Add script to build .pkg file for macos

According to
https://cmake.org/cmake/help/latest/command/add_custom_command.html this
more accurately matches the intent

.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

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() {
@@ -258,4 +258,4 @@ void benchConflictSet() {
   }
 }
 
-int main(void) { benchConflictSet(); }
+int main(void) { benchConflictSet(); }

CMakeLists.txt

@@ -1,15 +1,28 @@
 cmake_minimum_required(VERSION 3.18)
 project(
-  conflict_set
-  VERSION 0.0.1
+  conflict-set
+  VERSION 0.0.3
   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
@@ -23,16 +36,32 @@ if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
 endif()
 
 add_compile_options(-fdata-sections -ffunction-sections -Wswitch-enum
-                    -Werror=switch-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)
@@ -40,32 +69,46 @@ 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(${PROJECT_NAME}_object
+target_include_directories(${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")
@@ -73,24 +116,32 @@ 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 -Wunreachable-code -Wpedantic -UNDEBUG)
@@ -99,30 +150,51 @@ include(CTest)
 
 if(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 ${CMAKE_SOURCE_DIR}/include)
-  set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                                             "${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})
+  # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
+  # coverage with a small number of tests.
 
-  # Shared library version of a std::unordered_map-based conflict set (point
-  # queries only)
-  add_library(hash_table SHARED HashTable.cpp)
-  target_compile_options(hash_table PRIVATE -fPIC -fno-exceptions
-                                            -fvisibility=hidden)
-  target_include_directories(hash_table PUBLIC ${CMAKE_SOURCE_DIR}/include)
-  set_target_properties(hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                                              "${CMAKE_BINARY_DIR}/hash_table")
-  set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
-  set_target_properties(
-    hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
-                                                     ${PROJECT_VERSION_MAJOR})
+  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 -fno-exceptions
+                                             -fvisibility=hidden)
+    target_include_directories(skip_list PUBLIC ${CMAKE_SOURCE_DIR}/include)
+    set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                                               "${CMAKE_BINARY_DIR}/skip_list")
+    set_target_properties(skip_list PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
+    set_target_properties(
+      skip_list PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
+                                                      ${PROJECT_VERSION_MAJOR})
+
+    # Shared library version of a std::unordered_map-based conflict set (point
+    # queries only)
+    add_library(hash_table SHARED HashTable.cpp)
+    target_compile_options(hash_table PRIVATE -fno-exceptions
+                                              -fvisibility=hidden)
+    target_include_directories(hash_table PUBLIC ${CMAKE_SOURCE_DIR}/include)
+    set_target_properties(
+      hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                            "${CMAKE_BINARY_DIR}/hash_table")
+    set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
+    set_target_properties(
+      hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
+                                                       ${PROJECT_VERSION_MAJOR})
+
+    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)
@@ -149,10 +221,7 @@ 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 ${TEST_FLAGS})
   if(NOT CMAKE_CROSSCOMPILING)
@@ -167,24 +236,40 @@ if(BUILD_TESTING)
     add_test(NAME conflict_set_fuzz_${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()
 
+  # scripted tests. Written manually to fill in anything libfuzzer couldn't
+  # find.
   add_executable(script_test ScriptTest.cpp)
   target_compile_options(script_test PRIVATE ${TEST_FLAGS})
   target_link_libraries(script_test PRIVATE ${PROJECT_NAME})
-
   file(GLOB SCRIPT_TESTS ${CMAKE_SOURCE_DIR}/script_tests/*)
   foreach(TEST ${SCRIPT_TESTS})
     get_filename_component(name ${TEST} NAME)
     add_test(NAME conflict_set_script_${name} COMMAND script_test ${TEST})
   endforeach()
 
-  add_executable(driver_skip_list TestDriver.cpp)
-  target_compile_options(driver_skip_list PRIVATE ${TEST_FLAGS})
-  target_link_libraries(driver_skip_list PRIVATE skip_list)
-
   find_program(VALGRIND_EXE valgrind)
   if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
     add_test(NAME conflict_set_blackbox_valgrind
@@ -192,12 +277,6 @@ 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
@@ -218,31 +297,45 @@ 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 "${CMAKE_CURRENT_SOURCE_DIR}/LICENSE")
@@ -252,6 +345,29 @@ set(CPACK_RESOURCE_FILE_README "${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
+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)
 
@@ -263,7 +379,7 @@ 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}>)
 
@@ -272,13 +388,16 @@ 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)

ConflictSet.cpp

@@ -29,6 +29,7 @@ limitations under the License.
 #include <span>
 #include <string>
 #include <string_view>
+#include <type_traits>
 #include <utility>
 
 #ifdef HAS_AVX
@@ -39,6 +40,8 @@ limitations under the License.
 
 #include <memcheck.h>
 
+using namespace weaselab;
+
 // Use assert for checking potentially complex properties during tests.
 // Use assume to hint simple properties to the optimizer.
 
@@ -216,6 +219,7 @@ struct Node {
   /* end section that's copied to the next node */
 
   uint8_t *partialKey();
+  size_t size() const;
 
   Type getType() const { return type; }
   int32_t getCapacity() const { return partialKeyCapacity; }
@@ -249,6 +253,8 @@ struct Node0 : Node {
 
   void copyChildrenAndKeyFrom(const Node0 &other);
   void copyChildrenAndKeyFrom(const struct Node3 &other);
+
+  size_t size() const { return sizeof(Node0) + getCapacity(); }
 };
 
 struct Node3 : Node {
@@ -262,6 +268,8 @@ struct Node3 : Node {
   void copyChildrenAndKeyFrom(const Node0 &other);
   void copyChildrenAndKeyFrom(const Node3 &other);
   void copyChildrenAndKeyFrom(const struct Node16 &other);
+
+  size_t size() const { return sizeof(Node3) + getCapacity(); }
 };
 
 struct Node16 : Node {
@@ -275,6 +283,8 @@ struct Node16 : Node {
   void copyChildrenAndKeyFrom(const Node3 &other);
   void copyChildrenAndKeyFrom(const Node16 &other);
   void copyChildrenAndKeyFrom(const struct Node48 &other);
+
+  size_t size() const { return sizeof(Node16) + getCapacity(); }
 };
 
