Binary readers and writers: faster UTF-8 checks of strings

With error_handler_t::strict (and replace/ignore), every string a binary
reader reads is checked for valid UTF-8, and every string a binary writer
writes. The readers used the byte-at-a-time decoder; the writers skipped
ASCII eight bytes at a time.

- New valid_utf8_string() decides strings of up to 16 ASCII bytes, most
  keys and many values, with one or two word tests, and checks longer
  strings 16 bytes at a time with the NEON/SSSE3 UTF-8 check the lexer
  uses (vector_utf8_valid_blocks: the same lookup, but quotes and control
  characters are ordinary characters). Other targets and JSON_NO_SIMD use
  valid_utf8_prefix() as before.
- The readers check strings with it (for one-byte character types) in an
  out-of-line function, so that reading without a check does not pay for
  its code.
- The writers' check_utf8() uses it, and the CBOR and MessagePack writers
  write a checked string of up to max_text_in_reservation characters in
  one reservation, as they write unchecked ones.

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

docs/mkdocs/docs/home/debugging.md

@@ -10,13 +10,6 @@ at its root, a [Natvis](https://learn.microsoft.com/en-us/visualstudio/debugger/
 file that gives `json`/`ordered_json` values a friendly, key/value debugger view instead of showing raw internal
 fields, when debugging with the MSVC debug engine (`cppvsdbg`) in Visual Studio or VS Code.
 
-The file also has visualizers for `json_document` and `json_view` (and their `ordered_json`/editable variants). They
-show the value of the document or view, and its elements as an expandable list. Because a document is only a flat index
-into the source text, the strings of the direct elements are shown, but the elements of nested arrays and objects are
-shown as index nodes (their kind, size, and numbers; strings only by length). Arrays and objects that have been edited
-in a `json_editable_document` are not expanded. The visualizers read the internal layout of the MSVC STL (`std::array`,
-`std::unique_ptr`).
-
 The checked-in visualizers cover namespace variants with zero, one, or two ABI tags. Configurations with three or more
 ABI tags are not covered and may show raw internal fields instead. To generate a Natvis file for such a configuration:
 
@@ -36,7 +29,7 @@ ABI tags are not covered and may show raw internal fields instead. To generate a
 
 4. Use the generated file instead of `nlohmann_json.natvis`.
 
-`--abi-tags` can be repeated for several configurations, and `--all` generates every combination (about 60 MB for 11
+`--abi-tags` can be repeated for several configurations, and `--all` generates every combination (about 60 MB for 12
 ABI tags, doubling with each further tag). See
 [`tools/generate_natvis`](https://github.com/nlohmann/json/tree/develop/tools/generate_natvis) for all options.
 

include/nlohmann/detail/input/binary_reader.hpp

@@ -4293,6 +4293,19 @@ class binary_reader
         return get_bytes(len, "string", result);
     }
 
+    /// whether @a s is valid UTF-8: ASCII eight bytes at a time, the rest
+    /// sequence by sequence (as the binary writers check strings)
+    static bool is_valid_utf8_string(const string_t& s, std::true_type /*is_byte*/) noexcept
+    {
+        return valid_utf8_string(reinterpret_cast<const unsigned char*>(s.data()), s.size());
+    }
+
+    /// whether @a s, of a character type wider than a byte, is valid UTF-8
+    static bool is_valid_utf8_string(const string_t& s, std::false_type /*is_byte*/) noexcept
+    {
+        return is_valid_utf8(s);
+    }
+
     /*!
     @brief validate a decoded text string (value or object key) against @ref error_handler
 
@@ -4312,7 +4325,16 @@ class binary_reader
     */
     bool check_string_utf8(string_t& result, const char* context)
     {
-        if (error_handler == error_handler_t::keep || is_valid_utf8(result))
+        return error_handler == error_handler_t::keep || check_string_utf8_handler(result, context);
+    }
+
+    /// @ref check_string_utf8 for an error handler other than @ref
+    /// error_handler_t::keep; out of line, so that reading strings without a
+    /// check does not pay for the code of the check
+    JSON_HEDLEY_NEVER_INLINE
+    bool check_string_utf8_handler(string_t& result, const char* context)
+    {
+        if (is_valid_utf8_string(result, std::integral_constant<bool, sizeof(typename string_t::value_type) == 1>()))
         {
             return true;
         }

