Keep only the beginning and the end of a long token for error messages

For "last read" in parse errors, the lexer keeps the text of the token it
reads. The token start moves only at strings and numbers, so everything
after a string or number belongs to it until the next one starts. Input
that cannot be read again copies that text while parsing (the istream and
FILE* adapters in their carry buffer, the lexer in token_string for
iterators), so valid input of 1 followed by 100,000,000 newlines peaked
at about 225 MB, and a long string was kept a second time.

Replace the carry buffer and token_string by token_text_buffer, which
keeps the first 4096 and at least the last 4112 characters of a token and
drops the characters in between where it would otherwise grow, so
appending stays as cheap as before. get_token_string() shows a token of
more than 8192 characters as its first and last 4096 characters around
"...", the same for every kind of input; contiguous input copies only
these parts.

The example input now peaks at 1.8 MB. Instruction counts of parsing
from std::string, std::istream and FILE* are unchanged (clang 21 and
GCC 15); parsing from std::list<char> needs 1 to 17 % fewer.

Fixes #6120.

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

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

@@ -58,9 +58,7 @@ is not UTF-8 encoded and `error_handler` is set to `strict`
 
 ## Complexity
 
-Linear in the size of the result. With an `indent`, every line starts with up to `indent` characters per nesting level,
-so the result can be larger than the value by a factor of its nesting depth (see
-[Parsing untrusted input](../../features/parsing/untrusted_input.md#nesting-depth)).
+Linear.
 
 ## Notes
 

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

@@ -18,8 +18,7 @@ Strong exception safety: if an exception occurs, the original value stays intact
 
 ## Complexity
 
-Linear in the size of the result. Every key repeats the path of its value, so the keys can be larger than the value by a
-factor of its nesting depth (see [Parsing untrusted input](../../features/parsing/untrusted_input.md#nesting-depth)).
+Linear in the size of the JSON value.
 
 ## Notes
 

docs/mkdocs/docs/api/ordered_map.md

@@ -89,21 +89,19 @@ This differs from `#!cpp std::map`, where the same operations are O(log n).
 !!! warning "Quadratic cost of building large objects"
 
     Because every insertion scans all elements inserted so far, building an object of `n` distinct keys costs
-    **O(n²)** in total. This applies to filling an [`ordered_json`](ordered_json.md) object key by key (with
-    `operator[]`, `emplace`, or `insert`) and, for now, to reading one with a binary reader such as
-    [`from_cbor`](basic_json/from_cbor.md) ([#6104](https://github.com/nlohmann/json/issues/6104)). Parsing JSON text
-    is not affected: the parser finds the keys of an object with many members in a hash table and takes linear time.
+    **O(n²)** in total. This applies to filling an [`ordered_json`](ordered_json.md) object key by key as well as to
+    parsing one, since the parser inserts each key as it is read.
 
     The cost is negligible for the object sizes typically found in configuration files or API payloads, but it grows
-    steeply for machine-generated objects with many thousands of keys. Measured with Apple clang 21, `-O2 -DNDEBUG`,
-    arm64, for a flat object of `n` keys (best of seven runs):
+    steeply for machine-generated objects with many thousands of keys. Measured with `-O2 -DNDEBUG` for parsing a flat
+    object of `n` keys, relative to `#!cpp nlohmann::json` (which uses `#!cpp std::map`):
 
-    | `n`    | `json::parse` | `ordered_json::parse` | `ordered_json`, key by key | `ordered_json::from_cbor` |
-    |--------|---------------|-----------------------|----------------------------|---------------------------|
-    | 2000   | 0.3 ms        | 0.1 ms                | 4.5 ms                     | 4.4 ms                    |
-    | 4000   | 0.5 ms        | 0.2 ms                | 17.1 ms                    | 16.8 ms                   |
-    | 8000   | 0.9 ms        | 0.3 ms                | 67.0 ms                    | 66.8 ms                   |
-    | 16 000 | 2.0 ms        | 0.7 ms                | 282.4 ms                   | 282.4 ms                  |
+    | `n`    | `json` | `ordered_json` | factor |
+    |--------|--------|----------------|--------|
+    | 2000   | 0.7 ms | 3.6 ms         | 5×     |
+    | 4000   | 0.8 ms | 14.0 ms        | 19×    |
+    | 8000   | 1.6 ms | 67.8 ms        | 43×    |
+    | 16 000 | 3.3 ms | 181.6 ms       | 54×    |
 
     If key order matters for objects of that size, consider a container with a lookup index, such as
     [`nlohmann::fifo_map`](https://github.com/nlohmann/fifo_map)

docs/mkdocs/docs/features/parsing/untrusted_input.md

@@ -81,20 +81,6 @@ result. See the [documentation of the macro](../../api/macros/json_recursion_dep
 the speed, and the rules for choosing a value. Destroying a value (its destructor) recurses over the same number of
 levels and then continues without the call stack, and needs 32 KiB or less at every limit.
 
