1/*
2 * This file is part of the MicroPython project, http://micropython.org/
3 *
4 * The MIT License (MIT)
5 *
6 * Copyright (c) 2013, 2014 Damien P. George
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 */
26
27#include <stdint.h>
28#include <stdlib.h>
29#include <string.h>
30#include <assert.h>
31
32#include "py/mpconfig.h"
33#include "py/misc.h"
34#include "py/runtime.h"
35
36#if MICROPY_DEBUG_VERBOSE // print debugging info
37#define DEBUG_PRINT (1)
38#else // don't print debugging info
39#define DEBUG_PRINT (0)
40#define DEBUG_printf(...) (void)0
41#endif
42
43// This table of sizes is used to control the growth of hash tables.
44// The first set of sizes are chosen so the allocation fits exactly in a
45// 4-word GC block, and it's not so important for these small values to be
46// prime. The latter sizes are prime and increase at an increasing rate.
47STATIC const uint16_t hash_allocation_sizes[] = {
48 0, 2, 4, 6, 8, 10, 12, // +2
49 17, 23, 29, 37, 47, 59, 73, // *1.25
50 97, 127, 167, 223, 293, 389, 521, 691, 919, 1223, 1627, 2161, // *1.33
51 3229, 4831, 7243, 10861, 16273, 24407, 36607, 54907, // *1.5
52};
53
54STATIC size_t get_hash_alloc_greater_or_equal_to(size_t x) {
55 for (size_t i = 0; i < MP_ARRAY_SIZE(hash_allocation_sizes); i++) {
56 if (hash_allocation_sizes[i] >= x) {
57 return hash_allocation_sizes[i];
58 }
59 }
60 // ran out of primes in the table!
61 // return something sensible, at least make it odd
62 return (x + x / 2) | 1;
63}
64
65/******************************************************************************/
66/* map */
67
68void mp_map_init(mp_map_t *map, size_t n) {
69 if (n == 0) {
70 map->alloc = 0;
71 map->table = NULL;
72 } else {
73 map->alloc = n;
74 map->table = m_new0(mp_map_elem_t, map->alloc);
75 }
76 map->used = 0;
77 map->all_keys_are_qstrs = 1;
78 map->is_fixed = 0;
79 map->is_ordered = 0;
80}
81
82void mp_map_init_fixed_table(mp_map_t *map, size_t n, const mp_obj_t *table) {
83 map->alloc = n;
84 map->used = n;
85 map->all_keys_are_qstrs = 1;
86 map->is_fixed = 1;
87 map->is_ordered = 1;
88 map->table = (mp_map_elem_t *)table;
89}
90
91// Differentiate from mp_map_clear() - semantics is different
92void mp_map_deinit(mp_map_t *map) {
93 if (!map->is_fixed) {
94 m_del(mp_map_elem_t, map->table, map->alloc);
95 }
96 map->used = map->alloc = 0;
97}
98
99void mp_map_clear(mp_map_t *map) {
100 if (!map->is_fixed) {
101 m_del(mp_map_elem_t, map->table, map->alloc);
102 }
103 map->alloc = 0;
104 map->used = 0;
105 map->all_keys_are_qstrs = 1;
106 map->is_fixed = 0;
107 map->table = NULL;
108}
109
110STATIC void mp_map_rehash(mp_map_t *map) {
111 size_t old_alloc = map->alloc;
112 size_t new_alloc = get_hash_alloc_greater_or_equal_to(map->alloc + 1);
113 DEBUG_printf("mp_map_rehash(%p): " UINT_FMT " -> " UINT_FMT "\n", map, old_alloc, new_alloc);
114 mp_map_elem_t *old_table = map->table;
115 mp_map_elem_t *new_table = m_new0(mp_map_elem_t, new_alloc);
116 // If we reach this point, table resizing succeeded, now we can edit the old map.
