1 | /* |
2 | ** $Id: ltable.c $ |
3 | ** Lua tables (hash) |
4 | ** See Copyright Notice in lua.h |
5 | */ |
6 | |
7 | #define ltable_c |
8 | #define LUA_CORE |
9 | |
10 | #include "lprefix.h" |
11 | |
12 | |
13 | /* |
14 | ** Implementation of tables (aka arrays, objects, or hash tables). |
15 | ** Tables keep its elements in two parts: an array part and a hash part. |
16 | ** Non-negative integer keys are all candidates to be kept in the array |
17 | ** part. The actual size of the array is the largest 'n' such that |
18 | ** more than half the slots between 1 and n are in use. |
19 | ** Hash uses a mix of chained scatter table with Brent's variation. |
20 | ** A main invariant of these tables is that, if an element is not |
21 | ** in its main position (i.e. the 'original' position that its hash gives |
22 | ** to it), then the colliding element is in its own main position. |
23 | ** Hence even when the load factor reaches 100%, performance remains good. |
24 | */ |
25 | |
26 | #include <math.h> |
27 | #include <limits.h> |
28 | |
29 | #include "lua.h" |
30 | |
31 | #include "ldebug.h" |
32 | #include "ldo.h" |
33 | #include "lgc.h" |
34 | #include "lmem.h" |
35 | #include "lobject.h" |
36 | #include "lstate.h" |
37 | #include "lstring.h" |
38 | #include "ltable.h" |
39 | #include "lvm.h" |
40 | |
41 | |
42 | /* |
43 | ** MAXABITS is the largest integer such that MAXASIZE fits in an |
44 | ** unsigned int. |
45 | */ |
46 | #define MAXABITS cast_int(sizeof(int) * CHAR_BIT - 1) |
47 | |
48 | |
49 | /* |
50 | ** MAXASIZE is the maximum size of the array part. It is the minimum |
51 | ** between 2^MAXABITS and the maximum size that, measured in bytes, |
52 | ** fits in a 'size_t'. |
53 | */ |
54 | #define MAXASIZE luaM_limitN(1u << MAXABITS, TValue) |
55 | |
56 | /* |
57 | ** MAXHBITS is the largest integer such that 2^MAXHBITS fits in a |
58 | ** signed int. |
59 | */ |
60 | #define MAXHBITS (MAXABITS - 1) |
61 | |
62 | |
63 | /* |
64 | ** MAXHSIZE is the maximum size of the hash part. It is the minimum |
65 | ** between 2^MAXHBITS and the maximum size such that, measured in bytes, |
66 | ** it fits in a 'size_t'. |
67 | */ |
68 | #define MAXHSIZE luaM_limitN(1u << MAXHBITS, Node) |
69 | |
70 | |
71 | #define hashpow2(t,n) (gnode(t, lmod((n), sizenode(t)))) |
72 | |
73 | #define hashstr(t,str) hashpow2(t, (str)->hash) |
74 | #define hashboolean(t,p) hashpow2(t, p) |
75 | #define hashint(t,i) hashpow2(t, i) |
76 | |
77 | |
78 | /* |
79 | ** for some types, it is better to avoid modulus by power of 2, as |
80 | ** they tend to have many 2 factors. |
81 | */ |
82 | #define hashmod(t,n) (gnode(t, ((n) % ((sizenode(t)-1)|1)))) |
83 | |
84 | |
85 | #define hashpointer(t,p) hashmod(t, point2uint(p)) |
86 | |
87 | |
88 | #define dummynode (&dummynode_) |
89 | |
90 | static const Node dummynode_ = { |
91 | {{NULL}, LUA_VEMPTY, /* value's value and type */ |
92 | LUA_VNIL, 0, {NULL}} /* key type, next, and key value */ |
93 | }; |
94 | |
95 | |
96 | static const TValue absentkey = {ABSTKEYCONSTANT}; |
97 | |
98 | |
99 | |
100 | /* |
101 | ** Hash for floating-point numbers. |
102 | ** The main computation should be just |
103 | ** n = frexp(n, &i); return (n * INT_MAX) + i |
104 | ** but there are some numerical subtleties. |
105 | ** In a two-complement representation, INT_MAX does not has an exact |
106 | ** representation as a float, but INT_MIN does; because the absolute |
107 | ** value of 'frexp' is smaller than 1 (unless 'n' is inf/NaN), the |
108 | ** absolute value of the product 'frexp * -INT_MIN' is smaller or equal |
109 | ** to INT_MAX. Next, the use of 'unsigned int' avoids overflows when |
