Restore unflatten() speed after the linear-time rewrite

The prefix tree introduced for linear unflatten() kept a std::map per prefix,
copied each token into it and looked every token up a second time while
creating the values. Replace the per-prefix maps by one open-addressing table
keyed by (parent, token) that refers to the parsed tokens, skip the lookups for
the prefix shared with the previous key (flattened objects are mostly sorted),
and derive the prefix numbers of the second pass from parent links instead of
lookups.

Time and memory stay linear in the pointer depth, and the result is still
independent of the iteration order of the flattened object.

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

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(),
+        // one buffer: every token adds its escaped form (at least its length) and a slash
-                               string_t{},
+        std::size_t length = 0;
-                               [](const string_t& a, const string_t& b)
+        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;
+        json_pointer res;
-        res.pop_back();
+        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
+    The prefixes form a tree and are numbered (number 0 is the empty prefix),
-    once (as a node) rather than as a copy of all of its reference tokens.
+    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
+        // open-addressing hash table from (parent, last reference token) to
-        // extended by that token; number 0 is the empty prefix
+        // the number of the node; empty slots have token == nullptr
-        std::vector<std::map<string_t, std::size_t>> children;
+        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
+            // keep the load factor below 1/2
-            // reallocate children and invalidate the iterator
+            if (++used * 2 > slots.size())
-            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)
             {
-                children.emplace_back();
+                grow();
-                is_array.push_back(false);
             }
             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
+    @param[in] j  the value to create the pointed to value in
-    of the number of siblings for the prefix lookup).
+    @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
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
-                    ptr = &ptr->at(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);
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
-                    if (JSON_HEDLEY_UNLIKELY(it == ptr->end()))
+                    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
@@ -913,25 +1101,29 @@ 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
+            // check reference tokens are properly escaped; a token without
-            for (std::size_t pos = reference_token.find_first_of('~');
+            // a '~' (the common case) needs neither check nor unescaping
-                    pos != string_t::npos;
+            std::size_t pos = reference_token.find_first_of('~');
-                    pos = reference_token.find_first_of('~', pos + 1))
+            if (pos != string_t::npos)
             {
-                JSON_ASSERT(reference_token[pos] == '~');
+                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
-
-                // ~ 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));
+                    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(std::move(reference_token));
-            result.push_back(reference_token);
         }
 
         return result;
@@ -1116,21 +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 common = 0;
-            std::size_t id = 0;
+            if (previous != nullptr)
-            for (auto& reference_token : ptr.reference_tokens)
             {
-                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()))
@@ -1138,11 +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;
@@ -1181,7 +1415,7 @@ class json_pointer
         = delete;
 #else
     {
-        return *this == json_pointer(rhs);
+        return equals_string(rhs);
     }
 #endif
 
@@ -1265,7 +1499,7 @@ inline bool operator==(const json_pointer<RefStringTypeLhs>& lhs,
     = delete;
 #else
 {
-    return lhs == json_pointer<RefStringTypeLhs>(rhs);
+    return lhs.equals_string(rhs);
 }
 #endif
 
@@ -1278,7 +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
+ * characters that need escaping, while repeated replace() calls would move
- * common case -- nothing to escape -- as fast as a single search, while
+ * the tail of the string once per escaped character. @ref escape_into appends
- * repeated replace() calls would move the tail of the string once per
+ * the escaped string to an existing one.
- * escaped character.
  */
+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);
+        ++i;
-        const auto slash = s.find_first_of('/', from);
+    }
-        return tilde < slash ? tilde : slash; // npos is the largest value
+    if (i == size)
-    };
-
-    auto pos = next_special(0);
-    if (pos == StringType::npos)
     {
         return s;
     }
 
     StringType result;
-    result.reserve(s.size() + 2);
+    result.reserve(size + 2);
-
+    escape_into(result, s);
-    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);
     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/json.hpp

@@ -6562,24 +6562,18 @@ public:
         };
 
         // wrapper for "add" operation; add value at ptr
-        const auto operation_add = [&result](json_pointer & ptr, const basic_json & val)
+        const auto operation_add = [&result](json_pointer & ptr, basic_json val)
         {
             // adding to the root of the target document means replacing it
             if (ptr.empty())
             {
-                result = val;
+                result = std::move(val);
                 return;
             }
 
