2 Copyright (C) 2000-2003 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or (at
9 your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
30 /* With -DSTANDALONE, this file can be compiled outside Wget source
31 tree. To test, also use -DTEST. */
44 /* Get Wget's utility headers. */
48 /* Make do without them. */
49 # define xnew(x) xmalloc (sizeof (x))
50 # define xnew_array(type, x) xmalloc (sizeof (type) * (x))
51 # define xmalloc malloc
53 # define countof(x) (sizeof (x) / sizeof ((x)[0]))
54 # define TOLOWER(x) ('A' <= (x) && (x) <= 'Z' ? (x) - 32 : (x))
61 Hash tables are a technique used to implement mapping between
62 objects with near-constant-time access and storage. The table
63 associates keys to values, and a value can be very quickly
64 retrieved by providing the key. Fast lookup tables are typically
65 implemented as hash tables.
68 hash_table_new -- creates the table.
69 hash_table_destroy -- destroys the table.
70 hash_table_put -- establishes or updates key->value mapping.
71 hash_table_get -- retrieves value of key.
72 hash_table_get_pair -- get key/value pair for key.
73 hash_table_contains -- test whether the table contains key.
74 hash_table_remove -- remove the key->value mapping for key.
75 hash_table_map -- iterate through table mappings.
76 hash_table_clear -- clear hash table contents.
77 hash_table_count -- return the number of entries in the table.
79 The hash table grows internally as new entries are added and is not
80 limited in size, except by available memory. The table doubles
81 with each resize, which ensures that the amortized time per
82 operation remains constant.
84 By default, tables created by hash_table_new consider the keys to
85 be equal if their pointer values are the same. You can use
86 make_string_hash_table to create tables whose keys are considered
87 equal if their string contents are the same. In the general case,
88 the criterion of equality used to compare keys is specified at
89 table creation time with two callback functions, "hash" and "test".
90 The hash function transforms the key into an arbitrary number that
91 must be the same for two equal keys. The test function accepts two
92 keys and returns non-zero if they are to be considered equal.
94 Note that neither keys nor values are copied when inserted into the
95 hash table, so they must exist for the lifetime of the table. This
96 means that e.g. the use of static strings is OK, but objects with a
97 shorter life-time need to be copied (with strdup() or the like in
98 the case of strings) before being inserted. */
102 The hash table is implemented as an open-addressed table with
103 linear probing collision resolution.
105 The above means that all the hash entries (pairs of pointers, key
106 and value) are stored in a contiguous array. The position of each
107 mapping is determined by the hash value of its key and the size of
108 the table: location := hash(key) % size. If two different keys end
109 up on the same position (collide), the one that came second is
110 placed at the next empty position following the occupied place.
111 This collision resolution technique is called "linear probing".
113 There are more advanced collision resolution methods (quadratic
114 probing, double hashing), but we don't use them because they incur
115 more non-sequential access to the array, which results in worse CPU
116 cache behavior. Linear probing works well as long as the
117 count/size ratio (fullness) is kept below 75%. We make sure to
118 grow and rehash the table whenever this threshold is exceeded.
120 Collisions make deletion tricky because clearing a position
121 followed by a colliding entry would make the position seem empty
122 and the colliding entry not found. One solution is to leave a
123 "tombstone" instead of clearing the entry, and another is to
124 carefully rehash the entries immediately following the deleted one.
125 We use the latter method because it results in less bookkeeping and
126 faster retrieval at the (slight) expense of deletion. */
128 /* Maximum allowed fullness: when hash table's fullness exceeds this
129 value, the table is resized. */
130 #define HASH_MAX_FULLNESS 0.75
132 /* The hash table size is multiplied by this factor (and then rounded
133 to the next prime) with each resize. This guarantees infrequent
135 #define HASH_RESIZE_FACTOR 2
142 typedef unsigned long (*hashfun_t) (const void *);
143 typedef int (*testfun_t) (const void *, const void *);
146 hashfun_t hash_function;
147 testfun_t test_function;
149 struct mapping *mappings; /* pointer to the table entries. */
150 int size; /* size of the array. */
152 int count; /* number of non-empty entries. */
153 int resize_threshold; /* after size exceeds this number of
154 entries, resize the table. */
155 int prime_offset; /* the offset of the current prime in
159 /* We use the all-bits-set constant (INVALID_PTR) marker to mean that
160 a mapping is empty. It is unaligned and therefore illegal as a
161 pointer. INVALID_PTR_BYTE (0xff) is the one-byte value used to
162 initialize the mappings array as empty.
