Bound diff()'s descent with a depth count instead of scanning the source

Now that operator== no longer recurses (#5390), diff() can keep its per-level
equality shortcut all the way down. It diffs recursively for the first
detail::recursion_depth_limit() levels, as merge_patch() does, and hands
anything deeper to diff_iteratively(). The nesting_exceeds() scan, which
cost about 30% on equal documents, is gone, and diff() is on par with
develop again.

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

include/nlohmann/detail/output/binary_writer.hpp

@@ -168,20 +168,92 @@ class binary_writer
                 if (j.m_data.m_value.number_integer >= 0)
                 {
                     // CBOR does not differentiate between positive signed
-                    // integers and unsigned integers
-                    write_cbor_head(0x00, static_cast<std::uint64_t>(j.m_data.m_value.number_integer));
+                    // integers and unsigned integers. Therefore, we used the
+                    // code from the value_t::number_unsigned case here.
+                    if (j.m_data.m_value.number_integer <= 0x17)
+                    {
+                        write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint8_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x18));
+                        write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint16_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x19));
+                        write_number(static_cast<std::uint16_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint32_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x1A));
+                        write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
+                    }
+                    else
+                    {
+                        oa.write_character(to_char_type(0x1B));
+                        write_number(static_cast<std::uint64_t>(j.m_data.m_value.number_integer));
+                    }
                 }
                 else
                 {
-                    // a negative integer n is encoded as -1 - n
-                    write_cbor_head(0x20, static_cast<std::uint64_t>(-1 - j.m_data.m_value.number_integer));
+                    // The conversions below encode the sign in the first
+                    // byte, and the value is converted to a positive number.
+                    const auto positive_number = -1 - j.m_data.m_value.number_integer;
+                    if (j.m_data.m_value.number_integer >= -24)
+                    {
+                        write_number(static_cast<std::uint8_t>(0x20 + positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint8_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x38));
+                        write_number(static_cast<std::uint8_t>(positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint16_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x39));
+                        write_number(static_cast<std::uint16_t>(positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint32_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x3A));
+                        write_number(static_cast<std::uint32_t>(positive_number));
+                    }
+                    else
+                    {
+                        oa.write_character(to_char_type(0x3B));
+                        write_number(static_cast<std::uint64_t>(positive_number));
+                    }
                 }
                 break;
             }
 
             case value_t::number_unsigned:
             {
-                write_cbor_head(0x00, j.m_data.m_value.number_unsigned);
+                if (j.m_data.m_value.number_unsigned <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x18));
+                    write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x19));
+                    write_number(static_cast<std::uint16_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x1A));
+                    write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_unsigned));
+                }
+                else
+                {
+                    oa.write_character(to_char_type(0x1B));
+                    write_number(static_cast<std::uint64_t>(j.m_data.m_value.number_unsigned));
+                }
                 break;
             }
 
@@ -211,7 +283,33 @@ class binary_writer
             case value_t::string:
             {
                 // step 1: write control byte and the string length
-                write_cbor_head(0x60, j.m_data.m_value.string->size());
+                const auto N = j.m_data.m_value.string->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x60 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x78));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x79));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x7A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x7B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write the string
                 oa.write_characters(
@@ -223,7 +321,33 @@ class binary_writer
             case value_t::array:
             {
                 // step 1: write control byte and the array size
-                write_cbor_head(0x80, j.m_data.m_value.array->size());
+                const auto N = j.m_data.m_value.array->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x80 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x98));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x99));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x9A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x9B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 for (const auto& el : *j.m_data.m_value.array)
@@ -252,7 +376,7 @@ class binary_writer
                         write_number(static_cast<std::uint8_t>(0xda));
                         write_number(static_cast<std::uint32_t>(j.m_data.m_value.binary->subtype()));
                     }
-                    else
+                    else if (j.m_data.m_value.binary->subtype() <= (std::numeric_limits<std::uint64_t>::max)())
                     {
                         write_number(static_cast<std::uint8_t>(0xdb));
                         write_number(static_cast<std::uint64_t>(j.m_data.m_value.binary->subtype()));
@@ -261,7 +385,32 @@ class binary_writer
 
                 // step 1: write control byte and the binary array size
                 const auto N = j.m_data.m_value.binary->size();
-                write_cbor_head(0x40, N);
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x40 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x58));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x59));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x5A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x5B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 oa.write_characters(
@@ -274,7 +423,33 @@ class binary_writer
             case value_t::object:
             {
                 // step 1: write control byte and the object size
-                write_cbor_head(0xA0, j.m_data.m_value.object->size());
+                const auto N = j.m_data.m_value.object->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0xA0 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xB8));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xB9));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xBA));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xBB));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 for (const auto& el : *j.m_data.m_value.object)
@@ -342,7 +517,7 @@ class binary_writer
                         oa.write_character(to_char_type(0xCE));
                         write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
                     }
-                    else
+                    else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                     {
                         // uint 64
                         oa.write_character(to_char_type(0xCF));
@@ -377,7 +552,8 @@ class binary_writer
                         oa.write_character(to_char_type(0xD2));
                         write_number(static_cast<std::int32_t>(j.m_data.m_value.number_integer));
                     }
-                    else
+                    else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int64_t>::min)() &&
+                             j.m_data.m_value.number_integer <= (std::numeric_limits<std::int64_t>::max)())
                     {
                         // int 64
                         oa.write_character(to_char_type(0xD3));
@@ -412,7 +588,7 @@ class binary_writer
                     oa.write_character(to_char_type(0xCE));
                     write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
                 }
-                else
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                 {
                     // uint 64
                     oa.write_character(to_char_type(0xCF));
@@ -1350,46 +1526,6 @@ class binary_writer
     // CBOR //
     //////////
 
