conflict-set/Internal.h

#pragma once

#include "ConflictSet.h"

#include <bit>
#include <cassert>
#include <cstdint>
#include <cstdlib>
#include <cstring>
#include <inttypes.h>
#include <map>
#include <set>
#include <span>
#include <string>
#include <utility>
#include <vector>

#define DEBUG_VERBOSE 0

// This header contains code that we want to reuse outside of ConflictSet.cpp or
// want to exclude from coverage since it's only testing related.

// GCOVR_EXCL_START

__attribute__((always_inline)) inline void *safe_malloc(size_t s) {
  if (void *p = malloc(s)) {
    return p;
  }
  abort();
}

// ==================== BEGIN ARENA IMPL ====================

/// Group allocations with similar lifetimes to amortize the cost of malloc/free
struct Arena {
  explicit Arena(int initialSize = 0);
  /// O(log n) in the number of allocations
  ~Arena();
  struct ArenaImpl;
  Arena(const Arena &) = delete;
  Arena &operator=(const Arena &) = delete;
  Arena(Arena &&other) noexcept;
  Arena &operator=(Arena &&other) noexcept;

  ArenaImpl *impl = nullptr;
};

[[maybe_unused]] inline void operator delete(void *, std::align_val_t,
                                             Arena &) {}
inline void *operator new(size_t size, std::align_val_t align, Arena &arena);
void *operator new(size_t size, std::align_val_t align, Arena *arena) = delete;

[[maybe_unused]] inline void operator delete(void *, Arena &) {}
inline void *operator new(size_t size, Arena &arena) {
  return operator new(size, std::align_val_t(alignof(std::max_align_t)), arena);
}
inline void *operator new(size_t size, Arena *arena) = delete;

[[maybe_unused]] inline void operator delete[](void *, Arena &) {}
inline void *operator new[](size_t size, Arena &arena) {
  return operator new(size, arena);
}
inline void *operator new[](size_t size, Arena *arena) = delete;

[[maybe_unused]] inline void operator delete[](void *, std::align_val_t,
                                               Arena &) {}
inline void *operator new[](size_t size, std::align_val_t align, Arena &arena) {
  return operator new(size, align, arena);
}
inline void *operator new[](size_t size, std::align_val_t align,
                            Arena *arena) = delete;

/// align must be a power of two
template <class T> T *align_up(T *t, size_t align) {
  auto unaligned = uintptr_t(t);
  auto aligned = (unaligned + align - 1) & ~(align - 1);
  return reinterpret_cast<T *>(reinterpret_cast<char *>(t) + aligned -
                               unaligned);
}

/// align must be a power of two
constexpr inline int align_up(uint32_t unaligned, uint32_t align) {
  return (unaligned + align - 1) & ~(align - 1);
}

/// Returns the smallest power of two >= x
[[maybe_unused]] constexpr inline uint32_t nextPowerOfTwo(uint32_t x) {
  return x <= 1 ? 1 : 1 << (32 - std::countl_zero(x - 1));
}

struct Arena::ArenaImpl {
  Arena::ArenaImpl *prev;
  int capacity;
  int used;
  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);
    impl = (Arena::ArenaImpl *)safe_malloc(allocationSize);
    impl->prev = nullptr;
    impl->capacity = allocationSize - sizeof(ArenaImpl);
    impl->used = 0;
  }
}

inline void onDestroy(Arena::ArenaImpl *impl) {
  while (impl) {
    auto *prev = impl->prev;
    free(impl);
    impl = prev;
  }
}

[[maybe_unused]] inline Arena::Arena(Arena &&other) noexcept
    : impl(std::exchange(other.impl, nullptr)) {}
[[maybe_unused]] inline Arena &Arena::operator=(Arena &&other) noexcept {
  onDestroy(impl);
  impl = std::exchange(other.impl, nullptr);
  return *this;
}

inline Arena::~Arena() { onDestroy(impl); }

inline void *operator new(size_t size, std::align_val_t align, Arena &arena) {
  int64_t aligned_size = size + size_t(align) - 1;
  if (arena.impl == nullptr ||
      (arena.impl->capacity - arena.impl->used) < aligned_size) {
    auto allocationSize = align_up(
        sizeof(Arena::ArenaImpl) +
            std::max<int>(aligned_size,
                          (arena.impl ? std::max<int>(sizeof(Arena::ArenaImpl),
                                                      arena.impl->capacity * 2)
                                      : 0)),
        16);
    auto *impl = (Arena::ArenaImpl *)safe_malloc(allocationSize);
    impl->prev = arena.impl;
    impl->capacity = allocationSize - sizeof(Arena::ArenaImpl);
    impl->used = 0;
    arena.impl = impl;
  }
  auto *result =
      align_up(arena.impl->begin() + arena.impl->used, size_t(align));
  auto usedDelta = (result - arena.impl->begin()) + size - arena.impl->used;
  arena.impl->used += usedDelta;
  return result;
}

