summaryrefslogtreecommitdiffstats
path: root/lib/hash.c
blob: 2b123be75433cf01c294cbc663987646526bbe76 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
/* hash - hashing table processing.

   Copyright (C) 1998-2004, 2006-2007, 2009-2024 Free Software Foundation, Inc.

   Written by Jim Meyering, 1992.

   This file is free software: you can redistribute it and/or modify
   it under the terms of the GNU Lesser General Public License as
   published by the Free Software Foundation; either version 2.1 of the
   License, or (at your option) any later version.

   This file is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public License
   along with this program.  If not, see <https://www.gnu.org/licenses/>.  */

/* A generic hash table package.  */

/* Define USE_OBSTACK to 1 if you want the allocator to use obstacks instead
   of malloc.  If you change USE_OBSTACK, you have to recompile!  */

#include <config.h>

#include "hash.h"

#include "bitrotate.h"
#include "xalloc-oversized.h"

#include <errno.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

#if USE_OBSTACK
# include "obstack.h"
# ifndef obstack_chunk_alloc
#  define obstack_chunk_alloc malloc
# endif
# ifndef obstack_chunk_free
#  define obstack_chunk_free free
# endif
#endif

struct hash_entry
  {
    void *data;
    struct hash_entry *next;
  };

struct hash_table
  {
    /* The array of buckets starts at BUCKET and extends to BUCKET_LIMIT-1,
       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 const *bucket_limit;
    size_t n_buckets;
    size_t n_buckets_used;
    size_t n_entries;

    /* Tuning arguments, kept in a physically separate structure.  */
    const Hash_tuning *tuning;

    /* Three functions are given to 'hash_initialize', see the documentation
       block for this function.  In a word, HASHER randomizes a user entry
       into a number up from 0 up to some maximum minus 1; COMPARATOR returns
       true if two user entries compare equally; and DATA_FREER is the cleanup
       function for a user entry.  */
    Hash_hasher hasher;
    Hash_comparator comparator;
    Hash_data_freer data_freer;

    /* A linked list of freed struct hash_entry structs.  */
    struct hash_entry *free_entry_list;

#if USE_OBSTACK
    /* Whenever obstacks are used, it is possible to allocate all overflowed
       entries into a single stack, so they all can be freed in a single
       operation.  It is not clear if the speedup is worth the trouble.  */
    struct obstack entry_stack;
#endif
  };

/* A hash table contains many internal entries, each holding a pointer to
   some user-provided data (also called a user entry).  An entry indistinctly
   refers to both the internal entry and its associated user entry.  A user
   entry contents may be hashed by a randomization function (the hashing
   function, or just "hasher" for short) into a number (or "slot") between 0
   and the current table size.  At each slot position in the hash table,
   starts a linked chain of entries for which the user data all hash to this
   slot.  A bucket is the collection of all entries hashing to the same slot.

   A good "hasher" function will distribute entries rather evenly in buckets.
   In the ideal case, the length of each bucket is roughly the number of
   entries divided by the table size.  Finding the slot for a data is usually
   done in constant time by the "hasher", and the later finding of a precise
   entry is linear in time with the size of the bucket.  Consequently, a
   larger hash table size (that is, a larger number of buckets) is prone to
   yielding shorter chains, *given* the "hasher" function behaves properly.

   Long buckets slow down the lookup algorithm.  One might use big hash table
   sizes in hope to reduce the average length of buckets, but this might
   become inordinate, as unused slots in the hash table take some space.  The
   best bet is to make sure you are using a good "hasher" function (beware
   that those are not that easy to write! :-), and to use a table size
   larger than the actual number of entries.  */

/* If an insertion makes the ratio of nonempty buckets to table size larger
   than the growth threshold (a number between 0.0 and 1.0), then increase
   the table size by multiplying by the growth factor (a number greater than
   1.0).  The growth threshold defaults to 0.8, and the growth factor
   defaults to 1.414, meaning that the table will have doubled its size
   every second time 80% of the buckets get used.  */
#define DEFAULT_GROWTH_THRESHOLD 0.8f
#define DEFAULT_GROWTH_FACTOR 1.414f

