Update corpus

This is the default now

.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: 6d365699efc33b1b432eab5b4ae331a19e1857de  # frozen: v18.1.2
+    rev: 6d365699efc33b1b432eab5b4ae331a19e1857de # frozen: v18.1.2
     hooks:
       - id: clang-format
         exclude: ".*third_party/.*"
   - repo: https://github.com/cheshirekow/cmake-format-precommit
-    rev: e2c2116d86a80e72e7146a06e68b7c228afc6319  # frozen: v0.6.13
+    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
@@ -21,10 +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
+    rev: a23f6b85d0fdd5bb9d564e2579e678033debbdff # frozen: v0.10.0.1
     hooks:
-    -   id: shellcheck
+      - id: shellcheck
+  - repo: https://github.com/psf/black
+    rev: 552baf822992936134cbd31a38f69c8cfe7c0f05 # frozen: 24.3.0
+    hooks:
+      - id: black

Bench.cpp

@@ -258,4 +258,107 @@ void benchConflictSet() {
   }
 }
 
-int main(void) { benchConflictSet(); }
+constexpr int kKeyLenForWorstCase = 50;
+
+ConflictSet worstCaseConflictSetForRadixRangeRead(int cardinality) {
+  ConflictSet cs{0};
+
+  for (int i = 0; i < kKeyLenForWorstCase; ++i) {
+    for (int j = 0; j < cardinality; ++j) {
+      auto b = std::vector<uint8_t>(i, 0);
+      b.push_back(j);
+      auto e = std::vector<uint8_t>(i, 255);
+      e.push_back(255 - j);
+      weaselab::ConflictSet::WriteRange w[] = {{
+                                                   {b.data(), int(b.size())},
+                                                   {nullptr, 0},
+                                               },
+                                               {
+                                                   {e.data(), int(e.size())},
+                                                   {nullptr, 0},
+                                               }};
+      std::sort(std::begin(w), std::end(w),
+                [](const auto &lhs, const auto &rhs) {
+                  int cl = std::min(lhs.begin.len, rhs.begin.len);
+                  if (cl > 0) {
+                    int c = memcmp(lhs.begin.p, rhs.begin.p, cl);
+                    if (c != 0) {
+                      return c < 0;
+                    }
+                  }
+                  return lhs.begin.len < rhs.begin.len;
+                });
+      cs.addWrites(w, sizeof(w) / sizeof(w[0]), 0);
+    }
+  }
+
+  // Defeat short-circuiting on the left
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 0);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  // Defeat short-circuiting on the right
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 255);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  return cs;
+}
+
+void benchWorstCaseForRadixRangeRead() {
+  ankerl::nanobench::Bench bench;
+
+  std::unique_ptr<ConflictSet> cs[256];
+  for (int i = 0; i < 256; ++i) {
+    cs[i] =
+        std::make_unique<ConflictSet>(worstCaseConflictSetForRadixRangeRead(i));
+  }
+
+  auto begin = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 0);
+  begin.push_back(1);
+  auto end = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 255);
+  end.push_back(254);
+
+  weaselab::ConflictSet::Result result;
+  weaselab::ConflictSet::ReadRange r{
+      {begin.data(), int(begin.size())}, {end.data(), int(end.size())}, 0};
+
+  bench.run("worst case for radix tree", [&]() {
+    for (int i = 0; i < 256; ++i) {
+      result = weaselab::ConflictSet::TooOld;
+      cs[i]->check(&r, &result, 1);
+      if (result != weaselab::ConflictSet::Commit) {
+        abort();
+      }
+    }
+  });
+
+  // for (int i = 0; i < 256; ++i) {
+  //   bench.run("worst case for radix tree, span " + std::to_string(i), [&]() {
+  //     result = weaselab::ConflictSet::TooOld;
+  //     cs[i]->check(&r, &result, 1);
+  //     if (result != weaselab::ConflictSet::Commit) {
+  //       abort();
+  //     }
+  //   });
+  // }
+}
+
+int main(void) {
+  benchConflictSet();
+  benchWorstCaseForRadixRangeRead();
+}

CMakeLists.txt

@@ -1,13 +1,17 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.0.2
+  VERSION 0.0.8
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"
   LANGUAGES C CXX)
 set(CMAKE_CXX_STANDARD 20)
 
