Fix build for llvm 18

Instead of strictly increasing.

.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++]
+  - 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

Bench.cpp

@@ -34,7 +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) {
@@ -52,7 +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() {

CMakeLists.txt

@@ -1,15 +1,28 @@
 cmake_minimum_required(VERSION 3.18)
 project(
-  conflict_set
-  VERSION 0.0.1
+  conflict-set
+  VERSION 0.0.5
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"
   LANGUAGES C CXX)
 set(CMAKE_CXX_STANDARD 20)
 
+file(WRITE ${CMAKE_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+
+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
@@ -22,16 +35,33 @@ 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_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)
 
 # 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)
 
-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)
@@ -39,8 +69,21 @@ else()
   add_link_options(-Wl,--gc-sections)
 endif()
 
-include(CheckIncludeFileCXX)
-include(CMakePushCheckState)
+if(EMSCRIPTEN)
+  # https://github.com/emscripten-core/emscripten/issues/15377#issuecomment-1285167486
+  add_link_options(-lnodefs.js -lnoderawfs.js)
+  add_link_options(-s ALLOW_MEMORY_GROWTH)
+endif()
+
+if(NOT USE_SIMD_FALLBACK)
+  cmake_push_check_state()
+  list(APPEND CMAKE_REQUIRED_FLAGS -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)
@@ -55,16 +98,17 @@ 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(${PROJECT_NAME}_object
+target_include_directories(${PROJECT_NAME}-object
                            PRIVATE ${CMAKE_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")
@@ -72,56 +116,85 @@ if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
-if(NOT APPLE)
+if(HAS_VERSION_SCRIPT)
   target_link_options(${PROJECT_NAME} PRIVATE
                       LINKER:--version-script=${CMAKE_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_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_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)
+set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
 include(CTest)
 
 if(BUILD_TESTING)
 
+  # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
+  # coverage with a small number of tests.
+
+  file(GLOB CORPUS_TESTS ${CMAKE_SOURCE_DIR}/corpus/*)
+
+  # 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 -fPIC -fno-exceptions
+    target_compile_options(skip_list PRIVATE -fno-exceptions
                                              -fvisibility=hidden)
     target_include_directories(skip_list PUBLIC ${CMAKE_SOURCE_DIR}/include)
     set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                                                "${CMAKE_BINARY_DIR}/skip_list")
     set_target_properties(skip_list PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
-  set_target_properties(skip_list PROPERTIES VERSION ${PROJECT_VERSION}
-                                             SOVERSION ${PROJECT_VERSION_MAJOR})
+    set_target_properties(
+      skip_list PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
+                                                      ${PROJECT_VERSION_MAJOR})
 
     # Shared library version of a std::unordered_map-based conflict set (point
     # queries only)
     add_library(hash_table SHARED HashTable.cpp)
-  target_compile_options(hash_table PRIVATE -fPIC -fno-exceptions
+    target_compile_options(hash_table PRIVATE -fno-exceptions
                                               -fvisibility=hidden)
     target_include_directories(hash_table PUBLIC ${CMAKE_SOURCE_DIR}/include)
-  set_target_properties(hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+    set_target_properties(
+      hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                             "${CMAKE_BINARY_DIR}/hash_table")
     set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
     set_target_properties(
       hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
                                                        ${PROJECT_VERSION_MAJOR})
 
+    add_executable(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)
@@ -148,12 +221,13 @@ 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 ${CMAKE_CURRENT_SOURCE_DIR}/include)
@@ -162,13 +236,47 @@ if(BUILD_TESTING)
     add_test(NAME conflict_set_fuzz_${hash} COMMAND fuzz_driver ${TEST})
   endforeach()
 
+  # tsan tests
+  if(NOT CMAKE_CROSSCOMPILING AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+    add_executable(tsan_driver ConflictSet.cpp FuzzTestDriver.cpp)
+    target_compile_options(tsan_driver PRIVATE ${TEST_FLAGS} -fsanitize=thread)
+    target_link_options(tsan_driver PRIVATE -fsanitize=thread)
+    target_compile_definitions(tsan_driver PRIVATE ENABLE_FUZZ THREAD_TEST)
+    target_include_directories(tsan_driver
+                               PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    foreach(TEST ${CORPUS_TESTS})
+      get_filename_component(hash ${TEST} NAME)
+      add_test(NAME conflict_set_tsan_${hash} COMMAND tsan_driver ${TEST})
+    endforeach()
+  endif()
+
+  # 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_SOURCE_DIR}/test_conflict_set.py)
+    execute_process(
+      COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
+              list OUTPUT_VARIABLE SCRIPT_TESTS)
+    foreach(TEST ${SCRIPT_TESTS})
+      add_test(
+        NAME script_test_${TEST}
+        COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
+                test ${TEST} --build-dir ${CMAKE_BINARY_DIR})
+    endforeach()
+  endif()
 
