T>
+using is_rvalue_array_to_json = conjunction <
+                                is_movable_container_rvalue<T>,
+                                negation<std::is_same<T, typename BasicJsonType::array_t>>,
+                                negation<std::is_same<T, typename BasicJsonType::binary_t>>,
+                                is_compatible_array_type<BasicJsonType, T>,
+                                negation<is_compatible_object_type<BasicJsonType, T>>,
+                                negation<is_compatible_string_type<BasicJsonType, T>>,
+                                negation<is_compatible_binary_type<BasicJsonType, T>> >;
+
+// rvalue containers: their elements are moved instead of copied
+template < typename BasicJsonType, typename CompatibleArrayType,
+           enable_if_t < is_rvalue_array_to_json<BasicJsonType, CompatibleArrayType>::value, int > = 0 >
+inline void to_json(BasicJsonType& j, CompatibleArrayType && arr)
+{
+    external_constructor<value_t::array>::construct_moving(j, arr);
+}
+
 #if JSON_HAS_RANGE_VIEW_CONVERSION
 template < typename BasicJsonType, typename T,
            enable_if_t < is_compatible_range_view<std::remove_cvref_t<T>>::value
@@ -427,6 +509,20 @@ inline void to_json(BasicJsonType& j, const EnumKeyedMap& map)
 }
 #endif
 
+// rvalue object-like containers
+template<typename BasicJsonType, typename T>
+using is_rvalue_object_to_json = conjunction <
+                                 is_movable_container_rvalue<T>,
+                                 negation<std::is_same<T, typename BasicJsonType::object_t>>,
+                                 is_compatible_object_type<BasicJsonType, T> >;
+
+template < typename BasicJsonType, typename CompatibleObjectType,
+           enable_if_t < is_rvalue_object_to_json<BasicJsonType, CompatibleObjectType>::value, int > = 0 >
+inline void to_json(BasicJsonType& j, CompatibleObjectType && obj)
+{
+    external_constructor<value_t::object>::construct_moving(j, obj);
+}
+
 template<typename BasicJsonType>
 inline void to_json(BasicJsonType& j, typename BasicJsonType::object_t&& obj)
 {

include/nlohmann/detail/exceptions.hpp

@@ -428,7 +428,7 @@ void templated_json_throw(ExceptionType exception)
 @throw type_error.302 always
 */
 template<typename BasicJsonType>
-JSON_HEDLEY_NO_RETURN inline void throw_type_must_be(const char* expected, const BasicJsonType& j)
+JSON_HEDLEY_NO_RETURN JSON_HEDLEY_NEVER_INLINE void throw_type_must_be(const char* expected, const BasicJsonType& j)
 {
     static_cast<void>(expected); // unused when JSON_NOEXCEPTION is defined
     static_cast<void>(j);
@@ -443,7 +443,7 @@ JSON_HEDLEY_NO_RETURN inline void throw_type_must_be(const char* expected, const
 @throw type_error always
 */
 template<typename BasicJsonType>
-JSON_HEDLEY_NO_RETURN inline void throw_cannot_use_with(const exception_id id_, const char* operation, const BasicJsonType& j)
+JSON_HEDLEY_NO_RETURN JSON_HEDLEY_NEVER_INLINE void throw_cannot_use_with(const exception_id id_, const char* operation, const BasicJsonType& j)
 {
     static_cast<void>(id_); // unused when JSON_NOEXCEPTION is defined
     static_cast<void>(operation);
@@ -451,6 +451,18 @@ JSON_HEDLEY_NO_RETURN inline void throw_cannot_use_with(const exception_id id_,
     JSON_THROW(type_error::create(id_, concat("cannot use ", operation, " with ", j.type_name()), &j));
 }
 
+/*!
+@brief throws ExceptionType with a fixed message; kept out of line so that callers do not carry the exception construction
+*/
+template<typename ExceptionType, typename BasicJsonContext>
+JSON_HEDLEY_NO_RETURN JSON_HEDLEY_NEVER_INLINE void throw_literal(const exception_id id_, const char* what_arg, BasicJsonContext context)
+{
+    static_cast<void>(id_); // unused when JSON_NOEXCEPTION is defined
+    static_cast<void>(what_arg);
+    static_cast<void>(context);
+    JSON_THROW(ExceptionType::create(id_, what_arg, context));
+}
+
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END
 

include/nlohmann/detail/input/json_sax.hpp

@@ -674,16 +674,19 @@ class json_sax_dom_callback_parser
         if (keep && ref_stack.back())
         {
             auto& obj = *ref_stack.back()->m_data.m_value.object;
-            const auto it = obj.find(val);
-            if (it != obj.end())
+            // a single lookup: insert the placeholder, or find the existing member
+            const auto inserted = obj.emplace(val, discarded);
+            const auto it = inserted.first;
+            if (!inserted.second)
             {
                 // this is a duplicate key (legal in JSON); remember its
                 // current value so it can be restored later if the new
                 // value is rejected by the callback, instead of being
                 // erased together with the discarded placeholder
                 duplicate_key_stash.emplace_back(&(it->second), it->second);
+                it->second = discarded;
             }
-            object_element = &(obj[val] = discarded);
+            object_element = &(it->second);
         }
 
         return true;

include/nlohmann/detail/iterators/iter_impl.hpp

@@ -280,7 +280,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
     }
 
     /// @throw invalid_iterator.214 always
-    JSON_HEDLEY_NO_RETURN void throw_cannot_get_value() const
+    JSON_HEDLEY_NO_RETURN JSON_HEDLEY_NEVER_INLINE void throw_cannot_get_value() const
     {
         JSON_THROW(invalid_iterator::create(exception_id::iterator_value_unavailable, "cannot get value", m_object));
     }
@@ -294,40 +294,25 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
     {
         JSON_ASSERT(m_object != nullptr);
 
-        switch (m_object->m_data.m_type)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::array))
         {
-            case value_t::object:
-            {
-                JSON_ASSERT(m_it.object_iterator != m_object->m_data.m_value.object->end());
-                return m_it.object_iterator->second;
-            }
-
-            case value_t::array:
-            {
-                JSON_ASSERT(m_it.array_iterator != m_object->m_data.m_value.array->end());
-                return *m_it.array_iterator;
-            }
-
-            case value_t::null:
-                throw_cannot_get_value();
-
-            case value_t::string:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::binary:
-            case value_t::discarded:
-            default:
-            {
-                if (JSON_HEDLEY_LIKELY(m_it.primitive_iterator.is_begin()))
-                {
-                    return *m_object;
-                }
-
-                throw_cannot_get_value();
-            }
+            JSON_ASSERT(m_it.array_iterator != m_object->m_data.m_value.array->end());
+            return *m_it.array_iterator;
         }
+
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::object))
+        {
+            JSON_ASSERT(m_it.object_iterator != m_object->m_data.m_value.object->end());
+            return m_it.object_iterator->second;
+        }
+
+        // neither array nor object: a primitive value (null has no value to point to)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type != value_t::null && m_it.primitive_iterator.is_begin()))
+        {
+            return *m_object;
+        }
+
+        throw_cannot_get_value();
     }
 
     /*!
@@ -338,38 +323,25 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
     {
         JSON_ASSERT(m_object != nullptr);
 
-        switch (m_object->m_data.m_type)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::array))
         {
-            case value_t::object:
-            {
-                JSON_ASSERT(m_it.object_iterator != m_object->m_data.m_value.object->end());
-                return &(m_it.object_iterator->second);
-            }
-
-            case value_t::array:
-            {
-                JSON_ASSERT(m_it.array_iterator != m_object->m_data.m_value.array->end());
-                return &*m_it.array_iterator;
-            }
-
-            case value_t::null:
-            case value_t::string:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::binary:
-            case value_t::discarded:
-            default:
-            {
-                if (JSON_HEDLEY_LIKELY(m_it.primitive_iterator.is_begin()))
-                {
-                    return m_object;
-                }
-
-                throw_cannot_get_value();
-            }
+            JSON_ASSERT(m_it.array_iterator != m_object->m_data.m_value.array->end());
+            return &*m_it.array_iterator;
         }
+
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::object))
+        {
+            JSON_ASSERT(m_it.object_iterator != m_object->m_data.m_value.object->end());
+            return &(m_it.object_iterator->second);
+        }
+
+        // neither array nor object: a primitive value
+        if (JSON_HEDLEY_LIKELY(m_it.primitive_iterator.is_begin()))
+        {
+            return m_object;
+        }
+
+        throw_cannot_get_value();
     }
 
     /*!
@@ -391,35 +363,20 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
     {
         JSON_ASSERT(m_object != nullptr);
 
-        switch (m_object->m_data.m_type)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::array))
         {
-            case value_t::object:
-            {
-                std::advance(m_it.object_iterator, 1);
-                break;
-            }
-
-            case value_t::array:
-            {
-                std::advance(m_it.array_iterator, 1);
-                break;
-            }
-
-            case value_t::null:
-            case value_t::string:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::binary:
-            case value_t::discarded:
-            default:
-            {
-                ++m_it.primitive_iterator;
-                break;
-            }
+            ++m_it.array_iterator;
+            return *this;
         }
 
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::object))
+        {
+            std::advance(m_it.object_iterator, 1);
+            return *this;
+        }
+
+        // neither array nor object: a primitive value
+        ++m_it.primitive_iterator;
         return *this;
     }
 
@@ -442,35 +399,20 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
     {
         JSON_ASSERT(m_object != nullptr);
 
-        switch (m_object->m_data.m_type)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::array))
         {
-            case value_t::object:
-            {
-                std::advance(m_it.object_iterator, -1);
-                break;
-            }
-
-            case value_t::array:
-            {
-                std::advance(m_it.array_iterator, -1);
-                break;
-            }
-
-            case value_t::null:
-            case value_t::string:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::binary:
-            case value_t::discarded:
-            default:
-            {
-                --m_it.primitive_iterator;
-                break;
-            }
+            --m_it.array_iterator;
+            return *this;
         }
 
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::object))
+        {
+            std::advance(m_it.object_iterator, -1);
+            return *this;
+        }
+
+        // neither array nor object: a primitive value
+        --m_it.primitive_iterator;
         return *this;
     }
 