 struct Node48 : Node {
@@ -290,6 +300,8 @@ struct Node48 : Node {
   void copyChildrenAndKeyFrom(const Node16 &other);
   void copyChildrenAndKeyFrom(const Node48 &other);
   void copyChildrenAndKeyFrom(const struct Node256 &other);
+
+  size_t size() const { return sizeof(Node48) + getCapacity(); }
 };
 
 struct Node256 : Node {
@@ -299,6 +311,8 @@ struct Node256 : Node {
   uint8_t *partialKey() { return (uint8_t *)(this + 1); }
   void copyChildrenAndKeyFrom(const Node48 &other);
   void copyChildrenAndKeyFrom(const Node256 &other);
+
+  size_t size() const { return sizeof(Node256) + getCapacity(); }
 };
 
 inline void Node0::copyChildrenAndKeyFrom(const Node0 &other) {
@@ -535,7 +549,7 @@ template <class T> struct BoundedFreeListAllocator {
       } else {
         // The intent is to filter out too-small nodes in the freelist
         removeNode(n);
-        safe_free(n);
+        safe_free(n, sizeof(T) + n->partialKeyCapacity);
       }
     }
 
@@ -547,10 +561,9 @@ template <class T> struct BoundedFreeListAllocator {
   }
 
   void release(T *p) {
-    static_assert(std::is_trivially_destructible_v<T>);
     if (freeListBytes >= kFreeListMaxMemory) {
       removeNode(p);
-      return safe_free(p);
+      return safe_free(p, sizeof(T) + p->partialKeyCapacity);
     }
     memcpy((void *)p, &freeList, sizeof(freeList));
     freeList = p;
@@ -560,11 +573,11 @@ template <class T> struct BoundedFreeListAllocator {
 
   ~BoundedFreeListAllocator() {
     for (void *iter = freeList; iter != nullptr;) {
-      VALGRIND_MAKE_MEM_DEFINED(iter, sizeof(iter));
-      auto *tmp = iter;
+      VALGRIND_MAKE_MEM_DEFINED(iter, sizeof(Node));
+      auto *tmp = (T *)iter;
       memcpy(&iter, iter, sizeof(void *));
-      removeNode(((T *)tmp));
-      safe_free(tmp);
+      removeNode((tmp));
+      safe_free(tmp, sizeof(T) + tmp->partialKeyCapacity);
     }
   }
 
@@ -590,6 +603,23 @@ uint8_t *Node::partialKey() {
   }
 }
 
+size_t Node::size() const {
+  switch (type) {
+  case Type_Node0:
+    return ((Node0 *)this)->size();
+  case Type_Node3:
+    return ((Node3 *)this)->size();
+  case Type_Node16:
+    return ((Node16 *)this)->size();
+  case Type_Node48:
+    return ((Node48 *)this)->size();
+  case Type_Node256:
+    return ((Node256 *)this)->size();
+  default:                   // GCOVR_EXCL_LINE
+    __builtin_unreachable(); // GCOVR_EXCL_LINE
+  }
+}
+
 struct NodeAllocators {
   BoundedFreeListAllocator<Node0> node0;
   BoundedFreeListAllocator<Node3> node3;
@@ -925,6 +955,42 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     assert(self->getType() == Type_Node16);
 
     ++self->numChildren;
+#ifdef HAS_AVX
+    __m128i key_vec = _mm_set1_epi8(index);
+    __m128i indices;
+    memcpy(&indices, self16->index, sizeof(self16->index));
+    __m128i results = _mm_cmpeq_epi8(key_vec, _mm_min_epu8(key_vec, indices));
+    int mask = (1 << (self->numChildren - 1)) - 1;
+    uint32_t bitfield = _mm_movemask_epi8(results) & mask;
+    bitfield |= uint32_t(1) << (self->numChildren - 1);
+    int i = std::countr_zero(bitfield);
+    if (i < self->numChildren - 1) {
+      memmove(self16->index + i + 1, self16->index + i,
+              self->numChildren - (i + 1));
+      memmove(self16->children + i + 1, self16->children + i,
+              (self->numChildren - (i + 1)) * sizeof(Child));
+    }
+#elif defined(HAS_ARM_NEON)
+    uint8x16_t indices;
+    memcpy(&indices, self16->index, sizeof(self16->index));
+    // 0xff for each leq
+    auto results = vcleq_u8(vdupq_n_u8(index), indices);
+    uint64_t mask = (uint64_t(1) << ((self->numChildren - 1) * 4)) - 1;
+    // 0xf for each 0xff (within mask)
+    uint64_t bitfield =
+        vget_lane_u64(
+            vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)),
+            0) &
+        mask;
+    bitfield |= uint64_t(0xf) << ((self->numChildren - 1) * 4);
+    int i = std::countr_zero(bitfield) / 4;
+    if (i < self->numChildren - 1) {
+      memmove(self16->index + i + 1, self16->index + i,
+              self->numChildren - (i + 1));
+      memmove(self16->children + i + 1, self16->children + i,
+              (self->numChildren - (i + 1)) * sizeof(Child));
+    }
+#else
     int i = 0;
     for (; i < int(self->numChildren) - 1; ++i) {
       if (int(self16->index[i]) > int(index)) {
@@ -935,6 +1001,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
         break;
       }
     }
+#endif
     self16->index[i] = index;
     auto &result = self16->children[i].child;
     result = nullptr;
@@ -950,8 +1017,8 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
       self = newSelf;
       goto insert256;
     }
-  insert48:
 
+  insert48:
     auto *self48 = static_cast<Node48 *>(self);
     self48->bitSet.set(index);
     ++self->numChildren;
@@ -963,6 +1030,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     return result;
   }
   case Type_Node256: {
+
   insert256:
     auto *self256 = static_cast<Node256 *>(self);
     ++self->numChildren;
@@ -998,21 +1066,21 @@ Node *nextLogical(Node *node) {
 