include/nlohmann/detail/input/simd_scan.hpp

@@ -322,6 +322,78 @@ JSON_SCAN_SSSE3_TARGET JSON_HEDLEY_NEVER_INLINE inline bool vector_utf8_run(cons
     block_out = block;
     return false;
 }
+
+/*!
+@brief skip 16-byte blocks of well-formed UTF-8 at p (a character boundary)
+
+Like vector_utf8_run, but quotes, backslashes, and control characters are
+ordinary characters. Returns the start of the first block in which the check
+found an error (which may belong to a sequence begun in the block before) or
+where fewer than 16 bytes are left; a sequence may be cut at that point, so the
+caller resumes at its start, one sequence at a time.
+*/
+JSON_SCAN_SSSE3_TARGET JSON_HEDLEY_NEVER_INLINE inline const unsigned char* vector_utf8_valid_blocks(const unsigned char* block, const unsigned char* e) noexcept
+{
+    using lookup = utf8_lookup4<>;
+#if JSON_SCAN_NEON
+    const uint8x16_t t1h = vld1q_u8(lookup::byte_1_high.data());
+    const uint8x16_t t1l = vld1q_u8(lookup::byte_1_low.data());
+    const uint8x16_t t2h = vld1q_u8(lookup::byte_2_high.data());
+    uint8x16_t prev = vdupq_n_u8(0);
+    while (e - block >= 16)
+    {
+        const uint8x16_t in = vld1q_u8(block);
+        if (vmaxvq_u8(vorrq_u8(in, prev)) < 0x80)
+        {
+            // ASCII here, and no sequence from the block before to finish
+            prev = in;
+            block += 16;
+            continue;
+        }
+        const uint8x16_t prev1 = vextq_u8(prev, in, 15);
+        const uint8x16_t sc = vandq_u8(vandq_u8(vqtbl1q_u8(t1h, vshrq_n_u8(prev1, 4)), vqtbl1q_u8(t1l, vandq_u8(prev1, vdupq_n_u8(0x0F)))), vqtbl1q_u8(t2h, vshrq_n_u8(in, 4)));
+        const uint8x16_t must23 = vorrq_u8(vqsubq_u8(vextq_u8(prev, in, 14), vdupq_n_u8(0xE0 - 0x80)), vqsubq_u8(vextq_u8(prev, in, 13), vdupq_n_u8(0xF0 - 0x80)));
+        const uint8x16_t err = veorq_u8(vandq_u8(must23, vdupq_n_u8(0x80)), sc);
+        if (vmaxvq_u8(err) != 0)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#else
+    const __m128i t1h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t1l = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_low.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t2h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_2_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i nibble = _mm_set1_epi8(0x0F);
+    const __m128i zero = _mm_setzero_si128();
+    __m128i prev = zero;
+    while (e - block >= 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(block))); // NOLINT(bugprone-casting-through-void)
+        if (_mm_movemask_epi8(_mm_or_si128(in, prev)) == 0)
+        {
+            prev = in;
+            block += 16;
+            continue;
+        }
+        const __m128i prev1 = _mm_alignr_epi8(in, prev, 15);
+        const __m128i sc = _mm_and_si128(_mm_and_si128(_mm_shuffle_epi8(t1h, _mm_and_si128(_mm_srli_epi16(prev1, 4), nibble)),
+                                         _mm_shuffle_epi8(t1l, _mm_and_si128(prev1, nibble))),
+                                         _mm_shuffle_epi8(t2h, _mm_and_si128(_mm_srli_epi16(in, 4), nibble)));
+        const __m128i must23 = _mm_or_si128(_mm_subs_epu8(_mm_alignr_epi8(in, prev, 14), _mm_set1_epi8(0xE0 - 0x80)),
+                                            _mm_subs_epu8(_mm_alignr_epi8(in, prev, 13), _mm_set1_epi8(0xF0 - 0x80)));
+        const __m128i err = _mm_xor_si128(_mm_and_si128(must23, _mm_set1_epi8(static_cast<char>(-128))), sc);
+        if (_mm_movemask_epi8(_mm_cmpeq_epi8(err, zero)) != 0xFFFF)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#endif
+    return block;
+}
 #endif
 
 }  // namespace detail

include/nlohmann/detail/input/string_scan.hpp

@@ -236,6 +236,64 @@ inline std::size_t valid_utf8_prefix(const unsigned char* data, std::size_t n) n
     return pos;
 }
 
+/*!
+@brief whether [data, data+n) is valid UTF-8, for checking whole strings
+
+The same answer as valid_utf8_prefix(data, n) == n. Short ASCII strings, the
+most common keys and values, are decided in one test, and long strings are
+checked 16 bytes at a time where the vector check is available.
+*/
+inline bool valid_utf8_string(const unsigned char* data, std::size_t n) noexcept
+{
+    constexpr std::uint64_t high = 0x8080808080808080ull;
+    if (n <= 16)
+    {
+        // two overlapping words cover 8 to 16 bytes, one partial word fewer
+        std::uint64_t first = 0;
+        std::uint64_t last = 0;
+        if (n >= 8)
+        {
+            std::memcpy(&first, data, sizeof(first));
+            std::memcpy(&last, data + n - 8, sizeof(last));
+        }
+        else if (n != 0)
+        {
+            std::memcpy(&first, data, n);
+        }
+        if (((first | last) & high) == 0)
+        {
+            return true;
+        }
+    }
+#if JSON_SCAN_VECTOR_UTF8
+    else if (n >= 32
+#if JSON_SCAN_SSSE3_DISPATCH
+             && cpu_has_ssse3()
+#endif
+            )
+    {
+        const unsigned char* block = vector_utf8_valid_blocks(data, data + n);
+        // the rest from the start of a sequence that may be cut at block
+        for (int i = 1; i <= 3 && block - i >= data; ++i)
+        {
+            const unsigned char c = block[-i];
+            if (c < 0x80)
+            {
+                break;
+            }
+            if (c >= 0xC0)
+            {
+                block -= i;
+                break;
+            }
+        }
+        const auto checked = static_cast<std::size_t>(block - data);
+        return checked + valid_utf8_prefix(block, n - checked) == n;
+    }
+#endif
+    return valid_utf8_prefix(data, n) == n;
+}
+
 #if JSON_SCAN_VECTOR_UTF8
 /// the end of scan_string_vector from block, where the vector loop stopped
 /// (ill-formed UTF-8, or fewer than 16 bytes left): one byte or sequence at a

