forgot these lifebounds

* Initial work on JSON builder

* moving the files back to ondemand for now.

* tweak

* more later

* update

* minor edits

* dropping vs arm (missing support)

* adding tests. we still specialized write_string_escaped

* tweaking

* fix typo

* tweaking the approach

* minor fix

* missing store

* another missing store

* Attempt at fixing failing serialization tests. (#2292)

* Fixing appeand_float typo (#2294)

* applying a couple of fixes

* updating single header

* fix for pre C++17 if constexpr

* Fixing unused argument problem and updating the singleheader file

* various pedantic fixes

* Sketch of builder

* reordering.

* simplify

* Adding draft of static reflection based deserialization

* Updating simdjson singleheader

* patching the automated deserialization.

* automated

* Adding support for smart pointers of user defined types.

* Adding specialization for smart pointers for basic types. I think it is highly likely that this can be done in a more generic way.

* Referncing a later version of rapidjson that fixed the issue related with assignment attempt of a const variable for GenericStringRef class.

* guarding the tests

* adding documentation for string_builder

* saving

* rename to 'append'

* saving

* non-functional benchmarks (#2342)

* non-functional benchmarks

* Fix typo

* various fixes

* tweaking

---------

Co-authored-by: Daniel Lemire <dlemire@lemire.me>
Co-authored-by: Francisco Geiman Thiesen <franciscogthiesen@gmail.com>

* tuning

* various minor fixes

* minor tweak

* minor simplification

* updating amal

* adding a cast

* update

* fancy casting

* removing dead code

* Pushing latest changes. CITM benchmark is still not working.

* Still not working, but now I am getting only 10 errors.

* add static reflection benchmark to 'large random' benchmark and allows (#2349)

deserialization (with static reflection) from objects and arrays.

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

* Removing std::map from CitmCatalog definition, since that is not currently supported.

* Added free to rust bench, segfault is still happening..

* The syntax changed: ^E became ^^E. (#2350)

* The syntax changed: ^E became ^^E.

* guarding

---------

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

* Adding support for string_view_keyed_map types.

* Adding concepts as a conditional include.

* updating single-header

* Adding concepts to ondemand deps

* rust benchmark is finally working

* Fixing small typo in docs.

* adding docker config and instructions so that our users can test the static reflection (#2358)

* adding docker config and instructions so that our users can test the
static reflection

* completing the instructions

* pruning white spaces

---------

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

* minor optimizations on the JSON builder branch

* avoiding undef behaviour

* saving

* somewhat nicer builder

* make it possible to run just one benchmark

* adding linux perf

* fixing minor issue

* updating swar

* Adding real world compilation benchmark (#2379)

* Adding compilation benchmark for json parsing with and without reflection

* Moving it to the benchmark folder, also reducing a bit the number of iterations.

* Removing script from root folder.

* Reducing number of iterations

* Update benchmark/benchmark_reflection_usage_compilation.sh

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

* Update benchmark/benchmark_reflection_usage_compilation.sh

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

* Update benchmark/benchmark_reflection_usage_compilation.sh

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

* Making the script more customizable and also test whether the compiler being used supports reflection before actually running the benchmark

---------

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

* Using define_static_string from  https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2025/p3491r2.html (#2389)

* Applying changes needed after latest reflection paper updates.

* Working, but no template for yet.

* Updating single-header to incldue the use of define_static_string.

* copying over master

---------

Co-authored-by: Daniel Lemire <dlemire@lemire.me>
Co-authored-by: Francisco Geiman Thiesen <franciscogthiesen@gmail.com>

.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/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/emscripten.yml

@@ -0,0 +1,17 @@
+on: [push, pull_request]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+      - uses: mymindstorm/setup-emsdk@6ab9eb1bda2574c4ddb79809fc9247783eaf9021 # v14
+      - name: Verify
+        run: emcc -v
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v3.6.0
+      - name: Configure
+        run: emcmake cmake -B build
+      - name: Build # We build but do not test
+        run: cmake --build build

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

@@ -4,7 +4,7 @@ on: [push, pull_request]
 
 jobs:
   whitespace:
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - name: Remove whitespace and check the diff
@@ -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
 
@@ -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/msys2-clang.yml

@@ -20,7 +20,7 @@ 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"
+          - 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:

.github/workflows/msys2.yml

@@ -22,7 +22,7 @@ jobs:
           - msystem: "MINGW64"
             install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
             type: Debug
-            - msystem: "MINGW64"
+          - msystem: "MINGW64"
             install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
             type: RelWithDebInfo
     env:

.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-gcc8.yml

@@ -1,38 +0,0 @@
-name: Ubuntu 20.04 CI (GCC 8)
-
-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-20.04
-    env:
-      CXX: g++-8
-      CC: gcc-8
-    steps:
-      - uses: actions/checkout@v4
-      - uses: actions/cache@v4
-        with:
-          path: dependencies/.cache
-          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-      - name: Install GCC 8
-        run: sudo apt-get install -y g++-8
-      - name: Use cmake
-        run: |
-          mkdir builddebug &&
-          cd builddebug &&
-          cmake -DCMAKE_BUILD_TYPE=Debug -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF ..  &&
-          cmake --build .   &&
-          ctest --output-on-failure -LE explicitonly -j   &&
-          cd .. &&
-          mkdir build &&
-          cd build &&
-          cmake  -DSIMDJSON_GOOGLE_BENCHMARKS=ON -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_INSTALL_PREFIX:PATH=destination ..  &&
-          cmake --build .   &&
-          ctest --output-on-failure -LE explicitonly -j   &&
-          cmake --install . &&
-          echo -e '#include <simdjson.h>\nint main(int argc,char**argv) {simdjson::dom::parser parser;simdjson::dom::element tweets = parser.load(argv[1]); }' > tmp.cpp && c++ -Idestination/include -Ldestination/lib -std=c++17 -Wl,-rpath,destination/lib -o linkandrun tmp.cpp -lsimdjson && ./linkandrun jsonexamples/twitter.json &&
-          cd ../tests/installation_tests/find &&
-          mkdir build && cd build && cmake -DCMAKE_INSTALL_PREFIX:PATH=../../../build/destination .. &&  cmake --build .

.github/workflows/ubuntu20.yml

@@ -1,47 +0,0 @@
-name: Ubuntu 20.04 CI (GCC 9)
-
-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-20.04
-    steps:
-      - uses: actions/checkout@v4
-      - uses: actions/cache@v4
-        with:
-          path: dependencies/.cache
-          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-      - name: Use cmake to build just the library
-        run: |
-          mkdir buildjustlib &&
-          cd buildjustlib &&
-          cmake -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_DEVELOPER_MODE=OFF -DCMAKE_INSTALL_PREFIX:PATH=destination ..  &&
-          cmake --build .   &&
-          cmake --install . &&
-          echo -e '#include <simdjson.h>\nint main(int argc,char**argv) {simdjson::dom::parser parser;simdjson::dom::element tweets = parser.load(argv[1]); }' > tmp.cpp &&
-          c++ -Idestination/include -Ldestination/lib -std=c++17 -Wl,-rpath,destination/lib -o linkandrun tmp.cpp -lsimdjson &&
-          cd ../tests/installation_tests/find &&
-          mkdir buildjustlib &&
-          cd buildjustlib &&
-          cmake -DCMAKE_INSTALL_PREFIX:PATH=../../../buildjustlib/destination .. &&
-          cmake --build .
-      - name: Use cmake
-        run: |
-          mkdir builddebug &&
-          cd builddebug &&
-          cmake -DCMAKE_BUILD_TYPE=Debug -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF ..  &&
-          cmake --build .   &&
-          ctest --output-on-failure -LE explicitonly -j   &&
-          cd .. &&
-          mkdir build &&
-          cd build &&
-          cmake  -DSIMDJSON_GOOGLE_BENCHMARKS=ON -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_INSTALL_PREFIX:PATH=destination ..  &&
-          cmake --build .   &&
-          ctest --output-on-failure -LE explicitonly -j   &&
-          cmake --install . &&
-          echo -e '#include <simdjson.h>\nint main(int argc,char**argv) {simdjson::dom::parser parser;simdjson::dom::element tweets = parser.load(argv[1]); }' > tmp.cpp && c++ -Idestination/include -Ldestination/lib -std=c++17 -Wl,-rpath,destination/lib -o linkandrun tmp.cpp -lsimdjson && ./linkandrun jsonexamples/twitter.json &&
-          cd ../tests/installation_tests/find &&
-          mkdir build && cd build && cmake -DCMAKE_INSTALL_PREFIX:PATH=../../../build/destination .. &&  cmake --build .

.github/workflows/ubuntu24-checkperf.yml

@@ -13,7 +13,7 @@ jobs:
     if: >-
       ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
       ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - uses: actions/cache@v4

.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-noexcept.yml

@@ -7,7 +7,7 @@ jobs:
     if: >-
       ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
       ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - uses: actions/cache@v4

.github/workflows/ubuntu24-nothread.yml

@@ -7,7 +7,7 @@ jobs:
     if: >-
       ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
       ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - uses: actions/cache@v4

.github/workflows/ubuntu24-sani.yml

@@ -7,7 +7,7 @@ jobs:
     if: >-
       ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
       ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - uses: actions/cache@v4
@@ -25,7 +25,7 @@ jobs:
     if: >-
       ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
       ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@v4
       - uses: actions/cache@v4

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

@@ -10,7 +10,6 @@ jobs:
        fail-fast: false
        matrix:
          include:
-           - {arch: ARM}
            - {arch: ARM64}
            - {arch: ARM64EC}
      steps:
@@ -19,4 +18,4 @@ jobs:
        - name: Use cmake
          run: |
            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
+           cmake --build build --verbose

.github/workflows/vs17-ci.yml

@@ -26,15 +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 ${{build_type}} --verbose
+      run: cmake --build build --config ${{matrix.build_type}} --verbose
     - name: Run tests
       run: |
         cd build
-        ctest -C ${{build_type}}  -LE explicitonly  --output-on-failure
+        ctest -C ${{matrix.build_type}}  -LE explicitonly  --output-on-failure
     - name: Install
       run: |
-        cmake --install build --config ${{build_type}}
+        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 ${{build_type}}
+        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}}

.vscode/settings.json

@@ -109,11 +109,23 @@
         "numbers": "cpp",
         "semaphore": "cpp",
         "stop_token": "cpp",
-        "cfenv": "cpp"
+        "cfenv": "cpp",
-    },
+        "format": "cpp",
-    "cmake.configureSettings": {
+        "xlocmes": "cpp",
-        "CMAKE_EXPORT_COMPILE_COMMANDS": "YES",
+        "xlocmon": "cpp",
-        "SIMDJSON_DEVELOPER_MODE": "ON",
+        "xlocnum": "cpp",
-        "SIMDJSON_SINGLEHEADER": "OFF"
+        "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

@@ -1,9 +1,11 @@
 cmake_minimum_required(VERSION 3.14)
+cmake_policy(VERSION 3.5) # For doctest
+
 
 project(
     simdjson
     # The version number is modified by tools/release.py
-    VERSION 3.10.1
+    VERSION 3.13.0
     DESCRIPTION "Parsing gigabytes of JSON per second"
     HOMEPAGE_URL "https://simdjson.org/"
     LANGUAGES CXX C
@@ -20,8 +22,8 @@ string(
 # ---- Options, variables ----
 
 # These version numbers are modified by tools/release.py
-set(SIMDJSON_LIB_VERSION "23.0.0" CACHE STRING "simdjson library version")
+set(SIMDJSON_LIB_VERSION "26.0.0" CACHE STRING "simdjson library version")
-set(SIMDJSON_LIB_SOVERSION "23" CACHE STRING "simdjson library soversion")
+set(SIMDJSON_LIB_SOVERSION "26" CACHE STRING "simdjson library soversion")
 
 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)
@@ -43,6 +45,20 @@ 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)
@@ -84,7 +100,7 @@ set_target_properties(
 )
 
 # FIXME: Use proper CMake integration for exports
-if(MSVC AND BUILD_SHARED_LIBS)
+if(WIN32 AND BUILD_SHARED_LIBS)
   target_compile_definitions(
       simdjson
       PRIVATE SIMDJSON_BUILDING_WINDOWS_DYNAMIC_LIBRARY=1
@@ -98,7 +114,16 @@ 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(
@@ -111,6 +136,12 @@ if(
   )
 endif()
 
+option(SIMDJSON_MINUS_ZERO_AS_FLOAT "Treat -0 as a floating-point value" OFF)
+
+if(SIMDJSON_MINUS_ZERO_AS_FLOAT)
+  simdjson_add_props(target_compile_definitions PRIVATE SIMDJSON_MINUS_ZERO_AS_FLOAT=1)
+endif(SIMDJSON_MINUS_ZERO_AS_FLOAT)
+
 if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(loongarch64)$")
     option(SIMDJSON_PREFER_LSX "Prefer LoongArch SX" ON)
     include(CheckCXXCompilerFlag)
@@ -287,6 +318,7 @@ 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
 

CONTRIBUTING.md

@@ -92,7 +92,7 @@ We welcome contributions from women and less represented groups. If you need hel
 
 Consider the following points when engaging with the project:
 
-- We discourage arguments from authority: ideas are discusssed on their own merits and not based on who stated it.
+- We discourage arguments from authority: ideas are discussed on their own merits and not based on who stated it.
 - Be mindful that what you may view as an aggression is maybe merely a difference of opinion or a misunderstanding.
 - Be mindful that a collection of small aggressions, even if mild in isolation, can become harmful.
 

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         = "3.10.1"
+PROJECT_NUMBER         = "3.13.0"
 
 # 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

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)
@@ -61,6 +62,9 @@ Real-world usage
 - [ada-url](https://github.com/ada-url/ada)
 - [fastgron](https://github.com/adamritter/fastgron)
 - [WasmEdge](https://wasmedge.org)
+- [RonDB](https://github.com/logicalclocks/rondb)
+- [GreptimeDB](https://github.com/GreptimeTeam/greptimedb)
+
 
 If you are planning to use simdjson in a product, please work from one of our releases.
 
@@ -105,6 +109,7 @@ 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.
@@ -143,6 +148,9 @@ 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
@@ -167,7 +175,8 @@ We distinguish between "bindings" (which just wrap the C++ code) and a port to a
 - [simdjsone](https://github.com/saleyn/simdjsone): erlang bindings.
 - [lua-simdjson](https://github.com/FourierTransformer/lua-simdjson): lua bindings.
 - [hermes-json](https://hackage.haskell.org/package/hermes-json): haskell bindings.
-- [simdjzon](https://github.com/travisstaloch/simdjzon): zig port.
+- [zimdjson](https://github.com/EzequielRamis/zimdjson): Zig port.
+- [simdjzon](https://github.com/travisstaloch/simdjzon): Zig port.
 - [JSON-Simd](https://github.com/rawleyfowler/JSON-simd): Raku bindings.
 - [JSON::SIMD](https://metacpan.org/pod/JSON::SIMD): Perl bindings; fully-featured JSON module that uses simdjson for decoding.
 - [gemmaJSON](https://github.com/sainttttt/gemmaJSON): Nim JSON parser based on simdjson bindings.
@@ -207,6 +216,11 @@ 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
 ------------------------
 
@@ -216,7 +230,7 @@ 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/benchmark_reflection_usage_compilation.sh