@@ -484,7 +426,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         // if objects are not the same, the comparison is undefined
         if (JSON_HEDLEY_UNLIKELY(m_object != other.m_object))
         {
-            JSON_THROW(invalid_iterator::create(exception_id::iterators_compare_different_values, "cannot compare iterators of different containers", m_object));
+            throw_literal<invalid_iterator>(exception_id::iterators_compare_different_values, "cannot compare iterators of different containers", m_object);
         }
 
         // value-initialized forward iterators can be compared, and must compare equal to other value-initialized iterators of the same type #4493
@@ -493,25 +435,18 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
             return true;
         }
 
-        switch (m_object->m_data.m_type)
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::array))
         {
-            case value_t::object:
-                return (m_it.object_iterator == other.m_it.object_iterator);
-
-            case value_t::array:
-                return (m_it.array_iterator == other.m_it.array_iterator);
-
-            case value_t::null:
-            case value_t::string:
-            case value_t::boolean:
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            case value_t::number_float:
-            case value_t::binary:
-            case value_t::discarded:
-            default:
-                return (m_it.primitive_iterator == other.m_it.primitive_iterator);
+            return (m_it.array_iterator == other.m_it.array_iterator);
         }
+
+        if (JSON_HEDLEY_LIKELY(m_object->m_data.m_type == value_t::object))
+        {
+            return (m_it.object_iterator == other.m_it.object_iterator);
+        }
+
+        // neither array nor object: a primitive value
+        return (m_it.primitive_iterator == other.m_it.primitive_iterator);
     }
 
     /*!
@@ -533,7 +468,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         // if objects are not the same, the comparison is undefined
         if (JSON_HEDLEY_UNLIKELY(m_object != other.m_object))
         {
-            JSON_THROW(invalid_iterator::create(exception_id::iterators_compare_different_values, "cannot compare iterators of different containers", m_object));
+            throw_literal<invalid_iterator>(exception_id::iterators_compare_different_values, "cannot compare iterators of different containers", m_object);
         }
 
         // value-initialized forward iterators can be compared, and must compare equal to other value-initialized iterators of the same type #4493
@@ -546,7 +481,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         switch (m_object->m_data.m_type)
         {
             case value_t::object:
-                JSON_THROW(invalid_iterator::create(exception_id::iterator_order_on_object, "cannot compare order of object iterators", m_object));
+                throw_literal<invalid_iterator>(exception_id::iterator_order_on_object, "cannot compare order of object iterators", m_object);
 
             case value_t::array:
                 return (m_it.array_iterator < other.m_it.array_iterator);
@@ -602,7 +537,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         switch (m_object->m_data.m_type)
         {
             case value_t::object:
-                JSON_THROW(invalid_iterator::create(exception_id::iterator_arithmetic_on_object, "cannot use offsets with object iterators", m_object));
+                throw_literal<invalid_iterator>(exception_id::iterator_arithmetic_on_object, "cannot use offsets with object iterators", m_object);
 
             case value_t::array:
             {
@@ -681,7 +616,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         switch (m_object->m_data.m_type)
         {
             case value_t::object:
-                JSON_THROW(invalid_iterator::create(exception_id::iterator_arithmetic_on_object, "cannot use offsets with object iterators", m_object));
+                throw_literal<invalid_iterator>(exception_id::iterator_arithmetic_on_object, "cannot use offsets with object iterators", m_object);
 
             case value_t::array:
                 return m_it.array_iterator - other.m_it.array_iterator;
@@ -710,7 +645,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
         switch (m_object->m_data.m_type)
         {
             case value_t::object:
-                JSON_THROW(invalid_iterator::create(exception_id::iterator_subscript_on_object, "cannot use operator[] for object iterators", m_object));
+                throw_literal<invalid_iterator>(exception_id::iterator_subscript_on_object, "cannot use operator[] for object iterators", m_object);
 
             case value_t::array:
                 return *std::next(m_it.array_iterator, n);
@@ -750,7 +685,7 @@ class iter_impl // NOLINT(cppcoreguidelines-special-member-functions,hicpp-speci
             return m_it.object_iterator->first;
         }
 
-        JSON_THROW(invalid_iterator::create(exception_id::iterator_key_not_object, "cannot use key() for non-object iterators", m_object));
+        throw_literal<invalid_iterator>(exception_id::iterator_key_not_object, "cannot use key() for non-object iterators", m_object);
     }
 
     /*!

include/nlohmann/detail/json_pointer.hpp

@@ -16,8 +16,6 @@
     #include <iosfwd> // ostream
 #endif  // JSON_NO_IO
 #include <limits> // max
-#include <map> // map
-#include <numeric> // accumulate
 #include <string> // string
 #include <utility> // move
 #include <vector> // vector
@@ -69,12 +67,21 @@ class json_pointer
     /// @sa https://json.nlohmann.me/api/json_pointer/to_string/
     string_t to_string() const
     {
-        return std::accumulate(reference_tokens.begin(), reference_tokens.end(),
-                               string_t{},
-                               [](const string_t& a, const string_t& b)
+        // one buffer: every token adds its escaped form (at least its length) and a slash
+        std::size_t length = 0;
+        for (const auto& reference_token : reference_tokens)
         {
-            return detail::concat<string_t>(a, '/', detail::escape(b));
-        });
+            length += reference_token.size() + 1;
+        }
+
+        string_t result;
+        result.reserve(length);
+        for (const auto& reference_token : reference_tokens)
+        {
+            result.append("/", 1);
+            detail::escape_into(result, reference_token);
+        }
+        return result;
     }
 
     /// @brief return a string representation of the JSON pointer
@@ -129,21 +136,29 @@ class json_pointer
     friend json_pointer operator/(const json_pointer& lhs,
                                   const json_pointer& rhs)
     {
-        return json_pointer(lhs) /= rhs;
+        json_pointer result;
+        result.reference_tokens.reserve(lhs.reference_tokens.size() + rhs.reference_tokens.size());
+        result.reference_tokens.insert(result.reference_tokens.end(), lhs.reference_tokens.begin(), lhs.reference_tokens.end());
+        result.reference_tokens.insert(result.reference_tokens.end(), rhs.reference_tokens.begin(), rhs.reference_tokens.end());
+        return result;
     }
 
     /// @brief create a new JSON pointer by appending the unescaped token at the end of the JSON pointer
     /// @sa https://json.nlohmann.me/api/json_pointer/operator_slash/
     friend json_pointer operator/(const json_pointer& lhs, string_t token) // NOLINT(performance-unnecessary-value-param)
     {
-        return json_pointer(lhs) /= std::move(token);
+        json_pointer result;
+        result.reference_tokens.reserve(lhs.reference_tokens.size() + 1);
+        result.reference_tokens.insert(result.reference_tokens.end(), lhs.reference_tokens.begin(), lhs.reference_tokens.end());
+        result.reference_tokens.push_back(std::move(token));
+        return result;
     }
 
     /// @brief create a new JSON pointer by appending the array-index-token at the end of the JSON pointer
     /// @sa https://json.nlohmann.me/api/json_pointer/operator_slash/
     friend json_pointer operator/(const json_pointer& lhs, std::size_t array_idx)
     {
-        return json_pointer(lhs) /= array_idx;
+        return lhs / detail::to_string<string_t>(array_idx);
     }
 
     /// @brief returns the parent of this JSON pointer
@@ -155,8 +170,8 @@ class json_pointer
             return *this;
         }
 
-        json_pointer res = *this;
-        res.pop_back();
+        json_pointer res;
+        res.reference_tokens = std::vector<string_t>(reference_tokens.begin(), reference_tokens.end() - 1);
         return res;
     }
 
@@ -245,14 +260,14 @@ class json_pointer
 
   private:
     /// @throw out_of_range.405 always
-    JSON_HEDLEY_NO_RETURN static void throw_no_parent()
+    JSON_HEDLEY_NEVER_INLINE JSON_HEDLEY_NO_RETURN static void throw_no_parent()
     {
         JSON_THROW(detail::out_of_range::create(detail::exception_id::patch_on_root, "JSON pointer has no parent", nullptr));
     }
 
     /// @throw out_of_range.404 always
     template<typename BasicJsonContext>
-    JSON_HEDLEY_NO_RETURN static void throw_unresolved(const string_t& reference_token, BasicJsonContext context)
+    JSON_HEDLEY_NEVER_INLINE JSON_HEDLEY_NO_RETURN static void throw_unresolved(const string_t& reference_token, BasicJsonContext context)
     {
         static_cast<void>(reference_token); // unused when JSON_NOEXCEPTION is defined
         static_cast<void>(context);
@@ -305,6 +320,28 @@ class json_pointer
             return array_index_status::not_a_number;
         }
 
+        // fast path: up to 18 decimal digits cannot overflow unsigned long long or
+        // size_type (9 digits for a 32-bit size_type), so strtoull and errno are not needed
+        constexpr std::size_t max_fast_digits = sizeof(size_type) >= 8 ? 18 : 9;
+        if (!s.empty() && s.size() <= max_fast_digits)
+        {
+            size_type acc = 0;
+            for (std::size_t i = 0; i < s.size(); ++i)
+            {
+                const char c = s[i];
+                if (JSON_HEDLEY_UNLIKELY(c < '0' || c > '9'))
+                {
+                    break; // not a plain number: let strtoull decide
+                }
+                acc = static_cast<size_type>(acc * 10 + static_cast<size_type>(c - '0'));
+                if (JSON_HEDLEY_UNLIKELY(i == s.size() - 1))
+                {
+                    idx = acc;
+                    return array_index_status::ok;
+                }
+            }
+        }
+
         const char* p = s.data();
         char* p_end = nullptr; // NOLINT(misc-const-correctness)
         errno = 0; // strtoull doesn't reset errno
@@ -361,6 +398,83 @@ class json_pointer
         return idx;
     }
 