-    /*!
-    @brief write the head of a CBOR data item
-
-    The head is the major type in the upper three bits of the first byte and
-    an argument - an unsigned integer, the length of a string, the number of
-    elements of a container - in the shortest of its encodings: in the lower
-    five bits of the first byte itself if it is at most 23, otherwise in the
-    1, 2, 4, or 8 bytes that follow (RFC 8949, section 3).
-
-    @param[in] major_type  the major type, shifted into the upper three bits
-    @param[in] argument    the argument of the data item
-    */
-    void write_cbor_head(const std::uint8_t major_type, const std::uint64_t argument)
-    {
-        if (argument <= 0x17)
-        {
-            write_number(static_cast<std::uint8_t>(major_type + argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint8_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x18)));
-            write_number(static_cast<std::uint8_t>(argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint16_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x19)));
-            write_number(static_cast<std::uint16_t>(argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint32_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x1A)));
-            write_number(static_cast<std::uint32_t>(argument));
-        }
-        else
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x1B)));
-            write_number(argument);
-        }
-    }
-
     static constexpr CharType get_cbor_float_prefix(float /*unused*/)
     {
         return to_char_type(0xFA);  // Single-Precision Float
@@ -1495,7 +1631,7 @@ class binary_writer
             }
             write_number(static_cast<std::int64_t>(n), use_bjdata);
         }
-        else if (use_bjdata)
+        else if (use_bjdata && n <= (std::numeric_limits<uint64_t>::max)())
         {
             if (add_prefix)
             {
@@ -1575,59 +1711,30 @@ class binary_writer
             }
             write_number(static_cast<uint32_t>(n), use_bjdata);
         }
+        else if ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)())
+        {
+            if (add_prefix)
+            {
+                oa.write_character(to_char_type('L'));  // int64
+            }
+            write_number(static_cast<std::int64_t>(n), use_bjdata);
+        }
+        // LCOV_EXCL_START
         else
         {
-            // every value of an integer type of at most 64 bits fits into an
-            // int64; only a wider type needs a range check
-            write_ubjson_int64_or_high_precision(n, add_prefix, use_bjdata,
-                                                 std::integral_constant < bool, std::numeric_limits<NumberType>::digits <= std::numeric_limits<std::int64_t>::digits > {});
+            if (add_prefix)
+            {
+                oa.write_character(to_char_type('H'));  // high-precision number
+            }
+
+            const auto number = BasicJsonType(n).dump();
+            write_number_with_ubjson_prefix(number.size(), true, use_bjdata);
+            for (std::size_t i = 0; i < number.size(); ++i)
+            {
+                oa.write_character(to_char_type(static_cast<std::uint8_t>(number[i])));
+            }
         }
-    }
-
-    template<typename NumberType>
-    void write_ubjson_int64_or_high_precision(const NumberType n, const bool add_prefix, const bool use_bjdata, std::true_type /*fits_int64*/)
-    {
-        if (add_prefix)
-        {
-            oa.write_character(to_char_type('L'));  // int64
-        }
-        write_number(static_cast<std::int64_t>(n), use_bjdata);
-    }
-
-    template<typename NumberType>
-    void write_ubjson_int64_or_high_precision(const NumberType n, const bool add_prefix, const bool use_bjdata, std::false_type /*fits_int64*/)
-    {
-        if ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)())
-        {
-            write_ubjson_int64_or_high_precision(n, add_prefix, use_bjdata, std::true_type {});
-            return;
-        }
-
-        if (add_prefix)
-        {
-            oa.write_character(to_char_type('H'));  // high-precision number
-        }
-
-        const auto number = BasicJsonType(n).dump();
-        write_number_with_ubjson_prefix(number.size(), true, use_bjdata);
-        for (std::size_t i = 0; i < number.size(); ++i)
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(number[i])));
-        }
-    }
-
-    template<typename NumberType>
-    static constexpr CharType ubjson_int64_or_high_precision_prefix(const NumberType /*n*/, std::true_type /*fits_int64*/) noexcept
-    {
-        return 'L';
-    }
-
-    template<typename NumberType>
-    static CharType ubjson_int64_or_high_precision_prefix(const NumberType n, std::false_type /*fits_int64*/) noexcept
-    {
-        // anything outside of the range of an int64 is treated as a
-        // high-precision number
-        return ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)()) ? 'L' : 'H';
+        // LCOV_EXCL_STOP
     }
 
     /*!
@@ -1669,10 +1776,12 @@ class binary_writer
                 {
                     return 'm';
                 }
-                // every value of an integer type of at most 64 bits fits into
-                // an int64; only a wider type needs a range check
-                return ubjson_int64_or_high_precision_prefix(j.m_data.m_value.number_integer,
-                        std::integral_constant < bool, std::numeric_limits<typename BasicJsonType::number_integer_t>::digits <= std::numeric_limits<std::int64_t>::digits > {});
+                if ((std::numeric_limits<std::int64_t>::min)() <= j.m_data.m_value.number_integer && j.m_data.m_value.number_integer <= (std::numeric_limits<std::int64_t>::max)())
+                {
+                    return 'L';
+                }
+                // anything else is treated as a high-precision number
+                return 'H'; // LCOV_EXCL_LINE
             }
 
             case value_t::number_unsigned:
@@ -1705,12 +1814,12 @@ class binary_writer
                 {
                     return 'L';
                 }
-                if (use_bjdata)
+                if (use_bjdata && j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                 {
                     return 'M';
                 }
                 // anything else is treated as a high-precision number
-                return 'H';
+                return 'H'; // LCOV_EXCL_LINE
             }
 
             case value_t::number_float:

include/nlohmann/json.hpp

@@ -2157,6 +2157,17 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     // value access //
     //////////////////
 
+    /// get a boolean (explicit)
+    boolean_t get_impl(boolean_t* /*unused*/) const
+    {
+        if (JSON_HEDLEY_LIKELY(is_boolean()))
+        {
+            return m_data.m_value.boolean;
+        }
+
+        JSON_THROW(type_error::create(302, detail::concat("type must be boolean, but is ", type_name()), this));
+    }
+
     /// get a pointer to the value (object)
     object_t* get_impl_ptr(object_t* /*unused*/) noexcept
     {
@@ -5982,6 +5993,56 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     JSON_HEDLEY_WARN_UNUSED_RESULT
     static basic_json diff(const basic_json& source, const basic_json& target,
                            const string_t& path = "")
+    {
+        return diff_recursively(source, target, path, 0);
+    }
+
+  private:
+    /// @brief two arrays or two objects @ref diff_iteratively is diffing
+    struct diff_frame
+    {
+        diff_frame(const basic_json* source_, const basic_json* target_, const std::size_t path_length_) noexcept
+            : source(source_), target(target_), path_length(path_length_)
+        {}
+
+        // declared for GCC's -Weffc++, which asks for them in a class with
+        // pointer members and a non-trivial destructor; the exception
+        // specifications are left implicit, as GCC 4.8 rejects explicit ones
+        // that differ from them
+        diff_frame(const diff_frame&) = default;
+        diff_frame(diff_frame&&) = default;
+        diff_frame& operator=(const diff_frame&) = default;
+        diff_frame& operator=(diff_frame&&) = default;
+        ~diff_frame() = default;
+
+        /// the values being diffed, both arrays or both objects
+        const basic_json* source;
+        const basic_json* target;
+        /// the length of their path in `current_path`
+        std::size_t path_length;
+        /// arrays: the next index to diff
+        std::size_t index = 0;
+        /// objects: the next member of source to look at
+        const_iterator member{}; // NOLINT(readability-redundant-member-init)
+        /// objects: the keys common to both, in source's order
+        std::vector<typename object_t::key_type> common_keys{}; // NOLINT(readability-redundant-member-init)
+        /// objects: the next entry of common_keys
+        std::size_t next_common = 0;
+        /// objects: the "add" operations for keys only target has
+        basic_json added_ops{}; // NOLINT(readability-redundant-member-init)
+    };
+
+    /*!
+    @brief @ref diff, for values at nesting level @a depth
+
+    Diffing two arrays or objects calls this function again, once per nesting
+    level, so values nested deeply enough used to exhaust the call stack and
+    terminate the process. The descent is bounded here: once @ref
+    detail::recursion_depth_limit levels have been entered, @ref
+    diff_iteratively diffs what is left without the call stack.
+    */
+    static basic_json diff_recursively(const basic_json& source, const basic_json& target,
+                                       const string_t& path, const std::size_t depth)
     {
         // the patch
         basic_json result(value_t::array);
@@ -5992,6 +6053,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             return result;
         }
 