117 map->alloc = new_alloc;
118 map->used = 0;
119 map->all_keys_are_qstrs = 1;
120 map->table = new_table;
121 for (size_t i = 0; i < old_alloc; i++) {
122 if (old_table[i].key != MP_OBJ_NULL && old_table[i].key != MP_OBJ_SENTINEL) {
123 mp_map_lookup(map, old_table[i].key, MP_MAP_LOOKUP_ADD_IF_NOT_FOUND)->value = old_table[i].value;
124 }
125 }
126 m_del(mp_map_elem_t, old_table, old_alloc);
127}
128
129// MP_MAP_LOOKUP behaviour:
130// - returns NULL if not found, else the slot it was found in with key,value non-null
131// MP_MAP_LOOKUP_ADD_IF_NOT_FOUND behaviour:
132// - returns slot, with key non-null and value=MP_OBJ_NULL if it was added
133// MP_MAP_LOOKUP_REMOVE_IF_FOUND behaviour:
134// - returns NULL if not found, else the slot if was found in with key null and value non-null
135mp_map_elem_t *mp_map_lookup(mp_map_t *map, mp_obj_t index, mp_map_lookup_kind_t lookup_kind) {
136 // If the map is a fixed array then we must only be called for a lookup
137 assert(!map->is_fixed || lookup_kind == MP_MAP_LOOKUP);
138
139 // Work out if we can compare just pointers
140 bool compare_only_ptrs = map->all_keys_are_qstrs;
141 if (compare_only_ptrs) {
142 if (mp_obj_is_qstr(index)) {
143 // Index is a qstr, so can just do ptr comparison.
144 } else if (mp_obj_is_type(index, &mp_type_str)) {
145 // Index is a non-interned string.
146 // We can either intern the string, or force a full equality comparison.
147 // We chose the latter, since interning costs time and potentially RAM,
148 // and it won't necessarily benefit subsequent calls because these calls
149 // most likely won't pass the newly-interned string.
150 compare_only_ptrs = false;
151 } else if (lookup_kind != MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
152 // If we are not adding, then we can return straight away a failed
153 // lookup because we know that the index will never be found.
154 return NULL;
155 }
156 }
157
158 // if the map is an ordered array then we must do a brute force linear search
159 if (map->is_ordered) {
160 for (mp_map_elem_t *elem = &map->table[0], *top = &map->table[map->used]; elem < top; elem++) {
161 if (elem->key == index || (!compare_only_ptrs && mp_obj_equal(elem->key, index))) {
162 #if MICROPY_PY_COLLECTIONS_ORDEREDDICT
163 if (MP_UNLIKELY(lookup_kind == MP_MAP_LOOKUP_REMOVE_IF_FOUND)) {
164 // remove the found element by moving the rest of the array down
165 mp_obj_t value = elem->value;
166 --map->used;
167 memmove(elem, elem + 1, (top - elem - 1) * sizeof(*elem));
168 // put the found element after the end so the caller can access it if needed
169 // note: caller must NULL the value so the GC can clean up (e.g. see dict_get_helper).
