#include "ConflictSet.h"
#include "Internal.h"

#include <algorithm>
#include <bit>
#include <cassert>
#include <cstdint>
#include <cstring>
#include <inttypes.h>
#include <limits>
#include <span>
#include <string>
#include <string_view>
#include <utility>

#ifdef HAS_AVX
#include <immintrin.h>
#elif defined(HAS_ARM_NEON)
#include <arm_neon.h>
#endif

// ==================== BEGIN IMPLEMENTATION ====================

struct Entry {
  int64_t pointVersion;
  int64_t rangeVersion;
};

enum class Type : int8_t {
  Node4,
  Node16,
  Node48,
  Node256,
  Invalid,
};
struct Node {
  /* begin section that's copied to the next node */
  Node *parent = nullptr;
  int64_t maxVersion;
  Entry entry;
  int16_t numChildren = 0;
  bool entryPresent = false;
  uint8_t parentsIndex = 0;
  constexpr static auto kPartialKeyMaxLen = 10;
  uint8_t partialKey[kPartialKeyMaxLen];
  int8_t partialKeyLen = 0;
  /* end section that's copied to the next node */

  Type type = Type::Invalid;
};

struct Node4 : Node {
  // Sorted
  uint8_t index[4] = {};
  Node *children[4] = {};
  Node4() { this->type = Type::Node4; }
};

Node *newNode() { return new (safe_malloc(sizeof(Node4))) Node4; }

struct Node16 : Node {
  // Sorted
  uint8_t index[16] = {};
  Node *children[16] = {};
  Node16() { this->type = Type::Node16; }
};

struct BitSet {
  bool test(int i) const {
    assert(0 <= i);
    assert(i < 256);
    if (i < 128) {
      return (lo >> i) & 1;
    } else {
      return (hi >> (i - 128)) & 1;
    }
  }

  void set(int i) {
    assert(0 <= i);
    assert(i < 256);
    if (i < 128) {
      lo |= __uint128_t(1) << i;
    } else {
      hi |= __uint128_t(1) << (i - 128);
    }
  }

  void reset(int i) {
    assert(0 <= i);
    assert(i < 256);
    if (i < 128) {
      lo &= ~(__uint128_t(1) << i);
    } else {
      hi &= ~(__uint128_t(1) << (i - 128));
    }
  }

  int firstSetGeq(int i) const {
    if (i < 128) {
      int a = std::countr_zero(lo >> i);
      if (a < 128) {
        assert(i + a < 128);
        return i + a;
      }
      i = 128;
    }
    int b = std::countr_zero(hi >> (i - 128));
    if (b < 128) {
      assert(i + b < 256);
      return i + b;
    }
    return -1;
  }

  int lastSetLeq(int i) const {
    if (i >= 128) {
      int a = std::countl_zero(hi << (255 - i));
      if (a < 128) {
        return i - a;
      }
      i = 127;
    }
    int b = std::countl_zero(lo << (127 - i));
    if (b < 128) {
      return i - b;
    }
    return -1;
  }

private:
  __uint128_t lo = 0;
  __uint128_t hi = 0;
};

struct Node48 : Node {
  BitSet bitSet;
  Node *children[48] = {};
  int8_t nextFree = 0;
  int8_t index[256];
  Node48() {
    this->type = Type::Node48;
    memset(index, -1, 256);
  }
};

struct Node256 : Node {
  BitSet bitSet;
  Node *children[256] = {};
  Node256() { this->type = Type::Node256; }
};

int getNodeIndex(Node4 *self, uint8_t index) {
  for (int i = 0; i < self->numChildren; ++i) {
    if (self->index[i] == index) {
      return i;
    }
  }
  return -1;
}

int getNodeIndex(Node16 *self, uint8_t index) {
#ifdef HAS_AVX
  // Based on https://www.the-paper-trail.org/post/art-paper-notes/

  // key_vec is 16 repeated copies of the searched-for byte, one for every
  // possible position in child_keys that needs to be searched.
  __m128i key_vec = _mm_set1_epi8(index);

  // Compare all child_keys to 'index' in parallel. Don't worry if some of the
  // keys aren't valid, we'll mask the results to only consider the valid ones
  // below.
  __m128i indices;
  memcpy(&indices, self->index, sizeof(self->index));
  __m128i results = _mm_cmpeq_epi8(key_vec, indices);

  // Build a mask to select only the first node->num_children values from the
  // comparison (because the other values are meaningless)
  int mask = (1 << self->numChildren) - 1;

  // Change the results of the comparison into a bitfield, masking off any
  // invalid comparisons.
  int bitfield = _mm_movemask_epi8(results) & mask;

  // No match if there are no '1's in the bitfield.
  if (bitfield == 0)
    return -1;

