json_view: keep accept()'s string check out of run()

Clang 23 made run() slower on files without escapes (mesh.pretty +15 %) when
the check-only string path was a separate call target. Dispatch inside
slow_string() instead, so the call in run() is the same as before, and skip
the SIMD run scan between escapes that follow each other (twitterescaped
1.11 -> 0.82 against the previous commit).

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

docs/mkdocs/docs/api/basic_json_document/accept.md

@@ -13,9 +13,15 @@ function never throws an exception for invalid input, and the returned `#!cpp bo
 is returned.
 
 The result is the same as that of `BasicJsonType::accept()` by design; the difference is how the check is made. This
-function runs the document's index builder (and discards the index), while `BasicJsonType::accept()` runs the SAX
-parser with a handler that ignores every event. Use this one if you go on to parse the same input with
-[`parse()`](parse.md) anyway, so that both use the same code; either is fine for a plain yes/no answer.
+function runs the document's builder, which checks the text as [`parse()`](parse.md) does, while
+`BasicJsonType::accept()` runs the SAX parser with a handler that ignores every event. Use this one if you go on to
+parse the same input with [`parse()`](parse.md) anyway, so that both use the same code; either is fine for a plain
+yes/no answer.
+
+When both `ignore_comments` and `ignore_trailing_commas` are `#!cpp false` (the default), the builder writes no index.
+With either option set, it builds the index as [`parse()`](parse.md) does and discards it: the result is the same, only
+the time differs. (Neither option needs the index; the variant that skips it exists only for the default options to
+keep the code size down.)
 
 ## Template parameters
 

include/nlohmann/detail/view/builder.hpp

@@ -69,7 +69,8 @@ struct parse_failure
 };
 
 /// FloatType: the number_float_t of the document, whose overflow parse() rejects
-template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Sentinel>
+/// Store: false for accept(): the grammar, number, string and UTF-8 checks run as for parse(), but no node is written
+template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Sentinel, bool Store>
 class builder
 {
   public:
@@ -245,11 +246,87 @@ class builder
         return p;
     }
 
+    /// the string from p, the first backslash, control character, ill-formed UTF-8 byte, or the end, checked and not
+    /// decoded (accept() stores nothing, so the arena is not written): the checks, errors and error positions of
+    /// slow_string(), for everything up to the closing quote
+    NLOHMANN_VIEW_NOINLINE decoded check_string(const unsigned char* p)
+    {
+        const unsigned char* const end = m_end;
+        for (;;)
+        {
+            if (p == end)
+            {
+                return failed(error_code::string_missing_quote, p);
+            }
+            const unsigned char c = *p;
+            if (c == '"')
+            {
+                ++p;
+                return decoded{p, 0, 0};
+            }
+            if (c != '\\')
+            {
+                return failed(c < 0x20 ? error_code::string_control_character : error_code::string_utf8, p);
+            }
+            ++p;
+            if (p == end)
+            {
+                return failed(error_code::string_missing_quote, p);
+            }
+            const unsigned char d = *p++;
+            if (d == 'u')
+            {
+                const int cp = hex4(p); // (moves p past the four digits)
+                if (NLOHMANN_VIEW_UNLIKELY(cp < 0))
+                {
+                    return failed(error_code::string_unicode_hex, p);
+                }
+                if (NLOHMANN_VIEW_UNLIKELY((cp & 0xF800) == 0xD800)) // a surrogate
+                {
+                    if (cp >= 0xDC00)
+                    {
+                        return failed(error_code::string_surrogate_low, p);
+                    }
+                    if (end - p < 2 || p[0] != '\\' || p[1] != 'u')
+                    {
+                        return failed(error_code::string_surrogate_high, p);
+                    }
+                    p += 2;
+                    const int lo = hex4(p); // (moves p past the four digits)
+                    if (lo < 0)
+                    {
+                        return failed(error_code::string_unicode_hex, p);
+                    }
+                    if (lo < 0xDC00 || lo > 0xDFFF)
+                    {
+                        return failed(error_code::string_surrogate_high, p);
+                    }
+                }
+            }
+            else if (!(d == '"' || d == '\\' || d == '/' || d == 'b' || d == 'f' || d == 'n' || d == 'r' || d == 't'))
+            {
+                --p;
+                return failed(error_code::string_escape, p);
+            }
+            if (p != end && *p == '\\')
+            {
+                continue; // (escapes in a row: no run of text between them)
+            }
+            p = scan_string_run(p, end); // the text up to the next backslash, quote, control character, or ill-formed UTF-8
+        }
+    }
+
     /// s points to the start of the string, p to a backslash, a control
     /// character, an ill-formed UTF-8 byte, or the end: decode the string into
     /// the arena
     NLOHMANN_VIEW_NOINLINE decoded slow_string(const unsigned char* s, const unsigned char* p)
     {
+        if (!enabled(Store))
+        {
+            // (accept(): the string is checked, not decoded)
+            static_cast<void>(s);
+            return check_string(p);
+        }
         // single-character escapes; 0: invalid (and 'u', handled separately)
         static const std::array<char, 128> simple_escape =
         {
@@ -610,10 +687,13 @@ run_found:
 
         NLOHMANN_VIEW_ALWAYS_INLINE bool run()
         {
-            cold.m_doc.reserve(estimate_nodes(reinterpret_cast<const char*>(b), static_cast<std::size_t>(e - b)));
-            base = cold.m_doc.nodes;
-            out = base;
-            cap = base + cold.m_doc.node_capacity;
+            if (enabled(Store))
+            {
+                cold.m_doc.reserve(estimate_nodes(reinterpret_cast<const char*>(b), static_cast<std::size_t>(e - b)));
+                base = cold.m_doc.nodes;
+                out = base;
+                cap = base + cold.m_doc.node_capacity;
+            }
 
             if (e - p >= 3 && p[0] == 0xEF && p[1] == 0xBB && p[2] == 0xBF)
             {
@@ -797,7 +877,10 @@ close_object:
             // an array pays nothing for this)
             if (NLOHMANN_VIEW_UNLIKELY(cur_count >= document_data::index_min_members))
             {
-                cold.note_large_object(cur_idx);
+                if (enabled(Store))
+                {
+                    cold.note_large_object(cur_idx);
+                }
             }
 
 close_container:
@@ -957,6 +1040,12 @@ indent_done:
         /// an integer's value); two stores on little-endian targets
         NLOHMANN_VIEW_ALWAYS_INLINE node* push_node(node_kind k, std::uint8_t flags, std::uint16_t extra, std::size_t off, std::uint64_t second)
         {
+            if (!enabled(Store))
+            {
+                // accept(): nothing is stored, so what computes the arguments is not run either. base and out stay
+                // null: their difference, which open() takes, is 0.
+                return out;
+            }
             if (NLOHMANN_VIEW_UNLIKELY(out == cap))
             {
                 const auto n = static_cast<std::size_t>(out - base);
@@ -1015,9 +1104,12 @@ indent_done:
 
