Ignore itlb in grafana instead

Events in the same group should be associated with the same set of
instructions

Bench.cpp

@@ -258,4 +258,114 @@ void benchConflictSet() {
   }
 }
 
-int main(void) { benchConflictSet(); }
+constexpr int kKeyLenForWorstCase = 50;
+
+ConflictSet worstCaseConflictSetForRadixRangeRead(int cardinality) {
+  ConflictSet cs{0};
+
+  for (int i = 0; i < kKeyLenForWorstCase; ++i) {
+    for (int j = 0; j < cardinality; ++j) {
+      auto b = std::vector<uint8_t>(i, 0);
+      b.push_back(j);
+      auto e = std::vector<uint8_t>(i, 255);
+      e.push_back(255 - j);
+      weaselab::ConflictSet::WriteRange w[] = {{
+                                                   {b.data(), int(b.size())},
+                                                   {nullptr, 0},
+                                               },
+                                               {
+                                                   {e.data(), int(e.size())},
+                                                   {nullptr, 0},
+                                               }};
+      std::sort(std::begin(w), std::end(w),
+                [](const auto &lhs, const auto &rhs) {
+                  int cl = std::min(lhs.begin.len, rhs.begin.len);
+                  if (cl > 0) {
+                    int c = memcmp(lhs.begin.p, rhs.begin.p, cl);
+                    if (c != 0) {
+                      return c < 0;
+                    }
+                  }
+                  return lhs.begin.len < rhs.begin.len;
+                });
+      cs.addWrites(w, sizeof(w) / sizeof(w[0]), 0);
+    }
+  }
+
+  // Defeat short-circuiting on the left
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 0);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  // Defeat short-circuiting on the right
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 255);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  return cs;
+}
+
+void benchWorstCaseForRadixRangeRead() {
+  ankerl::nanobench::Bench bench;
+
+  std::unique_ptr<ConflictSet> cs[256];
+  for (int i = 0; i < 256; ++i) {
+    cs[i] =
+        std::make_unique<ConflictSet>(worstCaseConflictSetForRadixRangeRead(i));
+  }
+
+  auto begin = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 0);
+  begin.push_back(1);
+  auto end = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 255);
+  end.push_back(254);
+
+  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) {
+      result = weaselab::ConflictSet::TooOld;
+      cs[i]->check(&r, &result, 1);
+      if (result != weaselab::ConflictSet::Commit) {
+        abort();
+      }
+    }
+  });
+
+  // for (int i = 0; i < 256; ++i) {
+  //   bench.run("worst case for radix tree, span " + std::to_string(i), [&]() {
+  //     result = weaselab::ConflictSet::TooOld;
+  //     cs[i]->check(&r, &result, 1);
+  //     if (result != weaselab::ConflictSet::Commit) {
+  //       abort();
+  //     }
+  //   });
+  // }
+}
+
+void benchCreateAndDestroy() {
+  ankerl::nanobench::Bench bench;
+
+  bench.run("create and destroy", [&]() { ConflictSet cs{0}; });
+}
+
+int main(void) {
+  benchConflictSet();
+  benchWorstCaseForRadixRangeRead();
+  benchCreateAndDestroy();
+}

CMakeLists.txt

@@ -1,14 +1,16 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.0.4
+  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)
 set(CMAKE_CXX_STANDARD 20)
 
-file(WRITE ${CMAKE_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/version.txt.in
+               ${CMAKE_CURRENT_SOURCE_DIR}/paper/version.txt)
 
 include(CMakePushCheckState)
 include(CheckCXXCompilerFlag)
@@ -17,12 +19,6 @@ include(CheckCXXSourceCompiles)
 
 set(DEFAULT_BUILD_TYPE "Release")
 
-if(EMSCRIPTEN OR CMAKE_SYSTEM_NAME STREQUAL WASI)
-  set(WASM ON)
-else()
-  set(WASM OFF)
-endif()
-
 if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
   message(
     STATUS
@@ -37,6 +33,10 @@ 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(full_relro_flags "-pie;LINKER:-z,relro,-z,now,-z,noexecstack")
 cmake_push_check_state()
@@ -47,7 +47,8 @@ if(HAS_FULL_RELRO)
 endif()
 cmake_pop_check_state()
 
-set(version_script_flags LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+set(version_script_flags
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 cmake_push_check_state()
 list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${version_script_flags})
 check_cxx_source_compiles("int main(){}" HAS_VERSION_SCRIPT FAIL_REGEX
@@ -57,9 +58,11 @@ cmake_pop_check_state()
 option(USE_SIMD_FALLBACK
        "Use fallback implementations of functions that use SIMD" OFF)
 
+option(DISABLE_TSAN "Disable TSAN" OFF)
+
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
-include_directories(SYSTEM ${CMAKE_SOURCE_DIR}/third_party/valgrind)
+include_directories(SYSTEM ${CMAKE_CURRENT_SOURCE_DIR}/third_party/valgrind)
 
 add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
 
@@ -76,15 +79,6 @@ if(EMSCRIPTEN)
 endif()
 
 if(NOT USE_SIMD_FALLBACK)
-  cmake_push_check_state()
-  list(APPEND CMAKE_REQUIRED_FLAGS -msimd128)
-  check_include_file_cxx("wasm_simd128.h" HAS_WASM_SIMD)
-  if(HAS_WASM_SIMD)
-    add_compile_options(-msimd128)
-    add_compile_definitions(HAS_WASM_SIMD)
-  endif()
-  cmake_pop_check_state()
-
   cmake_push_check_state()
   list(APPEND CMAKE_REQUIRED_FLAGS -mavx)
   check_include_file_cxx("immintrin.h" HAS_AVX)
@@ -106,19 +100,20 @@ 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_SOURCE_DIR}/include)
+                           PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
 
 add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 set_target_properties(
   ${PROJECT_NAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                             "${CMAKE_BINARY_DIR}/radix_tree")
+                             "${CMAKE_CURRENT_BINARY_DIR}/radix_tree")
 if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
 if(HAS_VERSION_SCRIPT)
-  target_link_options(${PROJECT_NAME} PRIVATE
-                      LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+  target_link_options(
+    ${PROJECT_NAME} PRIVATE
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
 add_library(${PROJECT_NAME}-static STATIC
@@ -131,7 +126,7 @@ if(APPLE)
   add_custom_command(
     TARGET ${PROJECT_NAME}-static
     PRE_LINK
-    COMMAND ${CMAKE_SOURCE_DIR}/privatize_symbols_macos.sh
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
             $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 else()
   add_custom_command(
@@ -139,60 +134,57 @@ else()
     POST_BUILD
     COMMAND
       ${CMAKE_OBJCOPY}
-      --keep-global-symbols=${CMAKE_SOURCE_DIR}/symbol-exports.txt
+      --keep-global-symbols=${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt
       $<TARGET_FILE:${PROJECT_NAME}-static> || echo
       "Proceeding with all symbols global in static library")
 endif()
 
-set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
-
 include(CTest)
 
-if(BUILD_TESTING)
+# disable tests if this is being used through e.g. FetchContent
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
+
+  set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
   # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
   # coverage with a small number of tests.
 
