fix

* properly gitignore Visual Studio artifacts

* remove cmake_policy

.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

.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-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

.gitignore

@@ -38,7 +38,7 @@ cmake-build-release/
 .history/
 
 # Visual Studio artifacts
-/VS/
+/.vs/
 
 # C/C++ build outputs
 .build/
@@ -106,4 +106,4 @@ objs
 !.vscode/extensions.json
 
 # clangd
-.cache
+.cache

CMakeLists.txt

@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.14)
 project(
     simdjson
     # The version number is modified by tools/release.py
-    VERSION 3.12.2
+    VERSION 4.0.0
     DESCRIPTION "Parsing gigabytes of JSON per second"
     HOMEPAGE_URL "https://simdjson.org/"
     LANGUAGES CXX C
@@ -20,8 +20,8 @@ string(
 # ---- Options, variables ----
 
 # These version numbers are modified by tools/release.py
-set(SIMDJSON_LIB_VERSION "25.0.0" CACHE STRING "simdjson library version")
-set(SIMDJSON_LIB_SOVERSION "25" CACHE STRING "simdjson library soversion")
+set(SIMDJSON_LIB_VERSION "28.0.0" CACHE STRING "simdjson library version")
+set(SIMDJSON_LIB_SOVERSION "28" 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)
@@ -98,7 +98,7 @@ 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
@@ -134,6 +134,12 @@ 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)

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         = "3.12.2"
+PROJECT_NUMBER         = "4.0.0"
 
 # 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

README.md

@@ -63,6 +63,7 @@ Real-world usage
 - [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.
@@ -174,7 +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.

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/large_random/simdjson_ondemand.h

@@ -32,12 +32,11 @@ struct simdjson_ondemand_static_reflect {
 
   bool run(simdjson::padded_string &json, std::vector<point> &result) {
     auto doc = parser.iterate(json);
-    //ondemand::array arr = doc.get_array();
-    //arr.get<std::vector<point>>(result);
+    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>());
-    }
+    //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/static_reflect/CMakeLists.txt

@@ -39,6 +39,6 @@ else()
   endif()
 endif()
 
-
-
-add_subdirectory(twitter_data)
+# Add the benchmark executable targets
+add_subdirectory(twitter_benchmark)
+add_subdirectory(citm_catalog_benchmark)

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/lib.rs

@@ -1,11 +1,11 @@
 extern crate serde;
 extern crate serde_json;
 extern crate libc;
-use std::slice;
 
 use libc::{c_char, size_t};
 use serde::{Serialize, Deserialize};
-
+use std::{collections::HashMap, ffi::CString, ptr, slice};
+use serde::de::{self, Deserializer};
 /******************************************************/
 /******************************************************/
 /**
@@ -168,3 +168,315 @@ pub unsafe extern "C" fn free_twitter(raw: *mut TwitterData) {
 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

@@ -1,7 +1,7 @@
 #ifndef serde_benchmark_ffi_h
 #define serde_benchmark_ffi_h
 
-/* Generated with cbindgen:0.24.3 */
+/* Generated with cbindgen:0.28.0 */
 
 /* Warning, this file is autogenerated by cbindgen. Don't modify this manually. */
 
@@ -13,6 +13,8 @@
 
 namespace serde_benchmark {
 
+struct CitmCatalog;
+
 struct TwitterData;
 
 extern "C" {
@@ -25,8 +27,19 @@ void free_twitter(TwitterData *raw);
 
 void free_string(const char *ptr);
 
-} // extern "C"
+/// Creates a CitmCatalog from a JSON string (UTF-8 encoded).
+CitmCatalog *citm_from_str(const char *raw_input, uintptr_t raw_input_length);
 
-} // namespace serde_benchmark
+/// Serializes a CitmCatalog into a JSON string (UTF-8).
+char *str_from_citm(CitmCatalog *raw_catalog);
 
-#endif // serde_benchmark_ffi_h
+/// 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

@@ -8,7 +8,6 @@ add_executable(benchmark_serialization_twitter benchmark_serialization_twitter.c
     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)
 

benchmark/static_reflect/twitter_benchmark/benchmark_serialization_twitter.cpp

@@ -9,7 +9,7 @@
 #include <string>
 #include "twitter_data.h"
 #include "nlohmann_twitter_data.h"
-#include "benchmark_helper.h"
+#include "../benchmark_utils/benchmark_helper.h"
 #if SIMDJSON_BENCH_CPP_REFLECT
 #include <rfl.hpp>
 #include <rfl/json.hpp>
@@ -109,7 +109,25 @@ std::string read_file(std::string filename) {
   return out;
 }
 
-int main() {
+// 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);
 
@@ -127,16 +145,24 @@ int main() {
   }
 
   // Benchmarking the serialization
-  bench_nlohmann(my_struct);
-  bench_simdjson_static_reflection(my_struct);
+  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
-  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);
+  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
-  bench_reflect_cpp(my_struct);
+  if (matches_filter("reflect_cpp", filter)) {
+    bench_reflect_cpp(my_struct);
+  }
 #endif
   return EXIT_SUCCESS;
 }

benchmark/static_reflect/twitter_benchmark/nlohmann_twitter_data.h

@@ -105,7 +105,6 @@ std::string nlohmann_serialize(const std::vector<Status>& v) {
   return a.dump();
 }
 
-
 void to_json(nlohmann::json &j, const TwitterData &t) {
   j = nlohmann::json{{"statuses", t.statuses}};
 }

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(
@@ -164,9 +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

@@ -7,12 +7,12 @@ 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)
   CPMAddPackage(
         NAME google_benchmarks
-        URL https://github.com/google/benchmark/archive/refs/tags/v1.7.1.zip
+        URL https://github.com/google/benchmark/archive/refs/tags/v1.9.4.zip
         OPTIONS
         "BENCHMARK_ENABLE_TESTING OFF"
         "BENCHMARK_ENABLE_INSTALL OFF"
@@ -22,7 +22,7 @@ endif()
 
 CPMAddPackage(
         NAME simdjson-data
-        URL https://github.com/simdjson/simdjson-data/archive/a5b13babe65c1bba7186b41b43d4cbdc20a5c470.zip
+        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)
@@ -93,19 +93,11 @@ int main() {}
 
   CPMAddPackage(
       NAME nlohmann_json
-      URL https://github.com/nlohmann/json/archive/refs/tags/v3.10.5.zip
+      URL https://github.com/nlohmann/json/archive/refs/tags/v3.12.0.zip
   )
 
   set_property(TARGET nlohmann_json APPEND PROPERTY INTERFACE_COMPILE_DEFINITIONS SIMDJSON_COMPETITION_NLOHMANN_JSON)
 
-  CPMAddPackage(
-      NAME json11
-      URL https://github.com/dropbox/json11/archive/ec4e45219af1d7cde3d58b49ed762376fccf1ace.zip
-      DOWNLOAD_ONLY YES
-  )
-  add_library(json11 STATIC "${json11_SOURCE_DIR}/json11.cpp")
-  target_include_directories(json11 SYSTEM PUBLIC "${json11_SOURCE_DIR}")
-  target_compile_definitions(json11 INTERFACE SIMDJSON_COMPETITION_JSON11)
 
   set(jsoncpp_SOURCE_DIR "${simdjson_SOURCE_DIR}/dependencies/jsoncppdist")
   add_library(jsoncpp STATIC "${jsoncpp_SOURCE_DIR}/jsoncpp.cpp")
@@ -177,7 +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)

doc/basics.md

@@ -1,59 +1,64 @@
 The Basics
 ==========
 
-An overview of what you need to know to use simdjson, with examples.
+An overview of what you need to know to use simdjson to parse JSON documents, with examples.
+[Our documentation regarding the generation (serialization) of JSON documents is in a
+separate document](https://github.com/simdjson/simdjson/blob/master/doc/builder.md).
 
 - [The Basics](#the-basics)
-  - [Requirements](#requirements)
-  - [Including simdjson](#including-simdjson)
-  - [Using simdjson with package managers](#using-simdjson-with-package-managers)
-  - [Using simdjson as a CMake dependency](#using-simdjson-as-a-cmake-dependency)
-  - [Versions](#versions)
-  - [The basics: loading and parsing JSON documents](#the-basics-loading-and-parsing-json-documents)
-  - [Documents are iterators](#documents-are-iterators)
-    - [Parser, document and JSON scope](#parser-document-and-json-scope)
-  - [string_view](#string_view)
-  - [Avoiding pitfalls: enable development checks](#avoiding-pitfalls-enable-development-checks)
-  - [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)
-  - [JSONPath](#jsonpath)
-  - [Error handling](#error-handling)
-    - [Error handling examples without exceptions](#error-handling-examples-without-exceptions)
-    - [Disabling exceptions](#disabling-exceptions)
-    - [Exceptions](#exceptions)
-    - [Current location in document](#current-location-in-document)
-    - [Checking for trailing content](#checking-for-trailing-content)
-  - [Rewinding](#rewinding)
-  - [Newline-Delimited JSON (ndjson) and JSON lines](#newline-delimited-json-ndjson-and-json-lines)
-  - [Parsing numbers inside strings](#parsing-numbers-inside-strings)
-  - [Dynamic Number Types](#dynamic-number-types)
-  - [Raw strings from keys](#raw-strings-from-keys)
-  - [General direct access to the raw JSON string](#general-direct-access-to-the-raw-json-string)
-  - [Storing directly into an existing string instance](#storing-directly-into-an-existing-string-instance)
-  - [Thread safety](#thread-safety)
-  - [Standard compliance](#standard-compliance)
-  - [Backwards compatibility](#backwards-compatibility)
-  - [Examples](#examples)
-  - [Performance tips](#performance-tips)
-  - [Further reading](#further-reading)
+  * [Requirements](#requirements)
+  * [Including simdjson](#including-simdjson)
+  * [Using simdjson with package managers](#using-simdjson-with-package-managers)
+  * [Using simdjson as a CMake dependency](#using-simdjson-as-a-cmake-dependency)
+  * [Versions](#versions)
+  * [The basics: loading and parsing JSON documents](#the-basics--loading-and-parsing-json-documents)
+  * [Documents are iterators](#documents-are-iterators)
+    + [Parser, document and JSON scope](#parser--document-and-json-scope)
+  * [string_view](#string-view)
+  * [Avoiding pitfalls: enable development checks](#avoiding-pitfalls--enable-development-checks)
+  * [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--simdjson--ondemand--value--get--to-get-custom-types--pre-c--20-)
+    + [2. Use `tag_invoke` for custom types (C++20)](#2-use--tag-invoke--for-custom-types--c--20-)
+    + [3. Using static reflection (C++26)](#3-using-static-reflection--c--26-)
+  * [Minifying JSON strings without parsing](#minifying-json-strings-without-parsing)
+  * [UTF-8 validation (alone)](#utf-8-validation--alone-)
+  * [JSON Pointer](#json-pointer)
+  * [JSONPath](#jsonpath)
+  * [Error handling](#error-handling)
+    + [Error handling examples without exceptions](#error-handling-examples-without-exceptions)
+    + [Disabling exceptions](#disabling-exceptions)
+    + [Exceptions](#exceptions)
+    + [Current location in document](#current-location-in-document)
+    + [Checking for trailing content](#checking-for-trailing-content)
+  * [Rewinding](#rewinding)
+  * [Newline-Delimited JSON (ndjson) and JSON lines](#newline-delimited-json--ndjson--and-json-lines)
+  * [Parsing numbers inside strings](#parsing-numbers-inside-strings)
+  * [Dynamic Number Types](#dynamic-number-types)
+  * [Raw strings from keys](#raw-strings-from-keys)
+  * [General direct access to the raw JSON string](#general-direct-access-to-the-raw-json-string)
+  * [Storing directly into an existing string instance](#storing-directly-into-an-existing-string-instance)
+  * [Thread safety](#thread-safety)
+  * [Standard compliance](#standard-compliance)
+  * [Backwards compatibility](#backwards-compatibility)
+  * [Examples](#examples)
+  * [Performance tips](#performance-tips)
+  * [Further reading](#further-reading)
 
 
 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.
-- 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 (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.
+The simdjson library is widely deployed in popular systems such as the Node.js runtime
+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
 ------------------
@@ -320,13 +325,15 @@ 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.
-Alternatively, you can set the macro `SIMDJSON_DEVELOPMENT_CHECKS` to 1 prior to including
+The simdjson code will set `SIMDJSON_DEVELOPMENT_CHECKS=1` in debug mode. Because
+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
@@ -335,9 +342,12 @@ 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
-further encourage you to define `NDEBUG` in your Release builds to disable additional runtime
-testing and get the best performance.
+disable these checks by setting `SIMDJSON_DEVELOPMENT_CHECKS` to 0.
+
+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
 ---------------------
@@ -819,7 +829,7 @@ There are 3 main ways provided by simdjson to deserialize a value into a custom
    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)
+3. Using static reflection (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.
@@ -1083,6 +1093,32 @@ 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.
 
