Keep the serializer's buffers out of -Weffc++ and cpplint warnings

GCC's -Weffc++ asks for the three buffers in the member initialization
list, which would zero them again. They are now of a type with an empty
user-provided constructor (as the json_view builder's frame stack), so
they are listed there and stay uninitialized.

The 8-to-31-byte branch of the short-string writer no longer spans an
#if with a dangling else, which cpplint read as an else body without
braces.

dump() instructions (Apple clang, twitter/citm/canada, compact and
indented) are unchanged within noise (-0.3 % to +0.6 %).

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

.github/external_ci/appveyor.yml

@@ -18,7 +18,8 @@ environment:
   matrix:
     # The Visual Studio 2017 jobs compile everything with /std:c++17, so they
     # only build the C++17 variant of each test, split into two jobs each to
-    # stay below AppVeyor's 60-minute limit per job.
+    # stay below AppVeyor's 60-minute limit per job. The Visual Studio 2019
+    # job builds the tests for all standards and is split for the same reason.
     - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
       configuration: Debug
       platform: x86
@@ -50,7 +51,13 @@ environment:
       configuration: Release
       platform: x86
       CXX_FLAGS: "/W4 /WX"
-      CMAKE_OPTIONS: "-DJSON_ImplicitConversions=OFF"
+      CMAKE_OPTIONS: "-DJSON_ImplicitConversions=OFF -DJSON_TestShard=0/2"
+      GENERATOR: Visual Studio 16 2019
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2019
+      configuration: Release
+      platform: x86
+      CXX_FLAGS: "/W4 /WX"
+      CMAKE_OPTIONS: "-DJSON_ImplicitConversions=OFF -DJSON_TestShard=1/2"
       GENERATOR: Visual Studio 16 2019
 
     - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015

.github/labeler.yml

@@ -64,6 +64,7 @@ labels:
     - "tests/src/unit-json_view.*"
     - "tests/src/json_view_test_helpers\\.hpp"
     - "tests/src/fuzzer-(parse_json_view|json_view_image)\\.cpp"
+    - "tests/benchmarks/json_view/.*"
     - "tools/amalgamate/config_json_view\\.json"
     - "docs/mkdocs/docs/features/json_view\\.md"
     - "docs/mkdocs/docs/api/basic_json_(document|view)/.*"

.github/workflows/check_amalgamation.yml

@@ -154,6 +154,28 @@ jobs:
           # PR content in a job without credentials.
           python3 tools/generate_natvis/generate_natvis.py $MAIN_DIR
 
+      - name: Check the nlohmann_json.natvis generator options
+        run: |
+          cd $MAIN_DIR
+          # The options for configurations outside the checked-in file write
+          # to a temporary file, so they do not affect the diff below.
+          GENERATOR=tools/generate_natvis/generate_natvis.py
+          python3 $GENERATOR --list-tags
+          TAG_COUNT=$(python3 $GENERATOR --list-tags | wc -l)
+
+          python3 $GENERATOR --abi-tags diag,_dp,_vns --abi-tags _diag --output $RUNNER_TEMP/abi_tags.natvis
+          grep -q '<!-- Namespace nlohmann::json_abi_diag_dp_vns -->' $RUNNER_TEMP/abi_tags.natvis
+          grep -q '<!-- Namespace nlohmann::json_abi_diag_dp_vns_v[0-9_]* -->' $RUNNER_TEMP/abi_tags.natvis
+          test $(grep -c '<!-- Namespace ' $RUNNER_TEMP/abi_tags.natvis) -eq $(( $(grep -c '<!-- Namespace ' nlohmann_json.natvis) + 2 ))
+
+          python3 $GENERATOR --all --output $RUNNER_TEMP/all.natvis
+          test $(grep -c '<!-- Namespace ' $RUNNER_TEMP/all.natvis) -eq $(( (1 << TAG_COUNT) * 2 + 1 ))
+
+          if python3 $GENERATOR --abi-tags _unknown --output $RUNNER_TEMP/unknown.natvis; then
+            echo "generate_natvis.py accepted an unknown ABI tag"
+            exit 1
+          fi
+
       - name: Build patch and check for differences
         id: diff
         run: |

.github/workflows/json_view_benchmarks.yml

@@ -0,0 +1,84 @@
+name: "json_view benchmarks"
+
+# On demand only: runs the comparison of json_view with yyjson, simdjson, and
+# Boost.JSON (tests/benchmarks/json_view/compare.py) on GitHub-hosted runners,
+# for numbers from x86-64 and AArch64 Linux. It runs when started by hand, or
+# when a pull request gets the label "benchmark" (on both architectures, with
+# GCC and the default settings). Shared runners are noisy: the results show
+# where json_view stands, but published numbers need a quiet machine (see
+# tests/benchmarks/json_view/README.md).
+
+on:
+  pull_request:
+    types: [labeled]
+  workflow_dispatch:
+    inputs:
+      runner:
+        description: "Runner image"
+        type: choice
+        options:
+          - ubuntu-24.04
+          - ubuntu-24.04-arm
+        default: ubuntu-24.04
+      compiler:
+        description: "Compiler"
+        type: choice
+        options:
+          - g++
+          - clang++
+        default: g++
+      native:
+        description: "Compile for the runner's CPU (-march=native)"
+        type: boolean
+        default: false
+      rounds:
+        description: "Rounds of bench_view"
+        type: number
+        default: 30
+
+permissions:
+  contents: read
+
+concurrency:
+  # (the label is part of the group: adding an unrelated label must not cancel a running benchmark)
+  group: json-view-benchmarks-${{ github.workflow }}-${{ github.ref }}-${{ github.event.label.name }}
+  cancel-in-progress: true
+
+jobs:
+  compare:
+    if: github.event_name == 'workflow_dispatch' || github.event.label.name == 'benchmark'
+    strategy:
+      matrix:
+        runner: ${{ fromJSON(github.event_name == 'workflow_dispatch' && format('["{0}"]', inputs.runner) || '["ubuntu-24.04", "ubuntu-24.04-arm"]') }}
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 60
+    steps:
+      - name: Harden Runner
+        uses: step-security/harden-runner@e14015d583714f6e62063499dc959a02595150a1 # v2.21.1
+        with:
+          egress-policy: audit
+
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+
+      - name: Download test data
+        run: |
+          cmake -S . -B build -DJSON_BuildTests=On
+          cmake --build build --target download_test_data
+
+      - name: Run the comparison
+        env:
+          CXX: ${{ inputs.compiler || 'g++' }}
+          CC: ${{ inputs.compiler == 'clang++' && 'clang' || 'gcc' }}
+          ROUNDS: ${{ inputs.rounds || 30 }}
+          NATIVE: ${{ inputs.native && '--native' || '' }}
+        run: python3 tests/benchmarks/json_view/compare.py --data build/test_files --download --rounds "$ROUNDS" $NATIVE
+
+      - name: Summary
+        run: cat tests/benchmarks/json_view/results/*.md >> "$GITHUB_STEP_SUMMARY"
+
+      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: json_view-benchmarks-${{ matrix.runner }}-${{ inputs.compiler || 'g++' }}
+          path: tests/benchmarks/json_view/results/

BUILD.bazel

@@ -37,6 +37,7 @@ cc_library(
         "include/nlohmann/detail/input/parser.hpp",
         "include/nlohmann/detail/input/position_t.hpp",
         "include/nlohmann/detail/input/pow5_table.hpp",
+        "include/nlohmann/detail/input/simd_scan.hpp",
         "include/nlohmann/detail/input/string_scan.hpp",
         "include/nlohmann/detail/iterators/internal_iterator.hpp",
         "include/nlohmann/detail/iterators/iter_impl.hpp",

FILES.md

@@ -6,6 +6,7 @@ This file describes the source for supporting files; that is, files that are not
 - [GitHub](#github)
 - [REUSE](#reuse)
 - [Package Managers](#package-managers)
+- [Debugging](#debugging)
 
 ## Continuous Integration
 
@@ -289,3 +290,29 @@ The module definition for [Bazel](https://bazel.build)'s [Bzlmod](https://bazel.
 Further documentation:
 
 - [Bazel modules](https://bazel.build/external/module)
+
+## Debugging
+
+### `nlohmann_json.natvis`
+
+The [Natvis](https://learn.microsoft.com/en-us/visualstudio/debugger/create-custom-views-of-native-objects) file that
+gives `json` values a friendly view in the Visual Studio debugger. The file is generated by
+`tools/generate_natvis/generate_natvis.py` from the template `tools/generate_natvis/nlohmann_json.natvis.j2` and the ABI
+tags and version in `include/nlohmann/detail/abi_macros.hpp`; change the script or the template rather than editing
+the file by hand. The file can be updated by calling
+
+```shell
+make natvis
+```
+
+The "Check amalgamation" workflow fails if the file is out of date.
+
+The file covers the namespace variants with up to two ABI tags. A Natvis file for other configurations can be
+generated with the `--abi-tags` or `--all` options, for example
+`make natvis NATVIS_ARGS="--abi-tags _diag,_dp,_vns --output my_nlohmann_json.natvis"`; see
+[`tools/generate_natvis`](tools/generate_natvis/README.md). With MSVC, CMake adds the file to the `nlohmann_json`
+target and installs it.
+
+Further documentation:
+
+- [Create custom views of C++ objects in the debugger using the Natvis framework](https://learn.microsoft.com/en-us/visualstudio/debugger/create-custom-views-of-native-objects)

Makefile

@@ -46,7 +46,7 @@ all:
 	@echo "clean - remove built files"
 	@echo "(fuzzers) - see tests/fuzzing.md, or run 'make -C tests fuzzers' to build the AFL fuzzers"
 	@echo "macro_builder_check - check that macro_scope.hpp matches tools/macro_builder's output"
-	@echo "natvis - regenerate nlohmann_json.natvis from the current ABI tags and version"
+	@echo "natvis - regenerate nlohmann_json.natvis from the current ABI tags and version (pass generator options via NATVIS_ARGS, see tools/generate_natvis/README.md)"
 	@echo "pretty - beautify code with Artistic Style"
 	@echo "pvs_studio - run the PVS-Studio static analyzer and open the report"
 	@echo "release - create the release files (include.zip, json.tar.xz, single headers, signatures, hashes)"
@@ -120,13 +120,15 @@ $(AMALGAMATED_VIEW_FILE): $(SRCS)
 	tools/amalgamate/amalgamate.py -c tools/amalgamate/config_json_view.json -s . --verbose=yes
 
 NATVIS_VENV=tools/generate_natvis/venv
+# extra options for generate_natvis.py, e.g. NATVIS_ARGS="--abi-tags _diag,_dp,_vns --output my.natvis"
+NATVIS_ARGS=
 
 # regenerate nlohmann_json.natvis from the ABI tags and version in include/nlohmann/detail/abi_macros.hpp
 # (the dependencies are installed into a virtual environment first)
 natvis:
 	@test -d $(NATVIS_VENV) || python3 -mvenv $(NATVIS_VENV)
 	@$(NATVIS_VENV)/bin/python3 -c 'import jinja2' 2>/dev/null || $(NATVIS_VENV)/bin/pip3 install --quiet -r tools/generate_natvis/requirements.txt
-	$(NATVIS_VENV)/bin/python3 tools/generate_natvis/generate_natvis.py .
+	$(NATVIS_VENV)/bin/python3 tools/generate_natvis/generate_natvis.py $(NATVIS_ARGS) .
 
 # regenerate the two tools/macro_builder blocks of $(MACRO_SCOPE_HPP) (see its README.md) and diff against the
 # checked-in header; phony, because it never writes $(MACRO_SCOPE_HPP) itself

docs/mkdocs/docs/api/macros/index.md

@@ -38,6 +38,7 @@ header. See also the [macro overview page](../../features/macros.md).
 - [**JSON_HAS_THREE_WAY_COMPARISON**](json_has_three_way_comparison.md) - control 3-way comparison support
 - [**JSON_NO_AUTOMATIC_UDLS**](json_no_automatic_udls.md) - do not include the user-defined string literals (UDLs) automatically
 - [**JSON_NO_IO**](json_no_io.md) - switch off functions relying on certain C++ I/O headers
+- [**JSON_NO_SIMD**](json_no_simd.md) - use only portable code instead of vector instructions (SSE2, SSSE3, NEON)
 - [**JSON_NO_THREAD_LOCAL**](json_no_thread_local.md) - switch off the use of `thread_local` storage
 - [**JSON_SKIP_UNSUPPORTED_COMPILER_CHECK**](json_skip_unsupported_compiler_check.md) - do not warn about unsupported compilers
 - [**JSON_USE_GLOBAL_UDLS**](json_use_global_udls.md) - place user-defined string literals (UDLs) into the global namespace

docs/mkdocs/docs/api/macros/json_no_simd.md

@@ -0,0 +1,83 @@
+# JSON_NO_SIMD
+
+```cpp
+#define JSON_NO_SIMD
+```
+
+When defined, the library uses only portable C++ code where it would otherwise use vector instructions: when scanning
+strings during parsing, and when writing floating-point numbers during serialization.
+
+The macro is an escape hatch, for example for toolchains that cannot compile the vector code, for sanitizers or
+analyzers that do not understand it, or to compare the speed of both variants. It does not change any result: values,
+exceptions, and error positions are the same either way. Only the speed differs.
+
+## Default definition
+
+By default, `#!cpp JSON_NO_SIMD` is not defined, and vector instructions are used where the target has them.
+
+```cpp
+#undef JSON_NO_SIMD
+```
+
+## Notes
+
+!!! note "Which instructions are used"
+
+    - **x86-64:** SSE2, which every x86-64 CPU has, finds the end of runs of plain string characters 16 bytes at a time
+      and writes the digits of floating-point numbers. The UTF-8 check of non-ASCII text needs SSSE3, which is not part
+      of x86-64: it is compiled for SSSE3 with a function attribute and used when the CPU has SSSE3 (every x86-64 CPU
+      since about 2011), else the portable check. Code compiled for SSSE3 anyway (e.g., with `-mssse3` or `-march=native`)
+      skips this check of the CPU. 32-bit x86 uses SSE2 for strings when the compiler targets it.
+    - **AArch64:** NEON, which every 64-bit Arm CPU has, checks UTF-8 text and long runs of plain string characters 16
+      bytes at a time, and writes the digits of floating-point numbers. With MSVC, only the floating-point digits use
+      NEON.
+    - **Other targets:** portable code only.
+
+    Strings are only scanned this way for contiguous input and for the characters a `#!cpp std::istream` holds in its
+    stream buffer (see [`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md)); other inputs are read one
+    character at a time.
+
+!!! note "Interaction with `JSON_USE_SIMDUTF`"
+
+    With [`JSON_USE_SIMDUTF`](json_use_simdutf.md), runs of string characters are first validated by simdutf; the code
+    described here is used for what simdutf does not accept, so that errors are reported at the same position.
+    `JSON_NO_SIMD` does not affect simdutf.
+
+!!! note "ABI compatibility"
+
+    The macro selects between two definitions of the library's internal scanning and formatting functions. Whether it
+    is defined is encoded in the [namespace](../../features/namespace.md) (tag `_ns`), resulting in distinct symbol
+    names. Translation units compiled with and without it can therefore be linked into the same program without One
+    Definition Rule (ODR) violations, but they cannot exchange instances of library types.
+
+## Examples
+
+??? example "Example: switching off vector instructions"
+
+    The program parses and serializes the same values with and without the macro:
+
+    ```cpp
+    #define JSON_NO_SIMD
+    #include <iostream>
+    #include <nlohmann/json.hpp>
+
+    using json = nlohmann::json;
+
+    int main()
+    {
+        const json j = json::parse(R"({"name": "日本語 text", "pi": 3.141592653589793})");
+        std::cout << j.dump() << '\n';  // {"name":"日本語 text","pi":3.141592653589793}
+    }
+    ```
+
+## See also
+
+- [**parse**](../basic_json/parse.md) - deserialize from a compatible input
+- [**dump**](../basic_json/dump.md) - serialization
+- [**JSON_USE_SIMDUTF**](json_use_simdutf.md) - use the simdutf library to accelerate UTF-8 validation
+- [**JSON_DISABLE_STREAM_BULK_SCAN**](json_disable_stream_bulk_scan.md) - read a `std::istream` one character at a time
+- [Parsing: input sources and performance](../../features/parsing/index.md#input-sources-and-performance)
+
+## Version history
+
+- Added in version 3.13.0.

docs/mkdocs/docs/api/macros/json_use_simdutf.md

@@ -6,8 +6,9 @@
 
 When defined, the parser validates the UTF-8 content of JSON strings that come from a **contiguous byte input**
 (`std::string`, `std::vector<char>`/`<std::uint8_t>`, string literals, `const char*` ranges, …) using the
-[simdutf](https://github.com/simdutf/simdutf) library instead of the built-in scalar validator. On text with many
-non-ASCII characters (e.g. CJK or emoji) this can validate several times faster.
+[simdutf](https://github.com/simdutf/simdutf) library instead of the built-in validator (which uses SSSE3 or NEON where
+available, see [`JSON_NO_SIMD`](json_no_simd.md)). On text with many non-ASCII characters (e.g. CJK or emoji) this can
+validate faster, depending on the CPU.
 
 This is an **opt-in external dependency**. The library itself remains header-only and its behavior is unchanged: the
 same input is accepted or rejected either way, and every parse error is reported at the same position with the same

docs/mkdocs/docs/api/macros/nlohmann_json_namespace.md

@@ -39,6 +39,6 @@ the library.
 ## Version history
 
 - Added in version 3.11.0. Changed inline namespace name in version 3.11.2. Added the ABI tag `_dp` in
-  version 3.12.0, and the ABI tags `_bics`, `_psp`, `_snul`, `_sbu8`, `_ekmo`, `_dsbs`, `_dufio`, `_vns`, and
-  `_vssse3` in version 3.13.0; see
+  version 3.12.0, and the ABI tags `_bics`, `_psp`, `_snul`, `_sbu8`, `_ekmo`, `_dsbs`, `_dufio`, `_ns`, `_vns`,
+  and `_vssse3` in version 3.13.0; see
   [`nlohmann` Namespace](../../features/namespace.md#structure).

docs/mkdocs/docs/features/macros.md

@@ -174,6 +174,13 @@ security reasons (e.g., Intel Software Guard Extensions (SGX)).
 
 See [full documentation of `JSON_NO_IO`](../api/macros/json_no_io.md).
 
+## `JSON_NO_SIMD`
+
+When defined, the library uses only portable code instead of vector instructions (SSE2, SSSE3, NEON) for scanning
+strings and writing floating-point numbers. Results are the same either way. It is not defined by default.
+
+See [full documentation of `JSON_NO_SIMD`](../api/macros/json_no_simd.md).
+
 ## `JSON_NO_THREAD_LOCAL`
 
 When defined, the library does not use `#!cpp thread_local` storage. Copying a value and comparing two values then

docs/mkdocs/docs/features/namespace.md

@@ -27,6 +27,7 @@ The complete default namespace name is derived as follows:
       `_dsbs`.
     - [`JSON_DISABLE_UNLOCKED_FILE_IO`](../api/macros/json_disable_unlocked_file_io.md) defined non-zero appends
       `_dufio`.
+    - [`JSON_NO_SIMD`](../api/macros/json_no_simd.md) defined appends `_ns`.
     - [`JSON_VIEW_NO_SIMD`](../api/macros/json_view_no_simd.md) defined appends `_vns`.
     - [`JSON_VIEW_USE_SSSE3`](../api/macros/json_view_use_ssse3.md) defined appends `_vssse3`.
 - The inline namespace ends with the suffix `_v` followed by the 3 components of the version number separated by
@@ -114,6 +115,7 @@ follows:
   ([`JSON_STRICT_BINARY_UTF8`](../api/macros/json_strict_binary_utf8.md)), `_ekmo`
   ([`JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS`](../api/macros/json_use_objects_for_enum_keyed_maps.md)), `_dsbs`
   ([`JSON_DISABLE_STREAM_BULK_SCAN`](../api/macros/json_disable_stream_bulk_scan.md)), `_dufio`
-  ([`JSON_DISABLE_UNLOCKED_FILE_IO`](../api/macros/json_disable_unlocked_file_io.md)), `_vns`
+  ([`JSON_DISABLE_UNLOCKED_FILE_IO`](../api/macros/json_disable_unlocked_file_io.md)), `_ns`
+  ([`JSON_NO_SIMD`](../api/macros/json_no_simd.md)), `_vns`
   ([`JSON_VIEW_NO_SIMD`](../api/macros/json_view_no_simd.md)), and `_vssse3`
   ([`JSON_VIEW_USE_SSSE3`](../api/macros/json_view_use_ssse3.md)) in version 3.13.0.

docs/mkdocs/docs/features/parsing/index.md

@@ -79,6 +79,10 @@ follow the usual `std::streambuf` contract, and
 [`JSON_DISABLE_UNLOCKED_FILE_IO`](../../api/macros/json_disable_unlocked_file_io.md) for programs in which another
 thread must use the same `#!cpp FILE` object while it is parsed.
 
+Where the target has them, strings are scanned with vector instructions: SSE2 on x86-64 (and SSSE3 for UTF-8 text, when
+the CPU has it) and NEON on AArch64. [`JSON_NO_SIMD`](../../api/macros/json_no_simd.md) selects portable code instead,
+again without changing any result.
+
 ## SAX vs. DOM parsing
 
 The library offers two parsing models:

docs/mkdocs/docs/features/performance.md

@@ -20,32 +20,39 @@ below apply on every supported standard. This adapter exposes two optimizations
 
 - it can reconstruct already-consumed input on demand for error messages, instead of copying every character as it is
   read, and
-- the lexer can scan ordinary string characters directly out of the buffer, several bytes at a time, rather than one
-  character (and one function call) at a time.
+- the lexer can scan ordinary string characters, numbers, and indentation directly out of the buffer, several bytes at
+  a time, rather than one character (and one function call) at a time.
 
-A `#!cpp std::istream` (including `#!cpp std::ifstream`) or `#!cpp FILE*`, by contrast, is read through
-`input_stream_adapter` or `file_input_adapter`, which read one character (or one block, for binary formats) at a time
-and expose neither optimization -- the lexer falls back to the same byte-at-a-time path it uses for any
-non-contiguous, general-purpose iterator range. An iterator pair over non-contiguous but random-access storage (e.g.
-`#!cpp std::deque<char>::iterator`) gets the first optimization but not the second, since the byte-scanning fast path
-additionally requires contiguous storage.
+A `#!cpp std::istream` (including `#!cpp std::ifstream`) gets the same bulk scanning: `input_stream_adapter` scans the
+characters its stream buffer already holds (the *get area*) and asks the stream buffer for more only when they are used
+up. Nothing is read ahead, so the stream is left exactly where it would be after reading one character at a time, and
+[`operator>>`](../api/operator_gtgt.md) can still read several values from one stream. A stream buffer without a get
+area is read one character at a time. A `#!cpp FILE*` is still read one character at a time, but `file_input_adapter`
+locks the file once per parse instead of once per character. See
+[input sources and performance](parsing/index.md#input-sources-and-performance) for the full table. According to the
+description of [#5828](https://github.com/nlohmann/json/pull/5828), parsing from a stream is within about 10 to 20 % of
+parsing an equivalent `#!cpp std::string`.
 
-Practically: if the JSON text is already in memory, or small enough to read into memory, prefer passing a
-`#!cpp std::string`, a `#!cpp std::vector<char>`, or a pointer range to `parse` over a `#!cpp std::istream`. For a
-file, that means reading it into a string first and then parsing the string, rather than passing a
-`#!cpp std::ifstream` directly to `parse` -- the latter never benefits from either optimization:
+Inputs that are still read one character at a time: iterator ranges that are not contiguous (e.g.
+`#!cpp std::list<char>` or `#!cpp std::istreambuf_iterator<char>`), wide strings, and user-defined adapters. An
+iterator pair over non-contiguous but random-access storage (e.g. `#!cpp std::deque<char>::iterator`) gets the first
+optimization but not the second, since the byte-scanning fast path requires contiguous storage.
+
+Practically: a file can be passed directly to `parse`, which also avoids the memory for a copy of the file. Reading it
+into a string first is no longer necessary for speed:
 
 ```cpp
-// gets the contiguous fast paths
+// fast: the stream buffer's get area is scanned in bulk
 std::ifstream f("example.json");
-std::string contents((std::istreambuf_iterator<char>(f)), std::istreambuf_iterator<char>());
-json j = json::parse(contents);
-
-// does not: input_stream_adapter has no fast path
-std::ifstream f2("example.json");
-json j2 = json::parse(f2);
+json j = json::parse(f);
 ```
 
+Two macros switch the stream optimizations off without changing any result:
+[`JSON_DISABLE_STREAM_BULK_SCAN`](../api/macros/json_disable_stream_bulk_scan.md) for stream buffers that do not follow
+the usual `#!cpp std::streambuf` contract, and
+[`JSON_DISABLE_UNLOCKED_FILE_IO`](../api/macros/json_disable_unlocked_file_io.md) for programs in which another thread
+must use the same `#!cpp FILE` object while it is parsed.
+
 For contiguous input with many non-ASCII characters, [`JSON_USE_SIMDUTF`](../api/macros/json_use_simdutf.md) can
 additionally speed up UTF-8 validation by using the [simdutf](https://github.com/simdutf/simdutf) library instead of
 the built-in scalar validator; streaming inputs (files, `#!cpp std::istream`, wide strings, user-defined adapters)

docs/mkdocs/docs/home/architecture.md

@@ -64,7 +64,7 @@ below are relative to `include/nlohmann`.
 |-----------|----------|
 | Value type enumeration | [`detail/value_t.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/value_t.hpp) |
 | Input adapters | [`detail/input/input_adapters.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/input_adapters.hpp) |
-| Lexer | [`detail/input/lexer.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/lexer.hpp), [`detail/input/number_parse.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/number_parse.hpp), [`detail/input/string_scan.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/string_scan.hpp) |
+| Lexer | [`detail/input/lexer.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/lexer.hpp), [`detail/input/number_parse.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/number_parse.hpp), [`detail/input/string_scan.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/string_scan.hpp), [`detail/input/simd_scan.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/simd_scan.hpp) |
 | Parser | [`detail/input/parser.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/parser.hpp) |
 | SAX interface and DOM builders | [`detail/input/json_sax.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/json_sax.hpp) |
 | Binary format readers | [`detail/input/binary_reader.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/binary_reader.hpp) |

docs/mkdocs/docs/home/debugging.md

@@ -10,6 +10,29 @@ at its root, a [Natvis](https://learn.microsoft.com/en-us/visualstudio/debugger/
 file that gives `json`/`ordered_json` values a friendly, key/value debugger view instead of showing raw internal
 fields, when debugging with the MSVC debug engine (`cppvsdbg`) in Visual Studio or VS Code.
 
+The checked-in visualizers cover namespace variants with zero, one, or two ABI tags. Configurations with three or more
+ABI tags are not covered and may show raw internal fields instead. To generate a Natvis file for such a configuration:
+
+1. Check out the release tag of the library version you use.
+2. List the ABI tags and the macros that enable them:
+
+    ```shell
+    make natvis NATVIS_ARGS="--list-tags"
+    ```
+
+3. Generate a Natvis file with the tags of your configuration, for example for `JSON_DIAGNOSTICS`,
+    `JSON_DIAGNOSTIC_POSITIONS`, and `JSON_VIEW_NO_SIMD`:
+
+    ```shell
+    make natvis NATVIS_ARGS="--abi-tags _diag,_dp,_vns --output my_nlohmann_json.natvis"
+    ```
+
+4. Use the generated file instead of `nlohmann_json.natvis`.
+
+`--abi-tags` can be repeated for several configurations, and `--all` generates every combination (about 60 MB for 12
+ABI tags, doubling with each further tag). See
+[`tools/generate_natvis`](https://github.com/nlohmann/json/tree/develop/tools/generate_natvis) for all options.
+
 Debug engines that wrap LLDB instead of the MSVC debug engine (for example, `codelldb` in VS Code) only have
 partial/experimental Natvis support, and commonly fall back to showing raw internal fields even with the
 `.natvis` file present. Switching to `cppvsdbg` where available, or checking your debug extension's own Natvis

docs/mkdocs/mkdocs.yml

@@ -383,6 +383,7 @@ nav:
           - 'JSON_NO_AUTOMATIC_UDLS': api/macros/json_no_automatic_udls.md
           - 'JSON_NOEXCEPTION': api/macros/json_noexception.md
           - 'JSON_NO_IO': api/macros/json_no_io.md
+          - 'JSON_NO_SIMD': api/macros/json_no_simd.md
           - 'JSON_NO_THREAD_LOCAL': api/macros/json_no_thread_local.md
           - 'JSON_PRECISE_STREAM_POSITION': api/macros/json_precise_stream_position.md
           - 'JSON_SKIP_LIBRARY_VERSION_CHECK': api/macros/json_skip_library_version_check.md

include/nlohmann/detail/abi_macros.hpp

@@ -122,6 +122,12 @@
     #define NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO
 #endif
 
+#ifdef JSON_NO_SIMD
+    #define NLOHMANN_JSON_ABI_TAG_NO_SIMD _ns
+#else
+    #define NLOHMANN_JSON_ABI_TAG_NO_SIMD
+#endif
+
 #ifdef JSON_VIEW_NO_SIMD
     #define NLOHMANN_JSON_ABI_TAG_VIEW_NO_SIMD _vns
 #else
@@ -139,9 +145,9 @@
 #endif
 
 // Construct the namespace ABI tags component
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j ## k ## l
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j, k, l) \
-    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l)
+#define NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l, m) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j ## k ## l ## m
+#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l, m)
 
 #define NLOHMANN_JSON_ABI_TAGS                                       \
     NLOHMANN_JSON_ABI_TAGS_CONCAT(                                   \
@@ -155,6 +161,7 @@
             NLOHMANN_JSON_ABI_TAG_OBJECTS_FOR_ENUM_KEYED_MAPS,       \
             NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN,          \
             NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO,          \
+            NLOHMANN_JSON_ABI_TAG_NO_SIMD,                           \
             NLOHMANN_JSON_ABI_TAG_VIEW_NO_SIMD,                      \
             NLOHMANN_JSON_ABI_TAG_VIEW_USE_SSSE3)
 

include/nlohmann/detail/bit_ops.hpp

@@ -14,11 +14,24 @@
     #include <intrin0.h> // __umulh, _umul128, _BitScanForward64, _BitScanReverse64
 #endif
 
+#ifdef _MSC_VER
+    #include <cstdlib> // _byteswap_ushort, _byteswap_ulong, _byteswap_uint64
+#endif
+#if defined(_MSC_VER) && !defined(__clang__) && (defined(_M_IX86) || defined(_M_ARM))
+    #include <intrin.h> // _BitScanForward, _BitScanReverse
+#endif
+
 #include <nlohmann/detail/macro_scope.hpp> // JSON_HEDLEY_ALWAYS_INLINE, NLOHMANN_JSON_NAMESPACE_BEGIN
 
+#if defined(__cpp_lib_bit_cast) || defined(__cpp_lib_bitops) || defined(__cpp_lib_byteswap)
+    #include <bit> // bit_cast, byteswap, countl_zero, countr_zero
+#endif
+
 // Portable bit-level helpers for the number and string scanners. They use
-// compiler builtins or platform-specific intrinsics where available and plain
-// C++ otherwise, so they work regardless of byte order.
+// compiler builtins or platform-specific intrinsics where available, the
+// C++20 <bit> functions next, and plain C++ otherwise, so they work regardless
+// of byte order. (The builtins and intrinsics come first: MSVC's
+// std::countl_zero checks for LZCNT at run time unless AVX2 is enabled.)
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
@@ -33,6 +46,16 @@ inline int count_leading_zeros(std::uint64_t x) noexcept
     unsigned long index = 0; // NOLINT(runtime/int): the type _BitScan*64 takes
     _BitScanReverse64(&index, x);
     return 63 - static_cast<int>(index);
+#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_ARM))
+    unsigned long index = 0; // NOLINT(runtime/int): the type _BitScan* takes
+    if (_BitScanReverse(&index, static_cast<unsigned long>(x >> 32u))) // NOLINT(runtime/int)
+    {
+        return 31 - static_cast<int>(index);
+    }
+    _BitScanReverse(&index, static_cast<unsigned long>(x)); // NOLINT(runtime/int)
+    return 63 - static_cast<int>(index);
+#elif defined(__cpp_lib_bitops)
+    return std::countl_zero(x);
 #else
     int n = 0;
     for (int shift = 32; shift != 0; shift >>= 1)
@@ -56,6 +79,16 @@ inline int count_trailing_zeros(std::uint64_t x) noexcept
     unsigned long index = 0; // NOLINT(runtime/int): the type _BitScan*64 takes
     _BitScanForward64(&index, x);
     return static_cast<int>(index);
+#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_ARM))
+    unsigned long index = 0; // NOLINT(runtime/int): the type _BitScan* takes
+    if (_BitScanForward(&index, static_cast<unsigned long>(x))) // NOLINT(runtime/int)
+    {
+        return static_cast<int>(index);
+    }
+    _BitScanForward(&index, static_cast<unsigned long>(x >> 32u)); // NOLINT(runtime/int)
+    return 32 + static_cast<int>(index);
+#elif defined(__cpp_lib_bitops)
+    return std::countr_zero(x);
 #else
     int n = 0;
     for (int shift = 32; shift != 0; shift >>= 1)
@@ -70,6 +103,67 @@ inline int count_trailing_zeros(std::uint64_t x) noexcept
 #endif
 }
 
+/// the object representation of @a from as a @a To (std::bit_cast in C++20)
+template<typename To, typename From>
+JSON_HEDLEY_ALWAYS_INLINE To bit_cast(const From& from) noexcept
+{
+    static_assert(sizeof(To) == sizeof(From), "bit_cast needs types of the same size");
+#if defined(__cpp_lib_bit_cast)
+    return std::bit_cast<To>(from);
+#else
+    To to {};
+    std::memcpy(&to, &from, sizeof(To));
+    return to;
+#endif
+}
+
+/// x with its bytes in reverse order (std::byteswap in C++23); compilers lower
+/// it to a single bswap/rev/movbe
+inline std::uint16_t byteswap(std::uint16_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_bswap16(x);
+#elif defined(_MSC_VER)
+    return _byteswap_ushort(x);
+#elif defined(__cpp_lib_byteswap)
+    return std::byteswap(x);
+#else
+    return static_cast<std::uint16_t>((x >> 8u) | (x << 8u));
+#endif
+}
+
+/// x with its bytes in reverse order (std::byteswap in C++23)
+inline std::uint32_t byteswap(std::uint32_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_bswap32(x);
+#elif defined(_MSC_VER)
+    return _byteswap_ulong(x);
+#elif defined(__cpp_lib_byteswap)
+    return std::byteswap(x);
+#else
+    return ((x & 0x000000FFu) << 24u) | ((x & 0x0000FF00u) << 8u)
+           | ((x & 0x00FF0000u) >> 8u) | ((x & 0xFF000000u) >> 24u);
+#endif
+}
+
+/// x with its bytes in reverse order (std::byteswap in C++23)
+inline std::uint64_t byteswap(std::uint64_t x) noexcept
+{
+#if defined(__GNUC__) || defined(__clang__)
+    return __builtin_bswap64(x);
+#elif defined(_MSC_VER)
+    return _byteswap_uint64(x);
+#elif defined(__cpp_lib_byteswap)
+    return std::byteswap(x);
+#else
+    x = ((x & 0x00000000FFFFFFFFull) << 32u) | ((x & 0xFFFFFFFF00000000ull) >> 32u);
+    x = ((x & 0x0000FFFF0000FFFFull) << 16u) | ((x & 0xFFFF0000FFFF0000ull) >> 16u);
+    x = ((x & 0x00FF00FF00FF00FFull) << 8u) | ((x & 0xFF00FF00FF00FF00ull) >> 8u);
+    return x;
+#endif
+}
+
 /// the 128-bit product of two 64-bit numbers
 struct uint128_parts
 {
@@ -107,8 +201,8 @@ inline uint128_parts full_multiplication(std::uint64_t a, std::uint64_t b) noexc
 /// targets; always inlined, as GCC otherwise calls it in the number loops)
 JSON_HEDLEY_ALWAYS_INLINE std::uint64_t read_eight_bytes(const unsigned char* b) noexcept
 {
-#if defined(_MSC_VER) || defined(__x86_64__) || defined(__i386__) || (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
-    // the byte order already matches (all MSVC targets are little-endian)
+#if JSON_LITTLE_ENDIAN_TARGET
+    // the byte order already matches
     std::uint64_t result = 0;
     std::memcpy(&result, b, sizeof(result));
     return result;

include/nlohmann/detail/conversions/to_chars.hpp

@@ -18,17 +18,14 @@
 #include <limits> // numeric_limits
 #include <type_traits> // conditional
 
-#ifdef _MSC_VER
-    #include <cstdlib> // _byteswap_uint64
-#endif
 
 // SSE2 (every x86-64 CPU) and NEON (every 64-bit Arm CPU) convert the 16
-// digits of a double at once
-#if defined(__x86_64__) || (defined(_M_X64) && !defined(_M_ARM64EC))
+// digits of a double at once (unless JSON_NO_SIMD selects the portable code)
+#if !defined(JSON_NO_SIMD) && (defined(__x86_64__) || (defined(_M_X64) && !defined(_M_ARM64EC)))
     #include <emmintrin.h>
     #define JSON_DTOA_SSE2 1
     #define JSON_DTOA_NEON 0
-#elif (defined(__aarch64__) || defined(_M_ARM64)) && !defined(_M_ARM64EC) && !defined(__ARM_BIG_ENDIAN)
+#elif !defined(JSON_NO_SIMD) && (defined(__aarch64__) || defined(_M_ARM64)) && !defined(_M_ARM64EC) && !defined(__ARM_BIG_ENDIAN)
     #include <arm_neon.h>
     #define JSON_DTOA_SSE2 0
     #define JSON_DTOA_NEON 1
@@ -37,6 +34,7 @@
     #define JSON_DTOA_NEON 0
 #endif
 
+#include <nlohmann/detail/bit_ops.hpp>
 #include <nlohmann/detail/conversions/zmij.hpp>
 #include <nlohmann/detail/macro_scope.hpp>
 
@@ -66,16 +64,6 @@ For a detailed description of the algorithm see:
 namespace dtoa_impl
 {
 
-template<typename Target, typename Source>
-Target reinterpret_bits(const Source source)
-{
-    static_assert(sizeof(Target) == sizeof(Source), "size mismatch");
-
-    Target target;
-    std::memcpy(&target, &source, sizeof(Source));
-    return target;
-}
-
 struct diyfp // f * 2^e
 {
     static constexpr int kPrecision = 64; // = q
@@ -230,7 +218,7 @@ boundaries compute_boundaries(FloatType value)
 
     using bits_type = typename std::conditional<kPrecision == 24, std::uint32_t, std::uint64_t >::type;
 
-    const auto bits = static_cast<std::uint64_t>(reinterpret_bits<bits_type>(value));
+    const auto bits = static_cast<std::uint64_t>(bit_cast<bits_type>(value));
     const std::uint64_t E = bits >> (kPrecision - 1);
     const std::uint64_t F = bits & (kHiddenBit - 1);
 
@@ -1084,13 +1072,10 @@ inline std::uint64_t eight_digit_bytes(std::uint64_t abcdefgh) noexcept
 /// the byte stores once this is inlined)
 inline void store_msb_first(char* p, std::uint64_t v) noexcept
 {
-#if defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-    v = __builtin_bswap64(v);
+#if JSON_LITTLE_ENDIAN_TARGET
+    v = byteswap(v);
     std::memcpy(p, &v, sizeof(v));
-#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-    std::memcpy(p, &v, sizeof(v));
-#elif defined(_MSC_VER) // (little-endian on all its targets)
-    v = _byteswap_uint64(v);
+#elif JSON_BIG_ENDIAN_TARGET
     std::memcpy(p, &v, sizeof(v));
 #else
     for (unsigned i = 0; i < 8; ++i)
@@ -1462,8 +1447,7 @@ char* write_positive_zmij(char* first, const char* last, FloatType value)
 {
     static_assert(is_binary64<FloatType>::value,
                   "internal error: the conversion of Zmij needs IEEE 754 binary64 numbers");
-    std::uint64_t bits = 0;
-    std::memcpy(&bits, &value, sizeof(bits));
+    const auto bits = bit_cast<std::uint64_t>(value);
     const zmij::shortest_decimal d = zmij::to_shortest(bits);
     if (JSON_HEDLEY_LIKELY(last - first >= 41))
     {

include/nlohmann/detail/input/json_sax.hpp

@@ -418,8 +418,9 @@ class json_sax_dom_parser
         JSON_ASSERT(!ref_stack.empty());
         JSON_ASSERT(ref_stack.back()->is_object());
 
-        // add null at the given key and store the reference for later
-        object_element = &(ref_stack.back()->m_data.m_value.object->operator[](val));
+        // add null at the given key and store the reference for later; the
+        // key is moved, as the lexer starts the next token with reset()
+        object_element = &(ref_stack.back()->m_data.m_value.object->operator[](std::move(val)));
         return true;
     }
 

include/nlohmann/detail/input/lexer.hpp

@@ -12,6 +12,7 @@
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t
 #include <cstdio> // snprintf
+#include <cstring> // memcmp
 #include <initializer_list> // initializer_list
 #include <string> // char_traits, string
 #include <utility> // move
@@ -417,30 +418,43 @@ class lexer : public lexer_base<BasicJsonType>
     }
 
     /// contiguous input: bulk-append the run of ordinary characters and complete
-    /// well-formed UTF-8 sequences starting at the current read position, leaving
-    /// the first byte that needs individual handling (the closing quote, an
-    /// escape, a control character, or an ill-formed UTF-8 byte) for get()
-    void scan_string_bulk(std::true_type /*bulk*/)
+    /// well-formed UTF-8 sequences starting at the current read position, then
+    /// read the first byte that needs individual handling (the closing quote, an
+    /// escape, a control character, or an ill-formed UTF-8 byte) like get().
+    /// That byte is taken from the buffer that was just scanned. Through get(),
+    /// Apple Clang merged this read with the other get() calls of the scan
+    /// function, which load the read position again right after it was stored
+    /// (3% of json::accept() on jeopardy.json on Apple M1).
+    char_int_type get_after_string_run(std::true_type /*bulk*/)
     {
         // a pending unget must be consumed through the normal path first
         if (next_unget)
         {
-            return;
+            return get();
         }
         const std::size_t remaining = ia.bulk_remaining();
         if (remaining == 0)
         {
-            return;
+            return get();
         }
-        const auto* const data = reinterpret_cast<const unsigned char*>(ia.bulk_data());
+        const auto* const raw = ia.bulk_data();
+        const auto* const data = reinterpret_cast<const unsigned char*>(raw);
 