/* If a deletion empties a bucket and causes the ratio of used buckets to
   table size to become smaller than the shrink threshold (a number between
   0.0 and 1.0), then shrink the table by multiplying by the shrink factor (a
   number greater than the shrink threshold but smaller than 1.0).  The shrink
   threshold and factor default to 0.0 and 1.0, meaning that the table never
   shrinks.  */
#define DEFAULT_SHRINK_THRESHOLD 0.0f
#define DEFAULT_SHRINK_FACTOR 1.0f

/* Use this to initialize or reset a TUNING structure to
   some sensible values. */
static const Hash_tuning default_tuning =
  {
    DEFAULT_SHRINK_THRESHOLD,
    DEFAULT_SHRINK_FACTOR,
    DEFAULT_GROWTH_THRESHOLD,
    DEFAULT_GROWTH_FACTOR,
    false
  };

/* Information and lookup.  */

size_t
hash_get_n_buckets (const Hash_table *table)
{
  return table->n_buckets;
}

size_t
hash_get_n_buckets_used (const Hash_table *table)
{
  return table->n_buckets_used;
}

size_t
hash_get_n_entries (const Hash_table *table)
{
  return table->n_entries;
}

size_t
hash_get_max_bucket_length (const Hash_table *table)
{
  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 const *cursor = bucket;
          size_t bucket_length = 1;

          while (cursor = cursor->next, cursor)
            bucket_length++;

          if (bucket_length > max_bucket_length)
            max_bucket_length = bucket_length;
        }
    }

  return max_bucket_length;
}

bool
hash_table_ok (const Hash_table *table)
{
  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 const *cursor = bucket;

          /* Count bucket head.  */
          n_buckets_used++;
          n_entries++;

          /* Count bucket overflow.  */
          while (cursor = cursor->next, cursor)
            n_entries++;
        }
    }

  if (n_buckets_used == table->n_buckets_used && n_entries == table->n_entries)
    return true;

  return false;
}

void
hash_print_statistics (const Hash_table *table, FILE *stream)
{
  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: %lu\n",
           (unsigned long int) max_bucket_length);
}

/* Hash KEY and return a pointer to the selected bucket.
   If TABLE->hasher misbehaves, abort.  */
static struct hash_entry *
safe_hasher (const Hash_table *table, const void *key)
{
  size_t n = table->hasher (key, table->n_buckets);
  if (! (n < table->n_buckets))
    abort ();
  return table->bucket + n;
}

void *
hash_lookup (const Hash_table *table, const void *entry)
{
  struct hash_entry const *bucket = safe_hasher (table, entry);
  struct hash_entry const *cursor;

  if (bucket->data == NULL)
    return NULL;

  for (cursor = bucket; cursor; cursor = cursor->next)
    if (entry == cursor->data || table->comparator (entry, cursor->data))
      return cursor->data;

  return NULL;
}

/* Walking.  */

void *
hash_get_first (const Hash_table *table)
{
  struct hash_entry const *bucket;

  if (table->n_entries == 0)
    return NULL;

  for (bucket = table->bucket; ; bucket++)
    if (! (bucket < table->bucket_limit))
      abort ();
    else if (bucket->data)
      return bucket->data;
}

void *
hash_get_next (const Hash_table *table, const void *entry)
{
  struct hash_entry const *bucket = safe_hasher (table, entry);
  struct hash_entry const *cursor;

  /* Find next entry in the same bucket.  */
  cursor = bucket;
  do
    {
      if (cursor->data == entry && cursor->next)
        return cursor->next->data;
      cursor = cursor->next;
    }
  while (cursor != NULL);

  /* Find first entry in any subsequent bucket.  */
  while (++bucket < table->bucket_limit)
    if (bucket->data)
      return bucket->data;

  /* None found.  */
  return NULL;
}

size_t
hash_get_entries (const Hash_table *table, void **buffer,
                  size_t buffer_size)
{
  size_t counter = 0;
  struct hash_entry const *bucket;
  struct hash_entry const *cursor;

  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
    {
      if (bucket->data)
        {
          for (cursor = bucket; cursor; cursor = cursor->next)
            {
              if (counter >= buffer_size)
                return counter;
              buffer[counter++] = cursor->data;
            }
        }
    }

  return counter;
}

size_t
hash_do_for_each (const Hash_table *table, Hash_processor processor,
                  void *processor_data)
{
  size_t counter = 0;
  struct hash_entry const *bucket;
  struct hash_entry const *cursor;

