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. Assuming a good hashing function is used,
64 they provide near-constant-time access and storing of information.
65 Duplicate keys are not allowed.
67 This file defines the following entry points: hash_table_new
68 creates a hash table, and hash_table_destroy deletes it.
69 hash_table_put establishes a mapping between a key and a value.
70 hash_table_get retrieves the value that corresponds to a key.
71 hash_table_contains queries whether a key is stored in a table at
72 all. hash_table_remove removes a mapping that corresponds to a
73 key. hash_table_map allows you to map through all the entries in a
74 hash table. hash_table_clear clears all the entries from the hash
77 The number of mappings in a table is not limited, except by the
78 amount of memory. As you add new elements to a table, it regrows
79 as necessary. If you have an idea about how many elements you will
80 store, you can provide a hint to hash_table_new().
82 The hashing and equality functions depend on the type of key and
83 are normally provided by the user. For the special (and frequent)
84 case of using string keys, you can use the pre-canned
85 make_string_hash_table(), which provides an efficient string
86 hashing function, and a string equality wrapper around strcmp().
88 When specifying your hash and test functions, make sure the
91 - The test function returns non-zero for keys that are considered
92 "equal", zero otherwise.
94 - The hash function returns a number that represents the
95 "distinctness" of the object. In more precise terms, it means
96 that for any two objects that test "equal" under the test
97 function, the hash function MUST produce the same result.
99 This does not mean that each distinct object must produce a
100 distinct value, only that non-distinct objects must produce the
101 same values! For instance, a hash function that returns 0 for
102 any given object is a perfectly valid (albeit extremely bad) hash
103 function. A hash function that hashes a string by adding up all
104 its characters is another example of a valid (but quite bad) hash
107 The above stated rule is quite easy to enforce. For example, if
108 your testing function compares strings case-insensitively, all
109 your function needs to do is lower-case the string characters
110 before calculating a hash. That way you have easily guaranteed
111 that case differences will not result in a different hash.
113 - If you care about performance, choose a hash function with as
114 good "spreading" as possible. A good hash function will react to
115 even a small change in its input with a completely different
116 resulting hash. Finally, don't make the hash function itself
117 overly slow, because you'll be incurring a non-negligible
118 overhead to reads and writes to the hash table.
120 Note that neither keys nor values are copied when inserted into the
121 hash table, so they must exist for the lifetime of the table. This
122 means that e.g. the use of static strings is OK, but objects with a
123 shorter life-time need to be copied (with strdup() or the like in
124 the case of strings) before being inserted. */
128 All the hash mappings (key-value pairs of pointers) are stored in a
129 contiguous array. The position of each mapping is determined by
130 the hash value of its key and the size of the table: location :=
131 hash(key) % size. If two different keys end up on the same
132 position (hash collision), the one that came second is placed at
133 the next empty position following the occupied place. This
134 collision resolution technique is called "linear probing".
136 There are more advanced collision resolution mechanisms (quadratic
137 probing, double hashing), but we don't use them because they incur
138 more non-sequential access to the array, which results in worse
139 cache behavior. Linear probing works well as long as the
140 fullness/size ratio is kept below 75%. We make sure to regrow or
141 rehash the hash table whenever this threshold is exceeded.
143 Collisions make deletion tricky because finding collisions again
144 relies on new empty spots not being created. That's why
145 hash_table_remove is careful to rehash the mappings that follow the
154 unsigned long (*hash_function) PARAMS ((const void *));
155 int (*test_function) PARAMS ((const void *, const void *));
157 int size; /* size of the array */
158 int count; /* number of non-empty, non-deleted
161 int resize_threshold; /* after size exceeds this number of
162 entries, resize the table. */
163 int prime_offset; /* the offset of the current prime in
166 struct mapping *mappings; /* the array of mapping pairs. */
169 #define EMPTY_MAPPING_P(mp) ((mp)->key == NULL)
170 #define NEXT_MAPPING(mp, mappings, size) (mp == mappings + (size - 1) \
173 #define LOOP_NON_EMPTY(mp, mappings, size) \
174 for (; !EMPTY_MAPPING_P (mp); mp = NEXT_MAPPING (mp, mappings, size))
176 /* #### We might want to multiply with the "golden ratio" here to get
177 better randomness for keys that do not result from a good hash
178 function. This is currently not a problem in Wget because we only
179 use the string hash tables. */
181 #define HASH_POSITION(ht, key) (ht->hash_function (key) % ht->size)
183 /* Find a prime near, but greather than or equal to SIZE. Of course,
184 the primes are not calculated, but looked up from a table. The
185 table does not contain all primes in range, just a selection useful
188 PRIME_OFFSET is a micro-optimization: if specified, it starts the
189 search for the prime number beginning with the specific offset in
190 the prime number table. The final offset is stored in the same
194 prime_size (int size, int *prime_offset)
196 static const unsigned long primes [] = {
197 13, 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
198 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
199 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
200 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
201 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
202 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
203 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
204 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
205 1174703521, 1527114613, 1985248999,
206 (unsigned long)0x99d43ea5, (unsigned long)0xc7fa5177
208 int i = *prime_offset;
210 for (; i < countof (primes); i++)
211 if (primes[i] >= size)
213 /* Set the offset to the next prime. That is safe because,
214 next time we are called, it will be with a larger SIZE,
215 which means we could never return the same prime anyway.