+    /*!
+    @brief compare with a JSON pointer string without splitting it into tokens
+
+    @throw parse_error.107  if the string is not empty or begins with '/'
+    @throw parse_error.108  if character '~' is not followed by '0' or '1'
+    */
+    bool equals_string(const string_t& reference_string) const
+    {
+        if (reference_string.empty())
+        {
+            return reference_tokens.empty();
+        }
+
+        if (JSON_HEDLEY_UNLIKELY(reference_string[0] != '/'))
+        {
+            JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_missing_slash, 1, detail::concat("JSON pointer must be empty or begin with '/' - was: '", reference_string, "'"), nullptr));
+        }
+
+        // the whole string is validated, whether or not it matches
+        const std::size_t size = reference_string.size();
+        for (std::size_t pos = reference_string.find_first_of('~');
+                pos != string_t::npos;
+                pos = reference_string.find_first_of('~', pos + 1))
+        {
+            if (JSON_HEDLEY_UNLIKELY(pos == size - 1 || (reference_string[pos + 1] != '0' && reference_string[pos + 1] != '1')))
+            {
+                JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_invalid_escape, 0, "escape character '~' must be followed with '0' or '1'", nullptr));
+            }
+        }
+
+        // compare token by token, unescaping the string on the fly
+        std::size_t pos = 0; // the slash in front of the next token
+        for (const auto& reference_token : reference_tokens)
+        {
+            if (pos >= size)
+            {
+                return false;
+            }
+
+            std::size_t i = pos + 1;
+            std::size_t j = 0;
+            const std::size_t token_size = reference_token.size();
+            while (i < size && reference_string[i] != '/')
+            {
+                char c = reference_string[i++];
+                if (c == '~')
+                {
+                    c = (reference_string[i++] == '0') ? '~' : '/';
+                }
+                if (j == token_size || reference_token[j] != c)
+                {
+                    return false;
+                }
+                ++j;
+            }
+            if (j != token_size)
+            {
+                return false;
+            }
+            pos = i;
+        }
+        return pos == size;
+    }
+
+    /// @brief object member access with a key that is only copied if it is inserted
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    {
+        auto& members = *object.m_data.m_value.object;
+        const auto it = members.find(key);
+        if (it != members.end())
+        {
+            return it->second;
+        }
+        return object.operator[](key);
+    }
+
   JSON_PRIVATE_UNLESS_TESTED:
     json_pointer top() const
     {
@@ -384,84 +498,141 @@ class json_pointer
     the result independent of the order in which the flattened object is
     iterated, which is unspecified for some object types.
 
-    The prefixes form a tree and are numbered, so each of them is stored only
-    once (as a node) rather than as a copy of all of its reference tokens.
+    The prefixes form a tree and are numbered (number 0 is the empty prefix),
+    so each of them is stored only once (as a node) rather than as a copy of
+    all of its reference tokens. The tree does not own the reference tokens:
+    it refers to the tokens of the parsed keys, which must outlive it.
     */
     struct prefix_tree
     {
-        // children[id] maps a reference token to the number of the prefix
-        // extended by that token; number 0 is the empty prefix
-        std::vector<std::map<string_t, std::size_t>> children;
+        // open-addressing hash table from (parent, last reference token) to
+        // the number of the node; empty slots have token == nullptr
+        struct slot
+        {
+            std::size_t hash;
+            std::size_t parent;
+            const string_t* token;
+            std::size_t id;
+
+            slot() noexcept
+                : hash(0), parent(0), token(nullptr), id(0)
+            {}
+
+            slot(std::size_t h, std::size_t p, const string_t* t, std::size_t i) noexcept
+                : hash(h), parent(p), token(t), id(i)
+            {}
+        };
+
+        std::vector<slot> slots;
+        std::size_t used = 0;
+
+        static std::size_t hash_of(std::size_t parent, const string_t& token) noexcept
+        {
+            std::size_t h = parent * static_cast<std::size_t>(0x9E3779B97F4A7C15ULL);
+            for (std::size_t i = 0; i < token.size(); ++i)
+            {
+                h = (h ^ static_cast<unsigned char>(token[i])) * static_cast<std::size_t>(0x100000001B3ULL);
+            }
+            return h ^ (h >> 29u);
+        }
+
+        void grow()
+        {
+            std::vector<slot> old(slots.size() * 2);
+            old.swap(slots);
+            const std::size_t mask = slots.size() - 1;
+            for (const auto& s : old)
+            {
+                if (s.token != nullptr)
+                {
+                    std::size_t i = s.hash & mask;
+                    while (slots[i].token != nullptr)
+                    {
+                        i = (i + 1) & mask;
+                    }
+                    slots[i] = s;
+                }
+            }
+        }
+
+        // parent[id] is the number of the prefix without its last token
+        std::vector<std::size_t> parent;
         // is_array[id] is true iff some flattened key has the reference token
         // 0 directly below the prefix with number id
         std::vector<bool> is_array;
 
         // start with the empty prefix only
         prefix_tree()
-            : children(1)
+            : slots(16)
+            , parent(1, 0)
             , is_array(1, false)
         {}
 
         // return the number of the prefix with number id extended by
         // reference_token, adding it if it is new
-        std::size_t add_child(std::size_t id, string_t&& reference_token)
+        std::size_t add_child(std::size_t id, const string_t& reference_token)
         {
+            const std::size_t hash = hash_of(id, reference_token);
+            const std::size_t mask = slots.size() - 1;
+            std::size_t i = hash & mask;
+            while (slots[i].token != nullptr)
+            {
+                if (slots[i].hash == hash && slots[i].parent == id && *slots[i].token == reference_token)
+                {
+                    return slots[i].id;
+                }
+                i = (i + 1) & mask;
+            }
+
+            // new prefix
+            const std::size_t child = parent.size();
+            slots[i] = slot(hash, id, &reference_token, child);
             if (reference_token == "0")
             {
                 is_array[id] = true;
             }
+            parent.push_back(id);
+            is_array.push_back(false);
 
-            // read the number before the emplace_back below, which may
-            // reallocate children and invalidate the iterator
-            const std::size_t next = children.size();
-            const auto inserted = children[id].emplace(std::move(reference_token), next);
-            const std::size_t child = inserted.first->second;
-            if (inserted.second)
+            // keep the load factor below 1/2
+            if (++used * 2 > slots.size())
             {
-                children.emplace_back();
-                is_array.push_back(false);
+                grow();
             }
             return child;
         }
-
-        // return the number of the prefix with number id extended by
-        // reference_token, which must have been added before
-        std::size_t find_child(std::size_t id, const string_t& reference_token) const
-        {
-            const auto it = children[id].find(reference_token);
-            JSON_ASSERT(it != children[id].end());
-            return it->second;
-        }
     };
 
     /*!
     @brief create and return a reference to the pointed to value
 
-    Complexity: Linear in the number of reference tokens (times the logarithm
-    of the number of siblings for the prefix lookup).
+    @param[in] j  the value to create the pointed to value in
+    @param[in] tokens  the reference tokens of the pointer
+    @param[in] prefixes  prefixes[i] is the number in @a tree of the prefix
+                         consisting of the first i reference tokens
+
+    Complexity: Linear in the number of reference tokens.
 
     @throw parse_error.106 if an array index begins with '0'
     @throw parse_error.109 if array index is not a number
     @throw type_error.313 if value cannot be unflattened
     */
     template<typename BasicJsonType>
-    BasicJsonType& get_and_create(BasicJsonType& j, const prefix_tree& tree) const
+    static BasicJsonType& get_and_create(BasicJsonType& j, const std::vector<string_t>& tokens,
+                                         const std::vector<std::size_t>& prefixes, const prefix_tree& tree)
     {
         auto* result = &j;
 
-        // the number of the prefix consumed so far; used to look up whether
-        // the value to be created below is an array or an object
-        std::size_t id = 0;
-
         // in case no reference tokens exist, return a reference to the JSON value
         // j which will be overwritten by a primitive value
-        for (const auto& reference_token : reference_tokens)
+        for (std::size_t i = 0; i < tokens.size(); ++i)
         {
+            const auto& reference_token = tokens[i];
             switch (result->type())
             {
                 case detail::value_t::null:
                 {
-                    if (tree.is_array[id])
+                    if (tree.is_array[prefixes[i]])
                     {
                         // some reference token below this position is 0, so the
                         // value is an array
@@ -478,7 +649,7 @@ class json_pointer
                 case detail::value_t::object:
                 {
                     // create an entry in the object
-                    result = &result->operator[](reference_token);
+                    result = &object_member(*result, reference_token);
                     break;
                 }
 
@@ -505,8 +676,6 @@ class json_pointer
                 default:
                     JSON_THROW(detail::type_error::create(detail::exception_id::unflatten_invalid_value, "invalid value to unflatten", &j));
             }
-
-            id = tree.find_child(id, reference_token);
         }
 
         return *result;
@@ -558,7 +727,7 @@ class json_pointer
                 case detail::value_t::object:
                 {
                     // use unchecked object access
-                    ptr = &ptr->operator[](reference_token);
+                    ptr = &object_member(*ptr, reference_token);
                     break;
                 }
 
@@ -608,8 +777,13 @@ class json_pointer
             {
                 case detail::value_t::object:
                 {
-                    // note: at performs range check
-                    ptr = &ptr->at(reference_token);
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
+                    if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
+                    {
+                        // at() throws out_of_range.403 with the usual message
+                        ptr->at(reference_token);
+                    }
+                    ptr = &it->second;
                     break;
                 }
 
@@ -630,7 +804,7 @@ class json_pointer
                         JSON_THROW(detail::out_of_range::create(detail::exception_id::array_index_out_of_range, detail::concat(
                                 "array index ", std::to_string(idx), " is out of range"), ptr));
                     }
-                    ptr = &ptr->operator[](idx);
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
@@ -678,7 +852,9 @@ class json_pointer
                 {
                     // use unchecked object access; the const operator[]
                     // asserts that the key exists
-                    ptr = &ptr->operator[](reference_token);
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
+                    JSON_ASSERT(it != ptr->m_data.m_value.object->end());
+                    ptr = &it->second;
                     break;
                 }
 
@@ -692,7 +868,9 @@ class json_pointer
 
                     // use unchecked array access; the const operator[]
                     // asserts that the index exists
-                    ptr = &ptr->operator[](array_index<BasicJsonType>(reference_token));
+                    const auto idx = array_index<BasicJsonType>(reference_token);
+                    JSON_ASSERT(idx < ptr->m_data.m_value.array->size());
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
@@ -733,12 +911,12 @@ class json_pointer
             {
                 case detail::value_t::object:
                 {
-                    const auto it = ptr->find(reference_token);
-                    if (JSON_HEDLEY_UNLIKELY(it == ptr->end()))
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
+                    if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
                     {
                         return nullptr;
                     }
-                    ptr = &*it;
+                    ptr = &it->second;
                     break;
                 }
 
@@ -775,7 +953,7 @@ class json_pointer
                     {
                         return nullptr;
                     }
-                    ptr = &ptr->operator[](idx);
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
@@ -808,13 +986,14 @@ class json_pointer
             {
                 case detail::value_t::object:
                 {
-                    if (!ptr->contains(reference_token))
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
+                    if (it == ptr->m_data.m_value.object->end())
                     {
                         // we did not find the key in the object
                         return false;
                     }
 
