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>

include/nlohmann/detail/input/binary_reader.hpp

@@ -22,7 +22,6 @@
     #include <bit>  //byteswap
 #endif
 
-#include <nlohmann/detail/bit_ops.hpp>
 #include <nlohmann/detail/exceptions.hpp>
 #include <nlohmann/detail/input/input_adapters.hpp>
 #include <nlohmann/detail/input/json_sax.hpp>
@@ -742,13 +741,11 @@ class binary_reader
     bool get_cbor_negative_integer()
     {
         NumberType number{};
-        return get_number(number) && emit_cbor_negative_integer(number);
+        if (JSON_HEDLEY_UNLIKELY(!get_number(number)))
+        {
+            return false;
         }
 
-    /// pass the CBOR negative integer -1 - @a number to the SAX parser
-    template<typename NumberType>
-    bool emit_cbor_negative_integer(const NumberType number)
-    {
         // the value is -1 - number, which fits into number_integer_t
         // whenever number does; the outer cast undoes the integral promotion
         // for number_integer_t types narrower than int
@@ -973,7 +970,7 @@ class binary_reader
             case 0x95:
             case 0x96:
             case 0x97:
-                return enter_cbor_array(conditional_static_cast<std::size_t>(static_cast<unsigned int>(current) & 0x1Fu));
+                return enter_array(conditional_static_cast<std::size_t>(static_cast<unsigned int>(current) & 0x1Fu));
 
             case 0x98: // array (one-byte uint8_t for n follows)
             case 0x99: // array (two-byte uint16_t for n follow)
@@ -982,7 +979,7 @@ class binary_reader
             {
                 std::uint64_t len{};
                 std::size_t size{};
-                return get_cbor_argument(len) && get_cbor_container_size(len, size, "array") && enter_cbor_array(size);
+                return get_cbor_argument(len) && get_cbor_container_size(len, size, "array") && enter_array(size);
             }
 
             case 0x9F: // array (indefinite length)
@@ -1136,10 +1133,16 @@ class binary_reader
                 return get_half_float(false);
 
             case 0xFA: // Single-Precision Float (four-byte IEEE 754)
-                return get_cbor_float<float>();
+            {
+                float number{};
+                return get_number(number) && emit_float(number);
+            }
 
             case 0xFB: // Double-Precision Float (eight-byte IEEE 754)
-                return get_cbor_float<double>();
+            {
+                double number{};
+                return get_number(number) && emit_float(number);
+            }
 
             default: // anything else (0xFF is handled inside the other types)
             {
@@ -1148,16 +1151,6 @@ class binary_reader
         }
     }
 
-    /// read a single- or double-precision float (kept out of
-    /// parse_cbor_value, which is inlined into parse_cbor_internal)
-    template<typename NumberType>
-    JSON_HEDLEY_NEVER_INLINE
-    bool get_cbor_float()
-    {
-        NumberType number{};
-        return get_number(number) && emit_float(number);
-    }
-
     /*!
     @brief reports a nested indefinite-length CBOR string or byte array
     @param[in] type_name  name of the rejected string type
@@ -1683,157 +1676,6 @@ class binary_reader
         return true;
     }
 