-!!! warning "Indented output and flattened keys grow with the depth"
-
-    Some results are larger than the value by a factor of its nesting depth.
-    [`dump`](../../api/basic_json/dump.md) with an `indent` of `n` (and likewise `operator<<` with `std::setw(n)` and
-    [`basic_json_view::dump`](../../api/basic_json_view/dump.md)) writes up to `n` characters per nesting level in
-    front of every line, and [`flatten`](../../api/basic_json/flatten.md) repeats the path of every value in its key.
-    For the 80,001-byte array `[1,[1,...[1,1]...]]` with 20,000 levels, `dump(1)` returns 600 MB, `dump(4)` 2.4 GB,
-    and the keys of `flatten()` hold 400 MB. A compact `dump()` without indentation stays the size of the value. If you
-    pretty-print, log, or flatten untrusted input, limit its nesting depth while parsing, for example with a
-    [parser callback](parser_callbacks.md#recipe-max-nesting-depth-via-a-callback): with a limit of `d`, every line has
-    at most `n * d` indentation characters, and every flattened key has at most `d` reference tokens.
-    [`JSON_RECURSION_DEPTH_LIMIT`](../../api/macros/json_recursion_depth_limit.md) does not help here: it changes only
-    how much of the call stack a walk uses, never its result, and it does not limit the depth that `parse()` accepts.
-
 ## Input size
 
 The library does not limit the overall size of a JSON text; a value nested or wide enough will use memory
@@ -154,20 +140,6 @@ values for a repeated key ends up in the parsed object. If your application must
 silently resolving them one way or another, see the
 [parser callback recipe for rejecting duplicate keys](parser_callbacks.md#recipe-rejecting-duplicate-object-keys).
 
-## Objects with many keys in `ordered_json`
-
-[`ordered_json`](../../api/ordered_json.md) keeps the members of an object in a vector in insertion order
-([`ordered_map`](../../api/ordered_map.md)) and finds a key by searching all members. Parsing JSON text into an
-`ordered_json` avoids that search for objects with many keys and takes linear time. The binary readers
-([`from_cbor`](../../api/basic_json/from_cbor.md), [`from_msgpack`](../../api/basic_json/from_msgpack.md),
-[`from_ubjson`](../../api/basic_json/from_ubjson.md), [`from_bjdata`](../../api/basic_json/from_bjdata.md),
-[`from_bson`](../../api/basic_json/from_bson.md), and [`from_bon8`](../../api/basic_json/from_bon8.md)) do not yet:
-reading an object of `n` keys into an `ordered_json` costs time proportional to `n²`. A flat object of 40,000 keys
-(about 300 KB of CBOR) takes about 1.3 s, against a few milliseconds as JSON text or into a `json`
-([#6104](https://github.com/nlohmann/json/issues/6104)). `json` is not affected. If you read untrusted binary input
-into `ordered_json`, bound the input size more tightly, read it into a `json`, or reject objects with many keys in a
-[SAX handler](sax_interface.md).
-
 ## Numbers
 
 A number whose value cannot be represented — for instance `1E1000`, which overflows `double` — is rejected while
@@ -199,11 +171,9 @@ indices (or the pointer as a whole) before using them, or use [`at`](../../api/b
 
 - Wrap parsing in a `#!cpp try`/`#!cpp catch` block, or use `allow_exceptions=false`/`accept()` if your environment
   cannot use exceptions; never let `JSON_NOEXCEPTION`'s `abort()` be the first time you think about error handling.
-- If the input's nesting depth matters to you (for example because you pretty-print, log, or flatten it), enforce your
-  own limit with a [parser callback](parser_callbacks.md#recipe-max-nesting-depth-via-a-callback) or a
+- If the input's nesting depth matters to you, enforce your own limit with a
+  [parser callback](parser_callbacks.md#recipe-max-nesting-depth-via-a-callback) or a
   [SAX handler](sax_interface.md); the library bounds the call stack but not memory use.
-- If you read untrusted binary input into `ordered_json`, mind that objects with many keys are
-  [slow to read](#objects-with-many-keys-in-ordered_json).
 - Bound the size of the input itself before parsing, independent of the library's own bounded allocations for binary
   format lengths.
 - Decide up front how you want a string with invalid UTF-8 — from any source, not only parsing — to be serialized

docs/mkdocs/docs/home/exceptions.md

@@ -140,6 +140,8 @@ Exceptions have ids 1xx.
 
 This error indicates a syntax error while deserializing a JSON text. The error message describes that an unexpected token (character) was encountered, and the member `byte` indicates the error position.
 