-                    ptr = &ptr->operator[](reference_token);
+                    ptr = &it->second;
                     break;
                 }
 
@@ -836,13 +1015,13 @@ class json_pointer
                         return false;
                     }
 
-                    if (idx >= ptr->size())
+                    if (idx >= ptr->m_data.m_value.array->size())
                     {
                         // index out of range
                         return false;
                     }
 
-                    ptr = &ptr->operator[](idx);
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
@@ -892,6 +1071,15 @@ class json_pointer
             JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_missing_slash, 1, detail::concat("JSON pointer must be empty or begin with '/' - was: '", reference_string, "'"), nullptr));
         }
 
+        // every slash starts a reference token
+        const std::size_t size = reference_string.size();
+        std::size_t token_count = 0;
+        for (std::size_t i = 0; i < size; ++i)
+        {
+            token_count += (reference_string[i] == '/') ? 1 : 0;
+        }
+        result.reserve(token_count);
+
         // extract the reference tokens:
         // - slash: position of the last read slash (or end of string)
         // - start: position after the previous slash
@@ -913,31 +1101,52 @@ class json_pointer
             const auto count = (slash == string_t::npos ? reference_string.size() : slash) - start;
             auto reference_token = string_t(reference_string.data() + start, count);
 
-            // check reference tokens are properly escaped
-            for (std::size_t pos = reference_token.find_first_of('~');
-                    pos != string_t::npos;
-                    pos = reference_token.find_first_of('~', pos + 1))
+            // check reference tokens are properly escaped; a token without
+            // a '~' (the common case) needs neither check nor unescaping
+            std::size_t pos = reference_token.find_first_of('~');
+            if (pos != string_t::npos)
             {
-                JSON_ASSERT(reference_token[pos] == '~');
-
-                // ~ must be followed by 0 or 1
-                if (JSON_HEDLEY_UNLIKELY(pos == reference_token.size() - 1 ||
-                                         (reference_token[pos + 1] != '0' &&
-                                          reference_token[pos + 1] != '1')))
+                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
                 {
-                    JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_invalid_escape, 0, "escape character '~' must be followed with '0' or '1'", nullptr));
+                    JSON_ASSERT(reference_token[pos] == '~');
+
+                    // ~ must be followed by 0 or 1
+                    if (JSON_HEDLEY_UNLIKELY(pos == reference_token.size() - 1 ||
+                                             (reference_token[pos + 1] != '0' &&
+                                              reference_token[pos + 1] != '1')))
+                    {
+                        JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_invalid_escape, 0, "escape character '~' must be followed with '0' or '1'", nullptr));
+                    }
                 }
+
+                detail::unescape(reference_token);
             }
 
             // finally, store the reference token
-            detail::unescape(reference_token);
-            result.push_back(reference_token);
+            result.push_back(std::move(reference_token));
         }
 
         return result;
     }
 
   private:
+    // Add the entry (key, v) to the object `obj`. Flattened paths are pairwise
+    // distinct, so objects that can append a key known to be absent
+    // (ordered_map) skip the lookup, which is linear and would make flatten
+    // quadratic; all other object types use operator[].
+    template<typename ObjectType, typename ValueType>
+    static auto add_flattened(ObjectType& obj, const string_t& key, ValueType&& v, int /*unused*/)
+    -> decltype(obj.emplace_new_key(key, std::forward<ValueType>(v)), void())
+    {
+        obj.emplace_new_key(key, std::forward<ValueType>(v));
+    }
+
+    template<typename ObjectType, typename ValueType>
+    static void add_flattened(ObjectType& obj, const string_t& key, ValueType&& v, ...)
+    {
+        obj[key] = std::forward<ValueType>(v);
+    }
+
     /*!
     @param[in] reference_string  the reference string to the current value
     @param[in] value             the value to consider
@@ -982,7 +1191,8 @@ class json_pointer
 
         // flatten `v`, whose path is `path`: primitives and empty containers
         // are added to the result right away; other containers get a frame
-        const auto enter = [&stack, &path, &result](const BasicJsonType & v)
+        auto& object = *result.m_data.m_value.object;
+        const auto enter = [&stack, &path, &object](const BasicJsonType & v)
         {
             switch (v.type())
             {
@@ -991,7 +1201,7 @@ class json_pointer
                     if (v.m_data.m_value.array->empty())
                     {
                         // flatten empty array as null
-                        result[path] = nullptr;
+                        add_flattened(object, path, nullptr, 0);
                     }
                     else
                     {
@@ -1005,7 +1215,7 @@ class json_pointer
                     if (v.m_data.m_value.object->empty())
                     {
                         // flatten empty object as null
-                        result[path] = nullptr;
+                        add_flattened(object, path, nullptr, 0);
                     }
                     else
                     {
@@ -1025,7 +1235,7 @@ class json_pointer
                 default:
                 {
                     // add a primitive value with its reference string
-                    result[path] = v;
+                    add_flattened(object, path, v, 0);
                     return;
                 }
             }
@@ -1071,6 +1281,10 @@ class json_pointer
                 enter(it->second);
             }
         }
+
+        // the entries were added without going through basic_json, so the
+        // parent pointers (JSON_DIAGNOSTICS) have to be set now
+        result.set_parents();
     }
 
     /*!
@@ -1094,21 +1308,53 @@ class json_pointer
 
         BasicJsonType result;
 
+        // parse every key once; the reference tokens are kept for the second
+        // pass (the vectors are never resized, so the tree may refer to them)
+        std::vector<std::vector<string_t>> keys;
+        keys.reserve(value.m_data.m_value.object->size());
+        for (const auto& element : *value.m_data.m_value.object)
+        {
+            json_pointer ptr(element.first);
+            keys.push_back(std::move(ptr.reference_tokens));
+        }
+
         // collect the pointer prefixes that have a reference token 0 among
         // their children; the values below them are arrays, all others are
         // objects (see prefix_tree)
         prefix_tree tree;
-        for (const auto& element : *value.m_data.m_value.object)
+        // the number of the prefix with i + 1 tokens of the previous key; keys
+        // are often sorted, so most of the prefix is the same as for the
+        // previous key and need not be looked up again
+        std::vector<std::size_t> path;
+        const std::vector<string_t>* previous = nullptr;
+        // the number of the complete prefix (i.e., the key) of each key
+        std::vector<std::size_t> leaves;
+        leaves.reserve(keys.size());
+        for (const auto& tokens : keys)
         {
-            json_pointer ptr(element.first);
-            std::size_t id = 0;
-            for (auto& reference_token : ptr.reference_tokens)
+            std::size_t common = 0;
+            if (previous != nullptr)
             {
-                id = tree.add_child(id, std::move(reference_token));
+                const std::size_t limit = (std::min)(tokens.size(), previous->size());
+                while (common < limit && tokens[common] == (*previous)[common])
+                {
+                    ++common;
+                }
             }
+            path.resize(common);
+            std::size_t id = common == 0 ? 0 : path[common - 1];
+            for (std::size_t i = common; i < tokens.size(); ++i)
+            {
+                id = tree.add_child(id, tokens[i]);
+                path.push_back(id);
+            }
+            leaves.push_back(id);
+            previous = &tokens;
         }
 
         // iterate the JSON object values
+        std::vector<std::size_t> prefixes;
+        std::size_t key_index = 0;
         for (const auto& element : *value.m_data.m_value.object)
         {
             if (JSON_HEDLEY_UNLIKELY(!element.second.is_primitive()))
@@ -1116,11 +1362,21 @@ class json_pointer
                 JSON_THROW(detail::type_error::create(detail::exception_id::unflatten_value_not_primitive, "values in object must be primitive", &element.second));
             }
 
+            const auto& tokens = keys[key_index];
+            // number the prefixes of this key from its last one to the first
+            prefixes.resize(tokens.size());
+            for (std::size_t i = tokens.size(), id = leaves[key_index]; i > 0; --i)
+            {
+                id = tree.parent[id];
+                prefixes[i - 1] = id;
+            }
+            ++key_index;
+
             // Assign the value to the reference pointed to by JSON pointer. Note
             // that if the JSON pointer is "" (i.e., points to the whole value),
             // function get_and_create returns a reference to the result itself.
             // An assignment will then create a primitive value.
-            json_pointer(element.first).get_and_create(result, tree) = element.second;
+            get_and_create(result, tokens, prefixes, tree) = element.second;
         }
 
         return result;
@@ -1159,7 +1415,7 @@ class json_pointer
         = delete;
 #else
     {
-        return *this == json_pointer(rhs);
+        return equals_string(rhs);
     }
 #endif
 
@@ -1243,7 +1499,7 @@ inline bool operator==(const json_pointer<RefStringTypeLhs>& lhs,
     = delete;
 #else
 {
-    return lhs == json_pointer<RefStringTypeLhs>(rhs);
+    return lhs.equals_string(rhs);
 }
 #endif
 
@@ -1256,7 +1512,7 @@ inline bool operator==(const StringType& lhs,
     = delete;
 #else
 {
-    return json_pointer<RefStringTypeRhs>(lhs) == rhs;
+    return rhs.equals_string(lhs);
 }
 #endif
 

include/nlohmann/detail/string_escape.hpp

@@ -9,6 +9,7 @@
 #pragma once
 
 #include <cstddef> // size_t
+#include <utility> // move
 
 #include <nlohmann/detail/abi_macros.hpp>
 
@@ -24,39 +25,46 @@ namespace detail
  * Note the order of escaping "~" to "~0" and "/" to "~1" is important.
  *
  * The string is rebuilt in a single pass, appending whole runs between the
- * characters that need escaping. Scanning with find_first_of() keeps the
- * common case -- nothing to escape -- as fast as a single search, while
- * repeated replace() calls would move the tail of the string once per
- * escaped character.
+ * characters that need escaping, while repeated replace() calls would move
+ * the tail of the string once per escaped character. @ref escape_into appends
+ * the escaped string to an existing one.
  */
+template<typename StringType>
+inline void escape_into(StringType& out, const StringType& s)
+{
+    const std::size_t size = s.size();
+    std::size_t run = 0;
+    for (std::size_t i = 0; i < size; ++i)
+    {
+        const char c = s[i];
+        if (c == '~' || c == '/')
+        {
+            out.append(s.data() + run, i - run);
+            out.append(c == '~' ? "~0" : "~1", 2);
+            run = i + 1;
+        }
+    }
+    out.append(s.data() + run, size - run);
+}
+
 template<typename StringType>
 inline StringType escape(const StringType& s)
 {
-    auto next_special = [&s](std::size_t from)
+    // a single scan for both special characters; nothing to escape is the common case
+    const std::size_t size = s.size();
+    std::size_t i = 0;
+    while (i < size && s[i] != '~' && s[i] != '/')
     {
-        const auto tilde = s.find_first_of('~', from);
-        const auto slash = s.find_first_of('/', from);
-        return tilde < slash ? tilde : slash; // npos is the largest value
-    };
-
-    auto pos = next_special(0);
-    if (pos == StringType::npos)
+        ++i;
+    }
+    if (i == size)
     {
         return s;
     }
 
