Merge develop into perf/view-compare

Resolves the conflict in unit-json_view_image.cpp, where develop (#5889)
and this branch each added a section to "json_view images: errors": both
are kept. single_include is a fresh amalgamation.

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

include/nlohmann/detail/input/binary_reader.hpp

@@ -86,7 +86,7 @@ JSON_INLINE_VARIABLE constexpr std::size_t max_valueless_container_size = 1 << 2
 /*!
 @brief deserialization of BJData, BON8, BSON, CBOR, MessagePack, and UBJSON values
 */
-template<typename BasicJsonType, typename InputAdapterType, typename SAX = json_sax_dom_parser<BasicJsonType, InputAdapterType>>
+template<typename BasicJsonType, typename InputAdapterType, typename SAX = json_sax_dom_binary_parser<BasicJsonType, InputAdapterType>>
 class binary_reader
 {
     using number_integer_t = typename BasicJsonType::number_integer_t;

include/nlohmann/detail/input/input_adapters.hpp

@@ -8,17 +8,18 @@
 
 #pragma once
 
-#include <algorithm> // count, find, min
+#include <algorithm> // count, find, max, min
 #include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t
 #include <cstring> // strlen
 #include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
 #include <limits> // numeric_limits
+#include <memory> // unique_ptr
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
-#include <utility> // pair, declval
+#include <utility> // declval, move, pair
 #include <vector> // vector
 
 #ifndef JSON_NO_IO
@@ -72,6 +73,82 @@ enum class input_format_t { json, cbor, msgpack, ubjson, bson, bjdata, bon8 };
 
 #ifndef JSON_NO_IO
 /*!
+A growable array of characters for the part of a token that an input adapter
+copied out of its window: unlike std::vector<char>::push_back(), appending a
+character takes no reference to a local, and is a few instructions when there
+is room.
+*/
+class token_carry_buffer
+{
+  public:
+    token_carry_buffer() = default;
+    token_carry_buffer(const token_carry_buffer&) = delete;
+    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
+    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+        : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
+    {
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+    }
+    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
+    ~token_carry_buffer() = default;
+
+    void clear() noexcept
+    {
+        m_end = m_data.get();
+    }
+
+    void push_back(char c)
+    {
+        if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
+        {
+            grow(1);
+        }
+        *m_end++ = c;
+    }
+
+    void append(const char* first, const char* last)
+    {
+        const auto n = static_cast<std::size_t>(last - first);
+        const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
+        if (n > room)
+        {
+            grow(n);
+        }
+        if (n != 0)
+        {
+            std::memcpy(m_end, first, n);
+        }
+        m_end += n;
+    }
+
+    /// replace the contents of @a out
+    void copy_to(std::vector<char>& out) const
+    {
+        out.assign(m_data.get(), m_end);
+    }
+
+  private:
+    /// make room for @a extra more characters
+    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    {
+        const auto size = static_cast<std::size_t>(m_end - m_data.get());
+        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
+        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
+        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+        if (size != 0)
+        {
+            std::memcpy(data.get(), m_data.get(), size);
+        }
+        m_data = std::move(data);
+        m_end = m_data.get() + size;
+        m_capacity_end = m_data.get() + new_capacity;
+    }
+
+    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    char* m_end = nullptr;
+    char* m_capacity_end = nullptr;
+};
+/*!
 Input adapter for stdio file access. This adapter read only 1 byte and do not use any
  buffer. This adapter is a very low level adapter.
 */
@@ -209,6 +286,7 @@ class input_stream_adapter
         , token_segment(rhs.token_segment)
         , current_in_get_area(rhs.current_in_get_area)
         , tracking_token(rhs.tracking_token)
+        , no_get_area(rhs.no_get_area)
 #endif
     {
         rhs.is = nullptr;
@@ -295,37 +373,20 @@ class input_stream_adapter
     std::char_traits<char>::int_type get_character()
     {
 #if !JSON_DISABLE_STREAM_BULK_SCAN
-        // sbumpc() refills the get area if it is exhausted
-        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
-        if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            save_token_segment();
-        }
-#endif
-
-        auto res = sb->sbumpc();
-        // set eof manually, as we don't use the istream interface.
-        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
-        {
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-            current_in_get_area = false;
-#endif
-            is->clear(is->rdstate() | std::ios::eofbit);
-        }
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-        else if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            // the consumed character is at gptr() - 1, unless the streambuf
-            // has no get area and returned it from uflow()
-            const char* const next = streambuf_access::get_ptr(sb);
-            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
-        }
-        else
+        // the character is in the get area: step over it, as sbumpc() does
+        const char* const next = streambuf_access::get_ptr(sb);
+        if (JSON_HEDLEY_LIKELY(next != streambuf_access::get_end(sb)))
         {
             current_in_get_area = true;
+            streambuf_access::bump(sb, 1);
+            return std::char_traits<char>::to_int_type(*next);
+        }
+        if (no_get_area)
+        {
+            return get_character_no_get_area();
         }
 #endif
-        return res;
+        return get_character_slow();
     }
 #endif
 
@@ -334,6 +395,18 @@ class input_stream_adapter
     {
 #if JSON_PRECISE_STREAM_POSITION
         commit_lookahead();
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // the elements are in the get area: copy them, as sgetn() does, but
+        // without the virtual call
+        const char* const first = streambuf_access::get_ptr(sb);
+        const auto available = static_cast<std::size_t>(streambuf_access::get_end(sb) - first);
+        if (JSON_HEDLEY_LIKELY(count * sizeof(T) <= available && count * sizeof(T) <= static_cast<std::size_t>((std::numeric_limits<int>::max)())))
+        {
+            std::memcpy(dest, first, count * sizeof(T));
+            streambuf_access::bump(sb, static_cast<int>(count * sizeof(T)));
+            return count * sizeof(T);
+        }
 #endif
         auto res = static_cast<std::size_t>(sb->sgetn(reinterpret_cast<char*>(dest), static_cast<std::streamsize>(count * sizeof(T))));
         if (JSON_HEDLEY_UNLIKELY(res < count * sizeof(T)))
@@ -421,7 +494,7 @@ class input_stream_adapter
     // last one if @a drop_last.
     void copy_token(std::vector<char>& out, bool drop_last) const
     {
-        out = token_carry;
+        token_carry.copy_to(out);
         if (token_segment != nullptr)
         {
             out.insert(out.end(), token_segment, current_end());
@@ -434,6 +507,47 @@ class input_stream_adapter
 #endif
 
   private:
+#if !JSON_PRECISE_STREAM_POSITION
+    // the get area is exhausted, or there is none
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    JSON_HEDLEY_NEVER_INLINE
+#endif
+    std::char_traits<char>::int_type get_character_slow()
+    {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // sbumpc() refills the get area if it is exhausted
+        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
+        if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            save_token_segment();
+        }
+#endif
+
+        auto res = sb->sbumpc();
+        // set eof manually, as we don't use the istream interface.
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
+        {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            current_in_get_area = false;
+#endif
+            is->clear(is->rdstate() | std::ios::eofbit);
+        }
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        else if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            // the consumed character is at gptr() - 1, unless the streambuf
+            // has no get area and returned it from uflow()
+            const char* const next = streambuf_access::get_ptr(sb);
+            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
+        }
+        else
+        {
+            current_in_get_area = true;
+        }
+#endif
+        return res;
+    }
+#endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
     // Reads the protected get-area pointers of any std::streambuf. Naming the
     // members through a derived class makes forming the member pointers
@@ -457,6 +571,11 @@ class input_stream_adapter
         {
             (s->*(&streambuf_access::gbump))(n);
         }
+        // sbumpc() for an exhausted get area
+        static std::streambuf::int_type next_character(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::uflow))();
+        }
     };
 
     // one past the character last returned by get_character(), if that is in
@@ -477,16 +596,53 @@ class input_stream_adapter
     {
         if (token_segment != nullptr)
         {
-            token_carry.insert(token_carry.end(), token_segment, streambuf_access::get_end(sb));
+            token_carry.append(token_segment, streambuf_access::get_end(sb));
             token_segment = nullptr;
         }
     }
 
+#if !JSON_PRECISE_STREAM_POSITION
+    // The streambuf has no get area (e.g., std::cin synchronized with stdio):
+    // every character comes from uflow(), and the token is kept by value.
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area()
+    {
+        // the get area is exhausted, so this is what sbumpc() does
+        const auto res = streambuf_access::next_character(sb);
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof() || streambuf_access::get_end(sb) != nullptr))
+        {
+            return get_character_no_get_area_other(res);
+        }
+        // current_in_get_area is false: only the first character of a get
+        // area, or the end of the input, is read here
+        if (tracking_token)
+        {
+            token_carry.push_back(std::char_traits<char>::to_char_type(res));
+        }
+        return res;
+    }
+
+    // @a res came from uflow() of a streambuf without a get area: it is the
+    // end of the input, or the streambuf has a get area now
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area_other(std::char_traits<char>::int_type res)
+    {
+        if (res == std::char_traits<char>::eof())
+        {
+            current_in_get_area = false;
+            is->clear(is->rdstate() | std::ios::eofbit);
+            return res;
+        }
+        // the consumed character is at gptr() - 1
+        after_refill(res, streambuf_access::get_ptr(sb) != streambuf_access::get_begin(sb));
+        return res;
+    }
+#endif
+
     // character @a c was just read into a new get area (@a in_get_area) or
     // returned without one
     void after_refill(std::char_traits<char>::int_type c, bool in_get_area)
     {
         current_in_get_area = in_get_area;
+        no_get_area = !in_get_area;
         if (!tracking_token)
         {
             return;
@@ -525,13 +681,16 @@ class input_stream_adapter
 #endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
     /// the part of the current token that was in earlier get areas
-    std::vector<char> token_carry{}; // NOLINT(readability-redundant-member-init)
+    token_carry_buffer token_carry{}; // NOLINT(readability-redundant-member-init)
     /// start of the current token's part in the get area, or nullptr
     const char* token_segment = nullptr;
     /// whether the character last returned by get_character() is in the get area
     bool current_in_get_area = false;
     /// whether a token is tracked at all; the binary readers never start one
     bool tracking_token = false;
+    /// whether the streambuf had no get area when it was last asked for a
+    /// character
+    bool no_get_area = false;
 #endif
 };
 #endif  // JSON_NO_IO

include/nlohmann/detail/input/lexer.hpp

@@ -14,6 +14,7 @@
 #include <cstdio> // snprintf
 #include <cstring> // memcmp
 #include <initializer_list> // initializer_list
+#include <limits> // numeric_limits
 #include <string> // char_traits, string
 #include <utility> // move
 #include <vector> // vector
@@ -1816,6 +1817,15 @@ scan_number_done:
     /// significand (see scan_number_bulk_contiguous())
     bool convert_accumulated_float(const float_significand& s, std::true_type /*native*/) noexcept
     {
+        // accept() only needs to know that the value is finite (see the parser):
+        // w < 10^19, so a value below 10^max_exponent10 cannot overflow, and a
+        // value that underflows is a finite number as well. Only a token that
+        // may overflow has to be converted to find out.
+        if (JSON_HEDLEY_UNLIKELY(discard_number_values && s.exponent + 19 <= std::numeric_limits<number_float_t>::max_exponent10))
+        {
+            value_float = 0;
+            return true;
+        }
         value_float = decimal_to_float<number_float_t>(s);
         return true;
     }

include/nlohmann/detail/input/parser.hpp

@@ -56,8 +56,8 @@ using parser_callback_t =
 template<typename SAX>
 struct sax_ignores_number_text : std::false_type {};
 
-template<typename BasicJsonType, typename InputAdapterType>
-struct sax_ignores_number_text<json_sax_dom_parser<BasicJsonType, InputAdapterType>> : std::true_type {};
+template<typename BasicJsonType, typename InputAdapterType, bool Collect>
+struct sax_ignores_number_text<json_sax_dom_parser<BasicJsonType, InputAdapterType, Collect>> : std::true_type {};
 
 template<typename BasicJsonType, typename InputAdapterType>
 struct sax_ignores_number_text<json_sax_dom_callback_parser<BasicJsonType, InputAdapterType>> : std::true_type {};
@@ -131,7 +131,7 @@ class parser
         }
         else
         {
-            json_sax_dom_parser<BasicJsonType, InputAdapterType> sdp(result, allow_exceptions, &m_lexer);
+            json_sax_dom_parser<BasicJsonType, InputAdapterType, true> sdp(result, allow_exceptions, &m_lexer);
 
             // in case of an error, return a discarded value
             if (!parse_dom(sdp, strict))
@@ -243,12 +243,19 @@ class parser
     JSON_HEDLEY_NON_NULL(2)
     bool sax_parse_internal(SAX* sax)
     {
+        // value to avoid a goto (see comment where set to true)
+        bool skip_to_state_evaluation = false;
+
         // stack to remember the hierarchy of structured values we are parsing
         // true = array; false = object (bytes, not std::vector<bool>, whose
         // push_back() is a call with GCC)
         std::vector<std::uint8_t> states;
-        // value to avoid a goto (see comment where set to true)
-        bool skip_to_state_evaluation = false;
+        // a document that starts with a container pushes onto the stack at
+        // once: allocate it in one step instead of growing it from 1 byte
+        if (last_token == token_type::begin_object || last_token == token_type::begin_array)
+        {
+            states.reserve(32);
+        }
 
         while (true)
         {

include/nlohmann/detail/json_pointer.hpp

@@ -8,14 +8,14 @@
 
 #pragma once
 
-#include <algorithm> // all_of
+#include <algorithm> // all_of, copy
 #include <array> // array
 #include <cctype> // isdigit
 #include <cerrno> // errno, ERANGE
 #include <cstddef> // size_t
 #include <cstdint> // uint64_t
 #include <cstdlib> // strtoull
-#include <cstring> // memcmp, memcpy
+#include <cstring> // memchr, memcmp, memcpy
 #ifndef JSON_NO_IO
     #include <iosfwd> // ostream
 #endif  // JSON_NO_IO
@@ -80,14 +80,38 @@ class json_pointer
     /// @sa https://json.nlohmann.me/api/json_pointer/to_string/
     string_t to_string() const
     {
+        if (reference_tokens.empty())
+        {
+            return string_t();
+        }
+
         // one buffer: every token adds its escaped form (at least its length) and a slash
         std::size_t length = 0;
+        bool plain = true;
         for (const auto& reference_token : reference_tokens)
         {
             length += reference_token.size() + 1;
+            const char* const data = reference_token.data();
+            for (std::size_t i = 0, size = reference_token.size(); plain && i < size; ++i)
+            {
+                plain = data[i] != '~' && data[i] != '/';
+            }
         }
 
         string_t result;
+        if (JSON_HEDLEY_LIKELY(plain))
+        {
+            // nothing to escape: the tokens are copied into the string
+            result.resize(length);
+            char* out = &result[0];
+            for (const auto& reference_token : reference_tokens)
+            {
+                *out++ = '/';
+                out = std::copy(reference_token.data(), reference_token.data() + reference_token.size(), out);
+            }
+            return result;
+        }
+
         result.reserve(length);
         for (const auto& reference_token : reference_tokens)
         {
@@ -303,6 +327,12 @@ class json_pointer
         exceeds_size_type  ///< @a s converts to an integer that exceeds size_type
     };
 
