Opt-in to rpm/deb default package filenames

Closes #13

.clangd

@@ -1,2 +1,2 @@
 CompileFlags:
-  Add: [-DENABLE_MAIN, -UNDEBUG, -DENABLE_FUZZ, -fexceptions]
+  Add: [-DENABLE_MAIN, -UNDEBUG, -DENABLE_FUZZ, -DTHREAD_TEST, -fexceptions]

.vscode/settings.json

@@ -1,5 +1,9 @@
 {
   "files.associations": {
     "*.tikz": "latex"
-  }
-}
+  },
+  "latex-workshop.view.pdf.invertMode.enabled": "compat",
+  "latex-workshop.view.pdf.invert": 1,
+  "latex-workshop.view.pdf.invertMode.sepia": 1,
+  "latex-workshop.view.pdf.invertMode.grayscale": 0.5
+}

Bench.cpp

@@ -2,7 +2,6 @@
 #include "Internal.h"
 #include <cstdint>
 #include <cstring>
-#include <string>
 
 #if SHOW_MEMORY
 void showMemory(const ConflictSet &cs);
@@ -35,7 +34,7 @@ ConflictSet::ReadRange singleton(Arena &arena, std::span<const uint8_t> key) {
       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())};
+  return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
 ConflictSet::ReadRange prefixRange(Arena &arena, std::span<const uint8_t> key) {
@@ -53,7 +52,7 @@ ConflictSet::ReadRange prefixRange(Arena &arena, std::span<const uint8_t> key) {
   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())};
+  return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
 void benchConflictSet() {
@@ -61,7 +60,6 @@ void benchConflictSet() {
   ConflictSet cs{0};
 
   bench.batch(kOpsPerTx);
-  bench.minEpochIterations(10000);
 
   int64_t version = 0;
 
@@ -78,10 +76,9 @@ void benchConflictSet() {
       w.begin.len = r.begin.len;
       w.end.p = r.end.p;
       w.end.len = 0;
-      w.writeVersion = version + 1;
       writes.push_back(w);
     }
-    cs.addWrites(writes.data(), writes.size());
+    cs.addWrites(writes.data(), writes.size(), version + 1);
     ++version;
   }
 
@@ -112,11 +109,10 @@ void benchConflictSet() {
       w.begin.len = begin.size();
       w.end.p = end.data();
       w.end.len = end.size();
-      w.writeVersion = version + 1;
       writes.push_back(w);
     }
+    cs.addWrites(writes.data(), kOpsPerTx, version + 1);
     ++version;
-    cs.addWrites(writes.data(), kOpsPerTx);
   }
 
   {
@@ -189,17 +185,12 @@ void benchConflictSet() {
 
     while (version < kMvccWindow) {
       auto v = ++version;
-      writes[0].writeVersion = v;
-      cs.addWrites(writes.data(), 1);
+      cs.addWrites(writes.data(), 1, v);
     }
 
     bench.run("point writes", [&]() {
       auto v = ++version;
-      for (auto &w : writes) {
-        w.writeVersion = v;
-      }
-      cs.addWrites(writes.data(), writes.size());
-      cs.setOldestVersion(version - kMvccWindow);
+      cs.addWrites(writes.data(), writes.size(), v);
     });
   }
 
@@ -219,11 +210,7 @@ void benchConflictSet() {
 
     bench.run("prefix writes", [&]() {
       auto v = ++version;
-      for (auto &w : writes) {
-        w.writeVersion = v;
-      }
-      cs.addWrites(writes.data(), writes.size());
-      cs.setOldestVersion(version - kMvccWindow);
+      cs.addWrites(writes.data(), writes.size(), v);
     });
   }
 
@@ -243,16 +230,14 @@ void benchConflictSet() {
 
     bench.run("range writes", [&]() {
       auto v = ++version;
-      for (auto &w : writes) {
-        w.writeVersion = v;
-      }
-      cs.addWrites(writes.data(), writes.size());
-      cs.setOldestVersion(version - kMvccWindow);
+      cs.addWrites(writes.data(), writes.size(), v);
     });
   }
 
   bench.batch(1);
 
+  bench.warmup(10000);
+
   {
     bench.run("monotonic increasing point writes", [&]() {
       auto v = ++version;
@@ -263,15 +248,14 @@ void benchConflictSet() {
       auto x = __builtin_bswap64(version);
       memcpy(b, &x, 8);
 
-      w.writeVersion = v;
       w.begin.p = b;
       w.begin.len = 8;
       w.end.len = 0;
       w.end.p = b;
-      cs.addWrites(&w, 1);
+      cs.addWrites(&w, 1, v);
       cs.setOldestVersion(version - kMvccWindow);
     });
   }
 }
 
-int main(void) { benchConflictSet(); }
+int main(void) { benchConflictSet(); }

CMakeLists.txt

@@ -1,9 +1,10 @@
 cmake_minimum_required(VERSION 3.18)
 project(
-  conflict_set
+  conflict-set
   VERSION 0.0.1
   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)
 set(CMAKE_CXX_STANDARD 20)
 
@@ -21,7 +22,11 @@ if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
                                                "MinSizeRel" "RelWithDebInfo")
 endif()
 
-add_compile_options(-fdata-sections -ffunction-sections)
+add_compile_options(-fdata-sections -ffunction-sections -Wswitch-enum
+                    -Werror=switch-enum)
+
+option(USE_SIMD_FALLBACK
+       "Use fallback implementations of functions that use SIMD" OFF)
 
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
@@ -41,18 +46,20 @@ endif()
 include(CheckIncludeFileCXX)
 include(CMakePushCheckState)
 
-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()
+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)
+  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 "")
@@ -76,22 +83,22 @@ if(NOT APPLE)
                       LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
 endif()
 
-add_library(${PROJECT_NAME}_static STATIC
+add_library(${PROJECT_NAME}-static STATIC
             $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
 if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
-  set_target_properties(${PROJECT_NAME}_static PROPERTIES LINKER_LANGUAGE C)
+  set_target_properties(${PROJECT_NAME}-static PROPERTIES LINKER_LANGUAGE C)
 endif()
 
 if(NOT APPLE AND CMAKE_OBJCOPY)
   add_custom_command(
-    TARGET conflict_set_static
+    TARGET conflict-set-static
     POST_BUILD
     COMMAND
       ${CMAKE_OBJCOPY} --keep-global-symbols=${CMAKE_SOURCE_DIR}/symbols.txt
-      $<TARGET_FILE:${PROJECT_NAME}_static>)
+      $<TARGET_FILE:${PROJECT_NAME}-static>)
 endif()
 
-set(TEST_FLAGS -Wall -Wextra -Wpedantic -Wunreachable-code -UNDEBUG)
+set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
 include(CTest)
 
@@ -102,13 +109,25 @@ if(BUILD_TESTING)
   target_compile_options(skip_list PRIVATE -fPIC -fno-exceptions
                                            -fvisibility=hidden)
   target_include_directories(skip_list PUBLIC ${CMAKE_SOURCE_DIR}/include)
-  set_target_properties(skip_list PROPERTIES LINKER_LANGUAGE C)
   set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                                              "${CMAKE_BINARY_DIR}/skip_list")
   set_target_properties(skip_list PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
   set_target_properties(skip_list PROPERTIES VERSION ${PROJECT_VERSION}
                                              SOVERSION ${PROJECT_VERSION_MAJOR})
 