  // Find the index of the first '1' in the bitfield by counting the leading
  // zeros.
  return std::countr_zero(bitfield);
#elif defined(HAS_ARM_NEON)
  // Based on
  // https://community.arm.com/arm-community-blogs/b/infrastructure-solutions-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon

  uint8x16_t indices;
  memcpy(&indices, self->index, sizeof(self->index));
  // 0xff for each match
  uint16x8_t results =
      vreinterpretq_u16_u8(vceqq_u8(vdupq_n_u8(index), indices));
  uint64_t mask = self->numChildren == 16
                      ? uint64_t(-1)
                      : (uint64_t(1) << (self->numChildren * 4)) - 1;
  // 0xf for each match in valid range
  uint64_t bitfield =
      vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(results, 4)), 0) & mask;
  if (bitfield == 0)
    return -1;
  return std::countr_zero(bitfield) / 4;
#else
  for (int i = 0; i < self->numChildren; ++i) {
    if (self->index[i] == index) {
      return i;
    }
  }
  return -1;
#endif
}

[[maybe_unused]] Node *getChild(Node *self, uint8_t index) {
  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    int i = getNodeIndex(self4, index);
    if (i >= 0) {
      return self4->children[i];
    }
    return nullptr;
  } else if (self->type == Type::Node16) {
    auto *self16 = static_cast<Node16 *>(self);
    int i = getNodeIndex(self16, index);
    if (i >= 0) {
      return self16->children[i];
    }
    return nullptr;
  } else if (self->type == Type::Node48) {
    auto *self48 = static_cast<Node48 *>(self);
    int secondIndex = self48->index[index];
    if (secondIndex >= 0) {
      return self48->children[secondIndex];
    }
    return nullptr;
  } else {
    auto *self256 = static_cast<Node256 *>(self);
    return self256->children[index];
  }
}

// Precondition - an entry for index must exist in the node
Node *&getChildExists(Node *self, uint8_t index) {
  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    return self4->children[getNodeIndex(self4, index)];
  } else if (self->type == Type::Node16) {
    auto *self16 = static_cast<Node16 *>(self);
    return self16->children[getNodeIndex(self16, index)];
  } else if (self->type == Type::Node48) {
    auto *self48 = static_cast<Node48 *>(self);
    int secondIndex = self48->index[index];
    if (secondIndex >= 0) {
      return self48->children[secondIndex];
    }
  } else {
    auto *self256 = static_cast<Node256 *>(self);
    return self256->children[index];
  }
  __builtin_unreachable(); // GCOVR_EXCL_LINE
}

int getChildGeq(Node *self, int child) {
  if (child > 255) {
    return -1;
  }
  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    for (int i = 0; i < self->numChildren; ++i) {
      if (i > 0) {
        assert(self4->index[i - 1] < self4->index[i]);
      }
      if (self4->index[i] >= child) {
        return self4->index[i];
      }
    }
  } else if (self->type == Type::Node16) {
    auto *self16 = static_cast<Node16 *>(self);
#ifdef HAS_AVX
    __m128i key_vec = _mm_set1_epi8(child);
    __m128i indices;
    memcpy(&indices, self16->index, sizeof(self16->index));
    __m128i results = _mm_cmpeq_epi8(key_vec, _mm_min_epu8(key_vec, indices));
    int mask = (1 << self16->numChildren) - 1;
    int bitfield = _mm_movemask_epi8(results) & mask;
    int result = bitfield == 0 ? -1 : self16->index[std::countr_zero(bitfield)];
    assert(result == [&]() -> int {
      for (int i = 0; i < self16->numChildren; ++i) {
        if (self16->index[i] >= child) {
          return self16->index[i];
        }
      }
      return -1;
    }());
    return result;
#elif defined(HAS_ARM_NEON)
    uint8x16_t indices;
    memcpy(&indices, self16->index, sizeof(self16->index));
    // 0xff for each leq
    auto results = vcleq_u8(vdupq_n_u8(child), indices);
    uint64_t mask = self->numChildren == 16
                        ? uint64_t(-1)
                        : (uint64_t(1) << (self->numChildren * 4)) - 1;
    // 0xf for each 0xff (within mask)
    uint64_t bitfield =
        vget_lane_u64(
            vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)),
            0) &
        mask;
    int simd =
        bitfield == 0 ? -1 : self16->index[std::countr_zero(bitfield) / 4];
    assert(simd == [&]() -> int {
      for (int i = 0; i < self->numChildren; ++i) {
        if (self16->index[i] >= child) {
          return self16->index[i];
        }
      }
      return -1;
    }());
    return simd;
#else
    for (int i = 0; i < self->numChildren; ++i) {
      if (i > 0) {
        assert(self16->index[i - 1] < self16->index[i]);
      }
      if (self16->index[i] >= child) {
        return self16->index[i];
      }
    }
#endif
  } else if (self->type == Type::Node48) {
    auto *self48 = static_cast<Node48 *>(self);
    return self48->bitSet.firstSetGeq(child);
  } else {
    auto *self256 = static_cast<Node256 *>(self);
    return self256->bitSet.firstSetGeq(child);
  }
  return -1;
}

