template for

deserialization (with static reflection) from objects and arrays.

Co-authored-by: Daniel Lemire <dlemire@lemire.me>

.clangd

@@ -25,6 +25,7 @@ CompileFlags:
 Diagnostics:
     Suppress:
         - pp_including_mainfile_in_preamble
+        - unused-includes
 ---
 # Amalgamated files that require or partly define an implementation
 If:

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1 @@
+blank_issues_enabled: false

.github/ISSUE_TEMPLATE/standard-issue-template.md

@@ -18,7 +18,7 @@ We do not make changes to simdjson without clearly identifiable benefits, which
 
 Is your issue:
 
-1. A bug report? If so, please point at a reproducible test. Indicate whether you are willing or able to provide a bug fix as a pull request.
+1. A bug report? If so, please point at a reproducible test. Indicate whether you are willing or able to provide a bug fix as a pull request. As a matter of policy, we do not consider a compiler warning to be a bug.
 
 2. A build issue? If so, provide all possible details regarding your system configuration. If we cannot reproduce your issue, we cannot fix it.
 

.github/workflows/aarch64.yml

@@ -0,0 +1,29 @@
+name: Ubuntu ppc64le (GCC 11)
+
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+    branches:
+      - master
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: uraimo/run-on-arch-action@v3
+        name: Test
+        id: runcmd
+        with:
+          arch: aarch64
+          distro: ubuntu_latest
+          githubToken: ${{ github.token }}
+          install: |
+            apt-get update -q -y
+            apt-get install -y cmake make g++
+          run: |
+            cmake -DSIMDJSON_SANITIZE_UNDEFINED=ON -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -B build
+            cmake --build build -j=2
+            ctest --output-on-failure --test-dir build

.github/workflows/alpine.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/cifuzz.yml

@@ -17,7 +17,7 @@ jobs:
         fuzz-seconds: 600
         dry-run: false
     - name: Upload Crash
-      uses: actions/upload-artifact@v1
+      uses: actions/upload-artifact@v4
       if: failure() && steps.build.outcome == 'success'
       with:
         name: artifacts

.github/workflows/documentation.yml

@@ -1,9 +1,8 @@
 name: Doxygen GitHub Pages
 
 on:
-  push:
+  release:
-    branches:
+    types: [created]
-      - master
   # Allows you to run this workflow manually from the Actions tab
   workflow_dispatch:
 

.github/workflows/fix-trailing-whitespace.yml

@@ -24,7 +24,7 @@ jobs:
              echo "no trailing whitespace found, good!"
           fi
       - name: Archive whitespace patch
-        uses: actions/upload-artifact@v2
+        uses: actions/upload-artifact@v4
         if: always()
         with:
           name: whitespace-patch

.github/workflows/fuzzers.yml

@@ -24,7 +24,7 @@ jobs:
       implementations: haswell westmere fallback
       UBSAN_OPTIONS: halt_on_error=1
       MAXLEN: -max_len=4000
-      CLANGVERSION: 15
+      CLANGVERSION: 19
       # which optimization level to use for the sanitizer build (see build_fuzzer.variants.sh)
       OPTLEVEL: -O3
 
@@ -34,18 +34,18 @@ jobs:
         sudo apt update
         sudo apt-get install --quiet ninja-build valgrind zip unzip lsb-release wget software-properties-common gnupg
         wget https://apt.llvm.org/llvm.sh
-        sudo apt-get purge --auto-remove llvm python3-lldb-14 llvm-14
+        sudo apt-get purge --auto-remove llvm python3-lldb-15 llvm-15
         chmod +x llvm.sh
         sudo ./llvm.sh $CLANGVERSION
 
     - uses: actions/checkout@v4
 
-    - uses: actions/cache@v3
+    - uses: actions/cache@v4
       with:
         path: dependencies/.cache
         key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
 
-    - uses: actions/cache@v3
+    - uses: actions/cache@v4
       id: cache-corpus
       with:
         path: out/
@@ -125,7 +125,7 @@ jobs:
         done
 
     - name: Save the corpus as a github artifact
-      uses: actions/upload-artifact@v3
+      uses: actions/upload-artifact@v4
       with:
         name: corpus
         path: corpus.tar
@@ -148,7 +148,7 @@ jobs:
       run: tar cf valgrind.tar valgrind-*.txt
 
     - name: Save valgrind output as a github artifact
-      uses: actions/upload-artifact@v3
+      uses: actions/upload-artifact@v4
       if: always()
       with:
         name: valgrindresults
@@ -156,7 +156,7 @@ jobs:
         if-no-files-found: ignore
 
     - name: Archive any crashes as an artifact
-      uses: actions/upload-artifact@v3
+      uses: actions/upload-artifact@v4
       if: always()
       with:
         name: crashes

.github/workflows/loongarch64.yml

@@ -0,0 +1,50 @@
+name: LoongArch64-CI
+
+on: [push, pull_request]
+
+jobs:
+  loongarch64:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        platform:
+          - { toolchain-version: 2023.08.08 }
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install build requirements
+        run: |
+          sudo apt-get update -y
+          sudo apt-get install -y --no-install-recommends cmake
+
+      - uses: actions/cache/restore@v4
+        id: restore-cache
+        with:
+          path: /opt/cross-tools
+          key: loongarch64-${{ matrix.platform.toolchain-version }}
+
+      - name: Download LoongArch64 gcc+glibc toolchain
+        if: ${{ !steps.restore-cache.outputs.cache-hit }}
+        run: |
+          url="https://github.com/loongson/build-tools/releases/download/${{ matrix.platform.toolchain-version }}/x86_64-cross-tools-loongarch64-gcc-libc.tar.xz"
+
+          wget "$url" -O /tmp/toolchain.tar.xz
+
+          mkdir -p /opt
+          tar -C /opt -x -f /tmp/toolchain.tar.xz
+
+      - uses: actions/cache/save@v3
+        if: ${{ !steps.restore-cache.outputs.cache-hit }}
+        with:
+          path: /opt/cross-tools
+          key: loongarch64-${{ matrix.platform.toolchain-version }}
+
+      - name: setup Loongarch64 build environment
+        run: |
+          echo "/opt/cross-tools/bin" >> $GITHUB_PATH
+          echo "CC=loongarch64-unknown-linux-gnu-gcc" >> $GITHUB_ENV
+          echo "CXX=loongarch64-unknown-linux-gnu-g++" >> $GITHUB_ENV
+      - name: configure
+        run: cmake -B build -DCMAKE_SYSTEM_PROCESSOR=loongarch64 -DARCH=lonngarch64 -DCMAKE_SYSTEM_NAME=Linux -DCMAKE_C_COMPILER=loongarch64-unknown-linux-gnu-gcc -DCMAKE_CXX_COMPILER=loongarch64-unknown-linux-gnu-g++
+      - name: build
+        run: cmake --build build

.github/workflows/macos.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: macos-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/msys2-clang.yml

@@ -20,12 +20,15 @@ jobs:
           - msystem: "MINGW64"
             install: mingw-w64-x86_64-libxml2 mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-clang
             type: Debug
+          - msystem: "MINGW64"
+            install: mingw-w64-x86_64-libxml2 mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-clang
+            type: RelWithDebInfo
     env:
       CMAKE_GENERATOR: Ninja
 
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/msys2.yml

@@ -22,12 +22,15 @@ jobs:
           - msystem: "MINGW64"
             install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
             type: Debug
+          - msystem: "MINGW64"
+            install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
+            type: RelWithDebInfo
     env:
       CMAKE_GENERATOR: Ninja
 
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/power-fuzz.yml

@@ -1,71 +0,0 @@
-name: short fuzz on the power arch
-
-on:
-  push:
-    branches: [ master ]
-  pull_request:
-    branches: [ master ]
-
-jobs:
-  armv7_job:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    # The host should always be Linux
-    runs-on: ubuntu-20.04
-    name: Build on ubuntu-20.04 ppc64le
-    steps:
-      - uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2.0.5
-        name: Run commands
-        id: runcmd
-        env:
-          DEBIAN_FRONTEND: noninteractive
-        with:
-          arch: ppc64le
-          distro: buster
-
-          # Not required, but speeds up builds by storing container images in
-          # a GitHub package registry.
-          githubToken: ${{ github.token }}
-
-          run: |
-            export CLANGSUFFIX="-7"
-            apt-get -qq update
-            apt-get install -q -y clang-7 libfuzzer-7-dev git wget zip ninja-build gnupg software-properties-common
-            wget -q -O - "https://raw.githubusercontent.com/simdjson/debian-ppa/master/key.gpg" | apt-key add -
-            apt-add-repository "deb https://raw.githubusercontent.com/simdjson/debian-ppa/master simdjson main"
-            apt-get -qq update
-            apt-get purge cmake cmake-data
-            apt-get -t simdjson -y install cmake
-            mkdir -p build ; cd build
-            cmake .. -GNinja \
-                  -DCMAKE_CXX_COMPILER=clang++$CLANGSUFFIX \
-                  -DCMAKE_C_COMPILER=clang$CLANGSUFFIX \
-                  -DBUILD_SHARED_LIBS=OFF \
-                  -DSIMDJSON_DEVELOPER_MODE=ON \
-                  -DSIMDJSON_ENABLE_FUZZING=On \
-                  -DSIMDJSON_COMPETITION=OFF \
-                  -DSIMDJSON_GOOGLE_BENCHMARKS=OFF \
-                  -DSIMDJSON_DISABLE_DEPRECATED_API=On \
-                  -DSIMDJSON_FUZZ_LDFLAGS=-lFuzzer \
-                  -DCMAKE_CXX_FLAGS="-fsanitize=fuzzer-no-link -DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION=" \
-                  -DCMAKE_C_FLAGS="-fsanitize=fuzzer-no-link" \
-                  -DCMAKE_BUILD_TYPE=Release \
-                  -DSIMDJSON_FUZZ_LINKMAIN=Off
-            cd ..
-            builddir=build
-            cmake --build $builddir
-            wget -O corpus.tar.gz https://readonly:readonly@www.pauldreik.se/fuzzdata/index.php?project=simdjson
-            tar xf corpus.tar.gz
-            fuzzernames=$(cmake --build $builddir --target print_all_fuzzernames |tail -n1)
-            for fuzzer in $fuzzernames ; do
-               exe=$builddir/fuzz/$fuzzer
-               shortname=$(echo $fuzzer |cut -f2- -d_)
-               echo found fuzzer $shortname with executable $exe
-               mkdir -p out/$shortname
-               others=$(find out -type d -not -name $shortname -not -name out -not -name cmin)
-               $exe -max_total_time=20  -max_len=4000 out/$shortname $others
-               echo "*************************************************************************"
-            done
-            echo "all is good, no errors found in any of these fuzzers: $fuzzernames"

.github/workflows/ppc64.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2
+      - uses: uraimo/run-on-arch-action@v3
         name: Test
         id: runcmd
         with:
@@ -24,6 +24,6 @@ jobs:
             apt-get update -q -y
             apt-get install -y cmake make g++
           run: |
-            cmake -DCMAKE_BUILD_TYPE=Release -B build
+            cmake -DCMAKE_BUILD_TYPE=Release -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF  -B build
             cmake --build build -j=2
             ctest --output-on-failure --test-dir build

.github/workflows/riscv64.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2
+      - uses: uraimo/run-on-arch-action@v3
         name: Test
         id: runcmd
         with:
@@ -24,6 +24,6 @@ jobs:
             apt-get update -q -y
             apt-get install -y cmake make g++
           run: |
-            cmake -DCMAKE_BUILD_TYPE=Release -B build
+            cmake -DCMAKE_BUILD_TYPE=Release -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF  -B build
             cmake --build build -j=2
             ctest --output-on-failure --test-dir build

.github/workflows/s390x.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2
+      - uses: uraimo/run-on-arch-action@v3
         name: Test
         id: runcmd
         with:
@@ -24,6 +24,6 @@ jobs:
             apt-get update -q -y
             apt-get install -y cmake make g++
           run: |
-            cmake -DCMAKE_BUILD_TYPE=Release -B build
+            cmake -DCMAKE_BUILD_TYPE=Release  -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -B build
             cmake --build build -j=2
             ctest --output-on-failure --test-dir build

.github/workflows/ubuntu20-checkperf.yml

@@ -16,7 +16,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu20-gcc8.yml

@@ -13,7 +13,7 @@ jobs:
       CC: gcc-8
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu20-noexcept.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu20-nothread.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu20-sani.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
@@ -28,7 +28,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu20.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-clang13.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-clang14.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-cxx20.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-gcc12.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-gcc13.yml

@@ -1,23 +0,0 @@
-name: Ubuntu 22.04 CI (GCC 13)
-
-on: [push, pull_request]
-
-jobs:
-  ubuntu-build:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-22.04
-    steps:
-      - uses: actions/checkout@v4
-      - uses: actions/cache@v3
-        with:
-          path: dependencies/.cache
-          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-      - name: Use cmake
-        run: |
-          mkdir build &&
-          cd build &&
-          CXX=g++-13 cmake -DSIMDJSON_DEVELOPER_MODE=ON ..  &&
-          cmake --build .   &&
-          ctest --output-on-failure -LE explicitonly -j

.github/workflows/ubuntu22-glibcxxassertions.yml

@@ -9,7 +9,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu22-threadsani.yml

@@ -1,4 +1,4 @@
-name: Ubuntu 22.04 CI (GCC 11) with Thread Sanitizer
+name: Ubuntu 22.04 CI (GCC 12) with Thread Sanitizer
 
 on: [push, pull_request]
 
@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
@@ -18,7 +18,7 @@ jobs:
         run: |
           mkdir build &&
           cd build &&
-          cmake  -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_SANITIZE_THREADS=ON ..  &&
+          CXX=g++-12 cmake  -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_SANITIZE_THREADS=ON ..  &&
           cmake --build . --target document_stream_tests  --target ondemand_document_stream_tests --target parse_many_test  &&
           ctest --output-on-failure  -R parse_many_test  &&
           ctest --output-on-failure  -R document_stream_tests

.github/workflows/ubuntu22.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/cache@v3
+      - uses: actions/cache@v4
         with:
           path: dependencies/.cache
           key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.github/workflows/ubuntu24-cxx20.yml

@@ -0,0 +1,22 @@
+name: Ubuntu 24.04 CI (CXX 20)
+
+on: [push, pull_request]
+jobs:
+  ubuntu-build:
+    if: >-
+      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
+      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
+    runs-on: ubuntu-24.04
+    strategy:
+      matrix:
+        cxx: [g++-13, clang++-16]
+    steps:
+      - uses: actions/checkout@a5ac7e51b41094c92402da3b24376905380afc29 # v4.1.6
+      - name: Prepare
+        run: cmake -DSIMDJSON_CXX_STANDARD=20 -DSIMDJSON_DEVELOPER_MODE=ON  -B build
+        env:
+          CXX: ${{matrix.cxx}}
+      - name: Build
+        run: cmake --build build -j=2
+      - name: Test
+        run: ctest --output-on-failure --test-dir build

.github/workflows/ubuntu24.yml

@@ -0,0 +1,25 @@
+name: Ubuntu 24.04 CI
+
+on: [push, pull_request]
+jobs:
+  ubuntu-build:
+    if: >-
+      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
+      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
+    runs-on: ubuntu-24.04
+    strategy:
+      matrix:
+        shared: [ON, OFF]
+        cxx: [g++-13, clang++-16]
+        sanitizer: [ON, OFF]
+        build_type: [RelWithDebInfo, Debug, Release]
+    steps:
+      - uses: actions/checkout@a5ac7e51b41094c92402da3b24376905380afc29 # v4.1.6
+      - name: Prepare
+        run: cmake -DCMAKE_BUILD_TYPE=${{matrix.build_type}} -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_SANITIZE=${{matrix.sanitizer}} -DBUILD_SHARED_LIBS=${{matrix.shared}}  -B build
+        env:
+          CXX: ${{matrix.cxx}}
+      - name: Build
+        run: cmake --build build -j=2
+      - name: Test
+        run: ctest --output-on-failure --test-dir build

.github/workflows/vs17-arm-ci.yml

@@ -10,12 +10,12 @@ jobs:
        fail-fast: false
        matrix:
          include:
-           - {arch: ARM}
            - {arch: ARM64}
+           - {arch: ARM64EC}
      steps:
        - name: checkout
          uses: actions/checkout@v4
        - name: Use cmake
          run: |
-           cmake -A ${{ matrix.arch }} -DCMAKE_CROSSCOMPILING=1 -DSIMDJSON_DEVELOPER_MODE=ON -D SIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_EXCEPTIONS=OFF -B build  &&
+           cmake -A ${{ matrix.arch }} -DCMAKE_SYSTEM_VERSION="10.0.22621.0" -DCMAKE_CROSSCOMPILING=1 -DSIMDJSON_DEVELOPER_MODE=ON -D SIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_EXCEPTIONS=OFF -B build  &&
            cmake --build build --verbose

.github/workflows/vs17-ci.yml

@@ -13,10 +13,12 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - {gen: Visual Studio 17 2022, arch: Win32, shared: ON}
+          - {gen: Visual Studio 17 2022, arch: Win32, shared: ON, build_type: Release}
-          - {gen: Visual Studio 17 2022, arch: Win32, shared: OFF}
+          - {gen: Visual Studio 17 2022, arch: Win32, shared: OFF, build_type: Release}
-          - {gen: Visual Studio 17 2022, arch: x64, shared: ON}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: ON, build_type: Release}
-          - {gen: Visual Studio 17 2022, arch: x64, shared: OFF}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: OFF, build_type: Debug}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: OFF, build_type: Release}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: OFF, build_type: RelWithDebInfo}
     steps:
     - name: checkout
       uses: actions/checkout@v4
@@ -24,21 +26,15 @@ jobs:
       run: |
         cmake -G "${{matrix.gen}}" -A ${{matrix.arch}} -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -DBUILD_SHARED_LIBS=${{matrix.shared}} -B build
     - name: Build Debug
-      run: cmake --build build --config Debug --verbose
+      run: cmake --build build --config ${{matrix.build_type}} --verbose
-    - name: Build Release
+    - name: Run tests
-      run: cmake --build build --config Release --verbose
-    - name: Run Release tests
       run: |
         cd build
-        ctest -C Release  -LE explicitonly  --output-on-failure
+        ctest -C ${{matrix.build_type}}  -LE explicitonly  --output-on-failure
-    - name: Run Debug tests
-      run: |
-        cd build
-        ctest -C Debug  -LE explicitonly  --output-on-failure
     - name: Install
       run: |
-        cmake --install build --config Release
+        cmake --install build --config ${{matrix.build_type}}
     - name: Test Installation
       run: |
         cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -B build_install_test tests/installation_tests/find
-        cmake --build build_install_test --config Release
+        cmake --build build_install_test --config ${{matrix.build_type}}

.github/workflows/vs17-clang-ci-cxx20.yml

@@ -0,0 +1,37 @@
+name: VS17-CLANG-CI
+
+on: [push, pull_request]
+
+jobs:
+  ci:
+    if: >-
+      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
+      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
+    name: windows-vs17
+    runs-on: windows-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: Debug}
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: Release}
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: RelWithDebInfo}
+    steps:
+    - name: checkout
+      uses: actions/checkout@v4
+    - name: Configure
+      run: |
+        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -T ClangCL -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -B build
+    - name: Build
+      run: cmake --build build --config ${{matrix.build_type}} --verbose
+    - name: Run tests
+      run: |
+        cd build
+        ctest -C ${{matrix.build_type}}  -LE explicitonly  --output-on-failure
+    - name: Install
+      run: |
+        cmake --install build --config ${{matrix.build_type}}
+    - name: Test Installation
+      run: |
+        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -B build_install_test tests/installation_tests/find
+        cmake --build build_install_test --config ${{matrix.build_type}}

