+2003-10-25 Paul Eggert <eggert@twinsun.com>
+
+ Fix several address-calculation bugs in the hash modules,
+ plus some minor code cleanup.
+
+ * hash.h: Include <stdbool.h>, for bool.
+ * hash.c: Don't include <stdbool.h>, since hash.h does it now.
+ * hash.h (Hash_hasher, hash_get_n_buckets, hash_get_n_buckets_used,
+ hash_get_n_entries, hash_get_max_bucket_length,
+ hash_get_entries, hash_do_for_each, hash_string, hash_initialize,
+ hash_rehash): Use size_t rather than unsigned.
+ * hash.c (struct hash_table, hash_get_n_buckets,
+ hash_get_n_buckets_used, hash_get_n_entries,
+ hash_get_max_bucket_length, hash_table_ok, hash_print_statistics,
+ hash_get_entries, hash_do_for_each, hash_string, is_prime,
+ next_prime, hash_initialize, hash_rehash, hash_delete, hash_print):
+ Likewise.
+ (SIZE_MAX): Define if not defined.
+ (hash_get_max_bucket_length, hash_table_ok, hash_lookup,
+ hash_get_first, hash_get_next, hash_get_entries, hash_do_for_each,
+ hash_print):
+ Use const * when possible.
+ (hash_string): Use (unsigned char) *P rather than *(unsigned char *) P.
+ (check_tuning): Fix bug: if tuning parameters were very close to
+ 0 or 1, rounding errors could have caused subscript violations.
+ (hash_initialize, allocate_entry, hash_print): Remove unnecessary cast.
+ (hash_initialize): Add 'fail:' label
+ to free table and return NULL, and use it to simplify code.
+ Use calloc rather than clearing the storage ourself.
+ (hash_initialize, hash_rehash): Check for arithmetic overflow in
+ buffer size calculations.
+ * hash-pjw.h (hash_pjw): Use size_t, not unsigned.
+ Include <stddef.h>, for size_t.
+ * hash-pjw.c (hash_pjw): Likewise.
+ Switch to method described by Bruno Haible.
+ Include <limits.h>, for CHAR_BIT.
+ (SIZE_BITS): New macro.
+
2003-10-21 Paul Eggert <eggert@twinsun.com>
* getndelim2.c (getndelim2): When size calculation overflows,
# include <config.h>
#endif
+#include "hash.h"
+
#include <limits.h>
-#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
# endif
#endif
-#include "hash.h"
+#ifndef SIZE_MAX
+# define SIZE_MAX ((size_t) -1)
+#endif
struct hash_table
{
for a possibility of N_BUCKETS. Among those, N_BUCKETS_USED buckets
are not empty, there are N_ENTRIES active entries in the table. */
struct hash_entry *bucket;
- struct hash_entry *bucket_limit;
- unsigned n_buckets;
- unsigned n_buckets_used;
- unsigned n_entries;
+ struct hash_entry const *bucket_limit;
+ size_t n_buckets;
+ size_t n_buckets_used;
+ size_t n_entries;
/* Tuning arguments, kept in a physicaly separate structure. */
const Hash_tuning *tuning;
number of buckets (used plus unused), or the maximum number of slots, are
the same quantity. */
-unsigned
+size_t
hash_get_n_buckets (const Hash_table *table)
{
return table->n_buckets;
/* Return the number of slots in use (non-empty buckets). */
-unsigned
+size_t
hash_get_n_buckets_used (const Hash_table *table)
{
return table->n_buckets_used;
/* Return the number of active entries. */
-unsigned
+size_t
hash_get_n_entries (const Hash_table *table)
{
return table->n_entries;
/* Return the length of the longest chain (bucket). */
-unsigned
+size_t
hash_get_max_bucket_length (const Hash_table *table)
{
- struct hash_entry *bucket;
- unsigned max_bucket_length = 0;
+ struct hash_entry const *bucket;
+ size_t max_bucket_length = 0;
for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
{
if (bucket->data)
{
- struct hash_entry *cursor = bucket;
- unsigned bucket_length = 1;
+ struct hash_entry const *cursor = bucket;
+ size_t bucket_length = 1;
while (cursor = cursor->next, cursor)
bucket_length++;
bool
hash_table_ok (const Hash_table *table)
{
- struct hash_entry *bucket;
- unsigned n_buckets_used = 0;
- unsigned n_entries = 0;
+ struct hash_entry const *bucket;
+ size_t n_buckets_used = 0;
+ size_t n_entries = 0;
for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
