Speed up the lexer: own float parser, string scan, and \u table

README.md

@@ -1393,7 +1393,7 @@ THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR I
 - The class contains a slightly modified version of the Grisu2 algorithm from Florian Loitsch which is licensed under the [MIT License](https://opensource.org/licenses/MIT) (see above). Copyright © 2009 [Florian Loitsch](https://florian.loitsch.com/)
 - The class contains a copy of [Hedley](https://nemequ.github.io/hedley/) from Evan Nemerson which is licensed as [CC0-1.0](https://creativecommons.org/publicdomain/zero/1.0/).
 - The class contains parts of [Google Abseil](https://github.com/abseil/abseil-cpp) which is licensed under the [Apache 2.0 License](https://opensource.org/licenses/Apache-2.0).
-- The class contains an adapted version of the Eisel-Lemire algorithm and its table of powers of five from [fast_float](https://github.com/fastfloat/fast_float) by Daniel Lemire and contributors, which is available under the [MIT License](https://opensource.org/licenses/MIT) (used here), the Apache 2.0 License, and the Boost Software License. Copyright © 2021 The fast_float authors
+- The class contains an adapted version of the Eisel-Lemire algorithm, its table of powers of five, and its digit comparison for long numbers from [fast_float](https://github.com/fastfloat/fast_float) by Daniel Lemire and contributors, which is available under the [MIT License](https://opensource.org/licenses/MIT) (used here), the Apache 2.0 License, and the Boost Software License. Copyright © 2021 The fast_float authors
 
 <img align="right" src="https://git.fsfe.org/reuse/reuse-ci/raw/branch/master/reuse-horizontal.png" alt="REUSE Software">
 

docs/mkdocs/docs/api/basic_json/number_float_t.md

@@ -23,9 +23,10 @@ type to use.
 ## Template parameters
 
 `NumberFloatType`
-:   the type to store floating-point numbers. Parsing and serialization are implemented in terms of
-    `#!cpp std::strtof`/`#!cpp std::strtod`/`#!cpp std::strtold` and `#!cpp std::snprintf`, so the type must be
-    `#!cpp float`, `#!cpp double`, or `#!cpp long double`. The
+:   the type to store floating-point numbers. The parser converts `#!cpp float`, `#!cpp double`, and a
+    `#!cpp long double` that is IEEE 754 binary64 itself and other `#!cpp long double` formats with
+    `#!cpp std::from_chars` or `#!cpp std::strtold`, and serialization falls back to `#!cpp std::snprintf`, so the
+    type must be `#!cpp float`, `#!cpp double`, or `#!cpp long double`. The
     [binary formats](../../features/binary_formats/index.md) additionally require `#!cpp float` or `#!cpp double`,
     because they have no encoding for `#!cpp long double`. See
     [Template Parameter Requirements](../../features/types/template_parameters.md#numberfloattype).

docs/mkdocs/docs/features/types/number_handling.md

@@ -82,12 +82,13 @@ flowchart TD
 
     - Numbers with a decimal digit or scientific notation are always stored as `#!c double`.
     - The number types can be changed, see [Template number types](#template-number-types). 
-    - Integers are converted by the library's own digit parser. Floating-point numbers are converted with
-      [`std::from_chars`](https://en.cppreference.com/w/cpp/utility/from_chars) if the library is compiled with C++17
-      and the standard library supports it, then with an exact fast path for `#!c double` values with few significant
-      digits, and otherwise with the locale-aware
-      [`std::strtod`](https://en.cppreference.com/w/cpp/string/byte/strtof) (`std::strtof`/`std::strtold` for the
-      other floating-point types). Before version 3.13.0, the conversion was realized by
+    - The library converts integers and floating-point numbers itself, independent of the locale. Floating-point
+      numbers are correctly rounded (to nearest, ties to even). Only a `#!c long double` that is not IEEE 754 binary64
+      (e.g., the 80-bit x87 format) is converted with `#!cpp std::from_chars` where available, or else with
+      [`std::strtold`](https://en.cppreference.com/w/cpp/string/byte/strtof). For that call, the library temporarily
+      replaces the `.` with the decimal point of the current locale (which may be longer than one byte, e.g., in
+      `fa_IR.UTF-8`), so the result does not depend on the locale either. Changing the locale in another thread during
+      parsing is undefined behavior of the C library, though. Before version 3.13.0, the conversion was realized by
       [`std::strtoull`](https://en.cppreference.com/w/cpp/string/byte/strtoul),
       [`std::strtoll`](https://en.cppreference.com/w/cpp/string/byte/strtol), and `std::strtod`, respectively.
 