include/nlohmann/detail/output/binary_writer.hpp

@@ -3080,6 +3080,25 @@ class binary_writer
         }
     }
 
+    /// write the head and the characters of @a value, at most
+    /// max_text_in_reservation of them, in one reservation
+    JSON_HEDLEY_ALWAYS_INLINE void write_cbor_string_reserved(const string_t& value)
+    {
+        CharType* const out = oa.reserve(value.size() + 2);
+        CharType* p = out;
+        if (value.size() <= 0x17)
+        {
+            *p++ = to_char_type(static_cast<std::uint8_t>(0x60 + value.size()));
+        }
+        else
+        {
+            *p++ = to_char_type(0x78);
+            *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
+        }
+        copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
+        oa.commit(static_cast<std::size_t>(p - out) + value.size());
+    }
+
     /*!
     @brief write a CBOR text string
 
@@ -3101,19 +3120,7 @@ class binary_writer
             // the string is written as it is: head and characters in one reservation
             if (JSON_HEDLEY_LIKELY(value.size() <= max_text_in_reservation))
             {
-                CharType* const out = oa.reserve(value.size() + 2);
-                CharType* p = out;
-                if (value.size() <= 0x17)
-                {
-                    *p++ = to_char_type(static_cast<std::uint8_t>(0x60 + value.size()));
-                }
-                else
-                {
-                    *p++ = to_char_type(0x78);
-                    *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
-                }
-                copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
-                oa.commit(static_cast<std::size_t>(p - out) + value.size());
+                write_cbor_string_reserved(value);
                 return;
             }
         }
@@ -3127,6 +3134,13 @@ class binary_writer
     JSON_HEDLEY_NEVER_INLINE
     void write_cbor_checked_string(const string_t& value, const BasicJsonType& context)
     {
+        if (error_handler == error_handler_t::strict && value.size() <= max_text_in_reservation)
+        {
+            check_utf8(value, context);
+            write_cbor_string_reserved(value);
+            return;
+        }
+
         string_t storage;
         const string_t& sanitized = sanitize_utf8_for_write(value, context, storage);
 
@@ -3141,6 +3155,27 @@ class binary_writer
     // MsgPack //
     /////////////
 
+    /// write the head and the characters of @a value, at most
+    /// max_text_in_reservation of them, in one reservation
+    JSON_HEDLEY_ALWAYS_INLINE void write_msgpack_string_reserved(const string_t& value)
+    {
+        CharType* const out = oa.reserve(value.size() + 2);
+        CharType* p = out;
+        if (value.size() <= 31)
+        {
+            // fixstr
+            *p++ = to_char_type(static_cast<std::uint8_t>(0xA0 | value.size()));
+        }
+        else
+        {
+            // str 8
+            *p++ = to_char_type(0xD9);
+            *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
+        }
+        copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
+        oa.commit(static_cast<std::size_t>(p - out) + value.size());
+    }
+
     /*!
     @brief write a MessagePack str
 
@@ -3158,21 +3193,7 @@ class binary_writer
             // the string is written as it is: head and characters in one reservation
             if (JSON_HEDLEY_LIKELY(value.size() <= max_text_in_reservation))
             {
-                CharType* const out = oa.reserve(value.size() + 2);
-                CharType* p = out;
-                if (value.size() <= 31)
-                {
-                    // fixstr
-                    *p++ = to_char_type(static_cast<std::uint8_t>(0xA0 | value.size()));
-                }
-                else
-                {
-                    // str 8
-                    *p++ = to_char_type(0xD9);
-                    *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
-                }
-                copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
-                oa.commit(static_cast<std::size_t>(p - out) + value.size());
+                write_msgpack_string_reserved(value);
                 return;
             }
         }
@@ -3185,6 +3206,13 @@ class binary_writer
     JSON_HEDLEY_NEVER_INLINE
     void write_msgpack_checked_string(const string_t& value, const BasicJsonType& context)
     {
+        if (error_handler == error_handler_t::strict && value.size() <= max_text_in_reservation)
+        {
+            check_utf8(value, context);
+            write_msgpack_string_reserved(value);
+            return;
+        }
+
         string_t storage;
         const string_t& sanitized = sanitize_utf8_for_write(value, context, storage);
 
