Update InternalVersionT::zero in addWrite call

There can be other conflict sets in the same thread

.clangd

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

.gitignore

@@ -1,2 +1,3 @@
 .cache
+__pycache__
 build

.pre-commit-config.yaml

@@ -1,11 +1,11 @@
 repos:
   - repo: https://github.com/pre-commit/mirrors-clang-format
-    rev: b111689e7b5cba60be3c62d5db2bd1357f4d36ca
+    rev: 6d365699efc33b1b432eab5b4ae331a19e1857de # frozen: v18.1.2
     hooks:
       - id: clang-format
         exclude: ".*third_party/.*"
   - repo: https://github.com/cheshirekow/cmake-format-precommit
-    rev: e2c2116d86a80e72e7146a06e68b7c228afc6319
+    rev: e2c2116d86a80e72e7146a06e68b7c228afc6319 # frozen: v0.6.13
     hooks:
       - id: cmake-format
   - repo: local
@@ -13,7 +13,7 @@ repos:
       - id: debug verbose check
         name: disallow checking in DEBUG_VERBOSE=1
         description: disallow checking in DEBUG_VERBOSE=1
-        entry: '^#define DEBUG_VERBOSE 1$'
+        entry: "^#define DEBUG_VERBOSE 1$"
         language: pygrep
         types: [c++]
   - repo: local
@@ -21,6 +21,14 @@ repos:
       - id: debug verbose check
         name: disallow checking in SHOW_MEMORY=1
         description: disallow checking in SHOW_MEMORY=1
-        entry: '^#define SHOW_MEMORY 1$'
+        entry: "^#define SHOW_MEMORY 1$"
         language: pygrep
-        types: [c++]
+        types: [c++]
+  - repo: https://github.com/shellcheck-py/shellcheck-py
+    rev: a23f6b85d0fdd5bb9d564e2579e678033debbdff # frozen: v0.10.0.1
+    hooks:
+      - id: shellcheck
+  - repo: https://github.com/psf/black
+    rev: 552baf822992936134cbd31a38f69c8cfe7c0f05 # frozen: 24.3.0
+    hooks:
+      - id: black

.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 +248,124 @@ 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(); }
+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 +1,30 @@
 cmake_minimum_required(VERSION 3.18)
 project(
-  conflict_set
-  VERSION 0.0.1
+  conflict-set
+  VERSION 0.0.9
   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_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)
+include(CheckIncludeFileCXX)
+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
@@ -21,16 +37,36 @@ 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 -fPIC)
+
+set(full_relro_flags "-pie;LINKER:-z,relro,-z,now,-z,noexecstack")
+cmake_push_check_state()
+list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${full_relro_flags})
+check_cxx_source_compiles("int main(){}" HAS_FULL_RELRO FAIL_REGEX "warning:")
+if(HAS_FULL_RELRO)
+  add_link_options(${full_relro_flags})
+endif()
+cmake_pop_check_state()
+
+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
+                          "warning:")
+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)
 
-if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-  add_compile_options(-Wno-maybe-uninitialized
-                      $<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
-endif()
+add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
 
 if(APPLE)
   add_link_options(-Wl,-dead_strip)
@@ -38,77 +74,137 @@ else()
   add_link_options(-Wl,--gc-sections)
 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)
+if(EMSCRIPTEN)
+  # https://github.com/emscripten-core/emscripten/issues/15377#issuecomment-1285167486
+  add_link_options(-lnodefs.js -lnoderawfs.js)
+  add_link_options(-s ALLOW_MEMORY_GROWTH)
 endif()
-cmake_pop_check_state()
 
-check_include_file_cxx("arm_neon.h" HAS_ARM_NEON)
-if(HAS_ARM_NEON)
-  add_compile_definitions(HAS_ARM_NEON)
+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)
+  if(HAS_AVX)
+    add_compile_options(-mavx)
+    add_compile_definitions(HAS_AVX)
+  endif()
+  cmake_pop_check_state()
+
+  check_include_file_cxx("arm_neon.h" HAS_ARM_NEON)
+  if(HAS_ARM_NEON)
+    add_compile_definitions(HAS_ARM_NEON)
+  endif()
 endif()
 
 set(CMAKE_CXX_IMPLICIT_LINK_LIBRARIES "")
 
