Update corpus

To match new "interleaving" test setup

Bench.cpp

@@ -358,7 +358,14 @@ void benchWorstCaseForRadixRangeRead() {
   // }
 }
 
+void benchCreateAndDestroy() {
+  ankerl::nanobench::Bench bench;
+
+  bench.run("create and destroy", [&]() { ConflictSet cs{0}; });
+}
+
 int main(void) {
   benchConflictSet();
   benchWorstCaseForRadixRangeRead();
+  benchCreateAndDestroy();
 }

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.0.7
+  VERSION 0.0.10
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"
@@ -60,6 +60,8 @@ cmake_pop_check_state()
 option(USE_SIMD_FALLBACK
        "Use fallback implementations of functions that use SIMD" OFF)
 
+option(DISABLE_TSAN "Disable TSAN" OFF)
+
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
 include_directories(SYSTEM ${CMAKE_CURRENT_SOURCE_DIR}/third_party/valgrind)
@@ -245,7 +247,7 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
   endforeach()
 
   # tsan tests
-  if(NOT CMAKE_CROSSCOMPILING AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
+  if(NOT CMAKE_CROSSCOMPILING AND NOT DISABLE_TSAN)
     add_executable(tsan_driver ConflictSet.cpp FuzzTestDriver.cpp)
     target_compile_options(tsan_driver PRIVATE ${TEST_FLAGS} -fsanitize=thread)
     target_link_options(tsan_driver PRIVATE -fsanitize=thread)
@@ -351,6 +353,11 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
   add_executable(real_data_bench RealDataBench.cpp)
   target_link_libraries(real_data_bench PRIVATE ${PROJECT_NAME})
   set_target_properties(real_data_bench PROPERTIES SKIP_BUILD_RPATH ON)
+
+  # fuzzer-based perf
+  add_executable(driver_perf TestDriver.cpp)
+  target_compile_definitions(driver_perf PRIVATE PERF_TEST=1)
+  target_link_libraries(driver_perf PRIVATE ${PROJECT_NAME})
 endif()
 
 # packaging

HashTable.cpp

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

Internal.h

@@ -578,12 +578,16 @@ inline const char *resultToStr(ConflictSet::Result r) {
 
 namespace {
 
-template <class ConflictSetImpl> struct TestDriver {
+template <class ConflictSetImpl, bool kEnableAssertions = true>
-  Arbitrary arbitrary;
+struct TestDriver {
-  explicit TestDriver(const uint8_t *data, size_t size)
+  Arbitrary *arbitrary;
-      : arbitrary({data, size}) {}
+  explicit TestDriver(Arbitrary &a) : arbitrary(&a) {
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+    fprintf(stderr, "%p Initial version: {%" PRId64 "}\n", this, writeVersion);
+#endif
+  }
 
-  int64_t oldestVersion = arbitrary.next();
+  int64_t oldestVersion = arbitrary->next();
   int64_t writeVersion = oldestVersion;
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
@@ -592,33 +596,34 @@ template <class ConflictSetImpl> struct TestDriver {
 
   bool ok = true;
 