         NLOHMANN_VIEW_ALWAYS_INLINE void close()
         {
-            node& n = base[cur_idx];
-            n.len = cur_count;
-            n.next = static_cast<std::uint32_t>(out - base) - cur_idx;
+            if (enabled(Store))
+            {
+                node& n = base[cur_idx];
+                n.len = cur_count;
+                n.next = static_cast<std::uint32_t>(out - base) - cur_idx;
+            }
             if (--depth != 0)
             {
                 if (NLOHMANN_VIEW_LIKELY(depth <= 64))
@@ -1082,7 +1174,7 @@ indent_done:
             {
                 ++p;
                 const unsigned char* const f0 = p;
-                p = skip_fraction_digits(p, e);
+                p = skip_digits(p, e);
                 if (NLOHMANN_VIEW_UNLIKELY(p == f0))
                 {
                     return fail(error_code::number_after_dot);
@@ -1192,34 +1284,44 @@ indent_done:
 };
 
 /// run the builder with compile-time options
-template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd>
+template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Store>
 inline bool build_with(document_data& d, const char* src, std::size_t size, bool sentinel, parse_failure& failure)
 {
     if (sentinel)
     {
-        builder<FloatType, Comments, TrailingCommas, NulIsEnd, true> bld(d, src, size);
+        builder<FloatType, Comments, TrailingCommas, NulIsEnd, true, Store> bld(d, src, size);
         const bool ok = bld.run();
         failure = bld.failure();
         return ok;
     }
-    builder<FloatType, Comments, TrailingCommas, NulIsEnd, false> bld(d, src, size);
+    builder<FloatType, Comments, TrailingCommas, NulIsEnd, false, Store> bld(d, src, size);
     const bool ok = bld.run();
     failure = bld.failure();
     return ok;
 }
 
 /// sentinel: src[size] is readable and 0 (e.g. std::string); FloatType: the
-/// number_float_t of the document
+/// number_float_t of the document; validate_only: accept(), only the result is
+/// used
 template<typename FloatType, bool NulIsEnd>
-inline bool build(document_data& d, const char* src, std::size_t size, bool comments, bool trailing_commas, bool sentinel, parse_failure& failure)
+inline bool build(document_data& d, const char* src, std::size_t size, bool comments, bool trailing_commas, bool sentinel, bool validate_only,
+                  parse_failure& failure)
 {
+    if (validate_only && !comments && !trailing_commas)
+    {
+        // accept() of plain JSON: nothing is stored. The other option sets
+        // could skip the stores just as well; they keep the builder that
+        // stores only so that there are two more instantiations of the parse
+        // loop, not eight.
+        return build_with<FloatType, false, false, NulIsEnd, false>(d, src, size, sentinel, failure);
+    }
     if (comments)
     {
-        return trailing_commas ? build_with<FloatType, true, true, NulIsEnd>(d, src, size, sentinel, failure)
-               : build_with<FloatType, true, false, NulIsEnd>(d, src, size, sentinel, failure);
+        return trailing_commas ? build_with<FloatType, true, true, NulIsEnd, true>(d, src, size, sentinel, failure)
+               : build_with<FloatType, true, false, NulIsEnd, true>(d, src, size, sentinel, failure);
     }
-    return trailing_commas ? build_with<FloatType, false, true, NulIsEnd>(d, src, size, sentinel, failure)
-           : build_with<FloatType, false, false, NulIsEnd>(d, src, size, sentinel, failure);
+    return trailing_commas ? build_with<FloatType, false, true, NulIsEnd, true>(d, src, size, sentinel, failure)
+           : build_with<FloatType, false, false, NulIsEnd, true>(d, src, size, sentinel, failure);
 }
 
 }  // namespace view

include/nlohmann/detail/view/scan.hpp

@@ -198,57 +198,6 @@ inline const unsigned char* find_block_comment_end(const unsigned char* p, const
     return p;
 }
 
-/// The digits of a fraction, whose number varies from one number to the next: skip_digits() with 16 bytes compared at
-/// once and the end of the run taken from a bit mask (SSE2), instead of a compare and a branch for each digit. That
-/// is about a third of the instructions for a fraction of 12 digits, and no branch that is mispredicted when the
-/// lengths vary. Where there is no SSE2, skip_digits() is used.
-#if NLOHMANN_VIEW_SSE2
-// The last bytes of the input, one at a time. GCC unrolls this loop in every inlined copy of the function below, which
-// makes the parse loop 4 KiB larger and slows down documents that have hardly any fractions; out of line, it does
-// not. Clang does better with the loop inline (a call in the parse loop costs it more than the unrolling).
-#if defined(__GNUC__) && !defined(__clang__)
-inline NLOHMANN_VIEW_NOINLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (p != e && is_digit(*p))
-    {
-        ++p;
-    }
-    return p;
-}
-#else
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (p != e && is_digit(*p))
-    {
-        ++p;
-    }
-    return p;
-}
-#endif
-
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (e - p >= 16)
-    {
-        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
-        // a digit is above '/' and below ':' (signed compares: a byte with the top bit set is negative, and neither)
-        const __m128i digits = _mm_and_si128(_mm_cmpgt_epi8(in, _mm_set1_epi8('/')), _mm_cmplt_epi8(in, _mm_set1_epi8(':')));
-        const auto not_digit = static_cast<std::uint64_t>(~static_cast<unsigned>(_mm_movemask_epi8(digits)) & 0xFFFFu);
-        if (not_digit != 0)
-        {
-            return p + count_trailing_zeros(not_digit);
-        }
-        p += 16;
-    }
-    return skip_digits_tail(p, e);
-}
-#else
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
-{
-    return skip_digits(p, e);
-}
-#endif
-
 /// powers of ten up to 10^19 as integers
 inline std::uint64_t int_pow10(unsigned k) noexcept
 {

include/nlohmann/json_view.hpp

@@ -1060,7 +1060,7 @@ class basic_json_document
                                      const bool ignore_trailing_commas = false)
     {
         basic_json_document d;
-        d.read_range(first, last, allow_exceptions, ignore_comments, ignore_trailing_commas);
+        d.read_range(first, last, error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas);
         return d;
     }
 
