Mercurial > hg > xemacs-beta
annotate src/elhash.c @ 4779:fd98353950a4
Make my last change to elhash.c more kosher, comparing pointers not ints
2009-12-17 Aidan Kehoe <kehoea@parhasard.net>
* elhash.c (HASH_TABLE_DEFAULT_REHASH_THRESHOLD):
That last change was wrong in theory though not in practice; we're
comparing function pointers, not enumeration values.
author | Aidan Kehoe <kehoea@parhasard.net> |
---|---|
date | Thu, 17 Dec 2009 17:29:32 +0000 |
parents | 0081fd36b783 |
children | e6dec75ded0e e0db3c197671 |
rev | line source |
---|---|
428 | 1 /* Implementation of the hash table lisp object type. |
2 Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. | |
2421 | 3 Copyright (C) 1995, 1996, 2002, 2004 Ben Wing. |
428 | 4 Copyright (C) 1997 Free Software Foundation, Inc. |
5 | |
6 This file is part of XEmacs. | |
7 | |
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 | |
11 later version. | |
12 | |
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 | |
16 for more details. | |
17 | |
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. */ | |
22 | |
23 /* Synched up with: Not in FSF. */ | |
24 | |
1292 | 25 /* Author: Lost in the mists of history. At least back to Lucid 19.3, |
26 circa Sep 1992. Early hash table implementation allowed only `eq' as a | |
27 test -- other tests possible only when these objects were created from | |
28 the C code. | |
29 | |
30 Expansion to allow general `equal'-test Lisp-creatable tables, and hash | |
31 methods for the various Lisp objects in existence at the time, added | |
32 during 19.12 I think (early 1995?), by Ben Wing. | |
33 | |
34 Weak hash tables added by Jamie (maybe?) early on, perhaps around 19.6, | |
35 maybe earlier; again, only possible through the C code, and only | |
36 supported fully weak hash tables. Expansion to other kinds of weakness, | |
37 and exporting of the interface to Lisp, by Ben Wing during 19.12 | |
38 (early-mid 1995) or maybe 19.13 cycle (mid 1995). | |
39 | |
40 Expansion to full Common Lisp spec and interface, redoing of the | |
41 implementation, by Martin Buchholz, 1997? (Former hash table | |
42 implementation used "double hashing", I'm pretty sure, and was weirdly | |
43 tied into the generic hash.c code. Martin completely separated them.) | |
44 */ | |
45 | |
489 | 46 /* This file implements the hash table lisp object type. |
47 | |
504 | 48 This implementation was mostly written by Martin Buchholz in 1997. |
49 | |
50 The Lisp-level API (derived from Common Lisp) is almost completely | |
51 compatible with GNU Emacs 21, even though the implementations are | |
52 totally independent. | |
53 | |
489 | 54 The hash table technique used is "linear probing". Collisions are |
55 resolved by putting the item in the next empty place in the array | |
56 following the collision. Finding a hash entry performs a linear | |
57 search in the cluster starting at the hash value. | |
58 | |
59 On deletions from the hash table, the entries immediately following | |
60 the deleted entry are re-entered in the hash table. We do not have | |
61 a special way to mark deleted entries (known as "tombstones"). | |
62 | |
63 At the end of the hash entries ("hentries"), we leave room for an | |
64 entry that is always empty (the "sentinel"). | |
65 | |
66 The traditional literature on hash table implementation | |
67 (e.g. Knuth) suggests that too much "primary clustering" occurs | |
68 with linear probing. However, this literature was written when | |
69 locality of reference was not a factor. The discrepancy between | |
70 CPU speeds and memory speeds is increasing, and the speed of access | |
71 to memory is highly dependent on memory caches which work best when | |
72 there is high locality of data reference. Random access to memory | |
73 is up to 20 times as expensive as access to the nearest address | |
74 (and getting worse). So linear probing makes sense. | |
75 | |
76 But the representation doesn't actually matter that much with the | |
77 current elisp engine. Funcall is sufficiently slow that the choice | |
78 of hash table implementation is noise. */ | |
79 | |
428 | 80 #include <config.h> |
81 #include "lisp.h" | |
82 #include "bytecode.h" | |
83 #include "elhash.h" | |
489 | 84 #include "opaque.h" |
428 | 85 |
86 Lisp_Object Qhash_tablep; | |
87 static Lisp_Object Qhashtable, Qhash_table; | |
442 | 88 static Lisp_Object Qweakness, Qvalue, Qkey_or_value, Qkey_and_value; |
428 | 89 static Lisp_Object Vall_weak_hash_tables; |
90 static Lisp_Object Qrehash_size, Qrehash_threshold; | |
91 static Lisp_Object Q_size, Q_test, Q_weakness, Q_rehash_size, Q_rehash_threshold; | |
92 | |
93 /* obsolete as of 19990901 in xemacs-21.2 */ | |
442 | 94 static Lisp_Object Qweak, Qkey_weak, Qvalue_weak, Qkey_or_value_weak; |
95 static Lisp_Object Qnon_weak, Q_type; | |
428 | 96 |
97 struct Lisp_Hash_Table | |
98 { | |
3017 | 99 struct LCRECORD_HEADER header; |
665 | 100 Elemcount size; |
101 Elemcount count; | |
102 Elemcount rehash_count; | |
428 | 103 double rehash_size; |
104 double rehash_threshold; | |
665 | 105 Elemcount golden_ratio; |
428 | 106 hash_table_hash_function_t hash_function; |
107 hash_table_test_function_t test_function; | |
1204 | 108 htentry *hentries; |
428 | 109 enum hash_table_weakness weakness; |
110 Lisp_Object next_weak; /* Used to chain together all of the weak | |
111 hash tables. Don't mark through this. */ | |
112 }; | |
113 | |
1204 | 114 #define CLEAR_HTENTRY(htentry) \ |
115 ((*(EMACS_UINT*)(&((htentry)->key))) = 0, \ | |
116 (*(EMACS_UINT*)(&((htentry)->value))) = 0) | |
428 | 117 |
118 #define HASH_TABLE_DEFAULT_SIZE 16 | |
119 #define HASH_TABLE_DEFAULT_REHASH_SIZE 1.3 | |
120 #define HASH_TABLE_MIN_SIZE 10 | |
4778
0081fd36b783
Cast enumerations to int before comparing them for the sake of VC++.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4777
diff
changeset
|
121 #define HASH_TABLE_DEFAULT_REHASH_THRESHOLD(size, test_function) \ |
4779
fd98353950a4
Make my last change to elhash.c more kosher, comparing pointers not ints
Aidan Kehoe <kehoea@parhasard.net>
parents:
4778
diff
changeset
|
122 (((size) > 4096 && NULL == (test_function)) ? 0.7 : 0.6) |
428 | 123 |
665 | 124 #define HASHCODE(key, ht) \ |
444 | 125 ((((ht)->hash_function ? (ht)->hash_function (key) : LISP_HASH (key)) \ |
126 * (ht)->golden_ratio) \ | |
127 % (ht)->size) | |
428 | 128 |
129 #define KEYS_EQUAL_P(key1, key2, testfun) \ | |
434 | 130 (EQ (key1, key2) || ((testfun) && (testfun) (key1, key2))) |
428 | 131 |
132 #define LINEAR_PROBING_LOOP(probe, entries, size) \ | |
133 for (; \ | |
1204 | 134 !HTENTRY_CLEAR_P (probe) || \ |
428 | 135 (probe == entries + size ? \ |
1204 | 136 (probe = entries, !HTENTRY_CLEAR_P (probe)) : 0); \ |
428 | 137 probe++) |
138 | |
800 | 139 #ifdef ERROR_CHECK_STRUCTURES |
428 | 140 static void |
141 check_hash_table_invariants (Lisp_Hash_Table *ht) | |
142 { | |
143 assert (ht->count < ht->size); | |
144 assert (ht->count <= ht->rehash_count); | |
145 assert (ht->rehash_count < ht->size); | |
146 assert ((double) ht->count * ht->rehash_threshold - 1 <= (double) ht->rehash_count); | |
1204 | 147 assert (HTENTRY_CLEAR_P (ht->hentries + ht->size)); |
428 | 148 } |
149 #else | |
150 #define check_hash_table_invariants(ht) | |
151 #endif | |
152 | |
153 /* Return a suitable size for a hash table, with at least SIZE slots. */ | |
665 | 154 static Elemcount |
155 hash_table_size (Elemcount requested_size) | |
428 | 156 { |
157 /* Return some prime near, but greater than or equal to, SIZE. | |
158 Decades from the time of writing, someone will have a system large | |
159 enough that the list below will be too short... */ | |
665 | 160 static const Elemcount primes [] = |
428 | 161 { |
162 19, 29, 41, 59, 79, 107, 149, 197, 263, 347, 457, 599, 787, 1031, | |
163 1361, 1777, 2333, 3037, 3967, 5167, 6719, 8737, 11369, 14783, | |
164 19219, 24989, 32491, 42257, 54941, 71429, 92861, 120721, 156941, | |
165 204047, 265271, 344857, 448321, 582821, 757693, 985003, 1280519, | |
166 1664681, 2164111, 2813353, 3657361, 4754591, 6180989, 8035301, | |
167 10445899, 13579681, 17653589, 22949669, 29834603, 38784989, | |
168 50420551, 65546729, 85210757, 110774011, 144006217, 187208107, | |
169 243370577, 316381771, 411296309, 534685237, 695090819, 903618083, | |
647 | 170 1174703521, 1527114613, 1985248999 /* , 2580823717UL, 3355070839UL */ |
428 | 171 }; |
172 /* We've heard of binary search. */ | |
173 int low, high; | |
174 for (low = 0, high = countof (primes) - 1; high - low > 1;) | |
175 { | |
176 /* Loop Invariant: size < primes [high] */ | |
177 int mid = (low + high) / 2; | |
178 if (primes [mid] < requested_size) | |
179 low = mid; | |
180 else | |
181 high = mid; | |
182 } | |
183 return primes [high]; | |
184 } | |
185 | |
186 | |
187 #if 0 /* I don't think these are needed any more. | |
188 If using the general lisp_object_equal_*() functions | |
189 causes efficiency problems, these can be resurrected. --ben */ | |
190 /* equality and hash functions for Lisp strings */ | |
191 int | |
192 lisp_string_equal (Lisp_Object str1, Lisp_Object str2) | |
193 { | |
194 /* This is wrong anyway. You can't use strcmp() on Lisp strings, | |
195 because they can contain zero characters. */ | |
196 return !strcmp ((char *) XSTRING_DATA (str1), (char *) XSTRING_DATA (str2)); | |
