/* Hash tables.
- Copyright (C) 2000, 2001 Free Software Foundation, Inc.
+ Copyright (C) 2000-2003 Free Software Foundation, Inc.
This file is part of GNU Wget.
file, but you are not obligated to do so. If you do not wish to do
so, delete this exception statement from your version. */
+/* With -DSTANDALONE, this file can be compiled outside Wget source
+ tree. To test, also use -DTEST. */
+
#ifdef HAVE_CONFIG_H
# include <config.h>
#endif
-#ifdef HAVE_STRING_H
-# include <string.h>
-#else
-# include <strings.h>
-#endif /* HAVE_STRING_H */
+#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
+#include <string.h>
+#include <limits.h>
-#include "wget.h"
-#include "utils.h"
-
-#include "hash.h"
-
-#ifdef STANDALONE
-# undef xmalloc
-# undef xrealloc
-# undef xfree
-
+#ifndef STANDALONE
+/* Get Wget's utility headers. */
+# include "wget.h"
+# include "utils.h"
+#else
+/* Make do without them. */
+# define xnew(x) xmalloc (sizeof (x))
+# define xnew_array(type, x) xmalloc (sizeof (type) * (x))
# define xmalloc malloc
-# define xrealloc realloc
# define xfree free
-
-# undef TOLOWER
+# define countof(x) (sizeof (x) / sizeof ((x)[0]))
# define TOLOWER(x) ('A' <= (x) && (x) <= 'Z' ? (x) - 32 : (x))
#endif
+#include "hash.h"
+
/* INTERFACE:
Hash tables are a technique used to implement mapping between
The hash table is implemented as an open-addressed table with
linear probing collision resolution.
- In regular language, it means that all the hash entries (pairs of
- pointers key and value) are stored in a contiguous array. The
- position of each mapping is determined by the hash value of its key
- and the size of the table: location := hash(key) % size. If two
- different keys end up on the same position (collide), the one that
- came second is placed at the next empty position following the
- occupied place. This collision resolution technique is called
- "linear probing".
+ The above means that all the hash entries (pairs of pointers, key
+ and value) are stored in a contiguous array. The position of each
+ mapping is determined by the hash value of its key and the size of
+ the table: location := hash(key) % size. If two different keys end
+ up on the same position (collide), the one that came second is
+ placed at the next empty position following the occupied place.
+ This collision resolution technique is called "linear probing".
There are more advanced collision resolution methods (quadratic
probing, double hashing), but we don't use them because they incur
void *value;
};
-typedef unsigned long (*hashfun_t) PARAMS ((const void *));
-typedef int (*testfun_t) PARAMS ((const void *, const void *));
+typedef unsigned long (*hashfun_t) (const void *);
+typedef int (*testfun_t) (const void *, const void *);
struct hash_table {
hashfun_t hash_function;
the prime table. */
};
-/* We use all-bit-set marker to mean that a mapping is empty. It is
- (hopefully) illegal as a pointer, and it allows the users to use
- NULL (as well as any non-negative integer) as key. */
+/* We use the all-bits-set constant (INVALID_PTR) marker to mean that
+ a mapping is empty. It is unaligned and therefore illegal as a
+ pointer. INVALID_PTR_BYTE (0xff) is the one-byte value used to
+ initialize the mappings array as empty.
-#define NON_EMPTY(mp) (mp->key != (void *)~(unsigned long)0)
-#define MARK_AS_EMPTY(mp) (mp->key = (void *)~(unsigned long)0)
+ The all-bits-set value is a better choice than NULL because it
+ allows the use of NULL/0 keys. Since the keys are either integers
+ or pointers, the only key that cannot be used is the integer value
+ -1. This is acceptable because it still allows the use of
+ nonnegative integer keys. */
+
+#define INVALID_PTR ((void *) ~0UL)
+#ifndef UCHAR_MAX
+# define UCHAR_MAX 0xff
+#endif
+#define INVALID_PTR_BYTE UCHAR_MAX
+
+#define NON_EMPTY(mp) ((mp)->key != INVALID_PTR)
+#define MARK_AS_EMPTY(mp) ((mp)->key = INVALID_PTR)
/* "Next" mapping is the mapping after MP, but wrapping back to
MAPPINGS when MP would reach MAPPINGS+SIZE. */
being HASHFUN. */
#define HASH_POSITION(key, hashfun, size) ((hashfun) (key) % size)
-/* Find a prime near, but greather than or equal to SIZE. Of course,
- the primes are not calculated, but looked up from a table. The
- table does not contain all primes in range, just a selection useful
+/* Find a prime near, but greather than or equal to SIZE. The primes
+ are looked up from a table with a selection of primes convenient
for this purpose.
PRIME_OFFSET is a minor optimization: it specifies start position
static int
prime_size (int size, int *prime_offset)
{
- static const unsigned long primes [] = {
+ static const int primes[] = {
13, 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
- 1174703521, 1527114613, 1985248999,
- (unsigned long)0x99d43ea5, (unsigned long)0xc7fa5177
+ 1174703521, 1527114613, 1837299131, 2147483647
};
int i;
}
abort ();
- return 0;
}
-static unsigned long ptrhash PARAMS ((const void *));
-static int ptrcmp PARAMS ((const void *, const void *));
+static int cmp_pointer (const void *, const void *);
/* Create a hash table with hash function HASH_FUNCTION and test
function TEST_FUNCTION. The table is empty (its count is 0), but
needed, simply specify zero ITEMS.