-  file(GLOB CORPUS_TESTS ${CMAKE_SOURCE_DIR}/corpus/*)
+  file(GLOB CORPUS_TESTS ${CMAKE_CURRENT_SOURCE_DIR}/corpus/*)
 
-  # extra testing that relies on shared libraries, which aren't available with
-  # wasm
-  if(NOT WASM)
-    # Shared library version of FoundationDB's skip list implementation
-    add_library(skip_list SHARED SkipList.cpp)
-    target_compile_options(skip_list PRIVATE -fno-exceptions
-                                             -fvisibility=hidden)
-    target_include_directories(skip_list PUBLIC ${CMAKE_SOURCE_DIR}/include)
-    set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                                               "${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 FoundationDB's skip list implementation
+  add_library(skip_list SHARED SkipList.cpp)
+  target_compile_options(skip_list PRIVATE -fno-exceptions -fvisibility=hidden)
+  target_include_directories(skip_list
+                             PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+  set_target_properties(
+    skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                         "${CMAKE_CURRENT_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 -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})
+  # 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 -fno-exceptions -fvisibility=hidden)
+  target_include_directories(hash_table
+                             PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+  set_target_properties(
+    hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                          "${CMAKE_CURRENT_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(driver_skip_list TestDriver.cpp)
-    target_compile_options(driver_skip_list PRIVATE ${TEST_FLAGS})
-    target_link_libraries(driver_skip_list PRIVATE skip_list)
+  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)
 
-    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)
@@ -237,7 +229,7 @@ if(BUILD_TESTING)
   endforeach()
 
   # tsan tests
-  if(NOT CMAKE_CROSSCOMPILING AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+  if(NOT CMAKE_CROSSCOMPILING AND NOT DISABLE_TSAN)
     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)
@@ -266,15 +258,19 @@ if(BUILD_TESTING)
     set_property(
       DIRECTORY
       APPEND
-      PROPERTY CMAKE_CONFIGURE_DEPENDS ${CMAKE_SOURCE_DIR}/test_conflict_set.py)
+      PROPERTY CMAKE_CONFIGURE_DEPENDS
+               ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py)
     execute_process(
-      COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
-              list OUTPUT_VARIABLE SCRIPT_TESTS)
+      COMMAND ${Python3_EXECUTABLE}
+              ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py list
+      OUTPUT_VARIABLE SCRIPT_TESTS)
     foreach(TEST ${SCRIPT_TESTS})
       add_test(
         NAME script_test_${TEST}
-        COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
-                test ${TEST} --build-dir ${CMAKE_BINARY_DIR})
+        COMMAND
+          ${Python3_EXECUTABLE}
+          ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
+          --build-dir ${CMAKE_CURRENT_BINARY_DIR})
     endforeach()
   endif()
 
@@ -306,27 +302,28 @@ if(BUILD_TESTING)
   add_test(NAME conflict_set_cxx_api_test COMMAND conflict_set_cxx_api_test)
 
   # symbol visibility tests
-  if(NOT WASM AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+  if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
     if(APPLE)
-      set(symbol_exports ${CMAKE_SOURCE_DIR}/apple-symbol-exports.txt)
-      set(symbol_imports ${CMAKE_SOURCE_DIR}/apple-symbol-imports.txt)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-exports.txt)
+      set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-imports.txt)
     else()
-      set(symbol_exports ${CMAKE_SOURCE_DIR}/symbol-exports.txt)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt)
       if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/aarch64-symbol-imports.txt)
+        set(symbol_imports
+            ${CMAKE_CURRENT_SOURCE_DIR}/aarch64-symbol-imports.txt)
       else()
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/symbol-imports.txt)
+        set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-imports.txt)
       endif()
     endif()
     add_test(
       NAME conflict_set_shared_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh $<TARGET_FILE:${PROJECT_NAME}>
-        ${symbol_exports} ${symbol_imports})
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
+        $<TARGET_FILE:${PROJECT_NAME}> ${symbol_exports} ${symbol_imports})
     add_test(
       NAME conflict_set_static_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
         $<TARGET_FILE:${PROJECT_NAME}-static> ${symbol_exports}
         ${symbol_imports})
   endif()
@@ -338,6 +335,16 @@ if(BUILD_TESTING)
   add_executable(real_data_bench RealDataBench.cpp)
   target_link_libraries(real_data_bench PRIVATE ${PROJECT_NAME})
   set_target_properties(real_data_bench PROPERTIES SKIP_BUILD_RPATH ON)
+
+  # fuzzer-based perf
+  add_executable(driver_perf TestDriver.cpp)
+  target_compile_definitions(driver_perf PRIVATE PERF_TEST=1)
+  target_link_libraries(driver_perf PRIVATE ${PROJECT_NAME})
+
+  # server bench
+  add_executable(server_bench ServerBench.cpp)
+  target_link_libraries(server_bench PRIVATE ${PROJECT_NAME})
+  set_target_properties(server_bench PROPERTIES SKIP_BUILD_RPATH ON)
 endif()
 
 # packaging
@@ -369,13 +376,13 @@ set(CMAKE_OSX_DEPLOYMENT_TARGET 11.0)
 if(APPLE)
   find_program(PANDOC_EXE pandoc)
   if(PANDOC_EXE)
-    execute_process(COMMAND ${PANDOC_EXE} ${CMAKE_SOURCE_DIR}/README.md -o
-                            ${CMAKE_BINARY_DIR}/README.txt)
-    set(CPACK_RESOURCE_FILE_README ${CMAKE_BINARY_DIR}/README.txt)
+    execute_process(COMMAND ${PANDOC_EXE} ${CMAKE_CURRENT_SOURCE_DIR}/README.md
+                            -o ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
+    set(CPACK_RESOURCE_FILE_README ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
   endif()
-  configure_file(${CMAKE_SOURCE_DIR}/LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt
-                 COPYONLY)
-  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt)
+  configure_file(${CMAKE_CURRENT_SOURCE_DIR}/LICENSE
+                 ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt COPYONLY)
+  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt)
 endif()
 
 include(CPack)

HashTable.cpp

@@ -98,6 +98,13 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
 
 int64_t ConflictSet::getBytes() const { return -1; }
 
+void ConflictSet::getMetricsV1(MetricsV1 **metrics, int *count) const {
+  *metrics = nullptr;
+  *count = 0;
+}
+
+double ConflictSet::MetricsV1::getValue() const { return 0; }
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
     : impl(new(safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
 

Internal.h

@@ -99,8 +99,7 @@ __attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
   mallocBytesDelta -= s;
 #if SHOW_MEMORY
   mallocBytes -= s;
-  free(p);
-#else
+#endif
 #ifndef NDEBUG
   (char *&)p -= kMallocHeaderSize;
   size_t expected;
@@ -108,7 +107,6 @@ __attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
   assert(s == expected);
 #endif
   free(p);
-#endif
 }
 
 // ==================== BEGIN ARENA IMPL ====================
@@ -257,10 +255,75 @@ template <class T> struct ArenaAlloc {
   void deallocate(T *, size_t) noexcept {}
 };
 
-template <class T> using Vector = std::vector<T, ArenaAlloc<T>>;
-template <class T> auto vector(Arena &arena) {
-  return Vector<T>(ArenaAlloc<T>(&arena));
-}
+template <class T> struct Vector {
+  static_assert(std::is_trivially_destructible_v<T>);
+  static_assert(std::is_trivially_copyable_v<T>);
+
+  explicit Vector(Arena *arena)
+      : arena(arena), t(nullptr), size_(0), capacity(0) {}
+
+  void append(std::span<const T> slice) {
+    if (size_ + int(slice.size()) > capacity) {
+      grow(std::max<int>(size_ + slice.size(), capacity * 2));
+    }
+    if (slice.size() > 0) {
+      memcpy(const_cast<std::remove_const_t<T> *>(t) + size_, slice.data(),
+             slice.size() * sizeof(T));
+    }
+    size_ += slice.size();
+  }
+
+  // Caller must write to the returned slice
+  std::span<T> unsafePrepareAppend(int appendSize) {
+    if (size_ + appendSize > capacity) {
+      grow(std::max<int>(size_ + appendSize, capacity * 2));
+    }
+    auto result = std::span<T>(t + size_, appendSize);
+    size_ += appendSize;
+    return result;
+  }
+
+  void push_back(const T &t) { append(std::span<const T>(&t, 1)); }
+
+  T *begin() { return t; }
+  T *end() { return t + size_; }
+  T *data() { return t; }
+  T &back() {
+    assert(size_ > 0);
+    return t[size_ - 1];
+  }
+  T &operator[](int i) {
+    assert(i >= 0 && i < size_);
+    return t[i];
+  }
+
+  void pop_back() {
+    assert(size_ > 0);
+    --size_;
+  }
+
+  int size() const { return size_; }
+
+  operator std::span<const T>() const { return std::span(t, size_); }
+
+private:
+  void grow(int newCapacity) {
+    capacity = newCapacity;
+    auto old = std::span<const T>(*this);
+    t = (T *)new (std::align_val_t(alignof(T)), *arena)
+        uint8_t[capacity * sizeof(T)];
+    size_ = 0;
+    append(old);
+  }
+
+  Arena *arena;
+  T *t;
+  int size_;
+  int capacity;
+};
+
+template <class T> auto vector(Arena &arena) { return Vector<T>(&arena); }
+
 template <class T, class C> using Set = std::set<T, C, ArenaAlloc<T>>;
 template <class T, class C = std::less<T>> auto set(Arena &arena) {
   return Set<T, C>(ArenaAlloc<T>(&arena));
@@ -414,13 +477,15 @@ inline uint32_t Arbitrary::bounded(uint32_t s) {
 // ==================== END ARBITRARY IMPL ====================
 