-        const std::size_t pos = string_bulk_run(data, remaining);
-        if (pos == 0)
+        const std::size_t pos = lexer_string_bulk_run(data, remaining);
+        if (pos != 0)
         {
-            return;
+            token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
+            bulk_consume(reinterpret_cast<const char*>(data), pos);
         }
-        token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
-        bulk_consume(reinterpret_cast<const char*>(data), pos);
+        if (pos == remaining)
+        {
+            return get(); // end of the contiguous input
+        }
+        // what get() does, with the byte at hand
+        advance_position();
+        ia.bulk_skip(1);
+        current = char_traits<char_type>::to_int_type(raw[pos]);
+        return track_after_read();
     }
 
     /// consume @a n characters at @a data (the adapter's bulk_data()) and do
@@ -473,7 +487,10 @@ class lexer : public lexer_base<BasicJsonType>
     }
 
     /// streaming input: no bulk fast path
-    void scan_string_bulk(std::false_type /*bulk*/) const noexcept {}
+    char_int_type get_after_string_run(std::false_type /*bulk*/)
+    {
+        return get();
+    }
 
     /*!
     @brief scan a string literal
@@ -501,11 +518,8 @@ class lexer : public lexer_base<BasicJsonType>
         while (true)
         {
             // bulk-consume ordinary characters from contiguous input, then
-            // handle the next special byte through the switch below
-            scan_string_bulk(std::integral_constant<bool, bulk_scan> {});
-
-            // get the next character
-            switch (get())
+            // get the next character, the one that ends the run
+            switch (get_after_string_run(std::integral_constant<bool, bulk_scan>()))
             {
                 // end of file while parsing the string
                 case char_traits<char_type>::eof():
@@ -1625,9 +1639,12 @@ scan_number_done:
         const std::size_t avail = rem + 1;
 
         // validate + classify the number extent (mirrors scan_number()'s grammar)
+        // and accumulate the mantissa digits on the way, so that tokens with up
+        // to 19 digits need no second pass for their value
         std::size_t i = 0;
         std::size_t dot_index = std::string::npos;
         token_type number_type = token_type::value_unsigned;
+        std::uint64_t w = 0;
         if (data[0] == '-')
         {
             number_type = token_type::value_integer;
@@ -1637,22 +1654,21 @@ scan_number_done:
                 return token_type::uninitialized;
             }
         }
+        const std::size_t integer_begin = i;
         if (data[i] == '0')
         {
             ++i;
         }
         else if (data[i] >= '1' && data[i] <= '9')
         {
-            ++i;
-            while (i < avail && data[i] >= '0' && data[i] <= '9')
-            {
-                ++i;
-            }
+            i = accumulate_digits(data, i, avail, w);
         }
         else
         {
             return token_type::uninitialized;
         }
+        std::size_t digit_count = i - integer_begin;
+        std::size_t fraction_digits = 0;
         if (i < avail && data[i] == '.')
         {
             number_type = token_type::value_float;
@@ -1662,29 +1678,39 @@ scan_number_done:
             {
                 return token_type::uninitialized;
             }
-            while (i < avail && data[i] >= '0' && data[i] <= '9')
-            {
-                ++i;
-            }
+            const std::size_t fraction_begin = i;
+            i = accumulate_digits(data, i, avail, w);
+            fraction_digits = i - fraction_begin;
+            digit_count += fraction_digits;
         }
         // the mantissa ends here, whether or not an exponent part follows
         const std::size_t mantissa_end = i;
+        std::int64_t exponent = 0;
         if (i < avail && (data[i] == 'e' || data[i] == 'E'))
         {
             number_type = token_type::value_float;
             ++i;
+            bool negative_exponent = false;
             if (i < avail && (data[i] == '+' || data[i] == '-'))
             {
+                negative_exponent = data[i] == '-';
                 ++i;
             }
             if (i >= avail || !(data[i] >= '0' && data[i] <= '9'))
             {
                 return token_type::uninitialized;
             }
+            // saturated far beyond every range, see parse_float_exponent()
+            constexpr std::int64_t saturation = 100000000000000000; // 10^17
             while (i < avail && data[i] >= '0' && data[i] <= '9')
             {
+                if (exponent < saturation)
+                {
+                    exponent = (exponent * 10) + (data[i] - '0');
+                }
                 ++i;
             }
+            exponent = negative_exponent ? -exponent : exponent;
         }
         const std::size_t len = i;
         if (number_may_continue(len == avail, std::integral_constant<bool, bulk_window> {}))
@@ -1705,7 +1731,10 @@ scan_number_done:
 #if !JSON_DIAGNOSTIC_POSITIONS
         if (number_type != token_type::value_float)
         {
-            const token_type integer_result = convert_integer(number_type, data, data + len);
+            // up to 18 digits always fit into std::uint64_t and std::int64_t
+            const token_type integer_result = (digit_count <= 18)
+                                              ? store_integer(number_type, w)
+                                              : convert_integer(number_type, data, data + len);
             if (JSON_HEDLEY_LIKELY(integer_result != token_type::uninitialized))
             {
                 bulk_consume(data + 1, len - 1);
@@ -1719,15 +1748,42 @@ scan_number_done:
         }
 #endif
 
-        // materialize the token exactly as scan_number() would. reset() already
-        // cleared token_buffer, so append() fills it (assign() is avoided
-        // because custom string_t types need not provide it)
-        token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), len);
         decimal_point_position = dot_index;
 
         bulk_consume(data + 1, len - 1);
         read_number_terminator(std::integral_constant<bool, bulk_window> {});
 
+        // a float whose digits all fit into w converts from the accumulated
+        // value; longer ones (and other float formats) take the general path
+        if (number_type == token_type::value_float && digit_count <= 19)
+        {
+            float_significand s;
+            s.w = w;
+            s.exponent = exponent - static_cast<std::int64_t>(fraction_digits);
+            s.negative = data[0] == '-';
+            if (convert_accumulated_float(s, std::integral_constant<bool, has_native_float_format<number_float_t>::value> {}))
+            {
+                // the token text is only copied into token_buffer when
+                // get_string() asks for it; the library's own SAX handlers
+                // ignore it. A window into a stream may be refilled by the
+                // next read, so there the text is copied right away.
+                if (bulk_window)
+                {
+                    token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), len);
+                }
+                else
+                {
+                    pending_number_text = data;
+                    pending_number_length = len;
+                }
+                return token_type::value_float;
+            }
+        }
+
+        // materialize the token exactly as scan_number() would. reset() already
+        // cleared token_buffer, so append() fills it (assign() is avoided
+        // because custom string_t types need not provide it)
+        token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), len);
         return convert_number(number_type, mantissa_end);
     }
 
@@ -1756,6 +1812,35 @@ scan_number_done:
     /// the input is not handed back, so the terminator can stay unread
     void read_number_terminator(std::false_type /*window*/) const noexcept {}
 
+    /// binary32/binary64: the value of a float token from its accumulated
+    /// significand (see scan_number_bulk_contiguous())
+    bool convert_accumulated_float(const float_significand& s, std::true_type /*native*/) noexcept
+    {
+        value_float = decimal_to_float<number_float_t>(s);
+        return true;
+    }
+
+    /// other formats (e.g. x87 long double): not instantiated for them, as
+    /// decimal_to_float() only supports binary32/binary64
+    static bool convert_accumulated_float(const float_significand& /*s*/, std::false_type /*native*/) noexcept
+    {
+        return false;
+    }
+
+    /// the value of an integer token with at most 18 digits, accumulated in @a w
+    token_type store_integer(token_type number_type, std::uint64_t w) noexcept
+    {
+        if (number_type == token_type::value_unsigned)
+        {
+            value_unsigned = static_cast<number_unsigned_t>(w);
+            // reject values that do not round-trip into a narrower type
+            return (static_cast<std::uint64_t>(value_unsigned) == w) ? token_type::value_unsigned : token_type::uninitialized;
+        }
+        const std::int64_t x = -static_cast<std::int64_t>(w);
+        value_integer = static_cast<number_integer_t>(x);
+        return (static_cast<std::int64_t>(value_integer) == x) ? token_type::value_integer : token_type::uninitialized;
+    }
+
     /// contiguous input: try the number fast path, else the byte-path scanner
     token_type scan_number_dispatch(std::true_type /*bulk*/)
     {
@@ -1770,18 +1855,23 @@ scan_number_done:
     }
 
     /*!
-    @param[in] literal_text  the literal text to expect
-    @param[in] length        the length of the passed literal text
-    @param[in] return_type   the token type to return on success
+    @param[in] rest         the literal text after its first character, which
+                            is current (its length is a template parameter, so
+                            that the comparison of the contiguous fast path is
+                            a fixed-size load, not a call of memcmp)
+    @param[in] return_type  the token type to return on success
     */
-    JSON_HEDLEY_NON_NULL(2)
-    token_type scan_literal(const char_type* literal_text, const std::size_t length,
-                            token_type return_type)
+    template<std::size_t N>
+    JSON_HEDLEY_NEVER_INLINE token_type scan_literal(const char (&rest)[N], token_type return_type) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
     {
-        JSON_ASSERT(char_traits<char_type>::to_char_type(current) == literal_text[0]);
-        for (std::size_t i = 1; i < length; ++i)
+        constexpr std::size_t length = N - 1;
+        if (scan_literal_bulk(rest, length, std::integral_constant<bool, bulk_scan> {}))
         {
-            if (JSON_HEDLEY_UNLIKELY(char_traits<char_type>::to_char_type(get()) != literal_text[i]))
+            return return_type;
+        }
+        for (std::size_t i = 0; i < length; ++i)
+        {
+            if (JSON_HEDLEY_UNLIKELY(char_traits<char_type>::to_char_type(get()) != static_cast<char_type>(rest[i])))
             {
                 error_message = "invalid literal";
                 return token_type::parse_error;
@@ -1790,6 +1880,27 @@ scan_number_done:
         return return_type;
     }
 
+    /// contiguous input: compare the rest of a literal at once; false (and
+    /// nothing consumed) if it does not match, so that the byte path reports
+    /// the error at the right character
+    JSON_HEDLEY_ALWAYS_INLINE bool scan_literal_bulk(const char* rest, const std::size_t length, std::true_type /*bulk*/)
+    {
+        if (next_unget || ia.bulk_remaining() < length
+                || std::memcmp(ia.bulk_data(), rest, length) != 0)
+        {
+            return false;
+        }
+        bulk_consume(reinterpret_cast<const char*>(ia.bulk_data()), length);
+        current = static_cast<char_int_type>(static_cast<unsigned char>(rest[length - 1]));
+        return true;
+    }
+
+    /// streaming input: no bulk fast path
+    bool scan_literal_bulk(const char* /*rest*/, const std::size_t /*length*/, std::false_type /*bulk*/) const noexcept
+    {
+        return false;
+    }
+
     /////////////////////
     // input management
     /////////////////////
@@ -1798,6 +1909,7 @@ scan_number_done:
     void reset() noexcept
     {
         token_buffer.clear();
+        pending_number_text = nullptr;
         decimal_point_position = std::string::npos;
 
 #if JSON_DIAGNOSTIC_POSITIONS
@@ -2057,10 +2169,30 @@ scan_number_done:
     /// return current string value
     string_t& get_string()
     {
+        if (pending_number_text != nullptr)
+        {
+            token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(pending_number_text), pending_number_length);
+            pending_number_text = nullptr;
+        }
         // a number token holds '.' regardless of the locale (#4084)
         return token_buffer;
     }
 
+    /// the value of a string token (key or string), which never has pending
+    /// number text, so get_string()'s check is not needed
+    string_t& get_string_token() noexcept
+    {
+        JSON_ASSERT(pending_number_text == nullptr);
+        return token_buffer;
+    }
+
+    /// the text of a float token for a SAX handler that ignores it (the
+    /// library's own handlers): empty, without copying the token
+    const string_t& get_ignored_number_text() const noexcept
+    {
+        return token_buffer;
+    }
+
     /////////////////////
     // diagnostics
     /////////////////////
@@ -2394,21 +2526,14 @@ scan_number_done:
                 return token_type::value_separator;
 
             // literals
+            // (the text after the first character; no local arrays, which
+            // would give this function a stack canary on every token)
             case 't':
-            {
-                std::array<char_type, 4> true_literal = {{static_cast<char_type>('t'), static_cast<char_type>('r'), static_cast<char_type>('u'), static_cast<char_type>('e')}};
-                return scan_literal(true_literal.data(), true_literal.size(), token_type::literal_true);
-            }
+                return scan_literal("rue", token_type::literal_true);
             case 'f':
-            {
-                std::array<char_type, 5> false_literal = {{static_cast<char_type>('f'), static_cast<char_type>('a'), static_cast<char_type>('l'), static_cast<char_type>('s'), static_cast<char_type>('e')}};
-                return scan_literal(false_literal.data(), false_literal.size(), token_type::literal_false);
-            }
+                return scan_literal("alse", token_type::literal_false);
             case 'n':
-            {
-                std::array<char_type, 4> null_literal = {{static_cast<char_type>('n'), static_cast<char_type>('u'), static_cast<char_type>('l'), static_cast<char_type>('l')}};
-                return scan_literal(null_literal.data(), null_literal.size(), token_type::literal_null);
-            }
+                return scan_literal("ull", token_type::literal_null);
 
             // string
             case '\"':
@@ -2456,6 +2581,11 @@ scan_number_done:
     /// whether the next get() call should just return current
     bool next_unget = false;
 
+    /// the text of the current float token in the input, not yet copied into
+    /// token_buffer (see get_string()), or nullptr
+    const char* pending_number_text = nullptr;
+    std::size_t pending_number_length = 0;
+
     /// the start position of the current token
     position_t position {};
 

include/nlohmann/detail/input/number_parse.hpp

@@ -16,7 +16,6 @@
 #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 <type_traits> // conditional, integral_constant, is_signed, true_type, false_type
@@ -275,13 +274,77 @@ JSON_HEDLEY_ALWAYS_INLINE std::uint32_t parse_eight_digits(std::uint64_t v) noex
     return static_cast<std::uint32_t>(((v & 0x0000FFFF0000FFFFu) * 42949672960001u) >> 32u);
 }
 
-/// whether [first, last) contains a digit other than '0'
-inline bool has_nonzero_digit(const char* first, const char* last) noexcept
+/// whether the eight bytes in @a v (see read_eight_bytes()) are all ASCII
+/// digits (after fast_float)
+JSON_HEDLEY_ALWAYS_INLINE bool is_eight_digits(std::uint64_t v) noexcept
 {
-    return std::any_of(first, last, [](char c) noexcept
+    return ((v & 0xF0F0F0F0F0F0F0F0u) | (((v + 0x0606060606060606u) & 0xF0F0F0F0F0F0F0F0u) >> 4u)) == 0x3333333333333333u;
+}
+
+/*!
+@brief skip the ASCII digits starting at data[i] and accumulate their value
+
+Eight digits are taken at a time while at least eight bytes remain. @a w wraps
+around after 19 digits; the caller only uses it for shorter digit runs.
+
+@return the index of the first byte at or after @a i that is not a digit
+*/
+JSON_HEDLEY_ALWAYS_INLINE std::size_t accumulate_digits(const char* data, std::size_t i, std::size_t avail, std::uint64_t& w) noexcept
+{
+    while (avail - i >= 8)
     {
-        return c != '0';
-    });
+        const std::uint64_t v = read_eight_bytes(data + i);
+        if (!is_eight_digits(v))
+        {
+            break;
+        }
+        w = (w * 100000000u) + parse_eight_digits(v);
+        i += 8;
+    }
+    while (i < avail && data[i] >= '0' && data[i] <= '9')
+    {
+        w = (w * 10u) + static_cast<std::uint64_t>(data[i] - '0');
+        ++i;
+    }
+    return i;
+}
+
+/// whether [first, last) contains a digit other than '0'; a plain loop, as
+/// GCC calls std::any_of out of line even for the usual empty range
+JSON_HEDLEY_ALWAYS_INLINE bool has_nonzero_digit(const char* first, const char* last) noexcept
+{
+    for (; first != last; ++first)
+    {
+        if (*first != '0')
+        {
+            return true;
+        }
+    }
+    return false;
+}
+
+/*!
+@brief accumulate the validated digits [p, last) into @a w, as long as
+       @a remaining digits still fit into it
+
+Eight digits are taken at a time while at least eight remain, and @a remaining
+is decreased by the number of digits taken.
+
+@return a pointer past the last digit taken
+*/
+JSON_HEDLEY_ALWAYS_INLINE const char* accumulate_significant_digits(const char* p, const char* last, int& remaining, std::uint64_t& w) noexcept
+{
+    while (remaining >= 8 && last - p >= 8)
+    {
+        w = (w * 100000000u) + parse_eight_digits(read_eight_bytes(p));
+        p += 8;
+        remaining -= 8;
+    }
+    for (; remaining > 0 && p != last; ++p, --remaining)
+    {
+        w = (w * 10u) + static_cast<std::uint64_t>(*p - '0');
+    }
+    return p;
 }
 
 /// the value of the validated exponent digits [+-]?[0-9]+ in [first, last),
@@ -327,7 +390,7 @@ truncated is set.
                                    std::string::npos if there is none
 @param[in] mantissa_end            index of the 'e'/'E', or the token length
 */
-inline float_significand parse_float_significand(const char* first, const char* last,
+JSON_HEDLEY_ALWAYS_INLINE float_significand parse_float_significand(const char* first, const char* last,
         std::size_t decimal_point_position, std::size_t mantissa_end) noexcept
 {
     float_significand s;
@@ -342,16 +405,7 @@ inline float_significand parse_float_significand(const char* first, const char*
     int remaining = 19; // digits that still fit into w
     if (*p != '0') // the integer part is "0" or [1-9][0-9]*
     {
-        while (remaining >= 8 && integer_last - p >= 8)
-        {
-            w = (w * 100000000u) + parse_eight_digits(read_eight_bytes(p));
-            p += 8;
-            remaining -= 8;
-        }
-        for (; remaining > 0 && p != integer_last; ++p, --remaining)
-        {
-            w = (w * 10u) + static_cast<std::uint64_t>(*p - '0');
-        }
+        p = accumulate_significant_digits(p, integer_last, remaining, w);
         s.exponent = integer_last - p;
         s.truncated = has_nonzero_digit(p, integer_last);
     }
@@ -370,16 +424,7 @@ inline float_significand parse_float_significand(const char* first, const char*
             s.exponent -= p - zeros;
         }
         const char* const digits = p;
-        while (remaining >= 8 && mantissa_last - p >= 8)
-        {
-            w = (w * 100000000u) + parse_eight_digits(read_eight_bytes(p));
-            p += 8;
-            remaining -= 8;
-        }
-        for (; remaining > 0 && p != mantissa_last; ++p, --remaining)
-        {
-            w = (w * 10u) + static_cast<std::uint64_t>(*p - '0');
-        }
+        p = accumulate_significant_digits(p, mantissa_last, remaining, w);
         s.exponent -= p - digits;
         s.truncated = s.truncated || has_nonzero_digit(p, mantissa_last);
     }
@@ -712,15 +757,14 @@ std::uint64_t digit_comparison(const char* first, const char* last, std::uint64_
 inline void float_from_bits(std::uint64_t bits, double& value) noexcept
 {
     static_assert(sizeof(double) == sizeof(std::uint64_t), "double must have 64 bits");
-    std::memcpy(&value, &bits, sizeof(value));
+    value = bit_cast<double>(bits);
 }
 
 /// the float with the IEEE-754 bits @a bits (the lower 32)
 inline void float_from_bits(std::uint64_t bits, float& value) noexcept
 {
     static_assert(sizeof(float) == sizeof(std::uint32_t), "float must have 32 bits");
-    const auto bits32 = static_cast<std::uint32_t>(bits);
-    std::memcpy(&value, &bits32, sizeof(bits32));
+    value = bit_cast<float>(static_cast<std::uint32_t>(bits));
 }
 
 /// the powers of ten that are exact in binary64 (up to 10^22)

include/nlohmann/detail/input/parser.hpp

@@ -51,6 +51,20 @@ template<typename BasicJsonType>
 using parser_callback_t =
     std::function<bool(int /*depth*/, parse_event_t /*event*/, BasicJsonType& /*parsed*/)>;
 
+/// whether a SAX handler ignores the text of float tokens (the library's own
+/// handlers do), so that the parser need not materialize it
+template<typename SAX>
+struct sax_ignores_number_text : std::false_type {};
+
+template<typename BasicJsonType, typename InputAdapterType>
+struct sax_ignores_number_text<json_sax_dom_parser<BasicJsonType, InputAdapterType>> : std::true_type {};
+
+template<typename BasicJsonType, typename InputAdapterType>
+struct sax_ignores_number_text<json_sax_dom_callback_parser<BasicJsonType, InputAdapterType>> : std::true_type {};
+
+template<typename BasicJsonType>
+struct sax_ignores_number_text<json_sax_acceptor<BasicJsonType>> : std::true_type {};
+
 /*!
 @brief syntax analysis
 
@@ -171,6 +185,23 @@ class parser
     }
 
   private:
+    /// the text of the current float token for @a sax (see sax_ignores_number_text)
+    template<typename SAX>
+    const string_t& number_float_text(SAX* /*sax*/)
+    {
+        return number_float_text(std::integral_constant<bool, sax_ignores_number_text<SAX>::value> {});
+    }
+
+    const string_t& number_float_text(std::true_type /*ignored*/) const noexcept
+    {
+        return m_lexer.get_ignored_number_text();
+    }
+
+    string_t& number_float_text(std::false_type /*ignored*/)
+    {
+        return m_lexer.get_string();
+    }
+
     /*!
     @brief run a DOM SAX parser to completion and position the lexer
 
@@ -213,8 +244,9 @@ class parser
     bool sax_parse_internal(SAX* sax)
     {
         // stack to remember the hierarchy of structured values we are parsing
-        // true = array; false = object
-        std::vector<bool> states;
+        // true = array; false = object (bytes, not std::vector<bool>, whose
+        // push_back() is a call with GCC)
+        std::vector<std::uint8_t> states;
         // value to avoid a goto (see comment where set to true)
         bool skip_to_state_evaluation = false;
 
@@ -247,7 +279,7 @@ class parser
                         {
                             return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
                         }
-                        if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string())))
+                        if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                         {
                             return false;
                         }
@@ -259,7 +291,7 @@ class parser
                         }
 
                         // remember we are now inside an object
-                        states.push_back(false);
+                        states.push_back(0);
 
                         // parse values
                         get_token();
@@ -284,7 +316,7 @@ class parser
                         }
 
                         // remember we are now inside an array
-                        states.push_back(true);
+                        states.push_back(1);
 
                         // parse values (no need to call get_token)
                         continue;
@@ -301,7 +333,7 @@ class parser
                                                     out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr));
                         }
 
-                        if (JSON_HEDLEY_UNLIKELY(!sax->number_float(res, m_lexer.get_string())))
+                        if (JSON_HEDLEY_UNLIKELY(!sax->number_float(res, number_float_text(sax))))
                         {
                             return false;
                         }
@@ -347,7 +379,7 @@ class parser
 
                     case token_type::value_string:
                     {
-                        if (JSON_HEDLEY_UNLIKELY(!sax->string(m_lexer.get_string())))
+                        if (JSON_HEDLEY_UNLIKELY(!sax->string(m_lexer.get_string_token())))
                         {
                             return false;
                         }
@@ -401,7 +433,7 @@ class parser
                 return true;
             }
 
-            if (states.back())  // array
+            if (states.back() != 0)  // array
             {
                 // comma -> next value
                 // or end of array (ignore_trailing_commas = true)
@@ -456,7 +488,7 @@ class parser
                         return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
                     }
 
-                    if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string())))
+                    if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                     {
                         return false;
                     }

include/nlohmann/detail/input/simd_scan.hpp

@@ -0,0 +1,321 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-FileCopyrightText: 2018-2025 The simdjson authors <https://github.com/simdjson/simdjson>
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <array> // array
+#include <atomic> // atomic
+#include <cstddef> // size_t
+#include <cstdint> // uint8_t, uint64_t
+
+#include <nlohmann/detail/bit_ops.hpp>
+#include <nlohmann/detail/macro_scope.hpp>
+
+// Vector code for runs of string bytes, used by the lexer for contiguous input
+// (see string_bulk_run()). NEON (AArch64) and SSE2 (x86-64) belong to the
+// baseline instruction sets and are used by default. The vector UTF-8 check
+// needs NEON or SSSE3. SSSE3 is not part of x86-64, and the code must not
+// depend on the flags of a translation unit (two translation units with
+// different flags would have different definitions of the same inline
+// functions): the check is compiled for SSSE3 with a function attribute and
+// used where the CPU has SSSE3 (all x86-64 CPUs since about 2011), else the
+// portable check. Code compiled for SSSE3 anyway (-mssse3 or more) skips the
+// CPU check. JSON_NO_SIMD selects the portable code everywhere.
+// NLOHMANN_JSON_NO_TARGET_ATTRIBUTE (for compilers that ignore the target
+// attribute and then reject the SSSE3 intrinsics) keeps the check portable
+// without disabling the SSE2 code.
+#if !defined(JSON_NO_SIMD) && defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__)) && defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    #include <arm_neon.h>
+    #define JSON_SCAN_NEON 1
+#else
+    #define JSON_SCAN_NEON 0
+#endif
+// (x86 only: other targets can define __SSE2__ as well, e.g., WebAssembly with -msse2, but have no <cpuid.h>)
+#if !defined(JSON_NO_SIMD) && !JSON_SCAN_NEON && (defined(__x86_64__) || defined(__i386__) || defined(_M_X64) || defined(_M_IX86)) && !defined(_M_ARM64EC) && (defined(__SSE2__) || defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 2))
+    #include <emmintrin.h>
+    #define JSON_SCAN_SSE2 1
+#else
+    #define JSON_SCAN_SSE2 0
+#endif
+#if JSON_SCAN_SSE2 && defined(__SSSE3__)
+    #include <tmmintrin.h>
+    #define JSON_SCAN_SSSE3 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+#else
+    #define JSON_SCAN_SSSE3 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+#endif
+#if JSON_SCAN_SSE2 && !JSON_SCAN_SSSE3 && !defined(NLOHMANN_JSON_NO_TARGET_ATTRIBUTE) && ((defined(__clang__) && __clang_major__ >= 4) || (defined(__GNUC__) && !defined(__clang__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9)) && !defined(__cpp_modules)))
+    // (GCC before 4.9 has no SSSE3 intrinsics without -mssse3; GCC with C++20
+    // modules drops the target attribute from functions it reads from a module
+    // interface and then rejects the intrinsics: "needs isa option -mssse3")
+    // <cpuid.h> of GCC before 11 has no include guard, and json.hpp and
+    // json_view.hpp both include it
+    #ifndef NLOHMANN_JSON_CPUID_INCLUDED
+        #define NLOHMANN_JSON_CPUID_INCLUDED
+        #include <cpuid.h>
+    #endif
+    #include <tmmintrin.h>
+    #define JSON_SCAN_SSSE3_DISPATCH 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_SCAN_SSSE3_TARGET __attribute__((target("ssse3")))
+#elif JSON_SCAN_SSE2 && !JSON_SCAN_SSSE3 && defined(_MSC_VER)
+    // (MSVC compiles intrinsics of any instruction set)
+    #include <intrin.h>
+    #include <tmmintrin.h>
+    #define JSON_SCAN_SSSE3_DISPATCH 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_SCAN_SSSE3_TARGET
+#else
+    #define JSON_SCAN_SSSE3_DISPATCH 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_SCAN_SSSE3_TARGET
+#endif
+#define JSON_SCAN_VECTOR (JSON_SCAN_NEON || JSON_SCAN_SSE2)
+#define JSON_SCAN_VECTOR_UTF8 (JSON_SCAN_NEON || JSON_SCAN_SSSE3 || JSON_SCAN_SSSE3_DISPATCH)
+
+NLOHMANN_JSON_NAMESPACE_BEGIN
+namespace detail
+{
+
+#if JSON_SCAN_VECTOR
+/*!
+@brief the first byte of a string run that is a quote, a backslash, a control
+character, or not ASCII, 16 bytes per step
+
+Stops at such a byte, or where fewer than 16 bytes are left (the caller tells
+the two apart). A signed compare with 0x20 finds control characters and
+non-ASCII bytes at once.
+*/
+JSON_HEDLEY_ALWAYS_INLINE const unsigned char* vector_plain_run(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (e - p >= 16)
+    {
+#if JSON_SCAN_NEON
+        const uint8x16_t in = vld1q_u8(p);
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))),
+                                            vcltq_s8(vreinterpretq_s8_u8(in), vdupq_n_s8(0x20)));
+        // one nibble per byte (the usual NEON replacement of x86's movemask, see
+        // D. Kutenin, "Porting x86 vector bitmask optimizations to Arm NEON", 2022)
+        const std::uint64_t bits = vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(special), 4)), 0);
+        if (bits != 0)
+        {
+            return p + (count_trailing_zeros(bits) >> 2u);
+        }
+#else
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                             _mm_cmplt_epi8(in, _mm_set1_epi8(0x20)));
+        const auto bits = static_cast<std::uint64_t>(static_cast<unsigned>(_mm_movemask_epi8(special)));
+        if (bits != 0)
+        {
+            return p + count_trailing_zeros(bits);
+        }
+#endif
+        p += 16;
+    }
+    return p;
+}
+#endif
+
+#if JSON_SCAN_VECTOR
+/// vector_plain_run() for runs that are already longer than a few words:
+/// inline with SSE2, out of line with NEON, where a larger string loop slows
+/// down the rest of the lexer
+#if JSON_SCAN_NEON
+    JSON_HEDLEY_NEVER_INLINE inline
+#else
+    JSON_HEDLEY_ALWAYS_INLINE
+#endif
+const unsigned char* vector_plain_run_long(const unsigned char* p, const unsigned char* e) noexcept
+{
+    return vector_plain_run(p, e);
+}
+#endif
+
+#if JSON_SCAN_SSSE3_DISPATCH
+/// whether the CPU has SSSE3 (CPUID leaf 1, ECX bit 9)
+inline bool cpu_ssse3() noexcept
+{
+#if defined(_MSC_VER) && !defined(__clang__)
+    std::array<int, 4> regs {{}};
+    __cpuid(regs.data(), 1);
+    return (static_cast<unsigned>(regs[2]) & (1u << 9u)) != 0;
+#else
+    unsigned eax = 0;
+    unsigned ebx = 0;
+    unsigned ecx = 0;
+    unsigned edx = 0;
+    return __get_cpuid(1, &eax, &ebx, &ecx, &edx) != 0 && (ecx & (1u << 9u)) != 0;
+#endif
+}
+
+/// whether the CPU has SSSE3, asked once: the answer is kept in an atomic
+/// that is initialized at compile time, so that neither a guard of a local
+/// static nor a global constructor is needed (threads that ask at the same
+/// time all store the same answer)
+JSON_HEDLEY_ALWAYS_INLINE bool cpu_has_ssse3() noexcept
+{
+    static std::atomic<int> known{0}; // 0: not asked yet, 1: no, 2: yes
+    int state = known.load(std::memory_order_relaxed);
+    if (JSON_HEDLEY_UNLIKELY(state == 0))
+    {
+        state = cpu_ssse3() ? 2 : 1;
+        known.store(state, std::memory_order_relaxed);
+    }
+    return state == 2;
+}
+#endif
+
+#if JSON_SCAN_VECTOR_UTF8
+/// Tables of the UTF-8 check of J. Keiser and D. Lemire, "Validating UTF-8 In
+/// Less Than One Instruction Per Byte" (2021), as in simdjson ("lookup4"): each
+/// maps a nibble (high and low nibble of the previous byte, high nibble of the
+/// current byte) to the errors it allows; a byte pair is ill-formed if all
+/// three have an error bit in common.
+template<typename Dummy = void>
+struct utf8_lookup4
+{
+    static constexpr std::uint8_t too_short = 1u << 0u, too_long = 1u << 1u, overlong_3 = 1u << 2u, too_large = 1u << 3u;
+    static constexpr std::uint8_t surrogate = 1u << 4u, overlong_2 = 1u << 5u, too_large_1000 = 1u << 6u, overlong_4 = 1u << 6u;
+    static constexpr std::uint8_t two_conts = 1u << 7u, carry = too_short | too_long | two_conts;
+    static const std::array<std::uint8_t, 16> byte_1_high;
+    static const std::array<std::uint8_t, 16> byte_1_low;
+    static const std::array<std::uint8_t, 16> byte_2_high;
+};
+
+template<typename Dummy>
+const std::array<std::uint8_t, 16> utf8_lookup4<Dummy>::byte_1_high =
+{
+    {
+        too_long, too_long, too_long, too_long, too_long, too_long, too_long, too_long,
+        two_conts, two_conts, two_conts, two_conts,
+        too_short | overlong_2, too_short, too_short | overlong_3 | surrogate, too_short | too_large | too_large_1000 | overlong_4
+    }
+};
+
+template<typename Dummy>
+const std::array<std::uint8_t, 16> utf8_lookup4<Dummy>::byte_1_low =
+{
+    {
+        carry | overlong_3 | overlong_2 | overlong_4, carry | overlong_2, carry, carry,
+        carry | too_large, carry | too_large | too_large_1000, carry | too_large | too_large_1000, carry | too_large | too_large_1000,
+        carry | too_large | too_large_1000, carry | too_large | too_large_1000, carry | too_large | too_large_1000, carry | too_large | too_large_1000,
+        carry | too_large | too_large_1000, carry | too_large | too_large_1000 | surrogate, carry | too_large | too_large_1000, carry | too_large | too_large_1000
+    }
+};
+
+template<typename Dummy>
+const std::array<std::uint8_t, 16> utf8_lookup4<Dummy>::byte_2_high =
+{
+    {
+        too_short, too_short, too_short, too_short, too_short, too_short, too_short, too_short,
+        static_cast<std::uint8_t>(too_long | overlong_2 | two_conts | overlong_3 | too_large_1000 | overlong_4),
+        static_cast<std::uint8_t>(too_long | overlong_2 | two_conts | overlong_3 | too_large),
+        static_cast<std::uint8_t>(too_long | overlong_2 | two_conts | surrogate | too_large),
+        static_cast<std::uint8_t>(too_long | overlong_2 | two_conts | surrogate | too_large),
+        too_short, too_short, too_short, too_short
+    }
+};
+
+/*!
+@brief the run of plain string bytes and well-formed UTF-8 at p (a character
+       boundary), 16 bytes per step
+
+The first quote, backslash, or control character is found with vector
+compares, and the UTF-8 check of J. Keiser and D. Lemire covers the bytes up
+to it. Returns true with @a block at that byte; returns false with @a block at
+the start of the 16 bytes where the check found ill-formed UTF-8 or where
+fewer than 16 bytes are left, for the caller to finish one sequence at a time.
+Out of line, so that no constants of the check occupy registers in the
+caller's loop. On x86-64, it is compiled for SSSE3 (see cpu_has_ssse3()).
+*/
+JSON_SCAN_SSSE3_TARGET JSON_HEDLEY_NEVER_INLINE inline bool vector_utf8_run(const unsigned char* p, const unsigned char* e, const unsigned char*& block_out) noexcept
+{
+    using lookup = utf8_lookup4<>;
+    const unsigned char* block = p;
+#if JSON_SCAN_NEON
+    const uint8x16_t t1h = vld1q_u8(lookup::byte_1_high.data());
+    const uint8x16_t t1l = vld1q_u8(lookup::byte_1_low.data());
+    const uint8x16_t t2h = vld1q_u8(lookup::byte_2_high.data());
+    uint8x16_t prev = vdupq_n_u8(0);
+    while (e - block >= 16)
+    {
+        const uint8x16_t in = vld1q_u8(block);
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))), vcltq_u8(in, vdupq_n_u8(0x20)));
+        const uint8x16_t prev1 = vextq_u8(prev, in, 15);
+        const uint8x16_t sc = vandq_u8(vandq_u8(vqtbl1q_u8(t1h, vshrq_n_u8(prev1, 4)), vqtbl1q_u8(t1l, vandq_u8(prev1, vdupq_n_u8(0x0F)))), vqtbl1q_u8(t2h, vshrq_n_u8(in, 4)));
+        const uint8x16_t must23 = vorrq_u8(vqsubq_u8(vextq_u8(prev, in, 14), vdupq_n_u8(0xE0 - 0x80)), vqsubq_u8(vextq_u8(prev, in, 13), vdupq_n_u8(0xF0 - 0x80)));
+        const uint8x16_t err = veorq_u8(vandq_u8(must23, vdupq_n_u8(0x80)), sc);
+        const std::uint64_t special_bits = vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(special), 4)), 0);
+        const std::uint64_t err_bits = vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(vtstq_u8(err, err)), 4)), 0);
+        if (special_bits != 0)
+        {
+            // errors up to the special byte count (an incomplete sequence
+            // before a quote shows at the quote); the bytes after it do not
+            const unsigned k = static_cast<unsigned>(count_trailing_zeros(special_bits)) >> 2u;
+            const std::uint64_t upto = k == 15 ? ~std::uint64_t{0} :
+                                       (std::uint64_t{1} << (4u * (k + 1u))) - 1u;
+            if ((err_bits & upto) == 0)
+            {
+                block_out = block + k;
+                return true;
+            }
+            break;
+        }
+        if (err_bits != 0)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#else
+    // the same with SSSE3 (pshufb for the table lookups; nibbles from 16-bit
+    // shifts, as there are no byte shifts)
+    const __m128i t1h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t1l = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_1_low.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i t2h = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(lookup::byte_2_high.data()))); // NOLINT(bugprone-casting-through-void)
+    const __m128i nibble = _mm_set1_epi8(0x0F);
+    const __m128i zero = _mm_setzero_si128();
+    __m128i prev = zero;
+    while (e - block >= 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(block))); // NOLINT(bugprone-casting-through-void)
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                             _mm_cmpeq_epi8(_mm_subs_epu8(in, _mm_set1_epi8(0x1F)), zero)); // in < 0x20
+        const __m128i prev1 = _mm_alignr_epi8(in, prev, 15);
+        const __m128i sc = _mm_and_si128(_mm_and_si128(_mm_shuffle_epi8(t1h, _mm_and_si128(_mm_srli_epi16(prev1, 4), nibble)),
+                                         _mm_shuffle_epi8(t1l, _mm_and_si128(prev1, nibble))),
+                                         _mm_shuffle_epi8(t2h, _mm_and_si128(_mm_srli_epi16(in, 4), nibble)));
+        const __m128i must23 = _mm_or_si128(_mm_subs_epu8(_mm_alignr_epi8(in, prev, 14), _mm_set1_epi8(0xE0 - 0x80)),
+                                            _mm_subs_epu8(_mm_alignr_epi8(in, prev, 13), _mm_set1_epi8(0xF0 - 0x80)));
+        const __m128i err = _mm_xor_si128(_mm_and_si128(must23, _mm_set1_epi8(static_cast<char>(-128))), sc);
+        const auto special_bits = static_cast<unsigned>(_mm_movemask_epi8(special));
+        const auto err_bits = ~static_cast<unsigned>(_mm_movemask_epi8(_mm_cmpeq_epi8(err, zero))) & 0xFFFFu;
+        if (special_bits != 0)
+        {
+            const auto k = static_cast<unsigned>(count_trailing_zeros(static_cast<std::uint64_t>(special_bits)));
+            if ((err_bits & ((2u << k) - 1u)) == 0)
+            {
+                block_out = block + k;
+                return true;
+            }
+            break;
+        }
+        if (err_bits != 0)
+        {
+            break;
+        }
+        prev = in;
+        block += 16;
+    }
+#endif
+    block_out = block;
+    return false;
+}
+#endif
+
+}  // namespace detail
+NLOHMANN_JSON_NAMESPACE_END

include/nlohmann/detail/input/string_scan.hpp

@@ -14,6 +14,7 @@
 #include <cstring> // memcpy
 
 #include <nlohmann/detail/bit_ops.hpp>
+#include <nlohmann/detail/input/simd_scan.hpp>
 #include <nlohmann/detail/macro_scope.hpp>
 
 // Optional SIMD backend for bulk UTF-8 validation. This is an opt-in external
@@ -34,7 +35,7 @@
 // This file contains the byte-level string-scanning helpers used by the lexer's
 // contiguous fast path. They operate purely on raw bytes (no dependency on the
 // lexer's template parameters) so they are free functions, keeping the lexer
-// itself focused on the state machine; see lexer::scan_string_bulk().
+// itself focused on the state machine; see lexer::get_after_string_run().
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
@@ -66,7 +67,7 @@ inline std::uint64_t swar_string_special(std::uint64_t v) noexcept
 
 // return the index of the first is_string_special() byte in [data, data+n), or
 // n if every byte is ordinary; scans 8 bytes at a time
-inline std::size_t find_string_special(const unsigned char* data, std::size_t n) noexcept
+JSON_HEDLEY_ALWAYS_INLINE std::size_t find_string_special(const unsigned char* data, std::size_t n) noexcept
 {
     std::size_t i = 0;
     for (; i + 8 <= n; i += 8)
@@ -235,35 +236,143 @@ inline std::size_t valid_utf8_prefix(const unsigned char* data, std::size_t n) n
     return pos;
 }
 
