Bump Semgrep to 1.182.1 and allow two unfixable advisories

Semgrep 1.168.0 pins pyjwt~=2.13.0, which dependency-review flags
(GHSA-ffc3-869f-jxw9 and 14 more). Semgrep 1.182.1 requires
pyjwt>=2.15.0; the lock file now has pyjwt 2.15.1, mcp 1.29.0 and
click 8.4.2.

Semgrep 1.182.1 still pins mcp==1.29.0 (GHSA-84m7-p3x7-pcfv) and
opentelemetry-api~=1.37.0 (GHSA-xcq7-p3pf-p88h). Neither code path is
used by `semgrep scan` in CI, so dependency-review allows both until
Semgrep lifts its pins.

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

.github/dependabot.yml

@@ -29,6 +29,13 @@ updates:
       - dependency-name: mermaid
         update-types: ["version-update:semver-major"]
 
+  - package-ecosystem: pip
+    directory: /.github/requirements
+    schedule:
+      interval: daily
+    cooldown:
+      default-days: 7
+
   - package-ecosystem: pip
     directory: /tools/astyle
     schedule:
@@ -36,7 +43,7 @@ updates:
     cooldown:
       default-days: 7
     ignore:
-      # astyle is deliberately pinned (see tools/astyle/requirements.txt): newer versions
+      # astyle is deliberately pinned (see tools/astyle/requirements.in): newer versions
       # reformat the code, and the pinned version defines the formatting CI enforces
       - dependency-name: astyle
 

.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/requirements/meson.in

@@ -0,0 +1 @@
+meson==1.12.1

.github/requirements/meson.txt

@@ -0,0 +1,10 @@
+#
+# This file is autogenerated by pip-compile with Python 3.14
+# by the following command:
+#
+#    pip-compile --generate-hashes --no-emit-index-url --output-file=.github/requirements/meson.txt --strip-extras .github/requirements/meson.in
+#
+meson==1.12.1 \
+    --hash=sha256:930bc7542cbd9f57009e182fd014eba48cf1a0180a7b9006d2d32d8f168d8b02 \
+    --hash=sha256:ab0a6ca09f8ef70c564c8241fb5a23957886a0b53fb58412b5e07eaf07dba743
+    # via -r .github/requirements/meson.in

.github/requirements/semgrep.in

@@ -0,0 +1 @@
+semgrep==1.182.1

.github/workflows/check_amalgamation.yml

@@ -37,7 +37,7 @@ jobs:
           path: pr/
 
   check:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     env:
       MAIN_DIR: ${{ github.workspace }}/main
       INCLUDE_DIR: ${{ github.workspace }}/main/single_include/nlohmann
@@ -67,23 +67,22 @@ jobs:
       - name: Install astyle
         run: |
           python3 -mvenv venv
-          venv/bin/pip3 install -r $MAIN_DIR/tools/astyle/requirements.txt
+          venv/bin/pip3 install --require-hashes -r $MAIN_DIR/tools/astyle/requirements.txt
 
       - name: Install generate_natvis dependencies
-        run: pip3 install -r $NATVIS_TOOL_DIR/requirements.txt
+        run: pip3 install --require-hashes -r $MAIN_DIR/tools/generate_natvis/requirements.txt
 
       - name: Regenerate the tools/macro_builder tables in macro_scope.hpp
         run: |
           cd $MAIN_DIR
 
           # Built from this PR's own tools/macro_builder/main.cpp, not a
-          # develop checkout: unlike amalgamate.py and generate_natvis.py,
-          # this tool has no other source of truth to check against (its
-          # own README documents that it must reproduce macro_scope.hpp
-          # byte for byte), so there is nothing to gain from checking out a
-          # separate copy, and doing so would make this step fail on a PR
-          # that adds support for a new dispatch table until that PR itself
-          # merges to develop.
+          # develop checkout: unlike amalgamate.py, this tool has no other
+          # source of truth to check against (its own README documents that
+          # it must reproduce macro_scope.hpp byte for byte), so there is
+          # nothing to gain from checking out a separate copy, and doing so
+          # would make this step fail on a PR that adds support for a new
+          # dispatch table until that PR itself merges to develop.
           TMPDIR=$(mktemp -d ./macro_builder_check.XXXXXX)
           c++ -std=c++11 tools/macro_builder/main.cpp -o "$TMPDIR/macro_builder"
           "$TMPDIR/macro_builder" > "$TMPDIR/paste.hpp"
@@ -146,14 +145,36 @@ jobs:
       - name: Regenerate nlohmann_json.natvis
         run: |
           cd $MAIN_DIR
-          # The template is what this check verifies the committed natvis
-          # file against, so take it from this PR rather than from the
-          # develop checkout: generate_natvis.py loads the template from its
-          # own directory, and with develop's copy a PR that changes the
-          # template could never pass. Like the macro_builder step above,
-          # this only renders PR content in a job without credentials.
-          cp tools/generate_natvis/nlohmann_json.natvis.j2 $NATVIS_TOOL_DIR/
-          python3 $NATVIS_TOOL_DIR/generate_natvis.py $MAIN_DIR
+          # The generator and its template are what this check verifies the
+          # committed natvis file against (which namespaces it lists, and
+          # what each entry looks like), so run this PR's copy rather than
+          # the develop checkout's: with develop's, a PR that changes either
+          # could never pass. generate_natvis.py loads the template from its
+          # own directory. Like the macro_builder step above, this only runs
+          # 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

.github/workflows/codeql-analysis.yml

@@ -21,7 +21,7 @@ permissions:
 jobs:
   CodeQL-Build:
 
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       security-events: write
 
@@ -38,14 +38,14 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/init@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
       with:
         languages: c-cpp
 
     # Autobuild attempts to build any compiled languages  (C/C++, C#, or Java).
     # If this step fails, then you should remove it and run the build manually (see below)
     - name: Autobuild
-      uses: github/codeql-action/autobuild@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/autobuild@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/analyze@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0

.github/workflows/comment_check_amalgamation.yml

@@ -37,6 +37,14 @@ jobs:
             var matchArtifact = artifacts.data.artifacts.filter((artifact) => {
               return artifact.name == "pr"
             })[0];
+            // a run that failed before its jobs started (e.g. a stale queued
+            // run cancelled long after its push) has no artifact to report on
+            if (!matchArtifact) {
+              core.notice('Run ${{ github.event.workflow_run.id }} has no "pr" artifact; nothing to comment on.');
+              core.setOutput('found', 'false');
+              return;
+            }
+            core.setOutput('found', 'true');
             var download = await github.rest.actions.downloadArtifact({
                owner: context.repo.owner,
                repo: context.repo.repo,
@@ -51,9 +59,11 @@ jobs:
       # Extract the untrusted PR artifact into a dedicated empty directory and
       # read only the two expected files by fixed path afterwards. This avoids a
       # malicious archive overwriting workspace files or escaping via ../ paths.
-      - run: unzip -o pr.zip -d ./pr_artifact
+      - if: steps.download.outputs.found == 'true'
+        run: unzip -o pr.zip -d ./pr_artifact
 
       - name: 'Comment on PR'
+        if: steps.download.outputs.found == 'true'
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
         with:
           github-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/dependency-review.yml

@@ -31,3 +31,12 @@ jobs:
           persist-credentials: false
       - name: 'Dependency Review'
         uses: actions/dependency-review-action@a1d282b36b6f3519aa1f3fc636f609c47dddb294 # v5.0.0
+        with:
+          # Semgrep pins mcp==1.29.0 and opentelemetry-api~=1.37.0 exactly
+          # (.github/requirements/semgrep.txt), so these cannot be upgraded
+          # until Semgrep lifts its pins. They do not apply to our use: CI
+          # only runs `semgrep scan` on the checkout and never starts the MCP
+          # server (GHSA-84m7-p3x7-pcfv, StreamableHTTP sessions) or parses
+          # incoming trace context (GHSA-xcq7-p3pf-p88h, tracestate parsing).
+          # Remove the entries once a Semgrep release allows the fixed versions.
+          allow-ghsas: GHSA-84m7-p3x7-pcfv, GHSA-xcq7-p3pf-p88h

.github/workflows/flawfinder.yml

@@ -47,6 +47,6 @@ jobs:
           output: 'flawfinder_results.sarif'
 
       - name: Upload analysis results to GitHub Security tab
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: ${{github.workspace}}/flawfinder_results.sarif

.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@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
+        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@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
+        with:
+          name: json_view-benchmarks-${{ matrix.runner }}-${{ inputs.compiler || 'g++' }}
+          path: tests/benchmarks/json_view/results/

.github/workflows/scorecards.yml

@@ -50,7 +50,7 @@ jobs:
           persist-credentials: false
 
       - name: "Run analysis"
-        uses: ossf/scorecard-action@2d1146689b8cda280b9bc96326124645441f03bc # v2.4.4
+        uses: ossf/scorecard-action@fa7117db83c52d787ae5a89f350999701505d78e # v2.5.0
         with:
           results_file: results.sarif
           results_format: sarif
@@ -80,6 +80,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: results.sarif

.github/workflows/semgrep.yml

@@ -33,7 +33,7 @@ jobs:
       security-events: write # for github/codeql-action/upload-sarif to upload SARIF results
       actions: read # only required for a private repository by github/codeql-action/upload-sarif to get the Action run status
     name: Scan
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
         uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
@@ -53,7 +53,7 @@ jobs:
       # SEMGREP_APP_TOKEN is still passed through so registry auth works if a
       # token is ever added.
       - name: Install Semgrep
-        run: python3 -m pip install --user semgrep==1.168.0
+        run: python3 -m pip install --user --require-hashes -r .github/requirements/semgrep.txt
 
       # `semgrep scan --sarif` always exits 0 even with findings; continue-on-error
       # is a safety net so the SARIF upload still runs if the scan itself errors.
@@ -65,7 +65,7 @@ jobs:
 
       # Upload SARIF file generated in previous step
       - name: Upload SARIF file
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
         with:
           sarif_file: semgrep.sarif
         if: always()

.github/workflows/ubuntu.yml

@@ -32,7 +32,7 @@ jobs:
         run: cmake --build build --target ci_test_gcc
 
   ci_meson_install:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
@@ -42,7 +42,7 @@ jobs:
       - name: Check that Meson and pkg-config offer the CMake options
         run: make check_build_options
       - name: Install Meson
-        run: pip install meson
+        run: pip install --require-hashes -r .github/requirements/meson.txt
       - name: Install with Meson
         run: |
           meson setup build-meson --prefix=${{ github.workspace }}/install
@@ -77,7 +77,7 @@ jobs:
           cmake --build build-import-split
 
   ci_infer:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
         uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
@@ -106,7 +106,7 @@ jobs:
           if-no-files-found: ignore
 
   ci_static_analysis_ubuntu:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       matrix:
         target: [ci_test_single_header, ci_cppcheck, ci_cpplint, ci_reproducible_tests, ci_non_git_tests, ci_offline_testdata, ci_reuse_compliance, ci_test_valgrind]
@@ -175,7 +175,7 @@ jobs:
         run: cmake --build build --target ${{ matrix.target }}
 
   ci_test_coverage:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
         uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
@@ -430,7 +430,7 @@ jobs:
         run: cmake --build build --target ci_nvhpc
 
   ci_emscripten:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
         uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
@@ -450,7 +450,7 @@ jobs:
         run: cmake --build build
 
   ci_test_documentation:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       matrix:
         target: [ci_test_examples, ci_test_build_documentation, ci_test_documentation_mermaid]

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

cmake/ci.cmake

@@ -361,10 +361,10 @@ add_custom_target(ci_test_coverage
     COMMAND ${CMAKE_COMMAND} --build ${PROJECT_BINARY_DIR}/build_coverage32
     COMMAND cd ${PROJECT_BINARY_DIR}/build_coverage32 && ${CMAKE_CTEST_COMMAND} --parallel ${N} --output-on-failure
 
-    COMMAND ${LCOV_TOOL} --directory . --capture --output-file json.info --rc branch_coverage=1 --rc geninfo_unexecuted_blocks=1 --ignore-errors mismatch --ignore-errors unused
-    COMMAND ${LCOV_TOOL} -e json.info ${SRC_FILES} --output-file json.info.filtered --rc branch_coverage=1 --ignore-errors unused
+    COMMAND ${LCOV_TOOL} --directory . --capture --output-file json.info --rc branch_coverage=1 --rc geninfo_unexecuted_blocks=1 --ignore-errors mismatch --ignore-errors unused --ignore-errors inconsistent
+    COMMAND ${LCOV_TOOL} -e json.info ${SRC_FILES} --output-file json.info.filtered --rc branch_coverage=1 --ignore-errors unused --ignore-errors inconsistent
     COMMAND ${CMAKE_SOURCE_DIR}/tests/thirdparty/imapdl/filterbr.py json.info.filtered > json.info.filtered.noexcept
-    COMMAND ${GENHTML_TOOL} --title "JSON for Modern C++" --legend --demangle-cpp --output-directory html --show-details --branch-coverage json.info.filtered.noexcept
+    COMMAND ${GENHTML_TOOL} --title "JSON for Modern C++" --legend --demangle-cpp --output-directory html --show-details --branch-coverage --ignore-errors inconsistent json.info.filtered.noexcept
 
     COMMENT "Compile and test with coverage"
 )

cmake/requirements/requirements-cppcheck.txt

@@ -1 +1 @@
-cppcheck==1.5.2
+cppcheck==1.5.3

docs/mkdocs/docs/api/basic_json/accept.md

@@ -132,6 +132,10 @@ outside of a string, invalid) byte; see the [FAQ entry](../../home/faq.md#nul-by
 - [operator>>](../operator_gtgt.md) - deserialize from stream
 - [`JSON_STRICT_NUL_HANDLING`](../macros/json_strict_nul_handling.md) - opt in to rejecting a NUL byte in the input
   instead of treating it as end of input
+- [`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md) - read a `std::istream` one character
+  at a time
+- [`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md) - read a `FILE*` with the locking stdio
+  functions
 
 ## Version history
 
@@ -143,6 +147,10 @@ outside of a string, invalid) byte; see the [FAQ entry](../../home/faq.md#nul-by
 - Extended overload (2) to accept heterogeneous iterator+sentinel pairs (C++20 ranges support) in version 3.13.0.
 - `JSON_STRICT_NUL_HANDLING` added in version 3.13.0 to optionally reject a NUL byte in the input instead of treating
   it as end of input; planned to become the default in version 4.0.0.
+- Parsing from an `std::istream` scans the characters in its stream buffer in bulk in version 3.13.0
+  ([`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md)).
+- Parsing from a `FILE*` locks the file once instead of once per character in version 3.13.0
+  ([`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md)).
 
 !!! warning "Deprecation"
 

docs/mkdocs/docs/api/basic_json/parse.md

@@ -112,6 +112,16 @@ By default, a `'\0'` (NUL) byte anywhere in the input is treated as end of input
 outside of a string, invalid) byte; see the [FAQ entry](../../home/faq.md#nul-bytes-in-the-input) for details and the
 [`JSON_STRICT_NUL_HANDLING`](../macros/json_strict_nul_handling.md) macro to opt into rejecting it instead.
 
+!!! tip "Performance"
+
+    Contiguous inputs (strings, string views, vectors, arrays, pointer ranges) are the fastest to parse. A
+    `#!cpp std::istream` comes close: the parser scans the characters its stream buffer holds in bulk, without reading
+    ahead, so the stream is left exactly where it would be after reading one character at a time. A `#!cpp FILE*` is
+    read one character at a time, but the file is locked only once, so other threads that use the same `FILE` object
+    wait until parsing is done. See [input sources and performance](../../features/parsing/index.md#input-sources-and-performance),
+    [`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md), and
+    [`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md).
+
 ## Examples
 
 ??? example "Example: (1) parse from a character array"
@@ -247,6 +257,10 @@ outside of a string, invalid) byte; see the [FAQ entry](../../home/faq.md#nul-by
 - [operator>>](../operator_gtgt.md) - deserialize from stream
 - [`JSON_STRICT_NUL_HANDLING`](../macros/json_strict_nul_handling.md) - opt in to rejecting a NUL byte in the input
   instead of treating it as end of input
+- [`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md) - read a `std::istream` one character
+  at a time
+- [`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md) - read a `FILE*` with the locking stdio
+  functions
 
 ## Version history
 
@@ -261,6 +275,11 @@ outside of a string, invalid) byte; see the [FAQ entry](../../home/faq.md#nul-by
   it as end of input; planned to become the default in version 4.0.0.
 - Extended empty-input detection to also cover an `std::istream` without a stream buffer, and fixed a crash
   (`std::terminate`) when parsing from an `std::istream` with `eofbit` in its exception mask, in version 3.13.0.
+- Parsing from an `std::istream` scans the characters in its stream buffer in bulk in version 3.13.0
+  ([`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md) restores reading one character at a
+  time).
+- Parsing from a `FILE*` locks the file once instead of once per character in version 3.13.0
+  ([`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md) restores `std::fgetc`).
 
 !!! warning "Deprecation"
 

docs/mkdocs/docs/api/basic_json/sax_parse.md

@@ -116,6 +116,10 @@ consumer `sax` has a super-linear complexity.
 
 A UTF-8 byte order mark is silently ignored.
 
+When reading from a `#!c FILE*`, the file is locked while it is parsed. The lock is recursive, so the SAX handler may
+use the same file with stdio functions, but it must not wait for another thread that uses the same `FILE` object; see
+[`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md).
+
 ## Examples
 
 ??? example
@@ -137,6 +141,10 @@ A UTF-8 byte order mark is silently ignored.
 
 - [parse](parse.md) - deserialize from a compatible input
 - [accept](accept.md) - check if the input is valid JSON
+- [`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md) - read a `std::istream` one character
+  at a time
+- [`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md) - read a `FILE*` with the locking stdio
+  functions
 
 ## Version history
 
@@ -148,6 +156,10 @@ A UTF-8 byte order mark is silently ignored.
 - Extended overload (2) to accept heterogeneous iterator+sentinel pairs (C++20 ranges support) in version 3.13.0.
 - `JSON_PRECISE_STREAM_POSITION` added in version 3.13.0 to optionally leave a `#!cpp std::istream` positioned right
   after the parsed value when `strict` is `#!cpp false`.
+- Parsing from an `std::istream` scans the characters in its stream buffer in bulk in version 3.13.0
+  ([`JSON_DISABLE_STREAM_BULK_SCAN`](../macros/json_disable_stream_bulk_scan.md)).
+- Parsing from a `FILE*` locks the file once instead of once per character in version 3.13.0
+  ([`JSON_DISABLE_UNLOCKED_FILE_IO`](../macros/json_disable_unlocked_file_io.md)).
 
 !!! warning "Deprecation"
 

docs/mkdocs/docs/api/basic_json/to_bjdata.md

@@ -123,4 +123,6 @@ Linear in the size of the JSON value `j`.
   that is not valid UTF-8 unchanged, as before; `strict` (the default if
   [`JSON_STRICT_BINARY_UTF8`](../macros/json_strict_binary_utf8.md) is enabled) throws `type_error.316`.
 - Throws `type_error.321` for a discarded value since version 3.13.0; previously, a discarded value nested in an
-  array or object was silently skipped, producing invalid BJData.
+  array or object was silently skipped, producing invalid BJData.
+- Writes unsigned integers wider than 64 bits as high-precision numbers since version 3.13.0; previously, they were
+  silently truncated to 64 bits.

docs/mkdocs/docs/api/basic_json/to_bon8.md

@@ -37,8 +37,9 @@ With (2), the bytes written before the exception remain in the output adapter.
 