 // Invalidates `self`, replacing it with a node of at least capacity.
 // Does not return nodes to freelists when kUseFreeList is false.
-template <bool kUseFreeList>
 void freeAndMakeCapacityAtLeast(Node *&self, int capacity,
                                 NodeAllocators *allocators,
-                                ConflictSet::Impl *impl) {
+                                ConflictSet::Impl *impl,
+                                const bool kUseFreeList) {
   switch (self->getType()) {
   case Type_Node0: {
     auto *self0 = (Node0 *)self;
     auto *newSelf = allocators->node0.allocate(capacity);
     newSelf->copyChildrenAndKeyFrom(*self0);
     getInTree(self, impl) = newSelf;
-    if constexpr (kUseFreeList) {
+    if (kUseFreeList) {
       allocators->node0.release(self0);
     } else {
       removeNode(self0);
-      safe_free(self0);
+      safe_free(self0, self0->size());
     }
     self = newSelf;
   } break;
@@ -1021,11 +1089,11 @@ void freeAndMakeCapacityAtLeast(Node *&self, int capacity,
     auto *newSelf = allocators->node3.allocate(capacity);
     newSelf->copyChildrenAndKeyFrom(*self3);
     getInTree(self, impl) = newSelf;
-    if constexpr (kUseFreeList) {
+    if (kUseFreeList) {
       allocators->node3.release(self3);
     } else {
       removeNode(self3);
-      safe_free(self3);
+      safe_free(self3, self3->size());
     }
     self = newSelf;
   } break;
@@ -1034,11 +1102,11 @@ void freeAndMakeCapacityAtLeast(Node *&self, int capacity,
     auto *newSelf = allocators->node16.allocate(capacity);
     newSelf->copyChildrenAndKeyFrom(*self16);
     getInTree(self, impl) = newSelf;
-    if constexpr (kUseFreeList) {
+    if (kUseFreeList) {
       allocators->node16.release(self16);
     } else {
       removeNode(self16);
-      safe_free(self16);
+      safe_free(self16, self16->size());
     }
     self = newSelf;
   } break;
@@ -1047,11 +1115,11 @@ void freeAndMakeCapacityAtLeast(Node *&self, int capacity,
     auto *newSelf = allocators->node48.allocate(capacity);
     newSelf->copyChildrenAndKeyFrom(*self48);
     getInTree(self, impl) = newSelf;
-    if constexpr (kUseFreeList) {
+    if (kUseFreeList) {
       allocators->node48.release(self48);
     } else {
       removeNode(self48);
-      safe_free(self48);
+      safe_free(self48, self48->size());
     }
     self = newSelf;
   } break;
@@ -1060,11 +1128,11 @@ void freeAndMakeCapacityAtLeast(Node *&self, int capacity,
     auto *newSelf = allocators->node256.allocate(capacity);
     newSelf->copyChildrenAndKeyFrom(*self256);
     getInTree(self, impl) = newSelf;
-    if constexpr (kUseFreeList) {
+    if (kUseFreeList) {
       allocators->node256.release(self256);
     } else {
       removeNode(self256);
-      safe_free(self256);
+      safe_free(self256, self256->size());
     }
     self = newSelf;
   } break;
@@ -1083,11 +1151,9 @@ void maybeDecreaseCapacity(Node *&self, NodeAllocators *allocators,
   if (self->getCapacity() <= maxCapacity) {
     return;
   }
-  freeAndMakeCapacityAtLeast</*kUseFreeList*/ false>(self, maxCapacity,
-                                                     allocators, impl);
+  freeAndMakeCapacityAtLeast(self, maxCapacity, allocators, impl, false);
 }
 
-template <Type kType>
 void maybeDownsize(Node *self, NodeAllocators *allocators,
                    ConflictSet::Impl *impl, Node *&dontInvalidate) {
 
@@ -1095,9 +1161,7 @@ void maybeDownsize(Node *self, NodeAllocators *allocators,
   fprintf(stderr, "maybeDownsize: %s\n", getSearchPathPrintable(self).c_str());
 #endif
 
-  assert(self->getType() == kType);
-  static_assert(kType != Type_Node0);
-  switch (kType) {
+  switch (self->getType()) {
   case Type_Node0:           // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
   case Type_Node3: {
@@ -1113,8 +1177,7 @@ void maybeDownsize(Node *self, NodeAllocators *allocators,
 
       if (minCapacity > child->getCapacity()) {
         const bool update = child == dontInvalidate;
-        freeAndMakeCapacityAtLeast</*kUseFreeList*/ true>(child, minCapacity,
-                                                          allocators, impl);
+        freeAndMakeCapacityAtLeast(child, minCapacity, allocators, impl, true);
         if (update) {
           dontInvalidate = child;
         }
@@ -1201,24 +1264,7 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
 
   if (self->numChildren != 0) {
     const bool update = result == dontInvalidate;
-    switch (self->getType()) {
-    case Type_Node0:           // GCOVR_EXCL_LINE
-      __builtin_unreachable(); // GCOVR_EXCL_LINE
-    case Type_Node3:
-      maybeDownsize<Type_Node3>(self, allocators, impl, result);
-      break;
-    case Type_Node16:
-      maybeDownsize<Type_Node16>(self, allocators, impl, result);
-      break;
-    case Type_Node48:
-      maybeDownsize<Type_Node48>(self, allocators, impl, result);
-      break;
-    case Type_Node256:
-      maybeDownsize<Type_Node256>(self, allocators, impl, result);
-      break;
-    default:                   // GCOVR_EXCL_LINE
-      __builtin_unreachable(); // GCOVR_EXCL_LINE
-    }
+    maybeDownsize(self, allocators, impl, result);
     if (update) {
       dontInvalidate = result;
     }
@@ -1244,11 +1290,6 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
 
     --parent->numChildren;
     assert(parent->numChildren > 0 || parent->entryPresent);
-    const bool update = result == dontInvalidate;
-    maybeDownsize<Type_Node3>(parent, allocators, impl, result);
-    if (update) {
-      dontInvalidate = result;
-    }
   } break;
   case Type_Node16: {
     auto *parent16 = static_cast<Node16 *>(parent);
@@ -1265,13 +1306,6 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
 
     // By kMinChildrenNode16
     assert(parent->numChildren > 0);
-
-    const bool update = result == dontInvalidate;
-    maybeDownsize<Type_Node16>(parent, allocators, impl, result);
-    if (update) {
-      dontInvalidate = result;
-    }
-
   } break;
   case Type_Node48: {
     auto *parent48 = static_cast<Node48 *>(parent);
@@ -1292,12 +1326,6 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
 