164 The all-bits-set value is a better choice than NULL because it
165 allows the use of NULL/0 keys. Since the keys are either integers
166 or pointers, the only key that cannot be used is the integer value
167 -1. This is acceptable because it still allows the use of
168 nonnegative integer keys. */
170 #define INVALID_PTR ((void *) ~0UL)
172 # define UCHAR_MAX 0xff
174 #define INVALID_PTR_BYTE UCHAR_MAX
176 #define NON_EMPTY(mp) ((mp)->key != INVALID_PTR)
177 #define MARK_AS_EMPTY(mp) ((mp)->key = INVALID_PTR)
179 /* "Next" mapping is the mapping after MP, but wrapping back to
180 MAPPINGS when MP would reach MAPPINGS+SIZE. */
181 #define NEXT_MAPPING(mp, mappings, size) (mp != mappings + (size - 1) \
184 /* Loop over non-empty mappings starting at MP. */
185 #define LOOP_NON_EMPTY(mp, mappings, size) \
186 for (; NON_EMPTY (mp); mp = NEXT_MAPPING (mp, mappings, size))
188 /* Return the position of KEY in hash table SIZE large, hash function
190 #define HASH_POSITION(key, hashfun, size) ((hashfun) (key) % size)
192 /* Find a prime near, but greather than or equal to SIZE. The primes
193 are looked up from a table with a selection of primes convenient
196 PRIME_OFFSET is a minor optimization: it specifies start position
197 for the search for the large enough prime. The final offset is
198 stored in the same variable. That way the list of primes does not
199 have to be scanned from the beginning each time around. */
202 prime_size (int size, int *prime_offset)
204 static const int primes[] = {
205 13, 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
206 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
207 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
208 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
209 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
210 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
211 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
212 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
213 1174703521, 1527114613, 1837299131, 2147483647
217 for (i = *prime_offset; i < countof (primes); i++)
218 if (primes[i] >= size)
220 /* Set the offset to the next prime. That is safe because,
221 next time we are called, it will be with a larger SIZE,
222 which means we could never return the same prime anyway.
223 (If that is not the case, the caller can simply reset
225 *prime_offset = i + 1;
232 static int cmp_pointer (const void *, const void *);
234 /* Create a hash table with hash function HASH_FUNCTION and test
235 function TEST_FUNCTION. The table is empty (its count is 0), but
236 pre-allocated to store at least ITEMS items.
238 ITEMS is the number of items that the table can accept without
239 needing to resize. It is useful when creating a table that is to
240 be immediately filled with a known number of items. In that case,
241 the regrows are a waste of time, and specifying ITEMS correctly
242 will avoid them altogether.
244 Note that hash tables grow dynamically regardless of ITEMS. The
245 only use of ITEMS is to preallocate the table and avoid unnecessary
246 dynamic regrows. Don't bother making ITEMS prime because it's not
247 used as size unchanged. To start with a small table that grows as
248 needed, simply specify zero ITEMS.
250 If hash and test callbacks are not specified, identity mapping is
251 assumed, i.e. pointer values are used for key comparison. (Common
252 Lisp calls such tables EQ hash tables, and Java calls them
253 IdentityHashMaps.) If your keys require different comparison,
254 specify hash and test functions. For easy use of C strings as hash
255 keys, you can use the convenience functions make_string_hash_table
256 and make_nocase_string_hash_table. */
259 hash_table_new (int items,
260 unsigned long (*hash_function) (const void *),
261 int (*test_function) (const void *, const void *))
264 struct hash_table *ht = xnew (struct hash_table);
266 ht->hash_function = hash_function ? hash_function : hash_pointer;
267 ht->test_function = test_function ? test_function : cmp_pointer;
269 /* If the size of struct hash_table ever becomes a concern, this
270 field can go. (Wget doesn't create many hashes.) */
271 ht->prime_offset = 0;
273 /* Calculate the size that ensures that the table will store at
274 least ITEMS keys without the need to resize. */
275 size = 1 + items / HASH_MAX_FULLNESS;
276 size = prime_size (size, &ht->prime_offset);
278 ht->resize_threshold = size * HASH_MAX_FULLNESS;
279 /*assert (ht->resize_threshold >= items);*/
281 ht->mappings = xnew_array (struct mapping, ht->size);
283 /* Mark mappings as empty. We use 0xff rather than 0 to mark empty
284 keys because it allows us to use NULL/0 as keys. */
285 memset (ht->mappings, INVALID_PTR_BYTE, size * sizeof (struct mapping));
292 /* Free the data associated with hash table HT. */
295 hash_table_destroy (struct hash_table *ht)
297 xfree (ht->mappings);
301 /* The heart of most functions in this file -- find the mapping whose
302 KEY is equal to key, using linear probing. Returns the mapping
303 that matches KEY, or the first empty mapping if none matches. */
305 static inline struct mapping *
306 find_mapping (const struct hash_table *ht, const void *key)
308 struct mapping *mappings = ht->mappings;
310 struct mapping *mp = mappings + HASH_POSITION (key, ht->hash_function, size);
311 testfun_t equals = ht->test_function;
313 LOOP_NON_EMPTY (mp, mappings, size)
314 if (equals (key, mp->key))
319 /* Get the value that corresponds to the key KEY in the hash table HT.