 ## Exceptions
 
-- Throws [out_of_range.407](../../home/exceptions.md#jsonexceptionout_of_range407) if `j` contains an unsigned integer
-  above 9223372036854775807, which BON8 cannot represent
+- Throws [out_of_range.407](../../home/exceptions.md#jsonexceptionout_of_range407) if `j` contains an integer outside
+  the range of int64 (an unsigned integer above 9223372036854775807, or, with a number type wider than 64 bits, any
+  integer beyond int64), which BON8 cannot represent
 - Throws [type_error.316](../../home/exceptions.md#jsonexceptiontype_error316) if `j` contains a string that is not
   valid UTF-8
 - Throws [type_error.321](../../home/exceptions.md#jsonexceptiontype_error321) if `j` or a value nested in it is

docs/mkdocs/docs/api/basic_json/to_bson.md

@@ -47,6 +47,10 @@ Strong guarantee: if an exception is thrown, there are no changes in the JSON va
 
 - Throws [`type_error.317`](../../home/exceptions.md#jsonexceptiontype_error317) if the top-level type of the JSON value
   is not an object; example: `"to serialize to BSON, top-level type must be object, but is string"`
+- Throws [`out_of_range.407`](../../home/exceptions.md#jsonexceptionout_of_range407) if `j` contains a signed integer
+  outside the range of int64 or an unsigned integer outside the range of uint64, which is only possible with a number
+  type wider than 64 bits; example:
+  `"integer number 9223372036854775808 cannot be represented by BSON as it does not fit int64"`
 - Throws [`out_of_range.409`](../../home/exceptions.md#jsonexceptionout_of_range409) if a key in the JSON object contains
   a null byte (code point U+0000); example: `"BSON key cannot contain code point U+0000 (at byte 2)"`
 - Throws [`out_of_range.412`](../../home/exceptions.md#jsonexceptionout_of_range412) if the length of a document, array,
@@ -119,3 +123,5 @@ pass before anything is written.
   that is not valid UTF-8 unchanged, as before; `strict` (the default if
   [`JSON_STRICT_BINARY_UTF8`](../macros/json_strict_binary_utf8.md) is enabled) throws `type_error.316` before anything
   is written.
+- Throws `out_of_range.407` for integers that do not fit 64 bits since version 3.13.0; previously, integers of a
+  number type wider than 64 bits were silently truncated.

docs/mkdocs/docs/api/basic_json/to_cbor.md

@@ -46,6 +46,9 @@ Strong guarantee: if an exception is thrown, there are no changes in the JSON va
 
 ## Exceptions
 
+- Throws [`out_of_range.407`](../../home/exceptions.md#jsonexceptionout_of_range407) if `j` contains an integer
+  outside [-2^64, 2^64-1], which is only possible with a number type wider than 64 bits; example:
+  `"integer number 18446744073709551616 cannot be represented by CBOR as it does not fit [-2^64, 2^64-1]"`
 - Throws [type_error.316](../../home/exceptions.md#jsonexceptiontype_error316) if a string or object key in `j` is
   not valid UTF-8 and `error_handler` is `strict` (the default only if
   [`JSON_STRICT_BINARY_UTF8`](../macros/json_strict_binary_utf8.md) is enabled)
@@ -90,3 +93,5 @@ Linear in the size of the JSON value `j`.
   [`JSON_STRICT_BINARY_UTF8`](../macros/json_strict_binary_utf8.md) is enabled) throws `type_error.316`.
 - Throws `type_error.321` for a discarded value since version 3.13.0; previously, a discarded value nested in an
   array or object was silently skipped, producing invalid CBOR.
+- Throws `out_of_range.407` for integers that do not fit 64 bits since version 3.13.0; previously, integers of a
+  number type wider than 64 bits were silently truncated.

docs/mkdocs/docs/api/basic_json/to_msgpack.md

@@ -46,6 +46,9 @@ Strong guarantee: if an exception is thrown, there are no changes in the JSON va
 
 ## Exceptions
 
+- Throws [`out_of_range.407`](../../home/exceptions.md#jsonexceptionout_of_range407) if `j` contains an integer
+  outside [-2^63, 2^64-1], which is only possible with a number type wider than 64 bits; example:
+  `"integer number 18446744073709551616 cannot be represented by MessagePack as it does not fit [-2^63, 2^64-1]"`
 - Throws [`out_of_range.412`](../../home/exceptions.md#jsonexceptionout_of_range412) if the length of a string, binary
   value, array, or object exceeds 4294967295, the maximum MessagePack can store; example:
   `"MessagePack length 4294967296 exceeds maximum of 4294967295"`
@@ -112,3 +115,5 @@ Linear in the size of the JSON value `j`.
   `number_unsigned_t`.
 - Throws `type_error.321` for a discarded value since version 3.13.0; previously, a discarded value nested in an
   array or object was silently skipped, producing invalid MessagePack.
+- Throws `out_of_range.407` for integers that do not fit 64 bits since version 3.13.0; previously, integers of a
+  number type wider than 64 bits were silently truncated.

docs/mkdocs/docs/api/basic_json_document/erase.md

@@ -18,10 +18,10 @@ has `erase`; calling it on a read-only `basic_json_document` fails to compile (`
 1. Removes every member of `object` whose key is `key` (see [Notes](#notes) on duplicate keys) and returns how many
    were removed; `#!cpp 0` if `object` has no member with this key.
 2. Removes the element at index `idx` of `array`, which must already exist (`#!cpp idx < array.size()`).
-3. Removes the value the JSON pointer `ptr` refers to, relative to [`root()`](root.md), and returns how many values
-   were removed: the *parent* of the target must already exist, and the target itself is removed as in 1. (an object
-   member; `#!cpp 0` or more) or 2. (an array element; always `#!cpp 1`). `ptr` must not be empty -- [`root()`](root.md)
-   itself cannot be erased.
+3. Removes the value the JSON pointer `ptr` refers to, relative to [`root()`](root.md). `ptr` must not be empty
+   ([`root()`](root.md) itself cannot be erased), and the *parent* of the target must already exist. The target is
+   removed as in 1. if the parent is an object (returns the number of removed members, `#!cpp 0` or more) and as in 2.
+   if it is an array (always returns `#!cpp 1`).
 
 ## Template parameters
 
@@ -69,10 +69,10 @@ has `erase`; calling it on a read-only `basic_json_document` fails to compile (`
    [`out_of_range.404`](../../home/exceptions.md#jsonexceptionout_of_range404) (an empty token); otherwise (an
    object, or a primitive value the pointer's parent resolves to) throws what 1. throws.
 
-Every overload also throws [`invalid_iterator.202`](../../home/exceptions.md#jsonexceptioninvalid_iterator202) ("view
-does not belong to this document") if `object`/`array` is a [discarded](../basic_json_view/is_discarded.md) view or a
-view of a *different* document (overloads 1-2 only; overload 3 always starts from this document's own
-[`root()`](root.md)).
+Overloads 1. and 2. also throw [`invalid_iterator.202`](../../home/exceptions.md#jsonexceptioninvalid_iterator202)
+("view does not belong to this document") if `object`/`array` is a [discarded](../basic_json_view/is_discarded.md)
+view or a view of a *different* document. Overload 3. never throws `invalid_iterator.202`: it takes no view, and it
+always starts from this document's own [`root()`](root.md).
 
 ## Complexity
 
@@ -92,9 +92,10 @@ view of a *different* document (overloads 1-2 only; overload 3 always starts fro
 
 Like [`set`](set.md) and [`push_back`](push_back.md), `erase` never moves an element's *value*: a view still
 referring to a removed member or element keeps showing what it last held (see [Edits](index.md#edits)) -- it just no
-longer appears when `array`/`object` is read, dumped, or iterated. Removing an element of `array` (2.) does shift the
-*links* to the elements after it, the same way `insert`, `set`, or `push_back` on the same array would; any iterator
-already taken over `array`/`object` is invalidated by an erase, since it was walking the old layout.
+longer appears when `array`/`object` is read, dumped, or iterated. What does change is the container's sequence of
+*links* to its elements: removing an element of `array` (2.) shifts the links to the elements after it by one, as
+[`insert`](insert.md) does in the other direction. *Every* iterator already taken over `array`/`object` is invalidated
+by an erase, not only those at or after the removed element, since the container may now use a different sequence.
 
 ## Examples
 

docs/mkdocs/docs/api/basic_json_document/load.md

@@ -110,17 +110,30 @@ enum class image_check
 
 How thoroughly `load()` validates `image` before trusting it.
 
-| value    | checks                                                                                                    | guarantees |
-|----------|--------------------------------------------------------------------------------------------------------------|------------|
-| `full`   | everything the parser itself guarantees: structure and bounds; that every string is valid UTF-8 (and, for a string still in the source text, that it contains no quote, backslash, or control character); and that every number token is well-formed and matches the value stored for it; strings and float tokens may share a range only if the ranges are identical (as nodes that share a value do), and never overlap otherwise | reading and serializing a checked image is safe and always produces valid JSON, exactly as for a parsed document |
-| `bounds` | structure and bounds only -- that every offset and count in the node index stays inside the image            | reading and serializing stay memory-safe, but a crafted image can hold strings that are not valid UTF-8 or that serialize to invalid JSON ([`dump()`](../basic_json_view/dump.md) writes them unchanged or throws [`type_error.316`](../../home/exceptions.md#jsonexceptiontype_error316)), and numbers whose values differ from their text |
-| `none`   | nothing                                                                                                       | images from a trusted source only -- reading a damaged image is undefined behavior |
+`full` (the default)
+:   Checks everything the parser itself guarantees: structure and bounds; that every string is valid UTF-8 (and, for a
+    string still in the source text, that it contains no quote, backslash, or control character); and that every number
+    token is well-formed and matches the value stored for it. Strings and float tokens may share a range only if the
+    ranges are identical (as nodes that share a value do), and never overlap otherwise. Reading and serializing a
+    checked image is safe and always produces valid JSON, exactly as for a parsed document. This is the right choice for
+    an image from anything you do not fully control -- a file, a cache shared with other processes, a peer on the
+    network.
 
-`full` is the default and the right choice for an image from anything you do not fully control -- a file, a cache
-shared with other processes, a peer on the network. `bounds` skips scanning the text and the decoded strings, so it
-fits a cache your own process just wrote and reads straight back, where damage would mean a bug or a hardware fault
-rather than adversarial input; it still cannot crash or read out of bounds. `none` skips validation entirely and
-should only be used for an image you trust as much as your own memory.
+`bounds`
+:   Checks structure and bounds only: that every offset and count in the node index stays inside the image. So
+    `load()`, [`root()`](root.md), element access, and [`materialize()`](../basic_json_view/materialize.md) can never
+    read outside the image, even on a damaged one. It does *not* guarantee that the image describes valid JSON: a
+    string that a `full` check would have rejected can make [`dump()`](../basic_json_view/dump.md) write invalid UTF-8
+    or invalid JSON, or throw [`type_error.316`](../../home/exceptions.md#jsonexceptiontype_error316), and a number can
+    read back with a value that does not match how it is spelled. An [editable document](../json_editable_document.md)
+    does not take such a string over: [`set`](set.md), [`push_back`](push_back.md) and [`insert`](insert.md) throw
+    `type_error.316` when they copy it from a view of the loaded document, and leave the editable document unchanged.
+    `bounds` skips scanning the text and the decoded strings, so it fits a cache your own process just wrote and reads
+    straight back, where damage would mean a bug or a hardware fault rather than adversarial input.
+
+`none`
+:   Checks nothing. Reading a damaged image is undefined behavior, so use it only for an image you trust as much as
+    your own memory.
 
 **Lifetime.** Overloads 1 and 2 borrow `image`: it must stay alive and byte-for-byte unchanged for as long as the
 returned document, and any [view](../basic_json_view/index.md) taken from it, is used -- exactly like a document
@@ -134,16 +147,6 @@ returned document, and any [view](../basic_json_view/index.md) taken from it, is
     ("unknown format"). Use images to cache a document within one build of the library, or to hand one to another
     process running the *same* build on the *same* (little-endian) machine -- not as a long-term storage format.
 
-**What `image_check::bounds` does not guarantee.** A bounds-checked image can never make `load()`,
-[`root()`](root.md), element access, or [`materialize()`](../basic_json_view/materialize.md) read outside the image,
-so those stay safe on a damaged one. It does *not* guarantee that the image describes valid JSON: a string
-that a `full` check would have rejected can make [`dump()`](../basic_json_view/dump.md) write invalid UTF-8 or invalid
-JSON, or throw `type_error.316`, and a number can read back with a value that does not match how it is spelled.
-Reserve `bounds` for images you already trust to be well-formed, and use it only to skip the extra scan.
-An [editable document](../json_editable_document.md) does not take such a string over either:
-[`set`](set.md), [`push_back`](push_back.md) and [`insert`](insert.md) throw `type_error.316` when they copy it from
-a view of the loaded document, and leave the editable document unchanged.
-
 ## Examples
 
 ??? example "Caching a document, ownership, and a rejected image"

docs/mkdocs/docs/api/basic_json_view/begin.md

@@ -6,7 +6,7 @@ iterator begin() const noexcept;
 
 Returns an iterator to the first element of an array, or the first member value of an object, in **document order**
 -- the order the values appear in the source text, not sorted by key. A primitive value iterates as a range of one
-element (itself); `#!json null` and a [discarded](is_discarded.md) view iterate as an empty range.
+element (itself), and a `#!json null` value or a [discarded](is_discarded.md) view iterates as an empty range.
 