-  const int prefixLen = arbitrary.bounded(512);
+  const int prefixLen = arbitrary->bounded(512);
-  const int prefixByte = arbitrary.randT<uint8_t>();
+  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()) {
+    if (!arbitrary->hasEntropy()) {
       return true;
     }
     Arena arena;
     {
-      int numPointWrites = arbitrary.bounded(100);
+      int numPointWrites = arbitrary->bounded(100);
-      int numRangeWrites = arbitrary.bounded(100);
+      int numRangeWrites = arbitrary->bounded(100);
-      int64_t v = (writeVersion += arbitrary.bounded(10) ? arbitrary.bounded(10)
+      int64_t v =
-                                                         : arbitrary.next());
+          (writeVersion +=
+           arbitrary->bounded(10) ? arbitrary->bounded(10) : arbitrary->next());
       auto *writes =
           new (arena) ConflictSet::WriteRange[numPointWrites + numRangeWrites];
       auto keys = set<std::string_view>(arena);
       while (int(keys.size()) < numPointWrites + numRangeWrites * 2) {
-        if (!arbitrary.hasEntropy()) {
+        if (!arbitrary->hasEntropy()) {
           return true;
         }
-        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
+        int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
         memset(begin, prefixByte, prefixLen);
-        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
+        arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -628,7 +633,7 @@ template <class ConflictSetImpl> struct TestDriver {
                rangesRemaining = numRangeWrites;
            pointsRemaining > 0 || rangesRemaining > 0; ++i) {
         bool pointRead = pointsRemaining > 0 && rangesRemaining > 0
-                             ? bool(arbitrary.bounded(2))
+                             ? bool(arbitrary->bounded(2))
                              : pointsRemaining > 0;
         if (pointRead) {
           assert(pointsRemaining > 0);
@@ -647,54 +652,120 @@ template <class ConflictSetImpl> struct TestDriver {
           ++iter;
           --rangesRemaining;
         }
-#if DEBUG_VERBOSE && !defined(NDEBUG)
-        if (writes[i].end.len == 0) {
-          fprintf(stderr, "Write: {%s}\n", printable(writes[i].begin).c_str());
-        } else {
-          fprintf(stderr, "Write: [%s, %s)\n",
-                  printable(writes[i].begin).c_str(),
-                  printable(writes[i].end).c_str());
-        }
-#endif
       }
       assert(iter == keys.end());
       assert(i == numPointWrites + numRangeWrites);
 
+      // Test non-canonical writes
+      if (numPointWrites > 0) {
+        int overlaps = arbitrary->bounded(numPointWrites);
+        for (int i = 0; i < numPointWrites + numRangeWrites && overlaps > 0;
+             ++i) {
+          if (writes[i].end.len == 0) {
+            int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
+            auto *begin = new (arena) uint8_t[keyLen];
+            memset(begin, prefixByte, prefixLen);
+            arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
+            writes[i].end.len = keyLen;
+            writes[i].end.p = begin;
+            auto c =
+                std::span<const uint8_t>(writes[i].begin.p,
+                                         writes[i].begin.len) <=>
+                std::span<const uint8_t>(writes[i].end.p, writes[i].end.len);
+            if (c > 0) {
+              using std::swap;
+              swap(writes[i].begin, writes[i].end);
+            } else if (c == 0) {
+              // It's a point write after all, I guess
+              writes[i].end.len = 0;
+            }
+            --overlaps;
+          }
+        }
+      }
+      if (arbitrary->bounded(2)) {
+        // Shuffle writes
+        for (int i = numPointWrites + numRangeWrites - 1; i > 0; --i) {
+          int j = arbitrary->bounded(i + 1);
+          if (i != j) {
+            using std::swap;
+            swap(writes[i], writes[j]);
+          }
+        }
+      }
+
+      oldestVersion +=
+          arbitrary->bounded(10) ? arbitrary->bounded(10) : arbitrary->next();
+      oldestVersion = std::min(oldestVersion, writeVersion);
+
+#ifdef THREAD_TEST
+      std::latch ready{1};
+      std::thread thread2{[&]() {
+        ready.count_down();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
+      }};
+      ready.wait();
+#endif
+
 #if DEBUG_VERBOSE && !defined(NDEBUG)
-      fprintf(stderr, "Write @ %" PRId64 "\n", v);
+      for (int i = 0; i < numPointWrites + numRangeWrites; ++i) {
+        if (writes[i].end.len == 0) {
+          fprintf(stderr, "%p Write: {%s}\n", this,
+                  printable(writes[i].begin).c_str());
+        } else {
+          fprintf(stderr, "%p Write: [%s, %s)\n", this,
+                  printable(writes[i].begin).c_str(),
+                  printable(writes[i].end).c_str());
+        }
+      }
+      fprintf(stderr, "%p Write @ %" PRId64 "\n", this, v);
 #endif
 