320 If no value is found, return NULL. Note that NULL is a legal value
321 for value; if you are storing NULLs in your hash table, you can use
322 hash_table_contains to be sure that a (possibly NULL) value exists
323 in the table. Or, you can use hash_table_get_pair instead of this
327 hash_table_get (const struct hash_table *ht, const void *key)
329 struct mapping *mp = find_mapping (ht, key);
336 /* Like hash_table_get, but writes out the pointers to both key and
337 value. Returns non-zero on success. */
340 hash_table_get_pair (const struct hash_table *ht, const void *lookup_key,
341 void *orig_key, void *value)
343 struct mapping *mp = find_mapping (ht, lookup_key);
347 *(void **)orig_key = mp->key;
349 *(void **)value = mp->value;
356 /* Return 1 if HT contains KEY, 0 otherwise. */
359 hash_table_contains (const struct hash_table *ht, const void *key)
361 struct mapping *mp = find_mapping (ht, key);
362 return NON_EMPTY (mp);
365 /* Grow hash table HT as necessary, and rehash all the key-value
369 grow_hash_table (struct hash_table *ht)
371 hashfun_t hasher = ht->hash_function;
372 struct mapping *old_mappings = ht->mappings;
373 struct mapping *old_end = ht->mappings + ht->size;
374 struct mapping *mp, *mappings;
377 newsize = prime_size (ht->size * HASH_RESIZE_FACTOR, &ht->prime_offset);
379 printf ("growing from %d to %d; fullness %.2f%% to %.2f%%\n",
381 100.0 * ht->count / ht->size,
382 100.0 * ht->count / newsize);
386 ht->resize_threshold = newsize * HASH_MAX_FULLNESS;
388 mappings = xnew_array (struct mapping, newsize);
389 memset (mappings, INVALID_PTR_BYTE, newsize * sizeof (struct mapping));
390 ht->mappings = mappings;
392 for (mp = old_mappings; mp < old_end; mp++)
395 struct mapping *new_mp;
396 /* We don't need to test for uniqueness of keys because they
397 come from the hash table and are therefore known to be
399 new_mp = mappings + HASH_POSITION (mp->key, hasher, newsize);
400 LOOP_NON_EMPTY (new_mp, mappings, newsize)
405 xfree (old_mappings);
408 /* Put VALUE in the hash table HT under the key KEY. This regrows the
409 table if necessary. */
412 hash_table_put (struct hash_table *ht, const void *key, void *value)
414 struct mapping *mp = find_mapping (ht, key);
417 /* update existing item */
418 mp->key = (void *)key; /* const? */
423 /* If adding the item would make the table exceed max. fullness,
424 grow the table first. */
425 if (ht->count >= ht->resize_threshold)
427 grow_hash_table (ht);
428 mp = find_mapping (ht, key);
433 mp->key = (void *)key; /* const? */
437 /* Remove a mapping that matches KEY from HT. Return 0 if there was
438 no such entry; return 1 if an entry was removed. */
441 hash_table_remove (struct hash_table *ht, const void *key)
443 struct mapping *mp = find_mapping (ht, key);
449 struct mapping *mappings = ht->mappings;
450 hashfun_t hasher = ht->hash_function;
455 /* Rehash all the entries following MP. The alternative
456 approach is to mark the entry as deleted, i.e. create a
457 "tombstone". That speeds up removal, but leaves a lot of
458 garbage and slows down hash_table_get and hash_table_put. */
460 mp = NEXT_MAPPING (mp, mappings, size);
461 LOOP_NON_EMPTY (mp, mappings, size)
463 const void *key2 = mp->key;
464 struct mapping *mp_new;
466 /* Find the new location for the key. */
467 mp_new = mappings + HASH_POSITION (key2, hasher, size);
468 LOOP_NON_EMPTY (mp_new, mappings, size)
469 if (key2 == mp_new->key)
470 /* The mapping MP (key2) is already where we want it (in
471 MP_NEW's "chain" of keys.) */
484 /* Clear HT of all entries. After calling this function, the count
485 and the fullness of the hash table will be zero. The size will
489 hash_table_clear (struct hash_table *ht)
491 memset (ht->mappings, INVALID_PTR_BYTE, ht->size * sizeof (struct mapping));
495 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
496 called with three arguments: the key, the value, and MAPARG.