197 } | |
198 | |
665 | 199 static Hashcode |
428 | 200 lisp_string_hash (Lisp_Object obj) |
201 { | |
202 return hash_string (XSTRING_DATA (str), XSTRING_LENGTH (str)); | |
203 } | |
204 | |
205 #endif /* 0 */ | |
206 | |
207 static int | |
208 lisp_object_eql_equal (Lisp_Object obj1, Lisp_Object obj2) | |
209 { | |
210 return EQ (obj1, obj2) || (FLOATP (obj1) && internal_equal (obj1, obj2, 0)); | |
211 } | |
212 | |
665 | 213 static Hashcode |
428 | 214 lisp_object_eql_hash (Lisp_Object obj) |
215 { | |
216 return FLOATP (obj) ? internal_hash (obj, 0) : LISP_HASH (obj); | |
217 } | |
218 | |
219 static int | |
220 lisp_object_equal_equal (Lisp_Object obj1, Lisp_Object obj2) | |
221 { | |
222 return internal_equal (obj1, obj2, 0); | |
223 } | |
224 | |
665 | 225 static Hashcode |
428 | 226 lisp_object_equal_hash (Lisp_Object obj) |
227 { | |
228 return internal_hash (obj, 0); | |
229 } | |
230 | |
231 | |
232 static Lisp_Object | |
233 mark_hash_table (Lisp_Object obj) | |
234 { | |
235 Lisp_Hash_Table *ht = XHASH_TABLE (obj); | |
236 | |
237 /* If the hash table is weak, we don't want to mark the keys and | |
238 values (we scan over them after everything else has been marked, | |
239 and mark or remove them as necessary). */ | |
240 if (ht->weakness == HASH_TABLE_NON_WEAK) | |
241 { | |
1204 | 242 htentry *e, *sentinel; |
428 | 243 |
244 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++) | |
1204 | 245 if (!HTENTRY_CLEAR_P (e)) |
428 | 246 { |
247 mark_object (e->key); | |
248 mark_object (e->value); | |
249 } | |
250 } | |
251 return Qnil; | |
252 } | |
253 | |
254 /* Equality of hash tables. Two hash tables are equal when they are of | |
255 the same weakness and test function, they have the same number of | |
256 elements, and for each key in the hash table, the values are `equal'. | |
257 | |
258 This is similar to Common Lisp `equalp' of hash tables, with the | |
259 difference that CL requires the keys to be compared with the test | |
260 function, which we don't do. Doing that would require consing, and | |
261 consing is a bad idea in `equal'. Anyway, our method should provide | |
262 the same result -- if the keys are not equal according to the test | |
263 function, then Fgethash() in hash_table_equal_mapper() will fail. */ | |
264 static int | |
265 hash_table_equal (Lisp_Object hash_table1, Lisp_Object hash_table2, int depth) | |
266 { | |
267 Lisp_Hash_Table *ht1 = XHASH_TABLE (hash_table1); | |
268 Lisp_Hash_Table *ht2 = XHASH_TABLE (hash_table2); | |
1204 | 269 htentry *e, *sentinel; |
428 | 270 |
271 if ((ht1->test_function != ht2->test_function) || | |
272 (ht1->weakness != ht2->weakness) || | |
273 (ht1->count != ht2->count)) | |
274 return 0; | |
275 | |
276 depth++; | |
277 | |
278 for (e = ht1->hentries, sentinel = e + ht1->size; e < sentinel; e++) | |
1204 | 279 if (!HTENTRY_CLEAR_P (e)) |
428 | 280 /* Look up the key in the other hash table, and compare the values. */ |
281 { | |
282 Lisp_Object value_in_other = Fgethash (e->key, hash_table2, Qunbound); | |
283 if (UNBOUNDP (value_in_other) || | |
284 !internal_equal (e->value, value_in_other, depth)) | |
285 return 0; /* Give up */ | |
286 } | |
287 | |
288 return 1; | |
289 } | |
442 | 290 |
291 /* This is not a great hash function, but it _is_ correct and fast. | |
292 Examining all entries is too expensive, and examining a random | |
293 subset does not yield a correct hash function. */ | |
665 | 294 static Hashcode |
2286 | 295 hash_table_hash (Lisp_Object hash_table, int UNUSED (depth)) |
442 | 296 { |
297 return XHASH_TABLE (hash_table)->count; | |
298 } | |
299 | |
428 | 300 |
301 /* Printing hash tables. | |
302 | |
303 This is non-trivial, because we use a readable structure-style | |
304 syntax for hash tables. This means that a typical hash table will be | |
305 readably printed in the form of: | |
306 | |
307 #s(hash-table size 2 data (key1 value1 key2 value2)) | |
308 | |
309 The supported hash table structure keywords and their values are: | |
310 `test' (eql (or nil), eq or equal) | |
311 `size' (a natnum or nil) | |
312 `rehash-size' (a float) | |
313 `rehash-threshold' (a float) | |
442 | 314 `weakness' (nil, key, value, key-and-value, or key-or-value) |
428 | 315 `data' (a list) |
316 | |
430 | 317 If `print-readably' is nil, then a simpler syntax is used, for example |
428 | 318 |
319 #<hash-table size 2/13 data (key1 value1 key2 value2) 0x874d> | |
320 | |
321 The data is truncated to four pairs, and the rest is shown with | |
322 `...'. This printer does not cons. */ | |
323 | |
324 | |
325 /* Print the data of the hash table. This maps through a Lisp | |
326 hash table and prints key/value pairs using PRINTCHARFUN. */ | |
327 static void | |
328 print_hash_table_data (Lisp_Hash_Table *ht, Lisp_Object printcharfun) | |
329 { | |
330 int count = 0; | |
1204 | 331 htentry *e, *sentinel; |
428 | 332 |
826 | 333 write_c_string (printcharfun, " data ("); |
428 | 334 |
335 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++) | |
1204 | 336 if (!HTENTRY_CLEAR_P (e)) |
428 | 337 { |
338 if (count > 0) | |
826 | 339 write_c_string (printcharfun, " "); |
428 | 340 if (!print_readably && count > 3) |
341 { | |
826 | 342 write_c_string (printcharfun, "..."); |
428 | 343 break; |
344 } | |
345 print_internal (e->key, printcharfun, 1); | |
800 | 346 write_fmt_string_lisp (printcharfun, " %S", 1, e->value); |
428 | 347 count++; |
348 } | |
349 | |
826 | 350 write_c_string (printcharfun, ")"); |
428 | 351 } |
352 | |
353 static void | |
2286 | 354 print_hash_table (Lisp_Object obj, Lisp_Object printcharfun, |
355 int UNUSED (escapeflag)) | |
428 | 356 { |
357 Lisp_Hash_Table *ht = XHASH_TABLE (obj); | |
4777
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
358 Ascbyte pigbuf[350]; |
428 | 359 |
826 | 360 write_c_string (printcharfun, |
361 print_readably ? "#s(hash-table" : "#<hash-table"); | |
428 | 362 |
363 /* These checks have a kludgy look to them, but they are safe. | |
364 Due to nature of hashing, you cannot use arbitrary | |
365 test functions anyway. */ | |
366 if (!ht->test_function) | |
826 | 367 write_c_string (printcharfun, " test eq"); |
428 | 368 else if (ht->test_function == lisp_object_equal_equal) |
826 | 369 write_c_string (printcharfun, " test equal"); |
428 | 370 else if (ht->test_function == lisp_object_eql_equal) |
371 DO_NOTHING; | |
372 else | |
2500 | 373 ABORT (); |
428 | 374 |
375 if (ht->count || !print_readably) | |
376 { | |
377 if (print_readably) | |
800 | 378 write_fmt_string (printcharfun, " size %ld", (long) ht->count); |
428 | 379 else |
800 | 380 write_fmt_string (printcharfun, " size %ld/%ld", (long) ht->count, |
381 (long) ht->size); | |
428 | 382 } |
383 | |
384 if (ht->weakness != HASH_TABLE_NON_WEAK) | |
385 { | |
800 | 386 write_fmt_string |
387 (printcharfun, " weakness %s", | |
388 (ht->weakness == HASH_TABLE_WEAK ? "key-and-value" : | |
389 ht->weakness == HASH_TABLE_KEY_WEAK ? "key" : | |
390 ht->weakness == HASH_TABLE_VALUE_WEAK ? "value" : | |
391 ht->weakness == HASH_TABLE_KEY_VALUE_WEAK ? "key-or-value" : | |
392 "you-d-better-not-see-this")); | |
428 | 393 } |
394 | |
4777
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
395 if (ht->rehash_size != HASH_TABLE_DEFAULT_REHASH_SIZE) |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
396 { |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
397 float_to_string (pigbuf, ht->rehash_size); |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
398 write_fmt_string (printcharfun, " rehash-size %s", pigbuf); |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
399 } |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
400 |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
401 if (ht->rehash_threshold |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
402 != HASH_TABLE_DEFAULT_REHASH_THRESHOLD (ht->size, |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
403 ht->test_function)) |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
404 { |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
405 float_to_string (pigbuf, ht->rehash_threshold); |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
406 write_fmt_string (printcharfun, " rehash-threshold %s", pigbuf); |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
407 } |
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
408 |
428 | 409 if (ht->count) |
410 print_hash_table_data (ht, printcharfun); | |
411 | |
412 if (print_readably) | |
826 | 413 write_c_string (printcharfun, ")"); |
428 | 414 else |
2421 | 415 write_fmt_string (printcharfun, " 0x%x>", ht->header.uid); |
428 | 416 } |
417 | |
4117 | 418 #ifndef NEW_GC |
428 | 419 static void |
4117 | 420 free_hentries (htentry *hentries, |
2333 | 421 #ifdef ERROR_CHECK_STRUCTURES |
422 size_t size | |
4117 | 423 #else /* not ERROR_CHECK_STRUCTURES) */ |
2333 | 424 size_t UNUSED (size) |
4117 | 425 #endif /* not ERROR_CHECK_STRUCTURES) */ |
2333 | 426 ) |
489 | 427 { |
800 | 428 #ifdef ERROR_CHECK_STRUCTURES |
489 | 429 /* Ensure a crash if other code uses the discarded entries afterwards. */ |
1204 | 430 htentry *e, *sentinel; |
489 | 431 |
432 for (e = hentries, sentinel = e + size; e < sentinel; e++) | |
1204 | 433 * (unsigned long *) e = 0xdeadbeef; /* -559038737 base 10 */ |