+  # Shared library version of a std::unordered_map-based conflict set (point
+  # queries only)
+  add_library(hash_table SHARED HashTable.cpp)
+  target_compile_options(hash_table PRIVATE -fPIC -fno-exceptions
+                                            -fvisibility=hidden)
+  target_include_directories(hash_table PUBLIC ${CMAKE_SOURCE_DIR}/include)
+  set_target_properties(hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                                              "${CMAKE_BINARY_DIR}/hash_table")
+  set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
+  set_target_properties(
+    hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
+                                                     ${PROJECT_VERSION_MAJOR})
+
   add_executable(conflict_set_main ConflictSet.cpp)
   target_include_directories(conflict_set_main
                              PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
@@ -141,6 +160,10 @@ if(BUILD_TESTING)
 
   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)
+    target_link_options(fuzz_driver PRIVATE -fsanitize=address,undefined)
+  endif()
   target_compile_definitions(fuzz_driver PRIVATE ENABLE_FUZZ)
   target_include_directories(fuzz_driver
                              PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
@@ -149,16 +172,41 @@ if(BUILD_TESTING)
     add_test(NAME conflict_set_fuzz_${hash} COMMAND fuzz_driver ${TEST})
   endforeach()
 
+  # tsan
+
+  if(NOT CMAKE_CROSSCOMPILING AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+    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)
+    target_include_directories(tsan_driver
+                               PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    foreach(TEST ${CORPUS_TESTS})
+      get_filename_component(hash ${TEST} NAME)
+      add_test(NAME conflict_set_tsan_${hash} COMMAND tsan_driver ${TEST})
+    endforeach()
+  endif()
+
   add_executable(driver TestDriver.cpp)
   target_compile_options(driver PRIVATE ${TEST_FLAGS})
   target_link_libraries(driver PRIVATE ${PROJECT_NAME})
 
+  add_executable(script_test ScriptTest.cpp)
+  target_compile_options(script_test PRIVATE ${TEST_FLAGS})
+  target_link_libraries(script_test PRIVATE ${PROJECT_NAME})
+
+  file(GLOB SCRIPT_TESTS ${CMAKE_SOURCE_DIR}/script_tests/*)
+  foreach(TEST ${SCRIPT_TESTS})
+    get_filename_component(name ${TEST} NAME)
+    add_test(NAME conflict_set_script_${name} COMMAND script_test ${TEST})
+  endforeach()
+
   add_executable(driver_skip_list TestDriver.cpp)
   target_compile_options(driver_skip_list PRIVATE ${TEST_FLAGS})
   target_link_libraries(driver_skip_list PRIVATE skip_list)
 
   find_program(VALGRIND_EXE valgrind)
-  if(VALGRIND_EXE)
+  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})
@@ -199,7 +247,7 @@ if(BUILD_TESTING)
       NAME conflict_set_static_symbols
       COMMAND
         ${CMAKE_SOURCE_DIR}/test_symbols.sh
-        $<TARGET_FILE:${PROJECT_NAME}_static> ${CMAKE_SOURCE_DIR}/symbols.txt)
+        $<TARGET_FILE:${PROJECT_NAME}-static> ${CMAKE_SOURCE_DIR}/symbols.txt)
   endif()
 
   # bench
@@ -207,10 +255,6 @@ if(BUILD_TESTING)
   add_executable(conflict_set_bench Bench.cpp)
   target_link_libraries(conflict_set_bench PRIVATE ${PROJECT_NAME})
   set_target_properties(conflict_set_bench PROPERTIES SKIP_BUILD_RPATH ON)
-  # target_compile_options(conflict_set_bench PRIVATE
-  # "-fsanitize=address,undefined,fuzzer")
-  # target_link_options(conflict_set_bench PRIVATE
-  # "-fsanitize=address,undefined,fuzzer")
 
   add_executable(real_data_bench RealDataBench.cpp)
   target_link_libraries(real_data_bench PRIVATE ${PROJECT_NAME})
@@ -219,7 +263,20 @@ endif()
 # packaging
 
 set(CPACK_PACKAGE_CONTACT andrew@weaselab.dev)
-set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+
+set(CPACK_PACKAGE_VENDOR "Weaselab")
+set(CPACK_RESOURCE_FILE_LICENSE "${CMAKE_CURRENT_SOURCE_DIR}/LICENSE")
+set(CPACK_RESOURCE_FILE_README "${CMAKE_CURRENT_SOURCE_DIR}/README.md")
+
+# rpm
+set(CPACK_RPM_PACKAGE_ARCHITECTURE ${CMAKE_SYSTEM_PROCESSOR})
+set(CPACK_RPM_SPEC_INSTALL_POST "/bin/true") # avoid stripping
+set(CPACK_RPM_PACKAGE_LICENSE "Apache 2.0")
+set(CPACK_RPM_FILE_NAME RPM-DEFAULT)
+
+# deb
+set(CPACK_DEBIAN_FILE_NAME DEB-DEFAULT)
+
 include(CPack)
 
 include(GNUInstallDirs)
@@ -230,7 +287,7 @@ target_include_directories(
          $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME}>)
 
 target_include_directories(
-  ${PROJECT_NAME}_static
+  ${PROJECT_NAME}-static
   PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
          $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME}>)
 
@@ -239,13 +296,13 @@ set_target_properties(
                              SOVERSION ${PROJECT_VERSION_MAJOR})
 
 install(
-  TARGETS ${PROJECT_NAME} ${PROJECT_NAME}_static
-  EXPORT ConflictSetConfig
+  TARGETS ${PROJECT_NAME} ${PROJECT_NAME}-static
+  EXPORT conflict-setConfig
   ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
   LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
   RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
 install(DIRECTORY include/
         DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME})
 
-install(EXPORT ConflictSetConfig
-        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/ConflictSet/cmake)
+install(EXPORT conflict-setConfig
+        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/conflict-set/cmake)

Dockerfile

@@ -15,6 +15,7 @@ RUN TZ=America/Los_Angeles DEBIAN_FRONTEND=noninteractive apt-get install -y \
     curl \
     doxygen \
     file \
+    g++-aarch64-linux-gnu \
     gcovr \
     git \
     gperf \
@@ -23,17 +24,20 @@ RUN TZ=America/Los_Angeles DEBIAN_FRONTEND=noninteractive apt-get install -y \
     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-3.20.0.tar.bz2 -o valgrind.tar.bz2 && \
-    echo "8536c031dbe078d342f121fa881a9ecd205cb5a78e639005ad570011bdb9f3c6  valgrind.tar.bz2" > valgrind-sha.txt && \
+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 && \
+    make -j`nproc` && \
     make install && \
     cd .. && \
     rm -rf /tmp/*
@@ -42,49 +46,6 @@ RUN curl -Ls https://sourceware.org/pub/valgrind/valgrind-3.20.0.tar.bz2 -o valg
 ENV CC=clang
 ENV CXX=clang++
 
