Add cycles_total

If numChildren + entryPresent is enough, we don't have to copy even if
it would fit in a smaller node.

If we have to copy, we might as well use the smallest acceptable node
type.

.gitea/workflows/ci.yml

@@ -1,322 +0,0 @@
-name: CI
-
-on: [push, pull_request]
-
-env:
-  CC: clang
-  CXX: clang++
-
-jobs:
-  pre-commit:
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -y git nodejs pre-commit
-
-      - uses: actions/cache@v4
-        with:
-          path: ~/.cache/pre-commit
-          key: pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
-
-      - name: Run pre-commit
-        run: |
-          git config --global --add safe.directory "$PWD"
-          pre-commit run --all-files --show-diff-on-failure
-
-  test:
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - name: 64-bit-versions
-            cmake_args: -DCMAKE_CXX_FLAGS=-DUSE_64_BIT=1
-          - name: debug
-            cmake_args: -DCMAKE_BUILD_TYPE=Debug -DMSAN_TOOLCHAIN_PATH=/opt/msan
-          - name: simd-fallback
-            cmake_args: -DUSE_SIMD_FALLBACK=ON
-          - name: gcc
-            cmake_args: -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install common dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache cmake gcc g++ libc6-dbg \
-            ninja-build python3 valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-
-      - name: Install LLVM toolchain
-        if: matrix.name != 'gcc'
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            clang-21 llvm-21 lld-21 mold
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - name: Download MSan toolchain
-        if: matrix.name == 'debug'
-        run: |
-          curl -Ls "https://minio.weaselab.dev/public/x86_64/msan-toolchain-21.1.8.tar.zst" -o /tmp/msan-toolchain.tar.zst
-          sudo mkdir -p /opt/msan
-          sudo tar --zstd -xf /tmp/msan-toolchain.tar.zst -C /opt/msan
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ${{ matrix.cmake_args }}
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Upload test results to MinIO
-        if: always()
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "minio/jenkins/conflict-set/${{ gitea.run_number }}/${{ matrix.name }}/"
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/${{ matrix.name }}/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-
-  release:
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - runner: ubuntu-latest-amd64
-            arch: amd64
-            cmake_args: -DMSAN_TOOLCHAIN_PATH=/opt/msan
-          - runner: ubuntu-latest-arm64
-            arch: arm64
-            cmake_args: ""
-    runs-on: ${{ matrix.runner }}
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-${{ matrix.arch }}-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache clang-21 cmake devscripts \
-            libc6-dbg llvm-21 lld-21 mold ninja-build rpm \
-            valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - name: Install paper build dependencies
-        if: matrix.arch == 'amd64'
-        run: |
-          sudo apt-get install -y \
-            biber latexmk \
-            texlive-bibtex-extra texlive-fonts-recommended \
-            texlive-latex-extra texlive-pictures
-
-      - uses: actions/cache@v4
-        with:
-          path: .ccache
-          key: ccache-release-${{ matrix.arch }}-${{ gitea.sha }}
-          restore-keys: |
-            ccache-release-${{ matrix.arch }}-
-
-      - name: Download MSan toolchain
-        if: matrix.arch == 'amd64'
-        run: |
-          curl -Ls "https://minio.weaselab.dev/public/x86_64/msan-toolchain-21.1.8.tar.zst" -o /tmp/msan-toolchain.tar.zst
-          sudo mkdir -p /opt/msan
-          sudo tar --zstd -xf /tmp/msan-toolchain.tar.zst -C /opt/msan
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_FLAGS=-DNVALGRIND ${{ matrix.cmake_args }}
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          # On arm64, valgrind needs the MAKE_MEM_DEFINED client requests for
-          # https://git.weaselab.dev/weaselab/conflict-set/issues/39, but this
-          # build has -DNVALGRIND, which compiles them out. Skip valgrind
-          # tests here; they run annotated in the test job.
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 ${{ matrix.arch == 'arm64' && '-E valgrind' || '' }} -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Package
-        run: |
-          cd build
-          cpack -G DEB
-          cpack -G RPM
-
-      - name: Build paper
-        if: matrix.arch == 'amd64'
-        run: |
-          cd paper
-          make
-
-      - name: Upload artifacts to MinIO
-        if: always()
-        shell: bash
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          dest="minio/jenkins/conflict-set/${{ gitea.run_number }}/release-${{ matrix.arch }}/"
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "$dest"
-          if compgen -G "build/*.deb" > /dev/null; then
-            mc cp build/*.deb "$dest"
-          fi
-          if compgen -G "build/*.rpm" > /dev/null; then
-            mc cp build/*.rpm "$dest"
-          fi
-          if compgen -G "paper/*.pdf" > /dev/null; then
-            mc cp paper/*.pdf "$dest"
-          fi
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/release-${{ matrix.arch }}/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-
-  coverage:
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache clang-21 cmake gcovr \
-            libc6-dbg llvm-21 lld-21 mold ninja-build python3 valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - uses: actions/cache@v4
-        with:
-          path: .ccache
-          key: ccache-coverage-${{ gitea.sha }}
-          restore-keys: |
-            ccache-coverage-
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-            -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage \
-            -DCMAKE_BUILD_TYPE=Debug -DDISABLE_TSAN=ON
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Coverage report
-        shell: bash
-        run: |
-          gcov_args=(-f ConflictSet.cpp -f LongestCommonPrefix.h -f Metrics.h
-                     --gcov-executable "llvm-cov gcov" --exclude-noncode-lines)
-          gcovr "${gcov_args[@]}" --cobertura > build/coverage.xml
-          gcovr "${gcov_args[@]}"
-          mkdir -p build/coverage_html
-          gcovr "${gcov_args[@]}" --html-details build/coverage_html/index.html
-          gcovr "${gcov_args[@]}" --fail-under-line 100 > /dev/null
-
-      - name: Upload results to MinIO
-        if: always()
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          dest="minio/jenkins/conflict-set/${{ gitea.run_number }}/coverage/"
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "$dest"
-          if [ -e build/coverage.xml ]; then
-            mc cp build/coverage.xml "$dest"
-          fi
-          if [ -d build/coverage_html ]; then
-            mc cp -r build/coverage_html "$dest"
-          fi
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/coverage/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-          echo "" | tee -a "$GITHUB_STEP_SUMMARY"
-          echo "📊 [Coverage report](https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/coverage/coverage_html/index.html)" | tee -a "$GITHUB_STEP_SUMMARY"

Bench.cpp

@@ -7,6 +7,7 @@
 void showMemory(const ConflictSet &cs);
 #endif
 
+#define ANKERL_NANOBENCH_IMPLEMENT
 #include "third_party/nanobench.h"
 
 constexpr int kNumKeys = 1000000;
@@ -17,48 +18,45 @@ constexpr int kPrefixLen = 0;
 
 constexpr int kMvccWindow = 100000;
 
-TrivialSpan makeKey(Arena &arena, int index) {
+std::span<const uint8_t> makeKey(Arena &arena, int index) {
 
-  uint8_t *buf = new (arena) uint8_t[4 + kPrefixLen];
-  auto result = TrivialSpan{buf, 4 + kPrefixLen};
+  auto result =
+      std::span<uint8_t>{new (arena) uint8_t[4 + kPrefixLen], 4 + kPrefixLen};
   index = __builtin_bswap32(index);
-  memset(buf, 0, kPrefixLen);
-  memcpy(buf, &index, 4);
+  memset(result.data(), 0, kPrefixLen);
+  memcpy(result.data() + kPrefixLen, &index, 4);
 
   return result;
 }
 
-ConflictSet::ReadRange singleton(Arena &arena, TrivialSpan key) {
-  uint8_t *buf = new (arena) uint8_t[key.size() + 1];
-  auto r = TrivialSpan(buf, key.size() + 1);
-  memcpy(buf, key.data(), key.size());
-  buf[key.size()] = 0;
+ConflictSet::ReadRange singleton(Arena &arena, std::span<const uint8_t> key) {
+  auto r =
+      std::span<uint8_t>(new (arena) uint8_t[key.size() + 1], key.size() + 1);
+  memcpy(r.data(), key.data(), key.size());
+  r[key.size()] = 0;
   return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
-ConflictSet::ReadRange prefixRange(Arena &arena, TrivialSpan key) {
-  int index = key.size() - 1;
-  for (; index >= 0; index--)
-    if (key[index] != 255)
+ConflictSet::ReadRange prefixRange(Arena &arena, std::span<const uint8_t> key) {
+  int index;
+  for (index = key.size() - 1; index >= 0; index--)
+    if ((key[index]) != 255)
       break;
 
   // Must not be called with a string that consists only of zero or more '\xff'
-  // bytes, or with an empty string (which has no finite upper bound).
+  // bytes.
   if (index < 0) {
     assert(false);
-    std::abort();
   }
 
-  uint8_t *buf = new (arena) uint8_t[index + 1];
-  auto r = TrivialSpan(buf, index + 1);
-  memcpy(buf, key.data(), index + 1);
-  buf[r.size() - 1]++;
+  auto r = std::span<uint8_t>(new (arena) uint8_t[index + 1], index + 1);
+  memcpy(r.data(), key.data(), index + 1);
+  r[r.size() - 1]++;
   return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
 void benchConflictSet() {
   ankerl::nanobench::Bench bench;
-  bench.minEpochIterations(10000);
   ConflictSet cs{0};
 
   bench.batch(kOpsPerTx);
@@ -84,7 +82,14 @@ void benchConflictSet() {
     ++version;
   }
 
-  auto points = set<TrivialSpan, std::less<>>(arena);
+  // I don't know why std::less didn't work /shrug
+  struct Less {
+    bool operator()(const std::span<const uint8_t> &lhs,
+                    const std::span<const uint8_t> &rhs) const {
+      return lhs < rhs;
+    }
+  };
+  auto points = set<std::span<const uint8_t>, Less>(arena);
 
   while (points.size() < kOpsPerTx * 2 + 1) {
     // TODO don't use rand?
@@ -328,22 +333,16 @@ void benchWorstCaseForRadixRangeRead() {
   auto end = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 255);
   end.push_back(254);
 
-  weaselab::ConflictSet::ReadRange r[] = {
-      {{begin.data(), int(begin.size())}, {end.data(), int(end.size())}, 0},
-  };
-  weaselab::ConflictSet::Result results[sizeof(r) / sizeof(r[0])];
-  for (auto &result : results) {
-    result = weaselab::ConflictSet::TooOld;
-  }
-  bench.batch(sizeof(r) / sizeof(r[0]));
+  weaselab::ConflictSet::Result result;
+  weaselab::ConflictSet::ReadRange r{
+      {begin.data(), int(begin.size())}, {end.data(), int(end.size())}, 0};
 
   bench.run("worst case for radix tree", [&]() {
     for (int i = 0; i < 256; ++i) {
-      cs[i]->check(r, results, sizeof(r) / sizeof(r[0]));
-      for (auto result : results) {
-        if (result != weaselab::ConflictSet::Commit) {
-          abort();
-        }
+      result = weaselab::ConflictSet::TooOld;
+      cs[i]->check(&r, &result, 1);
+      if (result != weaselab::ConflictSet::Commit) {
+        abort();
       }
     }
   });
@@ -362,7 +361,21 @@ void benchWorstCaseForRadixRangeRead() {
 void benchCreateAndDestroy() {
   ankerl::nanobench::Bench bench;
 
-  bench.run("create and destroy", [&]() { ConflictSet cs{0}; });
+  bench.run("create and destroy", [&]() {
+    ConflictSet cs{0};
+    ConflictSet::WriteRange w;
+    uint8_t b[9];
+    b[8] = 0;
+    for (int64_t i = 0; i < 1000; i += 7) {
+      auto x = __builtin_bswap64(i);
+      memcpy(b, &x, 8);
+      w.begin.p = b;
+      w.begin.len = 8;
+      w.end.len = 0;
+      w.end.p = b;
+      cs.addWrites(&w, 1, 1);
+    }
+  });
 }
 
 int main(void) {

CMakeLists.txt

@@ -1,11 +1,11 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.1.0
+  VERSION 0.0.12
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"
-  LANGUAGES C CXX ASM)
+  LANGUAGES C CXX)
 set(CMAKE_CXX_STANDARD 20)
 
 file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/version.txt ${PROJECT_VERSION})
@@ -31,65 +31,22 @@ if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
                                                "MinSizeRel" "RelWithDebInfo")
 endif()
 