+The message ends with the last read token (`last read: '...'`), with control characters written as `<U+XXXX>`. A token of more than 8192 characters is shown by its first and its last 4096 characters, separated by `...`.
+
 !!! failure "Example message"
 
     Input ended prematurely:

include/nlohmann/detail/input/input_adapters.hpp

@@ -110,87 +110,164 @@ inline const char* empty_input_message() noexcept
     return "attempting to parse an empty input; check that your input string or stream contains the expected JSON";
 }
 
-////////////////////
-// input adapters //
-////////////////////
-
-#ifndef JSON_NO_IO
 /*!
+@brief the characters of the last read token, kept for error messages
+
 A growable array of characters for the part of a token that an input adapter
-copied out of its window: unlike std::vector<char>::push_back(), appending a
-character takes no reference to a local, and is a few instructions when there
-is room.
+copied out of its window, or that the lexer copies while it scans input it
+cannot read again: unlike std::vector::push_back(), appending a character
+takes no reference to a local, and is a few instructions when there is room.
+
+The token is only needed for error messages, which show its first head()
+characters and its last head() characters (see lexer::get_token_string). The
+buffer therefore keeps the first head() and at least the last tail()
+characters: when it is full and holds more than limit() characters, it drops
+the characters in between instead of growing. That happens where it would
+otherwise reallocate, so appending stays as cheap as before, and a long string
+or a long run of whitespace read from a stream is not kept a second time.
 */
-class token_carry_buffer
+template<typename CharType>
+class token_text_buffer
 {
   public:
-    token_carry_buffer() = default;
-    token_carry_buffer(const token_carry_buffer&) = delete;
-    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
-    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+    static constexpr std::size_t head() noexcept
+    {
+        return 4096;
+    }
+    /// a little more than head(), as the last character may be handed back
+    static constexpr std::size_t tail() noexcept
+    {
+        return head() + 16;
+    }
+    static constexpr std::size_t limit() noexcept
+    {
+        return head() + (2 * tail());
+    }
+
+    token_text_buffer() = default;
+    token_text_buffer(const token_text_buffer&) = delete;
+    token_text_buffer& operator=(const token_text_buffer&) = delete;
+    token_text_buffer(token_text_buffer&& rhs) noexcept
         : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
     {
         rhs.m_end = rhs.m_capacity_end = nullptr;
     }
-    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
-    ~token_carry_buffer() = default;
+    token_text_buffer& operator=(token_text_buffer&& rhs) noexcept
+    {
+        m_data = std::move(rhs.m_data);
+        m_end = rhs.m_end;
+        m_capacity_end = rhs.m_capacity_end;
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+        return *this;
+    }
+    ~token_text_buffer() = default;
 
     void clear() noexcept
     {
         m_end = m_data.get();
     }
 
-    void push_back(char c)
+    bool empty() const noexcept
+    {
+        return m_end == m_data.get();
+    }
+
+    void push_back(CharType c)
     {
         if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
         {
-            grow(1);
+            make_room(1);
         }
         *m_end++ = c;
     }
 
-    void append(const char* first, const char* last)
+    /// drop the last character
+    void pop_back() noexcept
+    {
+        JSON_ASSERT(!empty());
+        --m_end;
+    }
+
+    void append(const CharType* first, const CharType* last)
     {
         const auto n = static_cast<std::size_t>(last - first);
         const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
-        if (n > room)
+        if (JSON_HEDLEY_UNLIKELY(n > room))
         {
-            grow(n);
+            append_long(first, n);
+            return;
         }
         if (n != 0)
         {
-            std::memcpy(m_end, first, n);
+            std::memcpy(m_end, first, n * sizeof(CharType));
         }
         m_end += n;
     }
 
     /// replace the contents of @a out
-    void copy_to(std::vector<char>& out) const
+    void copy_to(std::vector<CharType>& out) const
     {
         out.assign(m_data.get(), m_end);
     }
 
   private:
-    /// make room for @a extra more characters
-    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    /// append @a n characters that do not fit, in pieces of at most tail()
+    /// characters, so that dropping the middle keeps the last tail() of them
+    JSON_HEDLEY_NEVER_INLINE void append_long(const CharType* first, std::size_t n)
     {
-        const auto size = static_cast<std::size_t>(m_end - m_data.get());
-        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
-        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
-        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
-        if (size != 0)
+        while (n != 0)
         {
-            std::memcpy(data.get(), m_data.get(), size);
+            const std::size_t piece = n < tail() ? n : tail();
+            make_room(piece);
+            std::memcpy(m_end, first, piece * sizeof(CharType));
+            m_end += piece;
+            first += piece;
+            n -= piece;
         }
-        m_data = std::move(data);
-        m_end = m_data.get() + size;
-        m_capacity_end = m_data.get() + new_capacity;
     }
 