{
if (bucket->data)
{
- struct hash_entry *cursor = bucket;
+ struct hash_entry const *cursor = bucket;
/* Count bucket head. */
n_buckets_used++;
void
hash_print_statistics (const Hash_table *table, FILE *stream)
{
- unsigned n_entries = hash_get_n_entries (table);
- unsigned n_buckets = hash_get_n_buckets (table);
- unsigned n_buckets_used = hash_get_n_buckets_used (table);
- unsigned max_bucket_length = hash_get_max_bucket_length (table);
-
- fprintf (stream, "# entries: %u\n", n_entries);
- fprintf (stream, "# buckets: %u\n", n_buckets);
- fprintf (stream, "# buckets used: %u (%.2f%%)\n", n_buckets_used,
+ size_t n_entries = hash_get_n_entries (table);
+ size_t n_buckets = hash_get_n_buckets (table);
+ size_t n_buckets_used = hash_get_n_buckets_used (table);
+ size_t max_bucket_length = hash_get_max_bucket_length (table);
+
+ fprintf (stream, "# entries: %lu\n", (unsigned long int) n_entries);
+ fprintf (stream, "# buckets: %lu\n", (unsigned long int) n_buckets);
+ fprintf (stream, "# buckets used: %lu (%.2f%%)\n",
+ (unsigned long int) n_buckets_used,
(100.0 * n_buckets_used) / n_buckets);
- fprintf (stream, "max bucket length: %u\n", max_bucket_length);
+ fprintf (stream, "max bucket length: %lu\n",
+ (unsigned long int) max_bucket_length);
}
/* If ENTRY matches an entry already in the hash table, return the
void *
hash_lookup (const Hash_table *table, const void *entry)
{
- struct hash_entry *bucket
+ struct hash_entry const *bucket
= table->bucket + table->hasher (entry, table->n_buckets);
- struct hash_entry *cursor;
+ struct hash_entry const *cursor;
if (! (bucket < table->bucket_limit))
abort ();
void *
hash_get_first (const Hash_table *table)
{
- struct hash_entry *bucket;
+ struct hash_entry const *bucket;
if (table->n_entries == 0)
return NULL;
void *
hash_get_next (const Hash_table *table, const void *entry)
{
- struct hash_entry *bucket
+ struct hash_entry const *bucket
= table->bucket + table->hasher (entry, table->n_buckets);
- struct hash_entry *cursor;
+ struct hash_entry const *cursor;
if (! (bucket < table->bucket_limit))
abort ();
return the number of pointers copied. Do not copy more than BUFFER_SIZE
pointers. */
-unsigned
+size_t
hash_get_entries (const Hash_table *table, void **buffer,
- unsigned buffer_size)
+ size_t buffer_size)
{
- unsigned counter = 0;
- struct hash_entry *bucket;
- struct hash_entry *cursor;
+ size_t counter = 0;
+ struct hash_entry const *bucket;
+ struct hash_entry const *cursor;
for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
{
as received. The walking continue for as long as the PROCESSOR function
returns nonzero. When it returns zero, the walking is interrupted. */
-unsigned
+size_t
hash_do_for_each (const Hash_table *table, Hash_processor processor,
void *processor_data)
{
- unsigned counter = 0;
- struct hash_entry *bucket;
- struct hash_entry *cursor;
+ size_t counter = 0;
+ struct hash_entry const *bucket;
+ struct hash_entry const *cursor;
for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
{
algorithms tend to be domain-specific, so what's good for [diffutils'] io.c
may not be good for your application." */
-unsigned
-hash_string (const char *string, unsigned n_buckets)
+size_t
+hash_string (const char *string, size_t n_buckets)
{
# define ROTATE_LEFT(Value, Shift) \
- ((Value) << (Shift) | (Value) >> ((sizeof (unsigned) * CHAR_BIT) - (Shift)))
+ ((Value) << (Shift) | (Value) >> ((sizeof (size_t) * CHAR_BIT) - (Shift)))
# define HASH_ONE_CHAR(Value, Byte) \
((Byte) + ROTATE_LEFT (Value, 7))
- unsigned value = 0;
+ size_t value = 0;
for (; *string; string++)
- value = HASH_ONE_CHAR (value, *(const unsigned char *) string);
+ value = HASH_ONE_CHAR (value, (unsigned char) *string);
return value % n_buckets;
# undef ROTATE_LEFT
very old Cyber `snoop', itself written in typical Greg Mansfield style.