int getChildLeq(Node *self, int child) {
  if (child < 0) {
    return -1;
  }
  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    for (int i = self->numChildren - 1; i >= 0; --i) {
      if (i > 0) {
        assert(self4->index[i - 1] < self4->index[i]);
      }
      if (self4->index[i] <= child) {
        return self4->index[i];
      }
    }
  } else if (self->type == Type::Node16) {
    auto *self16 = static_cast<Node16 *>(self);
#ifdef HAS_AVX
    __m128i key_vec = _mm_set1_epi8(child);
    __m128i indices;
    memcpy(&indices, self16->index, sizeof(self16->index));
    __m128i results = _mm_cmpeq_epi8(key_vec, _mm_max_epu8(key_vec, indices));
    int mask = (1 << self16->numChildren) - 1;
    int bitfield = _mm_movemask_epi8(results) & mask;
    int result =
        bitfield == 0 ? -1 : self16->index[31 - std::countl_zero(bitfield)];
    assert(result == [&]() -> int {
      for (int i = self16->numChildren - 1; i >= 0; --i) {
        if (self16->index[i] <= child) {
          return self16->index[i];
        }
      }
      return -1;
    }());
    return result;
#elif defined(HAS_ARM_NEON)
    uint8x16_t indices;
    memcpy(&indices, self16->index, sizeof(self16->index));
    // 0xff for each leq
    auto results = vcleq_u8(indices, vdupq_n_u8(child));
    uint64_t mask = self->numChildren == 16
                        ? uint64_t(-1)
                        : (uint64_t(1) << (self->numChildren * 4)) - 1;
    // 0xf for each 0xff (within mask)
    uint64_t bitfield =
        vget_lane_u64(
            vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)),
            0) &
        mask;
    int simd =
        bitfield == 0 ? -1 : self16->index[15 - std::countl_zero(bitfield) / 4];
    assert(simd == [&]() -> int {
      for (int i = self->numChildren - 1; i >= 0; --i) {
        if (self16->index[i] <= child) {
          return self16->index[i];
        }
      }
      return -1;
    }());
    return simd;
#else
    for (int i = self->numChildren - 1; i >= 0; --i) {
      if (self16->index[i] <= child) {
        return self16->index[i];
      }
    }
    return -1;
#endif
  } else if (self->type == Type::Node48) {
    auto *self48 = static_cast<Node48 *>(self);
    return self48->bitSet.lastSetLeq(child);
  } else {
    auto *self256 = static_cast<Node256 *>(self);
    return self256->bitSet.lastSetLeq(child);
  }
  return -1;
}

void setChildrenParents(Node *node) {
  for (int i = getChildGeq(node, 0); i >= 0; i = getChildGeq(node, i + 1)) {
    getChildExists(node, i)->parent = node;
  }
}