@@ -1075,7 +1075,7 @@ class basic_json_document
         basic_json_document d;
         std::string buf;
         append_input(buf, std::forward<InputType>(input));
-        d.build_owned(std::move(buf), allow_exceptions, ignore_comments, ignore_trailing_commas);
+        d.build_owned(std::move(buf), error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas);
         return d;
     }
 
@@ -1088,7 +1088,8 @@ class basic_json_document
     static bool accept(InputType&& input, const bool ignore_comments = false, const bool ignore_trailing_commas = false)
     {
         basic_json_document d;
-        d.read(std::forward<InputType>(input), false, ignore_comments, ignore_trailing_commas);
+        d.read_kind(std::forward<InputType>(input), on_error::validate, ignore_comments, ignore_trailing_commas,
+                    std::integral_constant<detail::view::input_kind, detail::view::classify_input<InputType>::value> {});
         return !d.is_discarded();
     }
 
@@ -1104,7 +1105,7 @@ class basic_json_document
               const bool ignore_comments = false,
               const bool ignore_trailing_commas = false)
     {
-        read_kind(std::forward<InputType>(input), allow_exceptions, ignore_comments, ignore_trailing_commas,
+        read_kind(std::forward<InputType>(input), error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas,
                   std::integral_constant<detail::view::input_kind, detail::view::classify_input<InputType>::value> {});
     }
 
@@ -1463,6 +1464,22 @@ class basic_json_document
         return idx;
     }
 
+    /// what a failed parse does
+    enum class on_error : std::uint8_t
+    {
+        discard, ///< the document is discarded
+        raise,   ///< parse_error is thrown
+        validate ///< accept(): the document is discarded, and no node is stored while parsing
+    };
+
+    static on_error error_mode(bool allow_exceptions) noexcept
+    {
+        return allow_exceptions ? on_error::raise : on_error::discard;
+    }
+
+    /// NulIsEnd of the builder: a null byte ends the input unless strict_nul_handling
+    static constexpr bool nul_is_end = !detail::abi_config::strict_nul_handling;
+
     /// create the storage (sized for the input) on first use
     void ensure_data(const char* src, std::size_t size)
     {
@@ -1474,7 +1491,7 @@ class basic_json_document
 
     /// parse a buffer the document takes ownership of (it replaces the
     /// previous owned copy)
-    void build_owned(std::string&& buf, bool allow_exceptions, bool comments, bool trailing_commas)
+    void build_owned(std::string&& buf, on_error allow_exceptions, bool comments, bool trailing_commas)
     {
         ensure_data(buf.data(), buf.size());
         m_data->owned = std::move(buf);
@@ -1484,7 +1501,7 @@ class basic_json_document
     /// parse a copy of the input; the copy is appended to the cleared owned
     /// buffer, so that its capacity is reused
     template<typename T>
-    void build_copy(T&& input, bool allow_exceptions, bool comments, bool trailing_commas)
+    void build_copy(T&& input, on_error allow_exceptions, bool comments, bool trailing_commas)
     {
         if (!m_data)
         {
@@ -1505,7 +1522,7 @@ class basic_json_document
     }
 
     /// sentinel: src[size] is readable and 0 (std::string, C strings)
-    void build(const char* src, std::size_t size, bool allow_exceptions, bool comments, bool trailing_commas, bool owned, bool sentinel)
+    void build(const char* src, std::size_t size, on_error allow_exceptions, bool comments, bool trailing_commas, bool owned, bool sentinel)
     {
         ensure_data(src, size);
         document_data& d = *m_data;
@@ -1532,7 +1549,8 @@ class basic_json_document
         }
         else
         {
-            ok = detail::view::build < typename BasicJsonType::number_float_t, !detail::abi_config::strict_nul_handling > (d, src, size, comments, trailing_commas, sentinel, failure);
+            ok = detail::view::build<typename BasicJsonType::number_float_t, nul_is_end>(d, src, size, comments, trailing_commas, sentinel,
+                 allow_exceptions == on_error::validate, failure);
         }
         if (NLOHMANN_VIEW_LIKELY(ok))
         {
@@ -1545,7 +1563,7 @@ class basic_json_document
             return;
         }
         std::vector<std::uint32_t>().swap(d.large_objects);
-        if (allow_exceptions)
+        if (allow_exceptions == on_error::raise)
         {
             detail::view::throw_parse_failure<BasicJsonType>(failure, src, size, comments, trailing_commas);
         }
@@ -1553,13 +1571,13 @@ class basic_json_document
 
     // --- input dispatch (see detail::view::input_kind) ---
 
-    void read_kind(std::string&& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::move_string> /*unused*/)
+    void read_kind(std::string&& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::move_string> /*unused*/)
     {
         build_owned(std::move(s), ae, c, tc);
     }
 
     template<typename CharT>
-    void read_kind(CharT* s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::c_string> /*unused*/)
+    void read_kind(CharT* s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::c_string> /*unused*/)
     {
         if (s == nullptr)
         {
@@ -1573,7 +1591,7 @@ class basic_json_document
     }
 
     template<typename Array>
-    void read_kind(Array& a, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::char_array> /*unused*/)
+    void read_kind(Array& a, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::char_array> /*unused*/)
     {
         using CharT = typename std::remove_cv<typename std::remove_extent<Array>::type>::type;
         const std::size_t n = std::extent<Array>::value;
@@ -1584,7 +1602,7 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(const T& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::borrow_range> /*unused*/)
+    void read_kind(const T& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::borrow_range> /*unused*/)
     {
         // std::basic_string guarantees data()[size()] == 0: use it as sentinel
         const auto size = static_cast<std::size_t>(s.size());
@@ -1593,13 +1611,13 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(const T& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::copy_range> /*unused*/)
+    void read_kind(const T& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::copy_range> /*unused*/)
     {
         build_copy(s, ae, c, tc);
     }
 