216 (If that is not the case, the caller can simply reset
218 *prime_offset = i + 1;
226 /* Create a hash table of INITIAL_SIZE with hash function
227 HASH_FUNCTION and test function TEST_FUNCTION. INITIAL_SIZE will
228 be rounded to the next prime, so you don't have to worry about it
229 being a prime number.
231 Consequently, if you wish to start out with a "small" table which
232 will be regrown as needed, specify INITIAL_SIZE 0.
234 If HASH_FUNCTION is not provided, identity table is assumed,
235 i.e. key pointers are compared as keys. If you want strings with
236 equal contents to hash the same, use make_string_hash_table. */
239 hash_table_new (int initial_size,
240 unsigned long (*hash_function) (const void *),
241 int (*test_function) (const void *, const void *))
243 struct hash_table *ht
244 = (struct hash_table *)xmalloc (sizeof (struct hash_table));
246 ht->hash_function = hash_function ? hash_function : ptrhash;
247 ht->test_function = test_function ? test_function : ptrcmp;
249 ht->prime_offset = 0;
250 ht->size = prime_size (initial_size, &ht->prime_offset);
251 ht->resize_threshold = ht->size * 3 / 4;
255 ht->mappings = xmalloc (ht->size * sizeof (struct mapping));
256 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
261 /* Free the data associated with hash table HT. */
264 hash_table_destroy (struct hash_table *ht)
266 xfree (ht->mappings);
270 /* The heart of almost all functions in this file -- find the mapping
271 whose KEY is equal to key, using linear probing. Returns the
272 mapping that matches KEY, or NULL if none matches. */
274 static inline struct mapping *
275 find_mapping (struct hash_table *ht, const void *key)
277 struct mapping *mappings = ht->mappings;
279 struct mapping *mp = mappings + HASH_POSITION (ht, key);
280 int (*equals) PARAMS ((const void *, const void *)) = ht->test_function;
282 LOOP_NON_EMPTY (mp, mappings, size)
283 if (equals (key, mp->key))
288 /* Get the value that corresponds to the key KEY in the hash table HT.
289 If no value is found, return NULL. Note that NULL is a legal value
290 for value; if you are storing NULLs in your hash table, you can use
291 hash_table_contains to be sure that a (possibly NULL) value exists
292 in the table. Or, you can use hash_table_get_pair instead of this
296 hash_table_get (struct hash_table *ht, const void *key)
298 struct mapping *mp = find_mapping (ht, key);
305 /* Like hash_table_get, but writes out the pointers to both key and
306 value. Returns non-zero on success. */
309 hash_table_get_pair (struct hash_table *ht, const void *lookup_key,
310 void *orig_key, void *value)
312 struct mapping *mp = find_mapping (ht, lookup_key);
317 *(void **)orig_key = mp->key;
319 *(void **)value = mp->value;
326 /* Return 1 if HT contains KEY, 0 otherwise. */
329 hash_table_contains (struct hash_table *ht, const void *key)
331 return find_mapping (ht, key) != NULL;
334 /* Grow hash table HT as necessary, and rehash all the key-value
338 grow_hash_table (struct hash_table *ht)
340 struct mapping *old_mappings = ht->mappings;
341 struct mapping *old_end = ht->mappings + ht->size;
342 struct mapping *mp, *mappings;
345 newsize = prime_size (ht->size * 2, &ht->prime_offset);
347 printf ("growing from %d to %d; fullness %.2f%% to %.2f%%\n",
349 (double)100 * ht->count / ht->size,
350 (double)100 * ht->count / newsize);
354 ht->resize_threshold = newsize * 3 / 4;
356 mappings = xmalloc (ht->size * sizeof (struct mapping));
357 memset (mappings, '\0', ht->size * sizeof (struct mapping));
358 ht->mappings = mappings;
360 for (mp = old_mappings; mp < old_end; mp++)
361 if (!EMPTY_MAPPING_P (mp))
363 struct mapping *new_mp = mappings + HASH_POSITION (ht, mp->key);
364 /* We don't need to call test function and worry about
365 collisions because all the keys come from the hash table
366 and are therefore guaranteed to be unique. */
367 LOOP_NON_EMPTY (new_mp, mappings, newsize)
372 xfree (old_mappings);
375 /* Put VALUE in the hash table HT under the key KEY. This regrows the
376 table if necessary. */
379 hash_table_put (struct hash_table *ht, const void *key, void *value)
381 struct mapping *mappings = ht->mappings;
383 int (*equals) PARAMS ((const void *, const void *)) = ht->test_function;
385 struct mapping *mp = mappings + HASH_POSITION (ht, key);
387 LOOP_NON_EMPTY (mp, mappings, size)
388 if (equals (key, mp->key))
390 mp->key = (void *)key; /* const? */
396 mp->key = (void *)key; /* const? */
399 if (ht->count > ht->resize_threshold)
400 /* When table is 75% full, regrow it. */
401 grow_hash_table (ht);
404 /* Remove a mapping that matches KEY from HT. Return 0 if there was
405 no such entry; return 1 if an entry was removed. */
408 hash_table_remove (struct hash_table *ht, const void *key)
410 struct mapping *mp = find_mapping (ht, key);
416 struct mapping *mappings = ht->mappings;
421 /* Rehash all the entries following MP. The alternative
422 approach is to mark the entry as deleted, i.e. create a
423 "tombstone". That makes remove faster, but leaves a lot of
424 garbage and slows down hash_table_get and hash_table_put. */
426 mp = NEXT_MAPPING (mp, mappings, size);
427 LOOP_NON_EMPTY (mp, mappings, size)
429 const void *key2 = mp->key;
430 struct mapping *mp_new = mappings + HASH_POSITION (ht, key2);
432 /* Find the new location for the key. */
434 LOOP_NON_EMPTY (mp_new, mappings, size)
435 if (key2 == mp_new->key)
436 /* The mapping MP (key2) is already where we want it (in
437 MP_NEW's "chain" of keys.) */
450 /* Clear HT of all entries. After calling this function, the count
451 and the fullness of the hash table will be zero. The size will
455 hash_table_clear (struct hash_table *ht)
457 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
461 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
462 called with three arguments: the key, the value, and the CLOSURE.