-    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
-    char* m_end = nullptr;
-    char* m_capacity_end = nullptr;
+    /// make room for @a extra (at most tail()) more characters: drop the
+    /// middle of a long token instead of growing beyond limit()
+    JSON_HEDLEY_NEVER_INLINE void make_room(std::size_t extra)
+    {
+        // nothing is allocated before the first character
+        std::size_t size = m_data ? static_cast<std::size_t>(m_end - m_data.get()) : 0;
+        if (size + extra > limit() && size > head() + tail())
+        {
+            std::memmove(m_data.get() + head(), m_end - tail(), tail() * sizeof(CharType));
+            m_end = m_data.get() + head() + tail();
+            size = head() + tail();
+        }
+        const std::size_t capacity = m_data ? static_cast<std::size_t>(m_capacity_end - m_data.get()) : 0;
+        if (size + extra > capacity)
+        {
+            // never more than limit(): from there on, the middle is dropped
+            const std::size_t doubled = capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity;
+            const std::size_t new_capacity = (std::max)(doubled < limit() ? doubled : limit(), size + extra);
+            std::unique_ptr<CharType[]> data(new CharType[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+            if (size != 0)
+            {
+                std::memcpy(data.get(), m_data.get(), size * sizeof(CharType));
+            }
+            m_data = std::move(data);
+            m_end = m_data.get() + size;
+            m_capacity_end = m_data.get() + new_capacity;
+        }
+    }
+
+    std::unique_ptr<CharType[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    CharType* m_end = nullptr;
+    CharType* m_capacity_end = nullptr;
 };
+
+/// the part of a token that an input adapter copied out of its window
+using token_carry_buffer = token_text_buffer<char>;
+
+////////////////////
+// input adapters //
+////////////////////
+
+#ifndef JSON_NO_IO
 /*!
 Input adapter for stdio file access. The file is locked for the lifetime of the
 adapter and read with the unlocked stdio functions (see

include/nlohmann/detail/input/lexer.hpp

@@ -494,7 +494,10 @@ class lexer : public lexer_base<BasicJsonType>
     /// streaming adapter: capture the characters
     void bulk_track_token(const char* data, std::size_t n, std::false_type /*lazy*/)
     {
-        token_string.insert(token_string.end(), data, data + n);
+        for (std::size_t i = 0; i < n; ++i)
+        {
+            token_string.push_back(static_cast<char_type>(data[i]));
+        }
     }
 
     /// streaming input: no bulk fast path
@@ -2237,15 +2240,28 @@ scan_number_done:
         const std::size_t stop = ia.get_consumed_count() - (pending_real_unget ? 1u : 0u);
         if (JSON_HEDLEY_LIKELY(stop >= token_string_start))
         {
-            ia.copy_consumed_range(token_string_start, stop, out);
+            // only the beginning and the end of a long token are shown (see
+            // get_token_string), so copy only these
+            constexpr std::size_t head = token_text_buffer<char_type>::head();
+            constexpr std::size_t tail = token_text_buffer<char_type>::tail();
+            if (stop - token_string_start > head + tail)
+            {
+                ia.copy_consumed_range(token_string_start, token_string_start + head, out);
+                ia.copy_consumed_range(stop - tail, stop, out);
+            }
+            else
+            {
+                ia.copy_consumed_range(token_string_start, stop, out);
+            }
         }
         return out;
     }
 
     /// streaming adapter: the token was copied eagerly while scanning
-    const std::vector<char_type>& collect_token_chars(std::vector<char_type>& /*out*/, std::false_type /*lazy*/) const
+    const std::vector<char_type>& collect_token_chars(std::vector<char_type>& out, std::false_type /*lazy*/) const
     {
-        return token_string;
+        token_string.copy_to(out);
+        return out;
     }
 
     /// return the last read token (for errors only).  Will never contain EOF
@@ -2256,10 +2272,32 @@ scan_number_done:
         std::vector<char_type> reconstructed;
         const std::vector<char_type>& chars = collect_token_chars(reconstructed, std::integral_constant<bool, lazy_token_string> {});
 
-        // escape control characters
+        // a long token is shown by its first and its last head() characters;
+        // the buffers that keep it while it is read keep no more than these
+        // and a few more (see token_text_buffer)
+        constexpr std::size_t shown = token_text_buffer<char_type>::head();
         std::string result;
-        for (const auto c : chars)
+        if (chars.size() > 2 * shown)
         {
+            append_escaped(result, chars.begin(), chars.begin() + static_cast<std::ptrdiff_t>(shown));
+            result += "...";
+            append_escaped(result, chars.end() - static_cast<std::ptrdiff_t>(shown), chars.end());
+        }
+        else
+        {
+            append_escaped(result, chars.begin(), chars.end());
+        }
+        return result;
+    }
+
+  private:
+    /// append the characters [first, last) to @a result, escaping control characters
+    template<typename Iterator>
+    static void append_escaped(std::string& result, Iterator first, Iterator last)
+    {
+        for (; first != last; ++first)
+        {
+            const auto c = *first;
             if (static_cast<unsigned char>(c) <= '\x1F')
             {
                 // escape control characters
@@ -2273,10 +2311,10 @@ scan_number_done:
                 result.push_back(static_cast<std::string::value_type>(c));
             }
         }
-
-        return result;
     }
 