-# Install recent flatbuffers from source
-RUN curl -Ls https://github.com/google/flatbuffers/archive/refs/tags/v23.3.3.tar.gz -o flatbuffers.tar.gz && \
-    echo "8aff985da30aaab37edf8e5b02fda33ed4cbdd962699a8e2af98fdef306f4e4d  flatbuffers.tar.gz" > flatbuffers-sha.txt && \
-    sha256sum --quiet -c flatbuffers-sha.txt && \
-    mkdir flatbuffers && \
-    tar --strip-components 1 --no-same-owner --no-same-permissions --directory flatbuffers  -xf flatbuffers.tar.gz && \
-    cd flatbuffers && \
-    cmake -S. -B build -G Ninja -DCMAKE_BUILD_TYPE=Release && \
-    ninja -C build install && \
-    rm -rf /tmp/*
-
-# Build msan-instrumented libc++ (llvmorg-16.0.0)
-RUN curl -Ls https://github.com/llvm/llvm-project/releases/download/llvmorg-16.0.0/llvm-project-16.0.0.src.tar.xz -o llvm-project.tar.gz && \
-    echo "9a56d906a2c81f16f06efc493a646d497c53c2f4f28f0cb1f3c8da7f74350254  llvm-project.tar.gz" > llvm-project-sha.txt && \
-    sha256sum --quiet -c llvm-project-sha.txt && \
-    mkdir llvm-project && \
-    tar --strip-components 1 --no-same-owner --no-same-permissions --directory llvm-project  -xf llvm-project.tar.gz && \
-    cmake -Sllvm-project/runtimes -B build -G Ninja \
-        -DCMAKE_BUILD_TYPE=Release \
-        -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi" \
-        -DCMAKE_C_COMPILER=clang \
-        -DCMAKE_CXX_COMPILER=clang++ \
-        -DCMAKE_INSTALL_PREFIX=/opt/llvm-msan \
-        -DLLVM_USE_SANITIZER=MemoryWithOrigins && \
-    ninja -C build cxx cxxabi && \
-    ninja -C build install-cxx install-cxxabi && \
-    rm -rf /tmp/*
-
-# Install bloaty from source
-RUN curl -Ls https://github.com/google/bloaty/releases/download/v1.1/bloaty-1.1.tar.bz2 -o bloaty.tar.bz2 && \
-    echo "a308d8369d5812aba45982e55e7c3db2ea4780b7496a5455792fb3dcba9abd6f  bloaty.tar.bz2" > bloaty-sha.txt && \
-    sha256sum --quiet -c bloaty-sha.txt && \
-    mkdir bloaty && \
-    tar --strip-components 1 --no-same-owner --no-same-permissions --directory bloaty  -xf bloaty.tar.bz2 && \
-    cd bloaty && \
-    cmake -S. -B build -G Ninja && \
-    ninja -C build install && \
-    rm -rf /tmp/*
-
-RUN apt-get update
-RUN apt-get upgrade -y
-RUN TZ=America/Los_Angeles DEBIAN_FRONTEND=noninteractive apt-get install -y texlive-full
-
 # 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

FuzzTestDriver.cpp

@@ -2,15 +2,25 @@
 #include <cstdint>
 #include <fstream>
 #include <sstream>
+#include <thread>
 
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size);
 
 int main(int argc, char **argv) {
-  for (int i = 1; i < argc; ++i) {
-    std::ifstream t(argv[i], std::ios::binary);
-    std::stringstream buffer;
-    buffer << t.rdbuf();
-    auto str = buffer.str();
-    LLVMFuzzerTestOneInput((const uint8_t *)str.data(), str.size());
-  }
+  auto doTest = [&]() {
+    for (int i = 1; i < argc; ++i) {
+      std::ifstream t(argv[i], std::ios::binary);
+      std::stringstream buffer;
+      buffer << t.rdbuf();
+      auto str = buffer.str();
+      LLVMFuzzerTestOneInput((const uint8_t *)str.data(), str.size());
+    }
+  };
+#ifdef THREAD_TEST
+  std::thread thread2{doTest};
+#endif
+  doTest();
+#ifdef THREAD_TEST
+  thread2.join();
+#endif
 }

HashTable.cpp

@@ -0,0 +1,154 @@
+#include "ConflictSet.h"
+#include "Internal.h"
+#include <functional>
+#include <string>
+#include <unordered_map>
+
+// This implementation isn't correct for range queries :). It's just intended as
+// a reference for performance comparison with point queries.
+
+// struct is from "https://www.cppstories.com/2021/heterogeneous-access-cpp20/"
+struct string_hash {
+  using is_transparent = void;
+  [[nodiscard]] size_t operator()(std::string_view txt) const {
+    return std::hash<std::string_view>{}(txt);
+  }
+  [[nodiscard]] size_t operator()(const std::string &txt) const {
+    return std::hash<std::string>{}(txt);
+  }
+};
+
+struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
+  Impl(int64_t oldestVersion) : oldestVersion(oldestVersion) {}
+  void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
+             int count) const {
+    for (int i = 0; i < count; ++i) {
+      auto key =
+          std::string_view((const char *)reads[i].begin.p, reads[i].begin.len);
+      auto version = reads[i].readVersion;
+      if (version < oldestVersion) {
+        results[i] = TooOld;
+        continue;
+      }
+      auto iter = map.find(key);
+      results[i] =
+          iter == map.end() || iter->second <= version ? Commit : Conflict;
+    }
+  }
+
+  void addWrites(const ConflictSet::WriteRange *writes, int count,
+                 int64_t writeVersion) {
+    for (int i = 0; i < count; ++i) {
+      auto &max = map[std::string((const char *)writes[i].begin.p,
+                                  writes[i].begin.len)];
+      assert(writeVersion >= max);
+      max = writeVersion;
+      keyUpdates += 2;
+    }
+  }
+
+  void setOldestVersion(int64_t oldestVersion) {
+    if (oldestVersion <= this->oldestVersion) {
+      return;
+    }
+    this->oldestVersion = oldestVersion;
+    if (keyUpdates < 100) {
+      return;
+    }
+    auto iter = map.find(removalKey);
+    while (keyUpdates > 0) {
+      if (iter == map.end()) {
+        iter = map.begin();
+      }
+      for (; iter != map.end(); --keyUpdates) {
+        if (iter->second <= oldestVersion) {
+          iter = map.erase(iter);
+        } else {
+          ++iter;
+        }
+      }
+    }
+    if (iter == map.end()) {
+      removalKey.clear();
+    } else {
+      removalKey = iter->first;
+    }
+  }
+
+private:
+  int64_t keyUpdates = 0;
+  int64_t oldestVersion;
+  std::unordered_map<std::string, int64_t, string_hash, std::equal_to<>> map;
+  std::string removalKey;
+};
+
+void ConflictSet::check(const ReadRange *reads, Result *results,
+                        int count) const {
+  return impl->check(reads, results, count);
+}
+
+void ConflictSet::addWrites(const WriteRange *writes, int count,
+                            int64_t writeVersion) {
+  return impl->addWrites(writes, count, writeVersion);
+}
+
+void ConflictSet::setOldestVersion(int64_t oldestVersion) {
+  return impl->setOldestVersion(oldestVersion);
+}
+
+int64_t ConflictSet::getBytes() const { return -1; }
+
+ConflictSet::ConflictSet(int64_t oldestVersion)
+    : impl(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
+
+ConflictSet::~ConflictSet() {
+  if (impl) {
+    impl->~Impl();
+    safe_free(impl, sizeof(Impl));
+  }
+}
+
+ConflictSet::ConflictSet(ConflictSet &&other) noexcept
+    : impl(std::exchange(other.impl, nullptr)) {}
+
+ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
+  impl = std::exchange(other.impl, nullptr);
+  return *this;
+}
+
+using ConflictSet_Result = ConflictSet::Result;
+using ConflictSet_Key = ConflictSet::Key;
+using ConflictSet_ReadRange = ConflictSet::ReadRange;
+using ConflictSet_WriteRange = ConflictSet::WriteRange;
+
+extern "C" {
+__attribute__((__visibility__("default"))) void
+ConflictSet_check(void *cs, const ConflictSet_ReadRange *reads,
+                  ConflictSet_Result *results, int count) {
+  ((ConflictSet::Impl *)cs)->check(reads, results, count);
+}
+__attribute__((__visibility__("default"))) void
+ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes, int count,
+                      int64_t writeVersion) {
+  ((ConflictSet::Impl *)cs)->addWrites(writes, count, writeVersion);
+}
+__attribute__((__visibility__("default"))) void
+ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
+  ((ConflictSet::Impl *)cs)->setOldestVersion(oldestVersion);
+}
+__attribute__((__visibility__("default"))) void *
+ConflictSet_create(int64_t oldestVersion) {
+  return new (safe_malloc(sizeof(ConflictSet::Impl)))
+      ConflictSet::Impl{oldestVersion};
+}
+__attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
+  using Impl = ConflictSet::Impl;
+  ((Impl *)cs)->~Impl();
+  safe_free(cs, sizeof(Impl));
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  using Impl = ConflictSet::Impl;
+  return -1;
+}
+}