 ## Return value
 

docs/mkdocs/docs/api/basic_json_view/end.md

@@ -5,9 +5,8 @@ iterator end() const noexcept;
 ```
 
 Returns an iterator to one past the last element of an array, one past the last member value of an object, in
-**document order** -- see [`begin()`](begin.md). A primitive value iterates as a range of one element (itself);
-`#!json null` and a [discarded](is_discarded.md) view iterate as an empty range, so `#!cpp begin() == end()` for
-them.
+**document order** -- see [`begin()`](begin.md). A primitive value iterates as a range of one element (itself), and a
+`#!json null` value or a [discarded](is_discarded.md) view iterates as an empty range (`#!cpp begin() == end()`).
 
 ## Return value
 

docs/mkdocs/docs/api/basic_json_view/index.md

@@ -22,15 +22,15 @@ Moving the document itself does not invalidate its views: the index is heap-allo
 
 `basic_json_view` provides the read-only part of the `BasicJsonType` interface:
 
-- the type-inspection functions,
-- element access,
-- lookup,
-- iteration,
+- the type-inspection functions
+- element access
+- lookup
+- iteration
 - conversion: [`get<T>()`](get.md), [`get_string()`](get_string.md), [`number_token()`](number_token.md), and
-  [`materialize()`](materialize.md) to build the `BasicJsonType` value of a subtree on demand, and
+  [`materialize()`](materialize.md) to build the `BasicJsonType` value of a subtree on demand
 - comparison: [`operator==`](operator_eq.md) and [`operator!=`](operator_ne.md) compare two views, or a view and a
   `BasicJsonType` value, without ever building a `BasicJsonType` value for a view; no ordering comparison
-  (`#!cpp operator<`) is provided.
+  (`#!cpp operator<`) is provided
 
 [`operator[]`](operator%5B%5D.md), [`at`](at.md), [`contains`](contains.md), and [`value`](value.md) also accept a
 [`json_pointer`](../json_pointer/index.md). Unlike `basic_json`, there are no implicit conversion operators (no

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

@@ -16,6 +16,10 @@ header. See also the [macro overview page](../../features/macros.md).
 
 ## Parsing
 
+- [**JSON_DISABLE_STREAM_BULK_SCAN**](json_disable_stream_bulk_scan.md) - read a `std::istream` one character at a time
+  instead of scanning its stream buffer in bulk
+- [**JSON_DISABLE_UNLOCKED_FILE_IO**](json_disable_unlocked_file_io.md) - read a `FILE*` with `std::fgetc` instead of
+  locking it once and using the unlocked stdio functions
 - [**JSON_PRECISE_STREAM_POSITION**](json_precise_stream_position.md) - opt in to leaving an input stream positioned
   right after a parsed number
 - [**JSON_STRICT_BINARY_UTF8**](json_strict_binary_utf8.md) - opt in to checking strings for valid UTF-8 in the CBOR,
@@ -34,6 +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_disable_stream_bulk_scan.md

@@ -0,0 +1,144 @@
+# JSON_DISABLE_STREAM_BULK_SCAN
+
+```cpp
+#define JSON_DISABLE_STREAM_BULK_SCAN /* value */
+```
+
+When defined to `1`, parsing from a `#!cpp std::istream` reads the input one character at a time through the stream
+buffer, as in version 3.12.0 and earlier, instead of scanning the characters that the stream buffer already holds in
+bulk.
+
+The macro is an escape hatch for stream buffers that do not follow the usual contract of `std::basic_streambuf` (see
+the requirements on the stream buffer in the notes below). It does not change any result: values, exceptions, and the
+position of the stream after parsing are the same either way. Only the speed differs.
+
+## Default definition
+
+The default value is `0`: streams are scanned in bulk.
+
+```cpp
+#define JSON_DISABLE_STREAM_BULK_SCAN 0
+```
+
+## Notes
+
+!!! note "How streams are scanned"
+
+    A `std::streambuf` keeps the part of the input it has read but not yet handed out in its *get area*, the range
+    from `gptr()` to `egptr()`. A `std::filebuf`, for instance, reads the file in blocks of a few kilobytes into its get
+    area, and a `std::stringbuf` exposes its whole string as one get area.
+
+    When parsing from a `std::istream`, the parser scans this get area directly, with the same fast paths that it uses
+    for a `std::string`: runs of ordinary string characters, numbers, and indentation are recognized several bytes at a
+    time instead of being requested from the stream buffer one character at a time. When the parser reaches the end of
+    the get area, it asks the stream buffer for the next character as before, which makes the stream buffer refill its
+    get area.
+
+    The parser only reads characters that it would also have read one at a time, and it advances the stream buffer
+    exactly over the characters it consumes. Nothing is read ahead, buffered elsewhere, or put back. Therefore, the
+    stream is left exactly where it was left before:
+
+    - [`operator>>`](../operator_gtgt.md) and [`sax_parse`](../basic_json/sax_parse.md) with `strict = false` can still
+      read several values from one stream, and the stream can still be used with other `std::istream` functions after a
+      value has been read.
+    - With [`JSON_PRECISE_STREAM_POSITION`](json_precise_stream_position.md), the stream is still positioned right after
+      a parsed number.
+    - Exception messages, including the line, the column, and the last read token, are unchanged.
+
+!!! note "Performance"
+
+    Parsing from a `std::ifstream` or `std::istringstream` is typically 1.5 to 2 times as fast as before, and reading
+    many small values from one stream with `operator>>` 1.4 to 4 times as fast. Parsing from a stream is then within
+    about 10 to 20 % of first reading the whole file into a `std::string` and parsing that.
+
+    A stream buffer without a get area (one that only overrides `underflow()` and `uflow()` and hands out every
+    character from there) offers nothing to scan in bulk, so it is read one character at a time as before.
+
+!!! note "Requirements on the stream buffer"
+
+    The bulk scan relies on the following behavior, which every stream buffer of the standard library and every stream
+    buffer that keeps its input in its get area has:
+
+    - When `sgetc()` returns a character while `gptr() < egptr()`, that character is `*gptr()`; when `sbumpc()` returns
+      one, it was at `gptr()` before the call.
+    - Characters in the get area are not changed while they are in the get area.
+    - `underflow()` or `uflow()` either sets up a get area that holds the returned character at `gptr()` (or at
+      `gptr() - 1` after `uflow()`), or leaves the get area empty.
+
+    A stream buffer that returns characters from `uflow()` while leaving unrelated characters in its get area, or that
+    changes the characters in its get area in place, does not meet these requirements. Define the macro to `1` for
+    such stream buffers.
+
+!!! note "ABI compatibility"
+
+    The value of this macro is encoded in the [namespace](../../features/namespace.md) (tag `_dsbs`), 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.
+
+!!! note "Other inputs"
+
+    Strings, string views, vectors, arrays, and pointer or iterator ranges over contiguous memory are scanned in bulk
+    anyway. Reading from a `#!cpp FILE*` is not affected by this macro; see
+    [`JSON_DISABLE_UNLOCKED_FILE_IO`](json_disable_unlocked_file_io.md) for that input.
+
+## Examples
+
+??? example "Example: reading values from a stream (default)"
+
+    The stream is scanned in bulk, and the values are read one after the other:
+
+    ```cpp
+    #include <iostream>
+    #include <sstream>
+    #include <nlohmann/json.hpp>
+
+    using json = nlohmann::json;
+
+    int main()
+    {
+        std::istringstream input(R"({"id": 1} {"id": 2} 42 rest)");
+        json j1, j2, j3;
+        input >> j1 >> j2 >> j3;  // {"id":1}, {"id":2}, 42
+
+        std::string word;
+        input >> word;            // "rest"
+    }
+    ```
+
+??? example "Example: reading one character at a time (macro defined to 1)"
+
+    With the macro, the same program reads the stream one character at a time, with the same results:
+
+    ```cpp
+    #define JSON_DISABLE_STREAM_BULK_SCAN 1
+    #include <iostream>
+    #include <sstream>
+    #include <nlohmann/json.hpp>
+
+    using json = nlohmann::json;
+
+    int main()
+    {
+        std::istringstream input(R"({"id": 1} {"id": 2} 42 rest)");
+        json j1, j2, j3;
+        input >> j1 >> j2 >> j3;  // {"id":1}, {"id":2}, 42
+
+        std::string word;
+        input >> word;            // "rest"
+    }
+    ```
+
+## See also
+
+- [**parse**](../basic_json/parse.md) - deserialize from a compatible input
+- [**operator>>**](../operator_gtgt.md) - deserialize from stream
+- [**sax_parse**](../basic_json/sax_parse.md) - generate SAX events
+- [**JSON_PRECISE_STREAM_POSITION**](json_precise_stream_position.md) - leave an input stream positioned right after a
+  parsed number
+- [**JSON_DISABLE_UNLOCKED_FILE_IO**](json_disable_unlocked_file_io.md) - read a `FILE*` with the locking stdio
+  functions
+- [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_disable_unlocked_file_io.md

@@ -0,0 +1,115 @@
+# JSON_DISABLE_UNLOCKED_FILE_IO
+
+```cpp
+#define JSON_DISABLE_UNLOCKED_FILE_IO /* value */
+```
+
+When defined to `1`, parsing from a `#!cpp FILE*` reads every character with `std::fgetc`, as in version 3.12.0 and
+earlier, instead of locking the file once and reading it with the unlocked stdio functions.
+
+It does not change any result: values, exceptions, and the position of the file after parsing are the same either way.
+Only the speed and the locking of the file differ (see the notes on locking below).
+
+Builds with MemorySanitizer use `std::fgetc` automatically, because the inline `getc_unlocked` of glibc reads the stdio
+buffer directly, which glibc (not instrumented) filled, and MemorySanitizer would report it as uninitialized.
+
+## Default definition
+
+The default value is `0`: the file is locked once and read with the unlocked stdio functions where they are available.
+
+```cpp
+#define JSON_DISABLE_UNLOCKED_FILE_IO 0
+```
+
+## Notes
+
+!!! note "Background"
+
+    The C standard library makes every `std::fgetc` call thread-safe by locking and unlocking the `FILE` object. When
+    parsing from a `FILE*`, the parser reads one character at a time, so this happens for every character of the input.
+
+    Instead, the parser locks the file once when parsing starts and unlocks it when parsing ends - also when an
+    exception is thrown - and reads the characters in between without locking:
+
+    | C library                                                  | functions                                          |
+    |------------------------------------------------------------|----------------------------------------------------|
+    | glibc (Linux), Apple (macOS, iOS), Android, FreeBSD, NetBSD, OpenBSD, DragonFly BSD | `flockfile`, `getc_unlocked`, `funlockfile` (POSIX) |
+    | Microsoft (MSVC, clang-cl)                                 | `_lock_file`, `_fgetc_nolock`, `_unlock_file`       |
+    | others (e.g., musl, MinGW, newlib)                         | `std::fgetc`, as before                             |
+
+    This makes parsing from a `FILE*` 3 to 4 times as fast with the Apple C library and 1.1 to 1.3 times as fast with
+    glibc, whose locking is cheaper. The binary formats ([`from_cbor`](../basic_json/from_cbor.md),
+    [`from_msgpack`](../basic_json/from_msgpack.md), ...) read from a `FILE*` the same way.
+
+!!! note "Locking"
+
+    While a value is parsed from a `FILE*`, the parsing thread holds the file's lock. The lock is recursive, so the
+    parsing thread itself - for instance, a [SAX](../json_sax/index.md) handler or a
+    [parser callback](../basic_json/parser_callback_t.md) - can keep using the file with stdio functions such as
+    `std::ftell`. Other threads that use the same `FILE` object wait until the parser is done. Before, they could
+    interleave their reads with the parser's, which made the parsed input unpredictable anyway.
+
+    A program in which a SAX handler or parser callback waits for **another** thread that uses the **same** `FILE`
+    object would deadlock. Define the macro to `1` for such programs.
+
+!!! note "ABI compatibility"
+
+    The value of this macro is encoded in the [namespace](../../features/namespace.md) (tag `_dufio`), 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.
+
+!!! note "Other inputs"
+
+    Reading from a `#!cpp std::istream` is not affected by this macro; see
+    [`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md) for that input.
+
+## Examples
+
+??? example "Example: parsing from a `FILE*` (default)"
+
+    The file is locked once while it is parsed:
+
+    ```cpp
+    #include <cstdio>
+    #include <nlohmann/json.hpp>
+
+    using json = nlohmann::json;
+
+    int main()
+    {
+        std::FILE* f = std::fopen("example.json", "rb");
+        json j = json::parse(f);  // f is locked until parse() returns
+        std::fclose(f);
+    }
+    ```
+
+??? example "Example: locking for every character (macro defined to 1)"
+
+    With the macro, every character is read with `std::fgetc`, with the same result:
+
+    ```cpp
+    #define JSON_DISABLE_UNLOCKED_FILE_IO 1
+    #include <cstdio>
+    #include <nlohmann/json.hpp>
+
+    using json = nlohmann::json;
+
+    int main()
+    {
+        std::FILE* f = std::fopen("example.json", "rb");
+        json j = json::parse(f);  // std::fgetc for every character
+        std::fclose(f);
+    }
+    ```
+
+## See also
+
+- [**parse**](../basic_json/parse.md) - deserialize from a compatible input
+- [**sax_parse**](../basic_json/sax_parse.md) - generate SAX events
+- [**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_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_precise_stream_position.md

@@ -72,6 +72,12 @@ The default value is `0` (disabled — existing behavior is preserved).
     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.
 
+!!! note "Bulk scanning"
+
+    The parser scans the characters in the stream buffer in bulk (see
+    [`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md)), but it never reads ahead. The stream is left
+    at the same position with and without bulk scanning, in both modes of this macro.
+
 !!! tip "Workaround without the macro"
 
     Separate the values in the stream with whitespace. The character consumed after a number is then the separator,

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

@@ -6,14 +6,16 @@
 
 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
 message (simdutf is only used to fast-path *valid* runs; anything it flags falls back to the scalar path so the exact
-diagnostic is preserved). Streaming inputs (files, `std::istream`, wide strings, user-defined adapters) always use the
-scalar path.
+diagnostic is preserved). A `std::istream` uses simdutf for the characters its stream buffer holds (see
+[`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md)); other streaming inputs (`FILE*`, wide strings,
+user-defined adapters) always use the scalar path.
 
 When `JSON_USE_SIMDUTF` is defined you must make the `simdutf.h` header available on the include path and link the
 simdutf library. When it is not defined, no simdutf header is included and there is no dependency.

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`, `_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/api/operator_gtgt.md

@@ -121,6 +121,7 @@ being read.
 - [`JSON_STRICT_NUL_HANDLING`](macros/json_strict_nul_handling.md) - opt in to rejecting a NUL byte in the input
   instead of treating it as end of input
 - [`JSON_PRECISE_STREAM_POSITION`](macros/json_precise_stream_position.md) - opt in to leaving the stream positioned right after a number
+- [`JSON_DISABLE_STREAM_BULK_SCAN`](macros/json_disable_stream_bulk_scan.md) - read the stream one character at a time
 
 ## Version history
 
@@ -133,3 +134,6 @@ being read.
   crash (`std::terminate`) when `i` has `eofbit` in its exception mask, in version 3.13.0.
 - Changed to the strong exception safety guarantee in version 3.13.0: `j` is no longer left with a partially parsed
   value if parsing throws.
+- Scans the characters in the stream buffer of `i` in bulk in version 3.13.0; the position of `i` after the call is
+  unchanged ([`JSON_DISABLE_STREAM_BULK_SCAN`](macros/json_disable_stream_bulk_scan.md) restores reading one character
+  at a time).

docs/mkdocs/docs/community/quality_assurance.md

@@ -135,6 +135,9 @@ Note: Some modern features (like C++20 ranges or filesystem support) may be disa
   headers are included.
 - [x] On Windows, the library is compiled with `<Windows.h>` being included to detect and avoid common bugs (see
   [`unit-windows_h.cpp`](https://github.com/nlohmann/json/blob/develop/tests/src/unit-windows_h.cpp)).
+- [x] The library is compiled with Qt's keyword macros (`slots`, `signals`, `emit`, `foreach`, `forever`) being defined
+  to make sure it does not use them as identifiers (see
+  [`unit-qt_keywords.cpp`](https://github.com/nlohmann/json/blob/develop/tests/src/unit-qt_keywords.cpp)).
 - [x] The library is compiled with exceptions disabled to support alternative means of error handling.
 
 ## Stable public API

docs/mkdocs/docs/features/index.md

@@ -14,8 +14,8 @@ C++ types, and finally serialize it again.
   [parsing untrusted input](parsing/untrusted_input.md).
 - [Zero-copy JSON views](json_view.md) — read a JSON text through a flat index instead of building a `json` tree;
   strings and numbers stay in the input and are only decoded when needed.
-  [Editable documents](json_view.md#editing-a-document) can also be modified, and [images](json_view.md#images) load a
-  parsed document again without parsing it.
+- [Editable documents](json_view.md#editing-a-document) — modify a parsed document without building a `json` tree.
+- [Images](json_view.md#images) — save a parsed document and load it again without parsing it.
 - [Comments](comments.md) and [trailing commas](trailing_commas.md) — opt-in relaxations of the JSON grammar.
 
 ## Accessing and modifying values

docs/mkdocs/docs/features/macros.md

@@ -133,6 +133,14 @@ When defined, default parse and serialize functions for enums are excluded and h
 
 See [full documentation of `JSON_DISABLE_ENUM_SERIALIZATION`](../api/macros/json_disable_enum_serialization.md).
 
+## `JSON_DISABLE_STREAM_BULK_SCAN`
+
+When defined to `1`, parsing from a `#!cpp std::istream` reads one character at a time instead of scanning the
+characters that the stream buffer holds in bulk. Results are the same either way; the macro is an escape hatch for
+stream buffers that do not follow the usual `std::streambuf` contract. The default value is `0`.
+
+See [full documentation of `JSON_DISABLE_STREAM_BULK_SCAN`](../api/macros/json_disable_stream_bulk_scan.md).
+
 ## `JSON_DISABLE_TUPLE_REFERENCE_CONVERSION`
 
 When defined to `1`, a JSON value can no longer be created from a one-element `std::tuple` holding a reference to a JSON
@@ -141,6 +149,14 @@ is planned to become the default in version 4.0.0.
 
 See [full documentation of `JSON_DISABLE_TUPLE_REFERENCE_CONVERSION`](../api/macros/json_disable_tuple_reference_conversion.md).
 
+## `JSON_DISABLE_UNLOCKED_FILE_IO`
+
+When defined to `1`, parsing from a `#!cpp FILE*` reads every character with `std::fgetc`, which locks the file each
+time, instead of locking the file once and reading it with the unlocked stdio functions. Results are the same either
+way. The default value is `0`.
+
+See [full documentation of `JSON_DISABLE_UNLOCKED_FILE_IO`](../api/macros/json_disable_unlocked_file_io.md).
+
 ## `JSON_NO_AUTOMATIC_UDLS`
 
 When defined, `<nlohmann/json.hpp>` does not include `<nlohmann/json_literals.hpp>` with the user-defined string literals
@@ -158,6 +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

@@ -23,6 +23,11 @@ The complete default namespace name is derived as follows:
     - [`JSON_STRICT_BINARY_UTF8`](../api/macros/json_strict_binary_utf8.md) defined non-zero appends `_sbu8`.
     - [`JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS`](../api/macros/json_use_objects_for_enum_keyed_maps.md) defined non-zero
       appends `_ekmo`.
+    - [`JSON_DISABLE_STREAM_BULK_SCAN`](../api/macros/json_disable_stream_bulk_scan.md) defined non-zero appends
+      `_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
@@ -108,6 +113,9 @@ follows:
   ([`JSON_PRECISE_STREAM_POSITION`](../api/macros/json_precise_stream_position.md)), `_snul`
   ([`JSON_STRICT_NUL_HANDLING`](../api/macros/json_strict_nul_handling.md)), `_sbu8`
   ([`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)), `_vns`
+  ([`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)), `_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

@@ -55,6 +55,34 @@ whitespace, because `operator>>` consumes the character that terminates a number
 [`JSON_PRECISE_STREAM_POSITION`](../../api/macros/json_precise_stream_position.md) is defined to `1` — see the
 [`operator>>` notes](../../api/operator_gtgt.md#notes) for details and examples.
 
+## Input sources and performance
+
+All inputs produce the same values and the same exceptions, but they are read in different ways, which makes some of
+them faster than others:
+
+| input                                                                                      | how it is read                                                                     |
+|--------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------|
+| strings, string views, string literals, contiguous containers, pointer or iterator ranges over contiguous memory | the whole input is scanned in bulk: runs of string characters, numbers, and indentation are recognized several bytes at a time |
+| `#!cpp std::istream` (`#!cpp std::ifstream`, `#!cpp std::istringstream`, ...)            | the characters the stream buffer holds are scanned in bulk like a string; the stream buffer is asked for more when they are used up |
+| `#!cpp FILE*`                                                                              | one character at a time, with the file locked once for the whole parse              |
+| other iterator ranges (e.g., `#!cpp std::list<char>`, `#!cpp std::istreambuf_iterator<char>`), wide strings | one character at a time                                                             |
+
+Parsing a file from a `#!cpp std::ifstream` is therefore only about 10 to 20 % slower than reading the whole file into a
+`#!cpp std::string` first and parsing that, and it needs no memory for a copy of the file. The stream is never read
+ahead: after parsing, it is positioned exactly as if every character had been read one at a time, so
+[`operator>>`](../../api/operator_gtgt.md) can still read several values from one stream (see also
+[JSON Lines](json_lines.md)).
+
+Two macros switch these 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 `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) |
@@ -194,8 +194,8 @@ packet-beta
 
 | Bytes | Field   | Type       | Contents                                                                                                                      |
 |-------|---------|------------|-------------------------------------------------------------------------------------------------------------------------------|
-| 0     | `kind`  | `uint8_t`  | the type, numbered as [`value_t`](../api/basic_json/value_t.md): 0 null, 1 object, 2 array, 3 string, 4 boolean, 5 signed integer, 6 unsigned integer, 7 float; 10 for a link (see below) |
-| 1     | `flags` | `uint8_t`  | bits 0-1: where a string's bytes (or a number's token) are (0: the source text, 1: the buffer of decoded strings, for strings with escapes, 2: the edit buffer); bit 2: the value of a boolean; bits 3 and 4: moved and new (see below) |