-add_compile_options(
-  -Werror=switch-enum
-  -Wswitch-enum
-  -Wunused-variable
-  -fPIC
-  -fdata-sections
-  -ffunction-sections
-  -fno-jump-tables # https://github.com/llvm/llvm-project/issues/54247
-)
-
-if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
-  add_link_options("-Wno-unused-command-line-argument")
-  find_program(LLVM_OBJCOPY llvm-objcopy)
-  if(LLVM_OBJCOPY)
-    set(CMAKE_OBJCOPY
-        ${LLVM_OBJCOPY}
-        CACHE FILEPATH "path to objcopy binary" FORCE)
-  endif()
-endif()
-
-if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-  add_compile_options("-Wno-maybe-uninitialized")
-  add_compile_options("-Wno-maybe-musttail-local-addr")
-endif()
-
+add_compile_options(-fdata-sections -ffunction-sections -Wswitch-enum
+                    -Werror=switch-enum -fPIC)
 if(NOT APPLE)
   # This causes some versions of clang to crash on macos
   add_compile_options(-g -fno-omit-frame-pointer)
 endif()
 
-set(relro_flags "LINKER:-z,relro,-z,now,-z,noexecstack")
-set(full_relro_flags "-pie;${relro_flags}")
+set(full_relro_flags "-pie;LINKER:-z,relro,-z,now,-z,noexecstack")
 cmake_push_check_state()
 list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${full_relro_flags})
 check_cxx_source_compiles("int main(){}" HAS_FULL_RELRO FAIL_REGEX "warning:")
 if(HAS_FULL_RELRO)
-  # -pie only applies to executables; passing it when linking a shared library
-  # makes the driver pull in Scrt1.o, which requires main.
-  add_link_options("$<$<STREQUAL:$<TARGET_PROPERTY:TYPE>,EXECUTABLE>:-pie>"
-                   ${relro_flags})
+  add_link_options(${full_relro_flags})
 endif()
 cmake_pop_check_state()
 
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64 OR CMAKE_SYSTEM_PROCESSOR STREQUAL
-                                              arm64)
-  add_compile_options(-mbranch-protection=standard)
-else()
-  add_compile_options(-fcf-protection)
-  set(rewrite_endbr_flags "-fuse-ld=mold;LINKER:-z,rewrite-endbr")
-  cmake_push_check_state()
-  list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${rewrite_endbr_flags})
-  check_cxx_source_compiles("int main(){}" HAS_REWRITE_ENDBR FAIL_REGEX
-                            "warning:")
-  if(HAS_REWRITE_ENDBR)
-    add_link_options(${rewrite_endbr_flags})
-  endif()
-  cmake_pop_check_state()
-endif()
-
 set(version_script_flags
     LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 cmake_push_check_state()
@@ -103,64 +60,53 @@ option(USE_SIMD_FALLBACK
 
 option(DISABLE_TSAN "Disable TSAN" OFF)
 
-set(MSAN_TOOLCHAIN_PATH
-    ""
-    CACHE
-      PATH
-      "Path to an MSan-instrumented libc++ toolchain. When set, an MSan fuzz_driver target is added. Must contain include/c++/v1 and lib. See build_msan_toolchain.sh to produce one."
-)
-
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
 include_directories(SYSTEM ${CMAKE_CURRENT_SOURCE_DIR}/third_party/valgrind)
 
+add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
+
 if(APPLE)
   add_link_options(-Wl,-dead_strip)
 else()
   add_link_options(-Wl,--gc-sections)
 endif()
 
-if(USE_SIMD_FALLBACK)
-  add_compile_definitions(USE_SIMD_FALLBACK)
-else()
-  if(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64)
+if(EMSCRIPTEN)
+  # https://github.com/emscripten-core/emscripten/issues/15377#issuecomment-1285167486
+  add_link_options(-lnodefs.js -lnoderawfs.js)
+  add_link_options(-s ALLOW_MEMORY_GROWTH)
+endif()
+
+if(NOT USE_SIMD_FALLBACK)
+  cmake_push_check_state()
+  list(APPEND CMAKE_REQUIRED_FLAGS -mavx)
+  check_include_file_cxx("immintrin.h" HAS_AVX)
+  if(HAS_AVX)
     add_compile_options(-mavx)
+    add_compile_definitions(HAS_AVX)
+  endif()
+  cmake_pop_check_state()
+
+  check_include_file_cxx("arm_neon.h" HAS_ARM_NEON)
+  if(HAS_ARM_NEON)
+    add_compile_definitions(HAS_ARM_NEON)
   endif()
 endif()
 
 set(CMAKE_CXX_IMPLICIT_LINK_LIBRARIES "")
 
-# Architecture-specific SIMD assembly. These functions operate on
-# potentially-indeterminate memory, which is UB in C++ but well-defined in
-# assembly.
-set(SIMD_ASM_FILES)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64 AND NOT USE_SIMD_FALLBACK)
-  set(SIMD_ASM_FILES ${CMAKE_CURRENT_SOURCE_DIR}/simd_x86_64.S)
-endif()
-
-add_library(${PROJECT_NAME}-object OBJECT ConflictSet.cpp ${SIMD_ASM_FILES})
+add_library(${PROJECT_NAME}-object OBJECT ConflictSet.cpp)
 target_compile_options(${PROJECT_NAME}-object PRIVATE -fno-exceptions
                                                       -fvisibility=hidden)
 target_include_directories(${PROJECT_NAME}-object
                            PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
-if(NOT LD_EXE)
-  set(LD_EXE ld)
-endif()
-add_custom_command(
-  OUTPUT ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o
-  COMMAND ${LD_EXE} -r $<TARGET_OBJECTS:${PROJECT_NAME}-object> -o
-          ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o
-  DEPENDS $<TARGET_OBJECTS:${PROJECT_NAME}-object>
-  COMMAND_EXPAND_LISTS)
 
-add_library(${PROJECT_NAME} SHARED ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o)
-add_dependencies(${PROJECT_NAME} ${PROJECT_NAME}-object)
+add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 set_target_properties(
   ${PROJECT_NAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                              "${CMAKE_CURRENT_BINARY_DIR}/radix_tree")
-if(CMAKE_BUILD_TYPE STREQUAL Debug)
-  set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE CXX)
-else()
+if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
@@ -170,14 +116,19 @@ if(HAS_VERSION_SCRIPT)
     LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
-add_library(${PROJECT_NAME}-static STATIC ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o)
-add_dependencies(${PROJECT_NAME}-static ${PROJECT_NAME}-object)
-if(CMAKE_BUILD_TYPE STREQUAL Debug)
-  set_target_properties(${PROJECT_NAME}-static PROPERTIES LINKER_LANGUAGE CXX)
-else()
+add_library(${PROJECT_NAME}-static STATIC
+            $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
+if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME}-static PROPERTIES LINKER_LANGUAGE C)
 endif()
-if(NOT APPLE)
+
+if(APPLE)
+  add_custom_command(
+    TARGET ${PROJECT_NAME}-static
+    PRE_LINK
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
+            $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
+else()
   add_custom_command(
     TARGET ${PROJECT_NAME}-static
     POST_BUILD
@@ -193,8 +144,6 @@ include(CTest)
 # disable tests if this is being used through e.g. FetchContent
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
 
-  add_library(nanobench ${CMAKE_CURRENT_SOURCE_DIR}/nanobench.cpp)
-
   set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
   # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
@@ -232,20 +181,16 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
   target_compile_options(driver_skip_list PRIVATE ${TEST_FLAGS})
   target_link_libraries(driver_skip_list PRIVATE skip_list)
 
-  # enable to test skip list
-  if(0)
-    foreach(TEST ${CORPUS_TESTS})
-      get_filename_component(hash ${TEST} NAME)
-      add_test(NAME skip_list_${hash} COMMAND driver_skip_list ${TEST})
-    endforeach()
-  endif()
+  foreach(TEST ${CORPUS_TESTS})
+    get_filename_component(hash ${TEST} NAME)
+    add_test(NAME skip_list_${hash} COMMAND driver_skip_list ${TEST})
+  endforeach()
 
   # ad hoc testing
-  add_executable(conflict_set_main ConflictSet.cpp ${SIMD_ASM_FILES})
+  add_executable(conflict_set_main ConflictSet.cpp)
   target_include_directories(conflict_set_main
                              PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
   target_compile_definitions(conflict_set_main PRIVATE ENABLE_MAIN)
-  target_link_libraries(conflict_set_main PRIVATE nanobench)
 
   if(NOT APPLE)
     # libfuzzer target, to generate/manage corpus
@@ -257,7 +202,7 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
     cmake_pop_check_state()
 
     if(HAS_LIB_FUZZER)
-      add_executable(conflict_set_fuzz_test ConflictSet.cpp ${SIMD_ASM_FILES})
+      add_executable(conflict_set_fuzz_test ConflictSet.cpp)
       target_include_directories(conflict_set_fuzz_test
                                  PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
       target_compile_definitions(conflict_set_fuzz_test PRIVATE ENABLE_FUZZ)
@@ -268,9 +213,8 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
     endif()
   endif()
 
-  # whitebox tests asan+ubsan
-  add_executable(fuzz_driver ConflictSet.cpp FuzzTestDriver.cpp
-                             ${SIMD_ASM_FILES})
+  # whitebox tests
+  add_executable(fuzz_driver ConflictSet.cpp FuzzTestDriver.cpp)
   target_compile_options(fuzz_driver PRIVATE ${TEST_FLAGS})
   if(NOT CMAKE_CROSSCOMPILING)
     target_compile_options(fuzz_driver PRIVATE -fsanitize=address,undefined)
@@ -284,39 +228,9 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
     add_test(NAME conflict_set_fuzz_${hash} COMMAND fuzz_driver ${TEST})
   endforeach()
 
-  # whitebox tests msan
-  if(MSAN_TOOLCHAIN_PATH)
-    add_executable(fuzz_driver_msan ConflictSet.cpp FuzzTestDriver.cpp
-                                    ${SIMD_ASM_FILES})
-    target_compile_options(fuzz_driver_msan PRIVATE ${TEST_FLAGS})
-    if(NOT CMAKE_CROSSCOMPILING)
-      target_compile_options(
-        fuzz_driver_msan
-        PRIVATE -fsanitize=memory -fsanitize-memory-track-origins=2
-                $<$<COMPILE_LANGUAGE:CXX>:-stdlib=libc++>
-                -I${MSAN_TOOLCHAIN_PATH}/include/c++/v1)
-      target_link_options(
-        fuzz_driver_msan
-        PRIVATE
-        -fsanitize=memory
-        -fsanitize-memory-track-origins=2
-        -stdlib=libc++
-        -L${MSAN_TOOLCHAIN_PATH}/lib
-        LINKER:-rpath,${MSAN_TOOLCHAIN_PATH}/lib)
-    endif()
-    target_compile_definitions(fuzz_driver_msan PRIVATE ENABLE_FUZZ)
-    target_include_directories(fuzz_driver_msan
-                               PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
-    foreach(TEST ${CORPUS_TESTS})
-      get_filename_component(hash ${TEST} NAME)
-      add_test(NAME conflict_set_msan_${hash} COMMAND fuzz_driver_msan ${TEST})
-    endforeach()
-  endif()
-
   # tsan tests
   if(NOT CMAKE_CROSSCOMPILING AND NOT DISABLE_TSAN)
-    add_executable(tsan_driver ConflictSet.cpp FuzzTestDriver.cpp
-                               ${SIMD_ASM_FILES})
+    add_executable(tsan_driver ConflictSet.cpp FuzzTestDriver.cpp)
     target_compile_options(tsan_driver PRIVATE ${TEST_FLAGS} -fsanitize=thread)
     target_link_options(tsan_driver PRIVATE -fsanitize=thread)
     target_compile_definitions(tsan_driver PRIVATE ENABLE_FUZZ THREAD_TEST)
@@ -337,19 +251,6 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
     add_test(NAME conflict_set_blackbox_${hash} COMMAND driver ${TEST})
   endforeach()
 
-  find_program(VALGRIND_EXE valgrind)
-  if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
-    list(LENGTH CORPUS_TESTS len)
-    math(EXPR last "${len} - 1")
-    set(partition_size 100)
-    foreach(i RANGE 0 ${last} ${partition_size})
-      list(SUBLIST CORPUS_TESTS ${i} ${partition_size} partition)
-      add_test(NAME conflict_set_blackbox_valgrind_${i}
-               COMMAND ${VALGRIND_EXE} --error-exitcode=99 --
-                       $<TARGET_FILE:driver> ${partition})
-    endforeach()
-  endif()
-
   # scripted tests. Written manually to fill in anything libfuzzer couldn't
   # find.
   if(NOT CMAKE_CROSSCOMPILING)
@@ -370,17 +271,16 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
           ${Python3_EXECUTABLE}
           ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
           --build-dir ${CMAKE_CURRENT_BINARY_DIR})
-      if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
-        add_test(
-          NAME script_test_${TEST}_valgrind
-          COMMAND
-            ${VALGRIND_EXE} ${Python3_EXECUTABLE}
-            ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
-            --build-dir ${CMAKE_CURRENT_BINARY_DIR})
-      endif()
     endforeach()
   endif()
 