Internal.h

@@ -2,6 +2,8 @@
 
 #include "ConflictSet.h"
 
+using namespace weaselab;
+
 #include <bit>
 #include <cassert>
 #include <compare>
@@ -10,10 +12,12 @@
 #include <cstdlib>
 #include <cstring>
 #include <inttypes.h>
+#include <latch>
 #include <map>
 #include <set>
 #include <span>
 #include <string>
+#include <thread>
 #include <unordered_set>
 #include <utility>
 #include <vector>
@@ -35,16 +39,76 @@ operator<=>(const std::span<const uint8_t> &lhs,
   return lhs.size() <=> rhs.size();
 }
 
+[[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) {
+    if (auto c = memcmp(lhs.data(), rhs.p, cl) <=> 0; c != 0) {
+      return c;
+    }
+  }
+  return lhs.size() <=> size_t(rhs.len);
+}
+
 // This header contains code that we want to reuse outside of ConflictSet.cpp or
 // want to exclude from coverage since it's only testing related.
 
 // GCOVR_EXCL_START
 
+#if SHOW_MEMORY
+inline int64_t mallocBytes = 0;
+inline int64_t peakMallocBytes = 0;
+#endif
+
+inline thread_local int64_t mallocBytesDelta = 0;
+
+#ifndef NDEBUG
+constexpr auto kMallocHeaderSize = 16;
+#endif
+
+// malloc that aborts on OOM and thus always returns a non-null pointer. Must be
+// paired with `safe_free`.
 __attribute__((always_inline)) inline void *safe_malloc(size_t s) {
-  if (void *p = malloc(s)) {
-    return p;
+  mallocBytesDelta += s;
+#if SHOW_MEMORY
+  mallocBytes += s;
+  if (mallocBytes > peakMallocBytes) {
+    peakMallocBytes = mallocBytes;
   }
-  abort();
+#endif
+  void *p = malloc(s
+#ifndef NDEBUG
+                   + kMallocHeaderSize
+#endif
+  );
+  if (p == nullptr) {
+    abort();
+  }
+#ifndef NDEBUG
+  memcpy(p, &s, sizeof(s));
+  (char *&)p += kMallocHeaderSize;
+#endif
+  return p;
+}
+
+// Must be paired with `safe_malloc`.
+//
+// There's nothing safer about this than free. Only called safe_free for
+// symmetry with safe_malloc.
+__attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
+  mallocBytesDelta -= s;
+#if SHOW_MEMORY
+  mallocBytes -= s;
+  free(p);
+#else
+#ifndef NDEBUG
+  (char *&)p -= kMallocHeaderSize;
+  size_t expected;
+  memcpy(&expected, p, sizeof(expected));
+  assert(s == expected);
+#endif
+  free(p);
+#endif
 }
 
 // ==================== BEGIN ARENA IMPL ====================
@@ -129,7 +193,7 @@ inline Arena::Arena(int initialSize) : impl(nullptr) {
 inline void onDestroy(Arena::ArenaImpl *impl) {
   while (impl) {
     auto *prev = impl->prev;
-    free(impl);
+    safe_free(impl, sizeof(Arena::ArenaImpl) + impl->capacity);
     impl = prev;
   }
 }
@@ -349,34 +413,6 @@ inline uint32_t Arbitrary::bounded(uint32_t s) {
 
 // ==================== END ARBITRARY IMPL ====================
 
-// ==================== BEGIN UTILITIES IMPL ====================
-
-// Call Stepwise::step for each element of remaining until it returns true.
-// Applies a permutation to `remaining` as a side effect.
-template <class Stepwise> void runInterleaved(std::span<Stepwise> remaining) {
-  while (remaining.size() > 0) {
-    for (int i = 0; i < int(remaining.size());) {
-      bool done = remaining[i].step();
-      if (done) {
-        if (i != int(remaining.size()) - 1) {
-          using std::swap;
-          swap(remaining[i], remaining.back());
-        }
-        remaining = remaining.subspan(0, remaining.size() - 1);
-      } else {
-        ++i;
-      }
-    }
-  }
-};
-
-template <class Stepwise> void runSequential(std::span<Stepwise> remaining) {
-  for (auto &r : remaining) {
-    while (!r.step()) {
-    }
-  }
-}
-
 struct ReferenceImpl {
   explicit ReferenceImpl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
     writeVersionMap[""] = oldestVersion;
@@ -404,7 +440,8 @@ struct ReferenceImpl {
                                                      : ConflictSet::Commit;
     }
   }
-  void addWrites(const ConflictSet::WriteRange *writes, int count) {
+  void addWrites(const ConflictSet::WriteRange *writes, int count,
+                 int64_t writeVersion) {
     for (int i = 0; i < count; ++i) {
       auto begin =
           std::string((const char *)writes[i].begin.p, writes[i].begin.len);
@@ -412,7 +449,6 @@ struct ReferenceImpl {
           writes[i].end.len == 0
               ? begin + std::string("\x00", 1)
               : std::string((const char *)writes[i].end.p, writes[i].end.len);
-      auto writeVersion = writes[i].writeVersion;
       auto prevVersion = (--writeVersionMap.upper_bound(end))->second;
       for (auto iter = writeVersionMap.lower_bound(begin),
                 endIter = writeVersionMap.lower_bound(end);
@@ -470,6 +506,18 @@ inline std::string printable(std::span<const uint8_t> key) {
   return printable(std::string_view((const char *)key.data(), key.size()));
 }
 
+inline const char *resultToStr(ConflictSet::Result r) {
+  switch (r) {
+  case ConflictSet::Commit:
+    return "commit";
+  case ConflictSet::Conflict:
+    return "conflict";
+  case ConflictSet::TooOld:
+    return "too old";
+  }
+  abort();
+}
+
 namespace {
 
 template <class ConflictSetImpl> struct TestDriver {
@@ -482,22 +530,10 @@ template <class ConflictSetImpl> struct TestDriver {
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
 
-  constexpr static auto kMaxKeyLen = 32;
+  constexpr static auto kMaxKeyLen = 8;
 
   bool ok = true;
 