// Caller is responsible for assigning a non-null pointer to the returned
// reference if null
Node *&getOrCreateChild(Node *&self, uint8_t index) {
  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    {
      int i = getNodeIndex(self4, index);
      if (i >= 0) {
        return self4->children[i];
      }
    }
    if (self->numChildren == 4) {
      auto *newSelf = new (safe_malloc(sizeof(Node16))) Node16;
      memcpy((void *)newSelf, self, offsetof(Node, type));
      memcpy(newSelf->index, self4->index, 4);
      memcpy(newSelf->children, self4->children, 4 * sizeof(void *));
      free(std::exchange(self, newSelf));
      setChildrenParents(self);
      goto insert16;
    } else {
      ++self->numChildren;
      for (int i = 0; i < int(self->numChildren) - 1; ++i) {
        if (int(self4->index[i]) > int(index)) {
          memmove(self4->index + i + 1, self4->index + i,
                  self->numChildren - (i + 1));
          memmove(self4->children + i + 1, self4->children + i,
                  (self->numChildren - (i + 1)) * sizeof(void *));
          self4->index[i] = index;
          self4->children[i] = nullptr;
          return self4->children[i];
        }
      }
      self4->index[self->numChildren - 1] = index;
      self4->children[self->numChildren - 1] = nullptr;
      return self4->children[self->numChildren - 1];
    }
  } else if (self->type == Type::Node16) {
  insert16:
    auto *self16 = static_cast<Node16 *>(self);
    {
      int i = getNodeIndex(self16, index);
      if (i >= 0) {
        return self16->children[i];
      }
    }
    if (self->numChildren == 16) {
      auto *newSelf = new (safe_malloc(sizeof(Node48))) Node48;
      memcpy((void *)newSelf, self, offsetof(Node, type));
      newSelf->nextFree = 16;
      int i = 0;
      for (auto x : self16->index) {
        newSelf->bitSet.set(x);
        newSelf->children[i] = self16->children[i];
        newSelf->index[x] = i;
        ++i;
      }
      assert(i == 16);
      free(std::exchange(self, newSelf));
      setChildrenParents(self);
      goto insert48;
    } else {
      ++self->numChildren;
      for (int i = 0; i < int(self->numChildren) - 1; ++i) {
        if (int(self16->index[i]) > int(index)) {
          memmove(self16->index + i + 1, self16->index + i,
                  self->numChildren - (i + 1));
          memmove(self16->children + i + 1, self16->children + i,
                  (self->numChildren - (i + 1)) * sizeof(void *));
          self16->index[i] = index;
          self16->children[i] = nullptr;
          return self16->children[i];
        }
      }
      self16->index[self->numChildren - 1] = index;
      self16->children[self->numChildren - 1] = nullptr;
      return self16->children[self->numChildren - 1];
    }
  } else if (self->type == Type::Node48) {
  insert48:
    auto *self48 = static_cast<Node48 *>(self);
    int secondIndex = self48->index[index];
    if (secondIndex >= 0) {
      return self48->children[secondIndex];
    }
    if (self->numChildren == 48) {
      auto *newSelf = new (safe_malloc(sizeof(Node256))) Node256;
      memcpy((void *)newSelf, self, offsetof(Node, type));
      for (int i = 0; i < 256; ++i) {
        if (self48->index[i] >= 0) {
          newSelf->bitSet.set(i);
          newSelf->children[i] = self48->children[self48->index[i]];
        }
      }
      free(std::exchange(self, newSelf));
      self = newSelf;
      setChildrenParents(self);
      goto insert256;
    } else {
      self48->bitSet.set(index);
      ++self->numChildren;
      assert(self48->nextFree < 48);
      self48->index[index] = self48->nextFree;
      self48->children[self48->nextFree] = nullptr;
      return self48->children[self48->nextFree++];
    }
  } else {
  insert256:
    auto *self256 = static_cast<Node256 *>(self);
    if (!self256->children[index]) {
      ++self->numChildren;
    }
    self256->bitSet.set(index);
    return self256->children[index];
  }
}

// Precondition - an entry for index must exist in the node
void eraseChild(Node *self, uint8_t index) {
  free(getChildExists(self, index));

  if (self->type == Type::Node4) {
    auto *self4 = static_cast<Node4 *>(self);
    int nodeIndex = getNodeIndex(self4, index);
    memmove(self4->index + nodeIndex, self4->index + nodeIndex + 1,
            sizeof(self4->index[0]) * (self->numChildren - (nodeIndex + 1)));
    memmove(self4->children + nodeIndex, self4->children + nodeIndex + 1,
            sizeof(self4->children[0]) * // NOLINT
                (self->numChildren - (nodeIndex + 1)));
  } else if (self->type == Type::Node16) {
    auto *self16 = static_cast<Node16 *>(self);
    int nodeIndex = getNodeIndex(self16, index);
    memmove(self16->index + nodeIndex, self16->index + nodeIndex + 1,
            sizeof(self16->index[0]) * (self->numChildren - (nodeIndex + 1)));
    memmove(self16->children + nodeIndex, self16->children + nodeIndex + 1,
            sizeof(self16->children[0]) * // NOLINT
                (self->numChildren - (nodeIndex + 1)));
  } else if (self->type == Type::Node48) {
    auto *self48 = static_cast<Node48 *>(self);
    self48->bitSet.reset(index);
    int8_t toRemoveChildrenIndex = std::exchange(self48->index[index], -1);
    int8_t lastChildrenIndex = --self48->nextFree;
    assert(toRemoveChildrenIndex >= 0);
    assert(lastChildrenIndex >= 0);
    if (toRemoveChildrenIndex != lastChildrenIndex) {
      self48->children[toRemoveChildrenIndex] =
          std::exchange(self48->children[lastChildrenIndex], nullptr);
      self48->index[self48->children[toRemoveChildrenIndex]->parentsIndex] =
          toRemoveChildrenIndex;
    }
  } else {
    auto *self256 = static_cast<Node256 *>(self);
    self256->bitSet.reset(index);
    self256->children[index] = nullptr;
  }
  --self->numChildren;
  if (self->numChildren == 0 && !self->entryPresent &&
      self->parent != nullptr) {
    eraseChild(self->parent, self->parentsIndex);
  }
}

Node *nextPhysical(Node *node) {
  int index = -1;
  for (;;) {
    auto nextChild = getChildGeq(node, index + 1);
    if (nextChild >= 0) {
      return getChildExists(node, nextChild);
    }
    index = node->parentsIndex;
    node = node->parent;
    if (node == nullptr) {
      return nullptr;
    }
  }
}

