Skip the indentation after a newline 8 bytes at a time

Pretty-printed JSON indents every line with a run of spaces, which
skip_whitespace() read one character at a time. For contiguous input
(bulk_scan) the lexer now skips the spaces after a newline 8 bytes at a
time directly in the input buffer, including the leading spaces of the
last, partial word. This bypasses get(), which is safe because
bulk_scan implies the lazy position from the previous commit: nothing
is counted per character.

Compared to lazy positions alone, pretty-printed input (twitter,
citm_catalog with indent 2 and 4) parses up to 13% faster and accept()
gets up to 30% faster (GCC 13 and clang 18 on x86-64, Apple clang 21 on
arm64). Minified input is unchanged within noise.

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

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

@@ -59,6 +59,7 @@ Currently, only `remove`, `add`, and `replace` operations are generated.
 - [patch](patch.md) applies a JSON Patch
 - [patch_inplace](patch_inplace.md) applies a JSON Patch in place
 - [merge_patch](merge_patch.md) applies a JSON Merge Patch
+- [merge_diff](merge_diff.md) creates a diff as a JSON Merge Patch
 
 ## Version history
 

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

@@ -285,6 +285,7 @@ Access to the JSON value
 ### JSON Merge Patch functions
 
 - [**merge_patch**](merge_patch.md) - applies a JSON Merge Patch
+- [**merge_diff**](merge_diff.md) (_static_) - creates a diff as a JSON Merge Patch
 
 ## Static functions
 

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

@@ -0,0 +1,79 @@
+# <small>nlohmann::basic_json::</small>merge_diff
+
+```cpp
+static basic_json merge_diff(const basic_json& source,
+                             const basic_json& target);
+```
+
+Creates a [JSON Merge Patch](https://tools.ietf.org/html/rfc7396) so that value `source` can be changed into the value
+`target` by calling the [`merge_patch`](merge_patch.md) function. The patch contains only what differs: changed and
+added members with their new values, and removed members with the value `#!json null`. Nested objects are compared
+member by member; any other value that differs, including an array, is replaced as a whole.
+
+For two JSON values `source` and `target`, where `target` contains no object member whose value is `#!json null`, the
+following code always yields `#!cpp true`:
+```cpp
+basic_json patched = source;
+patched.merge_patch(merge_diff(source, target));
+patched == target;
+```
+
+## Parameters
+
+`source` (in)
+:   JSON value to compare from
+
+`target` (in)
+:   JSON value to compare against
+
+## Return value
+
+a JSON Merge Patch to convert the `source` to `target`
+
+## Exception safety
+
+Strong guarantee: `source` and `target` are never modified.
+
+## Complexity
+
+Linear in the total number of values (including nested ones) in `source` and `target`, times the cost of a key-based
+object operation (looking up a key, or adding one to the patch): logarithmic for [`json`](../json.md), and linear for
+[`ordered_json`](../ordered_json.md) (see [`ordered_map` complexity](../ordered_map.md#complexity)). For
+`ordered_json`, diffing two objects with `n` and `m` members therefore takes O((n+m)²).
+
+## Notes
+
+- A merge patch uses `#!json null` to remove a member, so it cannot set a member to `#!json null`. As in RFC 7396, a
+  `#!json null` member of `target` is treated as absent: the patch removes the member from `source` instead of setting
+  it to `#!json null`. A member that is `#!json null` in both `source` and `target` is unchanged and left out of the
+  patch, so `source` keeps it.
+- A merge patch cannot reorder object members, and [`merge_patch`](merge_patch.md) appends added members at the end.
+  For [`ordered_json`](../ordered_json.md), the patched value therefore has the same members as `target`, but not
+  necessarily in the same order, and may compare unequal to it. The patch lists changed and removed members in the
+  order of `source`, followed by added members in the order of `target`.
+
+## Examples
+
+??? example
+
+    The following code shows how a JSON Merge Patch is created as a diff for two JSON values.
+
+    ```cpp
+    --8<-- "examples/merge_diff.cpp"
+    ```
+
+    Output:
+
+    ```json
+    --8<-- "examples/merge_diff.output"
+    ```
+
+## See also
+
+- [RFC 7396 (JSON Merge Patch)](https://tools.ietf.org/html/rfc7396)
+- [merge_patch](merge_patch.md) applies a JSON Merge Patch
+- [diff](diff.md) creates a diff as a JSON Patch
+
+## Version history
+
+- Added in version 3.13.0.

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

@@ -67,6 +67,7 @@ returned by `#!cpp (*this)[key]`); it is read as it was when `merge_patch()` was
 
 - [RFC 7396 (JSON Merge Patch)](https://tools.ietf.org/html/rfc7396)
 - [patch](patch.md) apply a JSON patch
+- [merge_diff](merge_diff.md) creates a diff as a JSON Merge Patch
 
 ## Version history
 

docs/mkdocs/docs/examples/merge_diff.cpp

@@ -0,0 +1,51 @@
+#include <iostream>
+#include <iomanip>
+#include <nlohmann/json.hpp>
+
+using json = nlohmann::json;
+using namespace nlohmann::literals;
+
+int main()
+{
+    // the source document
+    json source = R"(
+        {
+            "title": "Goodbye!",
+            "author": {
+                "givenName": "John",
+                "familyName": "Doe"
+            },
+            "tags": [
+                "example",
+                "sample"
+            ],
+            "content": "This will be unchanged"
+        }
+    )"_json;
+
+    // the target document
+    json target = R"(
+        {
+            "title": "Hello!",
+            "author": {
+                "givenName": "John"
+            },
+            "tags": [
+                "example"
+            ],
+            "content": "This will be unchanged",
+            "phoneNumber": "+01-123-456-7890"
+        }
+    )"_json;
+
+    // create the patch
+    json patch = json::merge_diff(source, target);
+
+    // roundtrip
+    json patched_source = source;
+    patched_source.merge_patch(patch);
+
+    // output patch and roundtrip result
+    std::cout << std::setw(4) << patch << "\n\n"
+              << std::setw(4) << patched_source << std::endl;
+}

docs/mkdocs/docs/examples/merge_diff.output

@@ -0,0 +1,22 @@
+{
+    "author": {
+        "familyName": null
+    },
+    "phoneNumber": "+01-123-456-7890",
+    "tags": [
+        "example"
+    ],
+    "title": "Hello!"
+}
+
+{
+    "author": {
+        "givenName": "John"
+    },
+    "content": "This will be unchanged",
+    "phoneNumber": "+01-123-456-7890",
+    "tags": [
+        "example"
+    ],
+    "title": "Hello!"
+}

docs/mkdocs/docs/features/merge_patch.md

@@ -9,6 +9,9 @@ syntax that closely mimics the document being modified. Unlike [JSON Patch](json
 express every kind of change (e.g., it cannot reorder array elements or remove a specific array element), but it is
 easier to read and write for object-shaped documents.
 
+The function [`merge_diff`](../api/basic_json/merge_diff.md) creates a merge patch that changes one JSON value into
+another.
+
 !!! tip "Not a general deep merge"
 
     A merge patch is not a general deep merge: a `#!json null` value in the patch deletes the key from the target.
@@ -35,4 +38,5 @@ easier to read and write for object-shaped documents.
 - [JSON Patch and Diff](json_patch.md) - a more expressive alternative that describes a sequence of operations
 - [JSON Pointer](json_pointer.md) - the addressing scheme used by JSON Patch
 - Function [`merge_patch`](../api/basic_json/merge_patch.md)
+- Function [`merge_diff`](../api/basic_json/merge_diff.md)
 - Function [`update`](../api/basic_json/update.md) - merge objects, optionally recursively

docs/mkdocs/docs/home/architecture.md

@@ -185,9 +185,11 @@ The lexer detects two optional extensions at compile time. Only `iterator_input_
 random-access input of single-byte characters:
 
 - `supports_seek`, `get_consumed_count()`, and `copy_consumed_range()` let the lexer reconstruct already consumed input
-  for error messages instead of copying every character it reads.
+  for error messages instead of copying every character it reads. With `get_read_count()` and
+  `count_consumed_lines()`, the lexer also derives the read position from the adapter and counts lines only when an
+  error needs them, instead of counting characters and lines on every read.
 - `supports_bulk_scan`, `bulk_data()`, `bulk_remaining()`, and `bulk_skip()` let the lexer scan strings directly in
-  contiguous memory, several bytes at a time.
+  contiguous memory, several bytes at a time, and skip the spaces that indent a line in 8-byte steps.
 