@@ -0,0 +1,618 @@
+#!/bin/bash
+
+# JSON Parsing Compilation Benchmark: Reflection Usage vs Manual Parsing
+# Compares compilation times when ACTUALLY USING reflection for parsing vs manual parsing
+# This measures the compile-time cost of reflection-based automatic deserialization
+
+set -e
+
+echo "=== simdjson Reflection Usage Compilation Benchmark ==="
+echo "Measuring compilation impact of ACTUALLY USING reflection for parsing"
+echo "Starting at: $(date)"
+echo
+echo "╔════════════════════════════════════════════════════════════════════════════╗"
+echo "║                              METHODOLOGY                                   ║"
+echo "╠════════════════════════════════════════════════════════════════════════════╣"
+echo "║                                                                            ║"
+echo "║ FAIR COMPARISON STRATEGY:                                                  ║"
+echo "║ This benchmark compares two DIFFERENT approaches to parsing the same JSON: ║"
+echo "║                                                                            ║"
+echo "║ • MANUAL PARSING: Traditional simdjson with explicit .get() calls          ║"
+echo "║   - Uses doc[\"field\"].get(variable) for each field                         ║"
+echo "║   - No reflection involved                                                 ║"
+echo "║                                                                            ║"
+echo "║ • REFLECTION PARSING: Automatic deserialization with reflection            ║"
+echo "║   - Uses doc.get<MyStruct>() for automatic field mapping                   ║"
+echo "║   - Relies on compile-time reflection to generate parsing code             ║"
+echo "║                                                                            ║"
+echo "║ WHAT WE'RE MEASURING:                                                      ║"
+echo "║ • Compile-time cost of reflection-based automatic deserialization          ║"
+echo "║ • Template instantiation overhead for reflection parsing                   ║"
+echo "║ • Code generation complexity from using reflection features                ║"
+echo "║                                                                            ║"
+echo "║ TEST SCENARIOS:                                                            ║"
+echo "║                                                                            ║"
+echo "║ 1. SIMPLE STRUCT: Basic fields (string, int, bool)                         ║"
+echo "║    - Measures baseline reflection overhead                                 ║"
+echo "║                                                                            ║"
+echo "║ 2. NESTED STRUCT: Multiple levels of nested objects                        ║"
+echo "║    - Measures reflection complexity scaling                                ║"
+echo "║                                                                            ║"
+echo "║ 3. COMPLEX STRUCT: Arrays, optional fields, mixed types                    ║"
+echo "║    - Measures real-world reflection usage impact                           ║"
+echo "║                                                                            ║"
+echo "║ WHY THIS IS MEANINGFUL:                                                    ║"
+echo "║ • Shows actual cost of using reflection features                           ║"
+echo "║ • Measures compile-time code generation overhead                           ║"
+echo "║ • Helps developers understand reflection's compilation impact              ║"
+echo "║ • Compares equivalent functionality implemented two different ways         ║"
+echo "║                                                                            ║"
+echo "╚════════════════════════════════════════════════════════════════════════════╝"
+echo
+
+# Configuration
+ITERATIONS=10
+JOBS=4
+
+# ───────────────────────────── BOX-PRINT HELPER ────────────────────────────
+BOX_WIDTH=74                               # characters between the pipes
+print_box_line() {                         # usage: print_box_line "text"
+  printf "║ %-*s ║\n" "${BOX_WIDTH}" "$1"
+}
+
+# Function to test if a compiler supports reflection with debug output
+test_reflection_support() {
+    local compiler="$1"
+    echo "  → Testing compiler: $compiler"
+
+    if [ ! -x "$compiler" ] && ! command -v "$compiler" >/dev/null 2>&1; then
+        echo "  → Compiler not found or not executable"
+        return 1
+    fi
+
+    # Check compiler version first
+    echo "  → Compiler version: $("$compiler" --version 2>/dev/null | head -n1 || echo "version check failed")"
+
+    # Simple test: check if compiler accepts reflection flags
+    local test_file=$(mktemp /tmp/reflection_test_XXXXXX.cpp)
+    cat > "$test_file" << 'EOF'
+int main() {
+    return 0;
+}
+EOF
+
+    echo "  → Testing basic reflection flags..."
+    local test_exe=$(mktemp /tmp/reflection_test_XXXXXX)
+    local basic_result=$("$compiler" -freflection -fexpansion-statements -std=c++26 "$test_file" -o "$test_exe" 2>&1)
+    local basic_exit_code=$?
+
+    if [ $basic_exit_code -ne 0 ]; then
+        echo "  → Basic flags FAILED with exit code $basic_exit_code"
+        echo "  → Error output: $basic_result"
+        rm -f "$test_file" "$test_exe"
+        return 1
+    fi
+    echo "  → Basic flags: OK"
+    rm -f "$test_exe"
+
+    # Test reflection syntax
+    echo "  → Testing reflection syntax..."
+    cat > "$test_file" << 'EOF'
+struct Test {
+    int x;
+};
+
+int main() {
+    auto refl = ^^Test;
+    return 0;
+}
+EOF
+
+    local syntax_result=$("$compiler" -freflection -fexpansion-statements -std=c++26 "$test_file" -o "$test_exe" 2>&1)
+    local syntax_exit_code=$?
+
+    if [ $syntax_exit_code -eq 0 ]; then
+        echo "  → Reflection syntax: OK"
+        echo "  → ✓ REFLECTION SUPPORT CONFIRMED"
+        rm -f "$test_file" "$test_exe"
+        return 0
+    else
+        echo "  → Reflection syntax FAILED with exit code $syntax_exit_code"
+        echo "  → Error output: $syntax_result"
+        rm -f "$test_file" "$test_exe"
+        return 1
+    fi
+}
+
+# Find a compiler with reflection support
+echo "Searching for clang++ with reflection support..."
+
+REFLECTION_CXX=""
+REFLECTION_CC=""
+
+# List of potential clang++ locations to check
+POTENTIAL_COMPILERS=(
+    "/usr/local/bin/clang++"
+    "/opt/clang/bin/clang++"
+    "/usr/bin/clang++"
+    "clang++"
+)
+
+# If CXX is already set, test it first
+if [ -n "$CXX" ]; then
+    echo "Testing user-specified compiler: $CXX"
+    if test_reflection_support "$CXX"; then
+        REFLECTION_CXX="$CXX"
+        echo "✓ User-specified compiler supports reflection: $CXX"
+    else
+        echo "✗ User-specified compiler does not support reflection: $CXX"
+        echo "Will search for alternative..."
+    fi
+fi
+
+# If we don't have a working compiler yet, search for one
+if [ -z "$REFLECTION_CXX" ]; then
+    for compiler in "${POTENTIAL_COMPILERS[@]}"; do
+        echo "Testing: $compiler"
+        if test_reflection_support "$compiler"; then
+            REFLECTION_CXX="$compiler"
+            echo "✓ Found reflection-enabled compiler: $compiler"
+            break
+        else
+            echo "✗ No reflection support: $compiler"
+        fi
+    done
+fi
+
+# Check if we found a working compiler
+if [ -z "$REFLECTION_CXX" ]; then
+    echo
+    echo "╔════════════════════════════════════════════════════════════════════════════╗"
+    echo "║                                 ERROR                                      ║"
+    echo "╠════════════════════════════════════════════════════════════════════════════╣"
+    echo "║                                                                            ║"
+    echo "║ No clang++ compiler with reflection support found!                        ║"
+    echo "║                                                                            ║"
+    echo "║ This benchmark requires a compiler that supports C++26 reflection.        ║"
+    echo "║                                                                            ║"
+    echo "║ Options:                                                                   ║"
+    echo "║ 1. Use the Docker container: ./p2996/run_docker.sh                        ║"
+    echo "║ 2. Build clang with reflection from: https://github.com/bloomberg/clang-p2996 ║"
+    echo "║ 3. Set CXX environment variable to point to reflection-enabled clang++    ║"
+    echo "║                                                                            ║"
+    echo "║ Example: CXX=/path/to/reflection-clang++ ./benchmark_script.sh            ║"
+    echo "║                                                                            ║"
+    echo "╚════════════════════════════════════════════════════════════════════════════╝"
+    exit 1
+fi
+
+# Set the compilers
+export CXX="$REFLECTION_CXX"
+
+# Find corresponding C compiler
+if [ -n "$CC" ]; then
+    REFLECTION_CC="$CC"
+elif [ "$REFLECTION_CXX" = "/usr/local/bin/clang++" ]; then
+    REFLECTION_CC="/usr/local/bin/clang"
+elif [ "$REFLECTION_CXX" = "/opt/clang/bin/clang++" ]; then
+    REFLECTION_CC="/opt/clang/bin/clang"
+elif [ "$REFLECTION_CXX" = "/usr/bin/clang++" ]; then
+    REFLECTION_CC="/usr/bin/clang"
+else
+    REFLECTION_CC="clang"
+fi
+
+export CC="$REFLECTION_CC"
+
+echo
+echo "Using reflection-enabled compiler: $($CXX --version | head -n1)"
+echo "Using C compiler: $($CC --version | head -n1)"
+echo
+
+# Function to create manual parsing test
+create_manual_parsing_test() {
+    local test_name="$1"
+    local struct_complexity="$2"
+
+    cat > "${test_name}_manual.cpp" << 'EOF'
+#include <simdjson.h>
+#include <iostream>
+#include <string>
+#include <vector>
+#include <optional>
+
+// Test structures
+struct Person {
+    std::string name;
+    int age;
+    bool active;
+};
+
+struct Address {
+    std::string street;
+    std::string city;
+    int zipcode;
+};
+
+struct Employee {
+    Person person;
+    Address address;
+    std::vector<std::string> skills;
+    std::optional<std::string> department;
+    double salary;
+};
+
+// Manual parsing functions
+bool parse_person_manual(simdjson::ondemand::value& val, Person& person) {
+    auto obj = val.get_object();
+    if (obj.error()) return false;
+
+    for (auto field : obj) {
+        std::string_view key = field.unescaped_key();
+        if (key == "name") {
+            std::string_view name_val;
+            if (field.value().get(name_val)) return false;
+            person.name = name_val;
+        } else if (key == "age") {
+            if (field.value().get(person.age)) return false;
+        } else if (key == "active") {
+            if (field.value().get(person.active)) return false;
+        }
+    }
+    return true;
+}
+
+bool parse_address_manual(simdjson::ondemand::value& val, Address& address) {
+    auto obj = val.get_object();
+    if (obj.error()) return false;
+
+    for (auto field : obj) {
+        std::string_view key = field.unescaped_key();
+        if (key == "street") {
+            std::string_view street_val;
+            if (field.value().get(street_val)) return false;
+            address.street = street_val;
+        } else if (key == "city") {
+            std::string_view city_val;
+            if (field.value().get(city_val)) return false;
+            address.city = city_val;
+        } else if (key == "zipcode") {
+            if (field.value().get(address.zipcode)) return false;
+        }
+    }
+    return true;
+}
+
+bool parse_employee_manual(simdjson::ondemand::document& doc, Employee& employee) {
+    auto obj = doc.get_object();
+    if (obj.error()) return false;
+
+    for (auto field : obj) {
+        std::string_view key = field.unescaped_key();
+        if (key == "person") {
+            auto person_val = field.value();
+            if (!parse_person_manual(person_val, employee.person)) return false;
+        } else if (key == "address") {
+            auto addr_val = field.value();
+            if (!parse_address_manual(addr_val, employee.address)) return false;
+        } else if (key == "skills") {
+            auto skills_array = field.value().get_array();
+            if (skills_array.error()) return false;
+            for (auto skill : skills_array) {
+                std::string_view skill_val;
+                if (skill.get(skill_val)) return false;
+                employee.skills.emplace_back(skill_val);
+            }
+        } else if (key == "department") {
+            std::string_view dept_val;
+            if (!field.value().get(dept_val)) {
+                employee.department = dept_val;
+            }
+        } else if (key == "salary") {
+            if (field.value().get(employee.salary)) return false;
+        }
+    }
+    return true;
+}
+
+int main() {
+    simdjson::ondemand::parser parser;
+    std::string json_str = R"({
+        "person": {
+            "name": "John Doe",
+            "age": 30,
+            "active": true
+        },
+        "address": {
+            "street": "123 Main St",
+            "city": "Anytown",
+            "zipcode": 12345
+        },
+        "skills": ["C++", "JSON", "Programming"],
+        "department": "Engineering",
+        "salary": 85000.50
+    })";
+
+    simdjson::ondemand::document doc;
+    auto error = parser.iterate(simdjson::pad(json_str)).get(doc);
+    if (error) {
+        std::cerr << "Parse error" << std::endl;
+        return 1;
+    }
+
+    Employee employee;
+    if (!parse_employee_manual(doc, employee)) {
+        std::cerr << "Manual parsing failed" << std::endl;
+        return 1;
+    }
+
+    std::cout << "Manual parsing successful: " << employee.person.name
+              << ", age " << employee.person.age << std::endl;
+    return 0;
+}
+EOF
+}
+
+# Function to create reflection parsing test
+create_reflection_parsing_test() {
+    local test_name="$1"
+    local struct_complexity="$2"
+
+    cat > "${test_name}_reflection.cpp" << 'EOF'
+#include <simdjson.h>
+#include <iostream>
+#include <string>
+#include <vector>
+#include <optional>
+
+// Test structures (same as manual version)
+struct Person {
+    std::string name;
+    int age;
+    bool active;
+};
+
+struct Address {
+    std::string street;
+    std::string city;
+    int zipcode;
+};
+
+struct Employee {
+    Person person;
+    Address address;
+    std::vector<std::string> skills;
+    std::optional<std::string> department;
+    double salary;
+};
+
+int main() {
+    simdjson::ondemand::parser parser;
+    std::string json_str = R"({
+        "person": {
+            "name": "John Doe",
+            "age": 30,
+            "active": true
+        },
+        "address": {
+            "street": "123 Main St",
+            "city": "Anytown",
+            "zipcode": 12345
+        },
+        "skills": ["C++", "JSON", "Programming"],
+        "department": "Engineering",
+        "salary": 85000.50
+    })";
+
+    simdjson::ondemand::document doc;
+    auto error = parser.iterate(simdjson::pad(json_str)).get(doc);
+    if (error) {
+        std::cerr << "Parse error" << std::endl;
+        return 1;
+    }
+
+    // Use reflection-based automatic deserialization
+    Employee employee;
+    auto result = doc.get<Employee>();
+    if (result.error()) {
+        std::cerr << "Reflection parsing failed" << std::endl;
+        return 1;
+    }
+    employee = result.value();
+
+    std::cout << "Reflection parsing successful: " << employee.person.name
+              << ", age " << employee.person.age << std::endl;
+    return 0;
+}
+EOF
+}
+
+# Function to time compilation of parsing approach
+time_parsing_compilation() {
+    local description="$1"
+    local test_file="$2"
+    local use_reflection="$3"
+    local iteration="$4"
+
+    echo "[$iteration] $description"
+
+    # Clean build
+    rm -rf build_parsing_test
+    mkdir build_parsing_test
+    cd build_parsing_test
+
+    # Copy test file
+    cp "../$test_file" .
+
+    echo "  Configuring..."
+    if [ "$use_reflection" = "true" ]; then
+        cmake -DCMAKE_CXX_COMPILER="$CXX" \
+              -DSIMDJSON_DEVELOPER_MODE=ON \
+              -DSIMDJSON_STATIC_REFLECTION=ON \
+              -DBUILD_SHARED_LIBS=OFF \
+              ../.. >/dev/null 2>&1
+    else
+        cmake -DCMAKE_CXX_COMPILER="$CXX" \
+              -DSIMDJSON_DEVELOPER_MODE=ON \
+              -DSIMDJSON_STATIC_REFLECTION=OFF \
+              -DBUILD_SHARED_LIBS=OFF \
+              ../.. >/dev/null 2>&1
+    fi
+
+    echo "  Building simdjson..."
+    cmake --build . --target simdjson >/dev/null 2>&1
+
+    echo "  Compiling parsing test..."
+    # Time just the test compilation
+    start_time=$(date +%s.%N)
+
+    "$CXX" -std=c++17 -I../../include "$test_file" -L. -lsimdjson -o parsing_test >/dev/null 2>&1
+
+    end_time=$(date +%s.%N)
+
+    # Calculate time duration
+    time_taken=$(echo "$end_time $start_time" | awk '{printf "%.3f", $1 - $2}')
+    echo "  Completed in: ${time_taken}s"
+
+    cd ..
+    rm -rf build_parsing_test
+
+    echo "$time_taken"
+}
+
+# Create test files
+echo "Creating test files..."
+create_manual_parsing_test "complex" "complex"
+create_reflection_parsing_test "complex" "complex"
+
+# Arrays to store times
+times_manual=""
+times_reflection=""
+
+echo
+echo "=== MANUAL PARSING COMPILATION ==="
+echo "Testing traditional simdjson parsing with explicit .get() calls"
+echo
+
+for i in $(seq 1 $ITERATIONS); do
+    time_result=$(time_parsing_compilation "Compiling manual parsing test" "complex_manual.cpp" "false" "$i" | tail -n1)
+    times_manual="$times_manual $time_result"
+done
+
+echo
+echo "=== REFLECTION PARSING COMPILATION ==="
+echo "Testing automatic deserialization with doc.get<Struct>()"
+echo
+
+for i in $(seq 1 $ITERATIONS); do
+    time_result=$(time_parsing_compilation "Compiling reflection parsing test" "complex_reflection.cpp" "true" "$i" | tail -n1)
+    times_reflection="$times_reflection $time_result"
+done
+
+echo
+echo "╔════════════════════════════════════════════════════════════════════════════╗"
+echo "║           REFLECTION USAGE COMPILATION RESULTS                             ║"
+echo "╠════════════════════════════════════════════════════════════════════════════╣"
+echo "║                                                                            ║"
+echo "║ MANUAL PARSING (explicit .get() calls):                                    ║"
+count=1
+for t in $times_manual; do
+    if [[ "$t" =~ ^[0-9]+\.?[0-9]*$ ]]; then
+        line=$(printf "Run %2d: %7.3f seconds" "$count" "$t")
+        print_box_line "$line"
+        count=$((count + 1))
+    fi
+done
+echo "║                                                                            ║"
+echo "║ REFLECTION PARSING (automatic doc.get<Struct>()):                          ║"
+count=1
+for t in $times_reflection; do
+    if [[ "$t" =~ ^[0-9]+\.?[0-9]*$ ]]; then
+        line=$(printf "Run %2d: %7.3f seconds" "$count" "$t")
+        print_box_line "$line"
+        count=$((count + 1))
+    fi
+done
+echo "╚════════════════════════════════════════════════════════════════════════════╝"
+
+echo
+echo "╔════════════════════════════════════════════════════════════════════════════╗"
+echo "║                              ANALYSIS SUMMARY                              ║"
+echo "╠════════════════════════════════════════════════════════════════════════════╣"
+echo "║                                                                            ║"
+
+# Calculate averages and percentages - filter to only numeric values first
+manual_numbers=""
+reflection_numbers=""
+
+for t in $times_manual; do
+    if [[ "$t" =~ ^[0-9]+\.?[0-9]*$ ]]; then
+        manual_numbers="$manual_numbers $t"
+    fi
+done
+
+for t in $times_reflection; do
+    if [[ "$t" =~ ^[0-9]+\.?[0-9]*$ ]]; then
+        reflection_numbers="$reflection_numbers $t"
+    fi
+done
+
+if [ -n "$manual_numbers" ] && [ -n "$reflection_numbers" ]; then
+    manual_avg=$(echo "$manual_numbers" | awk '{sum=0; for(i=1;i<=NF;i++) sum+=$i; print sum/NF}')
+    reflection_avg=$(echo "$reflection_numbers" | awk '{sum=0; for(i=1;i<=NF;i++) sum+=$i; print sum/NF}')
+    overhead=$(echo "$reflection_avg $manual_avg" | awk '{printf "%.3f", $1 - $2}')
+
+    if [ $(echo "$manual_avg > 0" | awk '{print ($1 > 0)}') -eq 1 ]; then
+        percent=$(echo "$reflection_avg $manual_avg" | awk '{printf "%.1f", ($1 - $2) / $2 * 100}')
+    else
+        percent="0"
+    fi
+
+    print_box_line "MANUAL PARSING RESULTS:"
+    print_box_line "$(printf "Average compilation time: %.3fs" "$manual_avg")"
+    print_box_line ""
+    print_box_line "REFLECTION PARSING RESULTS:"
+    print_box_line "$(printf "Average compilation time: %.3fs" "$reflection_avg")"
+    print_box_line ""
+    print_box_line "REFLECTION OVERHEAD:"
+    print_box_line "$(printf "Additional time: %.3fs (%+.1f%%)" "$overhead" "$percent")"
+    print_box_line ""
+else
+    echo "║ ERROR: Could not extract valid timing data                                 ║"
+    echo "║ Manual times: $times_manual"
+    echo "║ Reflection times: $times_reflection"
+    echo "║                                                                            ║"
+fi
+
+echo "╠════════════════════════════════════════════════════════════════════════════╣"
+echo "║                              INTERPRETATION                                ║"
+echo "╠════════════════════════════════════════════════════════════════════════════╣"
+echo "║                                                                            ║"
+echo "║ WHAT THESE RESULTS SHOW:                                                   ║"
+echo "║                                                                            ║"
+echo "║ • COMPILE-TIME COST: How much longer reflection parsing takes to compile   ║"
+echo "║   - Higher % = more expensive template instantiation and codegen           ║"
+echo "║                                                                            ║"
+echo "║ • CODE GENERATION OVERHEAD: Reflection creates parsing code at compile     ║"
+echo "║   time, which requires more template processing than manual parsing        ║"
+echo "║                                                                            ║"
+echo "║ • DEVELOPER TRADE-OFF: Reflection provides automatic deserialization       ║"
+echo "║   but at the cost of increased compilation time                            ║"
+echo "║                                                                            ║"
+echo "║ EVALUATION:                                                                ║"
+echo "║ • Low overhead (0-20%): Reflection is compile-time efficient               ║"
+echo "║ • Medium overhead (20-50%): Noticeable but potentially acceptable          ║"
+echo "║ • High overhead (50%+): Significant compilation cost for reflection        ║"
+echo "║                                                                            ║"
+echo "║ REAL-WORLD IMPACT:                                                         ║"
+echo "║ • Small projects: Absolute time matters more than percentage               ║"
+echo "║ • Large projects: Percentage overhead compounds across many files          ║"
+echo "║ • CI/CD pipelines: Longer builds affect development velocity               ║"
+echo "║                                                                            ║"
+echo "╚════════════════════════════════════════════════════════════════════════════╝"
+
+# Clean up test files
+rm -f complex_manual.cpp complex_reflection.cpp
+
+echo
+echo "Completed at: $(date)"

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 +23,29 @@ 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);
+    if(auto e = doc.get_array().get<std::vector<point>>(result); e) { return false; }
+    // 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/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 the benchmark executable targets
+add_subdirectory(twitter_benchmark)
+add_subdirectory(citm_catalog_benchmark)

benchmark/static_reflect/benchmark_utils/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/citm_catalog_benchmark/CMakeLists.txt

@@ -0,0 +1,14 @@
+add_executable(benchmark_serialization_citm_catalog benchmark_serialization_citm_catalog.cpp)
+
+# Link with Rust benchmarking code if available
+if(TARGET serde-benchmark)
+    message(STATUS "serde-benchmark target was created. Linking CITM catalog benchmark with serde-benchmark.")
+    target_link_libraries(benchmark_serialization_citm_catalog PRIVATE serde-benchmark)
+    target_compile_definitions(benchmark_serialization_citm_catalog PRIVATE SIMDJSON_RUST_VERSION="${Rust_VERSION}")
+endif()
+
+target_link_libraries(benchmark_serialization_citm_catalog PRIVATE simdjson::simdjson nlohmann_json)
+target_link_libraries(benchmark_serialization_citm_catalog PRIVATE reflectcpp)
+target_compile_definitions(benchmark_serialization_citm_catalog PRIVATE SIMDJSON_BENCH_CPP_REFLECT=1)
+
+target_compile_definitions(benchmark_serialization_citm_catalog PRIVATE JSON_FILE="${BENCH_CITM_JSON}")

benchmark/static_reflect/citm_catalog_benchmark/benchmark_serialization_citm_catalog.cpp

@@ -0,0 +1,177 @@
+#include <cassert>
+#include <cstdlib>
+#include <ctime>
+#include <format>
+#include <fstream>
+#include <iostream>
+#include <nlohmann/json.hpp>
+#include <simdjson.h>
+#include <string>
+#include "citm_catalog_data.h"
+#include "nlohmann_citm_catalog_data.h"
+#include "../benchmark_utils/benchmark_helper.h"
+
+#if SIMDJSON_BENCH_CPP_REFLECT
+#include <rfl.hpp>
+#include <rfl/json.hpp>
+void bench_reflect_cpp(CitmCatalog &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::CitmCatalog *data) {
+  const char * output = serde_benchmark::str_from_citm(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_citm(data);
+                 measured_volume = strlen(output);
+                 if (measured_volume != output_volume) {
+                   printf("mismatch\n");
+                 }
+                 serde_benchmark::free_str(const_cast<char*>(output));
+               }));
+  serde_benchmark::free_str(const_cast<char*>(output));
+}
+#endif // SIMDJSON_RUST_VERSION
+
+void bench_nlohmann(CitmCatalog &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");
+                 }
+               }));
+}
+
+void bench_simdjson_static_reflection(CitmCatalog &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");
+                 }
+               }));
+}
+
+std::string read_file(const std::string &file_path, size_t read_size = 65536) {
+  std::ifstream stream(file_path, std::ios::binary);
+  if(!stream) {
+    std::cerr << "Could not open file '" << file_path << "'" << std::endl;
+    exit(EXIT_FAILURE);
+  }
+  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;
+}
+
+// Function to check if benchmark name contains filter substring
+bool matches_filter(const std::string& benchmark_name, const std::string& filter) {
+  return filter.empty() || benchmark_name.find(filter) != std::string::npos;
+}
+
+int main(int argc, char* argv[]) {
+  std::string filter;
+
+  // Parse command-line arguments
+  for (int i = 1; i < argc; ++i) {
+      if (strcmp(argv[i], "-f") == 0 || strcmp(argv[i], "--filter") == 0) {
+          if (i + 1 < argc) {
+              filter = argv[++i];
+          } else {
+              std::cerr << "Error: -f/--filter requires an argument" << std::endl;
+              return EXIT_FAILURE;
+          }
+      }
+  }
+  // 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;
+  }
+  CitmCatalog my_struct;
+  if(doc.get<CitmCatalog>().get(my_struct)) {
+    std::cerr << "Error loading CitmCatalog!" << std::endl;
+    return EXIT_FAILURE;
+  }
+
+  // Benchmarking the serialization
+  if (matches_filter("nlohmann", filter)) {
+    bench_nlohmann(my_struct);
+  }
+  if (matches_filter("simdjson_static_reflection", filter)) {
+    bench_simdjson_static_reflection(my_struct);
+  }
+#ifdef SIMDJSON_RUST_VERSION
+  if (matches_filter("rust", filter)) {
+    printf("# WARNING: The Rust benchmark may not be directly comparable since it does not use an equivalent data structure.\n");
+    // Create a Rust-compatible CitmCatalog structure from the JSON string
+    serde_benchmark::CitmCatalog* rust_data =
+      serde_benchmark::citm_from_str(json_str.c_str(), json_str.size());
+
+    if (rust_data == nullptr) {
+      printf("# Failed to initialize Rust data structure\n");
+    } else {
+      bench_rust(rust_data);
+      serde_benchmark::free_citm(rust_data);
+    }
+  }
+#endif
+#if SIMDJSON_BENCH_CPP_REFLECT
+  if (matches_filter("reflect_cpp", filter)) {
+    bench_reflect_cpp(my_struct);
+  }
+#endif
+  return EXIT_SUCCESS;
+}

benchmark/static_reflect/citm_catalog_benchmark/citm_catalog_data.h

@@ -0,0 +1,82 @@
+#ifndef CITM_CATALOG_DATA_H
+#define CITM_CATALOG_DATA_H
+
+#include <string>
+#include <vector>
+#include <map>
+
+struct Area {
+  int64_t id;
+  std::string name;
+  int64_t parent;
+  std::vector<int64_t> childAreas;
+  bool operator==(const Area &other) const = default;
+};
+
+struct AudienceSubCategory {
+  int64_t id;
+  std::string name;
+  int64_t parent;
+  bool operator==(const AudienceSubCategory &other) const = default;
+};
+
+struct Event {
+  int64_t id;
+  std::string name;
+  std::string description;
+  int64_t subTopic;
+  int64_t topic;
+  std::vector<int64_t> audience;
+  bool operator==(const Event &other) const = default;
+};
+
+struct Performance {
+  int64_t id;
+  std::string name;
+  int64_t event;
+  std::string start;
+  int64_t venueCode;
+  bool operator==(const Performance &other) const = default;
+};
+
+struct SeatCategory {
+  int64_t id;
+  std::string name;
+  std::vector<int64_t> areas;
+  bool operator==(const SeatCategory &other) const = default;
+};
+
+struct SubTopic {
+  int64_t id;
+  std::string name;
+  int64_t parent;
+  bool operator==(const SubTopic &other) const = default;
+};
+
+struct Topic {
+  int64_t id;
+  std::string name;
+  bool operator==(const Topic &other) const = default;
+};
+
+struct Venue {
+  int64_t id;
+  std::string name;
+  int64_t address;
+  bool operator==(const Venue &other) const = default;
+};
+
+struct CitmCatalog {
+  std::map<std::string, Area> areas;
+  std::map<std::string, AudienceSubCategory> audienceSubCategory;
+  std::map<std::string, Event> events;
+  std::map<std::string, Performance> performances;
+  std::map<std::string, SeatCategory> seatCategory;
+  std::map<std::string, SubTopic> subTopic;
+  std::map<std::string, Topic> topic;
+  std::map<std::string, Venue> venue;
+
+  bool operator==(const CitmCatalog &other) const = default;
+};
+
+#endif

benchmark/static_reflect/citm_catalog_benchmark/nlohmann_citm_catalog_data.h

@@ -0,0 +1,171 @@
+// nlohmann_citm_catalog_data.h
+#ifndef NLOHMANN_CITM_CATALOG_DATA_H
+#define NLOHMANN_CITM_CATALOG_DATA_H
+
+#include "citm_catalog_data.h"
+#include <nlohmann/json.hpp>
+#include <string>
+
+using json = nlohmann::json;
+
+// ---- Area ----
+inline void to_json(json &j, const Area &a) {
+    j = json{
+      {"id", a.id},
+      {"name", a.name},
+      {"parent", a.parent},
+      {"childAreas", a.childAreas}
+    };
+}
+inline void from_json(const json &j, Area &a) {
+    j.at("id").get_to(a.id);
+    j.at("name").get_to(a.name);
+    j.at("parent").get_to(a.parent);
+    j.at("childAreas").get_to(a.childAreas);
+}
+
+// ---- AudienceSubCategory ----
+inline void to_json(json &j, const AudienceSubCategory &asc) {
+    j = json{
+      {"id", asc.id},
+      {"name", asc.name},
+      {"parent", asc.parent}
+    };
+}
+inline void from_json(const json &j, AudienceSubCategory &asc) {
+    j.at("id").get_to(asc.id);
+    j.at("name").get_to(asc.name);
+    j.at("parent").get_to(asc.parent);
+}
+
+// ---- Event ----
+inline void to_json(json &j, const Event &e) {
+    j = json{
+      {"id", e.id},
+      {"name", e.name},
+      {"description", e.description},
+      {"subTopic", e.subTopic},
+      {"topic", e.topic},
+      {"audience", e.audience}
+    };
+}
+inline void from_json(const json &j, Event &e) {
+    j.at("id").get_to(e.id);
+    j.at("name").get_to(e.name);
+    j.at("description").get_to(e.description);
+    j.at("subTopic").get_to(e.subTopic);
+    j.at("topic").get_to(e.topic);
+    j.at("audience").get_to(e.audience);
+}
+
+// ---- Performance ----
+inline void to_json(json &j, const Performance &p) {
+    j = json{
+      {"id", p.id},
+      {"name", p.name},
+      {"event", p.event},
+      {"start", p.start},
+      {"venueCode", p.venueCode}
+    };
+}
+inline void from_json(const json &j, Performance &p) {
+    j.at("id").get_to(p.id);
+    j.at("name").get_to(p.name);
+    j.at("event").get_to(p.event);
+    j.at("start").get_to(p.start);
+    j.at("venueCode").get_to(p.venueCode);
+}
+
+// ---- SeatCategory ----
+inline void to_json(json &j, const SeatCategory &sc) {
+    j = json{
+      {"id", sc.id},
+      {"name", sc.name},
+      {"areas", sc.areas}
+    };
+}
+inline void from_json(const json &j, SeatCategory &sc) {
+    j.at("id").get_to(sc.id);
+    j.at("name").get_to(sc.name);
+    j.at("areas").get_to(sc.areas);
+}
+
+// ---- SubTopic ----
+inline void to_json(json &j, const SubTopic &st) {
+    j = json{
+      {"id", st.id},
+      {"name", st.name},
+      {"parent", st.parent}
+    };
+}
+inline void from_json(const json &j, SubTopic &st) {
+    j.at("id").get_to(st.id);
+    j.at("name").get_to(st.name);
+    j.at("parent").get_to(st.parent);
+}
+
+// ---- Topic ----
+inline void to_json(json &j, const Topic &t) {
+    j = json{
+      {"id", t.id},
+      {"name", t.name}
+    };
+}
+inline void from_json(const json &j, Topic &t) {
+    j.at("id").get_to(t.id);
+    j.at("name").get_to(t.name);
+}
+
+// ---- Venue ----
+inline void to_json(json &j, const Venue &v) {
+    j = json{
+      {"id", v.id},
+      {"name", v.name},
+      {"address", v.address}
+    };
+}
+inline void from_json(const json &j, Venue &v) {
+    j.at("id").get_to(v.id);
+    j.at("name").get_to(v.name);
+    j.at("address").get_to(v.address);
+}
+
+// ---- CitmCatalog ----
+inline void to_json(json &j, const CitmCatalog &c) {
+    j = json{
+      {"areas", c.areas},
+      {"audienceSubCategory", c.audienceSubCategory},
+      {"events", c.events},
+      {"performances", c.performances},
+      {"seatCategory", c.seatCategory},
+      {"subTopic", c.subTopic},
+      {"topic", c.topic},
+      {"venue", c.venue}
+    };
+}
+inline void from_json(const json &j, CitmCatalog &c) {
+    j.at("areas").get_to(c.areas);
+    j.at("audienceSubCategory").get_to(c.audienceSubCategory);
+    j.at("events").get_to(c.events);
+    j.at("performances").get_to(c.performances);
+    j.at("seatCategory").get_to(c.seatCategory);
+    j.at("subTopic").get_to(c.subTopic);
+    j.at("topic").get_to(c.topic);
+    j.at("venue").get_to(c.venue);
+}
+
+// Optional convenience functions for benchmarking
+inline std::string nlohmann_serialize(const CitmCatalog &catalog) {
+    json j = catalog;
+    return j.dump();
+}
+inline bool nlohmann_deserialize(const std::string &json_in, CitmCatalog &catalog) {
+    try {
+        catalog = json::parse(json_in);
+        return false; // success
+    } catch(...) {
+        return true;  // failure
+    }
+}
+
+#endif // NLOHMANN_CITM_CATALOG_DATA_H

