1 /* Implementation of the hash table lisp object type.
2 Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc.
3 Copyright (C) 1995, 1996 Ben Wing.
4 Copyright (C) 1997 Free Software Foundation, Inc.
6 This file is part of XEmacs.
8 XEmacs is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by the
10 Free Software Foundation; either version 2, or (at your option) any
13 XEmacs is distributed in the hope that it will be useful, but WITHOUT
14 ANY WARRANTY; without even the implied warranty of MERCNTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
18 You should have received a copy of the GNU General Public License
19 along with XEmacs; see the file COPYING. If not, write to
20 the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
21 Boston, MA 02111-1307, USA. */
23 /* Synched up with: Not in FSF. */
30 Lisp_Object Qhash_tablep;
31 static Lisp_Object Qhashtable, Qhash_table;
32 static Lisp_Object Qweakness, Qvalue, Qkey_value;
33 static Lisp_Object Vall_weak_hash_tables;
34 static Lisp_Object Qrehash_size, Qrehash_threshold;
35 static Lisp_Object Q_size, Q_test, Q_weakness, Q_rehash_size, Q_rehash_threshold;
37 /* obsolete as of 19990901 in xemacs-21.2 */
38 static Lisp_Object Qweak, Qkey_weak, Qvalue_weak, Qkey_value_weak;
39 static Lisp_Object Qnon_weak, Q_type;
47 struct Lisp_Hash_Table
49 struct lcrecord_header header;
54 double rehash_threshold;
56 hash_table_hash_function_t hash_function;
57 hash_table_test_function_t test_function;
59 enum hash_table_weakness weakness;
60 Lisp_Object next_weak; /* Used to chain together all of the weak
61 hash tables. Don't mark through this. */
64 #define HENTRY_CLEAR_P(hentry) ((*(EMACS_UINT*)(&((hentry)->key))) == 0)
65 #define CLEAR_HENTRY(hentry) \
66 ((*(EMACS_UINT*)(&((hentry)->key))) = 0, \
67 (*(EMACS_UINT*)(&((hentry)->value))) = 0)
69 #define HASH_TABLE_DEFAULT_SIZE 16
70 #define HASH_TABLE_DEFAULT_REHASH_SIZE 1.3
71 #define HASH_TABLE_MIN_SIZE 10
73 #define HASH_CODE(key, ht) \
74 ((((ht)->hash_function ? (ht)->hash_function (key) : LISP_HASH (key)) \
75 * (ht)->golden_ratio) \
78 #define KEYS_EQUAL_P(key1, key2, testfun) \
79 (EQ (key1, key2) || ((testfun) && (testfun) (key1, key2)))
81 #define LINEAR_PROBING_LOOP(probe, entries, size) \
83 !HENTRY_CLEAR_P (probe) || \
84 (probe == entries + size ? \
85 (probe = entries, !HENTRY_CLEAR_P (probe)) : 0); \
88 #ifndef ERROR_CHECK_HASH_TABLE
89 # ifdef ERROR_CHECK_TYPECHECK
90 # define ERROR_CHECK_HASH_TABLE 1
92 # define ERROR_CHECK_HASH_TABLE 0
96 #if ERROR_CHECK_HASH_TABLE
98 check_hash_table_invariants (Lisp_Hash_Table *ht)
100 assert (ht->count < ht->size);
101 assert (ht->count <= ht->rehash_count);
102 assert (ht->rehash_count < ht->size);
103 assert ((double) ht->count * ht->rehash_threshold - 1 <= (double) ht->rehash_count);
104 assert (HENTRY_CLEAR_P (ht->hentries + ht->size));
107 #define check_hash_table_invariants(ht)
110 /* We use linear probing instead of double hashing, despite its lack
111 of blessing by Knuth and company, because, as a result of the
112 increasing discrepancy between CPU speeds and memory speeds, cache
113 behavior is becoming increasingly important, e.g:
115 For a trivial loop, the penalty for non-sequential access of an array is:
116 - a factor of 3-4 on Pentium Pro 200 Mhz
117 - a factor of 10 on Ultrasparc 300 Mhz */
119 /* Return a suitable size for a hash table, with at least SIZE slots. */
121 hash_table_size (size_t requested_size)
123 /* Return some prime near, but greater than or equal to, SIZE.
124 Decades from the time of writing, someone will have a system large
125 enough that the list below will be too short... */
126 static const size_t primes [] =
128 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031,
129 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783,
130 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941,
131 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519,
132 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301,
133 10445899, 13579681, 17653589, 22949669, 29834603, 38784989,
134 50420551, 65546729, 85210757, 110774011, 144006217, 187208107,
135 243370577, 316381771, 411296309, 534685237, 695090819, 903618083,
136 1174703521, 1527114613, 1985248999, 2580823717UL, 3355070839UL
138 /* We've heard of binary search. */
140 for (low = 0, high = countof (primes) - 1; high - low > 1;)
142 /* Loop Invariant: size < primes [high] */
143 int mid = (low + high) / 2;
144 if (primes [mid] < requested_size)
149 return primes [high];
153 #if 0 /* I don't think these are needed any more.