489 | 434 #endif |
435 | |
436 if (!DUMPEDP (hentries)) | |
1726 | 437 xfree (hentries, htentry *); |
489 | 438 } |
439 | |
440 static void | |
428 | 441 finalize_hash_table (void *header, int for_disksave) |
442 { | |
443 if (!for_disksave) | |
444 { | |
445 Lisp_Hash_Table *ht = (Lisp_Hash_Table *) header; | |
489 | 446 free_hentries (ht->hentries, ht->size); |
428 | 447 ht->hentries = 0; |
448 } | |
449 } | |
3263 | 450 #endif /* not NEW_GC */ |
428 | 451 |
1204 | 452 static const struct memory_description htentry_description_1[] = { |
453 { XD_LISP_OBJECT, offsetof (htentry, key) }, | |
454 { XD_LISP_OBJECT, offsetof (htentry, value) }, | |
428 | 455 { XD_END } |
456 }; | |
457 | |
1204 | 458 static const struct sized_memory_description htentry_description = { |
459 sizeof (htentry), | |
460 htentry_description_1 | |
428 | 461 }; |
462 | |
3092 | 463 #ifdef NEW_GC |
464 static const struct memory_description htentry_weak_description_1[] = { | |
465 { XD_LISP_OBJECT, offsetof (htentry, key), 0, { 0 }, XD_FLAG_NO_KKCC}, | |
466 { XD_LISP_OBJECT, offsetof (htentry, value), 0, { 0 }, XD_FLAG_NO_KKCC}, | |
467 { XD_END } | |
468 }; | |
469 | |
470 static const struct sized_memory_description htentry_weak_description = { | |
471 sizeof (htentry), | |
472 htentry_weak_description_1 | |
473 }; | |
474 | |
475 DEFINE_LRECORD_IMPLEMENTATION ("hash-table-entry", hash_table_entry, | |
476 1, /*dumpable-flag*/ | |
477 0, 0, 0, 0, 0, | |
478 htentry_description_1, | |
479 Lisp_Hash_Table_Entry); | |
480 #endif /* NEW_GC */ | |
481 | |
1204 | 482 static const struct memory_description htentry_union_description_1[] = { |
483 /* Note: XD_INDIRECT in this table refers to the surrounding table, | |
484 and so this will work. */ | |
3092 | 485 #ifdef NEW_GC |
486 { XD_LISP_OBJECT_BLOCK_PTR, HASH_TABLE_NON_WEAK, | |
487 XD_INDIRECT (0, 1), { &htentry_description } }, | |
488 { XD_LISP_OBJECT_BLOCK_PTR, 0, XD_INDIRECT (0, 1), | |
489 { &htentry_weak_description }, XD_FLAG_UNION_DEFAULT_ENTRY }, | |
490 #else /* not NEW_GC */ | |
2367 | 491 { XD_BLOCK_PTR, HASH_TABLE_NON_WEAK, XD_INDIRECT (0, 1), |
2551 | 492 { &htentry_description } }, |
493 { XD_BLOCK_PTR, 0, XD_INDIRECT (0, 1), { &htentry_description }, | |
1204 | 494 XD_FLAG_UNION_DEFAULT_ENTRY | XD_FLAG_NO_KKCC }, |
3092 | 495 #endif /* not NEW_GC */ |
1204 | 496 { XD_END } |
497 }; | |
498 | |
499 static const struct sized_memory_description htentry_union_description = { | |
500 sizeof (htentry *), | |
501 htentry_union_description_1 | |
502 }; | |
503 | |
504 const struct memory_description hash_table_description[] = { | |
505 { XD_ELEMCOUNT, offsetof (Lisp_Hash_Table, size) }, | |
506 { XD_INT, offsetof (Lisp_Hash_Table, weakness) }, | |
507 { XD_UNION, offsetof (Lisp_Hash_Table, hentries), XD_INDIRECT (1, 0), | |
2551 | 508 { &htentry_union_description } }, |
440 | 509 { XD_LO_LINK, offsetof (Lisp_Hash_Table, next_weak) }, |
428 | 510 { XD_END } |
511 }; | |
512 | |
3263 | 513 #ifdef NEW_GC |
514 DEFINE_LRECORD_IMPLEMENTATION ("hash-table", hash_table, | |
515 1, /*dumpable-flag*/ | |
516 mark_hash_table, print_hash_table, | |
517 0, hash_table_equal, hash_table_hash, | |
518 hash_table_description, | |
519 Lisp_Hash_Table); | |
520 #else /* not NEW_GC */ | |
934 | 521 DEFINE_LRECORD_IMPLEMENTATION ("hash-table", hash_table, |
522 1, /*dumpable-flag*/ | |
523 mark_hash_table, print_hash_table, | |
524 finalize_hash_table, | |
525 hash_table_equal, hash_table_hash, | |
526 hash_table_description, | |
527 Lisp_Hash_Table); | |
3263 | 528 #endif /* not NEW_GC */ |
428 | 529 |
530 static Lisp_Hash_Table * | |
531 xhash_table (Lisp_Object hash_table) | |
532 { | |
1123 | 533 /* #### What's going on here? Why the gc_in_progress check? */ |
428 | 534 if (!gc_in_progress) |
535 CHECK_HASH_TABLE (hash_table); | |
536 check_hash_table_invariants (XHASH_TABLE (hash_table)); | |
537 return XHASH_TABLE (hash_table); | |
538 } | |
539 | |
540 | |
541 /************************************************************************/ | |
542 /* Creation of Hash Tables */ | |
543 /************************************************************************/ | |
544 | |
545 /* Creation of hash tables, without error-checking. */ | |
546 static void | |
547 compute_hash_table_derived_values (Lisp_Hash_Table *ht) | |
548 { | |
665 | 549 ht->rehash_count = (Elemcount) |
438 | 550 ((double) ht->size * ht->rehash_threshold); |
665 | 551 ht->golden_ratio = (Elemcount) |
428 | 552 ((double) ht->size * (.6180339887 / (double) sizeof (Lisp_Object))); |
553 } | |
554 | |
555 Lisp_Object | |
450 | 556 make_standard_lisp_hash_table (enum hash_table_test test, |
665 | 557 Elemcount size, |
450 | 558 double rehash_size, |
559 double rehash_threshold, | |
560 enum hash_table_weakness weakness) | |
561 { | |
462 | 562 hash_table_hash_function_t hash_function = 0; |
450 | 563 hash_table_test_function_t test_function = 0; |
564 | |
565 switch (test) | |
566 { | |
567 case HASH_TABLE_EQ: | |
568 test_function = 0; | |
569 hash_function = 0; | |
570 break; | |
571 | |
572 case HASH_TABLE_EQL: | |
573 test_function = lisp_object_eql_equal; | |
574 hash_function = lisp_object_eql_hash; | |
575 break; | |
576 | |
577 case HASH_TABLE_EQUAL: | |
578 test_function = lisp_object_equal_equal; | |
579 hash_function = lisp_object_equal_hash; | |
580 break; | |
581 | |
582 default: | |
2500 | 583 ABORT (); |
450 | 584 } |
585 | |
586 return make_general_lisp_hash_table (hash_function, test_function, | |
587 size, rehash_size, rehash_threshold, | |
588 weakness); | |
589 } | |
590 | |
591 Lisp_Object | |
592 make_general_lisp_hash_table (hash_table_hash_function_t hash_function, | |
593 hash_table_test_function_t test_function, | |
665 | 594 Elemcount size, |
428 | 595 double rehash_size, |
596 double rehash_threshold, | |
597 enum hash_table_weakness weakness) | |
598 { | |
599 Lisp_Object hash_table; | |
3017 | 600 Lisp_Hash_Table *ht = ALLOC_LCRECORD_TYPE (Lisp_Hash_Table, &lrecord_hash_table); |
428 | 601 |
450 | 602 ht->test_function = test_function; |
603 ht->hash_function = hash_function; | |
438 | 604 ht->weakness = weakness; |
605 | |
606 ht->rehash_size = | |
607 rehash_size > 1.0 ? rehash_size : HASH_TABLE_DEFAULT_REHASH_SIZE; | |
608 | |
609 ht->rehash_threshold = | |
610 rehash_threshold > 0.0 ? rehash_threshold : | |
4777
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
611 HASH_TABLE_DEFAULT_REHASH_THRESHOLD (size, ht->test_function); |
438 | 612 |
428 | 613 if (size < HASH_TABLE_MIN_SIZE) |
614 size = HASH_TABLE_MIN_SIZE; | |
665 | 615 ht->size = hash_table_size ((Elemcount) (((double) size / ht->rehash_threshold) |
438 | 616 + 1.0)); |
428 | 617 ht->count = 0; |
438 | 618 |
428 | 619 compute_hash_table_derived_values (ht); |
620 | |
1204 | 621 /* We leave room for one never-occupied sentinel htentry at the end. */ |
3092 | 622 #ifdef NEW_GC |
623 ht->hentries = (htentry *) alloc_lrecord_array (sizeof (htentry), | |
624 ht->size + 1, | |
625 &lrecord_hash_table_entry); | |
626 #else /* not NEW_GC */ | |
1204 | 627 ht->hentries = xnew_array_and_zero (htentry, ht->size + 1); |
3092 | 628 #endif /* not NEW_GC */ |
428 | 629 |
793 | 630 hash_table = wrap_hash_table (ht); |
428 | 631 |
632 if (weakness == HASH_TABLE_NON_WEAK) | |
633 ht->next_weak = Qunbound; | |
634 else | |
635 ht->next_weak = Vall_weak_hash_tables, Vall_weak_hash_tables = hash_table; | |
636 | |
637 return hash_table; | |
638 } | |
639 | |
640 Lisp_Object | |
665 | 641 make_lisp_hash_table (Elemcount size, |
428 | 642 enum hash_table_weakness weakness, |
643 enum hash_table_test test) | |
644 { | |
450 | 645 return make_standard_lisp_hash_table (test, size, -1.0, -1.0, weakness); |
428 | 646 } |
647 | |
648 /* Pretty reading of hash tables. | |
649 | |
650 Here we use the existing structures mechanism (which is, | |
651 unfortunately, pretty cumbersome) for validating and instantiating | |
652 the hash tables. The idea is that the side-effect of reading a | |
653 #s(hash-table PLIST) object is creation of a hash table with desired | |
654 properties, and that the hash table is returned. */ | |
655 | |
656 /* Validation functions: each keyword provides its own validation | |
657 function. The errors should maybe be continuable, but it is | |
658 unclear how this would cope with ERRB. */ | |
659 static int | |
2286 | 660 hash_table_size_validate (Lisp_Object UNUSED (keyword), Lisp_Object value, |
661 Error_Behavior errb) | |
428 | 662 { |
663 if (NATNUMP (value)) | |
664 return 1; | |
665 | |
563 | 666 maybe_signal_error_1 (Qwrong_type_argument, list2 (Qnatnump, value), |
2286 | 667 Qhash_table, errb); |
428 | 668 return 0; |
669 } | |
670 | |
665 | 671 static Elemcount |
428 | 672 decode_hash_table_size (Lisp_Object obj) |
673 { | |
674 return NILP (obj) ? HASH_TABLE_DEFAULT_SIZE : XINT (obj); | |
675 } | |
676 | |
677 static int | |
2286 | 678 hash_table_weakness_validate (Lisp_Object UNUSED (keyword), Lisp_Object value, |
578 | 679 Error_Behavior errb) |
428 | 680 { |
442 | 681 if (EQ (value, Qnil)) return 1; |
682 if (EQ (value, Qt)) return 1; | |
683 if (EQ (value, Qkey)) return 1; | |
684 if (EQ (value, Qkey_and_value)) return 1; | |
685 if (EQ (value, Qkey_or_value)) return 1; | |
686 if (EQ (value, Qvalue)) return 1; | |
428 | 687 |