(By the way, what happened to this excellent man? Is he still alive?) */
-unsigned
-hash_string (const char *string, unsigned n_buckets)
+size_t
+hash_string (const char *string, size_t n_buckets)
{
- unsigned value = 0;
+ size_t value = 0;
while (*string)
- value = ((value * 31 + (int) *(const unsigned char *) string++)
- % n_buckets);
+ value = (value * 31 + (unsigned char) *string++) % n_buckets;
return value;
}
number at least equal to 11. */
static bool
-is_prime (unsigned long candidate)
+is_prime (size_t candidate)
{
- unsigned long divisor = 3;
- unsigned long square = divisor * divisor;
+ size_t divisor = 3;
+ size_t square = divisor * divisor;
while (square < candidate && (candidate % divisor))
{
/* Round a given CANDIDATE number up to the nearest prime, and return that
prime. Primes lower than 10 are merely skipped. */
-static unsigned long
-next_prime (unsigned long candidate)
+static size_t
+next_prime (size_t candidate)
{
/* Skip small primes. */
if (candidate < 10)
{
const Hash_tuning *tuning = table->tuning;
- if (tuning->growth_threshold > 0.0
- && tuning->growth_threshold < 1.0
- && tuning->growth_factor > 1.0
- && tuning->shrink_threshold >= 0.0
- && tuning->shrink_threshold < 1.0
- && tuning->shrink_factor > tuning->shrink_threshold
- && tuning->shrink_factor <= 1.0
- && tuning->shrink_threshold < tuning->growth_threshold)
+ /* Be a bit stricter than mathematics would require, so that
+ rounding errors in size calculations do not cause allocations to
+ fail to grow or shrink as they should. The smallest allocation
+ is 11 (due to next_prime's algorithm), so an epsilon of 0.1
+ should be good enough. */
+ float epsilon = 0.1f;
+
+ if (epsilon < tuning->growth_threshold
+ && tuning->growth_threshold < 1 - epsilon
+ && 1 + epsilon < tuning->growth_factor
+ && 0 <= tuning->shrink_threshold
+ && tuning->shrink_threshold + epsilon < tuning->shrink_factor
+ && tuning->shrink_factor <= 1
+ && tuning->shrink_threshold + epsilon < tuning->growth_threshold)
return true;
table->tuning = &default_tuning;
values. */
Hash_table *
-hash_initialize (unsigned candidate, const Hash_tuning *tuning,
+hash_initialize (size_t candidate, const Hash_tuning *tuning,
Hash_hasher hasher, Hash_comparator comparator,
Hash_data_freer data_freer)
{
Hash_table *table;
- struct hash_entry *bucket;
if (hasher == NULL || comparator == NULL)
return NULL;
- table = (Hash_table *) malloc (sizeof (Hash_table));
+ table = malloc (sizeof *table);
if (table == NULL)
return NULL;
if the user provides invalid tuning options, we silently revert to
using the defaults, and ignore further request to change the tuning
options. */
- free (table);
- return NULL;
+ goto fail;
}
- table->n_buckets
- = next_prime (tuning->is_n_buckets ? candidate
- : (unsigned) (candidate / tuning->growth_threshold));
-
- table->bucket = (struct hash_entry *)
- malloc (table->n_buckets * sizeof (struct hash_entry));
- if (table->bucket == NULL)
+ if (!tuning->is_n_buckets)
{
- free (table);
- return NULL;
+ float new_candidate = candidate / tuning->growth_threshold;
+ if (SIZE_MAX <= new_candidate)
+ goto fail;
+ candidate = new_candidate;
}
- table->bucket_limit = table->bucket + table->n_buckets;
- for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
- {
- bucket->data = NULL;
- bucket->next = NULL;
- }
+ if (SIZE_MAX / sizeof *table->bucket < candidate)
+ goto fail;
+ table->n_buckets = next_prime (candidate);
+ if (SIZE_MAX / sizeof *table->bucket < table->n_buckets)
+ goto fail;
+
+ table->bucket = calloc (table->n_buckets, sizeof *table->bucket);
+ table->bucket_limit = table->bucket + table->n_buckets;
table->n_buckets_used = 0;
table->n_entries = 0;
obstack_init (&table->entry_stack);
#endif
return table;
+
+ fail:
+ free (table);
+ return NULL;
}
/* Make all buckets empty, placing any chained entries on the free list.