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 +1,482 @@
+extern crate serde;
+extern crate serde_json;
+extern crate libc;
+
+use libc::{c_char, size_t};
+use serde::{Serialize, Deserialize};
+use std::{collections::HashMap, ffi::CString, ptr, slice};
+use serde::de::{self, Deserializer};
+/******************************************************/
+/******************************************************/
+/**
+ * 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 _);
+}
+
+// Functions associated with the CitmCatalog benchmark
+
+#[derive(Serialize, Deserialize)]
+pub struct Area {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+    pub parent: i64,
+    #[serde(rename = "childAreas")]
+    pub child_areas: Vec<i64>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct AudienceSubCategory {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+    pub parent: i64,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct Event {
+    #[serde(default)]
+    pub description: Option<String>,
+    pub id: i64,
+    #[serde(default)]
+    pub logo: Option<String>,
+    #[serde(default)]
+    pub name: Option<String>,
+    #[serde(default)]
+    pub subTopicIds: Vec<i64>,
+    #[serde(default)]
+    pub subjectCode: Option<String>,
+    #[serde(default)]
+    pub subtitle: Option<String>,
+    #[serde(default)]
+    pub topicIds: Vec<i64>,
+    // Add a catch-all for any other fields
+    #[serde(flatten)]
+    pub extra: HashMap<String, serde_json::Value>,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct Performance {
+    #[serde(default)]
+    pub id: i64,
+
+    #[serde(default)]
+    pub name: Option<String>,
+
+    #[serde(default)]
+    pub event: i64,
+
+    // This is the key fix - accept any JSON value type for timestamps
+    // This allows both string dates and integer timestamps (line 3511)
+    #[serde(default)]
+    pub start: serde_json::Value,
+
+    #[serde(rename = "venueCode")]
+    pub venue_code: String,
+
+    // Add a catch-all for any other fields
+    #[serde(flatten)]
+    pub extra: HashMap<String, serde_json::Value>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct SeatCategory {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+    pub areas: Vec<i64>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct SubTopic {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+    pub parent: i64,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Topic {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Venue {
+    pub id: i64,
+    pub name: Option<String>,  // Changed to Option
+    pub address: i64,
+}
+
+// Custom deserializers
+fn deserialize_string_to_area<'de, D>(deserializer: D) -> Result<HashMap<String, Area>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        let id_num = id.parse::<i64>().unwrap_or(0);
+        result.insert(id.clone(), Area {
+            id: id_num,
+            name: Some(name),
+            parent: 0,
+            child_areas: Vec::new(),
+        });
+    }
+
+    Ok(result)
+}
+
+fn deserialize_string_to_audience_subcategory<'de, D>(deserializer: D) -> Result<HashMap<String, AudienceSubCategory>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        let id_num = id.parse::<i64>().unwrap_or(0);
+        result.insert(id.clone(), AudienceSubCategory {
+            id: id_num,
+            name: Some(name),
+            parent: 0,
+        });
+    }
+
+    Ok(result)
+}
+
+fn deserialize_string_to_seat_category<'de, D>(deserializer: D) -> Result<HashMap<String, SeatCategory>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        let id_num = id.parse::<i64>().unwrap_or(0);
+        result.insert(id.clone(), SeatCategory {
+            id: id_num,
+            name: Some(name),
+            areas: Vec::new(),
+        });
+    }
+
+    Ok(result)
+}
+
+fn deserialize_string_to_subtopic<'de, D>(deserializer: D) -> Result<HashMap<String, SubTopic>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        let id_num = id.parse::<i64>().unwrap_or(0);
+        result.insert(id.clone(), SubTopic {
+            id: id_num,
+            name: Some(name),
+            parent: 0,
+        });
+    }
+
+    Ok(result)
+}
+
+fn deserialize_string_to_topic<'de, D>(deserializer: D) -> Result<HashMap<String, Topic>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        let id_num = id.parse::<i64>().unwrap_or(0);
+        result.insert(id.clone(), Topic {
+            id: id_num,
+            name: Some(name),
+        });
+    }
+
+    Ok(result)
+}
+
+fn deserialize_string_to_venue<'de, D>(deserializer: D) -> Result<HashMap<String, Venue>, D::Error>
+where D: Deserializer<'de> {
+    let string_map: HashMap<String, String> = HashMap::deserialize(deserializer)?;
+    let mut result = HashMap::new();
+
+    for (id, name) in string_map {
+        result.insert(id.clone(), Venue {
+            id: 0,
+            name: Some(name),
+            address: 0,
+        });
+    }
+
+    Ok(result)
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct CitmCatalog {
+    #[serde(rename = "areaNames")]
+    pub area_names: HashMap<String, String>,
+
+    #[serde(rename = "audienceSubCategoryNames")]
+    pub audience_subcategory_names: HashMap<String, String>,
+
+    #[serde(default)]
+    #[serde(rename = "blockNames")]
+    pub block_names: HashMap<String, String>,
+
+    pub events: HashMap<String, Event>,
+
+    #[serde(default)]
+    pub performances: Vec<Performance>,
+
+    #[serde(rename = "seatCategoryNames")]
+    pub seat_category_names: HashMap<String, String>,
+
+    #[serde(rename = "subTopicNames")]
+    pub subtopic_names: HashMap<String, String>,
+
+    #[serde(default)]
+    #[serde(rename = "subjectNames")]
+    pub subject_names: HashMap<String, String>,
+
+    #[serde(rename = "topicNames")]
+    pub topic_names: HashMap<String, String>,
+
+    #[serde(rename = "topicSubTopics")]
+    pub topic_subtopics: HashMap<String, Vec<i64>>,
+
+    #[serde(rename = "venueNames")]
+    pub venue_names: HashMap<String, String>,
+
+    // Catch-all for other fields
+    #[serde(flatten)]
+    pub extra: HashMap<String, serde_json::Value>,
+}
+
+/// Creates a CitmCatalog from a JSON string (UTF-8 encoded).
+#[no_mangle]
+pub unsafe extern "C" fn citm_from_str(
+    raw_input: *const c_char,
+    raw_input_length: usize
+) -> *mut CitmCatalog {
+    if raw_input.is_null() {
+        eprintln!("Error: Input pointer is null");
+        return ptr::null_mut();
+    }
+
+    // Convert the raw pointer + length into a Rust slice
+    let bytes = slice::from_raw_parts(raw_input as *const u8, raw_input_length);
+    let input_str = match std::str::from_utf8(bytes) {
+        Ok(s) => s,
+        Err(e) => {
+            eprintln!("Error: Invalid UTF-8 string: {}", e);
+            return ptr::null_mut();
+        }
+    };
+
+    // Try deserializing the input string into CitmCatalog
+    match serde_json::from_str::<CitmCatalog>(input_str) {
+        Ok(catalog) => Box::into_raw(Box::new(catalog)),
+        Err(e) => {
+            eprintln!("Error deserializing JSON: {}", e);
+            eprintln!("JSON snippet (first 200 chars): {:.200}...", input_str);
+            ptr::null_mut()
+        }
+    }
+}
+
+/// Serializes a CitmCatalog into a JSON string (UTF-8).
+#[no_mangle]
+pub unsafe extern "C" fn str_from_citm(raw_catalog: *mut CitmCatalog) -> *mut c_char {
+    if raw_catalog.is_null() {
+        eprintln!("Error: Catalog pointer is null");
+        return ptr::null_mut();
+    }
+
+    // Fix: Actually serialize the catalog
+    let catalog = &*raw_catalog;
+
+    match serde_json::to_string(catalog) {
+        Ok(serialized) => {
+            match CString::new(serialized) {
+                Ok(cstr) => cstr.into_raw(),
+                Err(e) => {
+                    eprintln!("Error creating CString: {}", e);
+                    ptr::null_mut()
+                }
+            }
+        },
+        Err(e) => {
+            eprintln!("Error serializing catalog to JSON: {}", e);
+            ptr::null_mut()
+        }
+    }
+}
+
+/// Frees the CitmCatalog pointer.
+#[no_mangle]
+pub unsafe extern "C" fn free_citm(raw_catalog: *mut CitmCatalog) {
+    if !raw_catalog.is_null() {
+        drop(Box::from_raw(raw_catalog));
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn free_str(ptr: *mut c_char) {
+    if !ptr.is_null() {
+        unsafe {
+            // Convert back into a CString, which automatically frees the memory
+            let _ = CString::from_raw(ptr);
+        }
+    }
+}

benchmark/static_reflect/serde-benchmark/serde_benchmark.h

@@ -0,0 +1,45 @@
+#ifndef serde_benchmark_ffi_h
+#define serde_benchmark_ffi_h
+
+/* Generated with cbindgen:0.28.0 */
+
+/* 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 CitmCatalog;
+
+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);
+
+/// Creates a CitmCatalog from a JSON string (UTF-8 encoded).
+CitmCatalog *citm_from_str(const char *raw_input, uintptr_t raw_input_length);
+
+/// Serializes a CitmCatalog into a JSON string (UTF-8).
+char *str_from_citm(CitmCatalog *raw_catalog);
+
+/// Frees the CitmCatalog pointer.
+void free_citm(CitmCatalog *raw_catalog);
+
+void free_str(char *ptr);
+
+}  // extern "C"
+
+}  // namespace serde_benchmark
+
+#endif  // serde_benchmark_ffi_h

benchmark/static_reflect/twitter_benchmark/CMakeLists.txt

@@ -0,0 +1,14 @@
+
+# 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_benchmark/benchmark_serialization_twitter.cpp

@@ -0,0 +1,168 @@
+#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_utils/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;
+}
+
+// Function to check if benchmark name contains filter substring
+bool matches_filter(const std::string& benchmark_name, const std::string& filter) {
+  return filter.empty() || benchmark_name.find(filter) != std::string::npos;
+}
+
+int main(int argc, char* argv[]) {
+  std::string filter;
+
+  // Parse command-line arguments
+  for (int i = 1; i < argc; ++i) {
+      if (strcmp(argv[i], "-f") == 0 || strcmp(argv[i], "--filter") == 0) {
+          if (i + 1 < argc) {
+              filter = argv[++i];
+          } else {
+              std::cerr << "Error: -f/--filter requires an argument" << std::endl;
+              return EXIT_FAILURE;
+          }
+      }
+  }
+  // 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
+  if (matches_filter("nlohmann", filter)) {
+    bench_nlohmann(my_struct);
+  }
+  if (matches_filter("simdjson_static_reflection", filter)) {
+    bench_simdjson_static_reflection(my_struct);
+  }
+#ifdef SIMDJSON_RUST_VERSION
+  if (matches_filter("rust", filter)) {
+    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
+  if (matches_filter("reflect_cpp", filter)) {
+    bench_reflect_cpp(my_struct);
+  }
+#endif
+  return EXIT_SUCCESS;
+}

benchmark/static_reflect/twitter_benchmark/nlohmann_twitter_data.h

@@ -0,0 +1,116 @@
+#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_benchmark/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

@@ -2,7 +2,10 @@
 # Flags used by exes and by the simdjson library (project-wide flags)
 #
 add_library(simdjson-internal-flags INTERFACE)
-
+if(NOT DEFINED CMAKE_POSITION_INDEPENDENT_CODE)
+  # We default to ON for all targets, so that we can use the library in shared libraries.
+  set_target_properties(simdjson-internal-flags PROPERTIES POSITION_INDEPENDENT_CODE ON)
+endif(NOT DEFINED CMAKE_POSITION_INDEPENDENT_CODE)
 
 option(SIMDJSON_CHECK_EOF "Check for the end of the input buffer. The setting is unnecessary since we require padding of the inputs. You should expect tests to fail with this option turned on." OFF)
 if(SIMDJSON_CHECK_EOF)
@@ -52,7 +55,6 @@ endif()
 
 option(SIMDJSON_SANITIZE_MEMORY "Sanitize memory" OFF)
 
-
 if(SIMDJSON_SANITIZE_MEMORY)
   message(STATUS "Setting the memory sanitizer.")
   add_compile_options(
@@ -110,8 +112,14 @@ 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)
@@ -158,9 +166,6 @@ We recommend Visual Studio 2019 or better on a 64-bit system.")
     add_compile_options(/Zi)
   endif()
 else()
-  if(NOT WIN32)
-    target_compile_options(simdjson-internal-flags INTERFACE -fPIC)
-  endif()
   target_compile_options(
       simdjson-internal-flags INTERFACE
       -Werror -Wall -Wextra -Weffc++ -Wsign-compare -Wshadow -Wwrite-strings

dependencies/CMakeLists.txt

@@ -1,5 +1,4 @@
 include(CMakeDependentOption)
-include(import.cmake)
 
 option(SIMDJSON_ALLOW_DOWNLOADS
         "Allow dependencies to be downloaded during configure time"
@@ -8,20 +7,24 @@ option(SIMDJSON_ALLOW_DOWNLOADS
 cmake_dependent_option(SIMDJSON_COMPETITION "Compile competitive benchmarks" ON
         SIMDJSON_ALLOW_DOWNLOADS OFF)
 cmake_dependent_option(SIMDJSON_GOOGLE_BENCHMARKS "compile the Google Benchmark benchmarks" ON
-        SIMDJSON_ALLOW_DOWNLOADS OFF)
+        "SIMDJSON_ALLOW_DOWNLOADS OR MINGW" 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/v1.9.4.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/351949906abde446f0314bf79606fb5d884f5be7.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"
@@ -60,38 +73,42 @@ 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/v3.12.0.zip
-    file(DOWNLOAD
+  )
-            "https://github.com/nlohmann/json/releases/download/v3.10.5/json.hpp"
+
-            "${nlohmann_json_SOURCE_DIR}/nlohmann/json.hpp")
+  set_property(TARGET nlohmann_json APPEND PROPERTY INTERFACE_COMPILE_DEFINITIONS SIMDJSON_COMPETITION_NLOHMANN_JSON)
-  endif()
-  add_library(nlohmann_json INTERFACE)
-  target_include_directories(nlohmann_json SYSTEM INTERFACE "${nlohmann_json_SOURCE_DIR}")
-  target_compile_definitions(nlohmann_json INTERFACE SIMDJSON_COMPETITION_NLOHMANN_JSON)
 
-  import_dependency(json11 dropbox/json11 ec4e452)
-  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)
 
   set(jsoncpp_SOURCE_DIR "${simdjson_SOURCE_DIR}/dependencies/jsoncppdist")
   add_library(jsoncpp STATIC "${jsoncpp_SOURCE_DIR}/jsoncpp.cpp")
   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 +127,22 @@ 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 +151,11 @@ 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)
@@ -140,7 +169,7 @@ int main() {}
   endif()
 
   add_library(competition-all INTERFACE)
-  target_link_libraries(competition-all INTERFACE competition-core jsoncpp json11 fastjson gason ujson4c)
+  target_link_libraries(competition-all INTERFACE competition-core jsoncpp fastjson gason ujson4c)
 endfunction()
 
 if(SIMDJSON_COMPETITION)
@@ -151,10 +180,12 @@ 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 5965670)
+        "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/basics.md

@@ -17,6 +17,8 @@ An overview of what you need to know to use simdjson, with examples.
   - [Using the parsed JSON](#using-the-parsed-json)
     - [Using the parsed JSON: additional examples](#using-the-parsed-json-additional-examples)
   - [Adding support for custom types](#adding-support-for-custom-types)
+    - [1. Specialize `simdjson::ondemand::value::get` to get custom types (pre-C++20)](#1-specialize-simdjsonondemandvalueget-to-get-custom-types-pre-c20)
+    - [2. Use `tag_invoke` for custom types (C++20)](#2-use-tag_invoke-for-custom-types-c20)
   - [Minifying JSON strings without parsing](#minifying-json-strings-without-parsing)
   - [UTF-8 validation (alone)](#utf-8-validation-alone)
   - [JSON Pointer](#json-pointer)
@@ -45,13 +47,15 @@ An overview of what you need to know to use simdjson, with examples.
 Requirements
 ------------------
 
-- A recent compiler (LLVM clang 6 or better, GNU GCC 7.4 or better, Xcode 11 or better) on a 64-bit (PPC, ARM or x64 Intel/AMD) POSIX systems such as macOS, freeBSD or Linux. We require that the compiler supports the C++11 standard or better.
+The simdjson library is widely deployed in popular systems such as the Node.js runtime
-- Visual Studio 2017 or better under 64-bit Windows. Users should target a 64-bit build (x64 or ARM64) instead of a 32-bit build (x86). We support the LLVM clang compiler under Visual Studio (clangcl) as well as as the regular Visual Studio compiler. We also support MinGW 64-bit under Windows.
+environment.
 
+- A recent compiler (LLVM clang 6 or better, GNU GCC 7.4 or better, Xcode 11 or better) on POSIX systems such as macOS, FreeBSD or Linux. We require that the compiler supports the C++11 standard or better. We test the library on a big-endian system (IBM s390x with Linux).
+- Visual Studio 2017 or better. We support the LLVM clang compiler under Visual Studio (clang-cl) as well as as the regular Visual Studio compiler. For better release performance (both compile time and execution time), we recommend Visual Studio users adopt LLVM (clang-cl). We also support MinGW 64-bit under Windows.
 
 Support for AVX-512 require a processor with AVX512-VBMI2 support (Ice Lake or better, AMD Zen 4 or better) under a 64-bit system and a recent compiler (LLVM clang 6 or better, GCC 8 or better, Visual Studio 2019 or better). You need a correspondingly recent assembler such as gas (2.30+) or nasm (2.14+): recent compilers usually come with recent assemblers. If you mix a recent compiler with an incompatible/old assembler (e.g., when using a recent compiler with an old Linux distribution), you may get errors at build time because the compiler produces instructions that the assembler does not recognize: you should update your assembler to match your compiler (e.g., upgrade binutils to version 2.30 or better under Linux) or use an older compiler matching the capabilities of your assembler.
 
-We test the library on a big-endian system (IBM s390x with Linux).
+
 
 Including simdjson
 ------------------
@@ -207,6 +211,25 @@ std::string data = "my data";
 simdjson::padded_string my_padded_data(data); // copies to a padded buffer
 ```
 
+You can then parse the JSON data from the `simdjson::padded_string` instance:
+
+
+```c++
+ondemand::document doc = parser.iterate(my_padded_data);
+```
+
+Whenever you pass an `std::string` reference to `parser::iterate`,
+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]";
+ondemand::document doc = parser.iterate(simdjson::pad(json));
+```
 
 We recommend against creating many `std::string` or many `std::padding_string` instances in your application to store your JSON data.
 Consider reusing the same buffers and limiting memory allocations.
@@ -247,7 +270,7 @@ documents.
 For code safety, you should keep (1) the `parser` instance, (2) the input string and (3) the document instance alive throughout your parsing. Additionally, you should follow the following rules:
 
 - A `parser` may have at most one document open at a time, since it holds allocated memory used for the parsing.
-- By design, you should only have one `document` instance per JSON document. Thus, if you must pass a document instance to a function, you should avoid passing it by value: choose to pass it by reference instance to avoid the copy. (We also provide a `document_reference` class if you need to pass by value.)
+- By design, you should only have one `document` instance per JSON document. Thus, if you must pass a document instance to a function, you should avoid passing it by value: choose to pass it by reference instance to avoid the copy. In any case, the `document` class does not have a copy constructor.
 
 During the `iterate` call, the original JSON text is never modified--only read. After you are done
 with the document, the source (whether file or string) can be safely discarded.
@@ -299,13 +322,15 @@ transcode the UTF-8 strings produced by the simdjson library to other formats. S
 Avoiding pitfalls: enable development checks
 --------------------
 
-We recommend that you first compile and run your code in Debug mode:
+We recommend that you first compile and run your code in debug mode:
 
 - under Visual Studio, it means having the `_DEBUG` macro defined,
--  for other compilers, it means leaving the `__OPTIMIZE__` macro undefined.
+- for many other compilers, it means leaving the `__OPTIMIZE__` macro undefined.
 
-The simdjson code will set `SIMDJSON_DEVELOPMENT_CHECKS=1` in debug mode.
+The simdjson code will set `SIMDJSON_DEVELOPMENT_CHECKS=1` in debug mode. Because
-Alternatively, you can set the macro `SIMDJSON_DEVELOPMENT_CHECKS` to 1 prior to including
+the C++ standard does not provide a direct way of checking for a debug build, and
+because you may want the checks while building with otimizations, you can set
+the  macro `SIMDJSON_DEVELOPMENT_CHECKS` to 1 prior to including
 the `simdjson.h` header to enable these additional checks: just make sure you remove the
 definition once your code has been tested. When `SIMDJSON_DEVELOPMENT_CHECKS` is set to 1, the
 simdjson library runs additional (expensive) tests on your code to help ensure that you are
@@ -314,9 +339,12 @@ using the library in a safe manner.
 Once your code has been tested, you can then run it in
 Release mode: under Visual Studio, it means having the `_DEBUG` macro undefined, and, for other
 compilers, it means setting `__OPTIMIZE__` to a positive integer. You can also forcefully
-disable these checks by setting `SIMDJSON_DEVELOPMENT_CHECKS` to 0. Once your code is tested, we
+disable these checks by setting `SIMDJSON_DEVELOPMENT_CHECKS` to 0.
-further encourage you to define `NDEBUG` in your Release builds to disable additional runtime
+
-testing and get the best performance.
+Once your code is tested, we further encourage you to define `NDEBUG` in your release
+builds to disable additional runtime testing and get the best performance. We
+disable these checks on a best-effort basis but the C++ standard does not provide
+a direct way to check for a release build.
 
 Using the parsed JSON
 ---------------------
@@ -401,9 +429,13 @@ support for users who avoid exceptions. See [the simdjson error handling documen
   of the object: to warn you, an OUT_OF_ORDER_ITERATION error is generated [when development checks](#avoiding-pitfalls-enable-development-checks) are active. If you need to access an object more
   than once, you may call `reset()` on it although we discourage this practice. Keep in mind that
   you should consume each value at most once.
+
+  When you are iterating through an object, you are advancing through its keys and values. You should not also access the object or other objects. E.g. within a loop over `myobject`, you should not be accessing `myobject`. The following is an anti-pattern: `for(auto value: myobject) {myobject["mykey"]}`.
+
+  You should never reset an object as you are iterating through it. The following is an anti-pattern: `for(auto value: myobject) {myobject.reset()}`.
 * **Array Index:** Because it is forward-only, you cannot look up an array element by index by index. Instead,
   you should iterate through the array and keep an index yourself. Exceptionally, if need a single value
-  out of the array, you may use an array access (e.g., `array[1]`).
+  out of the array, you may use an array access (e.g., `array[1]`). You should never reset an array as you are iterating through it. The following is an anti-pattern: `for(auto value: myarray) {myarray.reset()}`.
 * **Field Access:** To get the value of the "foo" field in an object, use `object["foo"]`. This will
   scan through the object looking for the field with the matching string, doing a character-by-character
   comparison. It may generate the error `simdjson::NO_SUCH_FIELD` if there is no such key in the object, it may throw an exception (see [Error handling](#error-handling)). For efficiency reason, you should avoid looking up the same field repeatedly: e.g., do
@@ -526,7 +558,7 @@ support for users who avoid exceptions. See [the simdjson error handling documen
   double value;
   auto doc = parser.iterate(abstract_json);
   auto error = doc["str"]["123"]["abc"].get(value);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   cout << value << endl; // Prints 3.14
   ```
   This examples also show how we can string several operations and only check for the error once, a strategy we call  *error chaining*.
@@ -536,7 +568,7 @@ support for users who avoid exceptions. See [the simdjson error handling documen
   For this purpose, `array` instances have a `count_elements` method. Users should be
   aware that the `count_elements` method can be costly since it requires scanning the
   whole array. You should only call `count_elements` as a last resort as it may
-  require scanning the document twice or more. You may use it as follows if your document is itself an array:
+  require scanning the document twice or more. You should never use the `count_elements` as part of an attempt to iterate through the array: use a `for` loop to iterate through arrays. In the spirit of On-Demand, the `count_elements` function does not validate the values in the array: they are validated when they are consumed. You may use it as follows if your document is itself an array:
 
   ```C++
   auto cars_json = R"( [ 40.1, 39.9, 37.7, 40.4 ] )"_padded;
@@ -788,6 +820,20 @@ for (ondemand::object points : parser.iterate(points_json)) {
 Adding support for custom types
 ----------------------
 
+There are 3 main ways provided by simdjson to deserialize a value into a custom type:
+
+1. Provide a [**template specialization** for member functions](https://en.cppreference.com/w/cpp/language/template_specialization#Members_of_specializations)
+   1. Specialize `simdjson::ondemand::document::get` for the whole document
+   2. Specialize `simdjson::ondemand::value::get` for each value
+2. Using `tag_invoke` *(the recommended way if your system supports C++20 or better)*
+3. Using static reflectioin (requires C++26 or better)
+
+We describe all of them in the following sections. Most users who have systems compatible with
+C++20 or better should skip ahead to [using `tag_invoke` for custom types (C++20)](#2-use-tag_invoke-for-custom-types-c20) as it is more powerful and simpler.
+The C++26 approach is even simpler.
+
+### 1. Specialize `simdjson::ondemand::value::get` to get custom types (pre-C++20)
+
 Suppose you have your own types, such as a `Car` struct:
 
 ```C++
@@ -821,9 +867,15 @@ type:
 ```
 
 We may do so by providing additional template definitions to the `ondemand::value` type.
-We may start by providing a definition for `std::vector<double>` as follows:
+We may start by providing a definition for `std::vector<double>` as follows. Observe
+how we guard the code with `#if !SIMDJSON_SUPPORTS_DESERIALIZATION`: that is because the necessary code
+is automatically provided by simdjson if C++20 (and concepts) are available.
+See [Use `tag_invoke` for custom types](#2-use-tag_invoke-for-custom-types-c20) if you have
+C++20 support.
 