       CALLGRIND_START_INSTRUMENTATION;
       cs.addWrites(writes, numPointWrites + numRangeWrites, v);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
+      if constexpr (kEnableAssertions) {
+        refImpl.addWrites(writes, numPointWrites + numRangeWrites, v);
+      }
 
-      oldestVersion +=
+#if DEBUG_VERBOSE && !defined(NDEBUG)
-          arbitrary.bounded(10) ? arbitrary.bounded(10) : arbitrary.next();
+      fprintf(stderr, "%p Set oldest version: %" PRId64 "\n", this,
-      oldestVersion = std::min(oldestVersion, writeVersion);
+              oldestVersion);
+#endif
       cs.setOldestVersion(oldestVersion);
-      refImpl.setOldestVersion(oldestVersion);
+      if constexpr (kEnableAssertions) {
+        refImpl.setOldestVersion(oldestVersion);
+      }
+
+#ifdef THREAD_TEST
+      thread2.join();
+#endif
     }
     {
-      int numPointReads = arbitrary.bounded(100);
+      int numPointReads = arbitrary->bounded(100);
-      int numRangeReads = arbitrary.bounded(100);
+      int numRangeReads = arbitrary->bounded(100);
 
-      int64_t v = std::max<int64_t>(writeVersion - (arbitrary.bounded(10)
+      int64_t v = std::max<int64_t>(writeVersion - (arbitrary->bounded(10)
-                                                        ? arbitrary.bounded(10)
+                                                        ? arbitrary->bounded(10)
-                                                        : arbitrary.next()),
+                                                        : arbitrary->next()),
                                     0);
       auto *reads =
           new (arena) ConflictSet::ReadRange[numPointReads + numRangeReads];
       auto keys = set<std::string_view>(arena);
       while (int(keys.size()) < numPointReads + numRangeReads * 2) {
-        if (!arbitrary.hasEntropy()) {
+        if (!arbitrary->hasEntropy()) {
           return true;
         }
-        int keyLen = prefixLen + arbitrary.bounded(kMaxKeySuffixLen);
+        int keyLen = prefixLen + arbitrary->bounded(kMaxKeySuffixLen);
         auto *begin = new (arena) uint8_t[keyLen];
         memset(begin, prefixByte, prefixLen);
-        arbitrary.randomBytes(begin + prefixLen, keyLen - prefixLen);
+        arbitrary->randomBytes(begin + prefixLen, keyLen - prefixLen);
         keys.insert(std::string_view((const char *)begin, keyLen));
       }
 
@@ -703,7 +774,7 @@ template <class ConflictSetImpl> struct TestDriver {
       for (int pointsRemaining = numPointReads, rangesRemaining = numRangeReads;
            pointsRemaining > 0 || rangesRemaining > 0; ++i) {
         bool pointRead = pointsRemaining > 0 && rangesRemaining > 0
-                             ? bool(arbitrary.bounded(2))
+                             ? bool(arbitrary->bounded(2))
                              : pointsRemaining > 0;
         if (pointRead) {
           assert(pointsRemaining > 0);
@@ -725,10 +796,10 @@ template <class ConflictSetImpl> struct TestDriver {
         reads[i].readVersion = v;
 #if DEBUG_VERBOSE && !defined(NDEBUG)
         if (reads[i].end.len == 0) {
-          fprintf(stderr, "Read: {%s} @ %" PRId64 "\n",
+          fprintf(stderr, "%p Read: {%s} @ %" PRId64 "\n", this,
                   printable(reads[i].begin).c_str(), reads[i].readVersion);
         } else {
-          fprintf(stderr, "Read: [%s, %s) @ %" PRId64 "\n",
+          fprintf(stderr, "%p Read: [%s, %s) @ %" PRId64 "\n", this,
                   printable(reads[i].begin).c_str(),
                   printable(reads[i].end).c_str(), reads[i].readVersion);
         }
@@ -747,7 +818,15 @@ template <class ConflictSetImpl> struct TestDriver {
       std::latch ready{1};
       std::thread thread2{[&]() {
         ready.count_down();
+        // Call all const methods
         cs.check(reads, results3, numPointReads + numRangeReads);
+        cs.getBytes();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
       }};
       ready.wait();
 #endif
@@ -756,26 +835,40 @@ template <class ConflictSetImpl> struct TestDriver {
       cs.check(reads, results1, numPointReads + numRangeReads);
       CALLGRIND_STOP_INSTRUMENTATION;
 