+  find_program(VALGRIND_EXE valgrind)
+  if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
+    add_test(NAME conflict_set_blackbox_valgrind
+             COMMAND ${VALGRIND_EXE} --error-exitcode=99 --
+                     $<TARGET_FILE:driver> ${CORPUS_TESTS})
+  endif()
+
   # api smoke tests
 
   # c90
@@ -428,39 +328,9 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
         ${symbol_imports})
   endif()
 
-  if(NOT CMAKE_CROSSCOMPILING)
-    find_program(HARDENING_CHECK hardening-check)
-    if(HARDENING_CHECK)
-      # Not all versions of hardening-check support the same options, so query
-      # the help output before using architecture-specific skips.
-      execute_process(
-        COMMAND ${HARDENING_CHECK} --help
-        OUTPUT_VARIABLE hardening_check_help
-        ERROR_VARIABLE hardening_check_help
-        OUTPUT_STRIP_TRAILING_WHITESPACE ERROR_STRIP_TRAILING_WHITESPACE)
-      set(hardening_check_arch_flags "")
-      # Control flow integrity (CET) is x86-only and branch protection (PAC/BTI)
-      # is arm64-only, so ignore whichever doesn't apply.
-      if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64 OR CMAKE_SYSTEM_PROCESSOR
-                                                    STREQUAL arm64)
-        if(hardening_check_help MATCHES "nocfprotection")
-          list(APPEND hardening_check_arch_flags --nocfprotection)
-        endif()
-      else()
-        if(hardening_check_help MATCHES "nobranchprotection")
-          list(APPEND hardening_check_arch_flags --nobranchprotection)
-        endif()
-      endif()
-      add_test(
-        NAME hardening_check
-        COMMAND ${HARDENING_CHECK} $<TARGET_FILE:${PROJECT_NAME}> --nofortify
-                --nostackprotector ${hardening_check_arch_flags})
-    endif()
-  endif()
-
   # bench
   add_executable(conflict_set_bench Bench.cpp)
-  target_link_libraries(conflict_set_bench PRIVATE ${PROJECT_NAME} nanobench)
+  target_link_libraries(conflict_set_bench PRIVATE ${PROJECT_NAME})
   set_target_properties(conflict_set_bench PROPERTIES SKIP_BUILD_RPATH ON)
   add_executable(real_data_bench RealDataBench.cpp)
   target_link_libraries(real_data_bench PRIVATE ${PROJECT_NAME})
@@ -475,14 +345,6 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
   add_executable(server_bench ServerBench.cpp)
   target_link_libraries(server_bench PRIVATE ${PROJECT_NAME})
   set_target_properties(server_bench PROPERTIES SKIP_BUILD_RPATH ON)
-
-  add_executable(interleaving_test InterleavingTest.cpp)
-  # work around lack of musttail for gcc
-  if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_BUILD_TYPE STREQUAL "Debug")
-    target_compile_options(interleaving_test PRIVATE -Og
-                                                     -foptimize-sibling-calls)
-  endif()
-  target_link_libraries(interleaving_test PRIVATE nanobench)
 endif()
 
 # packaging

Dockerfile

@@ -0,0 +1,51 @@
+FROM ubuntu:rolling
+
+WORKDIR /tmp
+ENV HOME=/tmp
+RUN chmod -R 777 /tmp
+
+# Install apt dependencies
+RUN apt-get update
+RUN apt-get upgrade -y
+RUN TZ=America/Los_Angeles DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    build-essential \
+    ccache \
+    clang \
+    cmake \
+    curl \
+    doxygen \
+    file \
+    g++-aarch64-linux-gnu \
+    gcovr \
+    git \
+    gperf \
+    graphviz \
+    libc6-dbg \
+    ninja-build \
+    pre-commit \
+    python3-requests \
+    qemu-user \
+    rpm \
+    texlive-full \
+    zstd
+
+# Install recent valgrind from source
+RUN curl -Ls https://sourceware.org/pub/valgrind/valgrind-3.22.0.tar.bz2 -o valgrind.tar.bz2 && \
+    echo "c811db5add2c5f729944caf47c4e7a65dcaabb9461e472b578765dd7bf6d2d4c  valgrind.tar.bz2" > valgrind-sha.txt && \
+    sha256sum --quiet -c valgrind-sha.txt && \
+    mkdir valgrind && \
+    tar --strip-components 1 --no-same-owner --no-same-permissions --directory valgrind -xjf valgrind.tar.bz2 && \
+    cd valgrind && \
+    ./configure --enable-only64bit --enable-lto && \
+    make -j`nproc` && \
+    make install && \
+    cd .. && \
+    rm -rf /tmp/*
+
+# Set after building valgrind, which doesn't build with clang for some reason
+ENV CC=clang
+ENV CXX=clang++
+
+# Try to have all the pre-commit hooks we'll need already initialized
+COPY .pre-commit-config.yaml /tmp/
+RUN git init && pre-commit install-hooks

HashTable.cpp

@@ -96,9 +96,7 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
   return impl->setOldestVersion(oldestVersion);
 }
 
-// The hash_table implementation does not track memory usage, so return 0 to
-// satisfy the API contract that getBytes() returns a non-negative value.
-int64_t ConflictSet::getBytes() const { return 0; }
+int64_t ConflictSet::getBytes() const { return -1; }
 
 void ConflictSet::getMetricsV1(MetricsV1 **metrics, int *count) const {
   *metrics = nullptr;
@@ -121,13 +119,7 @@ ConflictSet::ConflictSet(ConflictSet &&other) noexcept
     : impl(std::exchange(other.impl, nullptr)) {}
 
 ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
-  if (this != &other) {
-    if (impl) {
-      impl->~Impl();
-      safe_free(impl, sizeof(Impl));
-    }
-    impl = std::exchange(other.impl, nullptr);
-  }
+  impl = std::exchange(other.impl, nullptr);
   return *this;
 }
 
@@ -163,10 +155,7 @@ __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
 }
 __attribute__((__visibility__("default"))) int64_t
 ConflictSet_getBytes(void *cs) {
-  (void)cs;
-  // The hash_table implementation does not track memory usage, so return 0 to
-  // satisfy the API contract that ConflictSet_getBytes returns a non-negative
-  // value.
-  return 0;
+  using Impl = ConflictSet::Impl;
+  return -1;
 }
 }

InterleavingTest.cpp

@@ -1,256 +0,0 @@
-#include <alloca.h>
-#include <cassert>
-#ifdef __x86_64__
-#include <immintrin.h>
-#endif
-
-#include "third_party/nanobench.h"
-
-struct Job {
-  int *input;
-  // Returned void* is a function pointer to the next continuation. We have to
-  // use void* because otherwise the type would be recursive.
-  typedef void *(*continuation)(Job *);
-  continuation next;
-};
-
-void *stepJob(Job *j) {
-  auto done = --(*j->input) == 0;
-#ifdef __x86_64__
-  _mm_clflush(j->input);
-#endif
-  return done ? nullptr : (void *)stepJob;
-}
-
-void sequential(Job **jobs, int count) {
-  for (int i = 0; i < count; ++i) {
-    do {
-      jobs[i]->next = (Job::continuation)jobs[i]->next(jobs[i]);
-    } while (jobs[i]->next);
-  }
-}
-
-void sequentialNoFuncPtr(Job **jobs, int count) {
-  for (int i = 0; i < count; ++i) {
-    while (stepJob(jobs[i]))
-      ;
-  }
-}
-
-void interleaveSwapping(Job **jobs, int remaining) {
-  int current = 0;
-  while (remaining > 0) {
-    auto next = (Job::continuation)jobs[current]->next(jobs[current]);
-    jobs[current]->next = next;
-    if (next == nullptr) {
-      jobs[current] = jobs[remaining - 1];
-      --remaining;
-    } else {
-      ++current;
-    }
-    if (current == remaining) {
-      current = 0;
-    }
-  }
-}
-
-void interleaveBoundedCyclicList(Job **jobs, int count) {
-  if (count == 0) {
-    return;
-  }
-
-  constexpr int kConcurrent = 32;
-  Job *inProgress[kConcurrent];
-  int nextJob[kConcurrent];
-
-  int started = std::min(kConcurrent, count);
-  for (int i = 0; i < started; i++) {
-    inProgress[i] = jobs[i];
-    nextJob[i] = i + 1;
-  }
-  nextJob[started - 1] = 0;
-
-  int prevJob = started - 1;
-  int job = 0;
-  for (;;) {
-    auto next = (Job::continuation)inProgress[job]->next(inProgress[job]);
-    inProgress[job]->next = next;
-    if (next == nullptr) {
-      if (started == count) {
-        if (prevJob == job)
-          break;
-        nextJob[prevJob] = nextJob[job];
-        job = prevJob;
-      } else {
-        int temp = started++;
-        inProgress[job] = jobs[temp];
-      }
-    }
-    prevJob = job;
-    job = nextJob[job];
-  }
-}
-
-#ifndef __has_attribute
-#define __has_attribute(x) 0
-#endif
-
-#if __has_attribute(musttail)
-#define MUSTTAIL __attribute__((musttail))
-#else
-#define MUSTTAIL
-#endif
-
-struct Context {
-  constexpr static int kConcurrent = 32;
-  Job **jobs;
-  Job *inProgress[kConcurrent];
-  void (*continuation[kConcurrent])(Context *, int64_t prevJob, int64_t job,
-                                    int64_t started, int64_t count);
-  int nextJob[kConcurrent];
-};
-
-void keepGoing(Context *context, int64_t prevJob, int64_t job, int64_t started,
-               int64_t count) {
-  prevJob = job;
-  job = context->nextJob[job];
-  MUSTTAIL return context->continuation[job](context, prevJob, job, started,
-                                             count);
-}
-
-void stepJobTailCall(Context *context, int64_t prevJob, int64_t job,
-                     int64_t started, int64_t count);
-
-void complete(Context *context, int64_t prevJob, int64_t job, int64_t started,
-              int64_t count) {
-  if (started == count) {
-    if (prevJob == job) {
-      return;
-    }
-    context->nextJob[prevJob] = context->nextJob[job];
-    job = prevJob;
-  } else {
-    context->inProgress[job] = context->jobs[started++];
-    context->continuation[job] = stepJobTailCall;
-  }
-  prevJob = job;
-  job = context->nextJob[job];
-  MUSTTAIL return context->continuation[job](context, prevJob, job, started,
-                                             count);
-}
-
-void stepJobTailCall(Context *context, int64_t prevJob, int64_t job,
-                     int64_t started, int64_t count) {
-  auto *j = context->inProgress[job];
-  auto done = --(*j->input) == 0;
-#ifdef __x86_64__
-  _mm_clflush(j->input);
-#endif
-  if (done) {
-    MUSTTAIL return complete(context, prevJob, job, started, count);
-  } else {
-    context->continuation[job] = stepJobTailCall;
-    MUSTTAIL return keepGoing(context, prevJob, job, started, count);
-  }
-}
-
-void useTailCalls(Job **jobs, int count) {
-  if (count == 0) {
-    return;
-  }
-  Context context;
-  context.jobs = jobs;
-  int64_t started = std::min(Context::kConcurrent, count);
-  for (int i = 0; i < started; i++) {
-    context.inProgress[i] = jobs[i];
-    context.nextJob[i] = i + 1;
-    context.continuation[i] = stepJobTailCall;
-  }
-  context.nextJob[started - 1] = 0;
-  int prevJob = started - 1;
-  int job = 0;
-  return context.continuation[job](&context, prevJob, job, started, count);
-}
-
-void interleaveCyclicList(Job **jobs, int count) {
-  auto *nextJob = (int *)alloca(sizeof(int) * count);
-
-  for (int i = 0; i < count - 1; ++i) {
-    nextJob[i] = i + 1;
-  }
-  nextJob[count - 1] = 0;
-
-  int prevJob = count - 1;
-  int job = 0;
-  for (;;) {
-    auto next = (Job::continuation)jobs[job]->next(jobs[job]);
-    jobs[job]->next = next;
-    if (next == nullptr) {
-      if (prevJob == job)
-        break;
-      nextJob[prevJob] = nextJob[job];
-      job = prevJob;
-    }
-    prevJob = job;
-    job = nextJob[job];
-  }
-}
-
-int main() {
-  ankerl::nanobench::Bench bench;
-
-  constexpr int kNumJobs = 10000;
-  bench.relative(true);
-
-  Job jobs[kNumJobs];
-  Job jobsCopy[kNumJobs];
-  int iters = 0;
-  int originalInput[kNumJobs];
-  for (int i = 0; i < kNumJobs; ++i) {
-    originalInput[i] = rand() % 5 + 3;
-    jobs[i].input = new int{originalInput[i]};
-    jobs[i].next = stepJob;
-    iters += *jobs[i].input;
-  }
-  bench.batch(iters);
-
-  for (auto [scheduler, name] :
-       {std::make_pair(sequentialNoFuncPtr, "sequentialNoFuncPtr"),
-        std::make_pair(sequential, "sequential"),
-        std::make_pair(useTailCalls, "useTailCalls"),
-        std::make_pair(interleaveSwapping, "interleavingSwapping"),
-        std::make_pair(interleaveBoundedCyclicList,
-                       "interleaveBoundedCyclicList"),
-        std::make_pair(interleaveCyclicList, "interleaveCyclicList")}) {
-    for (int i = 0; i < kNumJobs; ++i) {
-      *jobs[i].input = originalInput[i];
-    }
-    memcpy(jobsCopy, jobs, sizeof(jobs));
-    Job *ps[kNumJobs];
-    for (int i = 0; i < kNumJobs; ++i) {
-      ps[i] = jobsCopy + i;
-    }
-    scheduler(ps, kNumJobs);
-    for (int i = 0; i < kNumJobs; ++i) {
-      if (*jobsCopy[i].input != 0) {
-        fprintf(stderr, "%s failed\n", name);
-        abort();
-      }
-    }
-
-    bench.run(name, [&]() {
-      for (int i = 0; i < kNumJobs; ++i) {
-        *jobs[i].input = originalInput[i];
-      }
-      memcpy(jobsCopy, jobs, sizeof(jobs));
-      Job *ps[kNumJobs];
-      for (int i = 0; i < kNumJobs; ++i) {
-        ps[i] = jobsCopy + i;
-      }
-      scheduler(ps, kNumJobs);
-    });
-  }
-  for (int i = 0; i < kNumJobs; ++i) {
-    delete jobs[i].input;
-  }
-}