@@ -1107,7 +1143,7 @@ struct Car {
 };
 ```
 
-Observe how we defined the class to use types that simdjson does not directly support (`float`, `int`).
+Observe how we define 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.
 
@@ -1245,7 +1281,6 @@ int main() {
 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 ] }
@@ -1260,6 +1295,23 @@ You may also conditionally fill in `std::optional` values.
   // 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.
@@ -1306,11 +1358,12 @@ your code with the `SIMDJSON_STATIC_REFLECTION` macro set:
 ```
 
 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);
+simdjson::ondemand::document doc = parser.iterate(simdjson::pad(json));
 Car c = doc.get<Car>();
 ```
 
@@ -1556,7 +1609,20 @@ 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
@@ -1945,6 +2011,8 @@ 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;
@@ -1959,6 +2027,21 @@ 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.
 
@@ -2225,6 +2308,11 @@ 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.
@@ -2552,6 +2640,20 @@ 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.
 
@@ -2570,7 +2672,7 @@ 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.

doc/builder.md

@@ -4,16 +4,29 @@ 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.
+define the structure of your JSON document. Our more advanced interface
+automates the process using C++26 static reflection: you get both high
+speed and high convenience.
+
+- [Builder](#builder)
+  * [Overview: string_builder](#overview--string-builder)
+  * [Example: string_builder](#example--string-builder)
+  * [C++26 static reflection](#c--26-static-reflection)
+    + [Without `string_buffer` instance](#without--string-buffer--instance)
+    + [Without `string_buffer` instance but with explicit error handling](#without--string-buffer--instance-but-with-explicit-error-handling)
 
 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).
+Note that JSON strings are always encoded as UTF-8.
 
 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:
+is reallocated when needed.
+The efficiency of `string_builder` stems from its internal use of a resizable array or buffer. When you append data, it adds the characters to this buffer, resizing it only when necessary, typically in a way that minimizes reallocations. This approach contrasts with regular string concatenation, where each operation creates a new string, copying all previous content, leading to quadratic time complexity for repeated concatenations.
+
+
+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.
@@ -32,6 +45,9 @@ After writting the content, if you have reasons to believe that the content migh
 
 - `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.
 
+You might need to do unicode validation if you have strings in your data structures containing
+malformed UTF-8.
+
 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.)
@@ -44,64 +60,75 @@ Example: string_builder
 ---------------------------
 
 ```C++
+struct Car {
+    std::string make;
+    std::string model;
+    int64_t year;
+    std::vector<double> tire_pressure;
+};
 
-    void serialize_car(const Car& car, simdjson::builder::string_builder& builder) {
-        // start of JSON
-        builder.append_raw("{");
+void serialize_car(const Car& car, simdjson::builder::string_builder& builder) {
+    // start of JSON
+    builder.start_object();
 
-        // "make"
-        builder.escape_and_append_with_quotes("make");
-        builder.append_raw(":");
-        builder.escape_and_append_with_quotes(car.make);
+    // "make"
+    builder.append_key_value("make", car.make);
+    builder.append_comma();
 
-        // "model"
-        builder.append_raw(",");
-        builder.escape_and_append_with_quotes("model");
-        builder.append_raw(":");
-        builder.escape_and_append_with_quotes(car.model);
+    // "model"
+    builder.append_key_value("model", car.model);
+    builder.append_comma();
 
-        // "year"
-        builder.append_raw(",");
-        builder.escape_and_append_with_quotes("year");
-        builder.append_raw(":");
-        builder.append(car.year);
+    // "year"
+    builder.append_key_value("year", car.year);
+    builder.append_comma();
 
-        // "tire_pressure"
-        builder.append_raw(",");
-        builder.escape_and_append_with_quotes("tire_pressure");
-        builder.append_raw(":[");
-
-        // vector tire_pressure
-        for (size_t i = 0; i < car.tire_pressure.size(); ++i) {
-            builder.append(car.tire_pressure[i]);
-            if (i < car.tire_pressure.size() - 1) {
-                builder.append_raw(",");
-            }
+    // "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();
         }
-        // end of array
-        builder.append_raw("]");
-
-        // end of object
-        builder.append_raw("}");
     }
+    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;
+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{sb};
+    // p holds the JSON:
+    // "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}"
+    return true;
+}
+```
+
+The `string_builder` constructor takes an optional parameter which specifies the initial
+memory allocation in byte. If you know approximately the size of your JSON output, you can
+pass this value as a parameter (e.g., `simdjson::builder::string_builder sb{1233213}`).
+
+The `string_builder` might throw an exception in case of error when you cast it result to `std::string_view`. If you wish to avoid exceptions, you can use the following programming pattern:
+
+```cpp
+    std::string_view p;
+    if(sb.view().get(p)) {
+        return false; // there was an error
     }
 ```
 
+In all cases, the `std::string_view` instance depends the corresponding `string_builder` instance.
+
 C++26 static reflection
 ------------------------
 
+Static reflection (or compile-time reflection) in C++26 introduces a powerful compile-time mechanism that allows a program to inspect and manipulate its own structure, such as types, variables, functions, and other program elements, during compilation. Unlike runtime reflection in languages like Java or Python, C++26’s static reflection operates entirely at compile time, aligning with C++’s emphasis on zero-overhead abstractions and high performance. It means
+that you can delegate much of the work to the library.
 If you have a compiler with support C++26 static reflection, you can compile
 your code with the `SIMDJSON_STATIC_REFLECTION` macro set:
 
@@ -115,21 +142,55 @@ And then you can append your data structures to a `string_builder` instance
 automatically. In most cases, it should work automatically:
 