-      refImpl.check(reads, results2, numPointReads + numRangeReads);
+      if constexpr (kEnableAssertions) {
+        // Call remaining const methods
+        cs.getBytes();
+        ConflictSet::MetricsV1 *m;
+        int count;
+        cs.getMetricsV1(&m, &count);
+        for (int i = 0; i < count; ++i) {
+          m[i].getValue();
+        }
 
-      auto compareResults = [reads](ConflictSet::Result *results1,
+        refImpl.check(reads, results2, numPointReads + numRangeReads);
-                                    ConflictSet::Result *results2, int count) {
+      }
+
+      auto compareResults = [reads, this](ConflictSet::Result *results1,
+                                          ConflictSet::Result *results2,
+                                          int count) {
         for (int i = 0; i < count; ++i) {
           if (results1[i] != results2[i]) {
             if (reads[i].end.len == 0) {
-              fprintf(stderr,
-                      "Expected %s, got %s for read of {%s} at version %" PRId64
-                      "\n",
-                      resultToStr(results2[i]), resultToStr(results1[i]),
-                      printable(reads[i].begin).c_str(), reads[i].readVersion);
-            } else {
               fprintf(
                   stderr,
-                  "Expected %s, got %s for read of [%s, %s) at version %" PRId64
+                  "%p Expected %s, got %s for read of {%s} at version %" PRId64
                   "\n",
-                  resultToStr(results2[i]), resultToStr(results1[i]),
+                  (void *)this, resultToStr(results2[i]),
-                  printable(reads[i].begin).c_str(),
+                  resultToStr(results1[i]), printable(reads[i].begin).c_str(),
-                  printable(reads[i].end).c_str(), reads[i].readVersion);
+                  reads[i].readVersion);
+            } else {
+              fprintf(stderr,
+                      "%p Expected %s, got %s for read of [%s, %s) at version "
+                      "%" PRId64 "\n",
+                      (void *)this, resultToStr(results2[i]),
+                      resultToStr(results1[i]),
+                      printable(reads[i].begin).c_str(),
+                      printable(reads[i].end).c_str(), reads[i].readVersion);
             }
             return false;
           }
@@ -783,9 +876,12 @@ template <class ConflictSetImpl> struct TestDriver {
         return true;
       };
 
-      if (!compareResults(results1, results2, numPointReads + numRangeReads)) {
+      if constexpr (kEnableAssertions) {
-        ok = false;
+        if (!compareResults(results1, results2,
-        return true;
+                            numPointReads + numRangeReads)) {
+          ok = false;
+          return true;
+        }
       }
 
 #ifdef THREAD_TEST

Jenkinsfile

@@ -48,6 +48,17 @@ pipeline {
                 recordIssues(tools: [clang()])
             }
         }
+        stage('Debug') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DCMAKE_BUILD_TYPE=Debug")
+            }
+        }
         stage('SIMD fallback') {
             agent {
                 dockerfile {
@@ -59,17 +70,6 @@ pipeline {
                 CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
             }
         }
-        stage('32-bit versions') {
-            agent {
-                dockerfile {
-                    args '-v /home/jenkins/ccache:/ccache'
-                    reuseNode true
-                }
-            }
-            steps {
-                CleanBuildAndTest("-DUSE_32_BIT_VERSIONS=ON")
-            }
-        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {
@@ -117,7 +117,7 @@ pipeline {
                 }
             }
             steps {
-                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug")
+                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug -DDISABLE_TSAN=ON")
                 sh '''
                 gcovr -f ConflictSet.cpp --cobertura > build/coverage.xml
                 '''

