1/* hash - hashing table processing.
2
3   Copyright (C) 1998-2004, 2006-2007, 2009-2012 Free Software Foundation, Inc.
4
5   Written by Jim Meyering, 1992.
6
7   This program is free software: you can redistribute it and/or modify
8   it under the terms of the GNU General Public License as published by
9   the Free Software Foundation; either version 3 of the License, or
10   (at your option) any later version.
11
12   This program is distributed in the hope that it will be useful,
13   but WITHOUT ANY WARRANTY; without even the implied warranty of
14   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15   GNU General Public License for more details.
16
17   You should have received a copy of the GNU General Public License
18   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
19
20/* A generic hash table package.  */
21
22/* Define USE_OBSTACK to 1 if you want the allocator to use obstacks instead
23   of malloc.  If you change USE_OBSTACK, you have to recompile!  */
24
25#include <config.h>
26
27#include "hash.h"
28
29#include "bitrotate.h"
30#include "xalloc-oversized.h"
31
32#include <stdint.h>
33#include <stdio.h>
34#include <stdlib.h>
35
36#if USE_OBSTACK
37# include "obstack.h"
38# ifndef obstack_chunk_alloc
39#  define obstack_chunk_alloc malloc
40# endif
41# ifndef obstack_chunk_free
42#  define obstack_chunk_free free
43# endif
44#endif
45
46struct hash_entry
47  {
48    void *data;
49    struct hash_entry *next;
50  };
51
52struct hash_table
53  {
54    /* The array of buckets starts at BUCKET and extends to BUCKET_LIMIT-1,
55       for a possibility of N_BUCKETS.  Among those, N_BUCKETS_USED buckets
56       are not empty, there are N_ENTRIES active entries in the table.  */
57    struct hash_entry *bucket;
58    struct hash_entry const *bucket_limit;
59    size_t n_buckets;
60    size_t n_buckets_used;
61    size_t n_entries;
62
63    /* Tuning arguments, kept in a physically separate structure.  */
64    const Hash_tuning *tuning;
65
66    /* Three functions are given to 'hash_initialize', see the documentation
67       block for this function.  In a word, HASHER randomizes a user entry
68       into a number up from 0 up to some maximum minus 1; COMPARATOR returns
69       true if two user entries compare equally; and DATA_FREER is the cleanup
70       function for a user entry.  */
71    Hash_hasher hasher;
72    Hash_comparator comparator;
73    Hash_data_freer data_freer;
74
75    /* A linked list of freed struct hash_entry structs.  */
76    struct hash_entry *free_entry_list;
77
78#if USE_OBSTACK
79    /* Whenever obstacks are used, it is possible to allocate all overflowed
80       entries into a single stack, so they all can be freed in a single
81       operation.  It is not clear if the speedup is worth the trouble.  */
82    struct obstack entry_stack;
83#endif
84  };
85
86/* A hash table contains many internal entries, each holding a pointer to
87   some user-provided data (also called a user entry).  An entry indistinctly
88   refers to both the internal entry and its associated user entry.  A user
89   entry contents may be hashed by a randomization function (the hashing
90   function, or just "hasher" for short) into a number (or "slot") between 0
91   and the current table size.  At each slot position in the hash table,
92   starts a linked chain of entries for which the user data all hash to this
93   slot.  A bucket is the collection of all entries hashing to the same slot.
94
95   A good "hasher" function will distribute entries rather evenly in buckets.
96   In the ideal case, the length of each bucket is roughly the number of
97   entries divided by the table size.  Finding the slot for a data is usually
98   done in constant time by the "hasher", and the later finding of a precise
99   entry is linear in time with the size of the bucket.  Consequently, a
100   larger hash table size (that is, a larger number of buckets) is prone to
101   yielding shorter chains, *given* the "hasher" function behaves properly.