  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
    {
      if (bucket->data)
        {
          for (cursor = bucket; cursor; cursor = cursor->next)
            {
              if (! processor (cursor->data, processor_data))
                return counter;
              counter++;
            }
        }
    }

  return counter;
}

/* Allocation and clean-up.  */

#if USE_DIFF_HASH

/* About hashings, Paul Eggert writes to me (FP), on 1994-01-01: "Please see
   B. J. McKenzie, R. Harries & T. Bell, Selecting a hashing algorithm,
   Software--practice & experience 20, 2 (Feb 1990), 209-224.  Good hash
   algorithms tend to be domain-specific, so what's good for [diffutils'] io.c
   may not be good for your application."  */

size_t
hash_string (const char *string, size_t n_buckets)
{
# define HASH_ONE_CHAR(Value, Byte) \
  ((Byte) + rotl_sz (Value, 7))

  size_t value = 0;
  unsigned char ch;

  for (; (ch = *string); string++)
    value = HASH_ONE_CHAR (value, ch);
  return value % n_buckets;

# undef HASH_ONE_CHAR
}

#else /* not USE_DIFF_HASH */

/* This one comes from 'recode', and performs a bit better than the above as
   per a few experiments.  It is inspired from a hashing routine found in the
   very old Cyber 'snoop', itself written in typical Greg Mansfield style.
   (By the way, what happened to this excellent man?  Is he still alive?)  */

size_t
hash_string (const char *string, size_t n_buckets)
{
  size_t value = 0;
  unsigned char ch;

  for (; (ch = *string); string++)
    value = (value * 31 + ch) % n_buckets;
  return value;
}

#endif /* not USE_DIFF_HASH */

/* Return true if CANDIDATE is a prime number.  CANDIDATE should be an odd
   number at least equal to 11.  */

static bool _GL_ATTRIBUTE_CONST
is_prime (size_t candidate)
{
  size_t divisor = 3;
  size_t square = divisor * divisor;

  while (square < candidate && (candidate % divisor))
    {
      divisor++;
      square += 4 * divisor;
      divisor++;
    }

  return (candidate % divisor ? true : false);
}

/* Round a given CANDIDATE number up to the nearest prime, and return that
   prime.  Primes lower than 10 are merely skipped.  */

static size_t _GL_ATTRIBUTE_CONST
next_prime (size_t candidate)
{
  /* Skip small primes.  */
  if (candidate < 10)
    candidate = 10;

  /* Make it definitely odd.  */
  candidate |= 1;

  while (SIZE_MAX != candidate && !is_prime (candidate))
    candidate += 2;

  return candidate;
}

void
hash_reset_tuning (Hash_tuning *tuning)
{
  *tuning = default_tuning;
}

/* If the user passes a NULL hasher, we hash the raw pointer.  */
static size_t
raw_hasher (const void *data, size_t n)
{
  /* When hashing unique pointers, it is often the case that they were
     generated by malloc and thus have the property that the low-order
     bits are 0.  As this tends to give poorer performance with small
     tables, we rotate the pointer value before performing division,
     in an attempt to improve hash quality.  */
  size_t val = rotr_sz ((size_t) data, 3);
  return val % n;
}

/* If the user passes a NULL comparator, we use pointer comparison.  */
static bool
raw_comparator (const void *a, const void *b)
{
  return a == b;
}


/* For the given hash TABLE, check the user supplied tuning structure for
   reasonable values, and return true if there is no gross error with it.
   Otherwise, definitively reset the TUNING field to some acceptable default
   in the hash table (that is, the user loses the right of further modifying
   tuning arguments), and return false.  */

static bool
check_tuning (Hash_table *table)
{
  const Hash_tuning *tuning = table->tuning;
  float epsilon;
  if (tuning == &default_tuning)
    return true;

  /* 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.  */
  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;
  return false;
}