 ```c++
+#if !SIMDJSON_SUPPORTS_DESERIALIZATION
+// The code is unnecessary with C++20:
 template <>
 simdjson_inline simdjson_result<std::vector<double>>
 simdjson::ondemand::value::get() noexcept {
@@ -839,6 +891,7 @@ simdjson::ondemand::value::get() noexcept {
   }
   return vec;
 }
+#endif
 ```
 
 We may then provide support for our `Car` struct:
@@ -850,26 +903,10 @@ simdjson_inline simdjson_result<Car> simdjson::ondemand::value::get() noexcept {
   auto error = get_object().get(obj);
   if (error) { return error; }
   Car car;
-  // Instead of repeatedly obj["something"], we iterate through the object which
+  if((error = obj["make"].get_string(car.make))) { return error; }
-  // we expect to be faster.
+  if((error = obj["model"].get_string(car.model))) { return error; }
-  for (auto field : obj) {
+  if((error = obj["year"].get_int64().get(car.year))) { return error; }
-    raw_json_string key;
+  if((error = obj["tire_pressure"].get<std::vector<double>>().get(car.tire_pressure))) { return error; }
-    error = field.key().get(key);
-    if (error) { return error; }
-    if (key == "make") {
-      error = field.value().get_string(car.make);
-      if (error) { return error; }
-    } else if (key == "model") {
-      error = field.value().get_string(car.model);
-      if (error) { return error; }
-    } else if (key == "year") {
-      error = field.value().get_int64().get(car.year);
-      if (error) { return error; }
-    } else if (key == "tire_pressure") {
-      error = field.value().get<std::vector<double>>().get(car.tire_pressure);
-      if (auto  error) { return error; }
-    }
-  }
   return car;
 }
 ```
@@ -893,6 +930,8 @@ struct Car {
   std::vector<double> tire_pressure;
 };
 
+#if !SIMDJSON_SUPPORTS_DESERIALIZATION
+// This code is not necessary if you have a C++20 compliant system:
 template <>
 simdjson_inline simdjson_result<std::vector<double>>
 simdjson::ondemand::value::get() noexcept {
@@ -908,7 +947,7 @@ simdjson::ondemand::value::get() noexcept {
   }
   return vec;
 }
-
+#endif
 
 template <>
 simdjson_inline simdjson_result<Car> simdjson::ondemand::value::get() noexcept {
@@ -916,26 +955,10 @@ simdjson_inline simdjson_result<Car> simdjson::ondemand::value::get() noexcept {
   auto error = get_object().get(obj);
   if (error) { return error; }
   Car car;
-  // Instead of repeatedly obj["something"], we iterate through the object which
+  if((error = obj["make"].get_string(car.make))) { return error; }
-  // we expect to be faster.
+  if((error = obj["model"].get_string(car.model))) { return error; }
-  for (auto field : obj) {
+  if((error = obj["year"].get_int64().get(car.year))) { return error; }
-    raw_json_string key;
+  if((error = obj["tire_pressure"].get<std::vector<double>>().get(car.tire_pressure))) { return error; }
-    error = field.key().get(key);
-    if (error) { return error; }
-    if (key == "make") {
-      error = field.value().get_string(car.make);
-      if (error) { return error; }
-    } else if (key == "model") {
-      error = field.value().get_string(car.model);
-      if (error) { return error; }
-    } else if (key == "year") {
-      error = field.value().get_int64().get(car.year);
-      if (error) { return error; }
-    } else if (key == "tire_pressure") {
-      error = field.value().get<std::vector<double>>().get(car.tire_pressure);
-      if (error) { return error; }
-    }
-  }
   return car;
 }
 
@@ -950,15 +973,14 @@ int main(void) {
   ondemand::parser parser;
   ondemand::document doc = parser.iterate(json);
   for (auto val : doc) {
-    Car c(val);
+    Car c(val); // an exception may be thrown
     std::cout << c.make << std::endl;
   }
-  direct();
   return EXIT_SUCCESS;
 }
 ```
 
-Observe that we require an explicit cast (`Car c(val)` instead of `for (Car c : doc) {`): it is by design.
+Observe that we require an explicit cast (`Car c(val)` instead of `for (Car c : doc) {`): it is by design. We require explicit casting.
 
 If you prefer to avoid exceptions, you may modify the `main` function as follows:
 
@@ -974,11 +996,11 @@ int main(void) {
   ondemand::parser parser;
   ondemand::document doc;
   auto error = parser.iterate(json).get(doc);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   for (auto val : doc) {
     Car c;
     error = val.get<Car>().get(c);
-    if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
     std::cout << c.make << std::endl;
   }
   return EXIT_SUCCESS;
@@ -1009,49 +1031,49 @@ struct Car {
   std::vector<double> tire_pressure;
 };
 
+#if !SIMDJSON_SUPPORTS_DESERIALIZATION
+// This code is not necessary if you have a C++20 compliant system:
 template <>
 simdjson_inline simdjson_result<std::vector<double>>
 simdjson::ondemand::value::get() noexcept {
   ondemand::array array;
-  if (auto error = get_array().get(array); error) { return error; }
+  auto error = get_array().get(array);
+  if (error) { return error; }
   std::vector<double> vec;
   for (auto v : array) {
     double val;
-    if (auto error = v.get_double().get(val); error) { return error; }
+    error = v.get_double().get(val);
+    if (error) { return error; }
     vec.push_back(val);
   }
   return vec;
 }
-
+#endif
 
 template <>
 simdjson_inline simdjson_result<Car> simdjson::ondemand::document::get() & noexcept {
   ondemand::object obj;
   auto error = get_object().get(obj);
-  if (error) {
+  if (error) { return error; }
-    return error;
-  }
   Car car;
-  // Instead of repeatedly obj["something"], we iterate through the object which
+  if((error = obj["make"].get_string(car.make))) { return error; }
-  // we expect to be faster.
+  if((error = obj["model"].get_string(car.model))) { return error; }
-  for (auto field : obj) {
+  if((error = obj["year"].get_int64().get(car.year))) { return error; }
-    raw_json_string key;
+  if((error = obj["tire_pressure"].get<std::vector<double>>().get(car.tire_pressure))) { return error; }
-    error = field.key().get(key);
+  return car;
-    if (error) { return error; }
+}
-    if (key == "make") {
+
-      error = field.value().get_string(car.make);
+
-      if (error) { return error; }
+template <>
-    } else if (key == "model") {
+simdjson_inline simdjson_result<Car> simdjson::ondemand::document::get() noexcept {
-      error = field.value().get_string(car.model);
+  ondemand::object obj;
-      if (error) { return error; }
+  auto error = get_object().get(obj);
-    } else if (key == "year") {
+  if (error) { return error; }
-      error = field.value().get_int64().get(car.year);
+  Car car;
-      if (error) { return error; }
+  if((error = obj["make"].get_string(car.make))) { return error; }
-    } else if (key == "tire_pressure") {
+  if((error = obj["model"].get_string(car.model))) { return error; }
-      error = field.value().get<std::vector<double>>().get(car.tire_pressure);
+  if((error = obj["year"].get_int64().get(car.year))) { return error; }
-      if (error) { return error; }
+  if((error = obj["tire_pressure"].get<std::vector<double>>().get(car.tire_pressure))) { return error; }
-    }
-  }
   return car;
 }
 
@@ -1066,6 +1088,284 @@ int main(void) {
 }
 ```
 
+### 2. Use `tag_invoke` for custom types (C++20)
+
+The simdjson library takes advantage of C++20. An immediate benefit
+is that you can deserialize JSON data directly in standard containers
+and other standard value types:
+
+```C++
+simdjson::padded_string json = R"({"data" : [1,2,3,4]})"_padded;
+
+simdjson::ondemand::parser parser;
+simdjson::ondemand::document d = parser.iterate(json);
+std::vector<uint8_t> array = d["data"].get<std::vector<uint8_t>>();
+```
+
+Appending to an existing container is just as easy:
+
+```C++
+std::vector<uint32_t> array = {0, 0};
+
+simdjson::padded_string json = R"({"data" : [1,2,3,4]})"_padded;
+
+simdjson::ondemand::parser parser;
+simdjson::ondemand::document d = parser.iterate(json);
+
+d["data"].get<std::vector<uint32_t>>(array);
+// array is now {0,0,1,2,3,4}
+```
+
+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.
+
+
+If your system supports C++20, we recommend that you adopt the `tag_invoke` approach
+instead to deserialize custom types. It may prove to be considerably simpler. When
+simdjson detects the necessary support, it sets the `SIMDJSON_SUPPORTS_DESERIALIZATION` macro
+to 1, otherwise it is set to 0.
+
+Consider a custom class `Car`:
+
+```C++
+struct Car {
+  std::string make;
+  std::string model;
+  int year;
+  std::vector<float> tire_pressure;
+};
+```
+
+Observe how we defined the class to use types that simdjson does not directly support (`float`, `int`).
+With C++20 support, the library grabs from the JSON the generic type (`double`, `int`) and then it
+casts it automatically.
+
+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
+```
+
+Observe how we call `get<std::vector<float>>()` even though we never defined support
+for `std::vector<float>` in the simdjson library: it is all automated thanks to C++20 concepts.
+
+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.
+
+You can use it like so:
+
+```cpp
+  simdjson::padded_string json =
+      R"( { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] })"_padded;
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc = parser.iterate(json);
+  Car c(doc);
+  std::cout << c.make << std::endl;
+```
+
+Observe how we first get an instance of `document` and then we cast.
+
+You can also handle errors explicitly:
+
+```cpp
+  Car c;
+  auto error = doc.get(c);
+  if(error) { std::cerr << simdjson::error_message(error); return false; }
+  std::cout << c.make << std::endl;
+```
+
+You can also read instances of `Car` from an array or an object:
+```cpp
+  simdjson::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;
+
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc = parser.iterate(json);
+  for (auto val : doc) {
+    Car c(val); // an exception may be thrown
+    std::cout << c.year << std::endl;
+  }
+```
+
+Observe how we first get a generic (`val`) which we cast to `Car`. It is by design: we require
+explicit casting. The cast may throw an exception.
+
+Once more, you can handle errors explicitly:
+
+```cpp
+  for (auto val : doc) {
+    Car c;
+    auto error = val.get(c);
+    if(error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
+  }
+```
+
+You can also use the custom `Car` type as part of a template such as `std::vector`:
+
+```cpp
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc = parser.iterate(json);
+  std::vector<Car> cars(doc);
+  // visual studio users need an explicit call:
+  //   std::vector<Car> cars = doc.get<std::vector<Car>>();
+  // because the compiler does not know whether to convert
+  // doc to an unsigned int or to a vector.
+  for(Car& c : cars) {
+    std::cout << c.year << std::endl;
+  }
+```
+
+By default, we support a wide range of standard templates such as
+`std::vector`,  `std::list`, `std::set`, `std::stack`, `std:queue`,
+`std:deque`, `std::priority_queue`, `std::unordered_set`, `std::multiset`,
+`std::unordered_multiset`, `std::unique_ptr`, `std::shared_ptr`, `std::optional`,
+etc. They are handled automatically.
+
+E.g., you can recover an `std::unique_ptr<Car>` like so:
+```C++
+int main() {
+  auto const json = R"( { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] })"_padded;
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc = parser.iterate(json);
+  std::unique_ptr<Car> c(doc);
+  std::cout << c->make << std::endl;
+  return EXIT_SUCCESS;
+}
+```
+
+You may also conditionally fill in `std::optional` values.
+
+```C++
+  padded_string json =
+      R"( { "car1": { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] }
+})"_padded;
+  ondemand::parser parser;
+  ondemand::document doc = parser.iterate(json);
+  std::optional<Car> car;
+  error = doc["key not found"].get<std::optional<Car>>().get(car);
+  // car has no value, error != simdjson::SUCCESS
+  error = doc["car1"].get<std::optional<Car>>().get(car);
+  // car has value Car{"Toyota", "Camry", 2018, {40.1f, 39.9f}}
+  // error is simdjson::SUCCESS
+```
+
+You can also deserialize to map-like types with keys that can be constructed
+from `std::string_view` instances:
+
+
+```C++
+  padded_string json =
+      R"( { "car1": { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] }
+})"_padded;
+  ondemand::parser parser;
+  ondemand::document doc = parser.iterate(json);
+  std::map<std::string,Car> cars;
+  error = doc.get<std::map<std::string,Car>>().get(cars);
+  // car has value car1->Car{"Toyota", "Camry", 2018, {40.1f, 39.9f}}
+  // error is simdjson::SUCCESS
+```
+
+And so forth.
+
+Advanced users may want to overwrite the defaults provided by the simdjson library.
+Suppose for example that you want to construct an instance of `std::list<Car>`, but
+you also want to filter out any car made by Toyota. You may provide your own
+`tag_invoke` function:
+
+```c++
+namespace simdjson {
+// suppose we want to filter out all Toyotas
+template <typename simdjson_value>
+auto tag_invoke(deserialize_tag, simdjson_value &val, std::list<Car>& car) {
+  ondemand::array arr;
+  auto error = val.get_array().get(arr);
+  if (error) {
+    return error;
+  }
+  for (auto v : arr) {
+    Car c;
+    if ((error = v.get<Car>().get(c))) {
+      return error;
+    }
+    if(c.make != "Toyota") {
+      car.push_back(c);
+    }
+  }
+  return simdjson::SUCCESS;
+}
+}
+```
+
+With this code, deserializing an `std::list<Car>` instance would capture only the cars
+that are not made by Toyota.
+
+### 3. Using static reflection (C++26)
+
+If you have a C++26 compatible compiler, you can compile
+your code with the `SIMDJSON_STATIC_REFLECTION` macro set:
+
+```cpp
+#define SIMDJSON_STATIC_REFLECTION 1
+//...
+#include "simdjson.h"
+```
+
+Then you can deserialize a type such as `Car` automatically:
+```cpp
+std::string json =  R"( { "make": "Toyota", "model": "Camry",  "year": 2018,
+       "tire_pressure": [ 40.1, 39.9 ] } )";
+simdjson::ondemand::parser parser;
+simdjson::ondemand::document doc = parser.iterate(simdjson::pad(json)).get(doc);
+Car c = doc.get<Car>();
+```
+
+You can also automatically serialize the `Car` instance to a JSON string, see
+our [Builder documentation](builder.md).
+
 
 Minifying JSON strings without parsing
 ----------------------
@@ -1081,6 +1381,7 @@ In some cases, you may have valid JSON strings that you do not wish to parse but
   std::unique_ptr<char[]> buffer{new char[length]};
   size_t new_length{}; // It will receive the minified length.
   auto error = simdjson::minify(some_string, length, buffer.get(), new_length);
+  if(error) { std::cerr << simdjson::error_message(error); }
   // The buffer variable now has "[1,2,3,4]" and new_length has value 9.
 ```
 
@@ -1224,9 +1525,11 @@ be represented as `value` instances. You can check that a document is a scalar w
 JSONPath
 ------------
 
-The simdjson library now 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.
+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.
 
-This implementation relies on `at_path()` converting its argument to JSON Pointer and then calling `at_pointer`, which makes use of [`rewind`](#rewind) to reset the parser at the beginning of the document. Hence, it invalidates all previously parsed values, objects and arrays: make sure to consume the values between each call to `at_path`.
+This implementation relies on `at_path()` converting its argument to JSON Pointer and then calling `at_pointer`, which makes use of
+[`rewind`](#rewind) to reset the parser at the beginning of the document. Hence, it invalidates all previously parsed values, objects
+ and arrays: make sure to consume the values between each call to `at_path`.
 
 Consider the following example:
 
@@ -1267,6 +1570,18 @@ doc.at_path(".\\u00E9") == 123; // true
 doc.at_path((const char*)u8".\u00E9") // returns an error (NO_SUCH_FIELD)
 ```
 
+
+We also support the `$` prefix. When you start a JSONPath expression with $, you are indicating that the path starts from the root of the JSON document. E.g.,
+
+```c++
+auto json = R"( { "c" :{ "foo": { "a": [ 10, 20, 30 ] }}, "d": { "foo2": { "a": [ 10, 20, 30 ] }} , "e": 120 })"_padded;
+ondemand::parser parser;
+ondemand::document doc = parser.iterate(json);
+ondemand::object obj = doc.get_object();
+int64_t x = obj.at_path("$.c.foo.a[1]"); // 20
+x = obj.at_path("$.d.foo2.a.2"); // 30
+```
+
 Error handling
 --------------
 
@@ -1278,7 +1593,7 @@ pair. You can retrieve the value with .get() without generating an exception, li
 ```c++
 ondemand::document doc;
 auto error = parser.iterate(json).get(doc);
-if (error) { cerr << error << endl; exit(1); }
+if(error) { std::cerr << simdjson::error_message(error); exit(1); }
 ```
 
 When there is no error, the error code `simdjson::SUCCESS`is returned: it evaluates as false as a Boolean.
@@ -1290,7 +1605,20 @@ Some errors are recoverable:
 * You may get the error `simdjson::INCORRECT_TYPE` after trying to convert a value to an incorrect type: e.g., you expected a number and try to convert the value to a number, but it is an array.
 * You may query a key from an object, but the key is missing in which case you get the error `simdjson::NO_SUCH_FIELD`: e.g., you call `obj["myname"]` and the object does not have a key `"myname"`.
 
-Other errors (e.g., `simdjson::INCOMPLETE_ARRAY_OR_OBJECT`) may indicate a fatal error and often follow from the fact that the document is not valid JSON. In which case, it is no longer possible to continue accessing the document: calling the method `is_alive()` on the document instance returns false. All following accesses will keep returning the same fatal error (e.g., `simdjson::INCOMPLETE_ARRAY_OR_OBJECT`).
+Other errors (`simdjson::INCOMPLETE_ARRAY_OR_OBJECT` and `simdjson::TAPE_ERROR`)  indicate a fatal error and follow from the fact that the document is not valid JSON. These errors are not recoverable: you cannot continue. In which case, it is no longer safe to continue accessing the document: calling the method `is_alive()` on the document instance returns false. It is your responsibility as a user to stop using the simdjson
+document after encountering these fatal errors. Consider the following example, after
+the fatal error, the document instance cannot be used. Observe how the JSON input is invalid.
+```cpp
+  simdjson::padded_string badjson = R"( { "make": "Toyota", "model": "Camry",  "year"})"_padded;
+  simdjson::ondemand::parser parser;
+  simdjson::ondemand::document doc;
+  auto errordoc = parser.iterate(badjson).get(doc);
+  // errordoc == simdjson::SUCCESS
+  simdjson::ondemand::value v;
+  auto error = doc.get_object()["year"].get(v);
+  // simdjson::is_fatal(error)) is true!
+  // doc.is_alive() is false
+```
 
 When you use the code without exceptions, it is your responsibility to check for error before using the
 result: if there is an error, the result value will not be valid and using it will caused undefined behavior. Most compilers should be able to help you if you activate the right
@@ -1309,7 +1637,7 @@ bool simple_error_example() {
     auto error = doc["bad number"].get_double().get(x);
     // returns "simdjson::NUMBER_ERROR"
     if (error != SUCCESS) {
-      std::cout << error << std::endl;
+      std::cerr << simdjson::error_message(error) << std::endl;
       return false;
     }
     std::cout << "Got " << x << std::endl;
@@ -1343,7 +1671,7 @@ Notice how we can retrieve the exact error condition (in this instance `simdjson
 from the exception.
 
 We can write a "quick start" example where we attempt to parse the following JSON file and access some data, without triggering exceptions:
-```JavaScript
+```JSON
 {
   "statuses": [
     {
@@ -1362,7 +1690,6 @@ We can write a "quick start" example where we attempt to parse the following JSO
 Our program loads the file, selects value corresponding to key `"search_metadata"` which expected to be an object, and then
 it selects the key `"count"` within that object.
 
-
 ```C++
 #include <iostream>
 #include "simdjson.h"
@@ -1370,10 +1697,10 @@ it selects the key `"count"` within that object.
 int main(void) {
   simdjson::ondemand::parser parser;
   auto error = padded_string::load("twitter.json").get(json);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   simdjson::ondemand::document tweets;
   error = parser.iterate(json).get(tweets);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   simdjson::ondemand::value res;
   error = tweets["search_metadata"]["count"].get(res);
   if (error != SUCCESS) {
@@ -1389,7 +1716,7 @@ The following is a similar example where one wants to get the id of the first tw
 triggering exceptions. To do this, we use `["statuses"].at(0)["id"]`. We break that expression down:
 
 - Get the list of tweets (the `"statuses"` key of the document) using `["statuses"]`). The result is expected to be an array.
-- Get the first tweet using `.at(0)`. The result is expected to be an object.
+- Get the first tweet using `.at(0)`. The result is expected to be an object. Observe that the `at` method can only be called once on an array (it cannot be used for iteration).
 - Get the id of the tweet using ["id"]. We expect the value to be a non-negative integer.
 
 Observe how we use the `at` method when querying an index into an array, and not the bracket operator.
@@ -1404,18 +1731,18 @@ int main(void) {
   simdjson::ondemand::document tweets;
   padded_string json;
   auto error = padded_string::load("twitter.json").get(json);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   error = parser.iterate(json).get(tweets);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   uint64_t identifier;
   error = tweets["statuses"].at(0)["id"].get(identifier);
-  if (error) { std::cerr << error << std::endl; return EXIT_FAILURE; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return EXIT_FAILURE; }
   std::cout << identifier << std::endl;
 }
 ```
 
-The `at` method can only be called once on an array. It cannot be used
+*Important remark*: The `at` method can only be called once on an array. It cannot be used
-to iterate through the values of an array.
+to iterate through the values of an array. We deliberately forbid this usage to avoid performance antipatterns. If you need to iterate through the values of an array, you should use a `for` loop.
 