-add_library(${PROJECT_NAME}_object OBJECT ConflictSet.cpp)
-target_compile_options(${PROJECT_NAME}_object PRIVATE -fPIC -fno-exceptions
+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 ${CMAKE_SOURCE_DIR}/include)
+target_include_directories(${PROJECT_NAME}-object
+                           PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
 
-add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
+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(NOT APPLE)
-  target_link_options(${PROJECT_NAME} PRIVATE
-                      LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+if(HAS_VERSION_SCRIPT)
+  target_link_options(
+    ${PROJECT_NAME} PRIVATE
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
-add_library(${PROJECT_NAME}_static STATIC
-            $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
+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)
+if(APPLE)
   add_custom_command(
-    TARGET conflict_set_static
+    TARGET ${PROJECT_NAME}-static
+    PRE_LINK
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
+            $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
+else()
+  add_custom_command(
+    TARGET ${PROJECT_NAME}-static
     POST_BUILD
     COMMAND
-      ${CMAKE_OBJCOPY} --keep-global-symbols=${CMAKE_SOURCE_DIR}/symbols.txt
-      $<TARGET_FILE:${PROJECT_NAME}_static>)
+      ${CMAKE_OBJCOPY}
+      --keep-global-symbols=${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 -Wpedantic -Wunreachable-code -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)
 
-  # Shared library version of FoundationDB's skip list implementation
-  add_library(skip_list SHARED SkipList.cpp)
-  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})
+  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_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 ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    set_target_properties(
+      skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                           "${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 ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    set_target_properties(
+      hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                            "${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)
+
+    foreach(TEST ${CORPUS_TESTS})
+      get_filename_component(hash ${TEST} NAME)
+      add_test(NAME skip_list_${hash} COMMAND driver_skip_list ${TEST})
+    endforeach()
+  endif()
+
+  # ad hoc testing
   add_executable(conflict_set_main ConflictSet.cpp)
   target_include_directories(conflict_set_main
                              PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
@@ -135,12 +231,13 @@ if(BUILD_TESTING)
     endif()
   endif()
 
-  # corpus tests
-
-  file(GLOB CORPUS_TESTS ${CMAKE_SOURCE_DIR}/corpus/*)
-
+  # whitebox tests
   add_executable(fuzz_driver ConflictSet.cpp FuzzTestDriver.cpp)
   target_compile_options(fuzz_driver PRIVATE ${TEST_FLAGS})
+  if(NOT CMAKE_CROSSCOMPILING)
+    target_compile_options(fuzz_driver PRIVATE -fsanitize=address,undefined)
+    target_link_options(fuzz_driver PRIVATE -fsanitize=address,undefined)
+  endif()
   target_compile_definitions(fuzz_driver PRIVATE ENABLE_FUZZ)
   target_include_directories(fuzz_driver
                              PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
@@ -149,27 +246,59 @@ if(BUILD_TESTING)
     add_test(NAME conflict_set_fuzz_${hash} COMMAND fuzz_driver ${TEST})
   endforeach()
 
+  # tsan tests
+  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)
+    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()
+
+  # blackbox tests
   add_executable(driver TestDriver.cpp)
   target_compile_options(driver PRIVATE ${TEST_FLAGS})
   target_link_libraries(driver PRIVATE ${PROJECT_NAME})
+  foreach(TEST ${CORPUS_TESTS})
+    get_filename_component(hash ${TEST} NAME)
+    add_test(NAME conflict_set_blackbox_${hash} COMMAND driver ${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)
+  # scripted tests. Written manually to fill in anything libfuzzer couldn't
+  # find.
+  if(NOT CMAKE_CROSSCOMPILING)
+    find_package(Python3 REQUIRED COMPONENTS Interpreter)
+    set_property(
+      DIRECTORY
+      APPEND
+      PROPERTY CMAKE_CONFIGURE_DEPENDS
+               ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py)
+    execute_process(
+      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_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
+          --build-dir ${CMAKE_CURRENT_BINARY_DIR})
+    endforeach()
+  endif()
 
   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})
   endif()
 