     // By kMinChildrenNode48
     assert(parent->numChildren > 0);
-
-    const bool update = result == dontInvalidate;
-    maybeDownsize<Type_Node48>(parent, allocators, impl, result);
-    if (update) {
-      dontInvalidate = result;
-    }
   } break;
   case Type_Node256: {
     auto *parent256 = static_cast<Node256 *>(parent);
@@ -1309,16 +1337,17 @@ Node *erase(Node *self, NodeAllocators *allocators, ConflictSet::Impl *impl,
     // By kMinChildrenNode256
     assert(parent->numChildren > 0);
 
-    const bool update = result == dontInvalidate;
-    maybeDownsize<Type_Node256>(parent, allocators, impl, result);
-    if (update) {
-      dontInvalidate = result;
-    }
   } break;
   default:                   // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
   }
 
+  const bool update = result == dontInvalidate;
+  maybeDownsize(parent, allocators, impl, result);
+  if (update) {
+    dontInvalidate = result;
+  }
+
   return result;
 }
 
@@ -1540,6 +1569,9 @@ bool checkPointRead(Node *n, const std::span<const uint8_t> key,
         goto downLeftSpine;
       } else {
         n = nextSibling(n);
+        if (n == nullptr) {
+          return true;
+        }
         goto downLeftSpine;
       }
     }
@@ -1556,6 +1588,9 @@ bool checkPointRead(Node *n, const std::span<const uint8_t> key,
           goto downLeftSpine;
         } else {
           n = nextSibling(n);
+          if (n == nullptr) {
+            return true;
+          }
           goto downLeftSpine;
         }
       }
@@ -1570,9 +1605,6 @@ bool checkPointRead(Node *n, const std::span<const uint8_t> key,
     }
   }
 downLeftSpine:
-  if (n == nullptr) {
-    return true;
-  }
   for (;;) {
     if (n->entryPresent) {
       return n->entry.rangeVersion <= readVersion;
@@ -1672,7 +1704,10 @@ Vector<uint8_t> getSearchPath(Arena &arena, Node *n) {
 } // GCOVR_EXCL_LINE
 
 // Return true if the max version among all keys that start with key + [child],
-// where begin < child < end, is <= readVersion
+// where begin < child < end, is <= readVersion.
+//
+// Precondition: transitively, no child of n has a search path that's a longer
+// prefix of key than n
 bool checkRangeStartsWith(Node *n, std::span<const uint8_t> key, int begin,
                           int end, int64_t readVersion,
                           ConflictSet::Impl *impl) {
@@ -1680,60 +1715,58 @@ bool checkRangeStartsWith(Node *n, std::span<const uint8_t> key, int begin,
   fprintf(stderr, "%s(%02x,%02x)*\n", printable(key).c_str(), begin, end);
 #endif
   auto remaining = key;
-  for (;;) {
-    if (maxVersion(n, impl) <= readVersion) {
-      return true;
-    }
-    if (remaining.size() == 0) {
-      return maxBetweenExclusive(n, begin, end) <= readVersion;
-    }
+  if (remaining.size() == 0) {
+    return maxBetweenExclusive(n, begin, end) <= readVersion;
+  }
 
-    auto *child = getChild(n, remaining[0]);
-    if (child == nullptr) {
-      int c = getChildGeq(n, remaining[0]);
-      if (c >= 0) {
-        n = getChildExists(n, c);
+  auto *child = getChild(n, remaining[0]);
+  if (child == nullptr) {
+    int c = getChildGeq(n, remaining[0]);
+    if (c >= 0) {
+      n = getChildExists(n, c);
+      goto downLeftSpine;
+    } else {
+      n = nextSibling(n);
+      if (n == nullptr) {
+        return true;
+      }
+      goto downLeftSpine;
+    }
+  }
+
+  n = child;
+  remaining = remaining.subspan(1, remaining.size() - 1);
+
+  assert(n->partialKeyLen > 0);
+  {
+    int commonLen = std::min<int>(n->partialKeyLen, remaining.size());
+    int i = longestCommonPrefix(n->partialKey(), remaining.data(), commonLen);
+    if (i < commonLen) {
+      auto c = n->partialKey()[i] <=> remaining[i];
+      if (c > 0) {
         goto downLeftSpine;
       } else {
         n = nextSibling(n);
+        if (n == nullptr) {
+          return true;
+        }
         goto downLeftSpine;
       }
     }
-
-    n = child;
-    remaining = remaining.subspan(1, remaining.size() - 1);
-
-    if (n->partialKeyLen > 0) {
-      int commonLen = std::min<int>(n->partialKeyLen, remaining.size());
-      int i = longestCommonPrefix(n->partialKey(), remaining.data(), commonLen);
-      if (i < commonLen) {
-        auto c = n->partialKey()[i] <=> remaining[i];
-        if (c > 0) {
-          goto downLeftSpine;
-        } else {
-          n = nextSibling(n);
-          goto downLeftSpine;
-        }
-      }
-      if (commonLen == n->partialKeyLen) {
-        // partial key matches
-        remaining = remaining.subspan(commonLen, remaining.size() - commonLen);
-      } else if (n->partialKeyLen > int(remaining.size())) {
-        if (begin < n->partialKey()[remaining.size()] &&
-            n->partialKey()[remaining.size()] < end) {
-          if (n->entryPresent && n->entry.rangeVersion > readVersion) {
-            return false;
-          }
-          return maxVersion(n, impl) <= readVersion;
-        }
-        return true;
+    assert(n->partialKeyLen > int(remaining.size()));
+    if (begin < n->partialKey()[remaining.size()] &&
+        n->partialKey()[remaining.size()] < end) {
+      if (n->entryPresent && n->entry.rangeVersion > readVersion) {
+        return false;
       }
+      return maxVersion(n, impl) <= readVersion;
     }
-  }
-downLeftSpine:
-  if (n == nullptr) {
     return true;
   }
+
+  __builtin_unreachable(); // GCOVR_EXCL_LINE
+
+downLeftSpine:
   for (;;) {
     if (n->entryPresent) {
       return n->entry.rangeVersion <= readVersion;
@@ -1801,6 +1834,10 @@ struct CheckRangeLeftSide {
           return true;
         } else {
           n = nextSibling(n);
+          if (n == nullptr) {
+            ok = true;
+            return true;
+          }
           return downLeftSpine();
         }
       }
@@ -1828,6 +1865,10 @@ struct CheckRangeLeftSide {
             return true;
           } else {
             n = nextSibling(n);
+            if (n == nullptr) {
+              ok = true;
+              return true;
+            }
             return downLeftSpine();
           }
         }
@@ -1836,9 +1877,7 @@ struct CheckRangeLeftSide {
           remaining =
               remaining.subspan(commonLen, remaining.size() - commonLen);
         } else if (n->partialKeyLen > int(remaining.size())) {
-          if (searchPathLen < prefixLen) {
-            return downLeftSpine();
-          }
+          assert(searchPathLen >= prefixLen);
           if (n->entryPresent && n->entry.rangeVersion > readVersion) {
             ok = false;
             return true;
@@ -1866,10 +1905,6 @@ struct CheckRangeLeftSide {
 