@@ -4794,10 +4822,9 @@ class binary_writer
     static void check_utf8(const string_t& s, const BasicJsonType& context)
     {
         const auto* data = reinterpret_cast<const unsigned char*>(s.data());
-        const std::size_t valid = valid_utf8_prefix(data, s.size());
-        if (JSON_HEDLEY_UNLIKELY(valid != s.size()))
+        if (JSON_HEDLEY_UNLIKELY(!valid_utf8_string(data, s.size())))
         {
-            throw_invalid_utf8(data, valid, context);
+            throw_invalid_utf8(data, valid_utf8_prefix(data, s.size()), context);
         }
     }
 

single_include/nlohmann/json.hpp

@@ -12377,6 +12377,78 @@ JSON_SCAN_SSSE3_TARGET JSON_HEDLEY_NEVER_INLINE inline bool vector_utf8_run(cons
     block_out = block;
     return false;
 }
+
+/*!
+@brief skip 16-byte blocks of well-formed UTF-8 at p (a character boundary)
+
+Like vector_utf8_run, but quotes, backslashes, and control characters are
+ordinary characters. Returns the start of the first block in which the check
+found an error (which may belong to a sequence begun in the block before) or
+where fewer than 16 bytes are left; a sequence may be cut at that point, so the
+caller resumes at its start, one sequence at a time.
+*/
+JSON_SCAN_SSSE3_TARGET JSON_HEDLEY_NEVER_INLINE inline const unsigned char* vector_utf8_valid_blocks(const unsigned char* block, const unsigned char* e) noexcept
+{
+    using lookup = utf8_lookup4<>;
+#if JSON_SCAN_NEON
+    const uint8x16_t t1h = vld1q_u8(lookup::byte_1_high.data());
+    const uint8x16_t t1l = vld1q_u8(lookup::byte_1_low.data());
+    const uint8x16_t t2h = vld1q_u8(lookup::byte_2_high.data());
+    uint8x16_t prev = vdupq_n_u8(0);
+    while (e - block >= 16)
+    {
+        const uint8x16_t in = vld1q_u8(block);
+        if (vmaxvq_u8(vorrq_u8(in, prev)) < 0x80)
+        {
+            // ASCII here, and no sequence from the block before to finish
+            prev = in;
+            block += 16;
+            continue;
+        }
+        const uint8x16_t prev1 = vextq_u8(prev, in, 15);
+        const uint8x16_t sc = vandq_u8(vandq_u8(vqtbl1q_u8(t1h, vshrq_n_u8(prev1, 4)), vqtbl1q_u8(t1l, vandq_u8(prev1, vdupq_n_u8(0x0F)))), vqtbl1q_u8(t2h, vshrq_n_u8(in, 4)));
+        const uint8x16_t must23 = vorrq_u8(vqsubq_u8(vextq_u8(prev, in, 14), vdupq_n_u8(0xE0 - 0x80)), vqsubq_u8(vextq_u8(prev, in, 13), vdupq_n_u8(0xF0 - 0x80)));
+        const uint8x16_t err = veorq_u8(vandq_u8(must23, vdupq_n_u8(0x80)), sc);
+        if (vmaxvq_u8(err) != 0)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#else
+    const __m128i t1h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t1l = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_low.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t2h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_2_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i nibble = _mm_set1_epi8(0x0F);
+    const __m128i zero = _mm_setzero_si128();
+    __m128i prev = zero;
+    while (e - block >= 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(block))); // NOLINT(bugprone-casting-through-void)
+        if (_mm_movemask_epi8(_mm_or_si128(in, prev)) == 0)
+        {
+            prev = in;
+            block += 16;
+            continue;
+        }
+        const __m128i prev1 = _mm_alignr_epi8(in, prev, 15);
+        const __m128i sc = _mm_and_si128(_mm_and_si128(_mm_shuffle_epi8(t1h, _mm_and_si128(_mm_srli_epi16(prev1, 4), nibble)),
+                                         _mm_shuffle_epi8(t1l, _mm_and_si128(prev1, nibble))),
+                                         _mm_shuffle_epi8(t2h, _mm_and_si128(_mm_srli_epi16(in, 4), nibble)));
+        const __m128i must23 = _mm_or_si128(_mm_subs_epu8(_mm_alignr_epi8(in, prev, 14), _mm_set1_epi8(0xE0 - 0x80)),
+                                            _mm_subs_epu8(_mm_alignr_epi8(in, prev, 13), _mm_set1_epi8(0xF0 - 0x80)));
+        const __m128i err = _mm_xor_si128(_mm_and_si128(must23, _mm_set1_epi8(static_cast<char>(-128))), sc);
+        if (_mm_movemask_epi8(_mm_cmpeq_epi8(err, zero)) != 0xFFFF)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#endif
+    return block;
+}
 #endif
 
 }  // namespace detail
@@ -12604,6 +12676,64 @@ inline std::size_t valid_utf8_prefix(const unsigned char* data, std::size_t n) n
     return pos;
 }
 