+    /// whether a reference token is "-" (a plain comparison with a string literal is not always inlined)
+    static bool is_dash(const string_t& s) noexcept
+    {
+        return s.size() == 1 && s[0] == '-';
+    }
+
     /*!
     @param[in] s    reference token to be converted into an array index
     @param[out] idx the integer representation of @a s if @ref array_index_status::ok
@@ -328,7 +358,7 @@ class json_pointer
         // error condition (cf. RFC 6901, Sect. 4); this also covers single-
         // character tokens, so "/x" and "/xy" fail alike; "-" and the empty
         // token are left to the conversion below and are reported as unresolved
-        if (JSON_HEDLEY_UNLIKELY(!s.empty() && s != "-" && !(s[0] >= '0' && s[0] <= '9')))
+        if (JSON_HEDLEY_UNLIKELY(!s.empty() && !is_dash(s) && !(s[0] >= '0' && s[0] <= '9')))
         {
             return array_index_status::not_a_number;
         }
@@ -475,9 +505,40 @@ class json_pointer
         return pos == size;
     }
 
+    /// reserve room for @a n elements of an array type that has reserve()
+    template<typename ArrayType>
+    static auto reserve_elements(ArrayType& array, const std::size_t n, detail::priority_tag<1> /*unused*/)
+    -> decltype(array.reserve(n), void())
+    {
+        array.reserve(n);
+    }
+
+    template<typename ArrayType>
+    static void reserve_elements(ArrayType& /*array*/, const std::size_t /*n*/, detail::priority_tag<0> /*unused*/)
+    {}
+
     /// @brief object member access with a key that is only copied if it is inserted
+    ///
+    /// Objects that can find the position of a key and insert there
+    /// (std::map) search once, others look the key up and add it if it is missing.
     template<typename BasicJsonType>
-    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    static auto object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<1> /*unused*/)
+    -> decltype(object.m_data.m_value.object->emplace_hint(object.m_data.m_value.object->lower_bound(key), key, nullptr), std::declval<BasicJsonType&>())
+    {
+        auto& members = *object.m_data.m_value.object;
+        auto it = members.lower_bound(key);
+        if (it != members.end() && !members.key_comp()(key, it->first))
+        {
+            return it->second;
+        }
+
+        const auto old_capacity = object.container_capacity();
+        it = members.emplace_hint(it, key, nullptr);
+        return object.set_parent_after_insert(it->second, old_capacity);
+    }
+
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<0> /*unused*/)
     {
         auto& members = *object.m_data.m_value.object;
         const auto it = members.find(key);
@@ -488,6 +549,12 @@ class json_pointer
         return object.operator[](key);
     }
 
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    {
+        return object_member_impl(object, key, detail::priority_tag<1> {});
+    }
+
   JSON_PRIVATE_UNLESS_TESTED:
     json_pointer top() const
     {
@@ -720,48 +787,103 @@ class json_pointer
         }
     }
 
+    /*!
+    @brief the values along the pointer last resolved by @ref resolve_plain
+
+    A patch from @ref basic_json::diff lists its operations in document order,
+    so consecutive pointers share most of their reference tokens; the values
+    along the previous pointer are kept to resume from the common prefix.
+    `steps[k]` is the value after the first `k` reference tokens together with
+    the length of the prefix of `*path` that holds them. A value stays valid as
+    long as no operation changes its container: callers keep only values up to
+    the parent of the pointer they modify, or call `reset()`.
+    */
+    template<typename BasicJsonType>
+    struct plain_steps
+    {
+        const string_t* path = nullptr;
+        std::vector<std::pair<std::size_t, BasicJsonType*>> steps{};
+
+        void reset() noexcept
+        {
+            path = nullptr;
+            steps.clear();
+        }
+    };
+
     /*!
     @brief resolve a plain JSON pointer without building its reference tokens
 
-    @param[in] ptr     the value the pointer starts at
+    @param[in] root    the value the pointer starts at
     @param[in] path    a JSON pointer string
     @param[in] length  the number of leading bytes of @a path to resolve; 0 or
                        the position of a slash
     @param[in,out] token  a string to hold the reference tokens, to reuse its storage
+    @param[in,out] cache  the values along the previously resolved pointer;
+                          @a path must outlive the next call
 
     @return the pointed to value, or `nullptr` if the pointer does not denote
-            an existing value, or does so in a way that is not plain: a
-            reference token with an escape, an array index that is malformed or
-            out of range, or a "-". No exception is thrown for these; callers
-            use @ref json_pointer instead, which handles them.
+            an existing value, or does so in a way that is not plain: an array
+            index that is malformed or out of range, or a "-", or a reference
+            token with an escape. No exception is
+            thrown for these; callers use @ref json_pointer instead, which
+            handles them.
     */
     template<typename BasicJsonType>
-    static BasicJsonType* resolve_plain(BasicJsonType* ptr, const string_t& path, const std::size_t length, string_t& token)
+    static BasicJsonType* resolve_plain(BasicJsonType* root, const string_t& path, const std::size_t length, string_t& token, plain_steps<BasicJsonType>& cache)
     {
-        for (std::size_t pos = 0; pos < length;)
+        auto& steps = cache.steps;
+
+        // resume from the deepest step that is also a prefix of this path
+        std::size_t depth = 1;
+        if (cache.path != nullptr && !steps.empty() && steps[0].second == root)
+        {
+            const string_t& previous = *cache.path;
+            for (depth = steps.size(); depth > 1; --depth)
+            {
+                // a step ends at a token boundary of the previous path
+                const std::size_t prefix = steps[depth - 1].first;
+                if (prefix <= length && (prefix == length || path[prefix] == '/') && std::memcmp(path.data(), previous.data(), prefix) == 0)
+                {
+                    break;
+                }
+            }
+            steps.resize(depth);
+        }
+        else
+        {
+            steps.assign(1, std::make_pair(std::size_t{0}, root));
+        }
+        cache.path = &path;
+
+        BasicJsonType* ptr = steps[depth - 1].second;
+        for (std::size_t pos = steps[depth - 1].first; pos < length;)
         {
             if (JSON_HEDLEY_UNLIKELY(path[pos] != '/'))
             {
                 return nullptr;
             }
 
-            // the reference token between the slash at pos and the next one
-            std::size_t end = path.find_first_of('/', pos + 1);
-            if (end == string_t::npos || end > length)
+            // the reference token between the slash at pos and the next one,
+            // which must not have an escape
+            const char* const begin = path.data() + pos + 1;
+            const char* const limit = path.data() + length;
+            const char* end = begin;
+            while (end < limit && *end != '/')
             {
-                end = length;
-            }
-            token.clear();
-            token.append(path.data() + pos + 1, end - pos - 1);
-            if (JSON_HEDLEY_UNLIKELY(token.find_first_of('~') != string_t::npos))
-            {
-                return nullptr;
+                if (JSON_HEDLEY_UNLIKELY(*end == '~'))
+                {
+                    return nullptr;
+                }
+                ++end;
             }
 
             switch (ptr->type())
             {
                 case detail::value_t::object:
                 {
+                    token.clear();
+                    token.append(begin, static_cast<std::size_t>(end - begin));
                     const auto it = ptr->m_data.m_value.object->find(token);
                     if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
                     {
@@ -773,8 +895,24 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    typename BasicJsonType::size_type idx{};
-                    if (JSON_HEDLEY_UNLIKELY(parse_array_index<BasicJsonType>(token, idx) != array_index_status::ok || idx >= ptr->m_data.m_value.array->size()))
+                    // an index of up to 18 digits without a leading zero fits into size_type
+                    // (9 digits for a 32-bit size_type); everything else is left to json_pointer
+                    constexpr std::size_t max_digits = sizeof(typename BasicJsonType::size_type) >= 8 ? 18 : 9;
+                    const auto digits = static_cast<std::size_t>(end - begin);
+                    if (JSON_HEDLEY_UNLIKELY(digits == 0 || digits > max_digits || (digits > 1 && *begin == '0')))
+                    {
+                        return nullptr;
+                    }
+                    typename BasicJsonType::size_type idx = 0;
+                    for (const char* c = begin; c != end; ++c)
+                    {
+                        if (JSON_HEDLEY_UNLIKELY(*c < '0' || *c > '9'))
+                        {
+                            return nullptr;
+                        }
+                        idx = static_cast<typename BasicJsonType::size_type>((idx * 10) + static_cast<typename BasicJsonType::size_type>(*c - '0'));
+                    }
+                    if (JSON_HEDLEY_UNLIKELY(idx >= ptr->m_data.m_value.array->size()))
                     {
                         return nullptr;
                     }
@@ -793,7 +931,8 @@ class json_pointer
                 default:
                     return nullptr;
             }
-            pos = end;
+            pos = static_cast<std::size_t>(end - path.data());
+            steps.emplace_back(pos, ptr);
         }
 
         return ptr;
@@ -907,7 +1046,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat(
@@ -978,7 +1117,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" cannot be used for const access
                         JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat("array index '-' (", std::to_string(ptr->m_data.m_value.array->size()), ") is out of range"), ptr));
@@ -1040,7 +1179,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         return nullptr;
@@ -1117,7 +1256,7 @@ class json_pointer
 
                 case detail::value_t::array:
                 {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                     {
                         // "-" always fails the range check
                         return false;
@@ -1198,33 +1337,32 @@ class json_pointer
         }
         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
-        for (
-            // search for the first slash after the first character
-            std::size_t slash = reference_string.find_first_of('/', 1),
-            // set the beginning of the first reference token
-            start = 1;
-            // we can stop if start == 0 (if slash == string_t::npos)
-            start != 0;
-            // set the beginning of the next reference token
-            // (will eventually be 0 if slash == string_t::npos)
-            start = (slash == string_t::npos) ? 0 : slash + 1,
-            // find next slash
-            slash = reference_string.find_first_of('/', start))
+        // extract the reference tokens: every one runs from after a slash to
+        // the next slash or the end of the string; a token without a '~' (the
+        // common case) is stored as it is, others are checked and unescaped
+        const char* const data = reference_string.data();
+        const char* const end = data + size;
+        for (const char* start = data + 1;; ++start)
         {
-            // use the text between the beginning of the reference token
-            // (start) and the last slash (slash).
-            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; 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)
+            const char* slash = start;
+            bool escaped = false;
+            while (slash != end && *slash != '/')
             {
-                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
+                escaped = escaped || *slash == '~';
+                ++slash;
+            }
+
+            const auto count = static_cast<std::size_t>(slash - start);
+            if (JSON_HEDLEY_LIKELY(!escaped))
+            {
+                result.emplace_back(start, count);
+            }
+            else
+            {
+                auto reference_token = string_t(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))
                 {
                     JSON_ASSERT(reference_token[pos] == '~');
 
@@ -1238,10 +1376,14 @@ class json_pointer
                 }
 
                 detail::unescape(reference_token);
+                result.push_back(std::move(reference_token));
             }
 
-            // finally, store the reference token
-            result.push_back(std::move(reference_token));
+            if (slash == end)
+            {
+                break;
+            }
+            start = slash;
         }
 
         return result;
@@ -1276,6 +1418,52 @@ class json_pointer
         obj[key] = std::forward<ValueType>(v);
     }
 
+    /// the number of entries that flattening @a value adds: its primitive
+    /// values and empty containers
+    template<typename BasicJsonType>
+    static std::size_t count_flattened(const BasicJsonType& value)
+    {
+        std::size_t count = 0;
+        std::vector<const BasicJsonType*> pending(1, &value);
+        while (!pending.empty())
+        {
+            const BasicJsonType* const v = pending.back();
+            pending.pop_back();
+            if (v->is_array() && !v->m_data.m_value.array->empty())
+            {
+                for (const auto& element : *v->m_data.m_value.array)
+                {
+                    pending.push_back(&element);
+                }
+            }
+            else if (v->is_object() && !v->m_data.m_value.object->empty())
+            {
+                for (const auto& member : *v->m_data.m_value.object)
+                {
+                    pending.push_back(&member.second);
+                }
+            }
+            else
+            {
+                ++count;
+            }
+        }
+        return count;
+    }
+
+    // Reserve the storage for @a value of objects that have a reserve(), so
+    // that it does not have to be moved while it grows
+    template<typename ObjectType, typename BasicJsonType>
+    static auto reserve_flattened(ObjectType& obj, const BasicJsonType& value, detail::priority_tag<1> /*unused*/)
+    -> decltype(obj.reserve(std::size_t {}), void())
+    {
+        obj.reserve(count_flattened(value));
+    }
+
+    template<typename ObjectType, typename BasicJsonType>
+    static void reserve_flattened(ObjectType& /*obj*/, const BasicJsonType& /*value*/, detail::priority_tag<0> /*unused*/)
+    {}
+
     /// append the decimal representation of @a i to @a out
     static void append_index(string_t& out, std::size_t i)
     {
@@ -1385,6 +1573,7 @@ class json_pointer
             }
         };
 
+        reserve_flattened(object, value, detail::priority_tag<1> {});
         enter(value);
         while (!stack.empty())
         {
@@ -1392,8 +1581,9 @@ class json_pointer
             // frame, which would invalidate a reference to it
             const BasicJsonType* const container = stack.back().container;
 
-            // drop the path of the previous child
-            path.resize(stack.back().path_length);
+            // the path of the previous child is replaced by the path of the
+            // next one; for an array element in one resize
+            const std::size_t path_length = stack.back().path_length;
 
             if (container->is_array())
             {
@@ -1407,8 +1597,20 @@ class json_pointer
 
                 // iterate array and use index as a reference string
                 ++stack.back().index;
-                path.push_back('/');
-                append_index(path, i);
+                std::array<char, 24> digits{};
+                std::size_t pos = digits.size();
+                std::size_t rest = i;
+                do
+                {
+                    digits[--pos] = static_cast<char>('0' + (rest % 10));
+                    rest /= 10;
+                }
+                while (rest != 0);
+                const std::size_t count = digits.size() - pos;
+                path.resize(path_length + 1 + count);
+                char* const out = &path[path_length];
+                out[0] = '/';
+                std::memcpy(out + 1, digits.data() + pos, count);
                 enter(array[i]);
             }
             else
@@ -1422,6 +1624,7 @@ class json_pointer
 