102
103   Long buckets slow down the lookup algorithm.  One might use big hash table
104   sizes in hope to reduce the average length of buckets, but this might
105   become inordinate, as unused slots in the hash table take some space.  The
106   best bet is to make sure you are using a good "hasher" function (beware
107   that those are not that easy to write! :-), and to use a table size
108   larger than the actual number of entries.  */
109
110/* If an insertion makes the ratio of nonempty buckets to table size larger
111   than the growth threshold (a number between 0.0 and 1.0), then increase
112   the table size by multiplying by the growth factor (a number greater than
113   1.0).  The growth threshold defaults to 0.8, and the growth factor
114   defaults to 1.414, meaning that the table will have doubled its size
115   every second time 80% of the buckets get used.  */
116#define DEFAULT_GROWTH_THRESHOLD 0.8f
117#define DEFAULT_GROWTH_FACTOR 1.414f
118
119/* If a deletion empties a bucket and causes the ratio of used buckets to
120   table size to become smaller than the shrink threshold (a number between
121   0.0 and 1.0), then shrink the table by multiplying by the shrink factor (a
122   number greater than the shrink threshold but smaller than 1.0).  The shrink
123   threshold and factor default to 0.0 and 1.0, meaning that the table never
124   shrinks.  */
125#define DEFAULT_SHRINK_THRESHOLD 0.0f
126#define DEFAULT_SHRINK_FACTOR 1.0f
127
128/* Use this to initialize or reset a TUNING structure to
129   some sensible values. */
130static const Hash_tuning default_tuning =
131  {
132    DEFAULT_SHRINK_THRESHOLD,
133    DEFAULT_SHRINK_FACTOR,
134    DEFAULT_GROWTH_THRESHOLD,
135    DEFAULT_GROWTH_FACTOR,
136    false
137  };
138
139/* Information and lookup.  */
140
141/* The following few functions provide information about the overall hash
142   table organization: the number of entries, number of buckets and maximum
143   length of buckets.  */
144
145/* Return the number of buckets in the hash table.  The table size, the total
146   number of buckets (used plus unused), or the maximum number of slots, are
147   the same quantity.  */
148
149size_t
150hash_get_n_buckets (const Hash_table *table)
151{
152  return table->n_buckets;
153}
154
155/* Return the number of slots in use (non-empty buckets).  */
156
157size_t
158hash_get_n_buckets_used (const Hash_table *table)
159{
160  return table->n_buckets_used;
161}
162
163/* Return the number of active entries.  */
164
165size_t
166hash_get_n_entries (const Hash_table *table)
167{
168  return table->n_entries;
169}
170
171/* Return the length of the longest chain (bucket).  */
172
173size_t
174hash_get_max_bucket_length (const Hash_table *table)
175{
176  struct hash_entry const *bucket;
177  size_t max_bucket_length = 0;
178
179  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
180    {
181      if (bucket->data)
182        {
183          struct hash_entry const *cursor = bucket;
184          size_t bucket_length = 1;
185
186          while (cursor = cursor->next, cursor)
187            bucket_length++;
188
189          if (bucket_length > max_bucket_length)
190            max_bucket_length = bucket_length;
191        }
192    }
193
194  return max_bucket_length;
195}
196
197/* Do a mild validation of a hash table, by traversing it and checking two
198   statistics.  */
199
200bool
201hash_table_ok (const Hash_table *table)
202{
203  struct hash_entry const *bucket;
204  size_t n_buckets_used = 0;
205  size_t n_entries = 0;
206
207  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
208    {
209      if (bucket->data)
210        {
211          struct hash_entry const *cursor = bucket;
212
213          /* Count bucket head.  */
214          n_buckets_used++;
215          n_entries++;
216
217          /* Count bucket overflow.  */
218          while (cursor = cursor->next, cursor)
219            n_entries++;
220        }
221    }
222
223  if (n_buckets_used == table->n_buckets_used && n_entries == table->n_entries)
224    return true;
225
226  return false;
227}
228
229void
230hash_print_statistics (const Hash_table *table, FILE *stream)
231{
232  size_t n_entries = hash_get_n_entries (table);
233  size_t n_buckets = hash_get_n_buckets (table);
234  size_t n_buckets_used = hash_get_n_buckets_used (table);
235  size_t max_bucket_length = hash_get_max_bucket_length (table);
236
237  fprintf (stream, "# entries:         %lu\n", (unsigned long int) n_entries);
238  fprintf (stream, "# buckets:         %lu\n", (unsigned long int) n_buckets);
239  fprintf (stream, "# buckets used:    %lu (%.2f%%)\n",
240           (unsigned long int) n_buckets_used,
241           (100.0 * n_buckets_used) / n_buckets);
242  fprintf (stream, "max bucket length: %lu\n",
243           (unsigned long int) max_bucket_length);
244}
245
246/* Hash KEY and return a pointer to the selected bucket.
247   If TABLE->hasher misbehaves, abort.  */
248static struct hash_entry *
249safe_hasher (const Hash_table *table, const void *key)
250{
251  size_t n = table->hasher (key, table->n_buckets);
252  if (! (n < table->n_buckets))
253    abort ();
254  return table->bucket + n;
255}
256
257/* If ENTRY matches an entry already in the hash table, return the
258   entry from the table.  Otherwise, return NULL.  */
259
260void *
261hash_lookup (const Hash_table *table, const void *entry)
262{
263  struct hash_entry const *bucket = safe_hasher (table, entry);
264  struct hash_entry const *cursor;
265
266  if (bucket->data == NULL)
267    return NULL;
268
269  for (cursor = bucket; cursor; cursor = cursor->next)
270    if (entry == cursor->data || table->comparator (entry, cursor->data))
271      return cursor->data;
272
273  return NULL;
274}
275
276/* Walking.  */
277
278/* The functions in this page traverse the hash table and process the
279   contained entries.  For the traversal to work properly, the hash table
280   should not be resized nor modified while any particular entry is being
281   processed.  In particular, entries should not be added, and an entry
282   may be removed only if there is no shrink threshold and the entry being
283   removed has already been passed to hash_get_next.  */
284
285/* Return the first data in the table, or NULL if the table is empty.  */
286
287void *
288hash_get_first (const Hash_table *table)
289{
290  struct hash_entry const *bucket;
291
292  if (table->n_entries == 0)
293    return NULL;
294
295  for (bucket = table->bucket; ; bucket++)
296    if (! (bucket < table->bucket_limit))
297      abort ();
298    else if (bucket->data)
299      return bucket->data;
300}
301
302/* Return the user data for the entry following ENTRY, where ENTRY has been
303   returned by a previous call to either 'hash_get_first' or 'hash_get_next'.
