2 Copyright (C) 2000 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. */
33 # define xmalloc malloc
34 # define xrealloc realloc
40 Hash tables are an implementation technique used to implement
41 mapping between objects. Provided a good hashing function is used,
42 they guarantee constant-time access and storing of information.
43 Duplicate keys are not allowed.
45 The basics are all covered. hash_table_new creates a hash table,
46 and hash_table_destroy deletes it. hash_table_put establishes a
47 mapping between a key and a value. hash_table_get retrieves the
48 value that corresponds to a key. hash_table_exists queries whether
49 a key is stored in a table at all. hash_table_remove removes a
50 mapping that corresponds to a key. hash_table_map allows you to
51 map through all the entries in a hash table. hash_table_clear
52 clears all the entries from the hash table.
54 The number of mappings in a table is not limited, except by the
55 amount of memory. As you add new elements to a table, it regrows
56 as necessary. If you have an idea about how many elements you will
57 store, you can provide a hint to hash_table_new().
59 The hashing and equality functions are normally provided by the
60 user. For the special (and frequent) case of hashing strings, you
61 can use the pre-canned make_string_hash_table(), which provides the
62 string hashing function from the Dragon Book, and a string equality
63 wrapper around strcmp().
65 When specifying your own hash and test functions, make sure the
68 - The test function returns non-zero for keys that are considered
69 "equal", zero otherwise.
71 - The hash function returns a number that represents the
72 "distinctness" of the object. In more precise terms, it means
73 that for any two objects that test "equal" under the test
74 function, the hash function MUST produce the same result.
76 This does not mean that each distinct object must produce a
77 distinct value, only that non-distinct objects must produce the
78 same values! For instance, a hash function that returns 0 for
79 any given object is a perfectly valid (albeit extremely bad) hash
82 The above stated rule is quite easy to enforce. For example, if
83 your testing function compares strings case-insensitively, all
84 your function needs to do is lower-case the string characters
85 before calculating a hash. That way you have easily guaranteed
86 that changes in case will not result in a different hash.
88 - (optional) Choose the hash function to get as good "spreading" as
89 possible. A good hash function will react to even a small change
90 in its input with a completely different resulting hash.
91 Finally, don't make your hash function extremely slow, because
92 you're then defeating the purpose of hashing.
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 All the hash mappings (key-value pairs of pointers) are stored in a
103 contiguous array. The position of each mapping is determined by
104 applying the hash function to the key: location = hash(key) % size.
105 If two different keys end up on the same position, the collision is
106 resolved by placing the second mapping at the next empty place in
107 the array following the occupied place. This method of collision
108 resolution is called "linear probing".
110 There are more advanced collision resolution mechanisms (quadratic
111 probing, double hashing), but we don't use them because they
112 involve more non-sequential access to the array, and therefore
113 worse cache behavior. Linear probing works well as long as the
114 fullness/size ratio is kept below 75%. We make sure to regrow or
115 rehash the hash table whenever this threshold is exceeded.
117 Collisions make deletion tricky because finding collisions again
118 relies on new empty spots not being created. That's why
119 hash_table_remove only marks the spot as deleted rather than really
128 unsigned long (*hash_function) (const void *);
129 int (*test_function) (const void *, const void *);
131 int size; /* size of the array */
132 int fullness; /* number of non-empty fields */
133 int count; /* number of non-empty, non-deleted
136 struct mapping *mappings;
139 #define ENTRY_DELETED ((void *)0xdeadbeef)
140 #define ENTRY_EMPTY NULL
142 #define DELETED_ENTRY_P(ptr) ((ptr) == ENTRY_DELETED)
143 #define EMPTY_ENTRY_P(ptr) ((ptr) == ENTRY_EMPTY)
145 /* Find a prime near, but greather than or equal to SIZE. */
148 prime_size (int size)
150 static const unsigned long primes [] = {