                 // iterate object and use keys as reference string
                 ++stack.back().member;
+                path.resize(path_length);
                 path.push_back('/');
                 detail::escape_into(path, it->first);
                 enter(it->second);
@@ -1517,8 +1720,28 @@ class json_pointer
                 {
                     end = key.size();
                 }
+                const std::size_t token_size = end - pos - 1;
+
+                // a token that is a plain index of up to 9 digits, which is
+                // the case for almost all array elements, is not copied
+                std::size_t index = 0;
+                bool plain_index = current->type() == detail::value_t::array && token_size != 0 && token_size <= 9 && (token_size == 1 || data[pos + 1] != '0');
+                for (std::size_t k = pos + 1; plain_index && k < end; ++k)
+                {
+                    plain_index = data[k] >= '0' && data[k] <= '9';
+                    index = (index * 10) + static_cast<std::size_t>(data[k] - '0');
+                }
+                if (plain_index)
+                {
+                    current = &current->operator[](static_cast<typename BasicJsonType::size_type>(index));
+                    values.push_back(current);
+                    ends.push_back(end);
+                    pos = end;
+                    continue;
+                }
+
                 token.clear();
-                token.append(data + pos + 1, end - pos - 1);
+                token.append(data + pos + 1, token_size);
                 if (token.find_first_of('~') != string_t::npos)
                 {
                     detail::unescape(token);
@@ -1528,10 +1751,13 @@ class json_pointer
                 {
                     case detail::value_t::null:
                     {
-                        if (arrays.contains(data, pos))
+                        // a first child 0 makes the value an array without a look-up
+                        if ((token_size == 1 && data[pos + 1] == '0') || arrays.contains(data, pos))
                         {
                             // some reference token below this position is 0, so the
                             // value is an array
+                            *current = BasicJsonType(detail::value_t::array);
+                            reserve_elements(*current->m_data.m_value.array, 2, detail::priority_tag<1> {});
                             current = &current->operator[](array_index<BasicJsonType>(token));
                         }
                         else

include/nlohmann/detail/output/output_adapters.hpp

@@ -55,6 +55,66 @@ using output_adapter_t = std::shared_ptr<output_adapter_protocol<CharType>>;
 /// output_vector_sink also appends writes of at least this size with insert()
 constexpr std::size_t max_reserve_size = 256;
 
+/// @brief copy @a length characters, at most 32, without a call of memcpy()
+///
+/// Short strings are the typical payload of a binary value, and a call of
+/// memcpy() costs more than copying them. A copy of 16 to 32 characters is two
+/// possibly overlapping 16-byte copies, one of 8 to 15 two 8-byte copies, and
+/// so on down to a single character; each is a fixed-size load and store.
+/// Nothing outside of [src, src + length) is read or written.
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::true_type /*is_byte*/) noexcept
+{
+    JSON_ASSERT(length <= 32);
+    if (length >= 16)
+    {
+        std::array<char, 16> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 16> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 16);
+        std::memcpy(last.data(), src + (length - 16), 16);
+        std::memcpy(dst, first.data(), 16);
+        std::memcpy(dst + (length - 16), last.data(), 16);
+    }
+    else if (length >= 8)
+    {
+        std::array<char, 8> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 8> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 8);
+        std::memcpy(last.data(), src + (length - 8), 8);
+        std::memcpy(dst, first.data(), 8);
+        std::memcpy(dst + (length - 8), last.data(), 8);
+    }
+    else if (length >= 4)
+    {
+        std::array<char, 4> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 4> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 4);
+        std::memcpy(last.data(), src + (length - 4), 4);
+        std::memcpy(dst, first.data(), 4);
+        std::memcpy(dst + (length - 4), last.data(), 4);
+    }
+    else if (length != 0)
+    {
+        // 1 to 3: the first, the middle, and the last character
+        const CharType a = src[0];
+        const CharType b = src[length >> 1];
+        const CharType c = src[length - 1];
+        dst[0] = a;
+        dst[length >> 1] = b;
+        dst[length - 1] = c;
+    }
+}
+
+/// @brief copy @a length characters, of a character type wider than a byte
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::false_type /*is_byte*/) noexcept
+{
+    std::memcpy(dst, src, length * sizeof(CharType));
+}
+
+/// the number of characters an empty vector is grown to by output_vector_sink
+constexpr std::size_t min_vector_sink_capacity = 64;
+
 /// @brief non-virtual output sink writing into a std::vector
 ///
 /// This sink is not part of the virtual output_adapter_protocol hierarchy: it is
@@ -78,7 +138,7 @@ class output_vector_sink
 {
   public:
     explicit output_vector_sink(std::vector<CharType, AllocatorType>& vec) noexcept
-        : v(&vec), position(vec.size())
+        : v(&vec), cursor(vec.data() + vec.size()), limit(cursor)
     {}
 
     output_vector_sink(const output_vector_sink&) = delete;
@@ -86,7 +146,7 @@ class output_vector_sink
     output_vector_sink& operator=(output_vector_sink&&) = delete;
 
     output_vector_sink(output_vector_sink&& other) noexcept
-        : v(other.v), position(other.position)
+        : v(other.v), cursor(other.cursor), limit(other.limit)
     {
         other.v = nullptr;
     }
@@ -95,7 +155,7 @@ class output_vector_sink
     {
         if (v != nullptr)
         {
-            v->resize(position);
+            v->resize(position());
         }
     }
 
@@ -104,11 +164,11 @@ class output_vector_sink
     JSON_HEDLEY_ALWAYS_INLINE
     void write_character(CharType c)
     {
-        if (JSON_HEDLEY_UNLIKELY(position == v->size()))
+        if (JSON_HEDLEY_UNLIKELY(cursor == limit))
         {
             grow(1);
         }
-        (*v)[position++] = c;
+        *cursor++ = c;
     }
 
     // no JSON_HEDLEY_NON_NULL here: binary_writer legitimately passes a null
@@ -117,7 +177,7 @@ class output_vector_sink
     JSON_HEDLEY_ALWAYS_INLINE
     void write_characters(const CharType* s, std::size_t length)
     {
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
         {
             if (length >= max_reserve_size)
             {
@@ -126,10 +186,15 @@ class output_vector_sink
             }
             grow(length);
         }
-        if (length != 0)
+        if (length <= 32)
         {
-            std::memcpy(v->data() + position, s, length * sizeof(CharType));
-            position += length;
+            copy_short(cursor, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            cursor += length;
+        }
+        else
+        {
+            std::memcpy(cursor, s, length * sizeof(CharType));
+            cursor += length;
         }
     }
 
@@ -144,21 +209,37 @@ class output_vector_sink
     CharType* reserve(const std::size_t length)
     {
         JSON_ASSERT(length <= max_reserve_size);
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
         {
             grow(length);
         }
-        return v->data() + position;
+        return cursor;
     }
 
     /// keep the first @a length characters written to the space from reserve()
     JSON_HEDLEY_ALWAYS_INLINE
     void commit(const std::size_t length) noexcept
     {
-        position += length;
+        cursor += length;
     }
 
+    /// nothing to pass on: the characters are in the vector (see output_adapter_sink)
+    void flush() noexcept
+    {}
+
   private:
+    /// number of characters of *v written so far
+    std::size_t position() const noexcept
+    {
+        return static_cast<std::size_t>(cursor - v->data());
+    }
+
+    /// number of characters that fit the vector behind the write position
+    std::size_t space() const noexcept
+    {
+        return static_cast<std::size_t>(limit - cursor);
+    }
+
     /// make room for at least @a length more characters after the position;
     /// kept out of line so that the write functions stay small enough for
     /// GCC to inline them into the writer
@@ -166,12 +247,15 @@ class output_vector_sink
     void grow(const std::size_t length)
     {
         // use the capacity the vector already has, else at least double it;
-        // never less than 64 characters, so that a small value does not
+        // never less than min_vector_sink_capacity, so that a small value does not
         // reallocate for every few bytes
-        const std::size_t needed = position + length;
-        v->resize((std::max)(std::size_t{64}, needed <= v->capacity()
+        const std::size_t written = position();
+        const std::size_t needed = written + length;
+        v->resize((std::max)(min_vector_sink_capacity, needed <= v->capacity()
                              ? v->capacity()
                              : (std::max)(needed, v->capacity() * 2)));
+        cursor = v->data() + written;
+        limit = v->data() + v->size();
     }
 
     /// append @a length (at least max_reserve_size) characters that do not fit
@@ -181,20 +265,24 @@ class output_vector_sink
     JSON_HEDLEY_NEVER_INLINE
     void append(const CharType* s, const std::size_t length)
     {
-        v->resize(position);
-        if (v->capacity() - position < length)
+        const std::size_t written = position();
+        v->resize(written);
+        if (v->capacity() - written < length)
         {
             // grow geometrically, with room for the writes that follow, so
             // that the next grow() need not reallocate right away
-            v->reserve((std::max)(v->capacity() * 2, position + length + max_reserve_size));
+            v->reserve((std::max)(v->capacity() * 2, written + length + max_reserve_size));
         }
         v->insert(v->end(), s, s + length);
-        position = v->size();
+        cursor = v->data() + v->size();
+        limit = cursor;
     }
 
     std::vector<CharType, AllocatorType>* v;
-    /// number of characters of *v written so far
-    std::size_t position;
+    /// the write position: in *v, at the number of characters written so far
+    CharType* cursor;
+    /// the end of *v
+    CharType* limit;
 };
 
 /// output adapter for byte vectors
@@ -277,9 +365,17 @@ class output_string_adapter : public output_adapter_protocol<CharType>
 ///
 /// Wraps the polymorphic output_adapter_t so the same binary_writer template can
 /// also target arbitrary adapters (output streams, strings, user-provided
-/// adapters) via the `output_adapter`-based overloads. Each write still goes
-/// through one virtual call, exactly as before; only the concrete sinks above
-/// avoid it.
+/// adapters) via the `output_adapter`-based overloads.
+///
+/// A write through the adapter is a virtual call, and for a stream a call with
+/// a sentry on top, which costs more than encoding the few bytes of most values.
+/// The sink therefore collects what is written in a buffer and passes it on in
+/// one piece when it is full, a write is too large for it, or @ref flush is
+/// called. Everything written reaches the adapter in the same order and with
+/// the same bytes; the destructor passes on what is left, so that a write that
+/// throws leaves the adapter with exactly the bytes written before it. Only
+/// @ref flush can report an error of the adapter (the destructor must not
+/// throw), so it has to be called once the value is written.
 template<typename CharType>
 class output_adapter_sink
 {
@@ -290,36 +386,121 @@ class output_adapter_sink
         JSON_ASSERT(oa);
     }
 
+    output_adapter_sink(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(output_adapter_sink&&) = delete;
+
+    output_adapter_sink(output_adapter_sink&& other) noexcept
+        : oa(std::move(other.oa)), used(other.used)
+    {
+        std::memcpy(buffer.data(), other.buffer.data(), other.used * sizeof(CharType));
+        other.used = 0;
+    }
+
+    ~output_adapter_sink()
+    {
+        // a normal return has flushed already, so this is for a write that threw
+        if (used != 0)
+        {
+            JSON_TRY
+            {
+                flush();
+            }
+            JSON_CATCH(...)
+            {
+                // nothing can be done about an adapter that throws here
+            }
+        }
+    }
+
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_character(CharType c)
     {
-        oa->write_character(c);
+        if (JSON_HEDLEY_UNLIKELY(used == buffer_size))
+        {
+            flush();
+        }
+        buffer[used++] = c;
     }
 
     // no JSON_HEDLEY_NON_NULL: forwards (null, 0) for empty payloads, exactly as
     // the type-erased path already did before this sink existed
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_characters(const CharType* s, std::size_t length)
     {
-        oa->write_characters(s, length);
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            write_large(s, length);
+            return;
+        }
+        if (length <= 32)
+        {
+            copy_short(buffer.data() + used, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            used += length;
+        }
+        else
+        {
+            std::memcpy(buffer.data() + used, s, length * sizeof(CharType));
+            used += length;
+        }
     }
 
     /// space for writing up to @a length characters (see
     /// output_vector_sink::reserve)
-    CharType* reserve(const std::size_t length) noexcept
+    JSON_HEDLEY_ALWAYS_INLINE
+    CharType* reserve(const std::size_t length)
     {
         JSON_ASSERT(length <= max_reserve_size);
-        static_cast<void>(length);
-        return reserved.data();
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            flush();
+        }
+        return buffer.data() + used;
     }
 
-    /// write the first @a length characters written to the space from reserve()
-    void commit(const std::size_t length)
+    /// keep the first @a length characters written to the space from reserve()
+    JSON_HEDLEY_ALWAYS_INLINE
+    void commit(const std::size_t length) noexcept
     {
-        oa->write_characters(reserved.data(), length);
+        used += length;
+    }
+
+    /// pass everything written so far on to the adapter
+    void flush()
+    {
+        if (used != 0)
+        {
+            // reset first: the characters count as written even if the adapter throws
+            const std::size_t length = used;
+            used = 0;
+            oa->write_characters(buffer.data(), length);
+        }
     }
 
   private:
+    /// a write that does not fit the space left in the buffer
+    JSON_HEDLEY_NEVER_INLINE
+    void write_large(const CharType* s, const std::size_t length)
+    {
+        flush();
+        if (length >= buffer_size)
+        {
+            oa->write_characters(s, length);
+        }
+        else
+        {
+            std::memcpy(buffer.data(), s, length * sizeof(CharType));
+            used = length;
+        }
+    }
+
+    /// large enough for reserve() at any time
+    static constexpr std::size_t buffer_size = 4 * max_reserve_size;
+
     output_adapter_t<CharType> oa;
-    std::array<CharType, max_reserve_size> reserved{};
+    /// number of characters of @ref buffer not yet passed on
+    std::size_t used = 0;
+    std::array<CharType, buffer_size> buffer{};
 };
 