@@ -100,10 +101,10 @@ flowchart TD
 ### Number limits
 
 - Any 64-bit signed or unsigned integer can be stored without loss of precision.
-- Numbers exceeding the limits of `#!c double` (i.e., numbers that after conversion via
-[`std::strtod`](https://en.cppreference.com/w/cpp/string/byte/strtof) are not satisfying
+- Numbers exceeding the limits of `#!c double` (i.e., numbers whose rounded value is not satisfying
 [`std::isfinite`](https://en.cppreference.com/w/cpp/numeric/math/isfinite) such as `#!c 1E400`) will throw exception
-[`json.exception.out_of_range.406`](../../home/exceptions.md#jsonexceptionout_of_range406) during parsing.
+[`json.exception.out_of_range.406`](../../home/exceptions.md#jsonexceptionout_of_range406) during parsing. Numbers too
+small for `#!c double` (such as `#!c 1E-400`) become zero, with the sign of the number.
 - Floating-point numbers are rounded to the next number representable as `double`. For instance
 `#!c 3.141592653589793238462643383279` is stored as [`0x400921fb54442d18`](https://float.exposed/0x400921fb54442d18).
 This is the same behavior as the code `#!c double x = 3.141592653589793238462643383279;`.

docs/mkdocs/docs/features/types/template_parameters.md

@@ -26,9 +26,9 @@ Requirements are split into two groups:
     diagnosed with dedicated error messages, and violating most of them results in a compiler error somewhere inside
     the library. Four violations are not caught at compile time at all:
 
-    - A [`StringType`](#stringtype) whose `data()` is not null-terminated compiles and can silently misparse
-      floating-point numbers, because the lexer may hand the buffer to `#!cpp std::strtod`, which reads up to the
-      terminating null character.
+    - A [`StringType`](#stringtype) whose `data()` is not null-terminated compiles and silently misparses numbers
+      stored as a `#!cpp long double` that is not IEEE 754 binary64 (e.g., the 80-bit x87 format), because the lexer
+      hands the buffer to `#!cpp std::strtold`.
     - A stateful [`AllocatorType`](#allocatortype) compiles and silently ignores its state: allocation, deallocation,
       and [`get_allocator()`](../../api/basic_json/get_allocator.md) each use a different default-constructed instance.
     - The two [cross-specialization conversions](#cross-specialization-conversions) below. These abort on an assertion
@@ -537,9 +537,10 @@ therefore silently changes parse results rather than raising an error. See
 
 `NumberFloatType` must be one of `#!cpp float`, `#!cpp double`, or `#!cpp long double`:
 
-- The [parser](../parsing/index.md) converts number literals with `#!cpp std::from_chars` or, as a fallback, with
-  `#!cpp std::strtof`, `#!cpp std::strtod`, or `#!cpp std::strtold`; the library provides overloads for exactly these
-  three types.
+- The [parser](../parsing/index.md) converts number literals to `#!cpp float`, `#!cpp double`, and a
+  `#!cpp long double` that is IEEE 754 binary64 itself; other `#!cpp long double` formats are converted with
+  `#!cpp std::from_chars` where available, or with `#!cpp std::strtold`. The library provides overloads for exactly
+  these three types.
 - [`dump`](../../api/basic_json/dump.md) falls back to `#!cpp std::snprintf` with the `%g` and `%Lg` conversion
   specifiers, for which the library likewise provides only `#!cpp double` and `#!cpp long double` overloads
   (`#!cpp float` is promoted to `#!cpp double`).

docs/mkdocs/docs/home/license.md

@@ -20,4 +20,4 @@ The class contains a slightly modified version of the Grisu2 algorithm from Flor
 
 The class contains a copy of [Hedley](https://nemequ.github.io/hedley/) from Evan Nemerson which is licensed as [CC0-1.0](https://creativecommons.org/publicdomain/zero/1.0/).
 