-  static const char *resultToStr(ConflictSet::Result r) {
-    switch (r) {
-    case ConflictSet::Commit:
-      return "commit";
-    case ConflictSet::Conflict:
-      return "conflict";
-    case ConflictSet::TooOld:
-      return "too old";
-    }
-    abort();
-  }
-
   // Call until it returns true, for "done". Check internal invariants etc
   // between calls to next.
   bool next() {
@@ -547,17 +583,14 @@ template <class ConflictSetImpl> struct TestDriver {
           ++iter;
           --rangesRemaining;
         }
-        writes[i].writeVersion = v;
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (writes[i].end.len == 0) {
-          fprintf(stderr, "Write: {%s} -> %d\n",
-                  printable(writes[i].begin).c_str(),
-                  int(writes[i].writeVersion));
+          fprintf(stderr, "Write: {%s} -> %" PRId64 "\n",
+                  printable(writes[i].begin).c_str(), writeVersion);
         } else {
-          fprintf(stderr, "Write: [%s, %s) -> %d\n",
+          fprintf(stderr, "Write: [%s, %s) -> %" PRId64 "\n",
                   printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str(),
-                  int(writes[i].writeVersion));
+                  printable(writes[i].end).c_str(), writeVersion);
         }
 #endif
       }
@@ -565,10 +598,10 @@ template <class ConflictSetImpl> struct TestDriver {
       assert(i == numPointWrites + numRangeWrites);
 
       CALLGRIND_START_INSTRUMENTATION;
-      cs.addWrites(writes, numPointWrites + numRangeWrites);
+      cs.addWrites(writes, numPointWrites + numRangeWrites, v);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.addWrites(writes, numPointWrites + numRangeWrites);
+      refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
 
       oldestVersion = std::max<int64_t>(writeVersion - arbitrary.bounded(10),
                                         oldestVersion);
@@ -635,32 +668,60 @@ template <class ConflictSetImpl> struct TestDriver {
       auto *results2 =
           new (arena) ConflictSet::Result[numPointReads + numRangeReads];
 
+#ifdef THREAD_TEST
+      auto *results3 =
+          new (arena) ConflictSet::Result[numPointReads + numRangeReads];
+      std::latch ready{1};
+      std::thread thread2{[&]() {
+        ready.count_down();
+        cs.check(reads, results3, numPointReads + numRangeReads);
+      }};
+      ready.wait();
+#endif
+
       CALLGRIND_START_INSTRUMENTATION;
       cs.check(reads, results1, numPointReads + numRangeReads);
       CALLGRIND_STOP_INSTRUMENTATION;
 
       refImpl.check(reads, results2, numPointReads + numRangeReads);
-      for (int i = 0; i < numPointReads + numRangeReads; ++i) {
-        if (results1[i] != results2[i]) {
-          if (reads[i].end.len == 0) {
-            fprintf(stderr,
-                    "Expected %s, got %s for read of {%s} at version %" PRId64
-                    "\n",
-                    resultToStr(results2[i]), resultToStr(results1[i]),
-                    printable(reads[i].begin).c_str(), reads[i].readVersion);
-          } else {
-            fprintf(
-                stderr,
-                "Expected %s, got %s for read of [%s, %s) at version %" PRId64
-                "\n",
-                resultToStr(results2[i]), resultToStr(results1[i]),
-                printable(reads[i].begin).c_str(),
-                printable(reads[i].end).c_str(), reads[i].readVersion);
+
+      auto compareResults = [reads](ConflictSet::Result *results1,
+                                    ConflictSet::Result *results2, int count) {
+        for (int i = 0; i < count; ++i) {
+          if (results1[i] != results2[i]) {
+            if (reads[i].end.len == 0) {
+              fprintf(stderr,
+                      "Expected %s, got %s for read of {%s} at version %" PRId64
+                      "\n",
+                      resultToStr(results2[i]), resultToStr(results1[i]),
+                      printable(reads[i].begin).c_str(), reads[i].readVersion);
+            } else {
+              fprintf(
+                  stderr,
+                  "Expected %s, got %s for read of [%s, %s) at version %" PRId64
+                  "\n",
+                  resultToStr(results2[i]), resultToStr(results1[i]),
+                  printable(reads[i].begin).c_str(),
+                  printable(reads[i].end).c_str(), reads[i].readVersion);
+            }
+            return false;
           }
-          ok = false;
-          return true;
         }
+        return true;
+      };
+
+      if (!compareResults(results1, results2, numPointReads + numRangeReads)) {
+        ok = false;
+        return true;
       }
+
+#ifdef THREAD_TEST
+      thread2.join();
+      if (!compareResults(results3, results2, numPointReads + numRangeReads)) {
+        ok = false;
+        return true;
+      }
+#endif
     }
     return false;
   }

Jenkinsfile

@@ -13,8 +13,6 @@ def CleanBuildAndTest(String cmakeArgs) {
         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
-        LD_LIBRARY_PATH=skip_list ./conflict_set_bench
-        LD_LIBRARY_PATH=radix_tree ./conflict_set_bench
         '''
     }
     xunit tools: [CTest(pattern: 'build/Testing/*/Test.xml')], reduceLog: false, skipPublishingChecks: false
@@ -38,6 +36,29 @@ pipeline {
                 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('SIMD fallback') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
+            }
+        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {
@@ -46,18 +67,35 @@ pipeline {
                 }
             }
             steps {
-                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_BUILD_TYPE=Release")
+                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
                 '''
-                minio bucket: 'jenkins', credentialsId: 'jenkins-minio', excludes: '', host: 'minio.weaselab.dev', includes: 'build/*.deb', targetFolder: '${JOB_NAME}/${BUILD_NUMBER}/${STAGE_NAME}/'
                 sh '''
                 cd paper
                 make
                 '''
-                minio bucket: 'jenkins', credentialsId: 'jenkins-minio', excludes: '', host: 'minio.weaselab.dev', includes: 'paper/*.pdf', targetFolder: '${JOB_NAME}/${BUILD_NUMBER}/${STAGE_NAME}/'
+                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') {
@@ -68,9 +106,9 @@ pipeline {
                 }
             }
             steps {
-                CleanBuildAndTest("-DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug")
+                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug")
                 sh '''
-                gcovr --gcov-executable "llvm-cov-15 gcov" --exclude '.*third_party.*' --cobertura > build/coverage.xml
+                gcovr -f ConflictSet.cpp --cobertura > build/coverage.xml
                 '''
                 cobertura autoUpdateHealth: false, autoUpdateStability: false, coberturaReportFile: 'build/coverage.xml', conditionalCoverageTargets: '70, 0, 0', failUnhealthy: false, failUnstable: false, lineCoverageTargets: '80, 0, 0', maxNumberOfBuilds: 0, methodCoverageTargets: '80, 0, 0', onlyStable: false, sourceEncoding: 'ASCII', zoomCoverageChart: false
             }

README.md

@@ -2,71 +2,104 @@ A data structure for optimistic concurrency control on ranges of bitwise-lexicog
 
 Intended to replace FoundationDB's skip list.
 