Node *nextLogical(Node *node) {
  for (node = nextPhysical(node); node != nullptr && !node->entryPresent;
       node = nextPhysical(node))
    ;
  return node;
}

Node *prevPhysical(Node *node) {
  assert(node->parent != nullptr);
  auto prevChild = getChildLeq(node->parent, node->parentsIndex - 1);
  assert(prevChild < node->parentsIndex);
  if (prevChild >= 0) {
    node = getChildExists(node->parent, prevChild);
    // Move down the right spine
    for (;;) {
      auto rightMostChild = getChildLeq(node, 255);
      if (rightMostChild >= 0) {
        node = getChildExists(node, rightMostChild);
      } else {
        return node;
      }
    }
  } else {
    return node->parent;
  }
}

Node *prevLogical(Node *node) {
  for (node = prevPhysical(node); node != nullptr && !node->entryPresent;
       node = prevPhysical(node))
    ;
  return node;
}

struct Iterator {
  Node *n;
  int cmp;
};

std::string_view getSearchPath(Arena &arena, Node *n) {
  auto result = vector<char>(arena);
  for (;;) {
    for (int i = n->partialKeyLen - 1; i >= 0; --i) {
      result.push_back(n->partialKey[i]);
    }
    if (n->parent == nullptr) {
      break;
    }
    result.push_back(n->parentsIndex);
    n = n->parent;
  }
  std::reverse(result.begin(), result.end());
  if (result.size() > 0) {
    return std::string_view((const char *)&result[0],
                            result.size()); // NOLINT
  } else {
    return std::string_view();
  }
}

Iterator lastLeq(Node *n, const std::span<const uint8_t> key) {
  auto remaining = key;
  for (;;) {
    if (n->partialKeyLen > 0) {
      int commonLen = std::min<int>(n->partialKeyLen, remaining.size());
      for (int i = 0; i < commonLen; ++i) {
        auto c = n->partialKey[i] <=> remaining[i];
        if (c == 0) {
          continue;
        }
        if (c > 0) {
          n = prevPhysical(n);
          goto scanBackward;
        } else {
          goto downRightSpine;
        }
      }
      if (commonLen == n->partialKeyLen) {
        // partial key matches
        remaining = remaining.subspan(commonLen, remaining.size() - commonLen);
      } else if (n->partialKeyLen > int(remaining.size())) {
        // n is the first physical node greater than remaining, and there's no
        // eq node
        n = prevPhysical(n);
        goto scanBackward;
      }
    }
    {
      Arena arena;
      assert((std::string(getSearchPath(arena, n)) +
              std::string((const char *)remaining.data(), remaining.size()))
                 .ends_with(std::string((const char *)key.data(), key.size())));
    }
    if (remaining.size() == 0) {
      // We've found the physical node corresponding to search path `key`
      if (n->entryPresent) {
        return {n, 0};
      } else {
        goto scanBackward;
      }
    } else {
      int c = getChildLeq(n, remaining[0]);
      if (c == remaining[0]) {
        n = getChildExists(n, c);
        remaining = remaining.subspan(1, remaining.size() - 1);
      } else {
        if (c >= 0) {
          n = getChildExists(n, c);
          goto downRightSpine;
        } else {
          goto scanBackward;
        }
      }
    }
  }
downRightSpine:
  // The physical node corresponding to search path `key` does not
  // exist. Let's find the physical node corresponding to the highest
  // search key (not necessarily present) less than key.
  // Move down the right spine
  for (;;) {
    int c = getChildLeq(n, 255);
    if (c >= 0) {
      n = getChildExists(n, c);
    } else {
      goto scanBackward;
    }
  }
scanBackward:
  // Iterate backwards along existing physical nodes until we find a present
  // entry
  for (; !n->entryPresent; n = prevPhysical(n)) {
  }
  return {n, -1};
}

// Returns a pointer to the newly inserted node.  caller is reponsible for
// setting 'entry' fields on the result, which may have !entryPresent. The
// search path for `key` will have maxVersion at least `writeVersion` as a
// postcondition.
[[nodiscard]] Node *insert(Node **self_, std::span<const uint8_t> key,
                           int64_t writeVersion) {
  for (;;) {
    auto &self = *self_;
    // Handle an existing partial key
    int partialKeyIndex = 0;
    for (; partialKeyIndex < self->partialKeyLen; ++partialKeyIndex) {
      if (partialKeyIndex == int(key.size()) ||
          self->partialKey[partialKeyIndex] != key[partialKeyIndex]) {
        auto *old = self;
        self = newNode();
        self->maxVersion = old->maxVersion;
        self->partialKeyLen = partialKeyIndex;
        self->parent = old->parent;
        self->parentsIndex = old->parentsIndex;
        memcpy(self->partialKey, old->partialKey, partialKeyIndex);

        getOrCreateChild(self, old->partialKey[partialKeyIndex]) = old;
        old->parent = self;
        old->parentsIndex = old->partialKey[partialKeyIndex];

        memmove(old->partialKey, old->partialKey + partialKeyIndex + 1,
                old->partialKeyLen - (partialKeyIndex + 1));
        old->partialKeyLen -= partialKeyIndex + 1;
        break;
      }
    }
    key = key.subspan(partialKeyIndex, key.size() - partialKeyIndex);