-The class contains an adapted version of the Eisel-Lemire algorithm and its table of powers of five from [fast_float](https://github.com/fastfloat/fast_float) by Daniel Lemire and contributors, which is available under the [MIT License](https://opensource.org/licenses/MIT) (used here), the Apache 2.0 License, and the Boost Software License. Copyright © 2021 The fast_float authors
+The class contains an adapted version of the Eisel-Lemire algorithm, its table of powers of five, and its digit comparison for long numbers from [fast_float](https://github.com/fastfloat/fast_float) by Daniel Lemire and contributors, which is available under the [MIT License](https://opensource.org/licenses/MIT) (used here), the Apache 2.0 License, and the Boost Software License. Copyright © 2021 The fast_float authors

include/nlohmann/detail/bit_ops.hpp

@@ -39,6 +39,25 @@ inline int count_leading_zeros(std::uint64_t x) noexcept
 #endif
 }
 
+/// number of trailing zero bits of x (x != 0)
+inline int count_trailing_zeros(std::uint64_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_ctzll(x);
+#else
+    int n = 0;
+    for (int shift = 32; shift != 0; shift >>= 1)
+    {
+        if ((x << (64 - shift)) == 0)
+        {
+            n += shift;
+            x >>= shift;
+        }
+    }
+    return n;
+#endif
+}
+
 /// the 128-bit product of two 64-bit numbers
 struct uint128_parts
 {
@@ -68,14 +87,19 @@ inline uint128_parts full_multiplication(std::uint64_t a, std::uint64_t b) noexc
 
 /// eight bytes as a little-endian word (compilers fold this into one load on
 /// little-endian targets)
-inline std::uint64_t read_eight_bytes(const char* p) noexcept
+inline std::uint64_t read_eight_bytes(const unsigned char* b) noexcept
 {
-    const auto* b = reinterpret_cast<const unsigned char*>(p); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
     return static_cast<std::uint64_t>(b[0]) | (static_cast<std::uint64_t>(b[1]) << 8u)
            | (static_cast<std::uint64_t>(b[2]) << 16u) | (static_cast<std::uint64_t>(b[3]) << 24u)
            | (static_cast<std::uint64_t>(b[4]) << 32u) | (static_cast<std::uint64_t>(b[5]) << 40u)
            | (static_cast<std::uint64_t>(b[6]) << 48u) | (static_cast<std::uint64_t>(b[7]) << 56u);
 }
 
+/// eight bytes as a little-endian word
+inline std::uint64_t read_eight_bytes(const char* p) noexcept
+{
+    return read_eight_bytes(reinterpret_cast<const unsigned char*>(p)); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+}
+
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END

include/nlohmann/detail/input/lexer.hpp

@@ -221,6 +221,44 @@ class lexer : public lexer_base<BasicJsonType>
     // scan functions
     /////////////////////
 
+    /// contiguous input: try to decode the 4 hex digits following `\\u`
+    /// directly from the input buffer via hex_codepoint(), instead of 4 calls
+    /// to get(). On success, advances the adapter and the position counters
+    /// exactly as those 4 get() calls would (a hex digit is never '\n', so
+    /// only the flat counters move) and leaves @a current holding the last of
+    /// the 4 digits, just as the last such get() would; the codepoint is
+    /// written to @a out. Makes no state change and returns false - for a
+    /// pending unget, fewer than 4 remaining bytes, or any of the 4 bytes not
+    /// being a hex digit - so the caller falls back unchanged to the
+    /// per-character loop, which then reports the same diagnostic (stopping
+    /// at the first invalid digit) as before this optimization.
+    bool get_codepoint_bulk(std::true_type /*bulk*/, int& out)
+    {
+        if (next_unget || ia.bulk_remaining() < 4)
+        {
+            return false;
+        }
+        const char_type* const raw = ia.bulk_data();
+        const int codepoint = hex_codepoint(reinterpret_cast<const unsigned char*>(raw));
+        if (codepoint < 0)
+        {
+            return false;
+        }
+        ia.bulk_skip(4);
+        // a hex digit is never a newline, so only the flat counters advance
+        position.chars_read_total += 4;
+        position.chars_read_current_line += 4;
+        current = char_traits<char_type>::to_int_type(raw[3]);
+        out = codepoint;
+        return true;
+    }
+
+    /// streaming input: no bulk fast path
+    bool get_codepoint_bulk(std::false_type /*bulk*/, int& /*out*/) const noexcept
+    {
+        return false;
+    }
+
     /*!
     @brief get codepoint from 4 hex characters following `\\u`
 