-            // make sure the top element of the pointer exists
+            // get reference to the parent of the JSON pointer ptr; a missing
-            json_pointer const top_pointer = ptr.top();
+            // parent (or one of its ancestors) is reported by at()
-            if (top_pointer != ptr)
+            auto last_path = std::move(ptr.reference_tokens.back());
-            {
-                result.at(top_pointer);
-            }
-
-            // get reference to the parent of the JSON pointer ptr
-            const auto last_path = ptr.back();
             ptr.pop_back();
             // parent must exist when performing patch add per RFC6902 specs
             basic_json& parent = result.at(ptr);
@@ -6590,7 +6584,7 @@ public:
                 case value_t::object:
                 {
                     // use operator[] to add value
-                    parent[last_path] = val;
+                    parent[std::move(last_path)] = std::move(val);
                     break;
                 }
 
@@ -6599,7 +6593,7 @@ public:
                     if (last_path == "-")
                     {
                         // special case: append to back
-                        parent.push_back(val);
+                        parent.push_back(std::move(val));
                     }
                     else
                     {
@@ -6611,7 +6605,7 @@ public:
                         }
 
                         // default case: insert add offset
-                        parent.insert(parent.begin() + static_cast<difference_type>(idx), val);
+                        parent.insert(parent.begin() + static_cast<difference_type>(idx), std::move(val));
                     }
                     break;
                 }
@@ -6634,7 +6628,8 @@ public:
         const auto operation_remove = [this, & result](json_pointer & ptr)
         {
             // get reference to the parent of the JSON pointer ptr
-            const auto last_path = ptr.back();
+            static_cast<void>(ptr.back()); // throws out_of_range.405 if ptr is the root
+            const auto last_path = std::move(ptr.reference_tokens.back());
             ptr.pop_back();
             basic_json& parent = result.at(ptr);
 
@@ -6734,8 +6729,8 @@ public:
             }
 
             // collect mandatory members
-            const auto op = get_value("op", "op", true).template get<string_t>();
+            const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
-            const auto path = get_value(op, "path", true).template get<string_t>();
+            const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
             json_pointer ptr(path);
 
             switch (get_op(op))
@@ -6761,7 +6756,7 @@ public:
 
                 case patch_operations::move:
                 {
-                    const auto from_path = get_value("move", "from", true).template get<string_t>();
+                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
                     json_pointer from_ptr(from_path);
 
                     if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -6770,29 +6765,28 @@ public:
                     }
 
                     // the "from" location must exist - use at()
-                    basic_json const v = result.at(from_ptr);
+                    // a root "from" is rejected by operation_remove below, which must leave result intact
+                    basic_json v = from_ptr.empty() ? basic_json(result) : std::move(result.at(from_ptr));
 
                     // The move operation is functionally identical to a
                     // "remove" operation on the "from" location, followed
                     // immediately by an "add" operation at the target
                     // location with the value that was just removed.
                     operation_remove(from_ptr);
-                    operation_add(ptr, v);
+                    operation_add(ptr, std::move(v));
                     break;
                 }
 
                 case patch_operations::copy:
                 {
-                    const auto from_path = get_value("copy", "from", true).template get<string_t>();
+                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
                     const json_pointer from_ptr(from_path);
 
-                    // the "from" location must exist - use at()
-                    basic_json const v = result.at(from_ptr);
-
                     // The copy is functionally identical to an "add"
                     // operation at the target location using the value
-                    // specified in the "from" member.
+                    // specified in the "from" member; the "from" location
-                    operation_add(ptr, v);
+                    // must exist - use at()
+                    operation_add(ptr, result.at(from_ptr));
                     break;
                 }
 

single_include/nlohmann/json.hpp

@@ -3415,6 +3415,7 @@ NLOHMANN_JSON_NAMESPACE_END
 
 
 #include <cstddef> // size_t
+#include <utility> // move
 
 // #include <nlohmann/detail/abi_macros.hpp>
 
@@ -3431,39 +3432,46 @@ 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
+ * characters that need escaping, while repeated replace() calls would move
- * common case -- nothing to escape -- as fast as a single search, while
+ * the tail of the string once per escaped character. @ref escape_into appends
- * repeated replace() calls would move the tail of the string once per
+ * the escaped string to an existing one.
- * escaped character.
  */
+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);
+        ++i;
-        const auto slash = s.find_first_of('/', from);
+    }
-        return tilde < slash ? tilde : slash; // npos is the largest value
+    if (i == size)
-    };
-
-    auto pos = next_special(0);
-    if (pos == StringType::npos)
     {
         return s;
     }
 
     StringType result;
-    result.reserve(s.size() + 2);
+    result.reserve(size + 2);
-
+    escape_into(result, s);
-    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);
     return result;
 }
 