+Hardware for all benchmarks is a mac m1 2020.
+
 # FoundationDB's benchmark
 
 ## Skip list
 
 ```
-New conflict set: 4.189 sec
-                  0.298 Mtransactions/sec
-                  1.194 Mkeys/sec
-Detect only:      3.990 sec
-                  0.313 Mtransactions/sec
-                  1.253 Mkeys/sec
-Skiplist only:    2.849 sec
-                  0.439 Mtransactions/sec
-                  1.755 Mkeys/sec
+New conflict set: 1.957 sec
+                  0.639 Mtransactions/sec
+                  2.555 Mkeys/sec
+Detect only:      1.845 sec
+                  0.678 Mtransactions/sec
+                  2.710 Mkeys/sec
+Skiplist only:    1.263 sec
+                  0.990 Mtransactions/sec
+                  3.960 Mkeys/sec
 Performance counters:
-               Build: 0.0913
-                 Add: 0.0998
-              Detect: 3.99
-              D.Sort: 0.808
-           D.Combine: 0.0309
-         D.CheckRead: 1.67
-   D.CheckIntraBatch: 0.0305
-        D.MergeWrite: 1.18
-      D.RemoveBefore: 0.265
+               Build: 0.0546
+                 Add: 0.0563
+              Detect: 1.84
+              D.Sort: 0.412
+           D.Combine: 0.0141
+         D.CheckRead: 0.671
+   D.CheckIntraBatch: 0.0068
+        D.MergeWrite: 0.592
+      D.RemoveBefore: 0.146
 ```
 
 ## Radix tree (this implementation)
 
 ```
-New conflict set: 2.965 sec
-                  0.422 Mtransactions/sec
-                  1.686 Mkeys/sec
-Detect only:      2.761 sec
-                  0.453 Mtransactions/sec
-                  1.811 Mkeys/sec
-Skiplist only:    1.580 sec
-                  0.791 Mtransactions/sec
-                  3.165 Mkeys/sec
+New conflict set: 1.366 sec
+                  0.915 Mtransactions/sec
+                  3.660 Mkeys/sec
+Detect only:      1.248 sec
+                  1.002 Mtransactions/sec
+                  4.007 Mkeys/sec
+Skiplist only:    0.573 sec
+                  2.182 Mtransactions/sec
+                  8.730 Mkeys/sec
 Performance counters:
-               Build: 0.0902
-                 Add: 0.107
-              Detect: 2.76
-              D.Sort: 0.809
-           D.Combine: 0.0309
-         D.CheckRead: 0.658
-   D.CheckIntraBatch: 0.0294
-        D.MergeWrite: 0.921
-      D.RemoveBefore: 0.305
+               Build: 0.0594
+                 Add: 0.0572
+              Detect: 1.25
+              D.Sort: 0.418
+           D.Combine: 0.0149
+         D.CheckRead: 0.232
+   D.CheckIntraBatch: 0.0067
+        D.MergeWrite: 0.341
+      D.RemoveBefore: 0.232
 ```
 
 # Our benchmark
 
+## Skip list
+
 |               ns/op |                op/s |    err% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------:|:----------
-|              325.60 |        3,071,225.77 |    4.8% |      0.77 | `skip list (point reads)`
-|              297.15 |        3,365,278.10 |    1.7% |      0.72 | `skip list (prefix reads)`
-|              408.79 |        2,446,222.23 |    1.0% |      1.03 | `skip list (range reads)`
-|              261.88 |        3,818,471.08 |    1.3% |      0.73 | `skip list (point writes)`
-|              253.54 |        3,944,191.08 |    0.1% |      0.61 | `skip list (prefix writes)`
-|              258.73 |        3,865,078.52 |    0.8% |      0.62 | `skip list (range writes)`
-|              489.56 |        2,042,648.19 |    1.8% |      0.01 | `skip list (monotonic increasing point writes)`
-|               14.83 |       67,446,579.75 |    0.1% |      0.04 | `radix tree (point reads)`
-|               59.68 |       16,756,917.37 |    0.1% |      0.14 | `radix tree (prefix reads)`
-|              287.32 |        3,480,485.22 |    1.2% |      0.69 | `radix tree (range reads)`
-|               46.59 |       21,461,855.59 |    0.2% |      0.12 | `radix tree (point writes)`
-|               83.70 |       11,946,755.99 |    0.1% |      0.20 | `radix tree (prefix writes)`
-|              100.75 |        9,925,723.26 |    0.6% |      0.25 | `radix tree (range writes)`
-|              118.37 |        8,448,345.29 |    0.6% |      0.01 | `radix tree (monotonic increasing point writes)`
+|              246.99 |        4,048,700.59 |    0.2% |      0.01 | `point reads`
+|              260.16 |        3,843,784.65 |    0.1% |      0.01 | `prefix reads`
+|              493.35 |        2,026,953.19 |    0.1% |      0.01 | `range reads`
+|              462.05 |        2,164,289.23 |    0.6% |      0.01 | `point writes`
+|              448.19 |        2,231,205.25 |    0.9% |      0.01 | `prefix writes`
+|              255.83 |        3,908,845.72 |    1.5% |      0.02 | `range writes`
+|              582.63 |        1,716,349.02 |    1.3% |      0.01 | `monotonic increasing point writes`
+
+## Radix tree (this implementation)
+
+|               ns/op |                op/s |    err% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------:|:----------
+|               19.42 |       51,483,206.67 |    0.3% |      0.01 | `point reads`
+|               58.43 |       17,115,612.57 |    0.1% |      0.01 | `prefix reads`
+|              216.09 |        4,627,766.60 |    0.2% |      0.01 | `range reads`
+|               28.35 |       35,267,567.72 |    0.2% |      0.01 | `point writes`
+|               43.43 |       23,026,226.17 |    0.2% |      0.01 | `prefix writes`
+|               50.00 |       20,000,000.00 |    0.0% |      0.01 | `range writes`
+|               92.38 |       10,824,863.69 |    4.1% |      0.01 | `monotonic increasing point writes`
+
+# "Real data" test
+
+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: 11.3385 seconds, 329.718 MB/s, Add: 5.35612 seconds, 131.072 MB/s, Gc ratio: 45.7173%
+```
+
+## radix tree
+
+```
+Check: 2.48583 seconds, 1503.93 MB/s, Add: 2.12768 seconds, 329.954 MB/s, Gc ratio: 41.7943%
+```
+
+## hash table
+
+(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: 1.83386 seconds, 2038.6 MB/s, Add: 0.601411 seconds, 1167.32 MB/s, Gc ratio: 48.9776%
+```

RealDataBench.cpp

@@ -3,12 +3,13 @@
 #include <chrono>
 #include <cstring>
 #include <fcntl.h>
-#include <fstream>
 #include <sys/mman.h>
 #include <sys/stat.h>
 #include <unistd.h>
 #include <vector>
 
+using namespace weaselab;
+
 double now() {
   return std::chrono::duration_cast<std::chrono::nanoseconds>(
              std::chrono::steady_clock::now().time_since_epoch())
@@ -29,7 +30,7 @@ constexpr inline size_t rightAlign(size_t offset, size_t alignment) {
 
 int main(int argc, const char **argv) {
   // Use with this dataset https://snap.stanford.edu/data/memetracker9.html
-  // Preprocess the files with `sed -i '' '/^Q/d'`
+  // Preprocess the files with `sed -i'' '/^Q/d'`
 
   double checkTime = 0;
   double addTime = 0;
@@ -41,6 +42,8 @@ int main(int argc, const char **argv) {
   int64_t version = 0;
   double timer = 0;
 