/// STL-friendly allocator using an arena
template <class T> struct ArenaAlloc {
  typedef T value_type;

  ArenaAlloc() = delete;
  explicit ArenaAlloc(Arena *arena) : arena(arena) {}

  Arena *arena;

  template <class U> constexpr ArenaAlloc(const ArenaAlloc<U> &other) noexcept {
    arena = other.arena;
  }

  [[nodiscard]] T *allocate(size_t n) {
    if (n > 0xfffffffffffffffful / sizeof(T)) { // NOLINT
      __builtin_unreachable();
    }

    return static_cast<T *>((void *)new (std::align_val_t(alignof(T)), *arena)
                                uint8_t[n * sizeof(T)]); // NOLINT
  }

  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> using Set = std::set<T, std::less<T>, ArenaAlloc<T>>;
template <class T> auto set(Arena &arena) {
  return Set<T>(ArenaAlloc<T>(&arena));
}

template <class T, class U>
bool operator==(const ArenaAlloc<T> &lhs, const ArenaAlloc<U> &rhs) {
  return lhs.arena == rhs.arena;
}

template <class T, class U>
bool operator!=(const ArenaAlloc<T> &lhs, const ArenaAlloc<U> &rhs) {
  return !(lhs == rhs);
}

// ==================== END ARENA IMPL ====================

// ==================== BEGIN ARBITRARY IMPL ====================

/// Think of `Arbitrary` as an attacker-controlled random number generator.
/// Usually you want your random number generator to be fair, so that you can
/// sensibly analyze probabilities. E.g. The analysis that shows that quicksort
/// is expected O(n log n) with a random pivot relies on the random pivot being
/// selected uniformly from a fair distribution.
///
/// Other times you want your randomness to be diabolically unfair, like when
/// looking for bugs and fuzzing. The random-number-like interface is still
/// convenient here, but you can potentially get much better coverage by
/// allowing the possibility of e.g. flipping heads 100 times in a row.
///
/// When it runs out of entropy, it always returns 0.
struct Arbitrary {
  Arbitrary() = default;

  explicit Arbitrary(std::span<const uint8_t> bytecode) : bytecode(bytecode) {}

  /// Draws an arbitrary uint32_t
  uint32_t next() { return consume<4>(); }

  /// Draws an arbitrary element from [0, s)
  uint32_t bounded(uint32_t s);

  /// Fill `bytes` with `size` arbitrary bytes
  void randomBytes(uint8_t *bytes, int size) {
    int toFill = std::min<int>(size, bytecode.size());
    if (toFill > 0) {
      memcpy(bytes, bytecode.data(), toFill);
    }
    bytecode = bytecode.subspan(toFill, bytecode.size() - toFill);
    memset(bytes + toFill, 0, size - toFill);
  }

  /// Fill `bytes` with `size` random hex bytes
  void randomHex(uint8_t *bytes, int size) {
    for (int i = 0; i < size;) {
      uint8_t arbitrary = consume<1>();
      bytes[i++] = "0123456789abcdef"[arbitrary & 0xf];
      arbitrary >>= 4;
      if (i < size) {
        bytes[i++] = "0123456789abcdef"[arbitrary & 0xf];
      }
    }
  }

  template <class T, class = std::enable_if_t<std::is_trivially_copyable_v<T>>>
  T randT() {
    T t;
    randomBytes((uint8_t *)&t, sizeof(T));
    return t;
  }

  bool hasEntropy() const { return bytecode.size() != 0; }

private:
  uint8_t consumeByte() {
    if (bytecode.size() == 0) {
      return 0;
    }
    auto result = bytecode[0];
    bytecode = bytecode.subspan(1, bytecode.size() - 1);
    return result;
  }