+  public:
+
     /// return syntax error message
     JSON_HEDLEY_RETURNS_NON_NULL
     constexpr const char* get_error_message() const noexcept
@@ -2525,7 +2563,7 @@ scan_number_done:
 
     /// raw input token string for error messages; only populated for streaming
     /// adapters (seekable adapters reconstruct it lazily via token_string_start)
-    std::vector<char_type> token_string {};
+    token_text_buffer<char_type> token_string {};
 
     /// start offset of the current token within the input, used to reconstruct
     /// the last read token on error for seekable adapters (see collect_token_chars)

single_include/nlohmann/json.hpp

@@ -9323,87 +9323,164 @@ inline const char* empty_input_message() noexcept
     return "attempting to parse an empty input; check that your input string or stream contains the expected JSON";
 }
 
-////////////////////
-// input adapters //
-////////////////////
-
-#ifndef JSON_NO_IO
 /*!
+@brief the characters of the last read token, kept for error messages
+
 A growable array of characters for the part of a token that an input adapter
-copied out of its window: unlike std::vector<char>::push_back(), appending a
-character takes no reference to a local, and is a few instructions when there
-is room.
+copied out of its window, or that the lexer copies while it scans input it
+cannot read again: unlike std::vector::push_back(), appending a character
+takes no reference to a local, and is a few instructions when there is room.
+
+The token is only needed for error messages, which show its first head()
+characters and its last head() characters (see lexer::get_token_string). The
+buffer therefore keeps the first head() and at least the last tail()
+characters: when it is full and holds more than limit() characters, it drops
+the characters in between instead of growing. That happens where it would
+otherwise reallocate, so appending stays as cheap as before, and a long string
+or a long run of whitespace read from a stream is not kept a second time.
 */
-class token_carry_buffer
+template<typename CharType>
+class token_text_buffer
 {
   public:
-    token_carry_buffer() = default;
-    token_carry_buffer(const token_carry_buffer&) = delete;
-    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
-    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+    static constexpr std::size_t head() noexcept
+    {
+        return 4096;
+    }
+    /// a little more than head(), as the last character may be handed back
+    static constexpr std::size_t tail() noexcept
+    {
+        return head() + 16;
+    }
+    static constexpr std::size_t limit() noexcept
+    {
+        return head() + (2 * tail());
+    }
+
+    token_text_buffer() = default;
+    token_text_buffer(const token_text_buffer&) = delete;
+    token_text_buffer& operator=(const token_text_buffer&) = delete;
+    token_text_buffer(token_text_buffer&& rhs) noexcept
         : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
     {
         rhs.m_end = rhs.m_capacity_end = nullptr;
     }
-    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
-    ~token_carry_buffer() = default;
+    token_text_buffer& operator=(token_text_buffer&& rhs) noexcept
+    {
+        m_data = std::move(rhs.m_data);
+        m_end = rhs.m_end;
+        m_capacity_end = rhs.m_capacity_end;
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+        return *this;
+    }
+    ~token_text_buffer() = default;
 
     void clear() noexcept
     {
         m_end = m_data.get();
     }
 
-    void push_back(char c)
+    bool empty() const noexcept
+    {
+        return m_end == m_data.get();
+    }
+
+    void push_back(CharType c)
     {
         if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
         {
-            grow(1);
+            make_room(1);
         }
         *m_end++ = c;
     }
 
-    void append(const char* first, const char* last)
+    /// drop the last character
+    void pop_back() noexcept
+    {
+        JSON_ASSERT(!empty());
+        --m_end;
+    }
+
+    void append(const CharType* first, const CharType* last)
     {
         const auto n = static_cast<std::size_t>(last - first);
         const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
-        if (n > room)
+        if (JSON_HEDLEY_UNLIKELY(n > room))
         {
-            grow(n);
+            append_long(first, n);
+            return;
         }
         if (n != 0)
         {
-            std::memcpy(m_end, first, n);
+            std::memcpy(m_end, first, n * sizeof(CharType));
         }
         m_end += n;
     }
 
     /// replace the contents of @a out
-    void copy_to(std::vector<char>& out) const
+    void copy_to(std::vector<CharType>& out) const
     {
         out.assign(m_data.get(), m_end);
     }
 
   private:
-    /// make room for @a extra more characters
-    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    /// append @a n characters that do not fit, in pieces of at most tail()
+    /// characters, so that dropping the middle keeps the last tail() of them
+    JSON_HEDLEY_NEVER_INLINE void append_long(const CharType* first, std::size_t n)
     {
-        const auto size = static_cast<std::size_t>(m_end - m_data.get());
-        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
-        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
-        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
-        if (size != 0)
+        while (n != 0)
         {
-            std::memcpy(data.get(), m_data.get(), size);
+            const std::size_t piece = n < tail() ? n : tail();
+            make_room(piece);
+            std::memcpy(m_end, first, piece * sizeof(CharType));
+            m_end += piece;
+            first += piece;
+            n -= piece;
         }
-        m_data = std::move(data);
-        m_end = m_data.get() + size;
-        m_capacity_end = m_data.get() + new_capacity;
     }
 