154 If using the general lisp_object_equal_*() functions
155 causes efficiency problems, these can be resurrected. --ben */
156 /* equality and hash functions for Lisp strings */
158 lisp_string_equal (Lisp_Object str1, Lisp_Object str2)
160 /* This is wrong anyway. You can't use strcmp() on Lisp strings,
161 because they can contain zero characters. */
162 return !strcmp ((char *) XSTRING_DATA (str1), (char *) XSTRING_DATA (str2));
166 lisp_string_hash (Lisp_Object obj)
168 return hash_string (XSTRING_DATA (str), XSTRING_LENGTH (str));
174 lisp_object_eql_equal (Lisp_Object obj1, Lisp_Object obj2)
176 return EQ (obj1, obj2) || (FLOATP (obj1) && internal_equal (obj1, obj2, 0));
180 lisp_object_eql_hash (Lisp_Object obj)
182 return FLOATP (obj) ? internal_hash (obj, 0) : LISP_HASH (obj);
186 lisp_object_equal_equal (Lisp_Object obj1, Lisp_Object obj2)
188 return internal_equal (obj1, obj2, 0);
192 lisp_object_equal_hash (Lisp_Object obj)
194 return internal_hash (obj, 0);
199 mark_hash_table (Lisp_Object obj)
201 Lisp_Hash_Table *ht = XHASH_TABLE (obj);
203 /* If the hash table is weak, we don't want to mark the keys and
204 values (we scan over them after everything else has been marked,
205 and mark or remove them as necessary). */
206 if (ht->weakness == HASH_TABLE_NON_WEAK)
208 hentry *e, *sentinel;
210 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
211 if (!HENTRY_CLEAR_P (e))
213 mark_object (e->key);
214 mark_object (e->value);
220 /* Equality of hash tables. Two hash tables are equal when they are of
221 the same weakness and test function, they have the same number of
222 elements, and for each key in the hash table, the values are `equal'.
224 This is similar to Common Lisp `equalp' of hash tables, with the
225 difference that CL requires the keys to be compared with the test
226 function, which we don't do. Doing that would require consing, and
227 consing is a bad idea in `equal'. Anyway, our method should provide
228 the same result -- if the keys are not equal according to the test
229 function, then Fgethash() in hash_table_equal_mapper() will fail. */
231 hash_table_equal (Lisp_Object hash_table1, Lisp_Object hash_table2, int depth)
233 Lisp_Hash_Table *ht1 = XHASH_TABLE (hash_table1);
234 Lisp_Hash_Table *ht2 = XHASH_TABLE (hash_table2);
235 hentry *e, *sentinel;
237 if ((ht1->test_function != ht2->test_function) ||
238 (ht1->weakness != ht2->weakness) ||
239 (ht1->count != ht2->count))
244 for (e = ht1->hentries, sentinel = e + ht1->size; e < sentinel; e++)
245 if (!HENTRY_CLEAR_P (e))
246 /* Look up the key in the other hash table, and compare the values. */
248 Lisp_Object value_in_other = Fgethash (e->key, hash_table2, Qunbound);
249 if (UNBOUNDP (value_in_other) ||
250 !internal_equal (e->value, value_in_other, depth))
251 return 0; /* Give up */
257 /* This is not a great hash function, but it _is_ correct and fast.
258 Examining all entries is too expensive, and examining a random
259 subset does not yield a correct hash function. */
261 hash_table_hash (Lisp_Object hash_table, int depth)
263 return XHASH_TABLE (hash_table)->count;
267 /* Printing hash tables.
269 This is non-trivial, because we use a readable structure-style
270 syntax for hash tables. This means that a typical hash table will be
271 readably printed in the form of:
273 #s(hash-table size 2 data (key1 value1 key2 value2))
275 The supported hash table structure keywords and their values are:
276 `test' (eql (or nil), eq or equal)
277 `size' (a natnum or nil)
278 `rehash-size' (a float)
279 `rehash-threshold' (a float)
280 `weakness' (nil, t, key or value)
283 If `print-readably' is nil, then a simpler syntax is used, for example
285 #<hash-table size 2/13 data (key1 value1 key2 value2) 0x874d>
287 The data is truncated to four pairs, and the rest is shown with
288 `...'. This printer does not cons. */
291 /* Print the data of the hash table. This maps through a Lisp
292 hash table and prints key/value pairs using PRINTCHARFUN. */
294 print_hash_table_data (Lisp_Hash_Table *ht, Lisp_Object printcharfun)
297 hentry *e, *sentinel;
299 write_c_string (" data (", printcharfun);
301 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
302 if (!HENTRY_CLEAR_P (e))
305 write_c_string (" ", printcharfun);
306 if (!print_readably && count > 3)
308 write_c_string ("...", printcharfun);
311 print_internal (e->key, printcharfun, 1);
312 write_c_string (" ", printcharfun);
313 print_internal (e->value, printcharfun, 1);
317 write_c_string (")", printcharfun);
321 print_hash_table (Lisp_Object obj, Lisp_Object printcharfun, int escapeflag)
323 Lisp_Hash_Table *ht = XHASH_TABLE (obj);
326 write_c_string (print_readably ? "#s(hash-table" : "#<hash-table",
329 /* These checks have a kludgy look to them, but they are safe.
