2 Copyright (C) 2000, 2001 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. */
38 #endif /* HAVE_STRING_H */
52 # define xmalloc malloc
53 # define xrealloc realloc
57 # define TOLOWER(x) ('A' <= (x) && (x) <= 'Z' ? (x) - 32 : (x))
62 Hash tables are an implementation technique used to implement
63 mapping between objects. Provided a good hashing function is used,
64 they guarantee constant-time access and storing of information.
65 Duplicate keys are not allowed.
67 The basics are all covered. hash_table_new creates a hash table,
68 and hash_table_destroy deletes it. hash_table_put establishes a
69 mapping between a key and a value. hash_table_get retrieves the
70 value that corresponds to a key. hash_table_contains queries
71 whether a key is stored in a table at all. hash_table_remove
72 removes a mapping that corresponds to a key. hash_table_map allows
73 you to map through all the entries in a hash table.
74 hash_table_clear clears all the entries from the hash table.
76 The number of mappings in a table is not limited, except by the
77 amount of memory. As you add new elements to a table, it regrows
78 as necessary. If you have an idea about how many elements you will
79 store, you can provide a hint to hash_table_new().
81 The hashing and equality functions are normally provided by the
82 user. For the special (and frequent) case of hashing strings, you
83 can use the pre-canned make_string_hash_table(), which provides an
84 efficient string hashing function, and a string equality wrapper
87 When specifying your own hash and test functions, make sure the
90 - The test function returns non-zero for keys that are considered
91 "equal", zero otherwise.
93 - The hash function returns a number that represents the
94 "distinctness" of the object. In more precise terms, it means
95 that for any two objects that test "equal" under the test
96 function, the hash function MUST produce the same result.
98 This does not mean that each distinct object must produce a
99 distinct value, only that non-distinct objects must produce the
100 same values! For instance, a hash function that returns 0 for
101 any given object is a perfectly valid (albeit extremely bad) hash
102 function. A hash function that hashes a string by adding up all
103 its characters is another example of a valid (but quite bad) hash
106 The above stated rule is quite easy to enforce. For example, if
107 your testing function compares strings case-insensitively, all
108 your function needs to do is lower-case the string characters
109 before calculating a hash. That way you have easily guaranteed
110 that case differences will not result in a different hash.
112 - (optional) Choose the hash function to get as good "spreading" as
113 possible. A good hash function will react to even a small change
114 in its input with a completely different resulting hash.
115 Finally, don't make your hash function extremely slow, because
116 you're then defeating the purpose of hashing.
118 Note that neither keys nor values are copied when inserted into the
119 hash table, so they must exist for the lifetime of the table. This
120 means that e.g. the use of static strings is OK, but objects with a
121 shorter life-time need to be copied (with strdup() or the like in
122 the case of strings) before being inserted. */
126 All the hash mappings (key-value pairs of pointers) are stored in a
127 contiguous array. The position of each mapping is determined by
128 applying the hash function to the key: location = hash(key) % size.
129 If two different keys end up on the same position, the collision is
130 resolved by placing the second mapping at the next empty place in
131 the array following the occupied place. This method of collision
132 resolution is called "linear probing".
134 There are more advanced collision resolution mechanisms (quadratic
135 probing, double hashing), but we don't use them because they
136 involve more non-sequential access to the array, and therefore
137 worse cache behavior. Linear probing works well as long as the
138 fullness/size ratio is kept below 75%. We make sure to regrow or
139 rehash the hash table whenever this threshold is exceeded.
141 Collisions make deletion tricky because finding collisions again
142 relies on new empty spots not being created. That's why
143 hash_table_remove is careful to rehash the mappings that follow the
152 unsigned long (*hash_function) PARAMS ((const void *));
153 int (*test_function) PARAMS ((const void *, const void *));
155 int size; /* size of the array */
156 int count; /* number of non-empty, non-deleted
159 int resize_threshold; /* after size exceeds this number of
160 entries, resize the table. */
161 int prime_offset; /* the offset of the current prime in
164 struct mapping *mappings; /* the array of mapping pairs. */
167 #define EMPTY_MAPPING_P(mp) ((mp)->key == NULL)
168 #define NEXT_MAPPING(mp, mappings, size) (mp == mappings + (size - 1) \
171 #define LOOP_NON_EMPTY(mp, mappings, size) \
172 for (; !EMPTY_MAPPING_P (mp); mp = NEXT_MAPPING (mp, mappings, size))
174 /* #### We might want to multiply with the "golden ratio" here to get
175 better randomness for keys that do not result from a good hash
176 function. This is currently not a problem in Wget because we only
177 use the string hash tables. */
179 #define HASH_POSITION(ht, key) (ht->hash_function (key) % ht->size)
181 /* Find a prime near, but greather than or equal to SIZE. Of course,
182 the primes are not calculated, but looked up from a table. The
183 table does not contain all primes in range, just a selection useful
186 PRIME_OFFSET is a micro-optimization: if specified, it starts the
187 search for the prime number beginning with the specific offset in
188 the prime number table. The final offset is stored in the same
192 prime_size (int size, int *prime_offset)
194 static const unsigned long primes [] = {
195 13, 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
196 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
197 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
198 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
199 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
200 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
201 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
202 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
203 1174703521, 1527114613, 1985248999,
204 (unsigned long)0x99d43ea5, (unsigned long)0xc7fa5177
206 int i = *prime_offset;
208 for (; i < ARRAY_SIZE (primes); i++)
209 if (primes[i] >= size)
211 /* Set the offset to the next prime. That is safe because,
212 next time we are called, it will be with a larger SIZE,
213 which means we could never return the same prime anyway.