304   Return NULL if there are no more entries.  */
305
306void *
307hash_get_next (const Hash_table *table, const void *entry)
308{
309  struct hash_entry const *bucket = safe_hasher (table, entry);
310  struct hash_entry const *cursor;
311
312  /* Find next entry in the same bucket.  */
313  cursor = bucket;
314  do
315    {
316      if (cursor->data == entry && cursor->next)
317        return cursor->next->data;
318      cursor = cursor->next;
319    }
320  while (cursor != NULL);
321
322  /* Find first entry in any subsequent bucket.  */
323  while (++bucket < table->bucket_limit)
324    if (bucket->data)
325      return bucket->data;
326
327  /* None found.  */
328  return NULL;
329}
330
331/* Fill BUFFER with pointers to active user entries in the hash table, then
332   return the number of pointers copied.  Do not copy more than BUFFER_SIZE
333   pointers.  */
334
335size_t
336hash_get_entries (const Hash_table *table, void **buffer,
337                  size_t buffer_size)
338{
339  size_t counter = 0;
340  struct hash_entry const *bucket;
341  struct hash_entry const *cursor;
342
343  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
344    {
345      if (bucket->data)
346        {
347          for (cursor = bucket; cursor; cursor = cursor->next)
348            {
349              if (counter >= buffer_size)
350                return counter;
351              buffer[counter++] = cursor->data;
352            }
353        }
354    }
355
356  return counter;
357}
358
359/* Call a PROCESSOR function for each entry of a hash table, and return the
360   number of entries for which the processor function returned success.  A
361   pointer to some PROCESSOR_DATA which will be made available to each call to
362   the processor function.  The PROCESSOR accepts two arguments: the first is
363   the user entry being walked into, the second is the value of PROCESSOR_DATA
364   as received.  The walking continue for as long as the PROCESSOR function
365   returns nonzero.  When it returns zero, the walking is interrupted.  */
366
367size_t
368hash_do_for_each (const Hash_table *table, Hash_processor processor,
369                  void *processor_data)
370{
371  size_t counter = 0;
372  struct hash_entry const *bucket;
373  struct hash_entry const *cursor;
374
375  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
376    {
377      if (bucket->data)
378        {
379          for (cursor = bucket; cursor; cursor = cursor->next)
380            {
381              if (! processor (cursor->data, processor_data))
382                return counter;
383              counter++;
384            }
385        }
386    }
387
388  return counter;
389}
390
391/* Allocation and clean-up.  */
392
393/* Return a hash index for a NUL-terminated STRING between 0 and N_BUCKETS-1.
394   This is a convenience routine for constructing other hashing functions.  */
395
396#if USE_DIFF_HASH
397
398/* About hashings, Paul Eggert writes to me (FP), on 1994-01-01: "Please see
399   B. J. McKenzie, R. Harries & T. Bell, Selecting a hashing algorithm,
400   Software--practice & experience 20, 2 (Feb 1990), 209-224.  Good hash
401   algorithms tend to be domain-specific, so what's good for [diffutils'] io.c
402   may not be good for your application."  */
403
404size_t
405hash_string (const char *string, size_t n_buckets)
406{
407# define HASH_ONE_CHAR(Value, Byte) \
408  ((Byte) + rotl_sz (Value, 7))
409
410  size_t value = 0;
411  unsigned char ch;
412
413  for (; (ch = *string); string++)
414    value = HASH_ONE_CHAR (value, ch);
415  return value % n_buckets;
416
417# undef HASH_ONE_CHAR
418}
419
420#else /* not USE_DIFF_HASH */
421
422/* This one comes from 'recode', and performs a bit better than the above as
423   per a few experiments.  It is inspired from a hashing routine found in the
424   very old Cyber 'snoop', itself written in typical Greg Mansfield style.