   bool downLeftSpine() {
     phase = DownLeftSpine;
-    if (n == nullptr) {
-      ok = true;
-      return true;
-    }
     return false;
   }
 };
@@ -2017,10 +2052,7 @@ struct CheckRangeRightSide {
 
   bool downLeftSpine() {
     phase = DownLeftSpine;
-    if (n == nullptr) {
-      ok = true;
-      return true;
-    }
+    assert(n != nullptr);
     return false;
   }
 };
@@ -2214,7 +2246,7 @@ void destroyTree(Node *root) {
       assert(c != nullptr);
       toFree.push_back(c);
     }
-    safe_free(n);
+    safe_free(n, n->size());
   }
 }
 
@@ -2356,6 +2388,9 @@ Iterator firstGeq(Node *n, const std::span<const uint8_t> key) {
         goto downLeftSpine;
       } else {
         n = nextSibling(n);
+        if (n == nullptr) {
+          return {nullptr, 1};
+        }
         goto downLeftSpine;
       }
     }
@@ -2386,9 +2421,6 @@ Iterator firstGeq(Node *n, const std::span<const uint8_t> key) {
     }
   }
 downLeftSpine:
-  if (n == nullptr) {
-    return {nullptr, 1};
-  }
   for (;;) {
     if (n->entryPresent) {
       return {n, 1};
@@ -2438,9 +2470,14 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
       return;
     }
     this->oldestVersion = oldestVersion;
+#ifdef NDEBUG
+    // This is here for performance reasons, since we want to amortize the cost
+    // of storing the search path as a string. In tests, we want to exercise the
+    // rest of the code often.
     if (keyUpdates < 100) {
       return;
     }
+#endif
     Node *n = firstGeq(root, removalKey).n;
     // There's no way to erase removalKey without introducing a key after it
     assert(n != nullptr);
@@ -2466,13 +2503,10 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     }
     if (n == nullptr) {
       removalKey = {};
-      return;
+    } else {
+      removalKeyArena = Arena();
+      removalKey = getSearchPath(removalKeyArena, n);
     }
-    if (keyUpdates == 0) {
-      keyUpdates = 10;
-    }
-    removalKeyArena = Arena();
-    removalKey = getSearchPath(removalKeyArena, n);
   }
 
   explicit Impl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
@@ -2501,6 +2535,7 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   Node *root;
   int64_t rootMaxVersion;
   int64_t oldestVersion;
+  int64_t totalBytes = 0;
 };
 
 // Precondition - an entry for index must exist in the node
@@ -2511,7 +2546,7 @@ int64_t &maxVersion(Node *n, ConflictSet::Impl *impl) {
     return impl->rootMaxVersion;
   }
   switch (n->getType()) {
-  case Type_Node0:
+  case Type_Node0:           // GCOVR_EXCL_LINE
     __builtin_unreachable(); // GCOVR_EXCL_LINE
   case Type_Node3: {
     auto *n3 = static_cast<Node3 *>(n);
@@ -2554,20 +2589,39 @@ void ConflictSet::check(const ReadRange *reads, Result *results,
 
 void ConflictSet::addWrites(const WriteRange *writes, int count,
                             int64_t writeVersion) {
-  return impl->addWrites(writes, count, writeVersion);
+  mallocBytesDelta = 0;
+  impl->addWrites(writes, count, writeVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
 void ConflictSet::setOldestVersion(int64_t oldestVersion) {
-  return impl->setOldestVersion(oldestVersion);
+  mallocBytesDelta = 0;
+  impl->setOldestVersion(oldestVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
+int64_t ConflictSet::getBytes() const { return impl->totalBytes; }
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
-    : impl(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
+    : impl((mallocBytesDelta = 0,
+            new(safe_malloc(sizeof(Impl))) Impl{oldestVersion})) {
+  impl->totalBytes += mallocBytesDelta;
+}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
     impl->~Impl();
-    safe_free(impl);
+    safe_free(impl, sizeof(*impl));
   }
 }
 
@@ -2593,24 +2647,44 @@ ConflictSet_check(void *cs, const ConflictSet_ReadRange *reads,
 __attribute__((__visibility__("default"))) void
 ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes, int count,
                       int64_t writeVersion) {
-  ((ConflictSet::Impl *)cs)->addWrites(writes, count, writeVersion);
+  auto *impl = ((ConflictSet::Impl *)cs);
+  mallocBytesDelta = 0;
+  impl->addWrites(writes, count, writeVersion);
+  impl->totalBytes += mallocBytesDelta;
 }
 __attribute__((__visibility__("default"))) void
 ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
-  ((ConflictSet::Impl *)cs)->setOldestVersion(oldestVersion);
+  auto *impl = ((ConflictSet::Impl *)cs);
+  mallocBytesDelta = 0;
+  impl->setOldestVersion(oldestVersion);
+  impl->totalBytes += mallocBytesDelta;
 }
 __attribute__((__visibility__("default"))) void *
 ConflictSet_create(int64_t oldestVersion) {
-  return new (safe_malloc(sizeof(ConflictSet::Impl)))
+  mallocBytesDelta = 0;
+  auto *result = new (safe_malloc(sizeof(ConflictSet::Impl)))
       ConflictSet::Impl{oldestVersion};
+  result->totalBytes += mallocBytesDelta;
+  return result;
 }
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
   using Impl = ConflictSet::Impl;
   ((Impl *)cs)->~Impl();
-  safe_free(cs);
+  safe_free(cs, sizeof(Impl));
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  using Impl = ConflictSet::Impl;
+  return ((Impl *)cs)->totalBytes;
 }
 }
 
+// Make sure abi is well-defined
+static_assert(std::is_standard_layout_v<ConflictSet::Result>);
+static_assert(std::is_standard_layout_v<ConflictSet::Key>);
+static_assert(std::is_standard_layout_v<ConflictSet::ReadRange>);
+static_assert(std::is_standard_layout_v<ConflictSet::WriteRange>);
+
 namespace {
 
 std::string getSearchPathPrintable(Node *n) {
@@ -2823,8 +2897,8 @@ Iterator firstGeq(Node *n, std::string_view key) {
   }
 }
 
-bool checkCorrectness(Node *node, int64_t oldestVersion,
-                      ConflictSet::Impl *impl) {
+[[maybe_unused]] bool checkCorrectness(Node *node, int64_t oldestVersion,
+                                       ConflictSet::Impl *impl) {
   bool success = true;
 
   checkParentPointers(node, success);

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

@@ -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}) {}
+    : impl(new(safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
     impl->~Impl();
-    safe_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();
-  safe_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>
@@ -35,6 +39,17 @@ 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.
 