   find_program(VALGRIND_EXE valgrind)
   if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
@@ -177,12 +285,6 @@ if(BUILD_TESTING)
                      $<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
@@ -203,31 +305,45 @@ 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_SOURCE_DIR}/apple-symbol-exports.txt)
+      set(symbol_imports ${CMAKE_SOURCE_DIR}/apple-symbol-imports.txt)
+    else()
+      set(symbol_exports ${CMAKE_SOURCE_DIR}/symbol-exports.txt)
+      if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
+        set(symbol_imports ${CMAKE_SOURCE_DIR}/aarch64-symbol-imports.txt)
+      else()
+        set(symbol_imports ${CMAKE_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_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)
+        $<TARGET_FILE:${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)
-
   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)
 endif()
+
 # packaging
 
 set(CPACK_PACKAGE_CONTACT andrew@weaselab.dev)
+set(CMAKE_INSTALL_DEFAULT_COMPONENT_NAME all)
 
 set(CPACK_PACKAGE_VENDOR "Weaselab")
 set(CPACK_RESOURCE_FILE_LICENSE "${CMAKE_CURRENT_SOURCE_DIR}/LICENSE")
@@ -237,6 +353,30 @@ set(CPACK_RESOURCE_FILE_README "${CMAKE_CURRENT_SOURCE_DIR}/README.md")
 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_SOURCE_DIR}/README.md -o
+                            ${CMAKE_BINARY_DIR}/README.txt)
+    set(CPACK_RESOURCE_FILE_README ${CMAKE_BINARY_DIR}/README.txt)
+  endif()
+  configure_file(${CMAKE_SOURCE_DIR}/LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt
+                 COPYONLY)
+  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt)
+endif()
 
 include(CPack)
 
@@ -248,7 +388,7 @@ 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}>)
 
@@ -257,13 +397,16 @@ 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)

FuzzTestDriver.cpp

@@ -2,10 +2,12 @@
 #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) {
+  auto doTest = [&]() {
     for (int i = 1; i < argc; ++i) {
       std::ifstream t(argv[i], std::ios::binary);
       std::stringstream buffer;
@@ -13,4 +15,12 @@ int main(int argc, char **argv) {
       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

@@ -96,13 +96,15 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
   return impl->setOldestVersion(oldestVersion);
 }
 
+int64_t ConflictSet::getBytes() const { return -1; }
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
     : impl(new(safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
     impl->~Impl();
-    free(impl);
+    safe_free(impl, sizeof(Impl));
   }
 }
 
@@ -142,6 +144,11 @@ ConflictSet_create(int64_t oldestVersion) {
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
   using Impl = ConflictSet::Impl;
   ((Impl *)cs)->~Impl();
-  free(cs);
+  safe_free(cs, sizeof(Impl));
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  using Impl = ConflictSet::Impl;
+  return -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)) {
+  mallocBytesDelta += s;
+#if SHOW_MEMORY
+  mallocBytes += s;
+  if (mallocBytes > peakMallocBytes) {
+    peakMallocBytes = mallocBytes;
+  }
+#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;
 }
-  abort();
+
+// 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));
@@ -349,34 +466,6 @@ inline uint32_t Arbitrary::bounded(uint32_t s) {
 
 // ==================== END ARBITRARY IMPL ====================
 
-// ==================== BEGIN UTILITIES IMPL ====================
-
-// Call Stepwise::step for each element of remaining until it returns true.
-// Applies a permutation to `remaining` as a side effect.
-template <class Stepwise> void runInterleaved(std::span<Stepwise> remaining) {
-  while (remaining.size() > 0) {
-    for (int i = 0; i < int(remaining.size());) {
-      bool done = remaining[i].step();
-      if (done) {
-        if (i != int(remaining.size()) - 1) {
-          using std::swap;
-          swap(remaining[i], remaining.back());
-        }
-        remaining = remaining.subspan(0, remaining.size() - 1);
-      } else {
-        ++i;
-      }
-    }
-  }
-};
-
-template <class Stepwise> void runSequential(std::span<Stepwise> remaining) {
-  for (auto &r : remaining) {
-    while (!r.step()) {
-    }
-  }
-}
-
 struct ReferenceImpl {
   explicit ReferenceImpl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
     writeVersionMap[""] = oldestVersion;
@@ -470,23 +559,7 @@ inline std::string printable(std::span<const uint8_t> key) {
   return printable(std::string_view((const char *)key.data(), key.size()));
 }
 