+/*!
+@brief whether [data, data+n) is valid UTF-8, for checking whole strings
+
+The same answer as valid_utf8_prefix(data, n) == n. Short ASCII strings, the
+most common keys and values, are decided in one test, and long strings are
+checked 16 bytes at a time where the vector check is available.
+*/
+inline bool valid_utf8_string(const unsigned char* data, std::size_t n) noexcept
+{
+    constexpr std::uint64_t high = 0x8080808080808080ull;
+    if (n <= 16)
+    {
+        // two overlapping words cover 8 to 16 bytes, one partial word fewer
+        std::uint64_t first = 0;
+        std::uint64_t last = 0;
+        if (n >= 8)
+        {
+            std::memcpy(&first, data, sizeof(first));
+            std::memcpy(&last, data + n - 8, sizeof(last));
+        }
+        else if (n != 0)
+        {
+            std::memcpy(&first, data, n);
+        }
+        if (((first | last) & high) == 0)
+        {
+            return true;
+        }
+    }
+#if JSON_SCAN_VECTOR_UTF8
+    else if (n >= 32
+#if JSON_SCAN_SSSE3_DISPATCH
+             && cpu_has_ssse3()
+#endif
+            )
+    {
+        const unsigned char* block = vector_utf8_valid_blocks(data, data + n);
+        // the rest from the start of a sequence that may be cut at block
+        for (int i = 1; i <= 3 && block - i >= data; ++i)
+        {
+            const unsigned char c = block[-i];
+            if (c < 0x80)
+            {
+                break;
+            }
+            if (c >= 0xC0)
+            {
+                block -= i;
+                break;
+            }
+        }
+        const auto checked = static_cast<std::size_t>(block - data);
+        return checked + valid_utf8_prefix(block, n - checked) == n;
+    }
+#endif
+    return valid_utf8_prefix(data, n) == n;
+}
+
 #if JSON_SCAN_VECTOR_UTF8
 /// the end of scan_string_vector from block, where the vector loop stopped
 /// (ill-formed UTF-8, or fewer than 16 bytes left): one byte or sequence at a
@@ -22069,6 +22199,19 @@ class binary_reader
         return get_bytes(len, "string", result);
     }
 
+    /// whether @a s is valid UTF-8: ASCII eight bytes at a time, the rest
+    /// sequence by sequence (as the binary writers check strings)
+    static bool is_valid_utf8_string(const string_t& s, std::true_type /*is_byte*/) noexcept
+    {
+        return valid_utf8_string(reinterpret_cast<const unsigned char*>(s.data()), s.size());
+    }
+
+    /// whether @a s, of a character type wider than a byte, is valid UTF-8
+    static bool is_valid_utf8_string(const string_t& s, std::false_type /*is_byte*/) noexcept
+    {
+        return is_valid_utf8(s);
+    }
+
     /*!
     @brief validate a decoded text string (value or object key) against @ref error_handler
 
@@ -22088,7 +22231,16 @@ class binary_reader
     */
     bool check_string_utf8(string_t& result, const char* context)
     {
-        if (error_handler == error_handler_t::keep || is_valid_utf8(result))
+        return error_handler == error_handler_t::keep || check_string_utf8_handler(result, context);
+    }
+
+    /// @ref check_string_utf8 for an error handler other than @ref
+    /// error_handler_t::keep; out of line, so that reading strings without a
+    /// check does not pay for the code of the check
+    JSON_HEDLEY_NEVER_INLINE
+    bool check_string_utf8_handler(string_t& result, const char* context)
+    {
+        if (is_valid_utf8_string(result, std::integral_constant<bool, sizeof(typename string_t::value_type) == 1>()))
         {
             return true;
         }
@@ -30306,6 +30458,25 @@ class binary_writer
         }
     }
 
+    /// write the head and the characters of @a value, at most
+    /// max_text_in_reservation of them, in one reservation
+    JSON_HEDLEY_ALWAYS_INLINE void write_cbor_string_reserved(const string_t& value)
+    {
+        CharType* const out = oa.reserve(value.size() + 2);
+        CharType* p = out;
+        if (value.size() <= 0x17)
+        {
+            *p++ = to_char_type(static_cast<std::uint8_t>(0x60 + value.size()));
+        }
+        else
+        {
+            *p++ = to_char_type(0x78);
+            *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
+        }
+        copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
+        oa.commit(static_cast<std::size_t>(p - out) + value.size());
+    }
+
     /*!
     @brief write a CBOR text string
 
@@ -30327,19 +30498,7 @@ class binary_writer
             // the string is written as it is: head and characters in one reservation
             if (JSON_HEDLEY_LIKELY(value.size() <= max_text_in_reservation))
             {
-                CharType* const out = oa.reserve(value.size() + 2);
-                CharType* p = out;
-                if (value.size() <= 0x17)
-                {
-                    *p++ = to_char_type(static_cast<std::uint8_t>(0x60 + value.size()));
-                }
-                else
-                {
-                    *p++ = to_char_type(0x78);
-                    *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
-                }
-                copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
-                oa.commit(static_cast<std::size_t>(p - out) + value.size());
+                write_cbor_string_reserved(value);
                 return;
             }
         }
@@ -30353,6 +30512,13 @@ class binary_writer
     JSON_HEDLEY_NEVER_INLINE
     void write_cbor_checked_string(const string_t& value, const BasicJsonType& context)
     {
+        if (error_handler == error_handler_t::strict && value.size() <= max_text_in_reservation)
+        {
+            check_utf8(value, context);
+            write_cbor_string_reserved(value);
+            return;
+        }
+
         string_t storage;
         const string_t& sanitized = sanitize_utf8_for_write(value, context, storage);
 