+file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/version.txt.in
+               ${CMAKE_CURRENT_SOURCE_DIR}/paper/version.txt)
+
 include(CMakePushCheckState)
 include(CheckCXXCompilerFlag)
 include(CheckIncludeFileCXX)
@@ -45,7 +49,8 @@ if(HAS_FULL_RELRO)
 endif()
 cmake_pop_check_state()
 
-set(version_script_flags LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+set(version_script_flags
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 cmake_push_check_state()
 list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${version_script_flags})
 check_cxx_source_compiles("int main(){}" HAS_VERSION_SCRIPT FAIL_REGEX
@@ -57,7 +62,7 @@ option(USE_SIMD_FALLBACK
 
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
-include_directories(SYSTEM ${CMAKE_SOURCE_DIR}/third_party/valgrind)
+include_directories(SYSTEM ${CMAKE_CURRENT_SOURCE_DIR}/third_party/valgrind)
 
 add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
 
@@ -104,19 +109,20 @@ add_library(${PROJECT_NAME}-object OBJECT ConflictSet.cpp)
 target_compile_options(${PROJECT_NAME}-object PRIVATE -fno-exceptions
                                                       -fvisibility=hidden)
 target_include_directories(${PROJECT_NAME}-object
-                           PRIVATE ${CMAKE_SOURCE_DIR}/include)
+                           PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
 
 add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 set_target_properties(
   ${PROJECT_NAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                             "${CMAKE_BINARY_DIR}/radix_tree")
+                             "${CMAKE_CURRENT_BINARY_DIR}/radix_tree")
 if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
 if(HAS_VERSION_SCRIPT)
-  target_link_options(${PROJECT_NAME} PRIVATE
-                      LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+  target_link_options(
+    ${PROJECT_NAME} PRIVATE
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
 add_library(${PROJECT_NAME}-static STATIC
@@ -128,8 +134,8 @@ endif()
 if(APPLE)
   add_custom_command(
     TARGET ${PROJECT_NAME}-static
-    PRE_BUILD
-    COMMAND ${CMAKE_SOURCE_DIR}/privatize_symbols_macos.sh
+    PRE_LINK
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
             $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 else()
   add_custom_command(
@@ -137,21 +143,22 @@ else()
     POST_BUILD
     COMMAND
       ${CMAKE_OBJCOPY}
-      --keep-global-symbols=${CMAKE_SOURCE_DIR}/symbol-exports.txt
+      --keep-global-symbols=${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt
       $<TARGET_FILE:${PROJECT_NAME}-static> || echo
       "Proceeding with all symbols global in static library")
 endif()
 
-set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
-
 include(CTest)
 
-if(BUILD_TESTING)
+# disable tests if this is being used through e.g. FetchContent
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
+
+  set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
   # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
   # coverage with a small number of tests.
 
-  file(GLOB CORPUS_TESTS ${CMAKE_SOURCE_DIR}/corpus/*)
+  file(GLOB CORPUS_TESTS ${CMAKE_CURRENT_SOURCE_DIR}/corpus/*)
 
   # extra testing that relies on shared libraries, which aren't available with
   # wasm
@@ -160,9 +167,11 @@ if(BUILD_TESTING)
     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")
+    target_include_directories(skip_list
+                               PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    set_target_properties(
+      skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                           "${CMAKE_CURRENT_BINARY_DIR}/skip_list")
     set_target_properties(skip_list PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
     set_target_properties(
       skip_list PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
@@ -173,10 +182,11 @@ if(BUILD_TESTING)
     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)
+    target_include_directories(hash_table
+                               PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
     set_target_properties(
       hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                            "${CMAKE_BINARY_DIR}/hash_table")
+                            "${CMAKE_CURRENT_BINARY_DIR}/hash_table")
     set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
     set_target_properties(
       hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
@@ -259,14 +269,26 @@ if(BUILD_TESTING)
 