 ```cpp
+    struct Car {
+        std::string make;
+        std::string model;
+        int64_t year;
+        std::vector<double> tire_pressure;
+    };
+
     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
-        }
+        sb << c;
+        std::string_view p{sb};
         // 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:
+
+### Without `string_buffer` instance
+
+In some instances, you might want to create a string directly from your own data type.
+You can create a string directly, without an explicit `string_builder` instance
+with the `simdjson::builder::to_json_string` function.
+(Under the hood a `string_builder` instance may still be created.)
+
+```cpp
+    struct Car {
+        std::string make;
+        std::string model;
+        int64_t year;
+        std::vector<double> tire_pressure;
+    };
+
+    void f() {
+        Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+        std::string json = simdjson::builder::to_json_string(c);
+    }
+```
+
+If you know the output size, in bytes, of your JSON string, you may
+pass it as a second parameter (e.g., `simdjson::builder::to_json_string(c, 31123)`).
+
+
+
+### Without `string_buffer` instance but with explicit error handling
+
+If prefer a version without exceptions and explicit error handling, you can use the following
+pattern:
 
 ```cpp
   std::string json;

doc/dom.md

@@ -1,12 +1,16 @@
 The Document-Object-Model (DOM) front-end
 ==========
 
-An overview of what you need to know to use simdjson, with examples.
+An overview of what you need to know to use simdjson to parse JSON documents with
+our DOM API, with examples. [Our documentation regarding the generation (serialization) of JSON documents is in a
+separate document](https://github.com/simdjson/simdjson/blob/master/doc/builder.md).
+
 
 * [DOM vs On-Demand](#dom-vs-on-demand)
 * [The Basics: Loading and Parsing JSON Documents](#the-basics-loading-and-parsing-json-documents-using-the-dom-front-end)
 * [Using the Parsed JSON](#using-the-parsed-json)
 * [C++17 Support](#c17-support)
+* [C++20 Support](#c20-support)
 * [JSON Pointer](#json-pointer)
 * [JSONPath](#jsonpath)
 * [Error Handling](#error-handling)
@@ -27,6 +31,10 @@ a conventional Document-Object-Model (DOM) front-end. In such a scenario, the JS
 entirely parsed, validated and materialized in memory as the first step. The programmer may
 then access the parsed data using this in-memory model.
 
+On-Demand is a different model where you parse just what you need, directly into your own
+data structure. The On-Demand approach, when well tuned, can provide superior performance.
+[We refer you to the On-Demand documentation for further details](https://github.com/simdjson/simdjson/blob/master/doc/basics.md).
+
 The Basics: Loading and Parsing JSON Documents using the DOM front-end
 ----------------------------------------------
 
@@ -90,7 +98,7 @@ 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
@@ -119,6 +127,12 @@ 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) { ... }`
@@ -232,6 +246,23 @@ 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
 ------------
@@ -733,5 +764,5 @@ 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

@@ -367,7 +367,7 @@ Please see our main documentation (`basics.md`) under
 "Use `tag_invoke` for custom types (C++20)" for details about
 tag_invoke functions.
 
-Given a stream of JSON documents, you can add them to a data struture
+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++

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
-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.
+We recommend that you set `NDEBUG` macro 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.
+In C/C++, the `NDEBUG` macro is not set by default.
+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
 -----------------------------------------

include/simdjson/arm64/implementation.h

@@ -23,7 +23,6 @@ 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/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
@@ -344,4 +356,16 @@ namespace std {
 #define SIMDJSON_AVX512_ALLOWED 1
 #endif
 
+
+#ifndef __has_cpp_attribute
+#define simdjson_lifetime_bound
+#elif __has_cpp_attribute(msvc::lifetimebound)
+#define simdjson_lifetime_bound [[msvc::lifetimebound]]
+#elif __has_cpp_attribute(clang::lifetimebound)
+#define simdjson_lifetime_bound [[clang::lifetimebound]]
+#elif __has_cpp_attribute(lifetimebound)
+#define simdjson_lifetime_bound [[lifetimebound]]
+#else
+#define simdjson_lifetime_bound
+#endif
 #endif // SIMDJSON_COMMON_DEFS_H

include/simdjson/concepts.h

@@ -32,14 +32,30 @@ 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_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>;
+    details::supports_insert<T>) && !string_view_keyed_map<T>;
 
 /// Insert into the container however possible
 template <appendable_containers T, typename... Args>
@@ -99,7 +115,6 @@ concept optional_type = requires(std::remove_cvref_t<T> obj) {
   { 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>;
@@ -108,6 +123,7 @@ concept optional_type = requires(std::remove_cvref_t<T> obj) {
 };
 
 
+
 } // namespace concepts
 } // namespace simdjson
 #endif // SIMDJSON_SUPPORTS_DESERIALIZATION

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;
   };
 
@@ -148,7 +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>

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.h

@@ -492,7 +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;

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;
   };
@@ -224,7 +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>;

include/simdjson/dom/parser.h

@@ -340,7 +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
@@ -434,7 +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

include/simdjson/dom/serialization.h

@@ -107,6 +107,9 @@ protected:
  * by a "formatter" which handles the details. Thus
  * the string_builder template could support both minification
  * and prettification, and various other tradeoffs.
+ *
+ * This is not to be confused with the simdjson::builder::string_builder
+ * which is a different class.
  */
 template <class formatter = mini_formatter>
 class string_builder {

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'
@@ -45,7 +45,7 @@ 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
@@ -53,6 +53,14 @@ enum error_code {
   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
@@ -189,6 +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
 
   /**
@@ -245,7 +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

@@ -26,7 +26,6 @@ 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/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.
   //
@@ -647,7 +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;
@@ -1108,6 +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/dependencies.h

@@ -8,6 +8,7 @@
 // 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/generic/ondemand/deserialize.h

@@ -77,7 +77,7 @@ 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>;
+concept deserializable = custom_deserializable<T, ValT> || is_builtin_deserializable_v<T> || concepts::optional_type<T>;
 
 template <typename T, typename ValT = SIMDJSON_IMPLEMENTATION::ondemand::value>
 concept nothrow_custom_deserializable = nothrow_tag_invocable<deserialize_tag, ValT&, T&>;

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

@@ -195,8 +195,8 @@ simdjson_inline document::operator object() & noexcept(false) { return get_objec
 simdjson_inline document::operator uint64_t() noexcept(false) { return get_uint64(); }
 simdjson_inline document::operator int64_t() noexcept(false) { return get_int64(); }
 simdjson_inline document::operator double() noexcept(false) { return get_double(); }
-simdjson_inline document::operator std::string_view() noexcept(false) { return get_string(false); }
-simdjson_inline document::operator raw_json_string() noexcept(false) { return get_raw_json_string(); }
+simdjson_inline document::operator std::string_view() noexcept(false) simdjson_lifetime_bound { return get_string(false); }
+simdjson_inline document::operator raw_json_string() noexcept(false) simdjson_lifetime_bound { return get_raw_json_string(); }
 simdjson_inline document::operator bool() noexcept(false) { return get_bool(); }
 simdjson_inline document::operator value() noexcept(false) { return get_value(); }
 

include/simdjson/generic/ondemand/document.h

@@ -239,20 +239,27 @@ public:
     if constexpr (custom_deserializable<T, document>) {
         return deserialize(*this, out);
     } else {
-#endif // 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_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;
-#if SIMDJSON_SUPPORTS_DESERIALIZATION
     }
-#endif
+#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;
 
@@ -315,7 +322,7 @@ public:
    *          time it parses a document or when it is destroyed.
    * @exception simdjson_error(INCORRECT_TYPE) if the JSON value is not a string.
    */
-  simdjson_inline operator std::string_view() noexcept(false);
+  simdjson_inline operator std::string_view() noexcept(false) simdjson_lifetime_bound;
   /**
    * Cast this JSON value to a raw_json_string.
    *
@@ -324,7 +331,7 @@ public:
    * @returns A pointer to the raw JSON for the given string.
    * @exception simdjson_error(INCORRECT_TYPE) if the JSON value is not a string.
    */
-  simdjson_inline operator raw_json_string() noexcept(false);
+  simdjson_inline operator raw_json_string() noexcept(false) simdjson_lifetime_bound;
   /**
    * Cast this JSON value to a bool.
    *
@@ -814,20 +821,27 @@ public:
     if constexpr (custom_deserializable<T, document_reference>) {
         return deserialize(*this, out);
     } else {
-#endif // 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_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;
-#if SIMDJSON_SUPPORTS_DESERIALIZATION
     }
-#endif
+#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;

include/simdjson/generic/ondemand/json_builder.h

@@ -7,15 +7,19 @@
 #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 <memory>
+#include <optional>
 #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 {
@@ -58,6 +62,27 @@ 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>
@@ -70,46 +95,161 @@ consteval std::string consteval_to_quoted_escaped(std::string_view input);
 
 template <class T>
   requires(std::is_class_v<T> && !container_but_not_string<T> &&
+           !concepts::string_view_keyed_map<T> &&
+           !concepts::optional_type<T> &&
+           !concepts::smart_pointer<T> &&
+           !concepts::appendable_containers<T> &&
            !std::is_same_v<T, std::string> &&
-           !std::is_same_v<T, std::string_view>)
+           !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) {
   int i = 0;
   b.append('{');
-  template for (constexpr auto mem : std::meta::nonstatic_data_members_of(^^T)) {
-    constexpr std::string_view key = std::string_view(std::meta::identifier_of(mem));
+  [:expand(std::meta::nonstatic_data_members_of(^^T, std::meta::access_context::unchecked())):] >> [&]<auto dm>() {
     if (i != 0)
       b.append(',');
-    b.append_raw(consteval_to_quoted_escaped(key));
-    b.append(':');
-    atom(b, t.*mem);
-    i++;
-  }
-  b.append('}');
-  /*[:expand(std::meta::nonstatic_data_members_of(^T)):] >> [&]<auto dm> {
-    if (i != 0)
-      b.append(',');
-    constexpr auto key = consteval_to_quoted_escaped(std::meta::identifier_of(dm));
+    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('}');*/
+  b.append('}');
+}
+
+// Support for optional types (std::optional, etc.)
+template <concepts::optional_type T>
+constexpr void atom(string_builder &b, const T &opt) {
+  if (opt) {
+    atom(b, opt.value());
+  } else {
+    b.append_raw("null");
+  }
+}
+
+// Support for smart pointers (std::unique_ptr, std::shared_ptr, etc.)
+template <concepts::smart_pointer T>
+constexpr void atom(string_builder &b, const T &ptr) {
+  if (ptr) {
+    atom(b, *ptr);
+  } else {
+    b.append_raw("null");
+  }
+}
+
+// Support for enums - serialize as string representation using expand approach from P2996R12
+template <typename T>
+  requires(std::is_enum_v<T>)
+void atom(string_builder &b, const T &e) {
+#if SIMDJSON_STATIC_REFLECTION
+  std::string_view result = "<unnamed>";
+  [:expand(std::meta::enumerators_of(^^T)):] >> [&]<auto enum_val>{
+    if (e == [:enum_val:]) {
+      result = std::meta::identifier_of(enum_val);
+    }
+  };
+
+  if (result != "<unnamed>") {
+    b.append_raw("\"");
+    b.append_raw(result);
+    b.append_raw("\"");
+  } else {
+    // Fallback to integer if enum value not found
+    atom(b, static_cast<std::underlying_type_t<T>>(e));
+  }
+#else
+  // Fallback: serialize as integer if reflection not available
+  atom(b, static_cast<std::underlying_type_t<T>>(e));
+#endif
+}
+
+// Support for appendable containers that don't have operator[] (sets, etc.)
+template <concepts::appendable_containers T>
+  requires(!container_but_not_string<T> && !concepts::string_view_keyed_map<T> &&
+           !concepts::optional_type<T> && !concepts::smart_pointer<T> &&
+           !std::is_same_v<T, std::string> &&
+           !std::is_same_v<T, std::string_view> && !std::is_same_v<T, const char*>)
+constexpr void atom(string_builder &b, const T &container) {
+  if (container.empty()) {
+    b.append_raw("[]");
+    return;
+  }
+  b.append('[');
+  bool first = true;
+  for (const auto& item : container) {
+    if (!first) {
+      b.append(',');
+    }
+    first = false;
+    atom(b, item);
+  }
+  b.append(']');
+}
+
+// append functions that delegate to atom functions for primitive types
+template <class T>
+  requires(std::is_arithmetic_v<T> && !std::is_same_v<T, char>)
+void append(string_builder &b, const T &t) {
+  atom(b, t);
+}
+
+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>)
+void append(string_builder &b, const T &t) {
+  atom(b, t);
+}
+
+template <concepts::optional_type T>
+void append(string_builder &b, const T &t) {
+  atom(b, t);
+}
+
+template <concepts::smart_pointer T>
+void append(string_builder &b, const T &t) {
+  atom(b, t);
+}
+
+template <concepts::appendable_containers T>
+  requires(!container_but_not_string<T> && !concepts::string_view_keyed_map<T> &&
+           !concepts::optional_type<T> && !concepts::smart_pointer<T> &&
+           !std::is_same_v<T, std::string> &&
+           !std::is_same_v<T, std::string_view> && !std::is_same_v<T, const char*>)
+void append(string_builder &b, const T &t) {
+  atom(b, t);
+}
+
+template <concepts::string_view_keyed_map T>
+void append(string_builder &b, const T &t) {
+  atom(b, t);
 }
 
 // works for struct
-template <class Z> void append(string_builder &b, const Z &z) {
+template <class Z>
+  requires(std::is_class_v<Z> && !container_but_not_string<Z> &&
+           !concepts::string_view_keyed_map<Z> &&
+           !concepts::optional_type<Z> &&
+           !concepts::smart_pointer<Z> &&
+           !concepts::appendable_containers<Z> &&
+           !std::is_same_v<Z, std::string> &&
+           !std::is_same_v<Z, std::string_view> &&
+           !std::is_same_v<Z, const char*> &&
+           !std::is_same_v<Z, char>)
+void append(string_builder &b, const Z &z) {
   int i = 0;
   b.append('{');
-  template for (constexpr auto dm : std::meta::nonstatic_data_members_of(^^Z)) {