425   (By the way, what happened to this excellent man?  Is he still alive?)  */
426
427size_t
428hash_string (const char *string, size_t n_buckets)
429{
430  size_t value = 0;
431  unsigned char ch;
432
433  for (; (ch = *string); string++)
434    value = (value * 31 + ch) % n_buckets;
435  return value;
436}
437
438#endif /* not USE_DIFF_HASH */
439
440/* Return true if CANDIDATE is a prime number.  CANDIDATE should be an odd
441   number at least equal to 11.  */
442
443static bool _GL_ATTRIBUTE_CONST
444is_prime (size_t candidate)
445{
446  size_t divisor = 3;
447  size_t square = divisor * divisor;
448
449  while (square < candidate && (candidate % divisor))
450    {
451      divisor++;
452      square += 4 * divisor;
453      divisor++;
454    }
455
456  return (candidate % divisor ? true : false);
457}
458
459/* Round a given CANDIDATE number up to the nearest prime, and return that
460   prime.  Primes lower than 10 are merely skipped.  */
461
462static size_t _GL_ATTRIBUTE_CONST
463next_prime (size_t candidate)
464{
465  /* Skip small primes.  */
466  if (candidate < 10)
467    candidate = 10;
468
469  /* Make it definitely odd.  */
470  candidate |= 1;
471
472  while (SIZE_MAX != candidate && !is_prime (candidate))
473    candidate += 2;
474
475  return candidate;
476}
477
478void
479hash_reset_tuning (Hash_tuning *tuning)
480{
481  *tuning = default_tuning;
482}
483
484/* If the user passes a NULL hasher, we hash the raw pointer.  */
485static size_t
486raw_hasher (const void *data, size_t n)
487{
488  /* When hashing unique pointers, it is often the case that they were
489     generated by malloc and thus have the property that the low-order
490     bits are 0.  As this tends to give poorer performance with small
491     tables, we rotate the pointer value before performing division,
492     in an attempt to improve hash quality.  */
493  size_t val = rotr_sz ((size_t) data, 3);
494  return val % n;
495}
496
497/* If the user passes a NULL comparator, we use pointer comparison.  */
498static bool
499raw_comparator (const void *a, const void *b)
500{
501  return a == b;
502}
503
504
505/* For the given hash TABLE, check the user supplied tuning structure for
506   reasonable values, and return true if there is no gross error with it.
507   Otherwise, definitively reset the TUNING field to some acceptable default
508   in the hash table (that is, the user loses the right of further modifying
509   tuning arguments), and return false.  */
510
511static bool
512check_tuning (Hash_table *table)
513{
514  const Hash_tuning *tuning = table->tuning;
515  float epsilon;
516  if (tuning == &default_tuning)
517    return true;
518
519  /* Be a bit stricter than mathematics would require, so that
520     rounding errors in size calculations do not cause allocations to
521     fail to grow or shrink as they should.  The smallest allocation
522     is 11 (due to next_prime's algorithm), so an epsilon of 0.1
523     should be good enough.  */
524  epsilon = 0.1f;
525
526  if (epsilon < tuning->growth_threshold
527      && tuning->growth_threshold < 1 - epsilon
528      && 1 + epsilon < tuning->growth_factor
529      && 0 <= tuning->shrink_threshold
530      && tuning->shrink_threshold + epsilon < tuning->shrink_factor
531      && tuning->shrink_factor <= 1
532      && tuning->shrink_threshold + epsilon < tuning->growth_threshold)
533    return true;
534
535  table->tuning = &default_tuning;
536  return false;
537}
538
539/* Compute the size of the bucket array for the given CANDIDATE and
540   TUNING, or return 0 if there is no possible way to allocate that
541   many entries.  */
542
543static size_t _GL_ATTRIBUTE_PURE
544compute_bucket_size (size_t candidate, const Hash_tuning *tuning)
545{
546  if (!tuning->is_n_buckets)
547    {
548      float new_candidate = candidate / tuning->growth_threshold;
549      if (SIZE_MAX <= new_candidate)
550        return 0;
551      candidate = new_candidate;
552    }
553  candidate = next_prime (candidate);
554  if (xalloc_oversized (candidate, sizeof (struct hash_entry *)))
555    return 0;
556  return candidate;
557}
558
559/* Allocate and return a new hash table, or NULL upon failure.  The initial
560   number of buckets is automatically selected so as to _guarantee_ that you
561   may insert at least CANDIDATE different user entries before any growth of
562   the hash table size occurs.  So, if have a reasonably tight a-priori upper
563   bound on the number of entries you intend to insert in the hash table, you
564   may save some table memory and insertion time, by specifying it here.  If
565   the IS_N_BUCKETS field of the TUNING structure is true, the CANDIDATE
566   argument has its meaning changed to the wanted number of buckets.
567
568   TUNING points to a structure of user-supplied values, in case some fine
569   tuning is wanted over the default behavior of the hasher.  If TUNING is
570   NULL, the default tuning parameters are used instead.  If TUNING is
571   provided but the values requested are out of bounds or might cause
572   rounding errors, return NULL.
573
574   The user-supplied HASHER function, when not NULL, accepts two
575   arguments ENTRY and TABLE_SIZE.  It computes, by hashing ENTRY contents, a
576   slot number for that entry which should be in the range 0..TABLE_SIZE-1.
577   This slot number is then returned.
578
579   The user-supplied COMPARATOR function, when not NULL, accepts two
580   arguments pointing to user data, it then returns true for a pair of entries
581   that compare equal, or false otherwise.  This function is internally called
582   on entries which are already known to hash to the same bucket index,
583   but which are distinct pointers.
584
585   The user-supplied DATA_FREER function, when not NULL, may be later called
586   with the user data as an argument, just before the entry containing the
587   data gets freed.  This happens from within 'hash_free' or 'hash_clear'.
