BSON writer: write to_bson(j) in one pass, with sizes written afterwards

BSON prefixes every document and array with its size. to_bson() therefore
walked the value twice: once to compute every size, once to write. When
the output is the vector that to_bson(j) returns, the writer now puts a
placeholder in front of each document and array and writes its size there
once the document is complete, so one walk is enough.

Every key and value is still checked as the size pass checked it, in the
same order, so the same exception is thrown; the partial output is
discarded with the vector. If a length would not fit BSON's 32-bit size
fields, the writer drops what it has written and writes the document in
two passes, which reports that error before writing anything, as before.

The two passes remain for output adapters (streams, strings, vectors
passed by reference), for error_handler_t::replace and ignore (which
change sizes), and for integer types wider than 64 bits.

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

include/nlohmann/detail/output/binary_writer.hpp

@@ -2088,7 +2088,8 @@ class binary_writer
         /// arrays: the first index with one more digit; only maintained by
         /// @ref calc_bson_sizes
         std::uint64_t index_boundary = 10;
-        /// @ref calc_bson_sizes only: where its size goes in the table
+        /// @ref calc_bson_sizes: where its size goes in the table; @ref
+        /// write_bson_document_single_pass: the position of its size field
         std::size_t size_slot;
         /// @ref calc_bson_sizes only: the size of its entries seen so far
         std::size_t entries_size = 0;
@@ -2339,6 +2340,197 @@ class binary_writer
         }
     }
 