-    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
-    char* m_end = nullptr;
-    char* m_capacity_end = nullptr;
+    /// make room for @a extra (at most tail()) more characters: drop the
+    /// middle of a long token instead of growing beyond limit()
+    JSON_HEDLEY_NEVER_INLINE void make_room(std::size_t extra)
+    {
+        // nothing is allocated before the first character
+        std::size_t size = m_data ? static_cast<std::size_t>(m_end - m_data.get()) : 0;
+        if (size + extra > limit() && size > head() + tail())
+        {
+            std::memmove(m_data.get() + head(), m_end - tail(), tail() * sizeof(CharType));
+            m_end = m_data.get() + head() + tail();
+            size = head() + tail();
+        }
+        const std::size_t capacity = m_data ? static_cast<std::size_t>(m_capacity_end - m_data.get()) : 0;
+        if (size + extra > capacity)
+        {
+            // never more than limit(): from there on, the middle is dropped
+            const std::size_t doubled = capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity;
+            const std::size_t new_capacity = (std::max)(doubled < limit() ? doubled : limit(), size + extra);
+            std::unique_ptr<CharType[]> data(new CharType[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+            if (size != 0)
+            {
+                std::memcpy(data.get(), m_data.get(), size * sizeof(CharType));
+            }
+            m_data = std::move(data);
+            m_end = m_data.get() + size;
+            m_capacity_end = m_data.get() + new_capacity;
+        }
+    }
+
+    std::unique_ptr<CharType[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    CharType* m_end = nullptr;
+    CharType* m_capacity_end = nullptr;
 };
+
+/// the part of a token that an input adapter copied out of its window
+using token_carry_buffer = token_text_buffer<char>;
+
+////////////////////
+// input adapters //
+////////////////////
+
+#ifndef JSON_NO_IO
 /*!
 Input adapter for stdio file access. The file is locked for the lifetime of the
 adapter and read with the unlocked stdio functions (see
@@ -13898,7 +13975,10 @@ class lexer : public lexer_base<BasicJsonType>
     /// streaming adapter: capture the characters
     void bulk_track_token(const char* data, std::size_t n, std::false_type /*lazy*/)
     {
-        token_string.insert(token_string.end(), data, data + n);
+        for (std::size_t i = 0; i < n; ++i)
+        {
+            token_string.push_back(static_cast<char_type>(data[i]));
+        }
     }
 
     /// streaming input: no bulk fast path
@@ -15641,15 +15721,28 @@ scan_number_done:
         const std::size_t stop = ia.get_consumed_count() - (pending_real_unget ? 1u : 0u);
         if (JSON_HEDLEY_LIKELY(stop >= token_string_start))
         {
-            ia.copy_consumed_range(token_string_start, stop, out);
+            // only the beginning and the end of a long token are shown (see
+            // get_token_string), so copy only these
+            constexpr std::size_t head = token_text_buffer<char_type>::head();
+            constexpr std::size_t tail = token_text_buffer<char_type>::tail();
+            if (stop - token_string_start > head + tail)
+            {
+                ia.copy_consumed_range(token_string_start, token_string_start + head, out);
+                ia.copy_consumed_range(stop - tail, stop, out);
+            }
+            else
+            {
+                ia.copy_consumed_range(token_string_start, stop, out);
+            }
         }
         return out;
     }
 
     /// streaming adapter: the token was copied eagerly while scanning
-    const std::vector<char_type>& collect_token_chars(std::vector<char_type>& /*out*/, std::false_type /*lazy*/) const
+    const std::vector<char_type>& collect_token_chars(std::vector<char_type>& out, std::false_type /*lazy*/) const
     {
-        return token_string;
+        token_string.copy_to(out);
+        return out;
     }
 
     /// return the last read token (for errors only).  Will never contain EOF
@@ -15660,10 +15753,32 @@ scan_number_done:
         std::vector<char_type> reconstructed;
         const std::vector<char_type>& chars = collect_token_chars(reconstructed, std::integral_constant<bool, lazy_token_string> {});
 