 The function `input_adapter` picks the right adapter for the argument passed to `parse`, `accept`, `sax_parse`, or the
 `from_*` functions:

docs/mkdocs/mkdocs.yml

@@ -182,6 +182,7 @@ nav:
           - 'json_serializer': api/basic_json/json_serializer.md
           - 'max_size': api/basic_json/max_size.md
           - 'meta': api/basic_json/meta.md
+          - 'merge_diff': api/basic_json/merge_diff.md
           - 'merge_patch': api/basic_json/merge_patch.md
           - 'number_float_t': api/basic_json/number_float_t.md
           - 'number_integer_t': api/basic_json/number_integer_t.md

include/nlohmann/detail/bit_ops.hpp

@@ -42,6 +42,25 @@ inline int count_leading_zeros(std::uint64_t x) noexcept
 #endif
 }
 
+/// number of trailing zero bits of x (x != 0)
+inline int count_trailing_zeros(std::uint64_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_ctzll(x);
+#else
+    int n = 0;
+    for (int shift = 32; shift != 0; shift >>= 1)
+    {
+        if ((x << (64 - shift)) == 0)
+        {
+            n += shift;
+            x >>= shift;
+        }
+    }
+    return n;
+#endif
+}
+
 /// the 128-bit product of two 64-bit numbers
 struct uint128_parts
 {

include/nlohmann/detail/input/input_adapters.hpp

@@ -8,12 +8,12 @@
 
 #pragma once
 
-#include <algorithm> // min
+#include <algorithm> // count, find, 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
+#include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
@@ -297,6 +297,7 @@ class iterator_input_adapter
             return result;
         }
 
+        ++eof_reads;
         return char_traits<char_type>::eof();
     }
 
@@ -306,6 +307,26 @@ class iterator_input_adapter
         return static_cast<std::size_t>(std::distance(begin, current));
     }
 
+    // number of get_character() calls so far, including those that returned
+    // EOF; the lexer derives its position from it (see lexer::lazy_position)
+    std::size_t get_read_count() const
+    {
+        return get_consumed_count() + eof_reads;
+    }
+
+    // number of newlines in the consumed range [0, last_index); @a line_start
+    // is set to the index after the last of them (0 if there is none); only
+    // valid when supports_seek
+    std::size_t count_consumed_lines(std::size_t last_index, std::size_t& line_start) const
+    {
+        const auto to = std::next(begin, static_cast<typename std::iterator_traits<IteratorType>::difference_type>(last_index));
+        const auto newline = char_traits<char_type>::to_char_type('\n');
+        using reverse_it = std::reverse_iterator<IteratorType>;
+        const auto last_newline = std::find(reverse_it(to), reverse_it(begin), newline);
+        line_start = static_cast<std::size_t>(std::distance(begin, last_newline.base()));
+        return static_cast<std::size_t>(std::count(begin, last_newline.base(), newline));
+    }
+
     // append the already-consumed characters in the half-open range
     // [first_index, last_index) to @a out; only valid when supports_seek
     template<typename ContainerType>
@@ -422,6 +443,8 @@ class iterator_input_adapter
     IteratorType begin;
     IteratorType current;
     SentinelType end;
+    /// number of get_character() calls that returned EOF
+    std::size_t eof_reads = 0;
 
     template<typename BaseInputAdapter, size_t T>
     friend struct wide_string_input_helper;

include/nlohmann/detail/input/json_sax.hpp

@@ -217,7 +217,7 @@ struct diagnostic_positions
         {
             // Lexer has read the first character of the container, so
             // subtract 1 from the position to get the correct start position.
-            v.start_position = lexer->get_position() - 1;
+            v.start_position = lexer->get_chars_read_total() - 1;
         }
     }
 
@@ -231,7 +231,7 @@ struct diagnostic_positions
         if (lexer)
         {
             // Lexer's position is past the closing brace or bracket, so set that as the end position.
-            v.end_position = lexer->get_position();
+            v.end_position = lexer->get_chars_read_total();
         }
     }
 
@@ -247,7 +247,7 @@ struct diagnostic_positions
             // Lexer has read past the current field value, so set the end position to the current position.
             // The start position will be set below based on the length of the string representation
             // of the value.
-            v.end_position = lexer->get_position();
+            v.end_position = lexer->get_chars_read_total();
 
             switch (v.type())
             {

include/nlohmann/detail/input/lexer.hpp

@@ -17,6 +17,7 @@
 #include <utility> // move
 #include <vector> // vector
 
+#include <nlohmann/detail/bit_ops.hpp>
 #include <nlohmann/detail/input/input_adapters.hpp>
 #include <nlohmann/detail/input/number_parse.hpp>
 #include <nlohmann/detail/input/position_t.hpp>
@@ -186,6 +187,15 @@ class lexer : public lexer_base<BasicJsonType>
     static constexpr bool lazy_token_string =
         input_adapter_supports_seek<InputAdapterType>(is_detected<detect_supports_seek, InputAdapterType> {});
 
+    /// whether the read position is derived from the input adapter on demand
+    /// instead of being counted on every scanned character. Seekable adapters
+    /// know how much input they consumed and can revisit it, so
+    /// chars_read_total is the adapter's read count (minus a pending unget),
+    /// and line and column are recomputed from the consumed input only when
+    /// get_position() is called (errors). Counting per character kept two
+    /// increments and a newline check in the hot path of every get().
+    static constexpr bool lazy_position = lazy_token_string;
+
     /// whether a simulated unget can be passed on to the input adapter, which
     /// then leaves the character in the input; see
     /// input_adapter_supports_lookahead
@@ -195,7 +205,9 @@ class lexer : public lexer_base<BasicJsonType>
     /// whether string scanning may bulk-consume runs of ordinary characters
     /// directly from a contiguous input buffer (SWAR fast path). This requires
     /// the token to be reconstructible lazily (lazy_token_string), so bypassing
-    /// the per-character capture in get() cannot lose error diagnostics.
+    /// the per-character capture in get() cannot lose error diagnostics. It
+    /// also implies lazy_position, so the bulk paths need not advance the
+    /// position counters.
     static constexpr bool bulk_scan =
         lazy_token_string
         && input_adapter_supports_bulk_scan<InputAdapterType>(is_detected<detect_supports_bulk_scan, InputAdapterType> {});
@@ -331,10 +343,6 @@ class lexer : public lexer_base<BasicJsonType>
         }
         token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
         ia.bulk_skip(pos);
-        // the run contains no newline (all bytes < 0x20 are treated as special),
-        // so only the flat character counters advance
-        position.chars_read_total += pos;
-        position.chars_read_current_line += pos;
     }
 
     /// streaming input: no bulk fast path
@@ -1567,8 +1575,6 @@ scan_number_done:
             if (JSON_HEDLEY_LIKELY(integer_result != token_type::uninitialized))
             {
                 ia.bulk_skip(len - 1);
-                position.chars_read_total += (len - 1);
-                position.chars_read_current_line += (len - 1);
                 return integer_result;
             }
             // The value does not fit an integer, so this token converts as a
@@ -1585,8 +1591,6 @@ scan_number_done:
         decimal_point_position = dot_index;
 
         ia.bulk_skip(len - 1);
-        position.chars_read_total += (len - 1);
-        position.chars_read_current_line += (len - 1);
 
         return convert_number(number_type, mantissa_end);
     }
@@ -1637,7 +1641,7 @@ scan_number_done:
 
 #if JSON_DIAGNOSTIC_POSITIONS
         // the first character of the token has already been read, hence the -1
-        token_start_position = position.chars_read_total - 1;
+        token_start_position = get_chars_read_total() - 1;
 #endif
 
         note_token_start(std::integral_constant<bool, lazy_token_string> {});
@@ -1690,6 +1694,15 @@ scan_number_done:
     /// per-character position counters (line-count-on-'\n' bookkeeping is
     /// handled afterwards, in track_after_read(), once `current` is known)
     void advance_position() noexcept
+    {
+        advance_position_impl(std::integral_constant<bool, lazy_position> {});
+    }
+
+    /// seekable adapter: the position is derived from the adapter on demand
+    void advance_position_impl(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: count the character
+    void advance_position_impl(std::false_type /*lazy*/) noexcept
     {
         ++position.chars_read_total;
         ++position.chars_read_current_line;
@@ -1703,7 +1716,16 @@ scan_number_done:
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
+        track_newline(std::integral_constant<bool, lazy_position> {});
+        return current;
+    }
 