588   You should specify this function only if you want these functions to free
589   all of your 'data' data.  This is typically the case when your data is
590   simply an auxiliary struct that you have malloc'd to aggregate several
591   values.  */
592
593Hash_table *
594hash_initialize (size_t candidate, const Hash_tuning *tuning,
595                 Hash_hasher hasher, Hash_comparator comparator,
596                 Hash_data_freer data_freer)
597{
598  Hash_table *table;
599
600  if (hasher == NULL)
601    hasher = raw_hasher;
602  if (comparator == NULL)
603    comparator = raw_comparator;
604
605  table = malloc (sizeof *table);
606  if (table == NULL)
607    return NULL;
608
609  if (!tuning)
610    tuning = &default_tuning;
611  table->tuning = tuning;
612  if (!check_tuning (table))
613    {
614      /* Fail if the tuning options are invalid.  This is the only occasion
615         when the user gets some feedback about it.  Once the table is created,
616         if the user provides invalid tuning options, we silently revert to
617         using the defaults, and ignore further request to change the tuning
618         options.  */
619      goto fail;
620    }
621
622  table->n_buckets = compute_bucket_size (candidate, tuning);
623  if (!table->n_buckets)
624    goto fail;
625
626  table->bucket = calloc (table->n_buckets, sizeof *table->bucket);
627  if (table->bucket == NULL)
628    goto fail;
629  table->bucket_limit = table->bucket + table->n_buckets;
630  table->n_buckets_used = 0;
631  table->n_entries = 0;
632
633  table->hasher = hasher;
634  table->comparator = comparator;
635  table->data_freer = data_freer;
636
637  table->free_entry_list = NULL;
638#if USE_OBSTACK
639  obstack_init (&table->entry_stack);
640#endif
641  return table;
642
643 fail:
644  free (table);
645  return NULL;
646}
647
648/* Make all buckets empty, placing any chained entries on the free list.
649   Apply the user-specified function data_freer (if any) to the datas of any
650   affected entries.  */
651
652void
653hash_clear (Hash_table *table)
654{
655  struct hash_entry *bucket;
656
657  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
658    {
659      if (bucket->data)
660        {
661          struct hash_entry *cursor;
662          struct hash_entry *next;
663
664          /* Free the bucket overflow.  */
665          for (cursor = bucket->next; cursor; cursor = next)
666            {
667              if (table->data_freer)
668                table->data_freer (cursor->data);
669              cursor->data = NULL;
670
671              next = cursor->next;
672              /* Relinking is done one entry at a time, as it is to be expected
673                 that overflows are either rare or short.  */
674              cursor->next = table->free_entry_list;
675              table->free_entry_list = cursor;
676            }
677
678          /* Free the bucket head.  */
679          if (table->data_freer)
680            table->data_freer (bucket->data);
681          bucket->data = NULL;
682          bucket->next = NULL;
683        }
684    }
685
686  table->n_buckets_used = 0;
687  table->n_entries = 0;
688}
689
690/* Reclaim all storage associated with a hash table.  If a data_freer
691   function has been supplied by the user when the hash table was created,
692   this function applies it to the data of each entry before freeing that
693   entry.  */
694
695void
696hash_free (Hash_table *table)
697{
698  struct hash_entry *bucket;
699  struct hash_entry *cursor;
700  struct hash_entry *next;
701
702  /* Call the user data_freer function.  */
703  if (table->data_freer && table->n_entries)
704    {
705      for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
706        {
707          if (bucket->data)
708            {
709              for (cursor = bucket; cursor; cursor = cursor->next)
710                table->data_freer (cursor->data);
711            }
712        }
713    }
714
715#if USE_OBSTACK
716
717  obstack_free (&table->entry_stack, NULL);
718
719#else
720
721  /* Free all bucket overflowed entries.  */
722  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
723    {
724      for (cursor = bucket->next; cursor; cursor = next)
725        {
726          next = cursor->next;
727          free (cursor);
728        }
729    }
730
731  /* Also reclaim the internal list of previously freed entries.  */
732  for (cursor = table->free_entry_list; cursor; cursor = next)
733    {
734      next = cursor->next;
735      free (cursor);
736    }
737
738#endif
739
740  /* Free the remainder of the hash table structure.  */
741  free (table->bucket);
742  free (table);
743}
744
745/* Insertion and deletion.  */
746
747/* Get a new hash entry for a bucket overflow, possibly by recycling a
748   previously freed one.  If this is not possible, allocate a new one.  */
749
750static struct hash_entry *
751allocate_entry (Hash_table *table)
752{
753  struct hash_entry *new;
754
755  if (table->free_entry_list)
756    {
757      new = table->free_entry_list;
758      table->free_entry_list = new->next;
759    }
760  else
761    {
762#if USE_OBSTACK
763      new = obstack_alloc (&table->entry_stack, sizeof *new);
764#else
765      new = malloc (sizeof *new);
766#endif
767    }
768
769  return new;
770}
771
772/* Free a hash entry which was part of some bucket overflow,
773   saving it for later recycling.  */
774
775static void
776free_entry (Hash_table *table, struct hash_entry *entry)
777{
778  entry->data = NULL;
779  entry->next = table->free_entry_list;
780  table->free_entry_list = entry;
781}
782
783/* This private function is used to help with insertion and deletion.  When
784   ENTRY matches an entry in the table, return a pointer to the corresponding
785   user data and set *BUCKET_HEAD to the head of the selected bucket.