@@ -43,43 +58,55 @@ operator<=>(const std::span<const uint8_t> &lhs,
 #if SHOW_MEMORY
 inline int64_t mallocBytes = 0;
 inline int64_t peakMallocBytes = 0;
-constexpr auto kIntMallocHeaderSize = 16;
+#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) {
+  mallocBytesDelta += s;
 #if SHOW_MEMORY
   mallocBytes += s;
   if (mallocBytes > peakMallocBytes) {
     peakMallocBytes = mallocBytes;
   }
-  void *p = malloc(s + kIntMallocHeaderSize);
-  if (p == nullptr) {
-    abort();
-  }
-  memcpy(p, &s, sizeof(s));
-  return (char *)p + kIntMallocHeaderSize;
-#else
-  void *p = malloc(s);
-  if (p == nullptr) {
-    abort();
-  }
-  return p;
 #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) {
+__attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
+  mallocBytesDelta -= s;
 #if SHOW_MEMORY
-  size_t s;
-  memcpy(&s, (char *)p - kIntMallocHeaderSize, sizeof(s));
   mallocBytes -= s;
-  free((char *)p - kIntMallocHeaderSize);
+  free(p);
 #else
+#ifndef NDEBUG
+  (char *&)p -= kMallocHeaderSize;
+  size_t expected;
+  memcpy(&expected, p, sizeof(expected));
+  assert(s == expected);
+#endif
   free(p);
 #endif
 }
@@ -127,6 +154,7 @@ 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 -
@@ -135,6 +163,7 @@ 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);
 }
 
@@ -150,12 +179,10 @@ 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);
@@ -166,7 +193,7 @@ inline Arena::Arena(int initialSize) : impl(nullptr) {
 inline void onDestroy(Arena::ArenaImpl *impl) {
   while (impl) {
     auto *prev = impl->prev;
-    safe_free(impl);
+    safe_free(impl, sizeof(Arena::ArenaImpl) + impl->capacity);
     impl = prev;
   }
 }
@@ -191,7 +218,7 @@ 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);
@@ -386,34 +413,6 @@ inline uint32_t Arbitrary::bounded(uint32_t s) {
 
 // ==================== END ARBITRARY IMPL ====================
 
-// ==================== BEGIN UTILITIES IMPL ====================
-
-// Call Stepwise::step for each element of remaining until it returns true.
-// Applies a permutation to `remaining` as a side effect.
-template <class Stepwise> void runInterleaved(std::span<Stepwise> remaining) {
-  while (remaining.size() > 0) {
-    for (int i = 0; i < int(remaining.size());) {
-      bool done = remaining[i].step();
-      if (done) {
-        if (i != int(remaining.size()) - 1) {
-          using std::swap;
-          swap(remaining[i], remaining.back());
-        }
-        remaining = remaining.subspan(0, remaining.size() - 1);
-      } else {
-        ++i;
-      }
-    }
-  }
-};
-
-template <class Stepwise> void runSequential(std::span<Stepwise> remaining) {
-  for (auto &r : remaining) {
-    while (!r.step()) {
-    }
-  }
-}
-
 struct ReferenceImpl {
   explicit ReferenceImpl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
     writeVersionMap[""] = oldestVersion;
@@ -669,32 +668,60 @@ template <class ConflictSetImpl> struct TestDriver {
       auto *results2 =
           new (arena) ConflictSet::Result[numPointReads + numRangeReads];
 
+#ifdef THREAD_TEST
+      auto *results3 =
+          new (arena) ConflictSet::Result[numPointReads + numRangeReads];
+      std::latch ready{1};
+      std::thread thread2{[&]() {
+        ready.count_down();
+        cs.check(reads, results3, numPointReads + numRangeReads);
+      }};
+      ready.wait();
+#endif
+
       CALLGRIND_START_INSTRUMENTATION;
       cs.check(reads, results1, numPointReads + numRangeReads);
       CALLGRIND_STOP_INSTRUMENTATION;
 
       refImpl.check(reads, results2, numPointReads + numRangeReads);
-      for (int i = 0; i < numPointReads + numRangeReads; ++i) {
-        if (results1[i] != results2[i]) {
-          if (reads[i].end.len == 0) {
-            fprintf(stderr,
-                    "Expected %s, got %s for read of {%s} at version %" PRId64
-                    "\n",
-                    resultToStr(results2[i]), resultToStr(results1[i]),
-                    printable(reads[i].begin).c_str(), reads[i].readVersion);
-          } else {
-            fprintf(
-                stderr,
-                "Expected %s, got %s for read of [%s, %s) at version %" PRId64
-                "\n",
-                resultToStr(results2[i]), resultToStr(results1[i]),
-                printable(reads[i].begin).c_str(),
-                printable(reads[i].end).c_str(), reads[i].readVersion);
+
+      auto compareResults = [reads](ConflictSet::Result *results1,
+                                    ConflictSet::Result *results2, int count) {
+        for (int i = 0; i < count; ++i) {
+          if (results1[i] != results2[i]) {
+            if (reads[i].end.len == 0) {
+              fprintf(stderr,
+                      "Expected %s, got %s for read of {%s} at version %" PRId64
+                      "\n",
+                      resultToStr(results2[i]), resultToStr(results1[i]),
+                      printable(reads[i].begin).c_str(), reads[i].readVersion);
+            } else {
+              fprintf(
+                  stderr,
+                  "Expected %s, got %s for read of [%s, %s) at version %" PRId64
+                  "\n",
+                  resultToStr(results2[i]), resultToStr(results1[i]),
+                  printable(reads[i].begin).c_str(),
+                  printable(reads[i].end).c_str(), reads[i].readVersion);
+            }
+            return false;
           }
-          ok = false;
-          return true;
         }
+        return true;
+      };
+
+      if (!compareResults(results1, results2, numPointReads + numRangeReads)) {
+        ok = false;
+        return true;
       }
+
+#ifdef THREAD_TEST
+      thread2.join();
+      if (!compareResults(results3, results2, numPointReads + numRangeReads)) {
+        ok = false;
+        return true;
+      }
+#endif
     }
     return false;
   }