-        // escape control characters
+        // a long token is shown by its first and its last head() characters;
+        // the buffers that keep it while it is read keep no more than these
+        // and a few more (see token_text_buffer)
+        constexpr std::size_t shown = token_text_buffer<char_type>::head();
         std::string result;
-        for (const auto c : chars)
+        if (chars.size() > 2 * shown)
         {
+            append_escaped(result, chars.begin(), chars.begin() + static_cast<std::ptrdiff_t>(shown));
+            result += "...";
+            append_escaped(result, chars.end() - static_cast<std::ptrdiff_t>(shown), chars.end());
+        }
+        else
+        {
+            append_escaped(result, chars.begin(), chars.end());
+        }
+        return result;
+    }
+
+  private:
+    /// append the characters [first, last) to @a result, escaping control characters
+    template<typename Iterator>
+    static void append_escaped(std::string& result, Iterator first, Iterator last)
+    {
+        for (; first != last; ++first)
+        {
+            const auto c = *first;
             if (static_cast<unsigned char>(c) <= '\x1F')
             {
                 // escape control characters
@@ -15677,10 +15792,10 @@ scan_number_done:
                 result.push_back(static_cast<std::string::value_type>(c));
             }
         }
-
-        return result;
     }
 
+  public:
+
     /// return syntax error message
     JSON_HEDLEY_RETURNS_NON_NULL
     constexpr const char* get_error_message() const noexcept
@@ -15929,7 +16044,7 @@ scan_number_done:
 
     /// raw input token string for error messages; only populated for streaming
     /// adapters (seekable adapters reconstruct it lazily via token_string_start)
-    std::vector<char_type> token_string {};
+    token_text_buffer<char_type> token_string {};
 
     /// start offset of the current token within the input, used to reconstruct
     /// the last read token on error for seekable adapters (see collect_token_chars)

tests/src/unit-class_lexer.cpp

@@ -12,6 +12,7 @@
 #include <nlohmann/json.hpp>
 using nlohmann::json;
 
+#include <algorithm> // min
 #include <array> // array
 #include <cctype> // toupper
 #include <cstdint> // uint32_t, uint64_t
@@ -2233,3 +2234,123 @@ TEST_CASE("serialization of short strings")
     CHECK(json("caf\xC3\xA9").dump() == "\"caf\xC3\xA9\"");
     CHECK(json("caf\xC3\xA9").dump(-1, ' ', true) == "\"caf\\u00e9\"");
 }
+
+TEST_CASE("token_text_buffer keeps the beginning and the end of a long token")
+{
+    using buffer_t = nlohmann::detail::token_text_buffer<char>;
+    constexpr std::size_t head = buffer_t::head();
+    constexpr std::size_t tail = buffer_t::tail();
+
+    // the characters 0, 1, ..., 250, 0, 1, ...
+    const auto at = [](std::size_t i)
+    {
+        return static_cast<char>(i % 251);
+    };
+    // whether out holds the first head and the last tail characters of a
+    // token of n characters (and maybe more characters of its middle)
+    const auto keeps_ends = [&](const std::vector<char>& out, std::size_t n)
+    {
+        if (out.size() < head + tail || out.size() > buffer_t::limit())
+        {
+            return false;
+        }
+        for (std::size_t i = 0; i < head; ++i)
+        {
+            if (out[i] != at(i))
+            {
+                return false;
+            }
+        }
+        for (std::size_t i = 0; i < tail; ++i)
+        {
+            if (out[out.size() - tail + i] != at(n - tail + i))
+            {
+                return false;
+            }
+        }
+        return true;
+    };
+
+    SECTION("a short token is kept as it is")
+    {
+        buffer_t buffer;
+        CHECK(buffer.empty());
+        for (std::size_t i = 0; i < head + tail; ++i)
+        {
+            buffer.push_back(at(i));
+        }
+        std::vector<char> out;
+        buffer.copy_to(out);
+        REQUIRE(out.size() == head + tail);
+        bool same = true;
+        for (std::size_t i = 0; i < out.size(); ++i)
+        {
+            same = same && out[i] == at(i);
+        }
+        CHECK(same);
+    }
+
+    SECTION("characters added one by one")
+    {
+        for (const std::size_t n :
+                {
+                    head + tail + 1, buffer_t::limit(), buffer_t::limit() + 1, static_cast<std::size_t>(100000)
+                })
+        {
+            CAPTURE(n)
+            buffer_t buffer;
+            for (std::size_t i = 0; i < n; ++i)
+            {
+                buffer.push_back(at(i));
+            }
+            std::vector<char> out;
+            buffer.copy_to(out);
+            CHECK(keeps_ends(out, n));
+
+            // the last character can be handed back
+            buffer.pop_back();
+            buffer.copy_to(out);
+            CHECK(keeps_ends(out, n - 1));
+        }
+    }
+
+    SECTION("characters added in pieces of different sizes")
+    {
+        std::vector<char> input(200000);
+        for (std::size_t i = 0; i < input.size(); ++i)
+        {
+            input[i] = at(i);
+        }
+        for (const std::size_t piece :
+                {
+                    static_cast<std::size_t>(1), static_cast<std::size_t>(100), tail - 1, tail, tail + 1, static_cast<std::size_t>(65536), input.size()
+                })
+        {
+            CAPTURE(piece)
+            buffer_t buffer;
+            for (std::size_t pos = 0; pos < input.size(); pos += piece)
+            {
+                const std::size_t n = (std::min)(piece, input.size() - pos);
+                buffer.append(input.data() + pos, input.data() + pos + n);
+            }
+            std::vector<char> out;
+            buffer.copy_to(out);
+            CHECK(keeps_ends(out, input.size()));
+        }
+    }
+
+    SECTION("clear starts a new token")
+    {
+        buffer_t buffer;
+        for (std::size_t i = 0; i < 100000; ++i)
+        {
+            buffer.push_back('x');
+        }
+        buffer.clear();
+        CHECK(buffer.empty());
+        buffer.push_back('a');
+        std::vector<char> out;
+        buffer.copy_to(out);
+        CHECK(out == std::vector<char> {'a'});
+    }
+}