330 Due to nature of hashing, you cannot use arbitrary
331 test functions anyway. */
332 if (!ht->test_function)
333 write_c_string (" test eq", printcharfun);
334 else if (ht->test_function == lisp_object_equal_equal)
335 write_c_string (" test equal", printcharfun);
336 else if (ht->test_function == lisp_object_eql_equal)
341 if (ht->count || !print_readably)
344 sprintf (buf, " size %lu", (unsigned long) ht->count);
346 sprintf (buf, " size %lu/%lu",
347 (unsigned long) ht->count,
348 (unsigned long) ht->size);
349 write_c_string (buf, printcharfun);
352 if (ht->weakness != HASH_TABLE_NON_WEAK)
354 sprintf (buf, " weakness %s",
355 (ht->weakness == HASH_TABLE_WEAK ? "t" :
356 ht->weakness == HASH_TABLE_KEY_WEAK ? "key" :
357 ht->weakness == HASH_TABLE_VALUE_WEAK ? "value" :
358 ht->weakness == HASH_TABLE_KEY_VALUE_WEAK ? "key-value" :
359 "you-d-better-not-see-this"));
360 write_c_string (buf, printcharfun);
364 print_hash_table_data (ht, printcharfun);
367 write_c_string (")", printcharfun);
370 sprintf (buf, " 0x%x>", ht->header.uid);
371 write_c_string (buf, printcharfun);
376 finalize_hash_table (void *header, int for_disksave)
380 Lisp_Hash_Table *ht = (Lisp_Hash_Table *) header;
382 xfree (ht->hentries);
387 static const struct lrecord_description hentry_description_1[] = {
388 { XD_LISP_OBJECT, offsetof (hentry, key) },
389 { XD_LISP_OBJECT, offsetof (hentry, value) },
393 static const struct struct_description hentry_description = {
398 const struct lrecord_description hash_table_description[] = {
399 { XD_SIZE_T, offsetof (Lisp_Hash_Table, size) },
400 { XD_STRUCT_PTR, offsetof (Lisp_Hash_Table, hentries), XD_INDIRECT(0, 1), &hentry_description },
401 { XD_LO_LINK, offsetof (Lisp_Hash_Table, next_weak) },
405 DEFINE_LRECORD_IMPLEMENTATION ("hash-table", hash_table,
406 mark_hash_table, print_hash_table,
408 hash_table_equal, hash_table_hash,
409 hash_table_description,
412 static Lisp_Hash_Table *
413 xhash_table (Lisp_Object hash_table)
416 CHECK_HASH_TABLE (hash_table);
417 check_hash_table_invariants (XHASH_TABLE (hash_table));
418 return XHASH_TABLE (hash_table);
422 /************************************************************************/
423 /* Creation of Hash Tables */
424 /************************************************************************/
426 /* Creation of hash tables, without error-checking. */
428 compute_hash_table_derived_values (Lisp_Hash_Table *ht)
430 ht->rehash_count = (size_t)
431 ((double) ht->size * ht->rehash_threshold);
432 ht->golden_ratio = (size_t)
433 ((double) ht->size * (.6180339887 / (double) sizeof (Lisp_Object)));
437 make_general_lisp_hash_table (enum hash_table_test test,
440 double rehash_threshold,
441 enum hash_table_weakness weakness)
443 Lisp_Object hash_table;
444 Lisp_Hash_Table *ht = alloc_lcrecord_type (Lisp_Hash_Table, &lrecord_hash_table);
449 ht->test_function = 0;
450 ht->hash_function = 0;
454 ht->test_function = lisp_object_eql_equal;
455 ht->hash_function = lisp_object_eql_hash;
458 case HASH_TABLE_EQUAL:
459 ht->test_function = lisp_object_equal_equal;
460 ht->hash_function = lisp_object_equal_hash;
467 ht->weakness = weakness;
470 rehash_size > 1.0 ? rehash_size : HASH_TABLE_DEFAULT_REHASH_SIZE;
472 ht->rehash_threshold =
473 rehash_threshold > 0.0 ? rehash_threshold :
474 size > 4096 && !ht->test_function ? 0.7 : 0.6;
476 if (size < HASH_TABLE_MIN_SIZE)
477 size = HASH_TABLE_MIN_SIZE;
478 ht->size = hash_table_size ((size_t) (((double) size / ht->rehash_threshold)
482 compute_hash_table_derived_values (ht);
484 /* We leave room for one never-occupied sentinel hentry at the end. */
485 ht->hentries = xnew_array (hentry, ht->size + 1);
488 hentry *e, *sentinel;
489 for (e = ht->hentries, sentinel = e + ht->size; e <= sentinel; e++)
493 XSETHASH_TABLE (hash_table, ht);
495 if (weakness == HASH_TABLE_NON_WEAK)
496 ht->next_weak = Qunbound;
498 ht->next_weak = Vall_weak_hash_tables, Vall_weak_hash_tables = hash_table;
504 make_lisp_hash_table (size_t size,
505 enum hash_table_weakness weakness,
506 enum hash_table_test test)
508 return make_general_lisp_hash_table (test, size, -1.0, -1.0, weakness);
511 /* Pretty reading of hash tables.
513 Here we use the existing structures mechanism (which is,
514 unfortunately, pretty cumbersome) for validating and instantiating
515 the hash tables. The idea is that the side-effect of reading a
516 #s(hash-table PLIST) object is creation of a hash table with desired
517 properties, and that the hash table is returned. */
519 /* Validation functions: each keyword provides its own validation
520 function. The errors should maybe be continuable, but it is
521 unclear how this would cope with ERRB. */
523 hash_table_size_validate (Lisp_Object keyword, Lisp_Object value,
529 maybe_signal_error (Qwrong_type_argument, list2 (Qnatnump, value),
535 decode_hash_table_size (Lisp_Object obj)
537 return NILP (obj) ? HASH_TABLE_DEFAULT_SIZE : XINT (obj);
541 hash_table_weakness_validate (Lisp_Object keyword, Lisp_Object value,
544 if (EQ (value, Qnil)) return 1;
545 if (EQ (value, Qt)) return 1;
546 if (EQ (value, Qkey)) return 1;
547 if (EQ (value, Qkey_value)) return 1;
548 if (EQ (value, Qvalue)) return 1;
550 /* Following values are obsolete as of 19990901 in xemacs-21.2 */
551 if (EQ (value, Qnon_weak)) return 1;
552 if (EQ (value, Qweak)) return 1;
553 if (EQ (value, Qkey_weak)) return 1;
554 if (EQ (value, Qkey_value_weak)) return 1;
555 if (EQ (value, Qvalue_weak)) return 1;