If hash and test callbacks are not specified, identity mapping is
- assumed, i.e. pointer values are used for key comparison. If,
- instead of that, you want strings with equal contents to hash the
- same, use make_string_hash_table. */
+ assumed, i.e. pointer values are used for key comparison. (Common
+ Lisp calls such tables EQ hash tables, and Java calls them
+ IdentityHashMaps.) If your keys require different comparison,
+ specify hash and test functions. For easy use of C strings as hash
+ keys, you can use the convenience functions make_string_hash_table
+ and make_nocase_string_hash_table. */
struct hash_table *
hash_table_new (int items,
int size;
struct hash_table *ht = xnew (struct hash_table);
- ht->hash_function = hash_function ? hash_function : ptrhash;
- ht->test_function = test_function ? test_function : ptrcmp;
+ ht->hash_function = hash_function ? hash_function : hash_pointer;
+ ht->test_function = test_function ? test_function : cmp_pointer;
/* If the size of struct hash_table ever becomes a concern, this
field can go. (Wget doesn't create many hashes.) */
/*assert (ht->resize_threshold >= items);*/
ht->mappings = xnew_array (struct mapping, ht->size);
+
/* Mark mappings as empty. We use 0xff rather than 0 to mark empty
- keys because it allows us to store NULL keys to the table. */
- memset (ht->mappings, 0xff, size * sizeof (struct mapping));
+ keys because it allows us to use NULL/0 as keys. */
+ memset (ht->mappings, INVALID_PTR_BYTE, size * sizeof (struct mapping));
ht->count = 0;
ht->resize_threshold = newsize * HASH_MAX_FULLNESS;
mappings = xnew_array (struct mapping, newsize);
- memset (mappings, 0xff, newsize * sizeof (struct mapping));
+ memset (mappings, INVALID_PTR_BYTE, newsize * sizeof (struct mapping));
ht->mappings = mappings;
for (mp = old_mappings; mp < old_end; mp++)
void
hash_table_clear (struct hash_table *ht)
{
- memset (ht->mappings, 0xff, ht->size * sizeof (struct mapping));
+ memset (ht->mappings, INVALID_PTR_BYTE, ht->size * sizeof (struct mapping));
ht->count = 0;
}
don't strictly belong to this file. However, this is as good a
place for them as any. */
-/* Rules for creating custom hash and test functions:
+/* Guidelines for creating custom hash and test functions:
- The test function returns non-zero for keys that are considered
"equal", zero otherwise.
We used to use the popular hash function from the Dragon Book, but
this one seems to perform much better. */
-unsigned long
-string_hash (const void *key)
+static unsigned long
+hash_string (const void *key)
{
const char *p = key;
unsigned int h = *p;
/* Frontend for strcmp usable for hash tables. */
-int
-string_cmp (const void *s1, const void *s2)
+static int
+cmp_string (const void *s1, const void *s2)
{
return !strcmp ((const char *)s1, (const char *)s2);
}
struct hash_table *
make_string_hash_table (int items)
{
- return hash_table_new (items, string_hash, string_cmp);
+ return hash_table_new (items, hash_string, cmp_string);
}
/*
*
*/
-/* Like string_hash, but produce the same hash regardless of the case. */
+/* Like hash_string, but produce the same hash regardless of the case. */
static unsigned long
-string_hash_nocase (const void *key)
+hash_string_nocase (const void *key)
{
const char *p = key;
unsigned int h = TOLOWER (*p);
struct hash_table *
make_nocase_string_hash_table (int items)
{
- return hash_table_new (items, string_hash_nocase, string_cmp_nocase);
+ return hash_table_new (items, hash_string_nocase, string_cmp_nocase);
}
-/* Hashing of numeric values, such as pointers and integers. Used for
- hash tables that are keyed by pointer identity. (Common Lisp calls
- them EQ hash tables, and Java calls them IdentityHashMaps.)
+/* Hashing of numeric values, such as pointers and integers.
This implementation is the Robert Jenkins' 32 bit Mix Function,
with a simple adaptation for 64-bit values. It offers excellent
spreading of values and doesn't need to know the hash table size to
work (unlike the very popular Knuth's multiplication hash). */
-static unsigned long
-ptrhash (const void *ptr)
+unsigned long
+hash_pointer (const void *ptr)
{
unsigned long key = (unsigned long)ptr;
key += (key << 12);
}
static int
-ptrcmp (const void *ptr1, const void *ptr2)
+cmp_pointer (const void *ptr1, const void *ptr2)
{
return ptr1 == ptr2;
}
\f
-#ifdef STANDALONE
+#ifdef TEST
#include <stdio.h>
#include <string.h>
#endif
return 0;
}
-#endif
+#endif /* TEST */
projects / wget / blobdiff