/* Compute the size of the bucket array for the given CANDIDATE and
   TUNING, or return 0 if there is no possible way to allocate that
   many entries.  */

static size_t _GL_ATTRIBUTE_PURE
compute_bucket_size (size_t candidate, const Hash_tuning *tuning)
{
  if (!tuning->is_n_buckets)
    {
      float new_candidate = candidate / tuning->growth_threshold;
      if ((float) SIZE_MAX <= new_candidate)
        goto nomem;
      candidate = new_candidate;
    }
  candidate = next_prime (candidate);
  if (xalloc_oversized (candidate, sizeof (struct hash_entry *)))
    goto nomem;
  return candidate;

 nomem:
  errno = ENOMEM;
  return 0;
}

Hash_table *
hash_initialize (size_t candidate, const Hash_tuning *tuning,
                 Hash_hasher hasher, Hash_comparator comparator,
                 Hash_data_freer data_freer)
{
  Hash_table *table;

  if (hasher == NULL)
    hasher = raw_hasher;
  if (comparator == NULL)
    comparator = raw_comparator;

  table = malloc (sizeof *table);
  if (table == NULL)
    return NULL;

  if (!tuning)
    tuning = &default_tuning;
  table->tuning = tuning;
  if (!check_tuning (table))
    {
      /* Fail if the tuning options are invalid.  This is the only occasion
         when the user gets some feedback about it.  Once the table is created,
         if the user provides invalid tuning options, we silently revert to
         using the defaults, and ignore further request to change the tuning
         options.  */
      errno = EINVAL;
      goto fail;
    }

  table->n_buckets = compute_bucket_size (candidate, tuning);
  if (!table->n_buckets)
    goto fail;

  table->bucket = calloc (table->n_buckets, sizeof *table->bucket);
  if (table->bucket == NULL)
    goto fail;
  table->bucket_limit = table->bucket + table->n_buckets;
  table->n_buckets_used = 0;
  table->n_entries = 0;

  table->hasher = hasher;
  table->comparator = comparator;
  table->data_freer = data_freer;

  table->free_entry_list = NULL;
#if USE_OBSTACK
  obstack_init (&table->entry_stack);
#endif
  return table;

 fail:
  free (table);
  return NULL;
}

void
hash_clear (Hash_table *table)
{
  struct hash_entry *bucket;

  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
    {
      if (bucket->data)
        {
          struct hash_entry *cursor;
          struct hash_entry *next;

          /* Free the bucket overflow.  */
          for (cursor = bucket->next; cursor; cursor = next)
            {
              if (table->data_freer)
                table->data_freer (cursor->data);
              cursor->data = NULL;

              next = cursor->next;
              /* Relinking is done one entry at a time, as it is to be expected
                 that overflows are either rare or short.  */
              cursor->next = table->free_entry_list;
              table->free_entry_list = cursor;
            }

          /* Free the bucket head.  */
          if (table->data_freer)
            table->data_freer (bucket->data);
          bucket->data = NULL;
          bucket->next = NULL;
        }
    }

  table->n_buckets_used = 0;
  table->n_entries = 0;
}

void
hash_free (Hash_table *table)
{
  struct hash_entry *bucket;
  struct hash_entry *cursor;
  struct hash_entry *next;
  int err = errno;

  /* Call the user data_freer function.  */
  if (table->data_freer && table->n_entries)
    {
      for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
        {
          if (bucket->data)
            {
              for (cursor = bucket; cursor; cursor = cursor->next)
                table->data_freer (cursor->data);
            }
        }
    }

#if USE_OBSTACK

  obstack_free (&table->entry_stack, NULL);

#else

  /* Free all bucket overflowed entries.  */
  for (bucket = table->bucket; bucket < table->bucket_limit; bucket++)
    {
      for (cursor = bucket->next; cursor; cursor = next)
        {
          next = cursor->next;
          free (cursor);
        }
    }

  /* Also reclaim the internal list of previously freed entries.  */
  for (cursor = table->free_entry_list; cursor; cursor = next)
    {
      next = cursor->next;
      free (cursor);
    }