+  [:expand(std::meta::nonstatic_data_members_of(^^Z, std::meta::access_context::unchecked())):] >> [&]<auto dm>() {
     if (i != 0)
       b.append(',');
-    constexpr auto key = consteval_to_quoted_escaped(std::meta::identifier_of(dm));
+    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('}');
 }
 
@@ -131,8 +271,8 @@ void append(string_builder &b, const Z &z) {
 }
 
 template <class Z>
-simdjson_result<std::string> to_json_string(const Z &z) {
-  string_builder b;
+simdjson_result<std::string> to_json_string(const Z &z, size_t initial_capacity = 1024) {
+  string_builder b(initial_capacity);
   append(b, z);
   std::string_view s;
   if(auto e = b.view().get(s); e) { return e; }
@@ -148,7 +288,13 @@ simdjson_error to_json(const Z &z, std::string &s) {
   s.assign(view);
   return SUCCESS;
 }
-} // namespace json_builder
+
+template <class Z>
+string_builder& operator<<(string_builder& b, const Z& z) {
+  append(b, z);
+  return b;
+}
+} // namespace builder
 } // namespace SIMDJSON_IMPLEMENTATION
 } // namespace simdjson
 #endif // SIMDJSON_STATIC_REFLECTION

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

@@ -42,10 +42,41 @@ 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) {
-    if ((uint8_t(c) < 32) | (c == '"') | (c == '\\')) {
+    // a table lookup is faster than a series of comparisons
+    if(json_quotable_character[static_cast<uint8_t>(c)]) {
       return true;
     }
   }
@@ -110,18 +141,6 @@ simdjson_inline bool fast_needs_escaping(std::string_view view) {
 }
 #endif
 
-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};
 
 SIMDJSON_CONSTEXPR_LAMBDA inline size_t
 find_next_json_quotable_character(const std::string_view view,
@@ -284,9 +303,9 @@ simdjson_inline void string_builder::clear() noexcept {
 namespace internal {
 
 // We could specialize further for 32-bit integers.
-int int_log2(uint32_t x) { return (63 - leading_zeroes(x | 1)); }
+simdjson_really_inline int int_log2(uint32_t x) { return (63 - leading_zeroes(x | 1)); }
 
-int fast_digit_count(uint32_t x) {
+simdjson_really_inline int fast_digit_count(uint32_t x) {
   static uint64_t table[] = {
       4294967296,  8589934582,  8589934582,  8589934582,  12884901788,
       12884901788, 12884901788, 17179868184, 17179868184, 17179868184,
@@ -298,9 +317,9 @@ int fast_digit_count(uint32_t x) {
   return uint32_t((x + table[int_log2(x)]) >> 32);
 }
 
-int int_log2(uint64_t x) { return 63 - leading_zeroes(x | 1); }
+simdjson_really_inline int int_log2(uint64_t x) { return 63 - leading_zeroes(x | 1); }
 
-int fast_digit_count(uint64_t x) {
+simdjson_really_inline int fast_digit_count(uint64_t x) {
   static uint64_t table[] = {9,
                              99,
                              999,
@@ -327,13 +346,31 @@ int fast_digit_count(uint64_t x) {
 
 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 {
+simdjson_really_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>
@@ -367,8 +404,12 @@ simdjson_inline void string_builder::append(number_type v) noexcept {
       unsigned_type pv = static_cast<unsigned_type>(v);
       size_t dc = internal::digit_count(pv);
       char *write_pointer = buffer.get() + position + dc - 1;
-      // optimization opportunity: if v is large, we can do better.
-      while (pv >= 10) {
+      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;
       }
@@ -381,14 +422,21 @@ simdjson_inline void string_builder::append(number_type v) noexcept {
     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>(negative ? -v : v);
+      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;
-      // optimization opportunity: if v is large, we can do better.
-      while (pv >= 10) {
+      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;
       }
@@ -436,6 +484,11 @@ string_builder::escape_and_append_with_quotes(char input) noexcept {
   }
 }
 
+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);
@@ -463,7 +516,7 @@ simdjson_inline string_builder::operator std::string() const noexcept(false) {
 }
 
 simdjson_inline string_builder::operator std::string_view() const
-    noexcept(false) {
+    noexcept(false) simdjson_lifetime_bound {
   return view();
 }
 #endif
@@ -488,6 +541,75 @@ 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

include/simdjson/generic/ondemand/json_string_builder.h

@@ -19,7 +19,7 @@ namespace builder {
  * supports atomic types (Booleans, strings), it does not support composed
  * types (arrays and objects).
  *
- * Ultimately, this class should support kernel-specific optimizations. E.g.,
+ * Ultimately, this class can support kernel-specific optimizations. E.g.,
  * it may make use of SIMD instructions to escape strings faster.
  */
 class string_builder {
@@ -68,12 +68,56 @@ public:
    * 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.
@@ -102,7 +146,7 @@ public:
    * 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);
+  simdjson_inline operator std::string_view() const noexcept(false) simdjson_lifetime_bound;
 #endif
 
   /**
@@ -156,7 +200,6 @@ private:
    */
   simdjson_inline void set_valid(bool valid) noexcept;
 
-
   std::unique_ptr<char[]> buffer{};
   size_t position{0};
   size_t capacity{0};

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

@@ -9,6 +9,9 @@
 #include "simdjson/generic/ondemand/object.h"
 #include "simdjson/generic/ondemand/serialization.h"
 #include "simdjson/generic/ondemand/value.h"
+#if SIMDJSON_STATIC_REFLECTION
+#include "simdjson/generic/ondemand/json_builder.h"
+#endif
 #endif // SIMDJSON_CONDITIONAL_INCLUDE
 
 namespace simdjson {

include/simdjson/generic/ondemand/serialization.h

@@ -35,6 +35,20 @@ inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON
 inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> x);
 inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::object> x);
 inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::array> x);
+
+#if SIMDJSON_STATIC_REFLECTION
+/**
+ * Create a JSON string from any user-defined type using static reflection.
+ * Only available when SIMDJSON_STATIC_REFLECTION is enabled.
+ */
+template<typename T>
+  requires(!std::same_as<T, SIMDJSON_IMPLEMENTATION::ondemand::document> &&
+           !std::same_as<T, SIMDJSON_IMPLEMENTATION::ondemand::value> &&
+           !std::same_as<T, SIMDJSON_IMPLEMENTATION::ondemand::object> &&
+           !std::same_as<T, SIMDJSON_IMPLEMENTATION::ondemand::array>)
+inline std::string to_json_string(const T& obj);
+#endif
+
 } // namespace simdjson
 
 /**

include/simdjson/generic/ondemand/std_deserialize.h

@@ -12,6 +12,7 @@
 #include <limits>
 #if SIMDJSON_STATIC_REFLECTION
 #include <experimental/meta>
+// #include <static_reflection> // for std::define_static_string - header not available yet
 #endif
 
 namespace simdjson {
@@ -63,6 +64,7 @@ error_code tag_invoke(deserialize_tag, auto &val, T &out) noexcept {
 // String deserialization
 //////////////////////////////
 
+// just a character!
 error_code tag_invoke(deserialize_tag, auto &val, char &out) noexcept {
   std::string_view x;
   SIMDJSON_TRY(val.get_string().get(x));
@@ -73,14 +75,17 @@ error_code tag_invoke(deserialize_tag, auto &val, char &out) noexcept {
   return SUCCESS;
 }
 
-
-error_code tag_invoke(deserialize_tag, auto &val, std::string &out) noexcept {
-  std::string_view x;
-  SIMDJSON_TRY(val.get_string().get(x));
-  out.assign(x.data(), x.size());
+// any string-like type (can be constructed from std::string_view)
+template <concepts::constructible_from_string_view T, typename ValT>
+  requires(!require_custom_serialization<T>)
+error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept(std::is_nothrow_constructible_v<T, std::string_view>) {
+  std::string_view str;
+  SIMDJSON_TRY(val.get_string().get(str));
+  out = T{str};
   return SUCCESS;
 }
 
+
 /**
  * STL containers have several constructors including one that takes a single
  * size argument. Thus, some compilers (Visual Studio) will not be able to
@@ -131,6 +136,77 @@ error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept(false) {
 }
 
 
+/**
+ * We want to support std::map and std::unordered_map but only for
+ * string-keyed types.
+ */
+ template <concepts::string_view_keyed_map T, typename ValT>
+ requires(!require_custom_serialization<T>)
+error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept(false) {
+  using value_type = typename std::remove_cvref_t<T>::mapped_type;
+  static_assert(
+     deserializable<value_type, ValT>,
+     "The specified value type inside the container must itself be deserializable");
+  static_assert(
+      std::is_default_constructible_v<value_type>,
+      "The specified value type inside the container must default constructible.");
+ SIMDJSON_IMPLEMENTATION::ondemand::object obj;
+ SIMDJSON_TRY(val.get_object().get(obj));
+ for (auto field : obj) {
+    std::string_view key;
+    SIMDJSON_TRY(field.unescaped_key().get(key));
+    value_type this_value;
+    SIMDJSON_TRY(field.value().get<value_type>().get(this_value));
+    [[maybe_unused]] std::pair<typename T::iterator, bool> result = out.emplace(key, this_value);
+    // unclear what to do if the key already exists
+    // if (result.second == false) {
+    //   // key already exists
+    // }
+ }
+ (void)out;
+ return SUCCESS;
+}
+
+template <concepts::string_view_keyed_map T>
+error_code tag_invoke(deserialize_tag, SIMDJSON_IMPLEMENTATION::ondemand::object &obj, T &out) noexcept {
+  using value_type = typename std::remove_cvref_t<T>::mapped_type;
+
+  out.clear();
+  for (auto field : obj) {
+    std::string_view key;
+    SIMDJSON_TRY(field.unescaped_key().get(key));
+
+    SIMDJSON_IMPLEMENTATION::ondemand::value value_obj;
+    SIMDJSON_TRY(field.value().get(value_obj));
+
+    value_type this_value;
+    SIMDJSON_TRY(value_obj.get(this_value));
+    out.emplace(typename T::key_type(key), std::move(this_value));
+  }
+  return SUCCESS;
+}
+
+template <concepts::string_view_keyed_map T>
+error_code tag_invoke(deserialize_tag, SIMDJSON_IMPLEMENTATION::ondemand::value &val, T &out) noexcept {
+  SIMDJSON_IMPLEMENTATION::ondemand::object obj;
+  SIMDJSON_TRY(val.get_object().get(obj));
+  return simdjson::deserialize(obj, out);
+}
+
+template <concepts::string_view_keyed_map T>
+error_code tag_invoke(deserialize_tag, SIMDJSON_IMPLEMENTATION::ondemand::document &doc, T &out) noexcept {
+  SIMDJSON_IMPLEMENTATION::ondemand::object obj;
+  SIMDJSON_TRY(doc.get_object().get(obj));
+  return simdjson::deserialize(obj, out);
+}
+
+template <concepts::string_view_keyed_map T>
+error_code tag_invoke(deserialize_tag, SIMDJSON_IMPLEMENTATION::ondemand::document_reference &doc, T &out) noexcept {
+  SIMDJSON_IMPLEMENTATION::ondemand::object obj;
+  SIMDJSON_TRY(doc.get_object().get(obj));
+  return simdjson::deserialize(obj, out);
+}
+
 
 /**
  * This CPO (Customization Point Object) will help deserialize into
@@ -172,17 +248,16 @@ error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept(nothrow_deser
 /**
  * This CPO (Customization Point Object) will help deserialize into optional types.
  */
-template <concepts::optional_type T, typename ValT>
+template <concepts::optional_type T>
   requires(!require_custom_serialization<T>)
-error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept(nothrow_deserializable<typename std::remove_cvref_t<T>::value_type, ValT>) {
+error_code tag_invoke(deserialize_tag, auto &val, T &out) noexcept(nothrow_deserializable<typename std::remove_cvref_t<T>::value_type, decltype(val)>) {
   using value_type = typename std::remove_cvref_t<T>::value_type;
 
-  static_assert(
-      deserializable<value_type, ValT>,
-      "The specified type inside the unique_ptr must itself be deserializable");
-  static_assert(
-      std::is_default_constructible_v<value_type>,
-      "The specified type inside the unique_ptr must default constructible.");
+  // Check if the value is null
+  if (val.is_null()) {
+    out.reset(); // Set to nullopt
+    return SUCCESS;
+  }
 
   if (!out) {
     out.emplace();
@@ -202,47 +277,81 @@ constexpr bool user_defined_type = (std::is_class_v<T>
 
 
 // workaround from
-// https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2024/p2996r3.html#back-and-forth
+// https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2024/p2996r10.html#back-and-forth
 // for missing expansion statements
-/*namespace __impl {
-template <auto... vals> struct replicator_type {
-  template <typename F> constexpr void operator>>(F body) const {
-    (body.template operator()<vals>(), ...);
-  }
-};
+namespace __impl {
+  template<auto... vals>
+  struct replicator_type {
+    template<typename F>
+      constexpr void operator>>(F body) const {
+        (body.template operator()<vals>(), ...);
+      }
+  };
 
-template <auto... vals> replicator_type<vals...> replicator = {};
-} // namespace __impl
+  template<auto... vals>
+  replicator_type<vals...> replicator = {};
+}
 
-template <typename R> consteval auto expand(R range) {
+template<typename R>
+consteval auto expand(R range) {
   std::vector<std::meta::info> args;
   for (auto r : range) {
-    args.push_back(std::meta::reflect_value(r));
+    args.push_back(reflect_constant(r));
   }
-  return substitute(^__impl::replicator, args);
-}*/
+  return substitute(^^__impl::replicator, args);
+}
 // end of workaround
 
 template <typename T, typename ValT>
   requires(user_defined_type<T> && std::is_class_v<T>)
-  constexpr error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept {
+error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept {
   SIMDJSON_IMPLEMENTATION::ondemand::object obj;
   if constexpr (std::is_same_v<std::remove_cvref_t<ValT>, SIMDJSON_IMPLEMENTATION::ondemand::object>) {
     obj = val;
   } else {
     SIMDJSON_TRY(val.get_object().get(obj));
   }
-  template for (constexpr auto mem : std::meta::nonstatic_data_members_of(^^T)) {
-    if(std::meta::is_const(mem)) {
-      continue;
+  error_code e = simdjson::SUCCESS;
+
+  [:expand(std::meta::nonstatic_data_members_of(^^T, std::meta::access_context::unchecked())):] >> [&]<auto mem>() {
+    if constexpr (!std::meta::is_const(mem) && std::meta::is_public(mem)) {
+      constexpr std::string_view key = std::define_static_string(std::meta::identifier_of(mem));
+      // Note: removed static assert as optional types are now handled generically
+      // as long we are succesful or the field is not found, we continue
+      if(e == simdjson::SUCCESS || e == simdjson::NO_SUCH_FIELD) {
+        e = obj[key].get(out.[:mem:]);
+      }
     }
-    std::string_view key = std::string_view(std::meta::identifier_of(mem));
-    if(error_code e = obj[key].get(out.[:mem:]); e != simdjson::SUCCESS && e != simdjson::NO_SUCH_FIELD) {
-      return e;
-    }
-  }
-  return simdjson::SUCCESS;
+  };
+  return e;
 }
+
+// Support for enum deserialization - deserialize from string representation using expand approach from P2996R12
+template <typename T, typename ValT>
+  requires(std::is_enum_v<T>)
+error_code tag_invoke(deserialize_tag, ValT &val, T &out) noexcept {
+#if SIMDJSON_STATIC_REFLECTION
+  std::string_view str;
+  SIMDJSON_TRY(val.get_string().get(str));
+
+  bool found = false;
+  [:expand(std::meta::enumerators_of(^^T)):] >> [&]<auto enum_val>{
+    if (!found && str == std::meta::identifier_of(enum_val)) {
+      out = [:enum_val:];
+      found = true;
+    }