 /// @brief whether std::basic_string<CharType> has a non-deprecated std::char_traits

include/nlohmann/detail/view/builder.hpp

@@ -275,11 +275,34 @@ class builder
                 0, 0, '\b', 0, 0, 0, '\f', 0, 0, 0, 0, 0, 0, 0, '\n', 0, 0, 0, '\r', 0, '\t', 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
             }
         };
+        // The write position is kept in locals: the stores of characters into
+        // the arena may alias the members of the builder, which would then be
+        // reloaded after every one. It is stored back where the arena grows,
+        // and by `write_back` on every return.
+        const unsigned char* const end = m_end;
         const std::size_t start = arena_used();
-        arena_run(s, static_cast<std::size_t>(p - s));
+        char* w = m_arena_write;
+        const char* w_end = m_arena_end;
+        const arena_write_back write_back(*this, w);
+        {
+            const auto n = static_cast<std::size_t>(p - s);
+            if (n <= 16 && end - s >= 16)
+            {
+                arena_ensure(w, w_end, 16);
+                std::memcpy(w, s, 16);
+            }
+            else if (n != 0)
+            {
+                // (an empty run leaves an empty arena unallocated, and memcpy()
+                // must not be passed its null pointer even for zero bytes)
+                arena_ensure(w, w_end, n);
+                std::memcpy(w, s, n);
+            }
+            w += n;
+        }
         for (;;)
         {
-            if (p == m_end)
+            if (p == end)
             {
                 return failed(error_code::string_missing_quote, p);
             }
@@ -287,7 +310,9 @@ class builder
             if (c == '"')
             {
                 ++p;
-                return decoded{p, start, arena_used() - start};
+                // (from w: `write_back` stores it only after the return value
+                // is made; the arena is allocated, since an escape was written)
+                return decoded{p, start, static_cast<std::size_t>(w - m_doc.arena.data()) - start};
             }
             if (c != '\\')
             {
@@ -297,7 +322,7 @@ class builder
                 return failed(c < 0x20 ? error_code::string_control_character : error_code::string_utf8, p);
             }
             ++p;
-            if (p == m_end)
+            if (p == end)
             {
                 return failed(error_code::string_missing_quote, p);
             }
@@ -315,7 +340,7 @@ class builder
                     {
                         return failed(error_code::string_surrogate_low, p);
                     }
-                    if (m_end - p < 2 || p[0] != '\\' || p[1] != 'u')
+                    if (end - p < 2 || p[0] != '\\' || p[1] != 'u')
                     {
                         return failed(error_code::string_surrogate_high, p);
                     }
@@ -331,26 +356,106 @@ class builder
                     }
                     cp = 0x10000 + ((cp - 0xD800) << 10) + (lo - 0xDC00);
                 }
-                arena_ensure(4);
-                m_arena_write = put_utf8(m_arena_write, cp);
+                arena_ensure(w, w_end, 4);
+                w = put_utf8(w, cp);
             }
             else if (NLOHMANN_VIEW_LIKELY(d < 128 && simple_escape[d] != 0))
             {
-                arena_ensure(1);
-                *m_arena_write++ = simple_escape[d];
+                arena_ensure(w, w_end, 1);
+                *w++ = simple_escape[d];
             }
             else
             {
                 --p;
                 return failed(error_code::string_escape, p);
             }
-            if (p != m_end && *p == '\\')
+#if NLOHMANN_VIEW_NEON
+            if (p != end && *p == '\\')
             {
                 continue; // consecutive escapes ("\u00e4\u00f6"): no run in between
             }
+            // The first 16 bytes of the run are checked one by one, in predicted branches, and copied
+            // as one block: where the runs between escapes are short, the vector step below would put
+            // its longer dependency chain (load, compare, mask, count, add) on p for each of them,
+            // and be slower than these steps (the scanner of the keys works the same way)
+            {
+                if (end - p >= 16)
+                {
+                    const std::uint8_t* const plain = string_plain();
+                    std::size_t steps = 16;
+#define NLOHMANN_VIEW_STEP(i) if (NLOHMANN_VIEW_LIKELY(plain[p[i]] != 0)) {} else { steps = (i); goto run_found; }
+                    NLOHMANN_VIEW_REPEAT16(NLOHMANN_VIEW_STEP)
+#undef NLOHMANN_VIEW_STEP
+run_found:
+                    arena_ensure(w, w_end, 16);
+                    std::memcpy(w, p, 16); // (only the first `steps` bytes are kept)
+                    w += steps;
+                    p += steps;
+                    if (steps < 16)
+                    {
+                        if (*p >= 0x80)
+                        {
+                            // non-ASCII: the UTF-8 check, and the rest of the run
+                            const unsigned char* const r = p;
+                            p = scan_string_run(p, end);
+                            const auto n = static_cast<std::size_t>(p - r);
+                            arena_ensure(w, w_end, n);
+                            std::memcpy(w, r, n);
+                            w += n;
+                        }
+                        continue;
+                    }
+                }
+            }
+#else
+            if (p != end && (*p == '\\' || *p == '"'))
+            {
+                continue; // consecutive escapes ("\u00e4\u00f6"), or the end of the string: no run in between
+            }
+#endif
+#if NLOHMANN_VIEW_VECTOR
+            // the plain bytes up to the next quote, backslash, or control
+            // character are copied as they are scanned, 16 bytes at a time
+            for (;;)
+            {
+                if (NLOHMANN_VIEW_UNLIKELY(end - p < 16))
+                {
+                    const unsigned char* const r = p;
+                    p = scan_string_run(p, end);
+                    const auto n = static_cast<std::size_t>(p - r);
+                    arena_ensure(w, w_end, n);
+                    std::memcpy(w, r, n);
+                    w += n;
+                    break;
+                }
+                arena_ensure(w, w_end, 16);
+                const unsigned k = copy_plain_run16(p, w);
+                w += k;
+                p += k;
+                if (k == 16)
+                {
+                    continue;
+                }
+                if (NLOHMANN_VIEW_UNLIKELY(*p >= 0x80))
+                {
+                    // non-ASCII: the UTF-8 check, and the rest of the run
+                    const unsigned char* const r = p;
+                    p = scan_string_run(p, end);
+                    const auto n = static_cast<std::size_t>(p - r);
+                    arena_ensure(w, w_end, n);
+                    std::memcpy(w, r, n);
+                    w += n;
+                }
+                break;
+            }
+#else
             const unsigned char* const r = p;
-            p = scan_string_run(p, m_end);
-            arena_run(r, static_cast<std::size_t>(p - r));
+            p = scan_string_run(p, end);
+            const auto n = static_cast<std::size_t>(p - r);
+            arena_ensure(w, w_end, n);
+            std::memcpy(w, r, n);
+            w += n;
+#endif
         }
     }
 
@@ -392,19 +497,49 @@ class builder
     char* m_arena_write = nullptr;
     char* m_arena_end = nullptr;
 
+    /// room for n more bytes at w, the write position kept in a local, with
+    /// w_end the end of the arena. (A lambda instead is not always inlined,
+    /// and then w would be kept in memory, as m_arena_write is.)
+    NLOHMANN_VIEW_ALWAYS_INLINE void arena_ensure(char*& w, const char*& w_end, std::size_t n)
+    {
+        if (NLOHMANN_VIEW_UNLIKELY(static_cast<std::size_t>(w_end - w) < n))
+        {
+            m_arena_write = w;
+            arena_grow(n);
+            w = m_arena_write;
+            w_end = m_arena_end;
+        }
+    }
+
+    /// stores a write position kept in a local back into m_arena_write when
+    /// the function that keeps it returns, on every path
+    class arena_write_back
+    {
+      public:
+        arena_write_back(builder& b, char*& w) noexcept
+            : m_builder(b)
+            , m_w(w)
+        {}
+        ~arena_write_back()
+        {
+            m_builder.m_arena_write = m_w;
+        }
+
+        arena_write_back(const arena_write_back&) = delete;
+        arena_write_back& operator=(const arena_write_back&) = delete;
+        arena_write_back(arena_write_back&&) = delete;
+        arena_write_back& operator=(arena_write_back&&) = delete;
+
+      private:
+        builder& m_builder;
+        char*& m_w;
+    };
+
     NLOHMANN_VIEW_ALWAYS_INLINE std::size_t arena_used() const noexcept
     {
         return m_arena_write != nullptr ? static_cast<std::size_t>(m_arena_write - m_doc.arena.data()) : 0;
     }
 
-    NLOHMANN_VIEW_ALWAYS_INLINE void arena_ensure(std::size_t n)
-    {
-        if (NLOHMANN_VIEW_UNLIKELY(static_cast<std::size_t>(m_arena_end - m_arena_write) < n))
-        {
-            arena_grow(n);
-        }
-    }
-
     NLOHMANN_VIEW_NOINLINE void arena_grow(std::size_t n)
     {
         const std::size_t used = arena_used();
@@ -413,25 +548,6 @@ class builder
         m_arena_end = m_doc.arena.data() + m_doc.arena.size();
     }
 
-    /// append the run [r, r + n) to the arena; short runs as one fixed-size
-    /// 16-byte move when both sides have the room (no library call)
-    NLOHMANN_VIEW_ALWAYS_INLINE void arena_run(const unsigned char* r, std::size_t n)
-    {
-        if (n <= 16 && m_end - r >= 16)
-        {
-            arena_ensure(16);
-            std::memcpy(m_arena_write, r, 16);
-        }
-        else if (n != 0)
-        {
-            // (an empty run leaves an empty arena unallocated, and memcpy()
-            // must not be passed its null pointer even for zero bytes)
-            arena_ensure(n);
-            std::memcpy(m_arena_write, r, n);
-        }
-        m_arena_write += n;
-    }
-
     /// UTF-8 encoding of cp at w (room for 4 bytes); returns the position after
     /// it. (A returned pointer stays in a register; writing through a
     /// reference to m_arena_write would store and reload it around every byte.)
@@ -954,11 +1070,11 @@ indent_done:
                 ++p;
             }
             const unsigned char* const int_start = p;
-            if (p != e && *p == '0')
+            if (cur() == '0')
             {
                 ++p;
             }
-            else if (NLOHMANN_VIEW_LIKELY(p != e && *p >= '1' && *p <= '9'))
+            else if (NLOHMANN_VIEW_LIKELY(cur() >= '1' && cur() <= '9'))
             {
                 p = skip_digits(p + 1, e);
             }
@@ -969,7 +1085,7 @@ indent_done:
             const auto int_digits = static_cast<std::size_t>(p - int_start);
             std::size_t frac_digits = 0;
             bool is_float = false;
-            if (p != e && *p == '.')
+            if (cur() == '.')
             {
                 ++p;
                 const unsigned char* const f0 = p;
@@ -982,20 +1098,20 @@ indent_done:
                 is_float = true;
             }
             std::int64_t exponent = 0;
-            if (p != e && (*p | 0x20) == 'e')
+            if ((cur() | 0x20) == 'e')
             {
                 ++p;
                 bool exp_negative = false;
-                if (p != e && (*p == '+' || *p == '-'))
+                if (cur() == '+' || cur() == '-')
                 {
                     exp_negative = *p == '-';
                     ++p;
                 }
-                if (NLOHMANN_VIEW_UNLIKELY(p == e || !is_digit(*p)))
+                if (NLOHMANN_VIEW_UNLIKELY(!is_digit(cur())))
                 {
                     return fail(error_code::number_after_exponent);
                 }
-                while (p != e && is_digit(*p))
+                while (is_digit(cur()))
                 {
                     if (exponent < 100000)
                     {
@@ -1057,7 +1173,7 @@ indent_done:
             ++p; // opening quote
             const unsigned char* const s = p;
             p = scan_string_run<Value>(p, e);
-            if (NLOHMANN_VIEW_LIKELY(p != e && *p == '"'))
+            if (NLOHMANN_VIEW_LIKELY(cur() == '"'))
             {
                 push_node(value_t::string, 0, 0, static_cast<std::size_t>(s - b), static_cast<std::uint64_t>(p - s));
                 ++p;

include/nlohmann/detail/view/input.hpp

@@ -10,6 +10,9 @@
 
 #include <algorithm> // max
 #include <cstddef> // size_t
+#ifndef JSON_NO_IO
+    #include <cstdio> // FILE, fread
+#endif
 #include <istream> // istream, streambuf
 #include <string> // basic_string, char_traits, string
 #include <type_traits> // decay, integral_constant, is_array, is_const, is_integral, is_lvalue_reference, is_pointer, is_same, remove_cv, remove_extent, remove_pointer, remove_reference
@@ -154,6 +157,37 @@ inline void append_stream(std::string& buf, std::istream& is)
     is.clear(is.rdstate() | std::ios::eofbit);
 }
 
+/// a pointer to a C stream (never with JSON_NO_IO, which leaves out <cstdio>)
+template<typename T>
+struct is_c_stream
+#ifndef JSON_NO_IO
+    : std::is_same<T, std::FILE*>
+#else
+    : std::false_type
+#endif
+{};
+
+#ifndef JSON_NO_IO
+/// append the rest of a C stream: in blocks with fread, not character by
+/// character (the library's file adapter reads with fgetc);
+/// f is not null
+inline void append_file(std::string& buf, std::FILE* f)
+{
+    for (;;)
+    {
+        const std::size_t used = buf.size();
+        const std::size_t room = (std::max)(buf.capacity() - used, std::size_t{4096});
+        buf.resize(used + room);
+        const std::size_t got = std::fread(&buf[used], 1, room, f);
+        buf.resize(used + got);
+        if (got < room)
+        {
+            break; // end of file, or an error (as fgetc, which returns EOF for both)
+        }
+    }
+}
+#endif
+
 }  // namespace view
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END

include/nlohmann/detail/view/iterator.hpp

@@ -147,20 +147,21 @@ class view_item
 
     view_item(const iterator& it, std::size_t index)
         : m_it(it)
-    {
-        if (!it.is_object_iterator())
-        {
-            m_index = std::to_string(index);
-        }
-    }
+        , m_position(index)
+    {}
 
-    /// the member key, or the element index for arrays
+    /// the member key, or the element index for arrays (as text made on the
+    /// first call, so that iterating over the values pays nothing for it)
     string_view_t key() const
     {
         if (m_it.is_object_iterator())
         {
             return m_it.key();
         }
+        if (m_index.empty())
+        {
+            m_index = std::to_string(m_position);
+        }
         return string_view_t(m_index.data(), m_index.size());
     }
 
@@ -183,7 +184,8 @@ class view_item
 
   private:
     iterator m_it;
-    std::string m_index{}; // NOLINT(readability-redundant-member-init)
+    std::size_t m_position; ///< the index of an array element
+    mutable std::string m_index{}; ///< the text of m_position, once key() was called // NOLINT(readability-redundant-member-init)
 };
 
 /// the range returned by basic_json_view::items()

include/nlohmann/detail/view/lookup.hpp

@@ -166,6 +166,10 @@ const node* element_at(const document_data& d, const node* array, std::size_t id
     {
         return e + idx; // a growable block: one link per element
     }
+    if (array->next == array->len + 1u && (!nav::editable() || (array->flags & node_flags::moved) == 0))
+    {
+        return e + idx; // no element has a subtree: the elements are the nodes after the array
+    }
     for (std::size_t i = 0; i < idx; ++i)
     {
         e = document_data::after(e);
@@ -181,6 +185,10 @@ const node* last_child(const document_data& d, const node* container) noexcept
     const std::size_t value_offset = container->kind == static_cast<std::uint8_t>(value_t::object) ? 1 : 0;
     const node* const end = navigation<Editable>::end(d, container);
     const node* last = navigation<Editable>::first(d, container);
+    if (value_offset == 0 && container->next == container->len + 1u && (!navigation<Editable>::editable() || (container->flags & node_flags::moved) == 0))
+    {
+        return last + (container->len - 1u); // an array without subtrees: the elements are the nodes after it
+    }
     for (const node* c = document_data::after(last + value_offset); c != end; c = document_data::after(c + value_offset))
     {
         last = c;

include/nlohmann/detail/view/materialize.hpp

@@ -8,7 +8,7 @@
 
 #pragma once
 
-#include <algorithm> // max
+#include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // int64_t, uint8_t
 #include <string> // string
@@ -40,12 +40,13 @@ template<typename BasicJsonType, bool Editable>
 BasicJsonType materialize(const document_data& d, const node* n)
 {
     using string_t = typename BasicJsonType::string_t;
-    using sax_t = json_sax_dom_parser<BasicJsonType, iterator_input_adapter<const char*>>;
+    using sax_t = json_sax_dom_parser<BasicJsonType, iterator_input_adapter<const char*>, true>;
     using nav = navigation<Editable>;
 