@@ -30367,6 +30533,27 @@ class binary_writer
     // MsgPack //
     /////////////
 
+    /// write the head and the characters of @a value, at most
+    /// max_text_in_reservation of them, in one reservation
+    JSON_HEDLEY_ALWAYS_INLINE void write_msgpack_string_reserved(const string_t& value)
+    {
+        CharType* const out = oa.reserve(value.size() + 2);
+        CharType* p = out;
+        if (value.size() <= 31)
+        {
+            // fixstr
+            *p++ = to_char_type(static_cast<std::uint8_t>(0xA0 | value.size()));
+        }
+        else
+        {
+            // str 8
+            *p++ = to_char_type(0xD9);
+            *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
+        }
+        copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
+        oa.commit(static_cast<std::size_t>(p - out) + value.size());
+    }
+
     /*!
     @brief write a MessagePack str
 
@@ -30384,21 +30571,7 @@ class binary_writer
             // the string is written as it is: head and characters in one reservation
             if (JSON_HEDLEY_LIKELY(value.size() <= max_text_in_reservation))
             {
-                CharType* const out = oa.reserve(value.size() + 2);
-                CharType* p = out;
-                if (value.size() <= 31)
-                {
-                    // fixstr
-                    *p++ = to_char_type(static_cast<std::uint8_t>(0xA0 | value.size()));
-                }
-                else
-                {
-                    // str 8
-                    *p++ = to_char_type(0xD9);
-                    *p++ = to_char_type(static_cast<std::uint8_t>(value.size()));
-                }
-                copy_characters(p, reinterpret_cast<const CharType*>(value.data()), value.size());
-                oa.commit(static_cast<std::size_t>(p - out) + value.size());
+                write_msgpack_string_reserved(value);
                 return;
             }
         }
@@ -30411,6 +30584,13 @@ class binary_writer
     JSON_HEDLEY_NEVER_INLINE
     void write_msgpack_checked_string(const string_t& value, const BasicJsonType& context)
     {
+        if (error_handler == error_handler_t::strict && value.size() <= max_text_in_reservation)
+        {
+            check_utf8(value, context);
+            write_msgpack_string_reserved(value);
+            return;
+        }
+
         string_t storage;
         const string_t& sanitized = sanitize_utf8_for_write(value, context, storage);
 
@@ -32020,10 +32200,9 @@ class binary_writer
     static void check_utf8(const string_t& s, const BasicJsonType& context)
     {
         const auto* data = reinterpret_cast<const unsigned char*>(s.data());
-        const std::size_t valid = valid_utf8_prefix(data, s.size());
-        if (JSON_HEDLEY_UNLIKELY(valid != s.size()))
+        if (JSON_HEDLEY_UNLIKELY(!valid_utf8_string(data, s.size())))
         {
-            throw_invalid_utf8(data, valid, context);
+            throw_invalid_utf8(data, valid_utf8_prefix(data, s.size()), context);
         }
     }
 

tests/src/unit-guard-page.cpp

@@ -403,6 +403,7 @@ TEST_CASE("guard page")
                     CHECK(nlohmann::detail::find_string_special(p, n) <= n);
                     CHECK(nlohmann::detail::find_ascii_copyable_run(p, n) <= n);
                     CHECK(nlohmann::detail::valid_utf8_prefix(p, n) <= n);
+                    CHECK(nlohmann::detail::valid_utf8_string(p, n) == (nlohmann::detail::valid_utf8_prefix(p, n) == n));
 #if defined(JSON_USE_SIMDUTF) && defined(JSON_HAS_CPP_17)
                     CHECK(nlohmann::detail::find_string_delimiter(p, n) <= n);
 #endif

tests/src/unit-unicode.cpp

@@ -13,6 +13,7 @@ using nlohmann::json;
 
 #include <algorithm>
 #include <fstream>
+#include <random>
 #include <string>
 #include "make_test_data_available.hpp"
 #include "test_utils.hpp"
@@ -1631,3 +1632,40 @@ TEST_CASE("Unicode RFC 3629")
 }
 
 DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+TEST_CASE("UTF-8 check of whole strings")
+{
+    // valid_utf8_string() decides short strings with a word test and long ones
+    // 16 bytes at a time where a vector check is available; it must agree with
+    // the byte-wise check for every mix of ASCII, sequences of each length,
+    // ill-formed and truncated sequences, at every length and offset
+    const char* const pieces[] =
+    {
+        "a", "\x7F", "\xC3\xA9", "\xE2\x82\xAC", "\xE0\xA0\x80", "\xEF\xBF\xBF", "\xF0\x9F\x98\x80", "\xF4\x8F\xBF\xBF",
+        "\x80", "\xC0\xAF", "\xC3", "\xE2\x82", "\xED\xA0\x80", "\xF0\x9F", "\xF4\x90\x80\x80", "\xFF"
+    };
+    // std::mt19937's sequence is fixed by the standard; no distribution is used
+    std::mt19937 rng(5);
+    std::size_t valid = 0;
+    for (int i = 0; i < 100000; ++i)
+    {
+        std::string s;
+        const std::size_t length = rng() % 100;
+        const std::uint32_t ascii = rng() % 101;
+        while (s.size() < length)
+        {
+            // mostly well-formed pieces (the first eight), so that long valid
+            // strings are common
+            s += (rng() % 100 < ascii) ? std::string("x") : std::string(pieces[rng() % (rng() % 8 == 0 ? 16 : 8)]);
+        }
+        const auto* data = reinterpret_cast<const unsigned char*>(s.data());
+        const bool expected = nlohmann::detail::is_valid_utf8(s);
+        valid += expected ? 1 : 0;
+        CAPTURE(s)
+        CHECK(nlohmann::detail::valid_utf8_string(data, s.size()) == expected);
+        CHECK((nlohmann::detail::valid_utf8_prefix(data, s.size()) == s.size()) == expected);
+    }
+    // both outcomes are common enough to be tested
+    CHECK(valid > 20000);
+    CHECK(valid < 80000);
+}

tools/generate_natvis/README.md

@@ -6,11 +6,6 @@ The checked-in `nlohmann_json.natvis` covers the default namespace (no ABI tags)
 of ABI tags, with both versioned and unversioned namespace variants. Configurations that enable three or more ABI tags
 are not covered by this reduced file; use `--abi-tags` or `--all` (see below) to generate a file for them.
 
-The visualizers of `basic_json_document` and `basic_json_view` live in the same inline namespaces as `basic_json`, so the
-ABI tags apply to them in the same way, and the `view_mode` template argument is covered by a wildcard: it does not
-multiply the namespaces. Each namespace gets three more types (`basic_json_view<*,*>`, `basic_json_document<*,*>`, and
-`detail::view::node`), which roughly doubles the size of the file.
-
 The ABI tag list and the library version are parsed from
 `include/nlohmann/detail/abi_macros.hpp`, so this script must be re-run (via
 `make natvis`) whenever an `NLOHMANN_JSON_ABI_TAG_*` macro is added to that
@@ -40,7 +35,7 @@ version other than the one in `abi_macros.hpp`.
 |-------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | `--list-tags`     | Print every ABI tag (in namespace order) with the macro that enables it and the `NLOHMANN_JSON_ABI_TAG_*` identifier, then exit.                                                                                                                    |
 | `--abi-tags TAGS` | Add the namespaces of one configuration, given as a comma-separated list of ABI tags (`_diag,_dp,_ns`; the leading underscore is optional, and the order does not matter). Can be repeated for several configurations. Unknown tags are an error. |
-| `--all`           | Generate the namespaces for every combination of ABI tags. The file size doubles with each tag (about 60 MB for 11 tags). Cannot be combined with `--abi-tags`.                                                                                    |
+| `--all`           | Generate the namespaces for every combination of ABI tags. The file size doubles with each tag (about 60 MB for 12 tags). Cannot be combined with `--abi-tags`.                                                                                    |
 | `--output FILE`   | Write the Natvis file to `FILE` instead of `nlohmann_json.natvis` in the repository root.                                                                                                                                                           |
 
 Each configuration gets both the versioned (`json_abi_diag_dp_ns_v3_12_0`) and the unversioned (`json_abi_diag_dp_ns`,

tools/generate_natvis/generate_natvis.py

@@ -95,7 +95,7 @@ if __name__ == '__main__':
                                 'optional) of one configuration to add to the default namespaces; can be repeated')
     selection.add_argument('--all', action='store_true',
                            help='Generate the namespaces for every combination of ABI tags; the file size doubles '
-                                'with each ABI tag (about 60 MB for 11 tags)')
+                                'with each ABI tag (about 60 MB for 12 tags)')
     parser.add_argument('--list-tags', action='store_true',
                         help='List the ABI tags with the macros that enable them, and exit')
     parser.add_argument('--output', metavar='FILE',