688 /* Following values are obsolete as of 19990901 in xemacs-21.2 */ | |
442 | 689 if (EQ (value, Qnon_weak)) return 1; |
690 if (EQ (value, Qweak)) return 1; | |
691 if (EQ (value, Qkey_weak)) return 1; | |
692 if (EQ (value, Qkey_or_value_weak)) return 1; | |
693 if (EQ (value, Qvalue_weak)) return 1; | |
428 | 694 |
563 | 695 maybe_invalid_constant ("Invalid hash table weakness", |
428 | 696 value, Qhash_table, errb); |
697 return 0; | |
698 } | |
699 | |
700 static enum hash_table_weakness | |
701 decode_hash_table_weakness (Lisp_Object obj) | |
702 { | |
442 | 703 if (EQ (obj, Qnil)) return HASH_TABLE_NON_WEAK; |
704 if (EQ (obj, Qt)) return HASH_TABLE_WEAK; | |
705 if (EQ (obj, Qkey_and_value)) return HASH_TABLE_WEAK; | |
706 if (EQ (obj, Qkey)) return HASH_TABLE_KEY_WEAK; | |
707 if (EQ (obj, Qkey_or_value)) return HASH_TABLE_KEY_VALUE_WEAK; | |
708 if (EQ (obj, Qvalue)) return HASH_TABLE_VALUE_WEAK; | |
428 | 709 |
710 /* Following values are obsolete as of 19990901 in xemacs-21.2 */ | |
442 | 711 if (EQ (obj, Qnon_weak)) return HASH_TABLE_NON_WEAK; |
712 if (EQ (obj, Qweak)) return HASH_TABLE_WEAK; | |
713 if (EQ (obj, Qkey_weak)) return HASH_TABLE_KEY_WEAK; | |
714 if (EQ (obj, Qkey_or_value_weak)) return HASH_TABLE_KEY_VALUE_WEAK; | |
715 if (EQ (obj, Qvalue_weak)) return HASH_TABLE_VALUE_WEAK; | |
428 | 716 |
563 | 717 invalid_constant ("Invalid hash table weakness", obj); |
1204 | 718 RETURN_NOT_REACHED (HASH_TABLE_NON_WEAK); |
428 | 719 } |
720 | |
721 static int | |
2286 | 722 hash_table_test_validate (Lisp_Object UNUSED (keyword), Lisp_Object value, |
723 Error_Behavior errb) | |
428 | 724 { |
725 if (EQ (value, Qnil)) return 1; | |
726 if (EQ (value, Qeq)) return 1; | |
727 if (EQ (value, Qequal)) return 1; | |
728 if (EQ (value, Qeql)) return 1; | |
729 | |
563 | 730 maybe_invalid_constant ("Invalid hash table test", |
2286 | 731 value, Qhash_table, errb); |
428 | 732 return 0; |
733 } | |
734 | |
735 static enum hash_table_test | |
736 decode_hash_table_test (Lisp_Object obj) | |
737 { | |
738 if (EQ (obj, Qnil)) return HASH_TABLE_EQL; | |
739 if (EQ (obj, Qeq)) return HASH_TABLE_EQ; | |
740 if (EQ (obj, Qequal)) return HASH_TABLE_EQUAL; | |
741 if (EQ (obj, Qeql)) return HASH_TABLE_EQL; | |
742 | |
563 | 743 invalid_constant ("Invalid hash table test", obj); |
1204 | 744 RETURN_NOT_REACHED (HASH_TABLE_EQ); |
428 | 745 } |
746 | |
747 static int | |
2286 | 748 hash_table_rehash_size_validate (Lisp_Object UNUSED (keyword), |
749 Lisp_Object value, Error_Behavior errb) | |
428 | 750 { |
751 if (!FLOATP (value)) | |
752 { | |
563 | 753 maybe_signal_error_1 (Qwrong_type_argument, list2 (Qfloatp, value), |
428 | 754 Qhash_table, errb); |
755 return 0; | |
756 } | |
757 | |
758 { | |
759 double rehash_size = XFLOAT_DATA (value); | |
760 if (rehash_size <= 1.0) | |
761 { | |
563 | 762 maybe_invalid_argument |
428 | 763 ("Hash table rehash size must be greater than 1.0", |
764 value, Qhash_table, errb); | |
765 return 0; | |
766 } | |
767 } | |
768 | |
769 return 1; | |
770 } | |
771 | |
772 static double | |
773 decode_hash_table_rehash_size (Lisp_Object rehash_size) | |
774 { | |
4585
871eb054b34a
Document non-obvious usages.
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4410
diff
changeset
|
775 /* -1.0 signals make_general_lisp_hash_table to use the default. */ |
428 | 776 return NILP (rehash_size) ? -1.0 : XFLOAT_DATA (rehash_size); |
777 } | |
778 | |
779 static int | |
2286 | 780 hash_table_rehash_threshold_validate (Lisp_Object UNUSED (keyword), |
781 Lisp_Object value, Error_Behavior errb) | |
428 | 782 { |
783 if (!FLOATP (value)) | |
784 { | |
563 | 785 maybe_signal_error_1 (Qwrong_type_argument, list2 (Qfloatp, value), |
428 | 786 Qhash_table, errb); |
787 return 0; | |
788 } | |
789 | |
790 { | |
791 double rehash_threshold = XFLOAT_DATA (value); | |
792 if (rehash_threshold <= 0.0 || rehash_threshold >= 1.0) | |
793 { | |
563 | 794 maybe_invalid_argument |
428 | 795 ("Hash table rehash threshold must be between 0.0 and 1.0", |
796 value, Qhash_table, errb); | |
797 return 0; | |
798 } | |
799 } | |
800 | |
801 return 1; | |
802 } | |
803 | |
804 static double | |
805 decode_hash_table_rehash_threshold (Lisp_Object rehash_threshold) | |
806 { | |
4585
871eb054b34a
Document non-obvious usages.
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4410
diff
changeset
|
807 /* -1.0 signals make_general_lisp_hash_table to use the default. */ |
428 | 808 return NILP (rehash_threshold) ? -1.0 : XFLOAT_DATA (rehash_threshold); |
809 } | |
810 | |
811 static int | |
2286 | 812 hash_table_data_validate (Lisp_Object UNUSED (keyword), Lisp_Object value, |
813 Error_Behavior errb) | |
428 | 814 { |
815 int len; | |
816 | |
4585
871eb054b34a
Document non-obvious usages.
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4410
diff
changeset
|
817 /* Check for improper lists while getting length. */ |
428 | 818 GET_EXTERNAL_LIST_LENGTH (value, len); |
819 | |
820 if (len & 1) | |
821 { | |
563 | 822 maybe_sferror |
428 | 823 ("Hash table data must have alternating key/value pairs", |
824 value, Qhash_table, errb); | |
825 return 0; | |
826 } | |
4585
871eb054b34a
Document non-obvious usages.
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4410
diff
changeset
|
827 |
428 | 828 return 1; |
829 } | |
830 | |
831 /* The actual instantiation of a hash table. This does practically no | |
832 error checking, because it relies on the fact that the paranoid | |
833 functions above have error-checked everything to the last details. | |
834 If this assumption is wrong, we will get a crash immediately (with | |
835 error-checking compiled in), and we'll know if there is a bug in | |
836 the structure mechanism. So there. */ | |
837 static Lisp_Object | |
838 hash_table_instantiate (Lisp_Object plist) | |
839 { | |
840 Lisp_Object hash_table; | |
841 Lisp_Object test = Qnil; | |
842 Lisp_Object size = Qnil; | |
843 Lisp_Object rehash_size = Qnil; | |
844 Lisp_Object rehash_threshold = Qnil; | |
845 Lisp_Object weakness = Qnil; | |
846 Lisp_Object data = Qnil; | |
847 | |
2421 | 848 PROPERTY_LIST_LOOP_3 (key, value, plist) |
428 | 849 { |
850 if (EQ (key, Qtest)) test = value; | |
851 else if (EQ (key, Qsize)) size = value; | |
852 else if (EQ (key, Qrehash_size)) rehash_size = value; | |
853 else if (EQ (key, Qrehash_threshold)) rehash_threshold = value; | |
854 else if (EQ (key, Qweakness)) weakness = value; | |
855 else if (EQ (key, Qdata)) data = value; | |
856 else if (EQ (key, Qtype))/*obsolete*/ weakness = value; | |
857 else | |
2500 | 858 ABORT (); |
428 | 859 } |
860 | |
861 /* Create the hash table. */ | |
450 | 862 hash_table = make_standard_lisp_hash_table |
428 | 863 (decode_hash_table_test (test), |
864 decode_hash_table_size (size), | |
865 decode_hash_table_rehash_size (rehash_size), | |
866 decode_hash_table_rehash_threshold (rehash_threshold), | |
867 decode_hash_table_weakness (weakness)); | |
868 | |
869 /* I'm not sure whether this can GC, but better safe than sorry. */ | |
870 { | |
871 struct gcpro gcpro1; | |
872 GCPRO1 (hash_table); | |
873 | |
874 /* And fill it with data. */ | |
875 while (!NILP (data)) | |
876 { | |
877 Lisp_Object key, value; | |
878 key = XCAR (data); data = XCDR (data); | |
879 value = XCAR (data); data = XCDR (data); | |
880 Fputhash (key, value, hash_table); | |
881 } | |
882 UNGCPRO; | |
883 } | |
884 | |
885 return hash_table; | |
886 } | |
887 | |
888 static void | |
889 structure_type_create_hash_table_structure_name (Lisp_Object structure_name) | |
890 { | |
891 struct structure_type *st; | |
892 | |
893 st = define_structure_type (structure_name, 0, hash_table_instantiate); | |
894 define_structure_type_keyword (st, Qtest, hash_table_test_validate); | |
895 define_structure_type_keyword (st, Qsize, hash_table_size_validate); | |
896 define_structure_type_keyword (st, Qrehash_size, hash_table_rehash_size_validate); | |
897 define_structure_type_keyword (st, Qrehash_threshold, hash_table_rehash_threshold_validate); | |
898 define_structure_type_keyword (st, Qweakness, hash_table_weakness_validate); | |
899 define_structure_type_keyword (st, Qdata, hash_table_data_validate); | |
900 | |
901 /* obsolete as of 19990901 in xemacs-21.2 */ | |
902 define_structure_type_keyword (st, Qtype, hash_table_weakness_validate); | |
903 } | |
904 | |
905 /* Create a built-in Lisp structure type named `hash-table'. | |
906 We make #s(hashtable ...) equivalent to #s(hash-table ...), | |
907 for backward compatibility. | |
908 This is called from emacs.c. */ | |
909 void | |
910 structure_type_create_hash_table (void) | |
911 { | |
912 structure_type_create_hash_table_structure_name (Qhash_table); | |
913 structure_type_create_hash_table_structure_name (Qhashtable); /* compat */ | |
914 } | |
915 | |
916 | |
917 /************************************************************************/ | |
918 /* Definition of Lisp-visible methods */ | |
919 /************************************************************************/ | |
920 | |
921 DEFUN ("hash-table-p", Fhash_table_p, 1, 1, 0, /* | |
922 Return t if OBJECT is a hash table, else nil. | |
923 */ | |
924 (object)) | |
925 { | |
926 return HASH_TABLEP (object) ? Qt : Qnil; | |
927 } | |
928 | |
929 DEFUN ("make-hash-table", Fmake_hash_table, 0, MANY, 0, /* | |