#endif

  /* Free the remainder of the hash table structure.  */
  free (table->bucket);
  free (table);

  errno = err;
}

/* Insertion and deletion.  */

/* Get a new hash entry for a bucket overflow, possibly by recycling a
   previously freed one.  If this is not possible, allocate a new one.  */

static struct hash_entry *
allocate_entry (Hash_table *table)
{
  struct hash_entry *new;

  if (table->free_entry_list)
    {
      new = table->free_entry_list;
      table->free_entry_list = new->next;
    }
  else
    {
#if USE_OBSTACK
      new = obstack_alloc (&table->entry_stack, sizeof *new);
#else
      new = malloc (sizeof *new);
#endif
    }

  return new;
}

/* Free a hash entry which was part of some bucket overflow,
   saving it for later recycling.  */

static void
free_entry (Hash_table *table, struct hash_entry *entry)
{
  entry->data = NULL;
  entry->next = table->free_entry_list;
  table->free_entry_list = entry;
}

/* This private function is used to help with insertion and deletion.  When
   ENTRY matches an entry in the table, return a pointer to the corresponding
   user data and set *BUCKET_HEAD to the head of the selected bucket.
   Otherwise, return NULL.  When DELETE is true and ENTRY matches an entry in
   the table, unlink the matching entry.  */

static void *
hash_find_entry (Hash_table *table, const void *entry,
                 struct hash_entry **bucket_head, bool delete)
{
  struct hash_entry *bucket = safe_hasher (table, entry);
  struct hash_entry *cursor;

  *bucket_head = bucket;

  /* Test for empty bucket.  */
  if (bucket->data == NULL)
    return NULL;

  /* See if the entry is the first in the bucket.  */
  if (entry == bucket->data || table->comparator (entry, bucket->data))
    {
      void *data = bucket->data;

      if (delete)
        {
          if (bucket->next)
            {
              struct hash_entry *next = bucket->next;

              /* Bump the first overflow entry into the bucket head, then save
                 the previous first overflow entry for later recycling.  */
              *bucket = *next;
              free_entry (table, next);
            }
          else
            {
              bucket->data = NULL;
            }
        }

      return data;
    }

  /* Scan the bucket overflow.  */
  for (cursor = bucket; cursor->next; cursor = cursor->next)
    {
      if (entry == cursor->next->data
          || table->comparator (entry, cursor->next->data))
        {
          void *data = cursor->next->data;

          if (delete)
            {
              struct hash_entry *next = cursor->next;

              /* Unlink the entry to delete, then save the freed entry for later
                 recycling.  */
              cursor->next = next->next;
              free_entry (table, next);
            }

          return data;
        }
    }

  /* No entry found.  */
  return NULL;
}

/* Internal helper, to move entries from SRC to DST.  Both tables must
   share the same free entry list.  If SAFE, only move overflow
   entries, saving bucket heads for later, so that no allocations will
   occur.  Return false (setting errno) if the free entry list is
   exhausted and an allocation fails.  */

static bool
transfer_entries (Hash_table *dst, Hash_table *src, bool safe)
{
  struct hash_entry *bucket;
  struct hash_entry *cursor;
  struct hash_entry *next;
  for (bucket = src->bucket; bucket < src->bucket_limit; bucket++)
    if (bucket->data)
      {
        void *data;
        struct hash_entry *new_bucket;

        /* Within each bucket, transfer overflow entries first and
           then the bucket head, to minimize memory pressure.  After
           all, the only time we might allocate is when moving the
           bucket head, but moving overflow entries first may create
           free entries that can be recycled by the time we finally
           get to the bucket head.  */
        for (cursor = bucket->next; cursor; cursor = next)
          {
            data = cursor->data;
            new_bucket = safe_hasher (dst, data);

            next = cursor->next;

            if (new_bucket->data)
              {
                /* Merely relink an existing entry, when moving from a
                   bucket overflow into a bucket overflow.  */
                cursor->next = new_bucket->next;
                new_bucket->next = cursor;
              }
            else
              {
                /* Free an existing entry, when moving from a bucket
                   overflow into a bucket header.  */
                new_bucket->data = data;
                dst->n_buckets_used++;
                free_entry (dst, cursor);
              }
          }
        /* Now move the bucket head.  Be sure that if we fail due to
           allocation failure that the src table is in a consistent
           state.  */
        data = bucket->data;
        bucket->next = NULL;
        if (safe)
          continue;
        new_bucket = safe_hasher (dst, data);

        if (new_bucket->data)
          {
            /* Allocate or recycle an entry, when moving from a bucket
               header into a bucket overflow.  */
            struct hash_entry *new_entry = allocate_entry (dst);