Internal.h

@@ -18,45 +18,18 @@ using namespace weaselab;
 #include <span>
 #include <string>
 #include <thread>
+#include <unordered_set>
 #include <utility>
+#include <vector>
 
 #include <callgrind.h>
 
 #define DEBUG_VERBOSE 0
 #define SHOW_MEMORY 0
 
-// std::span is not trivially constructible. We want a span that leaves its
-// members uninitialized for performance reasons.
-struct TrivialSpan {
-  TrivialSpan() = default;
-  TrivialSpan(const uint8_t *begin, int len) : begin(begin), len(len) {}
-
-  uint8_t back() const {
-    assert(len > 0);
-    return begin[len - 1];
-  }
-  uint8_t front() const {
-    assert(len > 0);
-    return begin[0];
-  }
-  uint8_t operator[](int i) const {
-    assert(0 <= i);
-    assert(i < len);
-    return begin[i];
-  }
-  int size() const { return len; }
-  TrivialSpan subspan(int offset, int len) { return {begin + offset, len}; }
-  const uint8_t *data() const { return begin; }
-
-private:
-  const uint8_t *begin;
-  int len;
-};
-
-static_assert(std::is_trivial_v<TrivialSpan>);
-
-[[nodiscard]] inline auto operator<=>(const TrivialSpan &lhs,
-                                      const TrivialSpan &rhs) noexcept {
+[[nodiscard]] inline auto
+operator<=>(const std::span<const uint8_t> &lhs,
+            const std::span<const uint8_t> &rhs) noexcept {
   int cl = std::min<int>(lhs.size(), rhs.size());
   if (cl > 0) {
     if (auto c = memcmp(lhs.data(), rhs.data(), cl) <=> 0; c != 0) {
@@ -66,7 +39,7 @@ static_assert(std::is_trivial_v<TrivialSpan>);
   return lhs.size() <=> rhs.size();
 }
 
-[[nodiscard]] inline auto operator<=>(const TrivialSpan &lhs,
+[[nodiscard]] inline auto operator<=>(const std::span<const uint8_t> &lhs,
                                       const ConflictSet::Key &rhs) noexcept {
   int cl = std::min<int>(lhs.size(), rhs.len);
   if (cl > 0) {
@@ -74,18 +47,7 @@ static_assert(std::is_trivial_v<TrivialSpan>);
       return c;
     }
   }
-  return lhs.size() <=> rhs.len;
-}
-
-[[nodiscard]] inline auto operator<=>(const ConflictSet::Key &lhs,
-                                      const ConflictSet::Key &rhs) noexcept {
-  int cl = std::min<int>(lhs.len, rhs.len);
-  if (cl > 0) {
-    if (auto c = memcmp(lhs.p, rhs.p, cl) <=> 0; c != 0) {
-      return c;
-    }
-  }
-  return lhs.len <=> rhs.len;
+  return lhs.size() <=> size_t(rhs.len);
 }
 
 // This header contains code that we want to reuse outside of ConflictSet.cpp or
@@ -367,6 +329,23 @@ template <class T, class C = std::less<T>> auto set(Arena &arena) {
   return Set<T, C>(ArenaAlloc<T>(&arena));
 }
 
+template <class T> struct MyHash;
+
+template <class T> struct MyHash<T *> {
+  size_t operator()(const T *t) const noexcept {
+    size_t result;
+    memcpy(&result, &t, sizeof(result));
+    return result;
+  }
+};
+
+template <class T>
+using HashSet =
+    std::unordered_set<T, MyHash<T>, std::equal_to<T>, ArenaAlloc<T>>;
+template <class T> auto hashSet(Arena &arena) {
+  return HashSet<T>(ArenaAlloc<T>(&arena));
+}
+
 template <class T, class U>
 bool operator==(const ArenaAlloc<T> &lhs, const ArenaAlloc<U> &rhs) {
   return lhs.arena == rhs.arena;
@@ -560,7 +539,7 @@ struct ReferenceImpl {
 
 using Key = ConflictSet::Key;
 
-inline Key operator""_s(const char *str, size_t size) {
+inline Key operator"" _s(const char *str, size_t size) {
   return {reinterpret_cast<const uint8_t *>(str), int(size)};
 }
 
@@ -591,7 +570,7 @@ inline std::string printable(const Key &key) {
   return printable(std::string_view((const char *)key.p, key.len));
 }
 
-inline std::string printable(TrivialSpan key) {
+inline std::string printable(std::span<const uint8_t> key) {
   return printable(std::string_view((const char *)key.data(), key.size()));
 }
 
@@ -699,8 +678,10 @@ struct TestDriver {
             arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
             writes[i].end.len = keyLen;
             writes[i].end.p = begin;
-            auto c = TrivialSpan(writes[i].begin.p, writes[i].begin.len) <=>
-                     TrivialSpan(writes[i].end.p, writes[i].end.len);
+            auto c =
+                std::span<const uint8_t>(writes[i].begin.p,
+                                         writes[i].begin.len) <=>
+                std::span<const uint8_t>(writes[i].end.p, writes[i].end.len);
             if (c > 0) {
               using std::swap;
               swap(writes[i].begin, writes[i].end);
@@ -767,10 +748,7 @@ struct TestDriver {
       fprintf(stderr, "%p Set oldest version: %" PRId64 "\n", this,
               oldestVersion);
 #endif
-      CALLGRIND_START_INSTRUMENTATION;
       cs.setOldestVersion(oldestVersion);
-      CALLGRIND_STOP_INSTRUMENTATION;
-
       if constexpr (kEnableAssertions) {
         refImpl.setOldestVersion(oldestVersion);
       }

Jenkinsfile

@@ -0,0 +1,140 @@
+def CleanBuildAndTest(String cmakeArgs) {
+    sh """
+    export CCACHE_DIR=/ccache
+    rm -rf build
+    mkdir build
+    cd build
+    cmake .. -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ${cmakeArgs} 
+    ninja
+    ccache -s
+    """
+    catchError {
+        sh '''
+        cd build
+        ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 -T Test -j `nproc` --timeout 90
+        zstd Testing/*/Test.xml
+        '''
+    }
+    xunit tools: [CTest(pattern: 'build/Testing/*/Test.xml')], reduceLog: false, skipPublishingChecks: false
+    minio bucket: 'jenkins', credentialsId: 'jenkins-minio', excludes: '', host: 'minio.weaselab.dev', includes: 'build/Testing/*/Test.xml.zst', targetFolder: '${JOB_NAME}/${BUILD_NUMBER}/${STAGE_NAME}/'
+}
+
+pipeline {
+    agent any
+    stages {
+        stage('Pre-commit') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                script {
+                    env.HOME = env.WORKSPACE
+                }
+                sh 'pre-commit run --all-files --show-diff-on-failure'
+            }
+        }
+        stage('Clang') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("")
+                recordIssues(tools: [clang()])
+            }
+        }
+        stage('Debug') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DCMAKE_BUILD_TYPE=Debug")
+            }
+        }
+        stage('SIMD fallback') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
+            }
+        }
+        stage('Release [gcc]') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_CXX_FLAGS=-DNVALGRIND")
+                recordIssues(tools: [gcc()])
+                sh '''
+                cd build
+                cpack -G DEB
+                cpack -G RPM
+                '''
+                sh '''
+                cd paper
+                make
+                '''
+                minio bucket: 'jenkins', credentialsId: 'jenkins-minio', excludes: '', host: 'minio.weaselab.dev', includes: 'build/*.deb,build/*.rpm,paper/*.pdf', targetFolder: '${JOB_NAME}/${BUILD_NUMBER}/${STAGE_NAME}/'
+            }
+        }
+        stage('Release [gcc,aarch64]') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DCMAKE_TOOLCHAIN_FILE=../aarch64-toolchain.cmake -DCMAKE_CXX_FLAGS=-DNVALGRIND")
+                sh '''
+                cd build
+                cpack -G DEB
+                cpack -G RPM
+                '''
+                minio bucket: 'jenkins', credentialsId: 'jenkins-minio', excludes: '', host: 'minio.weaselab.dev', includes: 'build/*.deb,build/*.rpm', targetFolder: '${JOB_NAME}/${BUILD_NUMBER}/${STAGE_NAME}'
+            }
+        }
+        stage('Coverage') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                script {
+                    filter_args = "-f ConflictSet.cpp -f LongestCommonPrefix.h"
+                }
+                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug -DDISABLE_TSAN=ON")
+                sh """
+                gcovr ${filter_args} --cobertura > build/coverage.xml
+                """
+                recordCoverage qualityGates: [[criticality: 'NOTE', metric: 'MODULE']], tools: [[parser: 'COBERTURA', pattern: 'build/coverage.xml']]
+                sh """
+                gcovr ${filter_args}
+                gcovr ${filter_args} --fail-under-line 100 > /dev/null
+                """
+            }
+        }
+    }
+    post {
+        always {
+            emailext mimeType: 'text/html', body: '${SCRIPT, template="groovy-html.template"}', subject: "${env.JOB_NAME} - Build# ${env.BUILD_NUMBER} - ${currentBuild.currentResult}", to: 'andrew@weaselab.dev'
+        }
+    }
+}

LongestCommonPrefix.h

@@ -129,7 +129,7 @@ longestCommonPrefix(const uint8_t *ap, const uint8_t *bp, int cl) {
   }
 
   int i = 0;
-  int end; // GCOVR_EXCL_LINE
+  int end;
 
   // kStride * kUnrollCount at a time
   end = cl & ~(kStride * kUnrollFactor - 1);

Metrics.h

@@ -1,64 +0,0 @@
-#pragma once
-
-#include "ConflictSet.h"
-#include "Internal.h"
-
-#include <assert.h>
-#include <atomic>
-#include <tuple>
-
-struct Metric {
-  Metric *prev;
-  const char *name;
-  const char *help;
-  weaselab::ConflictSet::MetricsV1::Type type;
-  std::atomic<int64_t> value;
-
-protected:
-  Metric(Metric *&metricList, int &metricsCount, const char *name,
-         const char *help, weaselab::ConflictSet::MetricsV1::Type type)
-      : prev(std::exchange(metricList, this)), name(name), help(help),
-        type(type), value(0) {
-    ++metricsCount;
-  }
-};
-
-struct Gauge : private Metric {
-  Gauge(Metric *&metricList, int &metricsCount, const char *name,
-        const char *help)
-      : Metric(metricList, metricsCount, name, help,
-               weaselab::ConflictSet::MetricsV1::Gauge) {}
-
-  void set(int64_t value) {
-    this->value.store(value, std::memory_order_relaxed);
-  }
-};
-
-struct Counter : private Metric {
-  Counter(Metric *&metricList, int &metricsCount, const char *name,
-          const char *help)
-      : Metric(metricList, metricsCount, name, help,
-               weaselab::ConflictSet::MetricsV1::Counter) {}
-  // Expensive. Accumulate locally and then call add instead of repeatedly
-  // calling add.
-  void add(int64_t value) {
-    assert(value >= 0);
-    static_assert(std::atomic<int64_t>::is_always_lock_free);
-    this->value.fetch_add(value, std::memory_order_relaxed);
-  }
-};
-
-inline weaselab::ConflictSet::MetricsV1 *initMetrics(Metric *metricsList,
-                                                     int metricsCount) {
-  weaselab::ConflictSet::MetricsV1 *metrics =
-      (weaselab::ConflictSet::MetricsV1 *)safe_malloc(metricsCount *
-                                                      sizeof(metrics[0]));
-  for (auto [i, m] = std::make_tuple(metricsCount - 1, metricsList); i >= 0;
-       --i, m = m->prev) {
-    metrics[i].name = m->name;
-    metrics[i].help = m->help;
-    metrics[i].p = m;
-    metrics[i].type = m->type;
-  }
-  return metrics;
-}

README.md

@@ -2,16 +2,7 @@ A data structure for optimistic concurrency control on ranges of bitwise-lexicog
 
 Intended as an alternative to FoundationDB's skip list.
 
-Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM.
-
-```
-$ clang++ --version
-
-Ubuntu clang version 21.1.8 (6ubuntu1)
-Target: x86_64-pc-linux-gnu
-Thread model: posix
-InstalledDir: /usr/lib/llvm-21/bin
-```
+Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM
 
 # Microbenchmark
 
@@ -19,45 +10,45 @@ InstalledDir: /usr/lib/llvm-21/bin
 
 |               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
-|              164.29 |        6,086,873.38 |    0.0% |        3,107.03 |          604.19 |  5.142 |         558.59 |    0.0% |      1.96 | `point reads`
-|              161.05 |        6,209,395.38 |    0.1% |        3,036.76 |          592.21 |  5.128 |         539.35 |    0.0% |      1.93 | `prefix reads`
-|              239.55 |        4,174,539.38 |    0.1% |        3,722.71 |          880.68 |  4.227 |         692.00 |    0.0% |      2.86 | `range reads`
-|              354.75 |        2,818,919.14 |    0.7% |        4,523.64 |        1,304.75 |  3.467 |         720.22 |    2.0% |      4.23 | `point writes`
-|              345.32 |        2,895,878.47 |    0.1% |        4,484.57 |        1,270.31 |  3.530 |         705.00 |    1.8% |      4.12 | `prefix writes`
-|              193.48 |        5,168,547.42 |    0.1% |        2,224.10 |          711.72 |  3.125 |         377.17 |    3.3% |      2.32 | `range writes`
-|              404.89 |        2,469,777.50 |    2.4% |        6,855.96 |        1,489.70 |  4.602 |       1,227.82 |    1.3% |      0.05 | `monotonic increasing point writes`
-|          134,231.80 |            7,449.80 |    1.9% |      812,045.25 |      495,770.40 |  1.638 |     151,246.50 |    0.9% |      0.01 | `worst case for radix tree`
-|               37.80 |       26,454,311.17 |    0.4% |          701.00 |          139.14 |  5.038 |         102.00 |    0.0% |      0.01 | `create and destroy`
+|              172.03 |        5,812,791.77 |    0.4% |        3,130.62 |          879.00 |  3.562 |         509.23 |    0.0% |      0.01 | `point reads`
+|              167.44 |        5,972,130.71 |    0.2% |        3,065.14 |          862.27 |  3.555 |         494.30 |    0.0% |      0.01 | `prefix reads`
+|              238.77 |        4,188,130.84 |    0.9% |        3,589.93 |        1,259.30 |  2.851 |         637.12 |    0.0% |      0.01 | `range reads`
+|              424.01 |        2,358,426.70 |    0.2% |        5,620.05 |        2,242.35 |  2.506 |         854.80 |    1.7% |      0.01 | `point writes`
+|              418.45 |        2,389,780.56 |    0.4% |        5,525.07 |        2,211.05 |  2.499 |         831.71 |    1.7% |      0.01 | `prefix writes`
+|              254.87 |        3,923,568.88 |    2.6% |        3,187.01 |        1,366.50 |  2.332 |         529.11 |    2.7% |      0.02 | `range writes`
+|              675.96 |        1,479,374.50 |    3.3% |        7,735.41 |        3,468.60 |  2.230 |       1,386.02 |    1.8% |      0.01 | `monotonic increasing point writes`
+|          137,986.20 |            7,247.10 |    0.6% |      789,752.33 |      699,462.00 |  1.129 |     144,824.14 |    0.0% |      0.01 | `worst case for radix tree`
+|               21.63 |       46,231,564.03 |    1.0% |          448.00 |          107.14 |  4.181 |          84.00 |    0.0% |      0.01 | `create and destroy`
 
 ## Radix tree (this implementation)
 
-
 |               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
-|               12.89 |       77,565,115.56 |    0.1% |          244.55 |           47.43 |  5.155 |          34.21 |    0.6% |      0.15 | `point reads`
-|               15.11 |       66,162,047.76 |    0.1% |          297.79 |           55.60 |  5.356 |          43.23 |    0.4% |      0.18 | `prefix reads`
-|               36.29 |       27,559,358.29 |    0.1% |          783.16 |          133.44 |  5.869 |         109.52 |    0.2% |      0.43 | `range reads`
-|               20.53 |       48,719,405.55 |    0.1% |          381.81 |           75.51 |  5.057 |          51.04 |    0.5% |      0.25 | `point writes`
-|               39.37 |       25,402,042.40 |    0.1% |          685.00 |          144.83 |  4.730 |         106.72 |    0.3% |      0.47 | `prefix writes`
-|               43.78 |       22,843,841.63 |    0.1% |          800.40 |          161.06 |  4.970 |         127.36 |    0.1% |      0.53 | `range writes`
-|               78.37 |       12,760,008.75 |    1.0% |        1,452.61 |          288.24 |  5.040 |         278.69 |    0.1% |      0.01 | `monotonic increasing point writes`
-|          322,885.50 |            3,097.07 |    1.5% |    4,362,382.00 |    1,183,852.00 |  3.685 |     765,301.00 |    0.1% |      0.01 | `worst case for radix tree`
-|               99.99 |       10,000,718.79 |    0.4% |        1,775.00 |          367.93 |  4.824 |         288.00 |    0.0% |      0.01 | `create and destroy`
+|               10.80 |       92,600,541.52 |    0.6% |          180.38 |           54.49 |  3.310 |          41.51 |    0.4% |      0.01 | `point reads`
+|               15.00 |       66,687,691.68 |    0.4% |          278.44 |           76.44 |  3.642 |          55.56 |    0.3% |      0.01 | `prefix reads`
+|               36.81 |       27,163,394.61 |    0.4% |          795.06 |          187.91 |  4.231 |         142.67 |    0.2% |      0.01 | `range reads`
+|               18.14 |       55,137,674.01 |    1.2% |          338.19 |           92.86 |  3.642 |          42.81 |    0.4% |      0.01 | `point writes`
+|               33.19 |       30,127,119.71 |    0.1% |          681.03 |          170.05 |  4.005 |          98.68 |    0.2% |      0.01 | `prefix writes`
+|               37.37 |       26,759,432.70 |    1.9% |          779.70 |          195.45 |  3.989 |         114.21 |    0.0% |      0.01 | `range writes`
+|               74.36 |       13,448,582.47 |    1.9% |        1,425.68 |          389.08 |  3.664 |         258.88 |    0.1% |      0.01 | `monotonic increasing point writes`
+|          316,928.00 |            3,155.29 |    1.5% |    3,992,986.00 |    1,699,813.00 |  2.349 |     806,226.50 |    0.0% |      0.01 | `worst case for radix tree`
+|               75.26 |       13,286,517.16 |    0.5% |        1,590.01 |          386.67 |  4.112 |         258.00 |    0.0% |      0.01 | `create and destroy`
+
 
 # "Real data" test
 
-Point queries only. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.
+Point queries only, best of three runs. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.
 
 ## skip list
 
 ```
-Check: 4.62967 seconds, 352.195 MB/s, Add: 3.34177 seconds, 167.771 MB/s, Gc ratio: 37.9399%, Peak idle memory: 5.51852e+06
+Check: 4.47891 seconds, 364.05 MB/s, Add: 4.55599 seconds, 123.058 MB/s, Gc ratio: 37.1145%
 ```
 
 ## radix tree
 
 ```
-Check: 1.00477 seconds, 1622.8 MB/s, Add: 1.21142 seconds, 462.808 MB/s, Gc ratio: 39.4716%, Peak idle memory: 2.0226e+06
+Check: 0.910234 seconds, 1791.35 MB/s, Add: 1.25908 seconds, 445.287 MB/s, Gc ratio: 44.0415%
 ```
 
 ## hash table
@@ -65,6 +56,5 @@ Check: 1.00477 seconds, 1622.8 MB/s, Add: 1.21142 seconds, 462.808 MB/s, Gc rati
 (The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)
 
 ```
-Check: 0.854254 seconds, 1908.74 MB/s, Add: 0.632626 seconds, 886.232 MB/s, Gc ratio: 41.0827%, Peak idle memory: 0
+Check: 0.804094 seconds, 2027.81 MB/s, Add: 0.652952 seconds, 858.645 MB/s, Gc ratio: 35.3885%
 ```
-

RealDataBench.cpp

@@ -1,12 +1,11 @@
 #include <ConflictSet.h>
 
-#include <algorithm>
 #include <cerrno>
 #include <chrono>
 #include <cstdio>
 #include <cstring>
 #include <fcntl.h>
-#include <span>
+#include <string_view>
 #include <sys/mman.h>
 #include <sys/stat.h>
 #include <unistd.h>
@@ -65,7 +64,7 @@ int main(int argc, const char **argv) {
     auto *const mapOriginal = begin;
     const auto sizeOriginal = size;
 
-    using StringView = std::span<const uint8_t>;
+    using StringView = std::basic_string_view<uint8_t>;
 
     StringView write;
     std::vector<StringView> reads;
@@ -78,10 +77,10 @@ int main(int argc, const char **argv) {
       begin = end + 1;
       end = (uint8_t *)memchr(begin, '\n', size);
 
-      if (line.size() >= 2 && line[0] == 'P') {
-        write = line.subspan(2, line.size() - 2);
-      } else if (line.size() >= 2 && line[0] == 'L') {
-        reads.push_back(line.subspan(2, line.size() - 2));
+      if (line.size() > 0 && line[0] == 'P') {
+        write = line.substr(2, line.size());
+      } else if (line.size() > 0 && line[0] == 'L') {
+        reads.push_back(line.substr(2, line.size()));
       } else if (line.empty()) {
         {
           readRanges.resize(reads.size());
@@ -91,7 +90,7 @@ int main(int argc, const char **argv) {
             iter->begin.len = read.size();
             checkBytes += read.size();
             iter->end.len = 0;
-            iter->readVersion = std::max<int64_t>(0, version - 100);
+            iter->readVersion = version - 100;
             ++iter;
           }
         }
@@ -122,7 +121,7 @@ int main(int argc, const char **argv) {
         }
 
         timer = now();
-        cs.setOldestVersion(std::max<int64_t>(0, version - 10000));
+        cs.setOldestVersion(version - 10000);
         gcTime += now() - timer;
       }
     }
@@ -134,10 +133,10 @@ int main(int argc, const char **argv) {
   int metricsCount;
   cs.getMetricsV1(&metrics, &metricsCount);
   for (int i = 0; i < metricsCount; ++i) {
-    fprintf(stderr, "# HELP %s %s\n", metrics[i].name, metrics[i].help);
-    fprintf(stderr, "# TYPE %s %s\n", metrics[i].name,
-            metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
-    fprintf(stderr, "%s %g\n", metrics[i].name, metrics[i].getValue());
+    printf("# HELP %s %s\n", metrics[i].name, metrics[i].help);
+    printf("# TYPE %s %s\n", metrics[i].name,
+           metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
+    printf("%s %g\n", metrics[i].name, metrics[i].getValue());
   }
 
   printf("Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc ratio: "

ServerBench.cpp

@@ -1,7 +1,4 @@
-#include <algorithm>
 #include <atomic>
-#include <cstdint>
-#include <cstdlib>
 #include <errno.h>
 #include <netdb.h>
 #include <stdio.h>
@@ -9,222 +6,91 @@
 #include <string.h>
 #include <string>
 #include <string_view>
-#include <sys/ioctl.h>
 #include <sys/resource.h>
 #include <sys/socket.h>
-#include <sys/syscall.h>
-#include <sys/types.h>
 #include <sys/uio.h>
 #include <thread>
 #include <unistd.h>
-#include <utility>
-#include <vector>
 
 #include "ConflictSet.h"
-#include "Internal.h"
 #include "third_party/nadeau.h"
 