    // Consider adding a partial key
    if (self->numChildren == 0 && !self->entryPresent) {
      self->partialKeyLen = std::min<int>(key.size(), self->kPartialKeyMaxLen);
      memcpy(self->partialKey, key.data(), self->partialKeyLen);
      key = key.subspan(self->partialKeyLen, key.size() - self->partialKeyLen);
    }

    self->maxVersion = std::max(self->maxVersion, writeVersion);

    if (key.size() == 0) {
      return self;
    }
    auto &child = getOrCreateChild(self, key.front());
    if (!child) {
      child = newNode();
      child->maxVersion = writeVersion;
      child->parent = self;
      child->parentsIndex = key.front();
    }
    self_ = &child;
    key = key.subspan(1, key.size() - 1);
  }
}

void destroyTree(Node *root) {
  Arena arena;
  auto toFree = vector<Node *>(arena);
  toFree.push_back(root);
  while (toFree.size() > 0) {
    auto *n = toFree.back();
    toFree.pop_back();
    // Add all children to toFree
    for (int child = getChildGeq(n, 0); child >= 0;
         child = getChildGeq(n, child + 1)) {
      auto *c = getChildExists(n, child);
      assert(c != nullptr);
      toFree.push_back(c);
    }
    free(n);
  }
}

void invalidateMax(Node *self) {
  int64_t expectedMax = std::numeric_limits<int64_t>::lowest();
  if (self->entryPresent) {
    expectedMax = std::max(expectedMax, self->entry.pointVersion);
    expectedMax = std::max(expectedMax, self->entry.rangeVersion);
  }
  for (int i = getChildGeq(self, 0); i >= 0; i = getChildGeq(self, i + 1)) {
    expectedMax = std::max(expectedMax, getChildExists(self, i)->maxVersion);
  }
  self->maxVersion = expectedMax;
  if (self->parent != nullptr) {
    invalidateMax(self->parent);
  }
}

struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
  void check(const ReadRange *reads, Result *result, int count) const {
    for (int i = 0; i < count; ++i) {
      const auto &r = reads[i];
      if (r.readVersion < oldestVersion) {
        result[i] = TooOld;
        continue;
      }
      // TODO support non-point reads
      assert(r.end.len == 0);
      auto [l, c] =
          lastLeq(root, std::span<const uint8_t>(r.begin.p, r.begin.len));
#if DEBUG_VERBOSE && !defined(NDEBUG)
      Arena arena;
      fprintf(stderr, "LastLeq for `%s' got `%s'\n", printable(r.begin).c_str(),
              printable(getSearchPath(arena, l)).c_str());
#endif
      assert(l != nullptr);
      assert(l->entryPresent);
      result[i] = (c == 0 ? l->entry.pointVersion : l->entry.rangeVersion) >
                          r.readVersion
                      ? Conflict
                      : Commit;
    }
  }
  void addWrites(const WriteRange *writes, int count) {
    for (int i = 0; i < count; ++i) {
      const auto &w = writes[i];
      if (w.end.len > 0) {
        auto *n = insert(&root, std::span<const uint8_t>(w.end.p, w.end.len),
                         std::numeric_limits<int64_t>::lowest());
        if (!n->entryPresent) {
          auto *p = prevLogical(n);
          assert(p != nullptr);
          n->entryPresent = true;
          n->entry.pointVersion = p->entry.rangeVersion;
          n->entry.rangeVersion = p->entry.rangeVersion;
        }

        auto *end = n;
        n = insert(&root, std::span<const uint8_t>(w.begin.p, w.begin.len),
                   std::numeric_limits<int64_t>::lowest());
        auto *begin = n;
        n->entryPresent = true;
        n->entry.pointVersion = w.writeVersion;
        n->entry.rangeVersion = w.writeVersion;
        for (n = nextLogical(n); n != end;) {
          auto *old = n;
          n = nextLogical(n);
          old->entryPresent = false;
          if (old->numChildren == 0 && old->parent != nullptr) {
            eraseChild(old->parent, old->parentsIndex);
          }
        }
        invalidateMax(begin);
        invalidateMax(end);
      } else {
        auto *n =
            insert(&root, std::span<const uint8_t>(w.begin.p, w.begin.len),
                   w.writeVersion);
        if (!n->entryPresent) {
          auto *p = prevLogical(n);
          assert(p != nullptr);
          n->entryPresent = true;
          n->entry.pointVersion = w.writeVersion;
          n->entry.rangeVersion = p->entry.rangeVersion;
        } else {
          n->entry.pointVersion =
              std::max(n->entry.pointVersion, w.writeVersion);
        }
      }
    }
  }
  void setOldestVersion(int64_t oldestVersion) {
    this->oldestVersion = oldestVersion;
  }
  explicit Impl(int64_t oldestVersion) : oldestVersion(oldestVersion) {
    // Insert ""
    root = newNode();
    root->maxVersion = oldestVersion;
    root->entry.pointVersion = oldestVersion;
    root->entry.rangeVersion = oldestVersion;
    root->entryPresent = true;
  }
  ~Impl() { destroyTree(root); }