  template <int kBytes> uint32_t consume() {
    uint32_t result = 0;
    static_assert(kBytes <= 4);
    for (int i = 0; i < kBytes; ++i) {
      result <<= 8;
      result |= consumeByte();
    }
    return result;
  }

  std::span<const uint8_t> bytecode;
};

inline uint32_t Arbitrary::bounded(uint32_t s) {
  if (s == 1) {
    return 0;
  }
  switch (32 - std::countl_zero(s - 1)) {
  case 1:
  case 2:
  case 3:
  case 4:
  case 5:
  case 6:
  case 7:
  case 8:
    return consume<1>() % s;
  case 9:
  case 10:
  case 11:
  case 12:
  case 13:
  case 14:
  case 15:
  case 16:
    return consume<2>() % s;
  case 17:
  case 18:
  case 19:
  case 20:
  case 21:
  case 22:
  case 23:
  case 24:
    return consume<3>() % s;
  default:
    return consume<4>() % 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;
  }
  void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
             int count) const {
    for (int i = 0; i < count; ++i) {
      if (reads[i].readVersion < oldestVersion) {
        results[i] = ConflictSet::TooOld;
        continue;
      }
      auto begin =
          std::string((const char *)reads[i].begin.p, reads[i].begin.len);
      auto end =
          reads[i].end.len == 0
              ? begin + std::string("\x00", 1)
              : std::string((const char *)reads[i].end.p, reads[i].end.len);
      int64_t maxVersion = oldestVersion;
      for (auto iter = --writeVersionMap.upper_bound(begin),
                endIter = writeVersionMap.lower_bound(end);
           iter != endIter; ++iter) {
        maxVersion = std::max(maxVersion, iter->second);
      }
      results[i] = maxVersion > reads[i].readVersion ? ConflictSet::Conflict
                                                     : ConflictSet::Commit;
    }
  }
  void addWrites(const ConflictSet::WriteRange *writes, int count) {
    for (int i = 0; i < count; ++i) {
      auto begin =
          std::string((const char *)writes[i].begin.p, writes[i].begin.len);
      auto end =
          writes[i].end.len == 0
              ? begin + std::string("\x00", 1)
              : std::string((const char *)writes[i].end.p, writes[i].end.len);
      auto writeVersion = writes[i].writeVersion;
      auto prevVersion = (--writeVersionMap.upper_bound(end))->second;
      for (auto iter = writeVersionMap.lower_bound(begin),
                endIter = writeVersionMap.lower_bound(end);
           iter != endIter;) {
        iter = writeVersionMap.erase(iter);
      }
      writeVersionMap[begin] = writeVersion;
      writeVersionMap[end] = prevVersion;
    }
  }

  void setOldestVersion(int64_t oldestVersion) {
    this->oldestVersion = oldestVersion;
  }

  void printLogical(std::string &result) {
    for (const auto &[k, v] : writeVersionMap) {
      std::string key;
      for (uint8_t c : k) {
        key += "x";
        key += "0123456789abcdef"[c / 16];
        key += "0123456789abcdef"[c % 16];
      }
      result += key + " -> " + std::to_string(v) + "\n";
    }
  }

  int64_t oldestVersion;
  std::map<std::string, int64_t> writeVersionMap;
};

using Key = ConflictSet::Key;

[[maybe_unused]] static Key toKey(Arena &arena, int n) {
  uint8_t *buf = new (arena) uint8_t[sizeof(n)];
  memcpy(buf, &n, sizeof(n));
  return Key{buf, sizeof(n)};
}

[[maybe_unused]] static Key toKeyAfter(Arena &arena, int n) {
  uint8_t *buf = new (arena) uint8_t[sizeof(n) + 1];
  memcpy(buf, &n, sizeof(n));
  buf[sizeof(n)] = 0;
  return Key{buf, sizeof(n) + 1};
}

inline std::string printable(std::string_view key) {
  std::string result;
  for (uint8_t c : key) {
    result += "x";
    result += "0123456789abcdef"[c / 16];
    result += "0123456789abcdef"[c % 16];
  }
  return result;
}

inline std::string printable(const Key &key) {
  return printable(std::string_view((const char *)key.p, key.len));
}

namespace {

template <class ConflictSetImpl> struct TestDriver {
  // TODO call setOldestVersion, and check range writes/reads
  Arbitrary arbitrary;
  explicit TestDriver(const uint8_t *data, size_t size)
      : arbitrary({data, size}) {}

  int64_t writeVersion = 0;
  ConflictSetImpl cs{writeVersion};
  ReferenceImpl refImpl{writeVersion};

  constexpr static auto kMaxKeyLen = 24;

  bool ok = true;

  static const char *resultToStr(ConflictSet::Result r) {
    switch (r) {
    case ConflictSet::Commit:
      return "commit";
    case ConflictSet::Conflict:
      return "conflict";
    case ConflictSet::TooOld:
      return "too old";
    }
    abort();
  }