     StringType result;
-    result.reserve(s.size() + 2);
-
-    std::size_t run = 0;
-    while (pos != StringType::npos)
-    {
-        result.append(s.data() + run, pos - run);
-        result.append(s[pos] == '~' ? "~0" : "~1", 2);
-        run = pos + 1;
-        pos = next_special(run);
-    }
-    result.append(s.data() + run, s.size() - run);
+    result.reserve(size + 2);
+    escape_into(result, s);
     return result;
 }
 
@@ -101,7 +109,7 @@ inline void unescape(StringType& s)
         pos = s.find_first_of('~', run);
     }
     result.append(s.data() + run, s.size() - run);
-    s = result;
+    s = std::move(result);
 }
 
 }  // namespace detail

include/nlohmann/ordered_map.hpp

@@ -92,13 +92,19 @@ private:
 
     /// @brief shared implementation of the const and non-const at() overloads
     /// @throw std::out_of_range if @a key is not found
+    JSON_HEDLEY_NO_RETURN JSON_HEDLEY_NEVER_INLINE
+    static void throw_key_not_found()
+    {
+        JSON_THROW(std::out_of_range("key not found"));
+    }
+
     template<typename Self, typename KeyType>
     static auto at_impl(Self& self, const KeyType& key) -> decltype((self.begin()->second))
     {
         const auto it = find_impl(self, key);
         if (it == self.end())
         {
-            JSON_THROW(std::out_of_range("key not found"));
+            throw_key_not_found();
         }
         return it->second;
     }
@@ -143,6 +149,17 @@ public:
         return {std::prev(this->end()), true};
     }
 
+    /// @brief add an element at the end without checking whether the key exists
+    /// @note The caller must guarantee that @a key is not yet contained, otherwise
+    ///       the container holds duplicate keys. This avoids the linear search of
+    ///       emplace() and operator[] when many keys known to be new are added.
+    template<class V, detail::enable_if_t<
+                 detail::is_constructible<T, V>::value, int> = 0>
+    void emplace_new_key(const key_type& key, V && t)
+    {
+        append(key, std::forward<V>(t));
+    }
+
     T& operator[](const key_type& key)
     {
         return emplace(key, T{}).first->second;

tests/benchmarks/src/benchmarks.cpp

@@ -9,7 +9,10 @@
 #include <benchmark/benchmark.h>
 #include <nlohmann/json.hpp>
 #include <fstream>
+#include <array>
+#include <deque>
 #include <numeric>
+#include <set>
 #include <vector>
 #include <test_data.hpp>
 
@@ -621,4 +624,27 @@ static void FromCborChunkedString(benchmark::State& state, const std::size_t chu
 
 BENCHMARK_CAPTURE(FromCborChunkedString, 10000 chunks, 10000);
 
+// conversion of a JSON array into standard containers
+template<typename Container>
+static void GetArray(benchmark::State& state)
+{
+    json j = json::array();
+    for (int i = 0; i < 1000; ++i)
+    {
+        j.push_back(i);
+    }
+
+    for (auto _ : state)
+    {
+        auto c = j.get<Container>();
+        benchmark::DoNotOptimize(c);
+    }
+    state.SetItemsProcessed(state.iterations() * 1000);
+}
+
+BENCHMARK_TEMPLATE(GetArray, std::vector<int>);
+BENCHMARK_TEMPLATE(GetArray, std::deque<int>);
+BENCHMARK_TEMPLATE(GetArray, std::set<int>);
+BENCHMARK_TEMPLATE(GetArray, std::array<int, 1000>);
+
 BENCHMARK_MAIN();

tests/src/unit-algorithms.cpp

@@ -9,6 +9,7 @@
 #include "doctest_compatibility.h"
 
 #include <algorithm>
+#include <numeric>
 #include <nlohmann/json.hpp>
 using nlohmann::json;
 

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 +187,193 @@ 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);
+    }
+}
+
+TEST_CASE_TEMPLATE("destructor of nested containers", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    SECTION("empty containers")
+    {
+        {
+            Json j = Json::array();
+        }
+        {
+            Json j = Json::object();
+        }
+        {
+            Json j = Json::array({Json::array(), Json::object(), Json::array({Json::array()})});
+        }
+    }
+
+    SECTION("leaves mixed with nested containers")
+    {
+        Json j = Json::object();
+        j["leaf"] = 1;
+        j["str"] = "text";
+        j["empty_a"] = Json::array();
+        j["empty_o"] = Json::object();
+        j["arr"] = Json::array({1, "a", Json::array({1, 2}), Json::object({{"k", Json::array({Json::object()})}}), nullptr});
+        j["obj"] = Json::object({{"x", Json::object({{"y", Json::array({1, Json::object({{"z", 1}})})}})}, {"w", 2}});
+        Json copy = j;
+        CHECK(copy == j);
+        j = Json::array({std::move(copy), j, 3, Json::array()});
+    }
+
+    SECTION("deep nesting")
+    {
+        for (int kind = 0; kind < 3; ++kind)
+        {
+            Json j = 1;
+            for (int i = 0; i < 5000; ++i)
+            {
+                switch (kind)
+                {
+                    case 0:
+                        j = Json::array({std::move(j)});
+                        break;
+                    case 1:
+                        j = Json::object({{"k", std::move(j)}});
+                        break;
+                    default:
+                        // siblings next to the nested child
+                        j = (i % 2 == 0) ? Json::array({1, Json::array(), std::move(j)}) : Json::object({{"a", 1}, {"b", Json::object()}, {"c", std::move(j)}});
+                        break;
+                }
+            }
+        }
+    }
+}

tests/src/unit-diagnostics.cpp

@@ -13,9 +13,11 @@
 #endif
 
 #define JSON_DIAGNOSTICS 1
+#define JSON_TESTS_PRIVATE
 
 #include <nlohmann/json.hpp>
 using nlohmann::json;
+using nlohmann::ordered_json;
 
 #include <map>
 #include <unordered_map>
@@ -631,3 +633,175 @@ TEST_CASE("Better diagnostics past the descent bound of update(), merge_patch()
         CHECK_THROWS_WITH_AS(p->at("y").at(0), expected_y.c_str(), json::type_error);
     }
 }
+
+namespace
+{
+// check that every child's parent pointer points to the containing value
+template<typename BasicJsonType>
+bool parents_consistent(const BasicJsonType& j)
+{
+    bool ok = true;
+    if (j.is_array() || j.is_object())
+    {
+        for (const auto& child : j)
+        {
+            ok = ok && child.m_parent == &j && parents_consistent(child);
+        }
+    }
+    return ok;
+}
+} // namespace
+
+TEST_CASE_TEMPLATE("Parent pointers stay consistent", J, json, ordered_json)
+{
+    SECTION("parse, copy, move, assign, swap")
+    {
+        J j = J::parse(R"({"a":[1,{"b":[2,3]},[4]],"c":{"d":{"e":[]}}})");
+        CHECK(parents_consistent(j));
+
+        J copy(j);
+        CHECK(parents_consistent(copy));
+
+        J moved(std::move(copy));
+        CHECK(parents_consistent(moved));
+
+        J assigned;
+        assigned = j;
+        CHECK(parents_consistent(assigned));
+
+        J other = {1, 2, {{"k", J::array({1, 2})}}};
+        other.swap(assigned);
+        CHECK(parents_consistent(other));
+        CHECK(parents_consistent(assigned));
+    }
+
+    SECTION("push_back growth")
+    {
+        J arr = J::array();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr.push_back(J::array({i}));
+        }
+        CHECK(parents_consistent(arr));
+
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            obj.push_back({std::to_string(i), J::array({i})});
+        }
+        CHECK(parents_consistent(obj));
+    }
+
+    SECTION("emplace_back and emplace growth")
+    {
+        J arr = J::array();
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr.emplace_back(J::array({i}));
+            obj.emplace(std::to_string(i), J::array({i}));
+        }
+        CHECK(parents_consistent(arr));
+        CHECK(parents_consistent(obj));
+    }
+
+    SECTION("operator[] growth and at()")
+    {
+        J obj = J::object();
+        for (int i = 0; i < 500; ++i)
+        {
+            obj[std::to_string(i)] = J::array({i});
+            CHECK(obj[std::to_string(i / 2)].m_parent == &obj);
+        }
+        CHECK(parents_consistent(obj));
+
+        J arr = J::array();
+        for (int i = 0; i < 500; ++i)
+        {
+            arr[static_cast<size_t>(i)] = J::object({{"x", i}});
+        }
+        for (size_t i = 0; i < 500; ++i)
+        {
+            CHECK(arr.at(i).m_parent == &arr);
+        }
+        CHECK(obj.at("3").m_parent == &obj);
+        CHECK(parents_consistent(arr));
+        CHECK(parents_consistent(obj));
+    }
+}
+
+TEST_CASE_TEMPLATE("update() and merge_patch() keep parent pointers and diagnostic paths", J, json, ordered_json)
+{
+    // enough keys to make the vector behind ordered_json reallocate several times
+    const auto make_object = [](int first, int last)
+    {
+        J o = J::object();
+        for (int i = first; i < last; ++i)
+        {
+            o[std::to_string(i)] = J::object({{"x", J::array({i})}});
+        }
+        return o;
+    };
+
+    // const access, so operator[] cannot repair the parent pointers
+    const auto check_path = [](const J & j, const std::string & key)
+    {
+        const std::string expected = "[json.exception.type_error.304] (/" + key + "/x) cannot use at() with array";
+        CHECK_THROWS_WITH_AS(j[key]["x"].at("y"), expected.c_str(), json::type_error);
+    };
+
+    SECTION("update(const_reference)")
+    {
+        J j = make_object(0, 200);
+        j.update(make_object(100, 400));
+        CHECK(j.size() == 400);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "150");
+        check_path(j, "399");
+    }
+
+    SECTION("update(const_reference, merge_objects)")
+    {
+        J j = make_object(0, 200);
+        j["50"]["y"] = 1;
+        J other = make_object(100, 400);
+        other["50"] = J::object({{"z", J::array({1})}});
+        j.update(other, true);
+        CHECK(j.size() == 400);
+        CHECK(j["50"].size() == 3);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "399");
+    }
+
+    SECTION("update(first, last)")
+    {
+        J j = make_object(0, 200);
+        J other = make_object(100, 400);
+        j.update(other.begin(), other.end());
+        CHECK(j.size() == 400);
+        CHECK(parents_consistent(j));
+        check_path(j, "0");
+        check_path(j, "399");
+
+        J k = make_object(0, 200);
+        k.update(other.begin(), other.end(), true);
+        CHECK(k.size() == 400);
+        CHECK(parents_consistent(k));
+        check_path(k, "399");
+    }
+
+    SECTION("merge_patch()")
+    {
+        J j = make_object(0, 200);
+        J patch = make_object(100, 400);
+        patch["0"] = nullptr;
+        j.merge_patch(patch);
+        CHECK(j.size() == 399);
+        CHECK_FALSE(j.contains("0"));
+        CHECK(parents_consistent(j));
+        check_path(j, "1");
+        check_path(j, "399");
+    }
+}