     template<typename CharT>
-    void read_kind(CharT* s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::wide_c_string> /*unused*/)
+    void read_kind(CharT* s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::wide_c_string> /*unused*/)
     {
         if (s == nullptr)
         {
@@ -1611,13 +1629,13 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(T&& input, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::adapter> /*unused*/)
+    void read_kind(T&& input, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::adapter> /*unused*/)
     {
         build_copy(std::forward<T>(input), ae, c, tc);
     }
 
     template<typename IteratorType>
-    void read_range(IteratorType first, IteratorType last, bool ae, bool c, bool tc)
+    void read_range(IteratorType first, IteratorType last, on_error ae, bool c, bool tc)
     {
         using value_type = typename std::remove_cv<typename std::iterator_traits<IteratorType>::value_type>::type;
         // borrow the range where the library's input adapter scans it as one
@@ -1628,14 +1646,14 @@ class basic_json_document
     }
 
     template<typename IteratorType>
-    void read_range_impl(IteratorType first, IteratorType last, bool ae, bool c, bool tc, std::true_type /*contiguous bytes*/)
+    void read_range_impl(IteratorType first, IteratorType last, on_error ae, bool c, bool tc, std::true_type /*contiguous bytes*/)
     {
         const auto size = static_cast<std::size_t>(std::distance(first, last));
         build(size == 0 ? "" : reinterpret_cast<const char*>(&*first), size, ae, c, tc, false, size == 0); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
     }
 
     template<typename IteratorType>
-    void read_range_impl(IteratorType first, IteratorType last, bool ae, bool c, bool tc, std::false_type /*other*/)
+    void read_range_impl(IteratorType first, IteratorType last, on_error ae, bool c, bool tc, std::false_type /*other*/)
     {
         build_copy(detail::view::make_adapter_input(detail::input_adapter(first, last)), ae, c, tc);
     }

single_include/nlohmann/json_view.hpp

@@ -1218,57 +1218,6 @@ inline const unsigned char* find_block_comment_end(const unsigned char* p, const
     return p;
 }
 
-/// The digits of a fraction, whose number varies from one number to the next: skip_digits() with 16 bytes compared at
-/// once and the end of the run taken from a bit mask (SSE2), instead of a compare and a branch for each digit. That
-/// is about a third of the instructions for a fraction of 12 digits, and no branch that is mispredicted when the
-/// lengths vary. Where there is no SSE2, skip_digits() is used.
-#if NLOHMANN_VIEW_SSE2
-// The last bytes of the input, one at a time. GCC unrolls this loop in every inlined copy of the function below, which
-// makes the parse loop 4 KiB larger and slows down documents that have hardly any fractions; out of line, it does
-// not. Clang does better with the loop inline (a call in the parse loop costs it more than the unrolling).
-#if defined(__GNUC__) && !defined(__clang__)
-inline NLOHMANN_VIEW_NOINLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (p != e && is_digit(*p))
-    {
-        ++p;
-    }
-    return p;
-}
-#else
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (p != e && is_digit(*p))
-    {
-        ++p;
-    }
-    return p;
-}
-#endif
-
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
-{
-    while (e - p >= 16)
-    {
-        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
-        // a digit is above '/' and below ':' (signed compares: a byte with the top bit set is negative, and neither)
-        const __m128i digits = _mm_and_si128(_mm_cmpgt_epi8(in, _mm_set1_epi8('/')), _mm_cmplt_epi8(in, _mm_set1_epi8(':')));
-        const auto not_digit = static_cast<std::uint64_t>(~static_cast<unsigned>(_mm_movemask_epi8(digits)) & 0xFFFFu);
-        if (not_digit != 0)
-        {
-            return p + count_trailing_zeros(not_digit);
-        }
-        p += 16;
-    }
-    return skip_digits_tail(p, e);
-}
-#else
-NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
-{
-    return skip_digits(p, e);
-}
-#endif
-
 /// powers of ten up to 10^19 as integers
 inline std::uint64_t int_pow10(unsigned k) noexcept
 {
@@ -1362,7 +1311,8 @@ struct parse_failure
 };
 
 /// FloatType: the number_float_t of the document, whose overflow parse() rejects
-template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Sentinel>
+/// Store: false for accept(): the grammar, number, string and UTF-8 checks run as for parse(), but no node is written
+template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Sentinel, bool Store>
 class builder
 {
   public:
@@ -1538,11 +1488,87 @@ class builder
         return p;
     }
 
+    /// the string from p, the first backslash, control character, ill-formed UTF-8 byte, or the end, checked and not
+    /// decoded (accept() stores nothing, so the arena is not written): the checks, errors and error positions of
+    /// slow_string(), for everything up to the closing quote
+    NLOHMANN_VIEW_NOINLINE decoded check_string(const unsigned char* p)
+    {
+        const unsigned char* const end = m_end;
+        for (;;)
+        {
+            if (p == end)
+            {
+                return failed(error_code::string_missing_quote, p);
+            }
+            const unsigned char c = *p;
+            if (c == '"')
+            {
+                ++p;
+                return decoded{p, 0, 0};
+            }
+            if (c != '\\')
+            {
+                return failed(c < 0x20 ? error_code::string_control_character : error_code::string_utf8, p);
+            }
+            ++p;
+            if (p == end)
+            {
+                return failed(error_code::string_missing_quote, p);
+            }
+            const unsigned char d = *p++;
+            if (d == 'u')
+            {
+                const int cp = hex4(p); // (moves p past the four digits)
+                if (NLOHMANN_VIEW_UNLIKELY(cp < 0))
+                {
+                    return failed(error_code::string_unicode_hex, p);
+                }
+                if (NLOHMANN_VIEW_UNLIKELY((cp & 0xF800) == 0xD800)) // a surrogate
+                {
+                    if (cp >= 0xDC00)
+                    {
+                        return failed(error_code::string_surrogate_low, p);
+                    }
+                    if (end - p < 2 || p[0] != '\\' || p[1] != 'u')
+                    {
+                        return failed(error_code::string_surrogate_high, p);
+                    }
+                    p += 2;
+                    const int lo = hex4(p); // (moves p past the four digits)
+                    if (lo < 0)
+                    {
+                        return failed(error_code::string_unicode_hex, p);
+                    }
+                    if (lo < 0xDC00 || lo > 0xDFFF)
+                    {
+                        return failed(error_code::string_surrogate_high, p);
+                    }
+                }
+            }
+            else if (!(d == '"' || d == '\\' || d == '/' || d == 'b' || d == 'f' || d == 'n' || d == 'r' || d == 't'))
+            {
+                --p;
+                return failed(error_code::string_escape, p);
+            }
+            if (p != end && *p == '\\')
+            {
+                continue; // (escapes in a row: no run of text between them)
+            }
+            p = scan_string_run(p, end); // the text up to the next backslash, quote, control character, or ill-formed UTF-8
+        }
+    }
+
     /// s points to the start of the string, p to a backslash, a control
     /// character, an ill-formed UTF-8 byte, or the end: decode the string into
     /// the arena
     NLOHMANN_VIEW_NOINLINE decoded slow_string(const unsigned char* s, const unsigned char* p)
     {
+        if (!enabled(Store))
+        {
+            // (accept(): the string is checked, not decoded)
+            static_cast<void>(s);
+            return check_string(p);
+        }
         // single-character escapes; 0: invalid (and 'u', handled separately)
         static const std::array<char, 128> simple_escape =
         {
@@ -1903,10 +1929,13 @@ run_found:
 