+  };
+
+  return found ? SUCCESS : INCORRECT_TYPE;
+#else
+  // Fallback: deserialize as integer if reflection not available
+  std::underlying_type_t<T> int_val;
+  SIMDJSON_TRY(val.get(int_val));
+  out = static_cast<T>(int_val);
+  return SUCCESS;
+#endif
+}
+
 template <typename simdjson_value, typename T>
   requires(user_defined_type<std::remove_cvref_t<T>>)
 error_code tag_invoke(deserialize_tag, simdjson_value &val, std::unique_ptr<T> &out) noexcept {
@@ -281,6 +390,10 @@ error_code tag_invoke(deserialize_tag, simdjson_value &val, std::shared_ptr<T> &
 // Unique pointers
 ////////////////////////////////////////
 error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<bool> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_unique<bool>();
     if (!out) { return MEMALLOC; }
@@ -290,6 +403,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<bool> &out) no
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<int64_t> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_unique<int64_t>();
     if (!out) { return MEMALLOC; }
@@ -299,6 +416,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<int64_t> &out)
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<uint64_t> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_unique<uint64_t>();
     if (!out) { return MEMALLOC; }
@@ -308,6 +429,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<uint64_t> &out
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<double> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_unique<double>();
     if (!out) { return MEMALLOC; }
@@ -317,6 +442,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<double> &out)
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<std::string_view> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_unique<std::string_view>();
     if (!out) { return MEMALLOC; }
@@ -330,6 +459,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<std::string_vi
 // Shared pointers
 ////////////////////////////////////////
 error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<bool> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_shared<bool>();
     if (!out) { return MEMALLOC; }
@@ -339,6 +472,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<bool> &out) no
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<int64_t> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_shared<int64_t>();
     if (!out) { return MEMALLOC; }
@@ -348,6 +485,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<int64_t> &out)
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<uint64_t> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_shared<uint64_t>();
     if (!out) { return MEMALLOC; }
@@ -357,6 +498,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<uint64_t> &out
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<double> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_shared<double>();
     if (!out) { return MEMALLOC; }
@@ -366,6 +511,10 @@ error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<double> &out)
 }
 
 error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<std::string_view> &out) noexcept {
+  if (val.is_null()) {
+    out.reset();
+    return SUCCESS;
+  }
   if (!out) {
     out = std::make_shared<std::string_view>();
     if (!out) { return MEMALLOC; }
@@ -375,6 +524,61 @@ error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<std::string_vi
 }
 
 
+////////////////////////////////////////
+// Explicit optional specializations
+////////////////////////////////////////
+
+////////////////////////////////////////
+// Explicit smart pointer specializations for string and int types
+////////////////////////////////////////
+error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<std::string> &out) noexcept {
+  // Check if the value is null
+  if (val.is_null()) {
+    out.reset(); // Set to nullptr
+    return SUCCESS;
+  }
+
+  if (!out) {
+    out = std::make_unique<std::string>();
+  }
+  std::string_view str;
+  SIMDJSON_TRY(val.get_string().get(str));
+  *out = std::string{str};
+  return SUCCESS;
+}
+
+error_code tag_invoke(deserialize_tag, auto &val, std::shared_ptr<std::string> &out) noexcept {
+  // Check if the value is null
+  if (val.is_null()) {
+    out.reset(); // Set to nullptr
+    return SUCCESS;
+  }
+
+  if (!out) {
+    out = std::make_shared<std::string>();
+  }
+  std::string_view str;
+  SIMDJSON_TRY(val.get_string().get(str));
+  *out = std::string{str};
+  return SUCCESS;
+}
+
+error_code tag_invoke(deserialize_tag, auto &val, std::unique_ptr<int> &out) noexcept {
+  // Check if the value is null
+  if (val.is_null()) {
+    out.reset(); // Set to nullptr
+    return SUCCESS;
+  }
+
+  if (!out) {
+    out = std::make_unique<int>();
+  }
+  int64_t temp;
+  SIMDJSON_TRY(val.get_int64().get(temp));
+  *out = static_cast<int>(temp);
+  return SUCCESS;
+}
+
 } // namespace simdjson
 
 #endif // SIMDJSON_ONDEMAND_DESERIALIZE_H

include/simdjson/generic/ondemand/value.h

@@ -72,22 +72,41 @@ public:
     noexcept
 #endif
  {
-#if SIMDJSON_SUPPORTS_DESERIALIZATION
-    if constexpr (custom_deserializable<T, value>) {
+  #if SIMDJSON_SUPPORTS_DESERIALIZATION
+  if constexpr (custom_deserializable<T, value>) {
       return deserialize(*this, out);
-    } else {
-#endif // 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;
-#if SIMDJSON_SUPPORTS_DESERIALIZATION
-    }
+  } else if constexpr (concepts::optional_type<T>) {
+      using value_type = typename std::remove_cvref_t<T>::value_type;
+
+      // Check if the value is null
+      if (is_null()) {
+        out.reset(); // Set to nullopt
+        return SUCCESS;
+      }
+
+      if (!out) {
+        out.emplace();
+      }
+      return get<value_type>(out.value());
+  } 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
   }
 

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

@@ -778,19 +778,23 @@ simdjson_warn_unused simdjson_inline simdjson_result<double> value_iterator::get
   }
   return result;
 }
+
 simdjson_warn_unused simdjson_inline simdjson_result<bool> value_iterator::get_root_bool(bool check_trailing) noexcept {
   auto max_len = peek_root_length();
   auto json = peek_root_scalar("bool");
-  uint8_t tmpbuf[5+1+1]; // +1 for null termination
-  tmpbuf[5+1] = '\0'; // make sure that buffer is always null terminated.
-  if (!_json_iter->copy_to_buffer(json, max_len, tmpbuf, 5+1)) { return incorrect_type_error("Not a boolean"); }
-  auto result = parse_bool(tmpbuf);
-  if(result.error() == SUCCESS) {
-    if (check_trailing && !_json_iter->is_single_token()) { return TRAILING_CONTENT; }
-    advance_root_scalar("bool");
-  }
-  return result;
+  // We have a boolean if we have either "true" or "false" and the next character is either
+  // a structural character or whitespace. We also check that the length is correct:
+  // "true" and "false" are 4 and 5 characters long, respectively.
+  bool value_true = (max_len >= 4 && !atomparsing::str4ncmp(json, "true") &&
+  (max_len == 4 || jsoncharutils::is_structural_or_whitespace(json[4])));
+  bool value_false = (max_len >= 5 && !atomparsing::str4ncmp(json, "false") &&
+  (max_len == 5 || jsoncharutils::is_structural_or_whitespace(json[5])));
+  if(value_true == false && value_false == false) { return incorrect_type_error("Not a boolean"); }
+  if (check_trailing && !_json_iter->is_single_token()) { return TRAILING_CONTENT; }
+  advance_root_scalar("bool");
+  return value_true;
 }
+
 simdjson_inline simdjson_result<bool> value_iterator::is_root_null(bool check_trailing) noexcept {
   auto max_len = peek_root_length();
   auto json = peek_root_scalar("null");

include/simdjson/generic/ondemand/value_iterator.h

@@ -125,7 +125,8 @@ public:
    *
    * @returns Whether the object had any fields (returns false for empty).
    * @error INCOMPLETE_ARRAY_OR_OBJECT If there are no more tokens (implying the *parent*
-   *        array or object is incomplete).
+   *        array or object is incomplete). An INCOMPLETE_ARRAY_OR_OBJECT is an unrecoverable error that
+   *        invalidates the document.
    */
   simdjson_warn_unused simdjson_inline simdjson_result<bool> started_object() noexcept;
   /**
@@ -135,7 +136,8 @@ public:
    *
    * @returns Whether the object had any fields (returns false for empty).
    * @error INCOMPLETE_ARRAY_OR_OBJECT If there are no more tokens (implying the *parent*
-   *        array or object is incomplete).
+   *        array or object is incomplete). An INCOMPLETE_ARRAY_OR_OBJECT is an unrecoverable error that
+   *        invalidates the document.
    */
   simdjson_warn_unused simdjson_inline simdjson_result<bool> started_root_object() noexcept;
 
@@ -257,7 +259,8 @@ public:
    *
    * @returns Whether the array had any elements (returns false for empty).
    * @error INCOMPLETE_ARRAY_OR_OBJECT If there are no more tokens (implying the *parent*
-   *        array or object is incomplete).
+   *        array or object is incomplete). An INCOMPLETE_ARRAY_OR_OBJECT is an unrecoverable error that
+   *        invalidates the document.
    */
   simdjson_warn_unused simdjson_inline simdjson_result<bool> started_array() noexcept;
   /**
@@ -267,7 +270,8 @@ public:
    *
    * @returns Whether the array had any elements (returns false for empty).
    * @error INCOMPLETE_ARRAY_OR_OBJECT If there are no more tokens (implying the *parent*
-   *        array or object is incomplete).
+   *        array or object is incomplete). An INCOMPLETE_ARRAY_OR_OBJECT is an unrecoverable error that
+   *        invalidates the document.
    */
   simdjson_warn_unused simdjson_inline simdjson_result<bool> started_root_array() noexcept;
 

include/simdjson/haswell/implementation.h

@@ -28,7 +28,6 @@ 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 haswell

include/simdjson/icelake/implementation.h

@@ -28,7 +28,6 @@ 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 icelake

include/simdjson/implementation.h

@@ -26,15 +26,6 @@ simdjson_inline simdjson_warn_unused bool validate_utf8(const std::string_view s
   return validate_utf8(sv.data(), sv.size());
 }
 
-/**
- * Write the string to the output buffer while escaping double-quote, backlash and ascii control characters.
- *
- * @param input the string_view to escape
- * @param out output buffer (for escaped string): to be safe, it should have 6 * input.size() allocated bytes.
- * @return number of bytes written
- */
-simdjson_warn_unused size_t write_string_escaped(const std::string_view input, char *out) noexcept;
-
 /**
  * Validate the UTF-8 string.
  *
@@ -136,14 +127,6 @@ public:
    */
   simdjson_warn_unused virtual bool validate_utf8(const char *buf, size_t len) const noexcept = 0;
 
-  /**
-   * Write the string to the output buffer while escaping double-quote, backlash and ascii control characters.
-   *
-   * @param input the string_view to escape
-   * @param out output buffer (for escaped string): to be safe, it should have 6 * input.size() allocated bytes.
-   * @return number of bytes written
-   */
-  simdjson_warn_unused virtual size_t write_string_escaped(const std::string_view input, char *out) const noexcept = 0;
 protected:
   /** @private Construct an implementation with the given name and description. For subclasses. */
   simdjson_inline implementation(

include/simdjson/jsonpathutil.h

@@ -2,7 +2,7 @@
 #define SIMDJSON_JSONPATHUTIL_H
 
 #include <string>
-#include <string_view>
+#include "simdjson/common_defs.h"
 
 namespace simdjson {
 /**

include/simdjson/lasx/implementation.h

@@ -23,7 +23,6 @@ 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 lasx

include/simdjson/lsx/implementation.h

@@ -23,7 +23,6 @@ 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 lsx

include/simdjson/ondemand.h

@@ -12,6 +12,28 @@ namespace simdjson {
    * @copydoc simdjson::builtin::builder
    */
   namespace builder = builtin::builder;
+
+#if SIMDJSON_STATIC_REFLECTION
+  /**
+   * Create a JSON string from any user-defined type using static reflection.
+   * Only available when SIMDJSON_STATIC_REFLECTION is enabled.
+   */
+  template<typename T>
+    requires(!std::same_as<T, builtin::ondemand::document> &&
+             !std::same_as<T, builtin::ondemand::value> &&
+             !std::same_as<T, builtin::ondemand::object> &&
+             !std::same_as<T, builtin::ondemand::array>)
+  inline std::string to_json_string(const T& obj) {
+    builder::string_builder str_builder;
+    append(str_builder, obj);
+    std::string_view view;
+    if (str_builder.view().get(view) == SUCCESS) {
+      return std::string(view);
+    }
+    return "";
+  }
+#endif
+
 } // namespace simdjson
 
 #endif // SIMDJSON_ONDEMAND_H

include/simdjson/padded_string-inl.h

@@ -125,9 +125,9 @@ inline const char *padded_string::data() const noexcept { return data_ptr; }
 
 inline char *padded_string::data() noexcept { return data_ptr; }
 
-inline padded_string::operator std::string_view() const { return std::string_view(data(), length()); }
+inline padded_string::operator std::string_view() const simdjson_lifetime_bound { return std::string_view(data(), length()); }
 
-inline padded_string::operator padded_string_view() const noexcept {
+inline padded_string::operator padded_string_view() const noexcept simdjson_lifetime_bound {
   return padded_string_view(data(), length(), length() + SIMDJSON_PADDING);
 }
 

include/simdjson/portability.h

@@ -191,6 +191,7 @@ using std::size_t;
 #if defined(NDEBUG) || defined(__OPTIMIZE__) || (defined(_MSC_VER) && !defined(_DEBUG))
 // If NDEBUG is set, or __OPTIMIZE__ is set, or we are under MSVC in release mode,
 // then do away with asserts and use __assume.
+// We still recommend that our users set NDEBUG in release mode.
 #if SIMDJSON_VISUAL_STUDIO
 #define SIMDJSON_UNREACHABLE() __assume(0)
 #define SIMDJSON_ASSUME(COND) __assume(COND)

include/simdjson/ppc64/implementation.h

@@ -32,7 +32,6 @@ public:
                                          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 ppc64

include/simdjson/simdjson_version.h

@@ -4,22 +4,22 @@
 #define SIMDJSON_SIMDJSON_VERSION_H
 
 /** The version of simdjson being used (major.minor.revision) */
-#define SIMDJSON_VERSION "3.12.2"
+#define SIMDJSON_VERSION "4.0.0"
 
 namespace simdjson {
 enum {
   /**
    * The major version (MAJOR.minor.revision) of simdjson being used.
    */
-  SIMDJSON_VERSION_MAJOR = 3,
+  SIMDJSON_VERSION_MAJOR = 4,
   /**
    * The minor version (major.MINOR.revision) of simdjson being used.
    */
-  SIMDJSON_VERSION_MINOR = 12,
+  SIMDJSON_VERSION_MINOR = 0,
   /**
    * The revision (major.minor.REVISION) of simdjson being used.
    */
-  SIMDJSON_VERSION_REVISION = 2
+  SIMDJSON_VERSION_REVISION = 0
 };
 } // namespace simdjson
 

include/simdjson/westmere/implementation.h

@@ -24,7 +24,6 @@ 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 westmere

jsonexamples/CMakeLists.txt

@@ -1,6 +1,7 @@
 # The bulk of our data files is found in https://github.com/simdjson/simdjson-data
 set(EXAMPLE_JSON ${CMAKE_CURRENT_BINARY_DIR}/twitter.json PARENT_SCOPE)
 set(EXAMPLE_NDJSON ${CMAKE_CURRENT_BINARY_DIR}/amazon_cellphones.ndjson PARENT_SCOPE)
+set(BENCH_CITM_JSON ${CMAKE_CURRENT_BINARY_DIR}/citm_catalog.json PARENT_SCOPE)
 
 # Copy static files to the build dir so they live alongside the generated ones
 file(GLOB_RECURSE example_files RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.json *.ndjson)