930 Return a new empty hash table object. | |
931 Use Common Lisp style keywords to specify hash table properties. | |
932 | |
933 Keyword :test can be `eq', `eql' (default) or `equal'. | |
934 Comparison between keys is done using this function. | |
935 If speed is important, consider using `eq'. | |
936 When storing strings in the hash table, you will likely need to use `equal'. | |
937 | |
938 Keyword :size specifies the number of keys likely to be inserted. | |
939 This number of entries can be inserted without enlarging the hash table. | |
940 | |
941 Keyword :rehash-size must be a float greater than 1.0, and specifies | |
942 the factor by which to increase the size of the hash table when enlarging. | |
943 | |
944 Keyword :rehash-threshold must be a float between 0.0 and 1.0, | |
945 and specifies the load factor of the hash table which triggers enlarging. | |
946 | |
442 | 947 Non-standard keyword :weakness can be `nil' (default), `t', `key-and-value', |
948 `key', `value' or `key-or-value'. `t' is an alias for `key-and-value'. | |
428 | 949 |
442 | 950 A key-and-value-weak hash table, also known as a fully-weak or simply |
951 as a weak hash table, is one whose pointers do not count as GC | |
952 referents: for any key-value pair in the hash table, if the only | |
953 remaining pointer to either the key or the value is in a weak hash | |
954 table, then the pair will be removed from the hash table, and the key | |
955 and value collected. A non-weak hash table (or any other pointer) | |
956 would prevent the object from being collected. | |
428 | 957 |
958 A key-weak hash table is similar to a fully-weak hash table except that | |
959 a key-value pair will be removed only if the key remains unmarked | |
960 outside of weak hash tables. The pair will remain in the hash table if | |
961 the key is pointed to by something other than a weak hash table, even | |
962 if the value is not. | |
963 | |
964 A value-weak hash table is similar to a fully-weak hash table except | |
965 that a key-value pair will be removed only if the value remains | |
966 unmarked outside of weak hash tables. The pair will remain in the | |
967 hash table if the value is pointed to by something other than a weak | |
968 hash table, even if the key is not. | |
442 | 969 |
970 A key-or-value-weak hash table is similar to a fully-weak hash table except | |
971 that a key-value pair will be removed only if the value and the key remain | |
972 unmarked outside of weak hash tables. The pair will remain in the | |
973 hash table if the value or key are pointed to by something other than a weak | |
974 hash table, even if the other is not. | |
4693
80cd90837ac5
Add argument information to remaining MANY or UNEVALLED C subrs.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4585
diff
changeset
|
975 |
4777
c69aeb86b2a3
Serialise non-default hash table rehash thresholds correctly; use this.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4693
diff
changeset
|
976 arguments: (&key TEST SIZE REHASH-SIZE REHASH-THRESHOLD WEAKNESS) |
428 | 977 */ |
978 (int nargs, Lisp_Object *args)) | |
979 { | |
980 int i = 0; | |
981 Lisp_Object test = Qnil; | |
982 Lisp_Object size = Qnil; | |
983 Lisp_Object rehash_size = Qnil; | |
984 Lisp_Object rehash_threshold = Qnil; | |
985 Lisp_Object weakness = Qnil; | |
986 | |
987 while (i + 1 < nargs) | |
988 { | |
989 Lisp_Object keyword = args[i++]; | |
990 Lisp_Object value = args[i++]; | |
991 | |
992 if (EQ (keyword, Q_test)) test = value; | |
993 else if (EQ (keyword, Q_size)) size = value; | |
994 else if (EQ (keyword, Q_rehash_size)) rehash_size = value; | |
995 else if (EQ (keyword, Q_rehash_threshold)) rehash_threshold = value; | |
996 else if (EQ (keyword, Q_weakness)) weakness = value; | |
997 else if (EQ (keyword, Q_type))/*obsolete*/ weakness = value; | |
563 | 998 else invalid_constant ("Invalid hash table property keyword", keyword); |
428 | 999 } |
1000 | |
1001 if (i < nargs) | |
563 | 1002 sferror ("Hash table property requires a value", args[i]); |
428 | 1003 |
1004 #define VALIDATE_VAR(var) \ | |
1005 if (!NILP (var)) hash_table_##var##_validate (Q##var, var, ERROR_ME); | |
1006 | |
1007 VALIDATE_VAR (test); | |
1008 VALIDATE_VAR (size); | |
1009 VALIDATE_VAR (rehash_size); | |
1010 VALIDATE_VAR (rehash_threshold); | |
1011 VALIDATE_VAR (weakness); | |
1012 | |
450 | 1013 return make_standard_lisp_hash_table |
428 | 1014 (decode_hash_table_test (test), |
1015 decode_hash_table_size (size), | |
1016 decode_hash_table_rehash_size (rehash_size), | |
1017 decode_hash_table_rehash_threshold (rehash_threshold), | |
1018 decode_hash_table_weakness (weakness)); | |
1019 } | |
1020 | |
1021 DEFUN ("copy-hash-table", Fcopy_hash_table, 1, 1, 0, /* | |
1022 Return a new hash table containing the same keys and values as HASH-TABLE. | |
1023 The keys and values will not themselves be copied. | |
1024 */ | |
1025 (hash_table)) | |
1026 { | |
442 | 1027 const Lisp_Hash_Table *ht_old = xhash_table (hash_table); |
3017 | 1028 Lisp_Hash_Table *ht = ALLOC_LCRECORD_TYPE (Lisp_Hash_Table, &lrecord_hash_table); |
1029 COPY_LCRECORD (ht, ht_old); | |
428 | 1030 |
3092 | 1031 #ifdef NEW_GC |
1032 ht->hentries = (htentry *) alloc_lrecord_array (sizeof (htentry), | |
1033 ht_old->size + 1, | |
1034 &lrecord_hash_table_entry); | |
1035 #else /* not NEW_GC */ | |
1204 | 1036 ht->hentries = xnew_array (htentry, ht_old->size + 1); |
3092 | 1037 #endif /* not NEW_GC */ |
1204 | 1038 memcpy (ht->hentries, ht_old->hentries, (ht_old->size + 1) * sizeof (htentry)); |
428 | 1039 |
793 | 1040 hash_table = wrap_hash_table (ht); |
428 | 1041 |
1042 if (! EQ (ht->next_weak, Qunbound)) | |
1043 { | |
1044 ht->next_weak = Vall_weak_hash_tables; | |
1045 Vall_weak_hash_tables = hash_table; | |
1046 } | |
1047 | |
1048 return hash_table; | |
1049 } | |
1050 | |
1051 static void | |
665 | 1052 resize_hash_table (Lisp_Hash_Table *ht, Elemcount new_size) |
428 | 1053 { |
1204 | 1054 htentry *old_entries, *new_entries, *sentinel, *e; |
665 | 1055 Elemcount old_size; |
428 | 1056 |
1057 old_size = ht->size; | |
1058 ht->size = new_size; | |
1059 | |
1060 old_entries = ht->hentries; | |
1061 | |
3092 | 1062 #ifdef NEW_GC |
1063 ht->hentries = (htentry *) alloc_lrecord_array (sizeof (htentry), | |
1064 new_size + 1, | |
1065 &lrecord_hash_table_entry); | |
1066 #else /* not NEW_GC */ | |
1204 | 1067 ht->hentries = xnew_array_and_zero (htentry, new_size + 1); |
3092 | 1068 #endif /* not NEW_GC */ |
428 | 1069 new_entries = ht->hentries; |
1070 | |
1071 compute_hash_table_derived_values (ht); | |
1072 | |
440 | 1073 for (e = old_entries, sentinel = e + old_size; e < sentinel; e++) |
1204 | 1074 if (!HTENTRY_CLEAR_P (e)) |
428 | 1075 { |
1204 | 1076 htentry *probe = new_entries + HASHCODE (e->key, ht); |
428 | 1077 LINEAR_PROBING_LOOP (probe, new_entries, new_size) |
1078 ; | |
1079 *probe = *e; | |
1080 } | |
1081 | |
4117 | 1082 #ifndef NEW_GC |
489 | 1083 free_hentries (old_entries, old_size); |
4117 | 1084 #endif /* not NEW_GC */ |
428 | 1085 } |
1086 | |
440 | 1087 /* After a hash table has been saved to disk and later restored by the |
1088 portable dumper, it contains the same objects, but their addresses | |
665 | 1089 and thus their HASHCODEs have changed. */ |
428 | 1090 void |
440 | 1091 pdump_reorganize_hash_table (Lisp_Object hash_table) |
428 | 1092 { |
442 | 1093 const Lisp_Hash_Table *ht = xhash_table (hash_table); |
3092 | 1094 #ifdef NEW_GC |
1095 htentry *new_entries = | |
1096 (htentry *) alloc_lrecord_array (sizeof (htentry), ht->size + 1, | |
1097 &lrecord_hash_table_entry); | |
1098 #else /* not NEW_GC */ | |
1204 | 1099 htentry *new_entries = xnew_array_and_zero (htentry, ht->size + 1); |
3092 | 1100 #endif /* not NEW_GC */ |
1204 | 1101 htentry *e, *sentinel; |
440 | 1102 |
1103 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++) | |
1204 | 1104 if (!HTENTRY_CLEAR_P (e)) |
440 | 1105 { |
1204 | 1106 htentry *probe = new_entries + HASHCODE (e->key, ht); |
440 | 1107 LINEAR_PROBING_LOOP (probe, new_entries, ht->size) |
1108 ; | |
1109 *probe = *e; | |
1110 } | |
1111 | |
1204 | 1112 memcpy (ht->hentries, new_entries, ht->size * sizeof (htentry)); |
440 | 1113 |
4117 | 1114 #ifndef NEW_GC |
1726 | 1115 xfree (new_entries, htentry *); |
3092 | 1116 #endif /* not NEW_GC */ |
428 | 1117 } |
1118 | |
1119 static void | |
1120 enlarge_hash_table (Lisp_Hash_Table *ht) | |
1121 { | |
665 | 1122 Elemcount new_size = |
1123 hash_table_size ((Elemcount) ((double) ht->size * ht->rehash_size)); | |
428 | 1124 resize_hash_table (ht, new_size); |
1125 } | |
1126 | |
4072 | 1127 htentry * |
1204 | 1128 find_htentry (Lisp_Object key, const Lisp_Hash_Table *ht) |
428 | 1129 { |
1130 hash_table_test_function_t test_function = ht->test_function; | |
1204 | 1131 htentry *entries = ht->hentries; |
1132 htentry *probe = entries + HASHCODE (key, ht); | |
428 | 1133 |
1134 LINEAR_PROBING_LOOP (probe, entries, ht->size) | |
1135 if (KEYS_EQUAL_P (probe->key, key, test_function)) | |
1136 break; | |
1137 | |
1138 return probe; | |
1139 } | |
1140 | |
2421 | 1141 /* A version of Fputhash() that increments the value by the specified |
1142 amount and dispenses will all error checks. Assumes that tables does | |
1143 comparison using EQ. Used by the profiling routines to avoid | |