214 (If that is not the case, the caller can simply reset
216 *prime_offset = i + 1;
224 /* Create a hash table of INITIAL_SIZE with hash function
225 HASH_FUNCTION and test function TEST_FUNCTION. INITIAL_SIZE will
226 be rounded to the next prime, so you don't have to worry about it
227 being a prime number.
229 Consequently, if you wish to start out with a "small" table which
230 will be regrown as needed, specify INITIAL_SIZE 0. */
233 hash_table_new (int initial_size,
234 unsigned long (*hash_function) (const void *),
235 int (*test_function) (const void *, const void *))
237 struct hash_table *ht
238 = (struct hash_table *)xmalloc (sizeof (struct hash_table));
240 ht->hash_function = hash_function;
241 ht->test_function = test_function;
243 ht->prime_offset = 0;
244 ht->size = prime_size (initial_size, &ht->prime_offset);
245 ht->resize_threshold = ht->size * 3 / 4;
249 ht->mappings = xmalloc (ht->size * sizeof (struct mapping));
250 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
255 /* Free the data associated with hash table HT. */
258 hash_table_destroy (struct hash_table *ht)
260 xfree (ht->mappings);
264 /* The heart of almost all functions in this file -- find the mapping
265 whose KEY is equal to key, using linear probing. Returns the
266 mapping that matches KEY, or NULL if none matches. */
268 static inline struct mapping *
269 find_mapping (struct hash_table *ht, const void *key)
271 struct mapping *mappings = ht->mappings;
273 struct mapping *mp = mappings + HASH_POSITION (ht, key);
274 int (*equals) PARAMS ((const void *, const void *)) = ht->test_function;
276 LOOP_NON_EMPTY (mp, mappings, size)
277 if (equals (key, mp->key))
282 /* Get the value that corresponds to the key KEY in the hash table HT.
283 If no value is found, return NULL. Note that NULL is a legal value
284 for value; if you are storing NULLs in your hash table, you can use
285 hash_table_contains to be sure that a (possibly NULL) value exists
286 in the table. Or, you can use hash_table_get_pair instead of this
290 hash_table_get (struct hash_table *ht, const void *key)
292 struct mapping *mp = find_mapping (ht, key);
299 /* Like hash_table_get, but writes out the pointers to both key and
300 value. Returns non-zero on success. */
303 hash_table_get_pair (struct hash_table *ht, const void *lookup_key,
304 void *orig_key, void *value)
306 struct mapping *mp = find_mapping (ht, lookup_key);
311 *(void **)orig_key = mp->key;
313 *(void **)value = mp->value;
320 /* Return 1 if HT contains KEY, 0 otherwise. */
323 hash_table_contains (struct hash_table *ht, const void *key)
325 return find_mapping (ht, key) != NULL;
328 /* Grow hash table HT as necessary, and rehash all the key-value
332 grow_hash_table (struct hash_table *ht)
334 struct mapping *old_mappings = ht->mappings;
335 struct mapping *old_end = ht->mappings + ht->size;
336 struct mapping *mp, *mappings;
339 newsize = prime_size (ht->size * 2, &ht->prime_offset);
341 printf ("growing from %d to %d; fullness %.2f%% to %.2f%%\n",
343 (double)100 * ht->count / ht->size,
344 (double)100 * ht->count / newsize);
348 ht->resize_threshold = newsize * 3 / 4;
350 mappings = xmalloc (ht->size * sizeof (struct mapping));
351 memset (mappings, '\0', ht->size * sizeof (struct mapping));
352 ht->mappings = mappings;
354 for (mp = old_mappings; mp < old_end; mp++)
355 if (!EMPTY_MAPPING_P (mp))
357 struct mapping *new_mp = mappings + HASH_POSITION (ht, mp->key);
358 /* We don't need to call test function and worry about
359 collisions because all the keys come from the hash table
360 and are therefore guaranteed to be unique. */
361 LOOP_NON_EMPTY (new_mp, mappings, newsize)
366 xfree (old_mappings);
369 /* Put VALUE in the hash table HT under the key KEY. This regrows the
370 table if necessary. */
373 hash_table_put (struct hash_table *ht, const void *key, void *value)
375 struct mapping *mappings = ht->mappings;
377 int (*equals) PARAMS ((const void *, const void *)) = ht->test_function;
379 struct mapping *mp = mappings + HASH_POSITION (ht, key);
381 LOOP_NON_EMPTY (mp, mappings, size)
382 if (equals (key, mp->key))
384 mp->key = (void *)key; /* const? */
390 mp->key = (void *)key; /* const? */
393 if (ht->count > ht->resize_threshold)
394 /* When table is 75% full, regrow it. */
395 grow_hash_table (ht);
398 /* Remove a mapping that matches KEY from HT. Return 0 if there was