   # 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()
+  if(NOT CMAKE_CROSSCOMPILING)
+    find_package(Python3 REQUIRED COMPONENTS Interpreter)
+    set_property(
+      DIRECTORY
+      APPEND
+      PROPERTY CMAKE_CONFIGURE_DEPENDS
+               ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py)
+    execute_process(
+      COMMAND ${Python3_EXECUTABLE}
+              ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py list
+      OUTPUT_VARIABLE SCRIPT_TESTS)
+    foreach(TEST ${SCRIPT_TESTS})
+      add_test(
+        NAME script_test_${TEST}
+        COMMAND
+          ${Python3_EXECUTABLE}
+          ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
+          --build-dir ${CMAKE_CURRENT_BINARY_DIR})
+    endforeach()
+  endif()
 
   find_program(VALGRIND_EXE valgrind)
   if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
@@ -298,25 +320,26 @@ if(BUILD_TESTING)
   # 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)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-exports.txt)
+      set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-imports.txt)
     else()
-      set(symbol_exports ${CMAKE_SOURCE_DIR}/symbol-exports.txt)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt)
       if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/aarch64-symbol-imports.txt)
+        set(symbol_imports
+            ${CMAKE_CURRENT_SOURCE_DIR}/aarch64-symbol-imports.txt)
       else()
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/symbol-imports.txt)
+        set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-imports.txt)
       endif()
     endif()
     add_test(
       NAME conflict_set_shared_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh $<TARGET_FILE:${PROJECT_NAME}>
-        ${symbol_exports} ${symbol_imports})
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
+        $<TARGET_FILE:${PROJECT_NAME}> ${symbol_exports} ${symbol_imports})
     add_test(
       NAME conflict_set_static_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
         $<TARGET_FILE:${PROJECT_NAME}-static> ${symbol_exports}
         ${symbol_imports})
   endif()
@@ -333,6 +356,7 @@ 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")
@@ -353,6 +377,20 @@ else()
   set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6 (>= 2.14)")
 endif()
 
+# macos
+set(CMAKE_OSX_DEPLOYMENT_TARGET 11.0)
+if(APPLE)
+  find_program(PANDOC_EXE pandoc)
+  if(PANDOC_EXE)
+    execute_process(COMMAND ${PANDOC_EXE} ${CMAKE_CURRENT_SOURCE_DIR}/README.md
+                            -o ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
+    set(CPACK_RESOURCE_FILE_README ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
+  endif()
+  configure_file(${CMAKE_CURRENT_SOURCE_DIR}/LICENSE
+                 ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt COPYONLY)
+  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt)
+endif()
+
 include(CPack)
 
 include(GNUInstallDirs)
@@ -377,8 +415,11 @@ install(
   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 ${PROJECT_NAME}Config
         DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/${PROJECT_NAME}/cmake)
+
+cpack_add_component(all)

Internal.h

@@ -99,8 +99,7 @@ __attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
   mallocBytesDelta -= s;
 #if SHOW_MEMORY
   mallocBytes -= s;
-  free(p);
-#else
+#endif
 #ifndef NDEBUG
   (char *&)p -= kMallocHeaderSize;
   size_t expected;
@@ -108,7 +107,6 @@ __attribute__((always_inline)) inline void safe_free(void *p, size_t s) {
   assert(s == expected);
 #endif
   free(p);
-#endif
 }
 
 // ==================== BEGIN ARENA IMPL ====================
@@ -257,10 +255,65 @@ 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));
@@ -414,13 +467,15 @@ inline uint32_t Arbitrary::bounded(uint32_t s) {
 // ==================== END ARBITRARY IMPL ====================
 
 struct ReferenceImpl {
-  explicit ReferenceImpl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
+  explicit ReferenceImpl(int64_t oldestVersion)
+      : oldestVersion(oldestVersion), newestVersion(oldestVersion) {
     writeVersionMap[""] = oldestVersion;
   }
   void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
              int count) const {
     for (int i = 0; i < count; ++i) {
-      if (reads[i].readVersion < oldestVersion) {
+      if (reads[i].readVersion < oldestVersion ||
+          reads[i].readVersion < newestVersion - 2e9) {
         results[i] = ConflictSet::TooOld;
         continue;
       }
@@ -442,6 +497,8 @@ struct ReferenceImpl {
   }
   void addWrites(const ConflictSet::WriteRange *writes, int count,
                  int64_t writeVersion) {
+    assert(writeVersion >= newestVersion);
+    newestVersion = writeVersion;
     for (int i = 0; i < count; ++i) {
       auto begin =
           std::string((const char *)writes[i].begin.p, writes[i].begin.len);
@@ -461,11 +518,12 @@ struct ReferenceImpl {
   }
 
   void setOldestVersion(int64_t oldestVersion) {
-    assert(oldestVersion >= oldestVersion);
+    assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
   }
 
   int64_t oldestVersion;
+  int64_t newestVersion;
   std::map<std::string, int64_t> writeVersionMap;
 };
 