tests/src/unit-iterators1.cpp

@@ -10,6 +10,8 @@
 
 #define JSON_TESTS_PRIVATE
 #include <nlohmann/json.hpp>
+#include <string>
+#include <vector>
 using nlohmann::json;
 
 TEST_CASE("iterators 1")
@@ -1628,3 +1630,108 @@ TEST_CASE("iterators 1")
         }
     }
 }
+
+TEST_CASE_TEMPLATE("reverse iteration and member access", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    SECTION("array: -- and rbegin/rend")
+    {
+        Json j = {1, 2, 3, 4};
+
+        auto it = j.end();
+        std::vector<int> seen;
+        while (it != j.begin())
+        {
+            --it;
+            seen.push_back(it->template get<int>());
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+
+        auto it2 = j.end();
+        auto old = it2--;
+        CHECK(old == j.end());
+        CHECK(*it2 == 4);
+
+        seen.clear();
+        for (auto r = j.rbegin(); r != j.rend(); ++r)
+        {
+            seen.push_back(r->template get<int>());
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+
+        seen.clear();
+        for (auto r = j.crbegin(); r != j.crend(); ++r)
+        {
+            seen.push_back(*r);
+        }
+        CHECK(seen == std::vector<int> {4, 3, 2, 1});
+    }
+
+    SECTION("object: -- and rbegin/rend")
+    {
+        Json j = {{"a", 1}, {"b", 2}, {"c", 3}};
+        // order of keys depends on the object type; use the forward order as reference
+        std::vector<std::string> fwd;
+        for (auto it = j.begin(); it != j.end(); ++it)
+        {
+            fwd.push_back(it.key());
+        }
+        std::vector<std::string> rev(fwd.rbegin(), fwd.rend());
+
+        std::vector<std::string> seen;
+        auto it = j.end();
+        while (it != j.begin())
+        {
+            --it;
+            seen.push_back(it.key());
+        }
+        CHECK(seen == rev);
+
+        auto it2 = j.end();
+        auto old = it2--;
+        CHECK(old == j.end());
+        CHECK(it2.key() == rev.front());
+
+        seen.clear();
+        for (auto r = j.rbegin(); r != j.rend(); ++r)
+        {
+            seen.push_back(r.key());
+        }
+        CHECK(seen == rev);
+    }
+
+    SECTION("object: operator-> reads and writes the mapped value")
+    {
+        Json j = {{"a", 1}, {"s", "str"}};
+        auto it = j.find("s");
+        REQUIRE(it != j.end());
+        CHECK(it->is_string());
+        CHECK(it->size() == 1);
+        *it = Json::array({1, 2});
+        CHECK(it->is_array());
+        it->push_back(3);
+        CHECK(j["s"].size() == 3);
+
+        const Json& cj = j;
+        auto cit = cj.find("a");
+        REQUIRE(cit != cj.end());
+        CHECK(cit->is_number_integer());
+        CHECK(cit->template get<int>() == 1);
+    }
+
+    SECTION("primitives: -- and ->")
+    {
+        Json j = 42;
+        auto it = j.end();
+        --it;
+        CHECK(it == j.begin());
+        CHECK(it->template get<int>() == 42);
+    }
+
+    SECTION("comparing iterators of different containers throws")
+    {
+        Json a = {1, 2};
+        Json b = {1, 2};
+        CHECK_THROWS_WITH_AS(a.begin() == b.begin(), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers", nlohmann::json::invalid_iterator);
+        CHECK_THROWS_WITH_AS(a.begin() < b.begin(), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers", nlohmann::json::invalid_iterator);
+    }
+}

tests/src/unit-json_patch.cpp

@@ -2066,3 +2066,40 @@ TEST_CASE("JSON patch - every operation on ordered_json")
         }
     }
 }
+
+TEST_CASE_TEMPLATE("JSON patch - diff round trip on hand-made documents", Json, nlohmann::json, nlohmann::ordered_json)
+{
+    const auto roundtrip = [](const Json & a, const Json & b)
+    {
+        CAPTURE(a.dump())
+        CAPTURE(b.dump())
+        const Json p = Json::diff(a, b);
+        CHECK(a.patch(p) == b);
+        CHECK(Json::diff(b, b).empty());
+    };
+
+    SECTION("reordered keys")
+    {
+        const Json a = Json::parse(R"({"x":1,"y":2,"z":{"p":1,"q":2}})");
+        const Json b = Json::parse(R"({"z":{"q":2,"p":1},"y":2,"x":1})");
+        roundtrip(a, b);
+        roundtrip(b, a);
+    }
+
+    SECTION("nested add, remove and replace")
+    {
+        const Json a = Json::parse(R"({"a":[1,2,{"k":"v","gone":true}],"b":{"c":{"d":1}},"e":"s"})");
+        const Json b = Json::parse(R"({"a":[1,3,{"k":"w","new":[]}],"b":{"c":{"d":[1,2]}},"f":null})");
+        roundtrip(a, b);
+        roundtrip(b, a);
+    }
+
+    SECTION("arrays of different length and type changes")
+    {
+        roundtrip(Json::parse("[1,2,3,4,5]"), Json::parse("[1,3]"));
+        roundtrip(Json::parse("[1]"), Json::parse("[1,[2],{\"a\":3}]"));
+        roundtrip(Json::parse("[]"), Json::parse("{}"));
+        roundtrip(Json::parse("{\"a\":[]}"), Json::parse("{\"a\":{}}"));
+        roundtrip(Json(1), Json::parse("[1]"));
+    }
+}

tests/src/unit-json_pointer.cpp

@@ -23,8 +23,14 @@ using nlohmann::json;
     using namespace nlohmann::literals; // NOLINT(google-build-using-namespace)
 #endif
 
+#include <cerrno>
+#include <cstdlib>
+#include <limits>
 #include <map>
+#include <random>
 #include <sstream>
+#include <string>
+#include <vector>
 