     // the open containers: their next element (or key of the next member),
-    // their end, and whether they are objects; sized once from the
-    // document's deepest nesting (for a subtree an upper bound)
+    // their end, and whether they are objects; for the first 16 levels on the
+    // stack, else sized once from the document's deepest nesting (for a
+    // subtree an upper bound)
     struct frame
     {
         const node* pos;
@@ -53,10 +54,43 @@ BasicJsonType materialize(const document_data& d, const node* n)
         bool object;
     };
 
+    // a scalar needs neither the handler nor the stack of open containers
+    switch (static_cast<value_t>(n->kind))
+    {
+        case value_t::string:
+            return BasicJsonType(string_t(d.str(*n), n->len));
+        case value_t::number_integer:
+            return BasicJsonType(static_cast<typename BasicJsonType::number_integer_t>(static_cast<std::int64_t>(integer_bits(*n))));
+        case value_t::number_unsigned:
+            return BasicJsonType(static_cast<typename BasicJsonType::number_unsigned_t>(integer_bits(*n)));
+        case value_t::number_float:
+            return BasicJsonType(float_value<typename BasicJsonType::number_float_t>(d, *n));
+        case value_t::boolean:
+            return BasicJsonType((n->flags & node_flags::is_true) != 0);
+        case value_t::null:
+            return BasicJsonType(nullptr);
+        case value_t::object:
+        case value_t::array:
+        case value_t::binary:
+        case value_t::discarded:
+        default:
+            break;
+    }
+
     BasicJsonType result;
     sax_t sax(result, true);
     const string_t no_token{};
-    std::vector<frame> open((std::max)(d.max_depth, std::size_t{1}));
+    // (shallow subtrees on the stack: a call for a small value then allocates only for the value)
+    std::array<frame, 16> shallow; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): filled as containers open
+    std::vector<frame> deep{};
+    frame* open = shallow.data();
+    std::size_t capacity = shallow.size();
+    if (d.max_depth > capacity)
+    {
+        deep.resize(d.max_depth);
+        open = deep.data();
+        capacity = deep.size();
+    }
     std::size_t depth = 0;
     for (;;)
     {
@@ -76,9 +110,13 @@ BasicJsonType materialize(const document_data& d, const node* n)
                 {
                     sax.start_array(n->len);
                 }
-                if (NLOHMANN_VIEW_UNLIKELY(depth == open.size()))
+                if (NLOHMANN_VIEW_UNLIKELY(depth == capacity))
                 {
-                    open.resize(depth * 2); // (deeper than the recorded nesting, e.g. after edits)
+                    // (deeper than the recorded nesting, e.g. after edits)
+                    deep.assign(open, open + depth);
+                    deep.resize(depth * 2);
+                    open = deep.data();
+                    capacity = deep.size();
                 }
                 open[depth++] = frame{nav::first(d, n), nav::end(d, n), object};
                 break;

include/nlohmann/detail/view/object_index.hpp

@@ -47,27 +47,96 @@ constexpr std::size_t index_max_displacement = 64;
 static_assert(document_data::index_min_members >= 2 * index_max_displacement,
               "a table must be much larger than the probe window of a lookup");
 
-/// hash of a key: its bytes, eight at a time, in a fixed byte order
+/// hash of a key: its bytes, eight at a time (in a fixed byte order), where the
+/// last word may overlap the one before (so that no loop over single bytes is
+/// needed); a key of fewer than eight bytes is read as two overlapping words of
+/// four or two bytes (in the native byte order: the hash is not stored, and the
+/// same on both sides of a lookup)
 inline std::uint64_t key_hash(const char* s, std::size_t n) noexcept
 {
     std::uint64_t h = 0x9E3779B97F4A7C15u * (n + 1);
     const auto* p = reinterpret_cast<const unsigned char*>(s); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-    while (n >= 8)
-    {
-        h = (h ^ read_eight_bytes(p)) * 0xBF58476D1CE4E5B9u;
-        h ^= h >> 29u;
-        p += 8;
-        n -= 8;
-    }
     std::uint64_t w = 0;
-    for (std::size_t i = 0; i < n; ++i)
+    if (n >= 8)
     {
-        w |= static_cast<std::uint64_t>(p[i]) << (8u * i);
+        const unsigned char* const last = p + n - 8;
+        while (p < last)
+        {
+            h = (h ^ read_eight_bytes(p)) * 0xBF58476D1CE4E5B9u;
+            h ^= h >> 29u;
+            p += 8;
+        }
+        w = read_eight_bytes(last);
     }
-    h = (h ^ w) * 0x94D049BB133111EBu;
+    else if (n >= 4)
+    {
+        std::uint32_t head = 0;
+        std::uint32_t tail = 0;
+        std::memcpy(&head, p, 4);
+        std::memcpy(&tail, p + n - 4, 4);
+        w = static_cast<std::uint64_t>(head) | (static_cast<std::uint64_t>(tail) << 32u);
+    }
+    else if (n >= 2)
+    {
+        std::uint16_t head = 0;
+        std::uint16_t tail = 0;
+        std::memcpy(&head, p, 2);
+        std::memcpy(&tail, p + n - 2, 2);
+        w = static_cast<std::uint64_t>(head) | (static_cast<std::uint64_t>(tail) << 16u);
+    }
+    else if (n == 1)
+    {
+        w = p[0];
+    }
+    // (a product only carries bits upwards, and the slot is taken from the low
+    // bits: the high bits of the last word are folded down before the multiplication)
+    h = (h ^ w ^ (w >> 29u)) * 0x94D049BB133111EBu;
     return h ^ (h >> 31u);
 }
 
+/// whether the n <= 16 bytes at a and b are equal, as two overlapping words
+/// (the first and the last 8, 4, or 2 bytes) and without a call
+NLOHMANN_VIEW_ALWAYS_INLINE bool same_short_bytes(const unsigned char* a, const unsigned char* b, std::size_t n) noexcept
+{
+    if (n >= 8)
+    {
+        std::uint64_t x0 = 0;
+        std::uint64_t x1 = 0;
+        std::uint64_t y0 = 0;
+        std::uint64_t y1 = 0;
+        std::memcpy(&x0, a, 8);
+        std::memcpy(&x1, a + n - 8, 8);
+        std::memcpy(&y0, b, 8);
+        std::memcpy(&y1, b + n - 8, 8);
+        return x0 == y0 && x1 == y1;
+    }
+    if (n >= 4)
+    {
+        std::uint32_t x0 = 0;
+        std::uint32_t x1 = 0;
+        std::uint32_t y0 = 0;
+        std::uint32_t y1 = 0;
+        std::memcpy(&x0, a, 4);
+        std::memcpy(&x1, a + n - 4, 4);
+        std::memcpy(&y0, b, 4);
+        std::memcpy(&y1, b + n - 4, 4);
+        return x0 == y0 && x1 == y1;
+    }
+    if (n >= 2)
+    {
+        std::uint16_t x0 = 0;
+        std::uint16_t x1 = 0;
+        std::uint16_t y0 = 0;
+        std::uint16_t y1 = 0;
+        std::memcpy(&x0, a, 2);
+        std::memcpy(&x1, a + n - 2, 2);
+        std::memcpy(&y0, b, 2);
+        std::memcpy(&y1, b + n - 2, 2);
+        return x0 == y0 && x1 == y1;
+    }
+    return n == 0 || *a == *b;
+}
+
 /// build the table of a large object
 inline void build_object_index(document_data& d, node* obj)
 {
@@ -132,6 +201,27 @@ inline const node* find_indexed(const document_data& d, const node* obj, const c
     const document_data::object_index& ix = d.indexes[obj->extra - 1u];
     const std::uint32_t* const table = d.index_slots.data() + ix.start;
     std::size_t slot = conditional_static_cast<std::size_t>(key_hash(key, n)) & ix.mask;
+    const auto* const k_bytes = reinterpret_cast<const unsigned char*>(key); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+    // (the loop for longer keys comes first, as the code is laid out; it is not the more common case)
+    if (NLOHMANN_VIEW_UNLIKELY(n <= 16))
+    {
+        // (up to 16 bytes, the bytes are compared as two words, without a call)
+        for (std::size_t distance = 0; distance <= index_max_displacement; ++distance)
+        {
+            const std::uint32_t s = table[slot];
+            if (s == 0)
+            {
+                return nullptr;
+            }
+            const node* const k = obj + s;
+            if (k->len == n && same_short_bytes(reinterpret_cast<const unsigned char*>(d.str(*k)), k_bytes, n)) // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+            {
+                return k;
+            }
+            slot = (slot + 1) & ix.mask;
+        }
+        return nullptr;
+    }
     for (std::size_t distance = 0; distance <= index_max_displacement; ++distance)
     {
         const std::uint32_t s = table[slot];

include/nlohmann/detail/view/serializer.hpp

@@ -23,6 +23,7 @@
 #include <nlohmann/detail/view/macro_scope.hpp>
 #include <nlohmann/detail/view/node.hpp>
 #include <nlohmann/detail/view/number.hpp>
+#include <nlohmann/detail/view/simd.hpp>
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
@@ -80,13 +81,6 @@ class output_buffer
         put(s, N - 1);
     }
 
-    void put_repeated(char c, std::size_t n)
-    {
-        reserve(n);
-        std::memset(m_pos, c, n);
-        m_pos += n;
-    }
-
     /// the write position and the end of the writable space, for a writer
     /// that keeps the position in a local variable (set_cursor() hands it back)
     char* cursor() const noexcept
@@ -143,9 +137,27 @@ class output_buffer
 /// image_check::bounds can have others, which are then written unchanged).
 inline std::size_t plain_output_run(const unsigned char* s, std::size_t n) noexcept
 {
+    std::size_t i = 0;
+#if NLOHMANN_VIEW_SSE2
+    // (16 bytes per step; with NEON, the mask of a step costs more than it saves on the
+    // short runs that are typical here)
+    // NOLINTBEGIN(portability-simd-intrinsics)
+    for (; i + 16 <= n; i += 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(s + i))); // NOLINT(bugprone-casting-through-void)
+        // (a byte is below 0x20 if it is its own minimum with 0x1F)
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                             _mm_cmpeq_epi8(_mm_min_epu8(in, _mm_set1_epi8(0x1F)), in));
+        const auto mask = static_cast<std::uint64_t>(static_cast<unsigned>(_mm_movemask_epi8(special)));
+        if (mask != 0)
+        {
+            return i + static_cast<std::size_t>(count_trailing_zeros(mask));
+        }
+    }
+    // NOLINTEND(portability-simd-intrinsics)
+#endif
     constexpr std::uint64_t ones = 0x0101010101010101ull;
     constexpr std::uint64_t high = 0x8080808080808080ull;
-    std::size_t i = 0;
     for (; i + 8 <= n; i += 8)
     {
         const std::uint64_t v = read_eight_bytes(s + i);
@@ -196,6 +208,70 @@ inline bool is_ascii_without_delete(const char* s, std::size_t n) noexcept
     return true;
 }
 
+#if NLOHMANN_VIEW_VECTOR
+/// the bytes of the 16 at p that ensure_ascii cannot copy (a quote, a
+/// backslash, a control character, DEL, or not ASCII; a signed compare with
+/// 0x20 finds control characters and non-ASCII bytes at once), one bit per
+/// byte (x86) or one nibble per byte (NEON)
+NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t ascii_stop_mask(const unsigned char* p) noexcept
+{
+#if NLOHMANN_VIEW_NEON
+    const uint8x16_t in = vld1q_u8(p);
+    const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))),
+                                        vorrq_u8(vceqq_u8(in, vdupq_n_u8(0x7F)), vcltq_s8(vreinterpretq_s8_u8(in), vdupq_n_s8(0x20))));
+    return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(special), 4)), 0);
+#else
+    const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+    const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                         _mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8(0x7F)), _mm_cmplt_epi8(in, _mm_set1_epi8(0x20))));
+    return static_cast<std::uint64_t>(static_cast<unsigned>(_mm_movemask_epi8(special)));
+#endif
+}
+
+/// the width in bits of a byte in the mask of ascii_stop_mask()
+constexpr unsigned ascii_stop_bits() noexcept
+{
+    return NLOHMANN_VIEW_NEON ? 4u : 1u;
+}
+#endif
+
+/// As find_ascii_copyable_run(), 16 bytes per step. end is the end of the
+/// buffer that holds the n bytes at p (a short run is read as 16 bytes if
+/// they are in the buffer).
+inline std::size_t ascii_copyable_run(const unsigned char* p, std::size_t n, const unsigned char* end) noexcept
+{
+#if NLOHMANN_VIEW_VECTOR
+    if (n >= 16)
+    {
+        std::size_t i = 0;
+        for (; i + 16 <= n; i += 16)
+        {
+            const std::uint64_t mask = ascii_stop_mask(p + i);
+            if (mask != 0)
+            {
+                return i + (static_cast<std::size_t>(count_trailing_zeros(mask)) / ascii_stop_bits());
+            }
+        }
+        if (i == n)
+        {
+            return n;
+        }
+        // the last 16 bytes: those before i are known to be copyable
+        const std::uint64_t mask = ascii_stop_mask(p + n - 16);
+        return mask != 0 ? n - 16 + (static_cast<std::size_t>(count_trailing_zeros(mask)) / ascii_stop_bits()) : n;
+    }
+    if (end - p >= 16)
+    {
+        // (the bytes after the run are masked out)
+        const std::uint64_t mask = ascii_stop_mask(p) & ((std::uint64_t{1} << (n * ascii_stop_bits())) - 1u);
+        return mask != 0 ? static_cast<std::size_t>(count_trailing_zeros(mask)) / ascii_stop_bits() : n;
+    }
+#else
+    static_cast<void>(end);
+#endif
+    return find_ascii_copyable_run(p, n);
+}
+
 /// A stack that starts in a buffer of the caller (a local array) and moves to
 /// the heap (a vector of the caller) only when that is full, so that dumps of
 /// shallow documents need no allocation. The top is a pointer, as in
@@ -275,6 +351,39 @@ NLOHMANN_VIEW_ALWAYS_INLINE char* put_literal(char* w, const char (&s)[N]) noexc
     return w + N - 1;
 }
 