.github/workflows/vs17-clang-ci.yml

@@ -13,29 +13,25 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - {gen: Visual Studio 17 2022, arch: x64}
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: Debug}
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: Release}
+          - {gen: Visual Studio 17 2022, arch: x64, build_type: RelWithDebInfo}
     steps:
     - name: checkout
       uses: actions/checkout@v4
     - name: Configure
       run: |
         cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -T ClangCL -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -B build
-    - name: Build Debug
+    - name: Build
-      run: cmake --build build --config Debug --verbose
+      run: cmake --build build --config ${{matrix.build_type}} --verbose
-    - name: Build Release
+    - name: Run tests
-      run: cmake --build build --config Release --verbose
-    - name: Run Release tests
       run: |
         cd build
-        ctest -C Release  -LE explicitonly  --output-on-failure
+        ctest -C ${{matrix.build_type}}  -LE explicitonly  --output-on-failure
-    - name: Run Debug tests
-      run: |
-        cd build
-        ctest -C Debug  -LE explicitonly  --output-on-failure
     - name: Install
       run: |
-        cmake --install build --config Release
+        cmake --install build --config ${{matrix.build_type}}
     - name: Test Installation
       run: |
         cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -B build_install_test tests/installation_tests/find
-        cmake --build build_install_test --config Release
+        cmake --build build_install_test --config ${{matrix.build_type}}

.github/workflows/vs17-noexcept-ci.yml

@@ -8,7 +8,7 @@ jobs:
     runs-on: windows-latest
     steps:
     - uses: actions/checkout@v4
-    - uses: actions/cache@v3
+    - uses: actions/cache@v4
       with:
         path: dependencies/.cache
         key: ${{ hashFiles('dependencies/CMakeLists.txt') }}

.vscode/settings.json

@@ -109,6 +109,23 @@
         "numbers": "cpp",
         "semaphore": "cpp",
         "stop_token": "cpp",
-        "cfenv": "cpp"
+        "cfenv": "cpp",
+        "format": "cpp",
+        "xlocmes": "cpp",
+        "xlocmon": "cpp",
+        "xlocnum": "cpp",
+        "xloctime": "cpp",
+        "xutility": "cpp",
+        "coroutine": "cpp",
+        "xfacet": "cpp",
+        "xhash": "cpp",
+        "xiosbase": "cpp",
+        "xlocale": "cpp",
+        "xlocbuf": "cpp",
+        "xlocinfo": "cpp",
+        "xmemory": "cpp",
+        "xstring": "cpp",
+        "xtr1common": "cpp",
+        "xtree": "cpp"
     }
 }

CMakeLists.txt

@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.14)
 project(
     simdjson
     # The version number is modified by tools/release.py
-    VERSION 3.8.0
+    VERSION 3.12.2
     DESCRIPTION "Parsing gigabytes of JSON per second"
     HOMEPAGE_URL "https://simdjson.org/"
     LANGUAGES CXX C
@@ -20,10 +20,14 @@ string(
 # ---- Options, variables ----
 
 # These version numbers are modified by tools/release.py
-set(SIMDJSON_LIB_VERSION "21.0.0" CACHE STRING "simdjson library version")
+set(SIMDJSON_LIB_VERSION "25.0.0" CACHE STRING "simdjson library version")
-set(SIMDJSON_LIB_SOVERSION "21" CACHE STRING "simdjson library soversion")
+set(SIMDJSON_LIB_SOVERSION "25" CACHE STRING "simdjson library soversion")
 
-option(SIMDJSON_BUILD_STATIC_LIB "Build simdjson_static library along with simdjson" OFF)
+option(SIMDJSON_BUILD_STATIC_LIB "Build simdjson_static library along with simdjson (only makes sense if BUILD_SHARED_LIBS=ON)" OFF)
+if(SIMDJSON_BUILD_STATIC_LIB AND NOT BUILD_SHARED_LIBS)
+  message(WARNING "SIMDJSON_BUILD_STATIC_LIB only makes sense if BUILD_SHARED_LIBS is set to ON")
+  message(WARNING "You might be building and installing a two identical static libraries.")
+endif()
 
 option(SIMDJSON_ENABLE_THREADS "Link with thread support" ON)
 
@@ -39,6 +43,20 @@ if(SIMDJSON_DISABLE_DEPRECATED_API)
   )
 endif()
 
+if(${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.25.0")
+    option(SIMDJSON_STATIC_REFLECTION "Enables static reflection (experimental), requires C++26" OFF)
+else()
+    set(SIMDJSON_STATIC_REFLECTION OFF CACHE BOOL "Enables static reflection (experimental)" FORCE)
+    message(WARNING "SIMDJSON_STATIC_REFLECTION is disabled because your CMake version is below 3.25")
+endif()
+
+if(SIMDJSON_STATIC_REFLECTION)
+  simdjson_add_props(
+      target_compile_definitions PUBLIC
+      SIMDJSON_STATIC_REFLECTION=1
+  )
+endif()
+
 option(SIMDJSON_DEVELOPMENT_CHECKS "Enable development-time aids, such as \
 checks for incorrect API usage. Enabled by default in DEBUG." OFF)
 if(SIMDJSON_DEVELOPMENT_CHECKS)
@@ -51,6 +69,7 @@ endif()
 if(is_top_project)
   option(SIMDJSON_DEVELOPER_MODE "Enable targets for developing simdjson" OFF)
   option(BUILD_SHARED_LIBS "Build simdjson as a shared library" OFF)
+  option(SIMDJSON_SINGLEHEADER "Disable singleheader generation" ON)
 endif()
 
 include(cmake/handle-deprecations.cmake)
@@ -93,7 +112,16 @@ simdjson_add_props(
     PRIVATE "$<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/src>"
 )
 
-simdjson_add_props(target_compile_features PUBLIC cxx_std_11)
+if(SIMDJSON_STATIC_REFLECTION)
+  # We would like to require C++26, but no compiler supports that!
+  # This is a hack:
+  simdjson_add_props(
+      target_compile_options PUBLIC
+      -freflection -fexpansion-statements -stdlib=libc++ -std=c++26
+  )
+else()
+  simdjson_add_props(target_compile_features PUBLIC cxx_std_11)
+endif()
 
 # workaround for GNU GCC poor AVX load/store code generation
 if(
@@ -106,6 +134,24 @@ if(
   )
 endif()
 
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(loongarch64)$")
+    option(SIMDJSON_PREFER_LSX "Prefer LoongArch SX" ON)
+    include(CheckCXXCompilerFlag)
+    check_cxx_compiler_flag(-mlasx COMPILER_SUPPORTS_LASX)
+    check_cxx_compiler_flag(-mlsx COMPILER_SUPPORTS_LSX)
+    if(COMPILER_SUPPORTS_LASX AND NOT SIMDJSON_PREFER_LSX)
+        simdjson_add_props(
+            target_compile_options PRIVATE
+            -mlasx
+        )
+    elseif(COMPILER_SUPPORTS_LSX)
+        simdjson_add_props(
+            target_compile_options PRIVATE
+            -mlsx
+        )
+    endif()
+endif()
+
 # GCC and Clang have horrendous Debug builds when using SIMD.
 # A common fix is to use '-Og' instead.
 # bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54412
@@ -137,11 +183,13 @@ endif()
 include(CMakePackageConfigHelpers)
 include(GNUInstallDirs)
 
-install(
+if(SIMDJSON_SINGLEHEADER)
+  install(
       FILES singleheader/simdjson.h
       DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
       COMPONENT simdjson_Development
-)
+  )
+endif()
 
 install(
     TARGETS simdjson
@@ -185,6 +233,7 @@ if(SIMDJSON_BUILD_STATIC_LIB)
       TARGETS simdjson_static
       EXPORT simdjson_staticTargets
       ARCHIVE COMPONENT simdjson_Development
+      INCLUDES DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}"
   )
   install(
       EXPORT simdjson_staticTargets
@@ -261,6 +310,7 @@ enable_testing()
 add_custom_target(all_tests)
 
 add_subdirectory(windows)
+include(cmake/CPM.cmake)
 add_subdirectory(dependencies) ## This needs to be before tools because of cxxopts
 add_subdirectory(tools)  ## This needs to be before tests because of cxxopts
 
@@ -268,8 +318,9 @@ add_subdirectory(tools)  ## This needs to be before tests because of cxxopts
 # most of the data has been moved to https://github.com/simdjson/simdjson-data
 add_subdirectory(jsonexamples)
 
-
+if(SIMDJSON_SINGLEHEADER)
 add_subdirectory(singleheader)
+endif()
 
 
 

Doxyfile

@@ -38,7 +38,7 @@ PROJECT_NAME           = simdjson
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = "3.8.0"
+PROJECT_NUMBER         = "3.12.2"
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

HACKING.md

@@ -88,8 +88,8 @@ simdjson's source structure, from the top level, looks like this:
   * simdjson/ondemand.h: the `simdjson::ondemand` namespace. Includes all public ondemand classes.
     * simdjson/builtin.h: the `simdjson::builtin` namespace. Aliased to the most universal implementation available.
     * simdjson/builtin/ondemand.h: the `simdjson::builtin::ondemand` namespace.
-    * simdjson/arm64|fallback|haswell|icelake|ppc64|westmere/ondemand.h: the `simdjson::<implementation>::ondemand` namespace. on demand compiled for the specific implementation.
+    * simdjson/arm64|fallback|haswell|icelake|ppc64|westmere/ondemand.h: the `simdjson::<implementation>::ondemand` namespace. On-Demand compiled for the specific implementation.
-    * simdjson/generic/ondemand/*.h: individual on demand classes, generically written.
+    * simdjson/generic/ondemand/*.h: individual On-Demand classes, generically written.
       * simdjson/generic/ondemand/dependencies.h: dependencies on common, non-implementation-specific simdjson classes. This will be included before including amalgamated.h.
       * simdjson/generic/ondemand/amalgamated.h: all generic ondemand classes for an implementation.
 * **src:** The source files for non-inlined functionality (e.g. the architecture-specific parser
@@ -99,7 +99,7 @@ simdjson's source structure, from the top level, looks like this:
   * *.cpp: other misc. implementations, such as `simdjson::implementation` and the minifier.
   * arm64|fallback|haswell|icelake|ppc64|westmere.cpp: Architecture-specific parser implementations.
     * generic/*.h: `simdjson::<implementation>` namespace. Generic implementation of the parser, particularly the `dom_parser_implementation`.
-    * generic/stage1/*.h: `simdjson::<implementation>::stage1` namespace. Generic implementation of the simd-heavy tokenizer/indexer pass of the simdjson parser. Used for the On Demand interface
+    * generic/stage1/*.h: `simdjson::<implementation>::stage1` namespace. Generic implementation of the simd-heavy tokenizer/indexer pass of the simdjson parser. Used for the On-Demand interface
     * generic/stage2/*.h: `simdjson::<implementation>::stage2` namespace. Generic implementation of the tape creator, which consumes the index from stage 1 and actually parses numbers and string and such. Used for the DOM interface.
 
 Other important files and directories:

LICENSE

@@ -186,7 +186,7 @@
       same "printed page" as the copyright notice for easier
       identification within third-party archives.
 
-   Copyright 2018-2023 The simdjson authors
+   Copyright 2018-2025 The simdjson authors
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.

LICENSE-MIT

@@ -0,0 +1,18 @@
+Copyright 2018-2025 The simdjson authors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+the Software, and to permit persons to whom the Software is furnished to do so,
+subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

README.md

@@ -1,7 +1,7 @@
 
-[![Ubuntu 20.04 CI](https://github.com/simdjson/simdjson/workflows/Ubuntu%2020.04%20CI%20(GCC%209)/badge.svg)](https://simdjson.org/plots.html)
 [![Fuzzing Status](https://oss-fuzz-build-logs.storage.googleapis.com/badges/simdjson.svg)](https://bugs.chromium.org/p/oss-fuzz/issues/list?sort=-opened&can=1&q=proj:simdjson)
-[![][license img]][license]
+[![][license img]][license] [![][licensemit img]][licensemit]
+
 
 [![Doxygen Documentation](https://img.shields.io/badge/docs-doxygen-green.svg)](https://simdjson.github.io/simdjson/)
 
@@ -31,6 +31,7 @@ Table of Contents
 * [Documentation](#documentation)
 * [Godbolt](#godbolt)
 * [Performance results](#performance-results)
+* [Packages](#packages)
 * [Bindings and Ports of simdjson](#bindings-and-ports-of-simdjson)
 * [About simdjson](#about-simdjson)
 * [Funding](#funding)
@@ -46,6 +47,7 @@ Real-world usage
 - [Meta Velox](https://velox-lib.io)
 - [Google Pax](https://github.com/google/paxml)
 - [milvus](https://github.com/milvus-io/milvus)
+- [QuestDB](https://questdb.io/blog/questdb-release-8-0-3/)
 - [Clang Build Analyzer](https://github.com/aras-p/ClangBuildAnalyzer)
 - [Shopify HeapProfiler](https://github.com/Shopify/heap-profiler)
 - [StarRocks](https://github.com/StarRocks/starrocks)
@@ -59,6 +61,9 @@ Real-world usage
 - [vast](https://github.com/tenzir/vast)
 - [ada-url](https://github.com/ada-url/ada)
 - [fastgron](https://github.com/adamritter/fastgron)
+- [WasmEdge](https://wasmedge.org)
+- [RonDB](https://github.com/logicalclocks/rondb)
+
 
 If you are planning to use simdjson in a product, please work from one of our releases.
 
@@ -103,6 +108,7 @@ Documentation
 Usage documentation is available:
 
 * [Basics](doc/basics.md) is an overview of how to use simdjson and its APIs.
+* [Builder](doc/builder.md) is an overview of how to efficiently write JSON strings using simdjson.
 * [Performance](doc/performance.md) shows some more advanced scenarios and how to tune for them.
 * [Implementation Selection](doc/implementation-selection.md) describes runtime CPU detection and
   how you can work with it.
@@ -141,6 +147,9 @@ speed for [synthetic files over various sizes generated with a script](https://g
 For NDJSON files, we can exceed 3 GB/s with [our  multithreaded parsing functions](https://github.com/simdjson/simdjson/blob/master/doc/parse_many.md).
 
 
+Packages
+------------------------------
+[![Packaging status](https://repology.org/badge/vertical-allrepos/simdjson.svg)](https://repology.org/project/simdjson/versions)
 
 
 Bindings and Ports of simdjson
@@ -178,9 +187,9 @@ The simdjson library takes advantage of modern microarchitectures, parallelizing
 instructions, reducing branch misprediction, and reducing data dependency to take advantage of each
 CPU's multiple execution cores.
 
-Our default front-end is called On Demand, and we wrote a paper about it:
+Our default front-end is called On-Demand, and we wrote a paper about it:
 
-- John Keiser, Daniel Lemire, [On-Demand JSON: A Better Way to Parse Documents?](http://arxiv.org/abs/2312.17149), Software: Practice and Experience (to appear)
+- John Keiser, Daniel Lemire, [On-Demand JSON: A Better Way to Parse Documents?](http://arxiv.org/abs/2312.17149), Software: Practice and Experience 54 (6), 2024.
 
 Some people [enjoy reading the first (2019) simdjson paper](https://arxiv.org/abs/1902.08318): A description of the design
 and implementation of simdjson is in our research article:
@@ -199,12 +208,17 @@ For the video inclined, <br />
 Funding
 -------
 
-The work is supported by the Natural Sciences and Engineering Research Council of Canada under grant
+The work is supported by the Natural Sciences and Engineering Research Council of Canada under grants
-number RGPIN-2017-03910.
+RGPIN-2017-03910 and RGPIN-2024-03787.
 
 [license]: LICENSE
 [license img]: https://img.shields.io/badge/License-Apache%202-blue.svg
 
+
+[licensemit]: LICENSE-MIT
+[licensemit img]: https://img.shields.io/badge/License-MIT-blue.svg
+
+
 Contributing to simdjson
 ------------------------
 
@@ -214,7 +228,7 @@ Head over to [CONTRIBUTING.md](CONTRIBUTING.md) for information on contributing
 License
 -------
 
-This code is made available under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0.html).
+This code is made available under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0.html) as well as under the MIT License. As a user, you can pick the license you prefer.
 
 Under Windows, we build some tools using the windows/dirent_portable.h file (which is outside our library code): it is under the liberal (business-friendly) MIT license.
 

benchmark/CMakeLists.txt

@@ -4,6 +4,9 @@ add_subdirectory(dom)
 include_directories( . linux )
 link_libraries(simdjson-windows-headers test-data)
 link_libraries(simdjson)
+if(SIMDJSON_STATIC_REFLECTION)
+  add_compile_definitions(SIMDJSON_STATIC_REFLECTION=1)
+endif(SIMDJSON_STATIC_REFLECTION)
 
 add_executable(benchfeatures benchfeatures.cpp)
 add_executable(get_corpus_benchmark get_corpus_benchmark.cpp)
@@ -32,3 +35,6 @@ if (TARGET benchmark::benchmark)
   endif()
 endif()
 
+if(SIMDJSON_STATIC_REFLECTION)
+  add_subdirectory(static_reflect)
+endif(SIMDJSON_STATIC_REFLECTION)

benchmark/event_counter.h

@@ -122,6 +122,7 @@ struct event_aggregate {
   }
 
   double elapsed_sec() const { return total.elapsed_sec() / iterations; }
+  double total_elapsed_ns() const { return total.elapsed_ns(); }
   double elapsed_ns() const { return total.elapsed_ns() / iterations; }
   double cycles() const { return total.cycles() / iterations; }
   double instructions() const { return total.instructions() / iterations; }

benchmark/find_tweet/nlohmann_json.h

@@ -13,7 +13,7 @@ struct nlohmann_json {
     auto root = nlohmann::json::parse(json.data(), json.data() + json.size());
     for (auto tweet : root["statuses"]) {
       if (tweet["id"] == find_id) {
-        result = tweet["text"];
+        result = to_string(tweet["text"]);
         return true;
       }
     }

benchmark/large_random/simdjson_ondemand.h

@@ -23,7 +23,30 @@ struct simdjson_ondemand {
 };
 
 BENCHMARK_TEMPLATE(large_random, simdjson_ondemand)->UseManualTime();
+#if SIMDJSON_STATIC_REFLECTION
 
+struct simdjson_ondemand_static_reflect {
+  static constexpr diff_flags DiffFlags = diff_flags::NONE;
+
+  ondemand::parser parser{};
+
+  bool run(simdjson::padded_string &json, std::vector<point> &result) {
+    auto doc = parser.iterate(json);
+    //ondemand::array arr = doc.get_array();
+    //arr.get<std::vector<point>>(result);
+    // We can also do it like so:
+    for (ondemand::object coord : doc) {
+      result.emplace_back(coord.get<point>());
+    }
+    // It seems that doing the reflection is slower than doing the manual lookup.
+    // E.g., it is faster if we do result.emplace_back(coord["x"], coord["y"], coord["z"]);
+    return true;
+  }
+};
+BENCHMARK_TEMPLATE(large_random, simdjson_ondemand_static_reflect)->UseManualTime();
+
+
+#endif
 } // namespace large_random
 
