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
148 /* When hash table's fullness exceeds this threshold, the hash table
150 #define HASH_FULLNESS_THRESHOLD 0.75
152 /* The hash table size is multiplied by this factor with each resize.
153 This guarantees infrequent resizes. */
154 #define HASH_RESIZE_FACTOR 2
162 unsigned long (*hash_function) PARAMS ((const void *));
163 int (*test_function) PARAMS ((const void *, const void *));
165 int size; /* size of the array */
166 int count; /* number of non-empty, non-deleted
169 int resize_threshold; /* after size exceeds this number of
170 entries, resize the table. */
171 int prime_offset; /* the offset of the current prime in
174 struct mapping *mappings; /* the array of mapping pairs. */
177 /* We use NULL key to mark a mapping as empty. It is consequently
178 illegal to store NULL keys. */
179 #define NON_EMPTY(mp) (mp->key != NULL)
181 /* "Next" mapping is the mapping after MP, but wrapping back to
182 MAPPINGS when MP would reach MAPPINGS+SIZE. */
183 #define NEXT_MAPPING(mp, mappings, size) (mp != mappings + (size - 1) \
186 /* Loop over non-empty mappings starting at MP. */
187 #define LOOP_NON_EMPTY(mp, mappings, size) \
188 for (; NON_EMPTY (mp); mp = NEXT_MAPPING (mp, mappings, size))
190 /* #### We might want to multiply with the "golden ratio" here to get
191 better randomness for keys that do not result from a good hash
192 function. This is currently not a problem in Wget because we only
193 use the string hash tables. */
195 #define HASH_POSITION(ht, key) (ht->hash_function (key) % ht->size)
197 /* Find a prime near, but greather than or equal to SIZE. Of course,
198 the primes are not calculated, but looked up from a table. The
199 table does not contain all primes in range, just a selection useful
202 PRIME_OFFSET is a minor optimization: if specified, it starts the
203 search for the prime number beginning with the specific offset in
204 the prime number table. The final offset is stored in the same
208 prime_size (int size, int *prime_offset)
210 static const unsigned long primes [] = {
211 13, 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
212 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
213 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
214 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
215 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
216 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
217 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
218 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
219 1174703521, 1527114613, 1985248999,
220 (unsigned long)0x99d43ea5, (unsigned long)0xc7fa5177
222 int i = *prime_offset;
224 for (; i < countof (primes); i++)
225 if (primes[i] >= size)
227 /* Set the offset to the next prime. That is safe because,
228 next time we are called, it will be with a larger SIZE,
229 which means we could never return the same prime anyway.
230 (If that is not the case, the caller can simply reset
232 *prime_offset = i + 1;
240 /* Create a hash table with hash function HASH_FUNCTION and test
241 function TEST_FUNCTION. The table is empty (its count is 0), but
242 pre-allocated to store at least ITEMS items.
244 ITEMS is the number of items that the table can accept without
245 needing to resize. It is useful when creating a table that is to
246 be immediately filled with a known number of items. In that case,
247 the regrows are a waste of time, and specifying ITEMS correctly
248 will avoid them altogether.
250 Note that hash tables grow dynamically regardless of ITEMS. The
251 only use of ITEMS is to preallocate the table and avoid unnecessary
252 dynamic regrows. Don't bother making ITEMS prime because it's not
253 used as size unchanged. To start with a small table that grows as
254 needed, simply specify zero ITEMS.
256 If HASH_FUNCTION is not provided, identity table is assumed,
257 i.e. key pointers are compared as keys. If you want strings with
258 equal contents to hash the same, use make_string_hash_table. */
261 hash_table_new (int items,
262 unsigned long (*hash_function) (const void *),
263 int (*test_function) (const void *, const void *))
266 struct hash_table *ht = xnew (struct hash_table);
268 ht->hash_function = hash_function ? hash_function : ptrhash;
269 ht->test_function = test_function ? test_function : ptrcmp;
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_FULLNESS_THRESHOLD;
276 size = prime_size (size, &ht->prime_offset);
278 ht->resize_threshold = size * HASH_FULLNESS_THRESHOLD;
279 /*assert (ht->resize_threshold >= items);*/
281 ht->mappings = xnew0_array (struct mapping, ht->size);
287 /* Free the data associated with hash table HT. */
290 hash_table_destroy (struct hash_table *ht)
292 xfree (ht->mappings);
296 /* The heart of most functions in this file -- find the mapping whose
297 KEY is equal to key, using linear probing. Returns the mapping
298 that matches KEY, or the first empty mapping if none matches. */
300 static inline struct mapping *
301 find_mapping (const struct hash_table *ht, const void *key)
303 struct mapping *mappings = ht->mappings;
305 struct mapping *mp = mappings + HASH_POSITION (ht, key);
306 int (*equals) PARAMS ((const void *, const void *)) = ht->test_function;
308 LOOP_NON_EMPTY (mp, mappings, size)
309 if (equals (key, mp->key))
314 /* Get the value that corresponds to the key KEY in the hash table HT.