 ### Error handling examples without exceptions
 
@@ -1433,40 +1760,40 @@ bool parse() {
 
   // Iterating through an array of objects
   auto error = parser.iterate(cars_json).get(doc);
-  if (error) { std::cerr << error << std::endl; return false; }
+  if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
   ondemand::array cars; // invalid until the get() succeeds
   error = doc.get_array().get(cars);
 
   for (auto car_value : cars) {
     ondemand::object car; // invalid until the get() succeeds
     error = car_value.get_object().get(car);
-    if (error) { std::cerr << error << std::endl; return false; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
 
     // Accessing a field by name
     std::string_view make;
     std::string_view model;
     error = car["make"].get(make);
-    if (error) { std::cerr << error << std::endl; return false; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
     error = car["model"].get(model);
-    if (error) { std::cerr << error << std::endl; return false; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
 
     cout << "Make/Model: " << make << "/" << model << endl;
 
     // Casting a JSON element to an integer
     uint64_t year{};
     error = car["year"].get(year);
-    if (error) { std::cerr << error << std::endl; return false; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
     cout << "- This car is " << 2020 - year << " years old." << endl;
 
     // Iterating through an array of floats
     double total_tire_pressure = 0;
     ondemand::array pressures;
     error = car["tire_pressure"].get_array().get(pressures);
-    if (error) { std::cerr << error << std::endl; return false; }
+    if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
     for (auto tire_pressure_value : pressures) {
       double tire_pressure;
       error = tire_pressure_value.get_double().get(tire_pressure);
-      if (error) { std::cerr << error << std::endl; return false; }
+      if (error) { std::cerr << simdjson::error_message(error) << std::endl; return false; }
       total_tire_pressure += tire_pressure;
     }
     cout << "- Average tire pressure: " << (total_tire_pressure / 4) << endl;
@@ -1588,7 +1915,7 @@ auto doc = parser.iterate(broken_json);
 int64_t i;
 auto error = doc["integer"].get_int64().get(i);    // Expect to get integer from "integer" key, but get TAPE_ERROR
 if (error) {
-  std::cout << error << std::endl;    // Prints TAPE_ERROR error message
+  std::cerr << simdjson::error_message(error) << std::endl;    // Prints TAPE_ERROR error message
   // Recover a pointer to the location of the first error:
   const char * ptr;
   doc.current_location().get(ptr);
@@ -1625,7 +1952,7 @@ auto doc = parser.iterate(json);
 int64_t i;
 auto error = doc["integer"].get_int64().get(i);    // Incorrect call on array, INCORRECT_TYPE error
 if (error) {
-  std::cout << error << std::endl;     // Prints INCORRECT_TYPE error message
+  std::cerr << simdjson::error_message(error) << std::endl;     // Prints INCORRECT_TYPE error message
   std::cout<< doc.current_location() << std::endl;  // Prints "[1,2,3] " (location of INCORRECT_TYPE error)
 }
 ```
@@ -1680,6 +2007,8 @@ conclude that you have trailing content and that your document is not valid JSON
 You may then use `doc.current_location()` to obtain a pointer to the start of the trailing
 content.
 
+Example 1.
+
 ```C++
   auto json = R"([1, 2] foo ])"_padded;
   ondemand::parser parser;
@@ -1694,6 +2023,21 @@ content.
   }
 ```
 
+Example 2.
+
+```cpp
+  auto json = R"(["extra close"]])"_padded;
+  ondemand::parser parser;
+  ondemand::document doc = parser.iterate(json);
+  ondemand::array array = doc.get_array();
+  for (std::string_view values : array) {
+    std::cout << values << std::endl;
+  }
+  if(!doc.at_end()) {
+    std::cerr << "trailing content at byte index " << doc.current_location() - json.data() << std::endl;
+  }
+```
+
 The `at_end()` method is equivalent to `doc.current_location().error() == simdjson::SUCCESS` but
 more convenient.
 
@@ -1734,6 +2078,7 @@ to the document `rewind()` method, except that it does not rewind the
 internal string buffer. Thus you should consume values only once
 even if you can iterate through the array or object more than once.
 If you unescape a string within an array more than once, you have unsafe code.
+You must not call `reset()` on an object or an array as you are iterating through it.
 
 
 Newline-Delimited JSON (ndjson) and JSON lines
@@ -1745,7 +2090,7 @@ serialize data into streams of multiple JSON documents. That is, instead of one
 write out multiple records as independent JSON documents, to be read one-by-one.
 
 The simdjson library also supports multithreaded JSON streaming through a large file
-containing many smaller JSON documents in either [ndjson](http://ndjson.org)
+containing many smaller JSON documents in either [ndjson](https://github.com/ndjson/ndjson-spec)
 or [JSON lines](http://jsonlines.org) format. If your JSON documents all contain arrays
 or objects, we even support direct file concatenation without whitespace. However, if there
 is content between your JSON documents, it should be exclusively ASCII white-space characters.
@@ -1798,7 +2143,7 @@ for (auto doc: stream) {
     error = doc.at_pointer("/4").get(val);
     // error == simdjson::CAPACITY
     if (error) {
-      std::cerr << error << std::endl;
+      std::cerr << simdjson::error_message(error) << std::endl;
       // We left 293 bytes unprocessed at the tail end of the input.
       std::cout << " unprocessed bytes at the end: " << stream.truncated_bytes() << std::endl;
       break;
@@ -1959,6 +2304,11 @@ Thus it is a dynamically typed number. Before accessing the value, you must dete
 such a number of `get_number()`, you get the error `BIGINT_ERROR`. You can access the underlying string of digits with the function `raw_json_token()` which returns a `std::string_view` instance starting at the beginning of the digit. You can also call `get_double()` to get a floating-point approximation.
 
 
+  By default, the string `-0` is parsed as the integer 0 as in Python or C++. If you set the macro
+  `SIMDJSON_MINUS_ZERO_AS_FLOAT` to `1` when building simdjson, you can get that `-0` is mapped to `-0.0`
+  as in JavaScript. You can get the desired effect by building simdjson with cmake setting the
+  `SIMDJSON_MINUS_ZERO_AS_FLOAT` to on: `cmake -B build -D SIMDJSON_MINUS_ZERO_AS_FLOAT=ON`.
+
 You must check the type before accessing the value: it is an error to call `get_int64()` when `number.get_number_type()` is not `number_type::signed_integer` and when `number.is_int64()` is false. You are responsible for this check as the user of the library.
 
 The `get_number()` function is designed with performance in mind. When calling `get_number()`, you scan the number string only once, determining efficiently the type and storing it in an efficient manner.
@@ -2286,6 +2636,20 @@ can use it with features such as `std::optional`:
   // value was populated with "3.1416"
 ```
 
+You can generally convert any answer that would return an `std::string_view`.
+
+```cpp
+  auto json = R"({"\u0062\u0065\u0062\u0065": 2} })"_padded;
+  ondemand::parser parser;
+  ondemand::document doc = parser.iterate(json);
+  ondemand::object object = doc.get_object();
+  for (auto field : object) {
+    std::string key;
+    error = field.unescaped_key().get(key);
+    if(error) { /* */ }
+  }
+```
+
 You should be mindful of the trade-off: allocating multiple
 `std::string` instances can become expensive.
 
@@ -2304,7 +2668,7 @@ The CPU detection, which runs the first time parsing is attempted and switches t
 parser for your CPU, is transparent and thread-safe.
 Our runtime dispatching is based on global objects that are instantiated at the beginning of the
 main thread and may be discarded at the end of the main thread. If you have multiple threads running
-and some threads use the library while the main thread is cleaning up ressources, you may encounter
+and some threads use the library while the main thread is cleaning up resources, you may encounter
 issues. If you expect such problems, you may consider using [std::quick_exit](https://en.cppreference.com/w/cpp/utility/program/quick_exit).
 