-    /*!
-    @brief read a big-endian number of type @a NumberType at @a data
-    */
-    template<typename NumberType>
-    NumberType load_big_endian(const unsigned char* data) const noexcept
-    {
-        NumberType result{};
-        std::memcpy(&result, data, sizeof(NumberType));
-        if (is_little_endian)
-        {
-            byte_swap(result);
-        }
-        return result;
-    }
-
-    /*!
-    @brief read the integers, floats, booleans, and nulls that follow in a
-           definite-length array, without the per-element bookkeeping of
-           @ref parse_cbor_internal
-
-    Stops at the end of the array, at any other kind of value, and when fewer
-    than 9 bytes (the longest of these values) are left in the input, leaving
-    those to @ref parse_cbor_value, which reads them the same way.
-
-    @param[in,out] remaining  the number of elements of the array not yet read
-    @return whether the SAX parser accepted the values
-    */
-    bool parse_cbor_scalar_run(std::size_t& remaining)
-    {
-        while (remaining != 0 && ia.bulk_remaining() >= 9)
-        {
-            const auto* const data = reinterpret_cast<const unsigned char*>(ia.bulk_data());
-            const unsigned char head = data[0];
-            std::size_t size = 0;
-            bool accepted = true;
-            switch (head)
-            {
-                case 0x18:
-                    size = 2;
-                    accepted = emit_unsigned(data[1]);
-                    break;
-                case 0x19:
-                    size = 3;
-                    accepted = emit_unsigned(load_big_endian<std::uint16_t>(data + 1));
-                    break;
-                case 0x1A:
-                    size = 5;
-                    accepted = emit_unsigned(load_big_endian<std::uint32_t>(data + 1));
-                    break;
-                case 0x1B:
-                    size = 9;
-                    accepted = emit_unsigned(load_big_endian<std::uint64_t>(data + 1));
-                    break;
-                case 0x38:
-                    size = 2;
-                    accepted = emit_cbor_negative_integer(data[1]);
-                    break;
-                case 0x39:
-                    size = 3;
-                    accepted = emit_cbor_negative_integer(load_big_endian<std::uint16_t>(data + 1));
-                    break;
-                case 0x3A:
-                    size = 5;
-                    accepted = emit_cbor_negative_integer(load_big_endian<std::uint32_t>(data + 1));
-                    break;
-                case 0x3B:
-                    size = 9;
-                    accepted = emit_cbor_negative_integer(load_big_endian<std::uint64_t>(data + 1));
-                    break;
-                case 0xF4:
-                    size = 1;
-                    accepted = sax->boolean(false);
-                    break;
-                case 0xF5:
-                    size = 1;
-                    accepted = sax->boolean(true);
-                    break;
-                case 0xF6:
-                    size = 1;
-                    accepted = sax->null();
-                    break;
-                case 0xFA:
-                    size = 5;
-                    accepted = emit_float(load_big_endian<float>(data + 1));
-                    break;
-                case 0xFB:
-                    size = 9;
-                    accepted = emit_float(load_big_endian<double>(data + 1));
-                    break;
-                default:
-                    if (head <= 0x17)
-                    {
-                        size = 1;
-                        accepted = sax->number_unsigned(static_cast<number_unsigned_t>(head));
-                    }
-                    else if (head >= 0x20 && head <= 0x37)
-                    {
-                        size = 1;
-                        accepted = sax->number_integer(static_cast<std::int8_t>(0x20 - 1 - head));
-                    }
-                    else
-                    {
-                        // any other value is read by parse_cbor_value
-                        return true;
-                    }
-                    break;
-            }
-            // as if read by get() and get_number(): current is the head byte
-            ia.bulk_skip(size);
-            chars_read += size;
-            current = head;
-            --remaining;
-            if (JSON_HEDLEY_UNLIKELY(!accepted))
-            {
-                return false;
-            }
-        }
-        return true;
-    }
-
-    /*!
-    @brief open a definite-length array and read the scalars it begins with
-           (see @ref parse_cbor_scalar_run)
-
-    Only the start of an array is tried, which covers arrays of numbers
-    (e.g. coordinates) but costs arrays of other values only one test, not one
-    per element.
-
-    @param[in] len  the number of elements
-    @return whether the SAX parser accepted the array and the values read
-    */
-    bool enter_cbor_array(const std::size_t len)
-    {
-        return enter_array(len) && (len == 0 || parse_cbor_array_start(std::integral_constant<bool, bulk_scan> {}));
-    }
-
-    /// read the scalars the array just opened begins with; out of line, so
-    /// that @ref parse_cbor_value stays small enough to be inlined into
-    /// @ref parse_cbor_internal
-    JSON_HEDLEY_NEVER_INLINE
-    bool parse_cbor_array_start(std::true_type /*bulk*/)
-    {
-        return parse_cbor_scalar_run(container_stack.back().remaining);
-    }
-
-    /// input that is not contiguous: every value is read by parse_cbor_value
-    bool parse_cbor_array_start(std::false_type /*bulk*/) const noexcept
-    {
-        return true;
-    }
-
     /*!
     @brief read a CBOR value and everything nested inside it
 
@@ -1863,7 +1705,6 @@ class binary_reader
                 // pop_back() below, which destroys the container_stack element
                 // it would otherwise alias
                 const container_frame top = container_stack.back();
-                const bool is_object = top.is_object;
                 bool at_end = false;
 
                 if (top.remaining != npos)
@@ -1875,7 +1716,7 @@ class binary_reader
                     {
                         // claim the element about to be read
                         --container_stack.back().remaining;
-                        if (is_object)
+                        if (top.is_object)
                         {
                             get();
                         }
@@ -1888,7 +1729,7 @@ class binary_reader
                     // Testing for it consumes a byte, which is the first byte
                     // of the next element when it is not one.
                     at_end = (get() == 0xFF);
-                    fetch = is_object;
+                    fetch = top.is_object;
                 }
 
                 if (at_end)
@@ -1905,7 +1746,7 @@ class binary_reader
                     continue;
                 }
 