 struct ReferenceImpl {
-  explicit ReferenceImpl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
+  explicit ReferenceImpl(int64_t oldestVersion)
+      : oldestVersion(oldestVersion), newestVersion(oldestVersion) {
     writeVersionMap[""] = oldestVersion;
   }
   void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
              int count) const {
     for (int i = 0; i < count; ++i) {
-      if (reads[i].readVersion < oldestVersion) {
+      if (reads[i].readVersion < oldestVersion ||
+          reads[i].readVersion < newestVersion - 2e9) {
         results[i] = ConflictSet::TooOld;
         continue;
       }
@@ -442,6 +507,8 @@ struct ReferenceImpl {
   }
   void addWrites(const ConflictSet::WriteRange *writes, int count,
                  int64_t writeVersion) {
+    assert(writeVersion >= newestVersion);
+    newestVersion = writeVersion;
     for (int i = 0; i < count; ++i) {
       auto begin =
           std::string((const char *)writes[i].begin.p, writes[i].begin.len);
@@ -461,11 +528,12 @@ struct ReferenceImpl {
   }
 
   void setOldestVersion(int64_t oldestVersion) {
-    assert(oldestVersion >= oldestVersion);
+    assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
   }
 
   int64_t oldestVersion;
+  int64_t newestVersion;
   std::map<std::string, int64_t> writeVersionMap;
 };
 
@@ -520,13 +588,17 @@ inline const char *resultToStr(ConflictSet::Result r) {
 
 namespace {
 
-template <class ConflictSetImpl> struct TestDriver {
-  Arbitrary arbitrary;
-  explicit TestDriver(const uint8_t *data, size_t size)
-      : arbitrary({data, size}) {}
+template <class ConflictSetImpl, bool kEnableAssertions = true>
+struct TestDriver {
+  Arbitrary *arbitrary;
+  explicit TestDriver(Arbitrary &a) : arbitrary(&a) {
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+    fprintf(stderr, "%p Initial version: {%" PRId64 "}\n", this, writeVersion);
+#endif
+  }
 
-  int64_t writeVersion = 0;
-  int64_t oldestVersion = 0;
+  int64_t oldestVersion = arbitrary->next();
+  int64_t writeVersion = oldestVersion;
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
 
@@ -534,31 +606,34 @@ template <class ConflictSetImpl> struct TestDriver {
 
   bool ok = true;
 
-  const int prefixLen = arbitrary.bounded(512);
-  const int prefixByte = arbitrary.randT<uint8_t>();
+  const int prefixLen = arbitrary->bounded(512);
+  const int prefixByte = arbitrary->randT<uint8_t>();
 
   // Call until it returns true, for "done". Check internal invariants etc
   // between calls to next.
   bool next() {
-    if (!arbitrary.hasEntropy()) {
+    assert(cs.getBytes() >= 0);
+    if (!arbitrary->hasEntropy()) {
       return true;
     }
     Arena arena;
     {
-      int numPointWrites = arbitrary.bounded(100);
-      int numRangeWrites = arbitrary.bounded(100);
-      int64_t v = ++writeVersion;
+      int numPointWrites = arbitrary->bounded(100);
+      int numRangeWrites = arbitrary->bounded(100);
+      int64_t v =
+          (writeVersion +=
+           arbitrary->bounded(10) ? arbitrary->bounded(10) : arbitrary->next());
       auto *writes =
           new (arena) ConflictSet::WriteRange[numPointWrites + numRangeWrites];
       auto keys = set<std::string_view>(arena);
       while (int(keys.size()) < numPointWrites + numRangeWrites * 2) {
-        if (!arbitrary.hasEntropy()) {
+        if (!arbitrary->hasEntropy()) {
           return true;
         }
-        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
+        int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
         memset(begin, prefixByte, prefixLen);
-        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
+        arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -568,7 +643,7 @@ template <class ConflictSetImpl> struct TestDriver {
                rangesRemaining = numRangeWrites;
            pointsRemaining > 0 || rangesRemaining > 0; ++i) {
         bool pointRead = pointsRemaining > 0 && rangesRemaining > 0
-                             ? bool(arbitrary.bounded(2))
+                             ? bool(arbitrary->bounded(2))
                              : pointsRemaining > 0;
         if (pointRead) {
           assert(pointsRemaining > 0);
@@ -587,46 +662,120 @@ template <class ConflictSetImpl> struct TestDriver {
           ++iter;
           --rangesRemaining;
         }
-#if DEBUG_VERBOSE && !defined(NDEBUG)
-        if (writes[i].end.len == 0) {
-          fprintf(stderr, "Write: {%s} -> %" PRId64 "\n",
-                  printable(writes[i].begin).c_str(), writeVersion);
-        } else {
-          fprintf(stderr, "Write: [%s, %s) -> %" PRId64 "\n",
-                  printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str(), writeVersion);
-        }
-#endif
       }
       assert(iter == keys.end());
       assert(i == numPointWrites + numRangeWrites);
 
+      // Test non-canonical writes
+      if (numPointWrites > 0) {
+        int overlaps = arbitrary->bounded(numPointWrites);
+        for (int i = 0; i < numPointWrites + numRangeWrites && overlaps > 0;
+             ++i) {
+          if (writes[i].end.len == 0) {
+            int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
+            auto *begin = new (arena) uint8_t[keyLen];
+            memset(begin, prefixByte, prefixLen);
+            arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
+            writes[i].end.len = keyLen;
+            writes[i].end.p = begin;
+            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);
+            } else if (c == 0) {
+              // It's a point write after all, I guess
+              writes[i].end.len = 0;
+            }
+            --overlaps;
+          }
+        }
+      }
+      if (arbitrary->bounded(2)) {
+        // Shuffle writes
+        for (int i = numPointWrites + numRangeWrites - 1; i > 0; --i) {
+          int j = arbitrary->bounded(i + 1);
+          if (i != j) {
+            using std::swap;
+            swap(writes[i], writes[j]);
+          }
+        }
+      }
+
+      oldestVersion +=
+          arbitrary->bounded(10) ? arbitrary->bounded(10) : arbitrary->next();
+      oldestVersion = std::min(oldestVersion, writeVersion);
+
+#ifdef THREAD_TEST
+      std::latch ready{1};
+      std::thread thread2{[&]() {
+        ready.count_down();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
+      }};
+      ready.wait();
+#endif
+
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+      for (int i = 0; i < numPointWrites + numRangeWrites; ++i) {
+        if (writes[i].end.len == 0) {
+          fprintf(stderr, "%p Write: {%s}\n", this,
+                  printable(writes[i].begin).c_str());
+        } else {
+          fprintf(stderr, "%p Write: [%s, %s)\n", this,
+                  printable(writes[i].begin).c_str(),
+                  printable(writes[i].end).c_str());
+        }
+      }
+      fprintf(stderr, "%p Write @ %" PRId64 "\n", this, v);
+#endif
+
       CALLGRIND_START_INSTRUMENTATION;
       cs.addWrites(writes, numPointWrites + numRangeWrites, v);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
+      if constexpr (kEnableAssertions) {
+        refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
+      }
 