-// Scalar (C++11) computation of the bulk run length: the number of leading
-// bytes in [data, data+n) that are ordinary ASCII or complete well-formed UTF-8
+#if JSON_SCAN_VECTOR_UTF8
+/// the end of scan_string_vector from block, where the vector loop stopped
+/// (ill-formed UTF-8, or fewer than 16 bytes left): one byte or sequence at a
+/// time, from the start of a sequence that crosses into the block
+inline const unsigned char* scan_string_finish(const unsigned char* p, const unsigned char* block, const unsigned char* e) noexcept
+{
+    for (int i = 1; i <= 3 && block - i >= p; ++i)
+    {
+        const unsigned char c = block[-i];
+        if (c < 0x80)
+        {
+            break;
+        }
+        if (c >= 0xC0)
+        {
+            const int len = 2 + static_cast<int>(c >= 0xE0) + static_cast<int>(c >= 0xF0);
+            if (len > i)
+            {
+                block -= i;
+            }
+            break;
+        }
+    }
+    for (p = block; p != e;)
+    {
+        if (*p < 0x80)
+        {
+            if (is_string_special(*p))
+            {
+                return p;
+            }
+            ++p;
+            continue;
+        }
+        const std::size_t n = validate_one_utf8(p, static_cast<std::size_t>(e - p));
+        if (n == 0)
+        {
+            return p;
+        }
+        p += n;
+    }
+    return p;
+}
+
+/// the rest of a string from p (a character boundary), with the vector UTF-8
+/// check and the scalar check for the end of the block where it stopped
+inline const unsigned char* scan_string_vector(const unsigned char* p, const unsigned char* e) noexcept
+{
+    const unsigned char* block = p;
+    if (vector_utf8_run(p, e, block))
+    {
+        return block;
+    }
+    return scan_string_finish(p, block, e);
+}
+#endif
+
+#if JSON_SCAN_VECTOR
+/// the end of the run of plain string characters at @a p: the first quote,
+/// backslash, control character, or non-ASCII byte, or @a e. The first words
+/// are checked with SWAR (one; four with NEON, where the vector mask costs
+/// more), as most keys and short values end there; longer runs continue 16
+/// bytes at a time.
+JSON_HEDLEY_ALWAYS_INLINE const unsigned char* plain_run_end(const unsigned char* p, const unsigned char* e) noexcept
+{
+    constexpr int prefix_words = JSON_SCAN_NEON ? 4 : 1;
+    for (int words = prefix_words; words > 0 && e - p >= 8; --words)
+    {
+        const std::uint64_t special = swar_string_special(read_eight_bytes(p));
+        if (special != 0)
+        {
+            return p + (static_cast<std::size_t>(count_trailing_zeros(special)) / 8);
+        }
+        p += 8;
+    }
+    // the vector loop only for 16 bytes or more: the serializer passes a
+    // string without its closing quote, so most of its runs end at e, and a
+    // call of the out-of-line NEON loop for a shorter rest made dump() up to
+    // 10% slower
+    if (e - p >= 16)
+    {
+        p = vector_plain_run_long(p, e);
+        if (e - p >= 16)
+        {
+            return p; // the vector loop stopped at a special byte
+        }
+    }
+    return p + find_string_special(p, static_cast<std::size_t>(e - p));
+}
+#endif
+
+// Computation of the bulk run length: the number of leading bytes in
+// [data, data+n) that are ordinary ASCII or complete well-formed UTF-8
 // sequences, stopping before the first byte that needs individual handling (the
 // closing quote, an escape, a control character, or an ill-formed/truncated
-// sequence). ASCII is skipped 8 bytes at a time.
-inline std::size_t scalar_string_bulk_run(const unsigned char* data, std::size_t n) noexcept
+// sequence). ASCII is skipped 16 bytes at a time with SSE2 and 8 bytes at a
+// time otherwise; runs of non-ASCII text use the vector UTF-8 check where
+// available.
+JSON_HEDLEY_ALWAYS_INLINE std::size_t scalar_string_bulk_run(const unsigned char* data, std::size_t n) noexcept
 {
-    std::size_t pos = 0;
-    while (pos < n)
+    const unsigned char* const e = data + n;
+    const unsigned char* p = data;
+    while (p != e)
     {
-        pos += find_string_special(data + pos, n - pos);
-        if (pos >= n || data[pos] < 0x80u)
+#if JSON_SCAN_VECTOR
+        p = plain_run_end(p, e);
+#else
+        p += find_string_special(p, static_cast<std::size_t>(e - p));
+#endif
+        if (p == e || *p < 0x80u)
         {
             break; // end of buffer, or a quote/escape/control byte
         }
+#if JSON_SCAN_VECTOR_UTF8
+#if JSON_SCAN_SSSE3_DISPATCH
+        if (JSON_HEDLEY_LIKELY(cpu_has_ssse3()))
+#endif
+        {
+            return static_cast<std::size_t>(scan_string_vector(p, e) - data);
+        }
+#endif
+#if !JSON_SCAN_VECTOR_UTF8 || JSON_SCAN_SSSE3_DISPATCH
         // a run of multi-byte sequences (e.g. CJK text) is validated sequence
         // by sequence without searching for the next special byte in between
         do
         {
-            const std::size_t seq = validate_one_utf8(data + pos, n - pos);
+            const std::size_t seq = validate_one_utf8(p, static_cast<std::size_t>(e - p));
             if (seq == 0)
             {
-                return pos; // ill-formed or truncated: let the byte path diagnose it
+                return static_cast<std::size_t>(p - data); // ill-formed or truncated: let the byte path diagnose it
             }
-            pos += seq;
+            p += seq;
         }
-        while (pos < n && data[pos] >= 0x80u);
+        while (p != e && *p >= 0x80u);
+#endif
     }
-    return pos;
+    return static_cast<std::size_t>(p - data);
 }
 
 #if defined(JSON_USE_SIMDUTF) && defined(JSON_HAS_CPP_17)
@@ -304,7 +413,7 @@ inline std::size_t find_string_delimiter(const unsigned char* data, std::size_t
 // next delimiter is validated in one shot by simdutf; on the rare failure the
 // scalar helper recomputes the exact valid prefix so the byte path still
 // produces the precise diagnostic. Without it, the pure scalar path is used.
-inline std::size_t string_bulk_run(const unsigned char* data, std::size_t n) noexcept
+JSON_HEDLEY_ALWAYS_INLINE std::size_t string_bulk_run(const unsigned char* data, std::size_t n) noexcept
 {
 #if defined(JSON_USE_SIMDUTF) && defined(JSON_HAS_CPP_17)
     const std::size_t run = find_string_delimiter(data, n);
@@ -316,6 +425,42 @@ inline std::size_t string_bulk_run(const unsigned char* data, std::size_t n) noe
     return scalar_string_bulk_run(data, n);
 }
 
+#if JSON_SCAN_NEON && !(defined(JSON_USE_SIMDUTF) && defined(JSON_HAS_CPP_17))
+/// scalar_string_bulk_run() from @a start on, out of line (see
+/// lexer_string_bulk_run())
+JSON_HEDLEY_NEVER_INLINE inline std::size_t string_bulk_run_from(const unsigned char* data, std::size_t start, std::size_t n) noexcept
+{
+    return start + scalar_string_bulk_run(data + start, n - start);
+}
+#endif
+
+// string_bulk_run() for the lexer. With NEON, only the first four words are
+// checked inline (most keys and short values end there), and longer runs and
+// non-ASCII text continue out of line. The lexer's scan function runs once per
+// token, and the vector code inlined into it needs more registers, which every
+// token pays for: 5% of json::accept() on jeopardy.json with Apple Clang on
+// Apple M1, even though no string of that file reaches the vector code. The
+// serializer keeps string_bulk_run() inline, which is faster there.
+JSON_HEDLEY_ALWAYS_INLINE std::size_t lexer_string_bulk_run(const unsigned char* data, std::size_t n) noexcept
+{
+#if JSON_SCAN_NEON && !(defined(JSON_USE_SIMDUTF) && defined(JSON_HAS_CPP_17))
+    std::size_t i = 0;
+    for (int words = 4; words > 0 && n - i >= 8; --words)
+    {
+        const std::uint64_t special = swar_string_special(read_eight_bytes(data + i));
+        if (special != 0)
+        {
+            const std::size_t pos = i + (static_cast<std::size_t>(count_trailing_zeros(special)) / 8);
+            return data[pos] < 0x80u ? pos : string_bulk_run_from(data, pos, n);
+        }
+        i += 8;
+    }
+    return string_bulk_run_from(data, i, n);
+#else
+    return string_bulk_run(data, n);
+#endif
+}
+
 // Decode the 4 hex digits at [data, data+4) - the digits following a `\u`
 // escape - into a codepoint 0x0000..0xFFFF via one table lookup per byte
 // (after yyjson's read_hex_u16), or return -1 if any of the 4 bytes is not a

include/nlohmann/detail/json_pointer.hpp

@@ -9,10 +9,13 @@
 #pragma once
 
 #include <algorithm> // all_of
+#include <array> // array
 #include <cctype> // isdigit
 #include <cerrno> // errno, ERANGE
 #include <cstddef> // size_t
+#include <cstdint> // uint64_t
 #include <cstdlib> // strtoull
+#include <cstring> // memcmp, memcpy
 #ifndef JSON_NO_IO
     #include <iosfwd> // ostream
 #endif  // JSON_NO_IO
@@ -24,6 +27,7 @@
 #include <nlohmann/detail/exceptions.hpp>
 #include <nlohmann/detail/macro_scope.hpp>
 #include <nlohmann/detail/meta/cpp_future.hpp>
+#include <nlohmann/detail/meta/type_traits.hpp>
 #include <nlohmann/detail/string_concat.hpp>
 #include <nlohmann/detail/string_escape.hpp>
 #include <nlohmann/detail/string_utils.hpp>
@@ -499,50 +503,111 @@ class json_pointer
 
   private:
     /*!
-    @brief the pointer prefixes of a flattened object, and which of them denote arrays
+    @brief the pointer prefixes of a flattened object that denote arrays
 
-    @ref unflatten collects the pointer prefixes that have a reference token 0
-    among their children; @ref get_and_create creates arrays exactly below
-    those prefixes and objects everywhere else. Deciding this up front keeps
-    the result independent of the order in which the flattened object is
-    iterated, which is unspecified for some object types.
+    @ref unflatten creates arrays exactly below the prefixes that have a
+    reference token 0 among their children, and objects everywhere else.
+    Deciding this up front keeps the result independent of the order in which
+    the flattened object is iterated, which is unspecified for some object
+    types.
 
-    The prefixes form a tree and are numbered (number 0 is the empty prefix),
-    so each of them is stored only once (as a node) rather than as a copy of
-    all of its reference tokens. The tree does not own the reference tokens:
-    it refers to the tokens of the parsed keys, which must outlive it.
+    A prefix is identified by the text of the key it is the beginning of:
+    escaping is unambiguous, so equal prefixes of valid keys are equal texts.
+    The set refers to the keys, which must outlive it.
     */
-    struct prefix_tree
+    class array_prefix_set
     {
-        // open-addressing hash table from (parent, last reference token) to
-        // the number of the node; empty slots have token == nullptr
+      public:
+        void insert(const char* data, std::size_t size)
+        {
+            // flattened objects are usually sorted, so most keys that have a
+            // 0 yield the prefix of the previous one
+            if (last_data != nullptr && last_size == size && std::memcmp(last_data, data, size) == 0)
+            {
+                return;
+            }
+            last_data = data;
+            last_size = size;
+
+            if (table.empty())
+            {
+                table.resize(16);
+            }
+            const std::size_t hash = hash_of(data, size);
+            const std::size_t mask = table.size() - 1;
+            std::size_t i = hash & mask;
+            while (table[i].data != nullptr)
+            {
+                if (table[i].hash == hash && table[i].size == size && std::memcmp(table[i].data, data, size) == 0)
+                {
+                    return;
+                }
+                i = (i + 1) & mask;
+            }
+            table[i] = slot(hash, data, size);
+
+            // keep the load factor below 1/2
+            if (++used * 2 > table.size())
+            {
+                grow();
+            }
+        }
+
+        bool empty() const noexcept
+        {
+            return used == 0;
+        }
+
+        bool contains(const char* data, std::size_t size) const noexcept
+        {
+            if (used == 0)
+            {
+                return false;
+            }
+            const std::size_t hash = hash_of(data, size);
+            const std::size_t mask = table.size() - 1;
+            for (std::size_t i = hash & mask; table[i].data != nullptr; i = (i + 1) & mask)
+            {
+                if (table[i].hash == hash && table[i].size == size && std::memcmp(table[i].data, data, size) == 0)
+                {
+                    return true;
+                }
+            }
+            return false;
+        }
+
+      private:
         struct slot
         {
-            std::size_t hash;
-            std::size_t parent;
-            const string_t* token;
-            std::size_t id;
+            slot() noexcept = default;
 
-            slot() noexcept
-                : hash(0), parent(0), token(nullptr), id(0)
+            slot(std::size_t h, const char* d, std::size_t s) noexcept
+                : hash(h), data(d), size(s)
             {}
 
-            slot(std::size_t h, std::size_t p, const string_t* t, std::size_t i) noexcept
-                : hash(h), parent(p), token(t), id(i)
-            {}
+            std::size_t hash = 0;
+            const char* data = nullptr; // nullptr: the slot is empty
+            std::size_t size = 0;
         };
 
-        std::vector<slot> table;
-        std::size_t used = 0;
-
-        static std::size_t hash_of(std::size_t parent, const string_t& token) noexcept
+        static std::size_t hash_of(const char* data, std::size_t size) noexcept
         {
-            std::size_t h = parent * static_cast<std::size_t>(0x9E3779B97F4A7C15ULL & (std::numeric_limits<std::size_t>::max)());
-            for (std::size_t i = 0; i < token.size(); ++i)
+            std::uint64_t h = 0x9E3779B97F4A7C15ULL ^ size;
+            std::size_t i = 0;
+            for (; i + 8 <= size; i += 8)
             {
-                h = (h ^ static_cast<unsigned char>(token[i])) * static_cast<std::size_t>(0x100000001B3ULL & (std::numeric_limits<std::size_t>::max)());
+                std::uint64_t w = 0;
+                std::memcpy(&w, data + i, 8);
+                h = (h ^ w) * 0x100000001B3ULL;
+                h ^= h >> 32u;
             }
-            return h ^ (h >> 29u);
+            std::uint64_t w = 0;
+            if (i < size)
+            {
+                std::memcpy(&w, data + i, size - i);
+            }
+            h = (h ^ w) * 0x9E3779B97F4A7C15ULL;
+            return detail::conditional_static_cast<std::size_t>(h ^ (h >> 29u));
         }
 
         void grow()
@@ -552,10 +617,10 @@ class json_pointer
             const std::size_t mask = table.size() - 1;
             for (const auto& s : old)
             {
-                if (s.token != nullptr)
+                if (s.data != nullptr)
                 {
                     std::size_t i = s.hash & mask;
-                    while (table[i].token != nullptr)
+                    while (table[i].data != nullptr)
                     {
                         i = (i + 1) & mask;
                     }
@@ -564,117 +629,160 @@ class json_pointer
             }
         }
 
-        // parent[id] is the number of the prefix without its last token
-        std::vector<std::size_t> parent;
-        // is_array[id] is true iff some flattened key has the reference token
-        // 0 directly below the prefix with number id
-        std::vector<bool> is_array;
-
-        // start with the empty prefix only
-        prefix_tree()
-            : table(16)
-            , parent(1, 0)
-            , is_array(1, false)
-        {}
-
-        // return the number of the prefix with number id extended by
-        // reference_token, adding it if it is new
-        std::size_t add_child(std::size_t id, const string_t& reference_token)
-        {
-            const std::size_t hash = hash_of(id, reference_token);
-            const std::size_t mask = table.size() - 1;
-            std::size_t i = hash & mask;
-            while (table[i].token != nullptr)
-            {
-                if (table[i].hash == hash && table[i].parent == id && *table[i].token == reference_token)
-                {
-                    return table[i].id;
-                }
-                i = (i + 1) & mask;
-            }
-
-            // new prefix
-            const std::size_t child = parent.size();
-            table[i] = slot(hash, id, &reference_token, child);
-            if (reference_token == "0")
-            {
-                is_array[id] = true;
-            }
-            parent.push_back(id);
-            is_array.push_back(false);
-
-            // keep the load factor below 1/2
-            if (++used * 2 > table.size())
-            {
-                grow();
-            }
-            return child;
-        }
+        std::vector<slot> table{};
+        std::size_t used = 0;
+        const char* last_data = nullptr;
+        std::size_t last_size = 0;
     };
 
+    /// @brief the length of the longest common prefix of the first @a limit bytes of @a a and @a b
+    static std::size_t common_prefix_length(const char* a, const char* b, const std::size_t limit) noexcept
+    {
+        std::size_t i = 0;
+        for (; i + 8 <= limit; i += 8)
+        {
+            std::uint64_t x = 0;
+            std::uint64_t y = 0;
+            std::memcpy(&x, a + i, 8);
+            std::memcpy(&y, b + i, 8);
+            if (x != y)
+            {
+                break;
+            }
+        }
+        while (i < limit && a[i] == b[i])
+        {
+            ++i;
+        }
+        return i;
+    }
+
     /*!
-    @brief create and return a reference to the pointed to value
+    @brief check that @a key is a valid JSON pointer, and add the prefixes of
+           it that denote arrays to @a arrays
 
-    @param[in] j  the value to create the pointed to value in
-    @param[in] tokens  the reference tokens of the pointer
-    @param[in] prefixes  prefixes[i] is the number in @a tree of the prefix
-                         consisting of the first i reference tokens
+    @throw parse_error.107  if the key is not empty or begins with '/'
+    @throw parse_error.108  if character '~' is not followed by '0' or '1'
 
-    Complexity: Linear in the number of reference tokens.
+    The same checks as @ref split, which throws the same exceptions, but
+    without building the reference tokens.
+    */
+    static void scan_key(const string_t& key, const string_t* previous, array_prefix_set& arrays)
+    {
+        const std::size_t size = key.size();
+        if (size == 0)
+        {
+            return;
+        }
 
-    @throw parse_error.106 if an array index begins with '0'
-    @throw parse_error.109 if array index is not a number
-    @throw type_error.313 if value cannot be unflattened
+        const char* const data = key.data();
+        if (JSON_HEDLEY_UNLIKELY(data[0] != '/'))
+        {
+            JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_missing_slash, 1, detail::concat("JSON pointer must be empty or begin with '/' - was: '", key, "'"), nullptr));
+        }
+
+        // The beginning of the key that it has in common with the previous
+        // one, which was checked already, is skipped: the scan starts with
+        // the token after the last slash in it.
+        std::size_t begin = 1;
+        if (previous != nullptr)
+        {
+            std::size_t same = common_prefix_length(data, previous->data(), (std::min)(size, previous->size()));
+            while (same > 1 && data[same - 1] != '/')
+            {
+                --same;
+            }
+            if (same > 1)
+            {
+                begin = same;
+            }
+        }
+
+        for (std::size_t i = begin; i <= size; ++i)
+        {
+            if (i == size || data[i] == '/')
+            {
+                // a token 0 makes the pointer before it an array
+                if (i - begin == 1 && data[begin] == '0')
+                {
+                    arrays.insert(data, begin - 1);
+                }
+                begin = i + 1;
+            }
+            else if (data[i] == '~')
+            {
+                // ~ must be followed by 0 or 1
+                if (JSON_HEDLEY_UNLIKELY(i + 1 == size || (data[i + 1] != '0' && data[i + 1] != '1')))
+                {
+                    JSON_THROW(detail::parse_error::create(detail::exception_id::pointer_invalid_escape, 0, "escape character '~' must be followed with '0' or '1'", nullptr));
+                }
+            }
+        }
+    }
+
+    /*!
+    @brief resolve a plain JSON pointer without building its reference tokens
+
+    @param[in] ptr     the value the pointer starts at
+    @param[in] path    a JSON pointer string
+    @param[in] length  the number of leading bytes of @a path to resolve; 0 or
+                       the position of a slash
+    @param[in,out] token  a string to hold the reference tokens, to reuse its storage
+
+    @return the pointed to value, or `nullptr` if the pointer does not denote
+            an existing value, or does so in a way that is not plain: a
+            reference token with an escape, an array index that is malformed or
+            out of range, or a "-". No exception is thrown for these; callers
+            use @ref json_pointer instead, which handles them.
     */
     template<typename BasicJsonType>
-    static BasicJsonType& get_and_create(BasicJsonType& j, const std::vector<string_t>& tokens,
-                                         const std::vector<std::size_t>& prefixes, const prefix_tree& tree)
+    static BasicJsonType* resolve_plain(BasicJsonType* ptr, const string_t& path, const std::size_t length, string_t& token)
     {
-        auto* result = &j;
-
-        // in case no reference tokens exist, return a reference to the JSON value
-        // j which will be overwritten by a primitive value
-        for (std::size_t i = 0; i < tokens.size(); ++i)
+        for (std::size_t pos = 0; pos < length;)
         {
-            const auto& reference_token = tokens[i];
-            switch (result->type())
+            if (JSON_HEDLEY_UNLIKELY(path[pos] != '/'))
             {
-                case detail::value_t::null:
-                {
-                    if (tree.is_array[prefixes[i]])
-                    {
-                        // some reference token below this position is 0, so the
-                        // value is an array
-                        result = &result->operator[](array_index<BasicJsonType>(reference_token));
-                    }
-                    else
-                    {
-                        // start a new object otherwise
-                        result = &result->operator[](reference_token);
-                    }
-                    break;
-                }
+                return nullptr;
+            }
 
+            // the reference token between the slash at pos and the next one
+            std::size_t end = path.find_first_of('/', pos + 1);
+            if (end == string_t::npos || end > length)
+            {
+                end = length;
+            }
+            token.clear();
+            token.append(path.data() + pos + 1, end - pos - 1);
+            if (JSON_HEDLEY_UNLIKELY(token.find_first_of('~') != string_t::npos))
+            {
+                return nullptr;
+            }
+
+            switch (ptr->type())
+            {
                 case detail::value_t::object:
                 {
-                    // create an entry in the object
-                    result = &object_member(*result, reference_token);
+                    const auto it = ptr->m_data.m_value.object->find(token);
+                    if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
+                    {
+                        return nullptr;
+                    }
+                    ptr = &it->second;
                     break;
                 }
 
                 case detail::value_t::array:
                 {
-                    // create an entry in the array
-                    result = &result->operator[](array_index<BasicJsonType>(reference_token));
+                    typename BasicJsonType::size_type idx{};
+                    if (JSON_HEDLEY_UNLIKELY(parse_array_index<BasicJsonType>(token, idx) != array_index_status::ok || idx >= ptr->m_data.m_value.array->size()))
+                    {
+                        return nullptr;
+                    }
+                    ptr = &(*ptr->m_data.m_value.array)[idx];
                     break;
                 }
 
-                /*
-                The following code is only reached if there exists a reference
-                token _and_ the current value is primitive. In this case, we have
-                an error situation, because primitive values may only occur as
-                a single value; that is, with an empty list of reference tokens.
-                */
+                case detail::value_t::null:
                 case detail::value_t::string:
                 case detail::value_t::boolean:
                 case detail::value_t::number_integer:
@@ -683,11 +791,12 @@ class json_pointer
                 case detail::value_t::binary:
                 case detail::value_t::discarded:
                 default:
-                    JSON_THROW(detail::type_error::create(detail::exception_id::unflatten_invalid_value, "invalid value to unflatten", &j));
+                    return nullptr;
             }
+            pos = end;
         }
 
-        return *result;
+        return ptr;
     }
 
     /*!
@@ -1142,20 +1251,45 @@ class json_pointer
     // Add the entry (key, v) to the object `obj`. Flattened paths are pairwise
     // distinct, so objects that can append a key known to be absent
     // (ordered_map) skip the lookup, which is linear and would make flatten
-    // quadratic; all other object types use operator[].
+    // quadratic. Objects with a positional insert (std::map) insert next to
+    // the entry added last: paths come out in nearly sorted order, so the
+    // position is usually right and no search from the root is needed.
+    // All other object types use operator[].
     template<typename ObjectType, typename ValueType>
-    static auto add_flattened(ObjectType& obj, const string_t& key, ValueType&& v, detail::priority_tag<1> /*unused*/)
+    static auto add_flattened(ObjectType& obj, typename ObjectType::iterator& /*hint*/, const string_t& key, ValueType&& v, detail::priority_tag<2> /*unused*/)
     -> decltype(obj.emplace_new_key(key, std::forward<ValueType>(v)), void())
     {
         obj.emplace_new_key(key, std::forward<ValueType>(v));
     }
 
     template<typename ObjectType, typename ValueType>
-    static void add_flattened(ObjectType& obj, const string_t& key, ValueType&& v, detail::priority_tag<0> /*unused*/)
+    static auto add_flattened(ObjectType& obj, typename ObjectType::iterator& hint, const string_t& key, ValueType&& v, detail::priority_tag<1> /*unused*/)
+    -> decltype(hint = obj.emplace_hint(hint, key, std::forward<ValueType>(v)), void())
+    {
+        hint = obj.emplace_hint(hint, key, std::forward<ValueType>(v));
+        ++hint;
+    }
+
+    template<typename ObjectType, typename ValueType>
+    static void add_flattened(ObjectType& obj, typename ObjectType::iterator& /*hint*/, const string_t& key, ValueType&& v, detail::priority_tag<0> /*unused*/)
     {
         obj[key] = std::forward<ValueType>(v);
     }
 
+    /// append the decimal representation of @a i to @a out
+    static void append_index(string_t& out, std::size_t i)
+    {
+        std::array<char, 24> buf{};
+        std::size_t pos = buf.size();
+        do
+        {
+            buf[--pos] = static_cast<char>('0' + (i % 10));
+            i /= 10;
+        }
+        while (i != 0);
+        out.append(buf.data() + pos, buf.size() - pos);
+    }
+
     /*!
     @param[in] reference_string  the reference string to the current value
     @param[in] value             the value to consider
@@ -1201,7 +1335,8 @@ class json_pointer
         // flatten `v`, whose path is `path`: primitives and empty containers
         // are added to the result right away; other containers get a frame
         auto& object = *result.m_data.m_value.object;
-        const auto enter = [&stack, &path, &object](const BasicJsonType & v)
+        auto hint = object.end();
+        const auto enter = [&stack, &path, &object, &hint](const BasicJsonType & v)
         {
             switch (v.type())
             {
@@ -1210,7 +1345,7 @@ class json_pointer
                     if (v.m_data.m_value.array->empty())
                     {
                         // flatten empty array as null
-                        add_flattened(object, path, nullptr, detail::priority_tag<1> {});
+                        add_flattened(object, hint, path, nullptr, detail::priority_tag<2> {});
                     }
                     else
                     {
@@ -1224,7 +1359,7 @@ class json_pointer
                     if (v.m_data.m_value.object->empty())
                     {
                         // flatten empty object as null
-                        add_flattened(object, path, nullptr, detail::priority_tag<1> {});
+                        add_flattened(object, hint, path, nullptr, detail::priority_tag<2> {});
                     }
                     else
                     {
@@ -1244,7 +1379,7 @@ class json_pointer
                 default:
                 {
                     // add a primitive value with its reference string
-                    add_flattened(object, path, v, detail::priority_tag<1> {});
+                    add_flattened(object, hint, path, v, detail::priority_tag<2> {});
                     return;
                 }
             }
@@ -1272,7 +1407,8 @@ class json_pointer
 
                 // iterate array and use index as a reference string
                 ++stack.back().index;
-                detail::concat_into(path, '/', detail::to_string<string_t>(i));
+                path.push_back('/');
+                append_index(path, i);
                 enter(array[i]);
             }
             else
@@ -1286,7 +1422,8 @@ class json_pointer
 
                 // iterate object and use keys as reference string
                 ++stack.back().member;
-                detail::concat_into(path, '/', detail::escape(it->first));
+                path.push_back('/');
+                detail::escape_into(path, it->first);
                 enter(it->second);
             }
         }
@@ -1317,53 +1454,30 @@ class json_pointer
 
         BasicJsonType result;
 
-        // parse every key once; the reference tokens are kept for the second
-        // pass (the vectors are never resized, so the tree may refer to them)
-        std::vector<std::vector<string_t>> keys;
-        keys.reserve(value.m_data.m_value.object->size());
+        // check every key, and collect the prefixes that have a reference
+        // token 0 among their children; the values below them are arrays,
+        // all others are objects (see array_prefix_set)
+        array_prefix_set arrays;
+        const string_t* previous_key = nullptr;
         for (const auto& element : *value.m_data.m_value.object)
         {
-            json_pointer ptr(element.first);
-            keys.push_back(std::move(ptr.reference_tokens));
+            scan_key(element.first, previous_key, arrays);
+            previous_key = &element.first;
         }
 
-        // collect the pointer prefixes that have a reference token 0 among
-        // their children; the values below them are arrays, all others are
-        // objects (see prefix_tree)
-        prefix_tree tree;
-        // the number of the prefix with i + 1 tokens of the previous key; keys
-        // are often sorted, so most of the prefix is the same as for the
-        // previous key and need not be looked up again
-        std::vector<std::size_t> path;
-        const std::vector<string_t>* previous = nullptr;
-        // the number of the complete prefix (i.e., the key) of each key
-        std::vector<std::size_t> leaves;
-        leaves.reserve(keys.size());
-        for (const auto& tokens : keys)
-        {
-            std::size_t common = 0;
-            if (previous != nullptr)
-            {
-                const std::size_t limit = (std::min)(tokens.size(), previous->size());
-                while (common < limit && tokens[common] == (*previous)[common])
-                {
-                    ++common;
-                }
-            }
-            path.resize(common);
-            std::size_t id = common == 0 ? 0 : path[common - 1];
-            for (std::size_t i = common; i < tokens.size(); ++i)
-            {
-                id = tree.add_child(id, tokens[i]);
-                path.push_back(id);
-            }
-            leaves.push_back(id);
-            previous = &tokens;
-        }
+        // values[i] is the value after the first i reference tokens of the
+        // previous key, ends[i] the length of that prefix of the key. Keys are
+        // often sorted, so most of the path to the value of a key is the
+        // same as for the previous key and need not be looked up again. Only
+        // the values below the common prefix are dropped, and only they can
+        // move when a value is added: the ones above are not inside the
+        // container that grows.
+        std::vector<BasicJsonType*> values(1, &result);
+        std::vector<std::size_t> ends(1, 0);
+        const string_t* previous = nullptr;
+        string_t token;
 
         // iterate the JSON object values
-        std::vector<std::size_t> prefixes;
-        std::size_t key_index = 0;
         for (const auto& element : *value.m_data.m_value.object)
         {
             if (JSON_HEDLEY_UNLIKELY(!element.second.is_primitive()))
@@ -1371,21 +1485,100 @@ class json_pointer
                 JSON_THROW(detail::type_error::create(detail::exception_id::unflatten_value_not_primitive, "values in object must be primitive", &element.second));
             }
 
-            const auto& tokens = keys[key_index];
-            // number the prefixes of this key from its last one to the first
-            prefixes.resize(tokens.size());
-            for (std::size_t i = tokens.size(), id = leaves[key_index]; i > 0; --i)
+            const string_t& key = element.first;
+            const char* const data = key.data();
+
+            // the number of reference tokens the key has in common with the
+            // previous one
+            std::size_t depth = 0;
+            if (previous != nullptr)
             {
-                id = tree.parent[id];
-                prefixes[i - 1] = id;
+                const std::size_t same = common_prefix_length(data, previous->data(), (std::min)(key.size(), previous->size()));
+                depth = ends.size() - 1;
+                while (depth > 0 && (ends[depth] > same || (ends[depth] == same && same < key.size() && data[same] != '/')))
+                {
+                    --depth;
+                }
             }
-            ++key_index;
+            values.resize(depth + 1);
+            ends.resize(depth + 1);
+            previous = &key;
 
             // Assign the value to the reference pointed to by JSON pointer. Note
             // that if the JSON pointer is "" (i.e., points to the whole value),
-            // function get_and_create returns a reference to the result itself.
-            // An assignment will then create a primitive value.
-            get_and_create(result, tokens, prefixes, tree) = element.second;
+            // the value is the result itself. An assignment will then create a
+            // primitive value.
+            BasicJsonType* current = values.back();
+            for (std::size_t pos = ends.back(); pos < key.size();)
+            {
+                // the reference token between the slash at pos and the next
+                std::size_t end = key.find_first_of('/', pos + 1);
+                if (end == string_t::npos)
+                {
+                    end = key.size();
+                }
+                token.clear();
+                token.append(data + pos + 1, end - pos - 1);
+                if (token.find_first_of('~') != string_t::npos)
+                {
+                    detail::unescape(token);
+                }
+
+                switch (current->type())
+                {
+                    case detail::value_t::null:
+                    {
+                        if (arrays.contains(data, pos))
+                        {
+                            // some reference token below this position is 0, so the
+                            // value is an array
+                            current = &current->operator[](array_index<BasicJsonType>(token));
+                        }
+                        else
+                        {
+                            // start a new object otherwise
+                            current = &current->operator[](token);
+                        }
+                        break;
+                    }
+
+                    case detail::value_t::object:
+                    {
+                        // create an entry in the object
+                        current = &object_member(*current, token);
+                        break;
+                    }
+
+                    case detail::value_t::array:
+                    {
+                        // create an entry in the array
+                        current = &current->operator[](array_index<BasicJsonType>(token));
+                        break;
+                    }
+
+                    /*
+                    The following code is only reached if there exists a reference
+                    token _and_ the current value is primitive. In this case, we have
+                    an error situation, because primitive values may only occur as
+                    a single value; that is, with an empty list of reference tokens.
+                    */
+                    case detail::value_t::string:
+                    case detail::value_t::boolean:
+                    case detail::value_t::number_integer:
+                    case detail::value_t::number_unsigned:
+                    case detail::value_t::number_float:
+                    case detail::value_t::binary:
+                    case detail::value_t::discarded:
+                    default:
+                        JSON_THROW(detail::type_error::create(detail::exception_id::unflatten_invalid_value, "invalid value to unflatten", &result));
+                }
+
+                values.push_back(current);
+                ends.push_back(end);
+                pos = end;
+            }
+
+            *current = element.second;
         }
 
         return result;

include/nlohmann/detail/macro_scope.hpp

@@ -187,6 +187,22 @@
     #endif
 #endif
 
+// the byte order of the target where the compiler tells it (std::endian is
+// C++20 and cannot select code in #if): JSON_LITTLE_ENDIAN_TARGET or
+// JSON_BIG_ENDIAN_TARGET is 1, or both are 0 where it is unknown (code then
+// handles the bytes one at a time); MSVC and x86 targets are little-endian
+#if defined(_MSC_VER) || defined(__x86_64__) || defined(__i386__) \
+    || (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+    #define JSON_LITTLE_ENDIAN_TARGET 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_BIG_ENDIAN_TARGET 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    #define JSON_LITTLE_ENDIAN_TARGET 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_BIG_ENDIAN_TARGET 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+#else
+    #define JSON_LITTLE_ENDIAN_TARGET 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+    #define JSON_BIG_ENDIAN_TARGET 0 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
+#endif
+
 #ifdef JSON_HAS_CPP_17
     #define JSON_INLINE_VARIABLE inline
 #else

include/nlohmann/detail/macro_unscope.hpp

@@ -24,6 +24,16 @@
 #undef JSON_UNLOCKED_FILE_IO
 #undef JSON_DTOA_SSE2
 #undef JSON_DTOA_NEON
+#undef JSON_LITTLE_ENDIAN_TARGET
+#undef JSON_BIG_ENDIAN_TARGET
+#undef JSON_SCAN_NEON
+#undef JSON_SCAN_SSE2
+#undef JSON_SCAN_SSSE3
+#undef JSON_SCAN_SSSE3_DISPATCH
+#undef JSON_SCAN_SSSE3_TARGET
+#undef JSON_SCAN_VECTOR
+#undef JSON_SCAN_VECTOR_UTF8
+#undef JSON_DUMP_SHORT_STRINGS
 
 #ifndef JSON_TEST_KEEP_MACROS
     #undef JSON_CATCH

include/nlohmann/detail/meta/cpp_future.hpp

@@ -10,17 +10,26 @@
 #pragma once
 
 #include <cstddef> // size_t
-#include <type_traits> // conditional, enable_if, false_type, integral_constant, is_constructible, is_integral, is_same, remove_cv, remove_reference, true_type
+#include <type_traits> // conditional, enable_if, false_type, integral_constant, is_arithmetic, is_constructible, is_integral, is_same, remove_cv, remove_reference, true_type
 #include <utility> // index_sequence, make_index_sequence, index_sequence_for
 
 #include <nlohmann/detail/macro_scope.hpp>
 
+#ifdef JSON_HAS_CPP_17
+    #include <algorithm> // clamp
+#endif
+
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
 {
 
-template<typename T>
-using uncvref_t = typename std::remove_cv<typename std::remove_reference<T>::type>::type;
+#ifdef __cpp_lib_remove_cvref
+    template<typename T>
+    using uncvref_t = std::remove_cvref_t<T>;
+#else
+    template<typename T>
+    using uncvref_t = typename std::remove_cv<typename std::remove_reference<T>::type>::type;
+#endif
 
 #ifdef JSON_HAS_CPP_14
 
@@ -160,5 +169,19 @@ struct static_const
     constexpr T static_const<T>::value;
 #endif
 
+/// @a v bounded to [@a lo, @a hi] (requires !(hi < lo)); like std::clamp,
+/// but by value, so that static constexpr members can be passed without an
+/// out-of-line definition before C++17
+template<typename T, enable_if_t<std::is_arithmetic<T>::value, int> = 0>
+constexpr T clamp(T v, T lo, T hi) noexcept
+{
+#ifdef __cpp_lib_clamp
+    return std::clamp(v, lo, hi);
+#else
+    // NOLINTNEXTLINE(readability-avoid-nested-conditional-operator): a constant expression of C++11
+    return v < lo ? lo : (hi < v ? hi : v);
+#endif
+}
+
 }  // namespace detail
 NLOHMANN_JSON_NAMESPACE_END

include/nlohmann/detail/meta/type_traits.hpp

@@ -313,6 +313,14 @@ struct char_traits<std::byte> : std::char_traits<char>
 // is_ functions //
 ///////////////////
 
+#ifdef __cpp_lib_logical_traits
+
+using std::conjunction;
+using std::disjunction;
+using std::negation;
+
+#else
+
 // https://en.cppreference.com/w/cpp/types/conjunction
 template<class...> struct conjunction : std::true_type { };
 template<class B> struct conjunction<B> : B { };
@@ -330,6 +338,8 @@ struct disjunction<B, Bn...>
 // https://en.cppreference.com/w/cpp/types/negation
 template<class B> struct negation : std::integral_constant < bool, !B::value > { };
 
+#endif
+
 // Reimplementation of is_constructible and is_default_constructible, due to them being broken for
 // std::pair and std::tuple until LWG 2367 fix (see https://cplusplus.github.io/LWG/lwg-defects.html#2367).
 // This causes compile errors in e.g., Clang 3.5 or GCC 4.9.
@@ -1060,8 +1070,12 @@ constexpr bool value_in_range_of(T val)
     return value_in_range_of_impl1<OfType, T>::test(val);
 }
 
-template<bool Value>
-using bool_constant = std::integral_constant<bool, Value>;
+#ifdef __cpp_lib_bool_constant
+    using std::bool_constant;
+#else
+    template<bool Value>
+    using bool_constant = std::integral_constant<bool, Value>;
+#endif
 
 // Stands in for number_unsigned_t in overload sets if it is the same type as
 // number_integer_t, so that the unsigned overloads do not collide with the

include/nlohmann/detail/output/binary_writer.hpp

@@ -19,10 +19,6 @@
 #include <utility> // move
 #include <vector> // vector
 
-#ifdef _MSC_VER
-    #include <cstdlib> // _byteswap_ushort, _byteswap_ulong, _byteswap_uint64
-#endif
-
 #include <nlohmann/detail/bit_ops.hpp>
 #include <nlohmann/detail/input/binary_reader.hpp>
 #include <nlohmann/detail/input/string_scan.hpp>
@@ -2073,7 +2069,7 @@ class binary_writer
     {
         out[0] = to_char_type(first_byte);
         const unsigned unused_bits = 64u - ((8u << size_class) >> 1u);
-        const std::uint64_t tail = is_little_endian ? byte_swap(value) >> unused_bits : value << unused_bits;
+        const std::uint64_t tail = is_little_endian ? byteswap(value) >> unused_bits : value << unused_bits;
         std::memcpy(out + 1, &tail, sizeof(tail));
         return ((std::size_t{1} << size_class) >> 1u) + 1;
     }
@@ -3517,53 +3513,14 @@ class binary_writer
     // Utility functions //
     ///////////////////////
 
-    // single-instruction byte swaps (compilers lower these to bswap/rev/movbe);
-    // used to emit big-endian numbers without a per-byte std::reverse loop
-    static std::uint16_t byte_swap(std::uint16_t x) noexcept
-    {
-#if defined(__GNUC__) || defined(__clang__)
-        return __builtin_bswap16(x);
-#elif defined(_MSC_VER)
-        return _byteswap_ushort(x);
-#else
-        return static_cast<std::uint16_t>((x >> 8) | (x << 8));
-#endif
-    }
-
-    static std::uint32_t byte_swap(std::uint32_t x) noexcept
-    {
-#if defined(__GNUC__) || defined(__clang__)
-        return __builtin_bswap32(x);
-#elif defined(_MSC_VER)
-        return _byteswap_ulong(x);
-#else
-        return ((x & 0x000000FFu) << 24) | ((x & 0x0000FF00u) << 8)
-               | ((x & 0x00FF0000u) >> 8) | ((x & 0xFF000000u) >> 24);
-#endif
-    }
-
-    static std::uint64_t byte_swap(std::uint64_t x) noexcept
-    {
-#if defined(__GNUC__) || defined(__clang__)
-        return __builtin_bswap64(x);
-#elif defined(_MSC_VER)
-        return _byteswap_uint64(x);
-#else
-        x = ((x & 0x00000000FFFFFFFFull) << 32) | ((x & 0xFFFFFFFF00000000ull) >> 32);
-        x = ((x & 0x0000FFFF0000FFFFull) << 16) | ((x & 0xFFFF0000FFFF0000ull) >> 16);
-        x = ((x & 0x00FF00FF00FF00FFull) << 8) | ((x & 0xFF00FF00FF00FF00ull) >> 8);
-        return x;
-#endif
-    }
-
     /*!
     @brief reverse the bytes of a buffer by byte-swapping it as UIntType
 
     Loading the buffer into an unsigned integer of the same width and swapping
     that is what lets the compiler emit a single bswap/rev/movbe; reversing the
     buffer element by element does not reliably get there (clang keeps a scalar
-    shuffle). The two memcpy calls are the only portable way to reinterpret the
-    bytes and are folded away by every optimizer.
+    shuffle). (The two memcpy calls, rather than bit_cast, keep GCC from
+    inlining differently in C++20, which made writing slower.)
     */
     template<typename UIntType, std::size_t N>
     static void byte_swap_buffer(std::array<CharType, N>& a) noexcept
@@ -3571,11 +3528,11 @@ class binary_writer
         static_assert(sizeof(UIntType) == N, "swap width must match the buffer size");
         UIntType v{};
         std::memcpy(&v, a.data(), sizeof(v));
-        v = byte_swap(v);
+        v = byteswap(v);
         std::memcpy(a.data(), &v, sizeof(v));
     }
 
-    // reverse the bytes of a fixed-size buffer; a single byte_swap() for the
+    // reverse the bytes of a fixed-size buffer; a single byteswap() for the
     // common 2/4/8-byte number payloads, std::reverse for any other size
     static void reverse_bytes(std::array<CharType, 2>& a) noexcept
     {
@@ -3700,9 +3657,7 @@ class binary_writer
 #endif
 
         static_assert(sizeof(std::uint8_t) == sizeof(CharType), "size of CharType must be equal to std::uint8_t");
-        CharType result;
-        std::memcpy(&result, &x, sizeof(x));
-        return result;
+        return bit_cast<CharType>(x);
     }
 
     template<typename C = CharType,

include/nlohmann/detail/output/serializer.hpp

@@ -34,6 +34,14 @@
 #include <nlohmann/detail/string_utils.hpp>
 #include <nlohmann/detail/value_t.hpp>
 