@@ -3508,7 +3516,7 @@ 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
@@ -20634,8 +20642,6 @@ NLOHMANN_JSON_NAMESPACE_END
     #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
@@ -20693,12 +20699,21 @@ 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(),
+        // one buffer: every token adds its escaped form (at least its length) and a slash
-                               string_t{},
+        std::size_t length = 0;
-                               [](const string_t& a, const string_t& b)
+        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
@@ -20753,21 +20768,29 @@ 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
@@ -20779,8 +20802,8 @@ class json_pointer
             return *this;
         }
 
-        json_pointer res = *this;
+        json_pointer res;
-        res.pop_back();
+        res.reference_tokens = std::vector<string_t>(reference_tokens.begin(), reference_tokens.end() - 1);
         return res;
     }
 
@@ -20869,14 +20892,14 @@ 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);
@@ -20929,6 +20952,28 @@ 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
@@ -20985,6 +21030,83 @@ 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
     {
@@ -21008,84 +21130,141 @@ 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
+    The prefixes form a tree and are numbered (number 0 is the empty prefix),
-    once (as a node) rather than as a copy of all of its reference tokens.
+    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
+        // open-addressing hash table from (parent, last reference token) to
-        // extended by that token; number 0 is the empty prefix
+        // the number of the node; empty slots have token == nullptr
-        std::vector<std::map<string_t, std::size_t>> children;
+        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
+            // keep the load factor below 1/2
-            // reallocate children and invalidate the iterator
+            if (++used * 2 > slots.size())
-            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)
             {
-                children.emplace_back();
+                grow();
-                is_array.push_back(false);
             }
             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
+    @param[in] j  the value to create the pointed to value in
-    of the number of siblings for the prefix lookup).
+    @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
@@ -21102,7 +21281,7 @@ 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;
                 }
 
@@ -21129,8 +21308,6 @@ 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;
@@ -21182,7 +21359,7 @@ class json_pointer
                 case detail::value_t::object:
                 {
                     // use unchecked object access
-                    ptr = &ptr->operator[](reference_token);
+                    ptr = &object_member(*ptr, reference_token);
                     break;
                 }
 
@@ -21232,8 +21409,13 @@ class json_pointer
             {
                 case detail::value_t::object:
                 {
-                    // note: at performs range check
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
-                    ptr = &ptr->at(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;
                 }
 
@@ -21254,7 +21436,7 @@ 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;
                 }
 
@@ -21302,7 +21484,9 @@ 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;
                 }
 
@@ -21316,7 +21500,9 @@ 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;
                 }
 
@@ -21357,12 +21543,12 @@ class json_pointer
             {
                 case detail::value_t::object:
                 {
-                    const auto it = ptr->find(reference_token);
+                    const auto it = ptr->m_data.m_value.object->find(reference_token);
-                    if (JSON_HEDLEY_UNLIKELY(it == ptr->end()))
+                    if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
                     {
                         return nullptr;
                     }
-                    ptr = &*it;
+                    ptr = &it->second;
                     break;
                 }
 
@@ -21399,7 +21585,7 @@ class json_pointer
                     {
                         return nullptr;
                     }
-                    ptr = &ptr->operator[](idx);
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
@@ -21432,13 +21618,14 @@ 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;
                 }
 
@@ -21460,13 +21647,13 @@ 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;
                 }
 
@@ -21516,6 +21703,15 @@ 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
@@ -21537,25 +21733,29 @@ 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
+            // check reference tokens are properly escaped; a token without
-            for (std::size_t pos = reference_token.find_first_of('~');
+            // a '~' (the common case) needs neither check nor unescaping
-                    pos != string_t::npos;
+            std::size_t pos = reference_token.find_first_of('~');
-                    pos = reference_token.find_first_of('~', pos + 1))
+            if (pos != string_t::npos)
             {
-                JSON_ASSERT(reference_token[pos] == '~');
+                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
-
-                // ~ 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));
+                    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(std::move(reference_token));
-            result.push_back(reference_token);
         }
 
         return result;
@@ -21740,21 +21940,53 @@ 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 common = 0;
-            std::size_t id = 0;
+            if (previous != nullptr)
-            for (auto& reference_token : ptr.reference_tokens)
             {
-                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()))
@@ -21762,11 +21994,21 @@ 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;
@@ -21805,7 +22047,7 @@ class json_pointer
         = delete;
 #else
     {
-        return *this == json_pointer(rhs);
+        return equals_string(rhs);
     }
 #endif
 
@@ -21889,7 +22131,7 @@ inline bool operator==(const json_pointer<RefStringTypeLhs>& lhs,
     = delete;
 #else
 {
-    return lhs == json_pointer<RefStringTypeLhs>(rhs);
+    return lhs.equals_string(rhs);
 }
 #endif
 