+| 0     | `kind`  | `uint8_t`  | the type, numbered as [`value_t`](../api/basic_json/value_t.md): 0 null, 1 object, 2 array, 3 string, 4 boolean, 5 signed integer, 6 unsigned integer, 7 float; 10 for a link (see below). There is no node for a discarded value: a [discarded](../api/basic_json_view/is_discarded.md) view has none |
+| 1     | `flags` | `uint8_t`  | bits 0-1: where a string's bytes (or a number's token) are (0: the source text, 1: the buffer of decoded strings, for strings with escapes, 2: the edit buffer); bit 2: the value of a boolean; bit 3: moved, bit 4: new, bit 5: linked (see below) |
 | 2-3   | `extra` | `uint16_t` | numbers: the number of integer digits (low byte) and fraction digits (high byte), 255 for more; objects: the number of their key hash index (1-based), or 0; otherwise 0 |
 | 4-7   | `off`   | `uint32_t` | where the value starts: the first byte after a string's opening quote (or its position in the buffer of decoded strings), the first byte of a number or literal, the bracket of an array or object |
 | 8-11  | `len`   | `uint32_t` | strings: the length after decoding; floats and literals: the length of the token; arrays and objects: the number of elements |
@@ -257,6 +257,8 @@ edited:
   sequence (a header node, then the entries), whose number is in `off`. The entries are links (`kind` 10), whose bytes
   8-15 hold the address of the value's node, so values never move.
 - A node written by an edit gets the flag *new* (bit 4): it has no position in the source text.
+- A value that an entry of such a sequence links to gets the flag *linked* (bit 5): its extent in the parsed layout
+  (`next`) no longer matters, so a scalar can replace an array or object there in place.
 - Views of read-only documents compile without any of this: how views walk the index is a template parameter
   (`navigation<Editable>`).
 
@@ -337,6 +339,11 @@ The `serializer` (used by [`dump`](../api/basic_json/dump.md) and [`operator<<`]
 - `output_stream_adapter` writes to a `std::ostream`.
 - `output_string_adapter` appends to a string.
 
+The `binary_writer` does not call an adapter for every byte. The `to_*` functions that return a `std::vector` write
+into the vector through a sink (`output_vector_sink`) that is not virtual. The functions that take an output adapter
+write through a sink (`output_adapter_sink`) that collects the output in a buffer of one kibibyte and passes it to the
+adapter in blocks, so that the adapter is called (and a stream locked) once per block instead of once per value.
+
 ## Value conversion
 
 Values are converted from and to other types with the `JSONSerializer` template parameter. The default,

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/docs/home/exceptions.md

@@ -972,9 +972,19 @@ double-precision number when `number_float_t` is `#!cpp float`.
 
 ### json.exception.out_of_range.407
 
-An unsigned integer number greater than 9223372036854775807 cannot be represented in the binary format. This exception
-is only thrown by [`to_bon8`](../api/basic_json/to_bon8.md), as [BON8](../features/binary_formats/bon8.md) has no
-unsigned 64-bit integer type.
+An integer number cannot be represented by the binary format it is serialized to:
+
+- [BON8](../features/binary_formats/bon8.md) only stores integers that fit into int64, so
+  [`to_bon8`](../api/basic_json/to_bon8.md) throws for unsigned integers greater than 9223372036854775807.
+- [CBOR](../features/binary_formats/cbor.md), [MessagePack](../features/binary_formats/messagepack.md), and
+  [BSON](../features/binary_formats/bson.md) store integers in at most 64 bits. With the default number types, every
+  integer fits, but a [`number_integer_t`](../api/basic_json/number_integer_t.md) or
+  [`number_unsigned_t`](../api/basic_json/number_unsigned_t.md) wider than 64 bits (e.g., `__int128`) can hold values
+  outside the range of the format: [-2^64, 2^64-1] for CBOR, [-2^63, 2^64-1] for MessagePack, and the range of int64
+  (signed integers) or uint64 (unsigned integers) for BSON.
+
+[UBJSON](../features/binary_formats/ubjson.md) and [BJData](../features/binary_formats/bjdata.md) never throw this
+exception, because they serialize integers beyond 64 bits as high-precision numbers.
 
 !!! note "Former occurrences"
 
@@ -983,11 +993,17 @@ unsigned 64-bit integer type.
     - Since version 3.9.0, integer numbers beyond int64 are serialized as high-precision UBJSON numbers.
     - Since version 3.12.0, integer numbers beyond int64 are serialized as uint64 BSON numbers.
 
-!!! failure "Example message"
+    Before version 3.13.0, CBOR, MessagePack, and BSON silently truncated integers wider than 64 bits, and BJData
+    truncated unsigned integers wider than 64 bits.
+
+!!! failure "Example messages"
 
     ```
     integer number 18446744073709551615 cannot be represented by BON8 as it does not fit int64
     ```
+    ```
+    integer number 1267650600228229401496703205376 cannot be represented by CBOR as it does not fit [-2^64, 2^64-1]
+    ```
 
 ### json.exception.out_of_range.408
 

docs/mkdocs/mkdocs.yml

@@ -370,7 +370,9 @@ nav:
           - 'JSON_DELETE_DEPRECATED_FUNCTIONS': api/macros/json_delete_deprecated_functions.md
           - 'JSON_DIAGNOSTIC_POSITIONS': api/macros/json_diagnostic_positions.md
           - 'JSON_DISABLE_ENUM_SERIALIZATION': api/macros/json_disable_enum_serialization.md
+          - 'JSON_DISABLE_STREAM_BULK_SCAN': api/macros/json_disable_stream_bulk_scan.md
           - 'JSON_DISABLE_TUPLE_REFERENCE_CONVERSION': api/macros/json_disable_tuple_reference_conversion.md
+          - 'JSON_DISABLE_UNLOCKED_FILE_IO': api/macros/json_disable_unlocked_file_io.md
           - 'JSON_HAS_CPP_11, JSON_HAS_CPP_14, JSON_HAS_CPP_17, JSON_HAS_CPP_20, JSON_HAS_CPP_23, JSON_HAS_CPP_26': api/macros/json_has_cpp_11.md
           - 'JSON_HAS_EXPERIMENTAL_FILESYSTEM, JSON_HAS_FILESYSTEM': api/macros/json_has_filesystem.md
           - 'JSON_HAS_RANGE_VIEW_CONVERSION': api/macros/json_has_range_view_conversion.md
@@ -381,6 +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

docs/mkdocs/scripts/mermaid/package-lock.json

@@ -6,7 +6,7 @@
     "": {
       "name": "check-mermaid",
       "dependencies": {
-        "jsdom": "30.1.2",
+        "jsdom": "30.1.3",
         "mermaid": "11.17.2"
       }
     },
@@ -40,9 +40,9 @@
       }
     },
     "node_modules/@asamuzakjp/dom-selector": {
-      "version": "9.2.4",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-9.2.4.tgz",
-      "integrity": "sha512-YnVzxDxLaqL0t7Q4wfcgHZjg55Wm6y5EaO7Bhu3q9J7wq/8iw3PhsnWiOQhYekx4VexnbkgWpTT6n+QIV7kczw==",
+      "version": "10.0.3",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-10.0.3.tgz",
+      "integrity": "sha512-3+777rEPsOOjoy5/sRmIzHnoRDaLpm3XWY2dmWn/N7f/XwDG0bcpnFaKL3X9FYJUbNwLmpcJC1bUYexJRxsFlw==",
       "license": "MIT",
       "dependencies": {
         "bidi-js": "^1.1.0",
@@ -1105,9 +1105,9 @@
       }
     },
     "node_modules/dompurify": {
-      "version": "3.4.16",
-      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.16.tgz",
-      "integrity": "sha512-sqo+pNp3qRhCIpbgRi1y8Tgk27Bo2Ry7w0dC1NBeNTdZChWjz9Xb/KOoZbRP/R6pQZ80Qw8YhXw13hWWBbMRnQ==",
+      "version": "3.4.17",
+      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.17.tgz",
+      "integrity": "sha512-B/pSlYVHQ58kqISD/yDjuJ1yfzI34m1gF4lAod2bwSRhiN4K5Cl3H8Af7hqB2PmTL911tdEudnnXn5FnhRoz0w==",
       "license": "(MPL-2.0 OR Apache-2.0)",
       "optionalDependencies": {
         "@types/trusted-types": "^2.0.7"
@@ -1202,13 +1202,13 @@
       "license": "MIT"
     },
     "node_modules/jsdom": {
-      "version": "30.1.2",
-      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.2.tgz",
-      "integrity": "sha512-0FFE/jE1rppmVfUrJUgxqXjcwolZYIGtAgj1pTussMhNZxIxi7W/PvfWaMNroGi2B36X1IKHbexpZ9DhkoOPiQ==",
+      "version": "30.1.3",
+      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.3.tgz",
+      "integrity": "sha512-hQMUPvRgN7KxsfIytDQji7bimuJeRQa6Qo5+iuiZQk5CafeRxLIA10/cG5YCBFNpD9RO3oqIiN4CO1Y44kxqXg==",
       "license": "MIT",
       "dependencies": {
         "@asamuzakjp/css-color": "^7.1.3",
-        "@asamuzakjp/dom-selector": "^9.2.4",
+        "@asamuzakjp/dom-selector": "^10.0.3",
         "@bramus/specificity": "^2.4.2",
         "@csstools/css-syntax-patches-for-csstree": "^1.1.15",
         "@exodus/bytes": "^1.16.0",
@@ -1221,7 +1221,7 @@
         "parse5": "^8.0.1",
         "saxes": "^6.0.0",
         "tough-cookie": "^6.0.2",
-        "undici": "^8.11.2",
+        "undici": "^8.12.1",
         "w3c-xmlserializer": "^6.0.0",
         "webidl-conversions": "^8.0.1",
         "whatwg-mimetype": "^5.0.0",
@@ -1241,28 +1241,29 @@
       }
     },
     "node_modules/katex": {
-      "version": "0.16.47",
-      "resolved": "https://registry.npmjs.org/katex/-/katex-0.16.47.tgz",
-      "integrity": "sha512-Eeo8Ys1doU1z+x8AZsPpQu+p/QcZBI5PeOo7QGQdy2x2m0MU/hYagBbGOmXwr5KVbEfVuWv9LpnQWeehogurjg==",
+      "version": "0.18.11",
+      "resolved": "https://registry.npmjs.org/katex/-/katex-0.18.11.tgz",
+      "integrity": "sha512-yvyz/2VMKqoho/35Hat5w37wIEdWJ5ElVMidTAlWrxr5uW4aBGsEpveTmKh18/zw08v/7lGswCPon4OhqmqrTA==",
+      "deprecated": "Accidentally published with breaking changes. Use 0.19.0 instead.",
       "funding": [
         "https://opencollective.com/katex",
         "https://github.com/sponsors/katex"
       ],
       "license": "MIT",
       "dependencies": {
-        "commander": "^8.3.0"
+        "commander": "^15.0.0"
       },
       "bin": {
         "katex": "cli.js"
       }
     },
     "node_modules/katex/node_modules/commander": {
-      "version": "8.3.0",
-      "resolved": "https://registry.npmjs.org/commander/-/commander-8.3.0.tgz",
-      "integrity": "sha512-OkTL9umf+He2DZkUq8f8J9of7yL6RJKI24dVITBmNfZBmri9zYZQrKkuXiKhyfPSu8tUhnVBB1iKXevvnlR4Ww==",
+      "version": "15.0.0",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-15.0.0.tgz",
+      "integrity": "sha512-z67u4ZhzCL/Tydu1lJARtEZYWbWaN7oYLHbsuzocr6y4N6WZAagG3RQ4FW61V1/0+jImpj293XfrcYnd1qxtPg==",
       "license": "MIT",
       "engines": {
-        "node": ">= 12"
+        "node": ">=22.12.0"
       }
     },
     "node_modules/khroma": {
@@ -1521,9 +1522,9 @@
       }
     },
     "node_modules/undici": {
-      "version": "8.11.2",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-8.11.2.tgz",
-      "integrity": "sha512-u4UB2/IrKdU6lFxumHmmo1a3fCQO5tzQllRorfoRS63txhrB7xTpSn1PftwC4qEHkOaqP95fCWW4lJzwErwzhQ==",
+      "version": "8.12.2",
+      "resolved": "https://registry.npmjs.org/undici/-/undici-8.12.2.tgz",
+      "integrity": "sha512-XtuDWVdqTLr3NAxm2ly9/hU6kUsJOgGh1RASzMBbsw6aWCOaQzHdW+QQDfdERjfJ2V5vL5pRPa4RetO9UfivDg==",
       "license": "MIT",
       "engines": {
         "node": ">=22.19.0"

docs/mkdocs/scripts/mermaid/package.json

@@ -4,7 +4,10 @@
   "description": "Validate the Mermaid diagrams of the documentation (see check_mermaid.mjs)",
   "type": "module",
   "dependencies": {
-    "jsdom": "30.1.2",
+    "jsdom": "30.1.3",
     "mermaid": "11.17.2"
+  },
+  "overrides": {
+    "katex": "0.18.11"
   }
 }

include/nlohmann/detail/abi_macros.hpp

@@ -54,6 +54,14 @@
     #define JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS 0
 #endif
 
+#ifndef JSON_DISABLE_STREAM_BULK_SCAN
+    #define JSON_DISABLE_STREAM_BULK_SCAN 0
+#endif
+
+#ifndef JSON_DISABLE_UNLOCKED_FILE_IO
+    #define JSON_DISABLE_UNLOCKED_FILE_IO 0
+#endif
+
 #if JSON_DIAGNOSTICS
     #define NLOHMANN_JSON_ABI_TAG_DIAGNOSTICS _diag
 #else
@@ -102,6 +110,24 @@
     #define NLOHMANN_JSON_ABI_TAG_OBJECTS_FOR_ENUM_KEYED_MAPS
 #endif
 
+#if JSON_DISABLE_STREAM_BULK_SCAN
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN _dsbs
+#else
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN
+#endif
+
+#if JSON_DISABLE_UNLOCKED_FILE_IO
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO _dufio
+#else
+    #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
@@ -119,9 +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) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j) \
-    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j)
+#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(                                   \
@@ -133,6 +159,9 @@
             NLOHMANN_JSON_ABI_TAG_STRICT_NUL_HANDLING,               \
             NLOHMANN_JSON_ABI_TAG_STRICT_BINARY_UTF8,                \
             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)
@@ -1342,14 +1327,32 @@ inline char* write_shortest(char* first, const zmij::shortest_decimal d) noexcep
     return end + (three ? 5 : 4);
 }
 
-/// the powers of ten up to 10^16
-inline const std::array<std::uint64_t, 17>& powers_of_ten_16() noexcept
+/// the powers of ten up to 10^19 (all that fit in std::uint64_t)
+inline const std::array<std::uint64_t, 20>& powers_of_ten() noexcept
 {
-    static const std::array<std::uint64_t, 17> powers =
+    static const std::array<std::uint64_t, 20> powers =
     {
         {
-            1u, 10u, 100u, 1000u, 10000u, 100000u, 1000000u, 10000000u, 100000000u, 1000000000u, 10000000000u,
-            100000000000u, 1000000000000u, 10000000000000u, 100000000000000u, 1000000000000000u, 10000000000000000u
+            1u,
+            10u,
+            100u,
+            1000u,
+            10000u,
+            100000u,
+            1000000u,
+            10000000u,
+            100000000u,
+            1000000000u,
+            10000000000u,
+            100000000000u,
+            1000000000000u,
+            10000000000000u,
+            100000000000000u,
+            1000000000000000u,
+            10000000000000000u,
+            100000000000000000u,
+            1000000000000000000u,
+            10000000000000000000u
         }
     };
     return powers;
@@ -1367,9 +1370,9 @@ JSON_HEDLEY_NON_NULL(1)
 JSON_HEDLEY_RETURNS_NON_NULL
 inline char* write_short_decimal(char* first, std::uint64_t digits, int count, int exp) noexcept
 {
-    JSON_ASSERT(digits >= powers_of_ten_16()[static_cast<std::size_t>(count - 1)] && count <= 15);
+    JSON_ASSERT(digits >= powers_of_ten()[static_cast<std::size_t>(count - 1)] && count <= 15);
     const int scale = 16 - count;
-    return write_shortest(first, zmij::shortest_decimal{digits * powers_of_ten_16()[static_cast<std::size_t>(scale)], exp - scale - 1, 0, false});
+    return write_shortest(first, zmij::shortest_decimal{digits * powers_of_ten()[static_cast<std::size_t>(scale)], exp - scale - 1, 0, false});
 }
 
 /// as write_short_decimal(), counting the digits (not 0, less than 10^15)
@@ -1380,7 +1383,7 @@ inline char* write_short_decimal(char* first, std::uint64_t digits, int exp) noe
     JSON_ASSERT(digits != 0 && digits < 1000000000000000u);
     // floor(log10(2^bits)) + 1 digits, or one less
     const int log2_bound = ((64 - count_leading_zeros(digits)) * 1233) >> 12;
-    const int count = log2_bound + (digits >= powers_of_ten_16()[static_cast<std::size_t>(log2_bound)] ? 1 : 0);
+    const int count = log2_bound + (digits >= powers_of_ten()[static_cast<std::size_t>(log2_bound)] ? 1 : 0);
     return write_short_decimal(first, digits, count, exp);
 }
 
@@ -1444,8 +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/input_adapters.hpp

@@ -8,16 +8,19 @@
 
 #pragma once
 
-#include <algorithm> // count, find, min
+#include <algorithm> // count, find, max, min
 #include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t
 #include <cstring> // strlen
 #include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
+#include <limits> // numeric_limits
+#include <memory> // unique_ptr
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
-#include <utility> // pair, declval
+#include <utility> // declval, move, pair
+#include <vector> // vector
 
 #ifndef JSON_NO_IO
     #include <cstdio>   // FILE *
@@ -30,6 +33,33 @@
 #include <nlohmann/detail/meta/type_traits.hpp>
 #include <nlohmann/detail/string_utils.hpp>
 
+#ifndef JSON_NO_IO
+    // Which unlocked stdio functions file_input_adapter uses (see
+    // JSON_DISABLE_UNLOCKED_FILE_IO): 1 = POSIX flockfile/getc_unlocked,
+    // 2 = Microsoft CRT _lock_file/_fgetc_nolock, 0 = none (std::fgetc, which
+    // locks the file for every character). Only C libraries known to declare
+    // the functions are listed; the macros are tested after <cstdio>, which
+    // defines __GLIBC__. MemorySanitizer builds always use std::fgetc: glibc's
+    // inline getc_unlocked reads the stdio buffer directly, which glibc (not
+    // instrumented) filled, so MemorySanitizer would report a false positive.
+    #if defined(__has_feature)
+        #if __has_feature(memory_sanitizer)
+            #define JSON_UNLOCKED_FILE_IO_MSAN 1
+        #endif
+    #endif
+
+    #if JSON_DISABLE_UNLOCKED_FILE_IO || defined(JSON_UNLOCKED_FILE_IO_MSAN)
+        #define JSON_UNLOCKED_FILE_IO 0
+    #elif defined(_MSC_VER)
+        #define JSON_UNLOCKED_FILE_IO 2
+    #elif defined(__GLIBC__) || defined(__APPLE__) || defined(__ANDROID__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__)
+        #define JSON_UNLOCKED_FILE_IO 1
+    #else
+        #define JSON_UNLOCKED_FILE_IO 0
+    #endif
+    #undef JSON_UNLOCKED_FILE_IO_MSAN
+#endif  // JSON_NO_IO
+
 NLOHMANN_JSON_NAMESPACE_BEGIN
 namespace detail
 {
@@ -43,6 +73,82 @@ enum class input_format_t { json, cbor, msgpack, ubjson, bson, bjdata, bon8 };
 
 #ifndef JSON_NO_IO
 /*!
+A growable array of characters for the part of a token that an input adapter
+copied out of its window: unlike std::vector<char>::push_back(), appending a
+character takes no reference to a local, and is a few instructions when there
+is room.
+*/
+class token_carry_buffer
+{
+  public:
+    token_carry_buffer() = default;
+    token_carry_buffer(const token_carry_buffer&) = delete;
+    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
+    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+        : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
+    {
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+    }
+    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
+    ~token_carry_buffer() = default;
+
+    void clear() noexcept
+    {
+        m_end = m_data.get();
+    }
+
+    void push_back(char c)
+    {
+        if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
+        {
+            grow(1);
+        }
+        *m_end++ = c;
+    }
+
+    void append(const char* first, const char* last)
+    {
+        const auto n = static_cast<std::size_t>(last - first);
+        const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
+        if (n > room)
+        {
+            grow(n);
+        }
+        if (n != 0)
+        {
+            std::memcpy(m_end, first, n);
+        }
+        m_end += n;
+    }
+
+    /// replace the contents of @a out
+    void copy_to(std::vector<char>& out) const
+    {
+        out.assign(m_data.get(), m_end);
+    }
+
+  private:
+    /// make room for @a extra more characters
+    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    {
+        const auto size = static_cast<std::size_t>(m_end - m_data.get());
+        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
+        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
+        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+        if (size != 0)
+        {
+            std::memcpy(data.get(), m_data.get(), size);
+        }
+        m_data = std::move(data);
+        m_end = m_data.get() + size;
+        m_capacity_end = m_data.get() + new_capacity;
+    }
+
+    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    char* m_end = nullptr;
+    char* m_capacity_end = nullptr;
+};
+/*!
 Input adapter for stdio file access. This adapter read only 1 byte and do not use any
  buffer. This adapter is a very low level adapter.
 */
@@ -56,18 +162,35 @@ class file_input_adapter
         : m_file(f)
     {
         JSON_ASSERT(m_file != nullptr);
+        lock();
     }
 
     // make class move-only
     file_input_adapter(const file_input_adapter&) = delete;
-    file_input_adapter(file_input_adapter&&) noexcept = default;
+    file_input_adapter(file_input_adapter&& rhs) noexcept
+        : m_file(rhs.m_file)
+    {
+        // the lock is owned by the thread, not the adapter: hand it over
+        rhs.m_file = nullptr;
+    }
     file_input_adapter& operator=(const file_input_adapter&) = delete;
     file_input_adapter& operator=(file_input_adapter&&) = delete;
-    ~file_input_adapter() = default;
+    ~file_input_adapter()
+    {
+        unlock();
+    }
 
     std::char_traits<char>::int_type get_character() noexcept
     {
+        // the file is locked for the lifetime of the adapter, so the
+        // per-character lock of std::fgetc can be skipped
+#if JSON_UNLOCKED_FILE_IO == 1
+        return getc_unlocked(m_file);
+#elif JSON_UNLOCKED_FILE_IO == 2
+        return _fgetc_nolock(m_file);
+#else
         return std::fgetc(m_file);
+#endif
     }
 
     // returns the number of characters successfully read
@@ -78,6 +201,32 @@ class file_input_adapter
     }
 
   private:
+    // Lock the file once for the whole parse instead of once per character.
+    // The lock is recursive, so fread() and the user's own stdio calls on the
+    // same thread keep working.
+    void lock() noexcept
+    {
+#if JSON_UNLOCKED_FILE_IO == 1
+        flockfile(m_file);
+#elif JSON_UNLOCKED_FILE_IO == 2
+        _lock_file(m_file);
+#endif
+    }
+
+    void unlock() noexcept
+    {
+        // a moved-from adapter has no file and holds no lock
+        if (m_file == nullptr)
+        {
+            return;
+        }
+#if JSON_UNLOCKED_FILE_IO == 1
+        funlockfile(m_file);
+#elif JSON_UNLOCKED_FILE_IO == 2
+        _unlock_file(m_file);
+#endif
+    }
+
     /// the file pointer to read from
     std::FILE* m_file;
 };
@@ -127,15 +276,27 @@ class input_stream_adapter
     input_stream_adapter& operator=(input_stream_adapter&) = delete;
     input_stream_adapter& operator=(input_stream_adapter&&) = delete;
 
-#if JSON_PRECISE_STREAM_POSITION
     input_stream_adapter(input_stream_adapter&& rhs) noexcept
-        : is(rhs.is), sb(rhs.sb), lookahead(rhs.lookahead)
+        : is(rhs.is), sb(rhs.sb)
+#if JSON_PRECISE_STREAM_POSITION
+        , lookahead(rhs.lookahead)
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        , token_carry(std::move(rhs.token_carry))
+        , token_segment(rhs.token_segment)
+        , current_in_get_area(rhs.current_in_get_area)
+        , tracking_token(rhs.tracking_token)
+        , no_get_area(rhs.no_get_area)
+#endif
     {
         rhs.is = nullptr;
         rhs.sb = nullptr;
+#if JSON_PRECISE_STREAM_POSITION
         rhs.lookahead = false;
+#endif
     }
 
+#if JSON_PRECISE_STREAM_POSITION
     // Whether the character last returned by get_character() can be given back
     // to the input with release_lookahead().
     static constexpr bool supports_lookahead = true;
@@ -155,17 +316,41 @@ class input_stream_adapter
             sb->sbumpc();
         }
 
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // sgetc() refills the get area if it is exhausted
+        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
+        if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            save_token_segment();
+        }
+#endif
+
         auto res = sb->sgetc();
         // set eof manually, as we don't use the istream interface.
         if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
         {
             // there is nothing to step over next time
             lookahead = false;
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            current_in_get_area = false;
+#endif
             is->clear(is->rdstate() | std::ios::eofbit);
         }
         else
         {
             lookahead = true;
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            if (JSON_HEDLEY_UNLIKELY(refill))
+            {
+                // the peeked character is at gptr(), unless the streambuf has
+                // no get area