-  foreach(TEST ${CORPUS_TESTS})
-    get_filename_component(hash ${TEST} NAME)
-    add_test(NAME conflict_set_blackbox_${hash} COMMAND driver ${TEST})
-    add_test(NAME skip_list_${hash} COMMAND driver_skip_list ${TEST})
-  endforeach()
-
   # api smoke tests
 
   # c90
@@ -190,36 +319,85 @@ if(BUILD_TESTING)
                         PROPERTIES CXX_STANDARD_REQUIRED ON)
   add_test(NAME conflict_set_cxx_api_test COMMAND conflict_set_cxx_api_test)
 
-  if(NOT APPLE AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+  # symbol visibility tests
+  if(NOT WASM AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+    if(APPLE)
+      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_CURRENT_SOURCE_DIR}/symbol-exports.txt)
+      if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
+        set(symbol_imports
+            ${CMAKE_CURRENT_SOURCE_DIR}/aarch64-symbol-imports.txt)
+      else()
+        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}> ${CMAKE_SOURCE_DIR}/symbols.txt)
+      COMMAND
+        ${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
-        $<TARGET_FILE:${PROJECT_NAME}_static> ${CMAKE_SOURCE_DIR}/symbols.txt)
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
+        $<TARGET_FILE:${PROJECT_NAME}-static> ${symbol_exports}
+        ${symbol_imports})
   endif()
 
   # bench
-
   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})
   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})
 endif()
+
 # packaging
 
 set(CPACK_PACKAGE_CONTACT andrew@weaselab.dev)
-set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME all)
+
+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)
+# see *-imports.txt - dependency versions need to be synced with symbol versions
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
+  set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6 (>= 2.17)")
+else()
+  set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6 (>= 2.14)")
+endif()
+
+# macos
+set(CMAKE_OSX_DEPLOYMENT_TARGET 11.0)
+if(APPLE)
+  find_program(PANDOC_EXE pandoc)
+  if(PANDOC_EXE)
+    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_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)
 
 include(GNUInstallDirs)
@@ -230,7 +408,7 @@ 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 +417,16 @@ set_target_properties(
                              SOVERSION ${PROJECT_VERSION_MAJOR})
 
 install(
-  TARGETS ${PROJECT_NAME} ${PROJECT_NAME}_static
-  EXPORT ConflictSetConfig
+  TARGETS ${PROJECT_NAME} ${PROJECT_NAME}-static
+  EXPORT ${PROJECT_NAME}Config
   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 ${PROJECT_NAME}Config
+        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/${PROJECT_NAME}/cmake)
+
+cpack_add_component(all)

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 +1,161 @@
+#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; }
+
+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}) {}
+
+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 +39,74 @@ 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;
+#endif
+#ifndef NDEBUG
+  (char *&)p -= kMallocHeaderSize;
+  size_t expected;
+  memcpy(&expected, p, sizeof(expected));
+  assert(s == expected);
+#endif
+  free(p);
 }
 
 // ==================== BEGIN ARENA IMPL ====================
@@ -90,6 +152,7 @@ inline void *operator new[](size_t size, std::align_val_t align,
 
 /// align must be a power of two
 template <class T> T *align_up(T *t, size_t align) {
+  assert(std::popcount(align) == 1);
   auto unaligned = uintptr_t(t);
   auto aligned = (unaligned + align - 1) & ~(align - 1);
   return reinterpret_cast<T *>(reinterpret_cast<char *>(t) + aligned -
@@ -98,6 +161,7 @@ template <class T> T *align_up(T *t, size_t align) {
 
 /// align must be a power of two
 constexpr inline int align_up(uint32_t unaligned, uint32_t align) {
+  assert(std::popcount(align) == 1);
   return (unaligned + align - 1) & ~(align - 1);
 }
 
@@ -113,12 +177,10 @@ struct Arena::ArenaImpl {
   uint8_t *begin() { return reinterpret_cast<uint8_t *>(this + 1); }
 };
 