+    /// seekable adapter: lines are counted on demand in get_position()
+    void track_newline(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: start a new line after reading a newline
+    void track_newline(std::false_type /*lazy*/) noexcept
+    {
         if (current == '\n')
         {
             ++position.lines_read;
@@ -1712,8 +1734,6 @@ scan_number_done:
             chars_read_before_newline = position.chars_read_current_line;
             position.chars_read_current_line = 0;
         }
-
-        return current;
     }
 
     /*!
@@ -1757,12 +1777,22 @@ scan_number_done:
     We implement unget by setting variable next_unget to true. The input is not
     changed - we just simulate ungetting by modifying chars_read_total,
     chars_read_current_line, and token_string. The next call to get() will
-    behave as if the unget character is read again.
+    behave as if the unget character is read again. With lazy_position, the
+    pending next_unget is all get_chars_read_total() needs.
     */
     void unget()
     {
         next_unget = true;
+        unget_position(std::integral_constant<bool, lazy_position> {});
+        uncapture_char(std::integral_constant<bool, lazy_token_string> {});
+    }
 
+    /// seekable adapter: get_chars_read_total() accounts for next_unget
+    void unget_position(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: step the counters back over the ungotten character
+    void unget_position(std::false_type /*lazy*/) noexcept
+    {
         --position.chars_read_total;
 
         // in case we "unget" a newline, we have to also decrement the lines_read
@@ -1785,8 +1815,6 @@ scan_number_done:
         {
             --position.chars_read_current_line;
         }
-
-        uncapture_char(std::integral_constant<bool, lazy_token_string> {});
     }
 
     /// adapter without lookahead: nothing to do (see release_lookahead)
@@ -1857,10 +1885,21 @@ scan_number_done:
     // diagnostics
     /////////////////////
 
-    /// return position of last read token
-    constexpr position_t get_position() const noexcept
+    /*!
+    @brief return the number of characters read so far
+
+    Unlike get_position(), this is O(1) for all adapters, so it may be called
+    for every value (JSON_DIAGNOSTIC_POSITIONS).
+    */
+    std::size_t get_chars_read_total() const noexcept
     {
-        return position;
+        return get_chars_read_total_impl(std::integral_constant<bool, lazy_position> {});
+    }
+
+    /// return position of last read token
+    position_t get_position() const
+    {
+        return get_position_impl(std::integral_constant<bool, lazy_position> {});
     }
 
     /*!
@@ -1897,6 +1936,43 @@ scan_number_done:
     }
 #endif
 
+  private:
+    /// seekable adapter: every get() either read from the adapter (including
+    /// reads at EOF, which get() counts as characters as well) or replayed an
+    /// ungotten character, so the count is the adapter's read count minus a
+    /// still pending unget
+    std::size_t get_chars_read_total_impl(std::true_type /*lazy*/) const noexcept
+    {
+        return ia.get_read_count() - (next_unget ? 1u : 0u);
+    }
+
+    /// streaming adapter: the counter is maintained by get() and unget()
+    std::size_t get_chars_read_total_impl(std::false_type /*lazy*/) const noexcept
+    {
+        return position.chars_read_total;
+    }
+
+    /// seekable adapter: count the lines in the consumed input; O(input read
+    /// so far), but only called on errors
+    position_t get_position_impl(std::true_type /*lazy*/) const
+    {
+        position_t result;
+        result.chars_read_total = get_chars_read_total();
+        // reads at EOF count as characters but are not part of the input
+        const std::size_t consumed = (std::min)(result.chars_read_total, ia.get_consumed_count());
+        std::size_t line_start = 0;
+        result.lines_read = ia.count_consumed_lines(consumed, line_start);
+        result.chars_read_current_line = result.chars_read_total - line_start;
+        return result;
+    }
+
+    /// streaming adapter: the position is maintained by get() and unget()
+    position_t get_position_impl(std::false_type /*lazy*/) const noexcept
+    {
+        return position;
+    }
+
+  public:
     /// seekable adapter: rebuild the last read token from the input on demand
     const std::vector<char_type>& collect_token_chars(std::vector<char_type>& out, std::true_type /*lazy*/) const
     {
@@ -2004,15 +2080,57 @@ scan_number_done:
         // into one that reloads the pointer from memory every character
         do
         {
+            if (current == '\n')
+            {
+                skip_indentation(std::integral_constant<bool, bulk_scan> {});
+            }
             get_ignoring_pending_unget();
         }
         while (current_is_whitespace());
     }
 
+    /*!
+    @brief contiguous input: skip the space indentation that follows a newline
+           8 bytes at a time
+
+    Pretty-printed JSON indents every line with a run of spaces, which the loop
+    in skip_whitespace() would otherwise read one character at a time. The
+    spaces are skipped 8 bytes at a time; the last 7 bytes of the input and any
+    other whitespace are left to that loop. Skipping bypasses get(), which is
+    fine because bulk_scan implies lazy_position: the position is derived from
+    the adapter, not counted per character.
+    */
+    void skip_indentation(std::true_type /*bulk*/)
+    {
+        JSON_ASSERT(!next_unget);
+        const std::size_t remaining = ia.bulk_remaining();
+        if (remaining < 8)
+        {
+            return;
+        }
+        const char* const data = reinterpret_cast<const char*>(ia.bulk_data());
+        std::size_t pos = 0;
+        for (; pos + 8 <= remaining; pos += 8)
+        {
+            // little-endian, so the first byte is the lowest
+            const std::uint64_t non_space = read_eight_bytes(data + pos) ^ 0x2020202020202020ull;
+            if (non_space != 0)
+            {
+                // skip the spaces before the first other byte of this word
+                pos += static_cast<std::size_t>(count_trailing_zeros(non_space)) / 8;
+                break;
+            }
+        }
+        ia.bulk_skip(pos);
+    }
+
+    /// streaming input: no bulk fast path
+    void skip_indentation(std::false_type /*bulk*/) const noexcept {}
+
     token_type scan()
     {
         // initially, skip the BOM
-        if (position.chars_read_total == 0 && !skip_bom())
+        if (get_chars_read_total() == 0 && !skip_bom())
         {
             error_message = "invalid BOM; must be 0xEF 0xBB 0xBF if given";
             return token_type::parse_error;
@@ -2039,7 +2157,7 @@ scan_number_done:
     token_type scan_expecting(token_type expected_type)
     {
         JSON_ASSERT(expected_type == token_type::name_separator || expected_type == token_type::value_separator);
-        JSON_ASSERT(position.chars_read_total > 0);
+        JSON_ASSERT(get_chars_read_total() > 0);
         const char_int_type expected_char = static_cast<unsigned char>((expected_type == token_type::name_separator) ? ':' : ',');
         skip_whitespace();
         if (JSON_HEDLEY_LIKELY(current == expected_char))

include/nlohmann/detail/input/parser.hpp

@@ -370,7 +370,7 @@ class parser
                     }
                     case token_type::end_of_input:
                     {
-                        if (JSON_HEDLEY_UNLIKELY(m_lexer.get_position().chars_read_total == 1))
+                        if (JSON_HEDLEY_UNLIKELY(m_lexer.get_chars_read_total() == 1))
                         {
                             return syntax_error(*sax, "attempting to parse an empty input; check that your input string or stream contains the expected JSON");
                         }
@@ -520,8 +520,10 @@ class parser
     {
         // MSVC 2015 reports C4100 (unreferenced parameter) when SAX::parse_error is static
         static_cast<void>(sax);
-        return sax.parse_error(m_lexer.get_position(), m_lexer.get_token_string(),
-                               parse_error::create(exception_id::syntax_error, m_lexer.get_position(), message, nullptr));
+        // computing the position is O(input read so far) for seekable input
+        const position_t position = m_lexer.get_position();
+        return sax.parse_error(position, m_lexer.get_token_string(),
+                               parse_error::create(exception_id::syntax_error, position, message, nullptr));
     }
 
     std::string exception_message(const token_type expected, const std::string& context)