-      oldestVersion = std::max<int64_t>(writeVersion - arbitrary.bounded(10),
-                                        oldestVersion);
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+      fprintf(stderr, "%p Set oldest version: %" PRId64 "\n", this,
+              oldestVersion);
+#endif
       cs.setOldestVersion(oldestVersion);
-      refImpl.setOldestVersion(oldestVersion);
+      if constexpr (kEnableAssertions) {
+        refImpl.setOldestVersion(oldestVersion);
+      }
+
+#ifdef THREAD_TEST
+      thread2.join();
+#endif
     }
     {
-      int numPointReads = arbitrary.bounded(100);
-      int numRangeReads = arbitrary.bounded(100);
-      int64_t v = std::max<int64_t>(writeVersion - arbitrary.bounded(10), 0);
+      int numPointReads = arbitrary->bounded(100);
+      int numRangeReads = arbitrary->bounded(100);
+
+      int64_t v = std::max<int64_t>(writeVersion - (arbitrary->bounded(10)
+                                                        ? arbitrary->bounded(10)
+                                                        : arbitrary->next()),
+                                    0);
       auto *reads =
           new (arena) ConflictSet::ReadRange[numPointReads + numRangeReads];
       auto keys = set<std::string_view>(arena);
       while (int(keys.size()) < numPointReads + numRangeReads * 2) {
-        if (!arbitrary.hasEntropy()) {
+        if (!arbitrary->hasEntropy()) {
           return true;
         }
-        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
+        int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
         memset(begin, prefixByte, prefixLen);
-        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
+        arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -635,7 +784,7 @@ template <class ConflictSetImpl> struct TestDriver {
       for (int pointsRemaining = numPointReads, rangesRemaining = numRangeReads;
            pointsRemaining > 0 || rangesRemaining > 0; ++i) {
         bool pointRead = pointsRemaining > 0 && rangesRemaining > 0
-                             ? bool(arbitrary.bounded(2))
+                             ? bool(arbitrary->bounded(2))
                              : pointsRemaining > 0;
         if (pointRead) {
           assert(pointsRemaining > 0);
@@ -657,12 +806,12 @@ template <class ConflictSetImpl> struct TestDriver {
         reads[i].readVersion = v;
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (reads[i].end.len == 0) {
-          fprintf(stderr, "Read: {%s} @ %d\n",
-                  printable(reads[i].begin).c_str(), int(reads[i].readVersion));
+          fprintf(stderr, "%p Read: {%s} @ %" PRId64 "\n", this,
+                  printable(reads[i].begin).c_str(), reads[i].readVersion);
         } else {
-          fprintf(stderr, "Read: [%s, %s) @ %d\n",
+          fprintf(stderr, "%p Read: [%s, %s) @ %" PRId64 "\n", this,
                   printable(reads[i].begin).c_str(),
-                  printable(reads[i].end).c_str(), int(reads[i].readVersion));
+                  printable(reads[i].end).c_str(), reads[i].readVersion);
         }
 #endif
       }
@@ -679,7 +828,15 @@ template <class ConflictSetImpl> struct TestDriver {
       std::latch ready{1};
       std::thread thread2{[&]() {
         ready.count_down();
+        // Call all const methods
         cs.check(reads, results3, numPointReads + numRangeReads);
+        cs.getBytes();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
       }};
       ready.wait();
 #endif
@@ -688,26 +845,40 @@ template <class ConflictSetImpl> struct TestDriver {
       cs.check(reads, results1, numPointReads + numRangeReads);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.check(reads, results2, numPointReads + numRangeReads);
+      if constexpr (kEnableAssertions) {
+        // Call remaining const methods
+        cs.getBytes();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
 
-      auto compareResults = [reads](ConflictSet::Result *results1,
-                                    ConflictSet::Result *results2, int count) {
+        refImpl.check(reads, results2, numPointReads + numRangeReads);
+      }
+
+      auto compareResults = [reads, this](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
+                  "%p Expected %s, got %s for read of {%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);
+                  (void *)this, resultToStr(results2[i]),
+                  resultToStr(results1[i]), printable(reads[i].begin).c_str(),
+                  reads[i].readVersion);
+            } else {
+              fprintf(stderr,
+                      "%p Expected %s, got %s for read of [%s, %s) at version "
+                      "%" PRId64 "\n",
+                      (void *)this, resultToStr(results2[i]),
+                      resultToStr(results1[i]),
+                      printable(reads[i].begin).c_str(),
+                      printable(reads[i].end).c_str(), reads[i].readVersion);
             }
             return false;
           }
@@ -715,9 +886,12 @@ template <class ConflictSetImpl> struct TestDriver {
         return true;
       };
 
-      if (!compareResults(results1, results2, numPointReads + numRangeReads)) {
-        ok = false;
-        return true;
+      if constexpr (kEnableAssertions) {
+        if (!compareResults(results1, results2,
+                            numPointReads + numRangeReads)) {
+          ok = false;
+          return true;
+        }
       }
 
 #ifdef THREAD_TEST

Jenkinsfile

@@ -48,6 +48,17 @@ pipeline {
                 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 {
@@ -106,14 +117,18 @@ pipeline {
                 }
             }
             steps {
-                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug")
-                sh '''
-                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
-                sh '''
-                gcovr -f ConflictSet.cpp --fail-under-line 100 > /dev/null
-                '''
+                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
+                """
             }
         }
     }