README.md

@@ -60,27 +60,27 @@ Performance counters:
 
 |               ns/op |                op/s |    err% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------:|:----------
-|              256.89 |        3,892,784.92 |    0.3% |      0.01 | `point reads`
+|              245.99 |        4,065,232.81 |    0.3% |      0.01 | `point reads`
-|              272.90 |        3,664,395.04 |    0.2% |      0.01 | `prefix reads`
+|              265.93 |        3,760,430.49 |    0.2% |      0.01 | `prefix reads`
-|              507.22 |        1,971,549.50 |    0.7% |      0.01 | `range reads`
+|              485.30 |        2,060,569.50 |    0.2% |      0.01 | `range reads`
-|              452.66 |        2,209,181.91 |    0.5% |      0.01 | `point writes`
+|              449.60 |        2,224,195.17 |    0.4% |      0.01 | `point writes`
-|              438.09 |        2,282,619.96 |    0.4% |      0.01 | `prefix writes`
+|              441.76 |        2,263,688.18 |    1.1% |      0.01 | `prefix writes`
-|              253.33 |        3,947,420.36 |    2.5% |      0.02 | `range writes`
+|              245.42 |        4,074,647.54 |    2.4% |      0.02 | `range writes`
-|              574.07 |        1,741,936.71 |    0.3% |      0.01 | `monotonic increasing point writes`
+|              572.80 |        1,745,810.06 |    1.3% |      0.01 | `monotonic increasing point writes`
-|          151,562.50 |            6,597.94 |    1.5% |      0.01 | `worst case for radix tree`
+|          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
 |--------------------:|--------------------:|--------:|----------:|:----------
-|               19.83 |       50,420,955.28 |    0.1% |      0.01 | `point reads`
+|               19.17 |       52,163,930.66 |    0.1% |      0.01 | `point reads`
-|               55.95 |       17,872,542.40 |    0.5% |      0.01 | `prefix reads`
+|               23.68 |       42,224,388.21 |    0.7% |      0.01 | `prefix reads`
-|               88.28 |       11,327,709.50 |    0.4% |      0.01 | `range reads`
+|               63.30 |       15,797,506.06 |    0.9% |      0.01 | `range reads`
-|               29.15 |       34,309,531.64 |    0.5% |      0.01 | `point writes`
+|               29.66 |       33,720,994.74 |    0.3% |      0.01 | `point writes`
-|               42.36 |       23,607,424.27 |    1.1% |      0.01 | `prefix writes`
+|               43.50 |       22,987,781.25 |    1.0% |      0.01 | `prefix writes`
-|               50.00 |       20,000,000.00 |    0.0% |      0.01 | `range writes`
+|               50.00 |       20,000,000.00 |    0.8% |      0.01 | `range writes`
-|               93.52 |       10,692,413.79 |    3.3% |      0.01 | `monotonic increasing point writes`
+|              103.25 |        9,684,786.47 |    2.9% |      0.01 | `monotonic increasing point writes`
-|        2,388,417.00 |              418.69 |    0.4% |      0.03 | `worst case for radix tree`
+|        1,181,500.00 |              846.38 |    2.3% |      0.01 | `worst case for radix tree`
 
 # "Real data" test
 

RealDataBench.cpp

@@ -129,6 +129,16 @@ int main(int argc, const char **argv) {
     close(fd);
   }
 
