| 1 | #include "duckdb/execution/index/art/iterator.hpp" |
| 2 | |
| 3 | #include "duckdb/common/limits.hpp" |
| 4 | #include "duckdb/execution/index/art/art.hpp" |
| 5 | #include "duckdb/execution/index/art/node.hpp" |
| 6 | #include "duckdb/execution/index/art/prefix.hpp" |
| 7 | |
| 8 | namespace duckdb { |
| 9 | |
| 10 | void IteratorCurrentKey::Push(const uint8_t byte) { |
| 11 | if (cur_key_pos == key.size()) { |
| 12 | key.push_back(x: byte); |
| 13 | } |
| 14 | D_ASSERT(cur_key_pos < key.size()); |
| 15 | key[cur_key_pos++] = byte; |
| 16 | } |
| 17 | |
| 18 | void IteratorCurrentKey::Pop(const idx_t n) { |
| 19 | cur_key_pos -= n; |
| 20 | D_ASSERT(cur_key_pos <= key.size()); |
| 21 | } |
| 22 | |
| 23 | uint8_t &IteratorCurrentKey::operator[](idx_t idx) { |
| 24 | if (idx >= key.size()) { |
| 25 | key.push_back(x: 0); |
| 26 | } |
| 27 | D_ASSERT(idx < key.size()); |
| 28 | return key[idx]; |
| 29 | } |
| 30 | |
| 31 | bool IteratorCurrentKey::operator>(const ARTKey &k) const { |
| 32 | for (idx_t i = 0; i < MinValue<idx_t>(a: cur_key_pos, b: k.len); i++) { |
| 33 | if (key[i] > k.data[i]) { |
| 34 | return true; |
| 35 | } else if (key[i] < k.data[i]) { |
| 36 | return false; |
| 37 | } |
| 38 | } |
| 39 | return cur_key_pos > k.len; |
| 40 | } |
| 41 | |
| 42 | bool IteratorCurrentKey::operator>=(const ARTKey &k) const { |
| 43 | for (idx_t i = 0; i < MinValue<idx_t>(a: cur_key_pos, b: k.len); i++) { |
| 44 | if (key[i] > k.data[i]) { |
| 45 | return true; |
| 46 | } else if (key[i] < k.data[i]) { |
| 47 | return false; |
| 48 | } |
| 49 | } |
| 50 | return cur_key_pos >= k.len; |
| 51 | } |
| 52 | |
| 53 | bool IteratorCurrentKey::operator==(const ARTKey &k) const { |
| 54 | if (cur_key_pos != k.len) { |
| 55 | return false; |
| 56 | } |
| 57 | for (idx_t i = 0; i < cur_key_pos; i++) { |
| 58 | if (key[i] != k.data[i]) { |
| 59 | return false; |
| 60 | } |
| 61 | } |
| 62 | return true; |
| 63 | } |
| 64 | |
| 65 | void Iterator::FindMinimum(Node &node) { |
| 66 | |
| 67 | // reconstruct the prefix |
| 68 | // FIXME: get all bytes at once to increase performance |
| 69 | auto &node_prefix = node.GetPrefix(art&: *art); |
| 70 | for (idx_t i = 0; i < node_prefix.count; i++) { |
| 71 | cur_key.Push(byte: node_prefix.GetByte(art: *art, position: i)); |
| 72 | } |
| 73 | |
| 74 | // found the minimum |
| 75 | if (node.DecodeARTNodeType() == NType::LEAF) { |
| 76 | last_leaf = Node::GetAllocator(art: *art, type: NType::LEAF).Get<Leaf>(ptr: node); |
| 77 | return; |
| 78 | } |
| 79 | |
| 80 | // go to the leftmost entry in the current node |
| 81 | uint8_t byte = 0; |
| 82 | auto next = node.GetNextChild(art&: *art, byte); |
| 83 | D_ASSERT(next); |
| 84 | cur_key.Push(byte); |
| 85 | |
| 86 | // recurse |
| 87 | nodes.emplace(args&: node, args&: byte); |
| 88 | FindMinimum(node&: *next); |
| 89 | } |
| 90 | |
| 91 | void Iterator::PushKey(const Node &node, const uint8_t byte) { |
| 92 | if (node.DecodeARTNodeType() != NType::LEAF) { |
| 93 | cur_key.Push(byte); |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | bool Iterator::Scan(const ARTKey &key, const idx_t &max_count, vector<row_t> &result_ids, const bool &is_inclusive) { |
| 98 | |
| 99 | bool has_next; |
| 100 | do { |
| 101 | if (!key.Empty()) { |
| 102 | // no more row IDs within the key bounds |
| 103 | if (is_inclusive) { |
| 104 | if (cur_key > key) { |
| 105 | return true; |