LongestCommonPrefix.h

@@ -0,0 +1,185 @@
+#pragma once
+
+#include <assert.h>
+#include <bit>
+#include <stdint.h>
+#include <string.h>
+
+#ifdef HAS_AVX
+#include <immintrin.h>
+#elif HAS_ARM_NEON
+#include <arm_neon.h>
+#endif
+
+#ifndef __SANITIZE_THREAD__
+#if defined(__has_feature)
+#if __has_feature(thread_sanitizer)
+#define __SANITIZE_THREAD__
+#endif
+#endif
+#endif
+
+#if defined(HAS_AVX) || defined(HAS_ARM_NEON)
+constexpr int kStride = 64;
+#else
+constexpr int kStride = 16;
+#endif
+
+constexpr int kUnrollFactor = 4;
+
+inline bool compareStride(const uint8_t *ap, const uint8_t *bp) {
+#if defined(HAS_ARM_NEON)
+  static_assert(kStride == 64);
+  uint8x16_t x[4]; // GCOVR_EXCL_LINE
+  for (int i = 0; i < 4; ++i) {
+    x[i] = vceqq_u8(vld1q_u8(ap + i * 16), vld1q_u8(bp + i * 16));
+  }
+  auto results = vreinterpretq_u16_u8(
+      vandq_u8(vandq_u8(x[0], x[1]), vandq_u8(x[2], x[3])));
+  bool eq = vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(results, 4)), 0) ==
+            uint64_t(-1);
+#elif defined(HAS_AVX)
+  static_assert(kStride == 64);
+  __m128i x[4]; // GCOVR_EXCL_LINE
+  for (int i = 0; i < 4; ++i) {
+    x[i] = _mm_cmpeq_epi8(_mm_loadu_si128((__m128i *)(ap + i * 16)),
+                          _mm_loadu_si128((__m128i *)(bp + i * 16)));
+  }
+  auto eq =
+      _mm_movemask_epi8(_mm_and_si128(_mm_and_si128(x[0], x[1]),
+                                      _mm_and_si128(x[2], x[3]))) == 0xffff;
+#else
+  // Hope it gets vectorized
+  auto eq = memcmp(ap, bp, kStride) == 0;
+#endif
+  return eq;
+}
+
+// Precondition: ap[:kStride] != bp[:kStride]
+inline int firstNeqStride(const uint8_t *ap, const uint8_t *bp) {
+#if defined(HAS_AVX)
+  static_assert(kStride == 64);
+  uint64_t c[kStride / 16]; // GCOVR_EXCL_LINE
+  for (int i = 0; i < kStride; i += 16) {
+    const auto a = _mm_loadu_si128((__m128i *)(ap + i));
+    const auto b = _mm_loadu_si128((__m128i *)(bp + i));
+    const auto compared = _mm_cmpeq_epi8(a, b);
+    c[i / 16] = _mm_movemask_epi8(compared) & 0xffff;
+  }
+  return std::countr_zero(~(c[0] | c[1] << 16 | c[2] << 32 | c[3] << 48));
+#elif defined(HAS_ARM_NEON)
+  static_assert(kStride == 64);
+  for (int i = 0; i < kStride; i += 16) {
+    // 0xff for each match
+    uint16x8_t results =
+        vreinterpretq_u16_u8(vceqq_u8(vld1q_u8(ap + i), vld1q_u8(bp + i)));
+    // 0xf for each mismatch
+    uint64_t bitfield =
+        ~vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(results, 4)), 0);
+    if (bitfield) {
+      return i + (std::countr_zero(bitfield) >> 2);
+    }
+  }
+  __builtin_unreachable(); // GCOVR_EXCL_LINE
+#else
+  int i = 0;
+  for (; i < kStride - 1; ++i) {
+    if (*ap++ != *bp++) {
+      break;
+    }
+  }
+  return i;
+#endif
+}
+
+// This gets covered in local development
+// GCOVR_EXCL_START
+#if defined(HAS_AVX) && !defined(__SANITIZE_THREAD__)
+__attribute__((target("avx512f,avx512bw"))) inline int
+longestCommonPrefix(const uint8_t *ap, const uint8_t *bp, int cl) {
+  int i = 0;
+  int end = cl & ~63;
+  while (i < end) {
+    const uint64_t eq =
+        _mm512_cmpeq_epi8_mask(_mm512_loadu_epi8(ap), _mm512_loadu_epi8(bp));
+    if (eq != uint64_t(-1)) {
+      return i + std::countr_one(eq);
+    }
+    i += 64;
+    ap += 64;
+    bp += 64;
+  }
+  if (i < cl) {
+    const uint64_t mask = (uint64_t(1) << (cl - i)) - 1;
+    const uint64_t eq = _mm512_cmpeq_epi8_mask(
+        _mm512_maskz_loadu_epi8(mask, ap), _mm512_maskz_loadu_epi8(mask, bp));
+    return i + std::countr_one(eq & mask);
+  }
+  assert(i == cl);
+  return i;
+}
+__attribute__((target("default")))
+#endif
+// GCOVR_EXCL_STOP
+
+inline int
+longestCommonPrefix(const uint8_t *ap, const uint8_t *bp, int cl) {
+  if (!(cl >= 0)) {
+    __builtin_unreachable(); // GCOVR_EXCL_LINE
+  }
+
+  int i = 0;
+  int end;
+
+  // kStride * kUnrollCount at a time
+  end = cl & ~(kStride * kUnrollFactor - 1);
+  while (i < end) {
+    for (int j = 0; j < kUnrollFactor; ++j) {
+      if (!compareStride(ap, bp)) {
+        return i + firstNeqStride(ap, bp);
+      }
+      i += kStride;
+      ap += kStride;
+      bp += kStride;
+    }
+  }
+
+  // kStride at a time
+  end = cl & ~(kStride - 1);
+  while (i < end) {
+    if (!compareStride(ap, bp)) {
+      return i + firstNeqStride(ap, bp);
+    }
+    i += kStride;
+    ap += kStride;
+    bp += kStride;
+  }
+
+  // word at a time
+  end = cl & ~(sizeof(uint64_t) - 1);
+  while (i < end) {
+    uint64_t a; // GCOVR_EXCL_LINE
+    uint64_t b; // GCOVR_EXCL_LINE
+    memcpy(&a, ap, 8);
+    memcpy(&b, bp, 8);
+    const auto mismatched = a ^ b;
+    if (mismatched) {
+      return i + std::countr_zero(mismatched) / 8;
+    }
+    i += 8;
+    ap += 8;
+    bp += 8;
+  }
+
+  // byte at a time
+  while (i < cl) {
+    if (*ap != *bp) {
+      break;
+    }
+    ++ap;
+    ++bp;
+    ++i;
+  }
+
+  return i;
+}

README.md

@@ -1,84 +1,38 @@
 A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys.
 
-Intended to replace FoundationDB's skip list.
+Intended as an alternative to FoundationDB's skip list.
 
-Hardware for all benchmarks is a mac m1 2020.
+Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM
 
-# FoundationDB's benchmark
+# Microbenchmark
 
 ## Skip list
 
-```
-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: 0.0546
-                 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
-```
+|               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
+|              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)
 
-```
-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: 0.0594
-                 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                           |
-| -----: | -----------: | ---: | ----: | :---------------------------------- |
-| 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` |
+|               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
+|               11.04 |       90,614,308.12 |    0.8% |          180.38 |           55.13 |  3.272 |          41.51 |    0.4% |      0.01 | `point reads`
+|               14.96 |       66,843,629.12 |    0.4% |          274.41 |           74.73 |  3.672 |          55.05 |    0.3% |      0.01 | `prefix reads`
+|               37.06 |       26,982,847.61 |    0.2% |          791.04 |          185.28 |  4.269 |         142.67 |    0.2% |      0.01 | `range reads`
+|               17.89 |       55,887,365.73 |    0.6% |          335.54 |           89.79 |  3.737 |          43.84 |    0.4% |      0.01 | `point writes`
+|               31.85 |       31,394,336.65 |    0.3% |          615.32 |          159.63 |  3.855 |          87.69 |    0.2% |      0.01 | `prefix writes`
+|               36.17 |       27,647,221.45 |    0.6% |          705.11 |          182.80 |  3.857 |         100.62 |    0.1% |      0.01 | `range writes`
+|               79.01 |       12,656,457.78 |    0.7% |        1,498.35 |          402.46 |  3.723 |         270.50 |    0.1% |      0.01 | `monotonic increasing point writes`
+|          303,667.50 |            3,293.08 |    1.1% |    3,931,273.00 |    1,612,702.50 |  2.438 |     806,223.33 |    0.0% |      0.01 | `worst case for radix tree`
+|               83.70 |       11,947,443.83 |    0.7% |        1,738.03 |          429.06 |  4.051 |         270.01 |    0.0% |      0.01 | `create and destroy`
 
 # "Real data" test
 
@@ -87,13 +41,13 @@ Point queries only, best of three runs. Gc ratio is the ratio of time spent doin
 ## skip list
 
 ```
-Check: 11.3385 seconds, 329.718 MB/s, Add: 5.35612 seconds, 131.072 MB/s, Gc ratio: 45.7173%
+Check: 4.47891 seconds, 364.05 MB/s, Add: 4.55599 seconds, 123.058 MB/s, Gc ratio: 37.1145%
 ```
 
 ## radix tree
 
 ```
-Check: 2.48583 seconds, 1503.93 MB/s, Add: 2.12768 seconds, 329.954 MB/s, Gc ratio: 41.7943%
+Check: 0.958985 seconds, 1700.28 MB/s, Add: 1.35083 seconds, 415.044 MB/s, Gc ratio: 44.4768%, Peak idle memory: 2.33588e+06
 ```
 