p2996/Dockerfile

@@ -0,0 +1,42 @@
+# Stage 1. Check out LLVM source code and run the build.
+FROM debian:12 AS builder
+# First, Update the apt's source list and include the sources of the packages.
+RUN grep deb /etc/apt/sources.list | \
+    sed 's/^deb/deb-src /g' >> /etc/apt/sources.list
+# Install compiler, python and subversion.
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends ca-certificates gnupg \
+           build-essential cmake make python3 zlib1g wget subversion unzip ninja-build git linux-perf && \
+    rm -rf /var/lib/apt/lists/*
+RUN git clone --depth=1 --branch p2996  https://github.com/bloomberg/clang-p2996.git /tmp/clang-source
+RUN cmake -S /tmp/clang-source/llvm -B /tmp/clang-source/build-llvm -DCMAKE_BUILD_TYPE=Release \
+    -DLLVM_ENABLE_ASSERTIONS=ON \
+    -DLLVM_UNREACHABLE_OPTIMIZE=ON \
+    -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" \
+    -DCLANG_DEFAULT_CXX_STDLIB=libc++ \
+    -DLLVM_ENABLE_PROJECTS=clang -G Ninja
+RUN cmake --build /tmp/clang-source/build-llvm -j
+RUN cmake --install /tmp/clang-source/build-llvm --prefix /tmp/clang-install
+
+
+# Stage 2. Produce a minimal release image with build results.
+FROM debian:12
+ARG USER_NAME
+ARG USER_ID
+ARG GROUP_ID
+LABEL maintainer "LLVM Developers"
+# Install packages for minimal useful image.
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends build-essential  ca-certificates rust-all libcurl4-openssl-dev cmake make wget python3 python3-dev sudo curl ninja-build vim git binutils linux-perf && \
+    rm -rf /var/lib/apt/lists/*
+# Copy build results of stage 1 to /usr/local.
+COPY --from=builder /tmp/clang-install/ /usr/local/
+RUN P=`/usr/local/bin/clang++ -v 2>&1 | grep Target | cut -d' ' -f2-`; echo /usr/local/lib/$P > /etc/ld.so.conf.d/$P.conf
+RUN ldconfig
+RUN addgroup --gid $GROUP_ID user; exit 0
+RUN adduser --disabled-password --gecos '' --uid $USER_ID --gid $GROUP_ID $USER_NAME; exit 0
+RUN echo "$USER_NAME:$USER_NAME" | chpasswd && adduser $USER_NAME sudo
+RUN echo '----->'
+RUN echo 'root:Docker!' | chpasswd
+ENV TERM xterm-256color
+USER $USER_NAME

p2996/README.md

@@ -0,0 +1,79 @@
+# Building the experimental LLVM compiler for static reflection
+
+The simdjson has experimental support for static reflection, when the compiler supports it.
+We focus on the [latest p2996 paper](https://isocpp.org/files/papers/P2996R10.html) that is targeting C++26.
+
+## Current status
+There are 2 versions of compiler that aim to support the C++26 reflection paper ([p2996](https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2024/p2996r10.html)).
+
+1. [Clang-p2996 llvm branch](https://github.com/bloomberg/clang-p2996/tree/p2996) that open-source and available in the [Compiler Explorer](https://godbolt.org/z/eoEej3E6j).
+2. EDG reflection branch that is only publicly available in the [Compiler Explorer](https://godbolt.org).
+
+For now, we will resort to #1, since it is open-source.
+
+We are excited to hear that this reflection proposal seems to be on-track for C++26. [As per Herb Sutter](https://herbsutter.com/2024/03/22/trip-report-winter-iso-c-standards-meeting-tokyo-japan/).
+
+## Instructions for building simdjson for static reflection (using docker)
+
+We are assuming that you are running Linux or macOS. We recommend that Windows users rely on WSL.
+
+Follow the following two steps:
+
+1. Clone the project
+
+```bash
+git clone https://github.com/simdjson/simdjson.git
+cd simdjson
+```
+
+As of March 2025, support for static reflection is available only in the `json_builder_init` branch, so you should switch:
+
+```bash
+git checkout -B json_builder_init
+```
+
+2. Make sure that you have [docker installed and running](https://docs.docker.com/engine/install/) on your system. Most Linux distributions support docker though some (like RedHat) have the equivalent (Podman). Users of Apple systems may want to [consider OrbStack](https://orbstack.dev). You do not need to familiar with docker, you just need to make sure that you are have it running.
+
+3. While inside the `simdjson` repository, run the following bash script:
+
+```bash
+bash ./p2996/run_docker.sh bash
+```
+
+This will enter a bash shell with access to the repo directory. Note that this will take some time when running it for the first time, since the specific container image has to be built.
+
+This step builds and executes a docker container defined by our Dockerfile which provides the necessary environment.
+
+Importantly, we build the experimental LLVM compiler based on the current state of the
+`p2996` branch of the `https://github.com/bloomberg/clang-p2996.git` repository. It is possible that the build could fail. Furthermore, you may need to refresh the build from time to time. We provide a script which allows you to delete the docker image you have built (`bash ./p2996/remove_docker.sh`) so you can start anew.
+
+
+
+
+4. While inside the docker shell, configure the build system with cmake:
+```bash
+CXX=clang++ cmake -B buildreflect -D SIMDJSON_STATIC_REFLECTION=ON -DSIMDJSON_DEVELOPER_MODE=ON
+```
+This only needs to be done once.
+
+5. Build the code...
+```bash
+cmake --build buildreflect --target benchmark_serialization_citm_catalog benchmark_serialization_twitter
+```
+
+
+6. Run the tests...
+```bash
+ctest --test-dir buildreflect --output-on-failure
+```
+
+7. Run the benchmarks.
+```bash
+./buildreflect/benchmark/static_reflect/citm_catalog_benchmark/benchmark_serialization_citm_catalog
+./buildreflect/benchmark/static_reflect/twitter_benchmark/benchmark_serialization_twitter
+```
+
+You can modify the source code with your favorite editor and run again steps 5 (Build the code) and 6 (Run the tests) and 7 (Run the benchmark). Importantly, you should remain in the docker shell.
+
+You can create a new docker shell at any time by running step 3 (bash script).
+

p2996/remove_docker.sh

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+tuser=$(echo $USER | tr -dc 'a-z')
+container_name=${CONTAINER_NAME:-"debian12-clang-p2996-programming_station-for-$tuser-simdjson"}
+echo $container_name
+echo " Removing the container "
+docker ps -a | awk '{ print $1,$2 }' | grep $container_name| awk '{print $1 }' | xargs -I {} docker stop -t 1 {} | xargs docker rm
+# removing image
+docker image rm $container_name

p2996/run_docker.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+
+if [ -z "$1" ]
+  then
+    echo "The run-docker-station script takes a command as a argument. E.g., try ./run-docker-station 'ls' "
+    exit 1
+fi
+
+set -o noglob
+SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )"
+
+COMMAND=$@
+
+tuser=$(echo $USER | tr -dc 'a-z')
+
+container_name=${CONTAINER_NAME:-"debian12-clang-p2996-programming_station-for-$tuser-simdjson"}
+
+command -v docker >/dev/null 2>&1 || { echo >&2 "Please install docker. E.g., go to https://www.docker.com/products/docker-desktop Type 'docker' to diagnose the problem."; exit 1; }
+
+docker info >/dev/null 2>&1 || { echo >&2 "Docker server is not running? type 'docker info'."; exit 1; }
+
+SSHPARAM=""
+[ -e "${SSH_AUTH_SOCK}" ] && SSHPARAM="--volume /run/host-services/ssh-auth.sock:/run/host-services/ssh-auth.sock  --env SSH_AUTH_SOCK=/run/host-services/ssh-auth.sock"
+[ -e "${HOME}/.ssh" ] && SSHPARAM="--volume ${HOME}/.ssh:/home/$tuser/.ssh:ro"
+
+
+docker image inspect $container_name >/dev/null 2>&1 || ( echo "instantiating the container" ; docker build --no-cache -t $container_name -f $SCRIPTPATH/Dockerfile --build-arg USER_NAME="$tuser"  --build-arg USER_ID=$(id -u)  --build-arg GROUP_ID=$(id -g) .  )
+
+if [ -t 0 ]; then DOCKER_ARGS=-it; fi
+docker run --rm $DOCKER_ARGS -h $container_name $SSHPARAM -v $(pwd):$(pwd):Z --privileged  -w $(pwd)  $container_name sh -c "$COMMAND"

singleheader/CMakeLists.txt

@@ -52,7 +52,7 @@ if (Python3_Interpreter_FOUND AND (NOT WIN32))
       # file regeneration in the source directory without the user's knowledge.
       # OUTPUT ${CMAKE_CURRENT_SOURCE_DIR}/simdjson.cpp ${CMAKE_CURRENT_SOURCE_DIR}/simdjson.h ${CMAKE_CURRENT_SOURCE_DIR}/amalgamate_demo.cpp ${CMAKE_CURRENT_SOURCE_DIR}/README.md
       COMMAND ${CMAKE_COMMAND} -E copy ${SINGLEHEADER_FILES} ${CMAKE_CURRENT_SOURCE_DIR}
-      DEPENDS ${SINGLEHEADER_FILES}
+      POST_BUILD ${SINGLEHEADER_FILES}
     )
   endif()
 

singleheader/amalgamate.py

@@ -296,10 +296,10 @@ class SimdjsonRepository:
 
 class Amalgamator:
     @classmethod
-    def amalgamate(cls, output_path: str, filename: str, roots: List[RelativeRoot], timestamp: str):
+    def amalgamate(cls, output_path: str, filename: str, roots: List[RelativeRoot], timestamp: str, version: str):
         print(f"Creating {output_path}")
         fid = open(output_path, 'w')
-        print(f"/* auto-generated on {timestamp}. Do not edit! */", file=fid)
+        print(f"/* auto-generated on {timestamp}. version {version} Do not edit! */", file=fid)
         amalgamator = cls(fid, SimdjsonRepository(PROJECTPATH, roots))
         file = amalgamator.repository[filename]
         assert file, f"{filename} not found in {[os.path.join(PROJECTPATH, root) for root in roots]}!"
@@ -480,8 +480,13 @@ AMAL_C = os.path.join(AMALGAMATE_OUTPUT_PATH, "simdjson.cpp")
 DEMOCPP = os.path.join(AMALGAMATE_OUTPUT_PATH, "amalgamate_demo.cpp")
 README = os.path.join(AMALGAMATE_OUTPUT_PATH, "README.md")
 
-Amalgamator.amalgamate(AMAL_H, "simdjson.h", ['include'], timestamp).validate_all_files_used('include')
-Amalgamator.amalgamate(AMAL_C, "simdjson.cpp", ['src', 'include'], timestamp).validate_all_files_used('src')
+def read_version():
+    with open(os.path.join(PROJECTPATH, 'include/simdjson/simdjson_version.h')) as f:
+        return re.search(r'\d+\.\d+\.\d+', f.read()).group(0)
+
+version = read_version()
+Amalgamator.amalgamate(AMAL_H, "simdjson.h", ['include'], timestamp, version).validate_all_files_used('include')
+Amalgamator.amalgamate(AMAL_C, "simdjson.cpp", ['src', 'include'], timestamp, version).validate_all_files_used('src')
 
 # copy the README and DEMOCPP
 if SCRIPTPATH != AMALGAMATE_OUTPUT_PATH:

src/arm64.cpp

@@ -165,10 +165,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return arm64::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace arm64
 } // namespace simdjson
 

src/fallback.cpp

@@ -121,10 +121,78 @@ simdjson_inline void validate_utf8_character() {
   idx += 4;
 }
 
+static const uint8_t CHAR_TYPE_SPACE     = 1 << 0;
+static const uint8_t CHAR_TYPE_OPERATOR  = 1 << 1;
+static const uint8_t CHAR_TYPE_ESC_ASCII = 1 << 2;
+static const uint8_t CHAR_TYPE_NON_ASCII = 1 << 3;
+
+const uint8_t char_table[256] = {
+  0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04,
+  0x04, 0x05, 0x05, 0x04, 0x04, 0x05, 0x04, 0x04,
+  0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04,
+  0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04,
+  0x01, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x02, 0x04, 0x02, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08,
+  0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08
+};
+
+simdjson_inline bool char_is_type(uint8_t c, uint8_t type) {
+  return (char_table[c] & type);
+}
+
+simdjson_inline bool char_is_space(uint8_t c) {
+  return char_is_type(c, CHAR_TYPE_SPACE);
+}
+
+simdjson_inline bool char_is_operator(uint8_t c) {
+  return char_is_type(c, CHAR_TYPE_OPERATOR);
+}
+
+simdjson_inline bool char_is_space_or_operator(uint8_t c) {
+  return char_is_type(c, CHAR_TYPE_SPACE | CHAR_TYPE_OPERATOR);
+}
+
+simdjson_inline bool char_is_ascii_stop(uint8_t c) {
+  return char_is_type(c, CHAR_TYPE_ESC_ASCII | CHAR_TYPE_NON_ASCII);
+}
+
 // Returns true if the string is unclosed.
 simdjson_inline bool validate_string() {
   idx++; // skip first quote
-  while (idx < len && buf[idx] != '"') {
+  while (idx < len) {
+    do {
+      if (char_is_ascii_stop(buf[idx])) { break; }
+      idx++;
+    } while (idx < len);
+    if (idx >= len) { return true; }
+    if (buf[idx] == '"') {
+      return false;
+    }
     if (buf[idx] == '\\') {
       idx += 2;
     } else if (simdjson_unlikely(buf[idx] & 0x80)) {
@@ -138,43 +206,31 @@ simdjson_inline bool validate_string() {
   return false;
 }
 
-simdjson_inline bool is_whitespace_or_operator(uint8_t c) {
-  switch (c) {
-    case '{': case '}': case '[': case ']': case ',': case ':':
-    case ' ': case '\r': case '\n': case '\t':
-      return true;
-    default:
-      return false;
-  }
-}
-
 //
 // Parse the entire input in STEP_SIZE-byte chunks.
 //
 simdjson_inline error_code scan() {
   bool unclosed_string = false;
   for (;idx<len;idx++) {
-    switch (buf[idx]) {
-      // String
-      case '"':
-        add_structural();
-        unclosed_string |= validate_string();
-        break;
-      // Operator
-      case '{': case '}': case '[': case ']': case ',': case ':':
-        add_structural();
-        break;
-      // Whitespace
-      case ' ': case '\r': case '\n': case '\t':
-        break;
-      // Primitive or invalid character (invalid characters will be checked in stage 2)
-      default:
-        // Anything else, add the structural and go until we find the next one
-        add_structural();
-        while (idx+1<len && !is_whitespace_or_operator(buf[idx+1])) {
-          idx++;
-        };
-        break;
+    do {
+      if (!char_is_space(buf[idx])) { break; }
+      idx++;
+    } while (idx < len);
+    if (idx >= len) { break; }
+    // String
+    if (buf[idx] == '"') {
+      add_structural();
+      unclosed_string |= validate_string();
+    // Operator
+    } else if (char_is_operator(buf[idx])) {
+      add_structural();
+    // Primitive or invalid character (invalid characters will be checked in stage 2)
+    } else {
+      // Anything else, add the structural and go until we find the next one
+      add_structural();
+      while (idx+1<len && !char_is_space_or_operator(buf[idx+1])) {
+        idx++;
+      };
     }
   }
   // We pad beyond.
@@ -403,10 +459,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return fallback::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace fallback
 } // namespace simdjson
 

src/generic/stage2/stringparsing.h

@@ -237,104 +237,6 @@ simdjson_warn_unused simdjson_inline uint8_t *parse_wobbly_string(const uint8_t
   }
 }
 
-/////////////
-/// TODO: This function is not used in the codebase. It is not clear if it is needed.
-/////////////
-simdjson_warn_unused size_t write_string_escaped(const std::string_view input, char *out) noexcept {
-  // We are making the following assumption: most strings will either be very short or they will not
-  // need escaping.
-  size_t i = 0;
-  size_t pos = 0;
-  /*if(input.size() >= escaping::BYTES_PROCESSED) {
-    auto vec_processing = [input,out]() -> size_t {
-      size_t index = 0;
-      size_t position = 0;
-      for(;input.size() - index >= escaping::BYTES_PROCESSED; index += escaping::BYTES_PROCESSED) {
-        escaping vinput = escaping::copy_and_find(reinterpret_cast<const uint8_t *>(input.data()) + index, reinterpret_cast<uint8_t *>(out) + position);
-        if(vinput.has_escape()) {
-          return index + vinput.escape_index(); // We have a character that needs escaping
-        }
-        position += escaping::BYTES_PROCESSED;
-      }
-      if(index == input.size()) { return input.size(); }
