2 Copyright (C) 2000, 2001 Free Software Foundation, Inc.
4 This file is part of Wget.
6 This program 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 This program 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 this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
28 #endif /* HAVE_STRING_H */
42 # define xmalloc malloc
43 # define xrealloc realloc
49 Hash tables are an implementation technique used to implement
50 mapping between objects. Provided a good hashing function is used,
51 they guarantee constant-time access and storing of information.
52 Duplicate keys are not allowed.
54 The basics are all covered. hash_table_new creates a hash table,
55 and hash_table_destroy deletes it. hash_table_put establishes a
56 mapping between a key and a value. hash_table_get retrieves the
57 value that corresponds to a key. hash_table_exists queries whether
58 a key is stored in a table at all. hash_table_remove removes a
59 mapping that corresponds to a key. hash_table_map allows you to
60 map through all the entries in a hash table. hash_table_clear
61 clears all the entries from the hash table.
63 The number of mappings in a table is not limited, except by the
64 amount of memory. As you add new elements to a table, it regrows
65 as necessary. If you have an idea about how many elements you will
66 store, you can provide a hint to hash_table_new().
68 The hashing and equality functions are normally provided by the
69 user. For the special (and frequent) case of hashing strings, you
70 can use the pre-canned make_string_hash_table(), which provides an
71 efficient string hashing function, and a string equality wrapper
74 When specifying your own hash and test functions, make sure the
77 - The test function returns non-zero for keys that are considered
78 "equal", zero otherwise.
80 - The hash function returns a number that represents the
81 "distinctness" of the object. In more precise terms, it means
82 that for any two objects that test "equal" under the test
83 function, the hash function MUST produce the same result.
85 This does not mean that each distinct object must produce a
86 distinct value, only that non-distinct objects must produce the
87 same values! For instance, a hash function that returns 0 for
88 any given object is a perfectly valid (albeit extremely bad) hash
89 function. A hash function that hashes a string by adding up all
90 its characters is another example of a valid (but quite bad) hash
93 The above stated rule is quite easy to enforce. For example, if
94 your testing function compares strings case-insensitively, all
95 your function needs to do is lower-case the string characters
96 before calculating a hash. That way you have easily guaranteed
97 that case differences will not result in a different hash.
99 - (optional) Choose the hash function to get as good "spreading" as
100 possible. A good hash function will react to even a small change
101 in its input with a completely different resulting hash.
102 Finally, don't make your hash function extremely slow, because
103 you're then defeating the purpose of hashing.
105 Note that neither keys nor values are copied when inserted into the
106 hash table, so they must exist for the lifetime of the table. This
107 means that e.g. the use of static strings is OK, but objects with a
108 shorter life-time need to be copied (with strdup() or the like in
109 the case of strings) before being inserted. */
113 All the hash mappings (key-value pairs of pointers) are stored in a
114 contiguous array. The position of each mapping is determined by
115 applying the hash function to the key: location = hash(key) % size.
116 If two different keys end up on the same position, the collision is
117 resolved by placing the second mapping at the next empty place in
118 the array following the occupied place. This method of collision
119 resolution is called "linear probing".
121 There are more advanced collision resolution mechanisms (quadratic
122 probing, double hashing), but we don't use them because they
123 involve more non-sequential access to the array, and therefore
124 worse cache behavior. Linear probing works well as long as the
125 fullness/size ratio is kept below 75%. We make sure to regrow or
126 rehash the hash table whenever this threshold is exceeded.
128 Collisions make deletion tricky because finding collisions again
129 relies on new empty spots not being created. That's why
130 hash_table_remove is careful to rehash the mappings that follow the
139 unsigned long (*hash_function) (const void *);
140 int (*test_function) (const void *, const void *);
142 int size; /* size of the array */
143 int count; /* number of non-empty, non-deleted
146 int resize_threshold; /* after size exceeds this number of
147 entries, resize the table. */
149 struct mapping *mappings;
152 #define EMPTY_MAPPING_P(mp) ((mp)->key == NULL)
153 #define NEXT_MAPPING(mp, mappings, size) (mp == mappings + (size - 1) \
156 #define LOOP_NON_EMPTY(mp, mappings, size) \
157 for (; !EMPTY_MAPPING_P (mp); mp = NEXT_MAPPING (mp, mappings, size))
159 #define HASH_POSITION(ht, key) (ht->hash_function (key) % ht->size)
161 /* Find a prime near, but greather than or equal to SIZE. */
164 prime_size (int size)
166 static const unsigned long primes [] = {
167 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
168 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
169 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
170 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
171 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
172 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
173 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
174 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
175 1174703521, 1527114613, 1985248999, 2580823717UL, 3355070839UL
178 for (i = 0; i < ARRAY_SIZE (primes); i++)
179 if (primes[i] >= size)
185 /* Create a hash table of INITIAL_SIZE with hash function
186 HASH_FUNCTION and test function TEST_FUNCTION. INITIAL_SIZE will
187 be rounded to the next prime, so you don't have to worry about it
188 being a prime number.