  // Call until it returns true, for "done". Check internal invariants etc
  // between calls to next.
  bool next() {
    if (!arbitrary.hasEntropy()) {
      return true;
    }
    Arena arena;
    {
      int numWriteKeys = arbitrary.bounded(10);
      int64_t v = ++writeVersion;
      auto *writes = new (arena) ConflictSet::WriteRange[numWriteKeys];
      auto keys = set<std::string_view>(arena);
      while (int(keys.size()) < numWriteKeys) {
        if (!arbitrary.hasEntropy()) {
          return true;
        }
        int keyLen = arbitrary.bounded(kMaxKeyLen);
        auto *begin = new (arena) uint8_t[keyLen];
        arbitrary.randomBytes(begin, keyLen);
        keys.insert(std::string_view((const char *)begin, keyLen));
      }
      auto iter = keys.begin();
      int numWrites = 0;
      for (int i = 0; i < numWriteKeys; ++i, ++numWrites) {
        writes[numWrites].begin.p = (const uint8_t *)iter->data();
        writes[numWrites].begin.len = iter->size();
        ++iter;
        if (i + 1 < numWriteKeys && arbitrary.bounded(2)) {
          ++i;
          writes[numWrites].end.p = (const uint8_t *)iter->data();
          writes[numWrites].end.len = iter->size();
          ++iter;
        } else {
          writes[numWrites].end.len = 0;
        }
        writes[numWrites].writeVersion = v;
#if DEBUG_VERBOSE && !defined(NDEBUG)
        if (writes[numWrites].end.len == 0) {
          fprintf(stderr, "Write: {%s} -> %d\n",
                  printable(writes[numWrites].begin).c_str(),
                  int(writes[numWrites].writeVersion));
        } else {
          fprintf(stderr, "Write: [%s, %s) -> %d\n",
                  printable(writes[numWrites].begin).c_str(),
                  printable(writes[numWrites].end).c_str(),
                  int(writes[numWrites].writeVersion));
        }
#endif
      }
      assert(iter == keys.end());
      cs.addWrites(writes, numWrites);
      refImpl.addWrites(writes, numWrites);
    }
    {
      int numReads = arbitrary.bounded(10);
      int64_t v = std::max<int64_t>(writeVersion - arbitrary.bounded(10), 0);
      auto *reads = new (arena) ConflictSet::ReadRange[numReads];
      auto keys = set<std::string_view>(arena);
      while (int(keys.size()) < numReads) {
        if (!arbitrary.hasEntropy()) {
          return true;
        }
        int keyLen = arbitrary.bounded(kMaxKeyLen);
        auto *begin = new (arena) uint8_t[keyLen];
        arbitrary.randomBytes(begin, keyLen);
        keys.insert(std::string_view((const char *)begin, keyLen));
      }
      auto iter = keys.begin();
      for (int i = 0; i < numReads; ++i) {
        reads[i].begin.p = (const uint8_t *)iter->data();
        reads[i].begin.len = iter->size();
        ++iter;
        reads[i].end.len = 0;
        reads[i].readVersion = v;
#if DEBUG_VERBOSE && !defined(NDEBUG)
        fprintf(stderr, "Read: {%s} at %d\n", printable(reads[i].begin).c_str(),
                int(reads[i].readVersion));
#endif
      }
      assert(iter == keys.end());
      auto *results1 = new (arena) ConflictSet::Result[numReads];
      auto *results2 = new (arena) ConflictSet::Result[numReads];
      cs.check(reads, results1, numReads);
      refImpl.check(reads, results2, numReads);
      for (int i = 0; i < numReads; ++i) {
        if (results1[i] != results2[i]) {
          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);
          ok = false;
          return true;
        }
      }
    }
    return false;
  }
};
} // namespace

// GCOVR_EXCL_STOP