| 106 | } |
| 107 | } else { |
| 108 | if (cur_key >= key) { |
| 109 | return true; |
| 110 | } |
| 111 | } |
| 112 | } |
| 113 | |
| 114 | // adding more elements would exceed the max count |
| 115 | if (result_ids.size() + last_leaf->count > max_count) { |
| 116 | return false; |
| 117 | } |
| 118 | |
| 119 | // FIXME: copy all at once to improve performance |
| 120 | for (idx_t i = 0; i < last_leaf->count; i++) { |
| 121 | row_t row_id = last_leaf->GetRowId(art: *art, position: i); |
| 122 | result_ids.push_back(x: row_id); |
| 123 | } |
| 124 | |
| 125 | // get the next leaf |
| 126 | has_next = Next(); |
| 127 | |
| 128 | } while (has_next); |
| 129 | |
| 130 | return true; |
| 131 | } |
| 132 | |
| 133 | void Iterator::PopNode() { |
| 134 | auto cur_node = nodes.top(); |
| 135 | idx_t elements_to_pop = cur_node.node.GetPrefix(art&: *art).count + (nodes.size() != 1); |
| 136 | cur_key.Pop(n: elements_to_pop); |
| 137 | nodes.pop(); |
| 138 | } |
| 139 | |
| 140 | bool Iterator::Next() { |
| 141 | if (!nodes.empty()) { |
| 142 | auto cur_node = nodes.top().node; |
| 143 | if (cur_node.DecodeARTNodeType() == NType::LEAF) { |
| 144 | // pop leaf |
| 145 | // we must pop the prefix size + the key to the node, unless we are popping the root |
| 146 | PopNode(); |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | // look for the next leaf |
| 151 | while (!nodes.empty()) { |
| 152 | |
| 153 | // cur_node |
| 154 | auto &top = nodes.top(); |
| 155 | Node node = top.node; |
| 156 | |
| 157 | // found a leaf: move to next node |
| 158 | if (node.DecodeARTNodeType() == NType::LEAF) { |
| 159 | last_leaf = Node::GetAllocator(art: *art, type: NType::LEAF).Get<Leaf>(ptr: node); |
| 160 | return true; |
| 161 | } |
| 162 | |
| 163 | // find next node |
| 164 | if (top.byte == NumericLimits<uint8_t>::Maximum()) { |
| 165 | // no node found: move up the tree, pop prefix and key of current node |
| 166 | PopNode(); |
| 167 | continue; |
| 168 | } |
| 169 | |
| 170 | top.byte == 0 ? top.byte : top.byte++; |
| 171 | auto next_node = node.GetNextChild(art&: *art, byte&: top.byte); |
| 172 | |
| 173 | if (next_node) { |
| 174 | // add the next node's key byte |
| 175 | PushKey(node, byte: top.byte); |
| 176 | |
| 177 | // add prefix of new node |
| 178 | // FIXME: get all bytes at once to increase performance |
| 179 | auto &next_node_prefix = next_node->GetPrefix(art&: *art); |
| 180 | for (idx_t i = 0; i < next_node_prefix.count; i++) { |
| 181 | cur_key.Push(byte: next_node_prefix.GetByte(art: *art, position: i)); |
| 182 | } |
| 183 | |
| 184 | // next node found: push it |
| 185 | nodes.emplace(args&: *next_node, args: 0); |
| 186 | } else { |
| 187 | |
| 188 | // no node found: move up the tree, pop prefix and key of current node |
| 189 | PopNode(); |
| 190 | } |
| 191 | } |
| 192 | return false; |
| 193 | } |
| 194 | |
| 195 | bool Iterator::LowerBound(Node node, const ARTKey &key, const bool &is_inclusive) { |
| 196 | |
| 197 | if (!node.IsSet()) { |
| 198 | return false; |
| 199 | } |
| 200 | |
| 201 | idx_t depth = 0; |
| 202 | bool equal = true; |
| 203 | while (true) { |
| 204 | |
| 205 | nodes.emplace(args&: node, args: 0); |
| 206 | auto &top = nodes.top(); |
| 207 | |
| 208 | // reconstruct the prefix |
| 209 | // FIXME: get all bytes at once to increase performance |
| 210 | reference<Prefix> node_prefix(top.node.GetPrefix(art&: *art)); |