+                after_refill(res, streambuf_access::get_ptr(sb) != streambuf_access::get_end(sb));
+            }
+            else
+            {
+                current_in_get_area = true;
+            }
+#endif
         }
         return res;
     }
@@ -179,13 +364,6 @@ class input_stream_adapter
         lookahead = false;
     }
 #else
-    input_stream_adapter(input_stream_adapter&& rhs) noexcept
-        : is(rhs.is), sb(rhs.sb)
-    {
-        rhs.is = nullptr;
-        rhs.sb = nullptr;
-    }
-
     // std::istream/std::streambuf use std::char_traits<char>::to_int_type, to
     // ensure that std::char_traits<char>::eof() and the character 0xFF do not
     // end up as the same value, e.g., 0xFFFFFFFF.
@@ -194,13 +372,21 @@ class input_stream_adapter
     // stays consumed after parsing; see JSON_PRECISE_STREAM_POSITION.
     std::char_traits<char>::int_type get_character()
     {
-        auto res = sb->sbumpc();
-        // set eof manually, as we don't use the istream interface.
-        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // the character is in the get area: step over it, as sbumpc() does
+        const char* const next = streambuf_access::get_ptr(sb);
+        if (JSON_HEDLEY_LIKELY(next != streambuf_access::get_end(sb)))
         {
-            is->clear(is->rdstate() | std::ios::eofbit);
+            current_in_get_area = true;
+            streambuf_access::bump(sb, 1);
+            return std::char_traits<char>::to_int_type(*next);
         }
-        return res;
+        if (no_get_area)
+        {
+            return get_character_no_get_area();
+        }
+#endif
+        return get_character_slow();
     }
 #endif
 
@@ -209,6 +395,18 @@ class input_stream_adapter
     {
 #if JSON_PRECISE_STREAM_POSITION
         commit_lookahead();
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // the elements are in the get area: copy them, as sgetn() does, but
+        // without the virtual call
+        const char* const first = streambuf_access::get_ptr(sb);
+        const auto available = static_cast<std::size_t>(streambuf_access::get_end(sb) - first);
+        if (JSON_HEDLEY_LIKELY(count * sizeof(T) <= available && count * sizeof(T) <= static_cast<std::size_t>((std::numeric_limits<int>::max)())))
+        {
+            std::memcpy(dest, first, count * sizeof(T));
+            streambuf_access::bump(sb, static_cast<int>(count * sizeof(T)));
+            return count * sizeof(T);
+        }
 #endif
         auto res = static_cast<std::size_t>(sb->sgetn(reinterpret_cast<char*>(dest), static_cast<std::streamsize>(count * sizeof(T))));
         if (JSON_HEDLEY_UNLIKELY(res < count * sizeof(T)))
@@ -218,7 +416,248 @@ class input_stream_adapter
         return res;
     }
 
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    // The lexer may scan the characters in the get area of the streambuf
+    // directly (SWAR fast paths). The get area is only a window into the
+    // input: it ends wherever the streambuf's buffer ends, and is refilled by
+    // the next get_character(). The lexer therefore keeps counting the
+    // position itself and treats the end of the window as "unknown", not as
+    // the end of the input.
+    static constexpr bool supports_bulk_scan = true;
+    static constexpr bool bulk_window = true;
+
+    // Pointer to the next unread character; only valid when
+    // bulk_remaining() > 0. The character before it is the one last returned
+    // by get_character().
+    const char* bulk_data() const
+    {
+        return current_end();
+    }
+
+    // Number of unread characters in the get area.
+    std::size_t bulk_remaining() const
+    {
+        if (!current_in_get_area)
+        {
+            return 0;
+        }
+        return static_cast<std::size_t>(streambuf_access::get_end(sb) - current_end());
+    }
+
+    // Consume @a n characters previously inspected via bulk_data(); the last
+    // of them becomes the character last returned by get_character().
+    void bulk_skip(std::size_t n)
+    {
+        JSON_ASSERT(n <= bulk_remaining());
+        // gbump() takes an int, but a get area (e.g., of a std::stringbuf) can
+        // be larger
+        constexpr auto max_bump = static_cast<std::size_t>((std::numeric_limits<int>::max)());
+        for (; n > max_bump; n -= max_bump)
+        {
+            streambuf_access::bump(sb, static_cast<int>(max_bump));
+        }
+        streambuf_access::bump(sb, static_cast<int>(n));
+    }
+
+    // The adapter keeps the current token for the lexer's error messages: the
+    // part in the get area is only remembered as a pointer, and copied out
+    // when the get area is about to be refilled.
+    static constexpr bool supports_token_tracking = true;
+
+    // Start a token with the characters read from now on. Until this is
+    // called, nothing is tracked (the binary readers never call it).
+    void begin_token() noexcept
+    {
+        tracking_token = true;
+        token_carry.clear();
+        token_segment = current_end();
+    }
+
+    // Start a new token with the character @a c last returned by
+    // get_character().
+    void mark_token_start(std::char_traits<char>::int_type c)
+    {
+        tracking_token = true;
+        token_carry.clear();
+        if (current_in_get_area)
+        {
+            token_segment = current_end() - 1;
+        }
+        else
+        {
+            token_segment = nullptr;
+            token_carry.push_back(std::char_traits<char>::to_char_type(c));
+        }
+    }
+
+    // Copy the characters read since mark_token_start() to @a out, without the
+    // last one if @a drop_last.
+    void copy_token(std::vector<char>& out, bool drop_last) const
+    {
+        token_carry.copy_to(out);
+        if (token_segment != nullptr)
+        {
+            out.insert(out.end(), token_segment, current_end());
+        }
+        if (drop_last && !out.empty())
+        {
+            out.pop_back();
+        }
+    }
+#endif
+
   private:
+#if !JSON_PRECISE_STREAM_POSITION
+    // the get area is exhausted, or there is none
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    JSON_HEDLEY_NEVER_INLINE
+#endif
+    std::char_traits<char>::int_type get_character_slow()
+    {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // sbumpc() refills the get area if it is exhausted
+        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
+        if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            save_token_segment();
+        }
+#endif
+
+        auto res = sb->sbumpc();
+        // set eof manually, as we don't use the istream interface.
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
+        {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            current_in_get_area = false;
+#endif
+            is->clear(is->rdstate() | std::ios::eofbit);
+        }
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        else if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            // the consumed character is at gptr() - 1, unless the streambuf
+            // has no get area and returned it from uflow()
+            const char* const next = streambuf_access::get_ptr(sb);
+            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
+        }
+        else
+        {
+            current_in_get_area = true;
+        }
+#endif
+        return res;
+    }
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    // Reads the protected get-area pointers of any std::streambuf. Naming the
+    // members through a derived class makes forming the member pointers
+    // accessible; calling them on another streambuf object is then ordinary
+    // member access through a pointer to member.
+    struct streambuf_access : std::streambuf
+    {
+        static const char* get_ptr(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::gptr))();
+        }
+        static const char* get_end(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::egptr))();
+        }
+        static const char* get_begin(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::eback))();
+        }
+        static void bump(std::streambuf* s, int n)
+        {
+            (s->*(&streambuf_access::gbump))(n);
+        }
+        // sbumpc() for an exhausted get area
+        static std::streambuf::int_type next_character(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::uflow))();
+        }
+    };
+
+    // one past the character last returned by get_character(), if that is in
+    // the get area
+    const char* current_end() const
+    {
+#if JSON_PRECISE_STREAM_POSITION
+        // the character is peeked at gptr() and not stepped over yet, unless
+        // it was given back with release_lookahead()
+        return streambuf_access::get_ptr(sb) + (lookahead ? 1 : 0);
+#else
+        return streambuf_access::get_ptr(sb);
+#endif
+    }
+
+    // the get area is about to be replaced: copy the token's part in it
+    void save_token_segment()
+    {
+        if (token_segment != nullptr)
+        {
+            token_carry.append(token_segment, streambuf_access::get_end(sb));
+            token_segment = nullptr;
+        }
+    }
+
+#if !JSON_PRECISE_STREAM_POSITION
+    // The streambuf has no get area (e.g., std::cin synchronized with stdio):
+    // every character comes from uflow(), and the token is kept by value.
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area()
+    {
+        // the get area is exhausted, so this is what sbumpc() does
+        const auto res = streambuf_access::next_character(sb);
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof() || streambuf_access::get_end(sb) != nullptr))
+        {
+            return get_character_no_get_area_other(res);
+        }
+        // current_in_get_area is false: only the first character of a get
+        // area, or the end of the input, is read here
+        if (tracking_token)
+        {
+            token_carry.push_back(std::char_traits<char>::to_char_type(res));
+        }
+        return res;
+    }
+
+    // @a res came from uflow() of a streambuf without a get area: it is the
+    // end of the input, or the streambuf has a get area now
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area_other(std::char_traits<char>::int_type res)
+    {
+        if (res == std::char_traits<char>::eof())
+        {
+            current_in_get_area = false;
+            is->clear(is->rdstate() | std::ios::eofbit);
+            return res;
+        }
+        // the consumed character is at gptr() - 1
+        after_refill(res, streambuf_access::get_ptr(sb) != streambuf_access::get_begin(sb));
+        return res;
+    }
+#endif
+
+    // character @a c was just read into a new get area (@a in_get_area) or
+    // returned without one
+    void after_refill(std::char_traits<char>::int_type c, bool in_get_area)
+    {
+        current_in_get_area = in_get_area;
+        no_get_area = !in_get_area;
+        if (!tracking_token)
+        {
+            return;
+        }
+        if (in_get_area)
+        {
+            token_segment = current_end() - 1;
+        }
+        else
+        {
+            token_carry.push_back(std::char_traits<char>::to_char_type(c));
+        }
+    }
+#endif
+
 #if JSON_PRECISE_STREAM_POSITION
     // Step over the character last returned by get_character(). The character
     // has already been peeked successfully, so for every streambuf with a get
@@ -240,6 +679,19 @@ class input_stream_adapter
     /// whether get_character() peeked a character that is not consumed yet
     bool lookahead = false;
 #endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    /// the part of the current token that was in earlier get areas
+    token_carry_buffer token_carry{}; // NOLINT(readability-redundant-member-init)
+    /// start of the current token's part in the get area, or nullptr
+    const char* token_segment = nullptr;
+    /// whether the character last returned by get_character() is in the get area
+    bool current_in_get_area = false;
+    /// whether a token is tracked at all; the binary readers never start one
+    bool tracking_token = false;
+    /// whether the streambuf had no get area when it was last asked for a
+    /// character
+    bool no_get_area = false;
+#endif
 };
 #endif  // JSON_NO_IO
 

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
@@ -166,6 +167,43 @@ constexpr bool input_adapter_supports_bulk_scan(std::false_type /*detected*/)
     return false;
 }
 
+// Detect whether an input adapter keeps the current token itself and can
+// hand it out on demand (see input_stream_adapter::supports_token_tracking),
+// so the lexer need not copy every scanned character into token_string.
+template<typename InputAdapterType>
+using detect_supports_token_tracking = decltype(InputAdapterType::supports_token_tracking);
+
+template<typename InputAdapterType>
+constexpr bool input_adapter_supports_token_tracking(std::true_type /*detected*/)
+{
+    return InputAdapterType::supports_token_tracking;
+}
+
+template<typename InputAdapterType>
+constexpr bool input_adapter_supports_token_tracking(std::false_type /*detected*/)
+{
+    return false;
+}
+
+// Detect whether the contiguous block of a bulk-scan adapter is only a window
+// into the input (see input_stream_adapter::bulk_window): its end is not the
+// end of the input, so a token that touches it must be finished character by
+// character.
+template<typename InputAdapterType>
+using detect_bulk_window = decltype(InputAdapterType::bulk_window);
+
+template<typename InputAdapterType>
+constexpr bool input_adapter_bulk_window(std::true_type /*detected*/)
+{
+    return InputAdapterType::bulk_window;
+}
+
+template<typename InputAdapterType>
+constexpr bool input_adapter_bulk_window(std::false_type /*detected*/)
+{
+    return false;
+}
+
 /*!
 @brief lexical analysis
 
@@ -184,9 +222,20 @@ class lexer : public lexer_base<BasicJsonType>
     /// whether the last read token can be reconstructed from the input adapter
     /// on demand (in error paths) instead of being copied on every scanned
     /// character; see input_adapter_supports_seek
-    static constexpr bool lazy_token_string =
+    static constexpr bool seekable =
         input_adapter_supports_seek<InputAdapterType>(is_detected<detect_supports_seek, InputAdapterType> {});
 
+    /// whether the adapter keeps the token itself (mark_token_start() /
+    /// copy_token()); see input_adapter_supports_token_tracking
+    static constexpr bool adapter_token =
+        !seekable && input_adapter_supports_token_tracking<InputAdapterType>(is_detected<detect_supports_token_tracking, InputAdapterType> {});
+
+    /// whether the last read token is obtained from the input adapter on
+    /// demand (in error paths) instead of being copied on every scanned
+    /// character: seekable adapters reconstruct it from the consumed input,
+    /// token-tracking adapters keep it themselves
+    static constexpr bool lazy_token_string = seekable || adapter_token;
+
     /// whether the read position is derived from the input adapter on demand
     /// instead of being counted on every scanned character. Seekable adapters
     /// know how much input they consumed and can revisit it, so
@@ -194,7 +243,7 @@ class lexer : public lexer_base<BasicJsonType>
     /// and line and column are recomputed from the consumed input only when
     /// get_position() is called (errors). Counting per character kept two
     /// increments and a newline check in the hot path of every get().
-    static constexpr bool lazy_position = lazy_token_string;
+    static constexpr bool lazy_position = seekable;
 
     /// whether a simulated unget can be passed on to the input adapter, which
     /// then leaves the character in the input; see
@@ -203,14 +252,17 @@ class lexer : public lexer_base<BasicJsonType>
         input_adapter_supports_lookahead<InputAdapterType>(is_detected<detect_supports_lookahead, InputAdapterType> {});
 
     /// whether string scanning may bulk-consume runs of ordinary characters
-    /// directly from a contiguous input buffer (SWAR fast path). This requires
-    /// the token to be reconstructible lazily (lazy_token_string), so bypassing
-    /// the per-character capture in get() cannot lose error diagnostics. It
-    /// also implies lazy_position, so the bulk paths need not advance the
-    /// position counters.
+    /// directly from a contiguous input buffer (SWAR fast path). Without
+    /// lazy_token_string / lazy_position, the bulk paths copy the consumed
+    /// characters into token_string and advance the position counters
+    /// themselves (see bulk_consume).
     static constexpr bool bulk_scan =
-        lazy_token_string
-        && input_adapter_supports_bulk_scan<InputAdapterType>(is_detected<detect_supports_bulk_scan, InputAdapterType> {});
+        input_adapter_supports_bulk_scan<InputAdapterType>(is_detected<detect_supports_bulk_scan, InputAdapterType> {});
+
+    /// whether the contiguous block is only a window into the input, so that
+    /// its end does not end a token; see input_adapter_bulk_window
+    static constexpr bool bulk_window =
+        input_adapter_bulk_window<InputAdapterType>(is_detected<detect_bulk_window, InputAdapterType> {});
 
   public:
     using token_type = typename lexer_base<BasicJsonType>::token_type;
@@ -219,7 +271,11 @@ class lexer : public lexer_base<BasicJsonType>
         : ia(std::move(adapter))
         , ignore_comments(ignore_comments_)
         , discard_number_values(discard_number_values_)
-    {}
+    {
+        // the token string of an error before the first token is everything
+        // read so far
+        begin_token(std::integral_constant<bool, adapter_token> {});
+    }
 
     // deleted because of pointer members
     lexer(const lexer&) = delete;
@@ -236,8 +292,8 @@ class lexer : public lexer_base<BasicJsonType>
     /// contiguous input: try to decode the 4 hex digits following `\\u`
     /// directly from the input buffer via hex_codepoint(), instead of 4 calls
     /// to get(). On success, advances the adapter exactly as those 4 get()
-    /// calls would (bulk_scan implies lazy_position, so there are no position
-    /// counters to advance) and leaves @a current holding the last of
+    /// calls would (including the position and token bookkeeping of a
+    /// streaming adapter, see bulk_consume) and leaves @a current holding the last of
     /// the 4 digits, just as the last such get() would; the codepoint is
     /// written to @a out. Makes no state change and returns false for a
     /// pending unget, fewer than 4 remaining bytes, or any of the 4 bytes not
@@ -256,7 +312,7 @@ class lexer : public lexer_base<BasicJsonType>
         {
             return false;
         }
-        ia.bulk_skip(4);
+        bulk_consume(reinterpret_cast<const char*>(raw), 4);
         current = char_traits<char_type>::to_int_type(raw[3]);
         out = codepoint;
         return true;
@@ -362,34 +418,79 @@ 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);
-        ia.bulk_skip(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
+    /// the bookkeeping get() would have done for them; the characters are
+    /// never newlines
+    void bulk_consume(const char* data, std::size_t n)
+    {
+        bulk_track_position(n, std::integral_constant<bool, lazy_position> {});
+        bulk_track_token(data, n, std::integral_constant<bool, lazy_token_string> {});
+        ia.bulk_skip(n);
+    }
+
+    /// seekable adapter: the position is derived from the adapter
+    void bulk_track_position(std::size_t /*n*/, std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: count the characters
+    void bulk_track_position(std::size_t n, std::false_type /*lazy*/) noexcept
+    {
+        position.chars_read_total += n;
+        position.chars_read_current_line += n;
+    }
+
+    /// the token is obtained from the adapter
+    void bulk_track_token(const char* /*data*/, std::size_t /*n*/, std::true_type /*lazy*/) const noexcept {}
+
+    /// streaming adapter: capture the characters
+    void bulk_track_token(const char* data, std::size_t n, std::false_type /*lazy*/)
+    {
+        token_string.insert(token_string.end(), data, data + n);
     }
 
     /// 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
@@ -417,11 +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():
@@ -1541,9 +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;
@@ -1553,22 +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;
@@ -1578,31 +1678,45 @@ 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> {}))
+        {
+            return token_type::uninitialized;
+        }
 
         // reset() records where this token starts (for diagnostics), so it has
         // to run before the input position advances below
@@ -1617,10 +1731,14 @@ 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))
             {
-                ia.bulk_skip(len - 1);
+                bulk_consume(data + 1, len - 1);
+                read_number_terminator(std::integral_constant<bool, bulk_window> {});
                 return integer_result;
             }
             // The value does not fit an integer, so this token converts as a
@@ -1630,17 +1748,99 @@ scan_number_done:
         }
 #endif
 
+        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);
-        decimal_point_position = dot_index;
-
-        ia.bulk_skip(len - 1);
-
         return convert_number(number_type, mantissa_end);
     }
 
+    /// window adapter: a number that reaches the end of the window may continue
+    /// after it, so scan_number() has to read it character by character
+    static constexpr bool number_may_continue(bool at_end, std::true_type /*window*/) noexcept
+    {
+        return at_end;
+    }
+
+    /// the end of the block is the end of the input, which ends the number
+    static constexpr bool number_may_continue(bool /*at_end*/, std::false_type /*window*/) noexcept
+    {
+        return false;
+    }
+
+    /// window adapter: read the character after a bulk-scanned number and unget
+    /// it, as scan_number() does. The adapter is a stream that is handed back to
+    /// the user, who must find the same input consumed either way.
+    void read_number_terminator(std::true_type /*window*/)
+    {
+        get();
+        unget();
+    }
+
+    /// 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*/)
     {
@@ -1655,18 +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;
@@ -1675,6 +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
     /////////////////////
@@ -1683,6 +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
@@ -1696,11 +1923,30 @@ scan_number_done:
     /// seekable adapter: remember where the current token starts so it can be
     /// reconstructed from the input on error; current has already been
     /// consumed, hence the -1
-    void note_token_start(std::true_type /*lazy*/) noexcept
+    void note_token_start(std::true_type /*lazy*/)
+    {
+        note_token_start_lazy(std::integral_constant<bool, adapter_token> {});
+    }
+
+    void note_token_start_lazy(std::false_type /*adapter_token*/) noexcept
     {
         token_string_start = ia.get_consumed_count() - 1;
     }
 
+    /// token-tracking adapter: start tracking the token at the first character
+    void begin_token(std::true_type /*adapter_token*/) noexcept
+    {
+        ia.begin_token();
+    }
+
+    void begin_token(std::false_type /*adapter_token*/) const noexcept {}
+
+    /// token-tracking adapter: the token starts with the current character
+    void note_token_start_lazy(std::true_type /*adapter_token*/)
+    {
+        ia.mark_token_start(current);
+    }
+
     /// streaming adapter: start copying the token eagerly, beginning with the
     /// already-read first character
     void note_token_start(std::false_type /*lazy*/) noexcept