786   Otherwise, return NULL.  When DELETE is true and ENTRY matches an entry in
787   the table, unlink the matching entry.  */
788
789static void *
790hash_find_entry (Hash_table *table, const void *entry,
791                 struct hash_entry **bucket_head, bool delete)
792{
793  struct hash_entry *bucket = safe_hasher (table, entry);
794  struct hash_entry *cursor;
795
796  *bucket_head = bucket;
797
798  /* Test for empty bucket.  */
799  if (bucket->data == NULL)
800    return NULL;
801
802  /* See if the entry is the first in the bucket.  */
803  if (entry == bucket->data || table->comparator (entry, bucket->data))
804    {
805      void *data = bucket->data;
806
807      if (delete)
808        {
809          if (bucket->next)
810            {
811              struct hash_entry *next = bucket->next;
812
813              /* Bump the first overflow entry into the bucket head, then save
814                 the previous first overflow entry for later recycling.  */
815              *bucket = *next;
816              free_entry (table, next);
817            }
818          else
819            {
820              bucket->data = NULL;
821            }
822        }
823
824      return data;
825    }
826
827  /* Scan the bucket overflow.  */
828  for (cursor = bucket; cursor->next; cursor = cursor->next)
829    {
830      if (entry == cursor->next->data
831          || table->comparator (entry, cursor->next->data))
832        {
833          void *data = cursor->next->data;
834
835          if (delete)
836            {
837              struct hash_entry *next = cursor->next;
838
839              /* Unlink the entry to delete, then save the freed entry for later
840                 recycling.  */
841              cursor->next = next->next;
842              free_entry (table, next);
843            }
844
845          return data;
846        }
847    }
848
849  /* No entry found.  */
850  return NULL;
851}
852
853/* Internal helper, to move entries from SRC to DST.  Both tables must
854   share the same free entry list.  If SAFE, only move overflow
855   entries, saving bucket heads for later, so that no allocations will
856   occur.  Return false if the free entry list is exhausted and an
857   allocation fails.  */
858
859static bool
860transfer_entries (Hash_table *dst, Hash_table *src, bool safe)
861{
862  struct hash_entry *bucket;
863  struct hash_entry *cursor;
864  struct hash_entry *next;
865  for (bucket = src->bucket; bucket < src->bucket_limit; bucket++)
866    if (bucket->data)
867      {
868        void *data;
869        struct hash_entry *new_bucket;
870
871        /* Within each bucket, transfer overflow entries first and
872           then the bucket head, to minimize memory pressure.  After
873           all, the only time we might allocate is when moving the
874           bucket head, but moving overflow entries first may create
875           free entries that can be recycled by the time we finally
876           get to the bucket head.  */
877        for (cursor = bucket->next; cursor; cursor = next)
878          {
879            data = cursor->data;
880            new_bucket = safe_hasher (dst, data);
881
882            next = cursor->next;
883
884            if (new_bucket->data)
885              {
886                /* Merely relink an existing entry, when moving from a
887                   bucket overflow into a bucket overflow.  */
888                cursor->next = new_bucket->next;
889                new_bucket->next = cursor;
890              }
891            else
892              {
893                /* Free an existing entry, when moving from a bucket
894                   overflow into a bucket header.  */
895                new_bucket->data = data;
896                dst->n_buckets_used++;
897                free_entry (dst, cursor);
898              }
899          }
900        /* Now move the bucket head.  Be sure that if we fail due to
901           allocation failure that the src table is in a consistent
902           state.  */
903        data = bucket->data;
904        bucket->next = NULL;
905        if (safe)
906          continue;
907        new_bucket = safe_hasher (dst, data);
908
909        if (new_bucket->data)
910          {
911            /* Allocate or recycle an entry, when moving from a bucket
912               header into a bucket overflow.  */
913            struct hash_entry *new_entry = allocate_entry (dst);
914
915            if (new_entry == NULL)
916              return false;
917
918            new_entry->data = data;
919            new_entry->next = new_bucket->next;
920            new_bucket->next = new_entry;
921          }
922        else
923          {
924            /* Move from one bucket header to another.  */
925            new_bucket->data = data;
926            dst->n_buckets_used++;
927          }
928        bucket->data = NULL;
929        src->n_buckets_used--;
930      }
931  return true;
932}
933
934/* For an already existing hash table, change the number of buckets through
935   specifying CANDIDATE.  The contents of the hash table are preserved.  The
936   new number of buckets is automatically selected so as to _guarantee_ that
937   the table may receive at least CANDIDATE different user entries, including
938   those already in the table, before any other growth of the hash table size
939   occurs.  If TUNING->IS_N_BUCKETS is true, then CANDIDATE specifies the
940   exact number of buckets desired.  Return true iff the rehash succeeded.  */
941
942bool
943hash_rehash (Hash_table *table, size_t candidate)
944{
945  Hash_table storage;
946  Hash_table *new_table;
947  size_t new_size = compute_bucket_size (candidate, table->tuning);
948
949  if (!new_size)
950    return false;
951  if (new_size == table->n_buckets)
952    return true;
953  new_table = &storage;
954  new_table->bucket = calloc (new_size, sizeof *new_table->bucket);
955  if (new_table->bucket == NULL)
956    return false;
957  new_table->n_buckets = new_size;
958  new_table->bucket_limit = new_table->bucket + new_size;
959  new_table->n_buckets_used = 0;
960  new_table->n_entries = 0;
961  new_table->tuning = table->tuning;
962  new_table->hasher = table->hasher;
963  new_table->comparator = table->comparator;
964  new_table->data_freer = table->data_freer;
965
966  /* In order for the transfer to successfully complete, we need
967     additional overflow entries when distinct buckets in the old
968     table collide into a common bucket in the new table.  The worst
969     case possible is a hasher that gives a good spread with the old
970     size, but returns a constant with the new size; if we were to
971     guarantee table->n_buckets_used-1 free entries in advance, then
972     the transfer would be guaranteed to not allocate memory.