         NLOHMANN_VIEW_ALWAYS_INLINE bool run()
         {
-            cold.m_doc.reserve(estimate_nodes(reinterpret_cast<const char*>(b), static_cast<std::size_t>(e - b)));
-            base = cold.m_doc.nodes;
-            out = base;
-            cap = base + cold.m_doc.node_capacity;
+            if (enabled(Store))
+            {
+                cold.m_doc.reserve(estimate_nodes(reinterpret_cast<const char*>(b), static_cast<std::size_t>(e - b)));
+                base = cold.m_doc.nodes;
+                out = base;
+                cap = base + cold.m_doc.node_capacity;
+            }
 
             if (e - p >= 3 && p[0] == 0xEF && p[1] == 0xBB && p[2] == 0xBF)
             {
@@ -2090,7 +2119,10 @@ close_object:
             // an array pays nothing for this)
             if (NLOHMANN_VIEW_UNLIKELY(cur_count >= document_data::index_min_members))
             {
-                cold.note_large_object(cur_idx);
+                if (enabled(Store))
+                {
+                    cold.note_large_object(cur_idx);
+                }
             }
 
 close_container:
@@ -2250,6 +2282,12 @@ indent_done:
         /// an integer's value); two stores on little-endian targets
         NLOHMANN_VIEW_ALWAYS_INLINE node* push_node(node_kind k, std::uint8_t flags, std::uint16_t extra, std::size_t off, std::uint64_t second)
         {
+            if (!enabled(Store))
+            {
+                // accept(): nothing is stored, so what computes the arguments is not run either. base and out stay
+                // null: their difference, which open() takes, is 0.
+                return out;
+            }
             if (NLOHMANN_VIEW_UNLIKELY(out == cap))
             {
                 const auto n = static_cast<std::size_t>(out - base);
@@ -2308,9 +2346,12 @@ indent_done:
 
         NLOHMANN_VIEW_ALWAYS_INLINE void close()
         {
-            node& n = base[cur_idx];
-            n.len = cur_count;
-            n.next = static_cast<std::uint32_t>(out - base) - cur_idx;
+            if (enabled(Store))
+            {
+                node& n = base[cur_idx];
+                n.len = cur_count;
+                n.next = static_cast<std::uint32_t>(out - base) - cur_idx;
+            }
             if (--depth != 0)
             {
                 if (NLOHMANN_VIEW_LIKELY(depth <= 64))
@@ -2375,7 +2416,7 @@ indent_done:
             {
                 ++p;
                 const unsigned char* const f0 = p;
-                p = skip_fraction_digits(p, e);
+                p = skip_digits(p, e);
                 if (NLOHMANN_VIEW_UNLIKELY(p == f0))
                 {
                     return fail(error_code::number_after_dot);
@@ -2485,34 +2526,44 @@ indent_done:
 };
 
 /// run the builder with compile-time options
-template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd>
+template<typename FloatType, bool Comments, bool TrailingCommas, bool NulIsEnd, bool Store>
 inline bool build_with(document_data& d, const char* src, std::size_t size, bool sentinel, parse_failure& failure)
 {
     if (sentinel)
     {
-        builder<FloatType, Comments, TrailingCommas, NulIsEnd, true> bld(d, src, size);
+        builder<FloatType, Comments, TrailingCommas, NulIsEnd, true, Store> bld(d, src, size);
         const bool ok = bld.run();
         failure = bld.failure();
         return ok;
     }
-    builder<FloatType, Comments, TrailingCommas, NulIsEnd, false> bld(d, src, size);
+    builder<FloatType, Comments, TrailingCommas, NulIsEnd, false, Store> bld(d, src, size);
     const bool ok = bld.run();
     failure = bld.failure();
     return ok;
 }
 
 /// sentinel: src[size] is readable and 0 (e.g. std::string); FloatType: the
-/// number_float_t of the document
+/// number_float_t of the document; validate_only: accept(), only the result is
+/// used
 template<typename FloatType, bool NulIsEnd>
-inline bool build(document_data& d, const char* src, std::size_t size, bool comments, bool trailing_commas, bool sentinel, parse_failure& failure)
+inline bool build(document_data& d, const char* src, std::size_t size, bool comments, bool trailing_commas, bool sentinel, bool validate_only,
+                  parse_failure& failure)
 {
+    if (validate_only && !comments && !trailing_commas)
+    {
+        // accept() of plain JSON: nothing is stored. The other option sets
+        // could skip the stores just as well; they keep the builder that
+        // stores only so that there are two more instantiations of the parse
+        // loop, not eight.
+        return build_with<FloatType, false, false, NulIsEnd, false>(d, src, size, sentinel, failure);
+    }
     if (comments)
     {
-        return trailing_commas ? build_with<FloatType, true, true, NulIsEnd>(d, src, size, sentinel, failure)
-               : build_with<FloatType, true, false, NulIsEnd>(d, src, size, sentinel, failure);
+        return trailing_commas ? build_with<FloatType, true, true, NulIsEnd, true>(d, src, size, sentinel, failure)
+               : build_with<FloatType, true, false, NulIsEnd, true>(d, src, size, sentinel, failure);
     }
-    return trailing_commas ? build_with<FloatType, false, true, NulIsEnd>(d, src, size, sentinel, failure)
-           : build_with<FloatType, false, false, NulIsEnd>(d, src, size, sentinel, failure);
+    return trailing_commas ? build_with<FloatType, false, true, NulIsEnd, true>(d, src, size, sentinel, failure)
+           : build_with<FloatType, false, false, NulIsEnd, true>(d, src, size, sentinel, failure);
 }
 