557 maybe_signal_simple_error ("Invalid hash table weakness",
558 value, Qhash_table, errb);
562 static enum hash_table_weakness
563 decode_hash_table_weakness (Lisp_Object obj)
565 if (EQ (obj, Qnil)) return HASH_TABLE_NON_WEAK;
566 if (EQ (obj, Qt)) return HASH_TABLE_WEAK;
567 if (EQ (obj, Qkey)) return HASH_TABLE_KEY_WEAK;
568 if (EQ (obj, Qkey_value)) return HASH_TABLE_KEY_VALUE_WEAK;
569 if (EQ (obj, Qvalue)) return HASH_TABLE_VALUE_WEAK;
571 /* Following values are obsolete as of 19990901 in xemacs-21.2 */
572 if (EQ (obj, Qnon_weak)) return HASH_TABLE_NON_WEAK;
573 if (EQ (obj, Qweak)) return HASH_TABLE_WEAK;
574 if (EQ (obj, Qkey_weak)) return HASH_TABLE_KEY_WEAK;
575 if (EQ (obj, Qkey_value_weak)) return HASH_TABLE_KEY_VALUE_WEAK;
576 if (EQ (obj, Qvalue_weak)) return HASH_TABLE_VALUE_WEAK;
578 signal_simple_error ("Invalid hash table weakness", obj);
579 return HASH_TABLE_NON_WEAK; /* not reached */
583 hash_table_test_validate (Lisp_Object keyword, Lisp_Object value,
586 if (EQ (value, Qnil)) return 1;
587 if (EQ (value, Qeq)) return 1;
588 if (EQ (value, Qequal)) return 1;
589 if (EQ (value, Qeql)) return 1;
591 maybe_signal_simple_error ("Invalid hash table test",
592 value, Qhash_table, errb);
596 static enum hash_table_test
597 decode_hash_table_test (Lisp_Object obj)
599 if (EQ (obj, Qnil)) return HASH_TABLE_EQL;
600 if (EQ (obj, Qeq)) return HASH_TABLE_EQ;
601 if (EQ (obj, Qequal)) return HASH_TABLE_EQUAL;
602 if (EQ (obj, Qeql)) return HASH_TABLE_EQL;
604 signal_simple_error ("Invalid hash table test", obj);
605 return HASH_TABLE_EQ; /* not reached */
609 hash_table_rehash_size_validate (Lisp_Object keyword, Lisp_Object value,
614 maybe_signal_error (Qwrong_type_argument, list2 (Qfloatp, value),
620 double rehash_size = XFLOAT_DATA (value);
621 if (rehash_size <= 1.0)
623 maybe_signal_simple_error
624 ("Hash table rehash size must be greater than 1.0",
625 value, Qhash_table, errb);
634 decode_hash_table_rehash_size (Lisp_Object rehash_size)
636 return NILP (rehash_size) ? -1.0 : XFLOAT_DATA (rehash_size);
640 hash_table_rehash_threshold_validate (Lisp_Object keyword, Lisp_Object value,
645 maybe_signal_error (Qwrong_type_argument, list2 (Qfloatp, value),
651 double rehash_threshold = XFLOAT_DATA (value);
652 if (rehash_threshold <= 0.0 || rehash_threshold >= 1.0)
654 maybe_signal_simple_error
655 ("Hash table rehash threshold must be between 0.0 and 1.0",
656 value, Qhash_table, errb);
665 decode_hash_table_rehash_threshold (Lisp_Object rehash_threshold)
667 return NILP (rehash_threshold) ? -1.0 : XFLOAT_DATA (rehash_threshold);
671 hash_table_data_validate (Lisp_Object keyword, Lisp_Object value,
676 GET_EXTERNAL_LIST_LENGTH (value, len);
680 maybe_signal_simple_error
681 ("Hash table data must have alternating key/value pairs",
682 value, Qhash_table, errb);
688 /* The actual instantiation of a hash table. This does practically no
689 error checking, because it relies on the fact that the paranoid
690 functions above have error-checked everything to the last details.
691 If this assumption is wrong, we will get a crash immediately (with
692 error-checking compiled in), and we'll know if there is a bug in
693 the structure mechanism. So there. */
695 hash_table_instantiate (Lisp_Object plist)
697 Lisp_Object hash_table;
698 Lisp_Object test = Qnil;
699 Lisp_Object size = Qnil;
700 Lisp_Object rehash_size = Qnil;
701 Lisp_Object rehash_threshold = Qnil;
702 Lisp_Object weakness = Qnil;
703 Lisp_Object data = Qnil;
705 while (!NILP (plist))
707 Lisp_Object key, value;
708 key = XCAR (plist); plist = XCDR (plist);
709 value = XCAR (plist); plist = XCDR (plist);
711 if (EQ (key, Qtest)) test = value;
712 else if (EQ (key, Qsize)) size = value;
713 else if (EQ (key, Qrehash_size)) rehash_size = value;
714 else if (EQ (key, Qrehash_threshold)) rehash_threshold = value;
715 else if (EQ (key, Qweakness)) weakness = value;
716 else if (EQ (key, Qdata)) data = value;
717 else if (EQ (key, Qtype))/*obsolete*/ weakness = value;
722 /* Create the hash table. */
723 hash_table = make_general_lisp_hash_table
724 (decode_hash_table_test (test),
725 decode_hash_table_size (size),
726 decode_hash_table_rehash_size (rehash_size),
727 decode_hash_table_rehash_threshold (rehash_threshold),
728 decode_hash_table_weakness (weakness));
730 /* I'm not sure whether this can GC, but better safe than sorry. */
735 /* And fill it with data. */
738 Lisp_Object key, value;
739 key = XCAR (data); data = XCDR (data);
740 value = XCAR (data); data = XCDR (data);
741 Fputhash (key, value, hash_table);
750 structure_type_create_hash_table_structure_name (Lisp_Object structure_name)
752 struct structure_type *st;
754 st = define_structure_type (structure_name, 0, hash_table_instantiate);
755 define_structure_type_keyword (st, Qtest, hash_table_test_validate);
756 define_structure_type_keyword (st, Qsize, hash_table_size_validate);
757 define_structure_type_keyword (st, Qrehash_size, hash_table_rehash_size_validate);
758 define_structure_type_keyword (st, Qrehash_threshold, hash_table_rehash_threshold_validate);
759 define_structure_type_keyword (st, Qweakness, hash_table_weakness_validate);
760 define_structure_type_keyword (st, Qdata, hash_table_data_validate);
762 /* obsolete as of 19990901 in xemacs-21.2 */
763 define_structure_type_keyword (st, Qtype, hash_table_weakness_validate);
766 /* Create a built-in Lisp structure type named `hash-table'.