include/nlohmann/json.hpp

@@ -7449,6 +7449,176 @@ public:
     }
 
   public:
+    /// @brief creates a diff as a JSON Merge Patch
+    /// @sa https://json.nlohmann.me/api/basic_json/merge_diff/
+    JSON_HEDLEY_WARN_UNUSED_RESULT
+    static basic_json merge_diff(const basic_json& source, const basic_json& target)
+    {
+        // unless both are objects, the patch is the target itself
+        if (!source.is_object() || !target.is_object())
+        {
+            return target;
+        }
+
+        return merge_diff_recursively(source, target, 0);
+    }
+
+  private:
+    /// @brief two objects @ref merge_diff_iteratively is diffing
+    struct merge_diff_frame
+    {
+        merge_diff_frame(const basic_json* source_, const basic_json* target_)
+            : source(source_), target(target_), member(source_->cbegin()), patch(value_t::object)
+        {}
+
+        // declared for GCC's -Weffc++, as in diff_frame
+        merge_diff_frame(const merge_diff_frame&) = default;
+        merge_diff_frame(merge_diff_frame&&) = default;
+        merge_diff_frame& operator=(const merge_diff_frame&) = default;
+        merge_diff_frame& operator=(merge_diff_frame&&) = default;
+        ~merge_diff_frame() = default;
+
+        /// the objects being diffed
+        const basic_json* source;
+        const basic_json* target;
+        /// the member of source being diffed; advanced once it is done
+        const_iterator member;
+        /// the patch for the members of source diffed so far
+        basic_json patch;
+    };
+
+    /// @brief add the members of object @a target that @a source does not
+    ///        have to the merge patch @a result
+    static void merge_diff_added_members(basic_json& result, const basic_json& source, const basic_json& target)
+    {
+        for (auto it = target.begin(); it != target.end(); ++it)
+        {
+            auto itf = source.find(it.key());
+            if (itf == source.end())
+            {
+                result[it.key()] = it.value();
+            }
+        }
+    }
+
+    /*!
+    @brief @ref merge_diff of two objects at nesting level @a depth
+
+    Diffing two nested objects calls this function again, once per nesting
+    level, so values nested deeply enough would exhaust the call stack and
+    terminate the process. The descent is bounded here: once @ref
+    detail::recursion_depth_limit levels have been entered, @ref
+    merge_diff_iteratively diffs what is left without the call stack.
+    */
+    static basic_json merge_diff_recursively(const basic_json& source, const basic_json& target, const std::size_t depth)
+    {
+        JSON_ASSERT(source.is_object() && target.is_object());
+
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return merge_diff_iteratively(source, target);
+        }
+
+        basic_json result(value_t::object);
+
+        for (auto it = source.begin(); it != source.end(); ++it)
+        {
+            auto itf = target.find(it.key());
+            if (itf != target.end())
+            {
+                if (it.value().is_object() && itf.value().is_object())
+                {
+                    auto diff = merge_diff_recursively(it.value(), itf.value(), depth + 1);
+                    if (!diff.empty())
+                    {
+                        result[it.key()] = std::move(diff);
+                    }
+                }
+                else if (it.value() != itf.value())
+                {
+                    result[it.key()] = itf.value();
+                }
+            }
+            else
+            {
+                result[it.key()] = value_t::null;
+            }
+        }
+
+        merge_diff_added_members(result, source, target);
+        return result;
+    }
+
+    /*!
+    @brief @ref merge_diff of two objects without the call stack
+
+    Produces the same patch as @ref merge_diff_recursively, keeping the objects
+    being diffed on an explicit stack instead of descending into them. Two
+    nested objects are still diffed completely before the next member, so the
+    patch has its members in the same order as in the recursive version. Only
+    reached for objects nested more deeply than @ref
+    detail::recursion_depth_limit.
+    */
+    static basic_json merge_diff_iteratively(const basic_json& source, const basic_json& target)
+    {
+        // the patch of source and target, set once their frame is popped
+        basic_json result;
+        std::vector<merge_diff_frame> stack;
+        stack.emplace_back(&source, &target);
+        while (!stack.empty())
+        {
+            // the frame's members are copied out and it is changed through
+            // stack.back(): pushing a frame for nested objects or popping
+            // this one would invalidate a reference to it
+            const basic_json* const s = stack.back().source;
+            const basic_json* const t = stack.back().target;
+            const const_iterator it = stack.back().member;
+
+            if (it != s->cend())
+            {
+                const const_iterator itf = t->find(it.key());
+                if (itf != t->cend())
+                {
+                    if (it.value().is_object() && itf.value().is_object())
+                    {
+                        // member is advanced once their patch is under its key
+                        stack.emplace_back(&it.value(), &itf.value());
+                        continue;
+                    }
+                    if (it.value() != itf.value())
+                    {
+                        stack.back().patch[it.key()] = itf.value();
+                    }
+                }
+                else
+                {
+                    stack.back().patch[it.key()] = value_t::null;
+                }
+                ++stack.back().member;
+                continue;
+            }
+
+            merge_diff_added_members(stack.back().patch, *s, *t);
+
+            // these objects are done: continue with the ones they are in
+            basic_json patch = std::move(stack.back().patch);
+            stack.pop_back();
+            if (stack.empty())
+            {
+                result = std::move(patch);
+            }
+            else
+            {
+                if (!patch.empty())
+                {
+                    stack.back().patch[stack.back().member.key()] = std::move(patch);
+                }
+                ++stack.back().member;
+            }
+        }
+        return result;
+    }
+
     /// @}
 };
 

single_include/nlohmann/json.hpp

@@ -7963,12 +7963,12 @@ NLOHMANN_JSON_NAMESPACE_END
 
 
 
-#include <algorithm> // min
+#include <algorithm> // count, find, 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
+#include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
@@ -8257,6 +8257,7 @@ class iterator_input_adapter
             return result;
         }
 
+        ++eof_reads;
         return char_traits<char_type>::eof();
     }
 
@@ -8266,6 +8267,26 @@ class iterator_input_adapter
         return static_cast<std::size_t>(std::distance(begin, current));
     }
 
+    // number of get_character() calls so far, including those that returned
+    // EOF; the lexer derives its position from it (see lexer::lazy_position)
+    std::size_t get_read_count() const
+    {
+        return get_consumed_count() + eof_reads;
+    }
+
+    // number of newlines in the consumed range [0, last_index); @a line_start
+    // is set to the index after the last of them (0 if there is none); only
+    // valid when supports_seek
+    std::size_t count_consumed_lines(std::size_t last_index, std::size_t& line_start) const
+    {
+        const auto to = std::next(begin, static_cast<typename std::iterator_traits<IteratorType>::difference_type>(last_index));
+        const auto newline = char_traits<char_type>::to_char_type('\n');
+        using reverse_it = std::reverse_iterator<IteratorType>;
+        const auto last_newline = std::find(reverse_it(to), reverse_it(begin), newline);
+        line_start = static_cast<std::size_t>(std::distance(begin, last_newline.base()));
+        return static_cast<std::size_t>(std::count(begin, last_newline.base(), newline));
+    }
+
     // append the already-consumed characters in the half-open range
     // [first_index, last_index) to @a out; only valid when supports_seek
     template<typename ContainerType>
@@ -8382,6 +8403,8 @@ class iterator_input_adapter
     IteratorType begin;
     IteratorType current;
     SentinelType end;
+    /// number of get_character() calls that returned EOF
+    std::size_t eof_reads = 0;
 
     template<typename BaseInputAdapter, size_t T>
     friend struct wide_string_input_helper;
@@ -8902,30 +8925,6 @@ NLOHMANN_JSON_NAMESPACE_END
 #include <utility> // move
 #include <vector> // vector
 
-// #include <nlohmann/detail/input/input_adapters.hpp>
-
-// #include <nlohmann/detail/input/number_parse.hpp>
-//     __ _____ _____ _____
-//  __|  |   __|     |   | |  JSON for Modern C++
-// |  |  |__   |  |  | | | |  version 3.12.0
-// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
-//
-// SPDX-FileCopyrightText: 2021 The fast_float authors <https://github.com/fastfloat/fast_float>
-// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
-// SPDX-License-Identifier: MIT
-
-
-
-#include <array> // array
-#include <cfloat> // FLT_EVAL_METHOD
-#include <clocale> // localeconv
-#include <cstddef> // size_t
-#include <cstdint> // int64_t, uint64_t
-#include <cstdlib> // strtof, strtod, strtold
-#include <cstring> // memcpy
-#include <limits> // numeric_limits
-#include <string> // string
-
 // #include <nlohmann/detail/bit_ops.hpp>
 //     __ _____ _____ _____
 //  __|  |   __|     |   | |  JSON for Modern C++
@@ -8972,6 +8971,25 @@ inline int count_leading_zeros(std::uint64_t x) noexcept
 #endif
 }
 