tests/src/unit-class_parser.cpp

@@ -3269,3 +3269,116 @@ TEST_CASE("a plain function is accepted as parser callback for containers")
     CHECK((nlohmann::detail::can_compare_ne<std::string::const_iterator, std::string::const_iterator>::value));
     CHECK((nlohmann::detail::can_compare_ne<const char*, const char*>::value));
 }
+
+#if !defined(JSON_NOEXCEPTION)
+TEST_CASE("parse error message of a long token")
+{
+    // the message of a parse error shows the last read token; a long token is
+    // shown by its first and its last 4096 characters, and the message is the
+    // same for every kind of input
+    const auto message = [](const std::string & text, int kind) -> std::string
+    {
+        json _;
+        try
+        {
+            if (kind == 0)
+            {
+                _ = json::parse(text);
+            }
+            else if (kind == 1)
+            {
+                std::istringstream stream(text);
+                _ = json::parse(stream);
+            }
+            else if (kind == 2)
+            {
+                FILE* const file = std::tmpfile(); // NOLINT(cppcoreguidelines-owning-memory)
+                REQUIRE(file != nullptr);
+                CHECK(std::fwrite(text.data(), 1, text.size(), file) == text.size());
+                CHECK(std::fseek(file, 0, SEEK_SET) == 0);
+                try
+                {
+                    _ = json::parse(file);
+                }
+                catch (...)
+                {
+                    std::fclose(file); // NOLINT(cert-err33-c,cppcoreguidelines-owning-memory)
+                    throw;
+                }
+                std::fclose(file); // NOLINT(cert-err33-c,cppcoreguidelines-owning-memory)
+            }
+            else
+            {
+                const std::list<char> chars(text.begin(), text.end());
+                _ = json::parse(chars.begin(), chars.end());
+            }
+        }
+        catch (const json::parse_error& e)
+        {
+            return e.what();
+        }
+        return "no error";
+    };
+
+    SECTION("a short token is shown completely")
+    {
+        const std::string text = "[\"" + std::string(8000, 'a') + "\x01\"]";
+        const std::string expected = "last read: '\"" + std::string(8000, 'a') + "<U+0001>'";
+        for (int kind = 0; kind < 4; ++kind)
+        {
+            CAPTURE(kind)
+            const std::string m = message(text, kind);
+            CHECK(m.size() > expected.size());
+            CHECK(m.compare(m.size() - expected.size(), expected.size(), expected) == 0);
+        }
+    }
+
+    SECTION("a long string")
+    {
+        const std::string text = "[\"" + std::string(100000, 'a') + "b\x01\"]";
+        // the token is '"', 100,000 times 'a', 'b', and the control character
+        const std::string expected = "last read: '\"" + std::string(4095, 'a') + "..." + std::string(4094, 'a') + "b<U+0001>'";
+        for (int kind = 0; kind < 4; ++kind)
+        {
+            CAPTURE(kind)
+            const std::string m = message(text, kind);
+            CHECK(m.size() > expected.size());
+            CHECK(m.compare(m.size() - expected.size(), expected.size(), expected) == 0);
+        }
+    }
+
+    SECTION("a long run of whitespace after a number")
+    {
+        // the token begins with the number and ends with the invalid character
+        const std::string text = "[1," + std::string(100000, '\n') + "?]";
+        std::string expected = "last read: '1,";
+        for (int i = 0; i < 4094; ++i)
+        {
+            expected += "<U+000A>";
+        }
+        expected += "...";
+        for (int i = 0; i < 4095; ++i)
+        {
+            expected += "<U+000A>";
+        }
+        expected += "?'";
+        for (int kind = 0; kind < 4; ++kind)
+        {
+            CAPTURE(kind)
+            const std::string m = message(text, kind);
+            CHECK(m.size() > expected.size());
+            CHECK(m.compare(m.size() - expected.size(), expected.size(), expected) == 0);
+        }
+    }
+
+    SECTION("valid input with a long run of whitespace")
+    {
+        const std::string text = "1" + std::string(100000, ' ');
+        for (int kind = 0; kind < 4; ++kind)
+        {
+            CAPTURE(kind)
+            CHECK(message(text, kind) == "no error");
+        }
+    }
+}
+#endif