+  ConflictSet::MetricsV1 *metrics;
+  int metricsCount;
+  cs.getMetricsV1(&metrics, &metricsCount);
+  for (int i = 0; i < metricsCount; ++i) {
+    printf("# HELP %s %s\n", metrics[i].name, metrics[i].help);
+    printf("# TYPE %s %s\n", metrics[i].name,
+           metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
+    printf("%s %g\n", metrics[i].name, metrics[i].getValue());
+  }
+
   printf("Check: %g seconds, %g MB/s, Add: %g seconds, %g MB/s, Gc ratio: "
          "%g%%, Peak idle memory: %g\n",
          checkTime, checkBytes / checkTime * 1e-6, addTime,

SkipList.cpp

@@ -22,6 +22,8 @@
 
 #include "ConflictSet.h"
 #include "Internal.h"
+
+#include <algorithm>
 #include <span>
 
 std::span<const uint8_t> keyAfter(Arena &arena, std::span<const uint8_t> key) {
@@ -40,7 +42,7 @@ std::span<const uint8_t> copyToArena(Arena &arena,
 }
 
 using Version = int64_t;
-#define force_inline __attribute__((always_inline))
+#define force_inline inline __attribute__((always_inline))
 using StringRef = std::span<const uint8_t>;
 
 struct KeyRangeRef {
@@ -52,6 +54,135 @@ struct KeyRangeRef {
       : begin(begin), end(keyAfter(arena, begin)) {}
 };
 
+struct KeyInfo {
+  StringRef key;
+  bool begin;
+  bool write;
+
+  KeyInfo() = default;
+  KeyInfo(StringRef key, bool begin, bool write)
+      : key(key), begin(begin), write(write) {}
+};
+
+force_inline int extra_ordering(const KeyInfo &ki) {
+  return ki.begin * 2 + (ki.write ^ ki.begin);
+}
+
+// returns true if done with string
+force_inline bool getCharacter(const KeyInfo &ki, int character,
+                               int &outputCharacter) {
+  // normal case
+  if (character < ki.key.size()) {
+    outputCharacter = 5 + ki.key.begin()[character];
+    return false;
+  }
+
+  // termination
+  if (character == ki.key.size()) {
+    outputCharacter = 0;
+    return false;
+  }
+
+  if (character == ki.key.size() + 1) {
+    // end/begin+read/write relative sorting
+    outputCharacter = extra_ordering(ki);
+    return false;
+  }
+
+  outputCharacter = 0;
+  return true;
+}
+
+bool operator<(const KeyInfo &lhs, const KeyInfo &rhs) {
+  int i = std::min(lhs.key.size(), rhs.key.size());
+  int c = memcmp(lhs.key.data(), rhs.key.data(), i);
+  if (c != 0)
+    return c < 0;
+
+  // Always sort shorter keys before longer keys.
+  if (lhs.key.size() < rhs.key.size()) {
+    return true;
+  }
+  if (lhs.key.size() > rhs.key.size()) {
+    return false;
+  }
+
+  // When the keys are the same length, use the extra ordering constraint.
+  return extra_ordering(lhs) < extra_ordering(rhs);
+}
+
+bool operator==(const KeyInfo &lhs, const KeyInfo &rhs) {
+  return !(lhs < rhs || rhs < lhs);
+}
+
+void swapSort(std::vector<KeyInfo> &points, int a, int b) {
+  if (points[b] < points[a]) {
+    KeyInfo temp;
+    temp = points[a];
+    points[a] = points[b];
+    points[b] = temp;
+  }
+}
+
+struct SortTask {
+  int begin;
+  int size;
+  int character;
+  SortTask(int begin, int size, int character)
+      : begin(begin), size(size), character(character) {}
+};
+
+void sortPoints(std::vector<KeyInfo> &points) {
+  std::vector<SortTask> tasks;
+  std::vector<KeyInfo> newPoints;
+  std::vector<int> counts;
+
+  tasks.emplace_back(0, points.size(), 0);
+
+  while (tasks.size()) {
+    SortTask st = tasks.back();
+    tasks.pop_back();
+
+    if (st.size < 10) {
+      std::sort(points.begin() + st.begin, points.begin() + st.begin + st.size);
+      continue;
+    }
+
+    newPoints.resize(st.size);
+    counts.assign(256 + 5, 0);
+
+    // get counts
+    int c;
+    bool allDone = true;
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      allDone &= getCharacter(points[i], st.character, c);
+      counts[c]++;
+    }
+    if (allDone)
+      continue;
+
+    // calculate offsets from counts and build next level of tasks
+    int total = 0;
+    for (int i = 0; i < counts.size(); i++) {
+      int temp = counts[i];
+      if (temp > 1)
+        tasks.emplace_back(st.begin + total, temp, st.character + 1);
+      counts[i] = total;
+      total += temp;
+    }
+
+    // put in their places
+    for (int i = st.begin; i < st.begin + st.size; i++) {
+      getCharacter(points[i], st.character, c);
+      newPoints[counts[c]++] = points[i];
+    }
+
+    // copy back into original points array
+    for (int i = 0; i < st.size; i++)
+      points[st.begin + i] = newPoints[i];
+  }
+}
+
 static thread_local uint32_t g_seed = 0;
 