+// the short-string fast path of the serializer assembles words from little-endian
+// loads, see serializer::put_short_quoted()
+#if defined(_MSC_VER) || defined(__x86_64__) || defined(__i386__) || (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+    #define JSON_DUMP_SHORT_STRINGS 1
+#else
+    #define JSON_DUMP_SHORT_STRINGS 0
+#endif
+
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
 {
@@ -80,11 +88,14 @@ class serializer
                const std::size_t indent_step_ = 0,
                error_handler_t error_handler_ = error_handler_t::strict)
         : o(&s)
+        , number_buffer() // NOLINT(readability-redundant-member-init)
+        , string_buffer() // NOLINT(readability-redundant-member-init)
         , indent_char(ichar)
         , pretty_print(pretty_print_)
         , ensure_ascii(ensure_ascii_)
         , indent_step(indent_step_)
         , error_handler(error_handler_)
+        , write_buffer() // NOLINT(readability-redundant-member-init)
     {}
 
     // deleted because of pointer members
@@ -159,34 +170,27 @@ class serializer
 
                 if (pretty_print)
                 {
-                    put_literal("{\n");
-
                     // variable to hold indentation for recursive calls
                     const auto new_indent = next_indent(current_indent, indent_step);
 
+                    put_literal_indent("{\n", new_indent);
+
                     // first n-1 elements
                     auto i = val.m_data.m_value.object->cbegin();
                     for (std::size_t cnt = 0; cnt < val.m_data.m_value.object->size() - 1; ++cnt, ++i)
                     {
-                        put_indent(new_indent);
-                        put_char('"');
-                        dump_escaped(i->first);
-                        put_literal("\": ");
-                        dump_internal(i->second, new_indent, depth + 1);
-                        put_literal(",\n");
+                        put_quoted(i->first, "\": ");
+                        dump_element(i->second, new_indent, depth + 1);
+                        put_literal_indent(",\n", new_indent);
                     }
 
                     // last element
                     JSON_ASSERT(i != val.m_data.m_value.object->cend());
                     JSON_ASSERT(std::next(i) == val.m_data.m_value.object->cend());
-                    put_indent(new_indent);
-                    put_char('"');
-                    dump_escaped(i->first);
-                    put_literal("\": ");
-                    dump_internal(i->second, new_indent, depth + 1);
+                    put_quoted(i->first, "\": ");
+                    dump_element(i->second, new_indent, depth + 1);
 
-                    put_char('\n');
-                    put_indent(current_indent);
+                    put_literal_indent("\n", current_indent);
                     put_char('}');
                 }
                 else
@@ -197,20 +201,16 @@ class serializer
                     auto i = val.m_data.m_value.object->cbegin();
                     for (std::size_t cnt = 0; cnt < val.m_data.m_value.object->size() - 1; ++cnt, ++i)
                     {
-                        put_char('"');
-                        dump_escaped(i->first);
-                        put_literal("\":");
-                        dump_internal(i->second, current_indent, depth + 1);
+                        put_quoted(i->first, "\":");
+                        dump_element(i->second, current_indent, depth + 1);
                         put_char(',');
                     }
 
                     // last element
                     JSON_ASSERT(i != val.m_data.m_value.object->cend());
                     JSON_ASSERT(std::next(i) == val.m_data.m_value.object->cend());
-                    put_char('"');
-                    dump_escaped(i->first);
-                    put_literal("\":");
-                    dump_internal(i->second, current_indent, depth + 1);
+                    put_quoted(i->first, "\":");
+                    dump_element(i->second, current_indent, depth + 1);
 
                     put_char('}');
                 }
@@ -234,27 +234,24 @@ class serializer
 
                 if (pretty_print)
                 {
-                    put_literal("[\n");
-
                     // variable to hold indentation for recursive calls
                     const auto new_indent = next_indent(current_indent, indent_step);
 
+                    put_literal_indent("[\n", new_indent);
+
                     // first n-1 elements
                     for (auto i = val.m_data.m_value.array->cbegin();
                             i != val.m_data.m_value.array->cend() - 1; ++i)
                     {
-                        put_indent(new_indent);
-                        dump_internal(*i, new_indent, depth + 1);
-                        put_literal(",\n");
+                        dump_element(*i, new_indent, depth + 1);
+                        put_literal_indent(",\n", new_indent);
                     }
 
                     // last element
                     JSON_ASSERT(!val.m_data.m_value.array->empty());
-                    put_indent(new_indent);
-                    dump_internal(val.m_data.m_value.array->back(), new_indent, depth + 1);
+                    dump_element(val.m_data.m_value.array->back(), new_indent, depth + 1);
 
-                    put_char('\n');
-                    put_indent(current_indent);
+                    put_literal_indent("\n", current_indent);
                     put_char(']');
                 }
                 else
@@ -265,13 +262,13 @@ class serializer
                     for (auto i = val.m_data.m_value.array->cbegin();
                             i != val.m_data.m_value.array->cend() - 1; ++i)
                     {
-                        dump_internal(*i, current_indent, depth + 1);
+                        dump_element(*i, current_indent, depth + 1);
                         put_char(',');
                     }
 
                     // last element
                     JSON_ASSERT(!val.m_data.m_value.array->empty());
-                    dump_internal(val.m_data.m_value.array->back(), current_indent, depth + 1);
+                    dump_element(val.m_data.m_value.array->back(), current_indent, depth + 1);
 
                     put_char(']');
                 }
@@ -294,6 +291,27 @@ class serializer
     }
 
   private:
+    /*!
+    @brief serialize an element of a container
+
+    Most elements are scalars, which need none of the state dump_internal() keeps
+    for a container; writing them from here spares each the entry and exit of
+    that function.
+    */
+    JSON_HEDLEY_ALWAYS_INLINE void dump_element(const BasicJsonType& val,
+            const std::size_t current_indent,
+            const std::size_t depth)
+    {
+        if (val.m_data.m_type == value_t::object || val.m_data.m_type == value_t::array)
+        {
+            dump_internal(val, current_indent, depth);
+        }
+        else
+        {
+            dump_scalar(val, current_indent);
+        }
+    }
+
     /*!
     @brief write out @a val and everything below it without the call stack
 
@@ -353,16 +371,13 @@ class serializer
                     put_indent(frame.child_indent);
                 }
 
-                put_char('"');
-                dump_escaped(frame.object_it->first);
-
                 if (pretty_print)
                 {
-                    put_literal("\": ");
+                    put_quoted(frame.object_it->first, "\": ");
                 }
                 else
                 {
-                    put_literal("\":");
+                    put_quoted(frame.object_it->first, "\":");
                 }
 
                 const BasicJsonType& element = frame.object_it->second;
@@ -537,78 +552,13 @@ class serializer
         {
             case value_t::string:
             {
-                put_char('"');
-                dump_escaped(*val.m_data.m_value.string);
-                put_char('"');
+                dump_string_value(*val.m_data.m_value.string);
                 return;
             }
 
             case value_t::binary:
             {
-                if (pretty_print)
-                {
-                    put_literal("{\n");
-
-                    // variable to hold indentation for the bytes
-                    const auto new_indent = next_indent(current_indent, indent_step);
-
-                    put_indent(new_indent);
-
-                    put_literal("\"bytes\": [");
-
-                    if (!val.m_data.m_value.binary->empty())
-                    {
-                        for (auto i = val.m_data.m_value.binary->cbegin();
-                                i != val.m_data.m_value.binary->cend() - 1; ++i)
-                        {
-                            dump_byte(*i);
-                            put_literal(", ");
-                        }
-                        dump_byte(val.m_data.m_value.binary->back());
-                    }
-
-                    put_literal("],\n");
-                    put_indent(new_indent);
-
-                    put_literal("\"subtype\": ");
-                    if (val.m_data.m_value.binary->has_subtype())
-                    {
-                        dump_integer(val.m_data.m_value.binary->subtype());
-                    }
-                    else
-                    {
-                        put_literal("null");
-                    }
-                    put_char('\n');
-                    put_indent(current_indent);
-                    put_char('}');
-                }
-                else
-                {
-                    put_literal("{\"bytes\":[");
-
-                    if (!val.m_data.m_value.binary->empty())
-                    {
-                        for (auto i = val.m_data.m_value.binary->cbegin();
-                                i != val.m_data.m_value.binary->cend() - 1; ++i)
-                        {
-                            dump_byte(*i);
-                            put_char(',');
-                        }
-                        dump_byte(val.m_data.m_value.binary->back());
-                    }
-
-                    put_literal("],\"subtype\":");
-                    if (val.m_data.m_value.binary->has_subtype())
-                    {
-                        dump_integer(val.m_data.m_value.binary->subtype());
-                        put_char('}');
-                    }
-                    else
-                    {
-                        put_literal("null}");
-                    }
-                }
+                dump_binary_value(current_indent, *val.m_data.m_value.binary);
                 return;
             }
 
@@ -627,19 +577,19 @@ class serializer
 
             case value_t::number_integer:
             {
-                dump_integer(val.m_data.m_value.number_integer);
+                dump_signed_value(val.m_data.m_value.number_integer);
                 return;
             }
 
             case value_t::number_unsigned:
             {
-                dump_integer(val.m_data.m_value.number_unsigned);
+                dump_unsigned_value(val.m_data.m_value.number_unsigned);
                 return;
             }
 
             case value_t::number_float:
             {
-                dump_float(val.m_data.m_value.number_float);
+                dump_float_value(val.m_data.m_value.number_float);
                 return;
             }
 
@@ -662,6 +612,107 @@ class serializer
         }
     }
 
+    /// @brief write the string @a s in quotes; kept out of line, so that the
+    /// other scalars do not pay for its register needs (and the other way round)
+    JSON_HEDLEY_NEVER_INLINE void dump_string_value(const string_t& s)
+    {
+#if JSON_DUMP_SHORT_STRINGS
+        if (s.size() <= short_string_max && put_short_quoted(s, "\""))
+        {
+            return;
+        }
+#endif
+        put_char('"');
+        dump_escaped(s);
+        put_char('"');
+    }
+
+    // The numbers, and the rest of what dump_scalar() writes, are left to
+    // functions of their own as well: a single function holding all of them
+    // saves and restores every register any of them needs, on every call. Now
+    // dump_scalar() only dispatches, and each writer pays for what it uses.
+    JSON_HEDLEY_NEVER_INLINE void dump_signed_value(const number_integer_t x)
+    {
+        dump_integer(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_unsigned_value(const number_unsigned_t x)
+    {
+        dump_integer(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_float_value(const number_float_t x)
+    {
+        dump_float(x);
+    }
+
+    JSON_HEDLEY_NEVER_INLINE void dump_binary_value(const std::size_t current_indent, const typename BasicJsonType::binary_t& binary)
+    {
+        if (pretty_print)
+        {
+            put_literal("{\n");
+
+            // variable to hold indentation for the bytes
+            const auto new_indent = next_indent(current_indent, indent_step);
+
+            put_indent(new_indent);
+
+            put_literal("\"bytes\": [");
+
+            if (!binary.empty())
+            {
+                for (auto i = binary.cbegin();
+                        i != binary.cend() - 1; ++i)
+                {
+                    dump_byte(*i);
+                    put_literal(", ");
+                }
+                dump_byte(binary.back());
+            }
+
+            put_literal("],\n");
+            put_indent(new_indent);
+
+            put_literal("\"subtype\": ");
+            if (binary.has_subtype())
+            {
+                dump_integer(binary.subtype());
+            }
+            else
+            {
+                put_literal("null");
+            }
+            put_char('\n');
+            put_indent(current_indent);
+            put_char('}');
+        }
+        else
+        {
+            put_literal("{\"bytes\":[");
+
+            if (!binary.empty())
+            {
+                for (auto i = binary.cbegin();
+                        i != binary.cend() - 1; ++i)
+                {
+                    dump_byte(*i);
+                    put_char(',');
+                }
+                dump_byte(binary.back());
+            }
+
+            put_literal("],\"subtype\":");
+            if (binary.has_subtype())
+            {
+                dump_integer(binary.subtype());
+                put_char('}');
+            }
+            else
+            {
+                put_literal("null}");
+            }
+        }
+    }
 
     /*!
     @brief the indentation level to use for the children of the current value
@@ -678,6 +729,217 @@ class serializer
         return new_indent;
     }
 
+  private:
+    /*!
+    @brief append the string @a s in quotes, followed by @a tail
+
+    The quote, the string and @a tail (the closing quote, possibly with the
+    colon of an object key) are the bulk of what most documents consist of, and
+    most keys and many values are short and plain. Those are handled in one go
+    by put_short_quoted(); everything else takes the general path.
+    */
+    template<std::size_t N>
+    void put_quoted(const string_t& s, const char (&tail)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+#if JSON_DUMP_SHORT_STRINGS
+        if (put_short_quoted(s, tail))
+        {
+            return;
+        }
+#endif
+        put_char('"');
+        dump_escaped(s);
+        put_literal(tail);
+    }
+
+#if JSON_DUMP_SHORT_STRINGS
+    /// the longest string put_short_quoted() takes
+    static constexpr std::size_t short_string_max = JSON_SCAN_VECTOR ? 32 : 16;
+
+    /// the high bit of every byte of @a v that is not printable ASCII or is a
+    /// quote, a backslash, or DEL (zero if the bytes can be copied as they are)
+    static std::uint64_t swar_not_plain(const std::uint64_t v) noexcept
+    {
+        constexpr std::uint64_t ones = 0x0101010101010101ull;
+        const std::uint64_t q = v ^ 0x2222222222222222ull; // '"'
+        const std::uint64_t b = v ^ 0x5C5C5C5C5C5C5C5Cull; // '\\'
+        return (((q - ones) & ~q) | ((b - ones) & ~b)         // == '"', == '\\'
+                | ((v - 0x2020202020202020ull) & ~v)          // < 0x20
+                | v | (v + ones))                             // >= 0x7F
+               & 0x8080808080808080ull;
+    }
+
+    template<typename WordType>
+    static WordType load_word(const char* p) noexcept
+    {
+        WordType w;
+        std::memcpy(&w, p, sizeof(w));
+        return w;
+    }
+
+    template<typename WordType>
+    static void store_word(char* p, const WordType w) noexcept
+    {
+        std::memcpy(p, &w, sizeof(w));
+    }
+
+    /*!
+    @brief write `"`, @a s and @a tail if @a s is short and needs no escaping
+
+    A string of up to 16 (32 with vector instructions) printable ASCII
+    characters (no quote, backslash, or DEL) is read with two overlapping loads
+    and written with two overlapping stores, instead of scanning it and calling
+    memcpy with a length that is only known at run time. Returns false, having
+    changed nothing, if @a s does not qualify or the write buffer lacks the
+    room.
+    */
+    template<std::size_t N>
+    bool put_short_quoted(const string_t& s, const char (&tail)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+        const std::size_t n = s.size();
+        if (n > short_string_max || write_buffer_pos + n + N > write_buffer.size())
+        {
+            return false;
+        }
+
+        char* d = write_buffer.data() + write_buffer_pos + 1;
+        const char* const p = s.data();
+
+        if (n >= 8)
+        {
+#if JSON_SCAN_VECTOR
+            if (!(n >= 16 ? copy_plain_vectors(d, p, n) : copy_plain_halves(d, p, n)))
+            {
+                return false;
+            }
+#else
+            const auto a = load_word<std::uint64_t>(p);
+            const auto b = load_word<std::uint64_t>(p + n - 8);
+            if ((swar_not_plain(a) | swar_not_plain(b)) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 8, b);
+#endif
+        }
+        else if (n >= 4)
+        {
+            const auto a = load_word<std::uint32_t>(p);
+            const auto b = load_word<std::uint32_t>(p + n - 4);
+            // the bytes above the string are filled with a plain character
+            const std::uint64_t w = a | (static_cast<std::uint64_t>(b) << (8 * (n - 4)))
+                                    | (0x4141414141414141ull << (8 * n));
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 4, b);
+        }
+        else if (n >= 2)
+        {
+            const auto a = load_word<std::uint16_t>(p);
+            const auto b = load_word<std::uint16_t>(p + n - 2);
+            const std::uint64_t w = a | (static_cast<std::uint64_t>(b) << (8 * (n - 2)))
+                                    | (0x4141414141414141ull << (8 * n));
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            store_word(d, a);
+            store_word(d + n - 2, b);
+        }
+        else if (n == 1)
+        {
+            const std::uint64_t w = static_cast<std::uint64_t>(static_cast<unsigned char>(p[0])) | (0x4141414141414141ull << 8u);
+            if (swar_not_plain(w) != 0)
+            {
+                return false;
+            }
+            *d = p[0];
+        }
+
+        d[-1] = '"';
+        std::memcpy(d + n, tail, N - 1);
+        write_buffer_pos += n + N;
+        return true;
+    }
+
+#if JSON_SCAN_VECTOR
+    /// copy the 8 to 15 bytes [@a p, @a p + @a n) to @a d with two overlapping
+    /// words, tested together as one vector, see copy_plain_vectors()
+    static bool copy_plain_halves(char* d, const char* p, const std::size_t n) noexcept
+    {
+#if JSON_SCAN_NEON
+        const uint8x8_t lo = vld1_u8(reinterpret_cast<const std::uint8_t*>(p));
+        const uint8x8_t hi = vld1_u8(reinterpret_cast<const std::uint8_t*>(p + n - 8));
+        const uint8x16_t in = vcombine_u8(lo, hi);
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vceqq_u8(in, vdupq_n_u8('"')), vceqq_u8(in, vdupq_n_u8('\\'))),
+                                            vorrq_u8(vcltq_s8(vreinterpretq_s8_u8(in), vdupq_n_s8(0x20)), vceqq_u8(in, vdupq_n_u8(0x7F))));
+        if (vmaxvq_u8(special) != 0)
+        {
+            return false;
+        }
+        vst1_u8(reinterpret_cast<std::uint8_t*>(d), lo);
+        vst1_u8(reinterpret_cast<std::uint8_t*>(d + n - 8), hi);
+#else
+        const __m128i lo = _mm_loadl_epi64(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        const __m128i hi = _mm_loadl_epi64(static_cast<const __m128i*>(static_cast<const void*>(p + n - 8))); // NOLINT(bugprone-casting-through-void)
+        const __m128i in = _mm_unpacklo_epi64(lo, hi);
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(in, _mm_set1_epi8('"')), _mm_cmpeq_epi8(in, _mm_set1_epi8('\\'))),
+                                             _mm_or_si128(_mm_cmplt_epi8(in, _mm_set1_epi8(0x20)), _mm_cmpeq_epi8(in, _mm_set1_epi8(0x7F))));
+        if (_mm_movemask_epi8(special) != 0)
+        {
+            return false;
+        }
+        _mm_storel_epi64(static_cast<__m128i*>(static_cast<void*>(d)), lo); // NOLINT(bugprone-casting-through-void)
+        _mm_storel_epi64(static_cast<__m128i*>(static_cast<void*>(d + n - 8)), hi); // NOLINT(bugprone-casting-through-void)
+#endif
+        return true;
+    }
+
+    /// copy the 16 to 32 bytes [@a p, @a p + @a n) to @a d with two overlapping
+    /// vectors, unless one of them is a quote, a backslash, a control
+    /// character, DEL or not ASCII
+    static bool copy_plain_vectors(char* d, const char* p, const std::size_t n) noexcept
+    {
+#if JSON_SCAN_NEON
+        const uint8x16_t a = vld1q_u8(reinterpret_cast<const std::uint8_t*>(p));
+        const uint8x16_t b = vld1q_u8(reinterpret_cast<const std::uint8_t*>(p + n - 16));
+        const uint8x16_t special = vorrq_u8(vorrq_u8(vorrq_u8(vceqq_u8(a, vdupq_n_u8('"')), vceqq_u8(a, vdupq_n_u8('\\'))),
+                                            vorrq_u8(vceqq_u8(b, vdupq_n_u8('"')), vceqq_u8(b, vdupq_n_u8('\\')))),
+                                            vorrq_u8(vorrq_u8(vcltq_s8(vreinterpretq_s8_u8(a), vdupq_n_s8(0x20)), vcltq_s8(vreinterpretq_s8_u8(b), vdupq_n_s8(0x20))),
+                                                vorrq_u8(vceqq_u8(a, vdupq_n_u8(0x7F)), vceqq_u8(b, vdupq_n_u8(0x7F)))));
+        if (vmaxvq_u8(special) != 0)
+        {
+            return false;
+        }
+        vst1q_u8(reinterpret_cast<std::uint8_t*>(d), a);
+        vst1q_u8(reinterpret_cast<std::uint8_t*>(d + n - 16), b);
+#else
+        const __m128i a = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        const __m128i b = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p + n - 16))); // NOLINT(bugprone-casting-through-void)
+        const __m128i quote = _mm_set1_epi8('"');
+        const __m128i backslash = _mm_set1_epi8('\\');
+        const __m128i space = _mm_set1_epi8(0x20);
+        const __m128i del = _mm_set1_epi8(0x7F);
+        const __m128i special = _mm_or_si128(_mm_or_si128(_mm_or_si128(_mm_cmpeq_epi8(a, quote), _mm_cmpeq_epi8(a, backslash)),
+                                             _mm_or_si128(_mm_cmpeq_epi8(b, quote), _mm_cmpeq_epi8(b, backslash))),
+                                             _mm_or_si128(_mm_or_si128(_mm_cmplt_epi8(a, space), _mm_cmplt_epi8(b, space)),
+                                                 _mm_or_si128(_mm_cmpeq_epi8(a, del), _mm_cmpeq_epi8(b, del))));
+        if (_mm_movemask_epi8(special) != 0)
+        {
+            return false;
+        }
+        _mm_storeu_si128(static_cast<__m128i*>(static_cast<void*>(d)), a); // NOLINT(bugprone-casting-through-void)
+        _mm_storeu_si128(static_cast<__m128i*>(static_cast<void*>(d + n - 16)), b); // NOLINT(bugprone-casting-through-void)
+#endif
+        return true;
+    }
+#endif
+#endif
+
   JSON_PRIVATE_UNLESS_TESTED:
     /*!
     @brief dump escaped string
@@ -690,9 +952,46 @@ class serializer
     @param[in] s  the string to escape
 
     Complexity: Linear in the length of string @a s.
+
+    @note kept out of line: with the short-string attempts in front of it gone
+          from here, compilers would otherwise inline it into @ref put_quoted
+          and then no longer into the loops that call that
     */
-    void dump_escaped(const string_t& s)
+    JSON_HEDLEY_NEVER_INLINE void dump_escaped(const string_t& s)
     {
+        // Short strings of printable ASCII - most object keys and many values -
+        // are checked byte by byte against a table and then copied in one go.
+        // For them, this is cheaper than starting the vector scanner below,
+        // whose setup costs more than such a string takes to scan. The same
+        // check serves both values of ensure_ascii.
+        //
+        // Where put_short_quoted() exists and takes strings of up to 32 bytes,
+        // the callers have tried it already, and repeating the attempt only
+        // costs time for strings that need escaping or are not ASCII.
+#if !(JSON_DUMP_SHORT_STRINGS && JSON_SCAN_VECTOR)
+        if (s.size() <= 32)
+        {
+            const auto& copyable = ascii_copyable();
+            const auto* const data = reinterpret_cast<const unsigned char*>(s.data());
+            const std::size_t n = s.size();
+            std::size_t k = 0;
+            while (k + 4 <= n && copyable[data[k]] != 0 && copyable[data[k + 1]] != 0
+                    && copyable[data[k + 2]] != 0 && copyable[data[k + 3]] != 0)
+            {
+                k += 4;
+            }
+            while (k < n && copyable[data[k]] != 0)
+            {
+                ++k;
+            }
+            if (k == n)
+            {
+                put_string(s, 0, n);
+                return;
+            }
+        }
+#endif
+
         // dispatch once here rather than test the flag inside the loop: it does
         // not change while a string is written, and folding it lets each of the
         // two scanners be inlined into a loop of its own
@@ -769,7 +1068,7 @@ class serializer
                         put_buffer(string_buffer, bytes);
                         bytes = 0;
                     }
-                    put_string(s, i, i + run);
+                    put_run(s, i, run);
                     bytes_after_last_accept = 0;
                     undumped_chars = 0;
                     i += run;
@@ -809,6 +1108,27 @@ class serializer
 
             const auto byte = static_cast<std::uint8_t>(s[i]);
 
+            if (EnsureAscii && state == UTF8_ACCEPT && byte >= 0x80)
+            {
+                // a run of well-formed multi-byte characters is escaped
+                // straight into the write buffer; an ill-formed sequence
+                // ends the run and is left to the byte-at-a-time path
+                if (bytes != 0)
+                {
+                    put_buffer(string_buffer, bytes);
+                    bytes = 0;
+                    bytes_after_last_accept = 0;
+                }
+                const std::size_t escaped = write_u_escapes(reinterpret_cast<const unsigned char*>(s.data()) + i, s.size() - i);
+                if (escaped != 0)
+                {
+                    bytes_after_last_accept = 0;
+                    undumped_chars = 0;
+                    i += escaped - 1;
+                    continue;
+                }
+            }
+
             switch (decode(state, codepoint, byte))
             {
                 case UTF8_ACCEPT:  // decode found a new code point
@@ -1073,8 +1393,12 @@ class serializer
     the indentation character once and flushing that same content repeatedly:
     flushing does not disturb what the buffer holds, so re-filling it between
     flushes would be redundant work.
+
+    @note inlined on purpose, as is @ref put_literal: GCC otherwise stops
+          inlining both into the pretty-printing loops once @ref dump_escaped
+          has grown, which made indented output up to 13% slower
     */
-    void put_indent(std::size_t indent)
+    JSON_HEDLEY_ALWAYS_INLINE void put_indent(std::size_t indent)
     {
         // closing braces at the outermost level ask for no indentation at all
         if (indent == 0)
@@ -1082,6 +1406,21 @@ class serializer
             return;
         }
 
+        // the usual indentation: four words of the indentation character are
+        // stored unconditionally and only the part wanted is claimed, which
+        // spares the call of memset
+        if (JSON_HEDLEY_LIKELY(indent <= 32 && write_buffer_pos + 32 <= write_buffer.size()))
+        {
+            const std::uint64_t word = 0x0101010101010101ull * static_cast<unsigned char>(indent_char);
+            char* const d = write_buffer.data() + write_buffer_pos;
+            std::memcpy(d, &word, sizeof(word));
+            std::memcpy(d + 8, &word, sizeof(word));
+            std::memcpy(d + 16, &word, sizeof(word));
+            std::memcpy(d + 24, &word, sizeof(word));
+            write_buffer_pos += indent;
+            return;
+        }
+
         const std::size_t capacity = write_buffer.size();
 
         // fill whatever room is left in the buffer; this is the whole job
@@ -1115,6 +1454,32 @@ class serializer
         write_buffer_pos = indent;
     }
 
+    /*!
+    @brief append a string literal and then @a indent indentation characters
+
+    The usual line break and indentation go into the buffer with one check of
+    its room and four stores of the indentation, see @ref put_indent.
+    */
+    template<std::size_t N>
+    JSON_HEDLEY_ALWAYS_INLINE void put_literal_indent(const char (&s)[N], const std::size_t indent) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    {
+        constexpr std::size_t length = N - 1;
+        if (JSON_HEDLEY_LIKELY(indent <= 32 && write_buffer_pos + length + 32 <= write_buffer.size()))
+        {
+            const std::uint64_t word = 0x0101010101010101ull * static_cast<unsigned char>(indent_char);
+            char* const d = write_buffer.data() + write_buffer_pos;
+            std::memcpy(d, s, length);
+            std::memcpy(d + length, &word, sizeof(word));
+            std::memcpy(d + length + 8, &word, sizeof(word));
+            std::memcpy(d + length + 16, &word, sizeof(word));
+            std::memcpy(d + length + 24, &word, sizeof(word));
+            write_buffer_pos += length + indent;
+            return;
+        }
+        put_literal(s);
+        put_indent(indent);
+    }
+
     /*!
     @brief append a string literal to the write buffer
 
@@ -1124,7 +1489,7 @@ class serializer
     write-through path for oversized runs.
     */
     template<std::size_t N>
-    void put_literal(const char (&s)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+    JSON_HEDLEY_ALWAYS_INLINE void put_literal(const char (&s)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
     {
         static_assert(N >= 2, "put_literal expects a non-empty string literal");
         // the array bound counts the terminating NUL, which is not written
@@ -1139,6 +1504,28 @@ class serializer
         write_buffer_pos += length;
     }
 
+    /*!
+    @brief append the characters of @a str in [@a start, @a start + @a length)
+
+    A run of up to 16 characters with 16 or more in the string from @a start on
+    and in the write buffer from its end on is copied as a whole block of 16,
+    which spares the call of memcpy with a length known only at run time; the
+    surplus characters in the buffer are overwritten by what follows.
+    */
+    template<typename StringType>
+    void put_run(const StringType& str, const std::size_t start, const std::size_t length)
+    {
+        // a short run is the exception in text without escapes (CJK, accented
+        // Latin), for which this keeps the long path the fast one
+        if (JSON_HEDLEY_UNLIKELY(length <= 16 && start + 16 <= str.size() && write_buffer_pos + 16 <= write_buffer.size()))
+        {
+            std::memcpy(write_buffer.data() + write_buffer_pos, str.data() + start, 16);
+            write_buffer_pos += length;
+            return;
+        }
+        put_string(str, start, start + length);
+    }
+
     /*!
     @brief append the characters of @a str in [@a start, @a end)
 
@@ -1233,6 +1620,39 @@ class serializer
         }
     }
 
+    /*!
+    @brief whether each byte can be written verbatim under either value of ensure_ascii
+
+    Entry c is 1 for printable ASCII other than '"' and '\\' (see
+    is_ascii_copyable()) and 0 otherwise. A table lookup per byte is the
+    cheapest test for the short strings handled first in dump_escaped().
+    */
+    static const std::array<std::uint8_t, 256>& ascii_copyable() noexcept
+    {
+        static const std::array<std::uint8_t, 256> copyable =
+        {
+            {
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x00
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x10
+                1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x20
+                1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x30
+                1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x40
+                1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, // 0x50
+                1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x60
+                1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, // 0x70
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x80
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x90
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xA0
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xB0
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xC0
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xD0
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xE0
+                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 // 0xF0
+            }
+        };
+        return copyable;
+    }
+
     /*!
     @brief the JSON form of each character below 0x60
 
@@ -1348,6 +1768,75 @@ class serializer
         return escapes;
     }
 
+    /*!
+    @brief escape the well-formed multi-byte characters at the start of @a data
+
+    Writes the "\\uXXXX" escape of each character (two of them for a character
+    above U+FFFF) straight into the write buffer, which is why the caller must
+    have written out what is pending in @ref string_buffer. Stops at the first
+    byte that is below 0x80 or starts an ill-formed or truncated sequence, which
+    the caller handles.
+
+    @return the number of bytes of @a data that have been escaped
+    */
+    JSON_HEDLEY_NEVER_INLINE std::size_t write_u_escapes(const unsigned char* data, std::size_t n)
+    {
+        constexpr const char* nibble_to_hex = "0123456789abcdef";
+        std::size_t i = 0;
+        while (i < n && data[i] >= 0x80)
+        {
+            const std::size_t length = validate_one_utf8(data + i, n - i);
+            if (length == 0)
+            {
+                break;
+            }
+
+            const auto b0 = static_cast<std::uint32_t>(data[i]);
+            const auto b1 = static_cast<std::uint32_t>(data[i + 1]);
+            std::uint32_t codepoint = 0;
+            switch (length)
+            {
+                case 2:
+                    codepoint = ((b0 & 0x1Fu) << 6u) | (b1 & 0x3Fu);
+                    break;
+                case 3:
+                    codepoint = ((b0 & 0x0Fu) << 12u) | ((b1 & 0x3Fu) << 6u) | (static_cast<std::uint32_t>(data[i + 2]) & 0x3Fu);
+                    break;
+                default:
+                    codepoint = ((b0 & 0x07u) << 18u) | ((b1 & 0x3Fu) << 12u) | ((static_cast<std::uint32_t>(data[i + 2]) & 0x3Fu) << 6u) | (static_cast<std::uint32_t>(data[i + 3]) & 0x3Fu);
+                    break;
+            }
+
+            // at most two escapes of six bytes each
+            if (JSON_HEDLEY_UNLIKELY(write_buffer.size() - write_buffer_pos < 12))
+            {
+                flush();
+            }
+            char* d = write_buffer.data() + write_buffer_pos;
+            if (codepoint > 0xFFFFu)
+            {
+                d = put_u_escape(d, static_cast<std::uint16_t>(0xD7C0u + (codepoint >> 10u)), nibble_to_hex);
+                codepoint = 0xDC00u + (codepoint & 0x3FFu);
+            }
+            d = put_u_escape(d, static_cast<std::uint16_t>(codepoint), nibble_to_hex);
+            write_buffer_pos = static_cast<std::size_t>(d - write_buffer.data());
+            i += length;
+        }
+        return i;
+    }
+
+    static char* put_u_escape(char* d, const std::uint16_t codeunit, const char* nibble_to_hex) noexcept
+    {
+        const auto unit = static_cast<std::uint32_t>(codeunit);
+        d[0] = '\\';
+        d[1] = 'u';
+        d[2] = nibble_to_hex[(unit >> 12u) & 0x0Fu];
+        d[3] = nibble_to_hex[(unit >> 8u) & 0x0Fu];
+        d[4] = nibble_to_hex[(unit >> 4u) & 0x0Fu];
+        d[5] = nibble_to_hex[unit & 0x0Fu];
+        return d + 6;
+    }
+
     /*!
      * @brief write a lowercase "\\uXXXX" escape sequence into @a string_buffer
      *
@@ -1705,11 +2194,22 @@ class serializer
     /// the output of the serializer (non-owning; the adapter lives at the call site)
     output_adapter_protocol<char>* o = nullptr;
 
+    /// a character buffer left uninitialized: the buffers below are written
+    /// before they are read (only the part written is ever read), and zeroing
+    /// 1.6 KiB took more than half of what dumping a small value costs (the
+    /// user-provided constructor keeps the member initialization, which
+    /// -Weffc++ asks for, from zeroing them)
+    template<std::size_t N>
+    struct uninitialized_buffer : std::array<char, N>
+    {
+        uninitialized_buffer() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+    };
+
     /// a (hopefully) large enough character buffer
-    std::array<char, 64> number_buffer{{}};
+    uninitialized_buffer<64> number_buffer;
 
     /// string buffer
-    std::array<char, 512> string_buffer{{}};
+    uninitialized_buffer<512> string_buffer;
 
     /// the indentation character
     const char indent_char;
@@ -1729,7 +2229,7 @@ class serializer
     /// buffer collecting output before it is flushed to the output adapter, so
     /// that the many small structural writes become few bulk writes
     static constexpr std::size_t write_buffer_size = 1024;
-    std::array<char, write_buffer_size> write_buffer{{}};
+    uninitialized_buffer<write_buffer_size> write_buffer;
     /// number of valid bytes currently held in @ref write_buffer
     std::size_t write_buffer_pos = 0;
 };

include/nlohmann/detail/view/serializer.hpp

@@ -105,17 +105,13 @@ class output_buffer
     }
 
   private:
-    /// the size of a growth step (a function: std::min() takes a reference,
-    /// which a static constexpr member does not have before C++17)
-    static constexpr std::size_t step() noexcept
-    {
-        return 0x10000;
-    }
+    /// the size of a growth step
+    static constexpr std::size_t step = 0x10000;
 
     static StringType& sized(StringType& out, std::size_t estimate)
     {
         out.reserve(estimate);
-        out.resize((std::min)((std::max)(estimate, static_cast<std::size_t>(64)), step()));
+        out.resize(detail::clamp(estimate, std::size_t{64}, step));
         return out;
     }
 
@@ -123,8 +119,9 @@ class output_buffer
     {
         const auto used = static_cast<std::size_t>(m_pos - m_out.data());
         // (a step does not go beyond the reserved estimate, so that a good
-        // estimate is never copied to a larger allocation)
-        const std::size_t size = (std::max)((std::min)(m_out.size() + step(), m_out.capacity()), used + n + 256);
+        // estimate is never copied to a larger allocation; not a clamp: the
+        // lower bound used + n + 256 can exceed capacity(), and then it must win)
+        const std::size_t size = (std::max)((std::min)(m_out.size() + step, m_out.capacity()), used + n + 256);
         if (size > m_out.capacity())
         {
             m_out.reserve((std::max)(m_out.capacity() * 2, size));
@@ -171,6 +168,34 @@ inline std::size_t plain_output_run(const unsigned char* s, std::size_t n) noexc
     return n;
 }
 
+/// Whether dump() with ensure_ascii writes the string at s unchanged, for a
+/// string of the source without escape sequences (which has no quote,
+/// backslash, or control character): no byte is DEL (0x7F) or above. Adding 1
+/// to each byte sets its high bit exactly for 0x7F, and a carry into the next
+/// byte only comes from 0xFF, whose own high bit is set anyway.
+inline bool is_ascii_without_delete(const char* s, std::size_t n) noexcept
+{
+    constexpr std::uint64_t ones = 0x0101010101010101ull;
+    constexpr std::uint64_t high = 0x8080808080808080ull;
+    std::size_t i = 0;
+    for (; i + 8 <= n; i += 8)
+    {
+        const std::uint64_t v = read_eight_bytes(s + i);
+        if (((v | (v + ones)) & high) != 0)
+        {
+            return false;
+        }
+    }
+    for (; i < n; ++i)
+    {
+        if (static_cast<unsigned char>(s[i]) >= 0x7F)
+        {
+            return false;
+        }
+    }
+    return true;
+}
+
 /// A stack that starts in a buffer of the caller (a local array) and moves to
 /// the heap (a vector of the caller) only when that is full, so that dumps of
 /// shallow documents need no allocation. The top is a pointer, as in
@@ -286,15 +311,19 @@ class view_serializer
 
     void dump(const node* root)
     {
-        if (!m_style.pretty && !m_style.ensure_ascii)
+        if (!m_style.pretty)
         {
-            if (m_style.source_numbers)
+            if (m_style.ensure_ascii)
             {
-                dump_compact<true>(root);
+                dump_compact_ascii(root);
+            }
+            else if (m_style.source_numbers)
+            {
+                dump_compact<true, false>(root);
             }
             else
             {
-                dump_compact<false>(root);
+                dump_compact<false, false>(root);
             }
             return;
         }
@@ -400,6 +429,41 @@ class view_serializer
         m_out.put('"');
     }
 
+    /// dump_compact() with ensure_ascii, out of line: the walk without it is
+    /// faster when it is not merged with this one into dump()
+    NLOHMANN_VIEW_NOINLINE void dump_compact_ascii(const node* root)
+    {
+        if (m_style.source_numbers)
+        {
+            dump_compact<true, true>(root);
+        }
+        else
+        {
+            dump_compact<false, true>(root);
+        }
+    }
+
+    /// whether dump_compact() copies a string or key from the source: one of
+    /// the source, and with ensure_ascii one without bytes >= 0x7F (an overload
+    /// each, so that the walk without ensure_ascii is compiled exactly as
+    /// before: with GCC, a condition that only folds away later changed its
+    /// block layout and made it 4-6% slower)
+    static bool copied(const node& n, const char* /*src*/, std::false_type /*ensure_ascii*/) noexcept
+    {
+        return (n.flags & node_flags::storage) == 0;
+    }
+
+    static bool copied(const node& n, const char* src, std::true_type /*ensure_ascii*/) noexcept
+    {
+        return (n.flags & node_flags::storage) == 0 && is_ascii_without_delete(src + n.off, n.len);
+    }
+
+    /// a string with ensure_ascii (dump_compact()), out of line
+    NLOHMANN_VIEW_NOINLINE void write_ascii_string(const node& n)
+    {
+        write_string(n);
+    }
+
     /// the copies of dump_compact() that are not fixed-size moves (long
     /// strings, or near the end of the source); out of line, so that the
     /// compiler does not merge the fixed-size moves into this call
@@ -417,16 +481,18 @@ class view_serializer
     }
 
     /*!
-    @brief the compact output without ensure_ascii (the default dump())
+    @brief the compact output (the default dump(), or with ensure_ascii)
 
     The same walk as dump(), with the write position in a local variable
     (stores through char pointers would otherwise force a reload of the
     buffer's members after each one), and with strings and number tokens of
     the source copied by fixed-size moves of 32 bytes where the source has
     that many bytes left, instead of a library call per token. The buffer
-    keeps 64 bytes of slack for the overshoot.
+    keeps 64 bytes of slack for the overshoot. With ensure_ascii (Ascii), a
+    string of the source is copied only if it has no byte >= 0x7F (see
+    is_ascii_without_delete()); any other string is written by write_string().
     */
-    template<bool SourceNumbers>
+    template<bool SourceNumbers, bool Ascii>
     void dump_compact(const node* root)
     {
         struct frame
@@ -478,7 +544,14 @@ class view_serializer
         const auto escaped = [&](const node & n)
         {
             m_out.set_cursor(w);
-            write_decoded(n);
+            if (enabled(Ascii))
+            {
+                write_ascii_string(n);
+            }
+            else
+            {
+                write_decoded(n);
+            }
             w = m_out.cursor();
             lim = m_out.limit();
         };
@@ -501,7 +574,7 @@ class view_serializer
             switch (static_cast<value_t>(n->kind))
             {
                 case value_t::string:
-                    if ((n->flags & node_flags::storage) == 0)
+                    if (copied(*n, src, std::integral_constant<bool, Ascii>()))
                     {
                         room(n->len + 2);
                         *w++ = '"';
@@ -632,7 +705,7 @@ class view_serializer
             if (cur.object)
             {
                 const node& key = *at;
-                if ((key.flags & node_flags::storage) == 0)
+                if (copied(key, src, std::integral_constant<bool, Ascii>()))
                 {
                     room(key.len + 3);
                     *w++ = '"';
@@ -794,8 +867,7 @@ class view_serializer
     static char* write_double_value_at(char* w, double x)
     {
         // (from the bits: without the checks of to_chars())
-        std::uint64_t bits = 0;
-        std::memcpy(&bits, &x, sizeof(bits));
+        auto bits = bit_cast<std::uint64_t>(x);
         if (NLOHMANN_VIEW_UNLIKELY((bits & 0x7FF0000000000000u) == 0x7FF0000000000000u))
         {
             return write_text_at(w, "null", 4);
@@ -843,9 +915,10 @@ class view_serializer
     {
         const char* const s = m_doc.str(n);
         m_out.put('"');
-        if ((n.flags & node_flags::storage) == 0 && !m_style.ensure_ascii)
+        if ((n.flags & node_flags::storage) == 0 && (!m_style.ensure_ascii || is_ascii_without_delete(s, n.len)))
         {
-            // a string of the source without escape sequences has nothing to escape
+            // a string of the source without escape sequences has nothing to
+            // escape (with ensure_ascii, if it is printable ASCII)
             m_out.put(s, n.len);
         }
         else if (m_style.ensure_ascii)

include/nlohmann/detail/view/simd.hpp

@@ -50,7 +50,12 @@
 #endif
 #if NLOHMANN_VIEW_SSE2 && !NLOHMANN_VIEW_SSSE3 && !defined(NLOHMANN_VIEW_NO_TARGET_ATTRIBUTE) && ((defined(__clang__) && __clang_major__ >= 4) || (defined(__GNUC__) && !defined(__clang__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9))))
     // (GCC before 4.9 has no SSSE3 intrinsics without -mssse3)
-    #include <cpuid.h>
+    // <cpuid.h> of GCC before 11 has no include guard, and json.hpp and
+    // json_view.hpp both include it
+    #ifndef NLOHMANN_JSON_CPUID_INCLUDED
+        #define NLOHMANN_JSON_CPUID_INCLUDED
+        #include <cpuid.h>
+    #endif
     #include <tmmintrin.h>
     #define NLOHMANN_VIEW_SSSE3_DISPATCH 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
     #define NLOHMANN_VIEW_SSSE3_TARGET __attribute__((target("ssse3")))

include/nlohmann/json.hpp

@@ -7045,6 +7045,125 @@ public:
             detail::throw_patch_not_an_array(json_patch);
         }
 
+        // the usual patch operations "add", "remove", "replace" and "test" on
+        // a pointer without escapes that denotes existing array elements and
+        // object members (as "add" does for the parent of the new value), as
+        // a diff creates them: apply the operation and return true; return
+        // false, with nothing changed, for all others
+        string_t token;
+        const string_t value_key = "value";
+        const auto try_plain = [&result, &token, &value_key](patch_operations operation, const string_t& path, const basic_json & val)
+        {
+            if (path.empty() || path.find_first_of('~') != string_t::npos)
+            {
+                return false;
+            }
+
+            // a missing "value" is reported below
+            const auto value_it = val.m_data.m_value.object->find(value_key);
+            const basic_json& value = value_it != val.m_data.m_value.object->end() ? value_it->second : val;
+            if (operation != patch_operations::remove && value_it == val.m_data.m_value.object->end())
+            {
+                return false;
+            }
+
+            switch (operation)
+            {
+                case patch_operations::replace:
+                {
+                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    if (target == nullptr)
+                    {
+                        return false;
+                    }
+                    *target = value;
+                    return true;
+                }
+
+                case patch_operations::test:
+                {
+                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    // a failing test is reported below
+                    return target != nullptr && *target == value;
+                }
+
+                case patch_operations::add:
+                case patch_operations::remove:
+                {
+                    // the parent of the pointer and its last reference token
+                    std::size_t slash = path.size();
+                    while (slash > 0 && path[slash - 1] != '/')
+                    {
+                        --slash;
+                    }
+                    if (slash == 0)
+                    {
+                        return false;
+                    }
+                    --slash;
+
+                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token);
+                    if (parent == nullptr)
+                    {
+                        return false;
+                    }
+                    token.clear();
+                    token.append(path.data() + slash + 1, path.size() - slash - 1);
+
+                    if (operation == patch_operations::add)
+                    {
+                        if (parent->is_object())
+                        {
+                            (*parent)[token] = value;
+                            return true;
+                        }
+                        if (parent->is_array())
+                        {
+                            typename basic_json::size_type idx{};
+                            if (token == "-")
+                            {
+                                parent->push_back(value);
+                                return true;
+                            }
+                            if (json_pointer::template parse_array_index<basic_json_t>(token, idx) == json_pointer::array_index_status::ok && idx <= parent->size())
+                            {
+                                parent->insert(parent->begin() + static_cast<difference_type>(idx), value);
+                                return true;
+                            }
+                        }
+                        return false;
+                    }
+
+                    if (parent->is_object())
+                    {
+                        const auto it = parent->find(token);
+                        if (it == parent->end())
+                        {
+                            return false;
+                        }
+                        parent->erase(it);
+                        return true;
+                    }
+                    if (parent->is_array())
+                    {
+                        typename basic_json::size_type idx{};
+                        if (json_pointer::template parse_array_index<basic_json_t>(token, idx) == json_pointer::array_index_status::ok && idx < parent->size())
+                        {
+                            parent->erase(idx);
+                            return true;
+                        }
+                    }
+                    return false;
+                }
+
+                case patch_operations::move:
+                case patch_operations::copy:
+                case patch_operations::invalid:
+                default:
+                    return false;
+            }
+        };
+
         // iterate and apply the operations
         for (const auto& val : json_patch)
         {
@@ -7081,9 +7200,19 @@ public:
             // collect mandatory members
             const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
             const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
+            const patch_operations operation = get_op(op);
+
+            // the usual operations on a plain pointer are applied without
+            // building the pointer's reference tokens; anything else (and
+            // every error) is handled below
+            if (try_plain(operation, path, val))
+            {
+                continue;
+            }
+
             json_pointer ptr(path);
 
-            switch (get_op(op))
+            switch (operation)
             {
                 case patch_operations::add:
                 {
@@ -7191,7 +7320,8 @@ public:
     {
         // the patch
         basic_json result(value_t::array);
-        diff_recursively(result, source, target, path, 0);
+        string_t buffer = path;
+        diff_recursively(result, source, target, buffer, 0);
         return result;
     }
 
@@ -7235,31 +7365,69 @@ public:
     // function it appears in. In diff_recursively, which is on the call stack
     // once per nesting level, that made every level cost kilobytes of stack.
 