1144 overhead -- profiling overhead was being recorded at up to 15% of the | |
1145 total time. */ | |
1146 | |
1147 void | |
1148 inchash_eq (Lisp_Object key, Lisp_Object table, EMACS_INT offset) | |
1149 { | |
1150 Lisp_Hash_Table *ht = XHASH_TABLE (table); | |
1151 htentry *entries = ht->hentries; | |
1152 htentry *probe = entries + HASHCODE (key, ht); | |
1153 | |
1154 LINEAR_PROBING_LOOP (probe, entries, ht->size) | |
1155 if (EQ (probe->key, key)) | |
1156 break; | |
1157 | |
1158 if (!HTENTRY_CLEAR_P (probe)) | |
1159 probe->value = make_int (XINT (probe->value) + offset); | |
1160 else | |
1161 { | |
1162 probe->key = key; | |
1163 probe->value = make_int (offset); | |
1164 | |
1165 if (++ht->count >= ht->rehash_count) | |
1166 enlarge_hash_table (ht); | |
1167 } | |
1168 } | |
1169 | |
428 | 1170 DEFUN ("gethash", Fgethash, 2, 3, 0, /* |
1171 Find hash value for KEY in HASH-TABLE. | |
1172 If there is no corresponding value, return DEFAULT (which defaults to nil). | |
1173 */ | |
1174 (key, hash_table, default_)) | |
1175 { | |
442 | 1176 const Lisp_Hash_Table *ht = xhash_table (hash_table); |
1204 | 1177 htentry *e = find_htentry (key, ht); |
428 | 1178 |
1204 | 1179 return HTENTRY_CLEAR_P (e) ? default_ : e->value; |
428 | 1180 } |
1181 | |
1182 DEFUN ("puthash", Fputhash, 3, 3, 0, /* | |
4410
aae1994dfeec
Document return values for #'puthash, #'clrhash.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4398
diff
changeset
|
1183 Hash KEY to VALUE in HASH-TABLE, and return VALUE. |
428 | 1184 */ |
1185 (key, value, hash_table)) | |
1186 { | |
1187 Lisp_Hash_Table *ht = xhash_table (hash_table); | |
1204 | 1188 htentry *e = find_htentry (key, ht); |
428 | 1189 |
1204 | 1190 if (!HTENTRY_CLEAR_P (e)) |
428 | 1191 return e->value = value; |
1192 | |
1193 e->key = key; | |
1194 e->value = value; | |
1195 | |
1196 if (++ht->count >= ht->rehash_count) | |
1197 enlarge_hash_table (ht); | |
1198 | |
1199 return value; | |
1200 } | |
1201 | |
1204 | 1202 /* Remove htentry pointed at by PROBE. |
428 | 1203 Subsequent entries are removed and reinserted. |
1204 We don't use tombstones - too wasteful. */ | |
1205 static void | |
1204 | 1206 remhash_1 (Lisp_Hash_Table *ht, htentry *entries, htentry *probe) |
428 | 1207 { |
665 | 1208 Elemcount size = ht->size; |
1204 | 1209 CLEAR_HTENTRY (probe); |
428 | 1210 probe++; |
1211 ht->count--; | |
1212 | |
1213 LINEAR_PROBING_LOOP (probe, entries, size) | |
1214 { | |
1215 Lisp_Object key = probe->key; | |
1204 | 1216 htentry *probe2 = entries + HASHCODE (key, ht); |
428 | 1217 LINEAR_PROBING_LOOP (probe2, entries, size) |
1218 if (EQ (probe2->key, key)) | |
1204 | 1219 /* htentry at probe doesn't need to move. */ |
428 | 1220 goto continue_outer_loop; |
1204 | 1221 /* Move htentry from probe to new home at probe2. */ |
428 | 1222 *probe2 = *probe; |
1204 | 1223 CLEAR_HTENTRY (probe); |
428 | 1224 continue_outer_loop: continue; |
1225 } | |
1226 } | |
1227 | |
1228 DEFUN ("remhash", Fremhash, 2, 2, 0, /* | |
1229 Remove the entry for KEY from HASH-TABLE. | |
1230 Do nothing if there is no entry for KEY in HASH-TABLE. | |
617 | 1231 Return non-nil if an entry was removed. |
428 | 1232 */ |
1233 (key, hash_table)) | |
1234 { | |
1235 Lisp_Hash_Table *ht = xhash_table (hash_table); | |
1204 | 1236 htentry *e = find_htentry (key, ht); |
428 | 1237 |
1204 | 1238 if (HTENTRY_CLEAR_P (e)) |
428 | 1239 return Qnil; |
1240 | |
1241 remhash_1 (ht, ht->hentries, e); | |
1242 return Qt; | |
1243 } | |
1244 | |
1245 DEFUN ("clrhash", Fclrhash, 1, 1, 0, /* | |
1246 Remove all entries from HASH-TABLE, leaving it empty. | |
4410
aae1994dfeec
Document return values for #'puthash, #'clrhash.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4398
diff
changeset
|
1247 Return HASH-TABLE. |
428 | 1248 */ |
1249 (hash_table)) | |
1250 { | |
1251 Lisp_Hash_Table *ht = xhash_table (hash_table); | |
1204 | 1252 htentry *e, *sentinel; |
428 | 1253 |
1254 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++) | |
1204 | 1255 CLEAR_HTENTRY (e); |
428 | 1256 ht->count = 0; |
1257 | |
1258 return hash_table; | |
1259 } | |
1260 | |
1261 /************************************************************************/ | |
1262 /* Accessor Functions */ | |
1263 /************************************************************************/ | |
1264 | |
1265 DEFUN ("hash-table-count", Fhash_table_count, 1, 1, 0, /* | |
1266 Return the number of entries in HASH-TABLE. | |
1267 */ | |
1268 (hash_table)) | |
1269 { | |
1270 return make_int (xhash_table (hash_table)->count); | |
1271 } | |
1272 | |
1273 DEFUN ("hash-table-test", Fhash_table_test, 1, 1, 0, /* | |
1274 Return the test function of HASH-TABLE. | |
1275 This can be one of `eq', `eql' or `equal'. | |
1276 */ | |
1277 (hash_table)) | |
1278 { | |
1279 hash_table_test_function_t fun = xhash_table (hash_table)->test_function; | |
1280 | |
1281 return (fun == lisp_object_eql_equal ? Qeql : | |
1282 fun == lisp_object_equal_equal ? Qequal : | |
1283 Qeq); | |
1284 } | |
1285 | |
1286 DEFUN ("hash-table-size", Fhash_table_size, 1, 1, 0, /* | |
1287 Return the size of HASH-TABLE. | |
1288 This is the current number of slots in HASH-TABLE, whether occupied or not. | |
1289 */ | |
1290 (hash_table)) | |
1291 { | |
1292 return make_int (xhash_table (hash_table)->size); | |
1293 } | |
1294 | |
1295 DEFUN ("hash-table-rehash-size", Fhash_table_rehash_size, 1, 1, 0, /* | |
1296 Return the current rehash size of HASH-TABLE. | |
1297 This is a float greater than 1.0; the factor by which HASH-TABLE | |
1298 is enlarged when the rehash threshold is exceeded. | |
1299 */ | |
1300 (hash_table)) | |
1301 { | |
1302 return make_float (xhash_table (hash_table)->rehash_size); | |
1303 } | |
1304 | |
1305 DEFUN ("hash-table-rehash-threshold", Fhash_table_rehash_threshold, 1, 1, 0, /* | |
1306 Return the current rehash threshold of HASH-TABLE. | |
1307 This is a float between 0.0 and 1.0; the maximum `load factor' of HASH-TABLE, | |
1308 beyond which the HASH-TABLE is enlarged by rehashing. | |
1309 */ | |
1310 (hash_table)) | |
1311 { | |
438 | 1312 return make_float (xhash_table (hash_table)->rehash_threshold); |
428 | 1313 } |
1314 | |
1315 DEFUN ("hash-table-weakness", Fhash_table_weakness, 1, 1, 0, /* | |
1316 Return the weakness of HASH-TABLE. | |
442 | 1317 This can be one of `nil', `key-and-value', `key-or-value', `key' or `value'. |
428 | 1318 */ |
1319 (hash_table)) | |
1320 { | |
1321 switch (xhash_table (hash_table)->weakness) | |
1322 { | |
442 | 1323 case HASH_TABLE_WEAK: return Qkey_and_value; |
1324 case HASH_TABLE_KEY_WEAK: return Qkey; | |
1325 case HASH_TABLE_KEY_VALUE_WEAK: return Qkey_or_value; | |
1326 case HASH_TABLE_VALUE_WEAK: return Qvalue; | |
1327 default: return Qnil; | |
428 | 1328 } |
1329 } | |
1330 | |
1331 /* obsolete as of 19990901 in xemacs-21.2 */ | |
1332 DEFUN ("hash-table-type", Fhash_table_type, 1, 1, 0, /* | |
1333 Return the type of HASH-TABLE. | |
1334 This can be one of `non-weak', `weak', `key-weak' or `value-weak'. | |
1335 */ | |
1336 (hash_table)) | |
1337 { | |
1338 switch (xhash_table (hash_table)->weakness) | |
1339 { | |
442 | 1340 case HASH_TABLE_WEAK: return Qweak; |
1341 case HASH_TABLE_KEY_WEAK: return Qkey_weak; | |
1342 case HASH_TABLE_KEY_VALUE_WEAK: return Qkey_or_value_weak; | |
1343 case HASH_TABLE_VALUE_WEAK: return Qvalue_weak; | |
1344 default: return Qnon_weak; | |
428 | 1345 } |
1346 } | |
1347 | |
1348 /************************************************************************/ | |
1349 /* Mapping Functions */ | |
1350 /************************************************************************/ | |
489 | 1351 |
1352 /* We need to be careful when mapping over hash tables because the | |
1353 hash table might be modified during the mapping operation: | |
1354 - by the mapping function | |
1355 - by gc (if the hash table is weak) | |
1356 | |
1357 So we make a copy of the hentries at the beginning of the mapping | |
497 | 1358 operation, and iterate over the copy. Naturally, this is |
1359 expensive, but not as expensive as you might think, because no | |
1360 actual memory has to be collected by our notoriously inefficient | |
1361 GC; we use an unwind-protect instead to free the memory directly. | |
1362 | |
1363 We could avoid the copying by having the hash table modifiers | |
1364 puthash and remhash check for currently active mapping functions. | |
1365 Disadvantages: it's hard to get right, and IMO hash mapping | |
1366 functions are basically rare, and no extra space in the hash table | |
1367 object and no extra cpu in puthash or remhash should be wasted to | |
1368 make maphash 3% faster. From a design point of view, the basic | |
1369 functions gethash, puthash and remhash should be implementable | |
1370 without having to think about maphash. | |
1371 | |
1372 Note: We don't (yet) have Common Lisp's with-hash-table-iterator. | |
1373 If you implement this naively, you cannot have more than one | |
1374 concurrently active iterator over the same hash table. The `each' | |
1375 function in perl has this limitation. | |
1376 | |
1377 Note: We GCPRO memory on the heap, not on the stack. There is no | |
1378 obvious reason why this is bad, but as of this writing this is the | |
1379 only known occurrence of this technique in the code. | |