+/// number of trailing zero bits of x (x != 0)
+inline int count_trailing_zeros(std::uint64_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_ctzll(x);
+#else
+    int n = 0;
+    for (int shift = 32; shift != 0; shift >>= 1)
+    {
+        if ((x << (64 - shift)) == 0)
+        {
+            n += shift;
+            x >>= shift;
+        }
+    }
+    return n;
+#endif
+}
+
 /// the 128-bit product of two 64-bit numbers
 struct uint128_parts
 {
@@ -9019,6 +9037,32 @@ inline std::uint64_t read_eight_bytes(const char* p) noexcept
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END
 
+// #include <nlohmann/detail/input/input_adapters.hpp>
+
+// #include <nlohmann/detail/input/number_parse.hpp>
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2021 The fast_float authors <https://github.com/fastfloat/fast_float>
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+
+
+#include <array> // array
+#include <cfloat> // FLT_EVAL_METHOD
+#include <clocale> // localeconv
+#include <cstddef> // size_t
+#include <cstdint> // int64_t, uint64_t
+#include <cstdlib> // strtof, strtod, strtold
+#include <cstring> // memcpy
+#include <limits> // numeric_limits
+#include <string> // string
+
+// #include <nlohmann/detail/bit_ops.hpp>
+
 // #include <nlohmann/detail/input/pow5_table.hpp>
 //     __ _____ _____ _____
 //  __|  |   __|     |   | |  JSON for Modern C++
@@ -10584,6 +10628,15 @@ class lexer : public lexer_base<BasicJsonType>
     static constexpr bool lazy_token_string =
         input_adapter_supports_seek<InputAdapterType>(is_detected<detect_supports_seek, InputAdapterType> {});
 
+    /// whether the read position is derived from the input adapter on demand
+    /// instead of being counted on every scanned character. Seekable adapters
+    /// know how much input they consumed and can revisit it, so
+    /// chars_read_total is the adapter's read count (minus a pending unget),
+    /// and line and column are recomputed from the consumed input only when
+    /// get_position() is called (errors). Counting per character kept two
+    /// increments and a newline check in the hot path of every get().
+    static constexpr bool lazy_position = lazy_token_string;
+
     /// whether a simulated unget can be passed on to the input adapter, which
     /// then leaves the character in the input; see
     /// input_adapter_supports_lookahead
@@ -10593,7 +10646,9 @@ class lexer : public lexer_base<BasicJsonType>
     /// whether string scanning may bulk-consume runs of ordinary characters
     /// directly from a contiguous input buffer (SWAR fast path). This requires
     /// the token to be reconstructible lazily (lazy_token_string), so bypassing
-    /// the per-character capture in get() cannot lose error diagnostics.
+    /// the per-character capture in get() cannot lose error diagnostics. It
+    /// also implies lazy_position, so the bulk paths need not advance the
+    /// position counters.
     static constexpr bool bulk_scan =
         lazy_token_string
         && input_adapter_supports_bulk_scan<InputAdapterType>(is_detected<detect_supports_bulk_scan, InputAdapterType> {});
@@ -10729,10 +10784,6 @@ class lexer : public lexer_base<BasicJsonType>
         }
         token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
         ia.bulk_skip(pos);
-        // the run contains no newline (all bytes < 0x20 are treated as special),
-        // so only the flat character counters advance
-        position.chars_read_total += pos;
-        position.chars_read_current_line += pos;
     }
 
     /// streaming input: no bulk fast path
@@ -11965,8 +12016,6 @@ scan_number_done:
             if (JSON_HEDLEY_LIKELY(integer_result != token_type::uninitialized))
             {
                 ia.bulk_skip(len - 1);
-                position.chars_read_total += (len - 1);
-                position.chars_read_current_line += (len - 1);
                 return integer_result;
             }
             // The value does not fit an integer, so this token converts as a
@@ -11983,8 +12032,6 @@ scan_number_done:
         decimal_point_position = dot_index;
 
         ia.bulk_skip(len - 1);
-        position.chars_read_total += (len - 1);
-        position.chars_read_current_line += (len - 1);
 
         return convert_number(number_type, mantissa_end);
     }
@@ -12035,7 +12082,7 @@ scan_number_done:
 
 #if JSON_DIAGNOSTIC_POSITIONS
         // the first character of the token has already been read, hence the -1
-        token_start_position = position.chars_read_total - 1;
+        token_start_position = get_chars_read_total() - 1;
 #endif
 
         note_token_start(std::integral_constant<bool, lazy_token_string> {});
@@ -12088,6 +12135,15 @@ scan_number_done:
     /// per-character position counters (line-count-on-'\n' bookkeeping is
     /// handled afterwards, in track_after_read(), once `current` is known)
     void advance_position() noexcept
+    {
+        advance_position_impl(std::integral_constant<bool, lazy_position> {});
+    }
+
+    /// seekable adapter: the position is derived from the adapter on demand
+    void advance_position_impl(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: count the character
+    void advance_position_impl(std::false_type /*lazy*/) noexcept
     {
         ++position.chars_read_total;
         ++position.chars_read_current_line;
@@ -12101,7 +12157,16 @@ scan_number_done:
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
+        track_newline(std::integral_constant<bool, lazy_position> {});
+        return current;
+    }
 
+    /// seekable adapter: lines are counted on demand in get_position()
+    void track_newline(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: start a new line after reading a newline
+    void track_newline(std::false_type /*lazy*/) noexcept
+    {
         if (current == '\n')
         {
             ++position.lines_read;
@@ -12110,8 +12175,6 @@ scan_number_done:
             chars_read_before_newline = position.chars_read_current_line;
             position.chars_read_current_line = 0;
         }
-
-        return current;
     }
 
     /*!
@@ -12155,12 +12218,22 @@ scan_number_done:
     We implement unget by setting variable next_unget to true. The input is not
     changed - we just simulate ungetting by modifying chars_read_total,
     chars_read_current_line, and token_string. The next call to get() will
-    behave as if the unget character is read again.
+    behave as if the unget character is read again. With lazy_position, the
+    pending next_unget is all get_chars_read_total() needs.
     */
     void unget()
     {
         next_unget = true;
+        unget_position(std::integral_constant<bool, lazy_position> {});
+        uncapture_char(std::integral_constant<bool, lazy_token_string> {});
+    }
 
+    /// seekable adapter: get_chars_read_total() accounts for next_unget
+    void unget_position(std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: step the counters back over the ungotten character
+    void unget_position(std::false_type /*lazy*/) noexcept
+    {
         --position.chars_read_total;
 
         // in case we "unget" a newline, we have to also decrement the lines_read
@@ -12183,8 +12256,6 @@ scan_number_done:
         {
             --position.chars_read_current_line;
         }
-
-        uncapture_char(std::integral_constant<bool, lazy_token_string> {});
     }
 
     /// adapter without lookahead: nothing to do (see release_lookahead)
@@ -12255,10 +12326,21 @@ scan_number_done:
     // diagnostics
     /////////////////////
 
-    /// return position of last read token
-    constexpr position_t get_position() const noexcept
+    /*!
+    @brief return the number of characters read so far
+
+    Unlike get_position(), this is O(1) for all adapters, so it may be called
+    for every value (JSON_DIAGNOSTIC_POSITIONS).
+    */
+    std::size_t get_chars_read_total() const noexcept
     {
-        return position;
+        return get_chars_read_total_impl(std::integral_constant<bool, lazy_position> {});
+    }
+
+    /// return position of last read token
+    position_t get_position() const
+    {
+        return get_position_impl(std::integral_constant<bool, lazy_position> {});
     }
 
     /*!
@@ -12295,6 +12377,43 @@ scan_number_done:
     }
 #endif
 
+  private:
+    /// seekable adapter: every get() either read from the adapter (including
+    /// reads at EOF, which get() counts as characters as well) or replayed an
+    /// ungotten character, so the count is the adapter's read count minus a
+    /// still pending unget
+    std::size_t get_chars_read_total_impl(std::true_type /*lazy*/) const noexcept
+    {
+        return ia.get_read_count() - (next_unget ? 1u : 0u);
+    }
+
+    /// streaming adapter: the counter is maintained by get() and unget()
+    std::size_t get_chars_read_total_impl(std::false_type /*lazy*/) const noexcept
+    {
+        return position.chars_read_total;
+    }
+
+    /// seekable adapter: count the lines in the consumed input; O(input read
+    /// so far), but only called on errors
+    position_t get_position_impl(std::true_type /*lazy*/) const
+    {
+        position_t result;
+        result.chars_read_total = get_chars_read_total();
+        // reads at EOF count as characters but are not part of the input
+        const std::size_t consumed = (std::min)(result.chars_read_total, ia.get_consumed_count());
+        std::size_t line_start = 0;
+        result.lines_read = ia.count_consumed_lines(consumed, line_start);
+        result.chars_read_current_line = result.chars_read_total - line_start;
+        return result;
+    }
+
+    /// streaming adapter: the position is maintained by get() and unget()
+    position_t get_position_impl(std::false_type /*lazy*/) const noexcept
+    {
+        return position;
+    }
+
+  public:
     /// seekable adapter: rebuild the last read token from the input on demand
     const std::vector<char_type>& collect_token_chars(std::vector<char_type>& out, std::true_type /*lazy*/) const
     {
@@ -12402,15 +12521,57 @@ scan_number_done:
         // into one that reloads the pointer from memory every character
         do
         {
+            if (current == '\n')
+            {
+                skip_indentation(std::integral_constant<bool, bulk_scan> {});
+            }
             get_ignoring_pending_unget();
         }
         while (current_is_whitespace());
     }
 