 static inline int skfastrand() {
@@ -602,10 +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;
+
+    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;
-    Arena arena;
+    const int stringCount = combinedWriteConflictRanges.size() * 2;
-    const int stringCount = count * 2;
 
     const int stripeSize = 16;
     SkipList::Finger fingers[stripeSize];
@@ -616,15 +777,9 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
     int ss = stringCount - (stripes - 1) * stripeSize;
     for (int s = stripes - 1; s >= 0; s--) {
       for (int i = 0; i * 2 < ss; ++i) {
-        const auto &w = writes[s * stripeSize / 2 + i];
+        const auto &w = combinedWriteConflictRanges[s * stripeSize / 2 + i];
-#if DEBUG_VERBOSE
+        values[i * 2] = w.first;
-        printf("Write begin: %s\n", printable(w.begin).c_str());
+        values[i * 2 + 1] = w.second;
-        fflush(stdout);
-#endif
-        values[i * 2] = {w.begin.p, size_t(w.begin.len)};
-        values[i * 2 + 1] = w.end.len > 0
-                                ? StringRef{w.end.p, size_t(w.end.len)}
-                                : keyAfter(arena, values[i * 2]);
         keyUpdates += 3;
       }
       skipList.find(values, fingers, temp, ss);
@@ -702,6 +857,16 @@ void internal_destroy(ConflictSet::Impl *impl) {
 
 int64_t internal_getBytes(ConflictSet::Impl *impl) { return impl->totalBytes; }
 
+void internal_getMetricsV1(ConflictSet::Impl *impl,
+                           ConflictSet::MetricsV1 **metrics, int *count) {
+  *metrics = nullptr;
+  *count = 0;
+}
+
+double internal_getMetricValue(const ConflictSet::MetricsV1 *metric) {
+  return 0;
+}
+
 void ConflictSet::check(const ReadRange *reads, Result *results,
                         int count) const {
   internal_check(impl, reads, results, count);
@@ -718,6 +883,14 @@ void ConflictSet::setOldestVersion(int64_t oldestVersion) {
 
 int64_t ConflictSet::getBytes() const { return internal_getBytes(impl); }
 
+void ConflictSet::getMetricsV1(MetricsV1 **metrics, int *count) const {
+  return internal_getMetricsV1(impl, metrics, count);
+}
+
+double ConflictSet::MetricsV1::getValue() const {
+  return internal_getMetricValue(this);
+}
+
 ConflictSet::ConflictSet(int64_t oldestVersion)
     : impl(internal_create(oldestVersion)) {}
 