@@ -1923,10 +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
     /////////////////////
@@ -2021,6 +2287,20 @@ scan_number_done:
   public:
     /// seekable adapter: rebuild the last read token from the input on demand
     const std::vector<char_type>& collect_token_chars(std::vector<char_type>& out, std::true_type /*lazy*/) const
+    {
+        return collect_token_chars_lazy(out, std::integral_constant<bool, adapter_token> {});
+    }
+
+    /// token-tracking adapter: the adapter hands out everything read since
+    /// mark_token_start(), except a pending unget (see below)
+    const std::vector<char_type>& collect_token_chars_lazy(std::vector<char_type>& out, std::true_type /*adapter_token*/) const
+    {
+        const bool pending_real_unget = next_unget && current != char_traits<char_type>::eof();
+        ia.copy_token(out, pending_real_unget);
+        return out;
+    }
+
+    const std::vector<char_type>& collect_token_chars_lazy(std::vector<char_type>& out, std::false_type /*adapter_token*/) const
     {
         // a pending unget of a real (non-EOF) character means that character
         // was consumed from the input but is not part of the token; EOF is
@@ -2142,9 +2422,8 @@ scan_number_done:
     Pretty-printed JSON indents every line with a run of spaces, which the loop
     in skip_whitespace() would otherwise read one character at a time. The
     spaces are skipped 8 bytes at a time; the last 7 bytes of the input and any
-    other whitespace are left to that loop. Skipping bypasses get(), which is
-    fine because bulk_scan implies lazy_position: the position is derived from
-    the adapter, not counted per character.
+    other whitespace are left to that loop. Skipping bypasses get();
+    bulk_consume() does its bookkeeping.
     */
     void skip_indentation(std::true_type /*bulk*/)
     {
@@ -2167,7 +2446,7 @@ scan_number_done:
                 break;
             }
         }
-        ia.bulk_skip(pos);
+        bulk_consume(data, pos);
     }
 
     /// streaming input: no bulk fast path
@@ -2247,21 +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 '\"':
@@ -2309,6 +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/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

@@ -21,8 +21,18 @@
 #undef JSON_NO_UNIQUE_ADDRESS
 #undef JSON_DISABLE_ENUM_SERIALIZATION
 #undef JSON_DISABLE_TUPLE_REFERENCE_CONVERSION
+#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
 
 #ifndef JSON_TEST_KEEP_MACROS
     #undef JSON_CATCH
@@ -47,6 +57,8 @@
     #undef JSON_STRICT_BINARY_UTF8
     #undef JSON_DELETE_DEPRECATED_FUNCTIONS
     #undef JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS
+    #undef JSON_DISABLE_STREAM_BULK_SCAN
+    #undef JSON_DISABLE_UNLOCKED_FILE_IO
 #endif
 
 #include <nlohmann/thirdparty/hedley/hedley_undef.hpp>

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/output_adapters.hpp

@@ -8,7 +8,10 @@
 
 #pragma once
 
+#include <algorithm> // max
+#include <array> // array
 #include <cstddef> // size_t
+#include <cstring> // memcpy
 #include <memory> // shared_ptr, make_shared
 #include <string> // basic_string
 #include <type_traits> // conditional, integral_constant, is_same
@@ -48,32 +51,258 @@ template<typename CharType> struct output_adapter_protocol
 template<typename CharType>
 using output_adapter_t = std::shared_ptr<output_adapter_protocol<CharType>>;
 
+/// the largest number of characters a writer may reserve() from a sink at once;
+/// output_vector_sink also appends writes of at least this size with insert()
+constexpr std::size_t max_reserve_size = 256;
+
+/// @brief copy @a length characters, at most 32, without a call of memcpy()
+///
+/// Short strings are the typical payload of a binary value, and a call of
+/// memcpy() costs more than copying them. A copy of 16 to 32 characters is two
+/// possibly overlapping 16-byte copies, one of 8 to 15 two 8-byte copies, and
+/// so on down to a single character; each is a fixed-size load and store.
+/// Nothing outside of [src, src + length) is read or written.
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::true_type /*is_byte*/) noexcept
+{
+    JSON_ASSERT(length <= 32);
+    if (length >= 16)
+    {
+        std::array<char, 16> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 16> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 16);
+        std::memcpy(last.data(), src + (length - 16), 16);
+        std::memcpy(dst, first.data(), 16);
+        std::memcpy(dst + (length - 16), last.data(), 16);
+    }
+    else if (length >= 8)
+    {
+        std::array<char, 8> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 8> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 8);
+        std::memcpy(last.data(), src + (length - 8), 8);
+        std::memcpy(dst, first.data(), 8);
+        std::memcpy(dst + (length - 8), last.data(), 8);
+    }
+    else if (length >= 4)
+    {
+        std::array<char, 4> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 4> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 4);
+        std::memcpy(last.data(), src + (length - 4), 4);
+        std::memcpy(dst, first.data(), 4);
+        std::memcpy(dst + (length - 4), last.data(), 4);
+    }
+    else if (length != 0)
+    {
+        // 1 to 3: the first, the middle, and the last character
+        const CharType a = src[0];
+        const CharType b = src[length >> 1];
+        const CharType c = src[length - 1];
+        dst[0] = a;
+        dst[length >> 1] = b;
+        dst[length - 1] = c;
+    }
+}
+
+/// @brief copy @a length characters, of a character type wider than a byte
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::false_type /*is_byte*/) noexcept
+{
+    std::memcpy(dst, src, length * sizeof(CharType));
+}
+
+/// the number of characters an empty vector is grown to by output_vector_sink
+constexpr std::size_t min_vector_sink_capacity = 64;
+
 /// @brief non-virtual output sink writing into a std::vector
 ///
 /// This sink is not part of the virtual output_adapter_protocol hierarchy: it is
 /// passed to binary_writer by value as a template parameter, so
 /// write_character()/write_characters() are ordinary (inlinable) calls with no
 /// vtable lookup and no shared_ptr. It is used for the common
-/// `to_cbor`/`to_msgpack`/... into a std::vector. output_vector_adapter below
-/// wraps this same sink to provide the virtual interface.
+/// `to_cbor`/`to_msgpack`/... into a std::vector.
+///
+/// Most writes are a few bytes - a type marker, or a number of 1, 2, 4, or 8
+/// bytes - and appending those with push_back() or insert() cost more than
+/// encoding them: insert() is an out-of-line call into a routine that can insert
+/// anywhere, and resize() is one as well with libstdc++. The sink therefore keeps
+/// its own write position: it grows the vector geometrically ahead of the
+/// position and copies each write to it, so a write is a capacity check and a
+/// short copy. The destructor shrinks the vector to the bytes actually written,
+/// also when a write throws; shrinking a vector never allocates. The sink owns
+/// that trim, so it can be moved but not copied, and the vector must not be used
+/// while the sink is alive.
 template<typename CharType, typename AllocatorType = std::allocator<CharType>>
 class output_vector_sink
 {
   public:
     explicit output_vector_sink(std::vector<CharType, AllocatorType>& vec) noexcept
+        : v(&vec), cursor(vec.data() + vec.size()), limit(cursor)
+    {}
+
+    output_vector_sink(const output_vector_sink&) = delete;
+    output_vector_sink& operator=(const output_vector_sink&) = delete;
+    output_vector_sink& operator=(output_vector_sink&&) = delete;
+
+    output_vector_sink(output_vector_sink&& other) noexcept
+        : v(other.v), cursor(other.cursor), limit(other.limit)
+    {
+        other.v = nullptr;
+    }
+
+    ~output_vector_sink()
+    {
+        if (v != nullptr)
+        {
+            v->resize(position());
+        }
+    }
+
+    // always inlined: GCC otherwise keeps these out of line in a large
+    // translation unit, and a call per byte or number costs more than the write
+    JSON_HEDLEY_ALWAYS_INLINE
+    void write_character(CharType c)
+    {
+        if (JSON_HEDLEY_UNLIKELY(cursor == limit))
+        {
+            grow(1);
+        }
+        *cursor++ = c;
+    }
+
+    // no JSON_HEDLEY_NON_NULL here: binary_writer legitimately passes a null
+    // pointer with length 0 for empty strings/binary values; memcpy must not
+    // see it.
+    JSON_HEDLEY_ALWAYS_INLINE
+    void write_characters(const CharType* s, std::size_t length)
+    {
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
+        {
+            if (length >= max_reserve_size)
+            {
+                append(s, length);
+                return;
+            }
+            grow(length);
+        }
+        if (length <= 32)
+        {
+            copy_short(cursor, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            cursor += length;
+        }
+        else
+        {
+            std::memcpy(cursor, s, length * sizeof(CharType));
+            cursor += length;
+        }
+    }
+
+    /// @brief space for writing up to @a length characters directly
+    ///
+    /// Returns a pointer to at least @a length (at most @ref max_reserve_size)
+    /// writable characters at the write position; @ref commit then keeps the
+    /// first ones of them. A writer can store whole words there and keep only
+    /// the bytes it needs, and keep its write pointer in a register across many
+    /// values instead of updating the sink after each.
+    JSON_HEDLEY_ALWAYS_INLINE
+    CharType* reserve(const std::size_t length)
+    {
+        JSON_ASSERT(length <= max_reserve_size);
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
+        {
+            grow(length);
+        }
+        return cursor;
+    }
+
+    /// keep the first @a length characters written to the space from reserve()
+    JSON_HEDLEY_ALWAYS_INLINE
+    void commit(const std::size_t length) noexcept
+    {
+        cursor += length;
+    }
+
+    /// nothing to pass on: the characters are in the vector (see output_adapter_sink)
+    void flush() noexcept
+    {}
+
+  private:
+    /// number of characters of *v written so far
+    std::size_t position() const noexcept
+    {
+        return static_cast<std::size_t>(cursor - v->data());
+    }
+
+    /// number of characters that fit the vector behind the write position
+    std::size_t space() const noexcept
+    {
+        return static_cast<std::size_t>(limit - cursor);
+    }
+
+    /// make room for at least @a length more characters after the position;
+    /// kept out of line so that the write functions stay small enough for
+    /// GCC to inline them into the writer
+    JSON_HEDLEY_NEVER_INLINE
+    void grow(const std::size_t length)
+    {
+        // use the capacity the vector already has, else at least double it;
+        // never less than min_vector_sink_capacity, so that a small value does not
+        // reallocate for every few bytes
+        const std::size_t written = position();
+        const std::size_t needed = written + length;
+        v->resize((std::max)(min_vector_sink_capacity, needed <= v->capacity()
+                             ? v->capacity()
+                             : (std::max)(needed, v->capacity() * 2)));
+        cursor = v->data() + written;
+        limit = v->data() + v->size();
+    }
+
+    /// append @a length (at least max_reserve_size) characters that do not fit
+    /// the space ahead of the position: insert() copies them once into the
+    /// grown vector, where grow() would first zero the new space and then copy
+    /// over it - noticeable for a large string or binary value
+    JSON_HEDLEY_NEVER_INLINE
+    void append(const CharType* s, const std::size_t length)
+    {
+        const std::size_t written = position();
+        v->resize(written);
+        if (v->capacity() - written < length)
+        {
+            // grow geometrically, with room for the writes that follow, so
+            // that the next grow() need not reallocate right away
+            v->reserve((std::max)(v->capacity() * 2, written + length + max_reserve_size));
+        }
+        v->insert(v->end(), s, s + length);
+        cursor = v->data() + v->size();
+        limit = cursor;
+    }
+
+    std::vector<CharType, AllocatorType>* v;
+    /// the write position: in *v, at the number of characters written so far
+    CharType* cursor;
+    /// the end of *v
+    CharType* limit;
+};
+
+/// output adapter for byte vectors
+template<typename CharType, typename AllocatorType = std::allocator<CharType>>
+class output_vector_adapter : public output_adapter_protocol<CharType>
+{
+  public:
+    explicit output_vector_adapter(std::vector<CharType, AllocatorType>& vec) noexcept
         : v(vec)
     {}
 
-    void write_character(CharType c)
+    void write_character(CharType c) override
     {
         v.push_back(c);
     }
 
     // no JSON_HEDLEY_NON_NULL here: binary_writer legitimately passes a null
-    // pointer with length 0 for empty strings/binary values. Appending an empty
-    // range is a no-op; the type-erased path tolerates this via the (unattributed)
-    // virtual base, and the concrete sink must do the same.
-    void write_characters(const CharType* s, std::size_t length)
+    // pointer with length 0 for empty strings/binary values; appending an empty
+    // range is a no-op
+    void write_characters(const CharType* s, std::size_t length) override
     {
         v.insert(v.end(), s, s + length);
     }
@@ -82,33 +311,6 @@ class output_vector_sink
     std::vector<CharType, AllocatorType>& v;
 };
 
-/// output adapter for byte vectors
-///
-/// The appending itself lives in output_vector_sink; this class only adds the
-/// virtual output_adapter_protocol interface on top of it, so both the
-/// type-erased and the templated path share one implementation.
-template<typename CharType, typename AllocatorType = std::allocator<CharType>>
-class output_vector_adapter : public output_adapter_protocol<CharType>
-{
-  public:
-    explicit output_vector_adapter(std::vector<CharType, AllocatorType>& vec) noexcept
-        : sink(vec)
-    {}
-
-    void write_character(CharType c) override
-    {
-        sink.write_character(c);
-    }
-
-    void write_characters(const CharType* s, std::size_t length) override
-    {
-        sink.write_characters(s, length);
-    }
-
-  private:
-    output_vector_sink<CharType, AllocatorType> sink;
-};
-
 #ifndef JSON_NO_IO
 /// output adapter for output streams
 template<typename CharType>
@@ -163,9 +365,17 @@ class output_string_adapter : public output_adapter_protocol<CharType>
 ///
 /// Wraps the polymorphic output_adapter_t so the same binary_writer template can
 /// also target arbitrary adapters (output streams, strings, user-provided
-/// adapters) via the `output_adapter`-based overloads. Each write still goes
-/// through one virtual call, exactly as before; only the concrete sinks above
-/// avoid it.
+/// adapters) via the `output_adapter`-based overloads.
+///
+/// A write through the adapter is a virtual call, and for a stream a call with
+/// a sentry on top, which costs more than encoding the few bytes of most values.
+/// The sink therefore collects what is written in a buffer and passes it on in
+/// one piece when it is full, a write is too large for it, or @ref flush is
+/// called. Everything written reaches the adapter in the same order and with
+/// the same bytes; the destructor passes on what is left, so that a write that
+/// throws leaves the adapter with exactly the bytes written before it. Only
+/// @ref flush can report an error of the adapter (the destructor must not
+/// throw), so it has to be called once the value is written.
 template<typename CharType>
 class output_adapter_sink
 {
@@ -176,20 +386,121 @@ class output_adapter_sink
         JSON_ASSERT(oa);
     }
 
+    output_adapter_sink(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(output_adapter_sink&&) = delete;
+
+    output_adapter_sink(output_adapter_sink&& other) noexcept
+        : oa(std::move(other.oa)), used(other.used)
+    {
+        std::memcpy(buffer.data(), other.buffer.data(), other.used * sizeof(CharType));
+        other.used = 0;
+    }
+
+    ~output_adapter_sink()
+    {
+        // a normal return has flushed already, so this is for a write that threw
+        if (used != 0)
+        {
+            JSON_TRY
+            {
+                flush();
+            }
+            JSON_CATCH(...)
+            {
+                // nothing can be done about an adapter that throws here
+            }
+        }
+    }
+
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_character(CharType c)
     {
-        oa->write_character(c);
+        if (JSON_HEDLEY_UNLIKELY(used == buffer_size))
+        {
+            flush();
+        }
+        buffer[used++] = c;
     }
 
     // no JSON_HEDLEY_NON_NULL: forwards (null, 0) for empty payloads, exactly as
     // the type-erased path already did before this sink existed
+    JSON_HEDLEY_ALWAYS_INLINE
     void write_characters(const CharType* s, std::size_t length)
     {
-        oa->write_characters(s, length);
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            write_large(s, length);
+            return;
+        }
+        if (length <= 32)
+        {
+            copy_short(buffer.data() + used, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            used += length;
+        }
+        else
+        {
+            std::memcpy(buffer.data() + used, s, length * sizeof(CharType));
+            used += length;
+        }
+    }
+
+    /// space for writing up to @a length characters (see
+    /// output_vector_sink::reserve)
+    JSON_HEDLEY_ALWAYS_INLINE
+    CharType* reserve(const std::size_t length)
+    {
+        JSON_ASSERT(length <= max_reserve_size);
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            flush();
+        }
+        return buffer.data() + used;
+    }
+
+    /// keep the first @a length characters written to the space from reserve()
+    JSON_HEDLEY_ALWAYS_INLINE
+    void commit(const std::size_t length) noexcept
+    {
+        used += length;
+    }
+
+    /// pass everything written so far on to the adapter
+    void flush()
+    {
+        if (used != 0)
+        {
+            // reset first: the characters count as written even if the adapter throws
+            const std::size_t length = used;
+            used = 0;
+            oa->write_characters(buffer.data(), length);
+        }
     }
 
   private:
+    /// a write that does not fit the space left in the buffer
+    JSON_HEDLEY_NEVER_INLINE
+    void write_large(const CharType* s, const std::size_t length)
+    {
+        flush();
+        if (length >= buffer_size)
+        {
+            oa->write_characters(s, length);
+        }
+        else
+        {
+            std::memcpy(buffer.data(), s, length * sizeof(CharType));
+            used = length;
+        }
+    }
+
+    /// large enough for reserve() at any time
+    static constexpr std::size_t buffer_size = 4 * max_reserve_size;
+
     output_adapter_t<CharType> oa;
+    /// number of characters of @ref buffer not yet passed on
+    std::size_t used = 0;
+    std::array<CharType, buffer_size> buffer{};
 };
 
 /// @brief whether std::basic_string<CharType> has a non-deprecated std::char_traits

include/nlohmann/detail/value_t.hpp

@@ -14,7 +14,6 @@
 #include <cstdint> // uint8_t
 #include <limits> // numeric_limits
 #include <string> // string
-#include <type_traits> // is_signed
 
 #include <nlohmann/detail/macro_scope.hpp>
 #if JSON_HAS_THREE_WAY_COMPARISON
@@ -208,13 +207,16 @@ FloatType compare_integer_with_float(const IntegerType i, const FloatType f) noe
 
     // values of IntegerType lie in [-bound, bound) when signed and in
     // [0, bound) when unsigned; digits excludes the sign bit, so bound is a
-    // power of two that the float represents exactly
-    const FloatType bound = std::ldexp(static_cast<FloatType>(1), std::numeric_limits<IntegerType>::digits);
+    // power of two that the float represents exactly; the signedness comes
+    // from numeric_limits as well, because std::is_signed is false for class
+    // types such as 128-bit or multiprecision integers
+    using limits = std::numeric_limits<IntegerType>;
+    const FloatType bound = std::ldexp(static_cast<FloatType>(1), limits::digits);
     if (f >= bound)
     {
         return ordered(-1);
     }
-    if (std::is_signed<IntegerType>::value ? (f < -bound) : (f < static_cast<FloatType>(0)))
+    if (limits::is_signed ? (f < -bound) : (f < static_cast<FloatType>(0)))
     {
         return ordered(1);
     }

include/nlohmann/detail/view/document_data.hpp

@@ -11,7 +11,7 @@
 #include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // uint8_t, uint32_t
-#include <cstring> // memcpy
+#include <cstring> // memcmp, memcpy
 #include <limits> // numeric_limits
 #include <map> // map
 #include <memory> // unique_ptr
@@ -56,14 +56,16 @@ struct document_data
     std::vector<object_index> indexes;
     std::vector<std::uint32_t> index_slots;
     std::vector<std::uint32_t> large_objects; ///< positions of the objects to index (noted while parsing)
-    std::array<const char*, 4> base = {{nullptr, nullptr, nullptr, nullptr}}; ///< string bases: source, arena, edit arena (indexed by flags & node_flags::storage)
+    std::array<const char*, 4> base = {{nullptr, nullptr, nullptr, nullptr}}; ///< string bases: source, arena, edit arena, unused (indexed by flags & node_flags::storage)
     bool discarded = true;
 
     /// The storage of edits (editable documents only; see edit_storage.hpp).
-    /// Edits never move or resize the parsed index, so views stay valid: an
-    /// array/object whose elements change gets node_flags::moved, and its
-    /// elements then live in a separate sequence (a header node, then the
-    /// entries), whose entries link to the values.
+    /// No edit moves or resizes the parsed index, and the nodes of values
+    /// written by edits never move either, so a view stays valid whenever it
+    /// was taken, before the first edit or between later ones: an array/object
+    /// whose elements change gets node_flags::moved, and its elements then live
+    /// in a separate sequence (a header node, then the entries), whose entries
+    /// link to the values.
     struct edit_state
     {
         std::vector<node*> moved{};                    ///< element sequences of moved arrays/objects (header node first) // NOLINT(readability-redundant-member-init)
@@ -183,6 +185,12 @@ struct document_data
         return base[n.flags & node_flags::storage] + n.off;
     }
 
+    /// whether the string of node k (a key) is the n bytes at key
+    bool key_equals(const node& k, const char* key, std::size_t n) const noexcept
+    {
+        return k.len == n && (n == 0 || std::memcmp(str(k), key, n) == 0);
+    }
+
     /// the node after n's subtree (containers span `next` nodes, scalars one)
     static NLOHMANN_VIEW_ALWAYS_INLINE const node* after(const node* n) noexcept
     {

include/nlohmann/detail/view/edit.hpp

@@ -12,7 +12,7 @@
 #include <cmath> // isinf, isnan
 #include <cstddef> // size_t
 #include <cstdint> // int64_t, uint8_t, uint32_t, uint64_t
-#include <cstring> // memcmp, memmove
+#include <cstring> // memmove
 #include <limits> // numeric_limits
 #include <string> // string, to_string
 #include <type_traits> // decay, enable_if, integral_constant, is_arithmetic, is_convertible, is_floating_point, is_same, is_signed
@@ -313,7 +313,7 @@ class editor
 
     bool key_equals(const node& k, string_view_t key) const noexcept
     {
-        return k.len == key.size() && (key.size() == 0 || std::memcmp(m_doc.str(k), key.data(), key.size()) == 0);
+        return m_doc.key_equals(k, key.data(), key.size());