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return diff_iteratively(source, target, path);
+        }
+
         if (source.type() != target.type())
         {
             // different types: replace value
@@ -6011,7 +6077,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                 while (i < source.size() && i < target.size())
                 {
                     // recursive call to compare array values at index i
-                    auto temp_diff = diff(source[i], target[i], detail::concat<string_t>(path, '/', detail::to_string<string_t>(i)));
+                    auto temp_diff = diff_recursively(source[i], target[i], detail::concat<string_t>(path, '/', detail::to_string<string_t>(i)), depth + 1);
                     result.insert(result.end(), temp_diff.begin(), temp_diff.end());
                     ++i;
                 }
@@ -6130,7 +6196,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                         if (common_it != common_keys_source_order.cend() && it.key() == *common_it)
                         {
                             const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
-                            auto temp_diff = diff(it.value(), target[it.key()], path_key);
+                            auto temp_diff = diff_recursively(it.value(), target[it.key()], path_key, depth + 1);
                             result.insert(result.end(), temp_diff.begin(), temp_diff.end());
                             ++common_it;
                         }
@@ -6214,6 +6280,301 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
 
         return result;
     }
+
+    /*!
+    @brief @ref diff without the call stack
+
+    Produces the same patch as @ref diff_recursively. Only reached for values
+    nested more deeply than @ref detail::recursion_depth_limit.
+    */
+    static basic_json diff_iteratively(const basic_json& source, const basic_json& target,
+                                       const string_t& path)
+    {
+        // the patch
+        basic_json result(value_t::array);
+
+        // The arrays and objects being diffed are kept on an explicit stack,
+        // and every pair of elements is still diffed completely before the
+        // next one, so the operations come out in the same order as in
+        // diff_recursively. The path of the values being diffed is kept in
+        // one buffer that grows and shrinks with the stack, rather than in a
+        // new string per level.
+        std::vector<diff_frame> stack;
+        string_t current_path = path;
+
+        // diff `s` against `t`, whose path is current_path: primitives,
+        // values of different types, and objects whose members were reordered
+        // are handled right away; arrays and other objects get a frame
+        const auto enter = [&result, &stack, &current_path](const basic_json & s, const basic_json & t)
+        {
+            // if the values are the same, there is nothing to do. Arrays and
+            // objects are not compared up front: comparing them visits
+            // everything below them, so doing that at every level would take
+            // quadratic time in the nesting depth - equal ones yield no
+            // operations anyway.
+            if ((!s.is_structured() || !t.is_structured()) && s == t)
+            {
+                return;
+            }
+
+            if (s.type() != t.type())
+            {
+                // different types: replace value
+                result.push_back(
+                {
+                    {"op", "replace"}, {"path", current_path}, {"value", t}
+                });
+                return;
+            }
+
+            switch (s.type())
+            {
+                case value_t::array:
+                {
+                    stack.emplace_back(&s, &t, current_path.size());
+                    return;
+                }
+
+                case value_t::object:
+                {
+                    // first pass: record, for every source key, whether it is
+                    // common to both objects (in source's iteration order) or
+                    // was deleted (i.e., in source but not in target) -- this is
+                    // a by-product of the t.find() call already needed to
+                    // tell the two cases apart, so it adds no extra lookups. The
+                    // "remove" ops themselves are emitted later, interleaved
+                    // with the per-key diffs in the fast path below, to match
+                    // source's original iteration order (as the original,
+                    // pre-reordering-aware implementation did) instead of
+                    // grouping all removes before all per-key diffs.
+                    std::vector<typename object_t::key_type> common_keys_source_order;
+                    for (auto it = s.cbegin(); it != s.cend(); ++it)
+                    {
+                        if (t.find(it.key()) != t.end())
+                        {
+                            common_keys_source_order.push_back(it.key());
+                        }
+                    }
+
+                    // second pass: find keys that were added (i.e., in target but
+                    // not in source), and record the keys common to both, in
+                    // target's iteration order -- again a by-product of the
+                    // s.find() call already needed to detect added keys. At
+                    // the same time, determine whether every added key comes
+                    // after every common key in target's order (a precondition
+                    // for the fast path below, which only ever appends new keys
+                    // at the very end): for an object_t whose iteration order is
+                    // a pure function of the key set (e.g. the default std::map,
+                    // which always iterates in sorted key order), the order
+                    // check further below is always true and this whole
+                    // mechanism is effectively a no-op; it only matters for a
+                    // reorderable object_t such as the one backing `ordered_json`.
+                    // patch ops for keys that were added (i.e., in target but not
+                    // in source); built here so the fast path below can reuse
+                    // them without a second s.find() per target key. Only
+                    // used by the fast path -- the slow (reordering) path
+                    // rebuilds "add" ops for every key itself.
+                    std::vector<typename object_t::key_type> common_keys_target_order;
+                    basic_json added_ops(value_t::array);
+                    bool new_keys_form_suffix = true;
+                    bool seen_new_key = false;
+                    for (auto it = t.cbegin(); it != t.cend(); ++it)
+                    {
+                        if (s.find(it.key()) == s.end())
+                        {
+                            seen_new_key = true;
+                            const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                            added_ops.push_back(
+                            {
+                                {"op", "add"}, {"path", path_key},
+                                {"value", it.value()}
+                            });
+                        }
+                        else
+                        {
+                            common_keys_target_order.push_back(it.key());
+                            if (seen_new_key)
+                            {
+                                new_keys_form_suffix = false;
+                            }
+                        }
+                    }
+
+                    if (common_keys_source_order == common_keys_target_order && new_keys_form_suffix)
+                    {
+                        // fast path: order of common keys already matches (or the
+                        // object_t's iteration order does not depend on
+                        // insertion history), so a plain per-key diff is correct
+                        // and minimal, as before. The frame walks source in
+                        // lockstep with common_keys_source_order, which is, by
+                        // construction, the subsequence of source's keys that
+                        // are common to both objects, in source's iteration
+                        // order -- so a cheap key comparison replaces another
+                        // lookup. Deleted keys are interleaved there too, in
+                        // source's original order, and the "add" ops collected
+                        // above are appended once all members are done.
+                        stack.emplace_back(&s, &t, current_path.size());
+                        stack.back().member = s.cbegin();
+                        stack.back().common_keys = std::move(common_keys_source_order);
+                        stack.back().added_ops = std::move(added_ops);
+                        return;
+                    }
+
+                    // slow path: the common keys are in a different relative
+                    // order in source and target (only possible for a
+                    // reorderable object_t like ordered_map). Building a
+                    // minimal reordering patch is a nontrivial (LCS-like)
+                    // problem; instead, remove every source key -- both
+                    // deleted keys (which must be removed regardless) and
+                    // common keys (removed so they can be re-added in
+                    // target's order) -- and re-add every key that should
+                    // remain, with its final target value, in target's
+                    // order. basic_json::patch()'s "add" operation on an
+                    // object uses operator[], which appends at the end for a
+                    // vector-backed insertion-ordered map when the key does
+                    // not already exist -- so removing a key and then adding
+                    // it moves it to the end, fixing its position.
+                    for (auto it = s.cbegin(); it != s.cend(); ++it)
+                    {
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(object(
+                        {
+                            {"op", "remove"}, {"path", path_key}
+                        }));
+                    }
+
+                    // add every key that is either common (just removed
+                    // above) or brand new, in target's iteration order, so
+                    // that the final order after applying the patch matches
+                    // target exactly
+                    for (auto it = t.cbegin(); it != t.cend(); ++it)
+                    {
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(
+                        {
+                            {"op", "add"}, {"path", path_key},
+                            {"value", it.value()}
+                        });
+                    }
+                    return;
+                }
+
+                case value_t::null:
+                case value_t::string:
+                case value_t::boolean:
+                case value_t::number_integer:
+                case value_t::number_unsigned:
+                case value_t::number_float:
+                case value_t::binary:
+                case value_t::discarded:
+                default:
+                {
+                    // both primitive types: replace value
+                    result.push_back(
+                    {
+                        {"op", "replace"}, {"path", current_path}, {"value", t}
+                    });
+                    return;
+                }
+            }
+        };
+
+        enter(source, target);
+        while (!stack.empty())
+        {
+            diff_frame& frame = stack.back();
+            const std::size_t path_length = frame.path_length;
+            const std::size_t depth = stack.size();
+
+            if (frame.source->is_array())
+            {
+                const auto& source_array = *frame.source->m_data.m_value.array;
+                const auto& target_array = *frame.target->m_data.m_value.array;
+
+                // first pass: traverse common elements
+                if (frame.index < source_array.size() && frame.index < target_array.size())
+                {
+                    const std::size_t i = frame.index++;
+                    detail::concat_into(current_path, '/', detail::to_string<string_t>(i));
+                    enter(source_array[i], target_array[i]); // may push, which invalidates `frame`
+                    if (stack.size() == depth)
+                    {
+                        current_path.resize(path_length);
+                    }
+                    continue;
+                }
+
+                // We now reached the end of at least one array
+                // in a second pass, traverse the remaining elements
+
+                // remove my remaining elements, highest index first; appending
+                // in that order avoids the quadratic reinsertion done before
+                for (std::size_t j = source_array.size(); j > frame.index; --j)
+                {
+                    result.push_back(object(
+                    {
+                        {"op", "remove"},
+                        {"path", detail::concat<string_t>(current_path, '/', detail::to_string<string_t>(j - 1))}
+                    }));
+                }
+
+                // add other remaining elements
+                for (std::size_t i = source_array.size(); i < target_array.size(); ++i)
+                {
+                    result.push_back(
+                    {
+                        {"op", "add"},
+                        {"path", detail::concat<string_t>(current_path, "/-")},
+                        {"value", target_array[i]}
+                    });
+                }
+            }
+            else
+            {
+                if (frame.member != frame.source->cend())
+                {
+                    const const_iterator it = frame.member;
+                    ++frame.member;
+                    if (frame.next_common < frame.common_keys.size() && it.key() == frame.common_keys[frame.next_common])
+                    {
+                        ++frame.next_common;
+                        const basic_json& target_value = (*frame.target)[it.key()];
+                        detail::concat_into(current_path, '/', detail::escape(it.key()));
+                        enter(it.value(), target_value); // may push, which invalidates `frame`
+                        if (stack.size() == depth)
+                        {
+                            current_path.resize(path_length);
+                        }
+                    }
+                    else
+                    {
+                        // found a key that is not in target -> remove it
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(object(
+                        {
+                            {"op", "remove"}, {"path", path_key}
+                        }));
+                    }
+                    continue;
+                }
+
+                // append the "add" ops for brand-new keys collected when the
+                // object was entered
+                result.insert(result.end(), frame.added_ops.begin(), frame.added_ops.end());
+            }
+
+            // this array or object is done: continue with the one it is in
+            stack.pop_back();
+            if (!stack.empty())
+            {
+                current_path.resize(stack.back().path_length);
+            }
+        }
+
+        return result;
+    }
+
+  public:
     /// @}
 