 In a threaded environment, stack space is often limited. Running code like simdjson in debug mode may require hundreds of kilobytes of stack memory. Thus stack overflows are a possibility. We recommend you turn on optimization when working in an environment where stack space is limited. If you must run your code in debug mode, we recommend you configure your system to have more stack space. We discourage you from running production code based on a debug build.
@@ -2673,6 +3037,7 @@ Performance tips
 	std::string_view year = data["year"];
 	std::string_view rating = data["rating"];
   ```
+- You will get better performance if you seek the keys in the order in which they appear in the document. So if processing `{"a":1, "b":2, "c":3}`, do `value1 = data["a"]; value2 = data["b"]; value3 data["c"];` and not `value2 = data["b"]; value1 = data["a"]; value3 data["c"];`. Of course, it is not always possible to know for sure in which order the keys appear.
 
 
 

doc/builder.md

@@ -0,0 +1,132 @@
+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.start_object();
+
+        // "make"
+        builder.append_key_value("make", car.make);
+        builder.append_comma();
+
+        // "model"
+        builder.append_key_value("model", car.model);
+        builder.append_comma();
+
+        // "year"
+        builder.append_key_value("year", car.year);
+        builder.append_comma();
+
+        // "tire_pressure"
+        builder.escape_and_append_with_quotes("tire_pressure");
+        builder.append_colon();
+        builder.start_array();
+        // 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_comma();
+            }
+        }
+        builder.end_array();
+        builder.end_object();
+    }
+
+    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

@@ -7,7 +7,9 @@ An overview of what you need to know to use simdjson, with examples.
 * [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)
+* [C++20 Support](#c20-support)
 * [JSON Pointer](#json-pointer)
+* [JSONPath](#jsonpath)
 * [Error Handling](#error-handling)
   * [Error Handling Example](#error-handling-example)
   * [Exceptions](#exceptions)
@@ -59,6 +61,26 @@ 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.
 
@@ -69,7 +91,7 @@ During the`load` or `parse` calls, neither the input file nor the input string a
 For best performance, a `parser` instance should be reused over several files: otherwise you will needlessly reallocate memory, an expensive process. It is also possible to avoid entirely memory allocations during parsing when using simdjson. [See our performance notes for details](performance.md).
 
 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).
+length of your input `l` in bytes.
 
 *Windows-specific*:  Windows users who need to read files with
 non-ANSI characters in the name should set their code page to
@@ -98,6 +120,12 @@ Once you have an element, you can navigate it with idiomatic C++ iterators, oper
   std::cout << "I parsed " << value << " from " << numberstring.data() << std::endl;
   ```
   The strings contain unescaped valid UTF-8 strings: no unmatched surrogate is allowed.
+  Internally, numbers are stored as either 64-bit integers or 64-bit floating-point numbers.
+  Thus it is possible to get the full 64-bit integer range (either signed or  unsigned).
+  By default, the string `-0` is parsed as the integer 0 as in Pytho or C++. If you set the macro
+  `SIMDJSON_MINUS_ZERO_AS_FLOAT` to `1` when building simdjson, you can get that `-0` is mapped to `-0.0`
+  as in JavaScript. You can get the desired effect by building simdjson with cmake setting the
+  `SIMDJSON_MINUS_ZERO_AS_FLOAT` to on: `cmake -B build -D SIMDJSON_MINUS_ZERO_AS_FLOAT=ON`.
 * **Field Access:** To get the value of the "foo" field in an object, use `object["foo"]`.
 * **Array Iteration:** To iterate through an array, use `for (auto value : array) { ... }`. If you
   know the type of the value, you can cast it right there, too! `for (double value : array) { ... }`
@@ -211,6 +239,23 @@ for (dom::key_value_pair field : object) {
 }
 ```
 
+C++20 Support
+------------
+
+simdjson library also supports some C++20 feature including `std::ranges`:
+
+```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;
+auto justmodel = [](auto car) { return car["model"]; };
+for (auto car : parser.parse(cars_json).get_array() | std::views::transform(justmodel)) {
+  std::cout << car << std::endl;
+}
+```
 
 JSON Pointer
 ------------
@@ -257,6 +302,49 @@ 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
@@ -669,5 +757,5 @@ Setting the `realloc_if_needed` parameter `false` in this manner may lead to bet
 Performance Tips
 ---------------------
 
-- For release builds, we recommend setting `NDEBUG` pre-processor directive when compiling the `simdjson` library. Importantly, using the optimization flags `-O2` or `-O3` under GCC and LLVM clang does not set the `NDEBUG` directrive, you must set it manually (e.g., `-DNDEBUG`).
+- For release builds, we recommend setting `NDEBUG` pre-processor directive when compiling the `simdjson` library. Importantly, using the optimization flags `-O2` or `-O3` under GCC and LLVM clang does not set the `NDEBUG` directive, you must set it manually (e.g., `-DNDEBUG`).
 - For long streams of JSON documents, consider [`iterate_many`](iterate_many.md) and [`parse_many`](parse_many.md) for better performance.

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 structure
+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

@@ -102,7 +102,7 @@ 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.

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

doc/performance.md

@@ -4,9 +4,9 @@ Performance Notes
 simdjson strives to be at its fastest *without tuning*, and generally achieves this. However, there
 are still some scenarios where tuning can enhance performance.
 Once your code is tested, we
-further encourage you to define `NDEBUG` in your Release builds to disable additional runtime
+further encourage you to define `NDEBUG` in your release builds to disable additional runtime
 testing and get the best performance.
-* [NDEBUG directive](#ndebug-directive)
+* [NDEBUG macro](#ndebug-macro)
 * [Reusing the parser for maximum efficiency](#reusing-the-parser-for-maximum-efficiency)
 * [Reusing string buffers](#reusing-string-buffers)
 * [Server Loops: Long-Running Processes and Memory Capacity](#server-loops-long-running-processes-and-memory-capacity)
@@ -17,16 +17,27 @@ testing and get the best performance.
 * [Free Padding](#free-padding)
 
 
-NDEBUG directive
+NDEBUG macro
 -------------
 
-In C/C++, the `NDEBUG` pre-processor directive is not set by default. When it is not set, the simdjson library does
+We recommend that you set `NDEBUG` macro in your release builds.
-many additional checks that may impact negatively the performance. We recommend that, once your code
-is well tested, you define `NDEBUG` directive in your Release builds. The `NDEBUG` directive should be defined
-prior to including the `simdjson.h` header.
 
-The `NDEBUG` directive is generally independent from optimization flags. For example, setting `-O3` under
+In C/C++, the `NDEBUG` macro is not set by default.
-GCC does not set the `NDEBUG` directive.
+When it is not set, the software may do many additional checks that may impact
+negatively the performance. We recommend that, once your code
+is well tested, you define `NDEBUG` directive in your release builds.
+
+The `NDEBUG` directive is generally independent from optimization flags.
+For example, setting `-O3` under GCC does not set the `NDEBUG` directive.
+However, tools like `CMake` automatically
+set `NDEBUG` for release builds.
+
+In the simdjson library, we check the `NDEBUG` macro as well as other
+macros to make performant release builds. However, the C++ standard
+does not provide a definitive approach to determine whether you are
+compiling for a release build. Thus we recommend that you follow
+the practice of setting the `NDEBUG` macro in release mode to make sure
+that you do not get undesirable expensive checks.
 
 Reusing the parser for maximum efficiency
 -----------------------------------------
@@ -158,7 +169,9 @@ 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).
 
@@ -241,8 +254,8 @@ long page_size() {
 // 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()) <
+  return ((reinterpret_cast<uintptr_t>(buf + len - 1) % page_size())
-          simdjson::SIMDJSON_PADDING);
+    + simdjson::SIMDJSON_PADDING > static_cast<uintptr_t>(page_size()));
 }
 
 simdjson::padded_string_view

examples/quickstart/CMakeLists.txt

@@ -20,10 +20,14 @@ IF(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
   endif()
 
   add_quickstart_test(quickstart_noexceptions   quickstart_noexceptions.cpp NO_EXCEPTIONS LABELS acceptance)
-  add_quickstart_test(quickstart_noexceptions11 quickstart_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11)
+  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)
-  add_quickstart_test(quickstart2_noexceptions11 quickstart2_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11)
+  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
   if (SIMDJSON_EXCEPTIONS)
@@ -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)
-  add_quickstart_test(quickstart_ondemand_noexceptions11 quickstart_ondemand_noexceptions.cpp NO_EXCEPTIONS CXX_STANDARD c++11 LABELS quickstart_ondemand)
+  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

@@ -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
@@ -134,6 +134,12 @@ namespace {
       tmp = vpaddq_u8(tmp, tmp);
       return vgetq_lane_u16(vreinterpretq_u16_u8(tmp), 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; }
   };
 
@@ -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/common_defs.h

@@ -23,7 +23,7 @@ double from_chars(const char *first, const char* end) noexcept;
 }
 
 #ifndef SIMDJSON_EXCEPTIONS
-#if __cpp_exceptions
+#if defined(__cpp_exceptions) || defined(_CPPUNWIND)
 #define SIMDJSON_EXCEPTIONS 1
 #else
 #define SIMDJSON_EXCEPTIONS 0
@@ -226,12 +226,14 @@ double from_chars(const char *first, const char* end) noexcept;
 // even if we do not have C++17 support.
 #ifdef __cpp_lib_string_view
 #define SIMDJSON_HAS_STRING_VIEW
+#include <string_view>
 #endif
 
 // Some systems have string_view even if we do not have C++17 support,
 // and even if __cpp_lib_string_view is undefined, it is the case
 // with Apple clang version 11.
 // We must handle it. *This is important.*
+#ifndef _MSC_VER
 #ifndef SIMDJSON_HAS_STRING_VIEW
 #if defined __has_include
 // do not combine the next #if with the previous one (unsafe)
@@ -247,6 +249,7 @@ double from_chars(const char *first, const char* end) noexcept;
 #endif // __has_include (<string_view>)
 #endif // defined __has_include
 #endif // def SIMDJSON_HAS_STRING_VIEW
+#endif // def _MSC_VER
 // end of complicated but important routine to try to detect string_view.
 
 //
@@ -279,16 +282,25 @@ namespace std {
 // It could also wrongly set SIMDJSON_DEVELOPMENT_CHECKS (e.g., if the programmer
 // sets _DEBUG in a release build under Visual Studio, or if some compiler fails to
 // set the __OPTIMIZE__ macro).
+// We make it so that if NDEBUG is defined, then SIMDJSON_DEVELOPMENT_CHECKS
+// is not defined, irrespective of the compiler.
+// We recommend that users set NDEBUG in release builds, so that
+// SIMDJSON_DEVELOPMENT_CHECKS is not defined in release builds by default,
+// irrespective of the compiler.
 #ifndef SIMDJSON_DEVELOPMENT_CHECKS
 #ifdef _MSC_VER
 // Visual Studio seems to set _DEBUG for debug builds.
-#ifdef _DEBUG
+// We set SIMDJSON_DEVELOPMENT_CHECKS to 1 if _DEBUG is defined
+// and NDEBUG is not defined.
+#if defined(_DEBUG) && !defined(NDEBUG)
 #define SIMDJSON_DEVELOPMENT_CHECKS 1
 #endif // _DEBUG
 #else // _MSC_VER
 // All other compilers appear to set __OPTIMIZE__ to a positive integer
 // when the compiler is optimizing.
-#ifndef __OPTIMIZE__
+// We only set SIMDJSON_DEVELOPMENT_CHECKS if both __OPTIMIZE__
+// and NDEBUG are not defined.
+#if !defined(__OPTIMIZE__) && !defined(NDEBUG)
 #define SIMDJSON_DEVELOPMENT_CHECKS 1
 #endif // __OPTIMIZE__
 #endif // _MSC_VER

include/simdjson/compiler_check.h

@@ -50,4 +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 +1,131 @@
+#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
+
+
+template <typename T>
+concept string_view_like = std::is_convertible_v<T, std::string_view> &&
+                           !std::is_convertible_v<T, const char*>;
+
+template<typename T>
+concept constructible_from_string_view = std::is_constructible_v<T, std::string_view>
+                                        && !std::is_same_v<T, std::string_view>
+                                        && std::is_default_constructible_v<T>;
+
+template<typename M>
+concept string_view_keyed_map = string_view_like<typename M::key_type>
+              && requires(std::remove_cvref_t<M>& m, typename M::key_type sv, typename M::mapped_type v) {
+    { m.emplace(sv, v) } -> std::same_as<std::pair<typename M::iterator, bool>>;
+};
+
+/// 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>) && !string_view_keyed_map<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;

include/simdjson/dom/array.h

@@ -57,7 +57,7 @@ public:
     iterator& operator=(const iterator&) noexcept = default;
   private:
     simdjson_inline iterator(const internal::tape_ref &tape) noexcept;
-    internal::tape_ref tape;
+    internal::tape_ref tape{};
     friend class array;
   };
 
@@ -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:
@@ -133,7 +148,7 @@ public:
 
 private:
   simdjson_inline array(const internal::tape_ref &tape) noexcept;
-  internal::tape_ref tape;
+  internal::tape_ref tape{};
   friend class element;
   friend struct simdjson_result<element>;
   template<typename T>
@@ -152,6 +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.h

@@ -70,7 +70,7 @@ public:
    * The memory allocation is strict: you
    * can you use this function to increase
    * or lower the amount of allocated memory.
-   * Passsing zero clears the memory.
+   * Passing zero clears the memory.
    */
   error_code allocate(size_t len) noexcept;
   /** @private Capacity in bytes, in terms

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 {
@@ -412,6 +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
   /**
    *
@@ -475,7 +492,7 @@ public:
 
 private:
   simdjson_inline element(const internal::tape_ref &tape) noexcept;
-  internal::tape_ref tape;
+  internal::tape_ref tape{};
   friend class document;
   friend class object;
   friend class array;
@@ -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) {

include/simdjson/dom/object.h

@@ -89,7 +89,7 @@ public:
   private:
     simdjson_inline iterator(const internal::tape_ref &tape) noexcept;
 
-    internal::tape_ref tape;
+    internal::tape_ref tape{};
 
     friend class object;
   };
@@ -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.
    *
@@ -208,7 +224,7 @@ public:
 private:
   simdjson_inline object(const internal::tape_ref &tape) noexcept;
 
-  internal::tape_ref tape;
+  internal::tape_ref tape{};
 
   friend class element;
   friend struct simdjson_result<element>;
@@ -243,7 +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.h

@@ -202,6 +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
@@ -324,7 +340,7 @@ public:
    * arrays or objects) MUST be separated with whitespace.
    *
    * The documents must not exceed batch_size bytes (by default 1MB) or they will fail to parse.
-   * Setting batch_size to excessively large or excesively small values may impact negatively the
+   * Setting batch_size to excessively large or excessively small values may impact negatively the
    * performance.
    *
    * ### Error Handling
@@ -418,7 +434,7 @@ public:
    * arrays or objects) MUST be separated with whitespace.
    *
    * The documents must not exceed batch_size bytes (by default 1MB) or they will fail to parse.
-   * Setting batch_size to excessively large or excesively small values may impact negatively the
+   * Setting batch_size to excessively large or excessively small values may impact negatively the
    * performance.
    *
    * ### Error Handling
@@ -549,9 +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;

include/simdjson/error-inl.h

@@ -6,6 +6,11 @@
 #include <iostream>
 
 namespace simdjson {
+
+inline bool is_fatal(error_code error) noexcept {
+  return error == TAPE_ERROR || error == INCOMPLETE_ARRAY_OR_OBJECT;
+}
+
 namespace internal {
   // We store the error code so we can validate the error message is associated with the right code
   struct error_code_info {
@@ -127,6 +132,23 @@ simdjson_warn_unused simdjson_inline error_code simdjson_result<T>::get(T &value
   return std::forward<internal::simdjson_result_base<T>>(*this).get(value);
 }
 
+template<typename T>
+simdjson_warn_unused simdjson_inline error_code
+simdjson_result<T>::get(std::string &value) && noexcept
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+requires (!std::is_same_v<T, std::string>)
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+{
+  // SFINAE : n'active que pour T = std::string_view
+  static_assert(std::is_same<T, std::string_view>::value, "simdjson_result<T>::get(std::string&) n'est disponible que pour T = std::string_view");
+  std::string_view v;
+  error_code error = std::forward<simdjson_result<T>>(*this).get(v);
+  if (!error) {
+    value.assign(v.data(), v.size());
+  }
+  return error;
+}
+
 template<typename T>
 simdjson_inline error_code simdjson_result<T>::error() const noexcept {
   return internal::simdjson_result_base<T>::error();

include/simdjson/error.h

@@ -20,7 +20,7 @@ enum error_code {
   SUCCESS = 0,                ///< No error
   CAPACITY,                   ///< This parser can't support a document that big
   MEMALLOC,                   ///< Error allocating memory, most likely out of memory
-  TAPE_ERROR,                 ///< Something went wrong, this is a generic error
+  TAPE_ERROR,                 ///< Something went wrong, this is a generic error. Fatal/unrecoverable error.
   DEPTH_ERROR,                ///< Your document exceeds the user-specified depth limitation
   STRING_ERROR,               ///< Problem while parsing a string
   T_ATOM_ERROR,               ///< Problem while parsing an atom starting with the letter 't'
@@ -45,13 +45,22 @@ enum error_code {
   PARSER_IN_USE,              ///< parser is already in use.
   OUT_OF_ORDER_ITERATION,     ///< tried to iterate an array or object out of order (checked when SIMDJSON_DEVELOPMENT_CHECKS=1)
   INSUFFICIENT_PADDING,       ///< The JSON doesn't have enough padding for simdjson to safely parse it.
-  INCOMPLETE_ARRAY_OR_OBJECT, ///< The document ends early.
+  INCOMPLETE_ARRAY_OR_OBJECT, ///< The document ends early. Fatal/unrecoverable error.
   SCALAR_DOCUMENT_AS_VALUE,   ///< A scalar document is treated as a value.
   OUT_OF_BOUNDS,              ///< Attempted to access location outside of document.
   TRAILING_CONTENT,           ///< Unexpected trailing content in the JSON input
-  NUM_ERROR_CODES
+  OUT_OF_CAPACITY,            ///< The capacity was exceeded, we cannot allocate enough memory.
+  NUM_ERROR_CODES             ///< Placeholder for end of error code list.
 };
 
+/**
+ * Some errors are fatal and invalidate the document. This function returns true if the
+ * error is fatal. It returns true for TAPE_ERROR and INCOMPLETE_ARRAY_OR_OBJECT.
+ * Once a fatal error is encountered, the on-demand document is no longer valid and
+ * processing should stop.
+ */
+ inline bool is_fatal(error_code error) noexcept;
+
 /**
  * It is the convention throughout the code that  the macro SIMDJSON_DEVELOPMENT_CHECKS determines whether
  * we check for OUT_OF_ORDER_ITERATION. The logic behind it is that these errors only occurs when the code
@@ -86,7 +95,7 @@ struct simdjson_error : public std::exception {
    */
   simdjson_error(error_code error) noexcept : _error{error} { }
   /** The error message */
-  const char *what() const noexcept { return error_message(error()); }
+  const char *what() const noexcept override { return error_message(error()); }
   /** The error code */
   error_code error() const noexcept { return _error; }
 private:
@@ -188,6 +197,7 @@ struct simdjson_result_base : protected std::pair<T, error_code> {
    * @throw simdjson_error if there was an error.
    */
   simdjson_inline operator T&&() && noexcept(false);
+
 #endif // SIMDJSON_EXCEPTIONS
 
   /**
@@ -244,7 +254,17 @@ struct simdjson_result : public internal::simdjson_result_base<T> {
    * @param value The variable to assign the value to. May not be set if there is an error.
    */
   simdjson_warn_unused simdjson_inline error_code get(T &value) && noexcept;
-
+//
+  /**
+   * Copy the value to a provided std::string, only enabled for std::string_view.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   */
+  simdjson_warn_unused simdjson_inline error_code get(std::string &value) && noexcept
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+  requires (!std::is_same_v<T, std::string>)
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+  ;
   /**
    * The error.
    */

include/simdjson/fallback/implementation.h

@@ -26,6 +26,7 @@ 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 fallback

include/simdjson/fallback/numberparsing_defs.h

@@ -75,6 +75,12 @@ simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t
 } // namespace fallback
 } // 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_FALLBACK_NUMBERPARSING_DEFS_H

include/simdjson/fallback/stringparsing_defs.h

@@ -29,6 +29,24 @@ simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uin
   return { src[0] };
 }
 
+
+struct escaping {
+  static constexpr uint32_t BYTES_PROCESSED = 1;
+  simdjson_inline static escaping copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_escape() { return escape_bits; }
+  simdjson_inline int escape_index() { return 0; }
+
+  bool escape_bits;
+}; // struct escaping
+
+
+
+simdjson_inline escaping escaping::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  dst[0] = src[0];
+  return { (src[0] == '\\') || (src[0] == '"') || (src[0] < 32) };
+}
+
 } // unnamed namespace
 } // namespace fallback
 } // namespace simdjson

include/simdjson/generic/dependencies.h

@@ -16,5 +16,4 @@
 #include "simdjson/internal/jsoncharutils_tables.h"
 #include "simdjson/internal/numberparsing_tables.h"
 #include "simdjson/internal/simdprune_tables.h"
-
 #endif // SIMDJSON_GENERIC_DEPENDENCIES_H

include/simdjson/generic/numberparsing.h

@@ -137,7 +137,7 @@ simdjson_inline bool compute_float_64(int64_t power, uint64_t i, bool negative,
   // floor(log(5**power)/log(2))
   //
   // Note that this is not magic: 152170/(1<<16) is
-  // approximatively equal to log(5)/log(2).
+  // approximately equal to log(5)/log(2).
   // The 1<<16 value is a power of two; we could use a
   // larger power of 2 if we wanted to.
   //
@@ -574,7 +574,6 @@ simdjson_unused simdjson_inline simdjson_result<number_type> get_number_type(con
 // Our objective is accurate parsing (ULP of 0) at high speed.
 template<typename W>
 simdjson_inline error_code parse_number(const uint8_t *const src, W &writer) {
-
   //
   // Check for minus sign
   //
@@ -648,7 +647,16 @@ simdjson_inline error_code parse_number(const uint8_t *const src, W &writer) {
   if (i > uint64_t(INT64_MAX)) {
     WRITE_UNSIGNED(i, src, writer);
   } else {
-    WRITE_INTEGER(negative ? (~i+1) : i, src, writer);
+#if SIMDJSON_MINUS_ZERO_AS_FLOAT
+    if(i == 0 && negative) {
+      // We have to write -0.0 instead of 0
+      WRITE_DOUBLE(-0.0, src, writer);
+    } else {
+      WRITE_INTEGER(negative ? (~i+1) : i, src, writer);
+    }
+#else
+  WRITE_INTEGER(negative ? (~i+1) : i, src, writer);
+#endif
   }
   if (jsoncharutils::is_not_structural_or_whitespace(*p)) { return INVALID_NUMBER(src); }
   return SUCCESS;
@@ -1109,6 +1117,12 @@ simdjson_unused simdjson_inline simdjson_result<number_type> get_number_type(con
       if (simdjson_unlikely(digit_count == 19 && memcmp(src, smaller_big_integer, 19) > 0)) {
         return number_type::big_integer;
       }
+#if SIMDJSON_MINUS_ZERO_AS_FLOAT
+      if(digit_count == 1 && src[0] == '0') {
+        // We have to write -0.0 instead of 0
+        return number_type::floating_point_number;
+      }
+#endif
       return number_type::signed_integer;
     }
     // Let us check if we have a big integer (>=2**64).

include/simdjson/generic/ondemand/amalgamated.h

@@ -4,6 +4,7 @@
 
 // Stuff other things depend on
 #include "simdjson/generic/ondemand/base.h"
+#include "simdjson/generic/ondemand/deserialize.h"
 #include "simdjson/generic/ondemand/value_iterator.h"
 #include "simdjson/generic/ondemand/value.h"
 #include "simdjson/generic/ondemand/logger.h"
@@ -23,9 +24,13 @@
 #include "simdjson/generic/ondemand/object_iterator.h"
 #include "simdjson/generic/ondemand/serialization.h"
 
+// Deserialization for standard types
+#include "simdjson/generic/ondemand/std_deserialize.h"
+
 // Inline definitions
 #include "simdjson/generic/ondemand/array-inl.h"
 #include "simdjson/generic/ondemand/array_iterator-inl.h"
+#include "simdjson/generic/ondemand/value-inl.h"
 #include "simdjson/generic/ondemand/document-inl.h"
 #include "simdjson/generic/ondemand/document_stream-inl.h"
 #include "simdjson/generic/ondemand/field-inl.h"
@@ -38,5 +43,11 @@
 #include "simdjson/generic/ondemand/raw_json_string-inl.h"
 #include "simdjson/generic/ondemand/serialization-inl.h"
 #include "simdjson/generic/ondemand/token_iterator-inl.h"
-#include "simdjson/generic/ondemand/value-inl.h"
 #include "simdjson/generic/ondemand/value_iterator-inl.h"
+
+// JSON builder, ideally they should not be part of the ondemand directory
+// but it is convenient for now to have them here.
+#include "simdjson/generic/ondemand/json_string_builder.h"
+#include "simdjson/generic/ondemand/json_string_builder-inl.h"
+#include "simdjson/generic/ondemand/json_builder.h"
+

include/simdjson/generic/ondemand/array-inl.h

@@ -2,6 +2,7 @@
 
 #ifndef SIMDJSON_CONDITIONAL_INCLUDE
 #define SIMDJSON_GENERIC_ONDEMAND_ARRAY_INL_H
+#include "simdjson/jsonpathutil.h"
 #include "simdjson/generic/ondemand/base.h"
 #include "simdjson/generic/ondemand/array.h"
 #include "simdjson/generic/ondemand/array_iterator-inl.h"
@@ -163,53 +164,6 @@ inline simdjson_result<value> array::at_pointer(std::string_view json_pointer) n
   return child;
 }
 
-inline std::string json_path_to_pointer_conversion(std::string_view json_path) {
-  if (json_path.empty() || (json_path.front() != '.' &&
-      json_path.front() != '[')) {
-    return "-1"; // This is just a sentinel value, the caller should check for this and return an error.
-  }
-
-  std::string result;
-  // Reserve space to reduce allocations, adjusting for potential increases due
-  // to escaping.
-  result.reserve(json_path.size() * 2);
-
-  size_t i = 0;
-
-  while (i < json_path.length()) {
-    if (json_path[i] == '.') {
-      result += '/';
-    } else if (json_path[i] == '[') {
-      result += '/';
-      ++i; // Move past the '['
-      while (i < json_path.length() && json_path[i] != ']') {
-          if (json_path[i] == '~') {
-            result += "~0";
-          } else if (json_path[i] == '/') {
-            result += "~1";
-          } else {
-            result += json_path[i];
-          }
-          ++i;
-      }
-      if (i == json_path.length() || json_path[i] != ']') {
-          return "-1"; // Using sentinel value that will be handled as an error by the caller.
-      }
-    } else {
-      if (json_path[i] == '~') {
-          result += "~0";
-      } else if (json_path[i] == '/') {
-          result += "~1";
-      } else {
-          result += json_path[i];
-      }
-    }
-    ++i;
-  }
-
-  return result;
-}
-
 inline simdjson_result<value> array::at_path(std::string_view json_path) noexcept {
   auto json_pointer = json_path_to_pointer_conversion(json_path);
   if (json_pointer == "-1") { return INVALID_JSON_POINTER; }

include/simdjson/generic/ondemand/array.h

@@ -44,7 +44,8 @@ public:
    * calling this function, if successful, the array is 'rewinded' at its
    * beginning as if it had never been accessed. If the JSON is malformed (e.g.,
    * there is a missing comma), then an error is returned and it is no longer
-   * safe to continue.
+   * safe to continue. Note that count_elements() does not validate the JSON values,
+   * only the structure of the array.
    *
    * To check that an array is empty, it is more performant to use
    * the is_empty() method.
@@ -130,6 +131,37 @@ public:
    *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
    */
   simdjson_inline simdjson_result<value> at(size_t index) noexcept;
+
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+  /**
+   * Get this array as the given type.
+   *
+   * @param out This is set to a value of the given type, parsed from the JSON. If there is an error, this may not be initialized.
+   * @returns INCORRECT_TYPE If the JSON array is not of the given type.
+   * @returns SUCCESS If the parse succeeded and the out parameter was set to the value.
+   */
+  template <typename T>
+  simdjson_inline error_code get(T &out)
+     noexcept(custom_deserializable<T, array> ? nothrow_custom_deserializable<T, array> : true) {
+    static_assert(custom_deserializable<T, array>);
+    return deserialize(*this, out);
+  }
+  /**
+   * Get this array as the given type.
+   *
+   * @returns A value of the given type, parsed from the JSON.
+   * @returns INCORRECT_TYPE If the JSON value is not the given type.
+   */
+  template <typename T>
+  simdjson_inline simdjson_result<T> get()
+    noexcept(custom_deserializable<T, value> ? nothrow_custom_deserializable<T, value> : true)
+  {
+    static_assert(std::is_default_constructible<T>::value, "The specified type is not default constructible.");
+    T out{};
+    SIMDJSON_TRY(get<T>(out));
+    return out;
+  }
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
 protected:
   /**
    * Go to the end of the array, no matter where you are right now.
@@ -208,7 +240,28 @@ public:
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> at_pointer(std::string_view json_pointer) noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> at_path(std::string_view json_path) noexcept;
   simdjson_inline simdjson_result<std::string_view> raw_json() noexcept;
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+  // TODO: move this code into object-inl.h
 
+  template<typename T>
+  simdjson_inline simdjson_result<T> get() noexcept {
+    if (error()) { return error(); }