110 | ** adding 'i'; the use of '~u' (instead of '-u') avoids problems with |
111 | ** INT_MIN. |
112 | */ |
113 | #if !defined(l_hashfloat) |
114 | static int l_hashfloat (lua_Number n) { |
115 | int i; |
116 | lua_Integer ni; |
117 | n = l_mathop(frexp)(n, &i) * -cast_num(INT_MIN); |
118 | if (!lua_numbertointeger(n, &ni)) { /* is 'n' inf/-inf/NaN? */ |
119 | lua_assert(luai_numisnan(n) || l_mathop(fabs)(n) == cast_num(HUGE_VAL)); |
120 | return 0; |
121 | } |
122 | else { /* normal case */ |
123 | unsigned int u = cast_uint(i) + cast_uint(ni); |
124 | return cast_int(u <= cast_uint(INT_MAX) ? u : ~u); |
125 | } |
126 | } |
127 | #endif |
128 | |
129 | |
130 | /* |
131 | ** returns the 'main' position of an element in a table (that is, |
132 | ** the index of its hash value). The key comes broken (tag in 'ktt' |
133 | ** and value in 'vkl') so that we can call it on keys inserted into |
134 | ** nodes. |
135 | */ |
136 | static Node *mainposition (const Table *t, int ktt, const Value *kvl) { |
137 | switch (withvariant(ktt)) { |
138 | case LUA_VNUMINT: |
139 | return hashint(t, ivalueraw(*kvl)); |
140 | case LUA_VNUMFLT: |
141 | return hashmod(t, l_hashfloat(fltvalueraw(*kvl))); |
142 | case LUA_VSHRSTR: |
143 | return hashstr(t, tsvalueraw(*kvl)); |
144 | case LUA_VLNGSTR: |
145 | return hashpow2(t, luaS_hashlongstr(tsvalueraw(*kvl))); |
146 | case LUA_VFALSE: |
147 | return hashboolean(t, 0); |
148 | case LUA_VTRUE: |
149 | return hashboolean(t, 1); |
150 | case LUA_VLIGHTUSERDATA: |
151 | return hashpointer(t, pvalueraw(*kvl)); |
152 | case LUA_VLCF: |
153 | return hashpointer(t, fvalueraw(*kvl)); |
154 | default: |
155 | return hashpointer(t, gcvalueraw(*kvl)); |
156 | } |
157 | } |
158 | |
159 | |
160 | /* |
161 | ** Returns the main position of an element given as a 'TValue' |
162 | */ |
163 | static Node *mainpositionTV (const Table *t, const TValue *key) { |
164 | return mainposition(t, rawtt(key), valraw(key)); |
165 | } |
166 | |
167 | |
168 | /* |
169 | ** Check whether key 'k1' is equal to the key in node 'n2'. |
170 | ** This equality is raw, so there are no metamethods. Floats |
171 | ** with integer values have been normalized, so integers cannot |
172 | ** be equal to floats. It is assumed that 'eqshrstr' is simply |
173 | ** pointer equality, so that short strings are handled in the |
174 | ** default case. |
175 | ** A true 'deadok' means to accept dead keys as equal to their original |
176 | ** values, which can only happen if the original key was collectable. |
177 | ** All dead values are compared in the default case, by pointer |
178 | ** identity. (Note that dead long strings are also compared by |
179 | ** identity). |
180 | */ |
181 | static int equalkey (const TValue *k1, const Node *n2, int deadok) { |
182 | if ((rawtt(k1) != keytt(n2)) && /* not the same variants? */ |
183 | !(deadok && keyisdead(n2) && iscollectable(k1))) |
184 | return 0; /* cannot be same key */ |
185 | switch (keytt(n2)) { |
186 | case LUA_VNIL: case LUA_VFALSE: case LUA_VTRUE: |
187 | return 1; |
188 | case LUA_VNUMINT: |
189 | return (ivalue(k1) == keyival(n2)); |
190 | case LUA_VNUMFLT: |
191 | return luai_numeq(fltvalue(k1), fltvalueraw(keyval(n2))); |
192 | case LUA_VLIGHTUSERDATA: |
193 | return pvalue(k1) == pvalueraw(keyval(n2)); |
194 | case LUA_VLCF: |
195 | return fvalue(k1) == fvalueraw(keyval(n2)); |
196 | case ctb(LUA_VLNGSTR): |
197 | return luaS_eqlngstr(tsvalue(k1), keystrval(n2)); |
198 | default: |
199 | return gcvalue(k1) == gcvalueraw(keyval(n2)); |
200 | } |
201 | } |
202 | |
203 | |
204 | /* |
205 | ** True if value of 'alimit' is equal to the real size of the array |
206 | ** part of table 't'. (Otherwise, the array part must be larger than |