315 If no value is found, return NULL. Note that NULL is a legal value
316 for value; if you are storing NULLs in your hash table, you can use
317 hash_table_contains to be sure that a (possibly NULL) value exists
318 in the table. Or, you can use hash_table_get_pair instead of this
322 hash_table_get (const struct hash_table *ht, const void *key)
324 struct mapping *mp = find_mapping (ht, key);
331 /* Like hash_table_get, but writes out the pointers to both key and
332 value. Returns non-zero on success. */
335 hash_table_get_pair (const struct hash_table *ht, const void *lookup_key,
336 void *orig_key, void *value)
338 struct mapping *mp = find_mapping (ht, lookup_key);
342 *(void **)orig_key = mp->key;
344 *(void **)value = mp->value;
351 /* Return 1 if HT contains KEY, 0 otherwise. */
354 hash_table_contains (const struct hash_table *ht, const void *key)
356 struct mapping *mp = find_mapping (ht, key);
357 return NON_EMPTY (mp);
360 /* Grow hash table HT as necessary, and rehash all the key-value
364 grow_hash_table (struct hash_table *ht)
366 struct mapping *old_mappings = ht->mappings;
367 struct mapping *old_end = ht->mappings + ht->size;
368 struct mapping *mp, *mappings;
371 newsize = prime_size (ht->size * HASH_RESIZE_FACTOR, &ht->prime_offset);
373 printf ("growing from %d to %d; fullness %.2f%% to %.2f%%\n",
375 100.0 * ht->count / ht->size,
376 100.0 * ht->count / newsize);
380 ht->resize_threshold = newsize * HASH_FULLNESS_THRESHOLD;
382 ht->mappings = mappings = xnew0_array (struct mapping, ht->size);
384 for (mp = old_mappings; mp < old_end; mp++)
387 struct mapping *new_mp = mappings + HASH_POSITION (ht, mp->key);
388 /* We don't need to test for uniqueness of keys because all
389 the keys come from the hash table and are therefore known
391 LOOP_NON_EMPTY (new_mp, mappings, newsize)
396 xfree (old_mappings);
399 /* Put VALUE in the hash table HT under the key KEY. This regrows the
400 table if necessary. */
403 hash_table_put (struct hash_table *ht, const void *key, void *value)
405 struct mapping *mp = find_mapping (ht, key);
408 /* update existing item */
409 mp->key = (void *)key; /* const? */
414 /* If adding the item would make the table exceed max. fullness,
415 grow the table first. */
416 if (ht->count >= ht->resize_threshold)
418 grow_hash_table (ht);
419 mp = find_mapping (ht, key);
424 mp->key = (void *)key; /* const? */
428 /* Remove a mapping that matches KEY from HT. Return 0 if there was
429 no such entry; return 1 if an entry was removed. */
432 hash_table_remove (struct hash_table *ht, const void *key)
434 struct mapping *mp = find_mapping (ht, key);
440 struct mapping *mappings = ht->mappings;
445 /* Rehash all the entries following MP. The alternative
446 approach is to mark the entry as deleted, i.e. create a
447 "tombstone". That makes remove faster, but leaves a lot of
448 garbage and slows down hash_table_get and hash_table_put. */
450 mp = NEXT_MAPPING (mp, mappings, size);
451 LOOP_NON_EMPTY (mp, mappings, size)
453 const void *key2 = mp->key;
454 struct mapping *mp_new = mappings + HASH_POSITION (ht, key2);
456 /* Find the new location for the key. */
458 LOOP_NON_EMPTY (mp_new, mappings, size)
459 if (key2 == mp_new->key)
460 /* The mapping MP (key2) is already where we want it (in
461 MP_NEW's "chain" of keys.) */
474 /* Clear HT of all entries. After calling this function, the count
475 and the fullness of the hash table will be zero. The size will
479 hash_table_clear (struct hash_table *ht)
481 memset (ht->mappings, '\0', ht->size * sizeof (struct mapping));
485 /* Map MAPFUN over all the mappings in hash table HT. MAPFUN is
486 called with three arguments: the key, the value, and MAPARG.