            if (new_entry == NULL)
              return false;

            new_entry->data = data;
            new_entry->next = new_bucket->next;
            new_bucket->next = new_entry;
          }
        else
          {
            /* Move from one bucket header to another.  */
            new_bucket->data = data;
            dst->n_buckets_used++;
          }
        bucket->data = NULL;
        src->n_buckets_used--;
      }
  return true;
}

bool
hash_rehash (Hash_table *table, size_t candidate)
{
  Hash_table storage;
  Hash_table *new_table;
  size_t new_size = compute_bucket_size (candidate, table->tuning);

  if (!new_size)
    return false;
  if (new_size == table->n_buckets)
    return true;
  new_table = &storage;
  new_table->bucket = calloc (new_size, sizeof *new_table->bucket);
  if (new_table->bucket == NULL)
    return false;
  new_table->n_buckets = new_size;
  new_table->bucket_limit = new_table->bucket + new_size;
  new_table->n_buckets_used = 0;
  new_table->n_entries = 0;
  new_table->tuning = table->tuning;
  new_table->hasher = table->hasher;
  new_table->comparator = table->comparator;
  new_table->data_freer = table->data_freer;

  /* In order for the transfer to successfully complete, we need
     additional overflow entries when distinct buckets in the old
     table collide into a common bucket in the new table.  The worst
     case possible is a hasher that gives a good spread with the old
     size, but returns a constant with the new size; if we were to
     guarantee table->n_buckets_used-1 free entries in advance, then
     the transfer would be guaranteed to not allocate memory.
     However, for large tables, a guarantee of no further allocation
     introduces a lot of extra memory pressure, all for an unlikely
     corner case (most rehashes reduce, rather than increase, the
     number of overflow entries needed).  So, we instead ensure that
     the transfer process can be reversed if we hit a memory
     allocation failure mid-transfer.  */

  /* Merely reuse the extra old space into the new table.  */
#if USE_OBSTACK
  new_table->entry_stack = table->entry_stack;
#endif
  new_table->free_entry_list = table->free_entry_list;

  if (transfer_entries (new_table, table, false))
    {
      /* Entries transferred successfully; tie up the loose ends.  */
      free (table->bucket);
      table->bucket = new_table->bucket;
      table->bucket_limit = new_table->bucket_limit;
      table->n_buckets = new_table->n_buckets;
      table->n_buckets_used = new_table->n_buckets_used;
      table->free_entry_list = new_table->free_entry_list;
      /* table->n_entries and table->entry_stack already hold their value.  */
      return true;
    }

  /* We've allocated new_table->bucket (and possibly some entries),
     exhausted the free list, and moved some but not all entries into
     new_table.  We must undo the partial move before returning
     failure.  The only way to get into this situation is if new_table
     uses fewer buckets than the old table, so we will reclaim some
     free entries as overflows in the new table are put back into
     distinct buckets in the old table.

     There are some pathological cases where a single pass through the
     table requires more intermediate overflow entries than using two
     passes.  Two passes give worse cache performance and takes
     longer, but at this point, we're already out of memory, so slow
     and safe is better than failure.  */
  int err = errno;
  table->free_entry_list = new_table->free_entry_list;
  if (! (transfer_entries (table, new_table, true)
         && transfer_entries (table, new_table, false)))
    abort ();
  /* table->n_entries already holds its value.  */
  free (new_table->bucket);
  errno = err;
  return false;
}

int
hash_insert_if_absent (Hash_table *table, void const *entry,
                       void const **matched_ent)
{
  void *data;
  struct hash_entry *bucket;

  /* The caller cannot insert a NULL entry, since hash_lookup returns NULL
     to indicate "not found", and hash_find_entry uses "bucket->data == NULL"
     to indicate an empty bucket.  */
  if (! entry)
    abort ();

  /* If there's a matching entry already in the table, return that.  */
  if ((data = hash_find_entry (table, entry, &bucket, false)) != NULL)
    {
      if (matched_ent)
        *matched_ent = data;
      return 0;
    }