     }
 
     /// remove the members with this key (all, or all but the first) from an object

include/nlohmann/detail/view/edit_storage.hpp

@@ -50,7 +50,19 @@ inline node* alloc_nodes(document_data& d, std::size_t k)
     document_data::edit_state& e = edit_state_of(d);
     if (NLOHMANN_VIEW_UNLIKELY(static_cast<std::size_t>(e.chunk_end - e.chunk_cur) < k))
     {
-        const std::size_t count = (std::max)(k, e.chunk_next);
+        if (k >= e.chunk_next / 2)
+        {
+            // a large request (half of the next chunk or more) gets a chunk of
+            // its own, so that the rest of the current chunk stays available
+            // for smaller ones
+            std::unique_ptr<node[]> own(new node[k]()); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+            e.chunks.push_back(std::move(own));
+            e.bytes += k * sizeof(node);
+            return e.chunks.back().get();
+        }
+        // a smaller request leaves the rest of the current chunk unused
+        // (fewer than k nodes, and k is less than half of the new chunk)
+        const std::size_t count = e.chunk_next;
         std::unique_ptr<node[]> fresh(new node[count]()); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
         e.chunks.push_back(std::move(fresh));
         e.chunk_cur = e.chunks.back().get();
@@ -81,7 +93,8 @@ inline std::size_t text_capacity(std::size_t cap, std::size_t used, std::size_t
 }
 
 /// copy n bytes into the edit arena and return their offset; a new buffer
-/// leaves the old one alive, so that string views into it remain valid
+/// leaves the old one alive, so that string views into it remain valid; n is
+/// 0 for an empty string (written by an edit, or the key of a new member)
 inline std::uint32_t append_text(document_data& d, const char* s, std::size_t n)
 {
     document_data::edit_state& e = edit_state_of(d);
@@ -89,7 +102,7 @@ inline std::uint32_t append_text(document_data& d, const char* s, std::size_t n)
     {
         const std::size_t cap = text_capacity(e.text_cap, e.text_used, n);
         std::unique_ptr<char[]> fresh(new char[cap]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
-        if (e.text_used != 0)
+        if (NLOHMANN_VIEW_LIKELY(e.text_used != 0)) // (0 only for the first buffer, and then texts is empty)
         {
             std::memcpy(fresh.get(), e.texts.back().get(), e.text_used);
         }
@@ -127,8 +140,10 @@ inline void reserve_moved(document_data& d)
     if (e.moved.size() == e.moved.capacity() || e.moved_cap.size() == e.moved_cap.capacity())
     {
         // both grow before either changes, so that the push_backs cannot throw
-        e.moved.reserve((2 * e.moved.size()) + 16);
-        e.moved_cap.reserve((2 * e.moved.size()) + 16);
+        // (both conditions are needed: reserve() may allocate more than asked,
+        // so the two capacities can differ)
+        e.moved.reserve((e.moved.size() * 2) + 16);
+        e.moved_cap.reserve((e.moved.size() * 2) + 16);
     }
 }
 

include/nlohmann/detail/view/image.hpp

@@ -8,11 +8,11 @@
 
 #pragma once
 
-#include <algorithm> // sort
+#include <algorithm> // min, sort
 #include <array> // array
-#include <cstddef> // size_t
+#include <cstddef> // offsetof, size_t
 #include <cstdint> // int64_t, uint8_t, uint16_t, uint32_t, uint64_t
-#include <cstring> // memcmp, memcpy
+#include <cstring> // memcmp, memcpy, memset
 #include <limits> // numeric_limits
 #include <string> // string
 #include <vector> // vector
@@ -162,7 +162,7 @@ inline void compact_nodes(const document_data& d, std::size_t arena_size, std::v
     copy_node(d.nodes);
     while (!stack.empty())
     {
-        frame& top = stack.back();
+        const frame top = stack.back();
         if (top.cur == top.end)
         {
             node& c = out[top.index];
@@ -171,20 +171,14 @@ inline void compact_nodes(const document_data& d, std::size_t arena_size, std::v
             stack.pop_back();
             continue;
         }
-        ++top.count;
-        const node* v = nullptr;
         if (top.object)
         {
             out.push_back(string_node(*top.cur));
-            v = document_data::deref(top.cur + 1);
-            top.cur = document_data::after(top.cur + 1);
         }
-        else
-        {
-            v = document_data::deref(top.cur);
-            top.cur = document_data::after(top.cur);
-        }
-        copy_node(v); // may grow the stack (top is not used afterwards)
+        const node* const value = top.cur + (top.object ? 1 : 0);
+        stack.back().cur = document_data::after(value);
+        ++stack.back().count;
+        copy_node(document_data::deref(value)); // may grow the stack
     }
 }
 
@@ -195,17 +189,15 @@ inline std::vector<std::uint8_t> save_image(const document_data& d)
     throw_type_error(320, "json_document images need a little-endian target"); // LCOV_EXCL_LINE
 #endif
     const std::size_t arena_size = d.arena_size;
-    const node* nodes = d.nodes;
-    std::size_t count = d.node_count;
     std::vector<node> compacted;
     std::string text_tail;
     std::string arena_tail;
     if (d.edits)
     {
         compact_nodes(d, arena_size, compacted, text_tail, arena_tail);
-        nodes = compacted.data();
-        count = compacted.size();
     }
+    const node* const nodes = d.edits ? compacted.data() : d.nodes;
+    const std::size_t count = d.edits ? compacted.size() : d.node_count;
     const std::size_t text_size = d.size + text_tail.size();
     const std::size_t total_arena = arena_size + arena_tail.size();
     if (NLOHMANN_VIEW_UNLIKELY(text_size >= image_limit || total_arena >= image_limit || count >= image_limit))
@@ -225,14 +217,13 @@ inline std::vector<std::uint8_t> save_image(const document_data& d)
     std::memcpy(o, &h, sizeof(h));
     o += sizeof(h);
     std::memcpy(o, nodes, count * sizeof(node));
-    // the hash indexes are rebuilt by load()
+    // the hash indexes are rebuilt by load(), whose check rejects any other
+    // extra of an object, and the same document saves to the same bytes
     for (std::size_t i = 0; i < count; ++i)
     {
         if (nodes[i].kind == static_cast<std::uint8_t>(value_t::object) && nodes[i].extra != 0)
         {
-            node n = nodes[i];
-            n.extra = 0;
-            std::memcpy(o + (i * sizeof(node)), &n, sizeof(node));
+            std::memset(o + (i * sizeof(node)) + offsetof(node, extra), 0, sizeof(nodes[i].extra));
         }
     }
     o += count * sizeof(node);
@@ -271,21 +262,22 @@ inline bool scan_number_token(const unsigned char* s, std::size_t len, number_to
     const unsigned char* p = s;
     t.negative = p != e && *p == '-';
     p += t.negative ? 1 : 0;
-    t.int_start = p;
     if (p == e)
     {
         return false;
     }
+    t.int_start = p;
     if (*p == '0')
     {
         ++p;
     }
     else if (*p >= '1' && *p <= '9')
     {
-        while (p != e && is_digit(*p))
+        do
         {
             ++p;
         }
+        while (p != e && is_digit(*p));
     }
     else
     {
@@ -318,11 +310,12 @@ inline bool scan_number_token(const unsigned char* s, std::size_t len, number_to
         {
             return false;
         }
-        while (p != e && is_digit(*p))
+        do
         {
             t.exponent = t.exponent < 100000 ? (t.exponent * 10) + (*p - '0') : t.exponent;
             ++p;
         }
+        while (p != e && is_digit(*p));
         t.exponent = exp_negative ? -t.exponent : t.exponent;
         t.is_float = true;
     }
@@ -333,7 +326,7 @@ inline bool scan_number_token(const unsigned char* s, std::size_t len, number_to
 /// writes "many" instead; the caller accepts both)
 inline std::uint16_t float_layout(const number_token& t)
 {
-    return static_cast<std::uint16_t>((t.int_digits < 255 ? t.int_digits : 255) | ((t.frac_digits < 255 ? t.frac_digits : 255) << 8u));
+    return static_cast<std::uint16_t>((std::min)(t.int_digits, std::size_t{255}) | ((std::min)(t.frac_digits, std::size_t{255}) << 8u));
 }
 
 /// whether a float token (well-formed, per scan_number_token) is finite as
@@ -735,14 +728,10 @@ inline void load_image(document_data& d, const std::uint8_t* image, std::size_t
     // the hash indexes of large objects, as after parsing
     for (std::size_t i = 0; i < count; ++i)
     {
-        node& n = d.nodes[i];
-        if (n.kind == static_cast<std::uint8_t>(value_t::object))
+        const node& n = d.nodes[i];
+        if (n.kind == static_cast<std::uint8_t>(value_t::object) && n.len >= document_data::index_min_members)
         {
-            n.extra = 0;
-            if (n.len >= document_data::index_min_members)
-            {
-                d.large_objects.push_back(static_cast<std::uint32_t>(i));
-            }
+            d.large_objects.push_back(static_cast<std::uint32_t>(i));
         }
     }
     build_object_indexes(d);

include/nlohmann/detail/view/lookup.hpp

@@ -112,7 +112,7 @@ const node* find_member(const document_data& d, const node* object, const char*
     }
     for (const node* m = nav::first(d, object); m != end; m = document_data::after(m + 1))
     {
-        if (m->len == n && std::memcmp(d.str(*m), key, n) == 0)
+        if (d.key_equals(*m, key, n))
         {
             return m;
         }
@@ -129,7 +129,7 @@ std::size_t count_members(const document_data& d, const node* object, const char
     const node* const end = nav::end(d, object);
     for (const node* m = nav::first(d, object); m != end; m = document_data::after(m + 1))
     {
-        if (m->len == n && (n == 0 || std::memcmp(d.str(*m), key, n) == 0))
+        if (d.key_equals(*m, key, n))
         {
             ++result;
         }

include/nlohmann/detail/view/node.hpp

@@ -34,7 +34,7 @@ struct node_flags
 {
     static constexpr std::uint8_t escaped = 1; ///< string payload lives in the decode arena, not the source
     static constexpr std::uint8_t edited = 2;  ///< string or number token lives in the edit arena (editable documents)
-    static constexpr std::uint8_t storage = 3; ///< mask: where a string or number token lives (index into document_data::base)
+    static constexpr std::uint8_t storage = escaped | edited; ///< mask: where a string or number token lives (index into document_data::base)
     static constexpr std::uint8_t is_true = 4; ///< boolean value
     static constexpr std::uint8_t moved = 8;   ///< array/object: the elements live in a separate sequence (editable documents)
     static constexpr std::uint8_t is_new = 16; ///< written by an edit: no source position

include/nlohmann/detail/view/number.hpp

@@ -45,10 +45,11 @@ inline bool float_token_layout(const char* first, const char* last, std::size_t&
     {
         return false;
     }
-    while (digit(p))
+    do
     {
         ++p;
     }
+    while (digit(p));
     dot = std::string::npos;
     if (p != last && *p == '.')
     {
@@ -57,10 +58,11 @@ inline bool float_token_layout(const char* first, const char* last, std::size_t&
         {
             return false;
         }
-        while (digit(p))
+        do
         {
             ++p;
         }
+        while (digit(p));
     }
     mantissa_end = static_cast<std::size_t>(p - first);
     if (p != last && (*p == 'e' || *p == 'E'))
@@ -71,10 +73,11 @@ inline bool float_token_layout(const char* first, const char* last, std::size_t&
         {
             return false;
         }
-        while (digit(p))
+        do
         {
             ++p;
         }
+        while (digit(p));
     }
     return p == last;
 }

include/nlohmann/detail/view/object_index.hpp

@@ -10,10 +10,8 @@
 
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t, uint64_t
-#include <cstring> // memcmp
 
 #include <nlohmann/json.hpp>
-#include <nlohmann/detail/meta/type_traits.hpp>
 #include <nlohmann/detail/view/document_data.hpp>
 #include <nlohmann/detail/view/macro_scope.hpp>
 #include <nlohmann/detail/view/node.hpp>
@@ -95,7 +93,7 @@ inline void build_object_index(document_data& d, node* obj)
         while (table[slot] != 0)
         {
             const node* const other = obj + table[slot];
-            if (other->len == k->len && (k->len == 0 || std::memcmp(d.str(*other), key, k->len) == 0))
+            if (d.key_equals(*other, key, k->len))
             {
                 duplicate = true; // keep the first
                 break;
@@ -142,7 +140,7 @@ inline const node* find_indexed(const document_data& d, const node* obj, const c
             return nullptr;
         }
         const node* const k = obj + s;
-        if (k->len == n && (n == 0 || std::memcmp(d.str(*k), key, n) == 0))
+        if (d.key_equals(*k, key, n))
         {
             return k;
         }

include/nlohmann/detail/view/scan.hpp

@@ -175,32 +175,7 @@ NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_digits(const unsigned char
 /// powers of ten up to 10^19 as integers
 inline std::uint64_t int_pow10(unsigned k) noexcept
 {
-    static const std::array<std::uint64_t, 20> table =
-    {
-        {
-            1u,
-            10u,
-            100u,
-            1000u,
-            10000u,
-            100000u,
-            1000000u,
-            10000000u,
-            100000000u,
-            1000000000u,
-            10000000000u,
-            100000000000u,
-            1000000000000u,
-            10000000000000u,
-            100000000000000u,
-            1000000000000000u,
-            10000000000000000u,
-            100000000000000000u,
-            1000000000000000000u,
-            10000000000000000000u
-        }
-    };
-    return table[k];
+    return ::nlohmann::detail::dtoa_impl::powers_of_ten()[k];
 }
 
 /// value of 0 < k < 8 digits at p in one step if [p, p + 8) lies below

include/nlohmann/detail/view/serializer.hpp

@@ -75,7 +75,7 @@ class output_buffer
 
     /// a string literal, without its terminating NUL
     template<std::size_t N>
-    NLOHMANN_VIEW_ALWAYS_INLINE void put(const char (&s)[N])
+    NLOHMANN_VIEW_ALWAYS_INLINE void put(const char (&s)[N]) // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
     {
         put(s, N - 1);
     }
@@ -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 65536;
-    }
+    /// 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));
@@ -171,17 +168,48 @@ 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
-/// std::vector. The address of the stack never escapes (the growth gets the
-/// vector and returns the new storage), so its pointers stay in registers.
+/// std::vector. grow() is static and out of line: it gets the vector and the
+/// used range and returns the new storage, so the address of the stack never
+/// reaches a call. Otherwise the compiler would have to keep the three
+/// pointers in memory and load and store them on every push_back() and
+/// pop_back(), not just when growing.
 template<typename T>
 class small_stack
 {
   public:
     small_stack(T* buffer, std::size_t capacity, std::vector<T>& heap) noexcept
-        : m_begin(buffer), m_top(buffer), m_end(buffer + capacity), m_heap(&heap)
+        : m_begin(buffer), m_top(buffer), m_end(buffer + capacity), m_heap(heap)
     {}
     small_stack(const small_stack&) = delete;
     small_stack(small_stack&&) = delete;
@@ -195,7 +223,7 @@ class small_stack
         {
             const std::size_t used = size();
             const std::size_t capacity = 2 * static_cast<std::size_t>(m_end - m_begin);
-            m_begin = grow(*m_heap, m_begin, used, capacity);
+            m_begin = grow(m_heap, m_begin, used, capacity);
             m_top = m_begin + used;
             m_end = m_begin + capacity;
         }
@@ -235,9 +263,18 @@ class small_stack
     T* m_begin;
     T* m_top;
     T* m_end;
-    std::vector<T>* m_heap;
+    std::vector<T>& m_heap;
 };
 
+/// a string literal at w, without its terminating NUL (after room())
+/// @return a pointer past it
+template<std::size_t N>
+NLOHMANN_VIEW_ALWAYS_INLINE char* put_literal(char* w, const char (&s)[N]) noexcept // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+{
+    std::memcpy(w, s, N - 1);
+    return w + N - 1;
+}
+
 /// how the view's dump() writes a value
 struct dump_style
 {
@@ -274,15 +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;
         }
@@ -326,15 +367,15 @@ class view_serializer
                     m_out.finish();
                     return;
                 }
-                frame& f = stack.back();
-                if (f.pos == f.end)
+                if (stack.back().pos == stack.back().end)
                 {
-                    const bool object = f.object; // (f dangles after pop_back)
+                    const bool object = stack.back().object;
                     stack.pop_back();
                     newline(stack.size());
                     m_out.put(object ? '}' : ']');
                     continue;
                 }
+                frame& f = stack.back();
                 if (!f.first)
                 {
                     m_out.put(',');
@@ -388,6 +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
@@ -405,16 +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
@@ -461,18 +539,19 @@ class view_serializer
             }
             w += n;
         };
-        // a literal of n bytes (after room(n))
-        const auto literal = [&](const char* text, std::size_t n)
-        {
-            std::memcpy(w, text, n);
-            w += n;
-        };
         // a string that is not a plain string of the source (out of line, so
         // that the cursor stays in a register here)
         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();
         };
@@ -495,7 +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++ = '"';
@@ -552,11 +631,11 @@ class view_serializer
                     room(8);
                     if ((n->flags & node_flags::is_true) != 0)
                     {
-                        literal("true", 4);
+                        w = put_literal(w, "true");
                     }
                     else
                     {
-                        literal("false", 5);
+                        w = put_literal(w, "false");
                     }
                     break;
                 case value_t::object:
@@ -566,7 +645,7 @@ class view_serializer
                     room(8);
                     if (n->len == 0)
                     {
-                        literal(object ? "{}" : "[]", 2);
+                        w = object ? put_literal(w, "{}") : put_literal(w, "[]");
                     }
                     else
                     {
@@ -586,7 +665,7 @@ class view_serializer
                 }
                 case value_t::null:
                     room(8);
-                    literal("null", 4);
+                    w = put_literal(w, "null");
                     break;
                 case value_t::binary:    // LCOV_EXCL_LINE (not in a document)
                 case value_t::discarded: // LCOV_EXCL_LINE
@@ -626,7 +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++ = '"';
@@ -766,7 +845,7 @@ class view_serializer
                 // check also keeps an image that was only checked for bounds,
                 // whose token may not be made of digits, from the counted
                 // overload)
-                const auto& powers = ::nlohmann::detail::dtoa_impl::powers_of_ten_16();
+                const auto& powers = ::nlohmann::detail::dtoa_impl::powers_of_ten();
                 const unsigned count = int_digits + frac_digits;
                 return count - 1u < 15u && d.w >= powers[count - 1u] && d.w < powers[count]
                        ? ::nlohmann::detail::dtoa_impl::write_short_decimal(w, d.w, static_cast<int>(count), static_cast<int>(d.exponent))
@@ -788,8 +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);
@@ -837,9 +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/detail/view/value.hpp

@@ -12,7 +12,7 @@
 #include <cstdint> // int64_t
 #include <map> // map
 #include <string> // basic_string
-#include <type_traits> // enable_if, is_constructible
+#include <type_traits> // decay, enable_if, is_arithmetic, is_constructible, is_integral, is_same
 #include <unordered_map> // unordered_map
 #include <vector> // vector
 
@@ -104,6 +104,20 @@ Map map_value(const View& v)
 template<typename Key>
 struct is_string_key : std::is_constructible<Key, const char*, std::size_t> {};
 
+/// the arithmetic types arithmetic_value() converts to (bool has its own
+/// conversion)
+template<typename T>
+struct is_non_bool_arithmetic : conjunction<std::is_arithmetic<T>, negation<std::is_same<T, bool>>> {};
+
+/// the default values of value() other than C strings, which have their own
+/// overload (returning string_t)
+template<typename T>
+struct is_non_c_string_default : negation<std::is_same<typename std::decay<T>::type, const char*>> {};
+
+/// the integral types accepted as an array index by the edits of a document
+template<typename T>
+struct is_non_bool_integral : conjunction<std::is_integral<T>, negation<std::is_same<T, bool>>> {};
+
 /// whether View::get<T>() converts to T without materialize(); get_to()
 /// assigns these, and passes the others to from_json
 template<typename T, typename BasicJsonType, typename View>

include/nlohmann/json.hpp

@@ -19,6 +19,7 @@
 #define INCLUDE_NLOHMANN_JSON_HPP_
 
 #include <algorithm> // all_of, find, for_each, none_of
+#include <array> // array
 #include <cmath> // isnan
 #include <cstddef> // nullptr_t, ptrdiff_t, size_t
 #include <cstdint> // uint8_t
@@ -6321,7 +6322,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_cbor(j);
         return result;
     }
@@ -6331,7 +6332,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_cbor(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a CBOR serialization of a given JSON value
@@ -6339,7 +6342,9 @@ public:
     static void to_cbor(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_cbor(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6348,7 +6353,7 @@ public:
             const error_handler_t error_handler = error_handler_t::keep)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_msgpack(j);
         return result;
     }
@@ -6358,7 +6363,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_msgpack(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a MessagePack serialization of a given JSON value
@@ -6366,7 +6373,9 @@ public:
     static void to_msgpack(const basic_json& j, detail::output_adapter<char> o,
                            const error_handler_t error_handler = error_handler_t::keep)
     {
-        binary_writer<char>(o, error_handler).write_msgpack(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6377,7 +6386,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type);
         return result;
     }
@@ -6388,7 +6397,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a UBJSON serialization of a given JSON value
@@ -6397,7 +6408,9 @@ public:
                           const bool use_size = false, const bool use_type = false,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6409,7 +6422,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
         return result;
     }
@@ -6421,7 +6434,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BJData serialization of a given JSON value
@@ -6431,7 +6446,9 @@ public:
                           const bjdata_version_t version = bjdata_version_t::draft2,
                           const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6440,7 +6457,7 @@ public:
             const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result, error_handler).write_bson(j);
         return result;
     }
@@ -6450,7 +6467,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<std::uint8_t>(o, error_handler).write_bson(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BSON serialization of a given JSON value
@@ -6458,7 +6477,9 @@ public:
     static void to_bson(const basic_json& j, detail::output_adapter<char> o,