+    /// @brief append an operation @a op for @a path to @a result, with @a value if it is not null
+    static void diff_operation(basic_json& result, const char* op, const string_t& path, const basic_json* value)
+    {
+        basic_json entry(value_t::object);
+        entry["op"] = op;
+        entry["path"] = path;
+        if (value != nullptr)
+        {
+            entry["value"] = *value;
+        }
+        result.push_back(std::move(entry));
+    }
+
     /// @brief append a "replace" operation for @a path with @a value to @a result
     static void diff_replace(basic_json& result, const string_t& path, const basic_json& value)
     {
-        result.push_back(
-        {
-            {"op", "replace"}, {"path", path}, {"value", value}
-        });
+        diff_operation(result, "replace", path, &value);
     }
 
     /// @brief append a "remove" operation for @a path to @a result
     static void diff_remove(basic_json& result, const string_t& path)
     {
-        result.push_back(object(
-        {
-            {"op", "remove"}, {"path", path}
-        }));
+        diff_operation(result, "remove", path, nullptr);
     }
 
     /// @brief append an "add" operation for @a path with @a value to @a result
     static void diff_add(basic_json& result, const string_t& path, const basic_json& value)
     {
-        result.push_back(
-        {
-            {"op", "add"}, {"path", path}, {"value", value}
-        });
+        diff_operation(result, "add", path, &value);
+    }
+
+    /// @brief append "/" and the decimal representation of @a index to @a path
+    static void diff_append_index(string_t& path, std::size_t index)
+    {
+        path.push_back('/');
+        json_pointer::append_index(path, index);
+    }
+
+    /// @brief append "/" and the escaped @a key to @a path
+    static void diff_append_key(string_t& path, const string_t& key)
+    {
+        path.push_back('/');
+        detail::escape_into(path, key);
+    }
+
+    /// @brief the path of the member @a key of the value at @a path
+    static string_t diff_key_path(const string_t& path, const string_t& key)
+    {
+        string_t result;
+        result.reserve(path.size() + key.size() + 1);
+        result.append(path.data(), path.size());
+        diff_append_key(result, key);
+        return result;
+    }
+
+    /// @brief the path of the element @a index of the value at @a path
+    static string_t diff_index_path(const string_t& path, const std::size_t index)
+    {
+        string_t result;
+        result.reserve(path.size() + 8);
+        result.append(path.data(), path.size());
+        diff_append_index(result, index);
+        return result;
     }
 
     /// @brief append the "remove" operations for the elements of array
@@ -7272,13 +7440,18 @@ public:
         // in that order avoids the quadratic reinsertion done before
         for (std::size_t j = source.size(); j > index; --j)
         {
-            diff_remove(result, detail::concat<string_t>(path, '/', detail::to_string<string_t>(j - 1)));
+            diff_remove(result, diff_index_path(path, j - 1));
         }
 
         // add other remaining elements
-        for (std::size_t i = source.size(); i < target.size(); ++i)
+        if (source.size() < target.size())
         {
-            diff_add(result, detail::concat<string_t>(path, "/-"), target[i]);
+            string_t tail_path = path;
+            tail_path.append("/-", 2);
+            for (std::size_t i = source.size(); i < target.size(); ++i)
+            {
+                diff_add(result, tail_path, target[i]);
+            }
         }
     }
 
@@ -7287,7 +7460,7 @@ public:
     {
         for (const auto& it : added)
         {
-            diff_add(result, detail::concat<string_t>(path, '/', detail::escape(it.key())), it.value());
+            diff_add(result, diff_key_path(path, it.key()), it.value());
         }
     }
 
@@ -7308,10 +7481,25 @@ public:
                                  std::vector<const_iterator>& added)
     {
         // first pass: one lookup per source key
+        // Objects of the same size usually have the same keys, in the same
+        // order: walk them together as long as that holds, which takes no lookups.
         target_values.reserve(source.size());
         std::size_t common = 0;
-        for (auto it = source.cbegin(); it != source.cend(); ++it)
+        bool together = source.size() == target.size();
+        for (auto it = source.cbegin(), partner = target.cbegin(); it != source.cend(); ++it)
         {
+            if (together)
+            {
+                if (it.key() == partner.key())
+                {
+                    target_values.push_back(&partner.value());
+                    ++common;
+                    ++partner;
+                    continue;
+                }
+                together = false;
+            }
+
             const auto found = target.find(it.key());
             if (found != target.cend())
             {
@@ -7324,6 +7512,12 @@ public:
             }
         }
 
+        // the same keys in the same order
+        if (together)
+        {
+            return true;
+        }
+
         // second pass: find the keys only target has. An object type that
         // keeps its members in insertion order, such as the one backing
         // ordered_json, also needs the common keys to be in the same order
@@ -7389,6 +7583,13 @@ public:
             return true;
         }
 
+        // equal objects, whose members are in a different order, need no
+        // operations
+        if (source == target)
+        {
+            return false;
+        }
+
         // slow path: the common keys are in a different relative
         // order in source and target (only possible for a
         // reorderable object_t like ordered_map). Building a
@@ -7405,7 +7606,7 @@ public:
         // it moves it to the end, fixing its position.
         for (auto it = source.cbegin(); it != source.cend(); ++it)
         {
-            diff_remove(result, detail::concat<string_t>(path, '/', detail::escape(it.key())));
+            diff_remove(result, diff_key_path(path, it.key()));
         }
 
         // add every key that is either common (just removed
@@ -7414,7 +7615,7 @@ public:
         // target exactly
         for (auto it = target.cbegin(); it != target.cend(); ++it)
         {
-            diff_add(result, detail::concat<string_t>(path, '/', detail::escape(it.key())), it.value());
+            diff_add(result, diff_key_path(path, it.key()), it.value());
         }
         return false;
     }
@@ -7423,6 +7624,9 @@ public:
     @brief @ref diff, for values at nesting level @a depth, appending the
            operations to @a result
 
+    @a path is the path of the values; it is a buffer that is extended and
+    shortened again below, and is the same when the function returns.
+
     Diffing two arrays or objects calls this function again, once per nesting
     level, so values nested deeply enough used to exhaust the call stack and
     terminate the process. The descent is bounded here: once @ref
@@ -7430,7 +7634,7 @@ public:
     diff_iteratively diffs what is left without the call stack.
     */
     static void diff_recursively(basic_json& result, const basic_json& source, const basic_json& target,
-                                 const string_t& path, const std::size_t depth)
+                                 string_t& path, const std::size_t depth)
     {
         // if the values are the same, there is nothing to do
         if (source == target)
@@ -7442,9 +7646,9 @@ public:
     }
 
     /// @brief @ref diff_recursively for values known to be unequal; callers
-    ///        compare first so they build no path for equal ones
+    ///        compare first so they extend the path only for unequal ones
     static void diff_unequal(basic_json& result, const basic_json& source, const basic_json& target,
-                             const string_t& path, const std::size_t depth)
+                             string_t& path, const std::size_t depth)
     {
         if (JSON_HEDLEY_UNLIKELY(depth >= detail::recursion_depth_limit()))
         {
@@ -7463,6 +7667,8 @@ public:
         {
             case value_t::array:
             {
+                const std::size_t path_length = path.size();
+
                 // first pass: traverse common elements
                 std::size_t i = 0;
                 while (i < source.size() && i < target.size())
@@ -7470,7 +7676,9 @@ public:
                     // recursive call to compare array values at index i
                     if (!(source[i] == target[i]))
                     {
-                        diff_unequal(result, source[i], target[i], detail::concat<string_t>(path, '/', detail::to_string<string_t>(i)), depth + 1);
+                        diff_append_index(path, i);
+                        diff_unequal(result, source[i], target[i], path, depth + 1);
+                        path.resize(path_length);
                     }
                     ++i;
                 }
@@ -7492,6 +7700,7 @@ public:
                     // lookup is needed. Deleted keys are interleaved here
                     // too, in source's original order, to match the
                     // historical (pre-reordering-aware) output order.
+                    const std::size_t path_length = path.size();
                     std::size_t i = 0;
                     for (auto it = source.cbegin(); it != source.cend(); ++it, ++i)
                     {
@@ -7499,13 +7708,15 @@ public:
                         {
                             if (!(it.value() == *target_values[i]))
                             {
-                                diff_unequal(result, it.value(), *target_values[i], detail::concat<string_t>(path, '/', detail::escape(it.key())), depth + 1);
+                                diff_append_key(path, it.key());
+                                diff_unequal(result, it.value(), *target_values[i], path, depth + 1);
+                                path.resize(path_length);
                             }
                         }
                         else
                         {
                             // found a key that is not in target -> remove it
-                            diff_remove(result, detail::concat<string_t>(path, '/', detail::escape(it.key())));
+                            diff_remove(result, diff_key_path(path, it.key()));
                         }
                     }
 
@@ -7636,7 +7847,7 @@ public:
                 if (i < source_array.size() && i < target_array.size())
                 {
                     ++stack.back().index;
-                    detail::concat_into(current_path, '/', detail::to_string<string_t>(i));
+                    diff_append_index(current_path, i);
                     enter(source_array[i], target_array[i]);
                     if (stack.size() == depth)
                     {
@@ -7659,7 +7870,7 @@ public:
                     if (target_value_ptr != nullptr)
                     {
                         const basic_json& target_value = *target_value_ptr;
-                        detail::concat_into(current_path, '/', detail::escape(it.key()));
+                        diff_append_key(current_path, it.key());
                         enter(it.value(), target_value);
                         if (stack.size() == depth)
                         {
@@ -7669,7 +7880,7 @@ public:
                     else
                     {
                         // found a key that is not in target -> remove it
-                        diff_remove(result, detail::concat<string_t>(current_path, '/', detail::escape(it.key())));
+                        diff_remove(result, diff_key_path(current_path, it.key()));
                     }
                     continue;
                 }

single_include/nlohmann/json_fwd.hpp

@@ -139,6 +139,12 @@
     #define NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO
 #endif
 
+#ifdef JSON_NO_SIMD
+    #define NLOHMANN_JSON_ABI_TAG_NO_SIMD _ns
+#else
+    #define NLOHMANN_JSON_ABI_TAG_NO_SIMD
+#endif
+
 #ifdef JSON_VIEW_NO_SIMD
     #define NLOHMANN_JSON_ABI_TAG_VIEW_NO_SIMD _vns
 #else
@@ -156,9 +162,9 @@
 #endif
 
 // Construct the namespace ABI tags component
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j ## k ## l
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j, k, l) \
-    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l)
+#define NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l, m) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j ## k ## l ## m
+#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j, k, l, m)
 
 #define NLOHMANN_JSON_ABI_TAGS                                       \
     NLOHMANN_JSON_ABI_TAGS_CONCAT(                                   \
@@ -172,6 +178,7 @@
             NLOHMANN_JSON_ABI_TAG_OBJECTS_FOR_ENUM_KEYED_MAPS,       \
             NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN,          \
             NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO,          \
+            NLOHMANN_JSON_ABI_TAG_NO_SIMD,                           \
             NLOHMANN_JSON_ABI_TAG_VIEW_NO_SIMD,                      \
             NLOHMANN_JSON_ABI_TAG_VIEW_USE_SSSE3)
 

single_include/nlohmann/json_view.hpp

@@ -690,7 +690,12 @@ NLOHMANN_JSON_NAMESPACE_END
 #endif
 #if NLOHMANN_VIEW_SSE2 && !NLOHMANN_VIEW_SSSE3 && !defined(NLOHMANN_VIEW_NO_TARGET_ATTRIBUTE) && ((defined(__clang__) && __clang_major__ >= 4) || (defined(__GNUC__) && !defined(__clang__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9))))
     // (GCC before 4.9 has no SSSE3 intrinsics without -mssse3)
-    #include <cpuid.h>
+    // <cpuid.h> of GCC before 11 has no include guard, and json.hpp and
+    // json_view.hpp both include it
+    #ifndef NLOHMANN_JSON_CPUID_INCLUDED
+        #define NLOHMANN_JSON_CPUID_INCLUDED
+        #include <cpuid.h>
+    #endif
     #include <tmmintrin.h>
     #define NLOHMANN_VIEW_SSSE3_DISPATCH 1 // NOLINT(cppcoreguidelines-macro-to-enum,modernize-macro-to-enum)
     #define NLOHMANN_VIEW_SSSE3_TARGET __attribute__((target("ssse3")))
@@ -6166,17 +6171,13 @@ class output_buffer
     }
 
   private:
-    /// the size of a growth step (a function: std::min() takes a reference,
-    /// which a static constexpr member does not have before C++17)
-    static constexpr std::size_t step() noexcept
-    {
-        return 0x10000;
-    }
+    /// the size of a growth step
+    static constexpr std::size_t step = 0x10000;
 
     static StringType& sized(StringType& out, std::size_t estimate)
     {
         out.reserve(estimate);
-        out.resize((std::min)((std::max)(estimate, static_cast<std::size_t>(64)), step()));
+        out.resize(detail::clamp(estimate, std::size_t{64}, step));
         return out;
     }
 
@@ -6184,8 +6185,9 @@ class output_buffer
     {
         const auto used = static_cast<std::size_t>(m_pos - m_out.data());
         // (a step does not go beyond the reserved estimate, so that a good
-        // estimate is never copied to a larger allocation)
-        const std::size_t size = (std::max)((std::min)(m_out.size() + step(), m_out.capacity()), used + n + 256);
+        // estimate is never copied to a larger allocation; not a clamp: the
+        // lower bound used + n + 256 can exceed capacity(), and then it must win)
+        const std::size_t size = (std::max)((std::min)(m_out.size() + step, m_out.capacity()), used + n + 256);
         if (size > m_out.capacity())
         {
             m_out.reserve((std::max)(m_out.capacity() * 2, size));
@@ -6232,6 +6234,34 @@ inline std::size_t plain_output_run(const unsigned char* s, std::size_t n) noexc
     return n;
 }
 
+/// Whether dump() with ensure_ascii writes the string at s unchanged, for a
+/// string of the source without escape sequences (which has no quote,
+/// backslash, or control character): no byte is DEL (0x7F) or above. Adding 1
+/// to each byte sets its high bit exactly for 0x7F, and a carry into the next
+/// byte only comes from 0xFF, whose own high bit is set anyway.
+inline bool is_ascii_without_delete(const char* s, std::size_t n) noexcept
+{
+    constexpr std::uint64_t ones = 0x0101010101010101ull;
+    constexpr std::uint64_t high = 0x8080808080808080ull;
+    std::size_t i = 0;
+    for (; i + 8 <= n; i += 8)
+    {
+        const std::uint64_t v = read_eight_bytes(s + i);
+        if (((v | (v + ones)) & high) != 0)
+        {
+            return false;
+        }
+    }
+    for (; i < n; ++i)
+    {
+        if (static_cast<unsigned char>(s[i]) >= 0x7F)
+        {
+            return false;
+        }
+    }
+    return true;
+}
+
 /// A stack that starts in a buffer of the caller (a local array) and moves to
 /// the heap (a vector of the caller) only when that is full, so that dumps of
 /// shallow documents need no allocation. The top is a pointer, as in
@@ -6347,15 +6377,19 @@ class view_serializer
 
     void dump(const node* root)
     {
-        if (!m_style.pretty && !m_style.ensure_ascii)
+        if (!m_style.pretty)
         {
-            if (m_style.source_numbers)
+            if (m_style.ensure_ascii)
             {
-                dump_compact<true>(root);
+                dump_compact_ascii(root);
+            }
+            else if (m_style.source_numbers)
+            {
+                dump_compact<true, false>(root);
             }
             else
             {
-                dump_compact<false>(root);
+                dump_compact<false, false>(root);
             }
             return;
         }
@@ -6461,6 +6495,41 @@ class view_serializer
         m_out.put('"');
     }
 
+    /// dump_compact() with ensure_ascii, out of line: the walk without it is
+    /// faster when it is not merged with this one into dump()
+    NLOHMANN_VIEW_NOINLINE void dump_compact_ascii(const node* root)
+    {
+        if (m_style.source_numbers)
+        {
+            dump_compact<true, true>(root);
+        }
+        else
+        {
+            dump_compact<false, true>(root);
+        }
+    }
+
+    /// whether dump_compact() copies a string or key from the source: one of
+    /// the source, and with ensure_ascii one without bytes >= 0x7F (an overload
+    /// each, so that the walk without ensure_ascii is compiled exactly as
+    /// before: with GCC, a condition that only folds away later changed its
+    /// block layout and made it 4-6% slower)
+    static bool copied(const node& n, const char* /*src*/, std::false_type /*ensure_ascii*/) noexcept
+    {
+        return (n.flags & node_flags::storage) == 0;
+    }
+
+    static bool copied(const node& n, const char* src, std::true_type /*ensure_ascii*/) noexcept
+    {
+        return (n.flags & node_flags::storage) == 0 && is_ascii_without_delete(src + n.off, n.len);
+    }
+
+    /// a string with ensure_ascii (dump_compact()), out of line
+    NLOHMANN_VIEW_NOINLINE void write_ascii_string(const node& n)
+    {
+        write_string(n);
+    }
+
     /// the copies of dump_compact() that are not fixed-size moves (long
     /// strings, or near the end of the source); out of line, so that the
     /// compiler does not merge the fixed-size moves into this call
@@ -6478,16 +6547,18 @@ class view_serializer
     }
 
     /*!
-    @brief the compact output without ensure_ascii (the default dump())
+    @brief the compact output (the default dump(), or with ensure_ascii)
 
     The same walk as dump(), with the write position in a local variable
     (stores through char pointers would otherwise force a reload of the
     buffer's members after each one), and with strings and number tokens of
     the source copied by fixed-size moves of 32 bytes where the source has
     that many bytes left, instead of a library call per token. The buffer
-    keeps 64 bytes of slack for the overshoot.
+    keeps 64 bytes of slack for the overshoot. With ensure_ascii (Ascii), a
+    string of the source is copied only if it has no byte >= 0x7F (see
+    is_ascii_without_delete()); any other string is written by write_string().
     */
-    template<bool SourceNumbers>
+    template<bool SourceNumbers, bool Ascii>
     void dump_compact(const node* root)
     {
         struct frame
@@ -6539,7 +6610,14 @@ class view_serializer
         const auto escaped = [&](const node & n)
         {
             m_out.set_cursor(w);
-            write_decoded(n);
+            if (enabled(Ascii))
+            {
+                write_ascii_string(n);
+            }
+            else
+            {
+                write_decoded(n);
+            }
             w = m_out.cursor();
             lim = m_out.limit();
         };
@@ -6562,7 +6640,7 @@ class view_serializer
             switch (static_cast<value_t>(n->kind))
             {
                 case value_t::string:
-                    if ((n->flags & node_flags::storage) == 0)
+                    if (copied(*n, src, std::integral_constant<bool, Ascii>()))
                     {
                         room(n->len + 2);
                         *w++ = '"';
@@ -6693,7 +6771,7 @@ class view_serializer
             if (cur.object)
             {
                 const node& key = *at;
-                if ((key.flags & node_flags::storage) == 0)
+                if (copied(key, src, std::integral_constant<bool, Ascii>()))
                 {
                     room(key.len + 3);
                     *w++ = '"';
@@ -6855,8 +6933,7 @@ class view_serializer
     static char* write_double_value_at(char* w, double x)
     {
         // (from the bits: without the checks of to_chars())
-        std::uint64_t bits = 0;
-        std::memcpy(&bits, &x, sizeof(bits));
+        auto bits = bit_cast<std::uint64_t>(x);
         if (NLOHMANN_VIEW_UNLIKELY((bits & 0x7FF0000000000000u) == 0x7FF0000000000000u))
         {
             return write_text_at(w, "null", 4);
@@ -6904,9 +6981,10 @@ class view_serializer
     {
         const char* const s = m_doc.str(n);
         m_out.put('"');
-        if ((n.flags & node_flags::storage) == 0 && !m_style.ensure_ascii)
+        if ((n.flags & node_flags::storage) == 0 && (!m_style.ensure_ascii || is_ascii_without_delete(s, n.len)))
         {
-            // a string of the source without escape sequences has nothing to escape
+            // a string of the source without escape sequences has nothing to
+            // escape (with ensure_ascii, if it is printable ASCII)
             m_out.put(s, n.len);
         }
         else if (m_style.ensure_ascii)

tests/CMakeLists.txt

@@ -373,6 +373,25 @@ json_test_add_test_for(src/unit-diagnostic-positions.cpp
     MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
 )
 
+# the lexer again with the portable string scanning instead of SSE2/NEON, and on
+# x86-64 with SSSE3 compiled in (no check of the CPU)
+json_test_set_test_options(test-class_lexer_no_simd
+    COMPILE_DEFINITIONS JSON_NO_SIMD
+)
+json_test_add_test_for(src/unit-class_lexer.cpp
+    NAME test-class_lexer_no_simd
+    MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
+)
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64|amd64)$" AND NOT MSVC)
+    json_test_set_test_options(test-class_lexer_ssse3
+        COMPILE_OPTIONS -mssse3
+    )
+    json_test_add_test_for(src/unit-class_lexer.cpp
+        NAME test-class_lexer_ssse3
+        MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
+    )
+endif()
+
 # test reading from a std::istream again without scanning the get area in bulk
 json_test_set_test_options(test-stream-bulk-scan_disabled
     COMPILE_DEFINITIONS JSON_DISABLE_STREAM_BULK_SCAN=1

tests/abi/config/CMakeLists.txt

@@ -24,6 +24,7 @@ foreach(test default noversion)
         JSON_USE_LEGACY_DISCARDED_VALUE_COMPARISON=1
         JSON_DISABLE_STREAM_BULK_SCAN=1
         JSON_DISABLE_UNLOCKED_FILE_IO=1
+        JSON_NO_SIMD
         JSON_VIEW_NO_SIMD
         JSON_VIEW_USE_SSSE3)
     target_link_libraries(abi_config_${test}_all_tags PRIVATE abi_compat_main)

tests/abi/config/default.cpp

@@ -60,6 +60,10 @@ TEST_CASE("default namespace")
         expected += "_dufio";
 #endif
 
+#ifdef JSON_NO_SIMD
+        expected += "_ns";
+#endif
+
 #ifdef JSON_VIEW_NO_SIMD
         expected += "_vns";
 #endif

tests/abi/config/noversion.cpp

@@ -61,6 +61,10 @@ TEST_CASE("default namespace without version component")
         expected += "_dufio";
 #endif
 
+#ifdef JSON_NO_SIMD
+        expected += "_ns";
+#endif
+
 #ifdef JSON_VIEW_NO_SIMD
         expected += "_vns";
 #endif

tests/benchmarks/README.md

@@ -22,6 +22,15 @@ Micro-benchmarks for parsing, serialization and the binary formats, written with
 | `ViewAccept` | validate with `json_document::accept`; compare with `Accept` |
 | `ViewMaterialize` | convert a parsed `json_document` into a `json` value |
 | `ViewDump` | serialize a parsed `json_document`; compare with `Dump` |
+| `SaxParse` | parse with a SAX handler that only counts events, so no values are built |
+| `Copy`, `CopyAssign`, `CopyRing` | copy a parsed document and destroy a copy, both timed: a temporary copy, an assignment over an earlier copy, and replacing the oldest of eight live copies |
+| `CopyOnly`, `DestroyOnly` | only the copy or only the destruction, the other half untimed; a fast destructor can leave work to the next allocations, so these tell the halves apart but do not compare versions |
+| `Equal`, `Iterate` | compare a parsed document with a copy, or visit every value with range-based `for` loops |
+| `Flatten`, `FlattenOrdered`, `Unflatten` | `flatten()` for `json` and `ordered_json`, and `unflatten()` |
+| `Diff`, `Patch` | `diff()` against the document with every eighth scalar changed, and applying that patch |
+| `JsonPointerParse`, `JsonPointerToString`, `JsonPointerAt`, `JsonPointerContains` | the JSON pointers of every scalar in a document: parse them from strings, turn them back into strings, look them up with `at()` and `contains()` |
+| `GetVectorInt`, `GetVectorDouble`, `GetMap` | `get<std::vector<std::int64_t>>()`, `get<std::vector<double>>()` and `get<std::map<std::string, json>>()` on large arrays and a large flat object |
+| `ToJsonStrings` | create a `json` array from a `std::vector` of 100,000 strings, copied or moved, and destroy it again |
 
 The input files are those of [nativejson-benchmark](https://github.com/miloyip/nativejson-benchmark) (`canada`,
 `citm_catalog`, `twitter`), a large `jeopardy` file, and number-heavy files (`floats`, `signed_ints`, ...).

tests/benchmarks/json_view/.gitignore

@@ -0,0 +1 @@
+build/

tests/benchmarks/json_view/README.md

@@ -0,0 +1,90 @@
+# json_view compared with other libraries
+
+The in-tree benchmarks in [`tests/benchmarks`](../README.md) measure `json_document` against `json::parse` only. The
+programs here compare it with [yyjson](https://github.com/ibireme/yyjson),
+[simdjson](https://github.com/simdjson/simdjson), and [Boost.JSON](https://github.com/boostorg/json): the question
+users ask when they pick a library. They are not built by CMake or run by CI.
+
+## Reproducing the numbers
+
+`compare.py` builds the programs against `include/` of this checkout, runs them, and writes the results together with
+everything needed to reproduce them to `results/<date>-<host>.md` (and `.csv`): the date, the commit, the CPU, the
+OS, the compiler, the flags, and the versions of all libraries.
+
+```sh
+python3 tests/benchmarks/json_view/compare.py --data <json_test_data directory> [--native] [--rounds 30]
+```
+
+- `--data` is the downloaded [test data](https://github.com/nlohmann/json_test_data), e.g. the `test_files` directory
+  of a CMake build directory. It needs `nativejson-benchmark/{twitter,citm_catalog,canada}.json` and
+  `jeopardy/jeopardy.json`.
+- The other libraries come from the system: pkg-config, or Homebrew (`brew install yyjson simdjson boost`). With
+  `--download`, pinned releases are downloaded instead and checked against their SHA-256. Without Boost headers (or
+  with `--no-boost`), the Boost.JSON columns are skipped, and the results say so.
+- `--corpus file...` adds files to the corpus benchmark, e.g. those of
+  [simdjson-data](https://github.com/simdjson/simdjson-data) or the
+  [yyjson benchmark](https://github.com/ibireme/yyjson_benchmark).
+- Only the Python 3 standard library is used; a C++17 compiler is needed (`CXX` and `CC` are honored).
+
+For numbers worth publishing, use a quiet machine (see [Getting stable numbers](../README.md#getting-stable-numbers)),
+the default 30 rounds or more, and `--native` only if the other libraries were built for the same CPU.
+
+### On GitHub-hosted runners
+
+The workflow [json_view benchmarks](../../../.github/workflows/json_view_benchmarks.yml) runs `compare.py --download`
+on demand: by hand (Actions → "json_view benchmarks" → "Run workflow"), on an x86-64 or AArch64 Ubuntu runner with GCC
+or Clang, or when a pull request gets the label `benchmark`, on both architectures with GCC. The results appear as the
+job summary and as an artifact. Shared runners are noisy, so these numbers show
+where `json_view` stands on another architecture; they are not meant for publication.
+
+## What is measured
+
+`bench_view.cpp` runs four workloads on twitter, citm_catalog, canada, jeopardy, a single tweet (`status`), and a
+JSON-RPC request (`rpc`):
+
+| workload | what it does |
+|---|---|
+| parse | build and free a document |
+| traverse | parse, then visit every value, convert every number, touch every string and key |
+| select | parse, then read a few fields per record (e.g. id, user name, and retweet count of each tweet) |
+| dump | serialize a parsed document (compact) |
+
+`bench_corpus.cpp` runs parse, traverse, and dump on any list of files, so that no library is tuned to a handful of
+documents. Its dump also writes the numbers as they are in the input: `json_view` with `number_format::source`, and
+yyjson with numbers read as raw text (`YYJSON_READ_NUMBER_AS_RAW`), without converting them.
+
+`bench_edit.cpp` measures read-modify-write: parse, apply the same logical edits with each library's own API, and
+serialize (compact). Workloads: `patch` (a handful of edits at fixed places) and `update` (edits in every record).
+An editable `json_document` edits in place; yyjson copies its immutable document into a mutable one first
+(`yyjson_doc_mut_copy`); Boost.JSON and `json::parse` build mutable DOMs; simdjson cannot edit a document. All
+outputs are checked to describe the same value.
+
+Before anything is timed, all engines must accept each document and agree on the traversal: the number of values, the
+bytes of all strings and keys, and the sum of all numbers. All engines run interleaved in every round, and the best
+round is reported, as time and as a factor of the `json_view` time (below 1 means faster than `json_view`). Each timed
+call follows an untimed call of the same engine: otherwise the engine after `json::parse` pays for the allocator
+cleaning up the tens of thousands of nodes `json::parse` just freed (with glibc, this made `json_view` look 1.7 times
+slower on citm_catalog traverse).
+
+The engines do not all offer the same features, which the numbers should be read with:
+
+| engine | document | random access | editable | notes |
+|---|---|---|---|---|
+| `json_view` | immutable index into the text | yes | no | a fresh document per parse; "reused" parses into the same document |
+| yyjson | immutable (`yyjson_read`) | yes | via a mutable copy | |
+| simdjson DOM | immutable, parser reused | yes | no | "fresh" uses a new parser per parse |
+| simdjson On-Demand | none: forward-only, lazy | no | no | only traverse and select |
+| Boost.JSON | owning, mutable DOM | yes | yes | monotonic resource |
+| `json::parse` | owning, mutable DOM | yes | yes | |
+
+Reusing memory matters as much as the parser. simdjson DOM reuses its parser, so it writes into memory it already
+touched; a fresh `json_view` document or yyjson document gets new memory for every parse. On Linux, glibc returns large
+blocks to the system when they are freed, so every fresh parse of a large document pays a page fault per 4 KiB page:
+on x86-64 Linux, a fresh `json_view` parse of jeopardy took about twice as long as a reused one. On macOS on Apple
+silicon, with 16 KiB pages, the difference is much smaller. Compare "json_view (reused)" with "simdjson DOM", and the
+fresh `json_view` with "simdjson DOM (fresh)" and yyjson.
+
+## Published results
+
+Results are only published with the file `compare.py` wrote, which names the machine and the versions; see
+`results/`. Numbers from one machine and compiler do not carry over to another: rerun the script.

tests/benchmarks/json_view/bench_corpus.cpp

@@ -0,0 +1,348 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+// Corpus benchmark: the read-only workloads of bench_view.cpp on any list of
+// JSON files (for example the benchmark sets of simdjson and yyjson).
+//
+//   ./bench_corpus [--rounds N] file...
+//
+// For every file, all engines must accept it and agree on a traversal (value
+// count, string bytes, sum of numbers) before anything is timed. Workloads:
+// parse (build and free a document), traverse (visit every value, convert
+// every number), dump (compact), and for json_view also dump with the source
+// number text, compared with yyjson writing numbers read as raw text
+// (YYJSON_READ_NUMBER_AS_RAW). Results go to bench_corpus.csv.
+#include <nlohmann/json_view.hpp>
+
+#if JSON_VIEW_BENCH_BOOST
+    #include <boost/json.hpp>
+    #include <boost/json/src.hpp>
+#endif
+#include <simdjson.h>
+#include <yyjson.h>
+
+#include <algorithm>
+#include <chrono>
+#include <cmath>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <fstream>
+#include <functional>
+#include <sstream>
+#include <string>
+#include <vector>
+
+using nlohmann::json;
+using nlohmann::json_document;
+using nlohmann::json_view;
+
+static volatile double g_sink;
+
+struct stats
+{
+    double num = 0;
+    std::size_t str = 0, nodes = 0;
+};
+
+static void walk(json_view v, stats& st)
+{
+    ++st.nodes;
+    switch (v.type())
+    {
+        case json::value_t::object:
+            for (auto it = v.begin(); it != v.end(); ++it)
+            {
+                st.str += it.key().size();
+                walk(*it, st);
+            }
+            break;
+        case json::value_t::array:
+            for (const json_view e : v)
+            {
+                walk(e, st);
+            }
+            break;
+        case json::value_t::string:
+            st.str += v.get_string().size();
+            break;
+        case json::value_t::number_integer:
+            st.num += static_cast<double>(v.get<std::int64_t>());
+            break;
+        case json::value_t::number_unsigned:
+            st.num += static_cast<double>(v.get<std::uint64_t>());
+            break;
+        case json::value_t::number_float:
+            st.num += v.get<double>();
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk(yyjson_val* v, stats& st)
+{
+    ++st.nodes;
+    switch (yyjson_get_type(v))
+    {
+        case YYJSON_TYPE_OBJ:
+        {
+            std::size_t idx, max;
+            yyjson_val* k, * val;
+            yyjson_obj_foreach(v, idx, max, k, val)
+            {
+                st.str += yyjson_get_len(k);
+                walk(val, st);
+            }
+            break;
+        }
+        case YYJSON_TYPE_ARR:
+        {
+            std::size_t idx, max;
+            yyjson_val* val;
+            yyjson_arr_foreach(v, idx, max, val)
+            {
+                walk(val, st);
+            }
+            break;
+        }
+        case YYJSON_TYPE_STR:
+            st.str += yyjson_get_len(v);
+            break;
+        case YYJSON_TYPE_NUM:
+            st.num += yyjson_is_sint(v) ? static_cast<double>(yyjson_get_sint(v)) : yyjson_is_uint(v) ? static_cast<double>(yyjson_get_uint(v)) : yyjson_get_real(v);
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk(simdjson::dom::element e, stats& st)
+{
+    ++st.nodes;
+    switch (e.type())
+    {
+        case simdjson::dom::element_type::OBJECT:
+            for (auto f : simdjson::dom::object(e))
+            {
+                st.str += f.key.size();
+                walk(f.value, st);
+            }
+            break;
+        case simdjson::dom::element_type::ARRAY:
+            for (auto c : simdjson::dom::array(e))
+            {
+                walk(c, st);
+            }
+            break;
+        case simdjson::dom::element_type::STRING:
+            st.str += std::string_view(e).size();
+            break;
+        case simdjson::dom::element_type::INT64:
+            st.num += static_cast<double>(int64_t(e));
+            break;
+        case simdjson::dom::element_type::UINT64:
+            st.num += static_cast<double>(uint64_t(e));
+            break;
+        case simdjson::dom::element_type::DOUBLE:
+            st.num += double(e);
+            break;
+        default:
+            break;
+    }
+}
+
+#if JSON_VIEW_BENCH_BOOST
+static void walk(const boost::json::value& v, stats& st)
+{
+    ++st.nodes;
+    switch (v.kind())
+    {
+        case boost::json::kind::object:
+            for (const auto& kv : v.get_object())
+            {
+                st.str += kv.key().size();
+                walk(kv.value(), st);
+            }
+            break;
+        case boost::json::kind::array:
+            for (const auto& c : v.get_array())
+            {
+                walk(c, st);
+            }
+            break;
+        case boost::json::kind::string:
+            st.str += v.get_string().size();
+            break;
+        case boost::json::kind::int64:
+            st.num += static_cast<double>(v.get_int64());
+            break;
+        case boost::json::kind::uint64:
+            st.num += static_cast<double>(v.get_uint64());
+            break;
+        case boost::json::kind::double_:
+            st.num += v.get_double();
+            break;
+        default:
+            break;
+    }
+}
+#endif
+
+static std::string slurp(const std::string& p)
+{
+    std::ifstream f(p, std::ios::binary);
+    if (!f)
+    {
+        std::fprintf(stderr, "cannot open %s\n", p.c_str());
+        std::exit(1);
+    }
+    std::stringstream ss;
+    ss << f.rdbuf();
+    return ss.str();
+}
+
+static bool same(const stats& a, const stats& b)
+{
+    return a.nodes == b.nodes && a.str == b.str && (a.num == b.num || std::fabs(a.num - b.num) <= 1e-9 * std::fabs(a.num));
+}
+
+int main(int argc, char** argv)
+{
+    int rounds = 0; // 0: by size
+    std::vector<std::string> files;
+    for (int i = 1; i < argc; ++i)
+    {
+        if (std::strcmp(argv[i], "--rounds") == 0 && i + 1 < argc)
+        {
+            // (clamped: 0 rounds by size, at most a million)
+            rounds = static_cast<int>((std::min)((std::max)(std::strtol(argv[++i], nullptr, 10), 0L), 1000000L));
+        }
+        else
+        {
+            files.push_back(argv[i]);
+        }
+    }
+    std::FILE* csv = std::fopen("bench_corpus.csv", "w");
+    if (csv == nullptr)
+    {
+        std::perror("bench_corpus.csv");
+        return 1;
+    }
+    std::fprintf(csv, "file,bytes,workload,engine,ns\n");
+    json_document reused;
+    simdjson::dom::parser sj;
+    for (const auto& path : files)
+    {
+        const std::string s = slurp(path);
+        const std::string name = path.substr(path.rfind('/') + 1);
+        const simdjson::padded_string ps(s);
+
+        // all engines must agree before timing
+        stats a, b, c, d;
+        const json_document doc = json_document::parse(s);
+        walk(doc.root(), a);
+        yyjson_doc* y = yyjson_read(s.data(), s.size(), 0);
+        auto sjr = sj.parse(ps);
+#if JSON_VIEW_BENCH_BOOST
+        boost::json::parse_options opt;
+        opt.numbers = boost::json::number_precision::precise;
+        boost::json::monotonic_resource mr0;
+        const boost::json::value bv = boost::json::parse(s, &mr0, opt);
+#endif
+        if (y == nullptr || sjr.error())
+        {
+            std::printf("%-34s skipped (an engine rejects it)\n", name.c_str());
+            yyjson_doc_free(y);
+            continue;
+        }
+        walk(yyjson_doc_get_root(y), b);
+        walk(sjr.value_unsafe(), c);
+#if JSON_VIEW_BENCH_BOOST
+        walk(bv, d);
+#else
+        d = a;
+#endif
+        yyjson_doc_free(y);
+        const bool ok = same(a, b) && same(a, c) && same(a, d);
+
+        const int r = rounds > 0 ? rounds : static_cast<int>(std::max<std::size_t>(3, std::min<std::size_t>(60, 400000000 / (s.size() + 1))));
+        struct engine
+        {
+            std::string name;
+            std::function<void()> fn;
+        };
+        json_document vd = json_document::parse(s);
+        yyjson_doc* yd = yyjson_read(s.data(), s.size(), 0);
+        yyjson_doc* yd_raw = yyjson_read(s.data(), s.size(), YYJSON_READ_NUMBER_AS_RAW);
+        simdjson::dom::parser sjd;
+        const simdjson::dom::element se = sjd.parse(ps).value_unsafe();
+        const std::vector<std::pair<std::string, std::vector<engine>>> workloads =
+        {
+            {
+                "parse", {
+                    {"json_view", [&] { auto x = json_document::parse(s); g_sink = static_cast<double>(x.node_count()); }},
+                    {"json_view (reused)", [&] { reused.read(s); g_sink = static_cast<double>(reused.node_count()); }},
+                    {"yyjson", [&] { yyjson_doc* x = yyjson_read(s.data(), s.size(), 0); g_sink = static_cast<double>(yyjson_doc_get_val_count(x)); yyjson_doc_free(x); }},
+                    {"simdjson DOM", [&] { auto e = sj.parse(ps).value_unsafe(); g_sink = e.is_object(); }},
+                    {"simdjson DOM (fresh)", [&] { simdjson::dom::parser p; auto e = p.parse(ps).value_unsafe(); g_sink = e.is_object(); }},
+#if JSON_VIEW_BENCH_BOOST
+                    {"Boost.JSON", [&] { boost::json::monotonic_resource mr; auto v = boost::json::parse(s, &mr); g_sink = v.is_object(); }},
+#endif
+                }
+            },
+            {
+                "traverse", {
+                    {"json_view", [&] { auto x = json_document::parse(s); stats st; walk(x.root(), st); g_sink = st.num; }},
+                    {"yyjson", [&] { yyjson_doc* x = yyjson_read(s.data(), s.size(), 0); stats st; walk(yyjson_doc_get_root(x), st); g_sink = st.num; yyjson_doc_free(x); }},
+                    {"simdjson DOM", [&] { stats st; walk(sj.parse(ps).value_unsafe(), st); g_sink = st.num; }},
+#if JSON_VIEW_BENCH_BOOST
+                    {"Boost.JSON", [&] { boost::json::monotonic_resource mr; auto v = boost::json::parse(s, &mr); stats st; walk(v, st); g_sink = st.num; }},
+#endif
+                }
+            },
+            {
+                "dump", {
+                    {"json_view", [&] { std::string o = vd.root().dump(); g_sink = static_cast<double>(o.size()); }},
+                    {"yyjson", [&] { std::size_t n = 0; char* o = yyjson_write(yd, 0, &n); g_sink = static_cast<double>(n); std::free(o); }},
+                    {"simdjson DOM", [&] { std::string o = simdjson::to_string(se); g_sink = static_cast<double>(o.size()); }},
+                    {"json_view (source numbers)", [&] { std::string o = vd.root().dump(-1, ' ', false, json_view::number_format::source); g_sink = static_cast<double>(o.size()); }},
+                    {"yyjson (raw numbers)", [&] { std::size_t n = 0; char* o = yyjson_write(yd_raw, 0, &n); g_sink = static_cast<double>(n); std::free(o); }},
+                }
+            },
+        };
+        std::printf("%-34s %9zu B%s\n", name.c_str(), s.size(), ok ? "" : "  [ENGINES DISAGREE]");
+        for (const auto& wl : workloads)
+        {
+            std::vector<double> best(wl.second.size(), 1e300);
+            for (int i = 0; i < r; ++i)
+            {
+                for (std::size_t k = 0; k < wl.second.size(); ++k)
+                {
+                    // an untimed call first: whatever the previous engine left to the allocator
+                    // (e.g. thousands of freed json nodes) is cleaned up here, not in the timing
+                    wl.second[k].fn();
+                    const auto t0 = std::chrono::steady_clock::now();
+                    wl.second[k].fn();
+                    best[k] = std::min(best[k], std::chrono::duration<double, std::nano>(std::chrono::steady_clock::now() - t0).count());
+                }
+            }
+            std::printf("  %-9s", wl.first.c_str());
+            for (std::size_t k = 0; k < wl.second.size(); ++k)
+            {
+                std::printf("  %s %.2f GB/s (%.2fx)", wl.second[k].name.c_str(), static_cast<double>(s.size()) / best[k], best[k] / best[0]);
+                std::fprintf(csv, "%s,%zu,%s,%s,%.1f\n", name.c_str(), s.size(), wl.first.c_str(), wl.second[k].name.c_str(), best[k]);
+            }
+            std::printf("\n");
+            std::fflush(stdout);
+        }
+        yyjson_doc_free(yd);
+        yyjson_doc_free(yd_raw);
+    }
+    std::fclose(csv);
+}