Jenkinsfile

@@ -48,6 +48,17 @@ pipeline {
                 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 {

README.md

@@ -9,76 +9,76 @@ Hardware for all benchmarks is a mac m1 2020.
 ## Skip list
 
 ```
-New conflict set: 1.962 sec
-                  0.637 Mtransactions/sec
-                  2.549 Mkeys/sec
-Detect only:      1.853 sec
-                  0.674 Mtransactions/sec
-                  2.698 Mkeys/sec
-Skiplist only:    1.269 sec
-                  0.985 Mtransactions/sec
-                  3.940 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: 0.0526
-                 Add: 0.054
-              Detect: 1.85
-              D.Sort: 0.413
-           D.Combine: 0.0148
-         D.CheckRead: 0.678
-   D.CheckIntraBatch: 0.00679
-        D.MergeWrite: 0.591
-      D.RemoveBefore: 0.147
+               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: 1.342 sec
-                  0.931 Mtransactions/sec
-                  3.726 Mkeys/sec
-Detect only:      1.227 sec
-                  1.018 Mtransactions/sec
-                  4.074 Mkeys/sec
-Skiplist only:    0.576 sec
-                  2.169 Mtransactions/sec
-                  8.676 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: 0.0568
-                 Add: 0.0564
-              Detect: 1.23
-              D.Sort: 0.414
-           D.Combine: 0.0162
-         D.CheckRead: 0.228
-   D.CheckIntraBatch: 0.00665
-        D.MergeWrite: 0.348
-      D.RemoveBefore: 0.211
+               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
-|--------------------:|--------------------:|--------:|----------:|:----------
-|              254.17 |        3,934,426.23 |    0.3% |      0.01 | `point reads`
-|              275.58 |        3,628,769.92 |    2.1% |      0.01 | `prefix reads`
-|              494.87 |        2,020,718.51 |    0.7% |      0.01 | `range reads`
-|              495.91 |        2,016,512.61 |    1.7% |      0.01 | `point writes`
-|              478.11 |        2,091,578.00 |    0.8% |      0.01 | `prefix writes`
-|              307.08 |        3,256,480.40 |    1.9% |      0.04 | `range writes`
-|              610.89 |        1,636,953.81 |    0.7% |      0.01 | `monotonic increasing point writes`
+|  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
-|--------------------:|--------------------:|--------:|----------:|:----------
-|               22.39 |       44,667,832.17 |    0.2% |      0.01 | `point reads`
-|               53.89 |       18,554,840.41 |    3.4% |      0.01 | `prefix reads`
-|              217.71 |        4,593,220.24 |    0.2% |      0.01 | `range reads`
-|               29.72 |       33,642,099.66 |    1.1% |      0.01 | `point writes`
-|               44.86 |       22,294,037.02 |    0.8% |      0.01 | `prefix writes`
-|               55.00 |       18,181,818.18 |    0.8% |      0.03 | `range writes`
-|              109.21 |        9,156,917.53 |    2.9% |      0.01 | `monotonic increasing point writes`
+|  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
 
@@ -87,13 +87,13 @@ Point queries only, best of three runs. Gc ratio is the ratio of time spent doin
 ## skip list
 
 ```
-Check: 11.7282 seconds, 318.763 MB/s, Add: 5.76499 seconds, 121.776 MB/s, Gc ratio: 48.0772%
+Check: 11.3385 seconds, 329.718 MB/s, Add: 5.35612 seconds, 131.072 MB/s, Gc ratio: 45.7173%
 ```
 
 ## radix tree
 
 ```
-Check: 3.27475 seconds, 1141.62 MB/s, Add: 2.13594 seconds, 328.678 MB/s, Gc ratio: 50.1739%
+Check: 2.48583 seconds, 1503.93 MB/s, Add: 2.12768 seconds, 329.954 MB/s, Gc ratio: 41.7943%
 ```
 
 ## hash table
@@ -101,5 +101,5 @@ Check: 3.27475 seconds, 1141.62 MB/s, Add: 2.13594 seconds, 328.678 MB/s, Gc rat
 (The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)
 
 ```
-Check: 1.86291 seconds, 2006.81 MB/s, Add: 0.923653 seconds, 760.067 MB/s, Gc ratio: 54.2605%
-```
+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())
@@ -28,7 +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;
@@ -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",
-      checkTime, checkBytes / checkTime * 1e-6, addTime,
-      addBytes / addTime * 1e-6, gcTime / (gcTime + addTime) * 1e2);
-}
+  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,
+         double(peakMemory));
+}

ScriptTest.cpp

@@ -51,6 +51,8 @@ int main(int argc, const char **argv) {
     StringView b;
     StringView e;
     int64_t v = 0;
+    int64_t lastWriteVersion = 0;
+    int64_t lastOldestVersion = 0;
     std::vector<ConflictSet::WriteRange> writeRanges;
     std::vector<ConflictSet::ReadRange> readRanges;
     std::vector<ConflictSet::Result> results;
@@ -84,7 +86,9 @@ int main(int argc, const char **argv) {
         readRanges.push_back(r);
       } else if (line.starts_with("pointwrite"_v)) {
         printf("pointwrite\n");
-        // TODO check sorted
+        assert(writeRanges.empty() ||
+               (writeRanges.back().end.len == 0 ? writeRanges.back().begin
+                                                : writeRanges.back().end) < b);
         ConflictSet::WriteRange w;
         w.begin.p = b.data();
         w.begin.len = b.size();
@@ -101,7 +105,10 @@ int main(int argc, const char **argv) {
         readRanges.push_back(r);
       } else if (line.starts_with("rangewrite"_v)) {
         printf("rangewrite\n");
-        // TODO check sorted
+        assert(b < e);
+        assert(writeRanges.empty() ||
+               (writeRanges.back().end.len == 0 ? writeRanges.back().begin
+                                                : writeRanges.back().end) < b);
         ConflictSet::WriteRange w;
         w.begin.p = b.data();
         w.begin.len = b.size();
@@ -125,11 +132,15 @@ int main(int argc, const char **argv) {
         readRanges = {};
       } else if (line.starts_with("addwrites"_v)) {
         printf("addwrites\n");
+        assert(v > lastWriteVersion);
+        lastWriteVersion = v;
         cs.addWrites(writeRanges.data(), writeRanges.size(), v);
         ref.addWrites(writeRanges.data(), writeRanges.size(), v);
         writeRanges = {};
       } else if (line.starts_with("setoldest"_v)) {
         printf("setoldest\n");
+        assert(v > lastOldestVersion);
+        lastOldestVersion = v;
         cs.setOldestVersion(v);
         ref.setOldestVersion(v);
       } else if (line.empty() || line.starts_with(";"_v)) {