                         const error_handler_t error_handler = detail::binary_writer_default_error_handler())
     {
-        binary_writer<char>(o, error_handler).write_bson(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
@@ -6466,7 +6487,7 @@ public:
     static std::vector<std::uint8_t> to_bon8(const basic_json& j)
     {
         std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
         vector_writer(result).write_bon8(j);
         return result;
     }
@@ -6475,14 +6496,18 @@ public:
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<std::uint8_t> o)
     {
-        binary_writer<std::uint8_t>(o).write_bon8(j);
+        binary_writer<std::uint8_t> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a BON8 serialization of a given JSON value
     /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
     static void to_bon8(const basic_json& j, detail::output_adapter<char> o)
     {
-        binary_writer<char>(o).write_bon8(j);
+        binary_writer<char> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
     }
 
     /// @brief create a JSON value from an input in CBOR format
@@ -6878,6 +6903,72 @@ public:
     /// @name JSON Patch functions
     /// @{
 
+  private:
+    /// the members of an operation of a JSON patch that are of interest
+    struct patch_members
+    {
+        basic_json* op = nullptr;
+        basic_json* path = nullptr;
+        basic_json* value = nullptr;
+        basic_json* from = nullptr;
+    };
+
+    /// @brief find the members "op", "path", "value", and "from" of the object @a val in one pass
+    static patch_members find_patch_members(const basic_json& val, std::true_type /*key_type is string_t*/)
+    {
+        patch_members members;
+        for (auto& member : *val.m_data.m_value.object)
+        {
+            const string_t& key = member.first;
+            switch (key.size())
+            {
+                case 2:
+                    if (key == "op")
+                    {
+                        members.op = &member.second;
+                    }
+                    break;
+                case 4:
+                    if (key == "path")
+                    {
+                        members.path = &member.second;
+                    }
+                    else if (key == "from")
+                    {
+                        members.from = &member.second;
+                    }
+                    break;
+                case 5:
+                    if (key == "value")
+                    {
+                        members.value = &member.second;
+                    }
+                    break;
+                default:
+                    break;
+            }
+        }
+        return members;
+    }
+
+    /// @brief find the members by lookup for key types other than string_t
+    static patch_members find_patch_members(const basic_json& val, std::false_type /*key_type is not string_t*/)
+    {
+        const auto find_member = [&val](const char* name) -> basic_json *
+        {
+            const auto it = val.m_data.m_value.object->find(string_t(name));
+            return it != val.m_data.m_value.object->end() ? &it->second : nullptr;
+        };
+
+        patch_members members;
+        members.op = find_member("op");
+        members.path = find_member("path");
+        members.value = find_member("value");
+        members.from = find_member("from");
+        return members;
+    }
+
+  public:
     /// @brief applies a JSON patch in-place without copying the object
     /// @sa https://json.nlohmann.me/api/basic_json/patch/
     void patch_inplace(const basic_json& json_patch)
@@ -7045,31 +7136,155 @@ 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;
+        typename json_pointer::template plain_steps<basic_json> steps;
+        const auto try_plain = [&result, &token, &steps](patch_operations operation, const string_t& path, const patch_members & members, const basic_json & val)
+        {
+            if (path.empty())
+            {
+                return false;
+            }
+
+            // a missing "value" is reported below
+            const basic_json* const value_ptr = members.value;
+            const basic_json& value = value_ptr != nullptr ? *value_ptr : val;
+            if (operation != patch_operations::remove && value_ptr == nullptr)
+            {
+                return false;
+            }
+
+            switch (operation)
+            {
+                case patch_operations::replace:
+                {
+                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
+                    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, steps);
+                    // 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, steps);
+                    if (parent == nullptr)
+                    {
+                        return false;
+                    }
+                    token.clear();
+                    token.append(path.data() + slash + 1, path.size() - slash - 1);
+                    if (token.find_first_of('~') != string_t::npos)
+                    {
+                        return false;
+                    }
+
+                    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;
+            }
+        };
+
+        // the name of the first member to check
+        const string_t op_member = "op";
+
         // iterate and apply the operations
         for (const auto& val : json_patch)
         {
             // wrapper to get a value for an operation
-            const auto get_value = [&val](const string_t& op,
-                                          const string_t& member,
+            const auto get_value = [&val](basic_json * found,
+                                          const string_t& op,
+                                          const char* member,
                                           bool string_type) -> basic_json &
             {
-                // find value
-                auto it = val.m_data.m_value.object->find(member);
-
                 // check if the desired value is present
-                if (JSON_HEDLEY_UNLIKELY(it == val.m_data.m_value.object->end()))
+                if (JSON_HEDLEY_UNLIKELY(found == nullptr))
                 {
-                    detail::throw_patch_missing_member(op, member, false, val);
+                    detail::throw_patch_missing_member(op, string_t(member), false, val);
                 }
 
                 // check if the result is of type string
-                if (JSON_HEDLEY_UNLIKELY(string_type && !it->second.is_string()))
+                if (JSON_HEDLEY_UNLIKELY(string_type && !found->is_string()))
                 {
-                    detail::throw_patch_missing_member(op, member, true, val);
+                    detail::throw_patch_missing_member(op, string_t(member), true, val);
                 }
 
                 // no error: return value
-                return it->second;
+                return *found;
             };
 
             // type check: every element of the array must be an object
@@ -7078,16 +7293,30 @@ public:
                 detail::throw_patch_not_an_array(val);
             }
 
+            const patch_members members = find_patch_members(val, std::is_same<typename object_t::key_type, string_t> {});
+
             // 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 string_t& op = *get_value(members.op, op_member, "op", true).m_data.m_value.string;
+            const string_t& path = *get_value(members.path, 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, members, val))
+            {
+                continue;
+            }
+
+            // the general operations may change any container
+            steps.reset();
             json_pointer ptr(path);
 
-            switch (get_op(op))
+            switch (operation)
             {
                 case patch_operations::add:
                 {
-                    operation_add(ptr, get_value("add", "value", false));
+                    operation_add(ptr, get_value(members.value, "add", "value", false));
                     break;
                 }
 
@@ -7100,13 +7329,13 @@ public:
                 case patch_operations::replace:
                 {
                     // the "path" location must exist - use at()
-                    result.at(ptr) = get_value("replace", "value", false);
+                    result.at(ptr) = get_value(members.value, "replace", "value", false);
                     break;
                 }
 
                 case patch_operations::move:
                 {
-                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "move", "from", true).m_data.m_value.string;
                     json_pointer from_ptr(from_path);
 
                     if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -7129,7 +7358,7 @@ public:
 
                 case patch_operations::copy:
                 {
-                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "copy", "from", true).m_data.m_value.string;
                     const json_pointer from_ptr(from_path);
 
                     // The copy is functionally identical to an "add"
@@ -7147,7 +7376,7 @@ public:
                     {
                         // check if "value" matches the one at "path"
                         // the "path" location must exist - use at()
-                        success = (result.at(ptr) == get_value("test", "value", false));
+                        success = (result.at(ptr) == get_value(members.value, "test", "value", false));
                     }
                     JSON_INTERNAL_CATCH (out_of_range&)
                     {
@@ -7191,7 +7420,8 @@ 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 +7465,130 @@ 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)
+    {
+        // the keys are distinct and in the order of std::map; the members are
+        // added without looking them up
+        basic_json entry(value_t::object);
+        auto& members = *entry.m_data.m_value.object;
+        auto hint = members.end();
+        json_pointer::add_flattened(members, hint, string_t("op"), string_t(op), detail::priority_tag<2> {});
+        json_pointer::add_flattened(members, hint, string_t("path"), path, detail::priority_tag<2> {});
+        if (value != nullptr)
+        {
+            json_pointer::add_flattened(members, hint, string_t("value"), *value, detail::priority_tag<2> {});
+        }
+        entry.set_parents();
+        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(
+        diff_operation(result, "add", path, &value);
+    }
+
+    /// @brief the values in the target of the members of a source object, for the
+    ///        objects of up to 16 members without allocating
+    ///
+    /// reserve() must be called with the number of values that are added.
+    class diff_target_values
+    {
+      public:
+        diff_target_values() noexcept
+            : inline_items() // NOLINT(readability-redundant-member-init)
+        {}
+        diff_target_values(const diff_target_values&) = delete;
+        diff_target_values& operator=(const diff_target_values&) = delete;
+        diff_target_values(diff_target_values&&) = delete;
+        diff_target_values& operator=(diff_target_values&&) = delete;
+        ~diff_target_values() = default;
+
+        void reserve(const std::size_t n)
         {
-            {"op", "add"}, {"path", path}, {"value", value}
-        });
+            if (n > inline_size)
+            {
+                large.resize(n);
+                items = large.data();
+            }
+        }
+
+        void push_back(const basic_json* value)
+        {
+            items[count++] = value;
+        }
+
+        std::size_t size() const noexcept
+        {
+            return count;
+        }
+
+        const basic_json* operator[](const std::size_t i) const noexcept
+        {
+            return items[i];
+        }
+
+      private:
+        static constexpr std::size_t inline_size = 16;
+        /// the values of a small object, left uninitialized: they are written
+        /// before they are read (the user-provided constructor keeps the
+        /// member initialization, which -Weffc++ asks for, from zeroing them)
+        struct uninitialized_items : std::array<const basic_json*, inline_size>
+        {
+            uninitialized_items() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+        };
+
+        std::size_t count = 0;
+        std::vector<const basic_json*> large{}; // NOLINT(readability-redundant-member-init)
+        uninitialized_items inline_items;
+        const basic_json** items = inline_items.data();
+    };
+
+    /// @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 +7601,18 @@ 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 +7621,7 @@ 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());
         }
     }
 
@@ -7303,15 +7637,31 @@ public:
     operations that remove every member of @a source and add every member of
     @a target to @a result, and returns false.
     */
+    template<typename TargetValues>
     static bool diff_object_keys(basic_json& result, const basic_json& source, const basic_json& target,
-                                 const string_t& path, std::vector<const basic_json*>& target_values,
+                                 const string_t& path, TargetValues& target_values,
                                  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 +7674,12 @@ 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 +7745,13 @@ 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 +7768,7 @@ 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 +7777,7 @@ 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 +7786,9 @@ 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 +7796,7 @@ 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 +7808,9 @@ 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 +7829,8 @@ 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 +7838,9 @@ 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;
                 }
@@ -7483,7 +7853,7 @@ public:
 
             case value_t::object:
             {
-                std::vector<const basic_json*> target_values;
+                diff_target_values target_values;
                 std::vector<const_iterator> added;
                 if (diff_object_keys(result, source, target, path, target_values, added))
                 {
@@ -7492,6 +7862,7 @@ 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 +7870,15 @@ 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 +8009,7 @@ 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 +8032,7 @@ 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 +8042,7 @@ 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;
                 }

include/nlohmann/json_view.hpp

@@ -379,7 +379,7 @@ class basic_json_view
     /// the member with this key converted to T, or the default value if there
     /// is no such member (the first one, should the key occur more than
     /// once). Throws type_error.306 if this is not an object.
-    template < typename T, typename std::enable_if < !std::is_same<typename std::decay<T>::type, const char*>::value, int >::type = 0 >
+    template<typename T, typename std::enable_if<detail::view::is_non_c_string_default<T>::value, int>::type = 0>
     T value(string_view_t key, const T& default_value) const
     {
         if (NLOHMANN_VIEW_UNLIKELY(!is_object()))
@@ -398,7 +398,7 @@ class basic_json_view
     /// the value a JSON pointer refers to converted to T, or the default
     /// value if the pointer cannot be resolved. Throws type_error.306 if this
     /// is neither an object nor an array.
-    template < typename T, typename std::enable_if < !std::is_same<typename std::decay<T>::type, const char*>::value, int >::type = 0 >
+    template<typename T, typename std::enable_if<detail::view::is_non_c_string_default<T>::value, int>::type = 0>
     T value(const json_pointer& ptr, const T& default_value) const
     {
         if (NLOHMANN_VIEW_UNLIKELY(!is_structured()))
@@ -750,8 +750,10 @@ class basic_json_view
     {
         if (editable() && m_doc->edits != nullptr)
         {
-            // positions of moved and new values are not source offsets
-            return m_node == m_doc->nodes ? m_doc->size + m_doc->edits->text_used : 64;
+            // positions of moved and new values are not source offsets; the root
+            // (the first node) spans the source text and every edit
+            const bool is_root = m_node == m_doc->nodes;
+            return is_root ? m_doc->size + m_doc->edits->text_used : 64;
         }
         const node* const end = m_doc->nodes + m_doc->node_count;
         if ((m_node->flags & detail::view::node_flags::storage) != 0)
@@ -798,7 +800,7 @@ class basic_json_view
         return (m_node->flags & detail::view::node_flags::is_true) != 0;
     }
 
-    template < typename T, typename std::enable_if < std::is_arithmetic<T>::value && !std::is_same<T, bool>::value, int >::type = 0 >
+    template<typename T, typename std::enable_if<detail::view::is_non_bool_arithmetic<T>::value, int>::type = 0>
     NLOHMANN_VIEW_ALWAYS_INLINE T get_impl(detail::view::value_tag<T> /*unused*/, detail::priority_tag<2> /*unused*/) const
     {
         if (NLOHMANN_VIEW_UNLIKELY(m_node == nullptr))
@@ -1038,7 +1040,7 @@ class basic_json_document
         return m_data ? m_data->node_count : 0;
     }
 
-    /// bytes held by the document (index, decoded strings, owned text or image)
+    /// bytes held by the document (index, decoded strings, and owned text or image)
     std::size_t memory_usage() const noexcept
     {
         if (!m_data)
@@ -1186,7 +1188,7 @@ class basic_json_document
     }
 
     /// assign an existing array element; returns a view of it
-    template < typename I, typename V, typename std::enable_if < std::is_integral<I>::value && !std::is_same<I, bool>::value, int >::type = 0 >
+    template<typename I, typename V, typename std::enable_if<detail::view::is_non_bool_integral<I>::value, int>::type = 0>
     view_type set(view_type array, I idx, V && value)
     {
         return editor().set(array, index(idx), std::forward<V>(value));
@@ -1208,14 +1210,14 @@ class basic_json_document
         const auto& token = ptr.back();
         if (parent.is_null())
         {
-            const bool digits = std::all_of(token.begin(), token.end(), [](const char c)
-            {
-                return c >= '0' && c <= '9';
-            });
             if (token == "-")
             {
                 return push_back(parent, std::forward<V>(value));
             }
+            const bool digits = std::all_of(token.begin(), token.end(), [](const char c)
+            {
+                return c >= '0' && c <= '9';
+            });
             if (digits)
             {
                 // (an invalid index is an error before the parent changes)
@@ -1253,7 +1255,7 @@ class basic_json_document
 
     /// insert into an array before position idx (idx <= size()); returns a
     /// view of the new element
-    template < typename I, typename V, typename std::enable_if < std::is_integral<I>::value && !std::is_same<I, bool>::value, int >::type = 0 >
+    template<typename I, typename V, typename std::enable_if<detail::view::is_non_bool_integral<I>::value, int>::type = 0>
     view_type insert(view_type array, I idx, V && value)
     {
         return editor().insert(array, index(idx), std::forward<V>(value));
@@ -1266,7 +1268,7 @@ class basic_json_document
     }
 
     /// remove an array element
-    template < typename I, typename std::enable_if < std::is_integral<I>::value && !std::is_same<I, bool>::value, int >::type = 0 >
+    template<typename I, typename std::enable_if<detail::view::is_non_bool_integral<I>::value, int>::type = 0>
     void erase(view_type array, I idx)
     {
         editor().erase(array, index(idx));

single_include/nlohmann/json_fwd.hpp

@@ -71,6 +71,14 @@
     #define JSON_USE_OBJECTS_FOR_ENUM_KEYED_MAPS 0
 #endif
 
+#ifndef JSON_DISABLE_STREAM_BULK_SCAN
+    #define JSON_DISABLE_STREAM_BULK_SCAN 0
+#endif
+
+#ifndef JSON_DISABLE_UNLOCKED_FILE_IO
+    #define JSON_DISABLE_UNLOCKED_FILE_IO 0
+#endif
+
 #if JSON_DIAGNOSTICS
     #define NLOHMANN_JSON_ABI_TAG_DIAGNOSTICS _diag
 #else
@@ -119,6 +127,24 @@
     #define NLOHMANN_JSON_ABI_TAG_OBJECTS_FOR_ENUM_KEYED_MAPS
 #endif
 
+#if JSON_DISABLE_STREAM_BULK_SCAN
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN _dsbs
+#else
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_STREAM_BULK_SCAN
+#endif
+
+#if JSON_DISABLE_UNLOCKED_FILE_IO
+    #define NLOHMANN_JSON_ABI_TAG_DISABLE_UNLOCKED_FILE_IO _dufio
+#else
+    #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
@@ -136,9 +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) json_abi ## a ## b ## c ## d ## e ## f ## g ## h ## i ## j
-#define NLOHMANN_JSON_ABI_TAGS_CONCAT(a, b, c, d, e, f, g, h, i, j) \
-    NLOHMANN_JSON_ABI_TAGS_CONCAT_EX(a, b, c, d, e, f, g, h, i, j)
+#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(                                   \
@@ -150,6 +176,9 @@
             NLOHMANN_JSON_ABI_TAG_STRICT_NUL_HANDLING,               \
             NLOHMANN_JSON_ABI_TAG_STRICT_BINARY_UTF8,                \
             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)
 

tests/CMakeLists.txt

@@ -285,6 +285,10 @@ if (JSON_TestSimdutf)
     endif()
 endif()
 
+# test-file_input checks from a second thread that parsing from a FILE* releases the file's lock
+find_package(Threads REQUIRED)
+json_test_set_test_options("test-file_input;test-file_input_unlocked_io_disabled" LINK_LIBRARIES Threads::Threads)
+
 # *DO* use json_test_set_test_options() above this line
 
 json_test_should_build_32bit_test(json_32bit_test json_32bit_test_only "${JSON_32bitTest}")
@@ -369,6 +373,52 @@ 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
+)
+json_test_add_test_for(src/unit-stream-bulk-scan.cpp
+    NAME test-stream-bulk-scan_disabled
+    MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
+)
+
+# test reading from a std::istream again with JSON_PRECISE_STREAM_POSITION
+json_test_set_test_options(test-stream-bulk-scan_precise
+    COMPILE_DEFINITIONS JSON_PRECISE_STREAM_POSITION=1
+)
+json_test_add_test_for(src/unit-stream-bulk-scan.cpp
+    NAME test-stream-bulk-scan_precise
+    MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
+)
+
+# test reading from a FILE* again with the locking stdio functions
+json_test_set_test_options(test-file_input_unlocked_io_disabled
+    COMPILE_DEFINITIONS JSON_DISABLE_UNLOCKED_FILE_IO=1
+)
+json_test_add_test_for(src/unit-file_input.cpp
+    NAME test-file_input_unlocked_io_disabled
+    MAIN test_main CXX_STANDARDS ${test_cxx_standards} ${test_force}
+)
+
 # the json_view parser again with the portable string scanning instead of
 # NEON/SSE2, and on x86-64 with the SSSE3 UTF-8 check (JSON_VIEW_USE_SSSE3)
 json_test_set_test_options(test-json_view_builder_portable

tests/abi/config/CMakeLists.txt

@@ -22,6 +22,9 @@ foreach(test default noversion)
         JSON_DIAGNOSTICS=1
         JSON_DIAGNOSTIC_POSITIONS=1
         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

@@ -52,6 +52,18 @@ TEST_CASE("default namespace")
         expected += "_ekmo";
 #endif
 
+#if JSON_DISABLE_STREAM_BULK_SCAN
+        expected += "_dsbs";
+#endif
+
+#if JSON_DISABLE_UNLOCKED_FILE_IO
+        expected += "_dufio";
+#endif
+
+#ifdef JSON_NO_SIMD
+        expected += "_ns";
+#endif
+
 #ifdef JSON_VIEW_NO_SIMD
         expected += "_vns";
 #endif

tests/abi/config/noversion.cpp

@@ -53,6 +53,18 @@ TEST_CASE("default namespace without version component")
         expected += "_ekmo";
 #endif
 
+#if JSON_DISABLE_STREAM_BULK_SCAN
+        expected += "_dsbs";
+#endif
+
+#if JSON_DISABLE_UNLOCKED_FILE_IO
+        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);
+}