 #endif // SIMDJSON_EXCEPTIONS

benchmark/partial_tweets/ondemand.h

@@ -13,7 +13,7 @@ class OnDemand {
 public:
   OnDemand() {
     if(!displayed_implementation) {
-      std::cout << "On Demand implementation: " << builtin_implementation()->name() << std::endl;
+      std::cout << "On-Demand implementation: " << builtin_implementation()->name() << std::endl;
       displayed_implementation = true;
     }
   }

benchmark/static_reflect/CMakeLists.txt

@@ -0,0 +1,44 @@
+# Include reflect-cpp
+CPMAddPackage(
+  NAME reflect-cpp
+  GITHUB_REPOSITORY getml/reflect-cpp
+  GIT_TAG v0.17.0
+  EXCLUDE_FROM_ALL YES
+)
+
+
+
+if(NOT WIN32)
+# We want the check whether Rust is available before trying to build a crate.
+CPMAddPackage(
+  NAME corrosion
+  GITHUB_REPOSITORY corrosion-rs/corrosion
+  VERSION 0.4.4
+  DOWNLOAD_ONLY ON
+  OPTIONS "Rust_FIND_QUIETLY OFF"
+)
+include("${corrosion_SOURCE_DIR}/cmake/FindRust.cmake")
+endif()
+
+if(RUST_FOUND)
+  message(STATUS "Rust found: " ${Rust_VERSION} )
+  add_subdirectory("${corrosion_SOURCE_DIR}" "${PROJECT_BINARY_DIR}/_deps/corrosion" EXCLUDE_FROM_ALL)
+  # Important: we want to build in release mode!
+  corrosion_import_crate(MANIFEST_PATH "serde-benchmark/Cargo.toml" NO_LINKER_OVERRIDE PROFILE release)
+else()
+  message(STATUS "Rust/Cargo is unavailable." )
+  message(STATUS "We will not benchmark serde-benchmark." )
+  if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+    message(STATUS "Under macOS, you may be able to install rust with")
+    message(STATUS "curl https://sh.rustup.rs -sSf | sh")
+  elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+    message(STATUS "Under Linux, you may be able to install rust with a command such as")
+    message(STATUS "apt-get install cargo" )
+    message(STATUS "or" )
+    message(STATUS "curl https://sh.rustup.rs -sSf | sh")
+  endif()
+endif()
+
+
+
+add_subdirectory(twitter_data)

benchmark/static_reflect/serde-benchmark/Cargo.lock

@@ -0,0 +1,103 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 3
+
+[[package]]
+name = "itoa"
+version = "1.0.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "49f1f14873335454500d59611f1cf4a4b0f786f9ac11f4312a78e4cf2566695b"
+
+[[package]]
+name = "libc"
+version = "0.2.158"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d8adc4bb1803a324070e64a98ae98f38934d91957a99cfb3a43dcbc01bc56439"
+
+[[package]]
+name = "memchr"
+version = "2.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78ca9ab1a0babb1e7d5695e3530886289c18cf2f87ec19a575a0abdce112e3a3"
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.86"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5e719e8df665df0d1c8fbfd238015744736151d4445ec0836b8e628aae103b77"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.37"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b5b9d34b8991d19d98081b46eacdd8eb58c6f2b201139f7c5f643cc155a633af"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "ryu"
+version = "1.0.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f3cb5ba0dc43242ce17de99c180e96db90b235b8a9fdc9543c96d2209116bd9f"
+
+[[package]]
+name = "serde"
+version = "1.0.209"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "99fce0ffe7310761ca6bf9faf5115afbc19688edd00171d81b1bb1b116c63e09"
+dependencies = [
+ "serde_derive",
+]
+
+[[package]]
+name = "serde-benchmark"
+version = "0.1.0"
+dependencies = [
+ "libc",
+ "serde",
+ "serde_json",
+]
+
+[[package]]
+name = "serde_derive"
+version = "1.0.209"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a5831b979fd7b5439637af1752d535ff49f4860c0f341d1baeb6faf0f4242170"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "serde_json"
+version = "1.0.127"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8043c06d9f82bd7271361ed64f415fe5e12a77fdb52e573e7f06a516dea329ad"
+dependencies = [
+ "itoa",
+ "memchr",
+ "ryu",
+ "serde",
+]
+
+[[package]]
+name = "syn"
+version = "2.0.76"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "578e081a14e0cefc3279b0472138c513f37b41a08d5a3cca9b6e4e8ceb6cd525"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3354b9ac3fae1ff6755cb6db53683adb661634f67557942dea4facebec0fee4b"

benchmark/static_reflect/serde-benchmark/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "serde-benchmark"
+version = "0.1.0"
+
+[lib]
+path = "lib.rs"
+crate-type = ["cdylib"]
+
+[dependencies]
+serde = { version = "1.0", features = ["derive"] }
+libc = "0.2"
+serde_json = "1.0"
+
+[profile.release]
+opt-level = 3
+debug = false
+lto = true

benchmark/static_reflect/serde-benchmark/README.md

@@ -0,0 +1,18 @@
+## Rust Serde FFI
+
+This folder includes FFI bindings for rust/serde.
+
+### Links
+
+- https://github.com/eqrion/cbindgen/blob/master/docs.md
+- https://gist.github.com/zbraniecki/b251714d77ffebbc73c03447f2b2c69f
+- https://michael-f-bryan.github.io/rust-ffi-guide/setting_up.html
+
+### Building
+
+- Generating cbindgen output
+  - Install dependencies with `brew install cbindgen` or `apt-get install cbindgen` or `cargo install cbindgen` or the equivalent: we used `cargo install --version 0.23.0 cbindgen`.
+  - Go to the directory where this README.md file is located
+  - Generate with `cbindgen --config cbindgen.toml --crate serde-benchmark --output serde_benchmark.h`
+- Building
+  - Run with `cargo build --release`

benchmark/static_reflect/serde-benchmark/cbindgen.toml

@@ -0,0 +1,12 @@
+autogen_warning = "/* Warning, this file is autogenerated by cbindgen. Don't modify this manually. */"
+include_version = true
+braces = "SameLine"
+line_length = 100
+tab_width = 2
+language = "C++"
+namespaces = ["serde_benchmark"]
+include_guard = "serde_benchmark_ffi_h"
+
+[parse]
+parse_deps = true
+include = ["serde_json", "serde"]

benchmark/static_reflect/serde-benchmark/lib.rs

@@ -0,0 +1,170 @@
+extern crate serde;
+extern crate serde_json;
+extern crate libc;
+use std::slice;
+
+use libc::{c_char, size_t};
+use serde::{Serialize, Deserialize};
+
+/******************************************************/
+/******************************************************/
+/**
+ * Warning: the C++ code may not generate the same JSON.
+ */
+ /******************************************************/
+ /******************************************************/
+
+// This has no equivalent in C++:
+#[derive(Serialize, Deserialize)]
+pub struct Metadata {
+    result_type: String,
+    iso_language_code: String,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct User {
+    id: i64,
+    id_str: String,
+    name: String,
+    screen_name: String,
+    location: String,
+    description: String,
+    // C++ does not have those:
+    // url: Option<String>,
+    //protected: bool,
+    //listed_count: i64,
+    //created_at: String,
+    //favourites_count: i64,
+    //utc_offset: Option<i64>,
+    //time_zone: Option<String>,
+    //geo_enabled: bool,
+    verified: bool,
+    followers_count: i64,
+    friends_count: i64,
+    statuses_count: i64,
+    // C++ does not have those:
+    //lang: String,
+    //profile_background_color: String,
+    //profile_background_image_url: String,
+    //profile_background_image_url_https: String,
+    //profile_background_tile: bool,
+    //profile_image_url: String,
+    //profile_image_url_https: String,
+    //profile_banner_url: Option<String>,
+    //profile_link_color: String,
+    //profile_sidebar_border_color: String,
+    //profile_sidebar_fill_color: String,
+    //profile_text_color: String,
+    //profile_use_background_image: bool,
+    //default_profile: bool,
+    //default_profile_image: bool,
+    //following: bool,
+    //follow_request_sent: bool,
+    //notifications: bool,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Hashtag {
+    text: String,
+
+    // C++ has those but D. Lemire does not know what they are, they don't appear in the JSON:
+    // int64_t indices_start;
+    // int64_t indices_end;
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Url {
+    url: String,
+    expanded_url: String,
+    display_url: String,
+    // C++ has those but D. Lemire does not know what they are, they don't appear in the JSON:
+    // int64_t indices_start;
+    // int64_t indices_end;
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct UserMention {
+    id: i64,
+    name: String,
+    screen_name: String,
+    // Not in the C++ equivalent:
+    //id_str: String,
+    //indices: Vec<i64>,
+    // C++ has those but D. Lemire does not know what they are, they don't appear in the JSON:
+    // int64_t indices_start;
+    // int64_t indices_end;
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Entities {
+    hashtags: Vec<Hashtag>,
+    urls: Vec<Url>,
+    user_mentions: Vec<UserMention>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Status {
+    created_at: String,
+    id: i64,
+    text: String,
+    user: User,
+    entities: Entities,
+    retweet_count: i64,
+    favorite_count: i64,
+    favorited: bool,
+    retweeted: bool,
+    // None of these are in the C++ equivalent:
+    /*
+    metadata: Metadata,
+    id_str: String,
+    source: String,
+    truncated: bool,
+    in_reply_to_status_id: Option<i64>,
+    in_reply_to_status_id_str: Option<String>,
+    in_reply_to_user_id: Option<i64>,
+    in_reply_to_user_id_str: Option<String>,
+    in_reply_to_screen_name: Option<String>,
+    geo: Option<String>,
+    coordinates: Option<String>,
+    place: Option<String>,
+    contributors: Option<String>,
+    lang: String,
+    */
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct TwitterData {
+    statuses: Vec<Status>,
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn  twitter_from_str(raw_input: *const c_char, raw_input_length: size_t) -> *mut TwitterData {
+  let input = std::str::from_utf8_unchecked(slice::from_raw_parts(raw_input as *const u8, raw_input_length));
+  match serde_json::from_str(&input) {
+    Ok(result) => Box::into_raw(Box::new(result)),
+    Err(_) => std::ptr::null_mut(),
+  }
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn str_from_twitter(raw: *mut TwitterData) -> *const c_char {
+  let twitter_thing = { &*raw };
+  let serialized = serde_json::to_string(&twitter_thing).unwrap();
+  return std::ffi::CString::new(serialized.as_str()).unwrap().into_raw()
+}
+
+
+#[no_mangle]
+pub unsafe extern "C" fn free_twitter(raw: *mut TwitterData) {
+  if raw.is_null() {
+    return;
+  }
+
+  drop(Box::from_raw(raw))
+}
+
+
+#[no_mangle]
+pub unsafe extern fn free_string(ptr: *const c_char) {
+    let _ = std::ffi::CString::from_raw(ptr as *mut _);
+}

benchmark/static_reflect/serde-benchmark/serde_benchmark.h

@@ -0,0 +1,32 @@
+#ifndef serde_benchmark_ffi_h
+#define serde_benchmark_ffi_h
+
+/* Generated with cbindgen:0.24.3 */
+
+/* Warning, this file is autogenerated by cbindgen. Don't modify this manually. */
+
+#include <cstdarg>
+#include <cstdint>
+#include <cstdlib>
+#include <ostream>
+#include <new>
+
+namespace serde_benchmark {
+
+struct TwitterData;
+
+extern "C" {
+
+TwitterData *twitter_from_str(const char *raw_input, size_t raw_input_length);
+
+const char *str_from_twitter(TwitterData *raw);
+
+void free_twitter(TwitterData *raw);
+
+void free_string(const char *ptr);
+
+} // extern "C"
+
+} // namespace serde_benchmark
+
+#endif // serde_benchmark_ffi_h

benchmark/static_reflect/twitter_data/CMakeLists.txt

@@ -0,0 +1,15 @@
+
+# Add executable targets
+add_executable(benchmark_serialization_twitter benchmark_serialization_twitter.cpp)
+
+ if(TARGET serde-benchmark)
+    message(STATUS "serde-benchmark target was created. Linking benchmarks and serde-benchmark.")
+    target_link_libraries(benchmark_serialization_twitter PRIVATE serde-benchmark)
+    target_compile_definitions(benchmark_serialization_twitter PRIVATE SIMDJSON_RUST_VERSION="${Rust_VERSION}")
+endif()
+target_link_libraries(benchmark_serialization_twitter PRIVATE simdjson::simdjson nlohmann_json)
+
+target_link_libraries(benchmark_serialization_twitter PRIVATE reflectcpp)
+target_compile_definitions(benchmark_serialization_twitter PRIVATE SIMDJSON_BENCH_CPP_REFLECT=1)
+
+target_compile_definitions(benchmark_serialization_twitter PRIVATE JSON_FILE="${EXAMPLE_JSON}")

benchmark/static_reflect/twitter_data/benchmark_helper.h

@@ -0,0 +1,52 @@
+#ifndef BENCHMARK_HELPER_HPP
+#define BENCHMARK_HELPER_HPP
+#include "event_counter.h"
+#include <atomic>
+
+inline event_collector &get_collector() {
+  static event_collector collector;
+  return collector;
+}
+
+template <class function_type>
+event_aggregate bench(const function_type &function, size_t min_repeat = 10,
+                      size_t min_time_ns = 1000000000,
+                      size_t max_repeat = 100000) {
+  event_collector &collector = get_collector();
+  event_aggregate aggregate{};
+  size_t N = min_repeat;
+  if (N == 0) {
+    N = 1;
+  }
+  for (size_t i = 0; i < N; i++) {
+    std::atomic_thread_fence(std::memory_order_acquire);
+    collector.start();
+    function();
+    std::atomic_thread_fence(std::memory_order_release);
+    event_count allocate_count = collector.end();
+    aggregate << allocate_count;
+    if ((i + 1 == N) && (aggregate.total_elapsed_ns() < min_time_ns) &&
+        (N < max_repeat)) {
+      N *= 10;
+    }
+  }
+  return aggregate;
+}
+
+// Source of the 2 functions below:
+// https://github.com/simdutf/simdutf/blob/master/benchmarks/base64/benchmark_base64.cpp
+inline void pretty_print(size_t strings, size_t bytes, std::string name,
+                         event_aggregate agg) {
+  event_collector &collector = get_collector();
+  printf("%-60s : ", name.c_str());
+  printf(" %5.2f MB/s ", bytes * 1000 / agg.elapsed_ns());
+  printf(" %5.2f Ms/s ", strings * 1000 / agg.elapsed_ns());
+  if (collector.has_events()) {
+    printf(" %5.2f GHz ", agg.cycles() / agg.elapsed_ns());
+    printf(" %5.2f c/b ", agg.cycles() / bytes);
+    printf(" %5.2f i/b ", agg.instructions() / bytes);
+    printf(" %5.2f i/c ", agg.instructions() / agg.cycles());
+  }
+  printf("\n");
+}
+#endif

benchmark/static_reflect/twitter_data/benchmark_serialization_twitter.cpp

@@ -0,0 +1,142 @@
+#include <cassert>
+#include <cstdlib>
+#include <ctime>
+#include <format>
+#include <fstream>
+#include <iostream>
+#include <nlohmann/json.hpp>
+#include <simdjson.h>
+#include <string>
+#include "twitter_data.h"
+#include "nlohmann_twitter_data.h"
+#include "benchmark_helper.h"
+#if SIMDJSON_BENCH_CPP_REFLECT
+#include <rfl.hpp>
+#include <rfl/json.hpp>
+void bench_reflect_cpp(TwitterData &data) {
+  std::string output = rfl::json::write(data);
+  size_t output_volume = output.size();
+  printf("# output volume: %zu bytes\n", output_volume);
+
+  volatile size_t measured_volume = 0;
+  pretty_print(1, output_volume, "bench_reflect_cpp",
+               bench([&data, &measured_volume, &output_volume]() {
+                 std::string output = rfl::json::write(data);
+                 measured_volume = output.size();
+                 if (measured_volume != output_volume) {
+                   printf("mismatch\n");
+                 }
+               }));
+}
+#endif // SIMDJSON_BENCH_CPP_REFLECT
+
+#ifdef SIMDJSON_RUST_VERSION
+#include "../serde-benchmark/serde_benchmark.h"
+
+
+void bench_rust(serde_benchmark::TwitterData *data) {
+  const char * output = serde_benchmark::str_from_twitter(data);
+  size_t output_volume = strlen(output);
+  printf("# output volume: %zu bytes\n", output_volume);
+  volatile size_t measured_volume = 0;
+  pretty_print(1, output_volume, "bench_rust",
+               bench([&data, &measured_volume, &output_volume]() {
+                 const char * output = serde_benchmark::str_from_twitter(data);
+                 serde_benchmark::free_string(output);
+               }));
+}
+#endif
+
+template <class T> void bench_simdjson_static_reflection(T &data) {
+  simdjson::builder::string_builder sb;
+  simdjson::builder::append(sb, data);
+  std::string_view p;
+  if(sb.view().get(p)) {
+    std::cerr << "Error!" << std::endl;
+  }
+  size_t output_volume = p.size();
+  sb.clear();
+  printf("# output volume: %zu bytes\n", output_volume);
+
+  volatile size_t measured_volume = 0;
+  pretty_print(sizeof(data), output_volume, "bench_simdjson_static_reflection",
+               bench([&data, &measured_volume, &output_volume, &sb]() {
+                 sb.clear();
+                 simdjson::builder::append(sb, data);
+                 std::string_view p;
+                 if(sb.view().get(p)) {
+                   std::cerr << "Error!" << std::endl;
+                 }
+                 measured_volume = sb.size();
+                 if (measured_volume != output_volume) {
+                   printf("mismatch\n");
+                 }
+               }));
+}
+
+void bench_nlohmann(TwitterData &data) {
+  std::string output = nlohmann_serialize(data);
+  size_t output_volume = output.size();
+  printf("# output volume: %zu bytes\n", output_volume);
+
+  volatile size_t measured_volume = 0;
+  pretty_print(1, output_volume, "bench_nlohmann",
+               bench([&data, &measured_volume, &output_volume]() {
+                 std::string output = nlohmann_serialize(data);
+                 measured_volume = output.size();
+                 if (measured_volume != output_volume) {
+                   printf("mismatch\n");
+                 }
+               }));
+}
+
+size_t WriteCallback(void *contents, size_t size, size_t nmemb, void *userp) {
+  ((std::string *)userp)->append((char *)contents, size * nmemb);
+  return size * nmemb;
+}
+
+std::string read_file(std::string filename) {
+  printf("# Reading file %s\n", filename.c_str());
+  constexpr size_t read_size = 4096;
+  auto stream = std::ifstream(filename.c_str());
+  stream.exceptions(std::ios_base::badbit);
+  std::string out;
+  std::string buf(read_size, '\0');
+  while (stream.read(&buf[0], read_size)) {
+    out.append(buf, 0, size_t(stream.gcount()));
+  }
+  out.append(buf, 0, size_t(stream.gcount()));
+  return out;
+}
+
+int main() {
+  // Testing correctness of round-trip (serialization + deserialization)
+  std::string json_str = read_file(JSON_FILE);
+
+  // Loading up the data into a structure.
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc;
+  if(parser.iterate(simdjson::pad(json_str)).get(doc)) {
+    std::cerr << "Error loading the document!" << std::endl;
+    return EXIT_FAILURE;
+  }
+  TwitterData my_struct;
+  if(doc.get<TwitterData>().get(my_struct)) {
+    std::cerr << "Error loading TwitterData!" << std::endl;
+    return EXIT_FAILURE;
+  }
+
+  // Benchmarking the serialization
+  bench_nlohmann(my_struct);
+  bench_simdjson_static_reflection(my_struct);
+#ifdef SIMDJSON_RUST_VERSION
+  printf("# WARNING: The Rust benchmark may not be directly comparable since it does not use an equivalent data structure.\n");
+  serde_benchmark::TwitterData * td = serde_benchmark::twitter_from_str(json_str.c_str(), json_str.size());
+  bench_rust(td);
+  serde_benchmark::free_twitter(td);
+#endif
+#if SIMDJSON_BENCH_CPP_REFLECT
+  bench_reflect_cpp(my_struct);
+#endif
+  return EXIT_SUCCESS;
+}