170 elem = &map->table[map->used];
171 elem->key = MP_OBJ_NULL;
172 elem->value = value;
173 }
174 #endif
175 return elem;
176 }
177 }
178 #if MICROPY_PY_COLLECTIONS_ORDEREDDICT
179 if (MP_LIKELY(lookup_kind != MP_MAP_LOOKUP_ADD_IF_NOT_FOUND)) {
180 return NULL;
181 }
182 if (map->used == map->alloc) {
183 // TODO: Alloc policy
184 map->alloc += 4;
185 map->table = m_renew(mp_map_elem_t, map->table, map->used, map->alloc);
186 mp_seq_clear(map->table, map->used, map->alloc, sizeof(*map->table));
187 }
188 mp_map_elem_t *elem = map->table + map->used++;
189 elem->key = index;
190 if (!mp_obj_is_qstr(index)) {
191 map->all_keys_are_qstrs = 0;
192 }
193 return elem;
194 #else
195 return NULL;
196 #endif
197 }
198
199 // map is a hash table (not an ordered array), so do a hash lookup
200
201 if (map->alloc == 0) {
202 if (lookup_kind == MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
203 mp_map_rehash(map);
204 } else {
205 return NULL;
206 }
207 }
208
209 // get hash of index, with fast path for common case of qstr
210 mp_uint_t hash;
211 if (mp_obj_is_qstr(index)) {
212 hash = qstr_hash(MP_OBJ_QSTR_VALUE(index));
213 } else {
214 hash = MP_OBJ_SMALL_INT_VALUE(mp_unary_op(MP_UNARY_OP_HASH, index));
215 }
216
217 size_t pos = hash % map->alloc;
218 size_t start_pos = pos;
219 mp_map_elem_t *avail_slot = NULL;
220 for (;;) {
221 mp_map_elem_t *slot = &map->table[pos];
222 if (slot->key == MP_OBJ_NULL) {
223 // found NULL slot, so index is not in table
224 if (lookup_kind == MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
225 map->used += 1;
226 if (avail_slot == NULL) {
227 avail_slot = slot;
228 }
229 avail_slot->key = index;
230 avail_slot->value = MP_OBJ_NULL;
231 if (!mp_obj_is_qstr(index)) {
232 map->all_keys_are_qstrs = 0;
233 }
234 return avail_slot;
235 } else {
236 return NULL;
237 }
238 } else if (slot->key == MP_OBJ_SENTINEL) {
239 // found deleted slot, remember for later
240 if (avail_slot == NULL) {
241 avail_slot = slot;
242 }
243 } else if (slot->key == index || (!compare_only_ptrs && mp_obj_equal(slot->key, index))) {
244 // found index
245 // Note: CPython does not replace the index; try x={True:'true'};x[1]='one';x
246 if (lookup_kind == MP_MAP_LOOKUP_REMOVE_IF_FOUND) {
247 // delete element in this slot
248 map->used--;
249 if (map->table[(pos + 1) % map->alloc].key == MP_OBJ_NULL) {
250 // optimisation if next slot is empty
251 slot->key = MP_OBJ_NULL;
252 } else {
253 slot->key = MP_OBJ_SENTINEL;
254 }
255 // keep slot->value so that caller can access it if needed
256 }
257 return slot;
258 }
259
260 // not yet found, keep searching in this table
261 pos = (pos + 1) % map->alloc;
262
263 if (pos == start_pos) {
264 // search got back to starting position, so index is not in table
265 if (lookup_kind == MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
266 if (avail_slot != NULL) {
267 // there was an available slot, so use that
268 map->used++;
269 avail_slot->key = index;
270 avail_slot->value = MP_OBJ_NULL;
271 if (!mp_obj_is_qstr(index)) {
272 map->all_keys_are_qstrs = 0;
273 }
274 return avail_slot;
275 } else {
276 // not enough room in table, rehash it
277 mp_map_rehash(map);
278 // restart the search for the new element
279 start_pos = pos = hash % map->alloc;
280 }
281 } else {
282 return NULL;
283 }
284 }
285 }
286}
287
288/******************************************************************************/
289/* set */
290
291#if MICROPY_PY_BUILTINS_SET
292
293void mp_set_init(mp_set_t *set, size_t n) {
294 set->alloc = n;
295 set->used = 0;
296 set->table = m_new0(mp_obj_t, set->alloc);
297}
298
299STATIC void mp_set_rehash(mp_set_t *set) {
300 size_t old_alloc = set->alloc;
301 mp_obj_t *old_table = set->table;
302 set->alloc = get_hash_alloc_greater_or_equal_to(set->alloc + 1);
303 set->used = 0;
304 set->table = m_new0(mp_obj_t, set->alloc);
305 for (size_t i = 0; i < old_alloc; i++) {
306 if (old_table[i] != MP_OBJ_NULL && old_table[i] != MP_OBJ_SENTINEL) {
307 mp_set_lookup(set, old_table[i], MP_MAP_LOOKUP_ADD_IF_NOT_FOUND);
308 }
309 }
310 m_del(mp_obj_t, old_table, old_alloc);
311}
312
313mp_obj_t mp_set_lookup(mp_set_t *set, mp_obj_t index, mp_map_lookup_kind_t lookup_kind) {
314 // Note: lookup_kind can be MP_MAP_LOOKUP_ADD_IF_NOT_FOUND_OR_REMOVE_IF_FOUND which
315 // is handled by using bitwise operations.