+    template<typename SinkType>
+    using sink_written_t = decltype(std::declval<const SinkType&>().written());
+
+    /*!
+    @brief whether BSON documents can be written in a single pass: the sink
+           can overwrite what it has written (output_vector_sink), and every
+           integer fits the 64 bits BSON has, so that writing an array of
+           numbers needs no range check
+    */
+    using bson_single_pass_sink = std::integral_constant < bool,
+          is_detected<sink_written_t, OutputSinkType>::value
+          && std::numeric_limits<typename BasicJsonType::number_integer_t>::is_specialized
+          && std::numeric_limits<typename BasicJsonType::number_integer_t>::digits <= std::numeric_limits<std::int64_t>::digits
+          && std::numeric_limits<typename BasicJsonType::number_unsigned_t>::is_specialized
+          && std::numeric_limits<typename BasicJsonType::number_unsigned_t>::digits <= std::numeric_limits<std::uint64_t>::digits >;
+
+    /// write a placeholder for the size of an object or array whose entries
+    /// follow; @return its position, for @ref patch_bson_size
+    std::size_t write_bson_size_placeholder()
+    {
+        const std::size_t pos = oa.written();
+        write_number<std::int32_t>(0, true);
+        return pos;
+    }
+
+    /// end the object or array whose size is at @a pos and write its size
+    /// there: everything from that field to the terminating null byte
+    /// @return false (and the size is not written) if it does not fit
+    bool close_bson_container(const std::size_t pos)
+    {
+        oa.write_character(to_char_type(0x00));
+        const std::size_t size = oa.written() - pos;
+        if (JSON_HEDLEY_UNLIKELY(!value_in_range_of<std::int32_t>(size)))
+        {
+            return false;
+        }
+        std::array<CharType, sizeof(std::int32_t)> bytes{};
+        store_little_endian<std::int32_t>(bytes.data(), static_cast<std::int32_t>(size));
+        oa.overwrite(pos, bytes.data(), bytes.size());
+        return true;
+    }
+
+    /*!
+    @brief check the value @a j of a BSON entry as @ref calc_bson_value_size
+           does, for @ref write_bson_document_single_pass
+    @return at least the size of the value
+    A number, boolean, or null needs no check there (its integers have at most
+    64 bits) and has at most 8 characters; a discarded value is rejected with
+    the same exception when it is written.
+    */
+    std::size_t check_bson_value(const BasicJsonType& j)
+    {
+        return (j.is_string() || j.is_binary()) ? calc_bson_value_size(j) : 8;
+    }
+
+    /// whether an entry of @a entry_size characters keeps the document that
+    /// begins at @a start, closed by @a open_containers null bytes, within
+    /// the 32-bit size of BSON
+    bool bson_entry_fits(const std::size_t start, const std::size_t entry_size, const std::size_t open_containers) const noexcept
+    {
+        const std::size_t limit = static_cast<std::size_t>((std::numeric_limits<std::int32_t>::max)());
+        const std::size_t written = oa.written() - start;
+        return written <= limit && entry_size <= limit - written && open_containers <= limit - written - entry_size;
+    }
+
+    /*!
+    @brief Serializes the JSON object @a document as a BSON document in a
+           single pass, writing the size of each object and array once it is
+           complete
+
+    Checks every key and value as @ref calc_bson_sizes does, in the same
+    order, so the same exception is thrown; the output written up to then is
+    discarded with the vector it was written to.
+
+    @return false as soon as the document or a string, binary value, object,
+            or array in it would be larger than the 32-bit BSON sizes allow;
+            nothing large has been written then, and the caller writes the
+            document in two passes, which reports that error
+    */
+    template < typename SinkType = OutputSinkType,
+               typename std::enable_if < is_detected<sink_written_t, SinkType>::value, int >::type = 0 >
+    bool write_bson_document_single_pass(const BasicJsonType& document)
+    {
+        const std::size_t start = oa.written();
+        // the object or array whose entries are being written (with the
+        // position of its size), and the ones it is in
+        bson_frame current(&document, write_bson_size_placeholder());
+        std::vector<bson_frame> parents;
+
+        while (true)
+        {
+            const BasicJsonType* nested = nullptr;
+            std::size_t nested_size_pos = 0;
+            if (current.value->is_object())
+            {
+                const auto& object = *current.value->m_data.m_value.object;
+                while (nested == nullptr && current.member != object.cend())
+                {
+                    const auto& el = *current.member;
+                    ++current.member;
+                    const std::size_t header_size = calc_bson_entry_header_size(el.first, el.second);
+                    const std::size_t value_size = el.second.is_structured() ? 0 : check_bson_value(el.second);
+                    if (JSON_HEDLEY_UNLIKELY(!bson_entry_fits(start, header_size + value_size, parents.size() + 1)))
+                    {
+                        return false;
+                    }
+                    if (el.second.is_structured())
+                    {
+                        write_bson_entry_header(el.first, el.second.is_object() ? 0x03 : 0x04);
+                        nested_size_pos = write_bson_size_placeholder();
+                        nested = &el.second;
+                    }
+                    else
+                    {
+                        write_bson_value(el.first, el.second);
+                    }
+                }
+            }
+            else
+            {
+                const auto& array = *current.value->m_data.m_value.array;
+                while (nested == nullptr && current.index < array.size())
+                {
+                    if (is_bson_fixed_size(array[current.index]))
+                    {
+                        write_bson_array_fixed_entries(current);
+                        if (current.index == array.size())
+                        {
+                            break;
+                        }
+                    }
+
+                    // a string, binary value, object or array
+                    const BasicJsonType& el = array[current.index];
+                    const string_t index_name(current.index_name.data(), current.index_name_size);
+                    current.next_index();
+                    const std::size_t value_size = el.is_structured() ? 0 : check_bson_value(el);
+                    if (JSON_HEDLEY_UNLIKELY(!bson_entry_fits(start, 2 + index_name.size() + value_size, parents.size() + 1)))
+                    {
+                        return false;
+                    }
+                    if (el.is_structured())
+                    {
+                        write_bson_entry_header(index_name, el.is_object() ? 0x03 : 0x04);
+                        nested_size_pos = write_bson_size_placeholder();
+                        nested = &el;
+                    }
+                    else
+                    {
+                        write_bson_value(index_name, el);
+                    }
+                }
+            }
+
+            if (nested != nullptr)
+            {
+                if (nested->is_array())
+                {
+                    // an array of booleans, numbers and null is written without a frame on the stack
+                    bson_frame leaf(nested, nested_size_pos);
+                    write_bson_array_fixed_entries(leaf);
+                    if (leaf.index == nested->m_data.m_value.array->size())
+                    {
+                        if (JSON_HEDLEY_UNLIKELY(!close_bson_container(nested_size_pos)))
+                        {
+                            return false;
+                        }
+                        continue;
+                    }
+                    parents.push_back(std::move(current));
+                    current = std::move(leaf);
+                    continue;
+                }
+                parents.push_back(std::move(current));
+                current = bson_frame(nested, nested_size_pos);
+                continue;
+            }
+
+            if (JSON_HEDLEY_UNLIKELY(!close_bson_container(current.size_slot)))
+            {
+                return false;
+            }
+            if (parents.empty())
+            {
+                return true;
+            }
+            current = std::move(parents.back());
+            parents.pop_back();
+        }
+    }
+
     /*!
     @brief Serializes the JSON object @a document as a BSON document
 