-                if (is_object)
+                if (top.is_object)
                 {
                     if (JSON_HEDLEY_UNLIKELY(!skip_cbor_key_tags(tag_handler)))
                     {
@@ -3850,13 +3691,13 @@ class binary_reader
                 return sax->null();
 
             case 0xFB:
-                return sax->number_float(static_cast<number_float_t>(-1.0), no_source);
+                return sax->number_float(static_cast<number_float_t>(-1.0), "");
 
             case 0xFC:
-                return sax->number_float(static_cast<number_float_t>(0.0), no_source);
+                return sax->number_float(static_cast<number_float_t>(0.0), "");
 
             case 0xFD:
-                return sax->number_float(static_cast<number_float_t>(1.0), no_source);
+                return sax->number_float(static_cast<number_float_t>(1.0), "");
 
             case 0xFF: // empty string
             {
@@ -4225,27 +4066,6 @@ class binary_reader
         return current;
     }
 
-    /// reverse the byte order of a double with a single byteswap(): GCC does
-    /// not merge the byte-wise swap of the generic version into one
-    static void byte_swap(double& number) noexcept
-    {
-        static_assert(sizeof(double) == sizeof(std::uint64_t), "double must have 64 bits");
-        std::uint64_t bits{};
-        std::memcpy(&bits, &number, sizeof(bits));
-        bits = byteswap(bits);
-        std::memcpy(&number, &bits, sizeof(bits));
-    }
-
-    /// reverse the byte order of a float with a single byteswap()
-    static void byte_swap(float& number) noexcept
-    {
-        static_assert(sizeof(float) == sizeof(std::uint32_t), "float must have 32 bits");
-        std::uint32_t bits{};
-        std::memcpy(&bits, &number, sizeof(bits));
-        bits = byteswap(bits);
-        std::memcpy(&number, &bits, sizeof(bits));
-    }
-
     template<class NumberType>
     static void byte_swap(NumberType& number)
     {
@@ -4381,7 +4201,7 @@ class binary_reader
         {
             return out_of_range_error(exception_id::number_overflow, "number overflow", "value");
         }
-        return sax->number_float(result, no_source);
+        return sax->number_float(result, "");
     }
 
     /*!
@@ -4441,7 +4261,7 @@ class binary_reader
         }();
         return sax->number_float((half & 0x8000u) != 0
                                  ? static_cast<number_float_t>(-val)
-                                 : static_cast<number_float_t>(val), no_source);
+                                 : static_cast<number_float_t>(val), "");
     }
 
     /*!
@@ -4473,6 +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
 
@@ -4492,7 +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;
         }
@@ -4766,10 +4608,6 @@ class binary_reader
     /// the SAX parser
     json_sax_t* sax = nullptr;
 
-    /// the (empty) source text of every number_float event: binary formats
-    /// have none, and a string_t built for each float would cost more
-    const string_t no_source{};
-
     /// the containers that have been opened and not closed yet; see @ref container_frame
     std::vector<container_frame> container_stack{};
 

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);
+                write_cbor_string_reserved(value);
-                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());
                 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);
+                write_msgpack_string_reserved(value);
-                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());
                 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_utf8_string(data, s.size())))
-        if (JSON_HEDLEY_UNLIKELY(valid != s.size()))
         {
-            throw_invalid_utf8(data, valid, context);
+            throw_invalid_utf8(data, valid_utf8_prefix(data, s.size()), context);
         }
     }
 

tests/src/unit-cbor.cpp

@@ -29,9 +29,7 @@ using nlohmann::json;
 #include "round_trip_corpus.hpp"
 #include "test_utils.hpp"
 #include "sax_countdown.hpp"
-#include "sax_event_loggers.hpp"
 using utils::SaxCountdown;
-using utils::SaxEventLogger;
 
 
 TEST_CASE("CBOR")
@@ -2448,78 +2446,6 @@ TEST_CASE("CBOR SAX parsing stops at every event")
     CHECK(count_events(std::vector<std::uint8_t>({0xBF, 0x61, 'a', 0x9F, 0x01, 0xFF, 0xFF})) == 6);
 }
 