@@ -21902,7 +22144,7 @@ inline bool operator==(const StringType& lhs,
     = delete;
 #else
 {
-    return json_pointer<RefStringTypeRhs>(lhs) == rhs;
+    return rhs.equals_string(lhs);
 }
 #endif
 
@@ -35946,24 +36188,18 @@ public:
         };
 
         // wrapper for "add" operation; add value at ptr
-        const auto operation_add = [&result](json_pointer & ptr, const basic_json & val)
+        const auto operation_add = [&result](json_pointer & ptr, basic_json val)
         {
             // adding to the root of the target document means replacing it
             if (ptr.empty())
             {
-                result = val;
+                result = std::move(val);
                 return;
             }
 
-            // make sure the top element of the pointer exists
+            // get reference to the parent of the JSON pointer ptr; a missing
-            json_pointer const top_pointer = ptr.top();
+            // parent (or one of its ancestors) is reported by at()
-            if (top_pointer != ptr)
+            auto last_path = std::move(ptr.reference_tokens.back());
-            {
-                result.at(top_pointer);
-            }
-
-            // get reference to the parent of the JSON pointer ptr
-            const auto last_path = ptr.back();
             ptr.pop_back();
             // parent must exist when performing patch add per RFC6902 specs
             basic_json& parent = result.at(ptr);
@@ -35974,7 +36210,7 @@ public:
                 case value_t::object:
                 {
                     // use operator[] to add value
-                    parent[last_path] = val;
+                    parent[std::move(last_path)] = std::move(val);
                     break;
                 }
 
@@ -35983,7 +36219,7 @@ public:
                     if (last_path == "-")
                     {
                         // special case: append to back
-                        parent.push_back(val);
+                        parent.push_back(std::move(val));
                     }
                     else
                     {
@@ -35995,7 +36231,7 @@ public:
                         }
 
                         // default case: insert add offset
-                        parent.insert(parent.begin() + static_cast<difference_type>(idx), val);
+                        parent.insert(parent.begin() + static_cast<difference_type>(idx), std::move(val));
                     }
                     break;
                 }
@@ -36018,7 +36254,8 @@ public:
         const auto operation_remove = [this, & result](json_pointer & ptr)
         {
             // get reference to the parent of the JSON pointer ptr
-            const auto last_path = ptr.back();
+            static_cast<void>(ptr.back()); // throws out_of_range.405 if ptr is the root
+            const auto last_path = std::move(ptr.reference_tokens.back());
             ptr.pop_back();
             basic_json& parent = result.at(ptr);
 
@@ -36118,8 +36355,8 @@ public:
             }
 
             // collect mandatory members
-            const auto op = get_value("op", "op", true).template get<string_t>();
+            const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
-            const auto path = get_value(op, "path", true).template get<string_t>();
+            const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
             json_pointer ptr(path);
 
             switch (get_op(op))
@@ -36145,7 +36382,7 @@ public:
 
                 case patch_operations::move:
                 {
-                    const auto from_path = get_value("move", "from", true).template get<string_t>();
+                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
                     json_pointer from_ptr(from_path);
 
                     if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -36154,29 +36391,28 @@ public:
                     }
 
                     // the "from" location must exist - use at()
-                    basic_json const v = result.at(from_ptr);
+                    // a root "from" is rejected by operation_remove below, which must leave result intact
+                    basic_json v = from_ptr.empty() ? basic_json(result) : std::move(result.at(from_ptr));
 
                     // The move operation is functionally identical to a
                     // "remove" operation on the "from" location, followed
                     // immediately by an "add" operation at the target
                     // location with the value that was just removed.
                     operation_remove(from_ptr);
-                    operation_add(ptr, v);
+                    operation_add(ptr, std::move(v));
                     break;
                 }
 
                 case patch_operations::copy:
                 {
-                    const auto from_path = get_value("copy", "from", true).template get<string_t>();
+                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
                     const json_pointer from_ptr(from_path);
 
-                    // the "from" location must exist - use at()
-                    basic_json const v = result.at(from_ptr);
-
                     // The copy is functionally identical to an "add"
                     // operation at the target location using the value
-                    // specified in the "from" member.
+                    // specified in the "from" member; the "from" location
-                    operation_add(ptr, v);
+                    // must exist - use at()
+                    operation_add(ptr, result.at(from_ptr));
                     break;
                 }
 

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-json_pointer.cpp

@@ -23,10 +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>
-#include <sstream>
 
 TEST_CASE("JSON pointers")
 {
@@ -1123,3 +1127,192 @@ TEST_CASE("flatten of ordered_json")
         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);
+}