@@ -2350,6 +2542,39 @@ class binary_writer
     @pre       document.type() == value_t::object
     */
     void write_bson_document(const BasicJsonType& document)
+    {
+        write_bson_document(document, bson_single_pass_sink());
+    }
+
+    /// the sink can overwrite: one pass, unless the error handler replaces or
+    /// drops ill-formed UTF-8, which changes the sizes the checks compute
+    void write_bson_document(const BasicJsonType& document, std::true_type /*single_pass*/)
+    {
+        const std::size_t start = oa.written();
+        if ((error_handler == error_handler_t::keep || error_handler == error_handler_t::strict)
+                && write_bson_document_single_pass(document))
+        {
+            return;
+        }
+        // a length does not fit: start over, so that it is reported as the
+        // two passes report it, before anything is written
+        oa.rewind(start);
+        write_bson_document_two_pass(document);
+    }
+
+    void write_bson_document(const BasicJsonType& document, std::false_type /*single_pass*/)
+    {
+        write_bson_document_two_pass(document);
+    }
+
+    /*!
+    @brief Serializes the JSON object @a document as a BSON document: the
+           sizes of all objects and arrays first, then the entries
+
+    @param[in] document  the JSON object to serialize
+    @pre       document.type() == value_t::object
+    */
+    void write_bson_document_two_pass(const BasicJsonType& document)
     {
         std::vector<std::size_t> nested_sizes;
         const std::size_t document_size = calc_bson_sizes(document, nested_sizes);

include/nlohmann/detail/output/output_adapters.hpp

@@ -227,6 +227,29 @@ class output_vector_sink
     void flush() noexcept
     {}
 
+    /// the number of characters written so far, the position of the next one
+    std::size_t written() const noexcept
+    {
+        return position();
+    }
+
+    /// @brief replace @a length characters written before, from position @a pos
+    ///
+    /// Lets a writer put a length in front of data it has written since, such
+    /// as the size of a BSON document, without computing it in advance.
+    void overwrite(const std::size_t pos, const CharType* s, const std::size_t length) noexcept
+    {
+        JSON_ASSERT(pos + length <= position());
+        std::memcpy(v->data() + pos, s, length * sizeof(CharType));
+    }
+
+    /// drop the characters written from position @a pos on
+    void rewind(const std::size_t pos) noexcept
+    {
+        JSON_ASSERT(pos <= position());
+        cursor = v->data() + pos;
+    }
+
   private:
     /// number of characters of *v written so far
     std::size_t position() const noexcept

include/nlohmann/detail/view/builder.hpp

@@ -1082,7 +1082,7 @@ 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);

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
 {

single_include/nlohmann/json.hpp

@@ -26870,6 +26870,29 @@ class output_vector_sink
     void flush() noexcept
     {}
 