399 no such entry; return 1 if an entry was removed. */
402 hash_table_remove (struct hash_table *ht, const void *key)
404 struct mapping *mp = find_mapping (ht, key);
410 struct mapping *mappings = ht->mappings;
415 /* Rehash all the entries following MP. The alternative
416 approach is to mark the entry as deleted, i.e. create a
417 "tombstone". That makes remove faster, but leaves a lot of
418 garbage and slows down hash_table_get and hash_table_put. */
420 mp = NEXT_MAPPING (mp, mappings, size);
421 LOOP_NON_EMPTY (mp, mappings, size)
423 const void *key2 = mp->key;
424 struct mapping *mp_new = mappings + HASH_POSITION (ht, key2);
426 /* Find the new location for the key. */
428 LOOP_NON_EMPTY (mp_new, mappings, size)
429 if (key2 == mp_new->key)
430 /* The mapping MP (key2) is already where we want it (in
431 MP_NEW's "chain" of keys.) */
444 /* Clear HT of all entries. After calling this function, the count
445 and the fullness of the hash table will be zero. The size will
449 hash_table_clear (struct hash_table *ht)
451 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
455 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
456 called with three arguments: the key, the value, and the CLOSURE.
458 It is undefined what happens if you add or remove entries in the
459 hash table while hash_table_map is running. The exception is the
460 entry you're currently mapping over; you may remove or change that
464 hash_table_map (struct hash_table *ht,
465 int (*mapfun) (void *, void *, void *),
468 struct mapping *mp = ht->mappings;
469 struct mapping *end = ht->mappings + ht->size;
471 for (; mp < end; mp++)
472 if (!EMPTY_MAPPING_P (mp))
477 if (mapfun (key, mp->value, closure))
479 /* hash_table_remove might have moved the adjacent
481 if (mp->key != key && !EMPTY_MAPPING_P (mp))
486 /* Return the number of elements in the hash table. This is not the
487 same as the physical size of the hash table, which is always
488 greater than the number of elements. */
491 hash_table_count (struct hash_table *ht)
496 /* Functions from this point onward are meant for convenience and
497 don't strictly belong to this file. However, this is as good a
498 place for them as any. */
501 Support for hash tables whose keys are strings.
504 /* 31 bit hash function. Taken from Gnome's glib, modified to use
507 We used to use the popular hash function from the Dragon Book, but
508 this one seems to perform much better. */
511 string_hash (const void *key)
517 for (p += 1; *p != '\0'; p++)
518 h = (h << 5) - h + *p;
523 /* Frontend for strcmp usable for hash tables. */
526 string_cmp (const void *s1, const void *s2)
528 return !strcmp ((const char *)s1, (const char *)s2);
531 /* Return a hash table of initial size INITIAL_SIZE suitable to use
535 make_string_hash_table (int initial_size)
537 return hash_table_new (initial_size, string_hash, string_cmp);
541 Support for hash tables whose keys are strings, but which are
542 compared case-insensitively.
545 /* Like string_hash, but produce the same hash regardless of the case. */
548 string_hash_nocase (const void *key)
551 unsigned int h = TOLOWER (*p);
554 for (p += 1; *p != '\0'; p++)
555 h = (h << 5) - h + TOLOWER (*p);
560 /* Like string_cmp, but doing case-insensitive compareison. */
563 string_cmp_nocase (const void *s1, const void *s2)
565 return !strcasecmp ((const char *)s1, (const char *)s2);
568 /* Like make_string_hash_table, but uses string_hash_nocase and
569 string_cmp_nocase. */
572 make_nocase_string_hash_table (int initial_size)
574 return hash_table_new (initial_size, string_hash_nocase, string_cmp_nocase);
578 /* If I ever need it: hashing of integers. */
581 inthash (unsigned int key)
601 print_hash_table_mapper (void *key, void *value, void *count)
604 printf ("%s: %s\n", (const char *)key, (char *)value);
609 print_hash (struct hash_table *sht)
612 hash_table_map (sht, print_hash_table_mapper, &debug_count);
613 assert (debug_count == sht->count);
619 struct hash_table *ht = make_string_hash_table (0);
621 while ((fgets (line, sizeof (line), stdin)))
623 int len = strlen (line);
627 if (!hash_table_contains (ht, line))
628 hash_table_put (ht, strdup (line), "here I am!");
633 if (hash_table_get_pair (ht, line, &line_copy, NULL))
635 hash_table_remove (ht, line);
645 printf ("%d %d\n", ht->count, ht->size);