-TEST_CASE("CBOR arrays of scalars from contiguous and non-contiguous input")
-{
-    // contiguous input reads the integers, floats, booleans, and nulls an
-    // array begins with in one loop, other input value by value; both must
-    // give the same events, value, or error for the complete document and for
-    // every truncation of it, and stop at every event a SAX handler rejects
-    const std::vector<std::uint8_t> cbor =
-    {
-        0xA3,                                                       // map(3)
-        0x61, 'a',                                                  // "a"
-        0x98, 0x16,                                                 // array(22)
-        0x00, 0x17, 0x18, 0xFF, 0x19, 0x01, 0x00,                   // 0, 23, 255, 256
-        0x1A, 0x00, 0x01, 0x00, 0x00,                               // 65536
-        0x1B, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,       // 18446744073709551615
-        0x20, 0x37, 0x38, 0xFF, 0x39, 0x01, 0x00,                   // -1, -24, -256, -257
-        0x3A, 0x00, 0x01, 0x00, 0x00,                               // -65537
-        0x3B, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,       // -18446744073709551616 (a float)
-        0x3B, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,       // -9223372036854775808
-        0xF4, 0xF5, 0xF6,                                           // false, true, null
-        0xFA, 0x3F, 0xC0, 0x00, 0x00,                               // 1.5 (float32)
-        0xFB, 0xBF, 0xB9, 0x99, 0x99, 0x99, 0x99, 0x99, 0x9A,       // -0.1
-        0xF9, 0x3C, 0x00,                                           // 1.0 (float16, ends the loop)
-        0x01, 0x61, 'a',                                            // 1, "a"
-        0xFB, 0x40, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,       // 2.5
-        0x61, 'b',                                                  // "b"
-        0x83,                                                       // array(3)
-        0x82, 0xFB, 0x3F, 0xF4, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // [1.25,
-        0xFB, 0xC0, 0x0E, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,       //  -3.75]
-        0x82, 0xFB, 0x3F, 0xF4, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // [1.25,
-        0xFB, 0xC0, 0x0E, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,       //  -3.75]
-        0x80,                                                       // []
-        0x61, 'c',                                                  // "c"
-        0x84, 0xA0, 0x01, 0xC1, 0x02, 0x81, 0x03                    // [{}, 1, tag(1) 2, [3]]
-    };
-    const auto tags = json::cbor_tag_handler_t::ignore;
-
-    const json j = json::from_cbor(cbor, true, true, tags);
-    CHECK(j["a"].size() == 22);
-    CHECK(j["a"][5] == 18446744073709551615u);
-    CHECK(j["a"][11].is_number_float());
-    CHECK(j["a"][11] == -18446744073709551616.0);
-    CHECK(j["a"][12] == (std::numeric_limits<std::int64_t>::min)());
-    CHECK(j["a"][17] == -0.1);
-    CHECK(j["b"] == json::parse("[[1.25, -3.75], [1.25, -3.75], []]"));
-    CHECK(j["c"] == json::parse("[{}, 1, 2, [3]]"));
-
-    for (std::size_t length = 0; length <= cbor.size(); ++length)
-    {
-        CAPTURE(length)
-        const std::vector<std::uint8_t> contiguous(cbor.begin(), cbor.begin() + static_cast<std::ptrdiff_t>(length));
-        const std::list<std::uint8_t> bytewise(contiguous.begin(), contiguous.end());
-
-        CHECK(json::from_cbor(contiguous, true, false, tags).dump() == json::from_cbor(bytewise.begin(), bytewise.end(), true, false, tags).dump());
-
-        SaxEventLogger from_contiguous;
-        SaxEventLogger from_bytewise;
-        CHECK(json::sax_parse(contiguous, &from_contiguous, json::input_format_t::cbor, true, false, false, tags)
-              == json::sax_parse(bytewise.begin(), bytewise.end(), &from_bytewise, json::input_format_t::cbor, true, false, false, tags));
-        CHECK(from_contiguous.events == from_bytewise.events);
-    }
-
-    for (int events = 0; events <= 60; ++events)
-    {
-        CAPTURE(events)
-        const std::list<std::uint8_t> bytewise(cbor.begin(), cbor.end());
-        SaxCountdown from_contiguous(events);
-        SaxCountdown from_bytewise(events);
-        CHECK(json::sax_parse(cbor, &from_contiguous, json::input_format_t::cbor, true, false, false, tags)
-              == json::sax_parse(bytewise.begin(), bytewise.end(), &from_bytewise, json::input_format_t::cbor, true, false, false, tags));
-    }
-}
-
 TEST_CASE("CBOR indefinite-length strings do not recurse per chunk")
 {
     // Reading an indefinite-length string or byte array used to call itself

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);
+}