+    /// the number of characters written so far, the position of the next one
+    std::size_t written() const noexcept
+    {
+        return position();
+    }
+
+    /// @brief replace @a length characters written before, from position @a pos
+    ///
+    /// Lets a writer put a length in front of data it has written since, such
+    /// as the size of a BSON document, without computing it in advance.
+    void overwrite(const std::size_t pos, const CharType* s, const std::size_t length) noexcept
+    {
+        JSON_ASSERT(pos + length <= position());
+        std::memcpy(v->data() + pos, s, length * sizeof(CharType));
+    }
+
+    /// drop the characters written from position @a pos on
+    void rewind(const std::size_t pos) noexcept
+    {
+        JSON_ASSERT(pos <= position());
+        cursor = v->data() + pos;
+    }
+
   private:
     /// number of characters of *v written so far
     std::size_t position() const noexcept
@@ -29314,7 +29337,8 @@ class binary_writer
         /// arrays: the first index with one more digit; only maintained by
         /// @ref calc_bson_sizes
         std::uint64_t index_boundary = 10;
-        /// @ref calc_bson_sizes only: where its size goes in the table
+        /// @ref calc_bson_sizes: where its size goes in the table; @ref
+        /// write_bson_document_single_pass: the position of its size field
         std::size_t size_slot;
         /// @ref calc_bson_sizes only: the size of its entries seen so far
         std::size_t entries_size = 0;
@@ -29565,6 +29589,197 @@ class binary_writer
         }
     }
 
