GeezerVisonThe World Wide Web, restored to 800 pixels of glory. |
| Home|About|Gaming cabinet|Archives|How it works|Starter sites |
GeezerVison is showing github.com — cpython/Python/hamt.c at main · python/cpython Expand file tree/ Copy pathhamt.cMore file actions 2907 lines (2319 loc) · 78.3 KB / Copy pathhamt.cFile metadata and controls2907 lines (2319 loc) · 78.3 KB 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 #include "Python.h" #include "pycore_bitutils.h" // _Py_popcount32() #include "pycore_hamt.h" #include "pycore_initconfig.h" // _PyStatus_OK() #include "pycore_long.h" // _PyLong_Format() #include "pycore_object.h" // _PyObject_GC_TRACK() #include "pycore_tuple.h" // _PyTuple_FromPair #include <stddef.h> // offsetof() /*[clinic input] class hamt "PyHamtObject *" "&_PyHamt_Type" [clinic start generated code]*/ /*[clinic end generated code: output=da39a3ee5e6b4b0d input=ab743c139c849b8e]*/ #include "clinic/hamt.c.h" /* This file provides an implementation of an immutable mapping using the Hash Array Mapped Trie (or HAMT) datastructure. This design allows to have: 1. Efficient copy: immutable mappings can be copied by reference, making it an O(1) operation. 2. Efficient mutations: due to structural sharing, only a portion of the trie needs to be copied when the collection is mutated. The cost of set/delete operations is O(log N). 3. Efficient lookups: O(log N). (where N is number of key/value items in the immutable mapping.) HAMT ==== The core idea of HAMT is that the shape of the trie is encoded into the hashes of keys. Say we want to store a K/V pair in our mapping. First, we calculate the hash of K, let's say it's 19830128, or in binary: 0b1001011101001010101110000 = 19830128 Now let's partition this bit representation of the hash into blocks of 5 bits each: 0b00_00000_10010_11101_00101_01011_10000 = 19830128 (6) (5) (4) (3) (2) (1) Each block of 5 bits represents a number between 0 and 31. So if we have a tree that consists of nodes, each of which is an array of 32 pointers, those 5-bit blocks will encode a position on a single tree level. For example, storing the key K with hash 19830128, results in the following tree structure: (array of 32 pointers) +---+ -- +----+----+----+ -- +----+ root node | 0 | .. | 15 | 16 | 17 | .. | 31 | 0b10000 = 16 (1) (level 1) +---+ -- +----+----+----+ -- +----+ | +---+ -- +----+----+----+ -- +----+ a 2nd level node | 0 | .. | 10 | 11 | 12 | .. | 31 | 0b01011 = 11 (2) +---+ -- +----+----+----+ -- +----+ | +---+ -- +----+----+----+ -- +----+ a 3rd level node | 0 | .. | 04 | 05 | 06 | .. | 31 | 0b00101 = 5 (3) +---+ -- +----+----+----+ -- +----+ | +---+ -- +----+----+----+----+ a 4th level node | 0 | .. | 04 | 29 | 30 | 31 | 0b11101 = 29 (4) +---+ -- +----+----+----+----+ | +---+ -- +----+----+----+ -- +----+ a 5th level node | 0 | .. | 17 | 18 | 19 | .. | 31 | 0b10010 = 18 (5) +---+ -- +----+----+----+ -- +----+ | +--------------+ | +---+ -- +----+----+----+ -- +----+ a 6th level node | 0 | .. | 15 | 16 | 17 | .. | 31 | 0b00000 = 0 (6) +---+ -- +----+----+----+ -- +----+ | V -- our value (or collision) To rehash: for a K/V pair, the hash of K encodes where in the tree V will be stored. To optimize memory footprint and handle hash collisions, our implementation uses three different types of nodes: * A Bitmap node; * An Array node; * A Collision node. Because we implement an immutable dictionary, our nodes are also immutable. Therefore, when we need to modify a node, we copy it, and do that modification to the