+    /*!
+    @brief contiguous input: skip the space indentation that follows a newline
+           8 bytes at a time
+
+    Pretty-printed JSON indents every line with a run of spaces, which the loop
+    in skip_whitespace() would otherwise read one character at a time. The
+    spaces are skipped 8 bytes at a time; the last 7 bytes of the input and any
+    other whitespace are left to that loop. Skipping bypasses get(), which is
+    fine because bulk_scan implies lazy_position: the position is derived from
+    the adapter, not counted per character.
+    */
+    void skip_indentation(std::true_type /*bulk*/)
+    {
+        JSON_ASSERT(!next_unget);
+        const std::size_t remaining = ia.bulk_remaining();
+        if (remaining < 8)
+        {
+            return;
+        }
+        const char* const data = reinterpret_cast<const char*>(ia.bulk_data());
+        std::size_t pos = 0;
+        for (; pos + 8 <= remaining; pos += 8)
+        {
+            // little-endian, so the first byte is the lowest
+            const std::uint64_t non_space = read_eight_bytes(data + pos) ^ 0x2020202020202020ull;
+            if (non_space != 0)
+            {
+                // skip the spaces before the first other byte of this word
+                pos += static_cast<std::size_t>(count_trailing_zeros(non_space)) / 8;
+                break;
+            }
+        }
+        ia.bulk_skip(pos);
+    }
+
+    /// streaming input: no bulk fast path
+    void skip_indentation(std::false_type /*bulk*/) const noexcept {}
+
     token_type scan()
     {
         // initially, skip the BOM
-        if (position.chars_read_total == 0 && !skip_bom())
+        if (get_chars_read_total() == 0 && !skip_bom())
         {
             error_message = "invalid BOM; must be 0xEF 0xBB 0xBF if given";
             return token_type::parse_error;
@@ -12437,7 +12598,7 @@ scan_number_done:
     token_type scan_expecting(token_type expected_type)
     {
         JSON_ASSERT(expected_type == token_type::name_separator || expected_type == token_type::value_separator);
-        JSON_ASSERT(position.chars_read_total > 0);
+        JSON_ASSERT(get_chars_read_total() > 0);
         const char_int_type expected_char = static_cast<unsigned char>((expected_type == token_type::name_separator) ? ':' : ',');
         skip_whitespace();
         if (JSON_HEDLEY_LIKELY(current == expected_char))
@@ -12791,7 +12952,7 @@ struct diagnostic_positions
         {
             // Lexer has read the first character of the container, so
             // subtract 1 from the position to get the correct start position.
-            v.start_position = lexer->get_position() - 1;
+            v.start_position = lexer->get_chars_read_total() - 1;
         }
     }
 
@@ -12805,7 +12966,7 @@ struct diagnostic_positions
         if (lexer)
         {
             // Lexer's position is past the closing brace or bracket, so set that as the end position.
-            v.end_position = lexer->get_position();
+            v.end_position = lexer->get_chars_read_total();
         }
     }
 
@@ -12821,7 +12982,7 @@ struct diagnostic_positions
             // Lexer has read past the current field value, so set the end position to the current position.
             // The start position will be set below based on the length of the string representation
             // of the value.
-            v.end_position = lexer->get_position();
+            v.end_position = lexer->get_chars_read_total();
 
             switch (v.type())
             {
@@ -18664,7 +18825,7 @@ class parser
                     }
                     case token_type::end_of_input:
                     {
-                        if (JSON_HEDLEY_UNLIKELY(m_lexer.get_position().chars_read_total == 1))
+                        if (JSON_HEDLEY_UNLIKELY(m_lexer.get_chars_read_total() == 1))
                         {
                             return syntax_error(*sax, "attempting to parse an empty input; check that your input string or stream contains the expected JSON");
                         }
@@ -18814,8 +18975,10 @@ class parser
     {
         // MSVC 2015 reports C4100 (unreferenced parameter) when SAX::parse_error is static
         static_cast<void>(sax);
-        return sax.parse_error(m_lexer.get_position(), m_lexer.get_token_string(),
-                               parse_error::create(exception_id::syntax_error, m_lexer.get_position(), message, nullptr));
+        // computing the position is O(input read so far) for seekable input
+        const position_t position = m_lexer.get_position();
+        return sax.parse_error(position, m_lexer.get_token_string(),
+                               parse_error::create(exception_id::syntax_error, position, message, nullptr));
     }
 
     std::string exception_message(const token_type expected, const std::string& context)
@@ -35570,6 +35733,176 @@ public:
     }
 
   public:
+    /// @brief creates a diff as a JSON Merge Patch
+    /// @sa https://json.nlohmann.me/api/basic_json/merge_diff/
+    JSON_HEDLEY_WARN_UNUSED_RESULT
+    static basic_json merge_diff(const basic_json& source, const basic_json& target)
+    {
+        // unless both are objects, the patch is the target itself
+        if (!source.is_object() || !target.is_object())
+        {
+            return target;
+        }
+
+        return merge_diff_recursively(source, target, 0);
+    }
+
+  private:
+    /// @brief two objects @ref merge_diff_iteratively is diffing
+    struct merge_diff_frame
+    {
+        merge_diff_frame(const basic_json* source_, const basic_json* target_)
+            : source(source_), target(target_), member(source_->cbegin()), patch(value_t::object)
+        {}
+
+        // declared for GCC's -Weffc++, as in diff_frame
+        merge_diff_frame(const merge_diff_frame&) = default;
+        merge_diff_frame(merge_diff_frame&&) = default;
+        merge_diff_frame& operator=(const merge_diff_frame&) = default;
+        merge_diff_frame& operator=(merge_diff_frame&&) = default;
+        ~merge_diff_frame() = default;
+
+        /// the objects being diffed
+        const basic_json* source;
+        const basic_json* target;
+        /// the member of source being diffed; advanced once it is done
+        const_iterator member;
+        /// the patch for the members of source diffed so far
+        basic_json patch;
+    };
+
+    /// @brief add the members of object @a target that @a source does not
+    ///        have to the merge patch @a result
+    static void merge_diff_added_members(basic_json& result, const basic_json& source, const basic_json& target)
+    {
+        for (auto it = target.begin(); it != target.end(); ++it)
+        {
+            auto itf = source.find(it.key());
+            if (itf == source.end())
+            {
+                result[it.key()] = it.value();
+            }
+        }
+    }
+
+    /*!
+    @brief @ref merge_diff of two objects at nesting level @a depth
+
+    Diffing two nested objects calls this function again, once per nesting
+    level, so values nested deeply enough would exhaust the call stack and
+    terminate the process. The descent is bounded here: once @ref
+    detail::recursion_depth_limit levels have been entered, @ref
+    merge_diff_iteratively diffs what is left without the call stack.
+    */
+    static basic_json merge_diff_recursively(const basic_json& source, const basic_json& target, const std::size_t depth)
+    {
+        JSON_ASSERT(source.is_object() && target.is_object());
+
+        if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
+        {
+            return merge_diff_iteratively(source, target);
+        }
+
+        basic_json result(value_t::object);
+
+        for (auto it = source.begin(); it != source.end(); ++it)
+        {
+            auto itf = target.find(it.key());
+            if (itf != target.end())
+            {
+                if (it.value().is_object() && itf.value().is_object())
+                {
+                    auto diff = merge_diff_recursively(it.value(), itf.value(), depth + 1);
+                    if (!diff.empty())
+                    {
+                        result[it.key()] = std::move(diff);
+                    }
+                }
+                else if (it.value() != itf.value())
+                {
+                    result[it.key()] = itf.value();
+                }
+            }
+            else
+            {
+                result[it.key()] = value_t::null;
+            }
+        }
+
+        merge_diff_added_members(result, source, target);
+        return result;
+    }
+
+    /*!
+    @brief @ref merge_diff of two objects without the call stack
+
+    Produces the same patch as @ref merge_diff_recursively, keeping the objects
+    being diffed on an explicit stack instead of descending into them. Two
+    nested objects are still diffed completely before the next member, so the
+    patch has its members in the same order as in the recursive version. Only
+    reached for objects nested more deeply than @ref
+    detail::recursion_depth_limit.
+    */
+    static basic_json merge_diff_iteratively(const basic_json& source, const basic_json& target)
+    {
+        // the patch of source and target, set once their frame is popped
+        basic_json result;
+        std::vector<merge_diff_frame> stack;
+        stack.emplace_back(&source, &target);
+        while (!stack.empty())
+        {
+            // the frame's members are copied out and it is changed through
+            // stack.back(): pushing a frame for nested objects or popping
+            // this one would invalidate a reference to it
+            const basic_json* const s = stack.back().source;
+            const basic_json* const t = stack.back().target;
+            const const_iterator it = stack.back().member;
+
+            if (it != s->cend())
+            {
+                const const_iterator itf = t->find(it.key());
+                if (itf != t->cend())
+                {
+                    if (it.value().is_object() && itf.value().is_object())
+                    {
+                        // member is advanced once their patch is under its key
+                        stack.emplace_back(&it.value(), &itf.value());
+                        continue;
+                    }
+                    if (it.value() != itf.value())
+                    {
+                        stack.back().patch[it.key()] = itf.value();
+                    }
+                }
+                else
+                {
+                    stack.back().patch[it.key()] = value_t::null;
+                }
+                ++stack.back().member;
+                continue;
+            }
+
+            merge_diff_added_members(stack.back().patch, *s, *t);
+
+            // these objects are done: continue with the ones they are in
+            basic_json patch = std::move(stack.back().patch);
+            stack.pop_back();
+            if (stack.empty())
+            {
+                result = std::move(patch);
+            }
+            else
+            {
+                if (!patch.empty())
+                {
+                    stack.back().patch[stack.back().member.key()] = std::move(patch);
+                }
+                ++stack.back().member;
+            }
+        }
+        return result;
+    }
+
     /// @}
 };
 