-static_assert(sizeof(Arena::ArenaImpl) == 16);
-static_assert(alignof(Arena::ArenaImpl) == 8);
-
 inline Arena::Arena(int initialSize) : impl(nullptr) {
   if (initialSize > 0) {
-    auto allocationSize = align_up(initialSize + sizeof(ArenaImpl), 16);
+    auto allocationSize =
+        align_up(initialSize + sizeof(ArenaImpl), alignof(ArenaImpl));
     impl = (Arena::ArenaImpl *)safe_malloc(allocationSize);
     impl->prev = nullptr;
     impl->capacity = allocationSize - sizeof(ArenaImpl);
@@ -129,7 +191,7 @@ 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;
   }
 }
@@ -154,7 +216,7 @@ inline void *operator new(size_t size, std::align_val_t align, Arena &arena) {
                           (arena.impl ? std::max<int>(sizeof(Arena::ArenaImpl),
                                                       arena.impl->capacity * 2)
                                       : 0)),
-        16);
+        alignof(Arena::ArenaImpl));
     auto *impl = (Arena::ArenaImpl *)safe_malloc(allocationSize);
     impl->prev = arena.impl;
     impl->capacity = allocationSize - sizeof(Arena::ArenaImpl);
@@ -193,10 +255,65 @@ 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();
+  }
+
+  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));
@@ -349,42 +466,16 @@ 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) {
+  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;
       }
@@ -404,7 +495,10 @@ struct ReferenceImpl {
                                                      : ConflictSet::Commit;
     }
   }
-  void addWrites(const ConflictSet::WriteRange *writes, int count) {
+  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);
@@ -412,7 +506,6 @@ 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);
@@ -425,11 +518,12 @@ 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;
 };
 
@@ -470,37 +564,42 @@ 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 {
+template <class ConflictSetImpl, bool kEnableAssertions = true>
+struct TestDriver {
   Arbitrary arbitrary;
   explicit TestDriver(const uint8_t *data, size_t size)
       : arbitrary({data, size}) {}
 
-  int64_t writeVersion = 0;
-  int64_t oldestVersion = 0;
+  int64_t oldestVersion = arbitrary.next();
+  int64_t writeVersion = oldestVersion;
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
 
-  constexpr static auto kMaxKeyLen = 32;
+  constexpr static auto kMaxKeySuffixLen = 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();
-  }
+  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() {
+    assert(cs.getBytes() >= 0);
     if (!arbitrary.hasEntropy()) {
       return true;
     }
@@ -508,7 +607,8 @@ template <class ConflictSetImpl> struct TestDriver {
     {
       int numPointWrites = arbitrary.bounded(100);
       int numRangeWrites = arbitrary.bounded(100);
-      int64_t v = ++writeVersion;
+      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);
@@ -516,9 +616,10 @@ template <class ConflictSetImpl> struct TestDriver {
         if (!arbitrary.hasEntropy()) {
           return true;
         }
-        int keyLen = arbitrary.bounded(kMaxKeyLen);
+        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
-        arbitrary.randomBytes(begin, keyLen);
+        memset(begin, prefixByte, prefixLen);
+        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -547,38 +648,104 @@ 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}\n", printable(writes[i].begin).c_str());
         } else {
-          fprintf(stderr, "Write: [%s, %s) -> %d\n",
+          fprintf(stderr, "Write: [%s, %s)\n",
                   printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str(),
-                  int(writes[i].writeVersion));
+                  printable(writes[i].end).c_str());
         }
 #endif
       }
       assert(iter == keys.end());
       assert(i == numPointWrites + numRangeWrites);
 
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+      fprintf(stderr, "Write @ %" PRId64 "\n", v);
+#endif
+
+      // 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
+
       CALLGRIND_START_INSTRUMENTATION;
-      cs.addWrites(writes, numPointWrites + numRangeWrites);
+      cs.addWrites(writes, numPointWrites + numRangeWrites, v);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.addWrites(writes, numPointWrites + numRangeWrites);
+      if constexpr (kEnableAssertions) {
+        refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
+      }
 