 TEST_CASE("JSON pointers")
 {
@@ -1071,3 +1077,242 @@ TEST_CASE("flatten of structured values")
               R"({"/z":"s","/a~1b/~0":1,"/a~1b/k":null,"/a/0":1,"/a/1/c":null,"/a/1/b":null,"/a/2/0/x~0~1/0":true,"/a/2/0/x~0~1/1":null,"/a/2/0/w":2})");
     }
 }
+
+TEST_CASE("flatten of ordered_json")
+{
+    using nlohmann::ordered_json;
+
+    SECTION("keeps the document order of the leaves")
+    {
+        const ordered_json j = ordered_json::parse(R"({
+            "z": 1, "a": {"y": [true, null, {}], "b": "x", "": []},
+            "m~n": 2.5, "o/p": {"k": {}}, "0": [[1], [2, 3]]
+        })");
+
+        const ordered_json flat = j.flatten();
+        std::vector<std::string> keys;
+        for (const auto& el : flat.items())
+        {
+            keys.push_back(el.key());
+        }
+        const std::vector<std::string> expected =
+        {
+            "/z", "/a/y/0", "/a/y/1", "/a/y/2", "/a/b", "/a/", "/m~0n", "/o~1p/k", "/0/0/0", "/0/1/0", "/0/1/1"
+        };
+        CHECK(keys == expected);
+        CHECK(flat.size() == expected.size());
+        CHECK(flat["/m~0n"] == 2.5);
+        CHECK(flat["/a/y/2"].is_null());
+    }
+
+    SECTION("round-trips and agrees with json")
+    {
+        ordered_json j;
+        for (int i = 0; i < 200; ++i)
+        {
+            j["k" + std::to_string(i)] = {{"v", i}, {"l", {i, i + 1, "s"}}};
+        }
+        const ordered_json flat = j.flatten();
+        CHECK(flat.size() == 800);
+        CHECK(flat.unflatten() == j);
+
+        // same entries as for a sorted object
+        const nlohmann::json plain = nlohmann::json::parse(j.dump());
+        CHECK(nlohmann::json::parse(flat.dump()) == plain.flatten());
+    }
+
+    SECTION("primitive and empty root")
+    {
+        CHECK(ordered_json(1).flatten() == ordered_json({{"", 1}}));
+        CHECK(ordered_json::array().flatten() == ordered_json({{"", nullptr}}));
+    }
+}
+namespace
+{
+// reference implementation of the array index rules (RFC 6901, Sect. 4) based on strtoull
+enum class index_class
+{
+    ok,
+    leading_zero,
+    not_a_number,
+    unresolved,
+    exceeds_size_type
+};
+
+index_class classify_index(const std::string& s, std::size_t& idx)
+{
+    if (s.size() > 1 && s[0] == '0')
+    {
+        return index_class::leading_zero;
+    }
+    if (!s.empty() && s != "-" && !(s[0] >= '0' && s[0] <= '9'))
+    {
+        return index_class::not_a_number;
+    }
+    char* end = nullptr;
+    errno = 0;
+    const unsigned long long res = std::strtoull(s.c_str(), &end, 10); // NOLINT(runtime/int)
+    if (end == s.c_str() || errno == ERANGE || static_cast<std::size_t>(end - s.c_str()) != s.size())
+    {
+        return index_class::unresolved;
+    }
+    if (res >= static_cast<unsigned long long>((std::numeric_limits<std::size_t>::max)())) // NOLINT(runtime/int)
+    {
+        return index_class::exceeds_size_type;
+    }
+    idx = static_cast<std::size_t>(res);
+    return index_class::ok;
+}
+
+// the observable behavior of json_pointer::at() and contains() for an array index token
+void check_index_token(const std::string& token)
+{
+    const json array = {10, 20, 30};
+    const json::json_pointer ptr("/" + token);
+
+    std::size_t idx = 0;
+    const auto cls = classify_index(token, idx);
+    const bool exists = cls == index_class::ok && idx < array.size();
+
+    CAPTURE(token)
+    CHECK(array.contains(ptr) == (exists && token != "-"));
+
+    if (token == "-")
+    {
+        CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+        return;
+    }
+
+    switch (cls)
+    {
+        case index_class::leading_zero:
+            CHECK_THROWS_AS(array.at(ptr), json::parse_error&);
+            break;
+        case index_class::not_a_number:
+            CHECK_THROWS_AS(array.at(ptr), json::parse_error&);
+            break;
+        case index_class::unresolved:
+        case index_class::exceeds_size_type:
+            CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+            break;
+        case index_class::ok:
+        default:
+            if (exists)
+            {
+                CHECK(array.at(ptr) == array[idx]);
+            }
+            else
+            {
+                CHECK_THROWS_AS(array.at(ptr), json::out_of_range&);
+            }
+            break;
+    }
+}
+} // namespace
+
+TEST_CASE("JSON pointer array index tokens match strtoull semantics")
+{
+    SECTION("boundary values")
+    {
+        const std::string max_digits = std::to_string((std::numeric_limits<std::size_t>::max)());
+        const std::vector<std::string> tokens =
+        {
+            "", "-", "0", "1", "2", "3", "00", "01", "007", "+1", "-1", " 1", "1 ", "1a", "a1", "0x1", "1e1", "1.0",
+            "999999999", "1000000000", "99999999999999999", "999999999999999999", "1000000000000000000",
+            "9999999999999999999", "18446744073709551614", max_digits, "18446744073709551616",
+            "99999999999999999999999", "000000000000000000000"
+        };
+        for (const auto& token : tokens)
+        {
+            check_index_token(token);
+        }
+    }
+
+    SECTION("random tokens")
+    {
+        std::mt19937 gen(42);
+        const std::string alphabet = "0123456789012345678901234567890123456789-+ ex";
+        for (int i = 0; i < 5000; ++i)
+        {
+            std::string token(static_cast<std::size_t>(gen() % 23), '0');
+            for (auto& c : token)
+            {
+                c = alphabet[gen() % alphabet.size()];
+            }
+            check_index_token(token);
+        }
+    }
+}
+
+TEST_CASE("JSON pointer operations on pointers with escaped tokens")
+{
+    const std::vector<std::string> strings =
+    {
+        "", "/", "//", "/a", "/a/b", "/a~1b", "/~0", "/~01", "/~10", "/a~0~1/b/", "/0/1/2", "/ ", "/a/", "/~1/~0"
+    };
+
+    for (const auto& str : strings)
+    {
+        CAPTURE(str)
+        const json::json_pointer ptr(str);
+        CHECK(ptr.to_string() == str);
+        CHECK(ptr / "x" / "y" / std::size_t{3} == json::json_pointer(str + "/x/y/3"));
+        CHECK(ptr.parent_pointer() == (str.empty() ? ptr : json::json_pointer(str.substr(0, str.rfind('/')))));
+
+        // comparison with a string, in place, and with every other pointer
+#if !defined(JSON_TEST_DEPRECATED_FUNCTIONS_DELETED)
+        for (const auto& other : strings)
+        {
+            CAPTURE(other)
+            CHECK((ptr == other) == (str == other));
+            CHECK((other == ptr) == (str == other));
+            CHECK((ptr != other) == (str != other));
+        }
+        CHECK_THROWS_AS(ptr == "x", json::parse_error&);
+        CHECK_THROWS_AS(ptr == "/~", json::parse_error&);
+        CHECK_THROWS_AS(ptr == "/b/~2", json::parse_error&);
+#endif
+    }
+}
+
+TEST_CASE("JSON pointer lookups")
+{
+    json j = {{"a", {{"b", {1, 2, 3}}}}, {"", 7}, {"~/", {{"x", nullptr}}}};
+    const json& cj = j;
+
+    CHECK(cj[json::json_pointer("/a/b/2")] == 3);
+    CHECK(cj.at(json::json_pointer("/a/b/1")) == 2);
+    CHECK(cj.at(json::json_pointer("/")) == 7);
+    CHECK(cj.contains(json::json_pointer("/~0~1/x")));
+    CHECK(!cj.contains(json::json_pointer("/a/c")));
+    CHECK(cj.value(json::json_pointer("/a/b/0"), 5) == 1);
+    CHECK(cj.value(json::json_pointer("/a/b/3"), 5) == 5);
+    CHECK(cj.value(json::json_pointer("/a/zz"), 5) == 5);
+
+    CHECK_THROWS_AS(cj.at(json::json_pointer("/a/c")), json::out_of_range&);
+    CHECK_THROWS_AS(cj.at(json::json_pointer("/a/b/3")), json::out_of_range&);
+
+    // non-const access reuses existing entries and creates missing ones
+    j[json::json_pointer("/a/b/1")] = 20;
+    j[json::json_pointer("/n/m")] = 1;
+    CHECK(j["a"]["b"] == json({1, 20, 3}));
+    CHECK(j["n"]["m"] == 1);
+}
+
+TEST_CASE("JSON patch with moves and copies")
+{
+    json doc = {{"a", {{"b", {1, 2, 3}}}}, {"c", {{"d", "x"}}}};
+
+    CHECK(doc.patch(json::parse(R"([{"op":"move","from":"/a/b","path":"/c/e"}])")) ==
+          json::parse(R"({"a":{},"c":{"d":"x","e":[1,2,3]}})"));
+    CHECK(doc.patch(json::parse(R"([{"op":"copy","from":"/a/b","path":"/c/e"}])")) ==
+          json::parse(R"({"a":{"b":[1,2,3]},"c":{"d":"x","e":[1,2,3]}})"));
+    CHECK(doc.patch(json::parse(R"([{"op":"move","from":"/a/b/0","path":"/a/b/2"}])")) ==
+          json::parse(R"({"a":{"b":[2,3,1]},"c":{"d":"x"}})"));
+
+    // a failing move leaves a copy-patched document untouched
+    CHECK_THROWS_AS(doc.patch(json::parse(R"([{"op":"move","from":"","path":""}])")), json::out_of_range&);
+    json inplace = doc;
+    CHECK_THROWS_AS(inplace.patch_inplace(json::parse(R"([{"op":"move","from":"","path":""}])")), json::out_of_range&);
+    CHECK(inplace == doc);
+}

tests/src/unit-throw_messages.cpp

@@ -0,0 +1,94 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+using nlohmann::json;
+
+// The exceptions in hot core functions are constructed out of line. These tests pin the
+// exact what() texts of the sites that moved, so the refactoring cannot change them.
+
+#define CHECK_MSG(expr, msg) CHECK_THROWS_WITH(expr, msg)
+
+TEST_CASE("exception messages of out-of-line throw sites")
+{
+    json arr = {1, 2, 3};
+    json arr2 = {4, 5};
+    json obj = {{"a", 1}, {"b", 2}};
+    json obj2 = {{"c", 3}};
+    json num = 1;
+
+    SECTION("iterators")
+    {
+        CHECK_MSG((void)(arr.begin() == arr2.begin()), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers");
+        CHECK_MSG((void)(arr.begin() < arr2.begin()), "[json.exception.invalid_iterator.212] cannot compare iterators of different containers");
+        CHECK_MSG((void)(obj.begin() < obj.end()), "[json.exception.invalid_iterator.213] cannot compare order of object iterators");
+        CHECK_MSG(obj.begin() += 1, "[json.exception.invalid_iterator.209] cannot use offsets with object iterators");
+        CHECK_MSG((void)(obj.begin() - 1), "[json.exception.invalid_iterator.209] cannot use offsets with object iterators");
+        CHECK_MSG((void)obj.begin()[0], "[json.exception.invalid_iterator.208] cannot use operator[] for object iterators");
+        CHECK_MSG((void)arr.begin().key(), "[json.exception.invalid_iterator.207] cannot use key() for non-object iterators");
+        json null;
+        CHECK_MSG((void)*null.end(), "[json.exception.invalid_iterator.214] cannot get value");
+    }
+
+    SECTION("constructors")
+    {
+        CHECK_MSG(json(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.201] iterators are not compatible");
+        CHECK_MSG(json(num.begin() + 1, num.end() + 1), "[json.exception.invalid_iterator.204] iterators out of range");
+        CHECK_MSG(json(json::initializer_list_t{1, 2}, false, json::value_t::object), "[json.exception.type_error.301] cannot create object from initializer list");
+    }
+
+    SECTION("get_ref and operator[]")
+    {
+        CHECK_MSG((void)num.get_ref<std::string&>(), "[json.exception.type_error.303] incompatible ReferenceType for get_ref, actual type is number");
+        CHECK_MSG((void)num[0], "[json.exception.type_error.305] cannot use operator[] with a numeric argument with number");
+    }
+
+    SECTION("at")
+    {
+        const json carr = arr;
+        const json cobj = obj;
+        CHECK_MSG((void)arr.at(static_cast<json::size_type>(10)), "[json.exception.out_of_range.401] array index 10 is out of range");
+        CHECK_MSG((void)carr.at(static_cast<json::size_type>(3)), "[json.exception.out_of_range.401] array index 3 is out of range");
+        CHECK_MSG((void)num.at(static_cast<json::size_type>(0)), "[json.exception.type_error.304] cannot use at() with number");
+        CHECK_MSG((void)obj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)cobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)obj.at(std::string("zz")), "[json.exception.out_of_range.403] key 'zz' not found");
+        CHECK_MSG((void)num.at("a"), "[json.exception.type_error.304] cannot use at() with number");
+
+        nlohmann::ordered_json oobj = {{"a", 1}};
+        const nlohmann::ordered_json coobj = oobj;
+        CHECK_MSG((void)oobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)coobj.at("z"), "[json.exception.out_of_range.403] key 'z' not found");
+        CHECK_MSG((void)coobj.at(std::string("zz")), "[json.exception.out_of_range.403] key 'zz' not found");
+    }
+
+    SECTION("erase")
+    {
+        CHECK_MSG(arr.erase(arr2.begin()), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(num.erase(num.end()), "[json.exception.invalid_iterator.205] iterator out of range");
+        CHECK_MSG(arr.erase(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.203] iterators do not fit current value");
+        CHECK_MSG(num.erase(num.begin() + 1, num.end()), "[json.exception.invalid_iterator.204] iterators out of range");
+        CHECK_MSG(arr.erase(static_cast<json::size_type>(10)), "[json.exception.out_of_range.401] array index 10 is out of range");
+    }
+
+    SECTION("insert")
+    {
+        CHECK_MSG(arr.insert(arr2.begin(), 1), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr2.begin(), 2, 1), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr2.begin(), arr2.begin(), arr2.end()), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(arr.insert(arr.begin(), arr.begin(), arr2.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+        CHECK_MSG(arr.insert(arr.begin(), arr.begin(), arr.end()), "[json.exception.invalid_iterator.211] passed iterators may not belong to container");
+        CHECK_MSG(arr.insert(arr.begin(), obj.begin(), obj.end()), "[json.exception.invalid_iterator.202] iterators first and last must point to arrays");
+        CHECK_MSG(arr.insert(arr2.begin(), {1, 2}), "[json.exception.invalid_iterator.202] iterator does not fit current value");
+        CHECK_MSG(obj.insert(obj2.begin(), arr.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+        CHECK_MSG(obj.insert(arr.begin(), arr.end()), "[json.exception.invalid_iterator.202] iterators first and last must point to objects");
+        CHECK_MSG(obj.update(arr.begin(), arr2.end()), "[json.exception.invalid_iterator.210] iterators do not fit");
+    }
+}