 }  // namespace view
@@ -10152,7 +10203,7 @@ class basic_json_document
                                      const bool ignore_trailing_commas = false)
     {
         basic_json_document d;
-        d.read_range(first, last, allow_exceptions, ignore_comments, ignore_trailing_commas);
+        d.read_range(first, last, error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas);
         return d;
     }
 
@@ -10167,7 +10218,7 @@ class basic_json_document
         basic_json_document d;
         std::string buf;
         append_input(buf, std::forward<InputType>(input));
-        d.build_owned(std::move(buf), allow_exceptions, ignore_comments, ignore_trailing_commas);
+        d.build_owned(std::move(buf), error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas);
         return d;
     }
 
@@ -10180,7 +10231,8 @@ class basic_json_document
     static bool accept(InputType&& input, const bool ignore_comments = false, const bool ignore_trailing_commas = false)
     {
         basic_json_document d;
-        d.read(std::forward<InputType>(input), false, ignore_comments, ignore_trailing_commas);
+        d.read_kind(std::forward<InputType>(input), on_error::validate, ignore_comments, ignore_trailing_commas,
+                    std::integral_constant<detail::view::input_kind, detail::view::classify_input<InputType>::value> {});
         return !d.is_discarded();
     }
 
@@ -10196,7 +10248,7 @@ class basic_json_document
               const bool ignore_comments = false,
               const bool ignore_trailing_commas = false)
     {
-        read_kind(std::forward<InputType>(input), allow_exceptions, ignore_comments, ignore_trailing_commas,
+        read_kind(std::forward<InputType>(input), error_mode(allow_exceptions), ignore_comments, ignore_trailing_commas,
                   std::integral_constant<detail::view::input_kind, detail::view::classify_input<InputType>::value> {});
     }
 
@@ -10555,6 +10607,22 @@ class basic_json_document
         return idx;
     }
 
+    /// what a failed parse does
+    enum class on_error : std::uint8_t
+    {
+        discard, ///< the document is discarded
+        raise,   ///< parse_error is thrown
+        validate ///< accept(): the document is discarded, and no node is stored while parsing
+    };
+
+    static on_error error_mode(bool allow_exceptions) noexcept
+    {
+        return allow_exceptions ? on_error::raise : on_error::discard;
+    }
+
+    /// NulIsEnd of the builder: a null byte ends the input unless strict_nul_handling
+    static constexpr bool nul_is_end = !detail::abi_config::strict_nul_handling;
+
     /// create the storage (sized for the input) on first use
     void ensure_data(const char* src, std::size_t size)
     {
@@ -10566,7 +10634,7 @@ class basic_json_document
 
     /// parse a buffer the document takes ownership of (it replaces the
     /// previous owned copy)
-    void build_owned(std::string&& buf, bool allow_exceptions, bool comments, bool trailing_commas)
+    void build_owned(std::string&& buf, on_error allow_exceptions, bool comments, bool trailing_commas)
     {
         ensure_data(buf.data(), buf.size());
         m_data->owned = std::move(buf);
@@ -10576,7 +10644,7 @@ class basic_json_document
     /// parse a copy of the input; the copy is appended to the cleared owned
     /// buffer, so that its capacity is reused
     template<typename T>
-    void build_copy(T&& input, bool allow_exceptions, bool comments, bool trailing_commas)
+    void build_copy(T&& input, on_error allow_exceptions, bool comments, bool trailing_commas)
     {
         if (!m_data)
         {
@@ -10597,7 +10665,7 @@ class basic_json_document
     }
 
     /// sentinel: src[size] is readable and 0 (std::string, C strings)
-    void build(const char* src, std::size_t size, bool allow_exceptions, bool comments, bool trailing_commas, bool owned, bool sentinel)
+    void build(const char* src, std::size_t size, on_error allow_exceptions, bool comments, bool trailing_commas, bool owned, bool sentinel)
     {
         ensure_data(src, size);
         document_data& d = *m_data;
@@ -10624,7 +10692,8 @@ class basic_json_document
         }
         else
         {
-            ok = detail::view::build < typename BasicJsonType::number_float_t, !detail::abi_config::strict_nul_handling > (d, src, size, comments, trailing_commas, sentinel, failure);
+            ok = detail::view::build<typename BasicJsonType::number_float_t, nul_is_end>(d, src, size, comments, trailing_commas, sentinel,
+                 allow_exceptions == on_error::validate, failure);
         }
         if (NLOHMANN_VIEW_LIKELY(ok))
         {
@@ -10637,7 +10706,7 @@ class basic_json_document
             return;
         }
         std::vector<std::uint32_t>().swap(d.large_objects);
-        if (allow_exceptions)
+        if (allow_exceptions == on_error::raise)
         {
             detail::view::throw_parse_failure<BasicJsonType>(failure, src, size, comments, trailing_commas);
         }
@@ -10645,13 +10714,13 @@ class basic_json_document
 