-constexpr int kCacheLine = 64; // TODO mac m1 is 128
-
-template <class T> struct TxQueue {
-
-  explicit TxQueue(int lgSlotCount)
-      : slotCount(1 << lgSlotCount), slotCountMask(slotCount - 1),
-        slots(new T[slotCount]) {
-    // Otherwise we can't tell the difference between full and empty.
-    assert(!(slotCountMask & 0x80000000));
-  }
-
-  /// Call from producer thread, after ensuring consumer is no longer accessing
-  /// it somehow
-  ~TxQueue() { delete[] slots; }
-
-  /// Must be called from the producer thread
-  void push(T t) {
-    if (wouldBlock()) {
-      // Wait for pops to change and try again
-      consumer.pops.wait(producer.lastPopRead, std::memory_order_relaxed);
-      producer.lastPopRead = consumer.pops.load(std::memory_order_acquire);
-    }
-    slots[producer.pushesNonAtomic++ & slotCountMask] = std::move(t);
-    // seq_cst so that the notify can't be ordered before the store
-    producer.pushes.store(producer.pushesNonAtomic, std::memory_order_seq_cst);
-    // We have to notify every time, since we don't know if this is the last
-    // push ever
-    producer.pushes.notify_one();
-  }
-
-  /// Must be called from the producer thread
-  uint32_t outstanding() {
-    return producer.pushesNonAtomic -
-           consumer.pops.load(std::memory_order_relaxed);
-  }
-
-  /// Returns true if a call to push might block. Must be called from the
-  /// producer thread.
-  bool wouldBlock() {
-    // See if we can determine that overflow won't happen entirely from state
-    // local to the producer
-    if (producer.pushesNonAtomic - producer.lastPopRead == slotCount - 1) {
-      // Re-read pops with memory order
-      producer.lastPopRead = consumer.pops.load(std::memory_order_acquire);
-      return producer.pushesNonAtomic - producer.lastPopRead == slotCount - 1;
-    }
-    return false;
-  }
-
-  /// Valid until the next pop, or until this queue is destroyed.
-  T *pop() {
-    // See if we can determine that there's an entry we can pop entirely from
-    // state local to the consumer
-    if (consumer.lastPushRead - consumer.popsNonAtomic == 0) {
-      // Re-read pushes with memory order and try again
-      consumer.lastPushRead = producer.pushes.load(std::memory_order_acquire);
-      if (consumer.lastPushRead - consumer.popsNonAtomic == 0) {
-        // Wait for pushes to change and try again
-        producer.pushes.wait(consumer.lastPushRead, std::memory_order_relaxed);
-        consumer.lastPushRead = producer.pushes.load(std::memory_order_acquire);
-      }
-    }
-    auto result = &slots[consumer.popsNonAtomic++ & slotCountMask];
-    // We only have to write pops with memory order if we've run out of items.
-    // We know that we'll eventually run out.
-    if (consumer.lastPushRead - consumer.popsNonAtomic == 0) {
-      // seq_cst so that the notify can't be ordered before the store
-      consumer.pops.store(consumer.popsNonAtomic, std::memory_order_seq_cst);
-      consumer.pops.notify_one();
-    }
-    return result;
-  }
-
-private:
-  const uint32_t slotCount;
-  const uint32_t slotCountMask;
-  T *slots;
-  struct alignas(kCacheLine) ProducerState {
-    std::atomic<uint32_t> pushes{0};
-    uint32_t pushesNonAtomic{0};
-    uint32_t lastPopRead{0};
-  };
-  struct alignas(kCacheLine) ConsumerState {
-    std::atomic<uint32_t> pops{0};
-    uint32_t popsNonAtomic{0};
-    uint32_t lastPushRead{0};
-  };
-  ProducerState producer;
-  ConsumerState consumer;
-};
-
 std::atomic<int64_t> transactions;
 
-int64_t safeUnaryMinus(int64_t x) {
-  return x == std::numeric_limits<int64_t>::min() ? x : -x;
-}
+constexpr int kBaseSearchDepth = 32;
+constexpr int kWindowSize = 10000000;
 
-void tupleAppend(std::string &output, int64_t value) {
-  if (value == 0) {
-    output.push_back(0x14);
-    return;
-  }
-  uint32_t size = 8 - __builtin_clrsbll(value) / 8;
-  int typeCode = 0x14 + (value < 0 ? -1 : 1) * size;
-  output.push_back(typeCode);
-  if (value < 0) {
-    value = ~safeUnaryMinus(value);
-  }
-  uint64_t swap = __builtin_bswap64(value);
-  output.insert(output.end(), (uint8_t *)&swap + 8 - size,
-                (uint8_t *)&swap + 8);
-}
-
-void tupleAppend(std::string &output, std::string_view value) {
-  output.push_back('\x02');
-  if (memchr(value.data(), '\x00', value.size()) != nullptr) {
-    for (auto c : value) {
-      if (c == '\x00') {
-        output.push_back('\x00');
-        output.push_back('\xff');
-      } else {
-        output.push_back(c);
-      }
-    }
-  } else {
-    output.insert(output.end(), value.begin(), value.end());
-  }
-  output.push_back('\x00');
-}
-
-template <class... Ts> std::string tupleKey(const Ts &...ts) {
-  std::string result;
-  (tupleAppend(result, ts), ...);
+std::basic_string<uint8_t> numToKey(int64_t num) {
+  std::basic_string<uint8_t> result;
+  result.resize(kBaseSearchDepth + sizeof(int64_t));
+  memset(result.data(), 0, kBaseSearchDepth);
+  int64_t be = __builtin_bswap64(num);
+  memcpy(result.data() + kBaseSearchDepth, &be, sizeof(int64_t));
   return result;
 }
 
-constexpr int kTotalKeyRange = 1'000'000'000;
-constexpr int kWindowSize = 1'000'000;
-constexpr int kNumReadKeysPerTx = 5;
-constexpr int kNumWriteKeysPerTx = 10;
-
-struct Transaction {
-  std::vector<std::string> keys;
-  std::vector<weaselab::ConflictSet::ReadRange> reads;
-  std::vector<weaselab::ConflictSet::WriteRange> writes;
-  int64_t version;
-  int64_t oldestVersion;
-  Transaction() = default;
-  explicit Transaction(int64_t version)
-      : version(version), oldestVersion(version - kWindowSize) {
-    std::vector<int64_t> keyIndices;
-    for (int i = 0; i < std::max(kNumReadKeysPerTx, kNumWriteKeysPerTx); ++i) {
-      keyIndices.push_back(rand() % kTotalKeyRange);
+void workload(weaselab::ConflictSet *cs) {
+  int64_t version = kWindowSize;
+  cs->addWrites(nullptr, 0, version);
+  for (;; transactions.fetch_add(1, std::memory_order_relaxed)) {
+    // Reads
+    {
+      auto beginK = numToKey(version - kWindowSize);
+      auto endK = numToKey(version - 1);
+      auto pointRv = version - kWindowSize + rand() % kWindowSize + 1;
+      auto pointK = numToKey(pointRv);
+      weaselab::ConflictSet::ReadRange reads[] = {
+          {
+              {pointK.data(), int(pointK.size())},
+              {nullptr, 0},
+              pointRv,
+          },
+          {
+              {beginK.data(), int(beginK.size())},
+              {endK.data(), int(endK.size())},
+              version - 2,
+          },
+      };
+      weaselab::ConflictSet::Result result[sizeof(reads) / sizeof(reads[0])];
+      cs->check(reads, result, sizeof(reads) / sizeof(reads[0]));
+      // for (int i = 0; i < sizeof(reads) / sizeof(reads[0]); ++i) {
+      //   if (result[i] != weaselab::ConflictSet::Commit) {
+      //     fprintf(stderr, "Unexpected conflict: [%s, %s) @ %" PRId64 "\n",
+      //             printable(reads[i].begin).c_str(),
+      //             printable(reads[i].end).c_str(), reads[i].readVersion);
+      //     abort();
+      //   }
+      // }
     }
-    std::sort(keyIndices.begin(), keyIndices.end());
-    constexpr std::string_view fullString =
-        "this is a string, where a prefix of it is used as an element of the "
-        "tuple forming the key";
-    for (int i = 0; i < int(keyIndices.size()); ++i) {
-      keys.push_back(
-          tupleKey(0x100, keyIndices[i] / fullString.size(),
-                   fullString.substr(0, keyIndices[i] % fullString.size())));
-      // printf("%s\n", printable(keys.back()).c_str());
-    }
-    for (int i = 0; i < kNumWriteKeysPerTx; ++i) {
-      writes.push_back({{(const uint8_t *)keys[i].data(), int(keys[i].size())},
-                        {nullptr, 0}});
-    }
-    reads.push_back({{(const uint8_t *)keys[0].data(), int(keys[0].size())},
-                     {(const uint8_t *)keys[1].data(), int(keys[1].size())},
-                     version - std::min(10, kWindowSize)});
-    static_assert(kNumReadKeysPerTx >= 3);
-    for (int i = 2; i < kNumReadKeysPerTx; ++i) {
-      reads.push_back({{(const uint8_t *)keys[i].data(), int(keys[i].size())},
-                       {nullptr, 0},
-                       version - kWindowSize});
+    // Writes
+    {
+      weaselab::ConflictSet::WriteRange w;
+      auto k = numToKey(version);
+      w.begin.p = k.data();
+      w.end.len = 0;
+      if (version % (kWindowSize / 2) == 0) {
+        for (int l = 0; l <= k.size(); ++l) {
+          w.begin.len = l;
+          cs->addWrites(&w, 1, version);
+        }
+      } else {
+        w.begin.len = k.size();
+        cs->addWrites(&w, 1, version);
+        int64_t beginN = version - kWindowSize + rand() % kWindowSize;
+        auto b = numToKey(beginN);
+        auto e = numToKey(beginN + 1000);
+        w.begin.p = b.data();
+        w.begin.len = b.size();
+        w.end.p = e.data();
+        w.end.len = e.size();
+        cs->addWrites(&w, 1, version);
+      }
     }
+    // GC
+    cs->setOldestVersion(version - kWindowSize);
+    ++version;
   }
-
-  Transaction(Transaction &&) = default;
-  Transaction &operator=(Transaction &&) = default;
-  Transaction(Transaction const &) = delete;
-  Transaction const &operator=(Transaction const &) = delete;
-};
-
-struct Resolver {
-
-  void resolve(const weaselab::ConflictSet::ReadRange *reads, int readCount,
-               const weaselab::ConflictSet::WriteRange *writes, int writeCount,
-               int64_t newVersion, int64_t newOldestVersion) {
-    results.resize(readCount);
-    cs.check(reads, results.data(), readCount);
-    cs.addWrites(writes, writeCount, newVersion);
-    cs.setOldestVersion(newOldestVersion);
-  }
-
-  ConflictSet cs{0};
-
-private:
-  std::vector<weaselab::ConflictSet::Result> results;
-};
+}
 
 // Adapted from getaddrinfo man page
 int getListenFd(const char *node, const char *service) {
@@ -298,29 +164,36 @@ double toSeconds(timeval t) {
   return double(t.tv_sec) + double(t.tv_usec) * 1e-6;
 }
 
-#ifdef __linux__
 #include <linux/perf_event.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#ifdef __linux__
 struct PerfCounter {
-  PerfCounter(int type, int config, const std::string &labels = {},
-              int groupLeaderFd = -1)
-      : labels(labels) {
+  explicit PerfCounter(int event) {
     struct perf_event_attr pe;
 
     memset(&pe, 0, sizeof(pe));
-    pe.type = type;
+    pe.type = PERF_TYPE_HARDWARE;
     pe.size = sizeof(pe);
-    pe.config = config;
+    pe.config = event;
     pe.inherit = 1;
     pe.exclude_kernel = 1;
     pe.exclude_hv = 1;
 
-    fd = perf_event_open(&pe, 0, -1, groupLeaderFd, 0);
-    if (fd < 0 && errno != ENOENT && errno != EINVAL) {
-      perror(labels.c_str());
+    fd = perf_event_open(&pe, 0, -1, -1, 0);
+    if (fd == -1) {
+      fprintf(stderr, "Error opening leader %llx\n", pe.config);
+      exit(EXIT_FAILURE);
     }
   }
 
-  int64_t total() const {
+  int64_t total() {
     int64_t count;
     if (read(fd, &count, sizeof(count)) != sizeof(count)) {
       perror("read instructions from perf");
@@ -329,27 +202,10 @@ struct PerfCounter {
     return count;
   }
 
-  PerfCounter(PerfCounter &&other)
-      : fd(std::exchange(other.fd, -1)), labels(std::move(other.labels)) {}
-  PerfCounter &operator=(PerfCounter &&other) {
-    fd = std::exchange(other.fd, -1);
-    labels = std::move(other.labels);
-    return *this;
-  }
-
-  ~PerfCounter() {
-    if (fd >= 0) {
-      close(fd);
-    }
-  }
-
-  bool ok() const { return fd >= 0; }
-  const std::string &getLabels() const { return labels; }
-  int getFd() const { return fd; }
+  ~PerfCounter() { close(fd); }
 
 private:
   int fd;
-  std::string labels;
   static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
                               int cpu, int group_fd, unsigned long flags) {
     int ret;
@@ -358,6 +214,11 @@ private:
     return ret;
   }
 };
+#else
+struct PerfCounter {
+  explicit PerPerfCounter(int) {}
+  int64_t total() { return 0; }
+};
 #endif
 
 int main(int argc, char **argv) {
@@ -367,72 +228,14 @@ int main(int argc, char **argv) {
   {
     int listenFd = getListenFd(argv[1], argv[2]);
 