@@ -240,6 +278,14 @@ class lexer : public lexer_base<BasicJsonType>
     {
         // this function only makes sense after reading `\u`
         JSON_ASSERT(current == 'u');
+
+        // contiguous input: decode all 4 hex digits directly from the buffer
+        int fast_codepoint = 0;
+        if (get_codepoint_bulk(std::integral_constant<bool, bulk_scan> {}, fast_codepoint))
+        {
+            return fast_codepoint;
+        }
+
         int codepoint = 0;
 
         const auto factors = { 12u, 8u, 4u, 0u };
@@ -1044,9 +1090,11 @@ class lexer : public lexer_base<BasicJsonType>
             token_type::parse_error otherwise
 
     @note The scanner is independent of the current locale: token_buffer
-          always holds `.`. Only the std::strtod fallback of convert_number()
-          depends on the locale, and it looks up the decimal point right
-          before converting (see detail::convert_float_locale_aware()).
+          always holds `.`. The conversion of float and double does not use
+          the locale either. Only the std::strtold fallback of
+          convert_number() for long double formats other than binary64
+          depends on it, and it looks up the decimal point right before
+          converting (see detail::convert_float_locale_aware()).
     */
     token_type scan_number()  // lgtm [cpp/use-of-goto] `goto` is used in this function to implement the number-parsing state machine described above. By design, any finite input will eventually reach the "done" state or return token_type::parse_error. In each intermediate state, 1 byte of the input is appended to the token_buffer vector, and only the already initialized variables token_buffer, number_type, and error_message are manipulated.
     {
@@ -1059,7 +1107,7 @@ class lexer : public lexer_base<BasicJsonType>
 
         // offset just past the last mantissa byte in token_buffer (i.e. the
         // index of 'e'/'E', or the whole token when there is no exponent).
-        // convert_number() uses it to count significant digits; npos means
+        // convert_number() uses it to split the token; npos means
         // "not seen an exponent yet" and is resolved at scan_number_done
         std::size_t mantissa_end = std::string::npos;
 
@@ -1389,8 +1437,8 @@ class lexer : public lexer_base<BasicJsonType>
     @param[in] mantissa_end  offset just past the last mantissa byte in
                              token_buffer (the index of 'e'/'E', or
                              token_buffer.size() when there is no exponent);
-                             used to skip Clinger's fast path when it cannot
-                             possibly succeed - see detail::mantissa_fits_clinger()
+                             with decimal_point_position, it locates the parts
+                             of a float token without scanning it again
     */
     token_type convert_number(token_type number_type, std::size_t mantissa_end)
     {
@@ -1444,10 +1492,11 @@ class lexer : public lexer_base<BasicJsonType>
         }
 
         // this code is reached if we parse a floating-point number or if an
-        // integer conversion above overflowed. Prefer std::from_chars
-        // (Eisel-Lemire, locale-independent, correctly rounded) when available;
-        // otherwise the exact Clinger fast path (double only); otherwise the
-        // locale-aware strtof/strtod/strtold.
+        // integer conversion above overflowed. float and double (and long
+        // double where it is binary64) are converted by the library itself,
+        // correctly rounded and independent of the locale; other long double
+        // formats use std::from_chars when available, otherwise the
+        // locale-aware strtold.
         if (convert_float_fast(num_begin, num_end, decimal_point_position, mantissa_end, value_float))
         {
             return token_type::value_float;