     ////////////////////////////////

single_include/nlohmann/json.hpp

@@ -19633,20 +19633,92 @@ class binary_writer
                 if (j.m_data.m_value.number_integer >= 0)
                 {
                     // CBOR does not differentiate between positive signed
-                    // integers and unsigned integers
-                    write_cbor_head(0x00, static_cast<std::uint64_t>(j.m_data.m_value.number_integer));
+                    // integers and unsigned integers. Therefore, we used the
+                    // code from the value_t::number_unsigned case here.
+                    if (j.m_data.m_value.number_integer <= 0x17)
+                    {
+                        write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint8_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x18));
+                        write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint16_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x19));
+                        write_number(static_cast<std::uint16_t>(j.m_data.m_value.number_integer));
+                    }
+                    else if (j.m_data.m_value.number_integer <= (std::numeric_limits<std::uint32_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x1A));
+                        write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
+                    }
+                    else
+                    {
+                        oa.write_character(to_char_type(0x1B));
+                        write_number(static_cast<std::uint64_t>(j.m_data.m_value.number_integer));
+                    }
                 }
                 else
                 {
-                    // a negative integer n is encoded as -1 - n
-                    write_cbor_head(0x20, static_cast<std::uint64_t>(-1 - j.m_data.m_value.number_integer));
+                    // The conversions below encode the sign in the first
+                    // byte, and the value is converted to a positive number.
+                    const auto positive_number = -1 - j.m_data.m_value.number_integer;
+                    if (j.m_data.m_value.number_integer >= -24)
+                    {
+                        write_number(static_cast<std::uint8_t>(0x20 + positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint8_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x38));
+                        write_number(static_cast<std::uint8_t>(positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint16_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x39));
+                        write_number(static_cast<std::uint16_t>(positive_number));
+                    }
+                    else if (positive_number <= (std::numeric_limits<std::uint32_t>::max)())
+                    {
+                        oa.write_character(to_char_type(0x3A));
+                        write_number(static_cast<std::uint32_t>(positive_number));
+                    }
+                    else
+                    {
+                        oa.write_character(to_char_type(0x3B));
+                        write_number(static_cast<std::uint64_t>(positive_number));
+                    }
                 }
                 break;
             }
 
             case value_t::number_unsigned:
             {
-                write_cbor_head(0x00, j.m_data.m_value.number_unsigned);
+                if (j.m_data.m_value.number_unsigned <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x18));
+                    write_number(static_cast<std::uint8_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x19));
+                    write_number(static_cast<std::uint16_t>(j.m_data.m_value.number_unsigned));
+                }
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x1A));
+                    write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_unsigned));
+                }
+                else
+                {
+                    oa.write_character(to_char_type(0x1B));
+                    write_number(static_cast<std::uint64_t>(j.m_data.m_value.number_unsigned));
+                }
                 break;
             }
 
@@ -19676,7 +19748,33 @@ class binary_writer
             case value_t::string:
             {
                 // step 1: write control byte and the string length
-                write_cbor_head(0x60, j.m_data.m_value.string->size());
+                const auto N = j.m_data.m_value.string->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x60 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x78));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x79));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x7A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x7B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write the string
                 oa.write_characters(
@@ -19688,7 +19786,33 @@ class binary_writer
             case value_t::array:
             {
                 // step 1: write control byte and the array size
-                write_cbor_head(0x80, j.m_data.m_value.array->size());
+                const auto N = j.m_data.m_value.array->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x80 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x98));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x99));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x9A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x9B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 for (const auto& el : *j.m_data.m_value.array)
@@ -19717,7 +19841,7 @@ class binary_writer
                         write_number(static_cast<std::uint8_t>(0xda));
                         write_number(static_cast<std::uint32_t>(j.m_data.m_value.binary->subtype()));
                     }
-                    else
+                    else if (j.m_data.m_value.binary->subtype() <= (std::numeric_limits<std::uint64_t>::max)())
                     {
                         write_number(static_cast<std::uint8_t>(0xdb));
                         write_number(static_cast<std::uint64_t>(j.m_data.m_value.binary->subtype()));
@@ -19726,7 +19850,32 @@ class binary_writer
 