benchmark/static_reflect/twitter_data/nlohmann_twitter_data.h

@@ -0,0 +1,117 @@
+#ifndef NLOHMANN_TWITTER_DATA_H
+#define NLOHMANN_TWITTER_DATA_H
+
+#include "twitter_data.h"
+#include <nlohmann/json.hpp>
+
+void to_json(nlohmann::json &j, const User &u) {
+  j = nlohmann::json{{"id", u.id},
+                     {"name", u.name},
+                     {"screen_name", u.screen_name},
+                     {"location", u.location},
+                     {"description", u.description},
+                     {"verified", u.verified},
+                     {"followers_count", u.followers_count},
+                     {"friends_count", u.friends_count},
+                     {"statuses_count", u.statuses_count}};
+}
+
+void to_json(nlohmann::json &j, const Hashtag &h) {
+  j = nlohmann::json{{"text", h.text},
+                     {"indices_start", h.indices_start},
+                     {"indices_end", h.indices_end}};
+}
+
+void to_json(nlohmann::json &j, const Url &u) {
+  j = nlohmann::json{{"url", u.url},
+                     {"expanded_url", u.expanded_url},
+                     {"display_url", u.display_url},
+                     {"indices_start", u.indices_start},
+                     {"indices_end", u.indices_end}};
+}
+
+void to_json(nlohmann::json &j, const UserMention &um) {
+  j = nlohmann::json{{"id", um.id},
+                     {"name", um.name},
+                     {"screen_name", um.screen_name},
+                     {"indices_start", um.indices_start},
+                     {"indices_end", um.indices_end}};
+}
+
+void to_json(nlohmann::json &j, const Entities &e) {
+  j = nlohmann::json{{"hashtags", e.hashtags},
+                     {"urls", e.urls},
+                     {"user_mentions", e.user_mentions}};
+}
+
+void to_json(nlohmann::json &j, const Status &s) {
+  j = nlohmann::json{{"created_at", s.created_at},
+                     {"id", s.id},
+                     {"text", s.text},
+                     {"user", s.user},
+                     {"entities", s.entities},
+                     {"retweet_count", s.retweet_count},
+                     {"favorite_count", s.favorite_count},
+                     {"favorited", s.favorited},
+                     {"retweeted", s.retweeted}};
+}
+
+
+std::string nlohmann_serialize(const std::vector<Hashtag>& v) {
+  nlohmann::json a = nlohmann::json::array();
+  for(const Hashtag & h : v) {
+    a.push_back(nlohmann::json{{"text", h.text},
+                     {"indices_start", h.indices_start},
+                     {"indices_end", h.indices_end}});
+  }
+  return a.dump();
+}
+std::string nlohmann_serialize(const std::vector<Url>& v) {
+  nlohmann::json a = nlohmann::json::array();
+  for(const Url & u : v) {
+    a.push_back(nlohmann::json{{"url", u.url},
+                     {"expanded_url", u.expanded_url},
+                     {"display_url", u.display_url},
+                     {"indices_start", u.indices_start},
+                     {"indices_end", u.indices_end}});
+  }
+  return a.dump();
+}
+std::string nlohmann_serialize(const std::vector<UserMention>& v) {
+  nlohmann::json a = nlohmann::json::array();
+  for(const UserMention & um : v) {
+    a.push_back(nlohmann::json{{"id", um.id},
+                     {"name", um.name},
+                     {"screen_name", um.screen_name},
+                     {"indices_start", um.indices_start},
+                     {"indices_end", um.indices_end}});
+  }
+  return a.dump();
+}
+
+std::string nlohmann_serialize(const std::vector<Status>& v) {
+  nlohmann::json a = nlohmann::json::array();
+  for(const Status & s : v) {
+    a.push_back(nlohmann::json{{"created_at", s.created_at},
+                     {"id", s.id},
+                     {"text", s.text},
+                     {"user", s.user},
+                     {"entities", s.entities},
+                     {"retweet_count", s.retweet_count},
+                     {"favorite_count", s.favorite_count},
+                     {"favorited", s.favorited},
+                     {"retweeted", s.retweeted}});
+  }
+  return a.dump();
+}
+
+
+void to_json(nlohmann::json &j, const TwitterData &t) {
+  j = nlohmann::json{{"statuses", t.statuses}};
+}
+
+std::string nlohmann_serialize(const TwitterData &data) {
+  return nlohmann_serialize(data.statuses);
+}
+
+#endif // NLOHMANN_TWITTER_DATA_H

benchmark/static_reflect/twitter_data/twitter_data.h

@@ -0,0 +1,71 @@
+#ifndef TWITTER_DATA_H
+#define TWITTER_DATA_H
+
+#include <string>
+#include <vector>
+
+struct User {
+  int64_t id;
+  std::string id_str;
+  std::string name;
+  std::string screen_name;
+  std::string location;
+  std::string description;
+  bool verified;
+  int64_t followers_count;
+  int64_t friends_count;
+  int64_t statuses_count;
+  bool operator<=>(const User &other) const = default;
+};
+
+struct Hashtag {
+  std::string text;
+  int64_t indices_start;
+  int64_t indices_end;
+  bool operator<=>(const Hashtag &other) const = default;
+};
+
+struct Url {
+  std::string url;
+  std::string expanded_url;
+  std::string display_url;
+  int64_t indices_start;
+  int64_t indices_end;
+  bool operator<=>(const Url &other) const = default;
+};
+
+struct UserMention {
+  int64_t id;
+  std::string name;
+  std::string screen_name;
+  int64_t indices_start;
+  int64_t indices_end;
+  bool operator<=>(const UserMention &other) const = default;
+};
+
+struct Entities {
+  std::vector<Hashtag> hashtags;
+  std::vector<Url> urls;
+  std::vector<UserMention> user_mentions;
+  bool operator==(const Entities &other) const = default;
+};
+
+struct Status {
+  std::string created_at;
+  int64_t id;
+  std::string text;
+  User user;
+  Entities entities;
+  int64_t retweet_count;
+  int64_t favorite_count;
+  bool favorited;
+  bool retweeted;
+  bool operator==(const Status &other) const = default;
+};
+
+struct TwitterData {
+  std::vector<Status> statuses;
+  bool operator==(const TwitterData &other) const = default;
+};
+
+#endif

benchmark/top_tweet/nlohmann_json.h

@@ -26,8 +26,8 @@ struct nlohmann_json {
       }
     }
 
-    result.text = top_tweet["text"];
+    result.text = to_string(top_tweet["text"]);
-    result.screen_name = top_tweet["user"]["screen_name"];
+    result.screen_name = to_string(top_tweet["user"]["screen_name"]);
     return result.retweet_count != -1;
   }
 };

cmake/CPM.cmake

@@ -0,0 +1,24 @@
+# SPDX-License-Identifier: MIT
+#
+# SPDX-FileCopyrightText: Copyright (c) 2019-2023 Lars Melchior and contributors
+
+set(CPM_DOWNLOAD_VERSION 0.40.2)
+set(CPM_HASH_SUM "c8cdc32c03816538ce22781ed72964dc864b2a34a310d3b7104812a5ca2d835d")
+
+if(CPM_SOURCE_CACHE)
+  set(CPM_DOWNLOAD_LOCATION "${CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+elseif(DEFINED ENV{CPM_SOURCE_CACHE})
+  set(CPM_DOWNLOAD_LOCATION "$ENV{CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+else()
+  set(CPM_DOWNLOAD_LOCATION "${CMAKE_BINARY_DIR}/cmake/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+endif()
+
+# Expand relative path. This is important if the provided path contains a tilde (~)
+get_filename_component(CPM_DOWNLOAD_LOCATION ${CPM_DOWNLOAD_LOCATION} ABSOLUTE)
+
+file(DOWNLOAD
+     https://github.com/cpm-cmake/CPM.cmake/releases/download/v${CPM_DOWNLOAD_VERSION}/CPM.cmake
+     ${CPM_DOWNLOAD_LOCATION} EXPECTED_HASH SHA256=${CPM_HASH_SUM}
+)
+
+include(${CPM_DOWNLOAD_LOCATION})

cmake/developer-options.cmake

@@ -110,8 +110,14 @@ endif()
 
 # We compile tools, tests, etc. with C++ 17. Override yourself if you need on a
 # target.
-set(SIMDJSON_CXX_STANDARD 17 CACHE STRING "the C++ standard to use for simdjson")
+if(SIMDJSON_STATIC_REFLECTION)
-set(CMAKE_CXX_STANDARD ${SIMDJSON_CXX_STANDARD})
+  # This is temporary.
+  set(SIMDJSON_CXX_STANDARD 26 CACHE STRING "the C++ standard to use for simdjson")
+  #set(CMAKE_CXX_STANDARD ${SIMDJSON_CXX_STANDARD})
+else()
+  set(SIMDJSON_CXX_STANDARD 17 CACHE STRING "the C++ standard to use for simdjson")
+  set(CMAKE_CXX_STANDARD ${SIMDJSON_CXX_STANDARD})
+endif()
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 set(CMAKE_THREAD_PREFER_PTHREAD ON)

dependencies/CMakeLists.txt

@@ -1,5 +1,4 @@
 include(CMakeDependentOption)
-include(import.cmake)
 
 option(SIMDJSON_ALLOW_DOWNLOADS
         "Allow dependencies to be downloaded during configure time"
@@ -11,17 +10,21 @@ cmake_dependent_option(SIMDJSON_GOOGLE_BENCHMARKS "compile the Google Benchmark
         SIMDJSON_ALLOW_DOWNLOADS OFF)
 
 if(SIMDJSON_GOOGLE_BENCHMARKS)
-  set_off(BENCHMARK_ENABLE_TESTING)
+  CPMAddPackage(
-  set_off(BENCHMARK_ENABLE_INSTALL)
+        NAME google_benchmarks
-  set_off(BENCHMARK_ENABLE_WERROR)
+        URL https://github.com/google/benchmark/archive/refs/tags/v1.7.1.zip
-
+        OPTIONS
-  import_dependency(google_benchmarks google/benchmark v1.7.1)
+        "BENCHMARK_ENABLE_TESTING OFF"
-  add_dependency(google_benchmarks)
+        "BENCHMARK_ENABLE_INSTALL OFF"
+        "BENCHMARK_ENABLE_WERROR OFF"
+  )
 endif()
 
-# The bulk of our benchmarking and testing data has been moved simdjson/simdjson-data
+CPMAddPackage(
-import_dependency(simdjson-data simdjson/simdjson-data a5b13babe65c1bba7186b41b43d4cbdc20a5c470)
+        NAME simdjson-data
-add_dependency(simdjson-data)
+        URL https://github.com/simdjson/simdjson-data/archive/a5b13babe65c1bba7186b41b43d4cbdc20a5c470.zip
+)
+
 option(SIMDJSON_USE_BOOST_JSON "Try to include BOOST_JSON, this may break your binaries under some systems." OFF)
 # This prevents variables declared with set() from unnecessarily escaping and
 # should not be called more than once
@@ -38,20 +41,30 @@ function(competition_scope_)
 int main() {}
 ]] SIMDJSON_FOUND_STRING_VIEW)
   if(SIMDJSON_FOUND_STRING_VIEW AND SIMDJSON_USE_BOOST_JSON)
-    import_dependency(boostjson boostorg/json ee8d72d)
+    CPMAddPackage(
+        NAME boostjson
+        URL https://github.com/boostorg/json/archive/ee8d72d8502b409b5561200299cad30ccdb91415.zip
+    )
     add_library(boostjson STATIC "${boostjson_SOURCE_DIR}/src/src.cpp")
     target_compile_definitions(boostjson PUBLIC BOOST_JSON_STANDALONE)
     target_include_directories(boostjson SYSTEM PUBLIC
             "${boostjson_SOURCE_DIR}/include")
     target_compile_definitions(boostjson INTERFACE SIMDJSON_COMPETITION_BOOSTJSON)
   endif()
-
+  CPMAddPackage(
-  import_dependency(cjson DaveGamble/cJSON c69134d)
+      NAME cjson
+      URL https://github.com/DaveGamble/cJSON/archive/c69134d01746dcf551dd7724b4edb12f922eb0d1.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(cjson STATIC "${cjson_SOURCE_DIR}/cJSON.c")
   target_include_directories(cjson SYSTEM PUBLIC "${cjson_SOURCE_DIR}")
   target_compile_definitions(cjson INTERFACE SIMDJSON_COMPETITION_CJSON)
 
-  import_dependency(fastjson mikeando/fastjson 485f994)
+  CPMAddPackage(
+      NAME fastjson
+      URL https://github.com/mikeando/fastjson/archive/485f994a61a64ac73fa6a40d4d639b99b463563b.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(fastjson STATIC
           "${fastjson_SOURCE_DIR}/src/fastjson.cpp"
           "${fastjson_SOURCE_DIR}/src/fastjson2.cpp"
@@ -60,28 +73,36 @@ int main() {}
           "${fastjson_SOURCE_DIR}/include")
   target_compile_definitions(fastjson INTERFACE SIMDJSON_COMPETITION_FASTJSON)
 
-  import_dependency(gason vivkin/gason 7aee524)
+  CPMAddPackage(
+      NAME gason
+      URL https://github.com/vivkin/gason/archive/7aee524189da1c1ecd19f67981e3d903dae25470.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(gason STATIC "${gason_SOURCE_DIR}/src/gason.cpp")
   target_include_directories(gason SYSTEM PUBLIC "${gason_SOURCE_DIR}/src")
   target_compile_definitions(gason INTERFACE SIMDJSON_COMPETITION_GASON)
 
-  import_dependency(jsmn zserge/jsmn 18e9fe4)
+  CPMAddPackage(
+      NAME jsmn
+      URL https://github.com/zserge/jsmn/archive/18e9fe42cbfe21d65076f5c77ae2be379ad1270f.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(jsmn STATIC "${jsmn_SOURCE_DIR}/jsmn.c")
   target_include_directories(jsmn SYSTEM PUBLIC "${jsmn_SOURCE_DIR}")
   target_compile_definitions(jsmn INTERFACE SIMDJSON_COMPETITION_JSMN)
 
-  message(STATUS "Importing json (nlohmann/json@v3.10.5)")
+  CPMAddPackage(
-  set(nlohmann_json_SOURCE_DIR "${dep_root}/json")
+      NAME nlohmann_json
-  if(NOT EXISTS "${nlohmann_json_SOURCE_DIR}")
+      URL https://github.com/nlohmann/json/archive/refs/tags/v3.10.5.zip
-    file(DOWNLOAD
+  )
-            "https://github.com/nlohmann/json/releases/download/v3.10.5/json.hpp"
-            "${nlohmann_json_SOURCE_DIR}/nlohmann/json.hpp")
-  endif()
-  add_library(nlohmann_json INTERFACE)
-  target_include_directories(nlohmann_json SYSTEM INTERFACE "${nlohmann_json_SOURCE_DIR}")
-  target_compile_definitions(nlohmann_json INTERFACE SIMDJSON_COMPETITION_NLOHMANN_JSON)
 
-  import_dependency(json11 dropbox/json11 ec4e452)
+  set_property(TARGET nlohmann_json APPEND PROPERTY INTERFACE_COMPILE_DEFINITIONS SIMDJSON_COMPETITION_NLOHMANN_JSON)
+
+  CPMAddPackage(
+      NAME json11
+      URL https://github.com/dropbox/json11/archive/ec4e45219af1d7cde3d58b49ed762376fccf1ace.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(json11 STATIC "${json11_SOURCE_DIR}/json11.cpp")
   target_include_directories(json11 SYSTEM PUBLIC "${json11_SOURCE_DIR}")
   target_compile_definitions(json11 INTERFACE SIMDJSON_COMPETITION_JSON11)
@@ -91,7 +112,11 @@ int main() {}
   target_include_directories(jsoncpp SYSTEM PUBLIC "${jsoncpp_SOURCE_DIR}")
   target_compile_definitions(jsoncpp INTERFACE SIMDJSON_COMPETITION_JSONCPP)
 
-  import_dependency(rapidjson Tencent/rapidjson f54b0e4)
+  CPMAddPackage(
+      NAME rapidjson
+      URL https://github.com/Tencent/rapidjson/archive/805d7ed5dfe97a39b8b0816fd5eeed8731dc4936.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(rapidjson INTERFACE)
   target_compile_definitions(rapidjson INTERFACE RAPIDJSON_HAS_STDSTRING)
   include (TestBigEndian)
@@ -110,14 +135,22 @@ int main() {}
     target_compile_definitions(rapidjson INTERFACE SIMDJSON_COMPETITION_RAPIDJSON)
   endif()
 
-  import_dependency(sajson chadaustin/sajson 2dcfd35)
+  CPMAddPackage(
+      NAME sajson
+      URL https://github.com/chadaustin/sajson/archive/2dcfd350586375f9910f74821d4f07d67ae455ba.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(sajson INTERFACE)
   target_compile_definitions(sajson INTERFACE SAJSON_UNSORTED_OBJECT_KEYS)
   target_include_directories(sajson SYSTEM INTERFACE
           "${sajson_SOURCE_DIR}/include")
   target_compile_definitions(sajson INTERFACE SIMDJSON_COMPETITION_SAJSON)
 
-  import_dependency(ujson4c esnme/ujson4c e14f3fd)
+  CPMAddPackage(
+      NAME ujson4c
+      URL https://github.com/esnme/ujson4c/archive/e14f3fd5207fe30d1bdea723f260609e69d1abfa.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(ujson4c STATIC
           "${ujson4c_SOURCE_DIR}/src/ujdecode.c"
           "${ujson4c_SOURCE_DIR}/3rdparty/ultrajsondec.c")
@@ -126,7 +159,11 @@ int main() {}
           "${ujson4c_SOURCE_DIR}/3rdparty")
   target_compile_definitions(ujson4c INTERFACE SIMDJSON_COMPETITION_UJSON4C)
 
-  import_dependency(yyjson ibireme/yyjson c385651)
+  CPMAddPackage(
+      NAME yyjson
+      URL https://github.com/ibireme/yyjson/archive/c3856514de0a67d7b66939bf3ed491a2d6e61277.zip
+      DOWNLOAD_ONLY YES
+  )
   add_library(yyjson STATIC "${yyjson_SOURCE_DIR}/src/yyjson.c")
   target_include_directories(yyjson SYSTEM PUBLIC "${yyjson_SOURCE_DIR}/src")
   target_compile_definitions(yyjson INTERFACE SIMDJSON_COMPETITION_YYJSON)
@@ -151,10 +188,12 @@ cmake_dependent_option(SIMDJSON_CXXOPTS "Download cxxopts (necessary for tools)"
         SIMDJSON_ALLOW_DOWNLOADS OFF)
 
 if(SIMDJSON_CXXOPTS)
-  set_off(CXXOPTS_BUILD_EXAMPLES)
+  CPMAddPackage(
-  set_off(CXXOPTS_BUILD_TESTS)
+        NAME cxxopts
-  set_off(CXXOPTS_ENABLE_INSTALL)
+        URL https://github.com/jarro2783/cxxopts/archive/59656709c0c58fcd0ed18b38e02938dbe05284c5.zip
-
+        OPTIONS
-  import_dependency(cxxopts jarro2783/cxxopts 794c975)
+        "CXXOPTS_BUILD_EXAMPLES OFF"
-  add_dependency(cxxopts)
+        "CXXOPTS_BUILD_TESTS OFF"
+        "CXXOPTS_ENABLE_INSTALL OFF"
+  )
 endif()

dependencies/import.cmake

@@ -1,48 +0,0 @@
-set(dep_root "${simdjson_SOURCE_DIR}/dependencies/.cache")
-if(DEFINED ENV{simdjson_DEPENDENCY_CACHE_DIR})
-  set(dep_root "$ENV{simdjson_DEPENDENCY_CACHE_DIR}")
-endif()
-
-function(import_dependency NAME GITHUB_REPO COMMIT)
-  message(STATUS "Importing ${NAME} (${GITHUB_REPO}@${COMMIT})")
-  set(target "${dep_root}/${NAME}")
-
-  # If the folder exists in the cache, then we assume that everything is as
-  # should be and do nothing
-  if(EXISTS "${target}")
-    set("${NAME}_SOURCE_DIR" "${target}" PARENT_SCOPE)
-    return()
-  endif()
-
-  set(zip_url "https://github.com/${GITHUB_REPO}/archive/${COMMIT}.zip")
-  set(archive "${dep_root}/archive.zip")
-  set(dest "${dep_root}/_extract")
-
-  file(DOWNLOAD "${zip_url}" "${archive}")
-  file(MAKE_DIRECTORY "${dest}")
-  execute_process(
-          WORKING_DIRECTORY "${dest}"
-          COMMAND "${CMAKE_COMMAND}" -E tar xf "${archive}")
-  file(REMOVE "${archive}")
-
-  # GitHub archives only ever have one folder component at the root, so this
-  # will always match that single folder
-  file(GLOB dir LIST_DIRECTORIES YES "${dest}/*")
-
-  file(RENAME "${dir}" "${target}")
-
-  set("${NAME}_SOURCE_DIR" "${target}" PARENT_SCOPE)
-endfunction()
-
-# Delegates to the dependency
-macro(add_dependency NAME)
-  if(NOT DEFINED "${NAME}_SOURCE_DIR")
-    message(FATAL_ERROR "Missing ${NAME}_SOURCE_DIR variable")
-  endif()
-
-  add_subdirectory("${${NAME}_SOURCE_DIR}" "${PROJECT_BINARY_DIR}/_deps/${NAME}" EXCLUDE_FROM_ALL)
-endmacro()
-
-function(set_off NAME)
-  set("${NAME}" OFF CACHE INTERNAL "")
-endfunction()