+/// the depth of the innermost open container, counted only for indented output
+template<bool Counted>
+struct depth_counter
+{
+    std::size_t value = 0;
+    void enter() noexcept
+    {
+        ++value;
+    }
+    std::size_t leave() noexcept
+    {
+        return --value;
+    }
+    std::size_t get() const noexcept
+    {
+        return value;
+    }
+};
+
+template<>
+struct depth_counter<false>
+{
+    void enter() noexcept {}
+    std::size_t leave() noexcept // NOLINT(readability-convert-member-functions-to-static): same interface as depth_counter<true>
+    {
+        return 0;
+    }
+    std::size_t get() const noexcept // NOLINT(readability-convert-member-functions-to-static): same interface as depth_counter<true>
+    {
+        return 0;
+    }
+};
+
 /// how the view's dump() writes a value
 struct dump_style
 {
@@ -311,139 +420,95 @@ class view_serializer
 
     void dump(const node* root)
     {
-        if (!m_style.pretty)
+        if (m_style.ensure_ascii)
         {
-            if (m_style.ensure_ascii)
-            {
-                dump_compact_ascii(root);
-            }
-            else if (m_style.source_numbers)
-            {
-                dump_compact<true, false>(root);
-            }
-            else
-            {
-                dump_compact<false, false>(root);
-            }
-            return;
+            dump_ascii(root);
         }
-        struct frame
+        else if (m_style.pretty)
         {
-            const node* pos; ///< next element, or key of the next member
-            const node* end;
-            bool object;
-            bool first;  ///< nothing written yet
-        };
-        std::array<frame, 32> buffer; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
-        std::vector<frame> heap;
-        small_stack<frame> stack(buffer.data(), buffer.size(), heap);
-        const node* n = root;
-        for (;;)
+            dump_pretty(root);
+        }
+        else if (m_style.source_numbers)
         {
-            // write the value at n
-            if (is_container(*n))
-            {
-                const bool object = n->kind == static_cast<std::uint8_t>(value_t::object);
-                if (n->len == 0)
-                {
-                    m_out.put(object ? "{}" : "[]", 2);
-                }
-                else
-                {
-                    m_out.put(object ? '{' : '[');
-                    stack.push_back(frame{nav::first(m_doc, n), nav::end(m_doc, n), object, true});
-                }
-            }
-            else
-            {
-                write_scalar(*n);
-            }
-
-            // go to the next value: close finished containers, then separate
-            for (;;)
-            {
-                if (stack.empty())
-                {
-                    m_out.finish();
-                    return;
-                }
-                if (stack.back().pos == stack.back().end)
-                {
-                    const bool object = stack.back().object;
-                    stack.pop_back();
-                    newline(stack.size());
-                    m_out.put(object ? '}' : ']');
-                    continue;
-                }
-                frame& f = stack.back();
-                if (!f.first)
-                {
-                    m_out.put(',');
-                }
-                f.first = false;
-                newline(stack.size());
-                if (f.object)
-                {
-                    write_string(*f.pos);
-                    if (m_style.pretty)
-                    {
-                        m_out.put(": ");
-                    }
-                    else
-                    {
-                        m_out.put(':');
-                    }
-                    n = nav::value(f.pos + 1);
-                    f.pos = document_data::after(f.pos + 1);
-                }
-                else
-                {
-                    n = nav::value(f.pos);
-                    f.pos = document_data::after(f.pos);
-                }
-                break;
-            }
+            dump_text<true, false, false>(root);
+        }
+        else
+        {
+            dump_text<false, false, false>(root);
         }
     }
 
   private:
+    /// dump_text() with ensure_ascii, out of line: the walk without it is
+    /// faster when it is not merged with this one into dump()
+    NLOHMANN_VIEW_NOINLINE void dump_ascii(const node* root)
+    {
+        if (m_style.pretty)
+        {
+            if (m_style.source_numbers)
+            {
+                dump_text<true, true, true>(root);
+            }
+            else
+            {
+                dump_text<false, true, true>(root);
+            }
+        }
+        else if (m_style.source_numbers)
+        {
+            dump_text<true, false, true>(root);
+        }
+        else
+        {
+            dump_text<false, false, true>(root);
+        }
+    }
+
+    /// dump_text() with indentation, out of line: the compact walk is faster
+    /// when it is not merged with the indented ones into dump()
+    NLOHMANN_VIEW_NOINLINE void dump_pretty(const node* root)
+    {
+        if (m_style.source_numbers)
+        {
+            dump_text<true, true, false>(root);
+        }
+        else
+        {
+            dump_text<false, true, false>(root);
+        }
+    }
+
     /// a string that is not a plain string of the source (decoded, or written
     /// by an edit), without ensure_ascii: runs without characters to escape
-    /// are copied
-    NLOHMANN_VIEW_NOINLINE void write_decoded(const node& n)
+    /// are copied. If validate is set, the string must be valid UTF-8: the
+    /// runs are found by the lexer's scan, which validates the sequences, and
+    /// invalid UTF-8 throws what basic_json::dump() throws. (The compact output
+    /// writes the bytes of a damaged image unchanged, and scans without
+    /// validating them.)
+    NLOHMANN_VIEW_NOINLINE void write_decoded(const node& n, bool validate)
     {
         const auto* const s = reinterpret_cast<const unsigned char*>(m_doc.str(n)); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
         m_out.put('"');
         for (std::size_t i = 0; i < n.len;)
         {
-            const std::size_t run = plain_output_run(s + i, n.len - i);
+            const std::size_t run = validate ? string_bulk_run(s + i, n.len - i) : plain_output_run(s + i, n.len - i);
             if (run != 0)
             {
                 m_out.put(reinterpret_cast<const char*>(s + i), run); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
                 i += run;
                 continue;
             }
+            if (validate && s[i] >= 0x80)
+            {
+                invalid_utf8(s, n.len); // (the scan stopped at ill-formed UTF-8)
+            }
             write_codepoint<false>(s[i], s + i, 1); // a quote, a backslash, or a control character
             ++i;
         }
         m_out.put('"');
     }
 
-    /// dump_compact() with ensure_ascii, out of line: the walk without it is
-    /// faster when it is not merged with this one into dump()
-    NLOHMANN_VIEW_NOINLINE void dump_compact_ascii(const node* root)
-    {
-        if (m_style.source_numbers)
-        {
-            dump_compact<true, true>(root);
-        }
-        else
-        {
-            dump_compact<false, true>(root);
-        }
-    }
-
-    /// whether dump_compact() copies a string or key from the source: one of
+    /// whether dump_text() copies a string or key from the source: one of
     /// the source, and with ensure_ascii one without bytes >= 0x7F (an overload
     /// each, so that the walk without ensure_ascii is compiled exactly as
     /// before: with GCC, a condition that only folds away later changed its
@@ -458,13 +523,155 @@ class view_serializer
         return (n.flags & node_flags::storage) == 0 && is_ascii_without_delete(src + n.off, n.len);
     }
 
-    /// a string with ensure_ascii (dump_compact()), out of line
-    NLOHMANN_VIEW_NOINLINE void write_ascii_string(const node& n)
+    /*!
+    @brief a string with ensure_ascii: the characters that are not printable ASCII as \\uXXXX
+    (a character beyond U+FFFF as a surrogate pair); strings of a document
+    are valid UTF-8, except in a damaged image loaded with image_check::bounds,
+    for which this throws what basic_json::dump() throws for the string
+    */
+    NLOHMANN_VIEW_NOINLINE void write_ascii(const node& n)
     {
-        write_string(n);
+        const auto* const s = reinterpret_cast<const unsigned char*>(m_doc.str(n)); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        const std::size_t size = n.len;
+        char* w = m_out.cursor();
+        char* lim = m_out.limit();
+        // room for k bytes
+        const auto room = [&](std::size_t k)
+        {
+            const auto left = static_cast<std::size_t>(lim - w);
+            if (NLOHMANN_VIEW_UNLIKELY(left < k))
+            {
+                m_out.set_cursor(w);
+                m_out.reserve(k);
+                w = m_out.cursor();
+                lim = m_out.limit();
+            }
+        };
+        static constexpr const char* hex = "0123456789abcdef";
+        // \\uXXXX
+        const auto u_escape = [&](std::uint32_t u)
+        {
+            w[0] = '\\';
+            w[1] = 'u';
+            w[2] = hex[(u >> 12u) & 0xFu];
+            w[3] = hex[(u >> 8u) & 0xFu];
+            w[4] = hex[(u >> 4u) & 0xFu];
+            w[5] = hex[u & 0xFu];
+            w += 6;
+        };
+        room(1);
+        *w++ = '"';
+        std::size_t i = 0;
+        // (a string of the source starts with a run that has nothing to escape, 16 bytes per step)
+        const std::size_t prefix = (n.flags & node_flags::storage) == 0 ? ascii_copyable_run(s, size, reinterpret_cast<const unsigned char*>(m_doc.src + m_doc.size)) : 0; // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        if (prefix != 0)
+        {
+            room(prefix);
+            std::memcpy(w, s, prefix);
+            w += prefix;
+            i = prefix;
+        }
+        while (i < size)
+        {
+            // (a character that needs an escape sequence is not worth a call)
+            const std::size_t run = is_ascii_copyable(s[i]) ? find_ascii_copyable_run(s + i, size - i) : 0;
+            if (run != 0)
+            {
+                room(run);
+                std::memcpy(w, s + i, run);
+                w += run;
+                i += run;
+                continue;
+            }
+            room(12);
+            if (s[i] >= 0x80)
+            {
+                const unsigned char lead = s[i];
+                // (the two-byte sequences of Latin text and the three-byte sequences of most other
+                // text, with continuation bytes of any value: the ranges that are restricted (E0,
+                // ED, and the sequences of four bytes) go through validate_one_utf8())
+                if (lead >= 0xE1 && lead <= 0xEF && lead != 0xED && size - i >= 3 && (s[i + 1] & 0xC0u) == 0x80 && (s[i + 2] & 0xC0u) == 0x80)
+                {
+                    u_escape(((lead & 0x0Fu) << 12u) | ((s[i + 1] & 0x3Fu) << 6u) | (s[i + 2] & 0x3Fu));
+                    i += 3;
+                    continue;
+                }
+                if (lead >= 0xC2 && lead <= 0xDF && size - i >= 2 && (s[i + 1] & 0xC0u) == 0x80)
+                {
+                    u_escape(((lead & 0x1Fu) << 6u) | (s[i + 1] & 0x3Fu));
+                    i += 2;
+                    continue;
+                }
+                const std::size_t len = validate_one_utf8(s + i, size - i);
+                if (NLOHMANN_VIEW_UNLIKELY(len == 0))
+                {
+                    m_out.set_cursor(w);
+                    invalid_utf8(s, size);
+                    return; // LCOV_EXCL_LINE (invalid_utf8 throws)
+                }
+                std::uint32_t codepoint = s[i] & (0xFFu >> (len + 1));
+                for (std::size_t k = 1; k < len; ++k)
+                {
+                    codepoint = (codepoint << 6u) | (s[i + k] & 0x3Fu);
+                }
+                i += len;
+                if (codepoint <= 0xFFFF)
+                {
+                    u_escape(codepoint);
+                }
+                else
+                {
+                    u_escape(0xD7C0u + (codepoint >> 10u));
+                    u_escape(0xDC00u + (codepoint & 0x3FFu));
+                }
+                continue;
+            }
+            // a quote, a backslash, or a control character (or DEL)
+            const unsigned char c = s[i++];
+            char escape = 0;
+            switch (c)
+            {
+                case 0x08:
+                    escape = 'b';
+                    break;
+                case 0x09:
+                    escape = 't';
+                    break;
+                case 0x0A:
+                    escape = 'n';
+                    break;
+                case 0x0C:
+                    escape = 'f';
+                    break;
+                case 0x0D:
+                    escape = 'r';
+                    break;
+                case 0x22:
+                    escape = '"';
+                    break;
+                case 0x5C:
+                    escape = '\\';
+                    break;
+                default:
+                    break;
+            }
+            if (escape != 0)
+            {
+                w[0] = '\\';
+                w[1] = escape;
+                w += 2;
+            }
+            else
+            {
+                u_escape(c);
+            }
+        }
+        room(1);
+        *w++ = '"';
+        m_out.set_cursor(w);
     }
 
-    /// the copies of dump_compact() that are not fixed-size moves (long
+    /// the copies of dump_text() that are not fixed-size moves (long
     /// strings, or near the end of the source); out of line, so that the
     /// compiler does not merge the fixed-size moves into this call
     NLOHMANN_VIEW_NOINLINE static void copy_long(char* to, const char* from, std::size_t n) noexcept
@@ -481,7 +688,7 @@ class view_serializer
     }
 
     /*!
-    @brief the compact output (the default dump(), or with ensure_ascii)
+    @brief the output, compact (the default dump()) or indented, with or without ensure_ascii
 
     The same walk as dump(), with the write position in a local variable
     (stores through char pointers would otherwise force a reload of the
@@ -489,11 +696,11 @@ class view_serializer
     the source copied by fixed-size moves of 32 bytes where the source has
     that many bytes left, instead of a library call per token. The buffer
     keeps 64 bytes of slack for the overshoot. With ensure_ascii (Ascii), a
-    string of the source is copied only if it has no byte >= 0x7F (see
-    is_ascii_without_delete()); any other string is written by write_string().
+    string of the source is copied only if it has nothing to escape (see
+    copied()); any other string is written by write_ascii().
     */
-    template<bool SourceNumbers, bool Ascii>
-    void dump_compact(const node* root)
+    template<bool SourceNumbers, bool Pretty, bool Ascii>
+    void dump_text(const node* root)
     {
         struct frame
         {
@@ -508,6 +715,16 @@ class view_serializer
         const char* const src_end = src + m_doc.size;
         char* w = m_out.cursor();
         char* lim = m_out.limit();
+        // (indented output) a line break and 63 indentation characters, from
+        // which the line break of a level of up to 63 characters is a copy
+        std::array<char, 64> ruler; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): filled below if used
+        if (Pretty)
+        {
+            ruler.fill(m_style.indent_char);
+            ruler[0] = '\n';
+        }
+        const std::size_t indent = m_style.indent;
+        depth_counter<Pretty> level; // depth of the innermost open container (indented output)
         // room for n bytes and the slack
         const auto room = [&](std::size_t n)
         {
@@ -539,6 +756,22 @@ class view_serializer
             }
             w += n;
         };
+        // (indented output) the line break and the indentation of a level
+        const auto newline = [&](std::size_t depth)
+        {
+            const std::size_t count = 1 + (depth * indent);
+            room(count);
+            if (count <= ruler.size())
+            {
+                std::memcpy(w, ruler.data(), ruler.size());
+            }
+            else
+            {
+                *w = '\n';
+                std::memset(w + 1, m_style.indent_char, count - 1);
+            }
+            w += count;
+        };
         // a string that is not a plain string of the source (out of line, so
         // that the cursor stays in a register here)
         const auto escaped = [&](const node & n)
@@ -546,11 +779,11 @@ class view_serializer
             m_out.set_cursor(w);
             if (enabled(Ascii))
             {
-                write_ascii_string(n);
+                write_ascii(n);
             }
             else
             {
-                write_decoded(n);
+                write_decoded(n, Pretty);
             }
             w = m_out.cursor();
             lim = m_out.limit();
@@ -651,6 +884,7 @@ class view_serializer
                     {
                         *w++ = object ? '{' : '[';
                         stack.push_back(cur);
+                        level.enter();
                         if (Editable)
                         {
                             cur = frame{nav::first(m_doc, n), nav::end(m_doc, n), object};
@@ -693,6 +927,10 @@ class view_serializer
                     {
                         break;
                     }
+                    if (Pretty)
+                    {
+                        newline(level.leave());
+                    }
                     room(1);
                     *w++ = cur.object ? '}' : ']';
                     cur = stack.back();
@@ -701,13 +939,17 @@ class view_serializer
                 room(1);
                 *w++ = ',';
             }
+            if (Pretty)
+            {
+                newline(level.get());
+            }
             const node* const at = Editable ? cur.pos : n;
             if (cur.object)
             {
                 const node& key = *at;
                 if (copied(key, src, std::integral_constant<bool, Ascii>()))
                 {
-                    room(key.len + 3);
+                    room(key.len + (Pretty ? 4 : 3));
                     *w++ = '"';
                     copy(src + key.off, key.len);
                     w[0] = '"';
@@ -717,9 +959,13 @@ class view_serializer
                 else
                 {
                     escaped(key);
-                    room(1);
+                    room(Pretty ? 2 : 1);
                     *w++ = ':';
                 }
+                if (Pretty)
+                {
+                    *w++ = ' ';
+                }
                 if (Editable)
                 {
                     n = nav::value(at + 1);
@@ -738,69 +984,6 @@ class view_serializer
         }
     }
 