tests/src/unit-class_lexer.cpp

@@ -938,6 +938,10 @@ TEST_CASE("Eisel-Lemire float conversion")
             const int k = i % 64;
             const std::uint64_t x = (std::uint64_t{1} << k) | (a & ((std::uint64_t{1} << k) - 1));
             CHECK(nlohmann::detail::count_leading_zeros(x) == 63 - k);
+
+            // the mirror image: bit k set, arbitrary bits above it
+            const std::uint64_t y = (std::uint64_t{1} << k) | (k == 63 ? 0 : (a << (k + 1)));
+            CHECK(nlohmann::detail::count_trailing_zeros(y) == k);
         }
     }
 

tests/src/unit-custom-binary-type.cpp

@@ -8,6 +8,13 @@
 
 #include "doctest_compatibility.h"
 
+// capture whether JSON_DISABLE_ENUM_SERIALIZATION was enabled on the command
+// line before including json.hpp, which #undefs it; std::byte is an enum, so
+// converting std::vector<std::byte> needs enum serialization
+#if defined(JSON_DISABLE_ENUM_SERIALIZATION) && (JSON_DISABLE_ENUM_SERIALIZATION == 1)
+    #define SKIP_TESTS_FOR_ENUM_SERIALIZATION
+#endif
+
 #include <nlohmann/json.hpp>
 
 #include <cstdint>
@@ -85,3 +92,53 @@ TEST_CASE("binary type whose value type is not std::uint8_t")
     }
 #endif
 }
+
+#ifdef JSON_HAS_CPP_17
+// moved from unit-regression3.cpp, so that file does not instantiate a second
+// basic_json specialization (MinGW section limit, see windows.yml)
+TEST_CASE("std::vector<std::byte> as binary type")
+{
+    SECTION("issue #4804: from_cbor incompatible with std::vector<std::byte> as binary_t")
+    {
+        const std::vector<std::uint8_t> data = {0x80};
+        const auto decoded = byte_binary_json::from_cbor(data);
+        CHECK((decoded == byte_binary_json::array()));
+    }
+
+#ifndef SKIP_TESTS_FOR_ENUM_SERIALIZATION
+    SECTION("discussion #4209 - custom BinaryType direct assignment and round-tripping")
+    {
+        // Test that assigning a custom BinaryType directly creates a binary value, not an array
+        const std::vector<std::byte> original{std::byte{1}, std::byte{2}, std::byte{3}};
+        const byte_binary_json j = original;
+        CHECK(j.is_binary());
+        CHECK(!j.is_array());
+
+        // Test round-tripping: extracting the binary value back as the custom container type
+        const auto extracted = j.get<std::vector<std::byte>>();
+        CHECK(extracted == original);
+
+        // Test that the default json alias behavior is unchanged: std::vector<uint8_t> -> array
+        const nlohmann::json default_json = std::vector<std::uint8_t> {1, 2, 3};
+        CHECK(default_json.is_array());
+        CHECK(!default_json.is_binary());
+    }
+
+    SECTION("discussion #4209 - custom BinaryType extraction from parsed array")
+    {
+        // Test that extracting a custom BinaryType from a parsed JSON array still works
+        // (not just from a binary-typed node)
+        const auto j = byte_binary_json::parse("[1,2,3]");
+        CHECK(j.is_array());
+        CHECK(!j.is_binary());
+
+        // Extracting as custom BinaryType should work from arrays
+        const auto extracted = j.get<std::vector<std::byte>>();
+        CHECK(extracted.size() == 3);
+        CHECK(extracted[0] == std::byte{1});
+        CHECK(extracted[1] == std::byte{2});
+        CHECK(extracted[2] == std::byte{3});
+    }
+#endif
+}
+#endif

tests/src/unit-deserialization.cpp

@@ -35,6 +35,7 @@ using nlohmann::json;
 #include <sstream>
 #include <string>
 #include <valarray>
+#include <vector>
 
 #include "sax_event_loggers.hpp"
 