-namespace {
-
-template <class ConflictSetImpl> struct TestDriver {
-  Arbitrary arbitrary;
-  explicit TestDriver(const uint8_t *data, size_t size)
-      : arbitrary({data, size}) {}
-
-  int64_t writeVersion = 0;
-  int64_t oldestVersion = 0;
-  ConflictSetImpl cs{oldestVersion};
-  ReferenceImpl refImpl{oldestVersion};
-
-  constexpr static auto kMaxKeyLen = 64;
-
-  bool ok = true;
-
-  static const char *resultToStr(ConflictSet::Result r) {
+inline const char *resultToStr(ConflictSet::Result r) {
   switch (r) {
   case ConflictSet::Commit:
     return "commit";
@@ -498,6 +571,25 @@ template <class ConflictSetImpl> struct TestDriver {
   abort();
 }
 
+namespace {
+
+template <class ConflictSetImpl> struct TestDriver {
+  Arbitrary arbitrary;
+  explicit TestDriver(const uint8_t *data, size_t size)
+      : arbitrary({data, size}) {}
+
+  int64_t writeVersion = 100;
+  int64_t oldestVersion = 0;
+  ConflictSetImpl cs{oldestVersion};
+  ReferenceImpl refImpl{oldestVersion};
+
+  constexpr static auto kMaxKeySuffixLen = 8;
+
+  bool ok = true;
+
+  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() {
@@ -508,7 +600,7 @@ 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));
       auto *writes =
           new (arena) ConflictSet::WriteRange[numPointWrites + numRangeWrites];
       auto keys = set<std::string_view>(arena);
@@ -516,9 +608,10 @@ 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));
       }
 
@@ -549,14 +642,12 @@ template <class ConflictSetImpl> struct TestDriver {
         }
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (writes[i].end.len == 0) {
-          fprintf(stderr, "Write: {%s} -> %d\n",
-                  printable(writes[i].begin).c_str(),
-                  int(writes[i].writeVersion));
+          fprintf(stderr, "Write: {%s} -> %" PRId64 "\n",
+                  printable(writes[i].begin).c_str(), writeVersion);
         } else {
-          fprintf(stderr, "Write: [%s, %s) -> %d\n",
+          fprintf(stderr, "Write: [%s, %s) -> %" PRId64 "\n",
                   printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str(),
-                  int(writes[i].writeVersion));
+                  printable(writes[i].end).c_str(), writeVersion);
         }
 #endif
       }
@@ -569,8 +660,8 @@ template <class ConflictSetImpl> struct TestDriver {
 
       refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
 
-      oldestVersion = std::max<int64_t>(writeVersion - arbitrary.bounded(10),
-                                        oldestVersion);
+      oldestVersion =
+          std::min(writeVersion - 10, oldestVersion + arbitrary.bounded(10));
       cs.setOldestVersion(oldestVersion);
       refImpl.setOldestVersion(oldestVersion);
     }
@@ -585,9 +676,10 @@ 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));
       }
 
@@ -634,12 +726,26 @@ template <class ConflictSetImpl> struct TestDriver {
       auto *results2 =
           new (arena) ConflictSet::Result[numPointReads + numRangeReads];
 
+#ifdef THREAD_TEST
+      auto *results3 =
+          new (arena) ConflictSet::Result[numPointReads + numRangeReads];
+      std::latch ready{1};
+      std::thread thread2{[&]() {
+        ready.count_down();
+        cs.check(reads, results3, numPointReads + numRangeReads);
+      }};
+      ready.wait();
+#endif
+
       CALLGRIND_START_INSTRUMENTATION;
       cs.check(reads, results1, numPointReads + numRangeReads);
       CALLGRIND_STOP_INSTRUMENTATION;
 
       refImpl.check(reads, results2, numPointReads + numRangeReads);
-      for (int i = 0; i < numPointReads + numRangeReads; ++i) {
+
+      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,
@@ -656,10 +762,24 @@ template <class ConflictSetImpl> struct TestDriver {
                   printable(reads[i].begin).c_str(),
                   printable(reads[i].end).c_str(), reads[i].readVersion);
             }
+            return false;
+          }
+        }
+        return true;
+      };
+
+      if (!compareResults(results1, results2, numPointReads + numRangeReads)) {
         ok = false;
         return true;
       }
+
+#ifdef THREAD_TEST
+      thread2.join();
+      if (!compareResults(results3, results2, numPointReads + numRangeReads)) {
+        ok = false;
+        return true;
       }
+#endif
     }
     return false;
   }