tools/generate_natvis/nlohmann_json.natvis.j2

@@ -5,39 +5,6 @@
 <!-- * * * * * * * * AUTO-GENERATED FILE  * * * * * * * * -->
 
 <AutoVisualizer xmlns="http://schemas.microsoft.com/vstudio/debugger/natvis/2010">
-{# The children of the container node `node` (a pointer expression) of the document `doc` (a document_data pointer
-   expression). A node is only an index into the source text: strings are read from `doc`, so the strings of the
-   direct children are readable here, but the children of a nested container are shown as nodes (see the node
-   visualizer below). Containers with edits (flag 8, "moved") are not expanded. The node kinds are the numbers of
-   node_kind: 1 object, 2 array, 3 string; the float token is shown only by the node visualizer. #}
-{% macro children(node, doc=none) %}
-            <CustomListItems MaxItemsPerView="1000">
-              <Variable Name="nd" InitialValue="{{ node }}" />
-              <Variable Name="obj" InitialValue="nd != 0 &amp;&amp; (int)nd-&gt;kind == 1" />
-              <Variable Name="n" InitialValue="(nd != 0 &amp;&amp; ((int)nd-&gt;kind == 1 || (int)nd-&gt;kind == 2) &amp;&amp; (nd-&gt;flags &amp; 8) == 0) ? (int)nd-&gt;len : 0" />
-              <Variable Name="p" InitialValue="nd + 1" />
-              <Variable Name="i" InitialValue="0" />
-              <Size>obj ? 2 * n : n</Size>
-              <Loop Condition="i &lt; n">
-{% if doc %}
-                <Item Name="key" Condition="obj">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% else %}
-                <Item Name="key" Condition="obj">*p</Item>
-{% endif %}
-                <Exec Condition="obj">p = p + 1</Exec>
-{% if doc %}
-                <Item Name="value" Condition="obj &amp;&amp; (int)p-&gt;kind == 3">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% endif %}
-                <Item Name="value" Condition="obj{% if doc %} &amp;&amp; (int)p-&gt;kind != 3{% endif %}">*p</Item>
-{% if doc %}
-                <Item Condition="!obj &amp;&amp; (int)p-&gt;kind == 3">{{ doc }}-&gt;base._Elems[p-&gt;flags &amp; 3] + p-&gt;off,[p-&gt;len]s8</Item>
-{% endif %}
-                <Item Condition="!obj{% if doc %} &amp;&amp; (int)p-&gt;kind != 3{% endif %}">*p</Item>
-                <Exec>p = p + (((int)p-&gt;kind == 1 || (int)p-&gt;kind == 2) ? p-&gt;next : 1)</Exec>
-                <Exec>i = i + 1</Exec>
-              </Loop>
-            </CustomListItems>
-{%- endmacro %}
 {% for ns in namespaces %}
     <!-- Namespace {{ ns }} -->
     <Type Name="{{ ns }}::basic_json&lt;*&gt;">
@@ -95,44 +62,5 @@
     </Type>
 
 {% endfor %}
-
-    <!-- json_view / json_document (include/nlohmann/json_view.hpp). They live in the same inline namespaces as
-         basic_json, so the ABI tags need no extra entries; the view_mode is a template argument that the wildcard
-         covers. A view is a pointer to the document's data and a pointer to a node of its flat index; the
-         strings of the source are found through document_data::base (std::array of the MSVC STL: _Elems). -->
-    <Type Name="{{ ns }}::detail::view::node">
-        <DisplayString Condition="(int)kind == 0">null</DisplayString>
-        <DisplayString Condition="(int)kind == 1">object, size={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 2">array, size={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 3">string, length={len}</DisplayString>
-        <DisplayString Condition="(int)kind == 4">boolean: {(flags &amp; 4) != 0}</DisplayString>
-        <DisplayString Condition="(int)kind == 5">int64: {(__int64)((unsigned __int64)len | ((unsigned __int64)next &lt;&lt; 32))}</DisplayString>
-        <DisplayString Condition="(int)kind == 6">uint64: {(unsigned __int64)len | ((unsigned __int64)next &lt;&lt; 32)}</DisplayString>
-        <DisplayString Condition="(int)kind == 7">float, length={len}</DisplayString>
-        <DisplayString>link</DisplayString>
-        <Expand>
-{{ children('this') }}
-        </Expand>
-    </Type>
-
-    <Type Name="{{ ns }}::basic_json_view&lt;*,*&gt;">
-        <DisplayString Condition="m_node == 0">discarded</DisplayString>
-        <DisplayString Condition="(int)m_node-&gt;kind == 3">{m_doc-&gt;base._Elems[m_node-&gt;flags &amp; 3] + m_node-&gt;off,[m_node-&gt;len]s8}</DisplayString>
-        <DisplayString Condition="(int)m_node-&gt;kind == 7">{m_doc-&gt;base._Elems[m_node-&gt;flags &amp; 3] + m_node-&gt;off,[m_node-&gt;len]s8b}</DisplayString>
-        <DisplayString>{*m_node}</DisplayString>
-        <Expand>
-{{ children('m_node', 'm_doc') }}
-        </Expand>
-    </Type>
-
-    <Type Name="{{ ns }}::basic_json_document&lt;*,*&gt;">
-        <DisplayString Condition="m_data._Mypair._Myval2 == 0 || m_data._Mypair._Myval2-&gt;discarded">discarded</DisplayString>
-        <DisplayString>{*m_data._Mypair._Myval2-&gt;nodes}</DisplayString>
-        <Expand>
-            <Item Name="[source]" Condition="m_data._Mypair._Myval2 != 0">m_data._Mypair._Myval2-&gt;src,[m_data._Mypair._Myval2-&gt;size]s8b</Item>
-            <Item Name="[nodes]" Condition="m_data._Mypair._Myval2 != 0">m_data._Mypair._Myval2-&gt;node_count</Item>
-{{ children('m_data._Mypair._Myval2->nodes', 'm_data._Mypair._Myval2') }}
-        </Expand>
-    </Type>
 {% endfor %}
 </AutoVisualizer>