973     However, for large tables, a guarantee of no further allocation
974     introduces a lot of extra memory pressure, all for an unlikely
975     corner case (most rehashes reduce, rather than increase, the
976     number of overflow entries needed).  So, we instead ensure that
977     the transfer process can be reversed if we hit a memory
978     allocation failure mid-transfer.  */
979
980  /* Merely reuse the extra old space into the new table.  */
981#if USE_OBSTACK
982  new_table->entry_stack = table->entry_stack;
983#endif
984  new_table->free_entry_list = table->free_entry_list;
985
986  if (transfer_entries (new_table, table, false))
987    {
988      /* Entries transferred successfully; tie up the loose ends.  */
989      free (table->bucket);
990      table->bucket = new_table->bucket;
991      table->bucket_limit = new_table->bucket_limit;
992      table->n_buckets = new_table->n_buckets;
993      table->n_buckets_used = new_table->n_buckets_used;
994      table->free_entry_list = new_table->free_entry_list;
995      /* table->n_entries and table->entry_stack already hold their value.  */
996      return true;
997    }
998
999  /* We've allocated new_table->bucket (and possibly some entries),
1000     exhausted the free list, and moved some but not all entries into
1001     new_table.  We must undo the partial move before returning
1002     failure.  The only way to get into this situation is if new_table
1003     uses fewer buckets than the old table, so we will reclaim some
1004     free entries as overflows in the new table are put back into
1005     distinct buckets in the old table.
1006
1007     There are some pathological cases where a single pass through the
1008     table requires more intermediate overflow entries than using two
1009     passes.  Two passes give worse cache performance and takes
1010     longer, but at this point, we're already out of memory, so slow
1011     and safe is better than failure.  */
1012  table->free_entry_list = new_table->free_entry_list;
1013  if (! (transfer_entries (table, new_table, true)
1014         && transfer_entries (table, new_table, false)))
1015    abort ();
1016  /* table->n_entries already holds its value.  */
1017  free (new_table->bucket);
1018  return false;
1019}
1020
1021/* Insert ENTRY into hash TABLE if there is not already a matching entry.
1022
1023   Return -1 upon memory allocation failure.
1024   Return 1 if insertion succeeded.
1025   Return 0 if there is already a matching entry in the table,
1026   and in that case, if MATCHED_ENT is non-NULL, set *MATCHED_ENT
1027   to that entry.
1028
1029   This interface is easier to use than hash_insert when you must
1030   distinguish between the latter two cases.  More importantly,
1031   hash_insert is unusable for some types of ENTRY values.  When using
1032   hash_insert, the only way to distinguish those cases is to compare
1033   the return value and ENTRY.  That works only when you can have two
1034   different ENTRY values that point to data that compares "equal".  Thus,
1035   when the ENTRY value is a simple scalar, you must use
1036   hash_insert_if_absent.  ENTRY must not be NULL.  */
1037int
1038hash_insert_if_absent (Hash_table *table, void const *entry,
1039                       void const **matched_ent)
1040{
1041  void *data;
1042  struct hash_entry *bucket;
1043
1044  /* The caller cannot insert a NULL entry, since hash_lookup returns NULL
1045     to indicate "not found", and hash_find_entry uses "bucket->data == NULL"
1046     to indicate an empty bucket.  */
1047  if (! entry)
1048    abort ();
1049
1050  /* If there's a matching entry already in the table, return that.  */
1051  if ((data = hash_find_entry (table, entry, &bucket, false)) != NULL)
1052    {
1053      if (matched_ent)
1054        *matched_ent = data;
1055      return 0;
1056    }
1057
1058  /* If the growth threshold of the buckets in use has been reached, increase
1059     the table size and rehash.  There's no point in checking the number of
1060     entries:  if the hashing function is ill-conditioned, rehashing is not
1061     likely to improve it.  */
1062
1063  if (table->n_buckets_used
1064      > table->tuning->growth_threshold * table->n_buckets)
1065    {
1066      /* Check more fully, before starting real work.  If tuning arguments
1067         became invalid, the second check will rely on proper defaults.  */
1068      check_tuning (table);
1069      if (table->n_buckets_used
1070          > table->tuning->growth_threshold * table->n_buckets)
1071        {
1072          const Hash_tuning *tuning = table->tuning;
1073          float candidate =
1074            (tuning->is_n_buckets
1075             ? (table->n_buckets * tuning->growth_factor)
1076             : (table->n_buckets * tuning->growth_factor
1077                * tuning->growth_threshold));
1078
1079          if (SIZE_MAX <= candidate)
1080            return -1;
1081
1082          /* If the rehash fails, arrange to return NULL.  */
1083          if (!hash_rehash (table, candidate))
1084            return -1;
1085
1086          /* Update the bucket we are interested in.  */
1087          if (hash_find_entry (table, entry, &bucket, false) != NULL)
1088            abort ();
1089        }
1090    }
1091
1092  /* ENTRY is not matched, it should be inserted.  */
1093
1094  if (bucket->data)
1095    {
1096      struct hash_entry *new_entry = allocate_entry (table);
1097
1098      if (new_entry == NULL)
1099        return -1;
1100
1101      /* Add ENTRY in the overflow of the bucket.  */
1102
1103      new_entry->data = (void *) entry;
1104      new_entry->next = bucket->next;
1105      bucket->next = new_entry;
1106      table->n_entries++;
1107      return 1;
1108    }
1109
1110  /* Add ENTRY right in the bucket head.  */
1111
1112  bucket->data = (void *) entry;
1113  table->n_entries++;
1114  table->n_buckets_used++;
1115
1116  return 1;
1117}
1118
1119/* hash_insert0 is the deprecated name for hash_insert_if_absent.