  Node *root;
  int64_t oldestVersion;
};

// ==================== END IMPLEMENTATION ====================

// GCOVR_EXCL_START

void ConflictSet::check(const ReadRange *reads, Result *results,
                        int count) const {
  return impl->check(reads, results, count);
}

void ConflictSet::addWrites(const WriteRange *writes, int count) {
  return impl->addWrites(writes, count);
}

void ConflictSet::setOldestVersion(int64_t oldestVersion) {
  return impl->setOldestVersion(oldestVersion);
}

ConflictSet::ConflictSet(int64_t oldestVersion)
    : impl(new (safe_malloc(sizeof(Impl))) Impl{oldestVersion}) {}

ConflictSet::~ConflictSet() {
  if (impl) {
    impl->~Impl();
    free(impl);
  }
}

ConflictSet::ConflictSet(ConflictSet &&other) noexcept
    : impl(std::exchange(other.impl, nullptr)) {}

ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
  impl = std::exchange(other.impl, nullptr);
  return *this;
}

using ConflictSet_Result = ConflictSet::Result;
using ConflictSet_Key = ConflictSet::Key;
using ConflictSet_ReadRange = ConflictSet::ReadRange;
using ConflictSet_WriteRange = ConflictSet::WriteRange;

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);
}
__attribute__((__visibility__("default"))) void
ConflictSet_addWrites(void *cs, const ConflictSet_WriteRange *writes,
                      int count) {
  ((ConflictSet::Impl *)cs)->addWrites(writes, count);
}
__attribute__((__visibility__("default"))) void
ConflictSet_setOldestVersion(void *cs, int64_t oldestVersion) {
  ((ConflictSet::Impl *)cs)->setOldestVersion(oldestVersion);
}
__attribute__((__visibility__("default"))) void *
ConflictSet_create(int64_t oldestVersion) {
  return new (safe_malloc(sizeof(ConflictSet::Impl)))
      ConflictSet::Impl{oldestVersion};
}
__attribute__((__visibility__("default"))) void ConflictSet_destroy(void *cs) {
  using Impl = ConflictSet::Impl;
  ((Impl *)cs)->~Impl();
  free(cs);
}
}

namespace {

void printLogical(std::string &result, Node *node) {
  Arena arena;
  for (Node *iter = node; iter != nullptr;) {
    auto *next = nextLogical(iter);
    std::string key;
    for (uint8_t c : getSearchPath(arena, iter)) {
      key += "x";
      key += "0123456789abcdef"[c / 16];
      key += "0123456789abcdef"[c % 16];
    }
    if (iter->entry.pointVersion == iter->entry.rangeVersion) {
      result += key + " -> " + std::to_string(iter->entry.pointVersion) + "\n";
    } else {
      result += key + " -> " + std::to_string(iter->entry.pointVersion) + "\n";
      if (next == nullptr || (getSearchPath(arena, next) !=
                              (std::string(getSearchPath(arena, iter)) +
                               std::string("\x00", 1)))) {
        result +=
            key + "x00 -> " + std::to_string(iter->entry.rangeVersion) + "\n";
      }
    }
    iter = next;
  }
}

[[maybe_unused]] void debugPrintDot(FILE *file, Node *node) {

  struct DebugDotPrinter {

    explicit DebugDotPrinter(FILE *file) : file(file) {}

    void print(Node *n) {
      assert(n != nullptr);
      auto partialKey = printable(Key{n->partialKey, n->partialKeyLen});
      if (n->entryPresent) {
        fprintf(file,
                " k_%p [label=\"m=%" PRId64 " p=%" PRId64 " r=%" PRId64
                " %s\"];\n",
                (void *)n, n->maxVersion, n->entry.pointVersion,
                n->entry.rangeVersion, partialKey.c_str());
      } else {
        fprintf(file, " k_%p [label=\"m=%" PRId64 " %s\"];\n", (void *)n,
                n->maxVersion, partialKey.c_str());
      }
      for (int child = getChildGeq(n, 0); child >= 0;
           child = getChildGeq(n, child + 1)) {
        auto *c = getChildExists(n, child);
        fprintf(file, " k_%p -> k_%p [label=\"x%02x\"];\n", (void *)n,
                (void *)c, child);
        print(c);
      }
    }
    FILE *file;
  };