504 | 1380 |
1381 -- Martin | |
1382 */ | |
1383 | |
1384 /* Ben disagrees with the "copying hentries" design, and says: | |
1385 | |
1386 Another solution is the same as I've already proposed -- when | |
1387 mapping, mark the table as "change-unsafe", and in this case, use a | |
1388 secondary table to maintain changes. this could be basically a | |
1389 standard hash table, but with entries only for added or deleted | |
1390 entries in the primary table, and a marker like Qunbound to | |
1391 indicate a deleted entry. puthash, gethash and remhash need a | |
1392 single extra check for this secondary table -- totally | |
1393 insignificant speedwise. if you really cared about making | |
1394 recursive maphashes completely correct, you'd have to do a bit of | |
1395 extra work here -- when maphashing, if the secondary table exists, | |
1396 make a copy of it, and use the copy in conjunction with the primary | |
1397 table when mapping. the advantages of this are | |
1398 | |
1399 [a] easy to demonstrate correct, even with weak hashtables. | |
1400 | |
1401 [b] no extra overhead in the general maphash case -- only when you | |
1402 modify the table while maphashing, and even then the overhead is | |
1403 very small. | |
497 | 1404 */ |
1405 | |
489 | 1406 static Lisp_Object |
1407 maphash_unwind (Lisp_Object unwind_obj) | |
1408 { | |
1409 void *ptr = (void *) get_opaque_ptr (unwind_obj); | |
1726 | 1410 xfree (ptr, void *); |
489 | 1411 free_opaque_ptr (unwind_obj); |
1412 return Qnil; | |
1413 } | |
1414 | |
1415 /* Return a malloced array of alternating key/value pairs from HT. */ | |
1416 static Lisp_Object * | |
1417 copy_compress_hentries (const Lisp_Hash_Table *ht) | |
1418 { | |
1419 Lisp_Object * const objs = | |
1420 /* If the hash table is empty, ht->count could be 0. */ | |
1421 xnew_array (Lisp_Object, 2 * (ht->count > 0 ? ht->count : 1)); | |
1204 | 1422 const htentry *e, *sentinel; |
489 | 1423 Lisp_Object *pobj; |
1424 | |
1425 for (e = ht->hentries, sentinel = e + ht->size, pobj = objs; e < sentinel; e++) | |
1204 | 1426 if (!HTENTRY_CLEAR_P (e)) |
489 | 1427 { |
1428 *(pobj++) = e->key; | |
1429 *(pobj++) = e->value; | |
1430 } | |
1431 | |
1432 type_checking_assert (pobj == objs + 2 * ht->count); | |
1433 | |
1434 return objs; | |
1435 } | |
1436 | |
428 | 1437 DEFUN ("maphash", Fmaphash, 2, 2, 0, /* |
1438 Map FUNCTION over entries in HASH-TABLE, calling it with two args, | |
1439 each key and value in HASH-TABLE. | |
1440 | |
489 | 1441 FUNCTION must not modify HASH-TABLE, with the one exception that FUNCTION |
428 | 1442 may remhash or puthash the entry currently being processed by FUNCTION. |
1443 */ | |
1444 (function, hash_table)) | |
1445 { | |
489 | 1446 const Lisp_Hash_Table * const ht = xhash_table (hash_table); |
1447 Lisp_Object * const objs = copy_compress_hentries (ht); | |
1448 Lisp_Object args[3]; | |
1449 const Lisp_Object *pobj, *end; | |
1450 int speccount = specpdl_depth (); | |
1451 struct gcpro gcpro1; | |
1452 | |
1453 record_unwind_protect (maphash_unwind, make_opaque_ptr ((void *)objs)); | |
1454 GCPRO1 (objs[0]); | |
1455 gcpro1.nvars = 2 * ht->count; | |
428 | 1456 |
489 | 1457 args[0] = function; |
1458 | |
1459 for (pobj = objs, end = pobj + 2 * ht->count; pobj < end; pobj += 2) | |
1460 { | |
1461 args[1] = pobj[0]; | |
1462 args[2] = pobj[1]; | |
1463 Ffuncall (countof (args), args); | |
1464 } | |
1465 | |
771 | 1466 unbind_to (speccount); |
489 | 1467 UNGCPRO; |
428 | 1468 |
1469 return Qnil; | |
1470 } | |
1471 | |
489 | 1472 /* Map *C* function FUNCTION over the elements of a non-weak lisp hash table. |
1473 FUNCTION must not modify HASH-TABLE, with the one exception that FUNCTION | |
1474 may puthash the entry currently being processed by FUNCTION. | |
1475 Mapping terminates if FUNCTION returns something other than 0. */ | |
428 | 1476 void |
489 | 1477 elisp_maphash_unsafe (maphash_function_t function, |
428 | 1478 Lisp_Object hash_table, void *extra_arg) |
1479 { | |
442 | 1480 const Lisp_Hash_Table *ht = XHASH_TABLE (hash_table); |
1204 | 1481 const htentry *e, *sentinel; |
428 | 1482 |
1483 for (e = ht->hentries, sentinel = e + ht->size; e < sentinel; e++) | |
1204 | 1484 if (!HTENTRY_CLEAR_P (e)) |
489 | 1485 if (function (e->key, e->value, extra_arg)) |
1486 return; | |
428 | 1487 } |
1488 | |
489 | 1489 /* Map *C* function FUNCTION over the elements of a lisp hash table. |
1490 It is safe for FUNCTION to modify HASH-TABLE. | |
1491 Mapping terminates if FUNCTION returns something other than 0. */ | |
1492 void | |
1493 elisp_maphash (maphash_function_t function, | |
1494 Lisp_Object hash_table, void *extra_arg) | |
1495 { | |
1496 const Lisp_Hash_Table * const ht = xhash_table (hash_table); | |
1497 Lisp_Object * const objs = copy_compress_hentries (ht); | |
1498 const Lisp_Object *pobj, *end; | |
1499 int speccount = specpdl_depth (); | |
1500 struct gcpro gcpro1; | |
1501 | |
1502 record_unwind_protect (maphash_unwind, make_opaque_ptr ((void *)objs)); | |
1503 GCPRO1 (objs[0]); | |
1504 gcpro1.nvars = 2 * ht->count; | |
1505 | |
1506 for (pobj = objs, end = pobj + 2 * ht->count; pobj < end; pobj += 2) | |
1507 if (function (pobj[0], pobj[1], extra_arg)) | |
1508 break; | |
1509 | |
771 | 1510 unbind_to (speccount); |
489 | 1511 UNGCPRO; |
1512 } | |
1513 | |
1514 /* Remove all elements of a lisp hash table satisfying *C* predicate PREDICATE. | |
1515 PREDICATE must not modify HASH-TABLE. */ | |
428 | 1516 void |
1517 elisp_map_remhash (maphash_function_t predicate, | |
1518 Lisp_Object hash_table, void *extra_arg) | |
1519 { | |
489 | 1520 const Lisp_Hash_Table * const ht = xhash_table (hash_table); |
1521 Lisp_Object * const objs = copy_compress_hentries (ht); | |
1522 const Lisp_Object *pobj, *end; | |
1523 int speccount = specpdl_depth (); | |
1524 struct gcpro gcpro1; | |
428 | 1525 |
489 | 1526 record_unwind_protect (maphash_unwind, make_opaque_ptr ((void *)objs)); |
1527 GCPRO1 (objs[0]); | |
1528 gcpro1.nvars = 2 * ht->count; | |
1529 | |
1530 for (pobj = objs, end = pobj + 2 * ht->count; pobj < end; pobj += 2) | |
1531 if (predicate (pobj[0], pobj[1], extra_arg)) | |
1532 Fremhash (pobj[0], hash_table); | |
1533 | |
771 | 1534 unbind_to (speccount); |
489 | 1535 UNGCPRO; |
428 | 1536 } |
1537 | |
1538 | |
1539 /************************************************************************/ | |
1540 /* garbage collecting weak hash tables */ | |
1541 /************************************************************************/ | |
1598 | 1542 #ifdef USE_KKCC |
2645 | 1543 #define MARK_OBJ(obj) do { \ |
1544 Lisp_Object mo_obj = (obj); \ | |
1545 if (!marked_p (mo_obj)) \ | |
1546 { \ | |
1547 kkcc_gc_stack_push_lisp_object (mo_obj, 0, -1); \ | |
1548 did_mark = 1; \ | |
1549 } \ | |
1598 | 1550 } while (0) |
1551 | |
1552 #else /* NO USE_KKCC */ | |
1553 | |
442 | 1554 #define MARK_OBJ(obj) do { \ |
1555 Lisp_Object mo_obj = (obj); \ | |
1556 if (!marked_p (mo_obj)) \ | |
1557 { \ | |
1558 mark_object (mo_obj); \ | |
1559 did_mark = 1; \ | |
1560 } \ | |
1561 } while (0) | |
1598 | 1562 #endif /*NO USE_KKCC */ |
442 | 1563 |
428 | 1564 |
1565 /* Complete the marking for semi-weak hash tables. */ | |
1566 int | |
1567 finish_marking_weak_hash_tables (void) | |
1568 { | |
1569 Lisp_Object hash_table; | |
1570 int did_mark = 0; | |
1571 | |
1572 for (hash_table = Vall_weak_hash_tables; | |
1573 !NILP (hash_table); | |
1574 hash_table = XHASH_TABLE (hash_table)->next_weak) | |
1575 { | |
442 | 1576 const Lisp_Hash_Table *ht = XHASH_TABLE (hash_table); |
1204 | 1577 const htentry *e = ht->hentries; |
1578 const htentry *sentinel = e + ht->size; | |
428 | 1579 |
1580 if (! marked_p (hash_table)) | |
1581 /* The hash table is probably garbage. Ignore it. */ | |
1582 continue; | |
1583 | |
1584 /* Now, scan over all the pairs. For all pairs that are | |
1585 half-marked, we may need to mark the other half if we're | |
1586 keeping this pair. */ | |
1587 switch (ht->weakness) | |
1588 { | |
1589 case HASH_TABLE_KEY_WEAK: | |
1590 for (; e < sentinel; e++) | |
1204 | 1591 if (!HTENTRY_CLEAR_P (e)) |
428 | 1592 if (marked_p (e->key)) |
1593 MARK_OBJ (e->value); | |
1594 break; | |
1595 | |
1596 case HASH_TABLE_VALUE_WEAK: | |
1597 for (; e < sentinel; e++) | |
1204 | 1598 if (!HTENTRY_CLEAR_P (e)) |
428 | 1599 if (marked_p (e->value)) |
1600 MARK_OBJ (e->key); | |
1601 break; | |
1602 | |
442 | 1603 case HASH_TABLE_KEY_VALUE_WEAK: |
1604 for (; e < sentinel; e++) | |
1204 | 1605 if (!HTENTRY_CLEAR_P (e)) |
442 | 1606 { |
1607 if (marked_p (e->value)) | |
1608 MARK_OBJ (e->key); | |
1609 else if (marked_p (e->key)) | |
1610 MARK_OBJ (e->value); | |
1611 } | |
1612 break; | |
1613 | |
428 | 1614 case HASH_TABLE_KEY_CAR_WEAK: |
1615 for (; e < sentinel; e++) | |
1204 | 1616 if (!HTENTRY_CLEAR_P (e)) |
428 | 1617 if (!CONSP (e->key) || marked_p (XCAR (e->key))) |
1618 { | |
1619 MARK_OBJ (e->key); | |
1620 MARK_OBJ (e->value); | |
1621 } | |
1622 break; | |
1623 | |
450 | 1624 /* We seem to be sprouting new weakness types at an alarming |
1625 rate. At least this is not externally visible - and in | |
1626 fact all of these KEY_CAR_* types are only used by the | |
1627 glyph code. */ | |
1628 case HASH_TABLE_KEY_CAR_VALUE_WEAK: | |
1629 for (; e < sentinel; e++) | |