-      oldestVersion = std::max<int64_t>(writeVersion - arbitrary.bounded(10),
-                                        oldestVersion);
       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);
+
+      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);
@@ -586,9 +753,10 @@ template <class ConflictSetImpl> struct TestDriver {
         if (!arbitrary.hasEntropy()) {
           return true;
         }
-        int keyLen = arbitrary.bounded(kMaxKeyLen);
+        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
-        arbitrary.randomBytes(begin, keyLen);
+        memset(begin, prefixByte, prefixLen);
+        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -619,12 +787,12 @@ 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, "Read: {%s} @ %" PRId64 "\n",
+                  printable(reads[i].begin).c_str(), reads[i].readVersion);
         } else {
-          fprintf(stderr, "Read: [%s, %s) @ %d\n",
+          fprintf(stderr, "Read: [%s, %s) @ %" PRId64 "\n",
                   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
       }
@@ -635,32 +803,82 @@ 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();
+        // 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
+
       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);
+      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();
+        }
+
+        refImpl.check(reads, results2, numPointReads + numRangeReads);
+      }
+
+      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;
           }
+        }
+        return true;
+      };
+
+      if constexpr (kEnableAssertions) {
+        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 +36,40 @@ 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('Debug') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DCMAKE_BUILD_TYPE=Debug")
+            }
+        }
+        stage('SIMD fallback') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
+            }
+        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {
@@ -46,18 +78,35 @@ 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') {
@@ -68,11 +117,15 @@ 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 -DDISABLE_TSAN=ON")
                 sh '''
-                gcovr --gcov-executable "llvm-cov-15 gcov" --exclude '.*third_party.*' --cobertura > build/coverage.xml
+                gcovr -f ConflictSet.cpp --cobertura > build/coverage.xml
+                '''
+                recordCoverage qualityGates: [[criticality: 'NOTE', metric: 'MODULE']], tools: [[parser: 'COBERTURA', pattern: 'build/coverage.xml']]
+                sh '''
+                # Suppress again, because we haven't dealt with function multi-versioning for x86 yet
+                # gcovr -f ConflictSet.cpp --fail-under-line 100 > /dev/null
                 '''
-                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 +2,106 @@ 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)`
+|              245.99 |        4,065,232.81 |    0.3% |      0.01 | `point reads`
+|              265.93 |        3,760,430.49 |    0.2% |      0.01 | `prefix reads`
+|              485.30 |        2,060,569.50 |    0.2% |      0.01 | `range reads`
+|              449.60 |        2,224,195.17 |    0.4% |      0.01 | `point writes`
+|              441.76 |        2,263,688.18 |    1.1% |      0.01 | `prefix writes`
+|              245.42 |        4,074,647.54 |    2.4% |      0.02 | `range writes`
+|              572.80 |        1,745,810.06 |    1.3% |      0.01 | `monotonic increasing point writes`
+|          154,819.33 |            6,459.14 |    0.9% |      0.01 | `worst case for radix tree`
+
+## Radix tree (this implementation)
+
+|               ns/op |                op/s |    err% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------:|:----------
+|               19.17 |       52,163,930.66 |    0.1% |      0.01 | `point reads`
+|               23.68 |       42,224,388.21 |    0.7% |      0.01 | `prefix reads`
+|               63.30 |       15,797,506.06 |    0.9% |      0.01 | `range reads`
+|               29.66 |       33,720,994.74 |    0.3% |      0.01 | `point writes`
+|               43.50 |       22,987,781.25 |    1.0% |      0.01 | `prefix writes`
+|               50.00 |       20,000,000.00 |    0.8% |      0.01 | `range writes`
+|              103.25 |        9,684,786.47 |    2.9% |      0.01 | `monotonic increasing point writes`
+|        1,181,500.00 |              846.38 |    2.3% |      0.01 | `worst case for radix tree`
+
+# "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

@@ -1,14 +1,18 @@
 #include <ConflictSet.h>
 
+#include <cerrno>
 #include <chrono>
+#include <cstdio>
 #include <cstring>
 #include <fcntl.h>
-#include <fstream>
+#include <string_view>
 #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 +33,7 @@ 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 +45,8 @@ 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 +61,8 @@ 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 +103,6 @@ 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 +110,38 @@ 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",
+  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,
-         addBytes / addTime * 1e-6, gcTime);
-}
+         addBytes / addTime * 1e-6, gcTime / (gcTime + addTime) * 1e2,
+         double(peakMemory));
+}