488 It is undefined what happens if you add or remove entries in the
489 hash table while hash_table_map is running. The exception is the
490 entry you're currently mapping over; you may remove or change that
494 hash_table_map (struct hash_table *ht,
495 int (*mapfun) (void *, void *, void *),
498 struct mapping *mp = ht->mappings;
499 struct mapping *end = ht->mappings + ht->size;
501 for (; mp < end; mp++)
507 if (mapfun (key, mp->value, maparg))
509 /* hash_table_remove might have moved the adjacent
511 if (mp->key != key && NON_EMPTY (mp))
516 /* Return the number of elements in the hash table. This is not the
517 same as the physical size of the hash table, which is always
518 greater than the number of elements. */
521 hash_table_count (const struct hash_table *ht)
526 /* Functions from this point onward are meant for convenience and
527 don't strictly belong to this file. However, this is as good a
528 place for them as any. */
531 * Support for hash tables whose keys are strings.
535 /* 31 bit hash function. Taken from Gnome's glib, modified to use
538 We used to use the popular hash function from the Dragon Book, but
539 this one seems to perform much better. */
542 string_hash (const void *key)
548 for (p += 1; *p != '\0'; p++)
549 h = (h << 5) - h + *p;
554 /* Frontend for strcmp usable for hash tables. */
557 string_cmp (const void *s1, const void *s2)
559 return !strcmp ((const char *)s1, (const char *)s2);
562 /* Return a hash table of preallocated to store at least ITEMS items
563 suitable to use strings as keys. */
566 make_string_hash_table (int items)
568 return hash_table_new (items, string_hash, string_cmp);
572 * Support for hash tables whose keys are strings, but which are
573 * compared case-insensitively.
577 /* Like string_hash, but produce the same hash regardless of the case. */
580 string_hash_nocase (const void *key)
583 unsigned int h = TOLOWER (*p);
586 for (p += 1; *p != '\0'; p++)
587 h = (h << 5) - h + TOLOWER (*p);
592 /* Like string_cmp, but doing case-insensitive compareison. */
595 string_cmp_nocase (const void *s1, const void *s2)
597 return !strcasecmp ((const char *)s1, (const char *)s2);
600 /* Like make_string_hash_table, but uses string_hash_nocase and
601 string_cmp_nocase. */
604 make_nocase_string_hash_table (int items)
606 return hash_table_new (items, string_hash_nocase, string_cmp_nocase);
609 /* Hashing of pointers. Used for hash tables that are keyed by
610 pointer identity. (Common Lisp calls them EQ hash tables, and Java
611 calls them IdentityHashMaps.) */
614 ptrhash (const void *ptr)
616 unsigned long key = (unsigned long)ptr;
639 ptrcmp (const void *ptr1, const void *ptr2)
645 /* Currently unused: hashing of integers. */
648 inthash (unsigned int key)
668 print_hash_table_mapper (void *key, void *value, void *count)
671 printf ("%s: %s\n", (const char *)key, (char *)value);
676 print_hash (struct hash_table *sht)
679 hash_table_map (sht, print_hash_table_mapper, &debug_count);
680 assert (debug_count == sht->count);
686 struct hash_table *ht = make_string_hash_table (0);
688 while ((fgets (line, sizeof (line), stdin)))
690 int len = strlen (line);
694 if (!hash_table_contains (ht, line))
695 hash_table_put (ht, strdup (line), "here I am!");
700 if (hash_table_get_pair (ht, line, &line_copy, NULL))
702 hash_table_remove (ht, line);
712 printf ("%d %d\n", ht->count, ht->size);