else
{
#if USE_OBSTACK
- new = (struct hash_entry *)
- obstack_alloc (&table->entry_stack, sizeof (struct hash_entry));
+ new = obstack_alloc (&table->entry_stack, sizeof *new);
#else
- new = (struct hash_entry *) malloc (sizeof (struct hash_entry));
+ new = malloc (sizeof *new);
#endif
}
exact number of buckets desired. */
bool
-hash_rehash (Hash_table *table, unsigned candidate)
+hash_rehash (Hash_table *table, size_t candidate)
{
Hash_table *new_table;
struct hash_entry *bucket;
> table->tuning->growth_threshold * table->n_buckets)
{
const Hash_tuning *tuning = table->tuning;
- unsigned candidate
- = (unsigned) (tuning->is_n_buckets
- ? (table->n_buckets * tuning->growth_factor)
- : (table->n_buckets * tuning->growth_factor
- * tuning->growth_threshold));
+ float candidate =
+ (tuning->is_n_buckets
+ ? (table->n_buckets * tuning->growth_factor)
+ : (table->n_buckets * tuning->growth_factor
+ * tuning->growth_threshold));
+
+ if (SIZE_MAX <= candidate)
+ return NULL;
/* If the rehash fails, arrange to return NULL. */
if (!hash_rehash (table, candidate))
< table->tuning->shrink_threshold * table->n_buckets)
{
const Hash_tuning *tuning = table->tuning;
- unsigned candidate
- = (unsigned) (tuning->is_n_buckets
- ? table->n_buckets * tuning->shrink_factor
- : (table->n_buckets * tuning->shrink_factor
- * tuning->growth_threshold));
+ size_t candidate =
+ (tuning->is_n_buckets
+ ? table->n_buckets * tuning->shrink_factor
+ : (table->n_buckets * tuning->shrink_factor
+ * tuning->growth_threshold));
hash_rehash (table, candidate);
}
void
hash_print (const Hash_table *table)
{
- struct hash_entry *bucket;
+ struct hash_entry const *bucket;
for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
{
struct hash_entry *cursor;
if (bucket)
- printf ("%d:\n", bucket - table->bucket);
+ printf ("%lu:\n", (unsigned long int) (bucket - table->bucket));
for (cursor = bucket; cursor; cursor = cursor->next)
{
- char *s = (char *) cursor->data;
+ char const *s = cursor->data;
/* FIXME */
if (s)
printf (" %s\n", s);
# define HASH_H_
# include <stdio.h>
+# include <stdbool.h>
-typedef unsigned (*Hash_hasher) (const void *, unsigned);
+typedef size_t (*Hash_hasher) (const void *, size_t);
typedef bool (*Hash_comparator) (const void *, const void *);
typedef void (*Hash_data_freer) (void *);
typedef bool (*Hash_processor) (void *, void *);
typedef struct hash_table Hash_table;
/* Information and lookup. */
-unsigned hash_get_n_buckets (const Hash_table *);
-unsigned hash_get_n_buckets_used (const Hash_table *);
-unsigned hash_get_n_entries (const Hash_table *);
-unsigned hash_get_max_bucket_length (const Hash_table *);
+size_t hash_get_n_buckets (const Hash_table *);
+size_t hash_get_n_buckets_used (const Hash_table *);
+size_t hash_get_n_entries (const Hash_table *);
+size_t hash_get_max_bucket_length (const Hash_table *);
bool hash_table_ok (const Hash_table *);
void hash_print_statistics (const Hash_table *, FILE *);
void *hash_lookup (const Hash_table *, const void *);
/* Walking. */
void *hash_get_first (const Hash_table *);
void *hash_get_next (const Hash_table *, const void *);
-unsigned hash_get_entries (const Hash_table *, void **, unsigned);
-unsigned hash_do_for_each (const Hash_table *, Hash_processor, void *);
+size_t hash_get_entries (const Hash_table *, void **, size_t);
+size_t hash_do_for_each (const Hash_table *, Hash_processor, void *);
/* Allocation and clean-up. */
-unsigned hash_string (const char *, unsigned);
+size_t hash_string (const char *, size_t);
void hash_reset_tuning (Hash_tuning *);
-Hash_table *hash_initialize (unsigned, const Hash_tuning *,
+Hash_table *hash_initialize (size_t, const Hash_tuning *,
Hash_hasher, Hash_comparator,
Hash_data_freer);
void hash_clear (Hash_table *);
void hash_free (Hash_table *);
/* Insertion and deletion. */
-bool hash_rehash (Hash_table *, unsigned);
+bool hash_rehash (Hash_table *, size_t);
void *hash_insert (Hash_table *, const void *);
void *hash_delete (Hash_table *, const void *);
projects / pspp / commitdiff