                 // step 1: write control byte and the binary array size
                 const auto N = j.m_data.m_value.binary->size();
-                write_cbor_head(0x40, N);
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0x40 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x58));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x59));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x5A));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0x5B));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 oa.write_characters(
@@ -19739,7 +19888,33 @@ class binary_writer
             case value_t::object:
             {
                 // step 1: write control byte and the object size
-                write_cbor_head(0xA0, j.m_data.m_value.object->size());
+                const auto N = j.m_data.m_value.object->size();
+                if (N <= 0x17)
+                {
+                    write_number(static_cast<std::uint8_t>(0xA0 + N));
+                }
+                else if (N <= (std::numeric_limits<std::uint8_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xB8));
+                    write_number(static_cast<std::uint8_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint16_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xB9));
+                    write_number(static_cast<std::uint16_t>(N));
+                }
+                else if (N <= (std::numeric_limits<std::uint32_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xBA));
+                    write_number(static_cast<std::uint32_t>(N));
+                }
+                // LCOV_EXCL_START
+                else if (N <= (std::numeric_limits<std::uint64_t>::max)())
+                {
+                    oa.write_character(to_char_type(0xBB));
+                    write_number(static_cast<std::uint64_t>(N));
+                }
+                // LCOV_EXCL_STOP
 
                 // step 2: write each element
                 for (const auto& el : *j.m_data.m_value.object)
@@ -19807,7 +19982,7 @@ class binary_writer
                         oa.write_character(to_char_type(0xCE));
                         write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
                     }
-                    else
+                    else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                     {
                         // uint 64
                         oa.write_character(to_char_type(0xCF));
@@ -19842,7 +20017,8 @@ class binary_writer
                         oa.write_character(to_char_type(0xD2));
                         write_number(static_cast<std::int32_t>(j.m_data.m_value.number_integer));
                     }
-                    else
+                    else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int64_t>::min)() &&
+                             j.m_data.m_value.number_integer <= (std::numeric_limits<std::int64_t>::max)())
                     {
                         // int 64
                         oa.write_character(to_char_type(0xD3));
@@ -19877,7 +20053,7 @@ class binary_writer
                     oa.write_character(to_char_type(0xCE));
                     write_number(static_cast<std::uint32_t>(j.m_data.m_value.number_integer));
                 }
-                else
+                else if (j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                 {
                     // uint 64
                     oa.write_character(to_char_type(0xCF));
@@ -20815,46 +20991,6 @@ class binary_writer
     // CBOR //
     //////////
 
-    /*!
-    @brief write the head of a CBOR data item
-
-    The head is the major type in the upper three bits of the first byte and
-    an argument - an unsigned integer, the length of a string, the number of
-    elements of a container - in the shortest of its encodings: in the lower
-    five bits of the first byte itself if it is at most 23, otherwise in the
-    1, 2, 4, or 8 bytes that follow (RFC 8949, section 3).
-
-    @param[in] major_type  the major type, shifted into the upper three bits
-    @param[in] argument    the argument of the data item
-    */
-    void write_cbor_head(const std::uint8_t major_type, const std::uint64_t argument)
-    {
-        if (argument <= 0x17)
-        {
-            write_number(static_cast<std::uint8_t>(major_type + argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint8_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x18)));
-            write_number(static_cast<std::uint8_t>(argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint16_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x19)));
-            write_number(static_cast<std::uint16_t>(argument));
-        }
-        else if (argument <= (std::numeric_limits<std::uint32_t>::max)())
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x1A)));
-            write_number(static_cast<std::uint32_t>(argument));
-        }
-        else
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(major_type + 0x1B)));
-            write_number(argument);
-        }
-    }
-
     static constexpr CharType get_cbor_float_prefix(float /*unused*/)
     {
         return to_char_type(0xFA);  // Single-Precision Float
@@ -20960,7 +21096,7 @@ class binary_writer
             }
             write_number(static_cast<std::int64_t>(n), use_bjdata);
         }
-        else if (use_bjdata)
+        else if (use_bjdata && n <= (std::numeric_limits<uint64_t>::max)())
         {
             if (add_prefix)
             {
@@ -21040,59 +21176,30 @@ class binary_writer
             }
             write_number(static_cast<uint32_t>(n), use_bjdata);
         }
+        else if ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)())
+        {
+            if (add_prefix)
+            {
+                oa.write_character(to_char_type('L'));  // int64
+            }
+            write_number(static_cast<std::int64_t>(n), use_bjdata);
+        }
+        // LCOV_EXCL_START
         else
         {
-            // every value of an integer type of at most 64 bits fits into an
-            // int64; only a wider type needs a range check
-            write_ubjson_int64_or_high_precision(n, add_prefix, use_bjdata,
-                                                 std::integral_constant < bool, std::numeric_limits<NumberType>::digits <= std::numeric_limits<std::int64_t>::digits > {});
+            if (add_prefix)
+            {
+                oa.write_character(to_char_type('H'));  // high-precision number
+            }
+
+            const auto number = BasicJsonType(n).dump();
+            write_number_with_ubjson_prefix(number.size(), true, use_bjdata);
+            for (std::size_t i = 0; i < number.size(); ++i)
+            {
+                oa.write_character(to_char_type(static_cast<std::uint8_t>(number[i])));
+            }
         }
-    }
-
-    template<typename NumberType>
-    void write_ubjson_int64_or_high_precision(const NumberType n, const bool add_prefix, const bool use_bjdata, std::true_type /*fits_int64*/)
-    {
-        if (add_prefix)
-        {
-            oa.write_character(to_char_type('L'));  // int64
-        }
-        write_number(static_cast<std::int64_t>(n), use_bjdata);
-    }
-
-    template<typename NumberType>
-    void write_ubjson_int64_or_high_precision(const NumberType n, const bool add_prefix, const bool use_bjdata, std::false_type /*fits_int64*/)
-    {
-        if ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)())
-        {
-            write_ubjson_int64_or_high_precision(n, add_prefix, use_bjdata, std::true_type {});
-            return;
-        }
-
-        if (add_prefix)
-        {
-            oa.write_character(to_char_type('H'));  // high-precision number
-        }
-
-        const auto number = BasicJsonType(n).dump();
-        write_number_with_ubjson_prefix(number.size(), true, use_bjdata);
-        for (std::size_t i = 0; i < number.size(); ++i)
-        {
-            oa.write_character(to_char_type(static_cast<std::uint8_t>(number[i])));
-        }
-    }
-
-    template<typename NumberType>
-    static constexpr CharType ubjson_int64_or_high_precision_prefix(const NumberType /*n*/, std::true_type /*fits_int64*/) noexcept
-    {
-        return 'L';
-    }
-
-    template<typename NumberType>
-    static CharType ubjson_int64_or_high_precision_prefix(const NumberType n, std::false_type /*fits_int64*/) noexcept
-    {
-        // anything outside of the range of an int64 is treated as a
-        // high-precision number
-        return ((std::numeric_limits<std::int64_t>::min)() <= n && n <= (std::numeric_limits<std::int64_t>::max)()) ? 'L' : 'H';
+        // LCOV_EXCL_STOP
     }
 