1120   .  */
1121int
1122hash_insert0 (Hash_table *table, void const *entry, void const **matched_ent)
1123{
1124  return hash_insert_if_absent (table, entry, matched_ent);
1125}
1126
1127/* If ENTRY matches an entry already in the hash table, return the pointer
1128   to the entry from the table.  Otherwise, insert ENTRY and return ENTRY.
1129   Return NULL if the storage required for insertion cannot be allocated.
1130   This implementation does not support duplicate entries or insertion of
1131   NULL.  */
1132
1133void *
1134hash_insert (Hash_table *table, void const *entry)
1135{
1136  void const *matched_ent;
1137  int err = hash_insert_if_absent (table, entry, &matched_ent);
1138  return (err == -1
1139          ? NULL
1140          : (void *) (err == 0 ? matched_ent : entry));
1141}
1142
1143/* If ENTRY is already in the table, remove it and return the just-deleted
1144   data (the user may want to deallocate its storage).  If ENTRY is not in the
1145   table, don't modify the table and return NULL.  */
1146
1147void *
1148hash_delete (Hash_table *table, const void *entry)
1149{
1150  void *data;
1151  struct hash_entry *bucket;
1152
1153  data = hash_find_entry (table, entry, &bucket, true);
1154  if (!data)
1155    return NULL;
1156
1157  table->n_entries--;
1158  if (!bucket->data)
1159    {
1160      table->n_buckets_used--;
1161
1162      /* If the shrink threshold of the buckets in use has been reached,
1163         rehash into a smaller table.  */
1164
1165      if (table->n_buckets_used
1166          < table->tuning->shrink_threshold * table->n_buckets)
1167        {
1168          /* Check more fully, before starting real work.  If tuning arguments
1169             became invalid, the second check will rely on proper defaults.  */
1170          check_tuning (table);
1171          if (table->n_buckets_used
1172              < table->tuning->shrink_threshold * table->n_buckets)
1173            {
1174              const Hash_tuning *tuning = table->tuning;
1175              size_t candidate =
1176                (tuning->is_n_buckets
1177                 ? table->n_buckets * tuning->shrink_factor
1178                 : (table->n_buckets * tuning->shrink_factor
1179                    * tuning->growth_threshold));
1180
1181              if (!hash_rehash (table, candidate))
1182                {
1183                  /* Failure to allocate memory in an attempt to
1184                     shrink the table is not fatal.  But since memory
1185                     is low, we can at least be kind and free any
1186                     spare entries, rather than keeping them tied up
1187                     in the free entry list.  */
1188#if ! USE_OBSTACK
1189                  struct hash_entry *cursor = table->free_entry_list;
1190                  struct hash_entry *next;
1191                  while (cursor)
1192                    {
1193                      next = cursor->next;
1194                      free (cursor);
1195                      cursor = next;
1196                    }
1197                  table->free_entry_list = NULL;
1198#endif
1199                }
1200            }
1201        }
1202    }
1203
1204  return data;
1205}
1206
1207/* Testing.  */
1208
1209#if TESTING
1210
1211void
1212hash_print (const Hash_table *table)
1213{
1214  struct hash_entry *bucket = (struct hash_entry *) table->bucket;
1215
1216  for ( ; bucket < table->bucket_limit; bucket++)
1217    {
1218      struct hash_entry *cursor;
1219
1220      if (bucket)
1221        printf ("%lu:\n", (unsigned long int) (bucket - table->bucket));
1222
1223      for (cursor = bucket; cursor; cursor = cursor->next)
1224        {
1225          char const *s = cursor->data;
1226          /* FIXME */
1227          if (s)
1228            printf ("  %s\n", s);
1229        }
1230    }
1231}
1232
1233#endif /* TESTING */
1234