tests/benchmarks/json_view/bench_edit.cpp

@@ -0,0 +1,589 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+// Read-modify-write benchmark: parse a document, apply the same logical edits
+// with each library's own API, and serialize it (compact).
+//
+//   json_view     json_editable_document: edits in place, unchanged values stay in the index
+//   yyjson        yyjson_read + yyjson_doc_mut_copy (the way to edit a parsed document)
+//   Boost.JSON    parse into a mutable DOM (monotonic resource, precise numbers), serialize
+//   json::parse   nlohmann::json today
+// simdjson has no mutable document and is not part of this comparison.
+//
+// Workloads:
+//   patch   a handful of edits at fixed places (scalars, a new member, a new array element)
+//   update  edits in every record (twitter: 100 statuses, citm: 243 performances,
+//           canada: 480 rings, jeopardy: 216,930 questions): set scalars, erase a
+//           member, add a member (canada: replace the first point of every ring)
+//
+// Build: see README.md (same flags as bench_view.cpp).
+#include <nlohmann/json_view.hpp>
+
+#if JSON_VIEW_BENCH_BOOST
+    #include <boost/json.hpp>
+    #include <boost/json/src.hpp>
+#endif
+#include <yyjson.h>
+
+#include <algorithm>
+#include <chrono>
+#include <cstdio>
+#include <cstdlib>
+#include <fstream>
+#include <functional>
+#include <sstream>
+
+using nlohmann::json;
+using nlohmann::json_editable_document;
+using nlohmann::json_editable_view;
+#if JSON_VIEW_BENCH_BOOST
+    namespace bj = boost::json;
+#endif
+
+static volatile std::size_t g_sink;
+
+// ---------------- json_view ----------------
+
+static std::string edit_view(const std::string& name, const std::string& s, bool update)
+{
+    json_editable_document d = json_editable_document::parse(s);
+    const json_editable_view r = d.root();
+    if (name == "twitter")
+    {
+        if (update)
+        {
+            std::int64_t i = 0;
+            for (const json_editable_view st : r["statuses"])
+            {
+                d.set(st, "retweet_count", i++);
+                d.set(st, "favorited", true);
+                d.set(st, "text", "redacted");
+                d.erase(st, "entities");
+                d.set(st, "edited", true);
+            }
+        }
+        else
+        {
+            d.set(r["search_metadata"], "count", 200);
+            d.set(r["statuses"][0], "text", "patched");
+            d.set(r["statuses"][0]["user"], "followers_count", 1);
+            d.set(r["statuses"][99], "favorited", true);
+            d.set(r, "patched", true);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        if (update)
+        {
+            for (const json_editable_view p : r["performances"])
+            {
+                d.set(p, "name", "performance");
+                d.set(p, "start", p["start"].get<std::int64_t>() + 1);
+                d.erase(p, "seatMapImage");
+                d.set(p, "edited", true);
+            }
+        }
+        else
+        {
+            d.set(r["events"]["138586341"], "name", "patched");
+            d.set(r["performances"][0], "start", 0);
+            d.set(r["venueNames"], "PLEYEL_PLEYEL", "Salle");
+            d.set(r, "patched", true);
+        }
+    }
+    else if (name == "canada")
+    {
+        const json_editable_view coords = r["features"][0]["geometry"]["coordinates"];
+        if (update)
+        {
+            for (const json_editable_view ring : coords)
+            {
+                d.set(ring, 0, json::array({0.5, 0.5}));
+            }
+        }
+        else
+        {
+            d.set(r["features"][0]["properties"], "name", "patched");
+            d.set(r, "type", "FeatureCollection2");
+            d.set(coords[0], 0, json::array({0.0, 0.0}));
+        }
+    }
+    else if (name == "jeopardy")
+    {
+        if (update)
+        {
+            for (const json_editable_view q : r)
+            {
+                d.set(q, "value", "$1");
+                d.erase(q, "air_date");
+            }
+        }
+        else
+        {
+            d.set(r[0], "value", "$0");
+            d.set(r[100000], "answer", "patched");
+            d.set(r[216929], "round", "x");
+            d.push_back(r, json::object({{"category", "NEW"}, {"value", "$5"}}));
+        }
+    }
+    else if (name == "status")
+    {
+        d.set(r, "retweet_count", 1);
+        d.set(r["user"], "name", "x");
+    }
+    else if (name == "rpc")
+    {
+        d.set(r, "id", 4);
+        d.set(r["params"], "subtrahend", 24);
+    }
+    return r.dump();
+}
+
+// ---------------- nlohmann::json ----------------
+
+static std::string edit_json(const std::string& name, const std::string& s, bool update)
+{
+    json r = json::parse(s);
+    if (name == "twitter")
+    {
+        if (update)
+        {
+            std::int64_t i = 0;
+            for (auto& st : r["statuses"])
+            {
+                st["retweet_count"] = i++;
+                st["favorited"] = true;
+                st["text"] = "redacted";
+                st.erase("entities");
+                st["edited"] = true;
+            }
+        }
+        else
+        {
+            r["search_metadata"]["count"] = 200;
+            r["statuses"][0]["text"] = "patched";
+            r["statuses"][0]["user"]["followers_count"] = 1;
+            r["statuses"][99]["favorited"] = true;
+            r["patched"] = true;
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        if (update)
+        {
+            for (auto& p : r["performances"])
+            {
+                p["name"] = "performance";
+                p["start"] = p["start"].get<std::int64_t>() + 1;
+                p.erase("seatMapImage");
+                p["edited"] = true;
+            }
+        }
+        else
+        {
+            r["events"]["138586341"]["name"] = "patched";
+            r["performances"][0]["start"] = 0;
+            r["venueNames"]["PLEYEL_PLEYEL"] = "Salle";
+            r["patched"] = true;
+        }
+    }
+    else if (name == "canada")
+    {
+        json& coords = r["features"][0]["geometry"]["coordinates"];
+        if (update)
+        {
+            for (auto& ring : coords)
+            {
+                ring[0] = json::array({0.5, 0.5});
+            }
+        }
+        else
+        {
+            r["features"][0]["properties"]["name"] = "patched";
+            r["type"] = "FeatureCollection2";
+            coords[0][0] = json::array({0.0, 0.0});
+        }
+    }
+    else if (name == "jeopardy")
+    {
+        if (update)
+        {
+            for (auto& q : r)
+            {
+                q["value"] = "$1";
+                q.erase("air_date");
+            }
+        }
+        else
+        {
+            r[0]["value"] = "$0";
+            r[100000]["answer"] = "patched";
+            r[216929]["round"] = "x";
+            r.push_back(json::object({{"category", "NEW"}, {"value", "$5"}}));
+        }
+    }
+    else if (name == "status")
+    {
+        r["retweet_count"] = 1;
+        r["user"]["name"] = "x";
+    }
+    else if (name == "rpc")
+    {
+        r["id"] = 4;
+        r["params"]["subtrahend"] = 24;
+    }
+    return r.dump();
+}
+
+// ---------------- yyjson ----------------
+
+static std::string edit_yyjson(const std::string& name, const std::string& s, bool update)
+{
+    yyjson_doc* idoc = yyjson_read(s.data(), s.size(), 0);
+    yyjson_mut_doc* d = yyjson_doc_mut_copy(idoc, nullptr);
+    yyjson_doc_free(idoc);
+    yyjson_mut_val* r = yyjson_mut_doc_get_root(d);
+    auto get = [](yyjson_mut_val * o, const char* k)
+    {
+        return yyjson_mut_obj_get(o, k);
+    };
+    if (name == "twitter")
+    {
+        yyjson_mut_val* sts = get(r, "statuses");
+        if (update)
+        {
+            std::size_t idx, max;
+            yyjson_mut_val* st;
+            std::int64_t i = 0;
+            yyjson_mut_arr_foreach(sts, idx, max, st)
+            {
+                yyjson_mut_set_sint(get(st, "retweet_count"), i++);
+                yyjson_mut_set_bool(get(st, "favorited"), true);
+                yyjson_mut_set_str(get(st, "text"), "redacted");
+                yyjson_mut_obj_remove_key(st, "entities");
+                yyjson_mut_obj_add_bool(d, st, "edited", true);
+            }
+        }
+        else
+        {
+            yyjson_mut_set_sint(get(get(r, "search_metadata"), "count"), 200);
+            yyjson_mut_val* s0 = yyjson_mut_arr_get(sts, 0);
+            yyjson_mut_set_str(get(s0, "text"), "patched");
+            yyjson_mut_set_sint(get(get(s0, "user"), "followers_count"), 1);
+            yyjson_mut_set_bool(get(yyjson_mut_arr_get(sts, 99), "favorited"), true);
+            yyjson_mut_obj_add_bool(d, r, "patched", true);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        if (update)
+        {
+            std::size_t idx, max;
+            yyjson_mut_val* p;
+            yyjson_mut_arr_foreach(get(r, "performances"), idx, max, p)
+            {
+                yyjson_mut_set_str(get(p, "name"), "performance");
+                yyjson_mut_val* start = get(p, "start");
+                yyjson_mut_set_sint(start, yyjson_mut_get_sint(start) + 1);
+                yyjson_mut_obj_remove_key(p, "seatMapImage");
+                yyjson_mut_obj_add_bool(d, p, "edited", true);
+            }
+        }
+        else
+        {
+            yyjson_mut_set_str(get(get(get(r, "events"), "138586341"), "name"), "patched");
+            yyjson_mut_set_sint(get(yyjson_mut_arr_get(get(r, "performances"), 0), "start"), 0);
+            yyjson_mut_set_str(get(get(r, "venueNames"), "PLEYEL_PLEYEL"), "Salle");
+            yyjson_mut_obj_add_bool(d, r, "patched", true);
+        }
+    }
+    else if (name == "canada")
+    {
+        yyjson_mut_val* f0 = yyjson_mut_arr_get(get(r, "features"), 0);
+        yyjson_mut_val* coords = get(get(f0, "geometry"), "coordinates");
+        if (update)
+        {
+            static const double half[2] = {0.5, 0.5};
+            std::size_t idx, max;
+            yyjson_mut_val* ring;
+            yyjson_mut_arr_foreach(coords, idx, max, ring)
+            {
+                yyjson_mut_arr_replace(ring, 0, yyjson_mut_arr_with_real(d, half, 2));
+            }
+        }
+        else
+        {
+            static const double zero[2] = {0.0, 0.0};
+            yyjson_mut_set_str(get(get(f0, "properties"), "name"), "patched");
+            yyjson_mut_set_str(get(r, "type"), "FeatureCollection2");
+            yyjson_mut_arr_replace(yyjson_mut_arr_get(coords, 0), 0, yyjson_mut_arr_with_real(d, zero, 2));
+        }
+    }
+    else if (name == "jeopardy")
+    {
+        if (update)
+        {
+            std::size_t idx, max;
+            yyjson_mut_val* q;
+            yyjson_mut_arr_foreach(r, idx, max, q)
+            {
+                yyjson_mut_set_str(get(q, "value"), "$1");
+                yyjson_mut_obj_remove_key(q, "air_date");
+            }
+        }
+        else
+        {
+            yyjson_mut_set_str(get(yyjson_mut_arr_get(r, 0), "value"), "$0");
+            yyjson_mut_set_str(get(yyjson_mut_arr_get(r, 100000), "answer"), "patched");
+            yyjson_mut_set_str(get(yyjson_mut_arr_get(r, 216929), "round"), "x");
+            yyjson_mut_val* o = yyjson_mut_obj(d);
+            yyjson_mut_obj_add_str(d, o, "category", "NEW");
+            yyjson_mut_obj_add_str(d, o, "value", "$5");
+            yyjson_mut_arr_append(r, o);
+        }
+    }
+    else if (name == "status")
+    {
+        yyjson_mut_set_sint(get(r, "retweet_count"), 1);
+        yyjson_mut_set_str(get(get(r, "user"), "name"), "x");
+    }
+    else if (name == "rpc")
+    {
+        yyjson_mut_set_sint(get(r, "id"), 4);
+        yyjson_mut_set_sint(get(get(r, "params"), "subtrahend"), 24);
+    }
+    std::size_t n = 0;
+    char* out = yyjson_mut_write(d, 0, &n);
+    std::string result(out, n);
+    std::free(out);
+    yyjson_mut_doc_free(d);
+    return result;
+}
+
+#if JSON_VIEW_BENCH_BOOST
+// ---------------- Boost.JSON ----------------
+
+static std::string edit_boost(const std::string& name, const std::string& s, bool update)
+{
+    bj::monotonic_resource mr;
+    bj::parse_options opt;
+    opt.numbers = bj::number_precision::precise; // correctly rounded, like the others
+    bj::value v = bj::parse(s, &mr, opt);
+    bj::object* const obj = v.if_object(); // nullptr for jeopardy (an array)
+    if (name == "twitter")
+    {
+        bj::array& sts = (*obj)["statuses"].as_array();
+        if (update)
+        {
+            std::int64_t i = 0;
+            for (auto& e : sts)
+            {
+                bj::object& st = e.as_object();
+                st["retweet_count"] = i++;
+                st["favorited"] = true;
+                st["text"] = "redacted";
+                st.erase("entities");
+                st["edited"] = true;
+            }
+        }
+        else
+        {
+            (*obj)["search_metadata"].as_object()["count"] = 200;
+            bj::object& s0 = sts[0].as_object();
+            s0["text"] = "patched";
+            s0["user"].as_object()["followers_count"] = 1;
+            sts[99].as_object()["favorited"] = true;
+            (*obj)["patched"] = true;
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        if (update)
+        {
+            for (auto& e : (*obj)["performances"].as_array())
+            {
+                bj::object& p = e.as_object();
+                p["name"] = "performance";
+                p["start"] = p["start"].as_int64() + 1;
+                p.erase("seatMapImage");
+                p["edited"] = true;
+            }
+        }
+        else
+        {
+            (*obj)["events"].as_object()["138586341"].as_object()["name"] = "patched";
+            (*obj)["performances"].as_array()[0].as_object()["start"] = 0;
+            (*obj)["venueNames"].as_object()["PLEYEL_PLEYEL"] = "Salle";
+            (*obj)["patched"] = true;
+        }
+    }
+    else if (name == "canada")
+    {
+        bj::object& f0 = (*obj)["features"].as_array()[0].as_object();
+        bj::array& coords = f0["geometry"].as_object()["coordinates"].as_array();
+        if (update)
+        {
+            for (auto& ring : coords)
+            {
+                ring.as_array()[0] = bj::array({0.5, 0.5});
+            }
+        }
+        else
+        {
+            f0["properties"].as_object()["name"] = "patched";
+            (*obj)["type"] = "FeatureCollection2";
+            coords[0].as_array()[0] = bj::array({0.0, 0.0});
+        }
+    }
+    else if (name == "jeopardy")
+    {
+        bj::array& a = v.as_array();
+        if (update)
+        {
+            for (auto& e : a)
+            {
+                bj::object& q = e.as_object();
+                q["value"] = "$1";
+                q.erase("air_date");
+            }
+        }
+        else
+        {
+            a[0].as_object()["value"] = "$0";
+            a[100000].as_object()["answer"] = "patched";
+            a[216929].as_object()["round"] = "x";
+            a.push_back(bj::object({{"category", "NEW"}, {"value", "$5"}}));
+        }
+    }
+    else if (name == "status")
+    {
+        (*obj)["retweet_count"] = 1;
+        (*obj)["user"].as_object()["name"] = "x";
+    }
+    else if (name == "rpc")
+    {
+        (*obj)["id"] = 4;
+        (*obj)["params"].as_object()["subtrahend"] = 24;
+    }
+    return bj::serialize(v);
+}
+#endif
+
+// ---------------- harness ----------------
+
+static std::string slurp(const std::string& p)
+{
+    std::ifstream f(p, std::ios::binary);
+    if (!f)
+    {
+        std::fprintf(stderr, "cannot open %s\n", p.c_str());
+        std::exit(1);
+    }
+    std::stringstream ss;
+    ss << f.rdbuf();
+    return ss.str();
+}
+
+int main(int argc, char** argv)
+{
+    if (argc < 2)
+    {
+        std::fprintf(stderr, "usage: %s <json_test_data directory> [rounds] [document]\n", argv[0]);
+        return 1;
+    }
+    const std::string T = std::string(argv[1]) + "/";
+    // (clamped to 1 to a million)
+    const int rounds = argc > 2 ? static_cast<int>((std::min)((std::max)(std::strtol(argv[2], nullptr, 10), 1L), 1000000L)) : 20;
+    const std::string only = argc > 3 ? argv[3] : "";
+    struct doc
+    {
+        std::string name, text;
+        int batch;
+    };
+    std::vector<doc> docs;
+    for (const char* f :
+            {"nativejson-benchmark/twitter.json", "nativejson-benchmark/citm_catalog.json", "nativejson-benchmark/canada.json", "jeopardy/jeopardy.json"
+            })
+    {
+        std::string n = std::string(f).substr(std::string(f).find('/') + 1);
+        docs.push_back({n.substr(0, n.size() - 5), slurp(T + f), 1});
+    }
+    docs.push_back({"status", json::parse(docs[0].text)["statuses"][0].dump(), 200});
+    docs.push_back({"rpc", R"({"jsonrpc": "2.0", "method": "subtract", "params": {"minuend": 42, "subtrahend": 23}, "id": 3})", 5000});
+
+    using fn = std::string (*)(const std::string&, const std::string&, bool);
+    const std::vector<std::pair<std::string, fn>> engines =
+    {
+        {"json_view", edit_view}, {"yyjson", edit_yyjson},
+#if JSON_VIEW_BENCH_BOOST
+        {"Boost.JSON", edit_boost},
+#endif
+        {"json::parse", edit_json}
+    };
+
+    std::FILE* csv = std::fopen("bench_edit.csv", "w");
+    if (csv == nullptr)
+    {
+        std::perror("bench_edit.csv");
+        return 1;
+    }
+    std::fprintf(csv, "doc,bytes,workload,engine,ns\n");
+    for (const auto& dc : docs)
+    {
+        if (!only.empty() && dc.name != only)
+        {
+            continue;
+        }
+        for (const bool update :
+                {
+                    false, true
+                })
+        {
+            if (update && (dc.name == "status" || dc.name == "rpc"))
+            {
+                continue;
+            }
+            // all engines must produce the same value
+            const json expected = json::parse(edit_json(dc.name, dc.text, update));
+            bool ok = true;
+            for (const auto& e : engines)
+            {
+                ok = ok && json::parse(e.second(dc.name, dc.text, update)) == expected;
+            }
+            std::vector<double> best(engines.size(), 1e300);
+            const int r = dc.text.size() > 10000000 ? std::max(3, rounds / 4) : rounds;
+            for (int i = 0; i < r; ++i)
+            {
+                for (std::size_t k = 0; k < engines.size(); ++k)
+                {
+                    // an untimed call first: whatever the previous engine left to the allocator
+                    // (e.g. thousands of freed json nodes) is cleaned up here, not in the timing
+                    g_sink = engines[k].second(dc.name, dc.text, update).size();
+                    const auto t0 = std::chrono::steady_clock::now();
+                    for (int b = 0; b < dc.batch; ++b)
+                    {
+                        g_sink = engines[k].second(dc.name, dc.text, update).size();
+                    }
+                    const double ns = std::chrono::duration<double, std::nano>(std::chrono::steady_clock::now() - t0).count() / dc.batch;
+                    best[k] = std::min(best[k], ns);
+                }
+            }
+            const char* wl = update ? "update" : "patch";
+            std::printf("%-13s %-7s %s", dc.name.c_str(), wl, ok ? "" : "[OUTPUT MISMATCH] ");
+            for (std::size_t k = 0; k < engines.size(); ++k)
+            {
+                const double us = best[k] / 1e3;
+                std::printf("  %s %.*fus (%.2fx)", engines[k].first.c_str(), us < 10 ? 3 : (us < 1000 ? 1 : 0), us, best[k] / best[0]);
+                std::fprintf(csv, "%s,%zu,%s,%s,%.1f\n", dc.name.c_str(), dc.text.size(), wl, engines[k].first.c_str(), best[k]);
+            }
+            std::printf("\n");
+            std::fflush(stdout);
+        }
+    }
+    std::fclose(csv);
+}

tests/benchmarks/json_view/bench_view.cpp

@@ -0,0 +1,755 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+// Same-feature-set benchmark: read-only JSON documents with random access.
+//
+//   json_view      nlohmann/json_view.hpp (fresh document per parse / reused)
+//   yyjson         yyjson_read(): immutable document, random access
+//   simdjson DOM   dom::parser (reused, as recommended; "fresh": a new parser
+//                  per parse): immutable, random access
+// references (different feature sets):
+//   simdjson OD    On-Demand: forward-only, lazy
+//   Boost.JSON     owning, mutable DOM (monotonic resource)
+//   json::parse    owning, mutable DOM (nlohmann today)
+//
+// Workloads: parse (build + free), traverse (visit everything, convert every
+// number, touch every string and key), select (a few fields per document),
+// dump (compact serialization of the parsed document).
+// All engines run interleaved in every round, each timed call after an untimed
+// one of the same engine; the best round is reported.
+#include <nlohmann/json_view.hpp>
+
+#if JSON_VIEW_BENCH_BOOST
+    #include <boost/json.hpp>
+    #include <boost/json/src.hpp>
+#endif
+#include <simdjson.h>
+#include <yyjson.h>
+
+#include <algorithm>
+#include <chrono>
+#include <cmath>
+#include <cstdio>
+#include <cstdlib>
+#include <fstream>
+#include <functional>
+#include <map>
+#include <sstream>
+
+using nlohmann::json;
+using nlohmann::json_document;
+using nlohmann::json_view;
+
+static volatile double g_sink;
+
+struct stats
+{
+    double num = 0;
+    std::size_t str = 0, nodes = 0;
+};
+
+// ---------------- traversal ----------------
+
+static void walk(json_view v, stats& st)
+{
+    ++st.nodes;
+    switch (v.type())
+    {
+        case json::value_t::object:
+            for (auto it = v.begin(); it != v.end(); ++it)
+            {
+                st.str += it.key().size();
+                walk(*it, st);
+            }
+            break;
+        case json::value_t::array:
+            for (const json_view e : v)
+            {
+                walk(e, st);
+            }
+            break;
+        case json::value_t::string:
+            st.str += v.get_string().size();
+            break;
+        case json::value_t::number_integer:
+            st.num += static_cast<double>(v.get<std::int64_t>());
+            break;
+        case json::value_t::number_unsigned:
+            st.num += static_cast<double>(v.get<std::uint64_t>());
+            break;
+        case json::value_t::number_float:
+            st.num += v.get<double>();
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk(const json& j, stats& st)
+{
+    ++st.nodes;
+    switch (j.type())
+    {
+        case json::value_t::object:
+            for (const auto& kv : j.get_ref<const json::object_t&>())
+            {
+                st.str += kv.first.size();
+                walk(kv.second, st);
+            }
+            break;
+        case json::value_t::array:
+            for (const auto& e : j.get_ref<const json::array_t&>())
+            {
+                walk(e, st);
+            }
+            break;
+        case json::value_t::string:
+            st.str += j.get_ref<const std::string&>().size();
+            break;
+        case json::value_t::number_integer:
+            st.num += static_cast<double>(*j.get_ptr<const json::number_integer_t*>());
+            break;
+        case json::value_t::number_unsigned:
+            st.num += static_cast<double>(*j.get_ptr<const json::number_unsigned_t*>());
+            break;
+        case json::value_t::number_float:
+            st.num += *j.get_ptr<const json::number_float_t*>();
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk(yyjson_val* v, stats& st)
+{
+    ++st.nodes;
+    switch (yyjson_get_type(v))
+    {
+        case YYJSON_TYPE_OBJ:
+        {
+            std::size_t idx, max;
+            yyjson_val* k, * val;
+            yyjson_obj_foreach(v, idx, max, k, val)
+            {
+                st.str += yyjson_get_len(k);
+                walk(val, st);
+            }
+            break;
+        }
+        case YYJSON_TYPE_ARR:
+        {
+            std::size_t idx, max;
+            yyjson_val* val;
+            yyjson_arr_foreach(v, idx, max, val)
+            {
+                walk(val, st);
+            }
+            break;
+        }
+        case YYJSON_TYPE_STR:
+            st.str += yyjson_get_len(v);
+            break;
+        case YYJSON_TYPE_NUM:
+            if (yyjson_is_sint(v))
+            {
+                st.num += static_cast<double>(yyjson_get_sint(v));
+            }
+            else if (yyjson_is_uint(v))
+            {
+                st.num += static_cast<double>(yyjson_get_uint(v));
+            }
+            else
+            {
+                st.num += yyjson_get_real(v);
+            }
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk(simdjson::dom::element e, stats& st)
+{
+    ++st.nodes;
+    switch (e.type())
+    {
+        case simdjson::dom::element_type::OBJECT:
+            for (auto f : simdjson::dom::object(e))
+            {
+                st.str += f.key.size();
+                walk(f.value, st);
+            }
+            break;
+        case simdjson::dom::element_type::ARRAY:
+            for (auto c : simdjson::dom::array(e))
+            {
+                walk(c, st);
+            }
+            break;
+        case simdjson::dom::element_type::STRING:
+            st.str += std::string_view(e).size();
+            break;
+        case simdjson::dom::element_type::INT64:
+            st.num += static_cast<double>(int64_t(e));
+            break;
+        case simdjson::dom::element_type::UINT64:
+            st.num += static_cast<double>(uint64_t(e));
+            break;
+        case simdjson::dom::element_type::DOUBLE:
+            st.num += double(e);
+            break;
+        default:
+            break;
+    }
+}
+
+static void walk_od(simdjson::ondemand::value v, stats& st)
+{
+    ++st.nodes;
+    switch (v.type())
+    {
+        case simdjson::ondemand::json_type::object:
+            for (auto f : v.get_object())
+            {
+                st.str += std::string_view(f.unescaped_key()).size();
+                walk_od(f.value(), st);
+            }
+            break;
+        case simdjson::ondemand::json_type::array:
+            for (auto c : v.get_array())
+            {
+                walk_od(c.value(), st);
+            }
+            break;
+        case simdjson::ondemand::json_type::string:
+            st.str += std::string_view(v.get_string()).size();
+            break;
+        case simdjson::ondemand::json_type::number:
+        {
+            simdjson::ondemand::number n = v.get_number();
+            switch (n.get_number_type())
+            {
+                case simdjson::ondemand::number_type::signed_integer:
+                    st.num += static_cast<double>(n.get_int64());
+                    break;
+                case simdjson::ondemand::number_type::unsigned_integer:
+                    st.num += static_cast<double>(n.get_uint64());
+                    break;
+                default:
+                    st.num += n.get_double();
+                    break;
+            }
+            break;
+        }
+        case simdjson::ondemand::json_type::boolean:
+            (void)bool(v.get_bool());
+            break;
+        default:
+            (void)v.is_null();
+            break;
+    }
+}
+
+#if JSON_VIEW_BENCH_BOOST
+static void walk(const boost::json::value& v, stats& st)
+{
+    ++st.nodes;
+    switch (v.kind())
+    {
+        case boost::json::kind::object:
+            for (const auto& kv : v.get_object())
+            {
+                st.str += kv.key().size();
+                walk(kv.value(), st);
+            }
+            break;
+        case boost::json::kind::array:
+            for (const auto& c : v.get_array())
+            {
+                walk(c, st);
+            }
+            break;
+        case boost::json::kind::string:
+            st.str += v.get_string().size();
+            break;
+        case boost::json::kind::int64:
+            st.num += static_cast<double>(v.get_int64());
+            break;
+        case boost::json::kind::uint64:
+            st.num += static_cast<double>(v.get_uint64());
+            break;
+        case boost::json::kind::double_:
+            st.num += v.get_double();
+            break;
+        default:
+            break;
+    }
+}
+#endif
+
+// ---------------- selective access ----------------
+// twitter: per status id, user.screen_name, retweet_count
+// citm: per performance id, eventId, #seatCategories; #events
+// canada: type, features[0].geometry.type, #coordinates
+// jeopardy: per question: round == "Final Jeopardy!", len(category)
+// status (one tweet): id, user.screen_name, retweet_count
+// rpc: method, params.minuend, id
+
+static double pick(const std::string& name, json_view r)
+{
+    double acc = 0;
+    if (name == "twitter" || name == "status")
+    {
+        auto one = [&](json_view s)
+        {
+            acc += static_cast<double>(s["id"].get<std::uint64_t>());
+            acc += static_cast<double>(s["user"]["screen_name"].get_string().size());
+            acc += static_cast<double>(s["retweet_count"].get<std::int64_t>());
+        };
+        if (name == "twitter")
+        {
+            for (const json_view s : r["statuses"])
+            {
+                one(s);
+            }
+        }
+        else
+        {
+            one(r);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        for (const json_view p : r["performances"])
+        {
+            acc += static_cast<double>(p["id"].get<std::uint64_t>() + p["eventId"].get<std::uint64_t>() + p["seatCategories"].size());
+        }
+        acc += static_cast<double>(r["events"].size());
+    }
+    else if (name == "canada")
+    {
+        const json_view g = r["features"][0]["geometry"];
+        acc += static_cast<double>(r["type"].get_string().size() + g["type"].get_string().size() + g["coordinates"].size());
+    }
+    else if (name == "jeopardy")
+    {
+        for (const json_view q : r)
+        {
+            acc += q["round"].get_string() == "Final Jeopardy!" ? 1 : 0;
+            acc += static_cast<double>(q["category"].get_string().size());
+        }
+    }
+    else if (name == "rpc")
+    {
+        acc += static_cast<double>(r["method"].get_string().size());
+        acc += static_cast<double>(r["params"]["minuend"].get<std::int64_t>() + r["id"].get<std::int64_t>());
+    }
+    return acc;
+}
+
+static double pick(const std::string& name, const json& r)
+{
+    double acc = 0;
+    if (name == "twitter" || name == "status")
+    {
+        auto one = [&](const json & s)
+        {
+            acc += static_cast<double>(s["id"].get<std::uint64_t>());
+            acc += static_cast<double>(s["user"]["screen_name"].get_ref<const std::string&>().size());
+            acc += static_cast<double>(s["retweet_count"].get<std::int64_t>());
+        };
+        if (name == "twitter")
+        {
+            for (const auto& s : r["statuses"])
+            {
+                one(s);
+            }
+        }
+        else
+        {
+            one(r);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        for (const auto& p : r["performances"])
+        {
+            acc += static_cast<double>(p["id"].get<std::uint64_t>() + p["eventId"].get<std::uint64_t>() + p["seatCategories"].size());
+        }
+        acc += static_cast<double>(r["events"].size());
+    }
+    else if (name == "canada")
+    {
+        const json& g = r["features"][0]["geometry"];
+        acc += static_cast<double>(r["type"].get_ref<const std::string&>().size() + g["type"].get_ref<const std::string&>().size() + g["coordinates"].size());
+    }
+    else if (name == "jeopardy")
+    {
+        for (const auto& q : r)
+        {
+            acc += q["round"].get_ref<const std::string&>() == "Final Jeopardy!" ? 1 : 0;
+            acc += static_cast<double>(q["category"].get_ref<const std::string&>().size());
+        }
+    }
+    else if (name == "rpc")
+    {
+        acc += static_cast<double>(r["method"].get_ref<const std::string&>().size());
+        acc += static_cast<double>(r["params"]["minuend"].get<std::int64_t>() + r["id"].get<std::int64_t>());
+    }
+    return acc;
+}
+
+static double pick(const std::string& name, yyjson_val* r)
+{
+    double acc = 0;
+    auto get = [](yyjson_val * o, const char* k)
+    {
+        return yyjson_obj_get(o, k);
+    };
+    if (name == "twitter" || name == "status")
+    {
+        auto one = [&](yyjson_val * s)
+        {
+            acc += static_cast<double>(yyjson_get_uint(get(s, "id")));
+            acc += static_cast<double>(yyjson_get_len(get(get(s, "user"), "screen_name")));
+            acc += static_cast<double>(yyjson_get_sint(get(s, "retweet_count")));
+        };
+        if (name == "twitter")
+        {
+            std::size_t idx, max;
+            yyjson_val* s;
+            yyjson_arr_foreach(get(r, "statuses"), idx, max, s)
+            {
+                one(s);
+            }
+        }
+        else
+        {
+            one(r);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        std::size_t idx, max;
+        yyjson_val* p;
+        yyjson_arr_foreach(get(r, "performances"), idx, max, p)
+        {
+            acc += static_cast<double>(yyjson_get_uint(get(p, "id")) + yyjson_get_uint(get(p, "eventId")) + yyjson_arr_size(get(p, "seatCategories")));
+        }
+        acc += static_cast<double>(yyjson_obj_size(get(r, "events")));
+    }
+    else if (name == "canada")
+    {
+        yyjson_val* g = get(yyjson_arr_get(get(r, "features"), 0), "geometry");
+        acc += static_cast<double>(yyjson_get_len(get(r, "type")) + yyjson_get_len(get(g, "type")) + yyjson_arr_size(get(g, "coordinates")));
+    }
+    else if (name == "jeopardy")
+    {
+        std::size_t idx, max;
+        yyjson_val* q;
+        yyjson_arr_foreach(r, idx, max, q)
+        {
+            acc += yyjson_equals_str(get(q, "round"), "Final Jeopardy!") ? 1 : 0;
+            acc += static_cast<double>(yyjson_get_len(get(q, "category")));
+        }
+    }
+    else if (name == "rpc")
+    {
+        acc += static_cast<double>(yyjson_get_len(get(r, "method")));
+        acc += static_cast<double>(yyjson_get_sint(get(get(r, "params"), "minuend")) + yyjson_get_sint(get(r, "id")));
+    }
+    return acc;
+}
+
+static double pick(const std::string& name, simdjson::dom::element r)
+{
+    double acc = 0;
+    if (name == "twitter" || name == "status")
+    {
+        auto one = [&](simdjson::dom::element s)
+        {
+            acc += static_cast<double>(uint64_t(s["id"]));
+            acc += static_cast<double>(std::string_view(s["user"]["screen_name"]).size());
+            acc += static_cast<double>(int64_t(s["retweet_count"]));
+        };
+        if (name == "twitter")
+        {
+            for (auto s : simdjson::dom::array(r["statuses"]))
+            {
+                one(s);
+            }
+        }
+        else
+        {
+            one(r);
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        for (auto p : simdjson::dom::array(r["performances"]))
+        {
+            acc += static_cast<double>(uint64_t(p["id"]) + uint64_t(p["eventId"]) + simdjson::dom::array(p["seatCategories"]).size());
+        }
+        acc += static_cast<double>(simdjson::dom::object(r["events"]).size());
+    }
+    else if (name == "canada")
+    {
+        auto g = r["features"].at(0)["geometry"];
+        acc += static_cast<double>(std::string_view(r["type"]).size() + std::string_view(g["type"]).size() + simdjson::dom::array(g["coordinates"]).size());
+    }
+    else if (name == "jeopardy")
+    {
+        for (auto q : simdjson::dom::array(r))
+        {
+            acc += std::string_view(q["round"]) == "Final Jeopardy!" ? 1 : 0;
+            acc += static_cast<double>(std::string_view(q["category"]).size());
+        }
+    }
+    else if (name == "rpc")
+    {
+        acc += static_cast<double>(std::string_view(r["method"]).size());
+        acc += static_cast<double>(int64_t(r["params"]["minuend"]) + int64_t(r["id"]));
+    }
+    return acc;
+}
+
+static double pick_od(const std::string& name, simdjson::ondemand::document& d)
+{
+    double acc = 0;
+    if (name == "twitter" || name == "status")
+    {
+        auto one = [&](simdjson::ondemand::object s)
+        {
+            acc += static_cast<double>(uint64_t(s["id"]));
+            acc += static_cast<double>(std::string_view(s["user"]["screen_name"]).size());
+            acc += static_cast<double>(int64_t(s["retweet_count"]));
+        };
+        if (name == "twitter")
+        {
+            for (auto s : d["statuses"])
+            {
+                one(s.get_object());
+            }
+        }
+        else
+        {
+            one(d.get_object());
+        }
+    }
+    else if (name == "citm_catalog")
+    {
+        simdjson::ondemand::object ev = d["events"].get_object();
+        acc += static_cast<double>(ev.count_fields());
+        for (auto p : d["performances"])
+        {
+            simdjson::ondemand::object o = p.get_object();
+            const auto a = uint64_t(o["eventId"]) + uint64_t(o["id"]);
+            simdjson::ondemand::array sc = o["seatCategories"].get_array();
+            acc += static_cast<double>(a + sc.count_elements());
+        }
+    }
+    else if (name == "canada")
+    {
+        acc += static_cast<double>(std::string_view(d["type"]).size());
+        auto g = d["features"].at(0)["geometry"];
+        acc += static_cast<double>(std::string_view(g["type"]).size());
+        simdjson::ondemand::array co = g["coordinates"].get_array();
+        acc += static_cast<double>(co.count_elements());
+    }
+    else if (name == "jeopardy")
+    {
+        for (auto q : d)
+        {
+            simdjson::ondemand::object o = q.get_object();
+            acc += static_cast<double>(std::string_view(o["category"]).size());
+            acc += std::string_view(o["round"]) == "Final Jeopardy!" ? 1 : 0;
+        }
+    }
+    else if (name == "rpc")
+    {
+        acc += static_cast<double>(std::string_view(d["method"]).size());
+        acc += static_cast<double>(int64_t(d["params"]["minuend"]));
+        acc += static_cast<double>(int64_t(d["id"]));
+    }
+    return acc;
+}
+
+// ---------------- harness ----------------
+
+static std::string slurp(const std::string& p)
+{
+    std::ifstream f(p, std::ios::binary);
+    if (!f)
+    {
+        std::fprintf(stderr, "cannot open %s\n", p.c_str());
+        std::exit(1);
+    }
+    std::stringstream ss;
+    ss << f.rdbuf();
+    return ss.str();
+}
+
+struct engine
+{
+    std::string name;
+    std::function<void()> fn;
+};
+
+int main(int argc, char** argv)
+{
+    if (argc < 2)
+    {
+        std::fprintf(stderr, "usage: %s <json_test_data directory> [rounds] [document]\n", argv[0]);
+        return 1;
+    }
+    const std::string T = std::string(argv[1]) + "/";
+    // (clamped to 1 to a million)
+    const int rounds = argc > 2 ? static_cast<int>((std::min)((std::max)(std::strtol(argv[2], nullptr, 10), 1L), 1000000L)) : 30;
+    const std::string only = argc > 3 ? argv[3] : "";
+    struct doc
+    {
+        std::string name, text;
+        int batch;
+    };
+    std::vector<doc> docs;
+    for (const char* f :
+            {"nativejson-benchmark/twitter.json", "nativejson-benchmark/citm_catalog.json", "nativejson-benchmark/canada.json", "jeopardy/jeopardy.json"
+            })
+    {
+        std::string n = std::string(f).substr(std::string(f).find('/') + 1);
+        docs.push_back({n.substr(0, n.size() - 5), slurp(T + f), 1});
+    }
+    docs.push_back({"status", json::parse(docs[0].text)["statuses"][0].dump(), 200});
+    docs.push_back({"rpc", R"({"jsonrpc": "2.0", "method": "subtract", "params": {"minuend": 42, "subtrahend": 23}, "id": 3})", 5000});
+
+    // correctness cross-check of the workloads
+    for (const auto& dc : docs)
+    {
+        stats a, b, c, dd;
+        walk(json::parse(dc.text), a);
+        auto d = json_document::parse(dc.text);
+        walk(d.root(), b);
+        yyjson_doc* y = yyjson_read(dc.text.data(), dc.text.size(), 0);
+        walk(yyjson_doc_get_root(y), c);
+        simdjson::dom::parser p;
+        walk(p.parse(dc.text).value(), dd);
+        const bool ok = a.nodes == b.nodes && a.nodes == c.nodes && a.nodes == dd.nodes && a.str == b.str && a.str == c.str && a.str == dd.str
+                        && std::fabs(a.num - b.num) <= 1e-9 * std::fabs(a.num) && std::fabs(a.num - c.num) <= 1e-9 * std::fabs(a.num);
+        const double pa = pick(dc.name, json::parse(dc.text)), pb = pick(dc.name, d.root()), pc = pick(dc.name, yyjson_doc_get_root(y)), pd = pick(dc.name, p.parse(dc.text).value());
+        std::printf("check %-13s traverse %s select %s\n", dc.name.c_str(), ok ? "OK" : "MISMATCH", (pa == pb && pa == pc && pa == pd) ? "OK" : "MISMATCH");
+        yyjson_doc_free(y);
+    }
+
+    std::FILE* csv = std::fopen("bench_view.csv", "w");
+    if (csv == nullptr)
+    {
+        std::perror("bench_view.csv");
+        return 1;
+    }
+    std::fprintf(csv, "doc,bytes,workload,engine,ns\n");
+    json_document reused;
+    simdjson::dom::parser sj;
+    simdjson::ondemand::parser od;
+    for (const auto& dc : docs)
+    {
+        if (!only.empty() && dc.name != only)
+        {
+            continue;
+        }
+        const std::string& s = dc.text;
+        const simdjson::padded_string ps(s);
+        const std::string name = dc.name;
+        std::vector<std::pair<std::string, std::vector<engine>>> workloads;
+
+        workloads.push_back({"parse", {
+                {"json_view", [&] { auto d = json_document::parse(s); g_sink = static_cast<double>(d.node_count()); }},
+                {"json_view (reused)", [&] { reused.read(s); g_sink = static_cast<double>(reused.node_count()); }},
+                {"yyjson", [&] { yyjson_doc* d = yyjson_read(s.data(), s.size(), 0); g_sink = static_cast<double>(yyjson_doc_get_val_count(d)); yyjson_doc_free(d); }},
+                {"simdjson DOM", [&] { auto e = sj.parse(ps).value_unsafe(); g_sink = e.is_object(); }},
+                {"simdjson DOM (fresh)", [&] { simdjson::dom::parser p; auto e = p.parse(ps).value_unsafe(); g_sink = e.is_object(); }},
+#if JSON_VIEW_BENCH_BOOST
+                {"Boost.JSON", [&] { boost::json::monotonic_resource mr; auto v = boost::json::parse(s, &mr); g_sink = v.is_object(); }},
+#endif
+                {"json::parse", [&] { json j = json::parse(s); g_sink = static_cast<double>(j.size()); }},
+            }});
+        workloads.push_back({"traverse", {
+                {"json_view", [&] { auto d = json_document::parse(s); stats st; walk(d.root(), st); g_sink = st.num; }},
+                {"json_view (reused)", [&] { reused.read(s); stats st; walk(reused.root(), st); g_sink = st.num; }},
+                {"yyjson", [&] { yyjson_doc* d = yyjson_read(s.data(), s.size(), 0); stats st; walk(yyjson_doc_get_root(d), st); g_sink = st.num; yyjson_doc_free(d); }},
+                {"simdjson DOM", [&] { stats st; walk(sj.parse(ps).value_unsafe(), st); g_sink = st.num; }},
+                {"simdjson OD", [&] { auto d = od.iterate(ps).value_unsafe(); stats st; walk_od(d.get_value().value_unsafe(), st); g_sink = st.num; }},
+#if JSON_VIEW_BENCH_BOOST
+                {"Boost.JSON", [&] { boost::json::monotonic_resource mr; auto v = boost::json::parse(s, &mr); stats st; walk(v, st); g_sink = st.num; }},
+#endif
+                {"json::parse", [&] { json j = json::parse(s); stats st; walk(j, st); g_sink = st.num; }},
+            }});
+        workloads.push_back({"select", {
+                {"json_view", [&] { auto d = json_document::parse(s); g_sink = pick(name, d.root()); }},
+                {"json_view (reused)", [&] { reused.read(s); g_sink = pick(name, reused.root()); }},
+                {"yyjson", [&] { yyjson_doc* d = yyjson_read(s.data(), s.size(), 0); g_sink = pick(name, yyjson_doc_get_root(d)); yyjson_doc_free(d); }},
+                {"simdjson DOM", [&] { g_sink = pick(name, sj.parse(ps).value_unsafe()); }},
+                {"simdjson OD", [&] { auto d = od.iterate(ps).value_unsafe(); g_sink = pick_od(name, d); }},
+                {"json::parse", [&] { json j = json::parse(s); g_sink = pick(name, j); }},
+            }});
+        {
+            // serialization of an already parsed document
+            static json_document vd;
+            vd.read(s);
+            static yyjson_doc* yd = nullptr;
+            if (yd)
+            {
+                yyjson_doc_free(yd);
+            }
+            yd = yyjson_read(s.data(), s.size(), 0);
+            static simdjson::dom::parser sjd;
+            static simdjson::dom::element se;
+            se = sjd.parse(ps).value_unsafe();
+            static json jd;
+            jd = json::parse(s);
+            workloads.push_back({"dump", {
+                    {"json_view", [&] { std::string o = vd.root().dump(); g_sink = static_cast<double>(o.size()); }},
+                    {"yyjson", [&] { std::size_t n = 0; char* o = yyjson_write(yd, 0, &n); g_sink = static_cast<double>(n); std::free(o); }},
+                    {"simdjson DOM", [&] { std::string o = simdjson::to_string(se); g_sink = static_cast<double>(o.size()); }},
+                    {"json::parse", [&] { std::string o = jd.dump(); g_sink = static_cast<double>(o.size()); }},
+                }});
+        }
+
+        for (auto& wl : workloads)
+        {
+            std::vector<double> best(wl.second.size(), 1e300);
+            const int r = s.size() > 10000000 ? std::max(3, rounds / 5) : rounds;
+            for (int i = 0; i < r; ++i)
+            {
+                for (std::size_t k = 0; k < wl.second.size(); ++k)
+                {
+                    // an untimed call first: whatever the previous engine left to the allocator
+                    // (e.g. thousands of freed json nodes) is cleaned up here, not in the timing
+                    wl.second[k].fn();
+                    const auto t0 = std::chrono::steady_clock::now();
+                    for (int b = 0; b < dc.batch; ++b)
+                    {
+                        wl.second[k].fn();
+                    }
+                    const double ns = std::chrono::duration<double, std::nano>(std::chrono::steady_clock::now() - t0).count() / dc.batch;
+                    best[k] = std::min(best[k], ns);
+                }
+            }
+            const double ref = best[0];
+            std::printf("%-13s %-9s", dc.name.c_str(), wl.first.c_str());
+            for (std::size_t k = 0; k < wl.second.size(); ++k)
+            {
+                const double us = best[k] / 1e3;
+                std::printf("  %s %s%s (%.2fx)", wl.second[k].name.c_str(), us >= 100 ? "" : "", (us >= 1000 ? std::to_string(static_cast<long>(us)) + "us" : (std::to_string(us).substr(0, 5) + "us")).c_str(), best[k] / ref);
+                std::fprintf(csv, "%s,%zu,%s,%s,%.1f\n", dc.name.c_str(), s.size(), wl.first.c_str(), wl.second[k].name.c_str(), best[k]);
+            }
+            std::printf("\n");
+            std::fflush(stdout);
+        }
+    }
+    std::fclose(csv);
+}