doc/builder.md

@@ -0,0 +1,141 @@
+Builder
+==========
+
+Sometimes you want to generate JSON string outputs efficiently.
+The simdjson library provides high-performance low-level facilities.
+When using these low-level functionalities, you are responsible to
+define the structure of your JSON document. However, string escaping
+and UTF-8 validation is automated.
+
+Overview: string_builder
+---------------------------
+
+The string_builder class is a low-level utility for constructing JSON strings representing documents. It is optimized for performance, potentially leveraging kernel-specific features like SIMD instructions for tasks such as string escaping. This class supports atomic types (e.g., booleans, numbers, strings) but does not handle composed types directly (like arrays or objects).
+
+An `string_builder` is created with an initial buffer capacity (e.g., 1kB). The memory
+is reallocated when needed. It has the following methods to add content to the string:
+
+
+- `append(number_type v)`: Appends a number (including booleans) to the JSON buffer. Booleans are converted to the strings "false" or "true". Numbers are formatted according to the JSON standard, with floating-point numbers using the shortest representation that accurately reflects the value.
+- `append(char c)`: Appends a single character to the JSON buffer.
+- `append_null()`: Appends the string "null" to the JSON buffer.
+- `clear()`: Clears the contents of the JSON buffer, resetting the position to 0 while retaining the allocated capacity.
+- `escape_and_append(std::string_view input)`: Appends a string view to the JSON buffer after escaping special characters (e.g., quotes, backslashes) as required by JSON.
+- `escape_and_append_with_quotes(std::string_view input)` Appends a string view surrounded by double quotes (e.g., "input") to the JSON buffer after escaping special characters.
+Parameters:
+- `escape_and_append_with_quotes(char input)`: Appends a single character surrounded by double quotes (e.g., "c") to the JSON buffer after escaping it if necessary.
+- `append_raw(const char *c)`: Appends a null-terminated C string directly to the JSON buffer without escaping.
+- `append_raw(std::string_view input)`: Appends a string view directly to the JSON buffer without escaping.
+- `append_raw(const char *str, size_t len)`: Appends a specified number of characters from a C string directly to the JSON
+
+After writting the content, if you have reasons to believe that the content might violate UTF-8 conventions, you can check it as follows:
+
+- `validate_unicode()`: Checks if the content in the JSON buffer is valid UTF-8. Returns: true if the content is valid UTF-8, false otherwise.
+
+Once you are satisfied, you can recover the string as follows:
+
+- `operator std::string()`: Converts the JSON buffer to an std::string. (Might throw if an error occurred.)
+- `operator std::string_view()`: Converts the JSON buffer to an std::string_view.  (Might throw if an error occurred.)
+- `view()`: Returns a view of the written JSON buffer as a `simdjson_result<std::string_view>`.
+
+The later method (`view()`) is recommended.
+
+Example: string_builder
+---------------------------
+
+```C++
+
+    void serialize_car(const Car& car, simdjson::builder::string_builder& builder) {
+        // start of JSON
+        builder.append_raw("{");
+
+        // "make"
+        builder.escape_and_append_with_quotes("make");
+        builder.append_raw(":");
+        builder.escape_and_append_with_quotes(car.make);
+
+        // "model"
+        builder.append_raw(",");
+        builder.escape_and_append_with_quotes("model");
+        builder.append_raw(":");
+        builder.escape_and_append_with_quotes(car.model);
+
+        // "year"
+        builder.append_raw(",");
+        builder.escape_and_append_with_quotes("year");
+        builder.append_raw(":");
+        builder.append(car.year);
+
+        // "tire_pressure"
+        builder.append_raw(",");
+        builder.escape_and_append_with_quotes("tire_pressure");
+        builder.append_raw(":[");
+
+        // vector tire_pressure
+        for (size_t i = 0; i < car.tire_pressure.size(); ++i) {
+            builder.append(car.tire_pressure[i]);
+            if (i < car.tire_pressure.size() - 1) {
+                builder.append_raw(",");
+            }
+        }
+        // end of array
+        builder.append_raw("]");
+
+        // end of object
+        builder.append_raw("}");
+    }
+
+    bool car_test() {
+        simdjson::builder::string_builder sb;
+        Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+        serialize_car(c, sb);
+        std::string_view p;
+        if(sb.view().get(p)) {
+            return false; // there was an error
+        }
+        // p holds the JSON:
+        // "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}"
+        return true;
+    }
+```
+
+C++26 static reflection
+------------------------
+
+If you have a compiler with support C++26 static reflection, you can compile
+your code with the `SIMDJSON_STATIC_REFLECTION` macro set:
+
+```cpp
+#define SIMDJSON_STATIC_REFLECTION 1
+//...
+#include "simdjson.h"
+```
+
+And then you can append your data structures to a `string_builder` instance
+automatically. In most cases, it should work automatically:
+
+```cpp
+    bool car_test() {
+        simdjson::builder::string_builder sb;
+        Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+        append(sb, c);
+        std::string_view p;
+        if(sb.view().get(p)) {
+            return false; // there was an error
+        }
+        // p holds the JSON:
+        // "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}"
+        return true;
+    }
+```
+
+If you prefer, you can also create a string directly:
+
+```cpp
+  std::string json;
+  if(simdjson::builder::to_json_string(c).get(json)) {
+    // there was an error
+  } else {
+    // json contain the serialized JSON
+  }
+```

doc/dom.md

@@ -3,11 +3,12 @@ The Document-Object-Model (DOM) front-end
 
 An overview of what you need to know to use simdjson, with examples.
 
-* [DOM vs On Demand](#dom-vs-on-demand)
+* [DOM vs On-Demand](#dom-vs-on-demand)
 * [The Basics: Loading and Parsing JSON Documents](#the-basics-loading-and-parsing-json-documents-using-the-dom-front-end)
 * [Using the Parsed JSON](#using-the-parsed-json)
 * [C++17 Support](#c17-support)
 * [JSON Pointer](#json-pointer)
+* [JSONPath](#jsonpath)
 * [Error Handling](#error-handling)
   * [Error Handling Example](#error-handling-example)
   * [Exceptions](#exceptions)
@@ -18,7 +19,7 @@ An overview of what you need to know to use simdjson, with examples.
 * [Padding and Temporary Copies](#padding-and-temporary-copies)
 * [Performance Tips](#performance-tips)
 
-DOM vs On Demand
+DOM vs On-Demand
 ----------------------------------------------
 
 The simdjson library offers two distinct approaches on how to access a JSON document. We support
@@ -59,6 +60,26 @@ std::string data = "my data";
 simdjson::padded_string my_padded_data(data); // copies to a padded buffer
 ```
 
+You can then parse the JSON document from the `simdjson::padded_string` instance:
+
+```cpp
+simdjson::dom::parser parser;
+simdjson::dom::element doc = parser.parse(my_padded_data);
+```
+
+Whenever you pass an `std::string` reference to `parser::parse`,
+the parser will access the bytes beyond the end of
+the string but before the end of the allocated memory (`std::string::capacity()`).
+If you are using a sanitizer that checks for reading uninitialized bytes or `std::string`'s
+container-overflow checks, you may encounter sanitizer warnings.
+You can safely ignore these warnings. Or you can call `simdjson::pad(std::string&)` to pad the
+string with `SIMDJSON_PADDING` spaces: this function returns a `simdjson::padding_string_view` which can be be passed to the parser's iterator function:
+
+```c++
+std::string json = "[1]";
+dom::element doc = parser.parse(simdjson::pad(json));
+```
+
 The parsed document resulting from the `parser.load` and `parser.parse` calls depends on the `parser` instance. Thus the `parser` instance must remain in scope. Furthermore, you must have at most one parsed document in play per `parser` instance.
 You cannot copy a `parser` instance, you may only move it.
 
@@ -71,6 +92,12 @@ For best performance, a `parser` instance should be reused over several files: o
 If you need a lower-level interface, you may call the function `parser.parse(const char * p, size_t l)` on a pointer `p` while specifying the
 length of your input `l` in bytes. To see how to get the very best performance from a low-level approach, you way want to read our [performance notes](https://github.com/simdjson/simdjson/blob/master/doc/performance.md#padding-and-temporary-copies) on this topic (see the Padding and Temporary Copies section).
 
+*Windows-specific*:  Windows users who need to read files with
+non-ANSI characters in the name should set their code page to
+UTF-8 (65001). This should be the default with Windows 11 and better.
+Further, they may use the AreFileApisANSI function to determine whether
+the filename is interpreted using the ANSI or the system default OEM
+codepage, and they may call SetFileApisToOEM accordingly.
 
 Using the Parsed JSON
 ---------------------
@@ -251,6 +278,49 @@ for (dom::element car_element : cars) {
 }
 ```
 
+JSONPath
+------------
+
+
+The simdjson library supports a subset of [JSONPath](https://datatracker.ietf.org/doc/html/draft-normington-jsonpath-00) through the `at_path()` method, allowing you to reach further into the document in a single call. The subset of JSONPath that is implemented is the subset that is trivially convertible into the JSON Pointer format, using `.` to access a field and `[]` to access a specific index.
+
+Consider the following example:
+
+```c++
+auto cars_json = R"( [
+  { "make": "Toyota", "model": "Camry",  "year": 2018, "tire_pressure": [ 40.1, 39.9, 37.7, 40.4 ] },
+  { "make": "Kia",    "model": "Soul",   "year": 2012, "tire_pressure": [ 30.1, 31.0, 28.6, 28.7 ] },
+  { "make": "Toyota", "model": "Tercel", "year": 1999, "tire_pressure": [ 29.8, 30.0, 30.2, 30.5 ] }
+] )"_padded;
+dom::parser parser;
+dom::element doc;
+auto error = parser.parse(cars_json).get(doc);
+if(error) { /*won't happen*/ }
+double p;
+error = doc.at_path("[0].tire_pressure[1]").get(p);
+if(error) { /*won't happen*/ }
+cout << p << endl; // Prints 39.9
+```
+
+
+We also support the `$` prefix. When you start a JSONPath expression with $, you are indicating that the path starts from the root of the JSON document. E.g.,
+
+```c++
+auto json = R"( { "c" :{ "foo": { "a": [ 10, 20, 30 ] }}, "d": { "foo2": { "a": [ 10, 20, 30 ] }} , "e": 120 })"_padded;
+dom::parser parser;
+dom::element doc;
+auto error = parser.parse(json).get(doc);
+if(error) { /*won't happen*/ }
+dom::object obj;
+error = doc.get_object().get(obj);
+if(error) { /*won't happen*/ }
+int64_t x;
+error = obj.at_path("$[3].foo.a[1]").get(x);
+if(error) { /*won't happen*/ }
+if(x != 20) { /*won't happen*/ }
+x = obj.at_path("$.d.foo2.a.2");
+if(error) { /*won't happen*/ }
+```
 
 
 Error Handling

doc/implementation-selection.md

@@ -24,6 +24,8 @@ The current implementations are:
 * westmere: SSE4.2 (2010 Westmere or later).
 * arm64: 64-bit ARMv8-A NEON
 * ppc64: 64-bit POWER8 and POWER9 with VSX and ALTIVEC extensions. Both big endian and little endian are implemented, depending on the compiler you are using. The library is tested on recent, little-endian, POWER systems.
+* lasx: Loongson Advanced SIMD EXtension (LASX),  a 256-bit vector expansion for the LoongArch architecture.
+* lsx: Loongson SIMD EXtension (LSX),  a 128-bit vector expansion for the LoongArch architecture.
 * fallback: A generic implementation that runs on any 64-bit processor.
 
 In many cases, you don't know where your compiled binary is going to run, so simdjson automatically
@@ -33,7 +35,7 @@ it will include 2 (ppc64 and fallback).
 
 If you know more about where you're going to run and want to save the space, you can disable any of
 these implementations at compile time with `-DSIMDJSON_IMPLEMENTATION_X=0` (where X is ICELAKE, HASWELL,
-WESTMERE, ARM64, PPC64 and FALLBACK).
+WESTMERE, ARM64, PPC64, LSX, LASX and FALLBACK).
 
 The simdjson library automatically sets header flags for each implementation as it compiles; there
 is no need to set architecture-specific flags yourself (e.g., `-mavx2`, `/AVX2`  or

doc/iterate_many.md

@@ -25,6 +25,7 @@ Contents
 - [Use cases](#use-cases)
 - [Tracking your position](#tracking-your-position)
 - [Incomplete streams](#incomplete-streams)
+- [C++20 features](#c20-features)
 
 Motivation
 -----------
@@ -102,7 +103,12 @@ remove almost entirely its cost and replaces it by the overhead of a thread, whi
 cheaper. Ain't that awesome!
 
 Thread support is only active if thread supported is detected in which case the macro
-SIMDJSON_THREADS_ENABLED is set. Otherwise the library runs in  single-thread mode.
+SIMDJSON_THREADS_ENABLED is set.  You can also manually pass `SIMDJSON_THREADS_ENABLED=1` flag
+to the library. Otherwise the library runs in single-thread mode.
+
+You should be consistent. If you link against the simdjson library built for multithreading
+(i.e., with `SIMDJSON_THREADS_ENABLED`), then you should build your application with multithreading
+system (setting `SIMDJSON_THREADS_ENABLED=1` and linking against a thread library).
 
 A `document_stream` instance uses at most two threads: there is a main thread and a worker thread.
 
@@ -124,7 +130,7 @@ If your documents are all objects or arrays, then you may even have nothing betw
 E.g., `[1,2]{"32":1}` is recognized as two documents.
 
 Some official formats **(non-exhaustive list)**:
-- [Newline-Delimited JSON (NDJSON)](http://ndjson.org/)
+- [Newline-Delimited JSON (NDJSON)](https://github.com/ndjson/ndjson-spec/)
 - [JSON lines (JSONL)](http://jsonlines.org/)
 - [Record separator-delimited JSON (RFC 7464)](https://tools.ietf.org/html/rfc7464) <- Not supported by JsonStream!
 - [More on Wikipedia...](https://en.wikipedia.org/wiki/JSON_streaming)
@@ -288,3 +294,111 @@ string
 object
 array
 ```
+
+
+C++20 features
+--------------------
+
+In C++20, the standard introduced the notion of *customization point*.
+A customization point is a function or function object that can be customized for different types. It allows library authors to provide default behavior while giving users the ability to override this behavior for specific types.
+
+A tag_invoke function serves as a mechanism for customization points. It is not directly part of the C++ standard library but is often used in libraries that implement customization points.
+The tag_invoke function is typically a generic function that takes a tag type and additional arguments.
+The first argument is usually a tag type (often an empty struct) that uniquely identifies the customization point (e.g., deserialization of custom types in simdjson). Users or library providers can specialize tag_invoke for their types by defining it in the appropriate namespace, often inline namespace.
+
+
+
+You can deserialize you own data structures conveniently if your system supports C++20.
+When it is the case, the macro `SIMDJSON_SUPPORTS_DESERIALIZATION` will be set to 1 by
+the simdjson library.
+
+Consider a custom class `Car`:
+
+```C++
+struct Car {
+  std::string make;
+  std::string model;
+  int year;
+  std::vector<float> tire_pressure;
+};
+```
+
+
+You may support deserializing directly from a JSON value or document to your own `Car` instance
+by defining a single `tag_invoke` function:
+
+
+```C++
+namespace simdjson {
+// This tag_invoke MUST be inside simdjson namespace
+template <typename simdjson_value>
+auto tag_invoke(deserialize_tag, simdjson_value &val, Car& car) {
+  ondemand::object obj;
+  auto error = val.get_object().get(obj);
+  if (error) {
+    return error;
+  }
+  if ((error = obj["make"].get_string(car.make))) {
+    return error;
+  }
+  if ((error = obj["model"].get_string(car.model))) {
+    return error;
+  }
+  if ((error = obj["year"].get(car.year))) {
+    return error;
+  }
+  if ((error = obj["tire_pressure"].get<std::vector<float>>().get(
+           car.tire_pressure))) {
+    return error;
+  }
+  return simdjson::SUCCESS;
+}
+} // namespace simdjson
+```
+
+Importantly, the `tag_invoke` function must be inside the `simdjson` namespace.
+Let us explain each argument of `tag_invoke` function.
+
+- `simdjson::deserialize_tag`: it is the tag for Customization Point Object (CPO). You may often ignore this parameter. It is used to indicate that you mean to provide a deserialization function for simdjson.
+- `var`: It receives automatically a `simdjson` value type (document, value, document_reference).
+- The third parameter is an instance of the type that you want to support.
+
+Please see our main documentation (`basics.md`) under
+"Use `tag_invoke` for custom types (C++20)" for details about
+tag_invoke functions.
+
+Given a stream of JSON documents, you can add them to a data struture
+such as a `std::vector<Car>` like so if you support exceptions:
+
+```C++
+  padded_string json =
+      R"( { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] }
+  { "make": "Kia",    "model": "Soul",   "year": 2012,
+       "tire_pressure": [ 30.1, 31.0 ] }
+  { "make": "Toyota", "model": "Tercel", "year": 1999,
+       "tire_pressure": [ 29.8, 30.0 ] }
+)"_padded;
+  ondemand::parser parser;
+  ondemand::document_stream stream;
+  [[maybe_unused]] auto error = parser.iterate_many(json).get(stream);
+  std::vector<Car> cars;
+  for(auto doc : stream) {
+    cars.push_back((Car)doc); // an exception may be thrown
+  }
+```
+
+Otherwise you may use this longer version for explicit handling of errors:
+
+
+```C++
+  std::vector<Car> cars;
+  for(auto doc : stream) {
+    Car c;
+    if ((error = doc.get<Car>().get(c))) {
+      std::cerr << simdjson::error_message(error); << std::endl;
+      return EXIT_FAILURE;
+    }
+    cars.push_back(c);
+  }
+```

doc/ondemand_design.md

@@ -9,8 +9,8 @@ Whether we parse JSON or XML, or any other serialized format, there are relative
 - Another popular approach is the schema-based deserialization model.
 
 We propose an approach that is as easy to use and often as flexible as the DOM approach, yet as fast and
-efficient as the schema-based or event-based approaches. We call this new approach "On Demand". The
+efficient as the schema-based or event-based approaches. We call this new approach "On-Demand". The
-simdjson On Demand API offers a familiar, friendly DOM API and
+simdjson On-Demand API offers a familiar, friendly DOM API and
 provides the performance of just-in-time parsing on top of the simdjson superior performance.
 
 To achieve ease of use, we mimicked the *form* of a traditional DOM API: you can iterate over
@@ -18,7 +18,7 @@ arrays, look up fields in objects, and extract native values like `double`, `uin
 
 To achieve performance, we introduced some key limitations that make the DOM API *streaming*:
 array/object iteration cannot be restarted, and string/number values can only be parsed once. If
-these limitations are acceptable to you, the On Demand API could help you write maintainable
+these limitations are acceptable to you, the On-Demand API could help you write maintainable
 applications with a computation efficiency that is difficult to surpass.
 
 A code example illustrates our API from a programmer's point of view:
@@ -72,24 +72,24 @@ This streaming approach means that unused fields and values are not parsed or
 converted, thus saving space and time. In our example, the `"name"`, `"followers_count"`,
 and `"friends_count"` keys and matching values are skipped.
 
-Further, the On Demand API does not parse a value *at all* until you try to convert it (e.g., to `double`,
+Further, the On-Demand API does not parse a value *at all* until you try to convert it (e.g., to `double`,
 `int`, `string`, or `bool`). In our example, when accessing the key-value pair `"retweet_count": 82`, the parser
 may not convert the pair of characters `82` to the binary integer 82. Because the programmer specifies the data
 type, we avoid branch mispredictions related to data type determination and improve the performance.
 
 
-We expect users of an On Demand API to work in terms of a JSON dialect, which is a set of expectations and
+We expect users of an On-Demand API to work in terms of a JSON dialect, which is a set of expectations and
 specifications that come in addition to the [JSON specification](https://www.rfc-editor.org/rfc/rfc8259.txt).
-The On Demand approach is designed around several principles:
+The On-Demand approach is designed around several principles:
 
 * **Streaming (\*):** It avoids preparsing values, keeping the memory usage and the latency down.
 * **Forward-Only:** To prevent reiteration of the same values and to keep the number of variables down (literally), only a single index is maintained and everything uses it (even if you have nested for loops). This means when you are going through an array of arrays, for example, that the inner array loop will advance the index to the next comma, and the array can just pick it up and look at it.
 * **Natural Iteration:** A JSON array or object can be iterated with a normal C++ for loop. Nested arrays and objects are supported by nested for loops.
 * **Use-Specific Parsing:** Parsing is always specific to the type required by the programmer. For example, if the programmer asks for an unsigned integer, we just start parsing digits. If there were no digits, we toss an error. There are even different parsers for `double`, `uint64_t` and `int64_t` values. This use-specific parsing avoids the branchiness of a generic "type switch," and makes the code more inlineable and compact.
-* **Validate What You Use:** On Demand deliberately validates the values you use and the structure leading to it, but nothing else. The goal is a guarantee that the value you asked for is the correct one and is not malformed: there must be no confusion over whether you got the right value.
+* **Validate What You Use:** On-Demand deliberately validates the values you use and the structure leading to it, but nothing else. The goal is a guarantee that the value you asked for is the correct one and is not malformed: there must be no confusion over whether you got the right value.
 
 
-To understand why On Demand is different, it is helpful to review the major
+To understand why On-Demand is different, it is helpful to review the major
 approaches to parsing and parser APIs in use today.
 
 ### DOM Parsers
@@ -102,11 +102,11 @@ or indexing (`object["key"]`). In some cases, the values are even deserialized d
 maps.
 
 The DOM approach is conceptually simple and "programmer friendly". Using the
-DOM tree is often easy enough that many users use the DOM as-is instead of creating
+DOM tree is often easy enough that many users process the DOM as-is instead of creating
 their own custom data structures.
 
 The DOM approach was the only way to parse JSON documents up to version 0.6 of the simdjson library.
-Our DOM API looks similar to our On Demand example, except
+Our DOM API looks similar to our On-Demand example, except
 it calls `parse` instead of `iterate`:
 
 ```c++
@@ -152,7 +152,7 @@ a tweet right now, or is this from some other place in the document
 entirely? Though an event-based approach may allow superior performance, it is demanding of the programmer
 who must efficiently keep track of its current state within the JSON input.
 