207 | ** 'alimit'.) |
208 | */ |
209 | #define limitequalsasize(t) (isrealasize(t) || ispow2((t)->alimit)) |
210 | |
211 | |
212 | /* |
213 | ** Returns the real size of the 'array' array |
214 | */ |
215 | LUAI_FUNC unsigned int luaH_realasize (const Table *t) { |
216 | if (limitequalsasize(t)) |
217 | return t->alimit; /* this is the size */ |
218 | else { |
219 | unsigned int size = t->alimit; |
220 | /* compute the smallest power of 2 not smaller than 'n' */ |
221 | size |= (size >> 1); |
222 | size |= (size >> 2); |
223 | size |= (size >> 4); |
224 | size |= (size >> 8); |
225 | size |= (size >> 16); |
226 | #if (UINT_MAX >> 30) > 3 |
227 | size |= (size >> 32); /* unsigned int has more than 32 bits */ |
228 | #endif |
229 | size++; |
230 | lua_assert(ispow2(size) && size/2 < t->alimit && t->alimit < size); |
231 | return size; |
232 | } |
233 | } |
234 | |
235 | |
236 | /* |
237 | ** Check whether real size of the array is a power of 2. |
238 | ** (If it is not, 'alimit' cannot be changed to any other value |
239 | ** without changing the real size.) |
240 | */ |
241 | static int ispow2realasize (const Table *t) { |
242 | return (!isrealasize(t) || ispow2(t->alimit)); |
243 | } |
244 | |
245 | |
246 | static unsigned int setlimittosize (Table *t) { |
247 | t->alimit = luaH_realasize(t); |
248 | setrealasize(t); |
249 | return t->alimit; |
250 | } |
251 | |
252 | |
253 | #define limitasasize(t) check_exp(isrealasize(t), t->alimit) |
254 | |
255 | |
256 | |
257 | /* |
258 | ** "Generic" get version. (Not that generic: not valid for integers, |
259 | ** which may be in array part, nor for floats with integral values.) |
260 | ** See explanation about 'deadok' in function 'equalkey'. |
261 | */ |
262 | static const TValue *getgeneric (Table *t, const TValue *key, int deadok) { |
263 | Node *n = mainpositionTV(t, key); |
264 | for (;;) { /* check whether 'key' is somewhere in the chain */ |
265 | if (equalkey(key, n, deadok)) |
266 | return gval(n); /* that's it */ |
267 | else { |
268 | int nx = gnext(n); |
269 | if (nx == 0) |
270 | return &absentkey; /* not found */ |
271 | n += nx; |
272 | } |
273 | } |
274 | } |
275 | |
276 | |
277 | /* |
278 | ** returns the index for 'k' if 'k' is an appropriate key to live in |
279 | ** the array part of a table, 0 otherwise. |
280 | */ |
281 | static unsigned int arrayindex (lua_Integer k) { |
282 | if (l_castS2U(k) - 1u < MAXASIZE) /* 'k' in [1, MAXASIZE]? */ |
283 | return cast_uint(k); /* 'key' is an appropriate array index */ |
284 | else |
285 | return 0; |
286 | } |
287 | |
288 | |
289 | /* |
290 | ** returns the index of a 'key' for table traversals. First goes all |
291 | ** elements in the array part, then elements in the hash part. The |
292 | ** beginning of a traversal is signaled by 0. |
293 | */ |
294 | static unsigned int findindex (lua_State *L, Table *t, TValue *key, |
295 | unsigned int asize) { |
296 | unsigned int i; |
297 | if (ttisnil(key)) return 0; /* first iteration */ |
298 | i = ttisinteger(key) ? arrayindex(ivalue(key)) : 0; |
299 | if (i - 1u < asize) /* is 'key' inside array part? */ |
300 | return i; /* yes; that's the index */ |
301 | else { |
302 | const TValue *n = getgeneric(t, key, 1); |
303 | if (unlikely(isabstkey(n))) |
304 | luaG_runerror(L, "invalid key to 'next'" ); /* key not found */ |
305 | i = cast_int(nodefromval(n) - gnode(t, 0)); /* key index in hash table */ |
306 | /* hash elements are numbered after array ones */ |
307 | return (i + 1) + asize; |
308 | } |
309 | } |
310 | |
311 | |
312 | int luaH_next (lua_State *L, Table *t, StkId key) { |
313 | unsigned int asize = luaH_realasize(t); |