Jenkinsfile

@@ -36,6 +36,29 @@ pipeline {
                 sh 'pre-commit run --all-files --show-diff-on-failure'
             }
         }
+        stage('Clang') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("")
+                recordIssues(tools: [clang()])
+            }
+        }
+        stage('SIMD fallback') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
+            }
+        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {
@@ -44,7 +67,7 @@ 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
@@ -66,7 +89,7 @@ pipeline {
                 }
             }
             steps {
-                CleanBuildAndTest("-DCMAKE_TOOLCHAIN_FILE=../aarch64-toolchain.cmake")
+                CleanBuildAndTest("-DCMAKE_TOOLCHAIN_FILE=../aarch64-toolchain.cmake -DCMAKE_CXX_FLAGS=-DNVALGRIND")
                 sh '''
                 cd build
                 cpack -G DEB
@@ -85,9 +108,12 @@ pipeline {
             steps {
                 CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug")
                 sh '''
-                gcovr --exclude '.*third_party.*' --cobertura > build/coverage.xml
+                gcovr -f ConflictSet.cpp --cobertura > build/coverage.xml
                 '''
                 cobertura autoUpdateHealth: false, autoUpdateStability: false, coberturaReportFile: 'build/coverage.xml', conditionalCoverageTargets: '70, 0, 0', failUnhealthy: false, failUnstable: false, lineCoverageTargets: '80, 0, 0', maxNumberOfBuilds: 0, methodCoverageTargets: '80, 0, 0', onlyStable: false, sourceEncoding: 'ASCII', zoomCoverageChart: false
+                sh '''
+                gcovr -f ConflictSet.cpp --fail-under-line 100 > /dev/null
+                '''
             }
         }
     }

README.md

@@ -2,71 +2,104 @@ A data structure for optimistic concurrency control on ranges of bitwise-lexicog
 
 Intended to replace FoundationDB's skip list.
 
+Hardware for all benchmarks is a mac m1 2020.
+
 # FoundationDB's benchmark
 
 ## Skip list
 
 ```
-New conflict set: 4.189 sec
-                  0.298 Mtransactions/sec
-                  1.194 Mkeys/sec
-Detect only:      3.990 sec
-                  0.313 Mtransactions/sec
-                  1.253 Mkeys/sec
-Skiplist only:    2.849 sec
-                  0.439 Mtransactions/sec
-                  1.755 Mkeys/sec
+New conflict set: 1.957 sec
+                  0.639 Mtransactions/sec
+                  2.555 Mkeys/sec
+Detect only:      1.845 sec
+                  0.678 Mtransactions/sec
+                  2.710 Mkeys/sec
+Skiplist only:    1.263 sec
+                  0.990 Mtransactions/sec
+                  3.960 Mkeys/sec
 Performance counters:
-               Build: 0.0913
-                 Add: 0.0998
-              Detect: 3.99
-              D.Sort: 0.808
-           D.Combine: 0.0309
-         D.CheckRead: 1.67
-   D.CheckIntraBatch: 0.0305
-        D.MergeWrite: 1.18
-      D.RemoveBefore: 0.265
+               Build: 0.0546
+                 Add: 0.0563
+              Detect: 1.84
+              D.Sort: 0.412
+           D.Combine: 0.0141
+         D.CheckRead: 0.671
+   D.CheckIntraBatch: 0.0068
+        D.MergeWrite: 0.592
+      D.RemoveBefore: 0.146
 ```
 