     // --- input dispatch (see detail::view::input_kind) ---
 
-    void read_kind(std::string&& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::move_string> /*unused*/)
+    void read_kind(std::string&& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::move_string> /*unused*/)
     {
         build_owned(std::move(s), ae, c, tc);
     }
 
     template<typename CharT>
-    void read_kind(CharT* s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::c_string> /*unused*/)
+    void read_kind(CharT* s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::c_string> /*unused*/)
     {
         if (s == nullptr)
         {
@@ -10665,7 +10734,7 @@ class basic_json_document
     }
 
     template<typename Array>
-    void read_kind(Array& a, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::char_array> /*unused*/)
+    void read_kind(Array& a, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::char_array> /*unused*/)
     {
         using CharT = typename std::remove_cv<typename std::remove_extent<Array>::type>::type;
         const std::size_t n = std::extent<Array>::value;
@@ -10676,7 +10745,7 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(const T& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::borrow_range> /*unused*/)
+    void read_kind(const T& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::borrow_range> /*unused*/)
     {
         // std::basic_string guarantees data()[size()] == 0: use it as sentinel
         const auto size = static_cast<std::size_t>(s.size());
@@ -10685,13 +10754,13 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(const T& s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::copy_range> /*unused*/)
+    void read_kind(const T& s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::copy_range> /*unused*/)
     {
         build_copy(s, ae, c, tc);
     }
 
     template<typename CharT>
-    void read_kind(CharT* s, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::wide_c_string> /*unused*/)
+    void read_kind(CharT* s, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::wide_c_string> /*unused*/)
     {
         if (s == nullptr)
         {
@@ -10703,13 +10772,13 @@ class basic_json_document
     }
 
     template<typename T>
-    void read_kind(T&& input, bool ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::adapter> /*unused*/)
+    void read_kind(T&& input, on_error ae, bool c, bool tc, std::integral_constant<input_kind, input_kind::adapter> /*unused*/)
     {
         build_copy(std::forward<T>(input), ae, c, tc);
     }
 
     template<typename IteratorType>
-    void read_range(IteratorType first, IteratorType last, bool ae, bool c, bool tc)
+    void read_range(IteratorType first, IteratorType last, on_error ae, bool c, bool tc)
     {
         using value_type = typename std::remove_cv<typename std::iterator_traits<IteratorType>::value_type>::type;
         // borrow the range where the library's input adapter scans it as one
@@ -10720,14 +10789,14 @@ class basic_json_document
     }
 
     template<typename IteratorType>
-    void read_range_impl(IteratorType first, IteratorType last, bool ae, bool c, bool tc, std::true_type /*contiguous bytes*/)
+    void read_range_impl(IteratorType first, IteratorType last, on_error ae, bool c, bool tc, std::true_type /*contiguous bytes*/)
     {
         const auto size = static_cast<std::size_t>(std::distance(first, last));
         build(size == 0 ? "" : reinterpret_cast<const char*>(&*first), size, ae, c, tc, false, size == 0); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
     }
 
     template<typename IteratorType>
-    void read_range_impl(IteratorType first, IteratorType last, bool ae, bool c, bool tc, std::false_type /*other*/)
+    void read_range_impl(IteratorType first, IteratorType last, on_error ae, bool c, bool tc, std::false_type /*other*/)
     {
         build_copy(detail::view::make_adapter_input(detail::input_adapter(first, last)), ae, c, tc);
     }

tests/src/unit-json_view_builder.cpp

@@ -59,7 +59,8 @@ built build(const std::string& text, bool comments, bool trailing_commas, bool s
         r.copy.assign(text.begin(), text.end());
         src = r.copy.data();
     }
-    r.ok = nlohmann::detail::view::build < FloatType, !nlohmann::detail::abi_config::strict_nul_handling > (*r.data, src, text.size(), comments, trailing_commas, sentinel, r.failure);
+    constexpr bool nul_is_end = !nlohmann::detail::abi_config::strict_nul_handling;
+    r.ok = nlohmann::detail::view::build<FloatType, nul_is_end>(*r.data, src, text.size(), comments, trailing_commas, sentinel, false, r.failure);
     r.data->src = src;
     r.data->base[0] = src;
     r.data->base[1] = r.data->arena.data();
@@ -172,6 +173,22 @@ void check_string(const std::string& content)
     }
 }
 
+// accept() stores no nodes: it must accept and reject what json::accept does, for all options and for inputs with
+// and without a terminating NUL
+void check_accept(const std::string& text)
+{
+    CAPTURE(text)
+    const std::vector<char> bytes(text.begin(), text.end());
+    for (int options = 0; options < 4; ++options)
+    {
+        const bool comments = (options & 1) != 0;
+        const bool trailing_commas = (options & 2) != 0;
+        const bool accepted = json::accept(text, comments, trailing_commas);
+        CHECK(nlohmann::json_document::accept(text, comments, trailing_commas) == accepted);
+        CHECK(nlohmann::json_document::accept(bytes, comments, trailing_commas) == accepted);
+    }
+}
+
 // a small deterministic generator of documents
 struct generator
 {
@@ -423,6 +440,133 @@ TEST_CASE("json_view builder")
         }
     }
 