SkipList.cpp

@@ -16,10 +16,14 @@
  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  * See the License for the specific language governing permissions and
  * limitations under the License.
+ *
+ * This source code is modified to compile outside of FoundationDB
  */
 
 #include "ConflictSet.h"
 #include "Internal.h"
+
+#include <algorithm>
 #include <span>
 
 std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
@@ -38,7 +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 {
@@ -50,6 +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() {
@@ -149,7 +282,7 @@ private:
       setMaxVersion(level, v);
     }
 
-    void destroy() { free(this); }
+    void destroy() { safe_free(this, getNodeSize()); }
 
   private:
     int getNodeSize() const {
@@ -268,13 +401,21 @@ public:
   }
 
   explicit SkipList(Version version = 0) {
+#if DEBUG_VERBOSE
+    fprintf(stderr, "skip_list: create\n");
+#endif
     header = Node::create(StringRef(), MaxLevels - 1);
     for (int l = 0; l < MaxLevels; l++) {
       header->setNext(l, nullptr);
       header->setMaxVersion(l, version);
     }
   }
-  ~SkipList() { destroy(); }
+  ~SkipList() {
+#if DEBUG_VERBOSE
+    fprintf(stderr, "skip_list: destroy\n");
+#endif
+    destroy();
+  }
   SkipList(SkipList &&other) noexcept : header(other.header) {
     other.header = nullptr;
   }
@@ -286,7 +427,7 @@ 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 +436,7 @@ public:
         insert(endF, endF.finger[0]->getMaxVersion(0));
 
       remove(startF, endF);
-      insert(startF, version[r]);
+      insert(startF, version);
     }
   }
 
@@ -567,7 +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;
@@ -582,15 +724,49 @@ 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;
       }
     }
   }
 
-  void addWrites(const ConflictSet::WriteRange *writes, int count) {
+  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];
@@ -601,60 +777,130 @@ 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];
-        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]);
-        writeVersions[i] = w.writeVersion;
-        keyUpdates += 2;
+        const auto &w = combinedWriteConflictRanges[s * stripeSize / 2 + i];
+#if DEBUG_VERBOSE
+        printf("Write begin: %s\n", printable(w.begin).c_str());
+        fflush(stdout);
+#endif
+        values[i * 2] = w.first;
+        values[i * 2 + 1] = w.second;
+        keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
-      skipList.addConflictRanges(fingers, ss / 2, writeVersions);
+      skipList.addConflictRanges(fingers, ss / 2, writeVersion);
       ss = stripeSize;
     }
   }
 
   void setOldestVersion(int64_t oldestVersion) {
+    assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
     SkipList::Finger finger;
     int temp;
     std::span<const uint8_t> key = removalKey;
     skipList.find(&key, &finger, &temp, 1);
-    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 0));
+    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 10));
     removalArena = Arena();
     removalKey = copyToArena(
         removalArena, {finger.getValue().data(), finger.getValue().size()});
   }
 
+  int64_t totalBytes = 0;
+
 private:
-  int64_t keyUpdates = 0;
+  int64_t keyUpdates = 10;
   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 {
-  return impl->check(reads, results, count);
+// 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 internal_addWrites(ConflictSet::Impl *impl,
+                        const ConflictSet::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::addWrites(const WriteRange *writes, int count) {
-  return impl->addWrites(writes, count);
+void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
+  mallocBytesDelta = 0;
+  impl->setOldestVersion(oldestVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#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;
+}
+
+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) {
-  return impl->setOldestVersion(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(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
+    : impl(internal_create(oldestVersion)) {}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
-    impl->~Impl();
-    free(impl);
+    internal_destroy(impl);
   }
 }
 