-    void newline(std::size_t level)
-    {
-        if (m_style.pretty)
-        {
-            m_out.put('\n');
-            m_out.put_repeated(m_style.indent_char, level * m_style.indent);
-        }
-    }
-
-    void write_scalar(const node& n)
-    {
-        switch (static_cast<value_t>(n.kind))
-        {
-            case value_t::null:
-                m_out.put("null");
-                break;
-            case value_t::boolean:
-                if ((n.flags & node_flags::is_true) != 0)
-                {
-                    m_out.put("true");
-                }
-                else
-                {
-                    m_out.put("false");
-                }
-                break;
-            case value_t::string:
-                write_string(n);
-                break;
-            case value_t::number_integer:
-            case value_t::number_unsigned:
-            {
-                const char* const token = m_doc.str(n);
-                const std::uint32_t len = number_length(n);
-                if (!m_style.source_numbers && len == 2 && token[0] == '-' && token[1] == '0')
-                {
-                    m_out.put('0'); // parse() reads -0 as the integer 0
-                }
-                else
-                {
-                    m_out.put(token, len);
-                }
-                break;
-            }
-            case value_t::number_float:
-                if (m_style.source_numbers && (n.flags & node_flags::storage) != node_flags::edited)
-                {
-                    m_out.put(m_doc.str(n), n.len); // (a float set by an edit is written as with shortest)
-                }
-                else
-                {
-                    write_float_node(n);
-                }
-                break;
-            case value_t::object:    // LCOV_EXCL_LINE (containers are written by dump())
-            case value_t::array:     // LCOV_EXCL_LINE
-            case value_t::binary:    // LCOV_EXCL_LINE (not in a document)
-            case value_t::discarded: // LCOV_EXCL_LINE
-            default:                 // LCOV_EXCL_LINE
-                break;               // LCOV_EXCL_LINE
-        }
-    }
-
     /// a float node as dump() writes it
     void write_float_node(const node& n)
     {
@@ -911,72 +1094,6 @@ class view_serializer
         m_out.put(s.data(), s.size());
     }
 
-    void write_string(const node& n)
-    {
-        const char* const s = m_doc.str(n);
-        m_out.put('"');
-        if ((n.flags & node_flags::storage) == 0 && (!m_style.ensure_ascii || is_ascii_without_delete(s, n.len)))
-        {
-            // a string of the source without escape sequences has nothing to
-            // escape (with ensure_ascii, if it is printable ASCII)
-            m_out.put(s, n.len);
-        }
-        else if (m_style.ensure_ascii)
-        {
-            write_escaped<true>(reinterpret_cast<const unsigned char*>(s), n.len); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-        }
-        else
-        {
-            write_escaped<false>(reinterpret_cast<const unsigned char*>(s), n.len); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-        }
-        m_out.put('"');
-    }
-
-    /// as serializer::dump_escaped(); strings of a document are valid UTF-8,
-    /// except in a damaged image loaded with image_check::bounds, for which
-    /// this throws what basic_json::dump() throws for the string
-    template<bool EnsureAscii>
-    void write_escaped(const unsigned char* s, std::size_t n)
-    {
-        std::size_t i = 0;
-        while (i < n)
-        {
-            std::size_t run = 0;
-            if (!EnsureAscii)
-            {
-                run = string_bulk_run(s + i, n - i);
-            }
-            else if (is_ascii_copyable(s[i]))
-            {
-                run = find_ascii_copyable_run(s + i, n - i);
-            }
-            if (run != 0)
-            {
-                m_out.put(reinterpret_cast<const char*>(s + i), run); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
-                i += run;
-                continue;
-            }
-            std::uint32_t codepoint = s[i];
-            std::size_t len = 1;
-            if (codepoint >= 0x80)
-            {
-                len = validate_one_utf8(s + i, n - i);
-                if (NLOHMANN_VIEW_UNLIKELY(len == 0))
-                {
-                    invalid_utf8(s, n);
-                    return;
-                }
-                codepoint &= 0xFFu >> (len + 1);
-                for (std::size_t k = 1; k < len; ++k)
-                {
-                    codepoint = (codepoint << 6u) | (s[i + k] & 0x3Fu);
-                }
-            }
-            write_codepoint<EnsureAscii>(codepoint, s + i, len);
-            i += len;
-        }
-    }
-
     /// throw what basic_json::dump() throws for a string that is not valid UTF-8
     NLOHMANN_VIEW_NOINLINE static void invalid_utf8(const unsigned char* s, std::size_t n)
     {

include/nlohmann/detail/view/simd.hpp

@@ -116,6 +116,34 @@ NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* vector_plain_run(const unsigned
     }
     return p;
 }
+
+/*!
+@brief copy 16 bytes from p to w, and find the first quote, backslash,
+control character, or non-ASCII byte among them
+
+The bytes are copied whether or not they are all plain (the caller has room
+for 16 bytes and keeps only the plain ones).
+
+@return the offset of that byte, or 16 if there is none
+*/
+NLOHMANN_VIEW_ALWAYS_INLINE unsigned copy_plain_run16(const unsigned char* p, char* w) noexcept
+{
+#if NLOHMANN_VIEW_NEON
+    const uint8x16_t in = vld1q_u8(p);
+    vst1q_u8(reinterpret_cast<std::uint8_t*>(w), in); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+    const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))),
+                                        vcltq_s8(vreinterpretq_s8_u8(in), vdupq_n_s8(0x20)));
+    const std::uint64_t bits = vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(special), 4)), 0);
+    return bits != 0 ? static_cast<unsigned>(count_trailing_zeros(bits) >> 2u) : 16u;
+#else
+    const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+    _mm_storeu_si128(static_cast<__m128i*>(static_cast<void*>(w)), in); // NOLINT(bugprone-casting-through-void)
+    const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                         _mm_cmplt_epi8(in, _mm_set1_epi8(0x20)));
+    const auto bits = static_cast<std::uint64_t>(static_cast<unsigned>(_mm_movemask_epi8(special)));
+    return bits != 0 ? static_cast<unsigned>(count_trailing_zeros(bits)) : 16u;
+#endif
+}
 #endif
 
 #if NLOHMANN_VIEW_SSSE3_DISPATCH

include/nlohmann/json.hpp

@@ -19,6 +19,7 @@
 #define INCLUDE_NLOHMANN_JSON_HPP_
 
 #include <algorithm> // all_of, find, for_each, none_of
+#include <array> // array
 #include <cmath> // isnan
 #include <cstddef> // nullptr_t, ptrdiff_t, size_t
 #include <cstdint> // uint8_t
@@ -158,9 +159,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     friend class ::nlohmann::detail::binary_writer;
     template<typename BasicJsonType, typename InputType, typename SAX>
     friend class ::nlohmann::detail::binary_reader;
-    template<typename BasicJsonType, typename InputAdapterType>
+    template<typename BasicJsonType, typename InputAdapterType, bool Collect>
     friend class ::nlohmann::detail::json_sax_dom_parser;
     template<typename BasicJsonType, typename InputAdapterType>
+    friend class ::nlohmann::detail::json_sax_dom_binary_parser;
+    template<typename BasicJsonType, typename InputAdapterType>
     friend class ::nlohmann::detail::json_sax_dom_callback_parser;
 #if JSON_DIAGNOSTIC_POSITIONS
     friend struct ::nlohmann::detail::diagnostic_positions;
@@ -6287,7 +6290,7 @@ public:
            from_ubjson/from_bjdata/from_bon8/from_bson overloads
 
     Building the @ref binary_reader as a named local, rather than as a
-    temporary that @ref detail::json_sax_dom_parser::parse is called on in
+    temporary that @ref detail::json_sax_dom_binary_parser::parse is called on in
     the same expression, avoids a false-positive cppcheck accessMoved
     warning in each of the 16 callers.
     */
@@ -6298,7 +6301,7 @@ public:
                                        const cbor_tag_handler_t tag_handler = cbor_tag_handler_t::error)
     {
         basic_json result;
-        detail::json_sax_dom_parser<basic_json, InputAdapterType> sdp(result, allow_exceptions);
+        detail::json_sax_dom_binary_parser<basic_json, InputAdapterType> sdp(result, allow_exceptions);
         binary_reader<InputAdapterType> reader(std::move(ia), format, error_handler);
         if (!reader.sax_parse(&sdp, strict, tag_handler))
         {
@@ -6321,7 +6324,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_cbor(j);
         return result;
     }
@@ -6331,7 +6334,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_cbor(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a CBOR serialization of a given JSON value
@@ -6339,7 +6344,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_cbor(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6348,7 +6355,7 @@ public:
             const error_handler_t error_handler = error_handler_t::keep)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_msgpack(j);
         return result;
     }
@@ -6358,7 +6365,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_msgpack(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6366,7 +6375,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<char> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<char>(o, error_handler).write_msgpack(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6377,7 +6388,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type);
         return result;
     }
@@ -6388,7 +6399,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6397,7 +6410,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6409,7 +6424,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
         return result;
     }
@@ -6421,7 +6436,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6431,7 +6448,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6440,7 +6459,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_bson(j);
         return result;
     }
@@ -6450,7 +6469,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_bson(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6458,7 +6479,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_bson(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
@@ -6466,7 +6489,7 @@ public:
     static std::vector<std::uint8_t> to_bon8(const basic_json& j)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result).write_bon8(j);
         return result;
     }
@@ -6475,14 +6498,18 @@ public:
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<std::uint8_t> o)
     {
-        binary_writer<std::uint8_t>(o).write_bon8(j);
+        binary_writer<std::uint8_t> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<char> o)
     {
-        binary_writer<char>(o).write_bon8(j);
+        binary_writer<char> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a JSON value from an input in CBOR format
@@ -6878,6 +6905,72 @@ public:
     /// @name JSON Patch functions
     /// @{
 
+  private:
+    /// the members of an operation of a JSON patch that are of interest
+    struct patch_members
+    {
+        basic_json* op = nullptr;
+        basic_json* path = nullptr;
+        basic_json* value = nullptr;
+        basic_json* from = nullptr;
+    };
+
+    /// @brief find the members "op", "path", "value", and "from" of the object @a val in one pass
+    static patch_members find_patch_members(const basic_json& val, std::true_type /*key_type is string_t*/)
+    {
+        patch_members members;
+        for (auto& member : *val.m_data.m_value.object)
+        {
+            const string_t& key = member.first;
+            switch (key.size())
+            {
+                case 2:
+                    if (key == "op")
+                    {
+                        members.op = &member.second;
+                    }
+                    break;
+                case 4:
+                    if (key == "path")
+                    {
+                        members.path = &member.second;
+                    }
+                    else if (key == "from")
+                    {
+                        members.from = &member.second;
+                    }
+                    break;
+                case 5:
+                    if (key == "value")
+                    {
+                        members.value = &member.second;
+                    }
+                    break;
+                default:
+                    break;
+            }
+        }
+        return members;
+    }
+
+    /// @brief find the members by lookup for key types other than string_t
+    static patch_members find_patch_members(const basic_json& val, std::false_type /*key_type is not string_t*/)
+    {
+        const auto find_member = [&val](const char* name) -> basic_json *
+        {
+            const auto it = val.m_data.m_value.object->find(string_t(name));
+            return it != val.m_data.m_value.object->end() ? &it->second : nullptr;
+        };
+
+        patch_members members;
+        members.op = find_member("op");
+        members.path = find_member("path");
+        members.value = find_member("value");
+        members.from = find_member("from");
+        return members;
+    }
+
+  public:
     /// @brief applies a JSON patch in-place without copying the object
     /// @sa https://json.nlohmann.me/api/basic_json/patch/
     void patch_inplace(const basic_json& json_patch)
@@ -7051,18 +7144,18 @@ public:
         // a diff creates them: apply the operation and return true; return
         // false, with nothing changed, for all others
         string_t token;
-        const string_t value_key = "value";
-        const auto try_plain = [&result, &token, &value_key](patch_operations operation, const string_t& path, const basic_json & val)
+        typename json_pointer::template plain_steps<basic_json> steps;
+        const auto try_plain = [&result, &token, &steps](patch_operations operation, const string_t& path, const patch_members & members, const basic_json & val)
         {
-            if (path.empty() || path.find_first_of('~') != string_t::npos)
+            if (path.empty())
             {
                 return false;
             }
 
             // a missing "value" is reported below
-            const auto value_it = val.m_data.m_value.object->find(value_key);
-            const basic_json& value = value_it != val.m_data.m_value.object->end() ? value_it->second : val;
-            if (operation != patch_operations::remove && value_it == val.m_data.m_value.object->end())
+            const basic_json* const value_ptr = members.value;
+            const basic_json& value = value_ptr != nullptr ? *value_ptr : val;
+            if (operation != patch_operations::remove && value_ptr == nullptr)
             {
                 return false;
             }
@@ -7071,7 +7164,7 @@ public:
             {
                 case patch_operations::replace:
                 {
-                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                     if (target == nullptr)
                     {
                         return false;
@@ -7082,7 +7175,7 @@ public:
 
                 case patch_operations::test:
                 {
-                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                     // a failing test is reported below
                     return target != nullptr && *target == value;
                 }
@@ -7102,13 +7195,17 @@ public:
                     }
                     --slash;
 
-                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token);
+                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token, steps);
                     if (parent == nullptr)
                     {
                         return false;
                     }
                     token.clear();
                     token.append(path.data() + slash + 1, path.size() - slash - 1);
+                    if (token.find_first_of('~') != string_t::npos)
+                    {
+                        return false;
+                    }
 
                     if (operation == patch_operations::add)
                     {
@@ -7164,31 +7261,32 @@ public:
             }
         };
 
+        // the name of the first member to check
+        const string_t op_member = "op";
+
         // iterate and apply the operations
         for (const auto& val : json_patch)
         {
             // wrapper to get a value for an operation
-            const auto get_value = [&val](const string_t& op,
-                                          const string_t& member,
+            const auto get_value = [&val](basic_json * found,
+                                          const string_t& op,
+                                          const char* member,
                                           bool string_type) -> basic_json &
             {
-                // find value
-                auto it = val.m_data.m_value.object->find(member);
-
                 // check if the desired value is present
-                if (JSON_HEDLEY_UNLIKELY(it == val.m_data.m_value.object->end()))
+                if (JSON_HEDLEY_UNLIKELY(found == nullptr))
                 {
-                    detail::throw_patch_missing_member(op, member, false, val);
+                    detail::throw_patch_missing_member(op, string_t(member), false, val);
                 }
 