+    template<typename SinkType>
+    using sink_written_t = decltype(std::declval<const SinkType&>().written());
+
+    /*!
+    @brief whether BSON documents can be written in a single pass: the sink
+           can overwrite what it has written (output_vector_sink), and every
+           integer fits the 64 bits BSON has, so that writing an array of
+           numbers needs no range check
+    */
+    using bson_single_pass_sink = std::integral_constant < bool,
+          is_detected<sink_written_t, OutputSinkType>::value
+          && std::numeric_limits<typename BasicJsonType::number_integer_t>::is_specialized
+          && std::numeric_limits<typename BasicJsonType::number_integer_t>::digits <= std::numeric_limits<std::int64_t>::digits
+          && std::numeric_limits<typename BasicJsonType::number_unsigned_t>::is_specialized
+          && std::numeric_limits<typename BasicJsonType::number_unsigned_t>::digits <= std::numeric_limits<std::uint64_t>::digits >;
+
+    /// write a placeholder for the size of an object or array whose entries
+    /// follow; @return its position, for @ref patch_bson_size
+    std::size_t write_bson_size_placeholder()
+    {
+        const std::size_t pos = oa.written();
+        write_number<std::int32_t>(0, true);
+        return pos;
+    }
+
+    /// end the object or array whose size is at @a pos and write its size
+    /// there: everything from that field to the terminating null byte
+    /// @return false (and the size is not written) if it does not fit
+    bool close_bson_container(const std::size_t pos)
+    {
+        oa.write_character(to_char_type(0x00));
+        const std::size_t size = oa.written() - pos;
+        if (JSON_HEDLEY_UNLIKELY(!value_in_range_of<std::int32_t>(size)))
+        {
+            return false;
+        }
+        std::array<CharType, sizeof(std::int32_t)> bytes{};
+        store_little_endian<std::int32_t>(bytes.data(), static_cast<std::int32_t>(size));
+        oa.overwrite(pos, bytes.data(), bytes.size());
+        return true;
+    }
+
+    /*!
+    @brief check the value @a j of a BSON entry as @ref calc_bson_value_size
+           does, for @ref write_bson_document_single_pass
+    @return at least the size of the value
+    A number, boolean, or null needs no check there (its integers have at most
+    64 bits) and has at most 8 characters; a discarded value is rejected with
+    the same exception when it is written.
+    */
+    std::size_t check_bson_value(const BasicJsonType& j)
+    {
+        return (j.is_string() || j.is_binary()) ? calc_bson_value_size(j) : 8;
+    }
+
+    /// whether an entry of @a entry_size characters keeps the document that
+    /// begins at @a start, closed by @a open_containers null bytes, within
+    /// the 32-bit size of BSON
+    bool bson_entry_fits(const std::size_t start, const std::size_t entry_size, const std::size_t open_containers) const noexcept
+    {
+        const std::size_t limit = static_cast<std::size_t>((std::numeric_limits<std::int32_t>::max)());
+        const std::size_t written = oa.written() - start;
+        return written <= limit && entry_size <= limit - written && open_containers <= limit - written - entry_size;
+    }
+
+    /*!
+    @brief Serializes the JSON object @a document as a BSON document in a
+           single pass, writing the size of each object and array once it is
+           complete
+
+    Checks every key and value as @ref calc_bson_sizes does, in the same
+    order, so the same exception is thrown; the output written up to then is
+    discarded with the vector it was written to.
+
+    @return false as soon as the document or a string, binary value, object,
+            or array in it would be larger than the 32-bit BSON sizes allow;
+            nothing large has been written then, and the caller writes the
+            document in two passes, which reports that error
+    */
+    template < typename SinkType = OutputSinkType,
+               typename std::enable_if < is_detected<sink_written_t, SinkType>::value, int >::type = 0 >
+    bool write_bson_document_single_pass(const BasicJsonType& document)
+    {
+        const std::size_t start = oa.written();
+        // the object or array whose entries are being written (with the
+        // position of its size), and the ones it is in
+        bson_frame current(&document, write_bson_size_placeholder());
+        std::vector<bson_frame> parents;
+
+        while (true)
+        {
+            const BasicJsonType* nested = nullptr;
+            std::size_t nested_size_pos = 0;
+            if (current.value->is_object())
+            {
+                const auto& object = *current.value->m_data.m_value.object;
+                while (nested == nullptr && current.member != object.cend())
+                {
+                    const auto& el = *current.member;
+                    ++current.member;
+                    const std::size_t header_size = calc_bson_entry_header_size(el.first, el.second);
+                    const std::size_t value_size = el.second.is_structured() ? 0 : check_bson_value(el.second);
+                    if (JSON_HEDLEY_UNLIKELY(!bson_entry_fits(start, header_size + value_size, parents.size() + 1)))
+                    {
+                        return false;
+                    }
+                    if (el.second.is_structured())
+                    {
+                        write_bson_entry_header(el.first, el.second.is_object() ? 0x03 : 0x04);
+                        nested_size_pos = write_bson_size_placeholder();
+                        nested = &el.second;
+                    }
+                    else
+                    {
+                        write_bson_value(el.first, el.second);
+                    }
+                }
+            }
+            else
+            {
+                const auto& array = *current.value->m_data.m_value.array;
+                while (nested == nullptr && current.index < array.size())
+                {
+                    if (is_bson_fixed_size(array[current.index]))
+                    {
+                        write_bson_array_fixed_entries(current);
+                        if (current.index == array.size())
+                        {
+                            break;
+                        }
+                    }
+
+                    // a string, binary value, object or array
+                    const BasicJsonType& el = array[current.index];
+                    const string_t index_name(current.index_name.data(), current.index_name_size);
+                    current.next_index();
+                    const std::size_t value_size = el.is_structured() ? 0 : check_bson_value(el);
+                    if (JSON_HEDLEY_UNLIKELY(!bson_entry_fits(start, 2 + index_name.size() + value_size, parents.size() + 1)))
+                    {
+                        return false;
+                    }
+                    if (el.is_structured())
+                    {
+                        write_bson_entry_header(index_name, el.is_object() ? 0x03 : 0x04);
+                        nested_size_pos = write_bson_size_placeholder();
+                        nested = &el;
+                    }
+                    else
+                    {
+                        write_bson_value(index_name, el);
+                    }
+                }
+            }
+
+            if (nested != nullptr)
+            {
+                if (nested->is_array())
+                {
+                    // an array of booleans, numbers and null is written without a frame on the stack
+                    bson_frame leaf(nested, nested_size_pos);
+                    write_bson_array_fixed_entries(leaf);
+                    if (leaf.index == nested->m_data.m_value.array->size())
+                    {
+                        if (JSON_HEDLEY_UNLIKELY(!close_bson_container(nested_size_pos)))
+                        {
+                            return false;
+                        }
+                        continue;
+                    }
+                    parents.push_back(std::move(current));
+                    current = std::move(leaf);
+                    continue;
+                }
+                parents.push_back(std::move(current));
+                current = bson_frame(nested, nested_size_pos);
+                continue;
+            }
+
+            if (JSON_HEDLEY_UNLIKELY(!close_bson_container(current.size_slot)))
+            {
+                return false;
+            }
+            if (parents.empty())
+            {
+                return true;
+            }
+            current = std::move(parents.back());
+            parents.pop_back();
+        }
+    }
+
     /*!
     @brief Serializes the JSON object @a document as a BSON document
 