@@ -675,25 +921,36 @@ 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) {
-  ((ConflictSet::Impl *)cs)->addWrites(writes, count);
+ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes, int count,
+                      int64_t writeVersion) {
+  internal_addWrites((ConflictSet::Impl *)cs, writes, count, writeVersion);
 }
 __attribute__((__visibility__("default"))) void
 ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
-  ((ConflictSet::Impl *)cs)->setOldestVersion(oldestVersion);
+  internal_setOldestVersion((ConflictSet::Impl *)cs, oldestVersion);
 }
 __attribute__((__visibility__("default"))) void *
 ConflictSet_create(int64_t oldestVersion) {
-  return new (safe_malloc(sizeof(ConflictSet::Impl)))
-      ConflictSet::Impl{oldestVersion};
+  return internal_create(oldestVersion);
 }
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
-  using Impl = ConflictSet::Impl;
-  ((Impl *)cs)->~Impl();
-  free(cs);
+  internal_destroy((ConflictSet::Impl *)cs);
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  return internal_getBytes((ConflictSet::Impl *)cs);
 }
 }
+
+#if SHOW_MEMORY
+struct __attribute__((visibility("default"))) PeakPrinter {
+  ~PeakPrinter() {
+    printf("--- skip_list ---\n");
+    printf("malloc bytes: %g\n", double(mallocBytes));
+    printf("Peak malloc bytes: %g\n", double(peakMallocBytes));
+  }
+} peakPrinter;
+#endif

TestDriver.cpp

@@ -3,18 +3,35 @@
 #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) {
-    printf("Running: %s\n", argv[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()};
+    TestDriver<ConflictSet, !PERF_TEST> driver{(const uint8_t *)str.data(),
+                                               str.size()};
     while (!driver.next())
       ;
     if (!driver.ok) {
       abort();
     }
+
+    ConflictSet::MetricsV1 *metrics;
+    int metricsCount;
+    driver.cs.getMetricsV1(&metrics, &metricsCount);
+    printf("#################### METRICS 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

@@ -0,0 +1,10 @@
+__aarch64_cas8_relax
+__getauxval@GLIBC_2.17
+__stack_chk_fail@GLIBC_2.17
+__stack_chk_guard@GLIBC_2.17
+abort@GLIBC_2.17
+free@GLIBC_2.17
+malloc@GLIBC_2.17
+memcpy@GLIBC_2.17
+memmove@GLIBC_2.17
+memset@GLIBC_2.17

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)

apple-symbol-exports.txt

@@ -0,0 +1,19 @@
+_ConflictSet_addWrites
+_ConflictSet_check
+_ConflictSet_create
+_ConflictSet_destroy
+_ConflictSet_getBytes
+_ConflictSet_setOldestVersion
+__ZN8weaselab11ConflictSet16setOldestVersionEx
+__ZN8weaselab11ConflictSet9addWritesEPKNS0_10WriteRangeEix
+__ZN8weaselab11ConflictSetC1EOS0_
+__ZN8weaselab11ConflictSetC1Ex
+__ZN8weaselab11ConflictSetC2EOS0_
+__ZN8weaselab11ConflictSetC2Ex
+__ZN8weaselab11ConflictSetD1Ev
+__ZN8weaselab11ConflictSetD2Ev
+__ZN8weaselab11ConflictSetaSEOS0_
+__ZNK8weaselab11ConflictSet12getMetricsV1EPPNS0_9MetricsV1EPi
+__ZNK8weaselab11ConflictSet5checkEPKNS0_9ReadRangeEPNS0_6ResultEi
+__ZNK8weaselab11ConflictSet8getBytesEv
+__ZNK8weaselab11ConflictSet9MetricsV18getValueEv

apple-symbol-imports.txt

@@ -0,0 +1,10 @@
+___stack_chk_fail
+___stack_chk_guard
+__tlv_bootstrap
+_abort
+_bzero
+_free
+_malloc
+_memcpy
+_memmove
+dyld_stub_binder