190 Consequently, if you wish to start out with a "small" table which
191 will be regrown as needed, specify INITIAL_SIZE 0. */
194 hash_table_new (int initial_size,
195 unsigned long (*hash_function) (const void *),
196 int (*test_function) (const void *, const void *))
198 struct hash_table *ht
199 = (struct hash_table *)xmalloc (sizeof (struct hash_table));
201 ht->hash_function = hash_function;
202 ht->test_function = test_function;
204 ht->size = prime_size (initial_size);
205 ht->resize_threshold = ht->size * 3 / 4;
209 ht->mappings = xmalloc (ht->size * sizeof (struct mapping));
210 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
215 /* Free the data associated with hash table HT. */
218 hash_table_destroy (struct hash_table *ht)
220 xfree (ht->mappings);
224 /* The heart of almost all functions in this file -- find the mapping
225 whose KEY is equal to key, using linear probing. Returns the
226 mapping that matches KEY, or NULL if none matches. */
228 static inline struct mapping *
229 find_mapping (struct hash_table *ht, const void *key)
231 struct mapping *mappings = ht->mappings;
233 struct mapping *mp = mappings + HASH_POSITION (ht, key);
234 int (*equals) (const void *, const void *) = ht->test_function;
236 LOOP_NON_EMPTY (mp, mappings, size)
237 if (equals (key, mp->key))
242 /* Get the value that corresponds to the key KEY in the hash table HT.
243 If no value is found, return NULL. Note that NULL is a legal value
244 for value; if you are storing NULLs in your hash table, you can use
245 hash_table_exists to be sure that a (possibly NULL) value exists in
246 the table. Or, you can use hash_table_get_pair instead of this
250 hash_table_get (struct hash_table *ht, const void *key)
252 struct mapping *mp = find_mapping (ht, key);
259 /* Like hash_table_get, but writes out the pointers to both key and
260 value. Returns non-zero on success. */
263 hash_table_get_pair (struct hash_table *ht, const void *lookup_key,
264 void *orig_key, void *value)
266 struct mapping *mp = find_mapping (ht, lookup_key);
271 *(void **)orig_key = mp->key;
273 *(void **)value = mp->value;
280 /* Return 1 if KEY exists in HT, 0 otherwise. */
283 hash_table_exists (struct hash_table *ht, const void *key)
285 return find_mapping (ht, key) != NULL;
288 /* Grow hash table HT as necessary, and rehash all the key-value
292 grow_hash_table (struct hash_table *ht)
294 struct mapping *old_mappings = ht->mappings;
295 struct mapping *old_end = ht->mappings + ht->size;
296 struct mapping *mp, *mappings;
299 newsize = prime_size (ht->size * 2);
301 printf ("growing from %d to %d\n", ht->size, newsize);
305 ht->resize_threshold = newsize * 3 / 4;
307 mappings = xmalloc (ht->size * sizeof (struct mapping));
308 memset (mappings, '\0', ht->size * sizeof (struct mapping));
309 ht->mappings = mappings;
311 for (mp = old_mappings; mp < old_end; mp++)
312 if (!EMPTY_MAPPING_P (mp))
314 struct mapping *new_mp = mappings + HASH_POSITION (ht, mp->key);
315 /* We don't need to call test function and worry about
316 collisions because all the keys come from the hash table
317 and are therefore guaranteed to be unique. */
318 LOOP_NON_EMPTY (new_mp, mappings, newsize)
323 xfree (old_mappings);
326 /* Put VALUE in the hash table HT under the key KEY. This regrows the
327 table if necessary. */
330 hash_table_put (struct hash_table *ht, const void *key, void *value)
332 struct mapping *mappings = ht->mappings;
334 int (*equals) (const void *, const void *) = ht->test_function;
336 struct mapping *mp = mappings + HASH_POSITION (ht, key);
338 LOOP_NON_EMPTY (mp, mappings, size)
339 if (equals (key, mp->key))