@@ -29576,6 +29791,39 @@ class binary_writer
     @pre       document.type() == value_t::object
     */
     void write_bson_document(const BasicJsonType& document)
+    {
+        write_bson_document(document, bson_single_pass_sink());
+    }
+
+    /// the sink can overwrite: one pass, unless the error handler replaces or
+    /// drops ill-formed UTF-8, which changes the sizes the checks compute
+    void write_bson_document(const BasicJsonType& document, std::true_type /*single_pass*/)
+    {
+        const std::size_t start = oa.written();
+        if ((error_handler == error_handler_t::keep || error_handler == error_handler_t::strict)
+                && write_bson_document_single_pass(document))
+        {
+            return;
+        }
+        // a length does not fit: start over, so that it is reported as the
+        // two passes report it, before anything is written
+        oa.rewind(start);
+        write_bson_document_two_pass(document);
+    }
+
+    void write_bson_document(const BasicJsonType& document, std::false_type /*single_pass*/)
+    {
+        write_bson_document_two_pass(document);
+    }
+
+    /*!
+    @brief Serializes the JSON object @a document as a BSON document: the
+           sizes of all objects and arrays first, then the entries
+
+    @param[in] document  the JSON object to serialize
+    @pre       document.type() == value_t::object
+    */
+    void write_bson_document_two_pass(const BasicJsonType& document)
     {
         std::vector<std::size_t> nested_sizes;
         const std::size_t document_size = calc_bson_sizes(document, nested_sizes);

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
 {
@@ -2375,7 +2324,7 @@ 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);