     /*!
@@ -21134,10 +21241,12 @@ class binary_writer
                 {
                     return 'm';
                 }
-                // every value of an integer type of at most 64 bits fits into
-                // an int64; only a wider type needs a range check
-                return ubjson_int64_or_high_precision_prefix(j.m_data.m_value.number_integer,
-                        std::integral_constant < bool, std::numeric_limits<typename BasicJsonType::number_integer_t>::digits <= std::numeric_limits<std::int64_t>::digits > {});
+                if ((std::numeric_limits<std::int64_t>::min)() <= j.m_data.m_value.number_integer && j.m_data.m_value.number_integer <= (std::numeric_limits<std::int64_t>::max)())
+                {
+                    return 'L';
+                }
+                // anything else is treated as a high-precision number
+                return 'H'; // LCOV_EXCL_LINE
             }
 
             case value_t::number_unsigned:
@@ -21170,12 +21279,12 @@ class binary_writer
                 {
                     return 'L';
                 }
-                if (use_bjdata)
+                if (use_bjdata && j.m_data.m_value.number_unsigned <= (std::numeric_limits<std::uint64_t>::max)())
                 {
                     return 'M';
                 }
                 // anything else is treated as a high-precision number
-                return 'H';
+                return 'H'; // LCOV_EXCL_LINE
             }
 
             case value_t::number_float:
@@ -27006,6 +27115,17 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     // value access //
     //////////////////
 
+    /// get a boolean (explicit)
+    boolean_t get_impl(boolean_t* /*unused*/) const
+    {
+        if (JSON_HEDLEY_LIKELY(is_boolean()))
+        {
+            return m_data.m_value.boolean;
+        }
+
+        JSON_THROW(type_error::create(302, detail::concat("type must be boolean, but is ", type_name()), this));
+    }
+
     /// get a pointer to the value (object)
     object_t* get_impl_ptr(object_t* /*unused*/) noexcept
     {
@@ -30831,6 +30951,56 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     JSON_HEDLEY_WARN_UNUSED_RESULT
     static basic_json diff(const basic_json& source, const basic_json& target,
                            const string_t& path = "")
+    {
+        return diff_recursively(source, target, path, 0);
+    }
+
+  private:
+    /// @brief two arrays or two objects @ref diff_iteratively is diffing
+    struct diff_frame
+    {
+        diff_frame(const basic_json* source_, const basic_json* target_, const std::size_t path_length_) noexcept
+            : source(source_), target(target_), path_length(path_length_)
+        {}
+
+        // declared for GCC's -Weffc++, which asks for them in a class with
+        // pointer members and a non-trivial destructor; the exception
+        // specifications are left implicit, as GCC 4.8 rejects explicit ones
+        // that differ from them
+        diff_frame(const diff_frame&) = default;
+        diff_frame(diff_frame&&) = default;
+        diff_frame& operator=(const diff_frame&) = default;
+        diff_frame& operator=(diff_frame&&) = default;
+        ~diff_frame() = default;
+
+        /// the values being diffed, both arrays or both objects
+        const basic_json* source;
+        const basic_json* target;
+        /// the length of their path in `current_path`
+        std::size_t path_length;
+        /// arrays: the next index to diff
+        std::size_t index = 0;
+        /// objects: the next member of source to look at
+        const_iterator member{}; // NOLINT(readability-redundant-member-init)
+        /// objects: the keys common to both, in source's order
+        std::vector<typename object_t::key_type> common_keys{}; // NOLINT(readability-redundant-member-init)
+        /// objects: the next entry of common_keys
+        std::size_t next_common = 0;
+        /// objects: the "add" operations for keys only target has
+        basic_json added_ops{}; // NOLINT(readability-redundant-member-init)
+    };
+
+    /*!
+    @brief @ref diff, for values at nesting level @a depth
+
+    Diffing two arrays or objects calls this function again, once per nesting
+    level, so values nested deeply enough used to exhaust the call stack and
+    terminate the process. The descent is bounded here: once @ref
+    detail::recursion_depth_limit levels have been entered, @ref
+    diff_iteratively diffs what is left without the call stack.
+    */
+    static basic_json diff_recursively(const basic_json& source, const basic_json& target,
+                                       const string_t& path, const std::size_t depth)
     {
         // the patch
         basic_json result(value_t::array);
@@ -30841,6 +31011,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
             return result;
         }
 
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return diff_iteratively(source, target, path);
+        }
+
         if (source.type() != target.type())
         {
             // different types: replace value
@@ -30860,7 +31035,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                 while (i < source.size() && i < target.size())
                 {
                     // recursive call to compare array values at index i
-                    auto temp_diff = diff(source[i], target[i], detail::concat<string_t>(path, '/', detail::to_string<string_t>(i)));
+                    auto temp_diff = diff_recursively(source[i], target[i], detail::concat<string_t>(path, '/', detail::to_string<string_t>(i)), depth + 1);
                     result.insert(result.end(), temp_diff.begin(), temp_diff.end());
                     ++i;
                 }
@@ -30979,7 +31154,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
                         if (common_it != common_keys_source_order.cend() && it.key() == *common_it)
                         {
                             const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
-                            auto temp_diff = diff(it.value(), target[it.key()], path_key);
+                            auto temp_diff = diff_recursively(it.value(), target[it.key()], path_key, depth + 1);
                             result.insert(result.end(), temp_diff.begin(), temp_diff.end());
                             ++common_it;
                         }
@@ -31063,6 +31238,301 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
 