-The following is event-based example of the Twitter problem we have reviewed in the DOM and On Demand
+The following is event-based example of the Twitter problem we have reviewed in the DOM and On-Demand
 examples. To make it short enough to use as an example at all, it has heavily redacted: it only solves
 a part of the problem (does not get user.screen_name), it has bugs (it does not handle sub-objects
 in a tweet at all), and it uses a theoretical, simple event-based API that minimizes ceremony.
@@ -257,7 +257,7 @@ stress the branch prediction. Though branch predictors improve with each new gen
 the cost of branch mispredictions also tends to increase as pipelines expand, and the processors become
 able to schedule longer streams of instructions.
 
-On Demand parsing is tailor-made to solve this problem at the source, parsing values only after the
+On-Demand parsing is tailor-made to solve this problem at the source, parsing values only after the
 user declares their type by asking for a `double`, an `int`, a `string`, etc. It attempts to do so while
 preserving most of the flexibility of DOM parsing.
 
@@ -297,7 +297,7 @@ To help visualize the algorithm, we'll walk through the example C++ given at the
 
    Since this is the first time this parser has been used, `iterate()` first allocates internal
    parser buffers if this is the first time through. When reusing an existing parser, allocation
-   only happens if the new document is bigger than internal buffers can handle. The On Demand
+   only happens if the new document is bigger than internal buffers can handle. The On-Demand
    API only ever allocates memory in the `iterate()` function call.
 
    The simdjson library then preprocesses the JSON text at high speed, finding all tokens (i.e. the starting
@@ -492,7 +492,7 @@ To help visualize the algorithm, we'll walk through the example C++ given at the
    Because of the cast to uint64_t, simdjson knows it's parsing an unsigned integer. This lets
    us use a fast parser which *only* knows how to parse digits. It validates that it is an integer
    by rejecting negative numbers, strings, and other values based on the fact that they are not the
-   digits 0-9. This type specificity is part of why parsing with on demand is so fast: you lose all
+   digits 0-9. This type specificity is part of why parsing with On-Demand is so fast: you lose all
    the code that has to understand those other types.
 
    The iterator is advanced to the `}`, and depth decreased back to 3 (root > statuses > tweet).
@@ -597,7 +597,7 @@ To help visualize the algorithm, we'll walk through the example C++ given at the
 
    This means you can very efficiently do things like read a single value from a JSON file, or take
    the top N, for example. It also means the things you don't use won't be fully validated. This is
-   a general principle of On Demand: don't validate what you don't use. We still fully validate
+   a general principle of On-Demand: don't validate what you don't use. We still fully validate
    values you do use, however, as well as the objects and arrays that lead to them, so that you can
    be sure you get the information you need.
 
@@ -654,7 +654,7 @@ for(auto field : doc.get_object())  {
 
 ### Iteration Safety
 
-The On Demand API is powerful. To compensate, we add some safeguards to ensure that it can be used without fear
+The On-Demand API is powerful. To compensate, we add some safeguards to ensure that it can be used without fear
 in production systems:
 
   - If the value fails to be parsed as one type, the program can try to parse it as something else until the program succeeds. Thus
@@ -667,7 +667,7 @@ in production systems:
     if it was `nullptr` but did not care what the actual value was--it will iterate. The destructor automates
     the iteration.
 
-  Some care is needed when using the On Demand API in scenarios where you need to access several sibling arrays or objects because
+  Some care is needed when using the On-Demand API in scenarios where you need to access several sibling arrays or objects because
   only one object or array can be active at any one time. Let us consider the following example:
 
 ```C++
@@ -709,36 +709,36 @@ A correct usage is given by the following example:
     }
 ```
 
-### Benefits of the On Demand Approach
+### Benefits of the On-Demand Approach
 
-We expect that the On Demand approach has many of the performance benefits of the schema-based approach, while providing a flexibility that is similar to that of the DOM-based approach.
+We expect that the On-Demand approach has many of the performance benefits of the schema-based approach, while providing a flexibility that is similar to that of the DOM-based approach.
 
 * Faster than DOM in some cases. Reduced memory usage.
 * Straightforward, programmer-friendly interface (arrays and objects).
 * Highly expressive, beyond deserialization and pointer queries: many tasks can be accomplished with little code.
 
-### Limitations of the On Demand Approach
+### Limitations of the On-Demand Approach
 
-The On Demand approach has some limitations:
+The On-Demand approach has some limitations:
 
 * Because it operates in streaming mode, you only have access to the current element in the JSON document. Furthermore, the document is traversed in order so the code is sensitive to the order of the JSON nodes in the same manner as an event-based approach (e.g., SAX). (The one exception to this is field lookup, which is more *performant* when the order of lookups matches the order of fields in the document, but which will still work with out-of-order fields, with a performance hit.)
-* The On Demand approach is less safe than DOM: we only validate the components of the JSON document that are used and it is possible to begin ingesting an invalid document only to find out later that the document is invalid. Are you fine ingesting a large JSON document that starts with well formed JSON but ends with invalid JSON content?
+* The On-Demand approach is less safe than DOM: we only validate the components of the JSON document that are used and it is possible to begin ingesting an invalid document only to find out later that the document is invalid. Are you fine ingesting a large JSON document that starts with well formed JSON but ends with invalid JSON content?
 
 There are currently additional technical limitations which we expect to resolve in future releases of the simdjson library:
 
-* The simdjson library offers runtime dispatching which allows you to compile one binary and have it run at full speed on different processors, taking advantage of the specific features of the processor. The On Demand API has limited runtime dispatch support. Under x64 systems, to fully benefit from the On Demand API, we recommend that you compile your code for a specific processor. E.g., if your processor supports AVX2 instructions, you should compile your binary executable with AVX2 instruction support (by using your compiler's commands). If you are sufficiently technically proficient, you can implement runtime dispatching within your application, by compiling your On Demand code for different processors.
+* The simdjson library offers runtime dispatching which allows you to compile one binary and have it run at full speed on different processors, taking advantage of the specific features of the processor. The On-Demand API has limited runtime dispatch support. Under x64 systems, to fully benefit from the On-Demand API, we recommend that you compile your code for a specific processor. E.g., if your processor supports AVX2 instructions, you should compile your binary executable with AVX2 instruction support (by using your compiler's commands). If you are sufficiently technically proficient, you can implement runtime dispatching within your application, by compiling your On-Demand code for different processors.
 * There is an initial phase which scans the entire document quickly, irrespective of the size of the document. We plan to break this phase into distinct steps for large files in a future release as we have done with other components of our API (e.g., `parse_many`).
 
-### Applicability of the On Demand Approach
+### Applicability of the On-Demand Approach
 
-At this time we recommend the On Demand API in the following cases:
+At this time we recommend the On-Demand API in the following cases:
 
 1. The 64-bit hardware (CPU) used to run the software is known at compile time. If you need runtime dispatching because you cannot be certain of the hardware used to run your software, you will be better served with the core simdjson API. (This only applies to x64 (AMD/Intel). On 64-bit ARM hardware, runtime dispatching is unnecessary.)
 2. The used parts of JSON files do not need to be validated and the layout of the nodes follows a strict JSON dialect. If you are receiving JSON from other systems, you might be better served with core simdjson API as it fully validates the JSON inputs and allows you to navigate through the document at will.
-3. Speed and efficiency are of the utmost importance. Keep in mind that the core simdjson API is highly efficient so adopting the On Demand API is not necessary for high efficiency.
+3. Speed and efficiency are of the utmost importance. Keep in mind that the core simdjson API is highly efficient so adopting the On-Demand API is not necessary for high efficiency.
 4. As a developer, you value a clean, flexible and maintainable API.
 
-Good applications for the On Demand API might be:
+Good applications for the On-Demand API might be:
 
 * You are working from pre-existing large JSON files that have been vetted. You expect them to be well formed according to a known JSON dialect and to have a consistent layout. For example, you might be doing biomedical research or machine learning on top of static data dumps in JSON.
 * Both the generation and the consumption of JSON data is within your system. Your team controls both the software that produces the JSON and the software the parses it, your team knows and control the hardware. Thus you can fully test your system.
@@ -746,13 +746,13 @@ Good applications for the On Demand API might be:
 
 ## Checking Your CPU Selection (x64 systems)
 
-The On Demand API uses advanced architecture-specific code for many common processors to make JSON preprocessing and string parsing faster. By default, however, most c++ compilers will compile to the	least common denominator (since the program could theoretically be run anywhere). Since On Demand is inlined into your own code, it cannot always use these advanced versions unless the compiler is told to target them.
+The On-Demand API uses advanced architecture-specific code for many common processors to make JSON preprocessing and string parsing faster. By default, however, most c++ compilers will compile to the	least common denominator (since the program could theoretically be run anywhere). Since On-Demand is inlined into your own code, it cannot always use these advanced versions unless the compiler is told to target them.
 
-On relevant systems, the On Demand API provides some support for runtime dispatching: that is, it will attempt to detect, at runtime, the instructions that your processor supports and optimize the code accordingly. However, it cannot always make full use of the features of your processor.
+On relevant systems, the On-Demand API provides some support for runtime dispatching: that is, it will attempt to detect, at runtime, the instructions that your processor supports and optimize the code accordingly. However, it cannot always make full use of the features of your processor.
 
 Some users wish to run at the best possible speed. Under recent Intel and AMD processors, these users should take additional steps to verify that their code is well optimized.
 
-Given that the On Demand API offer limited runtime dispatching, it matters that your code is compiled against a specific CPU target. You should verify that the code is compiled against the target you expect. Thankfully, the simdjson library will tell you exactly what it detects as an implementation: `icelake` (AVX512 x64 processors), `haswell` (AVX2 x64 processors), `westmere` (SSE4 x64 processors), `arm64` (64-bit ARM), `ppc64` (64-bit POWER), `fallback` (others). Under x64 processors, many programmers will want to target `haswell` whereas under ARM, most programmers will want to target `arm64` (and it should do so automatically). The `fallback` is probably only good for testing purposes, not for deployment.
+Given that the On-Demand API offer limited runtime dispatching, it matters that your code is compiled against a specific CPU target. You should verify that the code is compiled against the target you expect. Thankfully, the simdjson library will tell you exactly what it detects as an implementation: `icelake` (AVX512 x64 processors), `haswell` (AVX2 x64 processors), `westmere` (SSE4 x64 processors), `arm64` (64-bit ARM), `ppc64` (64-bit POWER), `lasx` (LoongArch), `lsx` (LoongArch),  `fallback` (others). Under x64 processors, many programmers will want to target `haswell` whereas under ARM, most programmers will want to target `arm64` (and it should do so automatically). The `fallback` is probably only good for testing purposes, not for deployment.
 
 ```C++
   std::cout << simdjson::builtin_implementation()->name() << std::endl;
@@ -777,6 +777,6 @@ In these examples, the `-march=haswell` flags targets a haswell processor and th
 
 Instead of specifying a specific microarchitecture, you can let your compiler do the work. The `-march=native` flags says "target the current computer," which is a reasonable default for many applications which both compile and run on the same processor.
 
-Passing `-march=native` to the compiler may make On Demand faster by allowing it to use optimizations specific to your machine. You cannot do this, however, if you are compiling code	that might be run on less advanced machines. That is, be mindful that when compiling with the `-march=native` flag, the resulting binary will run on the current system but may not run on other systems (e.g., on an old processor).
+Passing `-march=native` to the compiler may make On-Demand faster by allowing it to use optimizations specific to your machine. You cannot do this, however, if you are compiling code	that might be run on less advanced machines. That is, be mindful that when compiling with the `-march=native` flag, the resulting binary will run on the current system but may not run on other systems (e.g., on an old processor).
 
 If you are compiling on an ARM or POWER system, you do not need to be concerned with CPU selection during compilation. The `-march=native` flag is useful for best performance on x64 (e.g., Intel) systems but it is generally unsupported on some platforms such as ARM (aarch64) or POWER.

doc/parse_many.md

@@ -103,7 +103,12 @@ cases, remove almost entirely its cost and replaces it by the overhead of a thre
 of magnitude cheaper. Ain't that awesome!
 
 Thread support is only active if thread supported is detected in which case the macro
-SIMDJSON_THREADS_ENABLED is set. Otherwise the library runs in  single-thread mode.
+SIMDJSON_THREADS_ENABLED is set.  You can also manually pass `SIMDJSON_THREADS_ENABLED=1` flag
+to the library. Otherwise the library runs in single-thread mode.
+
+You should be consistent. If you link against the simdjson library built for multithreading
+(i.e., with `SIMDJSON_THREADS_ENABLED`), then you should build your application with multithreading
+system (setting `SIMDJSON_THREADS_ENABLED=1` and linking against a thread library).
 
 A `document_stream` instance uses at most two threads: there is a main thread and a worker thread.
 You should expect the main thread to be fully occupied while the worker thread is partially busy
@@ -125,7 +130,7 @@ Whitespace Characters:
 - **Nothing**
 
 Some official formats **(non-exhaustive list)**:
-- [Newline-Delimited JSON (NDJSON)](http://ndjson.org/)
+- [Newline-Delimited JSON (NDJSON)](https://github.com/ndjson/ndjson-spec)
 - [JSON lines (JSONL)](http://jsonlines.org/)
 - [Record separator-delimited JSON (RFC 7464)](https://tools.ietf.org/html/rfc7464) <- Not supported by JsonStream!
 - [More on Wikipedia...](https://en.wikipedia.org/wiki/JSON_streaming)

doc/performance.md

@@ -14,6 +14,7 @@ testing and get the best performance.
 * [Number parsing](#number-parsing)
 * [Visual Studio](#visual-studio)
 * [Power Usage and Downclocking](#power-usage-and-downclocking)
+* [Free Padding](#free-padding)
 
 
 NDEBUG directive
@@ -74,7 +75,7 @@ or simply
 Server Loops: Long-Running Processes and Memory Capacity
 ---------------------------------
 
-The On Demand approach also automatically expands its memory capacity when larger documents are parsed. However, for longer processes where very large files are processed (such as server loops), this capacity is not resized down. On Demand also lets you adjust the maximal capacity that the parser can process:
+The On-Demand approach also automatically expands its memory capacity when larger documents are parsed. However, for longer processes where very large files are processed (such as server loops), this capacity is not resized down. On-Demand also lets you adjust the maximal capacity that the parser can process:
 
 * You can set an upper bound (*max_capacity*) when construction the parser:
 ```C++
@@ -157,7 +158,9 @@ On Intel and AMD Windows platforms, Microsoft Visual Studio enables programmers
 
 When compiling with Visual Studio, we recommend the flags `/Ob2 /O2` or better. We do not recommend that you compile simdjson with architecture-specific flags such as  `arch:AVX2`. The simdjson library automatically selects the best execution kernel at runtime.
 
-Recent versions of Microsoft Visual Studio on Windows provides support for the LLVM Clang compiler. You  only need to install the "Clang compiler" optional component (ClangCL). You may also get a copy of the 64-bit LLVM CLang compiler for [Windows directly from LLVM](https://releases.llvm.org/download.html). The simdjson library fully supports the LLVM Clang compiler under Windows. In fact, you may get better performance out of simdjson with the LLVM Clang compiler than with the regular Visual Studio compiler. Meanwhile the [LLVM CLang compiler is binary compatible with Visual Studio](https://clang.llvm.org/docs/MSVCCompatibility.html) which means that you can combine their binaries (executables and libraries).
+Recent versions of Microsoft Visual Studio on Windows provides support for the LLVM Clang compiler. You  only need to install the "Clang compiler" optional component (clang-cl). You may also get a copy of the 64-bit LLVM CLang compiler for [Windows directly from LLVM](https://releases.llvm.org/download.html). The simdjson library fully supports the LLVM Clang compiler under Windows. In fact, you may get better performance out of simdjson with the LLVM Clang compiler than with the regular Visual Studio compiler. Meanwhile the [LLVM CLang compiler is binary compatible with Visual Studio](https://clang.llvm.org/docs/MSVCCompatibility.html) which means that you can combine their binaries (executables and libraries).
+
+We recommend Visual Studio users prefer LLVM (clang-cl). It compiles to faster release binaries. Furthermore, it compilers faster in release mode.
 
 Under Windows, we also support the GNU GCC compiler via MSYS2. The performance of 64-bit MSYS2 under Windows is excellent (on par with Linux).
 
@@ -179,3 +182,110 @@ The simdjson library does not generally make use of heavy 256-bit instructions.
 the macro `SIMDJSON_AVX512_ALLOWED` to `0` in C++ prior to importing the headers.
 