@@ -525,18 +583,22 @@ template <class ConflictSetImpl> struct TestDriver {
   explicit TestDriver(const uint8_t *data, size_t size)
       : arbitrary({data, size}) {}
 
-  int64_t writeVersion = 0;
-  int64_t oldestVersion = 0;
+  int64_t oldestVersion = arbitrary.next();
+  int64_t writeVersion = oldestVersion;
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
 
-  constexpr static auto kMaxKeyLen = 8;
+  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() {
+    assert(cs.getBytes() >= 0);
     if (!arbitrary.hasEntropy()) {
       return true;
     }
@@ -544,7 +606,8 @@ template <class ConflictSetImpl> struct TestDriver {
     {
       int numPointWrites = arbitrary.bounded(100);
       int numRangeWrites = arbitrary.bounded(100);
-      int64_t v = ++writeVersion;
+      int64_t v = (writeVersion += arbitrary.bounded(10) ? arbitrary.bounded(10)
+                                                         : arbitrary.next());
       auto *writes =
           new (arena) ConflictSet::WriteRange[numPointWrites + numRangeWrites];
       auto keys = set<std::string_view>(arena);
@@ -552,9 +615,10 @@ 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));
       }
 
@@ -585,33 +649,79 @@ template <class ConflictSetImpl> struct TestDriver {
         }
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (writes[i].end.len == 0) {
-          fprintf(stderr, "Write: {%s} -> %" PRId64 "\n",
-                  printable(writes[i].begin).c_str(), writeVersion);
+          fprintf(stderr, "Write: {%s}\n", printable(writes[i].begin).c_str());
         } else {
-          fprintf(stderr, "Write: [%s, %s) -> %" PRId64 "\n",
+          fprintf(stderr, "Write: [%s, %s)\n",
                   printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str(), writeVersion);
+                  printable(writes[i].end).c_str());
         }
 #endif
       }
       assert(iter == keys.end());
       assert(i == numPointWrites + numRangeWrites);
 
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+      fprintf(stderr, "Write @ %" PRId64 "\n", v);
+#endif
+
+      // Test non-canonical writes
+      if (numPointWrites > 0) {
+        int overlaps = arbitrary.bounded(numPointWrites);
+        for (int i = 0; i < numPointWrites + numRangeWrites && overlaps > 0;
+             ++i) {
+          if (writes[i].end.len == 0) {
+            int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
+            auto *begin = new (arena) uint8_t[keyLen];
+            memset(begin, prefixByte, prefixLen);
+            arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
+            writes[i].end.len = keyLen;
+            writes[i].end.p = begin;
+            auto c =
+                std::span<const uint8_t>(writes[i].begin.p,
+                                         writes[i].begin.len) <=>
+                std::span<const uint8_t>(writes[i].end.p, writes[i].end.len);
+            if (c > 0) {
+              using std::swap;
+              swap(writes[i].begin, writes[i].end);
+            } else if (c == 0) {
+              // It's a point write after all, I guess
+              writes[i].end.len = 0;
+            }
+            --overlaps;
+          }
+        }
+      }
+      if (arbitrary.bounded(2)) {
+        // Shuffle writes
+        for (int i = numPointWrites + numRangeWrites - 1; i > 0; --i) {
+          int j = arbitrary.bounded(i + 1);
+          if (i != j) {
+            using std::swap;
+            swap(writes[i], writes[j]);
+          }
+        }
+      }
+
       CALLGRIND_START_INSTRUMENTATION;
       cs.addWrites(writes, numPointWrites + numRangeWrites, v);
       CALLGRIND_STOP_INSTRUMENTATION;
 
       refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
 
-      oldestVersion = std::max<int64_t>(writeVersion - arbitrary.bounded(10),
-                                        oldestVersion);
+      oldestVersion +=
+          arbitrary.bounded(10) ? arbitrary.bounded(10) : arbitrary.next();
+      oldestVersion = std::min(oldestVersion, writeVersion);
       cs.setOldestVersion(oldestVersion);
       refImpl.setOldestVersion(oldestVersion);
     }
     {
       int numPointReads = arbitrary.bounded(100);
       int numRangeReads = arbitrary.bounded(100);
-      int64_t v = std::max<int64_t>(writeVersion - arbitrary.bounded(10), 0);
+
+      int64_t v = std::max<int64_t>(writeVersion - (arbitrary.bounded(10)
+                                                        ? arbitrary.bounded(10)
+                                                        : arbitrary.next()),
+                                    0);
       auto *reads =
           new (arena) ConflictSet::ReadRange[numPointReads + numRangeReads];
       auto keys = set<std::string_view>(arena);
@@ -619,9 +729,10 @@ 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));
       }
 
@@ -652,12 +763,12 @@ template <class ConflictSetImpl> struct TestDriver {
         reads[i].readVersion = v;
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (reads[i].end.len == 0) {
-          fprintf(stderr, "Read: {%s} @ %d\n",
-                  printable(reads[i].begin).c_str(), int(reads[i].readVersion));
+          fprintf(stderr, "Read: {%s} @ %" PRId64 "\n",
+                  printable(reads[i].begin).c_str(), reads[i].readVersion);
         } else {
-          fprintf(stderr, "Read: [%s, %s) @ %d\n",
+          fprintf(stderr, "Read: [%s, %s) @ %" PRId64 "\n",
                   printable(reads[i].begin).c_str(),
-                  printable(reads[i].end).c_str(), int(reads[i].readVersion));
+                  printable(reads[i].end).c_str(), reads[i].readVersion);
         }
 #endif
       }

Jenkinsfile

@@ -110,7 +110,11 @@ pipeline {
                 sh '''
                 gcovr -f ConflictSet.cpp --cobertura > build/coverage.xml
                 '''
-                cobertura autoUpdateHealth: false, autoUpdateStability: false, coberturaReportFile: 'build/coverage.xml', conditionalCoverageTargets: '70, 0, 0', failUnhealthy: false, failUnstable: false, lineCoverageTargets: '80, 0, 0', maxNumberOfBuilds: 0, methodCoverageTargets: '80, 0, 0', onlyStable: false, sourceEncoding: 'ASCII', zoomCoverageChart: false
+                recordCoverage qualityGates: [[criticality: 'NOTE', metric: 'MODULE']], tools: [[parser: 'COBERTURA', pattern: 'build/coverage.xml']]
+                sh '''
+                # Suppress again, because we haven't dealt with function multi-versioning for x86 yet
+                # gcovr -f ConflictSet.cpp --fail-under-line 100 > /dev/null
+                '''
             }
         }
     }

README.md

@@ -60,25 +60,27 @@ Performance counters:
 
 |               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`
+|              245.99 |        4,065,232.81 |    0.3% |      0.01 | `point reads`
+|              265.93 |        3,760,430.49 |    0.2% |      0.01 | `prefix reads`
+|              485.30 |        2,060,569.50 |    0.2% |      0.01 | `range reads`
+|              449.60 |        2,224,195.17 |    0.4% |      0.01 | `point writes`
+|              441.76 |        2,263,688.18 |    1.1% |      0.01 | `prefix writes`
+|              245.42 |        4,074,647.54 |    2.4% |      0.02 | `range writes`
+|              572.80 |        1,745,810.06 |    1.3% |      0.01 | `monotonic increasing point writes`
+|          154,819.33 |            6,459.14 |    0.9% |      0.01 | `worst case for radix tree`
 
 ## Radix tree (this implementation)
 
 |               ns/op |                op/s |    err% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------:|:----------
-|               19.42 |       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`
+|               19.17 |       52,163,930.66 |    0.1% |      0.01 | `point reads`
+|               23.68 |       42,224,388.21 |    0.7% |      0.01 | `prefix reads`
+|               63.30 |       15,797,506.06 |    0.9% |      0.01 | `range reads`
+|               29.66 |       33,720,994.74 |    0.3% |      0.01 | `point writes`
+|               43.50 |       22,987,781.25 |    1.0% |      0.01 | `prefix writes`
+|               50.00 |       20,000,000.00 |    0.8% |      0.01 | `range writes`
+|              103.25 |        9,684,786.47 |    2.9% |      0.01 | `monotonic increasing point writes`
+|        1,181,500.00 |              846.38 |    2.3% |      0.01 | `worst case for radix tree`
 