151 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
152 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
153 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
154 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
155 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
156 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
157 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
158 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
159 1174703521, 1527114613, 1985248999, 2580823717UL, 3355070839UL
162 for (i = 0; i < ARRAY_SIZE (primes); i++)
163 if (primes[i] >= size)
169 /* Create a hash table of INITIAL_SIZE with hash function
170 HASH_FUNCTION and test function TEST_FUNCTION. If you wish to
171 start out with a "small" table which will be regrown as needed,
172 specify 0 as INITIAL_SIZE. */
175 hash_table_new (int initial_size,
176 unsigned long (*hash_function) (const void *),
177 int (*test_function) (const void *, const void *))
179 struct hash_table *ht
180 = (struct hash_table *)xmalloc (sizeof (struct hash_table));
181 ht->hash_function = hash_function;
182 ht->test_function = test_function;
183 ht->size = prime_size (initial_size);
186 ht->mappings = xmalloc (ht->size * sizeof (struct mapping));
187 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
191 /* Free the data associated with hash table HT. */
194 hash_table_destroy (struct hash_table *ht)
196 xfree (ht->mappings);
200 /* The heart of almost all functions in this file -- find the mapping
201 whose KEY is equal to key, using a linear probing loop. Returns
202 the offset of the mapping in ht->mappings. This should probably be
206 find_mapping (struct hash_table *ht, const void *key)
208 struct mapping *mappings = ht->mappings;
210 int location = ht->hash_function (key) % size;
213 struct mapping *mp = mappings + location;
214 void *mp_key = mp->key;
216 if (EMPTY_ENTRY_P (mp_key))
218 else if (DELETED_ENTRY_P (mp_key)
219 || !ht->test_function (key, mp_key))
221 if (++location == size)
229 /* Get the value that corresponds to the key KEY in the hash table HT.
230 If no value is found, return NULL. Note that NULL is a legal value
231 for value; if you are storing NULLs in your hash table, you can use
232 hash_table_exists to be sure that a (possibly NULL) value exists in
233 the table. Or, you can use hash_table_get_pair instead of this
237 hash_table_get (struct hash_table *ht, const void *key)
239 int location = find_mapping (ht, key);
243 return ht->mappings[location].value;
246 /* Like hash_table_get, but writes out the pointers to both key and
247 value. Returns non-zero on success. */
250 hash_table_get_pair (struct hash_table *ht, const void *lookup_key,
251 void *orig_key, void *value)
253 int location = find_mapping (ht, lookup_key);
258 struct mapping *mp = ht->mappings + location;
260 *(void **)orig_key = mp->key;
262 *(void **)value = mp->value;
267 /* Return 1 if KEY exists in HT, 0 otherwise. */
270 hash_table_exists (struct hash_table *ht, const void *key)
272 return find_mapping (ht, key) >= 0;
275 #define MAX(i, j) (((i) >= (j)) ? (i) : (j))
277 /* Grow hash table HT as necessary, and rehash all the key-value
281 grow_hash_table (struct hash_table *ht)
284 struct mapping *old_mappings = ht->mappings;
285 int old_count = ht->count; /* for assert() below */
286 int old_size = ht->size;
288 /* To minimize the number of regrowth, we'd like to resize the hash
289 table exponentially. Normally, this would be done by doubling
290 ht->size (and round it to next prime) on each regrow:
292 ht->size = prime_size (ht->size * 2);
294 But it is possible that the table has large fullness because of
295 the many deleted entries. If that is the case, we don't want to
296 blindly grow the table; we just want to rehash it. For that
297 reason, we use ht->count as the relevant parameter. MAX is used
298 only because we don't want to actually shrink the table. (But
299 maybe that's wrong.) */
301 int needed_size = prime_size (ht->count * 3);
302 ht->size = MAX (old_size, needed_size);
305 printf ("growing from %d to %d\n", old_size, ht->size);
308 ht->mappings = xmalloc (ht->size * sizeof (struct mapping));
309 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
311 /* Need to reset these two; hash_table_put will reinitialize them. */
314 for (i = 0; i < old_size; i++)
316 struct mapping *mp = old_mappings + i;
317 void *mp_key = mp->key;
319 if (!EMPTY_ENTRY_P (mp_key)
320 && !DELETED_ENTRY_P (mp_key))