  /* If the growth threshold of the buckets in use has been reached, increase
     the table size and rehash.  There's no point in checking the number of
     entries:  if the hashing function is ill-conditioned, rehashing is not
     likely to improve it.  */

  if (table->n_buckets_used
      > table->tuning->growth_threshold * table->n_buckets)
    {
      /* Check more fully, before starting real work.  If tuning arguments
         became invalid, the second check will rely on proper defaults.  */
      check_tuning (table);
      if (table->n_buckets_used
          > table->tuning->growth_threshold * table->n_buckets)
        {
          const Hash_tuning *tuning = table->tuning;
          float candidate =
            (tuning->is_n_buckets
             ? (table->n_buckets * tuning->growth_factor)
             : (table->n_buckets * tuning->growth_factor
                * tuning->growth_threshold));

          if ((float) SIZE_MAX <= candidate)
            {
              errno = ENOMEM;
              return -1;
            }

          /* If the rehash fails, arrange to return NULL.  */
          if (!hash_rehash (table, candidate))
            return -1;

          /* Update the bucket we are interested in.  */
          if (hash_find_entry (table, entry, &bucket, false) != NULL)
            abort ();
        }
    }

  /* ENTRY is not matched, it should be inserted.  */

  if (bucket->data)
    {
      struct hash_entry *new_entry = allocate_entry (table);

      if (new_entry == NULL)
        return -1;

      /* Add ENTRY in the overflow of the bucket.  */

      new_entry->data = (void *) entry;
      new_entry->next = bucket->next;
      bucket->next = new_entry;
      table->n_entries++;
      return 1;
    }

  /* Add ENTRY right in the bucket head.  */

  bucket->data = (void *) entry;
  table->n_entries++;
  table->n_buckets_used++;

  return 1;
}

void *
hash_insert (Hash_table *table, void const *entry)
{
  void const *matched_ent;
  int err = hash_insert_if_absent (table, entry, &matched_ent);
  return (err == -1
          ? NULL
          : (void *) (err == 0 ? matched_ent : entry));
}

void *
hash_remove (Hash_table *table, const void *entry)
{
  void *data;
  struct hash_entry *bucket;

  data = hash_find_entry (table, entry, &bucket, true);
  if (!data)
    return NULL;

  table->n_entries--;
  if (!bucket->data)
    {
      table->n_buckets_used--;

      /* If the shrink threshold of the buckets in use has been reached,
         rehash into a smaller table.  */

      if (table->n_buckets_used
          < table->tuning->shrink_threshold * table->n_buckets)
        {
          /* Check more fully, before starting real work.  If tuning arguments
             became invalid, the second check will rely on proper defaults.  */
          check_tuning (table);
          if (table->n_buckets_used
              < table->tuning->shrink_threshold * table->n_buckets)
            {
              const Hash_tuning *tuning = table->tuning;
              size_t candidate =
                (tuning->is_n_buckets
                 ? table->n_buckets * tuning->shrink_factor
                 : (table->n_buckets * tuning->shrink_factor
                    * tuning->growth_threshold));

              if (!hash_rehash (table, candidate))
                {
                  /* Failure to allocate memory in an attempt to
                     shrink the table is not fatal.  But since memory
                     is low, we can at least be kind and free any
                     spare entries, rather than keeping them tied up
                     in the free entry list.  */
#if ! USE_OBSTACK
                  struct hash_entry *cursor = table->free_entry_list;
                  struct hash_entry *next;
                  while (cursor)
                    {
                      next = cursor->next;
                      free (cursor);
                      cursor = next;
                    }
                  table->free_entry_list = NULL;
#endif
                }
            }
        }
    }

  return data;
}

void *
hash_delete (Hash_table *table, const void *entry)
{
  return hash_remove (table, entry);
}

/* Testing.  */

#if TESTING

void
hash_print (const Hash_table *table)
{
  struct hash_entry *bucket = (struct hash_entry *) table->bucket;

  for ( ; bucket < table->bucket_limit; bucket++)
    {
      struct hash_entry *cursor;

      if (bucket)
        printf ("%lu:\n", (unsigned long int) (bucket - table->bucket));

      for (cursor = bucket; cursor; cursor = cursor->next)
        {
          char const *s = cursor->data;
          /* FIXME */
          if (s)
            printf ("  %s\n", s);
        }
    }
}

#endif /* TESTING */