464 It is undefined what happens if you add or remove entries in the
465 hash table while hash_table_map is running. The exception is the
466 entry you're currently mapping over; you may remove or change that
470 hash_table_map (struct hash_table *ht,
471 int (*mapfun) (void *, void *, void *),
474 struct mapping *mp = ht->mappings;
475 struct mapping *end = ht->mappings + ht->size;
477 for (; mp < end; mp++)
478 if (!EMPTY_MAPPING_P (mp))
483 if (mapfun (key, mp->value, closure))
485 /* hash_table_remove might have moved the adjacent
487 if (mp->key != key && !EMPTY_MAPPING_P (mp))
492 /* Return the number of elements in the hash table. This is not the
493 same as the physical size of the hash table, which is always
494 greater than the number of elements. */
497 hash_table_count (struct hash_table *ht)
502 /* Functions from this point onward are meant for convenience and
503 don't strictly belong to this file. However, this is as good a
504 place for them as any. */
507 * Support for hash tables whose keys are strings.
511 /* 31 bit hash function. Taken from Gnome's glib, modified to use
514 We used to use the popular hash function from the Dragon Book, but
515 this one seems to perform much better. */
518 string_hash (const void *key)
524 for (p += 1; *p != '\0'; p++)
525 h = (h << 5) - h + *p;
530 /* Frontend for strcmp usable for hash tables. */
533 string_cmp (const void *s1, const void *s2)
535 return !strcmp ((const char *)s1, (const char *)s2);
538 /* Return a hash table of initial size INITIAL_SIZE suitable to use
542 make_string_hash_table (int initial_size)
544 return hash_table_new (initial_size, string_hash, string_cmp);
548 * Support for hash tables whose keys are strings, but which are
549 * compared case-insensitively.
553 /* Like string_hash, but produce the same hash regardless of the case. */
556 string_hash_nocase (const void *key)
559 unsigned int h = TOLOWER (*p);
562 for (p += 1; *p != '\0'; p++)
563 h = (h << 5) - h + TOLOWER (*p);
568 /* Like string_cmp, but doing case-insensitive compareison. */
571 string_cmp_nocase (const void *s1, const void *s2)
573 return !strcasecmp ((const char *)s1, (const char *)s2);
576 /* Like make_string_hash_table, but uses string_hash_nocase and
577 string_cmp_nocase. */
580 make_nocase_string_hash_table (int initial_size)
582 return hash_table_new (initial_size, string_hash_nocase, string_cmp_nocase);
585 /* Hashing of pointers. Used for hash tables that are keyed by
586 pointer identity. (Common Lisp calls them EQ hash tables, and Java
587 calls them IdentityHashMaps.) */
590 ptrhash (const void *ptr)
592 unsigned long key = (unsigned long)ptr;
615 ptrcmp (const void *ptr1, const void *ptr2)
621 /* Currently unused: hashing of integers. */
624 inthash (unsigned int key)
644 print_hash_table_mapper (void *key, void *value, void *count)
647 printf ("%s: %s\n", (const char *)key, (char *)value);
652 print_hash (struct hash_table *sht)
655 hash_table_map (sht, print_hash_table_mapper, &debug_count);
656 assert (debug_count == sht->count);
662 struct hash_table *ht = make_string_hash_table (0);
664 while ((fgets (line, sizeof (line), stdin)))
666 int len = strlen (line);
670 if (!hash_table_contains (ht, line))
671 hash_table_put (ht, strdup (line), "here I am!");
676 if (hash_table_get_pair (ht, line, &line_copy, NULL))
678 hash_table_remove (ht, line);
688 printf ("%d %d\n", ht->count, ht->size);