+  int64_t peakMemory = 0;
+
   for (int i = 1; i < argc; ++i) {
     int fd = open(argv[i], O_RDONLY);
     struct stat st;
@@ -55,6 +58,8 @@ int main(int argc, const char **argv) {
     const uint8_t *begin =
         (uint8_t *)mmap(0, size, PROT_READ, MAP_PRIVATE, fd, 0);
     madvise((void *)begin, size, MADV_SEQUENTIAL);
+    auto *const mapOriginal = begin;
+    const auto sizeOriginal = size;
 
     using StringView = std::basic_string_view<uint8_t>;
 
@@ -95,7 +100,6 @@ int main(int argc, const char **argv) {
         // Add unconditionally so that the load doesn't actually depend on the
         // conflict rate
         ConflictSet::WriteRange w;
-        w.writeVersion = ++version;
         w.begin.p = (const uint8_t *)write.data();
         w.begin.len = write.size();
         w.end.len = 0;
@@ -103,22 +107,28 @@ int main(int argc, const char **argv) {
         addBytes += write.size();
 
         timer = now();
-        cs.addWrites(&w, 1);
+        cs.addWrites(&w, 1, ++version);
         addTime += now() - timer;
 
         write = {};
         reads.clear();
 
+        if (cs.getBytes() > peakMemory) {
+          peakMemory = cs.getBytes();
+        }
+
         timer = now();
         cs.setOldestVersion(version - 10000);
         gcTime += now() - timer;
       }
     }
+    munmap((void *)mapOriginal, sizeOriginal);
     close(fd);
   }
 
-  printf("Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc: %g "
-         "seconds\n",
+  printf("Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc ratio: "
+         "%g%%, Peak idle memory: %g\n",
          checkTime, checkBytes / checkTime * 1e-6, addTime,
-         addBytes / addTime * 1e-6, gcTime);
+         addBytes / addTime * 1e-6, gcTime / (gcTime + addTime) * 1e2,
+         double(peakMemory));
 }

ScriptTest.cpp

@@ -0,0 +1,155 @@
+#include "Internal.h"
+#include <ConflictSet.h>
+
+#include <chrono>
+#include <cstring>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <vector>
+
+inline size_t getPageSize() {
+  static size_t kPageSize = sysconf(_SC_PAGESIZE);
+  return kPageSize;
+}
+
+/// Helper for rounding up to page size (or some other alignment)
+constexpr inline size_t rightAlign(size_t offset, size_t alignment) {
+  return offset % alignment == 0 ? offset
+                                 : ((offset / alignment) + 1) * alignment;
+}
+
+using StringView = std::basic_string_view<uint8_t>;
+
+inline StringView operator"" _v(const char *str, size_t size) {
+  return {reinterpret_cast<const uint8_t *>(str), size};
+}
+
+int main(int argc, const char **argv) {
+
+  ConflictSet cs{0};
+  ReferenceImpl ref{0};
+
+  for (int i = 1; i < argc; ++i) {
+    int fd = open(argv[i], O_RDONLY);
+    struct stat st;
+    if (fstat(fd, &st) == -1) {
+      int err = errno;
+      fprintf(stderr, "stat error %s - %s\n", argv[i], strerror(err));
+      fflush(stderr);
+      abort();
+    }
+
+    int64_t size = rightAlign(st.st_size, getPageSize());
+    const uint8_t *begin =
+        (uint8_t *)mmap(0, size, PROT_READ, MAP_PRIVATE, fd, 0);
+    madvise((void *)begin, size, MADV_SEQUENTIAL);
+    auto *const mapOriginal = begin;
+    const auto sizeOriginal = size;
+
+    StringView b;
+    StringView e;
+    int64_t v = 0;
+    int64_t lastWriteVersion = 0;
+    int64_t lastOldestVersion = 0;
+    std::vector<ConflictSet::WriteRange> writeRanges;
+    std::vector<ConflictSet::ReadRange> readRanges;
+    std::vector<ConflictSet::Result> results;
+
+    for (uint8_t *end = (uint8_t *)memchr(begin, '\n', size); end != nullptr;) {
+      StringView line{begin, static_cast<size_t>(end - begin)};
+      size -= end - begin + 1;
+      begin = end + 1;
+      end = (uint8_t *)memchr(begin, '\n', size);
+
+      if (line.starts_with("begin"_v)) {
+        b = line.substr("begin "_v.size(), line.size());
+        printf("b <- %.*s\n", int(b.size()), b.data());
+      } else if (line.starts_with("end"_v)) {
+        e = line.substr("end "_v.size(), line.size());
+        printf("e <- %.*s\n", int(e.size()), e.data());
+      } else if (line.starts_with("version"_v)) {
+        line = line.substr("version "_v.size(), line.size());
+        v = 0;
+        for (auto c : line) {
+          v = v * 10 + int(c) - int('0');
+        }
+        printf("v <- %" PRId64 "\n", v);
+      } else if (line.starts_with("pointread"_v)) {
+        printf("pointread\n");
+        ConflictSet::ReadRange r;
+        r.begin.p = b.data();
+        r.begin.len = b.size();
+        r.end.len = 0;
+        r.readVersion = v;
+        readRanges.push_back(r);
+      } else if (line.starts_with("pointwrite"_v)) {
+        printf("pointwrite\n");
+        assert(writeRanges.empty() ||
+               (writeRanges.back().end.len == 0 ? writeRanges.back().begin
+                                                : writeRanges.back().end) < b);
+        ConflictSet::WriteRange w;
+        w.begin.p = b.data();
+        w.begin.len = b.size();
+        w.end.len = 0;
+        writeRanges.push_back(w);
+      } else if (line.starts_with("rangeread"_v)) {
+        printf("rangeread\n");
+        ConflictSet::ReadRange r;
+        r.begin.p = b.data();
+        r.begin.len = b.size();
+        r.end.p = e.data();
+        r.end.len = e.size();
+        r.readVersion = v;
+        readRanges.push_back(r);
+      } else if (line.starts_with("rangewrite"_v)) {
+        printf("rangewrite\n");
+        assert(b < e);
+        assert(writeRanges.empty() ||
+               (writeRanges.back().end.len == 0 ? writeRanges.back().begin
+                                                : writeRanges.back().end) < b);
+        ConflictSet::WriteRange w;
+        w.begin.p = b.data();
+        w.begin.len = b.size();
+        w.end.p = e.data();
+        w.end.len = e.size();
+        writeRanges.push_back(w);
+      } else if (line.starts_with("check"_v)) {
+        printf("check\n");
+        Arena arena;
+        auto *expected = new (arena) ConflictSet::Result[readRanges.size()];
+        auto *actual = new (arena) ConflictSet::Result[readRanges.size()];
+        ref.check(readRanges.data(), expected, readRanges.size());
+        cs.check(readRanges.data(), actual, readRanges.size());
+        for (int i = 0; i < int(readRanges.size()); ++i) {
+          if (expected[i] != actual[i]) {
+            fprintf(stderr, "Expected %s, got %s at index %d\n",
+                    resultToStr(expected[i]), resultToStr(actual[i]), i);
+            return 1;
+          }
+        }
+        readRanges = {};
+      } else if (line.starts_with("addwrites"_v)) {
+        printf("addwrites\n");
+        assert(v > lastWriteVersion);
+        lastWriteVersion = v;
+        cs.addWrites(writeRanges.data(), writeRanges.size(), v);
+        ref.addWrites(writeRanges.data(), writeRanges.size(), v);
+        writeRanges = {};
+      } else if (line.starts_with("setoldest"_v)) {
+        printf("setoldest\n");
+        assert(v > lastOldestVersion);
+        lastOldestVersion = v;
+        cs.setOldestVersion(v);
+        ref.setOldestVersion(v);
+      } else if (line.empty() || line.starts_with(";"_v)) {
+        // skip
+      } else {
+        printf("Unrecognized line: %.*s\n", int(line.size()), line.data());
+      }
+    }
+    munmap((void *)mapOriginal, sizeOriginal);
+    close(fd);
+  }
+}