 ## hash table
@@ -101,5 +55,5 @@ Check: 2.48583 seconds, 1503.93 MB/s, Add: 2.12768 seconds, 329.954 MB/s, Gc rat
 (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%
+Check: 0.804094 seconds, 2027.81 MB/s, Add: 0.652952 seconds, 858.645 MB/s, Gc ratio: 35.3885%
 ```

RealDataBench.cpp

@@ -129,6 +129,16 @@ int main(int argc, const char **argv) {
     close(fd);
   }
 
+  ConflictSet::MetricsV1 *metrics;
+  int metricsCount;
+  cs.getMetricsV1(&metrics, &metricsCount);
+  for (int i = 0; i < metricsCount; ++i) {
+    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: "
          "%g%%, Peak idle memory: %g\n",
          checkTime, checkBytes / checkTime * 1e-6, addTime,

ServerBench.cpp

@@ -0,0 +1,370 @@
+#include <atomic>
+#include <errno.h>
+#include <netdb.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <string>
+#include <string_view>
+#include <sys/ioctl.h>
+#include <sys/resource.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <thread>
+#include <unistd.h>
+#include <utility>
+#include <vector>
+
+#include "ConflictSet.h"
+#include "third_party/nadeau.h"
+
+std::atomic<int64_t> transactions;
+
+constexpr int kBaseSearchDepth = 32;
+constexpr int kWindowSize = 10000000;
+
+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;
+}
+
+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();
+      //   }
+      // }
+    }
+    // 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;
+  }
+}
+
+// Adapted from getaddrinfo man page
+int getListenFd(const char *node, const char *service) {
+
+  struct addrinfo hints;
+  struct addrinfo *result, *rp;
+  int sfd, s;
+
+  memset(&hints, 0, sizeof(hints));
+  hints.ai_family = AF_UNSPEC;     /* Allow IPv4 or IPv6 */
+  hints.ai_socktype = SOCK_STREAM; /* stream socket */
+  hints.ai_flags = AI_PASSIVE;     /* For wildcard IP address */
+  hints.ai_protocol = 0;           /* Any protocol */
+  hints.ai_canonname = nullptr;
+  hints.ai_addr = nullptr;
+  hints.ai_next = nullptr;
+
+  s = getaddrinfo(node, service, &hints, &result);
+  if (s != 0) {
+    fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(s));
+    abort();
+  }
+
+  /* getaddrinfo() returns a list of address structures.
+     Try each address until we successfully bind(2).
+     If socket(2) (or bind(2)) fails, we (close the socket
+     and) try the next address. */
+
+  for (rp = result; rp != nullptr; rp = rp->ai_next) {
+    sfd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
+    if (sfd == -1) {
+      continue;
+    }
+
+    int val = 1;
+    setsockopt(sfd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
+
+    if (bind(sfd, rp->ai_addr, rp->ai_addrlen) == 0) {
+      break; /* Success */
+    }
+
+    close(sfd);
+  }
+
+  freeaddrinfo(result); /* No longer needed */
+
+  if (rp == nullptr) { /* No address succeeded */
+    fprintf(stderr, "Could not bind\n");
+    abort();
+  }
+
+  int rv = listen(sfd, SOMAXCONN);
+  if (rv) {
+    perror("listen()");
+    abort();
+  }
+
+  return sfd;
+}
+
+// HTTP response
+//
+std::string_view part1 =
+    "HTTP/1.1 200 OK \r\nContent-type: text/plain; version=0.0.4; "
+    "charset=utf-8; escaping=values\r\nContent-Length: ";
+// Decimal content length
+std::string_view part2 = "\r\n\r\n";
+// Body
+
+double toSeconds(timeval t) {
+  return double(t.tv_sec) + double(t.tv_usec) * 1e-6;
+}
+
+#ifdef __linux__
+#include <linux/perf_event.h>
+struct PerfCounter {
+  PerfCounter(int type, int config, const std::string &labels = {},
+              int groupLeaderFd = -1)
+      : labels(labels) {
+    struct perf_event_attr pe;
+
+    memset(&pe, 0, sizeof(pe));
+    pe.type = type;
+    pe.size = sizeof(pe);
+    pe.config = config;
+    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());
+    }
+  }
+
+  int64_t total() const {
+    int64_t count;
+    if (read(fd, &count, sizeof(count)) != sizeof(count)) {
+      perror("read instructions from perf");
+      abort();
+    }
+    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; }
+
+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;
+
+    ret = syscall(SYS_perf_event_open, hw_event, pid, cpu, group_fd, flags);
+    return ret;
+  }
+};
+#else
+struct PerfCounter {
+  PerfCounter(int, int) {}
+  int64_t total() { return 0; }
+};
+#endif
+
+int main(int argc, char **argv) {
+  if (argc != 3) {
+    goto fail;
+  }
+  {
+    int listenFd = getListenFd(argv[1], argv[2]);
+
+    weaselab::ConflictSet cs{0};
+    weaselab::ConflictSet::MetricsV1 *metrics;
+    int metricsCount;
+    cs.getMetricsV1(&metrics, &metricsCount);
+
+    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();
+            }
+          }
+        }
+      }
+    }
+
+    auto w = std::thread{workload, &cs};
+
+    for (;;) {
+      struct sockaddr_storage peer_addr = {};
+      socklen_t peer_addr_len = sizeof(peer_addr);
+      const int connfd =
+          accept(listenFd, (struct sockaddr *)&peer_addr, &peer_addr_len);
+
+      std::string body;
+
+      rusage r;
+      getrusage(RUSAGE_SELF, &r);
+      body += "# HELP process_cpu_seconds_total Total user and system CPU time "
+              "spent in seconds.\n# TYPE process_cpu_seconds_total counter\n"
+              "process_cpu_seconds_total ";
+      body += std::to_string(toSeconds(r.ru_utime) + toSeconds(r.ru_stime));
+      body += "\n";
+      body += "# HELP process_resident_memory_bytes Resident memory size in "
+              "bytes.\n# TYPE process_resident_memory_bytes gauge\n"
+              "process_resident_memory_bytes ";
+      body += std::to_string(getCurrentRSS());
+      body += "\n";
+      body += "# HELP transactions_total Total number of transactions\n"
+              "# TYPE transactions_total counter\n"
+              "transactions_total ";
+      body += std::to_string(transactions.load(std::memory_order_relaxed));
+      body += "\n";
+      body += "# HELP instructions_total Total number of instructions\n"
+              "# TYPE instructions_total counter\n"
+              "instructions_total ";
+      body += std::to_string(instructions.total());
+      body += "\n";
+      body += "# HELP cycles_total Total number of cycles\n"
+              "# TYPE cycles_total counter\n"
+              "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";
+      }
+
+      for (int i = 0; i < metricsCount; ++i) {
+        body += "# HELP ";
+        body += metrics[i].name;
+        body += " ";
+        body += metrics[i].help;
+        body += "\n";
+        body += "# TYPE ";
+        body += metrics[i].name;
+        body += " ";
+        body += metrics[i].type == metrics[i].Counter ? "counter" : "gauge";
+        body += "\n";
+        body += metrics[i].name;
+        body += " ";
+        body += std::to_string(metrics[i].getValue());
+        body += "\n";
+      }
+
+      auto len = std::to_string(body.size());
+      iovec iov[] = {
+          {(void *)part1.data(), part1.size()},
+          {(void *)len.data(), len.size()},
+          {(void *)part2.data(), part2.size()},
+          {(void *)body.data(), body.size()},
+      };
+      int written;
+      do {
+        written = writev(connfd, iov, sizeof(iov) / sizeof(iov[0]));
+      } while (written < 0 && errno == EINTR);
+      close(connfd);
+    }
+  }
+fail:
+  fprintf(stderr, "Expected ./%s <host> <port>\n", argv[0]);
+  return 1;
+}

SkipList.cpp

@@ -22,6 +22,8 @@
 
 #include "ConflictSet.h"
 #include "Internal.h"
+
+#include <algorithm>
 #include <span>
 
 std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