| 211 | for (idx_t i = 0; i < node_prefix.get().count; i++) { |
| 212 | cur_key.Push(byte: node_prefix.get().GetByte(art: *art, position: i)); |
| 213 | } |
| 214 | |
| 215 | // greater case: find leftmost leaf node directly |
| 216 | if (!equal) { |
| 217 | while (node.DecodeARTNodeType() != NType::LEAF) { |
| 218 | |
| 219 | uint8_t byte = 0; |
| 220 | auto next_node = *node.GetNextChild(art&: *art, byte); |
| 221 | D_ASSERT(next_node.IsSet()); |
| 222 | |
| 223 | PushKey(node, byte); |
| 224 | nodes.emplace(args&: node, args&: byte); |
| 225 | node = next_node; |
| 226 | |
| 227 | // reconstruct the prefix |
| 228 | node_prefix = node.GetPrefix(art&: *art); |
| 229 | for (idx_t i = 0; i < node_prefix.get().count; i++) { |
| 230 | cur_key.Push(byte: node_prefix.get().GetByte(art: *art, position: i)); |
| 231 | } |
| 232 | |
| 233 | auto &c_top = nodes.top(); |
| 234 | c_top.node = node; |
| 235 | } |
| 236 | } |
| 237 | |
| 238 | if (node.DecodeARTNodeType() == NType::LEAF) { |
| 239 | // found a leaf node: check if it is bigger or equal than the current key |
| 240 | last_leaf = Node::GetAllocator(art: *art, type: NType::LEAF).Get<Leaf>(ptr: node); |
| 241 | |
| 242 | // if the search is not inclusive the leaf node could still be equal to the current value |
| 243 | // check if leaf is equal to the current key |
| 244 | if (cur_key == key) { |
| 245 | // if it's not inclusive check if there is a next leaf |
| 246 | if (!is_inclusive && !Next()) { |
| 247 | return false; |
| 248 | } else { |
| 249 | return true; |
| 250 | } |
| 251 | } |
| 252 | |
| 253 | if (cur_key > key) { |
| 254 | return true; |
| 255 | } |
| 256 | // Case1: When the ART has only one leaf node, the Next() will return false |
| 257 | // Case2: This means the previous node prefix(if any) + a_key(one element of of key array of previous node) |
| 258 | // == key[q..=w]. |
| 259 | // But key[w+1..=z] maybe greater than leaf node prefix. |
| 260 | // One fact is key[w] is alawys equal to a_key and the next element |
| 261 | // of key array of previous node is always > a_key So we just call Next() once. |
| 262 | |
| 263 | return Next(); |
| 264 | } |
| 265 | |
| 266 | // equal case: |
| 267 | node_prefix = node.GetPrefix(art&: *art); |
| 268 | auto mismatch_pos = node_prefix.get().KeyMismatchPosition(art: *art, key, depth); |
| 269 | if (mismatch_pos != node_prefix.get().count) { |
| 270 | if (node_prefix.get().GetByte(art: *art, position: mismatch_pos) < key[depth + mismatch_pos]) { |
| 271 | // less |
| 272 | PopNode(); |
| 273 | return Next(); |
| 274 | } |
| 275 | // greater |
| 276 | top.byte = 0; |
| 277 | return Next(); |
| 278 | } |
| 279 | |
| 280 | // prefix matches, search inside the child for the key |
| 281 | depth += node_prefix.get().count; |
| 282 | top.byte = key[depth]; |
| 283 | auto child = node.GetNextChild(art&: *art, byte&: top.byte); |
| 284 | equal = key[depth] == top.byte; |
| 285 | |
| 286 | // the maximum key byte of the current node is less than the key |
| 287 | // fall back to the previous node |
| 288 | if (!child) { |
| 289 | PopNode(); |
| 290 | return Next(); |
| 291 | } |
| 292 | |
| 293 | PushKey(node, byte: top.byte); |
| 294 | node = *child; |
| 295 | |
| 296 | // all children of this node qualify as greater or equal |
| 297 | depth++; |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | } // namespace duckdb |
| 302 | |