 # "Real data" test
 

ScriptTest.cpp

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

SkipList.cpp

@@ -16,10 +16,14 @@
  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  * See the License for the specific language governing permissions and
  * limitations under the License.
+ *
+ * This source code is modified to compile outside of FoundationDB
  */
 
 #include "ConflictSet.h"
 #include "Internal.h"
+
+#include <algorithm>
 #include <span>
 
 std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
@@ -50,6 +54,135 @@ struct KeyRangeRef {
       : begin(begin), end(keyAfter(arena, begin)) {}
 };
 
+struct KeyInfo {
+  StringRef key;
+  bool begin;
+  bool write;
+
+  KeyInfo() = default;
+  KeyInfo(StringRef key, bool begin, bool write)
+      : key(key), begin(begin), write(write) {}
+};
+
+force_inline int extra_ordering(const KeyInfo &ki) {
+  return ki.begin * 2 + (ki.write ^ ki.begin);
+}
+
+// returns true if done with string
+force_inline bool getCharacter(const KeyInfo &ki, int character,
+                               int &outputCharacter) {
+  // normal case
+  if (character < ki.key.size()) {
+    outputCharacter = 5 + ki.key.begin()[character];
+    return false;
+  }
+
+  // termination
+  if (character == ki.key.size()) {
+    outputCharacter = 0;
+    return false;
+  }
+
+  if (character == ki.key.size() + 1) {
+    // end/begin+read/write relative sorting
+    outputCharacter = extra_ordering(ki);
+    return false;
+  }
+
+  outputCharacter = 0;
+  return true;
+}
+
+bool operator<(const KeyInfo &lhs, const KeyInfo &rhs) {
+  int i = std::min(lhs.key.size(), rhs.key.size());
+  int c = memcmp(lhs.key.data(), rhs.key.data(), i);
+  if (c != 0)
+    return c < 0;
+
+  // Always sort shorter keys before longer keys.
+  if (lhs.key.size() < rhs.key.size()) {
+    return true;
+  }
+  if (lhs.key.size() > rhs.key.size()) {
+    return false;
+  }
+
+  // When the keys are the same length, use the extra ordering constraint.
+  return extra_ordering(lhs) < extra_ordering(rhs);
+}
+
+bool operator==(const KeyInfo &lhs, const KeyInfo &rhs) {
+  return !(lhs < rhs || rhs < lhs);
+}
+
+void swapSort(std::vector<KeyInfo> &points, int a, int b) {
+  if (points[b] < points[a]) {
+    KeyInfo temp;
+    temp = points[a];
+    points[a] = points[b];
+    points[b] = temp;
+  }
+}
+
+struct SortTask {
+  int begin;
+  int size;
+  int character;
+  SortTask(int begin, int size, int character)
+      : begin(begin), size(size), character(character) {}
+};
+
+void sortPoints(std::vector<KeyInfo> &points) {
+  std::vector<SortTask> tasks;
+  std::vector<KeyInfo> newPoints;
+  std::vector<int> counts;
+
+  tasks.emplace_back(0, points.size(), 0);
+
+  while (tasks.size()) {
+    SortTask st = tasks.back();
+    tasks.pop_back();
+
+    if (st.size < 10) {
+      std::sort(points.begin() + st.begin, points.begin() + st.begin + st.size);
+      continue;
+    }
+
+    newPoints.resize(st.size);
+    counts.assign(256 + 5, 0);
+
+    // get counts
+    int c;
+    bool allDone = true;
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      allDone &= getCharacter(points[i], st.character, c);
+      counts[c]++;
+    }
+    if (allDone)
+      continue;
+
+    // calculate offsets from counts and build next level of tasks
+    int total = 0;
+    for (int i = 0; i < counts.size(); i++) {
+      int temp = counts[i];
+      if (temp > 1)
+        tasks.emplace_back(st.begin + total, temp, st.character + 1);
+      counts[i] = total;
+      total += temp;
+    }
+
+    // put in their places
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      getCharacter(points[i], st.character, c);
+      newPoints[counts[c]++] = points[i];
+    }
+
+    // copy back into original points array
+    for (int i = 0; i < st.size; i++)
+      points[st.begin + i] = newPoints[i];
+  }
+}
+
 static thread_local uint32_t g_seed = 0;
 