767 We make #s(hashtable ...) equivalent to #s(hash-table ...),
768 for backward compatibility.
769 This is called from emacs.c. */
771 structure_type_create_hash_table (void)
773 structure_type_create_hash_table_structure_name (Qhash_table);
774 structure_type_create_hash_table_structure_name (Qhashtable); /* compat */
778 /************************************************************************/
779 /* Definition of Lisp-visible methods */
780 /************************************************************************/
782 DEFUN ("hash-table-p", Fhash_table_p, 1, 1, 0, /*
783 Return t if OBJECT is a hash table, else nil.
787 return HASH_TABLEP (object) ? Qt : Qnil;
790 DEFUN ("make-hash-table", Fmake_hash_table, 0, MANY, 0, /*
791 Return a new empty hash table object.
792 Use Common Lisp style keywords to specify hash table properties.
793 (make-hash-table &key test size rehash-size rehash-threshold weakness)
795 Keyword :test can be `eq', `eql' (default) or `equal'.
796 Comparison between keys is done using this function.
797 If speed is important, consider using `eq'.
798 When storing strings in the hash table, you will likely need to use `equal'.
800 Keyword :size specifies the number of keys likely to be inserted.
801 This number of entries can be inserted without enlarging the hash table.
803 Keyword :rehash-size must be a float greater than 1.0, and specifies
804 the factor by which to increase the size of the hash table when enlarging.
806 Keyword :rehash-threshold must be a float between 0.0 and 1.0,
807 and specifies the load factor of the hash table which triggers enlarging.
809 Non-standard keyword :weakness can be `nil' (default), `t', `key', `value'
812 A weak hash table is one whose pointers do not count as GC referents:
813 for any key-value pair in the hash table, if the only remaining pointer
814 to either the key or the value is in a weak hash table, then the pair
815 will be removed from the hash table, and the key and value collected.
816 A non-weak hash table (or any other pointer) would prevent the object
817 from being collected.
819 A key-weak hash table is similar to a fully-weak hash table except that
820 a key-value pair will be removed only if the key remains unmarked
821 outside of weak hash tables. The pair will remain in the hash table if
822 the key is pointed to by something other than a weak hash table, even
825 A value-weak hash table is similar to a fully-weak hash table except
826 that a key-value pair will be removed only if the value remains
827 unmarked outside of weak hash tables. The pair will remain in the
828 hash table if the value is pointed to by something other than a weak
829 hash table, even if the key is not.
831 A key-value-weak hash table is similar to a fully-weak hash table except
832 that a key-value pair will be removed only if the value and the key remain
833 unmarked outside of weak hash tables. The pair will remain in the
834 hash table if the value or key are pointed to by something other than a weak
835 hash table, even if the other is not.
837 (int nargs, Lisp_Object *args))
840 Lisp_Object test = Qnil;
841 Lisp_Object size = Qnil;
842 Lisp_Object rehash_size = Qnil;
843 Lisp_Object rehash_threshold = Qnil;
844 Lisp_Object weakness = Qnil;
846 while (i + 1 < nargs)
848 Lisp_Object keyword = args[i++];
849 Lisp_Object value = args[i++];
851 if (EQ (keyword, Q_test)) test = value;
852 else if (EQ (keyword, Q_size)) size = value;
853 else if (EQ (keyword, Q_rehash_size)) rehash_size = value;
854 else if (EQ (keyword, Q_rehash_threshold)) rehash_threshold = value;
855 else if (EQ (keyword, Q_weakness)) weakness = value;
856 else if (EQ (keyword, Q_type))/*obsolete*/ weakness = value;
857 else signal_simple_error ("Invalid hash table property keyword", keyword);
861 signal_simple_error ("Hash table property requires a value", args[i]);
863 #define VALIDATE_VAR(var) \
864 if (!NILP (var)) hash_table_##var##_validate (Q##var, var, ERROR_ME);
868 VALIDATE_VAR (rehash_size);
869 VALIDATE_VAR (rehash_threshold);
870 VALIDATE_VAR (weakness);
872 return make_general_lisp_hash_table
873 (decode_hash_table_test (test),
874 decode_hash_table_size (size),
875 decode_hash_table_rehash_size (rehash_size),
876 decode_hash_table_rehash_threshold (rehash_threshold),
877 decode_hash_table_weakness (weakness));