  fprintf(file, "digraph ConflictSet {\n");
  assert(node != nullptr);
  DebugDotPrinter printer{file};
  printer.print(node);
  fprintf(file, "}\n");
}

void checkParentPointers(Node *node, bool &success) {
  for (int i = getChildGeq(node, 0); i >= 0; i = getChildGeq(node, i + 1)) {
    auto *child = getChildExists(node, i);
    if (child->parent != node) {
      Arena arena;
      fprintf(stderr, "%s child %d has parent pointer %p. Expected %p\n",
              printable(getSearchPath(arena, node)).c_str(), i,
              (void *)child->parent, (void *)node);
      success = false;
    }
    checkParentPointers(child, success);
  }
}

[[maybe_unused]] int64_t checkMaxVersion(Node *node, bool &success) {
  int64_t expected =
      node->entryPresent
          ? std::max(node->entry.pointVersion, node->entry.rangeVersion)
          : std::numeric_limits<int64_t>::lowest();
  for (int i = getChildGeq(node, 0); i >= 0; i = getChildGeq(node, i + 1)) {
    auto *child = getChildExists(node, i);
    expected = std::max(expected, checkMaxVersion(child, success));
  }
  if (node->maxVersion != expected) {
    Arena arena;
    fprintf(stderr, "%s has max version %" PRId64 " . Expected %" PRId64 "\n",
            printable(getSearchPath(arena, node)).c_str(), node->maxVersion,
            expected);
    success = false;
  }
  return expected;
}

bool checkCorrectness(Node *node, ReferenceImpl &refImpl) {
  bool success = true;

  checkParentPointers(node, success);
  checkMaxVersion(node, success);

  std::string logicalMap;
  std::string referenceLogicalMap;
  printLogical(logicalMap, node);
  refImpl.printLogical(referenceLogicalMap);
  if (logicalMap != referenceLogicalMap) {
    fprintf(stderr,
            "Logical map not equal to reference logical map.\n\nActual:\n"
            "%s\nExpected:\n%s\n",
            logicalMap.c_str(), referenceLogicalMap.c_str());
    success = false;
  }

  return success;
}

} // namespace

namespace std {
void __throw_length_error(const char *) { __builtin_unreachable(); }
} // namespace std

#ifdef ENABLE_MAIN
#define ANKERL_NANOBENCH_IMPLEMENT
#include "third_party/nanobench.h"

void bench() {
  ankerl::nanobench::Bench bench;
  {
    auto *n = newNode();
    for (int i = 0; i < 64; ++i) {
      getOrCreateChild(n, i) = newNode();
      bench.run("getChildLeq" + std::to_string(i),
                [&]() { bench.doNotOptimizeAway(getChildLeq(n, 255)); });
    }
    destroyTree(n);
  }
  {
    auto *n = newNode();
    for (int i = 255; i >= 3 * 64; --i) {
      getOrCreateChild(n, i) = newNode();
      bench.run("getChildGeq" + std::to_string(i),
                [&]() { bench.doNotOptimizeAway(getChildGeq(n, 0)); });
    }
    destroyTree(n);
  }
}

void printTree() {
  int64_t writeVersion = 0;
  ConflictSet::Impl cs{writeVersion};
  ReferenceImpl refImpl{writeVersion};
  Arena arena;
  constexpr int kNumKeys = 5;
  auto *write = new (arena) ConflictSet::WriteRange[kNumKeys];
  for (int i = 0; i < kNumKeys; ++i) {
    write[i].begin = toKey(arena, i);
    write[i].end.len = 0;
    write[i].writeVersion = ++writeVersion;
  }
  cs.addWrites(write, kNumKeys);
  for (int i = 0; i < kNumKeys; ++i) {
    write[i].writeVersion = ++writeVersion;
  }
  cs.addWrites(write, kNumKeys);
  debugPrintDot(stdout, cs.root);
}

int main(void) {
  bench();
  // printTree();
  return 0;
}
#endif

#ifdef ENABLE_FUZZ
extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
  TestDriver<ConflictSet::Impl> driver{data, size};
  static_assert(driver.kMaxKeyLen > Node::kPartialKeyMaxLen);

  for (;;) {
    bool done = driver.next();
    if (!driver.ok) {
      debugPrintDot(stdout, driver.cs.root);
      fflush(stdout);
      abort();
    }
    bool success = checkCorrectness(driver.cs.root, driver.refImpl);
    if (!success) {
      debugPrintDot(stdout, driver.cs.root);
      fflush(stdout);
      abort();
    }
    if (done) {
      break;
    }
  }

  return 0;
}
#endif

// GCOVR_EXCL_STOP