314 | unsigned int i = findindex(L, t, s2v(key), asize); /* find original key */ |
315 | for (; i < asize; i++) { /* try first array part */ |
316 | if (!isempty(&t->array[i])) { /* a non-empty entry? */ |
317 | setivalue(s2v(key), i + 1); |
318 | setobj2s(L, key + 1, &t->array[i]); |
319 | return 1; |
320 | } |
321 | } |
322 | for (i -= asize; cast_int(i) < sizenode(t); i++) { /* hash part */ |
323 | if (!isempty(gval(gnode(t, i)))) { /* a non-empty entry? */ |
324 | Node *n = gnode(t, i); |
325 | getnodekey(L, s2v(key), n); |
326 | setobj2s(L, key + 1, gval(n)); |
327 | return 1; |
328 | } |
329 | } |
330 | return 0; /* no more elements */ |
331 | } |
332 | |
333 | |
334 | static void freehash (lua_State *L, Table *t) { |
335 | if (!isdummy(t)) |
336 | luaM_freearray(L, t->node, cast_sizet(sizenode(t))); |
337 | } |
338 | |
339 | |
340 | /* |
341 | ** {============================================================= |
342 | ** Rehash |
343 | ** ============================================================== |
344 | */ |
345 | |
346 | /* |
347 | ** Compute the optimal size for the array part of table 't'. 'nums' is a |
348 | ** "count array" where 'nums[i]' is the number of integers in the table |
349 | ** between 2^(i - 1) + 1 and 2^i. 'pna' enters with the total number of |
350 | ** integer keys in the table and leaves with the number of keys that |
351 | ** will go to the array part; return the optimal size. (The condition |
352 | ** 'twotoi > 0' in the for loop stops the loop if 'twotoi' overflows.) |
353 | */ |
354 | static unsigned int computesizes (unsigned int nums[], unsigned int *pna) { |
355 | int i; |
356 | unsigned int twotoi; /* 2^i (candidate for optimal size) */ |
357 | unsigned int a = 0; /* number of elements smaller than 2^i */ |
358 | unsigned int na = 0; /* number of elements to go to array part */ |
359 | unsigned int optimal = 0; /* optimal size for array part */ |
360 | /* loop while keys can fill more than half of total size */ |
361 | for (i = 0, twotoi = 1; |
362 | twotoi > 0 && *pna > twotoi / 2; |
363 | i++, twotoi *= 2) { |
364 | a += nums[i]; |
365 | if (a > twotoi/2) { /* more than half elements present? */ |
366 | optimal = twotoi; /* optimal size (till now) */ |
367 | na = a; /* all elements up to 'optimal' will go to array part */ |
368 | } |
369 | } |
370 | lua_assert((optimal == 0 || optimal / 2 < na) && na <= optimal); |
371 | *pna = na; |
372 | return optimal; |
373 | } |
374 | |
375 | |
376 | static int countint (lua_Integer key, unsigned int *nums) { |
377 | unsigned int k = arrayindex(key); |
378 | if (k != 0) { /* is 'key' an appropriate array index? */ |
379 | nums[luaO_ceillog2(k)]++; /* count as such */ |
380 | return 1; |
381 | } |
382 | else |
383 | return 0; |
384 | } |
385 | |
386 | |
387 | /* |
388 | ** Count keys in array part of table 't': Fill 'nums[i]' with |
389 | ** number of keys that will go into corresponding slice and return |
390 | ** total number of non-nil keys. |
391 | */ |
392 | static unsigned int numusearray (const Table *t, unsigned int *nums) { |
393 | int lg; |
394 | unsigned int ttlg; /* 2^lg */ |
395 | unsigned int ause = 0; /* summation of 'nums' */ |
396 | unsigned int i = 1; /* count to traverse all array keys */ |
397 | unsigned int asize = limitasasize(t); /* real array size */ |
398 | /* traverse each slice */ |
399 | for (lg = 0, ttlg = 1; lg <= MAXABITS; lg++, ttlg *= 2) { |
400 | unsigned int lc = 0; /* counter */ |
401 | unsigned int lim = ttlg; |
402 | if (lim > asize) { |
403 | lim = asize; /* adjust upper limit */ |
404 | if (i > lim) |
405 | break; /* no more elements to count */ |
406 | } |
407 | /* count elements in range (2^(lg - 1), 2^lg] */ |