880 DEFUN ("copy-hash-table", Fcopy_hash_table, 1, 1, 0, /*
881 Return a new hash table containing the same keys and values as HASH-TABLE.
882 The keys and values will not themselves be copied.
886 const Lisp_Hash_Table *ht_old = xhash_table (hash_table);
887 Lisp_Hash_Table *ht = alloc_lcrecord_type (Lisp_Hash_Table, &lrecord_hash_table);
889 copy_lcrecord (ht, ht_old);
891 ht->hentries = xnew_array (hentry, ht_old->size + 1);
892 memcpy (ht->hentries, ht_old->hentries, (ht_old->size + 1) * sizeof (hentry));
894 XSETHASH_TABLE (hash_table, ht);
896 if (! EQ (ht->next_weak, Qunbound))
898 ht->next_weak = Vall_weak_hash_tables;
899 Vall_weak_hash_tables = hash_table;
906 resize_hash_table (Lisp_Hash_Table *ht, size_t new_size)
908 hentry *old_entries, *new_entries, *sentinel, *e;
914 old_entries = ht->hentries;
916 ht->hentries = xnew_array_and_zero (hentry, new_size + 1);
917 new_entries = ht->hentries;
919 compute_hash_table_derived_values (ht);
921 for (e = old_entries, sentinel = e + old_size; e < sentinel; e++)
922 if (!HENTRY_CLEAR_P (e))
924 hentry *probe = new_entries + HASH_CODE (e->key, ht);
925 LINEAR_PROBING_LOOP (probe, new_entries, new_size)
930 if (!DUMPEDP (old_entries))
934 /* After a hash table has been saved to disk and later restored by the
935 portable dumper, it contains the same objects, but their addresses
936 and thus their HASH_CODEs have changed. */
938 pdump_reorganize_hash_table (Lisp_Object hash_table)
940 const Lisp_Hash_Table *ht = xhash_table (hash_table);
941 hentry *new_entries = xnew_array_and_zero (hentry, ht->size + 1);
942 hentry *e, *sentinel;
944 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
945 if (!HENTRY_CLEAR_P (e))
947 hentry *probe = new_entries + HASH_CODE (e->key, ht);
948 LINEAR_PROBING_LOOP (probe, new_entries, ht->size)
953 memcpy (ht->hentries, new_entries, ht->size * sizeof (hentry));
959 enlarge_hash_table (Lisp_Hash_Table *ht)
962 hash_table_size ((size_t) ((double) ht->size * ht->rehash_size));
963 resize_hash_table (ht, new_size);
967 find_hentry (Lisp_Object key, const Lisp_Hash_Table *ht)
969 hash_table_test_function_t test_function = ht->test_function;
970 hentry *entries = ht->hentries;
971 hentry *probe = entries + HASH_CODE (key, ht);
973 LINEAR_PROBING_LOOP (probe, entries, ht->size)
974 if (KEYS_EQUAL_P (probe->key, key, test_function))
980 DEFUN ("gethash", Fgethash, 2, 3, 0, /*
981 Find hash value for KEY in HASH-TABLE.
982 If there is no corresponding value, return DEFAULT (which defaults to nil).
984 (key, hash_table, default_))
986 const Lisp_Hash_Table *ht = xhash_table (hash_table);
987 hentry *e = find_hentry (key, ht);
989 return HENTRY_CLEAR_P (e) ? default_ : e->value;
992 DEFUN ("puthash", Fputhash, 3, 3, 0, /*
993 Hash KEY to VALUE in HASH-TABLE.
995 (key, value, hash_table))
997 Lisp_Hash_Table *ht = xhash_table (hash_table);
998 hentry *e = find_hentry (key, ht);
1000 if (!HENTRY_CLEAR_P (e))
1001 return e->value = value;
1006 if (++ht->count >= ht->rehash_count)
1007 enlarge_hash_table (ht);
1012 /* Remove hentry pointed at by PROBE.
1013 Subsequent entries are removed and reinserted.
1014 We don't use tombstones - too wasteful. */
1016 remhash_1 (Lisp_Hash_Table *ht, hentry *entries, hentry *probe)
1018 size_t size = ht->size;
1019 CLEAR_HENTRY (probe);
1023 LINEAR_PROBING_LOOP (probe, entries, size)
1025 Lisp_Object key = probe->key;
1026 hentry *probe2 = entries + HASH_CODE (key, ht);
1027 LINEAR_PROBING_LOOP (probe2, entries, size)
1028 if (EQ (probe2->key, key))
1029 /* hentry at probe doesn't need to move. */
1030 goto continue_outer_loop;
1031 /* Move hentry from probe to new home at probe2. */
1033 CLEAR_HENTRY (probe);
1034 continue_outer_loop: continue;
1038 DEFUN ("remhash", Fremhash, 2, 2, 0, /*
1039 Remove the entry for KEY from HASH-TABLE.
1040 Do nothing if there is no entry for KEY in HASH-TABLE.
1044 Lisp_Hash_Table *ht = xhash_table (hash_table);
1045 hentry *e = find_hentry (key, ht);
1047 if (HENTRY_CLEAR_P (e))
1050 remhash_1 (ht, ht->hentries, e);
1054 DEFUN ("clrhash", Fclrhash, 1, 1, 0, /*
1055 Remove all entries from HASH-TABLE, leaving it empty.
1059 Lisp_Hash_Table *ht = xhash_table (hash_table);
1060 hentry *e, *sentinel;
1062 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
1069 /************************************************************************/
1070 /* Accessor Functions */
1071 /************************************************************************/
1073 DEFUN ("hash-table-count", Fhash_table_count, 1, 1, 0, /*
1074 Return the number of entries in HASH-TABLE.
1078 return make_int (xhash_table (hash_table)->count);
1081 DEFUN ("hash-table-test", Fhash_table_test, 1, 1, 0, /*
1082 Return the test function of HASH-TABLE.
1083 This can be one of `eq', `eql' or `equal'.
1087 hash_table_test_function_t fun = xhash_table (hash_table)->test_function;
1089 return (fun == lisp_object_eql_equal ? Qeql :
1090 fun == lisp_object_equal_equal ? Qequal :
1094 DEFUN ("hash-table-size", Fhash_table_size, 1, 1, 0, /*
1095 Return the size of HASH-TABLE.
1096 This is the current number of slots in HASH-TABLE, whether occupied or not.
1100 return make_int (xhash_table (hash_table)->size);
1103 DEFUN ("hash-table-rehash-size", Fhash_table_rehash_size, 1, 1, 0, /*
1104 Return the current rehash size of HASH-TABLE.
1105 This is a float greater than 1.0; the factor by which HASH-TABLE
1106 is enlarged when the rehash threshold is exceeded.
1110 return make_float (xhash_table (hash_table)->rehash_size);
1113 DEFUN ("hash-table-rehash-threshold", Fhash_table_rehash_threshold, 1, 1, 0, /*
1114 Return the current rehash threshold of HASH-TABLE.
1115 This is a float between 0.0 and 1.0; the maximum `load factor' of HASH-TABLE,
1116 beyond which the HASH-TABLE is enlarged by rehashing.