341 mp->key = (void *)key; /* const? */
347 mp->key = (void *)key; /* const? */
350 if (ht->count > ht->resize_threshold)
351 /* When table is 75% full, regrow it. */
352 grow_hash_table (ht);
355 /* Remove a mapping that matches KEY from HT. Return 0 if there was
356 no such entry; return 1 if an entry was removed. */
359 hash_table_remove (struct hash_table *ht, const void *key)
361 struct mapping *mp = find_mapping (ht, key);
367 struct mapping *mappings = ht->mappings;
372 /* Rehash all the entries following MP. The alternative
373 approach is to mark the entry as deleted, i.e. create a
374 "tombstone". That makes remove faster, but leaves a lot of
375 garbage and slows down hash_table_get and hash_table_put. */
377 mp = NEXT_MAPPING (mp, mappings, size);
378 LOOP_NON_EMPTY (mp, mappings, size)
380 const void *key2 = mp->key;
381 struct mapping *mp_new = mappings + HASH_POSITION (ht, key2);
383 /* Find the new location for the key. */
385 LOOP_NON_EMPTY (mp_new, mappings, size)
386 if (key2 == mp_new->key)
387 /* The mapping MP (key2) is already where we want it (in
388 MP_NEW's "chain" of keys.) */
401 /* Clear HT of all entries. After calling this function, the count
402 and the fullness of the hash table will be zero. The size will
406 hash_table_clear (struct hash_table *ht)
408 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
412 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
413 called with three arguments: the key, the value, and the CLOSURE.
415 It is undefined what happens if you add or remove entries in the
416 hash table while hash_table_map is running. The exception is the
417 entry you're currently mapping over; you may remove or change that
421 hash_table_map (struct hash_table *ht,
422 int (*mapfun) (void *, void *, void *),
425 struct mapping *mp = ht->mappings;
426 struct mapping *end = ht->mappings + ht->size;
428 for (; mp < end; mp++)
429 if (!EMPTY_MAPPING_P (mp))
434 if (mapfun (key, mp->value, closure))
436 /* hash_table_remove might have moved the adjacent
438 if (mp->key != key && !EMPTY_MAPPING_P (mp))
443 /* Return the number of elements in the hash table. This is not the
444 same as the physical size of the hash table, which is always
445 greater than the number of elements. */
448 hash_table_count (struct hash_table *ht)
453 /* Support for hash tables whose keys are strings. */
455 /* 31 bit hash function. Taken from Gnome's glib, modified to use
458 We used to use the popular hash function from the Dragon Book, but
459 this one seems to perform much better. */
462 string_hash (const void *key)
468 for (p += 1; *p != '\0'; p++)
469 h = (h << 5) - h + *p;
475 /* If I ever need it: hashing of integers. */
478 inthash (unsigned int key)
493 string_cmp (const void *s1, const void *s2)
495 return !strcmp ((const char *)s1, (const char *)s2);
498 /* Return a hash table of initial size INITIAL_SIZE suitable to use
502 make_string_hash_table (int initial_size)
504 return hash_table_new (initial_size, string_hash, string_cmp);
514 print_hash_table_mapper (void *key, void *value, void *count)
517 printf ("%s: %s\n", (const char *)key, (char *)value);
522 print_hash (struct hash_table *sht)
525 hash_table_map (sht, print_hash_table_mapper, &debug_count);
526 assert (debug_count == sht->count);
532 struct hash_table *ht = make_string_hash_table (0);
534 while ((fgets (line, sizeof (line), stdin)))
536 int len = strlen (line);
540 if (!hash_table_exists (ht, line))
541 hash_table_put (ht, strdup (line), "here I am!");
546 if (hash_table_get_pair (ht, line, &line_copy, NULL))
548 hash_table_remove (ht, line);
558 printf ("%d %d\n", ht->count, ht->size);