1204 | 1630 if (!HTENTRY_CLEAR_P (e)) |
450 | 1631 { |
1632 if (!CONSP (e->key) || marked_p (XCAR (e->key))) | |
1633 { | |
1634 MARK_OBJ (e->key); | |
1635 MARK_OBJ (e->value); | |
1636 } | |
1637 else if (marked_p (e->value)) | |
1638 MARK_OBJ (e->key); | |
1639 } | |
1640 break; | |
1641 | |
428 | 1642 case HASH_TABLE_VALUE_CAR_WEAK: |
1643 for (; e < sentinel; e++) | |
1204 | 1644 if (!HTENTRY_CLEAR_P (e)) |
428 | 1645 if (!CONSP (e->value) || marked_p (XCAR (e->value))) |
1646 { | |
1647 MARK_OBJ (e->key); | |
1648 MARK_OBJ (e->value); | |
1649 } | |
1650 break; | |
1651 | |
1652 default: | |
1653 break; | |
1654 } | |
1655 } | |
1656 | |
1657 return did_mark; | |
1658 } | |
1659 | |
1660 void | |
1661 prune_weak_hash_tables (void) | |
1662 { | |
1663 Lisp_Object hash_table, prev = Qnil; | |
1664 for (hash_table = Vall_weak_hash_tables; | |
1665 !NILP (hash_table); | |
1666 hash_table = XHASH_TABLE (hash_table)->next_weak) | |
1667 { | |
1668 if (! marked_p (hash_table)) | |
1669 { | |
1670 /* This hash table itself is garbage. Remove it from the list. */ | |
1671 if (NILP (prev)) | |
1672 Vall_weak_hash_tables = XHASH_TABLE (hash_table)->next_weak; | |
1673 else | |
1674 XHASH_TABLE (prev)->next_weak = XHASH_TABLE (hash_table)->next_weak; | |
1675 } | |
1676 else | |
1677 { | |
1678 /* Now, scan over all the pairs. Remove all of the pairs | |
1679 in which the key or value, or both, is unmarked | |
1680 (depending on the weakness of the hash table). */ | |
1681 Lisp_Hash_Table *ht = XHASH_TABLE (hash_table); | |
1204 | 1682 htentry *entries = ht->hentries; |
1683 htentry *sentinel = entries + ht->size; | |
1684 htentry *e; | |
428 | 1685 |
1686 for (e = entries; e < sentinel; e++) | |
1204 | 1687 if (!HTENTRY_CLEAR_P (e)) |
428 | 1688 { |
1689 again: | |
1690 if (!marked_p (e->key) || !marked_p (e->value)) | |
1691 { | |
1692 remhash_1 (ht, entries, e); | |
1204 | 1693 if (!HTENTRY_CLEAR_P (e)) |
428 | 1694 goto again; |
1695 } | |
1696 } | |
1697 | |
1698 prev = hash_table; | |
1699 } | |
1700 } | |
1701 } | |
1702 | |
1703 /* Return a hash value for an array of Lisp_Objects of size SIZE. */ | |
1704 | |
665 | 1705 Hashcode |
428 | 1706 internal_array_hash (Lisp_Object *arr, int size, int depth) |
1707 { | |
1708 int i; | |
665 | 1709 Hashcode hash = 0; |
442 | 1710 depth++; |
428 | 1711 |
1712 if (size <= 5) | |
1713 { | |
1714 for (i = 0; i < size; i++) | |
442 | 1715 hash = HASH2 (hash, internal_hash (arr[i], depth)); |
428 | 1716 return hash; |
1717 } | |
1718 | |
1719 /* just pick five elements scattered throughout the array. | |
1720 A slightly better approach would be to offset by some | |
1721 noise factor from the points chosen below. */ | |
1722 for (i = 0; i < 5; i++) | |
442 | 1723 hash = HASH2 (hash, internal_hash (arr[i*size/5], depth)); |
428 | 1724 |
1725 return hash; | |
1726 } | |
1727 | |
1728 /* Return a hash value for a Lisp_Object. This is for use when hashing | |
1729 objects with the comparison being `equal' (for `eq', you can just | |
1730 use the Lisp_Object itself as the hash value). You need to make a | |
1731 tradeoff between the speed of the hash function and how good the | |
1732 hashing is. In particular, the hash function needs to be FAST, | |
1733 so you can't just traipse down the whole tree hashing everything | |
1734 together. Most of the time, objects will differ in the first | |
1735 few elements you hash. Thus, we only go to a short depth (5) | |
1736 and only hash at most 5 elements out of a vector. Theoretically | |
1737 we could still take 5^5 time (a big big number) to compute a | |
1738 hash, but practically this won't ever happen. */ | |
1739 | |
665 | 1740 Hashcode |
428 | 1741 internal_hash (Lisp_Object obj, int depth) |
1742 { | |
1743 if (depth > 5) | |
1744 return 0; | |
4398
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1745 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1746 if (CONSP(obj)) |
428 | 1747 { |
4398
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1748 Hashcode hash, h; |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1749 int s; |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1750 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1751 depth += 1; |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1752 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1753 if (!CONSP(XCDR(obj))) |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1754 { |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1755 /* special case for '(a . b) conses */ |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1756 return HASH2(internal_hash(XCAR(obj), depth), |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1757 internal_hash(XCDR(obj), depth)); |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1758 } |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1759 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1760 /* Don't simply tail recurse; we want to hash lists with the |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1761 same contents in distinct orders differently. */ |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1762 hash = internal_hash(XCAR(obj), depth); |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1763 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1764 obj = XCDR(obj); |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1765 for (s = 1; s < 6 && CONSP(obj); obj = XCDR(obj), s++) |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1766 { |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1767 h = internal_hash(XCAR(obj), depth); |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1768 hash = HASH3(hash, h, s); |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1769 } |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1770 |
479443c0f95a
Have list hashes depend on the order of the contents, as is the case for vectors.
Aidan Kehoe <kehoea@parhasard.net>
parents:
4117
diff
changeset
|
1771 return hash; |
428 | 1772 } |
1773 if (STRINGP (obj)) | |
1774 { | |
1775 return hash_string (XSTRING_DATA (obj), XSTRING_LENGTH (obj)); | |
1776 } | |
1777 if (LRECORDP (obj)) | |
1778 { | |
442 | 1779 const struct lrecord_implementation |
428 | 1780 *imp = XRECORD_LHEADER_IMPLEMENTATION (obj); |
1781 if (imp->hash) | |
1782 return imp->hash (obj, depth); | |
1783 } | |
1784 | |
1785 return LISP_HASH (obj); | |
1786 } | |
1787 | |
1788 DEFUN ("sxhash", Fsxhash, 1, 1, 0, /* | |
1789 Return a hash value for OBJECT. | |
444 | 1790 \(equal obj1 obj2) implies (= (sxhash obj1) (sxhash obj2)). |
428 | 1791 */ |
1792 (object)) | |
1793 { | |
1794 return make_int (internal_hash (object, 0)); | |
1795 } | |
1796 | |
1797 #if 0 | |
826 | 1798 DEFUN ("internal-hash-value", Finternal_hash_value, 1, 1, 0, /* |
428 | 1799 Hash value of OBJECT. For debugging. |
1800 The value is returned as (HIGH . LOW). | |
1801 */ | |
1802 (object)) | |
1803 { | |
1804 /* This function is pretty 32bit-centric. */ | |
665 | 1805 Hashcode hash = internal_hash (object, 0); |
428 | 1806 return Fcons (hash >> 16, hash & 0xffff); |
1807 } | |
1808 #endif | |
1809 | |
1810 | |
1811 /************************************************************************/ | |
1812 /* initialization */ | |
1813 /************************************************************************/ | |
1814 | |
1815 void | |
1816 syms_of_elhash (void) | |
1817 { | |
1818 DEFSUBR (Fhash_table_p); | |
1819 DEFSUBR (Fmake_hash_table); | |
1820 DEFSUBR (Fcopy_hash_table); | |
1821 DEFSUBR (Fgethash); | |
1822 DEFSUBR (Fremhash); | |
1823 DEFSUBR (Fputhash); | |
1824 DEFSUBR (Fclrhash); | |
1825 DEFSUBR (Fmaphash); | |
1826 DEFSUBR (Fhash_table_count); | |
1827 DEFSUBR (Fhash_table_test); | |
1828 DEFSUBR (Fhash_table_size); | |
1829 DEFSUBR (Fhash_table_rehash_size); | |
1830 DEFSUBR (Fhash_table_rehash_threshold); | |
1831 DEFSUBR (Fhash_table_weakness); | |
1832 DEFSUBR (Fhash_table_type); /* obsolete */ | |
1833 DEFSUBR (Fsxhash); | |
1834 #if 0 | |
1835 DEFSUBR (Finternal_hash_value); | |
1836 #endif | |
1837 | |
563 | 1838 DEFSYMBOL_MULTIWORD_PREDICATE (Qhash_tablep); |
1839 DEFSYMBOL (Qhash_table); | |
1840 DEFSYMBOL (Qhashtable); | |
1841 DEFSYMBOL (Qweakness); | |
1842 DEFSYMBOL (Qvalue); | |
1843 DEFSYMBOL (Qkey_or_value); | |
1844 DEFSYMBOL (Qkey_and_value); | |
1845 DEFSYMBOL (Qrehash_size); | |
1846 DEFSYMBOL (Qrehash_threshold); | |
428 | 1847 |
563 | 1848 DEFSYMBOL (Qweak); /* obsolete */ |
1849 DEFSYMBOL (Qkey_weak); /* obsolete */ | |
1850 DEFSYMBOL (Qkey_or_value_weak); /* obsolete */ | |
1851 DEFSYMBOL (Qvalue_weak); /* obsolete */ | |
1852 DEFSYMBOL (Qnon_weak); /* obsolete */ | |
428 | 1853 |
563 | 1854 DEFKEYWORD (Q_test); |
1855 DEFKEYWORD (Q_size); | |
1856 DEFKEYWORD (Q_rehash_size); | |
1857 DEFKEYWORD (Q_rehash_threshold); | |
1858 DEFKEYWORD (Q_weakness); | |
1859 DEFKEYWORD (Q_type); /* obsolete */ | |
428 | 1860 } |
1861 | |
1862 void | |
771 | 1863 init_elhash_once_early (void) |
428 | 1864 { |
771 | 1865 INIT_LRECORD_IMPLEMENTATION (hash_table); |
3092 | 1866 #ifdef NEW_GC |
1867 INIT_LRECORD_IMPLEMENTATION (hash_table_entry); | |
1868 #endif /* NEW_GC */ | |
771 | 1869 |
428 | 1870 /* This must NOT be staticpro'd */ |
1871 Vall_weak_hash_tables = Qnil; | |
452 | 1872 dump_add_weak_object_chain (&Vall_weak_hash_tables); |
428 | 1873 } |