408 | for (; i <= lim; i++) { |
409 | if (!isempty(&t->array[i-1])) |
410 | lc++; |
411 | } |
412 | nums[lg] += lc; |
413 | ause += lc; |
414 | } |
415 | return ause; |
416 | } |
417 | |
418 | |
419 | static int numusehash (const Table *t, unsigned int *nums, unsigned int *pna) { |
420 | int totaluse = 0; /* total number of elements */ |
421 | int ause = 0; /* elements added to 'nums' (can go to array part) */ |
422 | int i = sizenode(t); |
423 | while (i--) { |
424 | Node *n = &t->node[i]; |
425 | if (!isempty(gval(n))) { |
426 | if (keyisinteger(n)) |
427 | ause += countint(keyival(n), nums); |
428 | totaluse++; |
429 | } |
430 | } |
431 | *pna += ause; |
432 | return totaluse; |
433 | } |
434 | |
435 | |
436 | /* |
437 | ** Creates an array for the hash part of a table with the given |
438 | ** size, or reuses the dummy node if size is zero. |
439 | ** The computation for size overflow is in two steps: the first |
440 | ** comparison ensures that the shift in the second one does not |
441 | ** overflow. |
442 | */ |
443 | static void setnodevector (lua_State *L, Table *t, unsigned int size) { |
444 | if (size == 0) { /* no elements to hash part? */ |
445 | t->node = cast(Node *, dummynode); /* use common 'dummynode' */ |
446 | t->lsizenode = 0; |
447 | t->lastfree = NULL; /* signal that it is using dummy node */ |
448 | } |
449 | else { |
450 | int i; |
451 | int lsize = luaO_ceillog2(size); |
452 | if (lsize > MAXHBITS || (1u << lsize) > MAXHSIZE) |
453 | luaG_runerror(L, "table overflow" ); |
454 | size = twoto(lsize); |
455 | t->node = luaM_newvector(L, size, Node); |
456 | for (i = 0; i < (int)size; i++) { |
457 | Node *n = gnode(t, i); |
458 | gnext(n) = 0; |
459 | setnilkey(n); |
460 | setempty(gval(n)); |
461 | } |
462 | t->lsizenode = cast_byte(lsize); |
463 | t->lastfree = gnode(t, size); /* all positions are free */ |
464 | } |
465 | } |
466 | |
467 | |
468 | /* |
469 | ** (Re)insert all elements from the hash part of 'ot' into table 't'. |
470 | */ |
471 | static void reinsert (lua_State *L, Table *ot, Table *t) { |
472 | int j; |
473 | int size = sizenode(ot); |
474 | for (j = 0; j < size; j++) { |
475 | Node *old = gnode(ot, j); |
476 | if (!isempty(gval(old))) { |
477 | /* doesn't need barrier/invalidate cache, as entry was |
478 | already present in the table */ |
479 | TValue k; |
480 | getnodekey(L, &k, old); |
481 | setobjt2t(L, luaH_set(L, t, &k), gval(old)); |
482 | } |
483 | } |
484 | } |
485 | |
486 | |
487 | /* |
488 | ** Exchange the hash part of 't1' and 't2'. |
489 | */ |
490 | static void exchangehashpart (Table *t1, Table *t2) { |
491 | lu_byte lsizenode = t1->lsizenode; |
492 | Node *node = t1->node; |
493 | Node *lastfree = t1->lastfree; |
494 | t1->lsizenode = t2->lsizenode; |
495 | t1->node = t2->node; |
496 | t1->lastfree = t2->lastfree; |
497 | t2->lsizenode = lsizenode; |
498 | t2->node = node; |
499 | t2->lastfree = lastfree; |
500 | } |
501 | |
502 | |
503 | /* |
504 | ** Resize table 't' for the new given sizes. Both allocations (for |
505 | ** the hash part and for the array part) can fail, which creates some |
506 | ** subtleties. If the first allocation, for the hash part, fails, an |
507 | ** error is raised and that is it. Otherwise, it copies the elements from |
508 | ** the shrinking part of the array (if it is shrinking) into the new |
509 | ** hash. Then it reallocates the array part. If that fails, the table |
510 | ** is in its original state; the function frees the new hash part and then |
511 | ** raises the allocation error. Otherwise, it sets the new hash part |
512 | ** into the table, initializes the new part of the array (if any) with |