+    if constexpr (std::is_same_v<T, SIMDJSON_IMPLEMENTATION::ondemand::array>) {
+      return first;
+    }
+    return first.get<T>();
+  }
+  template<typename T>
+  simdjson_inline error_code get(T& out) noexcept {
+    if (error()) { return error(); }
+    if constexpr (std::is_same_v<T, SIMDJSON_IMPLEMENTATION::ondemand::array>) {
+      out = first;
+    } else {
+      SIMDJSON_TRY( first.get<T>(out) );
+    }
+    return SUCCESS;
+  }
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
 };
 
 } // namespace simdjson

include/simdjson/generic/ondemand/dependencies.h

@@ -8,10 +8,12 @@
 // Internal headers needed for ondemand generics.
 // All includes not under simdjson/generic/ondemand must be here!
 // Otherwise, amalgamation will fail.
+#include "simdjson/concepts.h"
 #include "simdjson/dom/base.h" // for MINIMAL_DOCUMENT_CAPACITY
 #include "simdjson/implementation.h"
 #include "simdjson/padded_string.h"
 #include "simdjson/padded_string_view.h"
 #include "simdjson/internal/dom_parser_implementation.h"
+#include "simdjson/jsonpathutil.h"
 
 #endif // SIMDJSON_GENERIC_ONDEMAND_DEPENDENCIES_H

include/simdjson/generic/ondemand/deserialize.h

@@ -0,0 +1,139 @@
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+
+#ifndef SIMDJSON_ONDEMAND_DESERIALIZE_H
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_ONDEMAND_DESERIALIZE_H
+#include "simdjson/generic/ondemand/base.h"
+#include "simdjson/generic/ondemand/array.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <concepts>
+namespace simdjson {
+
+namespace tag_invoke_fn_ns {
+void tag_invoke();
+
+struct tag_invoke_fn {
+  template <typename Tag, typename... Args>
+    requires requires(Tag tag, Args &&...args) {
+      tag_invoke(std::forward<Tag>(tag), std::forward<Args>(args)...);
+    }
+  constexpr auto operator()(Tag tag, Args &&...args) const
+      noexcept(noexcept(tag_invoke(std::forward<Tag>(tag),
+                                   std::forward<Args>(args)...)))
+          -> decltype(tag_invoke(std::forward<Tag>(tag),
+                                 std::forward<Args>(args)...)) {
+    return tag_invoke(std::forward<Tag>(tag), std::forward<Args>(args)...);
+  }
+};
+} // namespace tag_invoke_fn_ns
+
+inline namespace tag_invoke_ns {
+inline constexpr tag_invoke_fn_ns::tag_invoke_fn tag_invoke = {};
+} // namespace tag_invoke_ns
+
+template <typename Tag, typename... Args>
+concept tag_invocable = requires(Tag tag, Args... args) {
+  tag_invoke(std::forward<Tag>(tag), std::forward<Args>(args)...);
+};
+
+template <typename Tag, typename... Args>
+concept nothrow_tag_invocable =
+    tag_invocable<Tag, Args...> && requires(Tag tag, Args... args) {
+      {
+        tag_invoke(std::forward<Tag>(tag), std::forward<Args>(args)...)
+      } noexcept;
+    };
+
+template <typename Tag, typename... Args>
+using tag_invoke_result =
+    std::invoke_result<decltype(tag_invoke), Tag, Args...>;
+
+template <typename Tag, typename... Args>
+using tag_invoke_result_t =
+    std::invoke_result_t<decltype(tag_invoke), Tag, Args...>;
+
+template <auto &Tag> using tag_t = std::decay_t<decltype(Tag)>;
+
+
+struct deserialize_tag;
+
+/// These types are deserializable in a built-in way
+template <typename> struct is_builtin_deserializable : std::false_type {};
+template <> struct is_builtin_deserializable<int64_t> : std::true_type {};
+template <> struct is_builtin_deserializable<uint64_t> : std::true_type {};
+template <> struct is_builtin_deserializable<double> : std::true_type {};
+template <> struct is_builtin_deserializable<bool> : std::true_type {};
+template <> struct is_builtin_deserializable<SIMDJSON_IMPLEMENTATION::ondemand::array> : std::true_type {};
+template <> struct is_builtin_deserializable<SIMDJSON_IMPLEMENTATION::ondemand::object> : std::true_type {};
+template <> struct is_builtin_deserializable<SIMDJSON_IMPLEMENTATION::ondemand::value> : std::true_type {};
+template <> struct is_builtin_deserializable<SIMDJSON_IMPLEMENTATION::ondemand::raw_json_string> : std::true_type {};
+template <> struct is_builtin_deserializable<std::string_view> : std::true_type {};
+
+template <typename T>
+concept is_builtin_deserializable_v = is_builtin_deserializable<T>::value;
+
+template <typename T, typename ValT = SIMDJSON_IMPLEMENTATION::ondemand::value>
+concept custom_deserializable = tag_invocable<deserialize_tag, ValT&, T&>;
+
+template <typename T, typename ValT = SIMDJSON_IMPLEMENTATION::ondemand::value>
+concept deserializable = custom_deserializable<T, ValT> || is_builtin_deserializable_v<T>;
+
+template <typename T, typename ValT = SIMDJSON_IMPLEMENTATION::ondemand::value>
+concept nothrow_custom_deserializable = nothrow_tag_invocable<deserialize_tag, ValT&, T&>;
+
+// built-in types are noexcept and if an error happens, the value simply gets ignored and the error is returned.
+template <typename T, typename ValT = SIMDJSON_IMPLEMENTATION::ondemand::value>
+concept nothrow_deserializable = nothrow_custom_deserializable<T, ValT> || is_builtin_deserializable_v<T>;
+
+/// Deserialize Tag
+inline constexpr struct deserialize_tag {
+  using array_type = SIMDJSON_IMPLEMENTATION::ondemand::array;
+  using object_type = SIMDJSON_IMPLEMENTATION::ondemand::object;
+  using value_type = SIMDJSON_IMPLEMENTATION::ondemand::value;
+  using document_type = SIMDJSON_IMPLEMENTATION::ondemand::document;
+  using document_reference_type = SIMDJSON_IMPLEMENTATION::ondemand::document_reference;
+
+  // Customization Point for array
+  template <typename T>
+    requires custom_deserializable<T, value_type>
+  [[nodiscard]] constexpr /* error_code */ auto operator()(array_type &object, T& output) const noexcept(nothrow_custom_deserializable<T, value_type>) {
+    return tag_invoke(*this, object, output);
+  }
+
+  // Customization Point for object
+  template <typename T>
+    requires custom_deserializable<T, value_type>
+  [[nodiscard]] constexpr /* error_code */ auto operator()(object_type &object, T& output) const noexcept(nothrow_custom_deserializable<T, value_type>) {
+    return tag_invoke(*this, object, output);
+  }
+
+  // Customization Point for value
+  template <typename T>
+    requires custom_deserializable<T, value_type>
+  [[nodiscard]] constexpr /* error_code */ auto operator()(value_type &object, T& output) const noexcept(nothrow_custom_deserializable<T, value_type>) {
+    return tag_invoke(*this, object, output);
+  }
+
+  // Customization Point for document
+  template <typename T>
+    requires custom_deserializable<T, document_type>
+  [[nodiscard]] constexpr /* error_code */ auto operator()(document_type &object, T& output) const noexcept(nothrow_custom_deserializable<T, document_type>) {
+    return tag_invoke(*this, object, output);
+  }
+
+  // Customization Point for document reference
+  template <typename T>
+    requires custom_deserializable<T, document_reference_type>
+  [[nodiscard]] constexpr /* error_code */ auto operator()(document_reference_type &object, T& output) const noexcept(nothrow_custom_deserializable<T, document_reference_type>) {
+    return tag_invoke(*this, object, output);
+  }
+
+
+} deserialize{};
+
+} // namespace simdjson
+
+#endif // SIMDJSON_ONDEMAND_DESERIALIZE_H
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+

include/simdjson/generic/ondemand/document-inl.h

@@ -8,10 +8,12 @@
 #include "simdjson/generic/ondemand/json_type.h"
 #include "simdjson/generic/ondemand/raw_json_string.h"
 #include "simdjson/generic/ondemand/value.h"
+#include "simdjson/generic/ondemand/value-inl.h"
 #include "simdjson/generic/ondemand/array-inl.h"
 #include "simdjson/generic/ondemand/json_iterator-inl.h"
 #include "simdjson/generic/ondemand/object-inl.h"
 #include "simdjson/generic/ondemand/value_iterator-inl.h"
+#include "simdjson/generic/ondemand/deserialize.h"
 #endif // SIMDJSON_CONDITIONAL_INCLUDE
 
 namespace simdjson {
@@ -165,6 +167,16 @@ template<> simdjson_inline simdjson_result<int64_t> document::get() & noexcept {
 template<> simdjson_inline simdjson_result<bool> document::get() & noexcept { return get_bool(); }
 template<> simdjson_inline simdjson_result<value> document::get() & noexcept { return get_value(); }
 
+template<> simdjson_inline error_code document::get(array& out) & noexcept { return get_array().get(out); }
+template<> simdjson_inline error_code document::get(object& out) & noexcept { return get_object().get(out); }
+template<> simdjson_inline error_code document::get(raw_json_string& out) & noexcept { return get_raw_json_string().get(out); }
+template<> simdjson_inline error_code document::get(std::string_view& out) & noexcept { return get_string(false).get(out); }
+template<> simdjson_inline error_code document::get(double& out) & noexcept { return get_double().get(out); }
+template<> simdjson_inline error_code document::get(uint64_t& out) & noexcept { return get_uint64().get(out); }
+template<> simdjson_inline error_code document::get(int64_t& out) & noexcept { return get_int64().get(out); }
+template<> simdjson_inline error_code document::get(bool& out) & noexcept { return get_bool().get(out); }
+template<> simdjson_inline error_code document::get(value& out) & noexcept { return get_value().get(out); }
+
 template<> simdjson_deprecated simdjson_inline simdjson_result<raw_json_string> document::get() && noexcept { return get_raw_json_string(); }
 template<> simdjson_deprecated simdjson_inline simdjson_result<std::string_view> document::get() && noexcept { return get_string(false); }
 template<> simdjson_deprecated simdjson_inline simdjson_result<double> document::get() && noexcept { return std::forward<document>(*this).get_double(); }
@@ -173,13 +185,6 @@ template<> simdjson_deprecated simdjson_inline simdjson_result<int64_t> document
 template<> simdjson_deprecated simdjson_inline simdjson_result<bool> document::get() && noexcept { return std::forward<document>(*this).get_bool(); }
 template<> simdjson_deprecated simdjson_inline simdjson_result<value> document::get() && noexcept { return get_value(); }
 
-template<typename T> simdjson_inline error_code document::get(T &out) & noexcept {
-  return get<T>().get(out);
-}
-template<typename T> simdjson_deprecated simdjson_inline error_code document::get(T &out) && noexcept {
-  return std::forward<document>(*this).get<T>().get(out);
-}
-
 #if SIMDJSON_EXCEPTIONS
 template <class T>
 simdjson_deprecated simdjson_inline document::operator T() && noexcept(false) { return get<T>(); }
@@ -241,7 +246,14 @@ simdjson_inline simdjson_result<value> document::operator[](const char *key) & n
 }
 
 simdjson_inline error_code document::consume() noexcept {
-  auto error = iter.skip_child(0);
+  bool scalar = false;
+  auto error = is_scalar().get(scalar);
+  if(error) { return error; }
+  if(scalar) {
+    iter.return_current_and_advance();
+    return SUCCESS;
+  }
+  error = iter.skip_child(0);
   if(error) { iter.abandon(); }
   return error;
 }
@@ -263,6 +275,8 @@ simdjson_inline simdjson_result<json_type> document::type() noexcept {
 }
 
 simdjson_inline simdjson_result<bool> document::is_scalar() noexcept {
+  // For more speed, we could do:
+  // return iter.is_single_token();
   json_type this_type;
   auto error = type().get(this_type);
   if(error) { return error; }
@@ -818,6 +832,26 @@ simdjson_inline simdjson_result<bool> simdjson_result<SIMDJSON_IMPLEMENTATION::o
   if (error()) { return error(); }
   return first.is_null();
 }
+template<typename T>
+simdjson_inline simdjson_result<T> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get() & noexcept {
+  if (error()) { return error(); }
+  return first.get<T>();
+}
+template<typename T>
+simdjson_inline simdjson_result<T> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get() && noexcept {
+  if (error()) { return error(); }
+  return std::forward<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>(first).get<T>();
+}
+template <class T>
+simdjson_inline error_code simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get(T &out) & noexcept {
+  if (error()) { return error(); }
+  return first.get<T>(out);
+}
+template <class T>
+simdjson_inline error_code simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get(T &out) && noexcept {
+  if (error()) { return error(); }
+  return std::forward<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>(first).get<T>(out);
+}
 simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::json_type> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::type() noexcept {
   if (error()) { return error(); }
   return first.type();
@@ -830,6 +864,18 @@ simdjson_inline simdjson_result<bool> simdjson_result<SIMDJSON_IMPLEMENTATION::o
   if (error()) { return error(); }
   return first.is_string();
 }
+template <>
+simdjson_inline error_code simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get(SIMDJSON_IMPLEMENTATION::ondemand::document_reference &out) & noexcept {
+  if (error()) { return error(); }
+  out = first;
+  return SUCCESS;
+}
+template <>
+simdjson_inline error_code simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::get(SIMDJSON_IMPLEMENTATION::ondemand::document_reference &out) && noexcept {
+  if (error()) { return error(); }
+  out = first;
+  return SUCCESS;
+}
 simdjson_inline simdjson_result<bool> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::is_negative() noexcept {
   if (error()) { return error(); }
   return first.is_negative();
@@ -847,10 +893,12 @@ simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::number> simdj
   return first.get_number();
 }
 #if SIMDJSON_EXCEPTIONS
-template <class T, typename std::enable_if<std::is_same<T, SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::value == false>::type>
+template <class T>
 simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::operator T() noexcept(false) {
+  static_assert(std::is_same<T, SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::value == false, "You should not call get<T> when T is a document");
+  static_assert(std::is_same<T, SIMDJSON_IMPLEMENTATION::ondemand::document>::value == false, "You should not call get<T> when T is a document");
   if (error()) { throw simdjson_error(error()); }
-  return first;
+  return first.get<T>();
 }
 simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::operator SIMDJSON_IMPLEMENTATION::ondemand::array() & noexcept(false) {
   if (error()) { throw simdjson_error(error()); }

include/simdjson/generic/ondemand/document.h

@@ -4,8 +4,11 @@
 #define SIMDJSON_GENERIC_ONDEMAND_DOCUMENT_H
 #include "simdjson/generic/ondemand/base.h"
 #include "simdjson/generic/ondemand/json_iterator.h"
+#include "simdjson/generic/ondemand/deserialize.h"
+#include "simdjson/generic/ondemand/value.h"
 #endif // SIMDJSON_CONDITIONAL_INCLUDE
 
+
 namespace simdjson {
 namespace SIMDJSON_IMPLEMENTATION {
 namespace ondemand {
@@ -178,24 +181,39 @@ public:
    * @returns A value of the given type, parsed from the JSON.
    * @returns INCORRECT_TYPE If the JSON value is not the given type.
    */
-  template<typename T> simdjson_inline simdjson_result<T> get() & noexcept {
+  template <typename T>
-    // Unless the simdjson library or the user provides an inline implementation, calling this method should
+  simdjson_inline simdjson_result<T> get() &
-    // immediately fail.
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
-    static_assert(!sizeof(T), "The get method with given type is not implemented by the simdjson library. "
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document> : true)
-      "The supported types are ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+#else
-      "int64_t, double, and bool. We recommend you use get_double(), get_bool(), get_uint64(), get_int64(), "
+    noexcept
-      " get_object(), get_array(), get_raw_json_string(), or get_string() instead of the get template."
+#endif
-      " You may also add support for custom types, see our documentation.");
+  {
+    static_assert(std::is_default_constructible<T>::value, "Cannot initialize the specified type.");
+    T out{};
+    SIMDJSON_TRY(get<T>(out));
+    return out;
   }
-  /** @overload template<typename T> simdjson_result<T> get() & noexcept */
+  /**
-  template<typename T> simdjson_deprecated simdjson_inline simdjson_result<T> get() && noexcept {
+   * @overload template<typename T> simdjson_result<T> get() & noexcept
-    // Unless the simdjson library or the user provides an inline implementation, calling this method should
+   *
-    // immediately fail.
+   * We disallow the use tag_invoke CPO on a moved document; it may create UB
-    static_assert(!sizeof(T), "The get method with given type is not implemented by the simdjson library. "
+   * if user uses `ondemand::array` or `ondemand::object` in their custom type.
-      "The supported types are ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+   *
-      "int64_t, double, and bool. We recommend you use get_double(), get_bool(), get_uint64(), get_int64(), "
+   * The member function is still remains specialize-able for compatibility
-      " get_object(), get_array(), get_raw_json_string(), or get_string() instead of the get template."
+   * reasons, but we completely disallow its use when a tag_invoke customization
-      " You may also add support for custom types, see our documentation.");
+   * is provided.
+   */
+  template<typename T>
+  simdjson_inline simdjson_result<T> get() &&
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document> : true)
+#else
+    noexcept
+#endif
+  {
+      static_assert(!std::is_same<T, array>::value && !std::is_same<T, object>::value, "You should never hold either an ondemand::array or ondemand::object without a corresponding ondemand::document being alive; that would be Undefined Behaviour.");
+      return static_cast<document&>(*this).get<T>();
   }
 
   /**
@@ -206,10 +224,42 @@ public:
    * Be mindful that the document instance must remain in scope while you are accessing object, array and value instances.
    *
    * @param out This is set to a value of the given type, parsed from the JSON. If there is an error, this may not be initialized.
-   * @returns INCORRECT_TYPE If the JSON value is not an object.
+   * @returns INCORRECT_TYPE If the JSON value is of the given type.
    * @returns SUCCESS If the parse succeeded and the out parameter was set to the value.
    */
-  template<typename T> simdjson_inline error_code get(T &out) & noexcept;
+  template<typename T>
+  simdjson_inline error_code get(T &out) &
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document> : true)
+#else
+    noexcept
+#endif
+  {
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    if constexpr (custom_deserializable<T, document>) {
+        return deserialize(*this, out);
+    } else {
+      static_assert(!sizeof(T), "The get<T> method with type T is not implemented by the simdjson library. "
+        "And you do not seem to have added support for it. Indeed, we have that "
+        "simdjson::custom_deserializable<T> is false and the type T is not a default type "
+        "such as ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+        "int64_t, double, or bool.");
+      static_cast<void>(out); // to get rid of unused errors
+      return UNINITIALIZED;
+    }
+#else // SIMDJSON_SUPPORTS_DESERIALIZATION
+    // Unless the simdjson library or the user provides an inline implementation, calling this method should
+    // immediately fail.
+    static_assert(!sizeof(T), "The get method with given type is not implemented by the simdjson library. "
+      "The supported types are ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+      "int64_t, double, and bool. We recommend you use get_double(), get_bool(), get_uint64(), get_int64(), "
+      " get_object(), get_array(), get_raw_json_string(), or get_string() instead of the get template."
+      " You may also add support for custom types, see our documentation.");
+    static_cast<void>(out); // to get rid of unused errors
+    return UNINITIALIZED;
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+  }
+
   /** @overload template<typename T> error_code get(T &out) & noexcept */
   template<typename T> simdjson_deprecated simdjson_inline error_code get(T &out) && noexcept;
 
@@ -311,7 +361,8 @@ public:
    * calling this function, if successful, the array is 'rewinded' at its
    * beginning as if it had never been accessed. If the JSON is malformed (e.g.,
    * there is a missing comma), then an error is returned and it is no longer
-   * safe to continue.
+   * safe to continue. Note that count_elements() does not validate the JSON values,
+   * only the structure of the array.
    */
   simdjson_inline simdjson_result<size_t> count_elements() & noexcept;
    /**
@@ -423,6 +474,7 @@ public:
   simdjson_inline simdjson_result<value> operator[](std::string_view key) & noexcept;
   /** @overload simdjson_inline simdjson_result<value> find_field_unordered(std::string_view key) & noexcept; */
   simdjson_inline simdjson_result<value> operator[](const char *key) & noexcept;
+ simdjson_result<value> operator[](int) & noexcept = delete;
 
   /**
    * Get the type of this JSON value. It does not validate or consume the value.
@@ -691,6 +743,11 @@ protected:
 
 /**
  * A document_reference is a thin wrapper around a document reference instance.
+ * The document_reference instances are used primarily/solely for streams of JSON
+ * documents. They differ from document instances when parsing a scalar value
+ * (a document that is not an array or an object). In the case of a document,
+ * we expect the document to be fully consumed. In the case of a document_reference,
+ * we allow trailing content.
  */
 class document_reference {
 public:
@@ -716,7 +773,77 @@ public:
   simdjson_inline simdjson_result<value> get_value() noexcept;
 
   simdjson_inline simdjson_result<bool> is_null() noexcept;
-  template<typename T> simdjson_inline simdjson_result<T> get() & noexcept;
+  template <typename T>
+  simdjson_inline simdjson_result<T> get() &
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document> : true)
+#else
+    noexcept
+#endif
+  {
+    static_assert(std::is_default_constructible<T>::value, "Cannot initialize the specified type.");
+    T out{};
+    SIMDJSON_TRY(get<T>(out));
+    return out;
+  }
+  template<typename T>
+  simdjson_inline simdjson_result<T> get() &&
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document> : true)
+#else
+    noexcept
+#endif
+  {
+      static_assert(!std::is_same<T, array>::value && !std::is_same<T, object>::value, "You should never hold either an ondemand::array or ondemand::object without a corresponding ondemand::document_reference being alive; that would be Undefined Behaviour.");
+      return static_cast<document&>(*this).get<T>();
+  }
+
+  /**
+   * Get this value as the given type.
+   *
+   * Supported types: object, array, raw_json_string, string_view, uint64_t, int64_t, double, bool, value
+   *
+   * Be mindful that the document instance must remain in scope while you are accessing object, array and value instances.
+   *
+   * @param out This is set to a value of the given type, parsed from the JSON. If there is an error, this may not be initialized.
+   * @returns INCORRECT_TYPE If the JSON value is not an object.
+   * @returns SUCCESS If the parse succeeded and the out parameter was set to the value.
+   */
+  template<typename T>
+  simdjson_inline error_code get(T &out) &
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    noexcept(custom_deserializable<T, document> ? nothrow_custom_deserializable<T, document_reference> : true)
+#else
+    noexcept
+#endif
+  {
+#if SIMDJSON_SUPPORTS_DESERIALIZATION
+    if constexpr (custom_deserializable<T, document_reference>) {
+        return deserialize(*this, out);
+    } else {
+      static_assert(!sizeof(T), "The get<T> method with type T is not implemented by the simdjson library. "
+        "And you do not seem to have added support for it. Indeed, we have that "
+        "simdjson::custom_deserializable<T> is false and the type T is not a default type "
+        "such as ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+        "int64_t, double, or bool.");
+      static_cast<void>(out); // to get rid of unused errors
+      return UNINITIALIZED;
+    }
+#else // SIMDJSON_SUPPORTS_DESERIALIZATION
+    // Unless the simdjson library or the user provides an inline implementation, calling this method should
+    // immediately fail.
+    static_assert(!sizeof(T), "The get method with given type is not implemented by the simdjson library. "
+      "The supported types are ondemand::object, ondemand::array, raw_json_string, std::string_view, uint64_t, "
+      "int64_t, double, and bool. We recommend you use get_double(), get_bool(), get_uint64(), get_int64(), "
+      " get_object(), get_array(), get_raw_json_string(), or get_string() instead of the get template."
+      " You may also add support for custom types, see our documentation.");
+    static_cast<void>(out); // to get rid of unused errors
+    return UNINITIALIZED;
+#endif // SIMDJSON_SUPPORTS_DESERIALIZATION
+  }
+
+  /** @overload template<typename T> error_code get(T &out) & noexcept */
+  template<typename T> simdjson_inline error_code get(T &out) && noexcept;
   simdjson_inline simdjson_result<std::string_view> raw_json() noexcept;
   simdjson_inline operator document&() const noexcept;
 #if SIMDJSON_EXCEPTIONS
@@ -741,6 +868,7 @@ public:
   simdjson_inline simdjson_result<value> find_field(const char *key) & noexcept;
   simdjson_inline simdjson_result<value> operator[](std::string_view key) & noexcept;
   simdjson_inline simdjson_result<value> operator[](const char *key) & noexcept;
+  simdjson_result<value> operator[](int) & noexcept = delete;
   simdjson_inline simdjson_result<value> find_field_unordered(std::string_view key) & noexcept;
   simdjson_inline simdjson_result<value> find_field_unordered(const char *key) & noexcept;
 
@@ -819,6 +947,7 @@ public:
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field(const char *key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](std::string_view key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](const char *key) & noexcept;
+  simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](int) & noexcept = delete;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field_unordered(std::string_view key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field_unordered(const char *key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::json_type> type() noexcept;
@@ -868,8 +997,14 @@ public:
   simdjson_inline simdjson_result<bool> get_bool() noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> get_value() noexcept;
   simdjson_inline simdjson_result<bool> is_null() noexcept;
+
+  template<typename T> simdjson_inline simdjson_result<T> get() & noexcept;
+  template<typename T> simdjson_inline simdjson_result<T> get() && noexcept;
+
+  template<typename T> simdjson_inline error_code get(T &out) & noexcept;
+  template<typename T> simdjson_inline error_code get(T &out) && noexcept;
 #if SIMDJSON_EXCEPTIONS
-  template <class T, typename std::enable_if<std::is_same<T, SIMDJSON_IMPLEMENTATION::ondemand::document_reference>::value == false>::type>
+  template <class T>
   explicit simdjson_inline operator T() noexcept(false);
   simdjson_inline operator SIMDJSON_IMPLEMENTATION::ondemand::array() & noexcept(false);
   simdjson_inline operator SIMDJSON_IMPLEMENTATION::ondemand::object() & noexcept(false);
@@ -890,6 +1025,7 @@ public:
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field(const char *key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](std::string_view key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](const char *key) & noexcept;
+  simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> operator[](int) & noexcept = delete;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field_unordered(std::string_view key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> find_field_unordered(const char *key) & noexcept;
   simdjson_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::json_type> type() noexcept;

include/simdjson/generic/ondemand/document_stream-inl.h

@@ -152,7 +152,6 @@ simdjson_inline document_stream::iterator::iterator(document_stream* _stream, bo
 }
 
 simdjson_inline simdjson_result<ondemand::document_reference> document_stream::iterator::operator*() noexcept {
-  //if(stream->error) { return stream->error; }
   return simdjson_result<ondemand::document_reference>(stream->doc, stream->error);
 }
 

include/simdjson/generic/ondemand/json_builder.h

@@ -0,0 +1,170 @@
+/**
+ * This file is part of the builder API. It is temporarily in the ondemand directory
+ * but we will move it to a builder directory later.
+ */
+#ifndef SIMDJSON_GENERIC_BUILDER_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_STRING_BUILDER_H
+#include "simdjson/generic/builder/json_string_builder.h"
+#include "simdjson/concepts.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+#if SIMDJSON_STATIC_REFLECTION
+
+#include <charconv>
+#include <cstring>
+#include <experimental/meta>
+#include <string_view>
+#include <type_traits>
+#include <utility>
+// #include <static_reflection> // for std::define_static_string - header not available yet
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace builder {
+
+// Concept that checks if a type is a container but not a string (because
+// strings handling must be handled differently)
+template <typename T>
+concept container_but_not_string =
+    requires(T a) {
+      { a.size() } -> std::convertible_to<std::size_t>;
+      {
+        a[std::declval<std::size_t>()]
+      }; // check if elements are accessible for the subscript operator
+    } && !std::is_same_v<T, std::string> &&
+    !std::is_same_v<T, std::string_view> && !std::is_same_v<T, const char *>;
+
+template <class T>
+  requires(container_but_not_string<T>)
+constexpr void atom(string_builder &b, const T &t) {
+  if (t.size() == 0) {
+    b.append_raw("[]");
+    return;
+  }
+  b.append('[');
+  atom(b, t[0]);
+  for (size_t i = 1; i < t.size(); ++i) {
+    b.append(',');
+    atom(b, t[i]);
+  }
+  b.append(']');
+}
+
+template <class T>
+  requires(std::is_same_v<T, std::string> ||
+           std::is_same_v<T, std::string_view> ||
+           std::is_same_v<T, const char *> ||
+           std::is_same_v<T, char>)
+constexpr void atom(string_builder &b, const T &t) {
+  b.escape_and_append_with_quotes(t);
+}
+
+template <concepts::string_view_keyed_map T>
+constexpr void atom(string_builder &b, const T &m) {
