Skip the indentation after a newline 8 bytes at a time (#5810)

Pretty-printed JSON indents every line with a run of spaces, which
skip_whitespace() read one character at a time. For contiguous input
(bulk_scan) the lexer now skips the spaces after a newline 8 bytes at a
time directly in the input buffer, including the leading spaces of the
last, partial word. This bypasses get(), which is safe because
bulk_scan implies the lazy position from the previous commit: nothing
is counted per character.

Compared to lazy positions alone, pretty-printed input (twitter,
citm_catalog with indent 2 and 4) parses up to 13% faster and accept()
gets up to 30% faster (GCC 13 and clang 18 on x86-64, Apple clang 21 on
arm64). Minified input is unchanged within noise.

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

docs/mkdocs/docs/home/architecture.md

@@ -189,7 +189,7 @@ random-access input of single-byte characters:
   `count_consumed_lines()`, the lexer also derives the read position from the adapter and counts lines only when an
   error needs them, instead of counting characters and lines on every read.
 - `supports_bulk_scan`, `bulk_data()`, `bulk_remaining()`, and `bulk_skip()` let the lexer scan strings directly in
-  contiguous memory, several bytes at a time.
+  contiguous memory, several bytes at a time, and skip the spaces that indent a line in 8-byte steps.
 
 The function `input_adapter` picks the right adapter for the argument passed to `parse`, `accept`, `sax_parse`, or the
 `from_*` functions:

include/nlohmann/detail/bit_ops.hpp

@@ -42,6 +42,25 @@ 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
 {

include/nlohmann/detail/input/lexer.hpp

@@ -17,6 +17,7 @@
 #include <utility> // move
 #include <vector> // vector
 
+#include <nlohmann/detail/bit_ops.hpp>
 #include <nlohmann/detail/input/input_adapters.hpp>
 #include <nlohmann/detail/input/number_parse.hpp>
 #include <nlohmann/detail/input/position_t.hpp>
@@ -2079,11 +2080,53 @@ scan_number_done:
         // into one that reloads the pointer from memory every character
         do
         {
+            if (current == '\n')
+            {
+                skip_indentation(std::integral_constant<bool, bulk_scan> {});
+            }
             get_ignoring_pending_unget();
         }
         while (current_is_whitespace());
     }
 
+    /*!
+    @brief contiguous input: skip the space indentation that follows a newline
+           8 bytes at a time
+
+    Pretty-printed JSON indents every line with a run of spaces, which the loop
+    in skip_whitespace() would otherwise read one character at a time. The
+    spaces are skipped 8 bytes at a time; the last 7 bytes of the input and any
+    other whitespace are left to that loop. Skipping bypasses get(), which is
+    fine because bulk_scan implies lazy_position: the position is derived from
+    the adapter, not counted per character.
+    */
+    void skip_indentation(std::true_type /*bulk*/)
+    {
+        JSON_ASSERT(!next_unget);
+        const std::size_t remaining = ia.bulk_remaining();
+        if (remaining < 8)
+        {
+            return;
+        }
+        const char* const data = reinterpret_cast<const char*>(ia.bulk_data());
+        std::size_t pos = 0;
+        for (; pos + 8 <= remaining; pos += 8)
+        {
+            // little-endian, so the first byte is the lowest
+            const std::uint64_t non_space = read_eight_bytes(data + pos) ^ 0x2020202020202020ull;
+            if (non_space != 0)
+            {
+                // skip the spaces before the first other byte of this word
+                pos += static_cast<std::size_t>(count_trailing_zeros(non_space)) / 8;
+                break;
+            }
+        }
+        ia.bulk_skip(pos);
+    }
+
+    /// streaming input: no bulk fast path
+    void skip_indentation(std::false_type /*bulk*/) const noexcept {}
+
     token_type scan()
     {
         // initially, skip the BOM

single_include/nlohmann/json.hpp

@@ -8925,30 +8925,6 @@ NLOHMANN_JSON_NAMESPACE_END
 #include <utility> // move
 #include <vector> // vector
 
-// #include <nlohmann/detail/input/input_adapters.hpp>
-
-// #include <nlohmann/detail/input/number_parse.hpp>
-//     __ _____ _____ _____
-//  __|  |   __|     |   | |  JSON for Modern C++
-// |  |  |__   |  |  | | | |  version 3.12.0
-// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
-//
-// SPDX-FileCopyrightText: 2021 The fast_float authors <https://github.com/fastfloat/fast_float>
-// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
-// SPDX-License-Identifier: MIT
-
-
-
-#include <array> // array
-#include <cfloat> // FLT_EVAL_METHOD
-#include <clocale> // localeconv
-#include <cstddef> // size_t
-#include <cstdint> // int64_t, uint64_t
-#include <cstdlib> // strtof, strtod, strtold
-#include <cstring> // memcpy
-#include <limits> // numeric_limits
-#include <string> // string
-
 // #include <nlohmann/detail/bit_ops.hpp>
 //     __ _____ _____ _____
 //  __|  |   __|     |   | |  JSON for Modern C++
@@ -8995,6 +8971,25 @@ 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
 {
@@ -9042,6 +9037,32 @@ inline std::uint64_t read_eight_bytes(const char* p) noexcept
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END
 