tests/benchmarks/json_view/compare.py

@@ -0,0 +1,342 @@
+#!/usr/bin/env python3
+#     __ _____ _____ _____
+#  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+# |  |  |__   |  |  | | | |  version 3.12.0
+# |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+#
+# SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+# SPDX-License-Identifier: MIT
+
+"""Compare json_view with yyjson, simdjson, Boost.JSON, and json::parse.
+
+Builds bench_view.cpp, bench_corpus.cpp, and bench_edit.cpp against the include/ directory of
+this checkout, runs them, and writes the results with everything needed to
+reproduce them (date, commit, CPU, OS, compiler, library versions, flags) to
+results/<date>-<host>.md and .csv next to this script.
+
+The other libraries come from the system (--system, the default: pkg-config
+or Homebrew) or are downloaded as pinned releases and checked against their
+SHA-256 (--download). Boost.JSON is optional: without Boost headers, its
+columns are skipped, and the results say so.
+
+Only the Python 3 standard library is used; a C++17 compiler is needed.
+"""
+
+import argparse
+import datetime
+import hashlib
+import os
+import platform
+import re
+import shlex
+import shutil
+# runs only the compilers and benchmark binaries this script builds
+import subprocess  # nosec B404
+import sys
+import tarfile
+import tempfile
+import urllib.request
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+REPO = os.path.abspath(os.path.join(HERE, '..', '..', '..'))
+
+# seconds without data after which a download fails
+DOWNLOAD_TIMEOUT = 60
+
+# pinned releases for --download; the hashes are those of the archives
+#
+# The yyjson and simdjson archives are the tag archives GitHub generates
+# (archive/refs/tags/...), whose bytes GitHub does not guarantee to be stable. A SHA-256
+# mismatch for one of them means that GitHub regenerated the archive: check the upstream
+# tag, then pin the new hash.
+PINNED = {
+    'yyjson': {
+        'version': '0.13.0',
+        'url': 'https://github.com/ibireme/yyjson/archive/refs/tags/0.13.0.tar.gz',
+        'sha256': '34e0f62a2bc11ab20d601e8ca1cc2b2079503aa45119a19133d89d19b94a0fae',
+        'dir': 'yyjson-0.13.0',
+    },
+    'simdjson': {
+        'version': '4.6.11',
+        'url': 'https://github.com/simdjson/simdjson/archive/refs/tags/v4.6.11.tar.gz',
+        'sha256': '61d948fc24f0d793829ad658058e7597d064988a89b4607ea02e401a82df98ff',
+        'dir': 'simdjson-4.6.11',
+    },
+    'boost': {
+        'version': '1.92.0',
+        'url': 'https://archives.boost.io/release/1.92.0/source/boost_1_92_0.tar.gz',
+        'sha256': 'c4a3b310ddd2472416e091067166b0713be97c63f38c212c484ada022fd296ce',
+        'dir': 'boost_1_92_0',
+    },
+}
+
+# the documents of bench_view.cpp, relative to the json_test_data directory
+DEFAULT_CORPUS = [
+    'nativejson-benchmark/twitter.json',
+    'nativejson-benchmark/citm_catalog.json',
+    'nativejson-benchmark/canada.json',
+    'jeopardy/jeopardy.json',
+]
+
+
+def run(cmd, **kwargs):
+    print('+ ' + ' '.join(shlex.quote(c) for c in cmd), flush=True)
+    # cmd is an argument list built by this script, never a shell string
+    return subprocess.run(cmd, check=True, **kwargs)  # nosec B603
+
+
+def output(cmd):
+    try:
+        # cmd is an argument list built by this script, never a shell string
+        return subprocess.run(cmd, check=True, capture_output=True, text=True).stdout.strip()  # nosec B603
+    except (OSError, subprocess.CalledProcessError):
+        return ''
+
+
+# ---------------------------------------------------------------------------
+# libraries
+# ---------------------------------------------------------------------------
+
+class Library:
+    """include directories, sources to compile, and linker flags of a library"""
+
+    def __init__(self, name, include=None, sources=None, link=None, version=''):
+        self.name = name
+        self.include = include or []
+        self.sources = sources or []
+        self.link = link or []
+        self.version = version
+
+
+def header_version(path, pattern):
+    try:
+        with open(path, encoding='utf-8', errors='replace') as f:
+            m = re.search(pattern, f.read())
+        return m.group(1) if m else ''
+    except OSError:
+        return ''
+
+
+def library_version(name, include_dirs):
+    patterns = {
+        'yyjson': ('yyjson.h', r'#define\s+YYJSON_VERSION_STRING\s+"([^"]+)"'),
+        'simdjson': ('simdjson.h', r'#define\s+SIMDJSON_VERSION\s+"?([0-9.]+)"?'),
+        'boost': (os.path.join('boost', 'version.hpp'), r'#define\s+BOOST_LIB_VERSION\s+"([^"]+)"'),
+    }
+    header, pattern = patterns[name]
+    for d in include_dirs:
+        v = header_version(os.path.join(d, header), pattern)
+        if v:
+            return v.replace('_', '.')
+    return ''
+
+
+def system_library(name):
+    """a library found with pkg-config or Homebrew, or None"""
+    flags = output(['pkg-config', '--cflags', '--libs', name]).split()
+    if flags:
+        include = [f[2:] for f in flags if f.startswith('-I')]
+        link = [f for f in flags if f.startswith('-L') or f.startswith('-l')]
+        libdirs = [f[2:] for f in link if f.startswith('-L')]
+        link += ['-Wl,-rpath,' + d for d in libdirs]
+        return Library(name, include, [], link, library_version(name, include))
+    prefix = output(['brew', '--prefix', name]) if shutil.which('brew') else ''
+    if prefix and os.path.isdir(os.path.join(prefix, 'include')):
+        include = [os.path.join(prefix, 'include')]
+        link = []
+        if name != 'boost':
+            lib = os.path.join(prefix, 'lib')
+            link = ['-L' + lib, '-l' + name, '-Wl,-rpath,' + lib]
+        return Library(name, include, [], link, library_version(name, include))
+    if name == 'boost':
+        for d in ['/usr/include', '/usr/local/include']:
+            if os.path.isfile(os.path.join(d, 'boost', 'json.hpp')):
+                return Library(name, [d], [], [], library_version(name, [d]))
+    return None
+
+
+def sha256_file(path):
+    """SHA-256 of a file, read in chunks (the Boost archive is more than 100 MB)"""
+    h = hashlib.sha256()
+    with open(path, 'rb') as f:
+        for chunk in iter(lambda: f.read(1 << 20), b''):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def download_library(name, work):
+    """a pinned release, downloaded and checked, or an error"""
+    pin = PINNED[name]
+    archive = os.path.join(work, 'download', os.path.basename(pin['url']))
+    os.makedirs(os.path.dirname(archive), exist_ok=True)
+    if not os.path.isfile(archive):
+        print(f'downloading {pin["url"]}', flush=True)
+        # the URLs are the https constants in PINNED, and the SHA-256 is checked below
+        # (into a .part file first, so that an interrupted download is not kept)
+        with urllib.request.urlopen(pin['url'], timeout=DOWNLOAD_TIMEOUT) as response:  # nosec B310
+            with open(archive + '.part', 'wb') as f:
+                shutil.copyfileobj(response, f)
+        os.replace(archive + '.part', archive)
+    digest = sha256_file(archive)
+    if digest != pin['sha256']:
+        os.remove(archive)  # downloaded again by the next run
+        sys.exit(f'error: SHA-256 of {archive} is {digest}, expected {pin["sha256"]} (removed)')
+    src = os.path.join(work, 'download', pin['dir'])
+    if not os.path.isdir(src):
+        # extracted into a temporary directory first, so that an interrupted extraction
+        # is not mistaken for a complete one
+        tmp = tempfile.mkdtemp(prefix=pin['dir'] + '.part-', dir=os.path.join(work, 'download'))
+        try:
+            with tarfile.open(archive) as t:
+                # (the 'data' filter rejects links and paths outside the target where Python has it)
+                kwargs = {'filter': 'data'} if hasattr(tarfile, 'data_filter') else {}
+                t.extractall(tmp, **kwargs)  # noqa: S202 (checked archive)  # nosec B202
+            os.rename(os.path.join(tmp, pin['dir']), src)
+        finally:
+            shutil.rmtree(tmp, ignore_errors=True)
+    if name == 'yyjson':
+        return Library(name, [os.path.join(src, 'src')], [os.path.join(src, 'src', 'yyjson.c')], [], pin['version'])
+    if name == 'simdjson':
+        single = os.path.join(src, 'singleheader')
+        return Library(name, [single], [os.path.join(single, 'simdjson.cpp')], [], pin['version'])
+    return Library(name, [src], [], [], pin['version'])
+
+
+# ---------------------------------------------------------------------------
+# machine description
+# ---------------------------------------------------------------------------
+
+def cpu_model():
+    if sys.platform == 'darwin':
+        return output(['sysctl', '-n', 'machdep.cpu.brand_string'])
+    try:
+        with open('/proc/cpuinfo', encoding='utf-8') as f:
+            for line in f:
+                if line.startswith('model name') or line.startswith('Model'):
+                    return line.split(':', 1)[1].strip()
+    except OSError:
+        pass
+    # (AArch64 Linux: /proc/cpuinfo has no model name, lscpu knows it)
+    for line in output(['lscpu']).splitlines():
+        if line.startswith('Model name:'):
+            return line.split(':', 1)[1].strip()
+    return platform.processor() or platform.machine()
+
+
+def git_commit():
+    commit = output(['git', '-C', REPO, 'rev-parse', '--short=12', 'HEAD'])
+    dirty = output(['git', '-C', REPO, 'status', '--porcelain', '--untracked-files=no'])
+    return commit + (' (with local changes)' if dirty else '')
+
+
+# ---------------------------------------------------------------------------
+# main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument('--data', required=True, help='json_test_data directory (with nativejson-benchmark/ and jeopardy/)')
+    ap.add_argument('--download', action='store_true', help='use pinned downloads instead of system libraries')
+    ap.add_argument('--no-boost', action='store_true', help='skip Boost.JSON')
+    ap.add_argument('--native', action='store_true', help='compile for this CPU (-march=native / -mcpu=native)')
+    ap.add_argument('--rounds', type=int, default=30, help='rounds of bench_view (default: 30)')
+    ap.add_argument('--corpus', nargs='*', default=[], help='more files for bench_corpus')
+    ap.add_argument('--build-dir', default=os.path.join(HERE, 'build'), help='where to build (default: build/ next to this script)')
+    args = ap.parse_args()
+    # the benchmarks run in the build directory: make the paths absolute
+    args.data = os.path.abspath(args.data)
+    args.corpus = [os.path.abspath(f) for f in args.corpus]
+    args.build_dir = os.path.abspath(args.build_dir)
+
+    # (CXX and CC may contain arguments, e.g. 'ccache g++')
+    cxx = shlex.split(os.environ.get('CXX', 'c++'))
+    cc = shlex.split(os.environ.get('CC', 'cc'))
+    os.makedirs(args.build_dir, exist_ok=True)
+
+    libs = {}
+    for name in ['yyjson', 'simdjson', 'boost']:
+        if name == 'boost' and args.no_boost:
+            continue
+        lib = download_library(name, args.build_dir) if args.download else system_library(name)
+        if lib is None and name != 'boost':
+            sys.exit(f'error: {name} not found; install it, or use --download')
+        if lib is not None:
+            libs[name] = lib
+    with_boost = 'boost' in libs
+    if not with_boost:
+        print('Boost.JSON not found: its columns are skipped', flush=True)
+
+    flags = ['-std=c++17', '-O3', '-DNDEBUG', f'-DJSON_VIEW_BENCH_BOOST={1 if with_boost else 0}']
+    if args.native:
+        flags.append('-mcpu=native' if platform.machine().lower() in ('arm64', 'aarch64') else '-march=native')
+    include = ['-I' + os.path.join(REPO, 'include')] + ['-I' + d for lib in libs.values() for d in lib.include]
+    link = [f for lib in libs.values() for f in lib.link]
+
+    # C sources of downloaded libraries are compiled once
+    objects = []
+    for lib in libs.values():
+        for src in lib.sources:
+            obj = os.path.join(args.build_dir, os.path.basename(src) + '.o')
+            is_c = src.endswith('.c')
+            run((cc if is_c else cxx) + ([] if is_c else ['-std=c++17']) + ['-O3', '-DNDEBUG', '-c', src, '-o', obj]
+                + ['-I' + d for d in lib.include])
+            objects.append(obj)
+
+    binaries = {}
+    for bench in ['bench_view', 'bench_corpus', 'bench_edit']:
+        exe = os.path.join(args.build_dir, bench)
+        run(cxx + flags + include + [os.path.join(HERE, bench + '.cpp')] + objects + link + ['-o', exe])
+        binaries[bench] = exe
+
+    # run: bench_view on its documents, bench_corpus on those and the given files
+    corpus = [os.path.join(args.data, f) for f in DEFAULT_CORPUS] + args.corpus
+    outputs = {}
+    outputs['bench_view'] = run([binaries['bench_view'], args.data, str(args.rounds)], cwd=args.build_dir,
+                                capture_output=True, text=True).stdout
+    outputs['bench_corpus'] = run([binaries['bench_corpus']] + corpus, cwd=args.build_dir,
+                                  capture_output=True, text=True).stdout
+    outputs['bench_edit'] = run([binaries['bench_edit'], args.data, str(max(1, args.rounds // 2))], cwd=args.build_dir,
+                                capture_output=True, text=True).stdout
+    for name, text in outputs.items():
+        print(text)
+
+    # results with their metadata
+    now = datetime.datetime.now()
+    host = re.sub(r'[^A-Za-z0-9-]+', '-', platform.node().split('.')[0]) or 'host'
+    stem = os.path.join(HERE, 'results', f'{now:%Y-%m-%d}-{host}')
+    os.makedirs(os.path.dirname(stem), exist_ok=True)
+    cxx_version = output(cxx + ['--version'])
+    compiler_version = cxx_version.splitlines()[0] if cxx_version else ' '.join(cxx)
+    meta = [
+        ('date', f'{now:%Y-%m-%d %H:%M}'),
+        ('commit', git_commit()),
+        ('CPU', cpu_model()),
+        ('OS', f'{platform.system()} {platform.release()} ({platform.machine()})'),
+        ('compiler', compiler_version),
+        ('flags', ' '.join(flags)),
+        ('yyjson', libs['yyjson'].version),
+        ('simdjson', libs['simdjson'].version),
+        ('Boost.JSON', libs['boost'].version if with_boost else 'skipped (not found)'),
+        ('libraries from', 'pinned downloads' if args.download else 'the system'),
+        ('rounds', str(args.rounds)),
+    ]
+    with open(stem + '.md', 'w', encoding='utf-8') as f:
+        f.write(f'# json_view comparison, {now:%Y-%m-%d}\n\n')
+        f.write('Generated by `tests/benchmarks/json_view/compare.py`; best of the interleaved rounds.\n\n')
+        f.write('| | |\n|---|---|\n')
+        for key, value in meta:
+            f.write(f'| {key} | {value} |\n')
+        for name, text in outputs.items():
+            f.write(f'\n## {name}\n\n```\n{text.rstrip()}\n```\n')
+    with open(stem + '.csv', 'w', encoding='utf-8') as out:
+        out.write(''.join(f'# {key}: {value}\n' for key, value in meta))
+        for name in ['bench_view', 'bench_corpus', 'bench_edit']:
+            path = os.path.join(args.build_dir, name + '.csv')
+            if os.path.isfile(path):
+                with open(path, encoding='utf-8') as f:
+                    out.write(f'# {name}\n' + f.read())
+    print(f'results: {stem}.md, {stem}.csv')
+
+
+if __name__ == '__main__':
+    main()

tests/benchmarks/src/benchmarks.cpp

@@ -9,12 +9,15 @@
 #include <benchmark/benchmark.h>
 #include <nlohmann/json.hpp>
 #include <array>
+#include <cstdint>
 #include <cstdio>
 #include <deque>
 #include <fstream>
+#include <map>
 #include <numeric>
 #include <set>
 #include <sstream>
+#include <string>
 #include <vector>
 #include <test_data.hpp>
 
@@ -710,6 +713,584 @@ static void FromCborChunkedString(benchmark::State& state, const std::size_t chu
 
 BENCHMARK_CAPTURE(FromCborChunkedString, 10000 chunks, 10000);
 
+//////////////////////////////////////////////////////////////////////////////
+// SAX-parse JSON from string
+//
+// sax_parse() with a handler that does next to nothing runs the lexer and
+// parser without building values, so the gap to ParseString above is the cost
+// of building the DOM.
+//////////////////////////////////////////////////////////////////////////////
+
+static std::string read_file(const char* filename)
+{
+    std::ifstream f(filename);
+    return std::string((std::istreambuf_iterator<char>(f)), std::istreambuf_iterator<char>());
+}
+
+static json load_json(const char* filename)
+{
+    return json::parse(read_file(filename));
+}
+
+/// SAX handler that only counts the events, so that the handler costs next to nothing
+class counting_sax : public nlohmann::json_sax<json>
+{
+  public:
+    std::size_t events = 0;
+
+    bool null() override
+    {
+        return count();
+    }
+    bool boolean(bool /*val*/) override
+    {
+        return count();
+    }
+    bool number_integer(number_integer_t /*val*/) override
+    {
+        return count();
+    }
+    bool number_unsigned(number_unsigned_t /*val*/) override
+    {
+        return count();
+    }
+    bool number_float(number_float_t /*val*/, const string_t& /*s*/) override
+    {
+        return count();
+    }
+    bool string(string_t& /*val*/) override
+    {
+        return count();
+    }
+    bool binary(binary_t& /*val*/) override
+    {
+        return count();
+    }
+    bool start_object(std::size_t /*elements*/) override
+    {
+        return count();
+    }
+    bool key(string_t& /*val*/) override
+    {
+        return count();
+    }
+    bool end_object() override
+    {
+        return count();
+    }
+    bool start_array(std::size_t /*elements*/) override
+    {
+        return count();
+    }
+    bool end_array() override
+    {
+        return count();
+    }
+    bool parse_error(std::size_t /*position*/, const std::string& /*last_token*/, const json::exception& /*ex*/) override
+    {
+        return false;
+    }
+
+  private:
+    bool count()
+    {
+        ++events;
+        return true;
+    }
+};
+
+static void SaxParse(benchmark::State& state, const char* filename)
+{
+    const std::string str = read_file(filename);
+
+    for (auto _ : state)
+    {
+        counting_sax sax;
+        bool result = json::sax_parse(str, &sax);
+        benchmark::DoNotOptimize(result);
+        benchmark::DoNotOptimize(sax.events);
+    }
+
+    state.SetBytesProcessed(state.iterations() * str.size());
+}
+BENCHMARK_CAPTURE(SaxParse, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(SaxParse, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(SaxParse, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+//////////////////////////////////////////////////////////////////////////////
+// work with values: copy, destroy, compare, iterate
+//
+// The parsed test files stand in for a document an application keeps around.
+// Each benchmark times one whole-document operation; setup and cleanup run
+// outside the timed section.
+//
+// A copy is always destroyed again, and how a document is freed shapes the
+// heap the next copy allocates from: a destructor that frees quickly can leave
+// work to the allocator's next calls. Copy, CopyAssign and CopyRing therefore
+// time the copy together with the destruction of a copy; CopyOnly and
+// DestroyOnly time one half each and are meant for telling the halves apart,
+// not for comparing versions.
+//////////////////////////////////////////////////////////////////////////////
+
+/// copy a document and destroy the copy again, as a temporary copy is used
+static void Copy(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        json copy = j;
+        benchmark::DoNotOptimize(copy);
+    }
+}
+BENCHMARK_CAPTURE(Copy, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Copy, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Copy, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+/// copy-assign a document over an earlier copy of it, which is freed in the process
+static void CopyAssign(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+    json target = j;
+
+    for (auto _ : state)
+    {
+        target = j;
+        benchmark::DoNotOptimize(target);
+    }
+}
+BENCHMARK_CAPTURE(CopyAssign, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(CopyAssign, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(CopyAssign, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+/// keep eight copies alive and replace the oldest one, as a cache or an undo
+/// history does, so that the copy is not made from memory just freed by an
+/// identical document
+static void CopyRing(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+    std::vector<json> ring(8, j);
+    std::size_t oldest = 0;
+
+    for (auto _ : state)
+    {
+        ring[oldest] = j;
+        benchmark::DoNotOptimize(ring[oldest]);
+        oldest = (oldest + 1) % ring.size();
+    }
+}
+BENCHMARK_CAPTURE(CopyRing, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(CopyRing, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(CopyRing, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+/// only the copy; the copy is destroyed with the timer paused (diagnostic)
+static void CopyOnly(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        auto* copy = new json(j);
+
+        state.PauseTiming();
+        delete copy;
+        state.ResumeTiming();
+    }
+}
+BENCHMARK_CAPTURE(CopyOnly, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(CopyOnly, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(CopyOnly, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+/// only the destruction; the copy is made with the timer paused (diagnostic)
+static void DestroyOnly(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        state.PauseTiming();
+        auto* copy = new json(j);
+        state.ResumeTiming();
+
+        delete copy;
+    }
+}
+BENCHMARK_CAPTURE(DestroyOnly, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(DestroyOnly, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(DestroyOnly, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void Equal(benchmark::State& state, const char* filename)
+{
+    // two separate values, so that the comparison cannot stop at identical addresses
+    const json j = load_json(filename);
+    const json copy = j;
+
+    for (auto _ : state)
+    {
+        bool result = (j == copy);
+        benchmark::DoNotOptimize(result);
+    }
+}
+BENCHMARK_CAPTURE(Equal, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Equal, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Equal, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+/// visit every value with range-based for loops, as client code walks a document
+static std::size_t count_values(const json& j)
+{
+    std::size_t result = 1;
+    if (j.is_structured())
+    {
+        for (const auto& element : j)
+        {
+            result += count_values(element);
+        }
+    }
+    return result;
+}
+
+static void Iterate(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        std::size_t result = count_values(j);
+        benchmark::DoNotOptimize(result);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(count_values(j)));
+}
+BENCHMARK_CAPTURE(Iterate, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Iterate, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Iterate, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+//////////////////////////////////////////////////////////////////////////////
+// flatten and unflatten
+//////////////////////////////////////////////////////////////////////////////
+
+static void Flatten(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        json flat = j.flatten();
+        benchmark::DoNotOptimize(flat);
+    }
+}
+BENCHMARK_CAPTURE(Flatten, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Flatten, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Flatten, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+// ordered_json objects look keys up linearly, which flatten() used to do for
+// every value; canada (330,000 values in one flat object) is left out, as it
+// would take minutes with such versions
+static void FlattenOrdered(benchmark::State& state, const char* filename)
+{
+    const nlohmann::ordered_json j = nlohmann::ordered_json::parse(read_file(filename));
+
+    for (auto _ : state)
+    {
+        nlohmann::ordered_json flat = j.flatten();
+        benchmark::DoNotOptimize(flat);
+    }
+}
+BENCHMARK_CAPTURE(FlattenOrdered, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(FlattenOrdered, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void Unflatten(benchmark::State& state, const char* filename)
+{
+    const json flat = load_json(filename).flatten();
+
+    for (auto _ : state)
+    {
+        json j = flat.unflatten();
+        benchmark::DoNotOptimize(j);
+    }
+}
+BENCHMARK_CAPTURE(Unflatten, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Unflatten, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Unflatten, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+//////////////////////////////////////////////////////////////////////////////
+// diff and patch
+//
+// The target is the document with every eighth scalar changed, so the patch
+// has one "replace" operation per changed value and the structure stays the
+// same.
+//////////////////////////////////////////////////////////////////////////////
+
+static void change_every_eighth_scalar(json& j, std::size_t& counter)
+{
+    if (j.is_structured())
+    {
+        for (auto& element : j)
+        {
+            change_every_eighth_scalar(element, counter);
+        }
+        return;
+    }
+
+    if (++counter % 8 != 0)
+    {
+        return;
+    }
+
+    switch (j.type())
+    {
+        case json::value_t::number_integer:
+            j = j.get<json::number_integer_t>() + 1;
+            break;
+        case json::value_t::number_unsigned:
+            j = j.get<json::number_unsigned_t>() + 1;
+            break;
+        case json::value_t::number_float:
+            j = j.get<json::number_float_t>() + 0.5;
+            break;
+        case json::value_t::string:
+            j = j.get<std::string>() + "!";
+            break;
+        case json::value_t::boolean:
+            j = !j.get<bool>();
+            break;
+        case json::value_t::null:
+        case json::value_t::object:
+        case json::value_t::array:
+        case json::value_t::binary:
+        case json::value_t::discarded:
+        default:
+            j = 0;
+            break;
+    }
+}
+
+static json changed_copy(const json& j)
+{
+    json result = j;
+    std::size_t counter = 0;
+    change_every_eighth_scalar(result, counter);
+    return result;
+}
+
+static void Diff(benchmark::State& state, const char* filename)
+{
+    const json source = load_json(filename);
+    const json target = changed_copy(source);
+
+    for (auto _ : state)
+    {
+        json patch = json::diff(source, target);
+        benchmark::DoNotOptimize(patch);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(json::diff(source, target).size()));
+}
+BENCHMARK_CAPTURE(Diff, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Diff, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Diff, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void Patch(benchmark::State& state, const char* filename)
+{
+    const json source = load_json(filename);
+    const json patch = json::diff(source, changed_copy(source));
+
+    for (auto _ : state)
+    {
+        json result = source.patch(patch);
+        benchmark::DoNotOptimize(result);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(patch.size()));
+}
+BENCHMARK_CAPTURE(Patch, canada,       TEST_DATA_DIRECTORY "/nativejson-benchmark/canada.json");
+BENCHMARK_CAPTURE(Patch, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(Patch, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+//////////////////////////////////////////////////////////////////////////////
+// JSON Pointer
+//
+// The pointers are those of every scalar in the document, i.e., the keys of
+// its flattened form.
+//////////////////////////////////////////////////////////////////////////////
+
+static std::vector<std::string> pointer_strings(const json& j)
+{
+    std::vector<std::string> result;
+    const json flat = j.flatten();
+    result.reserve(flat.size());
+    for (auto it = flat.begin(); it != flat.end(); ++it)
+    {
+        result.push_back(it.key());
+    }
+    return result;
+}
+
+static std::vector<json::json_pointer> pointers(const json& j)
+{
+    const std::vector<std::string> strings = pointer_strings(j);
+    std::vector<json::json_pointer> result;
+    result.reserve(strings.size());
+    for (const auto& s : strings)
+    {
+        result.emplace_back(s);
+    }
+    return result;
+}
+
+static void JsonPointerParse(benchmark::State& state, const char* filename)
+{
+    const std::vector<std::string> strings = pointer_strings(load_json(filename));
+
+    for (auto _ : state)
+    {
+        for (const auto& s : strings)
+        {
+            json::json_pointer ptr(s);
+            benchmark::DoNotOptimize(ptr);
+        }
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(strings.size()));
+}
+BENCHMARK_CAPTURE(JsonPointerParse, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(JsonPointerParse, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void JsonPointerToString(benchmark::State& state, const char* filename)
+{
+    const std::vector<json::json_pointer> ptrs = pointers(load_json(filename));
+
+    for (auto _ : state)
+    {
+        for (const auto& ptr : ptrs)
+        {
+            std::string s = ptr.to_string();
+            benchmark::DoNotOptimize(s);
+        }
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(ptrs.size()));
+}
+BENCHMARK_CAPTURE(JsonPointerToString, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(JsonPointerToString, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void JsonPointerAt(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+    const std::vector<json::json_pointer> ptrs = pointers(j);
+
+    for (auto _ : state)
+    {
+        for (const auto& ptr : ptrs)
+        {
+            const json& value = j.at(ptr);
+            benchmark::DoNotOptimize(&value);
+        }
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(ptrs.size()));
+}
+BENCHMARK_CAPTURE(JsonPointerAt, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(JsonPointerAt, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+static void JsonPointerContains(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+    const std::vector<json::json_pointer> ptrs = pointers(j);
+
+    for (auto _ : state)
+    {
+        for (const auto& ptr : ptrs)
+        {
+            bool result = j.contains(ptr);
+            benchmark::DoNotOptimize(result);
+        }
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(ptrs.size()));
+}
+BENCHMARK_CAPTURE(JsonPointerContains, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+BENCHMARK_CAPTURE(JsonPointerContains, twitter,      TEST_DATA_DIRECTORY "/nativejson-benchmark/twitter.json");
+
+//////////////////////////////////////////////////////////////////////////////
+// conversions from and to C++ containers
+//////////////////////////////////////////////////////////////////////////////
+
+static void GetVectorInt(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        auto v = j.get<std::vector<std::int64_t>>();
+        benchmark::DoNotOptimize(v);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(j.size()));
+}
+BENCHMARK_CAPTURE(GetVectorInt, signed_ints, TEST_DATA_DIRECTORY "/regression/signed_ints.json");
+
+static void GetVectorDouble(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename);
+
+    for (auto _ : state)
+    {
+        auto v = j.get<std::vector<double>>();
+        benchmark::DoNotOptimize(v);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(j.size()));
+}
+BENCHMARK_CAPTURE(GetVectorDouble, floats, TEST_DATA_DIRECTORY "/regression/floats.json");
+
+// a large flat object: the flattened document, one member per scalar
+static void GetMap(benchmark::State& state, const char* filename)
+{
+    const json j = load_json(filename).flatten();
+
+    for (auto _ : state)
+    {
+        auto m = j.get<std::map<std::string, json>>();
+        benchmark::DoNotOptimize(m);
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(j.size()));
+}
+BENCHMARK_CAPTURE(GetMap, citm_catalog, TEST_DATA_DIRECTORY "/nativejson-benchmark/citm_catalog.json");
+
+/// 100,000 strings, each too long for the small string optimization
+static std::vector<std::string> make_strings()
+{
+    std::vector<std::string> result;
+    result.reserve(100000);
+    for (std::size_t i = 0; i < 100000; ++i)
+    {
+        result.push_back("a string that does not fit in place: " + std::to_string(i));
+    }
+    return result;
+}
+
+static void ToJsonStrings(benchmark::State& state, const bool move)
+{
+    const std::vector<std::string> strings = make_strings();
+
+    for (auto _ : state)
+    {
+        state.PauseTiming();
+        std::vector<std::string> input = strings;
+        state.ResumeTiming();
+
+        // destroyed inside the timed section, like the copies above
+        auto* j = move ? new json(std::move(input)) : new json(input);
+        delete j;
+    }
+
+    state.SetItemsProcessed(state.iterations() * static_cast<std::int64_t>(strings.size()));
+}
+BENCHMARK_CAPTURE(ToJsonStrings, copy, false);
+BENCHMARK_CAPTURE(ToJsonStrings, move, true);
+
 // conversion of a JSON array into standard containers
 template<typename Container>
 static void GetArray(benchmark::State& state)

tests/src/unit-class_lexer.cpp

@@ -13,11 +13,15 @@
 using nlohmann::json;
 
 #include <array> // array
+#include <cctype> // toupper
 #include <cstdint> // uint32_t, uint64_t
 #include <cstdio> // snprintf
+#include <cmath> // isfinite
 #include <cstdlib> // strtod
 #include <cstring> // memcpy
 #include <limits> // numeric_limits
+#include <initializer_list> // initializer_list
+#include <list> // list
 #include <map> // map
 #include <random> // mt19937
 #include <sstream> // stringstream
@@ -332,6 +336,112 @@ TEST_CASE("lexer number fast path")
         }
     }
 