TestDriver.cpp

@@ -3,17 +3,68 @@
 #include <fstream>
 #include <sstream>
 
+#ifndef PERF_TEST
+#define PERF_TEST 0
+#endif
+
 int main(int argc, char **argv) {
   for (int i = 1; i < argc; ++i) {
     std::ifstream t(argv[i], std::ios::binary);
     std::stringstream buffer;
     buffer << t.rdbuf();
     auto str = buffer.str();
-    TestDriver<ConflictSet> driver{(const uint8_t *)str.data(), str.size()};
+    Arbitrary arbitrary({(const uint8_t *)str.data(), str.size()});
-    while (!driver.next())
+    TestDriver<ConflictSet, !PERF_TEST> driver1{arbitrary};
-      ;
+    TestDriver<ConflictSet, !PERF_TEST> driver2{arbitrary};
-    if (!driver.ok) {
+    bool done1 = false;
-      abort();
+    bool done2 = false;
+    for (;;) {
+      if (!done1) {
+        done1 = driver1.next();
+        if (!driver1.ok) {
+          abort();
+        }
+      }
+      if (!done2) {
+        done2 = driver2.next();
+        if (!driver2.ok) {
+          abort();
+        }
+      }
+      if (done1 && done2) {
+        break;
+      }
+    }
+
+    {
+      ConflictSet::MetricsV1 *metrics;
+      int metricsCount;
+      driver1.cs.getMetricsV1(&metrics, &metricsCount);
+      printf("#################### METRICS for ConflictSet 1 for %s "
+             "####################\n",
+             argv[i]);
+      for (int i = 0; i < metricsCount; ++i) {
+        printf("# HELP %s %s\n", metrics[i].name, metrics[i].help);
+        printf("# TYPE %s %s\n", metrics[i].name,
+               metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
+        printf("%s %g\n", metrics[i].name, metrics[i].getValue());
+      }
+      puts("");
+    }
+    {
+      ConflictSet::MetricsV1 *metrics;
+      int metricsCount;
+      driver2.cs.getMetricsV1(&metrics, &metricsCount);
+      printf("#################### METRICS for ConflictSet 2 for %s "
+             "####################\n",
+             argv[i]);
+      for (int i = 0; i < metricsCount; ++i) {
+        printf("# HELP %s %s\n", metrics[i].name, metrics[i].help);
+        printf("# TYPE %s %s\n", metrics[i].name,
+               metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
+        printf("%s %g\n", metrics[i].name, metrics[i].getValue());
+      }
+      puts("");
     }
   }
 }

aarch64-symbol-imports.txt

@@ -1,3 +1,5 @@
+__aarch64_cas8_relax
+__getauxval@GLIBC_2.17
 __stack_chk_fail@GLIBC_2.17
 __stack_chk_guard@GLIBC_2.17
 abort@GLIBC_2.17

apple-symbol-exports.txt

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

conflict_set_cxx_api_test.cpp

@@ -1,6 +1,7 @@
 #include "ConflictSet.h"
 
 #include <cassert>
+#include <cstdio>
 
 using namespace weaselab;
 
@@ -21,4 +22,14 @@ int main(void) {
   assert(result == ConflictSet::Conflict);
   int64_t bytes = cs.getBytes();
   assert(bytes > 0);
+
+  ConflictSet::MetricsV1 *metrics;
+  int metricsCount;
+  cs.getMetricsV1(&metrics, &metricsCount);
+  for (int i = 0; i < metricsCount; ++i) {
+    printf("# HELP %s %s\n", metrics[i].name, metrics[i].help);
+    printf("# TYPE %s %s\n", metrics[i].name,
+           metrics[i].type == metrics[i].Counter ? "counter" : "gauge");
+    printf("%s %g\n", metrics[i].name, metrics[i].getValue());
+  }
 }

corpus/0e410d4488defd877d2c9690a26970f57ab78b0c

@@ -1,4 +0,0 @@
-*
-
-
-

corpus/22c23c2418773edaaff055ed387f82c8ad078f9f

@@ -1,2 +0,0 @@
-*
-