@@ -1250,3 +1251,139 @@ TEST_CASE_TEMPLATE("deserialization of different character types (UTF-32)", T, c
     CHECK(json::sax_parse(v, &l));
     CHECK(l.events.size() == 1);
 }
+
+namespace
+{
+// records the position and message of a parse error; reporting the error via
+// SAX does not throw, so the test also runs with JSON_NOEXCEPTION
+struct SaxErrorRecorder : public SaxEventLogger
+{
+    bool parse_error(std::size_t position, const std::string& /*last_token*/, const json::exception& ex) override
+    {
+        error = std::to_string(position) + " " + ex.what();
+        return false;
+    }
+
+    std::string error {}; // NOLINT(readability-redundant-member-init)
+};
+} // namespace
+
+TEST_CASE("parse error positions are the same for contiguous and stream input")
+{
+    // contiguous input derives the position from the input on error, stream
+    // input counts it while reading (see lexer::lazy_position); the cases
+    // cover newlines before the error, ungetting a newline after a number,
+    // newlines inside a string, a comment, and reads at the end of input
+    struct test_case
+    {
+        std::string input;
+        std::size_t byte;
+        std::string message;
+    };
+    const std::vector<test_case> cases =
+    {
+        {"[\n  1,\n  2\n", 12, "[json.exception.parse_error.101] parse error at line 4, column 1: syntax error while parsing array - unexpected end of input; expected ']'"},
+        {"[1\n,2,\n]", 8, "[json.exception.parse_error.101] parse error at line 3, column 1: syntax error while parsing value - unexpected ']'; expected '[', '{', or a literal"},
+        {"[1\n x]", 5, "[json.exception.parse_error.101] parse error at line 2, column 2: syntax error while parsing array - invalid literal; last read: '1<U+000A> x'; expected ']'"},
+        {"1\n2", 3, "[json.exception.parse_error.101] parse error at line 2, column 1: syntax error while parsing value - unexpected number literal; expected end of input"},
+        {
+            "{\"a\":\n\"b\nc\"}", 9, R"([json.exception.parse_error.101] parse error at line 3, column 0: syntax error while parsing value - invalid string: control character U+000A (LF) must be escaped to \u000A or \n; last read: '"b<U+000A>')"
+        },
+        {"/* a\n b */\n x", 13, "[json.exception.parse_error.101] parse error at line 3, column 2: syntax error while parsing value - invalid literal; last read: '/* a<U+000A> b */<U+000A> x'"},
+        {"\n\n", 3, "[json.exception.parse_error.101] parse error at line 3, column 1: syntax error while parsing value - unexpected end of input; expected '[', '{', or a literal"},
+        {"[\"a\",\n 1e\n]", 10, "[json.exception.parse_error.101] parse error at line 3, column 0: syntax error while parsing value - invalid number; expected '+', '-', or digit after exponent; last read: '1e<U+000A>'"},
+        {"[tru\n]", 5, "[json.exception.parse_error.101] parse error at line 2, column 0: syntax error while parsing value - invalid literal; last read: '[tru<U+000A>'"},
+    };
+
+    for (const auto& c : cases)
+    {
+        CAPTURE(c.input)
+        const std::string expected = std::to_string(c.byte) + " " + c.message;
+
+        SaxErrorRecorder string_sax;
+        CHECK_FALSE(json::sax_parse(c.input, &string_sax, json::input_format_t::json, true, true));
+        CHECK(string_sax.error == expected);
+
+        SaxErrorRecorder pointer_sax;
+        CHECK_FALSE(json::sax_parse(c.input.data(), c.input.data() + c.input.size(), &pointer_sax, json::input_format_t::json, true, true));
+        CHECK(pointer_sax.error == expected);
+
+        std::istringstream is(c.input);
+        SaxErrorRecorder stream_sax;
+        CHECK_FALSE(json::sax_parse(is, &stream_sax, json::input_format_t::json, true, true));
+        CHECK(stream_sax.error == expected);
+    }
+}
+
+#if !defined(JSON_NOEXCEPTION)
+TEST_CASE("indentation runs of every length around the 8-byte skip")
+{
+    // contiguous input skips the spaces after a newline 8 bytes at a time (see
+    // lexer::skip_indentation), stream input reads them one by one; results,
+    // byte offsets, and messages must be the same for runs that end anywhere
+    // relative to a word boundary, including at the end of the input
+    const auto parse_result = [](const std::string & input, bool contiguous)
+    {
+        try
+        {
+            if (contiguous)
+            {
+                return json::parse(input.data(), input.data() + input.size(), nullptr, true, true).dump();
+            }
+            std::istringstream is(input);
+            return json::parse(is, nullptr, true, true).dump();
+        }
+        catch (const json::parse_error& e)
+        {
+            return std::to_string(e.byte).append(" ").append(e.what());
+        }
+    };
+
+    // replace every '~' in a pattern by an indentation run
+    const auto indented = [](const std::string & pattern, const std::string & indent)
+    {
+        std::string result;
+        for (const char c : pattern)
+        {
+            if (c == '~')
+            {
+                result += indent;
+            }
+            else
+            {
+                result += c;
+            }
+        }
+        return result;
+    };
+
+    const std::vector<std::string> patterns =
+    {
+        "[\n~1,\n~2\n]",
+        "[\n~1,\n~x]",
+        "[\n~1,\n~",
+        "1\n~",
+        "[\n~\t1,\n~\r\n~2]",
+        "[\n~/* c */ 1]",
+        "[\n~1\n~2]",
+    };
+
+    for (std::size_t n = 0; n <= 33; ++n)
+    {
+        const std::string indent(n, ' ');
+        for (const auto& pattern : patterns)
+        {
+            const std::string input = indented(pattern, indent);
+            CAPTURE(n)
+            CAPTURE(input)
+            const std::string expected = parse_result(input, false);
+            CHECK(parse_result(input, true) == expected);
+            CHECK(json::accept(input.data(), input.data() + input.size(), true) == json::accept(input, true));
+        }
+    }
+
+    // and against fixed expectations, in case both paths broke the same way
+    CHECK(parse_result(indented("[\n~x]", std::string(16, ' ')), true) == "19 [json.exception.parse_error.101] parse error at line 2, column 17: syntax error while parsing value - invalid literal; last read: '[<U+000A>                x'");
+    CHECK(parse_result(indented("[\n~1]", std::string(17, ' ')), true) == "[1]");
+}
+#endif

tests/src/unit-diagnostics.cpp

@@ -567,9 +567,9 @@ TEST_CASE("Regression tests for extended diagnostics")
     }
 }
 
-TEST_CASE("Better diagnostics past the descent bound of update() and merge_patch()")
+TEST_CASE("Better diagnostics past the descent bound of update(), merge_patch() and merge_diff()")
 {
-    // Both merge objects nested more than detail::recursion_depth_limit()
+    // All three handle objects nested more than detail::recursion_depth_limit()
     // (128) levels deep without recursing; the values they add or replace
     // there must still know their parents.
     // The values are built rather than parsed, so that the expected messages
@@ -615,4 +615,19 @@ TEST_CASE("Better diagnostics past the descent bound of update() and merge_patch
         CHECK_THROWS_WITH_AS(p->at("x").at(0), expected_x.c_str(), json::type_error);
         CHECK_THROWS_WITH_AS(p->at("y").at(0), expected_y.c_str(), json::type_error);
     }
+
+    SECTION("merge_diff()")
+    {
+        // target and patch differ only in their innermost members, so the
+        // diff of the two is nested as deeply: {"a": ... {"x": null, "y": 2}}
+        const json diff = json::merge_diff(target, patch); // NOLINT(readability-suspicious-call-argument)
+
+        const json* p = &diff;
+        for (std::size_t i = 0; i < depth; ++i)
+        {
+            p = &p->at("a");
+        }
+        CHECK(p->at("x").is_null());
+        CHECK_THROWS_WITH_AS(p->at("y").at(0), expected_y.c_str(), json::type_error);
+    }
 }

tests/src/unit-regression3.cpp

@@ -58,11 +58,6 @@ using ordered_json = nlohmann::ordered_json;
         #include <optional>
     #elif __has_include(<experimental/optional>)
     #endif
-
-    /////////////////////////////////////////////////////////////////////
-    // for #4804
-    /////////////////////////////////////////////////////////////////////
-    using json_4804 = nlohmann::json::with_binary_t<std::vector<std::byte>>;
 #endif
 
 #ifdef JSON_HAS_CPP_20
@@ -458,49 +453,6 @@ TEST_CASE("regression tests 3")
 #endif
 
 #ifdef JSON_HAS_CPP_17
-    SECTION("issue #4804: from_cbor incompatible with std::vector<std::byte> as binary_t")
-    {
-        const std::vector<std::uint8_t> data = {0x80};
-        const auto decoded = json_4804::from_cbor(data);
-        CHECK((decoded == json_4804::array()));
-    }
-
-#ifndef SKIP_TESTS_FOR_ENUM_SERIALIZATION
-    SECTION("discussion #4209 - custom BinaryType direct assignment and round-tripping")
-    {
-        // Test that assigning a custom BinaryType directly creates a binary value, not an array
-        const std::vector<std::byte> original{std::byte{1}, std::byte{2}, std::byte{3}};
-        const json_4804 j = original;
-        CHECK(j.is_binary());
-        CHECK(!j.is_array());
-
-        // Test round-tripping: extracting the binary value back as the custom container type
-        const auto extracted = j.get<std::vector<std::byte>>();
-        CHECK(extracted == original);
-
-        // Test that the default json alias behavior is unchanged: std::vector<uint8_t> -> array
-        const json default_json = std::vector<std::uint8_t> {1, 2, 3};
-        CHECK(default_json.is_array());
-        CHECK(!default_json.is_binary());
-    }
-
-    SECTION("discussion #4209 - custom BinaryType extraction from parsed array")
-    {
-        // Test that extracting a custom BinaryType from a parsed JSON array still works
-        // (not just from a binary-typed node)
-        const auto j = json_4804::parse("[1,2,3]");
-        CHECK(j.is_array());
-        CHECK(!j.is_binary());
-
-        // Extracting as custom BinaryType should work from arrays
-        const auto extracted = j.get<std::vector<std::byte>>();
-        CHECK(extracted.size() == 3);
-        CHECK(extracted[0] == std::byte{1});
-        CHECK(extracted[1] == std::byte{2});
-        CHECK(extracted[2] == std::byte{3});
-    }
-#endif
-
     SECTION("issue #5046 - implicit conversion of return json to std::optional no longer implicit")
     {
         const json jval{};