@@ -40,7 +42,7 @@ std::span<const uint8_t> copyToArena(Arena &arena,
 }
 
 using Version = int64_t;
-#define force_inline __attribute__((always_inline))
+#define force_inline inline __attribute__((always_inline))
 using StringRef = std::span<const uint8_t>;
 
 struct KeyRangeRef {
@@ -52,6 +54,135 @@ struct KeyRangeRef {
       : begin(begin), end(keyAfter(arena, begin)) {}
 };
 
+struct KeyInfo {
+  StringRef key;
+  bool begin;
+  bool write;
+
+  KeyInfo() = default;
+  KeyInfo(StringRef key, bool begin, bool write)
+      : key(key), begin(begin), write(write) {}
+};
+
+force_inline int extra_ordering(const KeyInfo &ki) {
+  return ki.begin * 2 + (ki.write ^ ki.begin);
+}
+
+// returns true if done with string
+force_inline bool getCharacter(const KeyInfo &ki, int character,
+                               int &outputCharacter) {
+  // normal case
+  if (character < ki.key.size()) {
+    outputCharacter = 5 + ki.key.begin()[character];
+    return false;
+  }
+
+  // termination
+  if (character == ki.key.size()) {
+    outputCharacter = 0;
+    return false;
+  }
+
+  if (character == ki.key.size() + 1) {
+    // end/begin+read/write relative sorting
+    outputCharacter = extra_ordering(ki);
+    return false;
+  }
+
+  outputCharacter = 0;
+  return true;
+}
+
+bool operator<(const KeyInfo &lhs, const KeyInfo &rhs) {
+  int i = std::min(lhs.key.size(), rhs.key.size());
+  int c = memcmp(lhs.key.data(), rhs.key.data(), i);
+  if (c != 0)
+    return c < 0;
+
+  // Always sort shorter keys before longer keys.
+  if (lhs.key.size() < rhs.key.size()) {
+    return true;
+  }
+  if (lhs.key.size() > rhs.key.size()) {
+    return false;
+  }
+
+  // When the keys are the same length, use the extra ordering constraint.
+  return extra_ordering(lhs) < extra_ordering(rhs);
+}
+
+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;
+  int character;
+  SortTask(int begin, int size, int character)
+      : begin(begin), size(size), character(character) {}
+};
+
+void sortPoints(std::vector<KeyInfo> &points) {
+  std::vector<SortTask> tasks;
+  std::vector<KeyInfo> newPoints;
+  std::vector<int> counts;
+
+  tasks.emplace_back(0, points.size(), 0);
+
+  while (tasks.size()) {
+    SortTask st = tasks.back();
+    tasks.pop_back();
+
+    if (st.size < 10) {
+      std::sort(points.begin() + st.begin, points.begin() + st.begin + st.size);
+      continue;
+    }
+
+    newPoints.resize(st.size);
+    counts.assign(256 + 5, 0);
+
+    // get counts
+    int c;
+    bool allDone = true;
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      allDone &= getCharacter(points[i], st.character, c);
+      counts[c]++;
+    }
+    if (allDone)
+      continue;
+
+    // calculate offsets from counts and build next level of tasks
+    int total = 0;
+    for (int i = 0; i < counts.size(); i++) {
+      int temp = counts[i];
+      if (temp > 1)
+        tasks.emplace_back(st.begin + total, temp, st.character + 1);
+      counts[i] = total;
+      total += temp;
+    }
+
+    // put in their places
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      getCharacter(points[i], st.character, c);
+      newPoints[counts[c]++] = points[i];
+    }
+
+    // copy back into original points array
+    for (int i = 0; i < st.size; i++)
+      points[st.begin + i] = newPoints[i];
+  }
+}
+
 static thread_local uint32_t g_seed = 0;
 
 static inline int skfastrand() {
@@ -577,7 +708,8 @@ struct SkipListConflictSet {};
 
 struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   Impl(int64_t oldestVersion)
-      : oldestVersion(oldestVersion), skipList(oldestVersion) {}
+      : oldestVersion(oldestVersion), newestVersion(oldestVersion),
+        skipList(oldestVersion) {}
   void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
              int count) const {
     Arena arena;
@@ -592,7 +724,8 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     }
     skipList.detectConflicts(ranges, count, results);
     for (int i = 0; i < count; ++i) {
-      if (reads[i].readVersion < oldestVersion) {
+      if (reads[i].readVersion < oldestVersion ||
+          reads[i].readVersion < newestVersion - 2e9) {
         results[i] = TooOld;
       }
     }
@@ -600,8 +733,40 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
 
   void addWrites(const ConflictSet::WriteRange *writes, int count,
                  int64_t writeVersion) {
+    auto points = std::vector<KeyInfo>(count * 2);
     Arena arena;
-    const int stringCount = count * 2;
+
+    for (int r = 0; r < count; r++) {
+      points.emplace_back(StringRef(writes[r].begin.p, writes[r].begin.len),
+                          true, true);
+      points.emplace_back(
+          writes[r].end.len > 0
+              ? StringRef{writes[r].end.p, size_t(writes[r].end.len)}
+              : keyAfter(arena, points.back().key),
+          false, true);
+    }
+
+    sortPoints(points);
+
+    int activeWriteCount = 0;
+    std::vector<std::pair<StringRef, StringRef>> combinedWriteConflictRanges;
+    for (const KeyInfo &point : points) {
+      if (point.write) {
+        if (point.begin) {
+          activeWriteCount++;
+          if (activeWriteCount == 1)
+            combinedWriteConflictRanges.emplace_back(point.key, StringRef());
+        } else /*if (point.end)*/ {
+          activeWriteCount--;
+          if (activeWriteCount == 0)
+            combinedWriteConflictRanges.back().second = point.key;
+        }
+      }
+    }
+
+    assert(writeVersion >= newestVersion);
+    newestVersion = writeVersion;
+    const int stringCount = combinedWriteConflictRanges.size() * 2;
 
     const int stripeSize = 16;
     SkipList::Finger fingers[stripeSize];
@@ -612,15 +777,9 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     int ss = stringCount - (stripes - 1) * stripeSize;
     for (int s = stripes - 1; s >= 0; s--) {
       for (int i = 0; i * 2 < ss; ++i) {
-        const auto &w = writes[s * stripeSize / 2 + i];
-#if DEBUG_VERBOSE
-        printf("Write begin: %s\n", printable(w.begin).c_str());
-        fflush(stdout);
-#endif
-        values[i * 2] = {w.begin.p, size_t(w.begin.len)};
-        values[i * 2 + 1] = w.end.len > 0
-                                ? StringRef{w.end.p, size_t(w.end.len)}
-                                : keyAfter(arena, values[i * 2]);
+        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);
@@ -630,6 +789,7 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   }
 
   void setOldestVersion(int64_t oldestVersion) {
+    assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
     SkipList::Finger finger;
     int temp;
@@ -648,16 +808,20 @@ private:
   Arena removalArena;
   std::span<const uint8_t> removalKey;
   int64_t oldestVersion;
+  int64_t newestVersion;
   SkipList skipList;
 };
 
-void ConflictSet::check(const ReadRange *reads, Result *results,
-                        int count) const {
+// Internal entry points. Public entry points should just delegate to these
+
+void internal_check(ConflictSet::Impl *impl,
+                    const ConflictSet::ReadRange *reads,
+                    ConflictSet::Result *results, int count) {
   impl->check(reads, results, count);
 }
-
-void ConflictSet::addWrites(const WriteRange *writes, int count,
-                            int64_t writeVersion) {
+void internal_addWrites(ConflictSet::Impl *impl,
+                        const ConflictSet::WriteRange *writes, int count,
+                        int64_t writeVersion) {
   mallocBytesDelta = 0;
   impl->addWrites(writes, count, writeVersion);