+    SECTION("accept() without an index")
+    {
+        for (const char* text :
+                {
+                    "", " ", "\n", "null", "nul", "true", "[", "]", "[]", "{}", "[1,2,3]", "[1,2,3", "[1,2,3,]", "{\"a\":1,}", "{\"a\"}", "{\"a\":}",
+                    "[1] x", "[1][2]", "\xEF\xBB\xBF[1]", "[\"a\\", "[\"\\u12\"]", "[\"\\ud800\"]", "[\"\x01\"]", "[\"\xff\"]", "[01]", "[1.]",
+                    "[.5]", "[-]", "[1e]", "// c\n[1]", "/* c */ {\"a\" /* x */ : 1}", "/* c", "[1] // c", "1e400", "-1e400", "123456789012345678901234567890",
+                    "{\"a\":1,\"a\":2}", "[[[]]]", "\"\\u00e4\\n\\ud83d\\ude00\""
+                })
+        {
+            check_accept(text);
+        }
+        // a null byte inside the text, and after it
+        check_accept(std::string("[1,\0]", 5));
+        check_accept(std::string("[1]\0", 4));
+        check_accept(std::string("\0", 1));
+
+        // generated documents and damaged copies, as for the builder above
+        generator g;
+        for (int i = 0; i < 2000; ++i)
+        {
+            std::string text;
+            g.value(text, 0);
+            check_accept(text);
+            std::string damaged = text;
+            const auto at = static_cast<std::size_t>(g.r(static_cast<int>(damaged.size())));
+            static const char replacements[] = {'x', '"', '\\', ',', ':', ']', '}', '[', '{', '1', '-', '.', 'e', '\0', '\n', '/'}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+            damaged[at] = replacements[g.r(16)];
+            check_accept(damaged);
+            check_accept(text.substr(0, at));
+        }
+
+        // nesting beyond the 64 levels that are kept on the builder's own stack, closed and not closed
+        for (const std::size_t depth :
+                {
+                    63, 64, 65, 66, 200, 1000
+                })
+        {
+            check_accept(std::string(depth, '[') + std::string(depth, ']'));
+            check_accept(std::string(depth, '[') + std::string(depth - 1, ']'));
+            std::string objects;
+            for (std::size_t i = 0; i < depth; ++i)
+            {
+                objects += "{\"k\":";
+            }
+            objects += "1" + std::string(depth, '}');
+            check_accept(objects);
+            check_accept(objects.substr(0, objects.size() - 1));
+        }
+
+        // objects with enough members to get a hash index when they are indexed (and duplicate keys)
+        std::string large = "{";
+        for (int i = 0; i < 300; ++i)
+        {
+            large += (i == 0 ? "\"k" : ",\"k") + std::to_string(i % 200) + "\":" + std::to_string(i);
+        }
+        check_accept(large + "}");
+        check_accept(large);
+        check_accept("[" + large + "},{" + large.substr(1) + "}]");
+
+        // the overflow check uses the floating-point type of the document
+        using float_json = nlohmann::basic_json<std::map, std::vector, std::string, bool, std::int64_t, std::uint64_t, float>;
+        for (const char* text :
+                {
+                    "1e39", "-1e39", "3.4028235e38", "3.4028236e38", "[1.5e38, 3.5e38]", "{\"a\": 1e-50, \"b\": 1e39}", "340282356779733661637539395458142568448"
+                })
+        {
+            CAPTURE(text)
+            CHECK(nlohmann::basic_json_document<float_json>::accept(text) == float_json::accept(text));
+        }
+
+        // strings are checked and not decoded: runs of text of every length from 0 to 40 between valid and invalid
+        // escapes, UTF-8 sequences and control characters, in values and in keys, cut near the end of the text
+        const std::array<const char*, 9> valid = {{"\\n", "\\\"", "\\\\", "\\/", "\\t", "\\u00e4", "\\u20ac", "\\ud83d\\ude00", "\\u0000"}};
+        const std::array<const char*, 12> invalid = {{"\\x", "\\ud800", "\\udc00", "\\u12G4", "\\u12", "\\ud83dx", "\\ud83d\\u0041", "\\ud83d\\ud83d", "\\", "\x01", "\x1f", "\x7f"}};
+        const std::array<const char*, 6> utf8 = {{"\xc3\xa4", "\xe2\x82\xac", "\xf0\x9f\x98\x80", "\xc3", "\xff", "\x80"}};
+        std::mt19937 rng(2029); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed): reproducible
+        for (std::size_t run = 0; run <= 40; ++run)
+        {
+            for (int i = 0; i < 30; ++i)
+            {
+                std::string content = std::string(rng() % 20, 'a');
+                const auto pieces = 1 + (rng() % 6);
+                for (unsigned k = 0; k < pieces; ++k)
+                {
+                    content += valid[rng() % valid.size()];
+                    content += std::string(k == 0 ? run : rng() % 41, static_cast<char>('b' + (rng() % 20)));
+                    switch (rng() % 8)
+                    {
+                        case 0:
+                            content += utf8[rng() % 3];
+                            break;
+                        case 1:
+                            content += utf8[rng() % utf8.size()];
+                            break;
+                        case 2:
+                            content += invalid[rng() % invalid.size()];
+                            break;
+                        default:
+                            break;
+                    }
+                }
+                for (const std::string& text :
+                        {
+                            "[\"" + content + "\"]", "{\"" + content + "\":1}", "[\"" + content, "[\"" + content.substr(0, content.size() - (rng() % 4)) + "\"]"
+                        })
+                {
+                    check_accept(text);
+                }
+            }
+        }
+    }
+
+    SECTION("escapes in a row")
+    {
+        for (const char* content :
+                {
+                    "\\n", "\\n\\t\\\\", "\\u00e9\\u00e9\\u4e2d", "\\u00e9\\u00e", "\\u00e9\\", "\\n\\q", "\\n\\ud83d\\ude00\\n",
+                    "\\ud83d\\ude00\\ud83d", "\\n\\n\\n abc \\n\\n", "\\n\x01", "\\n\xC3\x28", "\\n\\"
+                })
+        {
+            check_accept(std::string("[\"") + content + "\"]");
+            check_accept(std::string("[\"") + content);
+            check_accept(std::string("{\"") + content + "\":1}");
+        }
+    }
+
     SECTION("test files")
     {
         for (const char* name :
@@ -523,47 +667,6 @@ TEST_CASE("json_view builder: strings across vector blocks")
     }
 }
 
-TEST_CASE("json_view builder: fraction digits")
-{
-    // The digits of a fraction are skipped 16 bytes at a time where there are
-    // that many left (SSE2), and one at a time after that. Fractions of every
-    // length from 0 to 40 digits, followed by every kind of byte that ends a
-    // number or lies next to the digits in ASCII ('/' and ':'), as values, as
-    // keys' neighbours, and at the very end of the text (a vector without a
-    // terminating NUL). The length of the number is part of the value that
-    // the test compares with json::parse().
-    const std::string followers = ",]}/:.eE+-x \n\t\"09";
-    std::mt19937 rng(2028); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed): reproducible
-    for (std::size_t digits = 0; digits <= 40; ++digits)
-    {
-        for (int i = 0; i < 3; ++i)
-        {
-            std::string fraction;
-            for (std::size_t k = 0; k < digits; ++k)
-            {
-                fraction += static_cast<char>('0' + (rng() % 10));
-            }
-            const std::string number = std::string(rng() % 2 == 0 ? "-" : "") + std::to_string(rng() % 100000) + "." + fraction;
-            check_same(number);
-            check_same("[" + number + "]");
-            check_same("[" + number + "," + number + "]");
-            check_same("{\"a\":" + number + "}");
-            for (const char follower : followers)
-            {
-                check_same(number + follower);
-                check_same("[" + number + follower + "]");
-                check_same("[" + number + follower + std::string(rng() % 20, '1'));
-            }
-            // everything near the end of the text: the number cut in every place
-            for (std::size_t cut = 1; cut <= 18 && cut <= number.size(); ++cut)
-            {
-                check_same(number.substr(0, number.size() - cut));
-                check_same("[" + number.substr(0, number.size() - cut));
-            }
-        }
-    }
-}
-
 TEST_CASE("json_view builder: escapes between runs of text")
 {
     // The text between escapes is copied 16 bytes at a time where there are