321 hash_table_put (ht, mp_key, mp->value);
323 assert (ht->count == old_count);
324 xfree (old_mappings);
327 /* Put VALUE in the hash table HT under the key KEY. This regrows the
328 table if necessary. */
331 hash_table_put (struct hash_table *ht, const void *key, void *value)
333 /* Cannot use find_mapping here because we're actually looking for
336 struct mapping *mappings = ht->mappings;
338 int location = ht->hash_function (key) % size;
341 struct mapping *mp = mappings + location;
342 void *mp_key = mp->key;
344 if (EMPTY_ENTRY_P (mp_key))
349 mp->key = (void *)key; /* const? */
353 else if (DELETED_ENTRY_P (mp_key)
354 || !ht->test_function (key, mp_key))
356 if (++location == size)
359 else /* equal to key and not deleted */
361 /* We're replacing an existing entry, so ht->count and
362 ht->fullness remain unchanged. */
366 if (ht->fullness * 4 > ht->size * 3)
367 /* When fullness exceeds 75% of size, regrow the table. */
368 grow_hash_table (ht);
371 /* Remove KEY from HT. */
374 hash_table_remove (struct hash_table *ht, const void *key)
376 int location = find_mapping (ht, key);
381 struct mapping *mappings = ht->mappings;
382 struct mapping *mp = mappings + location;
383 /* We don't really remove an entry from the hash table: we just
384 mark it as deleted. This is because there may be other
385 entries located after this entry whose hash points to a
386 location before this entry. (Example: keys A, B and C have
387 the same hash. If you were to really *delete* B from the
388 table, C could no longer be found.) */
390 /* Optimization addendum: if the mapping that follows LOCATION
391 is already empty, that is a sure sign that nobody depends on
392 LOCATION being non-empty. (This is because we're using
393 linear probing. This would not be the case with double
394 hashing.) In that case, we may safely delete the mapping. */
396 /* This could be generalized so that the all the non-empty
397 locations following LOCATION are simply shifted leftward. It
398 would make deletion a bit slower, but it would remove the
399 ugly DELETED_ENTRY_P checks from all the rest of the code,
400 making the whole thing faster. */
401 int location_after = (location + 1) == ht->size ? 0 : location + 1;
402 struct mapping *mp_after = mappings + location_after;
404 if (EMPTY_ENTRY_P (mp_after->key))
406 mp->key = ENTRY_EMPTY;
410 mp->key = ENTRY_DELETED;
417 /* Clear HT of all entries. After calling this function, the count
418 and the fullness of the hash table will be zero. The size will
422 hash_table_clear (struct hash_table *ht)
424 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
429 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
430 called with three arguments: the key, the value, and the CLOSURE.
431 Don't add or remove entries from HT while hash_table_map is being
432 called, or strange things may happen. */
435 hash_table_map (struct hash_table *ht,
436 int (*mapfun) (void *, void *, void *),
439 struct mapping *mappings = ht->mappings;
441 for (i = 0; i < ht->size; i++)
443 struct mapping *mp = mappings + i;
444 void *mp_key = mp->key;
446 if (!EMPTY_ENTRY_P (mp_key)
447 && !DELETED_ENTRY_P (mp_key))
448 if (mapfun (mp_key, mp->value, closure))
453 /* Support for hash tables whose keys are strings. */
455 /* supposedly from the Dragon Book P436. */
457 string_hash (const void *sv)
460 unsigned const char *x = (unsigned const char *) sv;
466 if ((g = h & 0xf0000000) != 0)
467 h = (h ^ (g >> 24)) ^ g;
474 /* If I ever need it: hashing of integers. */
477 inthash (unsigned int key)
492 string_cmp (const void *s1, const void *s2)
494 return !strcmp ((const char *)s1, (const char *)s2);
497 /* Return a hash table of initial size INITIAL_SIZE suitable to use
501 make_string_hash_table (int initial_size)
503 return hash_table_new (initial_size, string_hash, string_cmp);
513 print_hash_table_mapper (void *key, void *value, void *count)
516 printf ("%s: %s\n", (const char *)key, (char *)value);
521 print_hash (struct hash_table *sht)
524 hash_table_map (sht, print_hash_table_mapper, &debug_count);
525 assert (debug_count == sht->count);
531 struct hash_table *ht = make_string_hash_table (0);
533 while ((fgets (line, sizeof (line), stdin)))
535 int len = strlen (line);
539 if (!hash_table_exists (ht, line))
540 hash_table_put (ht, strdup (line), "here I am!");
545 if (hash_table_get_pair (ht, line, &line_copy, NULL))
547 hash_table_remove (ht, line);
557 printf ("%d %d %d\n", ht->count, ht->fullness, ht->size);