SkipList.cpp

@@ -149,7 +149,7 @@ private:
       setMaxVersion(level, v);
     }
 
-    void destroy() { safe_free(this); }
+    void destroy() { safe_free(this, getNodeSize()); }
 
   private:
     int getNodeSize() const {
@@ -627,6 +627,8 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
         removalArena, {finger.getValue().data(), finger.getValue().size()});
   }
 
+  int64_t totalBytes = 0;
+
 private:
   int64_t keyUpdates = 10;
   Arena removalArena;
@@ -637,25 +639,44 @@ private:
 
 void ConflictSet::check(const ReadRange *reads, Result *results,
                         int count) const {
-  return impl->check(reads, results, count);
+  impl->check(reads, results, count);
 }
 
 void ConflictSet::addWrites(const WriteRange *writes, int count,
                             int64_t writeVersion) {
-  return impl->addWrites(writes, count, writeVersion);
+  mallocBytesDelta = 0;
+  impl->addWrites(writes, count, writeVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
 void ConflictSet::setOldestVersion(int64_t oldestVersion) {
-  return impl->setOldestVersion(oldestVersion);
+  mallocBytesDelta = 0;
+  impl->setOldestVersion(oldestVersion);
+  impl->totalBytes += mallocBytesDelta;
+#if SHOW_MEMORY
+  if (impl->totalBytes != mallocBytes) {
+    abort();
+  }
+#endif
 }
 
+int64_t ConflictSet::getBytes() const { return impl->totalBytes; }
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
-    : impl(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}
+    : impl((mallocBytesDelta = 0,
+            new(safe_malloc(sizeof(Impl))) Impl{oldestVersion})) {
+  impl->totalBytes += mallocBytesDelta;
+}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
     impl->~Impl();
-    safe_free(impl);
+    safe_free(impl, sizeof(Impl));
   }
 }
 
@@ -695,7 +716,12 @@ ConflictSet_create(int64_t oldestVersion) {
 __attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
   using Impl = ConflictSet::Impl;
   ((Impl *)cs)->~Impl();
-  safe_free(cs);
+  safe_free(cs, sizeof(Impl));
+}
+__attribute__((__visibility__("default"))) int64_t
+ConflictSet_getBytes(void *cs) {
+  using Impl = ConflictSet::Impl;
+  return ((Impl *)cs)->totalBytes;
 }
 }
 

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 +1,7 @@
+__tlv_bootstrap
+_abort
+_bzero
+_free
+_malloc
+_memcpy
+_memmove

conflict_set.py

@@ -0,0 +1,120 @@
+import ctypes
+import enum
+import os
+
+from typing import Optional
+
+_lib = None
+for f in (
+    os.path.dirname(__file__) + "/build/radix_tree/libconflict-set.so.0",
+    os.path.dirname(__file__) + "/build/radix_tree/libconflict-set.0.dylib",
+):
+    try:
+        _lib = ctypes.cdll.LoadLibrary(f)
+    except:
+        pass
+
+if _lib is None:
+    import sys
+
+    print("Could not find libconflict-set", file=sys.stderr)
+    sys.exit(1)
+
+
+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)]
+
+
+_lib.ConflictSet_create.argtypes = (ctypes.c_int64,)
+_lib.ConflictSet_create.restype = ctypes.c_void_p
+
+_lib.ConflictSet_check.argtypes = (
+    ctypes.c_void_p,
+    ctypes.POINTER(ReadRange),
+    ctypes.POINTER(ctypes.c_int),
+    ctypes.c_int,
+)
+
+_lib.ConflictSet_addWrites.argtypes = (
+    ctypes.c_void_p,
+    ctypes.POINTER(WriteRange),
+    ctypes.c_int,
+    ctypes.c_int64,
+)
+
+_lib.ConflictSet_setOldestVersion.argtypes = (ctypes.c_void_p, ctypes.c_int64)
+
+_lib.ConflictSet_destroy.argtypes = (ctypes.c_void_p,)
+
+
+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))()
+    b.value = begin
+    if end is None:
+        e = (ctypes.c_ubyte * 0)()
+        e.value = b""
+    else:
+        e = (ctypes.c_ubyte * len(end))()
+        e.value = 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))()
+    b.value = begin
+    if end is None:
+        e = (ctypes.c_ubyte * 0)()
+        e.value = b""
+    else:
+        e = (ctypes.c_ubyte * len(end))()
+        e.value = end
+    return ReadRange(_Key(b, len(b)), _Key(e, len(e)), version)
+
+
+class ConflictSet:
+    def __init__(self, version: int = 0) -> None:
+        self.p = _lib.ConflictSet_create(version)
+
+    def addWrites(self, version: int, *writes: WriteRange):
+        _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))()
+        _lib.ConflictSet_check(self.p, *reads, r, 1)
+        return [Result(x) for x in r]
+
+    def setOldestVersion(self, version: int) -> None:
+        _lib.ConflictSet_setOldestVersion(self.p, version)
+
+    def __enter__(self):
+        return self
+
+    def close(self) -> None:
+        if self.p is not None:
+            _lib.ConflictSet_destroy(self.p)
+            self.p = None
+
+    def __exit__(self, exception_type, exception_value, exception_traceback):
+        if self.p is not None:
+            _lib.ConflictSet_destroy(self.p)
+            self.p = None

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);
 }