-      // We virtually backtrack so we can load a full vector register
-      index = input.size() - escaping::BYTES_PROCESSED;
-      position = index;
-      escaping vinput = escaping::copy_and_find(reinterpret_cast<const uint8_t *>(input.data()) + index, reinterpret_cast<uint8_t *>(out) + position);
-      if(vinput.has_escape()) {
-        return index + vinput.escape_index(); // We have a character that needs escaping
-      }
-      return input.size();
-    };
-    i = vec_processing();
-    pos = i;
-    if(i == input.size()) { return pos; }
-    // Here we only continue if there was a character that needed escaping.
-  }*/
-  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"};
-  static std::array<uint8_t, 256> json_quotable_character =
-    []() SIMDJSON_CONSTEXPR_LAMBDA {
-      std::array<uint8_t, 256> result{};
-      for (int index = 0; index < 32; index++) {
-        result[index] = 1;
-      }
-      for (int index : {'"', '\\'}) {
-        result[index] = 1;
-      }
-      return result;
-    }();
-  // The rest could possibly be vectorized, but consider that we expect most strings
-  // to be short or not to require escaping.
-  for (; i < input.size(); i++) {
-    uint8_t c = static_cast<uint8_t>(input[i]);
-    if(json_quotable_character[c]) {
-      switch (c) {
-      case '"':
-        out[pos++] = '\\';
-        out[pos++] = '"';
-        break;
-      case '\\':
-        out[pos++] = '\\';
-        out[pos++] = '\\';
-        break;
-      case '\b':
-        out[pos++] = '\\';
-        out[pos++] = 'b';
-        break;
-      case '\f':
-        out[pos++] = '\\';
-        out[pos++] = 'f';
-        break;
-      case '\n':
-        out[pos++] = '\\';
-        out[pos++] = 'n';
-        break;
-      case '\r':
-        out[pos++] = '\\';
-        out[pos++] = 'r';
-        break;
-      case '\t':
-        out[pos++] = '\\';
-        out[pos++] = 't';
-        break;
-      default:
-        control_chars[c].copy(out + pos, 6);
-        pos += 6;
-      }
-    } else {
-      out[pos++] = c;
-    }
-  }
-  return pos;
-}
-
-
-
 } // namespace stringparsing
 
 } // unnamed namespace

src/haswell.cpp

@@ -162,10 +162,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return haswell::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace SIMDJSON_IMPLEMENTATION
 } // namespace simdjson
 

src/icelake.cpp

@@ -208,10 +208,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return icelake::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace icelake
 } // namespace simdjson
 

src/implementation.cpp

@@ -186,9 +186,6 @@ public:
   simdjson_warn_unused bool validate_utf8(const char * buf, size_t len) const noexcept final override {
     return set_best()->validate_utf8(buf, len);
   }
-  simdjson_warn_unused size_t write_string_escaped(const std::string_view input, char *out) const noexcept final {
-    return set_best()->write_string_escaped(input, out);
-  }
   simdjson_inline detect_best_supported_implementation_on_first_use() noexcept : implementation("best_supported_detector", "Detects the best supported implementation and sets it", 0) {}
 private:
   const implementation *set_best() const noexcept;
@@ -239,9 +236,6 @@ public:
   simdjson_warn_unused error_code minify(const uint8_t *, size_t, uint8_t *, size_t &) const noexcept final override {
     return UNSUPPORTED_ARCHITECTURE;
   }
-  simdjson_warn_unused size_t write_string_escaped(const std::string_view, char *) const noexcept final override {
-    return 0; // TODO: Evaluate whether this is the right thing to do for unsupported architecture.
-  }
   simdjson_warn_unused bool validate_utf8(const char *, size_t) const noexcept final override {
     return false; // Just refuse to validate. Given that we have a fallback implementation
     // it seems unlikely that unsupported_implementation will ever be used. If it is used,
@@ -325,9 +319,6 @@ simdjson_warn_unused error_code minify(const char *buf, size_t len, char *dst, s
 simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) noexcept {
   return get_active_implementation()->validate_utf8(buf, len);
 }
-simdjson_warn_unused size_t write_string_escaped(const std::string_view input, char *out) noexcept {
-  return get_active_implementation()->write_string_escaped(input, out);
-}
 const implementation * builtin_implementation() {
   static const implementation * builtin_impl = get_available_implementations()[SIMDJSON_STRINGIFY(SIMDJSON_BUILTIN_IMPLEMENTATION)];
   assert(builtin_impl);

src/internal/error_tables.cpp

@@ -11,7 +11,7 @@ namespace internal {
     { SUCCESS, "SUCCESS: No error" },
     { CAPACITY, "CAPACITY: This parser can't support a document that big" },
     { MEMALLOC, "MEMALLOC: Error allocating memory, we're most likely out of memory" },
-    { TAPE_ERROR, "TAPE_ERROR: The JSON document has an improper structure: missing or superfluous commas, braces, missing keys, etc." },
+    { TAPE_ERROR, "TAPE_ERROR: The JSON document has an improper structure: missing or superfluous commas, braces, missing keys, etc.  This is a fatal and unrecoverable error." },
     { DEPTH_ERROR, "DEPTH_ERROR: The JSON document was too deep (too many nested objects and arrays)" },
     { STRING_ERROR, "STRING_ERROR: Problem while parsing a string" },
     { T_ATOM_ERROR, "T_ATOM_ERROR: Problem while parsing an atom starting with the letter 't'" },
@@ -36,7 +36,7 @@ namespace internal {
     { PARSER_IN_USE, "PARSER_IN_USE: Cannot parse a new document while a document is still in use." },
     { OUT_OF_ORDER_ITERATION, "OUT_OF_ORDER_ITERATION: Objects and arrays can only be iterated when they are first encountered." },
     { INSUFFICIENT_PADDING, "INSUFFICIENT_PADDING: simdjson requires the input JSON string to have at least SIMDJSON_PADDING extra bytes allocated, beyond the string's length. Consider using the simdjson::padded_string class if needed." },
-    { INCOMPLETE_ARRAY_OR_OBJECT, "INCOMPLETE_ARRAY_OR_OBJECT: JSON document ended early in the middle of an object or array." },
+    { INCOMPLETE_ARRAY_OR_OBJECT, "INCOMPLETE_ARRAY_OR_OBJECT: JSON document ended early in the middle of an object or array. This is a fatal and unrecoverable error." },
     { SCALAR_DOCUMENT_AS_VALUE, "SCALAR_DOCUMENT_AS_VALUE: A JSON document made of a scalar (number, Boolean, null or string) is treated as a value. Use get_bool(), get_double(), etc. on the document instead. "},
     { OUT_OF_BOUNDS, "OUT_OF_BOUNDS: Attempt to access location outside of document."},
     { TRAILING_CONTENT, "TRAILING_CONTENT: Unexpected trailing content in the JSON input."}

src/lasx.cpp

@@ -125,10 +125,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return lasx::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace lasx
 } // namespace simdjson
 

src/lsx.cpp

@@ -129,10 +129,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return lsx::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace lsx
 } // namespace simdjson
 

src/ppc64.cpp

@@ -135,10 +135,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   return stage2(_doc);
 }
 
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return ppc64::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace ppc64
 } // namespace simdjson
 

src/westmere.cpp

@@ -166,11 +166,6 @@ simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *
   if (error) { return error; }
   return stage2(_doc);
 }
-
-simdjson_warn_unused size_t implementation::write_string_escaped(const std::string_view input, char *out) const noexcept {
-  return westmere::stringparsing::write_string_escaped(input, out);
-}
-
 } // namespace westmere
 } // namespace simdjson
 

tests/CMakeLists.txt

@@ -13,7 +13,7 @@ add_cpp_test(minify_tests        LABELS other acceptance per_implementation)
 add_cpp_test(padded_string_tests LABELS other acceptance                   )
 add_cpp_test(prettify_tests      LABELS other acceptance per_implementation)
 
-if(MSVC AND BUILD_SHARED_LIBS)
+if(WIN32 AND BUILD_SHARED_LIBS)
   # Copy the simdjson dll into the tests directory
   add_custom_command(TARGET unicode_tests POST_BUILD        # Adds a post-build event
     COMMAND ${CMAKE_COMMAND} -E copy_if_different  # which executes "cmake -E copy_if_different..."

tests/builder/CMakeLists.txt

@@ -3,6 +3,9 @@ include_directories(..)
 add_cpp_test(builder_string_builder_tests              LABELS ondemand acceptance per_implementation)
 if(SIMDJSON_STATIC_REFLECTION)
   add_cpp_test(static_reflection_builder_tests           LABELS ondemand acceptance per_implementation)
+  add_cpp_test(static_reflection_comprehensive_tests     LABELS ondemand acceptance per_implementation)
+  add_cpp_test(static_reflection_edge_cases_tests        LABELS ondemand acceptance per_implementation)
+  add_cpp_test(static_reflection_enum_tests              LABELS ondemand acceptance per_implementation)
 endif(SIMDJSON_STATIC_REFLECTION)
 # Copy the simdjson dll into the tests directory
 if(MSVC AND BUILD_SHARED_LIBS)

tests/builder/builder_string_builder_tests.cpp

@@ -107,6 +107,65 @@ namespace builder_tests {
         TEST_SUCCEED();
     }
 
+    bool various_integers() {
+        TEST_START();
+        std::vector<std::pair<int64_t, std::string_view>> test_cases = {
+            {0, "0"},
+            {1, "1"},
+            {-1, "-1"},
+            {42, "42"},
+            {-42, "-42"},
+            {100, "100"},
+            {-100, "-100"},
+            {999, "999"},
+            {-999, "-999"},
+            {2147483647, "2147483647"},  // max 32-bit integer
+            {-2147483648, "-2147483648"}, // min 32-bit integer
+            {4294967296ULL, "4294967296"}, // 2^32
+            {-4294967296LL, "-4294967296"},
+            {10000000000LL, "10000000000"}, // 10 billion
+            {-10000000000LL, "-10000000000"},
+            {9223372036854775807LL, "9223372036854775807"}, // max 64-bit integer
+            {-9223372036854775807LL-1, "-9223372036854775808"}, // min 64-bit integer
+            {1234567890123LL, "1234567890123"},
+            {-1234567890123LL, "-1234567890123"},
+        };
+        for (const auto& [value, expected] : test_cases) {
+            simdjson::builder::string_builder sb;
+            sb.append(value);
+            std::string_view p;
+            auto result = sb.view().get(p);
+            ASSERT_SUCCESS(result);
+            ASSERT_EQUAL(p, expected);
+        }
+        TEST_SUCCEED();
+    }
+
+    bool various_unsigned_integers() {
+        TEST_START();
+        std::vector<std::pair<uint64_t, std::string_view>> test_cases = {
+            {0, "0"},
+            {1, "1"},
+            {42, "42"},
+            {100, "100"},
+            {999, "999"},
+            {2147483647, "2147483647"},  // max 32-bit integer
+            {4294967296ULL, "4294967296"}, // 2^32
+            {10000000000LL, "10000000000"}, // 10 billion
+            {9223372036854775807LL, "9223372036854775807"}, // max 64-bit integer
+            {1234567890123LL, "1234567890123"},
+        };
+        for (const auto& [value, expected] : test_cases) {
+            simdjson::builder::string_builder sb;
+            sb.append(value);
+            std::string_view p;
+            auto result = sb.view().get(p);
+            ASSERT_SUCCESS(result);
+            ASSERT_EQUAL(p, expected);
+        }
+        TEST_SUCCEED();
+    }
+
     bool append_raw() {
         TEST_START();
         simdjson::builder::string_builder sb;
@@ -158,7 +217,7 @@ namespace builder_tests {
         TEST_SUCCEED();
     }
 
-    void serialize_car(const Car& car, simdjson::builder::string_builder& builder) {
+    void serialize_car_long(const Car& car, simdjson::builder::string_builder& builder) {
         // start of JSON
         builder.append_raw("{");
 
@@ -198,6 +257,56 @@ namespace builder_tests {
         builder.append_raw("}");
     }
 
+    bool car_test_long() {
+        TEST_START();
+        simdjson::builder::string_builder sb;
+        Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+        serialize_car_long(c, sb);
+        std::string_view p;
+        auto result = sb.view().get(p);
+        ASSERT_SUCCESS(result);
+        ASSERT_EQUAL(p, "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}");
+        TEST_SUCCEED();
+    }
+
+
+
+    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();
+            }
+        }
+        // end of array
+        builder.end_array();
+
+        // end of object
+        builder.end_object();
+    }
+
     bool car_test() {
         TEST_START();
         simdjson::builder::string_builder sb;
@@ -210,11 +319,25 @@ namespace builder_tests {
         TEST_SUCCEED();
     }
 
-
+    #if SIMDJSON_EXCEPTIONS
+    bool car_test_exception() {
+        TEST_START();
+        simdjson::builder::string_builder sb;
+        Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+        serialize_car_long(c, sb);
+        std::string_view p{sb};
+        ASSERT_EQUAL(p, "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}");
+        TEST_SUCCEED();
+    }
+    #endif
     bool run() {
         return
+            various_integers() &&
+            various_unsigned_integers() &&
+            car_test_long() &&
             car_test() &&
     #if SIMDJSON_EXCEPTIONS
+            car_test_exception() &&
             string_convertion_except() &&
     #endif
             append_char() &&

tests/builder/static_reflection_builder_tests.cpp

@@ -48,14 +48,11 @@ namespace builder_tests {
 
   bool car_test() {
     TEST_START();
-    simdjson::builder::string_builder sb;
+#if SIMDJSON_STATIC_REFLECTION
     Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
-    append(sb, c);
-    std::string_view p;
-    auto result = sb.view().get(p);
+    auto result = builder::to_json_string(c);
     ASSERT_SUCCESS(result);
-    ASSERT_EQUAL(p, "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}");
-    std::string pstr(p.begin(), p.end());
+    std::string pstr = result.value();
     ASSERT_EQUAL(pstr, "{\"make\":\"Toyota\",\"model\":\"Corolla\",\"year\":2017,\"tire_pressure\":[30.0,30.2,30.513,30.79]}");
     simdjson::ondemand::parser parser;
     simdjson::ondemand::document doc;
@@ -70,12 +67,38 @@ namespace builder_tests {
     ASSERT_EQUAL(c2.tire_pressure[1], 30.2);
     ASSERT_EQUAL(c2.tire_pressure[2], 30.513);
     ASSERT_EQUAL(c2.tire_pressure[3], 30.79);
+#endif
     TEST_SUCCEED();
   }
+  #if SIMDJSON_EXCEPTIONS
+  bool car_test_exception() {
+      TEST_START();
+      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{sb};
+      TEST_SUCCEED();
+  }
+  bool car_test_exception2() {
+      TEST_START();
+      simdjson::builder::string_builder sb;
+      Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+      sb << c;
+      std::string_view p{sb};
+      TEST_SUCCEED();
+  }
+  void car_test_to_json_exception() {
+    TEST_START();
+    Car c = {"Toyota", "Corolla", 2017, {30.0,30.2,30.513,30.79}};
+    std::string json = simdjson::builder::to_json_string(c);
+    TEST_SUCCEED();
+  }
+  #endif // SIMDJSON_EXCEPTIONS
 
 
 bool serialize_deserialize_kid() {
   TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
   simdjson::padded_string json_str =
       R"({"age": 12, "name": "John", "toys": ["car", "ball"]})"_padded;
   simdjson::ondemand::parser parser;
@@ -90,7 +113,7 @@ bool serialize_deserialize_kid() {