 You may still be worried about which SIMD instruction set is used by simdjson.  Thankfully,  [you can always determine and change which architecture-specific implementation is used](implementation-selection.md) by simdjson. Thus even if your CPU supports AVX2, you do not need to use AVX2. You are in control.
+
+
+Free Padding
+-------
+
+For performance reasons, the simdjson library requires that the JSON input contain at least
+`simdjson::SIMDJSON_PADDING` bytes at the end of the stream. The value `simdjson::SIMDJSON_PADDING` is
+small (e.g., 64 bytes). On modern systems, you can safely read beyond an allocated buffers,
+as long as you remain within an allocated page. Pages on modern systems span at least 4 kilobytes,
+but can be significantly larger. E.g., Apple systems favour pages spanning 16 kilobytes.
+
+In effect, it means that you can almost always read a few bytes beyond your current buffer---without
+allocating extra memory. However, tools such as valgrind or memory sanitizers will flag such behavior as unsafe.
+Nevertheless, you can still make sure of this capability in your code if you are an expert
+programmer and you are willing to silence sanitizer warnings. The following code provides
+a portable example.
+
+
+The conditional compilation checks for the `_MSC_VER` macro (indicating Microsoft Visual Studio)
+and includes platform-specific headers accordingly.
+The `page_size()` function determines the default size of a memory page in bytes on the system.
+On Windows (when `_WIN32` is defined), it uses `GetSystemInfo()` to retrieve system information and obtain the page size.
+On other platforms (non-Windows), it uses `sysconf(_SC_PAGESIZE)` to get the page size.
+The function returns the page size.
+The `need_allocation()` function checks whether the buffer (given by `buf`) plus the specified length (`len`) is near a page boundary.
+If the buffer extends beyond the current page when padded by `simdjson::SIMDJSON_PADDING`, it returns true, indicating that reallocation is needed.
+Otherwise, it returns false.
+The `get_padded_string_view()`  creates a `padded_string_view` from the input buffer.
+If reallocation is needed (unlikely case), it allocates a new padded_string and assigns it to `jsonbuffer`.
+Otherwise (very likely), it creates a `padded_string_view` directly from the buffer.
+The `simdjson::SIMDJSON_PADDING` ensures that there is additional padding for parsing efficiency.
+The calling code just needs to provide `jsonbuffer` (an instance of `simdjson::padded_string`)
+and pass `get_padded_string_view(buf, len, jsonbuffer)` to  `parser.iterate`. Most of the time,
+this code will not allocate new memory.
+
+
+```cpp
+#ifdef _WIN32
+#include <windows.h>
+#include <sysinfoapi.h>
+#else
+#include <unistd.h>
+#endif
+#include "simdjson.h"
+#include <cstdio>
+
+// Returns the default size of the page in bytes on this system.
+long page_size() {
+#ifdef _WIN32
+  SYSTEM_INFO sysInfo;
+  GetSystemInfo(&sysInfo);
+  long pagesize = sysInfo.dwPageSize;
+#else
+  long pagesize = sysconf(_SC_PAGESIZE);
+#endif
+  return pagesize;
+}
+
+// Returns true if the buffer + len + simdjson::SIMDJSON_PADDING crosses the
+// page boundary.
+bool need_allocation(const char *buf, size_t len) {
+  return ((reinterpret_cast<uintptr_t>(buf + len - 1) % page_size())
+    + simdjson::SIMDJSON_PADDING > static_cast<uintptr_t>(page_size()));
+}
+
+simdjson::padded_string_view
+get_padded_string_view(const char *buf, size_t len,
+                       simdjson::padded_string &jsonbuffer) {
+  if (need_allocation(buf, len)) { // unlikely case
+    jsonbuffer = simdjson::padded_string(buf, len);
+    return jsonbuffer;
+  } else { // no reallcation needed (very likely)
+    return simdjson::padded_string_view(buf, len,
+                                            len + simdjson::SIMDJSON_PADDING);
+  }
+}
+
+int main() {
+  printf("page_size: %ld\n", page_size());
+  const char *jsonpoiner = R"(
+        {
+            "key": "value"
+        }
+    )";
+  size_t len = strlen(jsonpoiner);
+  simdjson::padded_string jsonbuffer; // only allocate if needed
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc;
+  simdjson::error_code error =
+      parser.iterate(get_padded_string_view(jsonpoiner, len, jsonbuffer))
+          .get(doc);
+  if (error) {
+    printf("error: %s\n", simdjson::error_message(error));
+    return EXIT_FAILURE;
+  }
+  std::string_view value;
+  error = doc["key"].get_string().get(value);
+  if (error) {
+    return EXIT_FAILURE;
+  }
+  printf("Value: \"%.*s\"\n", (int)value.size(), value.data());
+  if (value != "value") {
+    return EXIT_FAILURE;
+  }
+  return EXIT_SUCCESS;
+}
+```

examples/quickstart/CMakeLists.txt

@@ -20,12 +20,16 @@ IF(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
   endif()
 
   add_quickstart_test(quickstart_noexceptions   quickstart_noexceptions.cpp NO_EXCEPTIONS LABELS acceptance)
+  if(NOT SIMDJSON_STATIC_REFLECTION)
     add_quickstart_test(quickstart_noexceptions11 quickstart_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11)
+  endif(NOT SIMDJSON_STATIC_REFLECTION)
 
   add_quickstart_test(quickstart2_noexceptions   quickstart2_noexceptions.cpp NO_EXCEPTIONS LABELS acceptance)
+  if(NOT SIMDJSON_STATIC_REFLECTION)
     add_quickstart_test(quickstart2_noexceptions11 quickstart2_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11)
+  endif(NOT SIMDJSON_STATIC_REFLECTION)
 
-  # On Demand Quick Start
+  # On-Demand Quick Start
   if (SIMDJSON_EXCEPTIONS)
     add_quickstart_test(quickstart_ondemand   quickstart_ondemand.cpp LABELS quickstart_ondemand acceptance)
     add_quickstart_test(quickstart_ondemand11 quickstart_ondemand.cpp CXX_STANDARD c++11 LABELS quickstart_ondemand acceptance)
@@ -33,6 +37,8 @@ IF(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
   endif()
 
   add_quickstart_test(quickstart_ondemand_noexceptions   quickstart_ondemand_noexceptions.cpp NO_EXCEPTIONS LABELS quickstart_ondemand acceptance)
+  if(NOT SIMDJSON_STATIC_REFLECTION)
     add_quickstart_test(quickstart_ondemand_noexceptions11 quickstart_ondemand_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11 LABELS quickstart_ondemand)
+  endif(NOT SIMDJSON_STATIC_REFLECTION)
 
 endif()

fuzz/fuzz_implementations.cpp

@@ -72,7 +72,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
 
     // make this dynamic, so it works regardless of how it was compiled
     // or what hardware it runs on
-    constexpr std::size_t Nimplementations_max=3;
+    constexpr std::size_t Nimplementations_max=4;
     const std::size_t Nimplementations = supported_implementations.size();
 
     if(Nimplementations>Nimplementations_max) {

include/simdjson.h

@@ -53,5 +53,4 @@
 
 #include "simdjson/dom.h"
 #include "simdjson/ondemand.h"
-
 #endif // SIMDJSON_H

include/simdjson/arm64/bitmanipulation.h

@@ -19,7 +19,7 @@ SIMDJSON_NO_SANITIZE_UNDEFINED
 // See issue https://github.com/simdjson/simdjson/issues/1965
 SIMDJSON_NO_SANITIZE_MEMORY
 simdjson_inline int trailing_zeroes(uint64_t input_num) {
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
   unsigned long ret;
   // Search the mask data from least significant bit (LSB)
   // to the most significant bit (MSB) for a set bit (1).
@@ -35,9 +35,15 @@ simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
   return input_num & (input_num-1);
 }
 
+// We sometimes call leading_zeroes on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+// Applies only when SIMDJSON_PREFER_REVERSE_BITS is defined and true.
+// (See below.)
+SIMDJSON_NO_SANITIZE_UNDEFINED
 /* result might be undefined when input_num is zero */
 simdjson_inline int leading_zeroes(uint64_t input_num) {
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
   unsigned long leading_zero = 0;
   // Search the mask data from most significant bit (MSB)
   // to least significant bit (LSB) for a set bit (1).
@@ -90,7 +96,7 @@ simdjson_inline uint64_t zero_leading_bit(uint64_t rev_bits, int leading_zeroes)
 #endif
 
 simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2, uint64_t *result) {
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
   *result = value1 + value2;
   return *result < value1;
 #else

include/simdjson/arm64/implementation.h

@@ -23,6 +23,7 @@ public:
   ) const noexcept final;
   simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
   simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+  simdjson_warn_unused size_t write_string_escaped(const std::string_view input, char *out) const noexcept final;
 };
 
 } // namespace arm64

include/simdjson/arm64/numberparsing_defs.h

@@ -9,10 +9,10 @@
 
 #include <cstring>
 
-#if _M_ARM64
+#if SIMDJSON_REGULAR_VISUAL_STUDIO && SIMDJSON_IS_ARM64
 // __umulh requires intrin.h
 #include <intrin.h>
-#endif // _M_ARM64
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO && SIMDJSON_IS_ARM64
 
 namespace simdjson {
 namespace arm64 {
@@ -32,13 +32,13 @@ static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars
 simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
   internal::value128 answer;
 #if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
-#ifdef _M_ARM64
+#if SIMDJSON_IS_ARM64
   // ARM64 has native support for 64-bit multiplications, no need to emultate
   answer.high = __umulh(value1, value2);
   answer.low = value1 * value2;
 #else
   answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
-#endif // _M_ARM64
+#endif // SIMDJSON_IS_ARM64
 #else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
   __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
   answer.low = uint64_t(r);
@@ -51,6 +51,12 @@ simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t
 } // namespace arm64
 } // namespace simdjson
 
+#ifndef SIMDJSON_SWAR_NUMBER_PARSING
+#if SIMDJSON_IS_BIG_ENDIAN
+#define SIMDJSON_SWAR_NUMBER_PARSING 0
+#else
 #define SIMDJSON_SWAR_NUMBER_PARSING 1
+#endif
+#endif
 
 #endif // SIMDJSON_ARM64_NUMBERPARSING_DEFS_H

include/simdjson/arm64/simd.h

@@ -12,7 +12,7 @@ namespace arm64 {
 namespace {
 namespace simd {
 
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
 namespace {
 // Start of private section with Visual Studio workaround
 
@@ -121,7 +121,7 @@ namespace {
     // We return uint32_t instead of uint16_t because that seems to be more efficient for most
     // purposes (cutting it down to uint16_t costs performance in some compilers).
     simdjson_inline uint32_t to_bitmask() const {
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
       const uint8x16_t bit_mask =  simdjson_make_uint8x16_t(0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
                                                    0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80);
 #else
@@ -134,7 +134,13 @@ namespace {
       tmp = vpaddq_u8(tmp, tmp);
       return vgetq_lane_u16(vreinterpretq_u16_u8(tmp), 0);
     }
-    simdjson_inline bool any() const { return vmaxvq_u8(*this) != 0; }
+    // Returns 4-bit out of each byte, alternating between the high 4 bits and low
+    // bits result it is 64 bit.
+    simdjson_inline uint64_t to_bitmask64() const {
+      return vget_lane_u64(
+          vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(*this), 4)), 0);
+    }
+    simdjson_inline bool any() const { return vmaxvq_u32(vreinterpretq_u32_u8(*this)) != 0; }
   };
 
   // Unsigned bytes
@@ -152,7 +158,7 @@ namespace {
     // Splat constructor
     simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
     // Member-by-member initialization
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
     simdjson_inline simd8(
       uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
       uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
@@ -210,7 +216,7 @@ namespace {
 
     // Bit-specific operations
     simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return vtstq_u8(*this, bits); }
-    simdjson_inline bool any_bits_set_anywhere() const { return this->max_val() != 0; }
+    simdjson_inline bool any_bits_set_anywhere() const { return vmaxvq_u32(vreinterpretq_u32_u8(*this)) != 0; }
     simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return (*this & bits).any_bits_set_anywhere(); }
     template<int N>
     simdjson_inline simd8<uint8_t> shr() const { return vshrq_n_u8(*this, N); }
@@ -223,7 +229,12 @@ namespace {
       return lookup_table.apply_lookup_16_to(*this);
     }
 
-
+    // Returns 4-bit out of each byte, alternating between the high 4 bits and low
+    // bits result it is 64 bit.
+    simdjson_inline uint64_t to_bitmask64() const {
+      return vget_lane_u64(
+          vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(*this), 4)), 0);
+    }
     // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
     // Passing a 0 value for mask would be equivalent to writing out every byte to output.
     // Only the first 16 - count_ones(mask) bytes of the result are significant but 16 bytes
@@ -246,7 +257,7 @@ namespace {
       uint64x2_t shufmask64 = {thintable_epi8[mask1], thintable_epi8[mask2]};
       uint8x16_t shufmask = vreinterpretq_u8_u64(shufmask64);
       // we increment by 0x08 the second half of the mask
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
       uint8x16_t inc = simdjson_make_uint8x16_t(0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08);
 #else
       uint8x16_t inc = {0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08};
@@ -276,7 +287,7 @@ namespace {
       uint8x8_t compactmask1 = vcreate_u8(thintable_epi8[mask1]);
       uint8x8_t compactmask2 = vcreate_u8(thintable_epi8[mask2]);
       // we increment by 0x08 the second half of the mask
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
       uint8x8_t inc = simdjson_make_uint8x8_t(0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08);
 #else
       uint8x8_t inc = {0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08};
@@ -328,7 +339,7 @@ namespace {
     // Array constructor
     simdjson_inline simd8(const int8_t* values) : simd8(load(values)) {}
     // Member-by-member initialization
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
     simdjson_inline simd8(
       int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3, int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
       int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
@@ -449,7 +460,7 @@ namespace {
     }
 
     simdjson_inline uint64_t to_bitmask() const {
-#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
       const uint8x16_t bit_mask = simdjson_make_uint8x16_t(
         0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
         0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80

include/simdjson/arm64/stringparsing_defs.h

@@ -46,6 +46,32 @@ simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uin
   };
 }
 
+struct escaping {
+  static constexpr uint32_t BYTES_PROCESSED = 16;
+  simdjson_inline static escaping copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_escape() { return escape_bits != 0; }
+  simdjson_inline int escape_index() { return trailing_zeroes(escape_bits) / 4; }
+
+  uint64_t escape_bits;
+}; // struct escaping
+
+
+
+simdjson_inline escaping escaping::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "escaping finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v(src);
+  v.store(dst);
+  simd8<bool> is_quote = (v == '"');
+  simd8<bool> is_backslash = (v == '\\');
+  simd8<bool> is_control = (v < 32);
+  return {
+    (is_backslash | is_quote | is_control).to_bitmask64()
+  };
+}
+
+
+
 } // unnamed namespace
 } // namespace arm64
 } // namespace simdjson

include/simdjson/base.h

@@ -9,6 +9,7 @@
 #include "simdjson/compiler_check.h"
 #include "simdjson/error.h"
 #include "simdjson/portability.h"
+#include "simdjson/concepts.h"
 
 /**
  * @brief The top level simdjson namespace, containing everything the library provides.

include/simdjson/builtin.h

@@ -20,6 +20,10 @@
 #include "simdjson/ppc64.h"
 #elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(westmere)
 #include "simdjson/westmere.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lsx)
+#include "simdjson/lsx.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lasx)
+#include "simdjson/lasx.h"
 #else
 #error Unknown SIMDJSON_BUILTIN_IMPLEMENTATION
 #endif

include/simdjson/builtin/base.h

@@ -17,6 +17,10 @@ namespace simdjson {
   namespace ppc64 {}
 #elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(westmere)
   namespace westmere {}
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lsx)
+  namespace lsx {}
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lasx)
+  namespace lasx {}
 #else
 #error Unknown SIMDJSON_BUILTIN_IMPLEMENTATION
 #endif

include/simdjson/builtin/implementation.h

@@ -19,6 +19,10 @@
 #include "simdjson/ppc64/implementation.h"
 #elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(westmere)
 #include "simdjson/westmere/implementation.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lsx)
+#include "simdjson/lsx/implementation.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lasx)
+#include "simdjson/lasx/implementation.h"
 #else
 #error Unknown SIMDJSON_BUILTIN_IMPLEMENTATION
 #endif

include/simdjson/builtin/ondemand.h

@@ -20,6 +20,10 @@
 #include "simdjson/ppc64/ondemand.h"
 #elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(westmere)
 #include "simdjson/westmere/ondemand.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lsx)
+#include "simdjson/lsx/ondemand.h"
+#elif SIMDJSON_BUILTIN_IMPLEMENTATION_IS(lasx)
+#include "simdjson/lasx/ondemand.h"
 #else
 #error Unknown SIMDJSON_BUILTIN_IMPLEMENTATION
 #endif

include/simdjson/common_defs.h

@@ -50,6 +50,8 @@ double from_chars(const char *first, const char* end) noexcept;
 #define SIMDJSON_ISALIGNED_N(ptr, n) (((uintptr_t)(ptr) & ((n)-1)) == 0)
 
 #if SIMDJSON_REGULAR_VISUAL_STUDIO
+  // We could use [[deprecated]] but it requires C++14
+  #define simdjson_deprecated __declspec(deprecated)
 
   #define simdjson_really_inline __forceinline
   #define simdjson_never_inline __declspec(noinline)
@@ -88,6 +90,8 @@ double from_chars(const char *first, const char* end) noexcept;
   #define SIMDJSON_POP_DISABLE_UNUSED_WARNINGS
 
 #else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  // We could use [[deprecated]] but it requires C++14
+  #define simdjson_deprecated __attribute__((deprecated))
 
   #define simdjson_really_inline inline __attribute__((always_inline))
   #define simdjson_never_inline inline __attribute__((noinline))

include/simdjson/compiler_check.h

@@ -13,6 +13,16 @@
 #endif
 #endif
 
+// C++ 23
+#if !defined(SIMDJSON_CPLUSPLUS23) && (SIMDJSON_CPLUSPLUS >= 202302L)
+#define SIMDJSON_CPLUSPLUS23 1
+#endif
+
+// C++ 20
+#if !defined(SIMDJSON_CPLUSPLUS20) && (SIMDJSON_CPLUSPLUS >= 202002L)
+#define SIMDJSON_CPLUSPLUS20 1
+#endif
+
 // C++ 17
 #if !defined(SIMDJSON_CPLUSPLUS17) && (SIMDJSON_CPLUSPLUS >= 201703L)
 #define SIMDJSON_CPLUSPLUS17 1
@@ -40,4 +50,55 @@
 #endif
 #endif
 