+  if (m.empty()) {
+    b.append_raw("{}");
+    return;
+  }
+  b.append('{');
+  bool first = true;
+  for (const auto& [key, value] : m) {
+    if (!first) {
+      b.append(',');
+    }
+    first = false;
+    // Keys must be convertible to string_view per the concept
+    b.escape_and_append_with_quotes(key);
+    b.append(':');
+    atom(b, value);
+  }
+  b.append('}');
+}
+
+
+template<typename number_type,
+         typename = typename std::enable_if<std::is_arithmetic<number_type>::value && !std::is_same_v<number_type, char>>::type>
+constexpr void atom(string_builder &b, const number_type t) {
+  b.append(t);
+}
+#if SIMDJSON_CONSTEVAL
+consteval std::string consteval_to_quoted_escaped(std::string_view input);
+#endif
+
+template <class T>
+  requires(std::is_class_v<T> && !container_but_not_string<T> &&
+           !concepts::string_view_keyed_map<T> &&
+           !std::is_same_v<T, std::string> &&
+           !std::is_same_v<T, std::string_view>)
+constexpr void atom(string_builder &b, const T &t) {
+  int i = 0;
+  b.append('{');
+  [:expand(std::meta::nonstatic_data_members_of(^^T, std::meta::access_context::unchecked())):] >> [&]<auto dm>() {
+    if (i != 0)
+      b.append(',');
+    constexpr auto key = std::define_static_string(consteval_to_quoted_escaped(std::meta::identifier_of(dm)));
+    b.append_raw(key);
+    b.append(':');
+    atom(b, t.[:dm:]);
+    i++;
+  };
+  b.append('}');
+}
+
+// works for struct
+template <class Z> void append(string_builder &b, const Z &z) {
+  int i = 0;
+  b.append('{');
+  [:expand(std::meta::nonstatic_data_members_of(^^Z, std::meta::access_context::unchecked())):] >> [&]<auto dm>() {
+    if (i != 0)
+      b.append(',');
+    constexpr auto key = std::define_static_string(consteval_to_quoted_escaped(std::meta::identifier_of(dm)));
+    b.append_raw(key);
+    b.append(':');
+    atom(b, z.[:dm:]);
+    i++;
+  };
+  b.append('}');
+}
+
+// works for container
+template <class Z>
+  requires(container_but_not_string<Z>)
+void append(string_builder &b, const Z &z) {
+  if (z.size() == 0) {
+    b.append_raw("[]");
+    return;
+  }
+  b.append('[');
+  atom(b, z[0]);
+  for (size_t i = 1; i < z.size(); ++i) {
+    b.append(',');
+    atom(b, z[i]);
+  }
+  b.append(']');
+}
+
+template <class Z>
+simdjson_result<std::string> to_json_string(const Z &z) {
+  string_builder b;
+  append(b, z);
+  std::string_view s;
+  if(auto e = b.view().get(s); e) { return e; }
+  return std::string(s);
+}
+
+template <class Z>
+simdjson_error to_json(const Z &z, std::string &s) {
+  string_builder b;
+  append(b, z);
+  std::string_view view;
+  if(auto e = b.view().get(view); e) { return e; }
+  s.assign(view);
+  return SUCCESS;
+}
+} // namespace json_builder
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+#endif // SIMDJSON_STATIC_REFLECTION
+
+#endif

include/simdjson/generic/ondemand/json_string_builder-inl.h

@@ -0,0 +1,617 @@
+/**
+ * This file is part of the builder API. It is temporarily in the ondemand
+ * directory but we will move it to a builder directory later.
+ */
+#include <array>
+#include <cstring>
+#include <type_traits>
+#ifndef SIMDJSON_GENERIC_STRING_BUILDER_INL_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_STRING_BUILDER_INL_H
+#include "simdjson/generic/builder/json_string_builder.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+/*
+ * Empirically, we have found that an inlined optimization is important for
+ * performance. The following macros are not ideal. We should find a better
+ * way to inline the code.
+ */
+
+#if defined(__SSE2__) || defined(__x86_64__) || defined(__x86_64) ||           \
+    (defined(_M_AMD64) || defined(_M_X64) ||                                   \
+     (defined(_M_IX86_FP) && _M_IX86_FP == 2))
+#ifndef SIMDJSON_EXPERIMENTAL_HAS_SSE2
+#define SIMDJSON_EXPERIMENTAL_HAS_SSE2 1
+#endif
+#endif
+
+#if defined(__aarch64__) || defined(_M_ARM64)
+#ifndef SIMDJSON_EXPERIMENTAL_HAS_NEON
+#define SIMDJSON_EXPERIMENTAL_HAS_NEON 1
+#endif
+#endif
+#if SIMDJSON_EXPERIMENTAL_HAS_NEON
+#include <arm_neon.h>
+#endif
+#if SIMDJSON_EXPERIMENTAL_HAS_SSE2
+#include <emmintrin.h>
+#endif
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace builder {
+
+static SIMDJSON_CONSTEXPR_LAMBDA std::array<uint8_t, 256> json_quotable_character = {
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
+
+/**
+
+A possible SWAR implementation of has_json_escapable_byte. It is not used because
+it is slower than the current implementation. It is kept here for reference (to show
+that we tried it).
+
+inline bool has_json_escapable_byte(uint64_t x) {
+  uint64_t is_ascii = 0x8080808080808080ULL & ~x;
+  uint64_t xor2 = x ^ 0x0202020202020202ULL;
+  uint64_t lt32_or_eq34 = xor2 - 0x2121212121212121ULL;
+  uint64_t sub92 = x ^ 0x5C5C5C5C5C5C5C5CULL;
+  uint64_t eq92 = (sub92 - 0x0101010101010101ULL);
+  return ((lt32_or_eq34 | eq92) & is_ascii) != 0;
+}
+
+**/
+
+SIMDJSON_CONSTEXPR_LAMBDA simdjson_inline bool
+simple_needs_escaping(std::string_view v) {
+  for (char c : v) {
+    // a table lookup is faster than a series of comparisons
+    if(json_quotable_character[static_cast<uint8_t>(c)]) {
+      return true;
+    }
+  }
+  return false;
+}
+
+#if SIMDJSON_EXPERIMENTAL_HAS_NEON
+simdjson_inline bool fast_needs_escaping(std::string_view view) {
+  if (view.size() < 16) {
+    return simple_needs_escaping(view);
+  }
+  size_t i = 0;
+  uint8x16_t running = vdupq_n_u8(0);
+  uint8x16_t v34 = vdupq_n_u8(34);
+  uint8x16_t v92 = vdupq_n_u8(92);
+
+  for (; i + 15 < view.size(); i += 16) {
+    uint8x16_t word = vld1q_u8((const uint8_t *)view.data() + i);
+    running = vorrq_u8(running, vceqq_u8(word, v34));
+    running = vorrq_u8(running, vceqq_u8(word, v92));
+    running = vorrq_u8(running, vcltq_u8(word, vdupq_n_u8(32)));
+  }
+  if (i < view.size()) {
+    uint8x16_t word =
+        vld1q_u8((const uint8_t *)view.data() + view.length() - 16);
+    running = vorrq_u8(running, vceqq_u8(word, v34));
+    running = vorrq_u8(running, vceqq_u8(word, v92));
+    running = vorrq_u8(running, vcltq_u8(word, vdupq_n_u8(32)));
+  }
+  return vmaxvq_u32(vreinterpretq_u32_u8(running)) != 0;
+}
+#elif SIMDJSON_EXPERIMENTAL_HAS_SSE2
+simdjson_inline bool fast_needs_escaping(std::string_view view) {
+  if (view.size() < 16) {
+    return simple_needs_escaping(view);
+  }
+  size_t i = 0;
+  __m128i running = _mm_setzero_si128();
+  for (; i + 15 < view.size(); i += 16) {
+
+    __m128i word = _mm_loadu_si128(reinterpret_cast<const __m128i *>(view.data() + i));
+    running = _mm_or_si128(running, _mm_cmpeq_epi8(word, _mm_set1_epi8(34)));
+    running = _mm_or_si128(running, _mm_cmpeq_epi8(word, _mm_set1_epi8(92)));
+    running = _mm_or_si128(
+        running, _mm_cmpeq_epi8(_mm_subs_epu8(word, _mm_set1_epi8(31)),
+                                _mm_setzero_si128()));
+  }
+  if (i < view.size()) {
+    __m128i word =
+        _mm_loadu_si128(reinterpret_cast<const __m128i *>(view.data() + view.length() - 16));
+    running = _mm_or_si128(running, _mm_cmpeq_epi8(word, _mm_set1_epi8(34)));
+    running = _mm_or_si128(running, _mm_cmpeq_epi8(word, _mm_set1_epi8(92)));
+    running = _mm_or_si128(
+        running, _mm_cmpeq_epi8(_mm_subs_epu8(word, _mm_set1_epi8(31)),
+                                _mm_setzero_si128()));
+  }
+  return _mm_movemask_epi8(running) != 0;
+}
+#else
+simdjson_inline bool fast_needs_escaping(std::string_view view) {
+  return simple_needs_escaping(view);
+}
+#endif
+
+
+SIMDJSON_CONSTEXPR_LAMBDA inline size_t
+find_next_json_quotable_character(const std::string_view view,
+                                  size_t location) noexcept {
+
+  for (auto pos = view.begin() + location; pos != view.end(); ++pos) {
+    if (json_quotable_character[static_cast<uint8_t>(*pos)]) {
+      return pos - view.begin();
+    }
+  }
+  return size_t(view.size());
+}
+
+SIMDJSON_CONSTEXPR_LAMBDA static std::string_view control_chars[] = {
+    "\\x0000", "\\x0001", "\\x0002", "\\x0003", "\\x0004", "\\x0005", "\\x0006",
+    "\\x0007", "\\x0008", "\\t",     "\\n",     "\\x000b", "\\f",     "\\r",
+    "\\x000e", "\\x000f", "\\x0010", "\\x0011", "\\x0012", "\\x0013", "\\x0014",
+    "\\x0015", "\\x0016", "\\x0017", "\\x0018", "\\x0019", "\\x001a", "\\x001b",
+    "\\x001c", "\\x001d", "\\x001e", "\\x001f"};
+
+SIMDJSON_CONSTEXPR_LAMBDA void escape_json_char(char c, char *&out) {
+  if (c == '"') {
+    memcpy(out, "\\\"", 2);
+    out += 2;
+  } else if (c == '\\') {
+    memcpy(out, "\\\\", 2);
+    out += 2;
+  } else {
+    std::string_view v = control_chars[uint8_t(c)];
+    memcpy(out, v.data(), v.size());
+    out += v.size();
+  }
+}
+
+inline size_t write_string_escaped(const std::string_view input, char *out) {
+  size_t mysize = input.size();
+  if (!fast_needs_escaping(input)) { // fast path!
+    memcpy(out, input.data(), input.size());
+    return input.size();
+  }
+  const char *const initout = out;
+  size_t location = find_next_json_quotable_character(input, 0);
+  memcpy(out, input.data(), location);
+  out += location;
+  escape_json_char(input[location], out);
+  location += 1;
+  while (location < mysize) {
+    size_t newlocation = find_next_json_quotable_character(input, location);
+    memcpy(out, input.data() + location, newlocation - location);
+    out += newlocation - location;
+    location = newlocation;
+    if (location == mysize) {
+      break;
+    }
+    escape_json_char(input[location], out);
+    location += 1;
+  }
+  return out - initout;
+}
+
+#if SIMDJSON_CONSTEVAL
+// unoptimized, meant for compile-time execution
+consteval std::string consteval_to_quoted_escaped(std::string_view input) {
+  std::string out = "\"";
+  for (char c : input) {
+    if (json_quotable_character[uint8_t(c)]) {
+      if (c == '"') {
+        out.append("\\\"");
+      } else if (c == '\\') {
+        out.append("\\\\");
+      } else {
+        std::string_view v = control_chars[uint8_t(c)];
+        out.append(v);
+      }
+    } else {
+      out.push_back(c);
+    }
+  }
+  out.push_back('"');
+  return out;
+}
+#endif // SIMDJSON_CONSTEVAL
+
+simdjson_inline string_builder::string_builder(size_t initial_capacity)
+    : buffer(new(std::nothrow) char[initial_capacity]), position(0),
+      capacity(buffer.get() != nullptr ? initial_capacity : 0),
+      is_valid(buffer.get() != nullptr) {}
+
+simdjson_inline bool string_builder::capacity_check(size_t upcoming_bytes) {
+  // We use the convention that when is_valid is false, then the capacity and
+  // the position are 0.
+  // Most of the time, this function will return true.
+  if (simdjson_likely(upcoming_bytes <= capacity - position)) {
+    return true;
+  }
+  // check for overflow, most of the time there is no overflow
+  if (simdjson_likely(position + upcoming_bytes < position)) {
+    return false;
+  }
+  // We will rarely get here.
+  grow_buffer((std::max)(capacity * 2, position + upcoming_bytes));
+  // If the buffer allocation failed, we set is_valid to false.
+  return is_valid;
+}
+
+simdjson_inline void string_builder::grow_buffer(size_t desired_capacity) {
+  if (!is_valid) {
+    return;
+  }
+  std::unique_ptr<char[]> new_buffer(new (std::nothrow) char[desired_capacity]);
+  if (new_buffer.get() == nullptr) {
+    set_valid(false);
+    return;
+  }
+  std::memcpy(new_buffer.get(), buffer.get(), position);
+  buffer.swap(new_buffer);
+  capacity = desired_capacity;
+}
+
+simdjson_inline void string_builder::set_valid(bool valid) noexcept {
+  if (!valid) {
+    is_valid = false;
+    capacity = 0;
+    position = 0;
+    buffer.reset();
+  } else {
+    is_valid = true;
+  }
+}
+
+simdjson_inline size_t string_builder::size() const noexcept {
+  return position;
+}
+
+simdjson_inline void string_builder::append(char c) noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = c;
+  }
+}
+
+simdjson_inline void string_builder::append_null() noexcept {
+  constexpr char null_literal[] = "null";
+  constexpr size_t null_len = sizeof(null_literal) - 1;
+  if (capacity_check(null_len)) {
+    std::memcpy(buffer.get() + position, null_literal, null_len);
+    position += null_len;
+  }
+}
+
+simdjson_inline void string_builder::clear() noexcept {
+  position = 0;
+  // if it was invalid, we should try to repair it
+  if (!is_valid) {
+    capacity = 0;
+    buffer.reset();
+    is_valid = true;
+  }
+}
+
+namespace internal {
+
+// We could specialize further for 32-bit integers.
+int int_log2(uint32_t x) { return (63 - leading_zeroes(x | 1)); }
+
+int fast_digit_count(uint32_t x) {
+  static uint64_t table[] = {
+      4294967296,  8589934582,  8589934582,  8589934582,  12884901788,
+      12884901788, 12884901788, 17179868184, 17179868184, 17179868184,
+      21474826480, 21474826480, 21474826480, 21474826480, 25769703776,
+      25769703776, 25769703776, 30063771072, 30063771072, 30063771072,
+      34349738368, 34349738368, 34349738368, 34349738368, 38554705664,
+      38554705664, 38554705664, 41949672960, 41949672960, 41949672960,
+      42949672960, 42949672960};
+  return uint32_t((x + table[int_log2(x)]) >> 32);
+}
+
+int int_log2(uint64_t x) { return 63 - leading_zeroes(x | 1); }
+
+int fast_digit_count(uint64_t x) {
+  static uint64_t table[] = {9,
+                             99,
+                             999,
+                             9999,
+                             99999,
+                             999999,
+                             9999999,
+                             99999999,
+                             999999999,
+                             9999999999,
+                             99999999999,
+                             999999999999,
+                             9999999999999,
+                             99999999999999,
+                             999999999999999ULL,
+                             9999999999999999ULL,
+                             99999999999999999ULL,
+                             999999999999999999ULL,
+                             9999999999999999999ULL};
+  int y = (19 * int_log2(x) >> 6);
+  y += x > table[y];
+  return y + 1;
+}
+
+template <typename number_type, typename = typename std::enable_if<
+                                    std::is_unsigned<number_type>::value>::type>
+simdjson_inline size_t digit_count(number_type v) noexcept {
+  static_assert(sizeof(number_type) == 8 || sizeof(number_type) == 4 ||
+                    sizeof(number_type) == 2 || sizeof(number_type) == 1,
+                "We only support 8-bit, 16-bit, 32-bit and 64-bit numbers");
+  return fast_digit_count(v);
+}
+static const char decimal_table[200] = {
+  0x30, 0x30, 0x30, 0x31, 0x30, 0x32, 0x30, 0x33, 0x30, 0x34, 0x30, 0x35,
+  0x30, 0x36, 0x30, 0x37, 0x30, 0x38, 0x30, 0x39, 0x31, 0x30, 0x31, 0x31,
+  0x31, 0x32, 0x31, 0x33, 0x31, 0x34, 0x31, 0x35, 0x31, 0x36, 0x31, 0x37,
+  0x31, 0x38, 0x31, 0x39, 0x32, 0x30, 0x32, 0x31, 0x32, 0x32, 0x32, 0x33,
+  0x32, 0x34, 0x32, 0x35, 0x32, 0x36, 0x32, 0x37, 0x32, 0x38, 0x32, 0x39,
+  0x33, 0x30, 0x33, 0x31, 0x33, 0x32, 0x33, 0x33, 0x33, 0x34, 0x33, 0x35,
+  0x33, 0x36, 0x33, 0x37, 0x33, 0x38, 0x33, 0x39, 0x34, 0x30, 0x34, 0x31,
+  0x34, 0x32, 0x34, 0x33, 0x34, 0x34, 0x34, 0x35, 0x34, 0x36, 0x34, 0x37,
+  0x34, 0x38, 0x34, 0x39, 0x35, 0x30, 0x35, 0x31, 0x35, 0x32, 0x35, 0x33,
+  0x35, 0x34, 0x35, 0x35, 0x35, 0x36, 0x35, 0x37, 0x35, 0x38, 0x35, 0x39,
+  0x36, 0x30, 0x36, 0x31, 0x36, 0x32, 0x36, 0x33, 0x36, 0x34, 0x36, 0x35,
+  0x36, 0x36, 0x36, 0x37, 0x36, 0x38, 0x36, 0x39, 0x37, 0x30, 0x37, 0x31,
+  0x37, 0x32, 0x37, 0x33, 0x37, 0x34, 0x37, 0x35, 0x37, 0x36, 0x37, 0x37,
+  0x37, 0x38, 0x37, 0x39, 0x38, 0x30, 0x38, 0x31, 0x38, 0x32, 0x38, 0x33,
+  0x38, 0x34, 0x38, 0x35, 0x38, 0x36, 0x38, 0x37, 0x38, 0x38, 0x38, 0x39,
+  0x39, 0x30, 0x39, 0x31, 0x39, 0x32, 0x39, 0x33, 0x39, 0x34, 0x39, 0x35,
+  0x39, 0x36, 0x39, 0x37, 0x39, 0x38, 0x39, 0x39,
+};
+} // namespace internal
+
+template <typename number_type, typename>
+simdjson_inline void string_builder::append(number_type v) noexcept {
+  static_assert(std::is_same<number_type, bool>::value ||
+                    std::is_integral<number_type>::value ||
+                    std::is_floating_point<number_type>::value,
+                "Unsupported number type");
+  // If C++17 is available, we can 'if constexpr' here.
+  SIMDJSON_IF_CONSTEXPR(std::is_same<number_type, bool>::value) {
+    if (v) {
+      constexpr char true_literal[] = "true";
+      constexpr size_t true_len = sizeof(true_literal) - 1;
+      if (capacity_check(true_len)) {
+        std::memcpy(buffer.get() + position, true_literal, true_len);
+        position += true_len;
+      }
+    } else {
+      constexpr char false_literal[] = "false";
+      constexpr size_t false_len = sizeof(false_literal) - 1;
+      if (capacity_check(false_len)) {
+        std::memcpy(buffer.get() + position, false_literal, false_len);
+        position += false_len;
+      }
+    }
+  }
+  else SIMDJSON_IF_CONSTEXPR(std::is_unsigned<number_type>::value) {
+    constexpr size_t max_number_size = 20;
+    if (capacity_check(max_number_size)) {
+      using unsigned_type = typename std::make_unsigned<number_type>::type;
+      unsigned_type pv = static_cast<unsigned_type>(v);
+      size_t dc = internal::digit_count(pv);
+      char *write_pointer = buffer.get() + position + dc - 1;
+      while (pv >= 100) {
+        memcpy(write_pointer - 1, &internal::decimal_table[(pv % 100)*2], 2);
+        write_pointer -= 2;
+        pv /= 100;
+      }
+      if (pv >= 10) {
+        *write_pointer-- = char('0' + (pv % 10));
+        pv /= 10;
+      }
+      *write_pointer = char('0' + pv);
+      position += dc;
+    }
+  }
+  else SIMDJSON_IF_CONSTEXPR(std::is_integral<number_type>::value) {
+    constexpr size_t max_number_size = 20;
+    if (capacity_check(max_number_size)) {
+      using unsigned_type = typename std::make_unsigned<number_type>::type;
+      bool negative = v < 0;
+      unsigned_type pv = static_cast<unsigned_type>(v);
+      if (negative) {
+        pv = 0 - pv; // the 0 is for Microsoft
+      }
+      size_t dc = internal::digit_count(pv);
+      if (negative) {
+        buffer.get()[position++] = '-';
+      }
+      char *write_pointer = buffer.get() + position + dc - 1;
+      while (pv >= 100) {
+        memcpy(write_pointer - 1, &internal::decimal_table[(pv % 100)*2], 2);
+        write_pointer -= 2;
+        pv /= 100;
+      }
+      if (pv >= 10) {
+        *write_pointer-- = char('0' + (pv % 10));
+        pv /= 10;
+      }
+      *write_pointer = char('0' + pv);
+      position += dc;
+    }
+  }
+  else SIMDJSON_IF_CONSTEXPR(std::is_floating_point<number_type>::value) {
+    constexpr size_t max_number_size = 24;
+    if (capacity_check(max_number_size)) {
+      // We could specialize for float.
+      char *end = simdjson::internal::to_chars(buffer.get() + position, nullptr,
+                                               double(v));
+      position = end - buffer.get();
+    }
+  }
+}
+
+simdjson_inline void
+string_builder::escape_and_append(std::string_view input) noexcept {
+  // escaping might turn a control character into \x00xx so 6 characters.
+  if (capacity_check(6 * input.size())) {
+    position += write_string_escaped(input, buffer.get() + position);
+  }
+}
+
+simdjson_inline void
+string_builder::escape_and_append_with_quotes(std::string_view input) noexcept {
+  // escaping might turn a control character into \x00xx so 6 characters.
+  if (capacity_check(2 + 6 * input.size())) {
+    buffer.get()[position++] = '"';
+    position += write_string_escaped(input, buffer.get() + position);
+    buffer.get()[position++] = '"';
+  }
+}
+
+simdjson_inline void
+string_builder::escape_and_append_with_quotes(char input) noexcept {
+  // escaping might turn a control character into \x00xx so 6 characters.
+  if (capacity_check(2 + 6 * 1)) {
+    buffer.get()[position++] = '"';
+    std::string_view cinput(&input, 1);
+    position += write_string_escaped(cinput, buffer.get() + position);
+    buffer.get()[position++] = '"';
+  }
+}
+
+simdjson_inline void string_builder::escape_and_append_with_quotes(const char* input)  noexcept {
+  std::string_view cinput(input);
+  escape_and_append_with_quotes(cinput);
+}
+
+simdjson_inline void string_builder::append_raw(const char *c) noexcept {
+  size_t len = std::strlen(c);
+  append_raw(c, len);
+}
+
+simdjson_inline void
+string_builder::append_raw(std::string_view input) noexcept {
+  if (capacity_check(input.size())) {
+    std::memcpy(buffer.get() + position, input.data(), input.size());
+    position += input.size();
+  }
+}
+
+simdjson_inline void string_builder::append_raw(const char *str,
+                                                size_t len) noexcept {
+  if (capacity_check(len)) {
+    std::memcpy(buffer.get() + position, str, len);
+    position += len;
+  }
+}
+
+#if SIMDJSON_EXCEPTIONS
+simdjson_inline string_builder::operator std::string() const noexcept(false) {
+  return std::string(std::string_view());
+}
+
+simdjson_inline string_builder::operator std::string_view() const
+    noexcept(false) {
+  return view();
+}
+#endif
+
+simdjson_inline simdjson_result<std::string_view>
+string_builder::view() const noexcept {
+  if (!is_valid) {
+    return simdjson::OUT_OF_CAPACITY;
+  }
+  return std::string_view(buffer.get(), position);
+}
+
+simdjson_inline simdjson_result<const char *> string_builder::c_str() noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position] = '\0';
+    return buffer.get();
+  }
+  return simdjson::OUT_OF_CAPACITY;
+}
+
+simdjson_inline bool string_builder::validate_unicode() const noexcept {
+  return simdjson::validate_utf8(buffer.get(), position);
+}
+
+simdjson_inline void string_builder::start_object()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = '{';
+  }
+}
+
+
+simdjson_inline void string_builder::end_object()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = '}';
+  }
+}
+
+
+simdjson_inline void string_builder::start_array()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = '[';
+  }
+}
+
+
+simdjson_inline void string_builder::end_array()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = ']';
+  }
+}
+
+
+simdjson_inline void string_builder::append_comma()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = ',';
+  }
+}
+
+
+simdjson_inline void string_builder::append_colon()  noexcept {
+  if (capacity_check(1)) {
+    buffer.get()[position++] = ':';
+  }
+}
+
+template<typename key_type, typename value_type>
+simdjson_inline void string_builder::append_key_value(key_type key, value_type value) noexcept {
+  static_assert(
+    std::is_arithmetic<value_type>::value ||
+    std::is_same<value_type, char>::value ||
+    std::is_same<value_type, const char*>::value ||
+    std::is_convertible<value_type, std::string_view>::value ||
+    std::is_same<value_type, std::nullptr_t>::value,
+    "Unsupported value type");
+  static_assert(
+    std::is_same<key_type, const char*>::value ||
+    std::is_convertible<key_type, std::string_view>::value,
+    "Unsupported key type");
+  escape_and_append_with_quotes(key);
+  append_colon();
+  SIMDJSON_IF_CONSTEXPR(std::is_same<value_type, std::nullptr_t>::value) {
+    append_null();
+  } else SIMDJSON_IF_CONSTEXPR(std::is_same<value_type, char>::value) {
+    escape_and_append_with_quotes(value);
+  } else SIMDJSON_IF_CONSTEXPR(std::is_convertible<value_type, std::string_view>::value) {
+    escape_and_append_with_quotes(value);
+  } else SIMDJSON_IF_CONSTEXPR(std::is_same<value_type, const char*>::value) {
+    escape_and_append_with_quotes(value);
+  } else {
+    append(value);
+  }
+}
+
+} // namespace builder
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_STRING_BUILDER_INL_H

include/simdjson/generic/ondemand/json_string_builder.h

@@ -0,0 +1,215 @@
+/**
+ * This file is part of the builder API. It is temporarily in the ondemand directory
+ * but we will move it to a builder directory later.
+ */
+#ifndef SIMDJSON_GENERIC_STRING_BUILDER_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_STRING_BUILDER_H
+#include "simdjson/generic/implementation_simdjson_result_base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace builder {
+
+/**
+ * A builder for JSON strings representing documents. This is a low-level
+ * builder that is not meant to be used directly by end-users. Though it
+ * supports atomic types (Booleans, strings), it does not support composed
+ * types (arrays and objects).
+ *
+ * Ultimately, this class should support kernel-specific optimizations. E.g.,
+ * it may make use of SIMD instructions to escape strings faster.
+ */
+class string_builder {
+public:
+  simdjson_inline string_builder(size_t initial_capacity = 1024);
+
+  /**
+   * Append number (includes Booleans). Booleans are mapped to the strings
+   * false and true. Numbers are converted to strings abiding by the JSON standard.
+   * Floating-point numbers are converted to the shortest string that 'correctly'
+   * represents the number.
+   */
+  template<typename number_type,
+         typename = typename std::enable_if<std::is_arithmetic<number_type>::value>::type>
+  simdjson_inline void append(number_type v) noexcept;
+
+  /**
+   * Append character c.
+   */
+  simdjson_inline void append(char c)  noexcept;
+
+  /**
+   * Append the string 'null'.
+   */
+  simdjson_inline void append_null()  noexcept;
+
+  /**
+   * Clear the content.
+   */
+  simdjson_inline void clear()  noexcept;
+
+  /**
+   * Append the std::string_view, after escaping it.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void escape_and_append(std::string_view input)  noexcept;
+
+  /**
+   * Append the std::string_view surrounded by double quotes, after escaping it.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void escape_and_append_with_quotes(std::string_view input)  noexcept;
+
+  /**
+   * Append the character surrounded by double quotes, after escaping it.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void escape_and_append_with_quotes(char input)  noexcept;
+
+  /**
+   * Append the character surrounded by double quotes, after escaping it.
+   * There is no UTF-8 validation.
+   */
+   simdjson_inline void escape_and_append_with_quotes(const char* input)  noexcept;
+
+  /**
+   * Append the C string directly, without escaping.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void append_raw(const char *c)  noexcept;
+
+  /**
+   * Append "{" to the buffer.
+   */
+  simdjson_inline void start_object()  noexcept;
+
+  /**
+   * Append "}" to the buffer.
+   */
+  simdjson_inline void end_object()  noexcept;
+
+  /**
+   * Append "[" to the buffer.
+   */
+  simdjson_inline void start_array()  noexcept;
+
+  /**
+   * Append "]" to the buffer.
+   */
+  simdjson_inline void end_array()  noexcept;
+
+  /**
+   * Append "," to the buffer.
+   */
+  simdjson_inline void append_comma()  noexcept;
+
+  /**
+   * Append ":" to the buffer.
+   */
+  simdjson_inline void append_colon()  noexcept;
+
+  /**
+   * Append a key-value pair to the buffer.
+   * The key is escaped and surrounded by double quotes.
+   * The value is escaped if it is a string.
+   */
+   template<typename key_type, typename value_type>
+  simdjson_inline void append_key_value(key_type key, value_type value) noexcept;
+  /**
+   * Append the std::string_view directly, without escaping.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void append_raw(std::string_view input)  noexcept;
+
+  /**
+   * Append len characters from str.
+   * There is no UTF-8 validation.
+   */
+  simdjson_inline void append_raw(const char *str, size_t len) noexcept;
+#if SIMDJSON_EXCEPTIONS
+  /**
+   * Creates an std::string from the written JSON buffer.
+   * Throws if memory allocation failed
+   *
+   * The result may not be valid UTF-8 if some of your content was not valid UTF-8.
+   * Use validate_unicode() to check the content if needed.
+   */
+  simdjson_inline operator std::string() const noexcept(false);
+
+  /**
+   * Creates an std::string_view from the written JSON buffer.
+   * Throws if memory allocation failed.
+   *
+   * The result may not be valid UTF-8 if some of your content was not valid UTF-8.
+   * Use validate_unicode() to check the content if needed.
+   */
+  simdjson_inline operator std::string_view() const noexcept(false);
+#endif
+
+  /**
+   * Returns a view on the written JSON buffer. Returns an error
+   * if memory allocation failed.
+   *
+   * The result may not be valid UTF-8 if some of your content was not valid UTF-8.
+   * Use validate_unicode() to check the content.
+   */
+  simdjson_inline simdjson_result<std::string_view> view() const noexcept;
+
+  /**
+   * Appends the null character to the buffer and returns
+   * a pointer to the beginning of the written JSON buffer.
+   * Returns an error if memory allocation failed.
+   * The result is null-terminated.
+   *
+   * The result may not be valid UTF-8 if some of your content was not valid UTF-8.
+   * Use validate_unicode() to check the content.
+   */
+  simdjson_inline simdjson_result<const char *> c_str() noexcept;
+
+  /**
+   * Return true if the content is valid UTF-8.
+   */
+  simdjson_inline bool validate_unicode() const noexcept;
+
+  /**
+   * Returns the current size of the written JSON buffer.
+   * If an error occurred, returns 0.
+   */
+  simdjson_inline size_t size() const noexcept;
+
+private:
+  /**
+   * Returns true if we can write at least upcoming_bytes bytes.
+   * The underlying buffer is reallocated if needed. It is designed
+   * to be called before writing to the buffer. It should be fast.
+   */
+  simdjson_inline bool capacity_check(size_t upcoming_bytes);
+
+  /**
+   * Grow the buffer to at least desired_capacity bytes.
+   * If the allocation fails, is_valid is set to false. We expect
+   * that this function would not be repeatedly called.
+   */
+  simdjson_inline void grow_buffer(size_t desired_capacity);
+
+  /**
+   * We use this helper function to make sure that is_valid is kept consistent.
+   */
+  simdjson_inline void set_valid(bool valid) noexcept;
+
+  std::unique_ptr<char[]> buffer{};
+  size_t position{0};
+  size_t capacity{0};
+  bool is_valid{true};
+};
+
+
+
+}
+}
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_STRING_BUILDER_H

include/simdjson/generic/ondemand/object-inl.h

@@ -271,6 +271,7 @@ simdjson_inline  simdjson_result<std::string_view> simdjson_result<SIMDJSON_IMPL
   if (error()) { return error(); }
   return first.raw_json();
 }
+
 } // namespace simdjson
 
 #endif // SIMDJSON_GENERIC_ONDEMAND_OBJECT_INL_H