   ASSERT_EQUAL(k.toys[1], "ball");
   // Now, go the other direction:
   std::string json;
-  ASSERT_SUCCESS(simdjson::builder::to_json_string(k).get(json));
+  ASSERT_SUCCESS(builder::to_json_string(k).get(json));
   std::cout << json << std::endl;
   // Now we parse it back:
   simdjson::ondemand::parser parser2;
@@ -103,11 +126,13 @@ bool serialize_deserialize_kid() {
   ASSERT_EQUAL(k2.toys.size(), 2);
   ASSERT_EQUAL(k2.toys[0], "car");
   ASSERT_EQUAL(k2.toys[1], "ball");
+#endif
   TEST_SUCCEED();
 }
 
 bool serialize_deserialize_x_y_z() {
   TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
   X s1 = {.a = '1',
           .b = 10,
           .c = 0,
@@ -119,7 +144,7 @@ bool serialize_deserialize_x_y_z() {
                 .i = {1, 2, 3},
                 .z = {.x = 1000}}};
   std::string pstr;
-  ASSERT_SUCCESS(simdjson::builder::to_json_string(s1).get(pstr));
+  ASSERT_SUCCESS(builder::to_json_string(s1).get(pstr));
   ASSERT_EQUAL(
       pstr,
       R"({"a":"1","b":10,"c":0,"d":"test string\n\r\"","e":[1,2,3],"f":["ab","cd","fg"],"y":{"g":100,"h":"test string\n\r\"","i":[1,2,3],"z":{"x":1000}}})");
@@ -129,12 +154,23 @@ bool serialize_deserialize_x_y_z() {
   X s2;
   ASSERT_SUCCESS(doc.get<X>().get(s2));
   ASSERT_TRUE(s1 == s2);
+#endif
   TEST_SUCCEED();
 }
 
-bool run() {
-  return car_test() && serialize_deserialize_kid() && serialize_deserialize_x_y_z() && true;
-}
+  bool run() {
+    return
+
+  #if SIMDJSON_EXCEPTIONS
+          car_test_exception() &&
+          car_test_exception2() &&
+          car_test_to_json_exception() &&
+  #endif // SIMDJSON_EXCEPTIONS
+          car_test() &&
+          serialize_deserialize_kid() &&
+          serialize_deserialize_x_y_z() &&
+          true;
+  }
 
 } // namespace builder_tests
 

tests/builder/static_reflection_comprehensive_tests.cpp

@@ -0,0 +1,231 @@
+#include "simdjson.h"
+#include "test_builder.h"
+#include <string>
+#include <string_view>
+#include <vector>
+#include <set>
+#include <map>
+#include <optional>
+#include <memory>
+
+using namespace simdjson;
+
+namespace builder_tests {
+
+  bool test_primitive_types() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct PrimitiveTypes {
+      bool bool_val;
+      char char_val;
+      int int_val;
+      double double_val;
+      float float_val;
+    };
+
+    PrimitiveTypes test{true, 'X', 42, 3.14159, 2.71f};
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"bool_val\":true") != std::string::npos);
+    ASSERT_TRUE(json.find("\"char_val\":\"X\"") != std::string::npos);
+    ASSERT_TRUE(json.find("\"int_val\":42") != std::string::npos);
+    ASSERT_TRUE(json.find("\"double_val\":3.14159") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<PrimitiveTypes>();
+    ASSERT_SUCCESS(get_result);
+
+    PrimitiveTypes deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.bool_val, test.bool_val);
+    ASSERT_EQUAL(deserialized.char_val, test.char_val);
+    ASSERT_EQUAL(deserialized.int_val, test.int_val);
+    ASSERT_EQUAL(deserialized.double_val, test.double_val);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_string_types() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct StringTypes {
+      std::string string_val;
+      std::string_view string_view_val;
+    };
+
+    StringTypes test{"hello world", "test_view"};
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"string_val\":\"hello world\"") != std::string::npos);
+    ASSERT_TRUE(json.find("\"string_view_val\":\"test_view\"") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<StringTypes>();
+    ASSERT_SUCCESS(get_result);
+
+    StringTypes deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.string_val, test.string_val);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_optional_types() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct OptionalTypes {
+      std::optional<int> opt_int_with_value;
+      std::optional<std::string> opt_string_with_value;
+      std::optional<int> opt_int_null;
+      std::optional<std::string> opt_string_null;
+    };
+
+    OptionalTypes test;
+    test.opt_int_with_value = 42;
+    test.opt_string_with_value = "optional_test";
+    test.opt_int_null = std::nullopt;
+    test.opt_string_null = std::nullopt;
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"opt_int_with_value\":42") != std::string::npos);
+    ASSERT_TRUE(json.find("\"opt_string_with_value\":\"optional_test\"") != std::string::npos);
+    ASSERT_TRUE(json.find("\"opt_int_null\":null") != std::string::npos);
+    ASSERT_TRUE(json.find("\"opt_string_null\":null") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<OptionalTypes>();
+    ASSERT_SUCCESS(get_result);
+
+    OptionalTypes deserialized = std::move(get_result.value());
+    ASSERT_TRUE(deserialized.opt_int_with_value.has_value());
+    ASSERT_EQUAL(*deserialized.opt_int_with_value, 42);
+    ASSERT_TRUE(deserialized.opt_string_with_value.has_value());
+    ASSERT_EQUAL(*deserialized.opt_string_with_value, "optional_test");
+    ASSERT_FALSE(deserialized.opt_int_null.has_value());
+    ASSERT_FALSE(deserialized.opt_string_null.has_value());
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_smart_pointer_types() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct SmartPointerTypes {
+      std::unique_ptr<int> unique_int_with_value;
+      std::shared_ptr<std::string> shared_string_with_value;
+      std::unique_ptr<bool> unique_bool_with_value;
+      std::unique_ptr<int> unique_int_null;
+      std::shared_ptr<std::string> shared_string_null;
+    };
+
+    SmartPointerTypes test;
+    test.unique_int_with_value = std::make_unique<int>(123);
+    test.shared_string_with_value = std::make_shared<std::string>("shared_test");
+    test.unique_bool_with_value = std::make_unique<bool>(true);
+    test.unique_int_null = nullptr;
+    test.shared_string_null = nullptr;
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"unique_int_with_value\":123") != std::string::npos);
+    ASSERT_TRUE(json.find("\"shared_string_with_value\":\"shared_test\"") != std::string::npos);
+    ASSERT_TRUE(json.find("\"unique_bool_with_value\":true") != std::string::npos);
+    ASSERT_TRUE(json.find("\"unique_int_null\":null") != std::string::npos);
+    ASSERT_TRUE(json.find("\"shared_string_null\":null") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<SmartPointerTypes>();
+    ASSERT_SUCCESS(get_result);
+
+    SmartPointerTypes deserialized = std::move(get_result.value());
+    ASSERT_TRUE(deserialized.unique_int_with_value != nullptr);
+    ASSERT_EQUAL(*deserialized.unique_int_with_value, 123);
+    ASSERT_TRUE(deserialized.shared_string_with_value != nullptr);
+    ASSERT_EQUAL(*deserialized.shared_string_with_value, "shared_test");
+    ASSERT_TRUE(deserialized.unique_bool_with_value != nullptr);
+    ASSERT_EQUAL(*deserialized.unique_bool_with_value, true);
+    ASSERT_TRUE(deserialized.unique_int_null == nullptr);
+    ASSERT_TRUE(deserialized.shared_string_null == nullptr);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_container_types() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct ContainerTypes {
+      std::vector<int> int_vector;
+      std::set<std::string> string_set;
+      std::map<std::string, int> string_map;
+    };
+
+    ContainerTypes test;
+    test.int_vector = {1, 2, 3, 4, 5};
+    test.string_set = {"apple", "banana", "cherry"};
+    test.string_map = {{"key1", 10}, {"key2", 20}, {"key3", 30}};
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"int_vector\":[1,2,3,4,5]") != std::string::npos);
+    ASSERT_TRUE(json.find("\"string_set\":[") != std::string::npos);
+    ASSERT_TRUE(json.find("\"string_map\":{") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<ContainerTypes>();
+    ASSERT_SUCCESS(get_result);
+
+    ContainerTypes deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.int_vector.size(), 5);
+    ASSERT_EQUAL(deserialized.string_set.size(), 3);
+    ASSERT_EQUAL(deserialized.string_map.size(), 3);
+    ASSERT_EQUAL(deserialized.int_vector[0], 1);
+    ASSERT_EQUAL(deserialized.int_vector[4], 5);
+#endif
+    TEST_SUCCEED();
+  }
+
+
+  bool run() {
+    return test_primitive_types() &&
+           test_string_types() &&
+           test_optional_types() &&
+           test_smart_pointer_types() &&
+           test_container_types();
+  }
+
+} // namespace builder_tests
+
+int main(int argc, char *argv[]) {
+  return test_main(argc, argv, builder_tests::run);
+}

tests/builder/static_reflection_edge_cases_tests.cpp

@@ -0,0 +1,230 @@
+#include "simdjson.h"
+#include "test_builder.h"
+#include <string>
+#include <vector>
+#include <optional>
+#include <memory>
+#include <limits>
+
+using namespace simdjson;
+
+namespace builder_tests {
+
+  bool test_empty_values() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct EmptyValues {
+      std::string empty_string;
+      std::vector<int> empty_vector;
+      std::optional<int> null_optional;
+      std::unique_ptr<int> null_unique_ptr;
+      std::shared_ptr<std::string> null_shared_ptr;
+    };
+
+    EmptyValues test;
+    test.empty_string = "";
+    // empty_vector is already empty by default
+    test.null_optional = std::nullopt;
+    test.null_unique_ptr = nullptr;
+    test.null_shared_ptr = nullptr;
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"empty_string\":\"\"") != std::string::npos);
+    ASSERT_TRUE(json.find("\"empty_vector\":[]") != std::string::npos);
+    ASSERT_TRUE(json.find("\"null_optional\":null") != std::string::npos);
+    ASSERT_TRUE(json.find("\"null_unique_ptr\":null") != std::string::npos);
+    ASSERT_TRUE(json.find("\"null_shared_ptr\":null") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<EmptyValues>();
+    ASSERT_SUCCESS(get_result);
+
+    EmptyValues deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.empty_string, "");
+    ASSERT_EQUAL(deserialized.empty_vector.size(), 0);
+    ASSERT_FALSE(deserialized.null_optional.has_value());
+    ASSERT_TRUE(deserialized.null_unique_ptr == nullptr);
+    ASSERT_TRUE(deserialized.null_shared_ptr == nullptr);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_special_characters() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct SpecialChars {
+      std::string quotes;
+      std::string backslashes;
+      std::string newlines;
+      std::string unicode;
+      char null_char;
+    };
+
+    SpecialChars test;
+    test.quotes = "He said \"Hello\"";
+    test.backslashes = "Path\\to\\file";
+    test.newlines = "Line1\nLine2\tTabbed";
+    test.unicode = "Café résumé";
+    test.null_char = '\0';
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    // Test that quotes are properly escaped
+    ASSERT_TRUE(json.find("\\\"Hello\\\"") != std::string::npos);
+    // Test that backslashes are properly escaped
+    ASSERT_TRUE(json.find("\\\\to\\\\") != std::string::npos);
+    // Test that newlines are properly escaped
+    ASSERT_TRUE(json.find("\\n") != std::string::npos);
+    ASSERT_TRUE(json.find("\\t") != std::string::npos);
+
+    // Test round-trip (excluding null char which has special handling)
+    struct SpecialCharsNoNull {
+      std::string quotes;
+      std::string backslashes;
+      std::string newlines;
+      std::string unicode;
+    };
+
+    SpecialCharsNoNull test_no_null;
+    test_no_null.quotes = test.quotes;
+    test_no_null.backslashes = test.backslashes;
+    test_no_null.newlines = test.newlines;
+    test_no_null.unicode = test.unicode;
+
+    auto result_no_null = builder::to_json_string(test_no_null);
+    ASSERT_SUCCESS(result_no_null);
+
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(result_no_null.value()));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<SpecialCharsNoNull>();
+    ASSERT_SUCCESS(get_result);
+
+    SpecialCharsNoNull deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.quotes, test.quotes);
+    ASSERT_EQUAL(deserialized.backslashes, test.backslashes);
+    ASSERT_EQUAL(deserialized.newlines, test.newlines);
+    ASSERT_EQUAL(deserialized.unicode, test.unicode);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_numeric_limits() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct NumericLimits {
+      int max_int;
+      int min_int;
+      double max_double;
+      double min_double;
+      bool true_val;
+      bool false_val;
+    };
+
+    NumericLimits test;
+    test.max_int = std::numeric_limits<int>::max();
+    test.min_int = std::numeric_limits<int>::min();
+    test.max_double = 1e100;  // Large but safe double value
+    test.min_double = -1e100; // Large negative but safe double value
+    test.true_val = true;
+    test.false_val = false;
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"true_val\":true") != std::string::npos);
+    ASSERT_TRUE(json.find("\"false_val\":false") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<NumericLimits>();
+    ASSERT_SUCCESS(get_result);
+
+    NumericLimits deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.max_int, test.max_int);
+    ASSERT_EQUAL(deserialized.min_int, test.min_int);
+    ASSERT_EQUAL(deserialized.true_val, true);
+    ASSERT_EQUAL(deserialized.false_val, false);
+#endif
+    TEST_SUCCEED();
+  }
+
+  bool test_nested_structures() {
+    TEST_START();
+#if SIMDJSON_STATIC_REFLECTION
+    struct Inner {
+      int value;
+      std::string name;
+    };
+
+    struct Outer {
+      Inner inner_obj;
+      std::vector<Inner> inner_vector;
+      std::optional<Inner> optional_inner;
+      std::unique_ptr<Inner> unique_inner;
+    };
+
+    Outer test;
+    test.inner_obj = {42, "inner"};
+    test.inner_vector = {{1, "first"}, {2, "second"}};
+    test.optional_inner = Inner{99, "optional"};
+    test.unique_inner = std::make_unique<Inner>(Inner{123, "unique"});
+
+    auto result = builder::to_json_string(test);
+    ASSERT_SUCCESS(result);
+
+    std::string json = result.value();
+    ASSERT_TRUE(json.find("\"inner_obj\":{") != std::string::npos);
+    ASSERT_TRUE(json.find("\"inner_vector\":[") != std::string::npos);
+    ASSERT_TRUE(json.find("\"optional_inner\":{") != std::string::npos);
+    ASSERT_TRUE(json.find("\"unique_inner\":{") != std::string::npos);
+
+    // Test round-trip
+    ondemand::parser parser;
+    auto doc_result = parser.iterate(pad(json));
+    ASSERT_SUCCESS(doc_result);
+
+    auto get_result = doc_result.value().get<Outer>();
+    ASSERT_SUCCESS(get_result);
+
+    Outer deserialized = std::move(get_result.value());
+    ASSERT_EQUAL(deserialized.inner_obj.value, 42);
+    ASSERT_EQUAL(deserialized.inner_obj.name, "inner");
+    ASSERT_EQUAL(deserialized.inner_vector.size(), 2);
+    ASSERT_EQUAL(deserialized.inner_vector[0].value, 1);
+    ASSERT_EQUAL(deserialized.inner_vector[1].name, "second");
+    ASSERT_TRUE(deserialized.optional_inner.has_value());
+    ASSERT_EQUAL(deserialized.optional_inner->value, 99);
+    ASSERT_TRUE(deserialized.unique_inner != nullptr);
+    ASSERT_EQUAL(deserialized.unique_inner->value, 123);
+#endif
+    TEST_SUCCEED();
+  }
+
+
+  bool run() {
+    return test_empty_values() &&
+           test_special_characters() &&
+           test_numeric_limits() &&
+           test_nested_structures();
+  }
+
+} // namespace builder_tests
+
+int main(int argc, char *argv[]) {
+  return test_main(argc, argv, builder_tests::run);
+}