+#ifndef SIMDJSON_CONSTEXPR_LAMBDA
+#if SIMDJSON_CPLUSPLUS17
+#define SIMDJSON_CONSTEXPR_LAMBDA constexpr
+#else
+#define SIMDJSON_CONSTEXPR_LAMBDA
+#endif
+#endif
+
+
+
+#ifdef __has_include
+#if __has_include(<version>)
+#include <version>
+#endif
+#endif
+
+// The current specification is unclear on how we detect
+// static reflection, both __cpp_lib_reflection and
+// __cpp_impl_reflection are proposed in the draft specification.
+// For now, we disable static reflect by default. It must be
+// specified at compiler time.
+#ifndef SIMDJSON_STATIC_REFLECTION
+#define SIMDJSON_STATIC_REFLECTION 0 // disabled by default.
+#endif
+
+#if defined(__apple_build_version__)
+#if __apple_build_version__ < 14000000
+#define SIMDJSON_CONCEPT_DISABLED 1 // apple-clang/13 doesn't support std::convertible_to
+#endif
+#endif
+
+
+#if defined(__cpp_concepts) && !defined(SIMDJSON_CONCEPT_DISABLED)
+#if __cpp_concepts >= 201907L
+#include <utility>
+#define SIMDJSON_SUPPORTS_DESERIALIZATION 1
+#else
+#define SIMDJSON_SUPPORTS_DESERIALIZATION 0
+#endif
+#else // defined(__cpp_concepts) && !defined(SIMDJSON_CONCEPT_DISABLED)
+#define SIMDJSON_SUPPORTS_DESERIALIZATION 0
+#endif // defined(__cpp_concepts) && !defined(SIMDJSON_CONCEPT_DISABLED)
+
+#if !defined(SIMDJSON_CONSTEVAL)
+#if defined(__cpp_consteval) && __cpp_consteval >= 201811L
+#define SIMDJSON_CONSTEVAL 1
+#else
+#define SIMDJSON_CONSTEVAL 0
+#endif // defined(__cpp_consteval) && __cpp_consteval >= 201811L
+#endif // !defined(SIMDJSON_CONSTEVAL)
+
 #endif // SIMDJSON_COMPILER_CHECK_H

include/simdjson/concepts.h

@@ -0,0 +1,114 @@
+#ifndef SIMDJSON_CONCEPTS_H
+#define SIMDJSON_CONCEPTS_H
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+
+#include <concepts>
+#include <type_traits>
+
+namespace simdjson {
+namespace concepts {
+
+namespace details {
+#define SIMDJSON_IMPL_CONCEPT(name, method)                                    \
+  template <typename T>                                                        \
+  concept supports_##name = !std::is_const_v<T> && requires {                  \
+    typename std::remove_cvref_t<T>::value_type;                               \
+    requires requires(typename std::remove_cvref_t<T>::value_type &&val,       \
+                      T obj) {                                                 \
+      obj.method(std::move(val));                                              \
+      requires !requires { obj = std::move(val); };                            \
+    };                                                                         \
+  };
+
+SIMDJSON_IMPL_CONCEPT(emplace_back, emplace_back)
+SIMDJSON_IMPL_CONCEPT(emplace, emplace)
+SIMDJSON_IMPL_CONCEPT(push_back, push_back)
+SIMDJSON_IMPL_CONCEPT(add, add)
+SIMDJSON_IMPL_CONCEPT(push, push)
+SIMDJSON_IMPL_CONCEPT(append, append)
+SIMDJSON_IMPL_CONCEPT(insert, insert)
+SIMDJSON_IMPL_CONCEPT(op_append, operator+=)
+
+#undef SIMDJSON_IMPL_CONCEPT
+} // namespace details
+
+/// Check if T is a container that we can append to, including:
+///   std::vector, std::deque, std::list, std::string, ...
+template <typename T>
+concept appendable_containers =
+    details::supports_emplace_back<T> || details::supports_emplace<T> ||
+    details::supports_push_back<T> || details::supports_push<T> ||
+    details::supports_add<T> || details::supports_append<T> ||
+    details::supports_insert<T>;
+
+/// Insert into the container however possible
+template <appendable_containers T, typename... Args>
+constexpr decltype(auto) emplace_one(T &vec, Args &&...args) {
+  if constexpr (details::supports_emplace_back<T>) {
+    return vec.emplace_back(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_emplace<T>) {
+    return vec.emplace(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_push_back<T>) {
+    return vec.push_back(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_push<T>) {
+    return vec.push(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_add<T>) {
+    return vec.add(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_append<T>) {
+    return vec.append(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_insert<T>) {
+    return vec.insert(std::forward<Args>(args)...);
+  } else if constexpr (details::supports_op_append<T> && sizeof...(Args) == 1) {
+    return vec.operator+=(std::forward<Args>(args)...);
+  } else {
+    static_assert(!sizeof(T *),
+                  "We don't know how to add things to this container");
+  }
+}
+
+/// This checks if the container will return a reference to the newly added
+/// element after an insert which for example `std::vector::emplace_back` does
+/// since C++17; this will allow some optimizations.
+template <typename T>
+concept returns_reference = appendable_containers<T> && requires {
+  typename std::remove_cvref_t<T>::reference;
+  requires requires(typename std::remove_cvref_t<T>::value_type &&val, T obj) {
+    {
+      emplace_one(obj, std::move(val))
+    } -> std::same_as<typename std::remove_cvref_t<T>::reference>;
+  };
+};
+
+template <typename T>
+concept smart_pointer = requires(std::remove_cvref_t<T> ptr) {
+  // Check if T has a member type named element_type
+  typename std::remove_cvref_t<T>::element_type;
+
+  // Check if T has a get() member function
+  {
+    ptr.get()
+  } -> std::same_as<typename std::remove_cvref_t<T>::element_type *>;
+
+  // Check if T can be dereferenced
+  { *ptr } -> std::same_as<typename std::remove_cvref_t<T>::element_type &>;
+};
+
+template <typename T>
+concept optional_type = requires(std::remove_cvref_t<T> obj) {
+  typename std::remove_cvref_t<T>::value_type;
+  { obj.value() } -> std::same_as<typename std::remove_cvref_t<T>::value_type&>;
+  requires requires(typename std::remove_cvref_t<T>::value_type &&val) {
+    obj.emplace(std::move(val));
+    obj = std::move(val);
+    {
+      obj.value_or(val)
+    } -> std::convertible_to<typename std::remove_cvref_t<T>::value_type>;
+  };
+  { static_cast<bool>(obj) } -> std::same_as<bool>; // convertible to bool
+};
+
+
+} // namespace concepts
+} // namespace simdjson
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+#endif // SIMDJSON_CONCEPTS_H

include/simdjson/dom/array-inl.h

@@ -7,6 +7,7 @@
 #include "simdjson/dom/array.h"
 #include "simdjson/dom/element.h"
 #include "simdjson/error-inl.h"
+#include "simdjson/jsonpathutil.h"
 #include "simdjson/internal/tape_ref-inl.h"
 
 #include <limits>
@@ -44,6 +45,13 @@ inline simdjson_result<dom::element> simdjson_result<dom::array>::at_pointer(std
   if (error()) { return error(); }
   return first.at_pointer(json_pointer);
 }
+
+ inline simdjson_result<dom::element> simdjson_result<dom::array>::at_path(std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+ }
+
 inline simdjson_result<dom::element> simdjson_result<dom::array>::at(size_t index) const noexcept {
   if (error()) { return error(); }
   return first.at(index);
@@ -113,6 +121,12 @@ inline simdjson_result<element> array::at_pointer(std::string_view json_pointer)
   return child;
 }
 
+inline simdjson_result<element> array::at_path(std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+}
+
 inline simdjson_result<element> array::at(size_t index) const noexcept {
   SIMDJSON_DEVELOPMENT_ASSERT(tape.usable()); // https://github.com/simdjson/simdjson/issues/1914
   size_t i=0;
@@ -123,6 +137,10 @@ inline simdjson_result<element> array::at(size_t index) const noexcept {
   return INDEX_OUT_OF_BOUNDS;
 }
 
+inline array::operator element() const noexcept {
+	return element(tape);
+}
+
 //
 // array::iterator inline implementation
 //

include/simdjson/dom/array.h

@@ -108,6 +108,21 @@ public:
    */
   inline simdjson_result<element> at_pointer(std::string_view json_pointer) const noexcept;
 
+  /**
+   * Get the value associated with the given JSONPath expression. We only support
+   * JSONPath queries that trivially convertible to JSON Pointer queries: key
+   * names and array indices.
+   *
+   * https://datatracker.ietf.org/doc/html/draft-normington-jsonpath-00
+   *
+   * @return The value associated with the given JSONPath expression, or:
+   *         - INVALID_JSON_POINTER if the JSONPath to JSON Pointer conversion fails
+   *         - NO_SUCH_FIELD if a field does not exist in an object
+   *         - INDEX_OUT_OF_BOUNDS if an array index is larger than an array length
+   *         - INCORRECT_TYPE if a non-integer is used to access an array
+  */
+  inline simdjson_result<element> at_path(std::string_view json_path) const noexcept;
+
   /**
    * Get the value at the given index. This function has linear-time complexity and
    * is equivalent to the following:
@@ -126,6 +141,11 @@ public:
    */
   inline simdjson_result<element> at(size_t index) const noexcept;
 
+  /**
+   * Implicitly convert object to element
+   */
+  inline operator element() const noexcept;
+
 private:
   simdjson_inline array(const internal::tape_ref &tape) noexcept;
   internal::tape_ref tape;
@@ -147,6 +167,7 @@ public:
   simdjson_inline simdjson_result(error_code error) noexcept; ///< @private
 
   inline simdjson_result<dom::element> at_pointer(std::string_view json_pointer) const noexcept;
+  inline simdjson_result<dom::element> at_path(std::string_view json_path) const noexcept;
   inline simdjson_result<dom::element> at(size_t index) const noexcept;
 
 #if SIMDJSON_EXCEPTIONS

include/simdjson/dom/document_stream-inl.h

@@ -223,7 +223,11 @@ simdjson_inline std::string_view document_stream::iterator::source() const noexc
     return std::string_view(start, next_doc_index - current_index() + 1);
   } else {
     size_t next_doc_index = stream->batch_start + stream->parser->implementation->structural_indexes[stream->parser->implementation->next_structural_index];
-    return std::string_view(reinterpret_cast<const char*>(stream->buf) + current_index(), next_doc_index - current_index() - 1);
+    size_t svlen = next_doc_index - current_index();
+    while(svlen > 1 && (std::isspace(start[svlen-1]) || start[svlen-1] == '\0')) {
+      svlen--;
+    }
+    return std::string_view(start, svlen);
   }
 }
 

include/simdjson/dom/element-inl.h

@@ -9,6 +9,7 @@
 
 #include "simdjson/dom/object-inl.h"
 #include "simdjson/error-inl.h"
+#include "simdjson/jsonpathutil.h"
 
 #include <ostream>
 #include <limits>
@@ -122,6 +123,11 @@ simdjson_inline simdjson_result<dom::element> simdjson_result<dom::element>::at_
   if (error()) { return error(); }
   return first.at_pointer(json_pointer);
 }
+simdjson_inline simdjson_result<dom::element> simdjson_result<dom::element>::at_path(const std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+}
 #ifndef SIMDJSON_DISABLE_DEPRECATED_API
 [[deprecated("For standard compliance, use at_pointer instead, and prefix your pointers with a slash '/', see RFC6901 ")]]
 simdjson_inline simdjson_result<dom::element> simdjson_result<dom::element>::at(const std::string_view json_pointer) const noexcept {
@@ -375,6 +381,23 @@ inline simdjson_result<element> element::operator[](const char *key) const noexc
   return at_key(key);
 }
 
+inline bool is_pointer_well_formed(std::string_view json_pointer) noexcept {
+  if (simdjson_unlikely(json_pointer[0] != '/')) {
+    return false;
+  }
+  size_t escape = json_pointer.find('~');
+  if (escape == std::string_view::npos) {
+    return true;
+  }
+  if (escape == json_pointer.size() - 1) {
+    return false;
+  }
+  if (json_pointer[escape + 1] != '0' && json_pointer[escape + 1] != '1') {
+    return false;
+  }
+  return true;
+}
+
 inline simdjson_result<element> element::at_pointer(std::string_view json_pointer) const noexcept {
   SIMDJSON_DEVELOPMENT_ASSERT(tape.usable()); // https://github.com/simdjson/simdjson/issues/1914
   switch (tape.tape_ref_type()) {
@@ -383,7 +406,10 @@ inline simdjson_result<element> element::at_pointer(std::string_view json_pointe
     case internal::tape_type::START_ARRAY:
       return array(tape).at_pointer(json_pointer);
     default: {
-      if(!json_pointer.empty()) { // a non-empty string is invalid on an atom
+      if (!json_pointer.empty()) { // a non-empty string can be invalid, or accessing a primitive (issue 2154)
+        if (is_pointer_well_formed(json_pointer)) {
+          return NO_SUCH_FIELD;
+        }
         return INVALID_JSON_POINTER;
       }
       // an empty string means that we return the current node
@@ -392,6 +418,11 @@ inline simdjson_result<element> element::at_pointer(std::string_view json_pointe
     }
   }
 }
+inline simdjson_result<element> element::at_path(std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+}
 #ifndef SIMDJSON_DISABLE_DEPRECATED_API
 [[deprecated("For standard compliance, use at_pointer instead, and prefix your pointers with a slash '/', see RFC6901 ")]]
 inline simdjson_result<element> element::at(std::string_view json_pointer) const noexcept {

include/simdjson/dom/element.h

@@ -372,6 +372,8 @@ public:
    *         - INCORRECT_TYPE if this is not an object
    */
   inline simdjson_result<element> operator[](const char *key) const noexcept;
+  simdjson_result<element> operator[](int) const noexcept = delete;
+
 
   /**
    * Get the value associated with the given JSON pointer.  We use the RFC 6901
@@ -397,6 +399,21 @@ public:
    */
   inline simdjson_result<element> at_pointer(const std::string_view json_pointer) const noexcept;
 
+  /**
+   * Get the value associated with the given JSONPath expression. We only support
+   * JSONPath queries that trivially convertible to JSON Pointer queries: key
+   * names and array indices.
+   *
+   * https://datatracker.ietf.org/doc/html/draft-normington-jsonpath-00
+   *
+   * @return The value associated with the given JSONPath expression, or:
+   *         - INVALID_JSON_POINTER if the JSONPath to JSON Pointer conversion fails
+   *         - NO_SUCH_FIELD if a field does not exist in an object
+   *         - INDEX_OUT_OF_BOUNDS if an array index is larger than an array length
+   *         - INCORRECT_TYPE if a non-integer is used to access an array
+  */
+  inline simdjson_result<element> at_path(std::string_view json_path) const noexcept;
+
 #ifndef SIMDJSON_DISABLE_DEPRECATED_API
   /**
    *
@@ -525,7 +542,9 @@ public:
 
   simdjson_inline simdjson_result<dom::element> operator[](std::string_view key) const noexcept;
   simdjson_inline simdjson_result<dom::element> operator[](const char *key) const noexcept;
+  simdjson_result<dom::element> operator[](int) const noexcept = delete;
   simdjson_inline simdjson_result<dom::element> at_pointer(const std::string_view json_pointer) const noexcept;
+  simdjson_inline simdjson_result<dom::element> at_path(const std::string_view json_path) const noexcept;
   [[deprecated("For standard compliance, use at_pointer instead, and prefix your pointers with a slash '/', see RFC6901 ")]]
   simdjson_inline simdjson_result<dom::element> at(const std::string_view json_pointer) const noexcept;
   simdjson_inline simdjson_result<dom::element> at(size_t index) const noexcept;

include/simdjson/dom/object-inl.h

@@ -7,6 +7,7 @@
 
 #include "simdjson/dom/element-inl.h"
 #include "simdjson/error-inl.h"
+#include "simdjson/jsonpathutil.h"
 
 #include <cstring>
 
@@ -34,6 +35,11 @@ inline simdjson_result<dom::element> simdjson_result<dom::object>::at_pointer(st
   if (error()) { return error(); }
   return first.at_pointer(json_pointer);
 }
+inline simdjson_result<dom::element> simdjson_result<dom::object>::at_path(std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+}
 inline simdjson_result<dom::element> simdjson_result<dom::object>::at_key(std::string_view key) const noexcept {
   if (error()) { return error(); }
   return first.at_key(key);
@@ -131,6 +137,12 @@ inline simdjson_result<element> object::at_pointer(std::string_view json_pointer
   return child;
 }
 
+inline simdjson_result<element> object::at_path(std::string_view json_path) const noexcept {
+  auto json_pointer = json_path_to_pointer_conversion(json_path);
+  if (json_pointer == "-1") { return INVALID_JSON_POINTER; }
+  return at_pointer(json_pointer);
+}
+
 inline simdjson_result<element> object::at_key(std::string_view key) const noexcept {
   iterator end_field = end();
   for (iterator field = begin(); field != end_field; ++field) {
@@ -153,6 +165,10 @@ inline simdjson_result<element> object::at_key_case_insensitive(std::string_view
   return NO_SUCH_FIELD;
 }
 
+inline object::operator element() const noexcept {
+	return element(tape);
+}
+
 //
 // object::iterator inline implementation
 //

include/simdjson/dom/object.h

@@ -145,6 +145,7 @@ public:
    *         - INCORRECT_TYPE if this is not an object
    */
   inline simdjson_result<element> operator[](const char *key) const noexcept;
+  simdjson_result<element> operator[](int) const noexcept = delete;
 
   /**
    * Get the value associated with the given JSON pointer. We use the RFC 6901
@@ -171,6 +172,21 @@ public:
    */
   inline simdjson_result<element> at_pointer(std::string_view json_pointer) const noexcept;
 
+  /**
+   * Get the value associated with the given JSONPath expression. We only support
+   * JSONPath queries that trivially convertible to JSON Pointer queries: key
+   * names and array indices.
+   *
+   * https://datatracker.ietf.org/doc/html/draft-normington-jsonpath-00
+   *
+   * @return The value associated with the given JSONPath expression, or:
+   *         - INVALID_JSON_POINTER if the JSONPath to JSON Pointer conversion fails
+   *         - NO_SUCH_FIELD if a field does not exist in an object
+   *         - INDEX_OUT_OF_BOUNDS if an array index is larger than an array length
+   *         - INCORRECT_TYPE if a non-integer is used to access an array
+  */
+  inline simdjson_result<element> at_path(std::string_view json_path) const noexcept;
+
   /**
    * Get the value associated with the given key.
    *
@@ -200,6 +216,11 @@ public:
    */
   inline simdjson_result<element> at_key_case_insensitive(std::string_view key) const noexcept;
 
+  /**
+   * Implicitly convert object to element
+   */
+  inline operator element() const noexcept;
+
 private:
   simdjson_inline object(const internal::tape_ref &tape) noexcept;
 
@@ -238,7 +259,9 @@ public:
 
   inline simdjson_result<dom::element> operator[](std::string_view key) const noexcept;
   inline simdjson_result<dom::element> operator[](const char *key) const noexcept;
+  simdjson_result<dom::element> operator[](int) const noexcept = delete;
   inline simdjson_result<dom::element> at_pointer(std::string_view json_pointer) const noexcept;
+  inline simdjson_result<dom::element> at_path(std::string_view json_path) const noexcept;
   inline simdjson_result<dom::element> at_key(std::string_view key) const noexcept;
   inline simdjson_result<dom::element> at_key_case_insensitive(std::string_view key) const noexcept;
 

include/simdjson/dom/parser-inl.h

@@ -191,7 +191,7 @@ simdjson_inline size_t parser::capacity() const noexcept {
 simdjson_inline size_t parser::max_capacity() const noexcept {
   return _max_capacity;
 }
-simdjson_inline size_t parser::max_depth() const noexcept {
+simdjson_pure simdjson_inline size_t parser::max_depth() const noexcept {
   return implementation ? implementation->max_depth() : DEFAULT_MAX_DEPTH;
 }
 

include/simdjson/dom/parser.h

@@ -83,6 +83,15 @@ public:
    * If the parser's current capacity is less than the file length, it will allocate enough capacity
    * to handle it (up to max_capacity).
    *
+   * ## Windows and Unicode
+   *
+   * Windows users who need to read files with non-ANSI characters in the
+   * name should set their code page to UTF-8 (65001) before calling this
+   * function. This should be the default with Windows 11 and better.
+   * Further, they may use the AreFileApisANSI function to determine whether
+   * the filename is interpreted using the ANSI or the system default OEM
+   * codepage, and they may call SetFileApisToOEM accordingly.
+   *
    * @param path The path to load.
    * @return The document, or an error:
    *         - IO_ERROR if there was an error opening or reading the file.
@@ -193,6 +202,22 @@ public:
    *   simdjson::dom::parser parser;
    *   simdjson::dom::element element = parser.parse(padded_json_copy.get(), json_len, false);
    *
+   * ### std::string references
+   *
+   * If you pass a mutable std::string reference (std::string&), the parser will seek to extend
+   * its capacity to SIMDJSON_PADDING bytes beyond the end of the string.
+   *
+   * Whenever you pass an std::string reference, the parser will access the bytes beyond the end of
+   * the string but before the end of the allocated memory (std::string::capacity()).
+   * If you are using a sanitizer that checks for reading uninitialized bytes or std::string's
+   * container-overflow checks, you may encounter sanitizer warnings.
+   * You can safely ignore these warnings. Or you can call simdjson::pad(std::string&) to pad the
+   * string with SIMDJSON_PADDING spaces: this function returns a simdjson::padding_string_view
+   * which can be be passed to the parser's parse function:
+   *
+   *    std::string json = R"({ "foo": 1 } { "foo": 2 } { "foo": 3 } )";
+   *    element doc = parser.parse(simdjson::pad(json));
+   *
    * ### Parser Capacity
    *
    * If the parser's current capacity is less than len, it will allocate enough capacity
@@ -518,7 +543,7 @@ public:
    *
    * @return Maximum depth, in bytes.
    */
-  simdjson_inline size_t max_depth() const noexcept;
+  simdjson_pure simdjson_inline size_t max_depth() const noexcept;
 
   /**
    * Set max_capacity. This is the largest document this parser can automatically support.
@@ -540,9 +565,14 @@ public:
   /**
    * The parser instance can use threads when they are available to speed up some
    * operations. It is enabled by default. Changing this attribute will change the
-   * behavior of the parser for future operations.
+   * behavior of the parser for future operations. Set to true by default.
    */
   bool threaded{true};
+#else
+  /**
+   * When SIMDJSON_THREADS_ENABLED is not defined, the parser instance cannot use threads.
+   */
+  bool threaded{false};
 #endif
   /** @private Use the new DOM API instead */
   class Iterator;