 ## Radix tree (this implementation)
 
 ```
-New conflict set: 2.965 sec
-                  0.422 Mtransactions/sec
-                  1.686 Mkeys/sec
-Detect only:      2.761 sec
-                  0.453 Mtransactions/sec
-                  1.811 Mkeys/sec
-Skiplist only:    1.580 sec
-                  0.791 Mtransactions/sec
-                  3.165 Mkeys/sec
+New conflict set: 1.366 sec
+                  0.915 Mtransactions/sec
+                  3.660 Mkeys/sec
+Detect only:      1.248 sec
+                  1.002 Mtransactions/sec
+                  4.007 Mkeys/sec
+Skiplist only:    0.573 sec
+                  2.182 Mtransactions/sec
+                  8.730 Mkeys/sec
 Performance counters:
-               Build: 0.0902
-                 Add: 0.107
-              Detect: 2.76
-              D.Sort: 0.809
-           D.Combine: 0.0309
-         D.CheckRead: 0.658
-   D.CheckIntraBatch: 0.0294
-        D.MergeWrite: 0.921
-      D.RemoveBefore: 0.305
+               Build: 0.0594
+                 Add: 0.0572
+              Detect: 1.25
+              D.Sort: 0.418
+           D.Combine: 0.0149
+         D.CheckRead: 0.232
+   D.CheckIntraBatch: 0.0067
+        D.MergeWrite: 0.341
+      D.RemoveBefore: 0.232
 ```
 
 # Our benchmark
 
-|               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)`
+## Skip list
+
+|  ns/op |         op/s | err% | total | benchmark                           |
+| -----: | -----------: | ---: | ----: | :---------------------------------- |
+| 246.99 | 4,048,700.59 | 0.2% |  0.01 | `point reads`                       |
+| 260.16 | 3,843,784.65 | 0.1% |  0.01 | `prefix reads`                      |
+| 493.35 | 2,026,953.19 | 0.1% |  0.01 | `range reads`                       |
+| 462.05 | 2,164,289.23 | 0.6% |  0.01 | `point writes`                      |
+| 448.19 | 2,231,205.25 | 0.9% |  0.01 | `prefix writes`                     |
+| 255.83 | 3,908,845.72 | 1.5% |  0.02 | `range writes`                      |
+| 582.63 | 1,716,349.02 | 1.3% |  0.01 | `monotonic increasing point writes` |
+
+## Radix tree (this implementation)
+
+|  ns/op |          op/s | err% | total | benchmark                           |
+| -----: | ------------: | ---: | ----: | :---------------------------------- |
+|  19.42 | 51,483,206.67 | 0.3% |  0.01 | `point reads`                       |
+|  58.43 | 17,115,612.57 | 0.1% |  0.01 | `prefix reads`                      |
+| 216.09 |  4,627,766.60 | 0.2% |  0.01 | `range reads`                       |
+|  28.35 | 35,267,567.72 | 0.2% |  0.01 | `point writes`                      |
+|  43.43 | 23,026,226.17 | 0.2% |  0.01 | `prefix writes`                     |
+|  50.00 | 20,000,000.00 | 0.0% |  0.01 | `range writes`                      |
+|  92.38 | 10,824,863.69 | 4.1% |  0.01 | `monotonic increasing point writes` |
+
+# "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

@@ -1,13 +1,18 @@
 #include <ConflictSet.h>
 
+#include <cerrno>
 #include <chrono>
+#include <cstdio>
 #include <cstring>
 #include <fcntl.h>
+#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())
@@ -40,6 +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;
@@ -109,6 +116,10 @@ int main(int argc, const char **argv) {
         write = {};
         reads.clear();
 
+        if (cs.getBytes() > peakMemory) {
+          peakMemory = cs.getBytes();
+        }
+
         timer = now();
         cs.setOldestVersion(version - 10000);
         gcTime += now() - timer;
@@ -118,8 +129,9 @@ int main(int argc, const char **argv) {
     close(fd);
   }
 
-  printf(
-      "Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc ratio: %g%%\n",
+  printf("Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc ratio: "
+         "%g%%, Peak idle memory: %g\n",
          checkTime, checkBytes / checkTime * 1e-6, addTime,
-      addBytes / addTime * 1e-6, gcTime / (gcTime + addTime) * 1e2);
+         addBytes / addTime * 1e-6, gcTime / (gcTime + addTime) * 1e2,
+         double(peakMemory));
 }

SkipList.cpp

@@ -16,6 +16,8 @@
  * 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"
@@ -149,7 +151,7 @@ private:
       setMaxVersion(level, v);
     }
 