-    Resolver resolver;
-    auto &cs = resolver.cs;
+    weaselab::ConflictSet cs{0};
     weaselab::ConflictSet::MetricsV1 *metrics;
     int metricsCount;
     cs.getMetricsV1(&metrics, &metricsCount);
 
-#ifdef __linux__
-    PerfCounter instructions{PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS};
-    PerfCounter cycles{PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES, "",
-                       instructions.getFd()};
-
-    std::vector<PerfCounter> cacheCounters;
-    for (auto [id, idStr] : std::initializer_list<std::pair<int, std::string>>{
-             {PERF_COUNT_HW_CACHE_L1D, "l1d"},
-             {PERF_COUNT_HW_CACHE_L1I, "l1i"},
-             {PERF_COUNT_HW_CACHE_LL, "ll"},
-             {PERF_COUNT_HW_CACHE_DTLB, "dtlb"},
-             {PERF_COUNT_HW_CACHE_ITLB, "itlb"},
-             {PERF_COUNT_HW_CACHE_BPU, "bpu"},
-             {PERF_COUNT_HW_CACHE_NODE, "node"},
-         }) {
-      for (auto [op, opStr] :
-           std::initializer_list<std::pair<int, std::string>>{
-               {PERF_COUNT_HW_CACHE_OP_READ, "read"},
-               {PERF_COUNT_HW_CACHE_OP_WRITE, "write"},
-               {PERF_COUNT_HW_CACHE_OP_PREFETCH, "prefetch"},
-           }) {
-        int groupLeaderFd = -1;
-        for (auto [result, resultStr] :
-             std::initializer_list<std::pair<int, std::string>>{
-                 {PERF_COUNT_HW_CACHE_RESULT_MISS, "miss"},
-                 {PERF_COUNT_HW_CACHE_RESULT_ACCESS, "access"},
-             }) {
-          auto labels = "{id=\"" + idStr + "\", op=\"" + opStr +
-                        "\", result=\"" + resultStr + "\"}";
-          cacheCounters.emplace_back(PERF_TYPE_HW_CACHE,
-                                     id | (op << 8) | (result << 16), labels,
-                                     groupLeaderFd);
-          if (!cacheCounters.back().ok()) {
-            cacheCounters.pop_back();
-          } else {
-            if (groupLeaderFd == -1) {
-              groupLeaderFd = cacheCounters.back().getFd();
-            }
-          }
-        }
-      }
-    }
-#endif
-
-    TxQueue<Transaction> queue{10};
-
-    auto workloadThread = std::thread{[&]() {
-      for (int64_t version = kWindowSize;;
-           ++version, transactions.fetch_add(1, std::memory_order_relaxed)) {
-        queue.push(Transaction(version));
-      }
-    }};
-
-    auto resolverThread = std::thread{[&]() {
-      for (;;) {
-        auto tx = queue.pop();
-        resolver.resolve(tx->reads.data(), tx->reads.size(), tx->writes.data(),
-                         tx->writes.size(), tx->version, tx->oldestVersion);
-      }
-    }};
+    PerfCounter instructions{PERF_COUNT_HW_INSTRUCTIONS};
+    PerfCounter cycles{PERF_COUNT_HW_CPU_CYCLES};
+    auto w = std::thread{workload, &cs};
 
     for (;;) {
       struct sockaddr_storage peer_addr = {};
@@ -459,7 +262,6 @@ int main(int argc, char **argv) {
               "transactions_total ";
       body += std::to_string(transactions.load(std::memory_order_relaxed));
       body += "\n";
-#ifdef __linux__
       body += "# HELP instructions_total Total number of instructions\n"
               "# TYPE instructions_total counter\n"
               "instructions_total ";
@@ -470,13 +272,6 @@ int main(int argc, char **argv) {
               "cycles_total ";
       body += std::to_string(cycles.total());
       body += "\n";
-      body += "# HELP cache_event_total Total number of cache events\n"
-              "# TYPE cache_event_total counter\n";
-      for (const auto &counter : cacheCounters) {
-        body += "cache_event_total" + counter.getLabels() + " " +
-                std::to_string(counter.total()) + "\n";
-      }
-#endif
 
       for (int i = 0; i < metricsCount; ++i) {
         body += "# HELP ";

SkipList.cpp

@@ -22,11 +22,9 @@
 
 #include "ConflictSet.h"
 #include "Internal.h"
-#include "Metrics.h"
 
 #include <algorithm>
 #include <span>
-#include <vector>
 
 std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
   auto result =
@@ -38,9 +36,6 @@ std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
 
 std::span<const uint8_t> copyToArena(Arena &arena,
                                      std::span<const uint8_t> key) {
-  if (key.size() == 0) {
-    return std::span<const uint8_t>();
-  }
   auto result = std::span<uint8_t>(new (arena) uint8_t[key.size()], key.size());
   memcpy(result.data(), key.data(), key.size());
   return result;
@@ -100,7 +95,7 @@ force_inline bool getCharacter(const KeyInfo &ki, int character,
 
 bool operator<(const KeyInfo &lhs, const KeyInfo &rhs) {
   int i = std::min(lhs.key.size(), rhs.key.size());
-  int c = i > 0 ? memcmp(lhs.key.data(), rhs.key.data(), i) : 0;
+  int c = memcmp(lhs.key.data(), rhs.key.data(), i);
   if (c != 0)
     return c < 0;
 
@@ -120,6 +115,15 @@ bool operator==(const KeyInfo &lhs, const KeyInfo &rhs) {
   return !(lhs < rhs || rhs < lhs);
 }
 
+void swapSort(std::vector<KeyInfo> &points, int a, int b) {
+  if (points[b] < points[a]) {
+    KeyInfo temp;
+    temp = points[a];
+    points[a] = points[b];
+    points[b] = temp;
+  }
+}
+
 struct SortTask {
   int begin;
   int size;
@@ -179,6 +183,13 @@ void sortPoints(std::vector<KeyInfo> &points) {
   }
 }
 
+static thread_local uint32_t g_seed = 0;
+
+static inline int skfastrand() {
+  g_seed = g_seed * 1664525L + 1013904223L;
+  return g_seed;
+}
+
 static int compare(const StringRef &a, const StringRef &b) {
   int c = memcmp(a.data(), b.data(), std::min(a.size(), b.size()));
   if (c < 0)
@@ -204,24 +215,20 @@ struct ReadConflictRange {
   }
 };
 
-static constexpr int MaxLevels = 26;
-
-struct RandomLevel {
-  explicit RandomLevel(uint32_t seed) : seed(seed) {}
-
-  int next() {
-    int result = __builtin_clz(seed | (uint32_t(-1) >> (MaxLevels - 1)));
-    seed = seed * 1664525L + 1013904223L;
-    return result;
-  }
-
-private:
-  uint32_t seed;
-};
-
 class SkipList {
 private:
-  RandomLevel randomLevel{0};
+  static constexpr int MaxLevels = 26;
+
+  int randomLevel() const {
+    uint32_t i = uint32_t(skfastrand()) >> (32 - (MaxLevels - 1));
+    int level = 0;
+    while (i & 1) {
+      i >>= 1;
+      level++;
+    }
+    assert(level < MaxLevels);
+    return level;
+  }
 
   // Represent a node in the SkipList. The node has multiple (i.e., level)
   // pointers to other nodes, and keeps a record of the max versions for each
@@ -290,8 +297,7 @@ private:
 
   static force_inline bool less(const uint8_t *a, int aLen, const uint8_t *b,
                                 int bLen) {
-    int n = std::min(aLen, bLen);
-    int c = n > 0 ? memcmp(a, b, n) : 0;
+    int c = memcmp(a, b, std::min(aLen, bLen));
     if (c < 0)
       return true;
     if (c > 0)
@@ -420,33 +426,27 @@ public:
   }
   void swap(SkipList &other) { std::swap(header, other.header); }
 
-  // Returns the change in the number of entries
-  int64_t addConflictRanges(const Finger *fingers, int rangeCount,
-                            Version version) {
-    int64_t result = rangeCount;
+  void addConflictRanges(const Finger *fingers, int rangeCount,
+                         Version version) {
     for (int r = rangeCount - 1; r >= 0; r--) {
       const Finger &startF = fingers[r * 2];
       const Finger &endF = fingers[r * 2 + 1];
 
-      if (endF.found() == nullptr) {
-        ++result;
+      if (endF.found() == nullptr)
         insert(endF, endF.finger[0]->getMaxVersion(0));
-      }
 
-      result -= remove(startF, endF);
+      remove(startF, endF);
       insert(startF, version);
     }
-    return result;
   }
 
-  // Return number of iterations of main loop
-  int detectConflicts(ReadConflictRange *ranges, int count,
-                      ConflictSet::Result *transactionConflictStatus) const {
+  void detectConflicts(ReadConflictRange *ranges, int count,
+                       ConflictSet::Result *transactionConflictStatus) const {
     const int M = 16;
     int nextJob[M];
     CheckMax inProgress[M];
     if (!count)
-      return 0;
+      return;
 
     int started = std::min(M, count);
     for (int i = 0; i < started; i++) {
@@ -457,9 +457,8 @@ public:
 
     int prevJob = started - 1;
     int job = 0;
-    int iters = 0;
     // vtune: 340 parts
-    for (;; ++iters) {
+    while (true) {
       if (inProgress[job].advance()) {
         if (started == count) {
           if (prevJob == job)
@@ -475,7 +474,6 @@ public:
       prevJob = job;
       job = nextJob[job];
     }
-    return iters;
   }
 
   void find(const StringRef *values, Finger *results, int *temp, int count) {
@@ -569,10 +567,9 @@ public:
   }
 
 private:
-  // Returns the number of entries removed
-  int64_t remove(const Finger &start, const Finger &end) {
+  void remove(const Finger &start, const Finger &end) {
     if (start.finger[0] == end.finger[0])
-      return 0;
+      return;
 
     Node *x = start.finger[0]->getNext(0);
 
@@ -581,20 +578,17 @@ private:
       if (start.finger[i] != end.finger[i])
         start.finger[i]->setNext(i, end.finger[i]->getNext(i));
 
-    int64_t result = 0;
     while (true) {
       Node *next = x->getNext(0);
       x->destroy();
-      ++result;
       if (x == end.finger[0])
         break;
       x = next;
     }
-    return result;
   }
 
   void insert(const Finger &f, Version version) {
-    int level = randomLevel.next();
+    int level = randomLevel();
     // std::cout << std::string((const char*)value,length) << " level: " <<
     // level << std::endl;
     Node *x = Node::create(f.value, level);
@@ -710,27 +704,17 @@ private:
   };
 };
 
-struct ReadContext {
-  int64_t commits_accum = 0;
-  int64_t conflicts_accum = 0;
-  int64_t too_olds_accum = 0;
-  int64_t check_bytes_accum = 0;
-};
+struct SkipListConflictSet {};
 
 struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   Impl(int64_t oldestVersion)
       : oldestVersion(oldestVersion), newestVersion(oldestVersion),
-        skipList(oldestVersion) {
-    metrics = initMetrics(metricsList, metricsCount);
-  }
-  ~Impl() { safe_free(metrics, metricsCount * sizeof(metrics[0])); }
+        skipList(oldestVersion) {}
   void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
-             int count) {
-    ReadContext tls;
+             int count) const {
     Arena arena;
     auto *ranges = new (arena) ReadConflictRange[count];
     for (int i = 0; i < count; ++i) {
-      tls.check_bytes_accum += reads[i].begin.len + reads[i].end.len;
       ranges[i].begin = {reads[i].begin.p, size_t(reads[i].begin.len)};
       ranges[i].end = reads[i].end.len > 0
                           ? StringRef{reads[i].end.p, size_t(reads[i].end.len)}
@@ -738,28 +722,18 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
       ranges[i].version = reads[i].readVersion;
       results[i] = ConflictSet::Commit;
     }
-    int iters = skipList.detectConflicts(ranges, count, results);
+    skipList.detectConflicts(ranges, count, results);
     for (int i = 0; i < count; ++i) {
       if (reads[i].readVersion < oldestVersion ||
           reads[i].readVersion < newestVersion - 2e9) {
         results[i] = TooOld;
       }
-      tls.commits_accum += results[i] == Commit;
-      tls.conflicts_accum += results[i] == Conflict;
-      tls.too_olds_accum += results[i] == TooOld;
     }
-    range_read_iterations_total.add(iters);
-    range_read_total.add(count);
-    commits_total.add(tls.commits_accum);
-    conflicts_total.add(tls.conflicts_accum);
-    too_olds_total.add(tls.too_olds_accum);
-    check_bytes_total.add(tls.check_bytes_accum);
   }
 
   void addWrites(const ConflictSet::WriteRange *writes, int count,
                  int64_t writeVersion) {
-    auto points = std::vector<KeyInfo>();
-    points.reserve(count * 2);
+    auto points = std::vector<KeyInfo>(count * 2);
     Arena arena;
 
     for (int r = 0; r < count; r++) {
@@ -772,9 +746,7 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
           false, true);
     }
 