+#if !defined(JSON_NOEXCEPTION)
+    // this section reports parse errors, which abort when exceptions are off
+    SECTION("digits accumulated while scanning")
+    {
+        // The fast path accumulates the digits while it validates them, eight
+        // at a time where possible, and converts tokens with at most 19
+        // digits (floats) or 18 digits (integers) from that value. Every digit
+        // count from 1 to 25, in the integer and in the fraction part, with
+        // and without sign and exponent, must give the byte path's result. A
+        // std::list has no contiguous storage, so parsing it takes the byte
+        // path.
+        // type and value, or the error (values beyond the range of double
+        // are rejected with out_of_range.406 on both paths)
+        const auto outcome = [](const std::string & doc, bool byte_path) -> std::string
+        {
+            JSON_TRY
+            {
+                const std::list<char> chars(doc.begin(), doc.end());
+                const json j = byte_path ? json::parse(chars.begin(), chars.end()) : json::parse(doc);
+                return std::string(j[0].type_name()) + '|' + j.dump();
+            }
+            JSON_CATCH (const json::exception & e)
+            {
+                return e.what();
+            }
+        };
+
+        // the concatenation of @a parts
+        const auto join = [](std::initializer_list<std::string> parts)
+        {
+            std::string result;
+            for (const auto& part : parts)
+            {
+                result += part;
+            }
+            return result;
+        };
+
+        std::vector<std::string> numbers;
+        for (std::size_t digits = 1; digits <= 25; ++digits)
+        {
+            std::string run;
+            for (std::size_t i = 0; i < digits; ++i)
+            {
+                run += static_cast<char>('1' + static_cast<char>((i * 7) % 9));
+            }
+            const std::string nines(digits, '9');
+            for (const char* sign_text :
+                    {"", "-"
+                    })
+            {
+                const std::string sign = sign_text;
+                numbers.push_back(join({sign, run}));
+                numbers.push_back(join({sign, nines}));
+                numbers.push_back(join({sign, "1", std::string(digits - 1, '0')}));
+                numbers.push_back(join({sign, "0.", run}));
+                numbers.push_back(join({sign, "0.", std::string(digits, '0'), "1"}));
+                numbers.push_back(join({sign, run, ".", run}));
+                numbers.push_back(join({sign, run, "e-", std::to_string(digits)}));
+                numbers.push_back(join({sign, "1.", run, "E+", std::to_string(digits * 13)}));
+            }
+        }
+        // exponents beyond every range, and zeros
+        numbers.insert(numbers.end(), {"1e400", "1e-400", "-1e99999999999999999999", "1e-99999999999999999999",
+                                       "0e99999999999999999999", "0.0e-5", "-0.000", "123456789e-0000000000000000000000330"
+                                      });
+
+        for (const auto& n : numbers)
+        {
+            CAPTURE(n)
+            const std::string doc = "[" + n + "]";
+            const std::string a = outcome(doc, false);
+            CHECK(a == outcome(doc, true));
+            const double expected = std::strtod(n.c_str(), nullptr);
+            if (a.compare(0, 6, "number") == 0 && std::isfinite(expected) && n.find_first_of(".eE") != std::string::npos)
+            {
+                CHECK(json::parse(doc)[0].get<double>() == expected);
+            }
+        }
+    }
+#endif
+
+    SECTION("integers that do not fit a narrower type become floats")
+    {
+        // up to 18 digits, the fast path stores the accumulated value without
+        // an overflow check; a narrower number_integer_t/number_unsigned_t
+        // must still see the value rejected and converted as a float
+        using json32 = nlohmann::basic_json<std::map, std::vector, std::string, bool, std::int32_t, std::uint32_t, float>;
+        for (const char* n :
+                {"4294967295", "4294967296", "2147483647", "-2147483648", "-2147483649", "99999999999", "-99999999999"
+                })
+        {
+            CAPTURE(n)
+            const std::string doc = std::string("[") + n + "]";
+            const std::list<char> chars(doc.begin(), doc.end());
+            const json32 a = json32::parse(doc);
+            const json32 b = json32::parse(chars.begin(), chars.end());
+            CHECK(a[0].type() == b[0].type());
+            CHECK(a == b);
+        }
+        CHECK(json32::parse("[4294967295]")[0].is_number_unsigned());
+        CHECK(json32::parse("[4294967296]")[0].is_number_float());
+        CHECK(json32::parse("[-2147483648]")[0].is_number_integer());
+        CHECK(json32::parse("[-2147483649]")[0].is_number_float());
+    }
+
     SECTION("token type classification")
     {
         CHECK((scan_string("0") == json::lexer::token_type::value_unsigned));
@@ -490,6 +600,67 @@ TEST_CASE("lexer number fast path")
 #endif
 }
 
+#if !defined(JSON_NOEXCEPTION)
+// this test case compares the errors of both lexer paths, which abort when
+// exceptions are off
+TEST_CASE("lexer literal fast path")
+{
+    // For contiguous input, the rest of a literal is compared at once, and a
+    // mismatch is left to the byte path, which reports the error at the
+    // character that differs. Every truncation and every wrong character of
+    // each literal, at the end of the input or followed by more text, must
+    // give the byte path's result (a std::list is read one character at a
+    // time).
+    const auto outcome = [](const std::string & doc, bool byte_path) -> std::string
+    {
+        JSON_TRY
+        {
+            const std::list<char> chars(doc.begin(), doc.end());
+            return (byte_path ? json::parse(chars.begin(), chars.end()) : json::parse(doc)).dump();
+        }
+        JSON_CATCH (const json::exception & e)
+        {
+            return e.what();
+        }
+    };
+
+    std::vector<std::string> mismatches;
+    for (const std::string literal :
+            {"true", "false", "null"
+            })
+    {
+        std::vector<std::string> variants;
+        for (std::size_t length = 1; length <= literal.size(); ++length)
+        {
+            variants.push_back(literal.substr(0, length));
+        }
+        for (std::size_t i = 1; i < literal.size(); ++i)
+        {
+            std::string changed = literal;
+            changed[i] = 'X';
+            variants.push_back(changed);
+            changed[i] = static_cast<char>(std::toupper(static_cast<unsigned char>(literal[i])));
+            variants.push_back(changed);
+        }
+        for (const auto& variant : variants)
+        {
+            for (const std::string& doc :
+                    {
+                        variant, "[" + variant + "]", "[" + variant, variant + " ", variant + "1", "{\"k\":" + variant + "}"
+                    })
+            {
+                if (outcome(doc, false) != outcome(doc, true))
+                {
+                    mismatches.push_back(doc);
+                }
+            }
+        }
+    }
+    CAPTURE(mismatches)
+    CHECK(mismatches.empty());
+}
+#endif
+
 TEST_CASE("lexer string fast path")
 {
     // Build a byte string from explicit values: a hex escape in a string
@@ -610,6 +781,73 @@ TEST_CASE("lexer string fast path")
         CAPTURE(mismatches)
         CHECK(mismatches.empty());
     }
+
+    SECTION("special bytes at every offset of the vector blocks")
+    {
+        // With SSE2 or NEON, runs are scanned 16 bytes at a time (after one
+        // SWAR word, or four with NEON), and runs of non-ASCII text are
+        // checked 16 bytes at a time as well. Place every kind of byte that
+        // ends a run at each offset across several blocks, after ASCII and
+        // after CJK text, so sequences also straddle the block boundaries.
+        // The reference is a std::list, which has no contiguous storage and
+        // is therefore read one character at a time.
+        const auto byte_path = [](const std::string & doc) -> std::string
+        {
+            const std::list<char> chars(doc.begin(), doc.end());
+            JSON_TRY
+            {
+                return json::parse(chars.begin(), chars.end()).dump();
+            }
+            JSON_CATCH (const json::exception & e)
+            {
+                return e.what();
+            }
+        };
+        const std::vector<std::string> specials =
+        {
+            "\"", "\\", "\\u00e9", bytes({0x01}), bytes({0x1F}), bytes({0x7F}),
+            bytes({0xC3, 0xA9}), bytes({0xE4, 0xB8, 0xAD}), bytes({0xF0, 0x9F, 0x98, 0x80}),
+            bytes({0xFF}), bytes({0xC3}), bytes({0xE4, 0xB8}), bytes({0xED, 0xA0, 0x80}), bytes({0xC0, 0x80})
+        };
+        const std::string cjk = bytes({0xE4, 0xB8, 0xAD}); // U+4E2D
+
+        std::vector<std::string> mismatches;
+        for (std::size_t offset = 0; offset <= 80; ++offset)
+        {
+            std::string text;
+            while (text.size() < offset)
+            {
+                text += cjk;
+            }
+            for (const auto& prefix :
+                    {
+                        std::string(offset, 'a'), text
+                    })
+            {
+                for (const auto& special : specials)
+                {
+                    // the closing quote right behind, or in a later block
+                    for (const int tail :
+                            {
+                                0, 1, 3, 20
+                            })
+                    {
+                        std::string doc = "[\"";
+                        doc += prefix;
+                        doc += special;
+                        doc.append(static_cast<std::size_t>(tail), 'b');
+                        doc += "\"]";
+                        if (outcome(doc, false) != byte_path(doc))
+                        {
+                            mismatches.push_back(doc);
+                        }
+                    }
+                }
+            }
+        }
+        CAPTURE(mismatches)
+        CHECK(mismatches.empty());
+    }
 #endif
 
     // json::accept() never throws, so the ranges stay covered without exceptions

tests/src/unit-class_parser.cpp

@@ -30,7 +30,9 @@ using nlohmann::json;
 #include <valarray>
 #include <algorithm>
 #include <cstdio>
+#include <cstdint>
 #include <fstream>
+#include <iterator>
 #include <list>
 #include <sstream>
 #include <string>
@@ -2450,6 +2452,109 @@ std::string parse_error_message_range(IteratorType first, IteratorType last)
 }
 } // namespace
 
+TEST_CASE("object keys are moved into the value")
+{
+    // json_sax_dom_parser::key() moves the key out of the lexer's buffer into
+    // the object, and the lexer starts the next token with an empty buffer.
+    // Keys of every length (inline and allocated strings), keys with escapes,
+    // and duplicate keys must give the same objects as reading the input one
+    // character at a time, and as ordered_json, whose map copies the key.
+    const std::string long_key(100, 'k');
+    const std::string doc = R"({"a": 1, ")" + long_key + R"(": {"b": [2, 3], "sixteen_chars_xx": 4}, )"
+                            R"("esc\"apedé": 5, "a": 6, ")" + long_key + R"(x": 7, "": 8, "dup": 9, "dup": 10})";
+
+    const json j = json::parse(doc);
+    const std::list<char> chars(doc.begin(), doc.end());
+    CHECK(j == json::parse(chars.begin(), chars.end()));
+
+    CHECK(j.size() == 6);
+    CHECK(j["a"] == 6);
+    CHECK(j[long_key]["b"] == json({2, 3}));
+    CHECK(j[long_key]["sixteen_chars_xx"] == 4);
+    CHECK(j["esc\"aped\xc3\xa9"] == 5);
+    CHECK(j[long_key + "x"] == 7);
+    CHECK(j[""] == 8);
+    CHECK(j["dup"] == 10);
+
+    const auto o = nlohmann::ordered_json::parse(doc);
+    CHECK(o.size() == 6);
+    CHECK(o[long_key + "x"] == 7);
+    CHECK(json(o) == j);
+}
+
+TEST_CASE("text of float tokens")
+{
+    // For contiguous input, the lexer copies the text of a float token into
+    // its buffer only when a SAX handler asks for it (the library's own
+    // handlers do not). A user handler must still see the exact text, for
+    // every input adapter and in every position of a document.
+    const std::string doc = R"([1.5, -0.0, 0.1, 1e5, -2.5E-4, 12345678901234567890.5, 0.12345678901234567890123,
+                              {"a": 3.25, "b": [6.02e23, 1]}, "x", true, 7e-1])";
+    const std::vector<std::string> expected_floats =
+    {
+        "number_float(1.5)", "number_float(-0.0)", "number_float(0.1)", "number_float(1e5)", "number_float(-2.5E-4)",
+        "number_float(12345678901234567890.5)", "number_float(0.12345678901234567890123)", "number_float(3.25)",
+        "number_float(6.02e23)", "number_float(7e-1)"
+    };
+
+    const auto floats = [](const std::vector<std::string>& events)
+    {
+        std::vector<std::string> result;
+        std::copy_if(events.begin(), events.end(), std::back_inserter(result), [](const std::string & e)
+        {
+            return e.compare(0, 13, "number_float(") == 0;
+        });
+        return result;
+    };
+
+    SECTION("contiguous input")
+    {
+        SaxEventLogger l;
+        CHECK(json::sax_parse(doc, &l));
+        CHECK(floats(l.events) == expected_floats);
+    }
+
+    SECTION("byte input")
+    {
+        const std::vector<std::uint8_t> bytes(doc.begin(), doc.end());
+        SaxEventLogger l;
+        CHECK(json::sax_parse(bytes, &l));
+        CHECK(floats(l.events) == expected_floats);
+    }
+
+    SECTION("stream input")
+    {
+        std::istringstream is(doc);
+        SaxEventLogger l;
+        CHECK(json::sax_parse(is, &l));
+        CHECK(floats(l.events) == expected_floats);
+    }
+
+    SECTION("non-contiguous iterators")
+    {
+        const std::list<char> chars(doc.begin(), doc.end());
+        SaxEventLogger l;
+        CHECK(json::sax_parse(chars.begin(), chars.end(), &l));
+        CHECK(floats(l.events) == expected_floats);
+    }
+
+    SECTION("the library's own handlers do not need the text")
+    {
+        using lexer_adapter = nlohmann::detail::iterator_input_adapter<const char*>;
+        CHECK(nlohmann::detail::sax_ignores_number_text<nlohmann::detail::json_sax_dom_parser<json, lexer_adapter>>::value);
+        CHECK(nlohmann::detail::sax_ignores_number_text<nlohmann::detail::json_sax_dom_callback_parser<json, lexer_adapter>>::value);
+        CHECK(nlohmann::detail::sax_ignores_number_text<nlohmann::detail::json_sax_acceptor<json>>::value);
+        CHECK_FALSE(nlohmann::detail::sax_ignores_number_text<SaxEventLogger>::value);
+
+        // and the values are the same as when the text is read
+        const json j = json::parse(doc);
+        CHECK(j[0] == 1.5);
+        CHECK(j[5].get<double>() == 12345678901234567890.5);
+        CHECK(j[7]["b"][0] == 6.02e23);
+        CHECK(json::accept(doc));
+    }
+}
+
 TEST_CASE("last-read diagnostics are identical across input adapters")
 {
     // The lexer reconstructs the "last read" token lazily for seekable adapters

tests/src/unit-json_view_dump.cpp

@@ -12,6 +12,7 @@
 using nlohmann::json;
 using nlohmann::ordered_json;
 using nlohmann::json_document;
+using nlohmann::ordered_json_editable_document;
 using nlohmann::json_view;
 using nlohmann::ordered_json_document;
 using nlohmann::ordered_json_view;
@@ -219,6 +220,76 @@ TEST_CASE("json_view dump")
         CHECK(d.root()["a"].dump(-1, ' ', true).size() == 2 + (5000 * 6));
     }
 
+    SECTION("ensure_ascii with strings of the source, decoded strings, and edited strings")
+    {
+        // plain ASCII, non-ASCII, escaped (decoded), and mixed strings, as values and as keys;
+        // the second text has a DEL, which ensure_ascii writes as an escape
+        for (const char* const del :
+                {"", "\x7F"
+                })
+        {
+            const std::string text = std::string("{\"plain\":\"abc\",\"caf\xC3\xA9\":\"\xC3\xA9\xE2\x82\xAC\xF0\x9F\x98\x80\",")
+                                     + "\"esc\\n\\\"\":\"a\\tb\\u00e4\\ud83d\\ude00\",\"del\":\"a" + del + "b\",\"\":\"\"," // NOLINT(modernize-raw-string-literal)
+                                     + "\"list\":[\"x\",\"\xC3\xBC\",\"\\\\\",\"long ASCII string with more than thirty-two bytes in it\","
+                                     + "\"long string with more than thirty-two bytes and one non-ASCII byte \xC3\xB6\"],"
+                                     + "\"n\":[1,2.5,-0,true,null]}";
+            const ordered_json j = ordered_json::parse(text);
+            const ordered_json_document d = ordered_json_document::parse(text);
+            CHECK(d.root().dump(-1, ' ', true) == j.dump(-1, ' ', true));
+            CHECK(d.root().dump(4, ' ', true) == j.dump(4, ' ', true));
+            CHECK(d.root().dump() == j.dump());
+            CHECK(d.root()["list"].dump(-1, ' ', true) == j["list"].dump(-1, ' ', true));
+            CHECK(d.root()["del"].dump(-1, ' ', true) == j["del"].dump(-1, ' ', true));
+
+            // strings that an edit wrote (the source strings that they replace were plain)
+            ordered_json_editable_document e = ordered_json_editable_document::parse(text);
+            ordered_json k = j;
+            e.set(e.root()["plain"], std::string("\xC3\xA4gypten"));
+            k["plain"] = "\xC3\xA4gypten";
+            e.set(e.root()["list"][0], std::string("ascii after edit"));
+            k["list"][0] = "ascii after edit";
+            e.set(e.root()["list"][1], std::string("\xC3\xA4"));
+            k["list"][1] = "\xC3\xA4";
+            e.set(e.root()["del"], std::string("tab\there"));
+            k["del"] = "tab\there";
+            CHECK(e.root().dump(-1, ' ', true) == k.dump(-1, ' ', true));
+            CHECK(e.root().dump(2, ' ', true) == k.dump(2, ' ', true));
+            CHECK(e.root().dump() == k.dump());
+
+            // strings that were edited from or to ASCII text
+            e.set(e.root()["list"][2], std::string("\xC3\xA4"));
+            k["list"][2] = "\xC3\xA4";
+            e.set(e.root()["list"][4], std::string("plain again"));
+            k["list"][4] = "plain again";
+            CHECK(e.root().dump(-1, ' ', true) == k.dump(-1, ' ', true));
+        }
+    }
+
+    SECTION("ensure_ascii with DEL or a non-ASCII character at every position")
+    {
+        // the check whether a string of the source has no byte >= 0x7F reads
+        // eight bytes at a time and then the rest one by one
+        for (std::size_t length = 1; length <= 20; ++length)
+        {
+            for (std::size_t pos = 0; pos < length; ++pos)
+            {
+                for (const char* const c :
+                        {
+                            "\x7F", "\x7E", "\xC3\xA4", "\xC3\xBF" // NOLINT(modernize-raw-string-literal)
+                        })
+                {
+                    const std::string s = std::string(pos, 'a') + c + std::string(length - pos - 1, 'b');
+                    std::string text = "[\"";
+                    text.append(s).append("\",{\"").append(s).append("\":1}]");
+                    const json_document d = json_document::parse(text);
+                    const json j = json::parse(text);
+                    CHECK(d.root().dump(-1, ' ', true) == j.dump(-1, ' ', true));
+                    CHECK(d.root().dump(1, ' ', true) == j.dump(1, ' ', true));
+                }
+            }
+        }
+    }
+
     SECTION("streams and discarded views")
     {
         const json_document d = json_document::parse(R"({"a": [1, 2]})");

tests/src/unit-serialization.cpp

@@ -1417,6 +1417,27 @@ void check_dump_against_reference(const std::string& s, const bool with_strict =
 }
 } // namespace
 
+TEST_CASE("every ASCII character at every position of a short string is written like the reference")
+{
+    // dump_escaped() copies strings of up to 32 bytes of printable ASCII after
+    // a table lookup per byte, and hands everything else to the scanners; put
+    // each ASCII character and a 2-byte character at every position of
+    // strings of up to 40 bytes to cover both sides of that bound
+    for (std::size_t length = 1; length <= 40; ++length)
+    {
+        for (std::size_t pos = 0; pos < length; ++pos)
+        {
+            for (int c = 0; c < 0x80; ++c)
+            {
+                std::string s(length, 'a');
+                s[pos] = static_cast<char>(c);
+                check_dump_against_reference(s);
+            }
+            check_dump_against_reference(std::string(pos, 'a') + "\xC3\xA4" + std::string(length - pos - 1, 'a'));
+        }
+    }
+}
+
 TEST_CASE("invalid UTF-8 at every position is handled like the reference")
 {
     // ill-formed byte sequences; those that end in a truncated sequence are
@@ -1559,3 +1580,94 @@ TEST_CASE("invalid UTF-8 at every position is handled like the reference")
         }
     }
 }
+
+TEST_CASE("short strings are written like the reference, as values and as keys")
+{
+    // strings of up to 16 characters (32 with vector instructions) that need no
+    // escaping are written as a whole (together with the quotes and, for a key,
+    // the colon); everything else, including a string that is merely short,
+    // takes the general path.
+    // Every length around the edges of the size classes of the fast path is
+    // combined with each kind of byte that must send a string to the general
+    // path, at each position, and compared with the reference for values and,
+    // via the value, for keys.
+    const std::vector<std::string> specials =
+    {
+        "\"", "\\", "\n", "\x1f", "\x7f", "\xc3\xa9", "\xe2\x82\xac", "\xf0\x9f\x98\x80", "\xff", "\x80", "\xc3"
+    };
+
+    for (std::size_t length = 0; length <= 36; ++length)
+    {
+        const std::string plain(length, 'k');
+        check_dump_against_reference(plain, false);
+
+        for (const auto& special : specials)
+        {
+            for (std::size_t pos = 0; pos <= length; ++pos)
+            {
+                std::string s = plain;
+                s.insert(pos, special);
+                check_dump_against_reference(s, false);
+
+                for (const bool ensure_ascii :
+                        {
+                            false, true
+                        })
+                {
+                    // a key and a value are written alike, in compact and
+                    // in pretty-printed form
+                    const json value = s;
+                    const json object = json::object({{s, 1}});
+                    for (const auto error_handler :
+                            {
+                                json::error_handler_t::replace, json::error_handler_t::keep
+                            })
+                    {
+                        const std::string quoted = value.dump(-1, ' ', ensure_ascii, error_handler);
+                        CHECK(object.dump(-1, ' ', ensure_ascii, error_handler) == "{" + quoted + ":1}");
+                        CHECK(object.dump(2, ' ', ensure_ascii, error_handler) == "{\n  " + quoted + ": 1\n}");
+                    }
+                }
+            }
+        }
+    }
+}
+
+TEST_CASE("runs of multi-byte characters are escaped like the reference under ensure_ascii")
+{
+    // a run of well-formed multi-byte characters is escaped straight into the
+    // write buffer; it has to flush that mid-run, stop at an ASCII character or
+    // an ill-formed sequence and leave those to the general path, and pick up
+    // behind what the general path has pending
+    const std::vector<std::string> characters =
+    {
+        "\xc2\x80", "\xdf\xbf", "\xe0\xa0\x80", "\xe2\x82\xac", "\xed\x9f\xbf", "\xef\xbf\xbf",
+        "\xf0\x90\x80\x80", "\xf0\x9f\x98\x80", "\xf4\x8f\xbf\xbf"
+    };
+    const std::vector<std::string> interruptions =
+    {
+        "", "a", "\n", "\"", "\x7f", "\xff", "\x80", "\xc3", "\xe2\x82", "\xf0\x9f\x98", "\xed\xa0\x80"
+    };
+
+    for (std::size_t shift = 0; shift <= 13; ++shift)
+    {
+        for (const auto& interruption : interruptions)
+        {
+            for (const std::size_t count :
+                    {
+                        std::size_t{1}, std::size_t{2}, std::size_t{85}, std::size_t{200}
+                    })
+            {
+                std::string s(shift, 'x');
+                for (std::size_t i = 0; i < count; ++i)
+                {
+                    s += characters[(i + shift) % characters.size()];
+                }
+                s += interruption;
+                s += characters[shift % characters.size()];
+                s += characters[(shift + 3) % characters.size()];
+                check_dump_against_reference(s, shift < 2);
+            }
+        }
+    }
+}

tests/src/unit-stl_helpers.cpp

@@ -0,0 +1,129 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#include "doctest_compatibility.h"
+
+#include <nlohmann/json.hpp>
+
+#include <array>
+#include <cstdint>
+#include <limits>
+#include <type_traits>
+
+// The detail helpers that stand in for standard library functions of newer
+// standards: each uses the standard function where the library has it and a
+// C++11 implementation otherwise, so this file is also built with
+// JSON_HAS_CPP_17, JSON_HAS_CPP_20, and JSON_HAS_CPP_23 to test both.
+
+namespace
+{
+struct with_constexpr_member
+{
+    static constexpr std::size_t lower = 64; // no out-of-line definition
+};
+} // namespace
+
+TEST_CASE("detail::clamp")
+{
+    using nlohmann::detail::clamp;
+
+    SECTION("integers")
+    {
+        CHECK(clamp(5, 1, 10) == 5);
+        CHECK(clamp(0, 1, 10) == 1);
+        CHECK(clamp(11, 1, 10) == 10);
+        CHECK(clamp(1, 1, 10) == 1);
+        CHECK(clamp(10, 1, 10) == 10);
+        CHECK(clamp(7, 7, 7) == 7);
+        CHECK(clamp(-5, -3, 3) == -3);
+        CHECK(clamp(std::size_t{0}, std::size_t{64}, std::size_t{0x10000}) == 64);
+        CHECK(clamp(std::size_t{0x20000}, std::size_t{64}, std::size_t{0x10000}) == 0x10000);
+    }
+
+    SECTION("floating-point numbers")
+    {
+        CHECK(clamp(0.5, 0.0, 1.0) == 0.5);
+        CHECK(clamp(-0.5, 0.0, 1.0) == 0.0);
+        CHECK(clamp(1.5, 0.0, 1.0) == 1.0);
+        CHECK(clamp(-std::numeric_limits<double>::infinity(), 0.0, 1.0) == 0.0);
+    }
+
+    SECTION("by value: a static constexpr member needs no definition")
+    {
+        CHECK(clamp(std::size_t{1}, with_constexpr_member::lower, std::size_t{100}) == 64);
+    }
+
+    SECTION("constexpr and noexcept")
+    {
+        static_assert(clamp(3, 1, 2) == 2, "clamp must be constexpr");
+        static_assert(noexcept(clamp(3, 1, 2)), "clamp must be noexcept");
+    }
+}
+
+TEST_CASE("detail::bit_cast")
+{
+    using nlohmann::detail::bit_cast;
+
+    CHECK(bit_cast<std::uint64_t>(1.0) == 0x3FF0000000000000u);
+    CHECK(bit_cast<std::uint64_t>(-2.0) == 0xC000000000000000u);
+    CHECK(bit_cast<std::uint32_t>(1.0f) == 0x3F800000u);
+    CHECK(bit_cast<double>(std::uint64_t{0x4009000000000000u}) == 3.125);
+    CHECK(bit_cast<float>(std::uint32_t{0xC0490000u}) == -3.140625f);
+    CHECK(bit_cast<std::int8_t>(std::uint8_t{0xFF}) == -1);
+
+    // to and from byte arrays
+    const auto bytes = bit_cast<std::array<std::uint8_t, 4>>(std::uint32_t{0x01020304u});
+    CHECK(bit_cast<std::uint32_t>(bytes) == 0x01020304u);
+
+    static_assert(noexcept(bit_cast<std::uint64_t>(1.0)), "bit_cast must be noexcept");
+}
+
+TEST_CASE("detail::byteswap")
+{
+    using nlohmann::detail::byteswap;
+
+    CHECK(byteswap(std::uint16_t{0x0102u}) == 0x0201u);
+    CHECK(byteswap(std::uint16_t{0xFF00u}) == 0x00FFu);
+    CHECK(byteswap(std::uint32_t{0x01020304u}) == 0x04030201u);
+    CHECK(byteswap(std::uint32_t{0xFF000080u}) == 0x800000FFu);
+    CHECK(byteswap(std::uint64_t{0x0102030405060708u}) == 0x0807060504030201u);
+    CHECK(byteswap(std::uint64_t{0xFF00000000000080u}) == 0x80000000000000FFu);
+
+    // an involution
+    for (std::uint64_t x :
+            {
+                std::uint64_t{0}, std::uint64_t{1}, std::uint64_t{0x123456789ABCDEF0u}, (std::numeric_limits<std::uint64_t>::max)()
+            })
+    {
+        CHECK(byteswap(byteswap(x)) == x);
+        CHECK(byteswap(byteswap(static_cast<std::uint32_t>(x))) == static_cast<std::uint32_t>(x));
+        CHECK(byteswap(byteswap(static_cast<std::uint16_t>(x))) == static_cast<std::uint16_t>(x));
+    }
+
+    static_assert(std::is_same < decltype(byteswap(std::uint16_t{})), std::uint16_t >::value, "byteswap keeps the type");
+    static_assert(std::is_same < decltype(byteswap(std::uint32_t{})), std::uint32_t >::value, "byteswap keeps the type");
+    static_assert(std::is_same < decltype(byteswap(std::uint64_t{})), std::uint64_t >::value, "byteswap keeps the type");
+}
+
+TEST_CASE("detail::count_leading_zeros and count_trailing_zeros")
+{
+    using nlohmann::detail::count_leading_zeros;
+    using nlohmann::detail::count_trailing_zeros;
+
+    // a single bit, and that bit with all lower (resp. higher) bits set: both
+    // halves of the 32-bit fallbacks
+    for (int k = 0; k < 64; ++k)
+    {
+        const std::uint64_t bit = std::uint64_t{1} << k;
+        CAPTURE(k)
+        CHECK(count_leading_zeros(bit) == 63 - k);
+        CHECK(count_leading_zeros(bit | (bit - 1)) == 63 - k);
+        CHECK(count_trailing_zeros(bit) == k);
+        CHECK(count_trailing_zeros(bit | ~(bit | (bit - 1))) == k);
+    }
+}

tests/src/unit-stream-bulk-scan.cpp

@@ -171,7 +171,8 @@ std::vector<std::string> corpus()
         R"("unterminated)", "\"control\x01character\"", "\"ill-formed \xc3\x28 UTF-8\"", R"("bad \x escape")",
         R"("\ud800 lone surrogate")",
         // literals
-        "true", "false", "null", "tru", "nul", "falsey",
+        "true", "false", "null", "tru", "nul", "falsey", "[true,false,null]", "nulL", "truex", "[fals e]", "t",
+        "true1", "null null", "[false]",
         // structures
         "[]", "{}", "[1,2,3]", R"({"a":1,"b":[true,false,null],"c":{"d":"e"}})", "[1,]", R"({"a" 1})", "[[[[[]]]]]",
         // pretty-printed

tests/src/unit-to_chars.cpp

@@ -13,7 +13,7 @@
 #include "doctest_compatibility.h"
 
 #include <nlohmann/json.hpp>
-using nlohmann::detail::dtoa_impl::reinterpret_bits;
+using nlohmann::detail::bit_cast;
 
 #include <algorithm>
 #include <array>
@@ -44,7 +44,7 @@ float make_float(uint32_t sign_bit, uint32_t biased_exponent, uint32_t significa
     bits |= biased_exponent << 23;
     bits |= significand;
 
-    return reinterpret_bits<float>(bits);
+    return bit_cast<float>(bits);
 }
 
 // ldexp -- convert f * 2^e to IEEE single precision
@@ -81,7 +81,7 @@ float make_float(uint64_t f, int e)
                                      : static_cast<uint64_t>(e + kExponentBias);
 
     const uint64_t bits = (f & kSignificandMask) | (biased_exponent << kPhysicalSignificandSize);
-    return reinterpret_bits<float>(static_cast<uint32_t>(bits));
+    return bit_cast<float>(static_cast<uint32_t>(bits));
 }
 
 double make_double(uint64_t sign_bit, uint64_t biased_exponent, uint64_t significand)
@@ -96,7 +96,7 @@ double make_double(uint64_t sign_bit, uint64_t biased_exponent, uint64_t signifi
     bits |= biased_exponent << 52;
     bits |= significand;
 
-    return reinterpret_bits<double>(bits);
+    return bit_cast<double>(bits);
 }
 
 // ldexp -- convert f * 2^e to IEEE double precision
@@ -133,7 +133,7 @@ double make_double(uint64_t f, int e)
                                      : static_cast<uint64_t>(e + kExponentBias);
 
     const uint64_t bits = (f & kSignificandMask) | (biased_exponent << kPhysicalSignificandSize);
-    return reinterpret_bits<double>(bits);
+    return bit_cast<double>(bits);
 }
 } // namespace
 
@@ -652,7 +652,7 @@ bool reads_back(const std::string& digits, int e, double v)
 {
     const std::string text = digits + "e" + std::to_string(e);
     // (compared bit for bit: v is positive and finite, and -Wfloat-equal)
-    return reinterpret_bits<std::uint64_t>(parse_double(text)) == reinterpret_bits<std::uint64_t>(v);
+    return bit_cast<std::uint64_t>(parse_double(text)) == bit_cast<std::uint64_t>(v);
 }
 
 /// Check the representation of a positive finite double: it reads back as
@@ -665,7 +665,7 @@ void check_shortest(double v)
     CAPTURE(text)
     CHECK(parse_double(text) == v);
     // the layout is that of format_buffer() for the digits of Zmij
-    const auto sd = nlohmann::detail::zmij::to_shortest(reinterpret_bits<std::uint64_t>(v));
+    const auto sd = nlohmann::detail::zmij::to_shortest(bit_cast<std::uint64_t>(v));
     const std::uint64_t significand = sd.has_digit ? (sd.integral * 10) + sd.digit : sd.integral;
     int exponent = sd.has_digit ? sd.exponent : sd.exponent + 1;
     std::string significand_digits = std::to_string(significand);
@@ -779,7 +779,7 @@ TEST_CASE("shortest digits of doubles")
         for (int i = 0; i < 100000; ++i)
         {
             const std::uint64_t bits = rng() & 0x7FFFFFFFFFFFFFFFu;
-            const auto v = reinterpret_bits<double>(bits);
+            const auto v = bit_cast<double>(bits);
             if (std::isfinite(v) && bits != 0)
             {
                 check_shortest(v);

tools/generate_natvis/README.md

@@ -1,6 +1,10 @@
 # generate_natvis.py
 
-Generate the Natvis debugger visualization file for all supported namespace combinations.
+Generate the Natvis debugger visualization file for namespace combinations with up to two ABI tags.
+
+The checked-in `nlohmann_json.natvis` covers the default namespace (no ABI tags), every single ABI tag, and every pair
+of ABI tags, with both versioned and unversioned namespace variants. Configurations that enable three or more ABI tags
+are not covered by this reduced file; use `--abi-tags` or `--all` (see below) to generate a file for them.
 
 The ABI tag list and the library version are parsed from
 `include/nlohmann/detail/abi_macros.hpp`, so this script must be re-run (via
@@ -19,10 +23,43 @@ make natvis
 `requirements.txt` (Jinja2) and runs the script with it. Alternatively, install the requirements yourself and run:
 
 ```shell
-./generate_natvis.py [--version X.Y.Z] [repository_root/]
+./generate_natvis.py [--version X.Y.Z] [--abi-tags TAGS]... [--all] [--list-tags] [--output FILE] [repository_root/]
 ```
 
-`--version` and the output/repository-root directory both default to values
+`--version` and the repository root both default to values
 derived from this script's own location, so they only need to be given
 explicitly when generating a Natvis file for a different checkout or a
 version other than the one in `abi_macros.hpp`.
+
+| Option            | Description                                                                                                                                                                                                                                         |
+|-------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| `--list-tags`     | Print every ABI tag (in namespace order) with the macro that enables it and the `NLOHMANN_JSON_ABI_TAG_*` identifier, then exit.                                                                                                                    |
+| `--abi-tags TAGS` | Add the namespaces of one configuration, given as a comma-separated list of ABI tags (`_diag,_dp,_vns`; the leading underscore is optional, and the order does not matter). Can be repeated for several configurations. Unknown tags are an error. |
+| `--all`           | Generate the namespaces for every combination of ABI tags. The file size doubles with each tag (about 60 MB for 12 tags). Cannot be combined with `--abi-tags`.                                                                                    |
+| `--output FILE`   | Write the Natvis file to `FILE` instead of `nlohmann_json.natvis` in the repository root.                                                                                                                                                           |
+
+Each configuration gets both the versioned (`json_abi_diag_dp_vns_v3_12_0`) and the unversioned (`json_abi_diag_dp_vns`,
+used with `NLOHMANN_JSON_NAMESPACE_NO_VERSION`) namespace. The default namespaces (up to two ABI tags) are always
+included, so a generated file can replace the checked-in one.
+
+### Generating a Natvis file for your configuration
+
+1. Check out the release tag of the library version you use (the ABI tags and the version are read from the checkout).
+2. Find the ABI tags of the macros you enable:
+
+   ```shell
+   make natvis NATVIS_ARGS="--list-tags"
+   ```
+
+3. Generate a Natvis file with them, for example for `JSON_DIAGNOSTICS`, `JSON_DIAGNOSTIC_POSITIONS`, and
+   `JSON_VIEW_NO_SIMD`:
+
+   ```shell
+   make natvis NATVIS_ARGS="--abi-tags _diag,_dp,_vns --output my_nlohmann_json.natvis"
+   ```
+
+4. Use `my_nlohmann_json.natvis` instead of the shipped `nlohmann_json.natvis`, for example by adding it to your Visual
+   Studio project.
+
+Passing `--output` keeps the checked-in `nlohmann_json.natvis` unchanged; without it, `make check-amalgamation` would
+report the file as out of date.

tools/generate_natvis/generate_natvis.py

@@ -20,9 +20,11 @@ def semver(v):
 
 
 def abi_info(repo_root):
-    """Derive the ABI tag list (in NLOHMANN_JSON_ABI_TAGS_CONCAT order) and the
+    """Derive the ABI tags (in NLOHMANN_JSON_ABI_TAGS_CONCAT order) and the
     library version from <repo_root>/include/nlohmann/detail/abi_macros.hpp,
-    so this script cannot drift from the header it visualizes."""
+    so this script cannot drift from the header it visualizes. Each tag is
+    returned as a (suffix, configuration macro, tag macro) tuple, for example
+    ('_diag', 'JSON_DIAGNOSTICS', 'NLOHMANN_JSON_ABI_TAG_DIAGNOSTICS')."""
     abi_macros_hpp = os.path.join(repo_root, 'include', 'nlohmann', 'detail', 'abi_macros.hpp')
     with open(abi_macros_hpp) as f:
         content = f.read()
@@ -47,7 +49,9 @@ def abi_info(repo_root):
         match = re.search(r'#define\s+' + re.escape(ident) + r'\s+(_\w+)\s*\n', content)
         if not match:
             raise ValueError(f'could not find a non-empty #define for {ident} in {abi_macros_hpp}')
-        abi_tags.append(match.group(1))
+        # the option that enables the tag is the #if/#ifdef right above it
+        option = re.search(r'#\s*(?:if|ifdef)\s+(\w+)\s*\n\s*#define\s+' + re.escape(ident) + r'\s+_', content)
+        abi_tags.append((match.group(1), option.group(1) if option else '?', ident))
 
     version = {}
     for part in ('MAJOR', 'MINOR', 'PATCH'):
@@ -59,30 +63,79 @@ def abi_info(repo_root):
     return abi_tags, '{MAJOR}.{MINOR}.{PATCH}'.format(**version)
 
 
+def parse_tag_list(value, abi_tags):
+    """Turn one --abi-tags value ("_diag,dp") into the tuple of suffixes it
+    names, in namespace order; unknown tags and empty lists are errors."""
+    known = [suffix for suffix, _, _ in abi_tags]
+    requested = set()
+    for tag in value.split(','):
+        tag = tag.strip()
+        if not tag:
+            continue
+        if not tag.startswith('_'):
+            tag = '_' + tag
+        if tag not in known:
+            raise ValueError(f'unknown ABI tag {tag!r}; valid tags: {", ".join(known)} (see --list-tags)')
+        requested.add(tag)
+    if not requested:
+        raise ValueError(f'empty ABI tag list {value!r}; the namespace without tags is always generated')
+    return tuple(tag for tag in known if tag in requested)
+
+
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(
+        description='Generate nlohmann_json.natvis. By default, it covers the namespaces with up to two ABI tags; '
+                    'use --abi-tags or --all for other configurations.')
     parser.add_argument('--version', type=semver,
                          help='Library version number (default: parsed from '
-                              'include/nlohmann/detail/abi_macros.hpp below "output")')
-    parser.add_argument('output', nargs='?', default=REPO_ROOT,
-                         help='Repository root: where include/nlohmann/detail/abi_macros.hpp is '
-                              'read from and where nlohmann_json.natvis is written '
+                              'include/nlohmann/detail/abi_macros.hpp below "repo_root")')
+    selection = parser.add_mutually_exclusive_group()
+    selection.add_argument('--abi-tags', action='append', default=[], metavar='TAGS',
+                           help='Comma-separated list of ABI tags (e.g. "_diag,_dp,_vns", leading underscores are '
+                                'optional) of one configuration to add to the default namespaces; can be repeated')
+    selection.add_argument('--all', action='store_true',
+                           help='Generate the namespaces for every combination of ABI tags; the file size doubles '
+                                'with each ABI tag (about 60 MB for 12 tags)')
+    parser.add_argument('--list-tags', action='store_true',
+                        help='List the ABI tags with the macros that enable them, and exit')
+    parser.add_argument('--output', metavar='FILE',
+                        help='Where to write the Natvis file (default: nlohmann_json.natvis below "repo_root")')
+    parser.add_argument('repo_root', nargs='?', default=REPO_ROOT,
+                         help='Repository root: where include/nlohmann/detail/abi_macros.hpp is read from '
                               '(default: the repository root this script lives in)')
     args = parser.parse_args()
 
-    derived_tags, derived_version = abi_info(args.output)
+    abi_tags, derived_version = abi_info(args.repo_root)
+
+    if args.list_tags:
+        width = max(len(suffix) for suffix, _, _ in abi_tags)
+        option_width = max(len(option) for _, option, _ in abi_tags)
+        for suffix, option, ident in abi_tags:
+            print(f'{suffix:<{width}}  {option:<{option_width}}  {ident}')
+        sys.exit(0)
+
+    try:
+        extra_combinations = [parse_tag_list(value, abi_tags) for value in args.abi_tags]
+    except ValueError as e:
+        parser.error(str(e))
 
     namespaces = ['nlohmann']
     abi_prefix = 'json_abi'
-    abi_tags = derived_tags
+    suffixes = [suffix for suffix, _, _ in abi_tags]
     version = '_v' + (args.version or derived_version).replace('.', '_')
-    inline_namespaces = []
 
-    # generate all combinations of inline namespace names
-    for n in range(0, len(abi_tags) + 1):
-        for tags in itertools.combinations(abi_tags, n):
-            ns = abi_prefix + ''.join(tags)
-            inline_namespaces += [ns, ns + version]
+    # by default, generate the namespace names for up to two ABI tags, plus
+    # the combinations requested with --abi-tags
+    max_tags = len(suffixes) if args.all else min(len(suffixes), 2)
+    combinations = [tags for n in range(0, max_tags + 1) for tags in itertools.combinations(suffixes, n)]
+    for tags in extra_combinations:
+        if tags not in combinations:
+            combinations.append(tags)
+
+    inline_namespaces = []
+    for tags in combinations:
+        ns = abi_prefix + ''.join(tags)
+        inline_namespaces += [ns, ns + version]
 
     namespaces += [f'{namespaces[0]}::{ns}' for ns in inline_namespaces]
 
@@ -91,5 +144,10 @@ if __name__ == '__main__':
     template = env.get_template('nlohmann_json.natvis.j2')
     natvis = template.render(namespaces=namespaces)
 
-    with open(os.path.join(args.output, 'nlohmann_json.natvis'), 'w') as f:
+    output = args.output or os.path.join(args.repo_root, 'nlohmann_json.natvis')
+    with open(output, 'w') as f:
         f.write(natvis)
+
+    if args.all:
+        print(f'wrote {len(namespaces)} namespaces ({os.path.getsize(output) / 1e6:.1f} MB) to {output}',
+              file=sys.stderr)