316
317 if (set->alloc == 0) {
318 if (lookup_kind & MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
319 mp_set_rehash(set);
320 } else {
321 return MP_OBJ_NULL;
322 }
323 }
324 mp_uint_t hash = MP_OBJ_SMALL_INT_VALUE(mp_unary_op(MP_UNARY_OP_HASH, index));
325 size_t pos = hash % set->alloc;
326 size_t start_pos = pos;
327 mp_obj_t *avail_slot = NULL;
328 for (;;) {
329 mp_obj_t elem = set->table[pos];
330 if (elem == MP_OBJ_NULL) {
331 // found NULL slot, so index is not in table
332 if (lookup_kind & MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
333 if (avail_slot == NULL) {
334 avail_slot = &set->table[pos];
335 }
336 set->used++;
337 *avail_slot = index;
338 return index;
339 } else {
340 return MP_OBJ_NULL;
341 }
342 } else if (elem == MP_OBJ_SENTINEL) {
343 // found deleted slot, remember for later
344 if (avail_slot == NULL) {
345 avail_slot = &set->table[pos];
346 }
347 } else if (mp_obj_equal(elem, index)) {
348 // found index
349 if (lookup_kind & MP_MAP_LOOKUP_REMOVE_IF_FOUND) {
350 // delete element
351 set->used--;
352 if (set->table[(pos + 1) % set->alloc] == MP_OBJ_NULL) {
353 // optimisation if next slot is empty
354 set->table[pos] = MP_OBJ_NULL;
355 } else {
356 set->table[pos] = MP_OBJ_SENTINEL;
357 }
358 }
359 return elem;
360 }
361
362 // not yet found, keep searching in this table
363 pos = (pos + 1) % set->alloc;
364
365 if (pos == start_pos) {
366 // search got back to starting position, so index is not in table
367 if (lookup_kind & MP_MAP_LOOKUP_ADD_IF_NOT_FOUND) {
368 if (avail_slot != NULL) {
369 // there was an available slot, so use that
370 set->used++;
371 *avail_slot = index;
372 return index;
373 } else {
374 // not enough room in table, rehash it
375 mp_set_rehash(set);
376 // restart the search for the new element
377 start_pos = pos = hash % set->alloc;
378 }
379 } else {
380 return MP_OBJ_NULL;
381 }
382 }
383 }
384}
385
386mp_obj_t mp_set_remove_first(mp_set_t *set) {
387 for (size_t pos = 0; pos < set->alloc; pos++) {
388 if (mp_set_slot_is_filled(set, pos)) {
389 mp_obj_t elem = set->table[pos];
390 // delete element
391 set->used--;
392 if (set->table[(pos + 1) % set->alloc] == MP_OBJ_NULL) {
393 // optimisation if next slot is empty
394 set->table[pos] = MP_OBJ_NULL;
395 } else {
396 set->table[pos] = MP_OBJ_SENTINEL;
397 }
398 return elem;
399 }
400 }
401 return MP_OBJ_NULL;
402}
403
404void mp_set_clear(mp_set_t *set) {
405 m_del(mp_obj_t, set->table, set->alloc);
406 set->alloc = 0;
407 set->used = 0;
408 set->table = NULL;
409}
410
411#endif // MICROPY_PY_BUILTINS_SET
412
413#if defined(DEBUG_PRINT) && DEBUG_PRINT
414void mp_map_dump(mp_map_t *map) {
415 for (size_t i = 0; i < map->alloc; i++) {
416 if (map->table[i].key != MP_OBJ_NULL) {
417 mp_obj_print(map->table[i].key, PRINT_REPR);
418 } else {
419 DEBUG_printf("(nil)");
420 }
421 DEBUG_printf(": %p\n", map->table[i].value);
422 }
423 DEBUG_printf("---\n");
424}
425#endif
426