         return result;
     }
+
+    /*!
+    @brief @ref diff without the call stack
+
+    Produces the same patch as @ref diff_recursively. Only reached for values
+    nested more deeply than @ref detail::recursion_depth_limit.
+    */
+    static basic_json diff_iteratively(const basic_json& source, const basic_json& target,
+                                       const string_t& path)
+    {
+        // the patch
+        basic_json result(value_t::array);
+
+        // The arrays and objects being diffed are kept on an explicit stack,
+        // and every pair of elements is still diffed completely before the
+        // next one, so the operations come out in the same order as in
+        // diff_recursively. The path of the values being diffed is kept in
+        // one buffer that grows and shrinks with the stack, rather than in a
+        // new string per level.
+        std::vector<diff_frame> stack;
+        string_t current_path = path;
+
+        // diff `s` against `t`, whose path is current_path: primitives,
+        // values of different types, and objects whose members were reordered
+        // are handled right away; arrays and other objects get a frame
+        const auto enter = [&result, &stack, &current_path](const basic_json & s, const basic_json & t)
+        {
+            // if the values are the same, there is nothing to do. Arrays and
+            // objects are not compared up front: comparing them visits
+            // everything below them, so doing that at every level would take
+            // quadratic time in the nesting depth - equal ones yield no
+            // operations anyway.
+            if ((!s.is_structured() || !t.is_structured()) && s == t)
+            {
+                return;
+            }
+
+            if (s.type() != t.type())
+            {
+                // different types: replace value
+                result.push_back(
+                {
+                    {"op", "replace"}, {"path", current_path}, {"value", t}
+                });
+                return;
+            }
+
+            switch (s.type())
+            {
+                case value_t::array:
+                {
+                    stack.emplace_back(&s, &t, current_path.size());
+                    return;
+                }
+
+                case value_t::object:
+                {
+                    // first pass: record, for every source key, whether it is
+                    // common to both objects (in source's iteration order) or
+                    // was deleted (i.e., in source but not in target) -- this is
+                    // a by-product of the t.find() call already needed to
+                    // tell the two cases apart, so it adds no extra lookups. The
+                    // "remove" ops themselves are emitted later, interleaved
+                    // with the per-key diffs in the fast path below, to match
+                    // source's original iteration order (as the original,
+                    // pre-reordering-aware implementation did) instead of
+                    // grouping all removes before all per-key diffs.
+                    std::vector<typename object_t::key_type> common_keys_source_order;
+                    for (auto it = s.cbegin(); it != s.cend(); ++it)
+                    {
+                        if (t.find(it.key()) != t.end())
+                        {
+                            common_keys_source_order.push_back(it.key());
+                        }
+                    }
+
+                    // second pass: find keys that were added (i.e., in target but
+                    // not in source), and record the keys common to both, in
+                    // target's iteration order -- again a by-product of the
+                    // s.find() call already needed to detect added keys. At
+                    // the same time, determine whether every added key comes
+                    // after every common key in target's order (a precondition
+                    // for the fast path below, which only ever appends new keys
+                    // at the very end): for an object_t whose iteration order is
+                    // a pure function of the key set (e.g. the default std::map,
+                    // which always iterates in sorted key order), the order
+                    // check further below is always true and this whole
+                    // mechanism is effectively a no-op; it only matters for a
+                    // reorderable object_t such as the one backing `ordered_json`.
+                    // patch ops for keys that were added (i.e., in target but not
+                    // in source); built here so the fast path below can reuse
+                    // them without a second s.find() per target key. Only
+                    // used by the fast path -- the slow (reordering) path
+                    // rebuilds "add" ops for every key itself.
+                    std::vector<typename object_t::key_type> common_keys_target_order;
+                    basic_json added_ops(value_t::array);
+                    bool new_keys_form_suffix = true;
+                    bool seen_new_key = false;
+                    for (auto it = t.cbegin(); it != t.cend(); ++it)
+                    {
+                        if (s.find(it.key()) == s.end())
+                        {
+                            seen_new_key = true;
+                            const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                            added_ops.push_back(
+                            {
+                                {"op", "add"}, {"path", path_key},
+                                {"value", it.value()}
+                            });
+                        }
+                        else
+                        {
+                            common_keys_target_order.push_back(it.key());
+                            if (seen_new_key)
+                            {
+                                new_keys_form_suffix = false;
+                            }
+                        }
+                    }
+
+                    if (common_keys_source_order == common_keys_target_order && new_keys_form_suffix)
+                    {
+                        // fast path: order of common keys already matches (or the
+                        // object_t's iteration order does not depend on
+                        // insertion history), so a plain per-key diff is correct
+                        // and minimal, as before. The frame walks source in
+                        // lockstep with common_keys_source_order, which is, by
+                        // construction, the subsequence of source's keys that
+                        // are common to both objects, in source's iteration
+                        // order -- so a cheap key comparison replaces another
+                        // lookup. Deleted keys are interleaved there too, in
+                        // source's original order, and the "add" ops collected
+                        // above are appended once all members are done.
+                        stack.emplace_back(&s, &t, current_path.size());
+                        stack.back().member = s.cbegin();
+                        stack.back().common_keys = std::move(common_keys_source_order);
+                        stack.back().added_ops = std::move(added_ops);
+                        return;
+                    }
+
+                    // slow path: the common keys are in a different relative
+                    // order in source and target (only possible for a
+                    // reorderable object_t like ordered_map). Building a
+                    // minimal reordering patch is a nontrivial (LCS-like)
+                    // problem; instead, remove every source key -- both
+                    // deleted keys (which must be removed regardless) and
+                    // common keys (removed so they can be re-added in
+                    // target's order) -- and re-add every key that should
+                    // remain, with its final target value, in target's
+                    // order. basic_json::patch()'s "add" operation on an
+                    // object uses operator[], which appends at the end for a
+                    // vector-backed insertion-ordered map when the key does
+                    // not already exist -- so removing a key and then adding
+                    // it moves it to the end, fixing its position.
+                    for (auto it = s.cbegin(); it != s.cend(); ++it)
+                    {
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(object(
+                        {
+                            {"op", "remove"}, {"path", path_key}
+                        }));
+                    }
+
+                    // add every key that is either common (just removed
+                    // above) or brand new, in target's iteration order, so
+                    // that the final order after applying the patch matches
+                    // target exactly
+                    for (auto it = t.cbegin(); it != t.cend(); ++it)
+                    {
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(
+                        {
+                            {"op", "add"}, {"path", path_key},
+                            {"value", it.value()}
+                        });
+                    }
+                    return;
+                }
+
+                case value_t::null:
+                case value_t::string:
+                case value_t::boolean:
+                case value_t::number_integer:
+                case value_t::number_unsigned:
+                case value_t::number_float:
+                case value_t::binary:
+                case value_t::discarded:
+                default:
+                {
+                    // both primitive types: replace value
+                    result.push_back(
+                    {
+                        {"op", "replace"}, {"path", current_path}, {"value", t}
+                    });
+                    return;
+                }
+            }
+        };
+
+        enter(source, target);
+        while (!stack.empty())
+        {
+            diff_frame& frame = stack.back();
+            const std::size_t path_length = frame.path_length;
+            const std::size_t depth = stack.size();
+
+            if (frame.source->is_array())
+            {
+                const auto& source_array = *frame.source->m_data.m_value.array;
+                const auto& target_array = *frame.target->m_data.m_value.array;
+
+                // first pass: traverse common elements
+                if (frame.index < source_array.size() && frame.index < target_array.size())
+                {
+                    const std::size_t i = frame.index++;
+                    detail::concat_into(current_path, '/', detail::to_string<string_t>(i));
+                    enter(source_array[i], target_array[i]); // may push, which invalidates `frame`
+                    if (stack.size() == depth)
+                    {
+                        current_path.resize(path_length);
+                    }
+                    continue;
+                }
+
+                // We now reached the end of at least one array
+                // in a second pass, traverse the remaining elements
+
+                // remove my remaining elements, highest index first; appending
+                // in that order avoids the quadratic reinsertion done before
+                for (std::size_t j = source_array.size(); j > frame.index; --j)
+                {
+                    result.push_back(object(
+                    {
+                        {"op", "remove"},
+                        {"path", detail::concat<string_t>(current_path, '/', detail::to_string<string_t>(j - 1))}
+                    }));
+                }
+
+                // add other remaining elements
+                for (std::size_t i = source_array.size(); i < target_array.size(); ++i)
+                {
+                    result.push_back(
+                    {
+                        {"op", "add"},
+                        {"path", detail::concat<string_t>(current_path, "/-")},
+                        {"value", target_array[i]}
+                    });
+                }
+            }
+            else
+            {
+                if (frame.member != frame.source->cend())
+                {
+                    const const_iterator it = frame.member;
+                    ++frame.member;
+                    if (frame.next_common < frame.common_keys.size() && it.key() == frame.common_keys[frame.next_common])
+                    {
+                        ++frame.next_common;
+                        const basic_json& target_value = (*frame.target)[it.key()];
+                        detail::concat_into(current_path, '/', detail::escape(it.key()));
+                        enter(it.value(), target_value); // may push, which invalidates `frame`
+                        if (stack.size() == depth)
+                        {
+                            current_path.resize(path_length);
+                        }
+                    }
+                    else
+                    {
+                        // found a key that is not in target -> remove it
+                        const auto path_key = detail::concat<string_t>(current_path, '/', detail::escape(it.key()));
+                        result.push_back(object(
+                        {
+                            {"op", "remove"}, {"path", path_key}
+                        }));
+                    }
+                    continue;
+                }
+
+                // append the "add" ops for brand-new keys collected when the
+                // object was entered
+                result.insert(result.end(), frame.added_ops.begin(), frame.added_ops.end());
+            }
+
+            // this array or object is done: continue with the one it is in
+            stack.pop_back();
+            if (!stack.empty())
+            {
+                current_path.resize(stack.back().path_length);
+            }
+        }
+
+        return result;
+    }
+
+  public:
     /// @}
 