tests/src/unit-bson.cpp

@@ -2267,3 +2267,132 @@ TEST_CASE("BSON large strings and binaries (chunked reader)")
         }
     }
 }
+
+TEST_CASE("BSON written in one pass and in two passes")
+{
+    // to_bson() into a returned vector writes each size once the object or
+    // array is complete; through an output adapter, all sizes are computed
+    // first. Both must produce the same bytes and throw the same exceptions.
+    const auto two_pass = [](const json & j, const json::error_handler_t handler)
+    {
+        std::vector<std::uint8_t> result;
+        json::to_bson(j, result, handler);
+        return result;
+    };
+
+    const auto check_same = [&](const json & j, const json::error_handler_t handler)
+    {
+        CAPTURE(j.dump(-1, ' ', true, json::error_handler_t::replace))
+        CAPTURE(static_cast<int>(handler))
+#if !defined(JSON_NOEXCEPTION)
+        std::string one_pass_error;
+        std::string two_pass_error;
+        std::vector<std::uint8_t> one;
+        std::vector<std::uint8_t> two;
+        try
+        {
+            one = json::to_bson(j, handler);
+        }
+        catch (const json::exception& e)
+        {
+            one_pass_error = e.what();
+        }
+        try
+        {
+            two = two_pass(j, handler);
+        }
+        catch (const json::exception& e)
+        {
+            two_pass_error = e.what();
+        }
+        CHECK(one_pass_error == two_pass_error);
+        CHECK(one == two);
+#else
+        CHECK(json::to_bson(j, handler) == two_pass(j, handler));
+#endif
+    };
+
+    const json::error_handler_t handlers[] = {json::error_handler_t::keep, json::error_handler_t::strict,
+                                              json::error_handler_t::replace, json::error_handler_t::ignore
+                                             };
+
+    SECTION("corpus")
+    {
+        for (const auto& value : utils::round_trip_corpus::values())
+        {
+            const json document = value.is_object() ? value : json{{"value", value}};
+            for (const auto handler : handlers)
+            {
+                check_same(document, handler);
+            }
+        }
+    }
+
+    SECTION("nesting and arrays of every length class")
+    {
+        json document = json::object();
+        document["empty object"] = json::object();
+        document["empty array"] = json::array();
+        for (const std::size_t n :
+                {
+                    std::size_t(1), std::size_t(9), std::size_t(10), std::size_t(11), std::size_t(99), std::size_t(100), std::size_t(101), std::size_t(1000)
+                })
+        {
+            json numbers = json::array();
+            json mixed = json::array();
+            for (std::size_t i = 0; i < n; ++i)
+            {
+                numbers.push_back(i % 3 == 0 ? json(static_cast<double>(i) / 3) : json(static_cast<std::int64_t>(i) * 1000003));
+                mixed.push_back(i % 4 == 0 ? json("s" + std::to_string(i)) : i % 4 == 1 ? json::array({i, json::object({{"k", i}})}) : i % 4 == 2 ? json(nullptr) : json(i % 2 == 0));
+            }
+            document["numbers " + std::to_string(n)] = numbers;
+            document["mixed " + std::to_string(n)] = mixed;
+            document["nested " + std::to_string(n)] = json::object({{"a", numbers}, {"b", json::array({mixed, json::object()})}});
+        }
+        document["binary"] = json::binary({1, 2, 3}, 0x80);
+        document["binary old"] = json::binary({1, 2, 3}, 0x02);
+        for (const auto handler : handlers)
+        {
+            check_same(document, handler);
+        }
+    }
+
+    SECTION("errors at every position")
+    {
+        // an entry that cannot be written, at each place among valid ones
+        const std::string bad_utf8 = "\xC3\x28";
+        const json bad_entries[] =
+        {
+            json::object({{std::string("a\0b", 3), 1}}),
+            json::object({{"s", bad_utf8}}),
+            json::object({{bad_utf8, 1}}),
+            json::array({1, 2, bad_utf8}),
+            json::object({{"b", json::binary({1}, 0x100)}}),
+            json::object({{"d", json(json::value_t::discarded)}}),
+            json::array({json(json::value_t::discarded)}),
+        };
+        for (const auto& bad : bad_entries)
+        {
+            for (std::size_t pos = 0; pos < 4; ++pos)
+            {
+                json document = json::object();
+                json list = json::array();
+                for (std::size_t i = 0; i < 4; ++i)
+                {
+                    if (i == pos)
+                    {
+                        document["x" + std::to_string(i)] = bad;
+                        list.push_back(bad);
+                    }
+                    document["n" + std::to_string(i)] = json::array({i, "str", json::object({{"k", i}})});
+                    list.push_back(i);
+                }
+                document["list"] = list;
+                for (const auto handler : handlers)
+                {
+                    check_same(document, handler);
+                }
+            }
+        }
+    }
+}

tests/src/unit-json_view_builder.cpp

@@ -523,51 +523,6 @@ 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 + "]");
-            std::string pair = "[" + number;
-            pair += ",";
-            pair += number;
-            pair += "]";
-            check_same(pair);
-            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