   impl->totalBytes += mallocBytesDelta;
@@ -668,7 +832,7 @@ void ConflictSet::addWrites(const WriteRange *writes, int count,
 #endif
 }
 
-void ConflictSet::setOldestVersion(int64_t oldestVersion) {
+void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
   mallocBytesDelta = 0;
   impl->setOldestVersion(oldestVersion);
   impl->totalBytes += mallocBytesDelta;
@@ -678,19 +842,61 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
   }
 #endif
 }
+ConflictSet::Impl *internal_create(int64_t oldestVersion) {
+  mallocBytesDelta = 0;
+  auto *result = new (safe_malloc(sizeof(ConflictSet::Impl)))
+      ConflictSet::Impl{oldestVersion};
+  result->totalBytes += mallocBytesDelta;
+  return result;
+}
 
-int64_t ConflictSet::getBytes() const { return impl->totalBytes; }
+void internal_destroy(ConflictSet::Impl *impl) {
+  impl->~Impl();
+  safe_free(impl, sizeof(ConflictSet::Impl));
+}
+
+int64_t internal_getBytes(ConflictSet::Impl *impl) { return impl->totalBytes; }
+
+void internal_getMetricsV1(ConflictSet::Impl *impl,
+                           ConflictSet::MetricsV1 **metrics, int *count) {
+  *metrics = nullptr;
+  *count = 0;
+}
+
+double internal_getMetricValue(const ConflictSet::MetricsV1 *metric) {
+  return 0;
+}
+
+void ConflictSet::check(const ReadRange *reads, Result *results,
+                        int count) const {
+  internal_check(impl, reads, results, count);
+}
+
+void ConflictSet::addWrites(const WriteRange *writes, int count,
+                            int64_t writeVersion) {
+  internal_addWrites(impl, writes, count, writeVersion);
+}
+
+void ConflictSet::setOldestVersion(int64_t oldestVersion) {
+  internal_setOldestVersion(impl, oldestVersion);
+}
+
+int64_t ConflictSet::getBytes() const { return internal_getBytes(impl); }
+
+void ConflictSet::getMetricsV1(MetricsV1 **metrics, int *count) const {
+  return internal_getMetricsV1(impl, metrics, count);
+}
+
+double ConflictSet::MetricsV1::getValue() const {
+  return internal_getMetricValue(this);
+}
 
 ConflictSet::ConflictSet(int64_t oldestVersion)
-    : impl((mallocBytesDelta = 0,
-            new(safe_malloc(sizeof(Impl))) Impl{oldestVersion})) {
-  impl->totalBytes += mallocBytesDelta;
-}
+    : impl(internal_create(oldestVersion)) {}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
-    impl->~Impl();
-    safe_free(impl, sizeof(Impl));
+    internal_destroy(impl);
   }
 }
 
@@ -711,50 +917,27 @@ 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);
+  internal_check((ConflictSet::Impl *)cs, reads, results, count);
 }
 __attribute__((__visibility__("default"))) void
 ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes, int count,
                       int64_t writeVersion) {
-  auto *impl = (ConflictSet::Impl *)cs;
-  mallocBytesDelta = 0;
-  impl->addWrites(writes, count, writeVersion);
-  impl->totalBytes += mallocBytesDelta;
-#if SHOW_MEMORY
-  if (impl->totalBytes != mallocBytes) {
-    abort();
-  }
-#endif
+  internal_addWrites((ConflictSet::Impl *)cs, writes, count, writeVersion);
 }
 __attribute__((__visibility__("default"))) void
 ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
-  auto *impl = (ConflictSet::Impl *)cs;
-  mallocBytesDelta = 0;
-  impl->setOldestVersion(oldestVersion);
-  impl->totalBytes += mallocBytesDelta;
-#if SHOW_MEMORY
-  if (impl->totalBytes != mallocBytes) {
-    abort();
-  }
-#endif
+  internal_setOldestVersion((ConflictSet::Impl *)cs, oldestVersion);
 }
 __attribute__((__visibility__("default"))) void *
 ConflictSet_create(int64_t oldestVersion) {
-  mallocBytesDelta = 0;
-  auto *result = new (safe_malloc(sizeof(ConflictSet::Impl)))
-      ConflictSet::Impl{oldestVersion};
-  result->totalBytes += mallocBytesDelta;
-  return result;
+  return internal_create(oldestVersion);
 }
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
-  using Impl = ConflictSet::Impl;
-  ((Impl *)cs)->~Impl();
-  safe_free(cs, sizeof(Impl));
+  internal_destroy((ConflictSet::Impl *)cs);
 }
 __attribute__((__visibility__("default"))) int64_t
 ConflictSet_getBytes(void *cs) {
-  using Impl = ConflictSet::Impl;
-  return ((Impl *)cs)->totalBytes;
+  return internal_getBytes((ConflictSet::Impl *)cs);
 }
 }
 

TestDriver.cpp

@@ -3,17 +3,68 @@
 #include <fstream>
 #include <sstream>
 
+#ifndef PERF_TEST
+#define PERF_TEST 0
+#endif
+
 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();
-    TestDriver<ConflictSet> driver{(const uint8_t *)str.data(), str.size()};
-    while (!driver.next())
-      ;
-    if (!driver.ok) {
-      abort();
+    Arbitrary arbitrary({(const uint8_t *)str.data(), str.size()});
+    TestDriver<ConflictSet, !PERF_TEST> driver1{arbitrary};
+    TestDriver<ConflictSet, !PERF_TEST> driver2{arbitrary};
+    bool done1 = false;
+    bool done2 = false;
+    for (;;) {
+      if (!done1) {
+        done1 = driver1.next();
+        if (!driver1.ok) {
+          abort();
+        }
+      }
+      if (!done2) {
+        done2 = driver2.next();
+        if (!driver2.ok) {
+          abort();
+        }
+      }
+      if (done1 && done2) {
+        break;
+      }
+    }
+
+    {
+      ConflictSet::MetricsV1 *metrics;
+      int metricsCount;
+      driver1.cs.getMetricsV1(&metrics, &metricsCount);
+      printf("#################### METRICS for ConflictSet 1 for %s "
+             "####################\n",
+             argv[i]);
+      for (int i = 0; i < metricsCount; ++i) {
+        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());
+      }
+      puts("");
+    }
+    {
+      ConflictSet::MetricsV1 *metrics;
+      int metricsCount;
+      driver2.cs.getMetricsV1(&metrics, &metricsCount);
+      printf("#################### METRICS for ConflictSet 2 for %s "
+             "####################\n",
+             argv[i]);
+      for (int i = 0; i < metricsCount; ++i) {
+        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());
+      }
+      puts("");
     }
   }
 }

aarch64-symbol-imports.txt

@@ -1,3 +1,5 @@
+__aarch64_ldadd8_relax
+__getauxval@GLIBC_2.17
 __stack_chk_fail@GLIBC_2.17
 __stack_chk_guard@GLIBC_2.17
 abort@GLIBC_2.17

apple-symbol-exports.txt

@@ -13,5 +13,7 @@ __ZN8weaselab11ConflictSetC2Ex
 __ZN8weaselab11ConflictSetD1Ev
 __ZN8weaselab11ConflictSetD2Ev
 __ZN8weaselab11ConflictSetaSEOS0_
+__ZNK8weaselab11ConflictSet12getMetricsV1EPPNS0_9MetricsV1EPi
 __ZNK8weaselab11ConflictSet5checkEPKNS0_9ReadRangeEPNS0_6ResultEi
 __ZNK8weaselab11ConflictSet8getBytesEv
+__ZNK8weaselab11ConflictSet9MetricsV18getValueEv

apple-symbol-imports.txt

@@ -1,3 +1,5 @@
+___stack_chk_fail
+___stack_chk_guard
 __tlv_bootstrap
 _abort
 _bzero

conflict_set.py

@@ -115,7 +115,9 @@ class ConflictSet:
 
     def check(self, *reads: ReadRange) -> list[Result]:
         r = (ctypes.c_int * len(reads))()
-        self._lib.ConflictSet_check(self.p, *reads, r, 1)
+        self._lib.ConflictSet_check(
+            self.p, (ReadRange * len(reads))(*reads), r, len(reads)
+        )
         return [Result(x) for x in r]
 
     def setOldestVersion(self, version: int) -> None:

conflict_set_cxx_api_test.cpp

@@ -1,6 +1,7 @@
 #include "ConflictSet.h"
 
 #include <cassert>
+#include <cstdio>
 
 using namespace weaselab;
 
@@ -21,4 +22,14 @@ int main(void) {
   assert(result == ConflictSet::Conflict);
   int64_t bytes = cs.getBytes();
   assert(bytes > 0);
+
+  ConflictSet::MetricsV1 *metrics;
+  int metricsCount;
+  cs.getMetricsV1(&metrics, &metricsCount);
+  for (int i = 0; i < metricsCount; ++i) {
+    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());
+  }
 }