1120 return make_float (xhash_table (hash_table)->rehash_threshold);
1123 DEFUN ("hash-table-weakness", Fhash_table_weakness, 1, 1, 0, /*
1124 Return the weakness of HASH-TABLE.
1125 This can be one of `nil', `t', `key' or `value'.
1129 switch (xhash_table (hash_table)->weakness)
1131 case HASH_TABLE_WEAK: return Qt;
1132 case HASH_TABLE_KEY_WEAK: return Qkey;
1133 case HASH_TABLE_KEY_VALUE_WEAK: return Qkey_value;
1134 case HASH_TABLE_VALUE_WEAK: return Qvalue;
1135 default: return Qnil;
1139 /* obsolete as of 19990901 in xemacs-21.2 */
1140 DEFUN ("hash-table-type", Fhash_table_type, 1, 1, 0, /*
1141 Return the type of HASH-TABLE.
1142 This can be one of `non-weak', `weak', `key-weak' or `value-weak'.
1146 switch (xhash_table (hash_table)->weakness)
1148 case HASH_TABLE_WEAK: return Qweak;
1149 case HASH_TABLE_KEY_WEAK: return Qkey_weak;
1150 case HASH_TABLE_KEY_VALUE_WEAK: return Qkey_value_weak;
1151 case HASH_TABLE_VALUE_WEAK: return Qvalue_weak;
1152 default: return Qnon_weak;
1156 /************************************************************************/
1157 /* Mapping Functions */
1158 /************************************************************************/
1159 DEFUN ("maphash", Fmaphash, 2, 2, 0, /*
1160 Map FUNCTION over entries in HASH-TABLE, calling it with two args,
1161 each key and value in HASH-TABLE.
1163 FUNCTION may not modify HASH-TABLE, with the one exception that FUNCTION
1164 may remhash or puthash the entry currently being processed by FUNCTION.
1166 (function, hash_table))
1168 const Lisp_Hash_Table *ht = xhash_table (hash_table);
1169 const hentry *e, *sentinel;
1171 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
1172 if (!HENTRY_CLEAR_P (e))
1174 Lisp_Object args[3], key;
1180 Ffuncall (countof (args), args);
1181 /* Has FUNCTION done a remhash? */
1182 if (!EQ (key, e->key) && !HENTRY_CLEAR_P (e))
1189 /* Map *C* function FUNCTION over the elements of a lisp hash table. */
1191 elisp_maphash (maphash_function_t function,
1192 Lisp_Object hash_table, void *extra_arg)
1194 const Lisp_Hash_Table *ht = XHASH_TABLE (hash_table);
1195 const hentry *e, *sentinel;
1197 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
1198 if (!HENTRY_CLEAR_P (e))
1203 if (function (key, e->value, extra_arg))
1205 /* Has FUNCTION done a remhash? */
1206 if (!EQ (key, e->key) && !HENTRY_CLEAR_P (e))
1211 /* Remove all elements of a lisp hash table satisfying *C* predicate PREDICATE. */
1213 elisp_map_remhash (maphash_function_t predicate,
1214 Lisp_Object hash_table, void *extra_arg)
1216 Lisp_Hash_Table *ht = XHASH_TABLE (hash_table);
1217 hentry *e, *entries, *sentinel;
1219 for (e = entries = ht->hentries, sentinel = e + ht->size; e < sentinel; e++)
1220 if (!HENTRY_CLEAR_P (e))
1223 if (predicate (e->key, e->value, extra_arg))
1225 remhash_1 (ht, entries, e);
1226 if (!HENTRY_CLEAR_P (e))
1233 /************************************************************************/
1234 /* garbage collecting weak hash tables */
1235 /************************************************************************/
1236 #define MARK_OBJ(obj) do { \
1237 Lisp_Object mo_obj = (obj); \
1238 if (!marked_p (mo_obj)) \
1240 mark_object (mo_obj); \
1246 /* Complete the marking for semi-weak hash tables. */
1248 finish_marking_weak_hash_tables (void)
1250 Lisp_Object hash_table;
1253 for (hash_table = Vall_weak_hash_tables;
1255 hash_table = XHASH_TABLE (hash_table)->next_weak)
1257 const Lisp_Hash_Table *ht = XHASH_TABLE (hash_table);
1258 const hentry *e = ht->hentries;
1259 const hentry *sentinel = e + ht->size;
1261 if (! marked_p (hash_table))
1262 /* The hash table is probably garbage. Ignore it. */
1265 /* Now, scan over all the pairs. For all pairs that are
1266 half-marked, we may need to mark the other half if we're
1267 keeping this pair. */
1268 switch (ht->weakness)
1270 case HASH_TABLE_KEY_WEAK:
1271 for (; e < sentinel; e++)
1272 if (!HENTRY_CLEAR_P (e))
1273 if (marked_p (e->key))
1274 MARK_OBJ (e->value);
1277 case HASH_TABLE_VALUE_WEAK:
1278 for (; e < sentinel; e++)
1279 if (!HENTRY_CLEAR_P (e))
1280 if (marked_p (e->value))
1284 case HASH_TABLE_KEY_VALUE_WEAK:
1285 for (; e < sentinel; e++)
1286 if (!HENTRY_CLEAR_P (e))
1288 if (marked_p (e->value))
1290 else if (marked_p (e->key))
1291 MARK_OBJ (e->value);
1295 case HASH_TABLE_KEY_CAR_WEAK:
1296 for (; e < sentinel; e++)
1297 if (!HENTRY_CLEAR_P (e))
1298 if (!CONSP (e->key) || marked_p (XCAR (e->key)))
1301 MARK_OBJ (e->value);
1305 case HASH_TABLE_VALUE_CAR_WEAK:
1306 for (; e < sentinel; e++)
1307 if (!HENTRY_CLEAR_P (e))
1308 if (!CONSP (e->value) || marked_p (XCAR (e->value)))
1311 MARK_OBJ (e->value);
1324 prune_weak_hash_tables (void)
1326 Lisp_Object hash_table, prev = Qnil;
1327 for (hash_table = Vall_weak_hash_tables;
1329 hash_table = XHASH_TABLE (hash_table)->next_weak)
1331 if (! marked_p (hash_table))
1333 /* This hash table itself is garbage. Remove it from the list. */
1335 Vall_weak_hash_tables = XHASH_TABLE (hash_table)->next_weak;
1337 XHASH_TABLE (prev)->next_weak = XHASH_TABLE (hash_table)->next_weak;
1341 /* Now, scan over all the pairs. Remove all of the pairs
1342 in which the key or value, or both, is unmarked
1343 (depending on the weakness of the hash table). */
1344 Lisp_Hash_Table *ht = XHASH_TABLE (hash_table);
1345 hentry *entries = ht->hentries;
1346 hentry *sentinel = entries + ht->size;
1349 for (e = entries; e < sentinel; e++)
1350 if (!HENTRY_CLEAR_P (e))
1353 if (!marked_p (e->key) || !marked_p (e->value))
1355 remhash_1 (ht, entries, e);
1356 if (!HENTRY_CLEAR_P (e))
1366 /* Return a hash value for an array of Lisp_Objects of size SIZE. */
1369 internal_array_hash (Lisp_Object *arr, int size, int depth)
1372 hashcode_t hash = 0;
1377 for (i = 0; i < size; i++)
1378 hash = HASH2 (hash, internal_hash (arr[i], depth));
1382 /* just pick five elements scattered throughout the array.