                 // check if the result is of type string
-                if (JSON_HEDLEY_UNLIKELY(string_type && !it->second.is_string()))
+                if (JSON_HEDLEY_UNLIKELY(string_type && !found->is_string()))
                 {
-                    detail::throw_patch_missing_member(op, member, true, val);
+                    detail::throw_patch_missing_member(op, string_t(member), true, val);
                 }
 
                 // no error: return value
-                return it->second;
+                return *found;
             };
 
             // type check: every element of the array must be an object
@@ -7197,26 +7295,30 @@ public:
                 detail::throw_patch_not_an_array(val);
             }
 
+            const patch_members members = find_patch_members(val, std::is_same<typename object_t::key_type, string_t> {});
+
             // collect mandatory members
-            const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
-            const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
+            const string_t& op = *get_value(members.op, op_member, "op", true).m_data.m_value.string;
+            const string_t& path = *get_value(members.path, op, "path", true).m_data.m_value.string;
             const patch_operations operation = get_op(op);
 
             // the usual operations on a plain pointer are applied without
             // building the pointer's reference tokens; anything else (and
             // every error) is handled below
-            if (try_plain(operation, path, val))
+            if (try_plain(operation, path, members, val))
             {
                 continue;
             }
 
+            // the general operations may change any container
+            steps.reset();
             json_pointer ptr(path);
 
             switch (operation)
             {
                 case patch_operations::add:
                 {
-                    operation_add(ptr, get_value("add", "value", false));
+                    operation_add(ptr, get_value(members.value, "add", "value", false));
                     break;
                 }
 
@@ -7229,13 +7331,13 @@ public:
                 case patch_operations::replace:
                 {
                     // the "path" location must exist - use at()
-                    result.at(ptr) = get_value("replace", "value", false);
+                    result.at(ptr) = get_value(members.value, "replace", "value", false);
                     break;
                 }
 
                 case patch_operations::move:
                 {
-                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "move", "from", true).m_data.m_value.string;
                     json_pointer from_ptr(from_path);
 
                     if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -7258,7 +7360,7 @@ public:
 
                 case patch_operations::copy:
                 {
-                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "copy", "from", true).m_data.m_value.string;
                     const json_pointer from_ptr(from_path);
 
                     // The copy is functionally identical to an "add"
@@ -7276,7 +7378,7 @@ public:
                     {
                         // check if "value" matches the one at "path"
                         // the "path" location must exist - use at()
-                        success = (result.at(ptr) == get_value("test", "value", false));
+                        success = (result.at(ptr) == get_value(members.value, "test", "value", false));
                     }
                     JSON_INTERNAL_CATCH (out_of_range&)
                     {
@@ -7368,13 +7470,18 @@ public:
     /// @brief append an operation @a op for @a path to @a result, with @a value if it is not null
     static void diff_operation(basic_json& result, const char* op, const string_t& path, const basic_json* value)
     {
+        // the keys are distinct and in the order of std::map; the members are
+        // added without looking them up
         basic_json entry(value_t::object);
-        entry["op"] = op;
-        entry["path"] = path;
+        auto& members = *entry.m_data.m_value.object;
+        auto hint = members.end();
+        json_pointer::add_flattened(members, hint, string_t("op"), string_t(op), detail::priority_tag<2> {});
+        json_pointer::add_flattened(members, hint, string_t("path"), path, detail::priority_tag<2> {});
         if (value != nullptr)
         {
-            entry["value"] = *value;
+            json_pointer::add_flattened(members, hint, string_t("value"), *value, detail::priority_tag<2> {});
         }
+        entry.set_parents();
         result.push_back(std::move(entry));
     }
 
@@ -7396,6 +7503,62 @@ public:
         diff_operation(result, "add", path, &value);
     }
 
+    /// @brief the values in the target of the members of a source object, for the
+    ///        objects of up to 16 members without allocating
+    ///
+    /// reserve() must be called with the number of values that are added.
+    class diff_target_values
+    {
+      public:
+        diff_target_values() noexcept
+            : inline_items() // NOLINT(readability-redundant-member-init)
+        {}
+        diff_target_values(const diff_target_values&) = delete;
+        diff_target_values& operator=(const diff_target_values&) = delete;
+        diff_target_values(diff_target_values&&) = delete;
+        diff_target_values& operator=(diff_target_values&&) = delete;
+        ~diff_target_values() = default;
+
+        void reserve(const std::size_t n)
+        {
+            if (n > inline_size)
+            {
+                large.resize(n);
+                items = large.data();
+            }
+        }
+
+        void push_back(const basic_json* value)
+        {
+            items[count++] = value;
+        }
+
+        std::size_t size() const noexcept
+        {
+            return count;
+        }
+
+        const basic_json* operator[](const std::size_t i) const noexcept
+        {
+            return items[i];
+        }
+
+      private:
+        static constexpr std::size_t inline_size = 16;
+        /// the values of a small object, left uninitialized: they are written
+        /// before they are read (the user-provided constructor keeps the
+        /// member initialization, which -Weffc++ asks for, from zeroing them)
+        struct uninitialized_items : std::array<const basic_json*, inline_size>
+        {
+            uninitialized_items() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+        };
+
+        std::size_t count = 0;
+        std::vector<const basic_json*> large{}; // NOLINT(readability-redundant-member-init)
+        uninitialized_items inline_items;
+        const basic_json** items = inline_items.data();
+    };
+
     /// @brief append "/" and the decimal representation of @a index to @a path
     static void diff_append_index(string_t& path, std::size_t index)
     {
@@ -7476,8 +7639,9 @@ public:
     operations that remove every member of @a source and add every member of
     @a target to @a result, and returns false.
     */
+    template<typename TargetValues>
     static bool diff_object_keys(basic_json& result, const basic_json& source, const basic_json& target,
-                                 const string_t& path, std::vector<const basic_json*>& target_values,
+                                 const string_t& path, TargetValues& target_values,
                                  std::vector<const_iterator>& added)
     {
         // first pass: one lookup per source key
@@ -7691,7 +7855,7 @@ public:
 
             case value_t::object:
             {
-                std::vector<const basic_json*> target_values;
+                diff_target_values target_values;
                 std::vector<const_iterator> added;
                 if (diff_object_keys(result, source, target, path, target_values, added))
                 {

include/nlohmann/json_view.hpp

@@ -30,6 +30,7 @@
 #include <array> // array
 #include <cstddef> // nullptr_t, size_t // IWYU pragma: keep
 #include <cstdint> // uint8_t, uint32_t
+#include <cstdio> // FILE
 #include <cstring> // memcpy, strlen
 #include <istream> // istream // IWYU pragma: keep
 #include <iterator> // distance, input_iterator_tag, iterator_traits
@@ -295,7 +296,7 @@ class basic_json_view
             }
             detail::view::throw_type_error(305, "cannot use operator[] with a numeric argument with ", type_name());
         }
-        return idx < m_node->len ? basic_json_view(m_doc, navigation::value(detail::view::element_at<Editable>(*m_doc, m_node, idx))) : basic_json_view();
+        return idx < m_node->len ? element(idx) : basic_json_view();
     }
 
     /// any other integer type (int, unsigned, long, std::int64_t, ...; a
@@ -358,7 +359,7 @@ class basic_json_view
         {
             detail::view::throw_out_of_range(401, detail::concat("array index ", std::to_string(idx), " is out of range"));
         }
-        return basic_json_view(m_doc, navigation::value(detail::view::element_at<Editable>(*m_doc, m_node, idx)));
+        return element(idx);
     }
 
     /// any other integer type, see operator[]; negative values are out of
@@ -392,7 +393,13 @@ class basic_json_view
 
     string_t value(string_view_t key, const char* default_value) const
     {
-        return value(key, string_t(default_value));
+        if (NLOHMANN_VIEW_UNLIKELY(!is_object()))
+        {
+            detail::view::throw_type_error(306, "cannot use value() with ", type_name());
+        }
+        // (the default is made only if there is no such member)
+        const basic_json_view r = lookup(key);
+        return r.is_discarded() ? string_t(default_value) : r.template get<string_t>();
     }
 
     /// the value a JSON pointer refers to converted to T, or the default
@@ -411,7 +418,12 @@ class basic_json_view
 
     string_t value(const json_pointer& ptr, const char* default_value) const
     {
-        return value(ptr, string_t(default_value));
+        if (NLOHMANN_VIEW_UNLIKELY(!is_structured()))
+        {
+            detail::view::throw_type_error(306, "cannot use value() with ", type_name());
+        }
+        const basic_json_view r = detail::view::resolve_pointer(*this, detail::json_pointer_access::reference_tokens(ptr), detail::view::pointer_mode::value);
+        return r.is_discarded() ? string_t(default_value) : r.template get<string_t>();
     }
 
     /// the first element or member value; a primitive value itself. Throws
@@ -433,7 +445,8 @@ class basic_json_view
     {
         if (is_structured() && m_node->len != 0)
         {
-            return basic_json_view(m_doc, navigation::value(detail::view::last_child<Editable>(*m_doc, m_node) + (is_object() ? 1 : 0)));
+            return unedited() ? basic_json_view(m_doc, detail::view::last_child<false>(*m_doc, m_node) + (is_object() ? 1 : 0))
+                   : basic_json_view(m_doc, navigation::value(detail::view::last_child<Editable>(*m_doc, m_node) + (is_object() ? 1 : 0)));
         }
         return front();
     }
@@ -450,7 +463,7 @@ class basic_json_view
         {
             return end();
         }
-        const node* const k = detail::view::find_member<Editable>(*m_doc, m_node, key.data(), key.size());
+        const node* const k = find_key(key);
         return k != nullptr ? iterator(m_doc, k, true) : end();
     }
 
@@ -467,7 +480,7 @@ class basic_json_view
     /// whether this is an object with a member with this key
     bool contains(string_view_t key) const
     {
-        return is_object() && detail::view::find_member<Editable>(*m_doc, m_node, key.data(), key.size()) != nullptr;
+        return is_object() && find_key(key) != nullptr;
     }
 
     bool contains(const char* key) const
@@ -627,7 +640,14 @@ class basic_json_view
         // the compact writer keeps 64 bytes of slack, so that it does not grow
         // the buffer just before the end
         const std::size_t estimate = source_extent() + (style.pretty ? source_extent() / 2 : 0) + 160;
-        detail::view::view_serializer<BasicJsonType, Editable>(*m_doc, out, estimate, style).dump(m_node);
+        if (unedited())
+        {
+            detail::view::view_serializer<BasicJsonType, false>(*m_doc, out, estimate, style).dump(m_node);
+        }
+        else
+        {
+            detail::view::view_serializer<BasicJsonType, Editable>(*m_doc, out, estimate, style).dump(m_node);
+        }
         return out;
     }
 
@@ -707,7 +727,7 @@ class basic_json_view
         {
             return BasicJsonType(value_t::discarded);
         }
-        return detail::view::materialize<BasicJsonType, Editable>(*m_doc, m_node);
+        return unedited() ? detail::view::materialize<BasicJsonType, false>(*m_doc, m_node) : detail::view::materialize<BasicJsonType, Editable>(*m_doc, m_node);
     }
 
     /// byte offset of this value in the source text (for strings: of the
@@ -789,10 +809,50 @@ class basic_json_view
     /// (object required)
     NLOHMANN_VIEW_ALWAYS_INLINE basic_json_view lookup(string_view_t key) const noexcept
     {
+        if (unedited())
+        {
+            const node* const k = detail::view::find_member<false>(*m_doc, m_node, key.data(), key.size());
+            return k != nullptr ? basic_json_view(m_doc, k + 1) : basic_json_view();
+        }
         const node* const k = detail::view::find_member<Editable>(*m_doc, m_node, key.data(), key.size());
         return k != nullptr ? basic_json_view(m_doc, navigation::value(k + 1)) : basic_json_view();
     }
 
+    /// whether the document has not been edited yet: until the first edit, its index is walked like that of a
+    /// read-only document (nothing is moved or linked), which is cheaper
+    NLOHMANN_VIEW_ALWAYS_INLINE bool unedited() const noexcept
+    {
+        return !navigation::editable() || m_doc->edits == nullptr;
+    }
+
+    /// the member of this object with @a key (the last one if the key occurs more than once), or nullptr
+    NLOHMANN_VIEW_ALWAYS_INLINE const node* find_key(string_view_t key) const
+    {
+        return find_key(key, std::integral_constant<bool, Editable> {});
+    }
+
+    NLOHMANN_VIEW_ALWAYS_INLINE const node* find_key(string_view_t key, std::true_type /*editable*/) const
+    {
+        // (an editable document without edits is looked up as one that cannot be edited)
+        return unedited() ? detail::view::find_member<false>(*m_doc, m_node, key.data(), key.size())
+               : detail::view::find_member<true>(*m_doc, m_node, key.data(), key.size());
+    }
+
+    NLOHMANN_VIEW_ALWAYS_INLINE const node* find_key(string_view_t key, std::false_type /*editable*/) const
+    {
+        return detail::view::find_member<false>(*m_doc, m_node, key.data(), key.size());
+    }
+
+    /// the element of an array at an index below its size
+    NLOHMANN_VIEW_ALWAYS_INLINE basic_json_view element(size_type idx) const
+    {
+        if (unedited())
+        {
+            return basic_json_view(m_doc, detail::view::element_at<false>(*m_doc, m_node, idx));
+        }
+        return basic_json_view(m_doc, navigation::value(detail::view::element_at<Editable>(*m_doc, m_node, idx)));
+    }
+
     // --- get() dispatch ---
 
     bool get_impl(detail::view::value_tag<bool> /*unused*/, detail::priority_tag<2> /*unused*/) const
@@ -1532,7 +1592,8 @@ class basic_json_document
         // NOLINTBEGIN(readability-avoid-nested-conditional-operator): a constant expression of C++11
         append_input_impl(buf, std::forward<T>(input), std::integral_constant < int, detail::is_contiguous_byte_container<D>::value ? 0
                           : std::is_base_of<std::istream, D>::value ? 1
-                          : detail::view::is_adapter_input<D>::value ? 2 : 3 > {});
+                          : detail::view::is_adapter_input<D>::value ? 2
+                          : detail::view::is_c_stream<D>::value ? 4 : 3 > {});
         // NOLINTEND(readability-avoid-nested-conditional-operator)
     }
 
@@ -1554,6 +1615,19 @@ class basic_json_document
         detail::view::append_adapter(buf, std::move(input.ia));
     }
 
+#ifndef JSON_NO_IO
+    static void append_input_impl(std::string& buf, std::FILE* file, std::integral_constant<int, 4> /*C stream*/)
+    {
+        if (file == nullptr)
+        {
+            // (as before: the library's adapter decides what a null stream yields)
+            detail::view::append_adapter(buf, detail::input_adapter(file));
+            return;
+        }
+        detail::view::append_file(buf, file);
+    }
+#endif
+
     template<typename T>
     static void append_input_impl(std::string& buf, T&& input, std::integral_constant<int, 3> /*other*/)
     {