-    if (!std::is_sorted(points.begin(), points.end())) {
-      sortPoints(points);
-    }
+    sortPoints(points);
 
     int activeWriteCount = 0;
     std::vector<std::pair<StringRef, StringRef>> combinedWriteConflictRanges;
@@ -801,34 +773,29 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     int temp[stripeSize];
     int stripes = (stringCount + stripeSize - 1) / stripeSize;
     StringRef values[stripeSize];
+    int64_t writeVersions[stripeSize / 2];
     int ss = stringCount - (stripes - 1) * stripeSize;
-    int64_t entryDelta = 0;
     for (int s = stripes - 1; s >= 0; s--) {
       for (int i = 0; i * 2 < ss; ++i) {
         const auto &w = combinedWriteConflictRanges[s * stripeSize / 2 + i];
         values[i * 2] = w.first;
         values[i * 2 + 1] = w.second;
+        keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
-      entryDelta += skipList.addConflictRanges(fingers, ss / 2, writeVersion);
+      skipList.addConflictRanges(fingers, ss / 2, writeVersion);
       ss = stripeSize;
     }
-
-    // Run gc at least 200% the rate we're inserting entries
-    keyUpdates += std::max<int64_t>(entryDelta, 0) * 2;
   }
 
   void setOldestVersion(int64_t oldestVersion) {
-    // This isn't 100% accurate. It overcounts if you hit the end
-    gc_iterations_total.add(keyUpdates);
-
     assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
     SkipList::Finger finger;
     int temp;
     std::span<const uint8_t> key = removalKey;
     skipList.find(&key, &finger, &temp, 1);
-    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 0));
+    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 10));
     removalArena = Arena();
     removalKey = copyToArena(
         removalArena, {finger.getValue().data(), finger.getValue().size()});
@@ -836,56 +803,8 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
 
   int64_t totalBytes = 0;
 
-  MetricsV1 *metrics;
-  int metricsCount = 0;
-  Metric *metricsList = nullptr;
-
-#define GAUGE(name, help)                                                      \
-  Gauge name { metricsList, metricsCount, #name, help }
-#define COUNTER(name, help)                                                    \
-  Counter name { metricsList, metricsCount, #name, help }
-  // ==================== METRICS DEFINITIONS ====================
-  COUNTER(range_read_total, "Total number of range reads checked");
-  COUNTER(range_read_iterations_total,
-          "Total number of iterations of the main loops for range read checks");
-  COUNTER(commits_total,
-          "Total number of checks where the result is \"commit\"");
-  COUNTER(conflicts_total,
-          "Total number of checks where the result is \"conflict\"");
-  COUNTER(too_olds_total,
-          "Total number of checks where the result is \"too old\"");
-  COUNTER(check_bytes_total, "Total number of key bytes checked");
-  GAUGE(memory_bytes, "Total number of bytes in use");
-  COUNTER(nodes_allocated_total,
-          "The total number of physical tree nodes allocated");
-  COUNTER(nodes_released_total,
-          "The total number of physical tree nodes released");
-  COUNTER(insert_iterations_total,
-          "The total number of iterations of the main loop for insertion. "
-          "Includes searches where the entry already existed, and so insertion "
-          "did not take place");
-  COUNTER(entries_inserted_total,
-          "The total number of entries inserted in the tree");
-  COUNTER(entries_erased_total,
-          "The total number of entries erased from the tree");
-  COUNTER(
-      gc_iterations_total,
-      "The total number of iterations of the main loop for garbage collection");
-  COUNTER(write_bytes_total, "Total number of key bytes in calls to addWrites");
-  GAUGE(oldest_version,
-        "The lowest version that doesn't result in \"TooOld\" for checks");
-  GAUGE(newest_version, "The version of the most recent call to addWrites");
-  // ==================== END METRICS DEFINITIONS ====================
-#undef GAUGE
-#undef COUNTER
-
-  void getMetricsV1(MetricsV1 **metrics, int *count) {
-    *metrics = this->metrics;
-    *count = metricsCount;
-  }
-
 private:
-  int64_t keyUpdates = 0;
+  int64_t keyUpdates = 10;
   Arena removalArena;
   std::span<const uint8_t> removalKey;
   int64_t oldestVersion;
@@ -906,7 +825,6 @@ void internal_addWrites(ConflictSet::Impl *impl,
   mallocBytesDelta = 0;
   impl->addWrites(writes, count, writeVersion);
   impl->totalBytes += mallocBytesDelta;
-  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
   if (impl->totalBytes != mallocBytes) {
     abort();
@@ -918,7 +836,6 @@ void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
   mallocBytesDelta = 0;
   impl->setOldestVersion(oldestVersion);
   impl->totalBytes += mallocBytesDelta;
-  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
   if (impl->totalBytes != mallocBytes) {
     abort();
@@ -942,11 +859,12 @@ int64_t internal_getBytes(ConflictSet::Impl *impl) { return impl->totalBytes; }
 
 void internal_getMetricsV1(ConflictSet::Impl *impl,
                            ConflictSet::MetricsV1 **metrics, int *count) {
-  return impl->getMetricsV1(metrics, count);
+  *metrics = nullptr;
+  *count = 0;
 }
 
 double internal_getMetricValue(const ConflictSet::MetricsV1 *metric) {
-  return ((Metric *)metric->p)->value.load(std::memory_order_relaxed);
+  return 0;
 }
 
 void ConflictSet::check(const ReadRange *reads, Result *results,
@@ -986,12 +904,7 @@ ConflictSet::ConflictSet(ConflictSet &&other) noexcept
     : impl(std::exchange(other.impl, nullptr)) {}
 
 ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
-  if (this != &other) {
-    if (impl) {
-      internal_destroy(impl);
-    }
-    impl = std::exchange(other.impl, nullptr);
-  }
+  impl = std::exchange(other.impl, nullptr);
   return *this;
 }
 

aarch64-symbol-imports.txt

@@ -5,7 +5,6 @@ __stack_chk_guard@GLIBC_2.17
 abort@GLIBC_2.17
 free@GLIBC_2.17
 malloc@GLIBC_2.17
-memcmp@GLIBC_2.17
 memcpy@GLIBC_2.17
 memmove@GLIBC_2.17
 memset@GLIBC_2.17

aarch64-toolchain.cmake

@@ -1,8 +1,7 @@
 set(CMAKE_SYSTEM_NAME Linux)
 set(CMAKE_SYSTEM_PROCESSOR aarch64)
-set(CMAKE_C_COMPILER "clang;--target=aarch64-linux-gnu")
-set(CMAKE_CXX_COMPILER "clang++;--target=aarch64-linux-gnu")
+set(CMAKE_C_COMPILER "/usr/bin/aarch64-linux-gnu-gcc")
+set(CMAKE_CXX_COMPILER "/usr/bin/aarch64-linux-gnu-g++")
 set(CMAKE_FIND_ROOT_PATH /usr/aarch64-linux-gnu)
 set(CMAKE_CROSSCOMPILING_EMULATOR "qemu-aarch64;-L;/usr/aarch64-linux-gnu/")
 set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE arm64)
-set(LD_EXE "/usr/bin/aarch64-linux-gnu-ld")

apple-symbol-imports.txt

@@ -1,4 +1,3 @@
-___chkstk_darwin
 ___stack_chk_fail
 ___stack_chk_guard
 __tlv_bootstrap
@@ -6,7 +5,6 @@ _abort
 _bzero
 _free
 _malloc
-_memcmp
 _memcpy
 _memmove
 dyld_stub_binder

build_msan_toolchain.sh

@@ -1,43 +0,0 @@
-#!/bin/bash
-
-set -euxo pipefail
-
-LLVM_VERSION="${LLVM_VERSION:-21}"
-MSAN_PREFIX="${MSAN_PREFIX:-$PWD/msan}"
-JOBS="${JOBS:-$(nproc)}"
-
-cd /tmp
-rm -rf libcxx-msan
-mkdir libcxx-msan
-cd libcxx-msan
-
-git clone --depth=1 "https://github.com/llvm/llvm-project.git" -b "release/${LLVM_VERSION}.x"
-
-cmake -S llvm-project/runtimes -B build_msan \
-  -DCMAKE_BUILD_TYPE=Release \
-  -DCMAKE_CXX_COMPILER="clang++-${LLVM_VERSION}" \
-  -DCMAKE_C_COMPILER="clang-${LLVM_VERSION}" \
-  -DCMAKE_INSTALL_PREFIX="${MSAN_PREFIX}" \
-  -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" \
-  -DLLVM_USE_SANITIZER=MemoryWithOrigins
-
-cmake --build build_msan -j"${JOBS}"
-cmake --install build_msan
-
-cmake -S llvm-project/runtimes -B build_nomsan \
-  -DCMAKE_BUILD_TYPE=Release \
-  -DCMAKE_CXX_COMPILER="clang++-${LLVM_VERSION}" \
-  -DCMAKE_C_COMPILER="clang-${LLVM_VERSION}" \
-  -DCMAKE_INSTALL_PREFIX="${MSAN_PREFIX}" \
-  -DLLVM_ENABLE_RUNTIMES="libunwind"
-
-cmake --build build_nomsan -j"${JOBS}"
-cmake --install build_nomsan
-
-VERSION="$(clang-"${LLVM_VERSION}" --version | head -n1 | sed -E 's/.*clang version ([0-9.]+).*/\1/')"
-TARBALL="msan-toolchain-${VERSION}.tar.zst"
-
-tar --zstd -cf "${TARBALL}" -C "${MSAN_PREFIX}" .
-
-TARBALL_PATH="$(pwd)/${TARBALL}"
-echo "Created: ${TARBALL_PATH}"

conflict_set.py

@@ -27,37 +27,23 @@ class Result(enum.Enum):
     TOO_OLD = 2
 
 
-def _make_key(buf: bytes) -> tuple[_Key, bytearray]:
-    """Create a _Key and a backing bytearray that must be kept alive."""
-    backing = bytearray(buf)
-    array = (ctypes.c_ubyte * len(backing)).from_buffer(backing)
-    return _Key(array, len(array)), backing
-
-
 def write(begin: bytes, end: Optional[bytes] = None) -> WriteRange:
-    begin_key, begin_buf = _make_key(begin)
+    b = (ctypes.c_ubyte * len(begin)).from_buffer(bytearray(begin))
+
     if end is None:
-        end_key = _Key((ctypes.c_ubyte * 0)(), 0)
-        end_buf = None
+        e = (ctypes.c_ubyte * 0)()
     else:
-        end_key, end_buf = _make_key(end)
-    result = WriteRange(begin_key, end_key)
-    result._begin_buf = begin_buf
-    result._end_buf = end_buf
-    return result
+        e = (ctypes.c_ubyte * len(end)).from_buffer(bytearray(end))
+    return WriteRange(_Key(b, len(b)), _Key(e, len(e)))
 
 
 def read(version: int, begin: bytes, end: Optional[bytes] = None) -> ReadRange:
-    begin_key, begin_buf = _make_key(begin)
+    b = (ctypes.c_ubyte * len(begin)).from_buffer(bytearray(begin))
     if end is None:
-        end_key = _Key((ctypes.c_ubyte * 0)(), 0)
-        end_buf = None
+        e = (ctypes.c_ubyte * 0)()
     else:
-        end_key, end_buf = _make_key(end)
-    result = ReadRange(begin_key, end_key, version)
-    result._begin_buf = begin_buf
-    result._end_buf = end_buf
-    return result
+        e = (ctypes.c_ubyte * len(end)).from_buffer(bytearray(end))
+    return ReadRange(_Key(b, len(b)), _Key(e, len(e)), version)
 
 
 class ConflictSet:
@@ -102,7 +88,6 @@ class ConflictSet:
             ctypes.POINTER(ctypes.c_int),
             ctypes.c_int,
         )
-        self._lib.ConflictSet_check.restype = None
 
         self._lib.ConflictSet_addWrites.argtypes = (
             ctypes.c_void_p,
@@ -110,16 +95,13 @@ class ConflictSet:
             ctypes.c_int,
             ctypes.c_int64,
         )
-        self._lib.ConflictSet_addWrites.restype = None
 
         self._lib.ConflictSet_setOldestVersion.argtypes = (
             ctypes.c_void_p,
             ctypes.c_int64,
         )
-        self._lib.ConflictSet_setOldestVersion.restype = None
 
         self._lib.ConflictSet_destroy.argtypes = (ctypes.c_void_p,)
-        self._lib.ConflictSet_destroy.restype = None
 
         self._lib.ConflictSet_getBytes.argtypes = (ctypes.c_void_p,)
         self._lib.ConflictSet_getBytes.restype = ctypes.c_int64