+// #include <nlohmann/detail/input/input_adapters.hpp>
+
+// #include <nlohmann/detail/input/number_parse.hpp>
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2021 The fast_float authors <https://github.com/fastfloat/fast_float>
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+
+
+#include <array> // array
+#include <cfloat> // FLT_EVAL_METHOD
+#include <clocale> // localeconv
+#include <cstddef> // size_t
+#include <cstdint> // int64_t, uint64_t
+#include <cstdlib> // strtof, strtod, strtold
+#include <cstring> // memcpy
+#include <limits> // numeric_limits
+#include <string> // string
+
+// #include <nlohmann/detail/bit_ops.hpp>
+
 // #include <nlohmann/detail/input/pow5_table.hpp>
 //     __ _____ _____ _____
 //  __|  |   __|     |   | |  JSON for Modern C++
@@ -12500,11 +12521,53 @@ scan_number_done:
         // into one that reloads the pointer from memory every character
         do
         {
+            if (current == '\n')
+            {
+                skip_indentation(std::integral_constant<bool, bulk_scan> {});
+            }
             get_ignoring_pending_unget();
         }
         while (current_is_whitespace());
     }
 
+    /*!
+    @brief contiguous input: skip the space indentation that follows a newline
+           8 bytes at a time
+
+    Pretty-printed JSON indents every line with a run of spaces, which the loop
+    in skip_whitespace() would otherwise read one character at a time. The
+    spaces are skipped 8 bytes at a time; the last 7 bytes of the input and any
+    other whitespace are left to that loop. Skipping bypasses get(), which is
+    fine because bulk_scan implies lazy_position: the position is derived from
+    the adapter, not counted per character.
+    */
+    void skip_indentation(std::true_type /*bulk*/)
+    {
+        JSON_ASSERT(!next_unget);
+        const std::size_t remaining = ia.bulk_remaining();
+        if (remaining < 8)
+        {
+            return;
+        }
+        const char* const data = reinterpret_cast<const char*>(ia.bulk_data());
+        std::size_t pos = 0;
+        for (; pos + 8 <= remaining; pos += 8)
+        {
+            // little-endian, so the first byte is the lowest
+            const std::uint64_t non_space = read_eight_bytes(data + pos) ^ 0x2020202020202020ull;
+            if (non_space != 0)
+            {
+                // skip the spaces before the first other byte of this word
+                pos += static_cast<std::size_t>(count_trailing_zeros(non_space)) / 8;
+                break;
+            }
+        }
+        ia.bulk_skip(pos);
+    }
+
+    /// streaming input: no bulk fast path
+    void skip_indentation(std::false_type /*bulk*/) const noexcept {}
+
     token_type scan()
     {
         // initially, skip the BOM

tests/src/unit-class_lexer.cpp

@@ -938,6 +938,10 @@ TEST_CASE("Eisel-Lemire float conversion")
             const int k = i % 64;
             const std::uint64_t x = (std::uint64_t{1} << k) | (a & ((std::uint64_t{1} << k) - 1));
             CHECK(nlohmann::detail::count_leading_zeros(x) == 63 - k);
+
+            // the mirror image: bit k set, arbitrary bits above it
+            const std::uint64_t y = (std::uint64_t{1} << k) | (k == 63 ? 0 : (a << (k + 1)));
+            CHECK(nlohmann::detail::count_trailing_zeros(y) == k);
         }
     }
 

tests/src/unit-deserialization.cpp

@@ -1314,3 +1314,76 @@ TEST_CASE("parse error positions are the same for contiguous and stream input")
         CHECK(stream_sax.error == expected);
     }
 }
+
+#if !defined(JSON_NOEXCEPTION)
+TEST_CASE("indentation runs of every length around the 8-byte skip")
+{
+    // contiguous input skips the spaces after a newline 8 bytes at a time (see
+    // lexer::skip_indentation), stream input reads them one by one; results,
+    // byte offsets, and messages must be the same for runs that end anywhere
+    // relative to a word boundary, including at the end of the input
+    const auto parse_result = [](const std::string & input, bool contiguous)
+    {
+        try
+        {
+            if (contiguous)
+            {
+                return json::parse(input.data(), input.data() + input.size(), nullptr, true, true).dump();
+            }
+            std::istringstream is(input);
+            return json::parse(is, nullptr, true, true).dump();
+        }
+        catch (const json::parse_error& e)
+        {
+            return std::to_string(e.byte).append(" ").append(e.what());
+        }
+    };
+
+    // replace every '~' in a pattern by an indentation run
+    const auto indented = [](const std::string & pattern, const std::string & indent)
+    {
+        std::string result;
+        for (const char c : pattern)
+        {
+            if (c == '~')
+            {
+                result += indent;
+            }
+            else
+            {
+                result += c;
+            }
+        }
+        return result;
+    };
+
+    const std::vector<std::string> patterns =
+    {
+        "[\n~1,\n~2\n]",
+        "[\n~1,\n~x]",
+        "[\n~1,\n~",
+        "1\n~",
+        "[\n~\t1,\n~\r\n~2]",
+        "[\n~/* c */ 1]",
+        "[\n~1\n~2]",
+    };
+
+    for (std::size_t n = 0; n <= 33; ++n)
+    {
+        const std::string indent(n, ' ');
+        for (const auto& pattern : patterns)
+        {
+            const std::string input = indented(pattern, indent);
+            CAPTURE(n)
+            CAPTURE(input)
+            const std::string expected = parse_result(input, false);
+            CHECK(parse_result(input, true) == expected);
+            CHECK(json::accept(input.data(), input.data() + input.size(), true) == json::accept(input, true));
+        }
+    }
+
+    // and against fixed expectations, in case both paths broke the same way
+    CHECK(parse_result(indented("[\n~x]", std::string(16, ' ')), true) == "19 [json.exception.parse_error.101] parse error at line 2, column 17: syntax error while parsing value - invalid literal; last read: '[<U+000A>                x'");
+    CHECK(parse_result(indented("[\n~1]", std::string(17, ' ')), true) == "[1]");
+}
+#endif