498 It is undefined what happens if you add or remove entries in the
499 hash table while hash_table_map is running. The exception is the
500 entry you're currently mapping over; you may remove or change that
504 hash_table_map (struct hash_table *ht,
505 int (*mapfun) (void *, void *, void *),
508 struct mapping *mp = ht->mappings;
509 struct mapping *end = ht->mappings + ht->size;
511 for (; mp < end; mp++)
517 if (mapfun (key, mp->value, maparg))
519 /* hash_table_remove might have moved the adjacent
521 if (mp->key != key && NON_EMPTY (mp))
526 /* Return the number of elements in the hash table. This is not the
527 same as the physical size of the hash table, which is always
528 greater than the number of elements. */
531 hash_table_count (const struct hash_table *ht)
536 /* Functions from this point onward are meant for convenience and
537 don't strictly belong to this file. However, this is as good a
538 place for them as any. */
540 /* Guidelines for creating custom hash and test functions:
542 - The test function returns non-zero for keys that are considered
543 "equal", zero otherwise.
545 - The hash function returns a number that represents the
546 "distinctness" of the object. In more precise terms, it means
547 that for any two objects that test "equal" under the test
548 function, the hash function MUST produce the same result.
550 This does not mean that all different objects must produce
551 different values (that would be "perfect" hashing), only that
552 non-distinct objects must produce the same values! For instance,
553 a hash function that returns 0 for any given object is a
554 perfectly valid (albeit extremely bad) hash function. A hash
555 function that hashes a string by adding up all its characters is
556 another example of a valid (but quite bad) hash function.
558 It is not hard to make hash and test functions agree about
559 equality. For example, if the test function compares strings
560 case-insensitively, the hash function can lower-case the
561 characters when calculating the hash value. That ensures that
562 two strings differing only in case will hash the same.
564 - If you care about performance, choose a hash function with as
565 good "spreading" as possible. A good hash function will use all
566 the bits of the input when calculating the hash, and will react
567 to even small changes in input with a completely different
568 output. Finally, don't make the hash function itself overly
569 slow, because you'll be incurring a non-negligible overhead to
570 all hash table operations. */
573 * Support for hash tables whose keys are strings.
577 /* 31 bit hash function. Taken from Gnome's glib, modified to use
580 We used to use the popular hash function from the Dragon Book, but
581 this one seems to perform much better. */
584 hash_string (const void *key)
590 for (p += 1; *p != '\0'; p++)
591 h = (h << 5) - h + *p;
596 /* Frontend for strcmp usable for hash tables. */
599 cmp_string (const void *s1, const void *s2)
601 return !strcmp ((const char *)s1, (const char *)s2);
604 /* Return a hash table of preallocated to store at least ITEMS items
605 suitable to use strings as keys. */
608 make_string_hash_table (int items)
610 return hash_table_new (items, hash_string, cmp_string);
614 * Support for hash tables whose keys are strings, but which are
615 * compared case-insensitively.
619 /* Like hash_string, but produce the same hash regardless of the case. */
622 hash_string_nocase (const void *key)
625 unsigned int h = TOLOWER (*p);
628 for (p += 1; *p != '\0'; p++)
629 h = (h << 5) - h + TOLOWER (*p);
634 /* Like string_cmp, but doing case-insensitive compareison. */
637 string_cmp_nocase (const void *s1, const void *s2)
639 return !strcasecmp ((const char *)s1, (const char *)s2);
642 /* Like make_string_hash_table, but uses string_hash_nocase and
643 string_cmp_nocase. */
646 make_nocase_string_hash_table (int items)
648 return hash_table_new (items, hash_string_nocase, string_cmp_nocase);
651 /* Hashing of numeric values, such as pointers and integers.
653 This implementation is the Robert Jenkins' 32 bit Mix Function,
654 with a simple adaptation for 64-bit values. It offers excellent
655 spreading of values and doesn't need to know the hash table size to
656 work (unlike the very popular Knuth's multiplication hash). */
659 hash_pointer (const void *ptr)
661 unsigned long key = (unsigned long)ptr;
684 cmp_pointer (const void *ptr1, const void *ptr2)
695 print_hash_table_mapper (void *key, void *value, void *count)
698 printf ("%s: %s\n", (const char *)key, (char *)value);
703 print_hash (struct hash_table *sht)
706 hash_table_map (sht, print_hash_table_mapper, &debug_count);
707 assert (debug_count == sht->count);
713 struct hash_table *ht = make_string_hash_table (0);
715 while ((fgets (line, sizeof (line), stdin)))
717 int len = strlen (line);
721 if (!hash_table_contains (ht, line))
722 hash_table_put (ht, strdup (line), "here I am!");
727 if (hash_table_get_pair (ht, line, &line_copy, NULL))
729 hash_table_remove (ht, line);
739 printf ("%d %d\n", ht->count, ht->size);