tests/src/unit-to_json-rvalue.cpp

@@ -0,0 +1,257 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+using nlohmann::json;
+
+#include <array>
+#include <deque>
+#include <forward_list>
+#include <list>
+#include <map>
+#include <set>
+#include <string>
+#include <unordered_map>
+#include <utility>
+#include <valarray>
+#include <vector>
+
+#if (defined(__cplusplus) && __cplusplus >= 201703L) || (defined(_HAS_CXX17) && _HAS_CXX17 == 1)
+    #if __has_include(<filesystem>)
+        #include <filesystem>
+        #define TO_JSON_RVALUE_HAS_FS 1
+    #endif
+#endif
+
+namespace
+{
+// element type that counts how it is converted to JSON: a conversion from an
+// lvalue copies, a conversion from an rvalue moves
+struct tracked
+{
+    std::string payload;
+    bool moved_from = false;
+
+    explicit tracked(std::string p) : payload(std::move(p)) {}
+    tracked(const tracked&) = default;
+    tracked& operator=(const tracked&) = default;
+    tracked(tracked&& other) noexcept : payload(std::move(other.payload)), moved_from(other.moved_from)
+    {
+        other.moved_from = true;
+    }
+    tracked& operator=(tracked&& other) noexcept
+    {
+        payload = std::move(other.payload);
+        moved_from = other.moved_from;
+        other.moved_from = true;
+        return *this;
+    }
+    ~tracked() = default;
+};
+
+int tracked_copies = 0; // NOLINT(cppcoreguidelines-avoid-non-const-global-variables,misc-use-internal-linkage)
+int tracked_moves = 0;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables,misc-use-internal-linkage)
+
+void to_json(json& j, const tracked& t)
+{
+    ++tracked_copies;
+    j = t.payload;
+}
+
+void to_json(json& j, tracked&& t)
+{
+    ++tracked_moves;
+    j = std::move(t.payload);
+    t.moved_from = true;
+}
+
+// user type with its own rvalue-aware serializer
+struct udt_container
+{
+    std::vector<int> values;
+};
+} // namespace
+
+namespace nlohmann
+{
+template<>
+struct adl_serializer<udt_container>
+{
+    static void to_json(json& j, const udt_container& c)
+    {
+        j = c.values;
+        j.push_back("copied");
+    }
+    static void to_json(json& j, udt_container&& c)
+    {
+        j = std::move(c.values);
+        j.push_back("moved");
+    }
+};
+} // namespace nlohmann
+
+TEST_CASE("to_json from rvalue containers")
+{
+    const std::string long_a(64, 'a');
+    const std::string long_b(64, 'b');
+
+    SECTION("vector<std::string>: elements are moved")
+    {
+        std::vector<std::string> v{long_a, long_b};
+        const json j = std::move(v);
+        CHECK(j == json::array({long_a, long_b}));
+        for (const auto& s : v) // NOLINT(bugprone-use-after-move,hicpp-invalid-access-moved)
+        {
+            CHECK(s.empty()); // long strings leave the source element empty when moved
+        }
+    }
+
+    SECTION("vector<std::string>: lvalues are copied and left untouched")
+    {
+        const std::vector<std::string> v{long_a, long_b};
+        const json j = v;
+        CHECK(j == json::array({long_a, long_b}));
+        CHECK(v[0] == long_a);
+        CHECK(v[1] == long_b);
+
+        std::vector<std::string> v2{long_a, long_b};
+        const json j2 = v2;
+        CHECK(j2 == j);
+        CHECK(v2[0] == long_a);
+        CHECK(v2[1] == long_b);
+    }
+
+    SECTION("tracked elements: rvalue moves, lvalue copies")
+    {
+        tracked_copies = 0;
+        tracked_moves = 0;
+
+        std::vector<tracked> v;
+        v.emplace_back("x");
+        v.emplace_back("y");
+
+        const json copy = v;
+        CHECK(copy == json::array({"x", "y"}));
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 0);
+        CHECK_FALSE(v[0].moved_from);
+
+        const json moved = std::move(v);
+        CHECK(moved == copy);
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 2);
+    }
+
+    SECTION("const rvalue is copied")
+    {
+        tracked_copies = 0;
+        tracked_moves = 0;
+        const std::vector<tracked> v(2, tracked("z"));
+        const json j = std::move(v); // NOLINT(performance-move-const-arg,hicpp-move-const-arg)
+        CHECK(j == json::array({"z", "z"}));
+        CHECK(tracked_copies == 2);
+        CHECK(tracked_moves == 0);
+    }
+
+    SECTION("other sequence containers")
+    {
+        CHECK(json(std::deque<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::list<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::forward_list<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::array<std::string, 2> {{long_a, long_b}}) == json::array({long_a, long_b}));
+        CHECK(json(std::set<std::string> {long_a, long_b}) == json::array({long_a, long_b}));
+        CHECK(json(std::vector<bool> {true, false}) == json::array({true, false}));
+        CHECK(json(std::vector<int> {1, 2, 3}) == json::array({1, 2, 3}));
+        CHECK(json(std::vector<std::vector<std::string>> {{long_a}, {long_b}}) == json::array({json::array({long_a}), json::array({long_b})}));
+        CHECK(json(std::vector<json> {1, "a"}) == json::array({1, "a"}));
+    }
+
+    SECTION("valarray")
+    {
+        std::valarray<int> va(3);
+        va[0] = 1;
+        va[1] = 2;
+        va[2] = 3;
+        CHECK(json(std::move(va)) == json::array({1, 2, 3})); // NOLINT(performance-move-const-arg,hicpp-move-const-arg)
+        CHECK(json(std::valarray<int>(2)) == json::array({0, 0}));
+    }
+
+    SECTION("array_t and json arrays")
+    {
+        json::array_t arr{1, 2};
+        CHECK(json(std::move(arr)) == json::array({1, 2}));
+        json a = json::array({1, 2});
+        const json b = std::move(a);
+        CHECK(b == json::array({1, 2}));
+    }
+
+    SECTION("object-like containers")
+    {
+        std::map<std::string, std::string> m{{"a", long_a}, {"b", long_b}};
+        const json j = std::move(m);
+        CHECK(j == json({{"a", long_a}, {"b", long_b}}));
+        for (const auto& kv : m) // NOLINT(bugprone-use-after-move,hicpp-invalid-access-moved)
+        {
+            CHECK(kv.second.empty()); // values moved, keys (const) copied
+        }
+
+        const std::map<std::string, std::string> cm{{"a", long_a}};
+        const json jc = cm;
+        CHECK(jc["a"] == long_a);
+        CHECK(cm.at("a") == long_a);
+
+        std::unordered_map<std::string, std::string> um{{"a", long_a}};
+        CHECK(json(std::move(um)) == json({{"a", long_a}}));
+
+        std::multimap<std::string, int> mm{{"a", 1}};
+        CHECK(json(std::move(mm)) == json({{"a", 1}}));
+
+        tracked_copies = 0;
+        tracked_moves = 0;
+        std::map<std::string, tracked> mt;
+        mt.emplace("k", tracked("v"));
+        const json jt = std::move(mt);
+        CHECK(jt == json({{"k", "v"}}));
+        CHECK(tracked_moves == 1);
+        CHECK(tracked_copies == 0);
+
+        json::object_t obj{{"a", 1}};
+        CHECK(json(std::move(obj)) == json({{"a", 1}}));
+    }
+
+    SECTION("user type with its own to_json overloads")
+    {
+        udt_container c{{1, 2}};
+        const json lv = c;
+        CHECK(lv == json::array({1, 2, "copied"}));
+        const json rv = std::move(c);
+        CHECK(rv == json::array({1, 2, "moved"}));
+    }
+
+    SECTION("assignment and implicit conversion")
+    {
+        std::vector<std::string> v{long_a};
+        json j;
+        j = std::move(v);
+        CHECK(j == json::array({long_a}));
+        json k = {std::vector<std::string>{long_b}};
+        CHECK(k.size() == 1);
+    }
+
+#ifdef TO_JSON_RVALUE_HAS_FS
+    SECTION("std::filesystem::path rvalue is still a string")
+    {
+        CHECK(json(std::filesystem::path("a/b")) == json("a/b"));
+#ifdef __cpp_lib_char8_t
+        CHECK(json(std::u8string(u8"a/b")) == json("a/b"));
+#endif
+    }
+#endif
+}