1383 A slightly better approach would be to offset by some
1384 noise factor from the points chosen below. */
1385 for (i = 0; i < 5; i++)
1386 hash = HASH2 (hash, internal_hash (arr[i*size/5], depth));
1391 /* Return a hash value for a Lisp_Object. This is for use when hashing
1392 objects with the comparison being `equal' (for `eq', you can just
1393 use the Lisp_Object itself as the hash value). You need to make a
1394 tradeoff between the speed of the hash function and how good the
1395 hashing is. In particular, the hash function needs to be FAST,
1396 so you can't just traipse down the whole tree hashing everything
1397 together. Most of the time, objects will differ in the first
1398 few elements you hash. Thus, we only go to a short depth (5)
1399 and only hash at most 5 elements out of a vector. Theoretically
1400 we could still take 5^5 time (a big big number) to compute a
1401 hash, but practically this won't ever happen. */
1404 internal_hash (Lisp_Object obj, int depth)
1410 /* no point in worrying about tail recursion, since we're not
1412 return HASH2 (internal_hash (XCAR (obj), depth + 1),
1413 internal_hash (XCDR (obj), depth + 1));
1417 return hash_string (XSTRING_DATA (obj), XSTRING_LENGTH (obj));
1421 const struct lrecord_implementation
1422 *imp = XRECORD_LHEADER_IMPLEMENTATION (obj);
1424 return imp->hash (obj, depth);
1427 return LISP_HASH (obj);
1430 DEFUN ("sxhash", Fsxhash, 1, 1, 0, /*
1431 Return a hash value for OBJECT.
1432 (equal obj1 obj2) implies (= (sxhash obj1) (sxhash obj2)).
1436 return make_int (internal_hash (object, 0));
1440 xxDEFUN ("internal-hash-value", Finternal_hash_value, 1, 1, 0, /*
1441 Hash value of OBJECT. For debugging.
1442 The value is returned as (HIGH . LOW).
1446 /* This function is pretty 32bit-centric. */
1447 hashcode_t hash = internal_hash (object, 0);
1448 return Fcons (hash >> 16, hash & 0xffff);
1453 /************************************************************************/
1454 /* initialization */
1455 /************************************************************************/
1458 syms_of_elhash (void)
1460 INIT_LRECORD_IMPLEMENTATION (hash_table);
1462 DEFSUBR (Fhash_table_p);
1463 DEFSUBR (Fmake_hash_table);
1464 DEFSUBR (Fcopy_hash_table);
1470 DEFSUBR (Fhash_table_count);
1471 DEFSUBR (Fhash_table_test);
1472 DEFSUBR (Fhash_table_size);
1473 DEFSUBR (Fhash_table_rehash_size);
1474 DEFSUBR (Fhash_table_rehash_threshold);
1475 DEFSUBR (Fhash_table_weakness);
1476 DEFSUBR (Fhash_table_type); /* obsolete */
1479 DEFSUBR (Finternal_hash_value);
1482 defsymbol (&Qhash_tablep, "hash-table-p");
1483 defsymbol (&Qhash_table, "hash-table");
1484 defsymbol (&Qhashtable, "hashtable");
1485 defsymbol (&Qweakness, "weakness");
1486 defsymbol (&Qvalue, "value");
1487 defsymbol (&Qkey_value, "key-value");
1488 defsymbol (&Qrehash_size, "rehash-size");
1489 defsymbol (&Qrehash_threshold, "rehash-threshold");
1491 defsymbol (&Qweak, "weak"); /* obsolete */
1492 defsymbol (&Qkey_weak, "key-weak"); /* obsolete */
1493 defsymbol (&Qkey_value_weak, "key-value-weak"); /* obsolete */
1494 defsymbol (&Qvalue_weak, "value-weak"); /* obsolete */
1495 defsymbol (&Qnon_weak, "non-weak"); /* obsolete */
1497 defkeyword (&Q_test, ":test");
1498 defkeyword (&Q_size, ":size");
1499 defkeyword (&Q_rehash_size, ":rehash-size");
1500 defkeyword (&Q_rehash_threshold, ":rehash-threshold");
1501 defkeyword (&Q_weakness, ":weakness");
1502 defkeyword (&Q_type, ":type"); /* obsolete */
1506 vars_of_elhash (void)
1508 /* This must NOT be staticpro'd */
1509 Vall_weak_hash_tables = Qnil;
1510 pdump_wire_list (&Vall_weak_hash_tables);