copy. Array Nodes ----------- These nodes are very simple. Essentially they are arrays of 32 pointers we used to illustrate the high-level idea in the previous section. We use Array nodes only when we need to store more than 16 pointers in a single node. Array nodes do not store key objects or value objects. They are used only as an indirection level - their pointers point to other nodes in the tree. Bitmap Node ----------- Allocating a new 32-pointers array for every node of our tree would be very expensive. Unless we store millions of keys, most of tree nodes would be very sparse. When we have less than 16 elements in a node, we don't want to use the Array node, that would mean that we waste a lot of memory. Instead, we can use bitmap compression and can have just as many pointers as we need! Bitmap nodes consist of two fields: 1. An array of pointers. If a Bitmap node holds N elements, the array will be of N pointers. 2. A 32bit integer -- a bitmap field. If an N-th bit is set in the bitmap, it means that the node has an N-th element. For example, say we need to store a 3 elements sparse array: +---+ -- +---+ -- +----+ -- +----+ | 0 | .. | 4 | .. | 11 | .. | 17 | +---+ -- +---+ -- +----+ -- +----+ | | | o1 o2 o3 We allocate a three-pointer Bitmap node. Its bitmap field will be then set to: 0b_00100_00010_00000_10000 == (1 << 17) | (1 << 11) | (1 << 4) To check if our Bitmap node has an I-th element we can do: bitmap & (1 << I) And here's a formula to calculate a position in our pointer array which would correspond to an I-th element: popcount(bitmap & ((1 << I) - 1)) Let's break it down: * `popcount` is a function that returns a number of bits set to 1; * `((1 << I) - 1)` is a mask to filter the bitmask to contain bits set to the *right* of our bit. So for our 17, 11, and 4 indexes: * bitmap & ((1 << 17) - 1) == 0b100000010000 => 2 bits are set => index is 2. * bitmap & ((1 << 11) - 1) == 0b10000 => 1 bit is set => index is 1. * bitmap & ((1 << 4) - 1) == 0b0 => 0 bits are set => index is 0. To conclude: Bitmap nodes are just like Array nodes -- they can store a number of pointers, but use bitmap compression to eliminate unused pointers. Bitmap nodes have two pointers for each item: +----+----+----+----+ -- +----+----+ | k1 | v1 | k2 | v2 | .. | kN | vN | +----+----+----+----+ -- +----+----+ When kI == NULL, vI points to another tree level. When kI != NULL, the actual key object is stored in kI, and its value is stored in vI. Collision Nodes --------------- Collision nodes are simple arrays of pointers -- two pointers per key/value. When there's a hash collision, say for k1/v1 and k2/v2 we have `hash(k1)==hash(k2)`. Then our collision node will be: +----+----+----+----+ | k1 | v1 | k2 | v2 | +----+----+----+----+ Tree Structure -------------- All nodes are PyObjects. The `PyHamtObject` object has a pointer to the root node (h_root), and has a length field (h_count). High-level functions accept a PyHamtObject object and dispatch to lower-level functions depending on what kind of node h_root points to. Operations ========== There are three fundamental operations on an immutable dictionary: 1. "o.assoc(k, v)" will return a new immutable dictionary, that will be a copy of "o", but with the "k/v" item set. Functions in this file: hamt_node_assoc, hamt_node_bitmap_assoc, hamt_node_array_assoc, hamt_node_collision_assoc `hamt_node_assoc` function accepts a node object, and calls other functions depending on its actual type. 2. "o.find(k)" will lookup key "k" in "o". Functions: hamt_node_find, hamt_node_bitmap_find, hamt_node_array_find, hamt_node_collision_find 3. "o.without(k)" will return a new immutable dictionary, that will be a copy of "o", buth without the "k" key. Functions: hamt_node_without, hamt_node_bitmap_without, hamt_node_array_without, hamt_node_collision_without Further Reading =============== 1. http://blog.higher-order.net/2009/09/08/understanding-clojures-persistenthashmap-deftwice.html 2. http://blog.higher-order.net/2010/08/16/assoc-and-clojures-persistenthashmap-part-ii.html 3. Clojure's PersistentHashMap implementation: https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/PersistentHashMap.java Debug ===== The HAMT datatype is accessible for testing purposes under the `_testinternalcapi` module: >>> from _testinternalcapi import hamt >>> h = hamt() >>> h2 = h.set('a', 2) >>> h3 = h2.set('b', 3) >>> list(h3) ['a', 'b'] When CPython is built in debug mode, a '__dump__()' method is available to introspect the tree: >>> print(h3.__dump__()) HAMT(len=2): BitmapNode(size=4 count=2 bitmap=0b110 id=0x10eb9d9e8): 'a': 2 'b': 3 */ #define IS_ARRAY_NODE(node) Py_IS_TYPE(node, &_PyHamt_ArrayNode_Type) #define IS_BITMAP_NODE(node) Py_IS_TYPE(node, &_PyHamt_BitmapNode_Type) #define IS_COLLISION_NODE(node) Py_IS_TYPE(node, &_PyHamt_CollisionNode_Type) /* Return type for 'find' (lookup a key) functions. * F_ERROR - an error occurred; * F_NOT_FOUND - the key was not found; * F_FOUND - the key was found. */ typedef enum {F_ERROR, F_NOT_FOUND, F_FOUND} hamt_find_t; /* Return type for 'without' (delete a key) functions. * W_ERROR - an error occurred; * W_NOT_FOUND - the key was not found: there's nothing to delete; * W_EMPTY - the key was found: the node/tree would be empty if the key is deleted; * W_NEWNODE - the key was found: a new node/tree is returned without that key. */ typedef enum {W_ERROR, W_NOT_FOUND, W_EMPTY, W_NEWNODE} hamt_without_t; /* Low-level iterator protocol type. * I_ITEM - a new item has been yielded; * I_END - the whole tree was visited (similar to StopIteration). */ typedef enum {I_ITEM, I_END} hamt_iter_t; #define HAMT_ARRAY_NODE_SIZE 32 typedef struct { PyObject_HEAD PyHamtNode *a_array[HAMT_ARRAY_NODE_SIZE]; Py_ssize_t a_count; } PyHamtNode_Array; #define _PyHamtNode_Array_CAST(op) ((PyHamtNode_Array *)(op)) typedef struct { PyObject_VAR_HEAD int32_t c_hash; PyObject *c_array[1]; } PyHamtNode_Collision; #define _PyHamtNode_Collision_CAST(op) ((PyHamtNode_Collision *)(op)) static PyHamtObject * hamt_alloc(void); static PyHamtNode * hamt_node_assoc(PyHamtNode *node, uint32_t shift, int32_t hash, PyObject *key, PyObject *val, int* added_leaf); static hamt_without_t hamt_node_without(PyHamtNode *node, uint32_t shift, int32_t hash, PyObject *key, PyHamtNode **new_node); static hamt_find_t hamt_node_find(PyHamtNode *node, uint32_t shift, int32_t hash, PyObject *key, PyObject **val); #ifdef Py_DEBUG static int hamt_node_dump(PyHamtNode *node, PyUnicodeWriter *writer, int level); #endif static PyHamtNode * hamt_node_array_new(Py_ssize_t); static PyHamtNode * hamt_node_collision_new(int32_t hash, Py_ssize_t size); static inline Py_ssize_t hamt_node_collision_count(PyHamtNode_Collision *node); #ifdef Py_DEBUG static void _hamt_node_array_validate(void *obj_raw) { PyObject *obj = _PyObject_CAST(obj_raw); assert(IS_ARRAY_NODE(obj)); PyHamtNode_Array *node = (PyHamtNode_Array*)obj; Py_ssize_t i = 0, count = 0; for (; i < HAMT_ARRAY_NODE_SIZE; i++) { if (node->a_array[i] != NULL) { count++; } } assert(count == node->a_count); } #define VALIDATE_ARRAY_NODE(NODE) \ do { _hamt_node_array_validate(NODE); } while (0); #else #define VALIDATE_ARRAY_NODE(NODE) #endif /* Returns -1 on error */ static inline int32_t hamt_hash(PyObject *o) { Py_hash_t hash = PyObject_Hash(o); #if SIZEOF_PY_HASH_T <= 4 return hash; #else if (hash == -1) { /* exception */ return -1; } /* While it's somewhat suboptimal to reduce Python's 64 bit hash to 32 bits via XOR, it seems that the resulting hash function is good enough (this is also how Long type is hashed in Java.) Storing 10, 100, 1000 Python strings results in a relatively shallow and uniform tree structure. Also it's worth noting that it would be possible to adapt the tree structure to 64 bit hashes, but that would increase memory pressure and provide little to no performance benefits for collections with fewer than billions of key/value pairs. Important: do not change this hash reducing function. There are many tests that need an exact tree shape to cover all code paths and we do that by specifying concrete values for test data's `__hash__`. If this function is changed most of the regression tests would become useless. */ int32_t xored = (int32_t)(hash & 0xffffffffl) ^ (int32_t)(hash >> 32); return xored == -1 ? -2 : xored; #endif } static inline uint32_t hamt_mask(int32_t hash, uint32_t shift) { return (((uint32_t)hash >> shift) & 0x01f); } static inline uint32_t hamt_bitpos(int32_t hash, uint32_t shift) { return (uint32_t)1 << hamt_mask(hash, shift); } static inline uint32_t hamt_bitindex(uint32_t bitmap, uint32_t bit) { return (uint32_t)_Py_popcount32(bitmap & (bit - 1)); } /////////////////////////////////// Dump Helpers #ifdef Py_DEBUG static int _hamt_dump_ident(PyUnicodeWriter *writer, int level) { /* Write `' ' * level` to the `writer` */ PyObject *str = NULL; PyObject *num = NULL; PyObject *res = NULL; int ret = -1; str = PyUnicode_FromString(" "); if (str == NULL) { goto error; } num = PyLong_FromLong((long)level); if (num == NULL) { goto error; } res = PyNumber_Multiply(str, num); if (res == NULL) { goto error; } ret = PyUnicodeWriter_WriteStr(writer, res); error: Py_XDECREF(res); Py_XDECREF(str); Py_XDECREF(num); return ret; } #endif /* Py_DEBUG */ /////////////////////////////////// Bitmap Node #define _PyHamtNode_Bitmap_CAST(op) ((PyHamtNode_Bitmap *)(op)) static PyHamtNode * hamt_node_bitmap_new(Py_ssize_t size) { /* Create a new bitmap node of size 'size' */ PyHamtNode_Bitmap *node; Py_ssize_t i; if (size == 0) { /* Since bitmap nodes are immutable, we can cache the instance for size=0 and reuse it whenever we need an empty bitmap node. */ return (PyHamtNode *)&_Py_SINGLETON(hamt_bitmap_node_empty); } assert(size >= 0); assert(size % 2 == 0); /* No freelist; allocate a new bitmap node */ node = PyObject_GC_NewVar( PyHamtNode_Bitmap, &_PyHamt_BitmapNode_Type, size); if (node == NULL) { return NULL; } Py_SET_SIZE(node, size); for (i = 0; i < size; i++) { node->b_array[i] = NULL; } node->b_bitmap = 0; _PyObject_GC_TRACK(node); return (PyHamtNode *)node; } static inline Py_ssize_t hamt_node_bitmap_count(PyHamtNode_Bitmap *node) { return Py_SIZE(node) / 2; } static PyHamtNode_Bitmap * hamt_node_bitmap_clone(PyHamtNode_Bitmap *node) { /* Clone a bitmap node; return a new one with the same child notes. */ PyHamtNode_Bitmap *clone; Py_ssize_t i; clone = (PyHamtNode_Bitmap *)hamt_node_bitmap_new(Py_SIZE(node)); if (clone == NULL) { return NULL; } for (i = 0; i < Py_SIZE(node); i++) { clone->b_array[i] = Py_XNewRef(node->b_array[i]); } clone->b_bitmap = node->b_bitmap; return clone; } static PyHamtNode_Bitmap * hamt_node_bitmap_clone_without(PyHamtNode_Bitmap *o, uint32_t bit) { assert(bit & o->b_bitmap); assert(hamt_node_bitmap_count(o) > 1); PyHamtNode_Bitmap *new = (PyHamtNode_Bitmap *)hamt_node_bitmap_new( Py_SIZE(o) - 2); if (new == NULL) { return NULL; } uint32_t idx = hamt_bitindex(o->b_bitmap, bit); uint32_t key_idx = 2 * idx; uint32_t val_idx = key_idx + 1; uint32_t i; for (i = 0; i < key_idx; i++) { new->b_array[i] = Py_XNewRef(o->b_array[i]); } assert(Py_SIZE(o) >= 0 && Py_SIZE(o) <= 32); for (i = val_idx + 1; i < (uint32_t)Py_SIZE(o); i++) { new->b_array[i - 2] = Py_XNewRef(o->b_array[i]); } new->b_bitmap = o->b_bitmap & ~bit; return new; } static PyHamtNode * hamt_node_new_bitmap_or_collision(uint32_t shift, PyObject *key1, PyObject *val1, int32_t key2_hash, PyObject *key2, PyObject *val2) { /* Helper method. Creates a new node for key1/val and key2/val2 pairs. If key1 hash is equal to the hash of key2, a Collision node will be created. If they are not equal, a Bitmap node is created. */ int32_t key1_hash = hamt_hash(key1); if (key1_hash == -1) { return NULL; } if (key1_hash == key2_hash) { PyHamtNode_Collision *n; n = (PyHamtNode_Collision *)hamt_node_collision_new(key1_hash, 4); if (n == NULL) { return NULL; } n->c_array[0] = Py_NewRef(key1); n->c_array[1] = Py_NewRef(val1); n->c_array[2] = Py_NewRef(key2); n->c_array[3] = Py_NewRef(val2); return (PyHamtNode *)n; } else { int added_leaf = 0; PyHamtNode *n = hamt_node_bitmap_new(0); if (n == NULL) { return NULL; } PyHamtNode *n2 = hamt_node_assoc( n, shift, key1_hash, key1, val1, &added_leaf); Py_DECREF(n); if (n2 == NULL) { return NULL; } n = hamt_node_assoc(n2, shift, key2_hash, key2, val2, &added_leaf); Py_DECREF(n2); if (n == NULL) { return NULL; } return n; } } static PyHamtNode * hamt_node_bitmap_assoc(PyHamtNode_Bitmap *self, uint32_t shift, int32_t hash, PyObject *key, PyObject *val, int* added_leaf) { /* assoc operation for bitmap nodes. Return: a new node, or self if key/val already is in the collection. 'added_leaf' is later used in '_PyHamt_Assoc' to determine if `hamt.set(key, val)` increased the size of the collection. */ uint32_t bit = hamt_bitpos(hash, shift); uint32_t idx = hamt_bitindex(self->b_bitmap, bit); /* Bitmap node layout: +------+------+------+------+ --- +------+------+ | key1 | val1 | key2 | val2 | ... | keyN | valN | +------+------+------+------+ --- +------+------+ where `N < Py_SIZE(node)`. The `node->b_bitmap` field is a bitmap. For a given `(shift, hash)` pair we can determine: - If this node has the corresponding key/val slots. - The index of key/val slots. */ if (self->b_bitmap & bit) { /* The key is set in this node */ uint32_t key_idx = 2 * idx; uint32_t val_idx = key_idx + 1; assert(val_idx < (size_t)Py_SIZE(self)); PyObject *key_or_null = self->b_array[key_idx]; PyObject *val_or_node = self->b_array[val_idx]; if (key_or_null == NULL) { /* key is NULL. This means that we have a few keys that have the same (hash, shift) pair. */ assert(val_or_node != NULL); PyHamtNode *sub_node = hamt_node_assoc( (PyHamtNode *)val_or_node, shift + 5, hash, key, val, added_leaf); if (sub_node == NULL) { return NULL; } if (val_or_node == (PyObject *)sub_node) { Py_DECREF(sub_node); return (PyHamtNode *)Py_NewRef(self); } PyHamtNode_Bitmap *ret = hamt_node_bitmap_clone(self); if (ret == NULL) { Py_DECREF(sub_node); return NULL; } Py_SETREF(ret->b_array[val_idx], (PyObject*)sub_node); return (PyHamtNode *)ret; } assert(key != NULL); /* key is not NULL. This means that we have only one other key in this collection that matches our hash for this shift. */ int comp_err = PyObject_RichCompareBool(key, key_or_null, Py_EQ); if (comp_err < 0) { /* exception in __eq__ */ return NULL; } if (comp_err == 1) { /* key == key_or_null */ if (val == val_or_node) { /* we already have the same key/val pair; return self. */ return (PyHamtNode *)Py_NewRef(self); } /* We're setting a new value for the key we had before. Make a new bitmap node with a replaced value, and return it. */ PyHamtNode_Bitmap *ret = hamt_node_bitmap_clone(self); if (ret == NULL) { return NULL; } Py_SETREF(ret->b_array[val_idx], Py_NewRef(val)); return (PyHamtNode *)ret; } /* It's a new key, and it has the same index as *one* another key. We have a collision. We need to create a new node which will combine the existing key and the key we're adding. `hamt_node_new_bitmap_or_collision` will either create a new Collision node if the keys have identical hashes, or a new Bitmap node. */ PyHamtNode *sub_node = hamt_node_new_bitmap_or_collision( shift + 5, key_or_null, val_or_node, /* existing key/val */ hash, key, val /* new key/val */ ); if (sub_node == NULL) { return NULL; } PyHamtNode_Bitmap *ret = hamt_node_bitmap_clone(self); if (ret == NULL) { Py_DECREF(sub_node); return NULL; } Py_SETREF(ret->b_array[key_idx], NULL); Py_SETREF(ret->b_array[val_idx], (PyObject *)sub_node); *added_leaf = 1; return (PyHamtNode *)ret; } else { /* There was no key before with the same (shift,hash). */ uint32_t n = (uint32_t)_Py_popcount32(self->b_bitmap); if (n >= 16) { /* When we have a situation where we want to store more than 16 nodes at one level of the tree, we no longer want to use the Bitmap node with bitmap encoding. Instead we start using an Array node, which has simpler (faster) implementation at the expense of having preallocated 32 pointers for its keys/values pairs. Small hamt objects (<30 keys) usually don't have any Array nodes at all. Between ~30 and ~400 keys hamt objects usually have one Array node, and usually it's a root node. */ uint32_t jdx = hamt_mask(hash, shift); /* 'jdx' is the index of where the new key should be added in the new Array node we're about to create. */ PyHamtNode *empty = NULL; PyHamtNode_Array *new_node = NULL; PyHamtNode *res = NULL; /* Create a new Array node. */ new_node = (PyHamtNode_Array *)hamt_node_array_new(n + 1); if (new_node == NULL) { goto fin; } /* Create an empty bitmap node for the next hamt_node_assoc call. */ empty = hamt_node_bitmap_new(0); if (empty == NULL) { goto fin; } /* Make a new bitmap node for the key/val we're adding. Set that bitmap node to new-array-node[jdx]. */ new_node->a_array[jdx] = hamt_node_assoc( empty, shift + 5, hash, key, val, added_leaf); if (new_node->a_array[jdx] == NULL) { goto fin; } /* Copy existing key/value pairs from the current Bitmap node to the new Array node we've just created. */ Py_ssize_t i, j; for (i = 0, j = 0; i < HAMT_ARRAY_NODE_SIZE; i++) { if (((self->b_bitmap >> i) & 1) != 0) { /* Ensure we don't accidentally override `jdx` element we set few lines above. */ assert(new_node->a_array[i] == NULL); if (self->b_array[j] == NULL) { new_node->a_array[i] = (PyHamtNode *)Py_NewRef(self->b_array[j + 1]); } else { int32_t rehash = hamt_hash(self->b_array[j]); if (rehash == -1) { goto fin; } new_node->a_array[i] = hamt_node_assoc( empty, shift + 5, rehash, self->b_array[j], self->b_array[j + 1], added_leaf); if (new_node->a_array[i] == NULL) { goto fin; } } j += 2; } } VALIDATE_ARRAY_NODE(new_node) /* That's it! */ res = (PyHamtNode *)new_node; fin: Py_XDECREF(empty); if (res == NULL) { Py_XDECREF(new_node); } return res; } else { /* We have less than 16 keys at this level; let's just create a new bitmap node out of this node with the new key/val pair added. */ uint32_t key_idx = 2 * idx; uint32_t val_idx = key_idx + 1; uint32_t i; *added_leaf = 1; /* Allocate new Bitmap node which can have one more key/val pair in addition to what we have already. */ PyHamtNode_Bitmap *new_node = (PyHamtNode_Bitmap *)hamt_node_bitmap_new(2 * (n + 1)); if (new_node == NULL) { return NULL; } /* Copy all keys/values that will be before the new key/value we are adding. */ for (i = 0; i < key_idx; i++) { new_node->b_array[i] = Py_XNewRef(self->b_array[i]); } /* Set the new key/value to the new Bitmap node. */ new_node->b_array[key_idx] = Py_NewRef(key); new_node->b_array[val_idx] = Py_NewRef(val); /* Copy all keys/values that will be after the new key/value we are adding. */ assert(Py_SIZE(self) >= 0 && Py_SIZE(self) <= 32); for (i = key_idx; i < (uint32_t)Py_SIZE(self); i++) { new_node->b_array[i + 2] = Py_XNewRef(self->b_array[i]); } new_node->b_bitmap = self->b_bitmap | bit; return (PyHamtNode *)new_node; } } } static hamt_without_t hamt_node_bitmap_without(PyHamtNode_Bitmap *self, uint32_t shift, int32_t hash, PyObject *key, PyHamtNode **new_node) { uint32_t bit = hamt_bitpos(hash, shift); if ((self->b_bitmap & bit) == 0) { return W_NOT_FOUND; } uint32_t idx = hamt_bitindex(self->b_bitmap, bit); uint32_t key_idx = 2 * idx; uint32_t val_idx = key_idx + 1; PyObject *key_or_null = self->b_array[key_idx]; PyObject *val_or_node = self->b_array[val_idx]; if (key_or_null == NULL) { /* key == NULL means that 'value' is another tree node. */ PyHamtNode *sub_node = NULL; hamt_without_t res = hamt_node_without( (PyHamtNode *)val_or_node, shift + 5, hash, key, &sub_node); switch (res) { case W_EMPTY: /* It's impossible for us to receive a W_EMPTY here: - Array nodes are converted to Bitmap nodes when we delete 16th item from them; - Collision nodes are converted to Bitmap when there is one item in them; - Bitmap node's without() inlines single-item sub-nodes. So in no situation we can have a single-item Bitmap child of another Bitmap node. */ Py_UNREACHABLE(); case W_NEWNODE: { assert(sub_node != NULL); if (IS_BITMAP_NODE(sub_node)) { PyHamtNode_Bitmap *sub_tree = (PyHamtNode_Bitmap *)sub_node; if (hamt_node_bitmap_count(sub_tree) == 1 && sub_tree->b_array[0] != NULL) { /* A bitmap node with one key/value pair. Just merge it into this node. Note that we don't inline Bitmap nodes that have a NULL key -- those nodes point to another tree level, and we cannot simply move tree levels up or down. */ PyHamtNode_Bitmap *clone = hamt_node_bitmap_clone(self); if (clone == NULL) { Py_DECREF(sub_node); return W_ERROR; } PyObject *key = sub_tree->b_array[0]; PyObject *val = sub_tree->b_array[1]; Py_XSETREF(clone->b_array[key_idx], Py_NewRef(key)); Py_SETREF(clone->b_array[val_idx], Py_NewRef(val)); Py_DECREF(sub_tree); *new_node = (PyHamtNode *)clone; return W_NEWNODE; } } #ifdef Py_DEBUG /* Ensure that Collision.without implementation converts to Bitmap nodes itself. */ if (IS_COLLISION_NODE(sub_node)) { assert(hamt_node_collision_count( (PyHamtNode_Collision*)sub_node) > 1); } |
| You are visitor 01402503 — GeezerVison does not store the pages you look up. Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required. |