conflict_set.py

@@ -0,0 +1,138 @@
+import ctypes
+import enum
+import os
+
+from typing import Optional
+
+
+class _Key(ctypes.Structure):
+    _fields_ = [("p", ctypes.POINTER(ctypes.c_ubyte)), ("len", ctypes.c_int)]
+
+
+class ReadRange(ctypes.Structure):
+    _fields_ = [
+        ("begin", _Key),
+        ("end", _Key),
+        ("readVersion", ctypes.c_int64),
+    ]
+
+
+class WriteRange(ctypes.Structure):
+    _fields_ = [("begin", _Key), ("end", _Key)]
+
+
+class Result(enum.Enum):
+    COMMIT = 0
+    CONFLICT = 1
+    TOO_OLD = 2
+
+
+def write(begin: bytes, end: Optional[bytes] = None) -> WriteRange:
+    b = (ctypes.c_ubyte * len(begin)).from_buffer(bytearray(begin))
+
+    if end is None:
+        e = (ctypes.c_ubyte * 0)()
+    else:
+        e = (ctypes.c_ubyte * len(end)).from_buffer(bytearray(end))
+    return WriteRange(_Key(b, len(b)), _Key(e, len(e)))
+
+
+def read(version: int, begin: bytes, end: Optional[bytes] = None) -> ReadRange:
+    b = (ctypes.c_ubyte * len(begin)).from_buffer(bytearray(begin))
+    if end is None:
+        e = (ctypes.c_ubyte * 0)()
+    else:
+        e = (ctypes.c_ubyte * len(end)).from_buffer(bytearray(end))
+    return ReadRange(_Key(b, len(b)), _Key(e, len(e)), version)
+
+
+class ConflictSet:
+    def __init__(
+        self,
+        version: int = 0,
+        build_dir: Optional[str] = None,
+        implementation: Optional[str] = None,
+    ) -> None:
+        self._lib = None
+        if build_dir is None:
+            build_dir = os.path.dirname(__file__) + "/build"
+        if implementation is None:
+            implementation = "radix_tree"
+        for f in (
+            build_dir + "/" + implementation + "/libconflict-set.so.0",
+            os.path.dirname(__file__)
+            + "/build/"
+            + implementation
+            + "/libconflict-set.0.dylib",
+        ):
+            try:
+                self._lib = ctypes.cdll.LoadLibrary(f)
+            except:
+                pass
+
+        if self._lib is None:
+            import sys
+
+            print(
+                "Could not find libconflict-set implementation " + implementation,
+                file=sys.stderr,
+            )
+            sys.exit(1)
+
+        self._lib.ConflictSet_create.argtypes = (ctypes.c_int64,)
+        self._lib.ConflictSet_create.restype = ctypes.c_void_p
+
+        self._lib.ConflictSet_check.argtypes = (
+            ctypes.c_void_p,
+            ctypes.POINTER(ReadRange),
+            ctypes.POINTER(ctypes.c_int),
+            ctypes.c_int,
+        )
+
+        self._lib.ConflictSet_addWrites.argtypes = (
+            ctypes.c_void_p,
+            ctypes.POINTER(WriteRange),
+            ctypes.c_int,
+            ctypes.c_int64,
+        )
+
+        self._lib.ConflictSet_setOldestVersion.argtypes = (
+            ctypes.c_void_p,
+            ctypes.c_int64,
+        )
+
+        self._lib.ConflictSet_destroy.argtypes = (ctypes.c_void_p,)
+
+        self._lib.ConflictSet_getBytes.argtypes = (ctypes.c_void_p,)
+        self._lib.ConflictSet_getBytes.restype = ctypes.c_int64
+
+        self.p = self._lib.ConflictSet_create(version)
+
+    def addWrites(self, version: int, *writes: WriteRange):
+        self._lib.ConflictSet_addWrites(
+            self.p, (WriteRange * len(writes))(*writes), len(writes), version
+        )
+
+    def check(self, *reads: ReadRange) -> list[Result]:
+        r = (ctypes.c_int * len(reads))()
+        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:
+        self._lib.ConflictSet_setOldestVersion(self.p, version)
+
+    def getBytes(self) -> int:
+        return self._lib.ConflictSet_getBytes(self.p)
+
+    def __enter__(self):
+        return self
+
+    def close(self) -> None:
+        if self.p is not None:
+            self._lib.ConflictSet_destroy(self.p)
+            self.p = None
+
+    def __exit__(self, exception_type, exception_value, exception_traceback):
+        self.close()

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

@@ -1,6 +1,9 @@
 #include "ConflictSet.h"
 
 #include <cassert>
+#include <cstdio>
+
+using namespace weaselab;
 
 int main(void) {
   ConflictSet cs(0);
@@ -8,8 +11,7 @@ int main(void) {
   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 +20,16 @@ int main(void) {
   r.readVersion = 0;
   cs.check(&r, &result, 1);
   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());
+  }
 }