 static inline int skfastrand() {
@@ -268,13 +401,21 @@ public:
   }
 
   explicit SkipList(Version version = 0) {
+#if DEBUG_VERBOSE
+    fprintf(stderr, "skip_list: create\n");
+#endif
     header = Node::create(StringRef(), MaxLevels - 1);
     for (int l = 0; l < MaxLevels; l++) {
       header->setNext(l, nullptr);
       header->setMaxVersion(l, version);
     }
   }
-  ~SkipList() { destroy(); }
+  ~SkipList() {
+#if DEBUG_VERBOSE
+    fprintf(stderr, "skip_list: destroy\n");
+#endif
+    destroy();
+  }
   SkipList(SkipList &&other) noexcept : header(other.header) {
     other.header = nullptr;
   }
@@ -567,7 +708,8 @@ struct SkipListConflictSet {};
 
 struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   Impl(int64_t oldestVersion)
-      : oldestVersion(oldestVersion), skipList(oldestVersion) {}
+      : oldestVersion(oldestVersion), newestVersion(oldestVersion),
+        skipList(oldestVersion) {}
   void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
              int count) const {
     Arena arena;
@@ -582,7 +724,8 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     }
     skipList.detectConflicts(ranges, count, results);
     for (int i = 0; i < count; ++i) {
-      if (reads[i].readVersion < oldestVersion) {
+      if (reads[i].readVersion < oldestVersion ||
+          reads[i].readVersion < newestVersion - 2e9) {
         results[i] = TooOld;
       }
     }
@@ -590,8 +733,40 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
 
   void addWrites(const ConflictSet::WriteRange *writes, int count,
                  int64_t writeVersion) {
+    auto points = std::vector<KeyInfo>(count * 2);
     Arena arena;
-    const int stringCount = count * 2;
+
+    for (int r = 0; r < count; r++) {
+      points.emplace_back(StringRef(writes[r].begin.p, writes[r].begin.len),
+                          true, true);
+      points.emplace_back(
+          writes[r].end.len > 0
+              ? StringRef{writes[r].end.p, size_t(writes[r].end.len)}
+              : keyAfter(arena, points.back().key),
+          false, true);
+    }
+
+    sortPoints(points);
+
+    int activeWriteCount = 0;
+    std::vector<std::pair<StringRef, StringRef>> combinedWriteConflictRanges;
+    for (const KeyInfo &point : points) {
+      if (point.write) {
+        if (point.begin) {
+          activeWriteCount++;
+          if (activeWriteCount == 1)
+            combinedWriteConflictRanges.emplace_back(point.key, StringRef());
+        } else /*if (point.end)*/ {
+          activeWriteCount--;
+          if (activeWriteCount == 0)
+            combinedWriteConflictRanges.back().second = point.key;
+        }
+      }
+    }
+
+    assert(writeVersion >= newestVersion);
+    newestVersion = writeVersion;
+    const int stringCount = combinedWriteConflictRanges.size() * 2;
 
     const int stripeSize = 16;
     SkipList::Finger fingers[stripeSize];
@@ -602,11 +777,13 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     int ss = stringCount - (stripes - 1) * stripeSize;
     for (int s = stripes - 1; s >= 0; s--) {
       for (int i = 0; i * 2 < ss; ++i) {
-        const auto &w = writes[s * stripeSize / 2 + i];
-        values[i * 2] = {w.begin.p, size_t(w.begin.len)};
-        values[i * 2 + 1] = w.end.len > 0
-                                ? StringRef{w.end.p, size_t(w.end.len)}
-                                : keyAfter(arena, values[i * 2]);
+        const auto &w = combinedWriteConflictRanges[s * stripeSize / 2 + i];
+#if DEBUG_VERBOSE
+        printf("Write begin: %s\n", printable(w.begin).c_str());
+        fflush(stdout);
+#endif
+        values[i * 2] = w.first;
+        values[i * 2 + 1] = w.second;
         keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
@@ -616,6 +793,7 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
   }
 
   void setOldestVersion(int64_t oldestVersion) {
+    assert(oldestVersion >= this->oldestVersion);
     this->oldestVersion = oldestVersion;
     SkipList::Finger finger;
     int temp;
@@ -634,16 +812,20 @@ private:
   Arena removalArena;
   std::span<const uint8_t> removalKey;
   int64_t oldestVersion;
+  int64_t newestVersion;
   SkipList skipList;
 };
 