     ////////////////////////////////

tests/src/unit-json_patch.cpp

@@ -15,8 +15,65 @@ using nlohmann::json;
 #endif
 
 #include <fstream>
+#include <string>
+#include <vector>
 #include "make_test_data_available.hpp"
 
+namespace
+{
+// alternating objects and arrays nested `depth` levels deep, with members that
+// depend on `variant` at some levels, so diffing two variants yields
+// operations on many levels: replacing the innermost value, adding, removing,
+// and (for ordered_json) reordering members, and changing array lengths
+template<typename BasicJsonType>
+BasicJsonType nested(const std::size_t depth, const int variant)
+{
+    BasicJsonType value = variant;
+    for (std::size_t i = 0; i < depth; ++i)
+    {
+        if (i % 2 == 0)
+        {
+            BasicJsonType object = BasicJsonType::object();
+            if ((i + static_cast<std::size_t>(variant)) % 7 == 0)
+            {
+                object["x"] = i;
+            }
+            if (variant == 2 && i % 11 == 0)
+            {
+                object["z"] = "z";
+            }
+            object["a"] = std::move(value);
+            if (variant == 1 && i % 5 == 0)
+            {
+                object["y"] = 1;
+            }
+            value = std::move(object);
+        }
+        else
+        {
+            BasicJsonType array = BasicJsonType::array({std::move(value)});
+            if ((i + static_cast<std::size_t>(variant)) % 3 == 0)
+            {
+                array.push_back(i);
+            }
+            value = std::move(array);
+        }
+    }
+    return value;
+}
+
+// a path of `depth` reference tokens, as nested() nests its values
+std::string nested_path(const std::size_t depth)
+{
+    std::string path;
+    for (std::size_t i = depth; i > 0; --i)
+    {
+        path += (i - 1) % 2 == 0 ? "/a" : "/0";
+    }
+    return path;
+}
+} // namespace
+
 TEST_CASE("JSON patch")
 {
     SECTION("examples from RFC 6902")
@@ -1751,3 +1808,99 @@ TEST_CASE("JSON patch - diff emits array removals in descending index order")
         CHECK(source.patch(patch) == target);
     }
 }
+
+TEST_CASE("JSON patch: diff of deeply nested values")
+{
+    SECTION("the diff reproduces the target at every depth")
+    {
+        // depths on either side of the nesting depth up to which diff()
+        // recurses (detail::recursion_depth_limit(), 128); not every depth up
+        // to 300, as the test would then time out under Valgrind
+        std::vector<std::size_t> depths;
+        for (std::size_t depth = 0; depth <= 16; ++depth)
+        {
+            depths.push_back(depth);
+        }
+        for (std::size_t depth = 120; depth <= 136; ++depth)
+        {
+            depths.push_back(depth);
+        }
+        depths.push_back(300);
+
+        for (const auto depth : depths)
+        {
+            CAPTURE(depth);
+            for (int from = 0; from < 3; ++from)
+            {
+                for (int to = 0; to < 3; ++to)
+                {
+                    CAPTURE(from);
+                    CAPTURE(to);
+                    const auto source = nested<json>(depth, from);
+                    const auto target = nested<json>(depth, to);
+                    const auto patch = json::diff(source, target);
+                    CHECK(source.patch(patch) == target);
+                    CHECK(patch.empty() == (from == to));
+
+                    const auto ordered_source = nested<nlohmann::ordered_json>(depth, from);
+                    const auto ordered_target = nested<nlohmann::ordered_json>(depth, to);
+                    CHECK(ordered_source.patch(nlohmann::ordered_json::diff(ordered_source, ordered_target)) == ordered_target);
+                }
+            }
+        }
+    }
+
+    SECTION("a difference only in the innermost value is one replace operation")
+    {
+        for (std::size_t depth = 0; depth <= 300; ++depth)
+        {
+            CAPTURE(depth);
+            json source = 1;
+            json target = 2;
+            for (std::size_t i = 0; i < depth; ++i)
+            {
+                source = i % 2 == 0 ? json::object({{"a", std::move(source)}}) : json::array({std::move(source)});
+                target = i % 2 == 0 ? json::object({{"a", std::move(target)}}) : json::array({std::move(target)});
+            }
+            CHECK(json::diff(source, target, "/root") == json::array({{{"op", "replace"}, {"path", "/root" + nested_path(depth)}, {"value", 2}}}));
+        }
+    }
+
+    SECTION("values nested too deeply for the call stack (#5393)")
+    {
+        // diff() used to recurse once per nesting level, and compared the
+        // values with operator== on every level. The values are only
+        // parsed and diffed, never copied or compared, since those recurse
+        // too.
+        const std::size_t depth = 100000;
+        for (const bool objects :
+                {
+                    false, true
+                })
+        {
+            CAPTURE(objects);
+            std::string source_text;
+            std::string target_text;
+            std::string equal_text;
+            std::string path;
+            for (std::size_t i = 0; i < depth; ++i)
+            {
+                source_text += objects ? "{\"a\":" : "[";
+                path += objects ? "/a" : "/0";
+            }
+            target_text = source_text + "2";
+            equal_text = source_text + "1";
+            source_text += "1";
+            const std::string closing(depth, objects ? '}' : ']');
+            const auto source = json::parse(source_text + closing);
+
+            const auto patch = json::diff(source, json::parse(target_text + closing));
+            REQUIRE(patch.size() == 1);
+            CHECK(patch[0]["op"] == "replace");
+            CHECK(patch[0]["path"] == path);
+            CHECK(patch[0]["value"] == 2);
+
+            CHECK(json::diff(source, json::parse(equal_text + closing)).empty());
+        }
+    }
+}

tests/src/unit-ubjson.cpp

@@ -2981,18 +2981,3 @@ TEST_CASE("UBJSON roundtrips" * doctest::skip())
         }
     }
 }
-
-TEST_CASE("UBJSON optimized array of unsigned integers beyond int64")
-{
-    // UBJSON has no unsigned 64-bit type, so such values are written as
-    // high-precision numbers - also as the type of an optimized container
-    const json j = {18446744073709551615ULL, 9223372036854775808ULL};
-    const std::vector<std::uint8_t> expected =
-    {
-        '[', '$', 'H', '#', 'i', 2,
-        'i', 20, '1', '8', '4', '4', '6', '7', '4', '4', '0', '7', '3', '7', '0', '9', '5', '5', '1', '6', '1', '5',
-        'i', 19, '9', '2', '2', '3', '3', '7', '2', '0', '3', '6', '8', '5', '4', '7', '7', '5', '8', '0', '8'
-    };
-    CHECK(json::to_ubjson(j, true, true) == expected);
-    CHECK(json::from_ubjson(expected) == j);
-}