-    void destroy() { free(this); }
+    void destroy() { safe_free(this, getNodeSize()); }
 
   private:
     int getNodeSize() const {
@@ -268,13 +270,21 @@ 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;
   }
@@ -603,11 +613,15 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     for (int s = stripes - 1; s >= 0; s--) {
       for (int i = 0; i * 2 < ss; ++i) {
         const auto &w = writes[s * stripeSize / 2 + i];
+#if DEBUG_VERBOSE
+        printf("Write begin: %s\n", printable(w.begin).c_str());
+        fflush(stdout);
+#endif
         values[i * 2] = {w.begin.p, size_t(w.begin.len)};
         values[i * 2 + 1] = w.end.len > 0
                                 ? StringRef{w.end.p, size_t(w.end.len)}
                                 : keyAfter(arena, values[i * 2]);
-        keyUpdates += 2;
+        keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
       skipList.addConflictRanges(fingers, ss / 2, writeVersion);
@@ -621,41 +635,89 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     int temp;
     std::span<const uint8_t> key = removalKey;
     skipList.find(&key, &finger, &temp, 1);
-    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 0));
+    skipList.removeBefore(oldestVersion, finger, std::exchange(keyUpdates, 10));
     removalArena = Arena();
     removalKey = copyToArena(
         removalArena, {finger.getValue().data(), finger.getValue().size()});
   }
 
+  int64_t totalBytes = 0;
+
 private:
-  int64_t keyUpdates = 0;
+  int64_t keyUpdates = 10;
   Arena removalArena;
   std::span<const uint8_t> removalKey;
   int64_t oldestVersion;
   SkipList skipList;
 };
 
+// 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 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 ConflictSet::check(const ReadRange *reads, Result *results,
                         int count) const {
-  return impl->check(reads, results, count);
+  internal_check(impl, reads, results, count);
 }
 
 void ConflictSet::addWrites(const WriteRange *writes, int count,
                             int64_t writeVersion) {
-  return impl->addWrites(writes, count, 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); }
+
 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);
   }
 }
 
@@ -676,25 +738,36 @@ 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,
                       int64_t writeVersion) {
-  ((ConflictSet::Impl *)cs)->addWrites(writes, count, 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

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

aarch64-symbol-imports.txt

@@ -0,0 +1,8 @@
+__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

apple-symbol-exports.txt

@@ -0,0 +1,17 @@
+_ConflictSet_addWrites
+_ConflictSet_check
+_ConflictSet_create
+_ConflictSet_destroy
+_ConflictSet_getBytes
+_ConflictSet_setOldestVersion
+__ZN8weaselab11ConflictSet16setOldestVersionEx
+__ZN8weaselab11ConflictSet9addWritesEPKNS0_10WriteRangeEix
+__ZN8weaselab11ConflictSetC1EOS0_
+__ZN8weaselab11ConflictSetC1Ex
+__ZN8weaselab11ConflictSetC2EOS0_
+__ZN8weaselab11ConflictSetC2Ex
+__ZN8weaselab11ConflictSetD1Ev
+__ZN8weaselab11ConflictSetD2Ev
+__ZN8weaselab11ConflictSetaSEOS0_
+__ZNK8weaselab11ConflictSet5checkEPKNS0_9ReadRangeEPNS0_6ResultEi
+__ZNK8weaselab11ConflictSet8getBytesEv

apple-symbol-imports.txt

@@ -0,0 +1,8 @@
+__tlv_bootstrap
+_abort
+_bzero
+_free
+_malloc
+_memcpy
+_memmove
+dyld_stub_binder

conflict_set.py

@@ -0,0 +1,136 @@
+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, *reads, r, 1)
+        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

@@ -7,6 +7,7 @@ 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;
@@ -17,6 +18,8 @@ int main(void) {
   r.readVersion = 0;
   ConflictSet_check(cs, &r, &result, 1);
   assert(result == ConflictSet_Conflict);
+  bytes = ConflictSet_getBytes(cs);
+  assert(bytes > 0);
   ConflictSet_destroy(cs);
   return 0;
 }

conflict_set_cxx_api_test.cpp

@@ -2,6 +2,8 @@
 
 #include <cassert>
 
+using namespace weaselab;
+
 int main(void) {
   ConflictSet cs(0);
   ConflictSet::WriteRange w;
@@ -17,4 +19,6 @@ int main(void) {
   r.readVersion = 0;
   cs.check(&r, &result, 1);
   assert(result == ConflictSet::Conflict);
+  int64_t bytes = cs.getBytes();
+  assert(bytes > 0);
 }