SkipList.cpp

@@ -149,7 +149,7 @@ private:
       setMaxVersion(level, v);
     }
 
-    void destroy() { free(this); }
+    void destroy() { safe_free(this, getNodeSize()); }
 
   private:
     int getNodeSize() const {
@@ -286,7 +286,7 @@ public:
   void swap(SkipList &other) { std::swap(header, other.header); }
 
   void addConflictRanges(const Finger *fingers, int rangeCount,
-                         Version *version) {
+                         Version version) {
     for (int r = rangeCount - 1; r >= 0; r--) {
       const Finger &startF = fingers[r * 2];
       const Finger &endF = fingers[r * 2 + 1];
@@ -295,7 +295,7 @@ public:
         insert(endF, endF.finger[0]->getMaxVersion(0));
 
       remove(startF, endF);
-      insert(startF, version[r]);
+      insert(startF, version);
     }
   }
 
@@ -588,7 +588,8 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     }
   }
 
-  void addWrites(const ConflictSet::WriteRange *writes, int count) {
+  void addWrites(const ConflictSet::WriteRange *writes, int count,
+                 int64_t writeVersion) {
     Arena arena;
     const int stringCount = count * 2;
 
@@ -606,11 +607,10 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
         values[i * 2 + 1] = w.end.len > 0
                                 ? StringRef{w.end.p, size_t(w.end.len)}
                                 : keyAfter(arena, values[i * 2]);
-        writeVersions[i] = w.writeVersion;
-        keyUpdates += 2;
+        keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
-      skipList.addConflictRanges(fingers, ss / 2, writeVersions);
+      skipList.addConflictRanges(fingers, ss / 2, writeVersion);
       ss = stripeSize;
     }
   }
@@ -621,14 +621,16 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     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()});
   }
 
+  int64_t totalBytes = 0;
+
 private:
-  int64_t keyUpdates = 0;
+  int64_t keyUpdates = 10;
   Arena removalArena;
   std::span<const uint8_t> removalKey;
   int64_t oldestVersion;
@@ -637,24 +639,44 @@ private:
 
 void ConflictSet::check(const ReadRange *reads, Result *results,
                         int count) const {
-  return impl->check(reads, results, count);
+  impl->check(reads, results, count);
 }
 
-void ConflictSet::addWrites(const WriteRange *writes, int count) {
-  return impl->addWrites(writes, count);
+void ConflictSet::addWrites(const WriteRange *writes, int count,
+                            int64_t writeVersion) {
+  mallocBytesDelta = 0;
+  impl->addWrites(writes, count, writeVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
 void ConflictSet::setOldestVersion(int64_t oldestVersion) {
-  return impl->setOldestVersion(oldestVersion);
+  mallocBytesDelta = 0;
+  impl->setOldestVersion(oldestVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
+int64_t ConflictSet::getBytes() const { return impl->totalBytes; }
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
-    : impl(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
+    : impl((mallocBytesDelta = 0,
+            new (safe_malloc(sizeof(Impl))) Impl{oldestVersion})) {
+  impl->totalBytes += mallocBytesDelta;
+}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
     impl->~Impl();
-    free(impl);
+    safe_free(impl, sizeof(Impl));
   }
 }
 
@@ -678,9 +700,9 @@ ConflictSet_check(void *cs, const ConflictSet_ReadRange *reads,
   ((ConflictSet::Impl *)cs)->check(reads, results, count);
 }
 __attribute__((__visibility__("default"))) void
-ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes,
-                      int count) {
-  ((ConflictSet::Impl *)cs)->addWrites(writes, count);
+ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes, int count,
+                      int64_t writeVersion) {
+  ((ConflictSet::Impl *)cs)->addWrites(writes, count, writeVersion);
 }
 __attribute__((__visibility__("default"))) void
 ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
@@ -694,6 +716,20 @@ ConflictSet_create(int64_t oldestVersion) {
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
   using Impl = ConflictSet::Impl;
   ((Impl *)cs)->~Impl();
-  free(cs);
+  safe_free(cs, sizeof(Impl));
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  using Impl = ConflictSet::Impl;
+  return ((Impl *)cs)->totalBytes;
 }
 }
+
+#if SHOW_MEMORY
+struct __attribute__((visibility("default"))) PeakPrinter {
+  ~PeakPrinter() {
+    printf("malloc bytes: %g\n", double(mallocBytes));
+    printf("Peak malloc bytes: %g\n", double(peakMallocBytes));
+  }
+} peakPrinter;
+#endif

TestDriver.cpp

@@ -5,7 +5,6 @@
 
 int main(int argc, char **argv) {
   for (int i = 1; i < argc; ++i) {
-    printf("Running: %s\n", argv[i]);
     std::ifstream t(argv[i], std::ios::binary);
     std::stringstream buffer;
     buffer << t.rdbuf();

aarch64-toolchain.cmake

@@ -0,0 +1,7 @@
+set(CMAKE_SYSTEM_NAME Linux)
+set(CMAKE_SYSTEM_PROCESSOR aarch64)
+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)

conflict_set_c_api_test.c

@@ -7,17 +7,19 @@ int main(void) {
   ConflictSet_WriteRange w;
   ConflictSet_Result result;
   ConflictSet_ReadRange r;
+  int64_t bytes;
   w.begin.p = (const uint8_t *)"0000";
   w.begin.len = 4;
   w.end.len = 0;
-  w.writeVersion = 1;
-  ConflictSet_addWrites(cs, &w, 1);
+  ConflictSet_addWrites(cs, &w, 1, 1);
   r.begin.p = (const uint8_t *)"0000";
   r.begin.len = 4;
   r.end.len = 0;
   r.readVersion = 0;
   ConflictSet_check(cs, &r, &result, 1);
   assert(result == ConflictSet_Conflict);
+  bytes = ConflictSet_getBytes(cs);
+  assert(bytes > 0);
   ConflictSet_destroy(cs);
   return 0;
 }

conflict_set_cxx_api_test.cpp

@@ -2,14 +2,15 @@
 
 #include <cassert>
 
+using namespace weaselab;
+
 int main(void) {
   ConflictSet cs(0);
   ConflictSet::WriteRange w;
   w.begin.p = (const uint8_t *)"0000";
   w.begin.len = 4;
   w.end.len = 0;
-  w.writeVersion = 1;
-  cs.addWrites(&w, 1);
+  cs.addWrites(&w, 1, 1);
   ConflictSet::Result result;
   ConflictSet::ReadRange r;
   r.begin.p = (const uint8_t *)"0000";
@@ -18,4 +19,6 @@ int main(void) {
   r.readVersion = 0;
   cs.check(&r, &result, 1);
   assert(result == ConflictSet::Conflict);
+  int64_t bytes = cs.getBytes();
+  assert(bytes > 0);
 }