-void ConflictSet::check(const ReadRange *reads, Result *results,
-                        int count) const {
+// Internal entry points. Public entry points should just delegate to these
+
+void internal_check(ConflictSet::Impl *impl,
+                    const ConflictSet::ReadRange *reads,
+                    ConflictSet::Result *results, int count) {
   impl->check(reads, results, count);
 }
-
-void ConflictSet::addWrites(const WriteRange *writes, int count,
-                            int64_t writeVersion) {
+void internal_addWrites(ConflictSet::Impl *impl,
+                        const ConflictSet::WriteRange *writes, int count,
+                        int64_t writeVersion) {
   mallocBytesDelta = 0;
   impl->addWrites(writes, count, writeVersion);
   impl->totalBytes += mallocBytesDelta;
@@ -654,7 +836,7 @@ void ConflictSet::addWrites(const WriteRange *writes, int count,
 #endif
 }
 
-void ConflictSet::setOldestVersion(int64_t oldestVersion) {
+void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
   mallocBytesDelta = 0;
   impl->setOldestVersion(oldestVersion);
   impl->totalBytes += mallocBytesDelta;
@@ -664,19 +846,43 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
   }
 #endif
 }
+ConflictSet::Impl *internal_create(int64_t oldestVersion) {
+  mallocBytesDelta = 0;
+  auto *result = new (safe_malloc(sizeof(ConflictSet::Impl)))
+      ConflictSet::Impl{oldestVersion};
+  result->totalBytes += mallocBytesDelta;
+  return result;
+}
 
-int64_t ConflictSet::getBytes() const { return impl->totalBytes; }
+void internal_destroy(ConflictSet::Impl *impl) {
+  impl->~Impl();
+  safe_free(impl, sizeof(ConflictSet::Impl));
+}
+
+int64_t internal_getBytes(ConflictSet::Impl *impl) { return impl->totalBytes; }
+
+void ConflictSet::check(const ReadRange *reads, Result *results,
+                        int count) const {
+  internal_check(impl, reads, results, count);
+}
+
+void ConflictSet::addWrites(const WriteRange *writes, int count,
+                            int64_t writeVersion) {
+  internal_addWrites(impl, writes, count, writeVersion);
+}
+
+void ConflictSet::setOldestVersion(int64_t oldestVersion) {
+  internal_setOldestVersion(impl, oldestVersion);
+}
+
+int64_t ConflictSet::getBytes() const { return internal_getBytes(impl); }
 
 ConflictSet::ConflictSet(int64_t oldestVersion)
-    : impl((mallocBytesDelta = 0,
-            new(safe_malloc(sizeof(Impl))) Impl{oldestVersion})) {
-  impl->totalBytes += mallocBytesDelta;
-}
+    : impl(internal_create(oldestVersion)) {}
 
 ConflictSet::~ConflictSet() {
   if (impl) {
-    impl->~Impl();
-    safe_free(impl, sizeof(Impl));
+    internal_destroy(impl);
   }
 }
 
@@ -697,37 +903,34 @@ 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();
-  safe_free(cs, sizeof(Impl));
+  internal_destroy((ConflictSet::Impl *)cs);
 }
 __attribute__((__visibility__("default"))) int64_t
 ConflictSet_getBytes(void *cs) {
-  using Impl = ConflictSet::Impl;
-  return ((Impl *)cs)->totalBytes;
+  return internal_getBytes((ConflictSet::Impl *)cs);
 }
 }
 
 #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));
   }

apple-symbol-imports.txt

@@ -1,7 +1,10 @@
+___stack_chk_fail
+___stack_chk_guard
 __tlv_bootstrap
 _abort
 _bzero
 _free
 _malloc
 _memcpy
-_memmove
+_memmove
+dyld_stub_binder

conflict_set.py

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