513 | ** nils and reinserts the elements of the old hash back into the new |
514 | ** parts of the table. |
515 | */ |
516 | void luaH_resize (lua_State *L, Table *t, unsigned int newasize, |
517 | unsigned int nhsize) { |
518 | unsigned int i; |
519 | Table newt; /* to keep the new hash part */ |
520 | unsigned int oldasize = setlimittosize(t); |
521 | TValue *newarray; |
522 | /* create new hash part with appropriate size into 'newt' */ |
523 | setnodevector(L, &newt, nhsize); |
524 | if (newasize < oldasize) { /* will array shrink? */ |
525 | t->alimit = newasize; /* pretend array has new size... */ |
526 | exchangehashpart(t, &newt); /* and new hash */ |
527 | /* re-insert into the new hash the elements from vanishing slice */ |
528 | for (i = newasize; i < oldasize; i++) { |
529 | if (!isempty(&t->array[i])) |
530 | luaH_setint(L, t, i + 1, &t->array[i]); |
531 | } |
532 | t->alimit = oldasize; /* restore current size... */ |
533 | exchangehashpart(t, &newt); /* and hash (in case of errors) */ |
534 | } |
535 | /* allocate new array */ |
536 | newarray = luaM_reallocvector(L, t->array, oldasize, newasize, TValue); |
537 | if (unlikely(newarray == NULL && newasize > 0)) { /* allocation failed? */ |
538 | freehash(L, &newt); /* release new hash part */ |
539 | luaM_error(L); /* raise error (with array unchanged) */ |
540 | } |
541 | /* allocation ok; initialize new part of the array */ |
542 | exchangehashpart(t, &newt); /* 't' has the new hash ('newt' has the old) */ |
543 | t->array = newarray; /* set new array part */ |
544 | t->alimit = newasize; |
545 | for (i = oldasize; i < newasize; i++) /* clear new slice of the array */ |
546 | setempty(&t->array[i]); |
547 | /* re-insert elements from old hash part into new parts */ |
548 | reinsert(L, &newt, t); /* 'newt' now has the old hash */ |
549 | freehash(L, &newt); /* free old hash part */ |
550 | } |
551 | |
552 | |
553 | void luaH_resizearray (lua_State *L, Table *t, unsigned int nasize) { |
554 | int nsize = allocsizenode(t); |
555 | luaH_resize(L, t, nasize, nsize); |
556 | } |
557 | |
558 | /* |
559 | ** nums[i] = number of keys 'k' where 2^(i - 1) < k <= 2^i |
560 | */ |
561 | static void rehash (lua_State *L, Table *t, const TValue *ek) { |
562 | unsigned int asize; /* optimal size for array part */ |
563 | unsigned int na; /* number of keys in the array part */ |
564 | unsigned int nums[MAXABITS + 1]; |
565 | int i; |
566 | int totaluse; |
567 | for (i = 0; i <= MAXABITS; i++) nums[i] = 0; /* reset counts */ |
568 | setlimittosize(t); |
569 | na = numusearray(t, nums); /* count keys in array part */ |
570 | totaluse = na; /* all those keys are integer keys */ |
571 | totaluse += numusehash(t, nums, &na); /* count keys in hash part */ |
572 | /* count extra key */ |
573 | if (ttisinteger(ek)) |
574 | na += countint(ivalue(ek), nums); |
575 | totaluse++; |
576 | /* compute new size for array part */ |
577 | asize = computesizes(nums, &na); |
578 | /* resize the table to new computed sizes */ |
579 | luaH_resize(L, t, asize, totaluse - na); |
580 | } |
581 | |
582 | |
583 | |
584 | /* |
585 | ** }============================================================= |
586 | */ |
587 | |
588 | |
589 | Table *luaH_new (lua_State *L) { |
590 | GCObject *o = luaC_newobj(L, LUA_VTABLE, sizeof(Table)); |
591 | Table *t = gco2t(o); |
592 | t->metatable = NULL; |
593 | t->flags = cast_byte(maskflags); /* table has no metamethod fields */ |
594 | t->array = NULL; |
595 | t->alimit = 0; |
596 | setnodevector(L, t, 0); |
597 | return t; |
598 | } |
599 | |
600 | |
601 | void luaH_free (lua_State *L, Table *t) { |
602 | freehash(L, t); |
603 | luaM_freearray(L, t->array, luaH_realasize(t)); |
604 | luaM_free(L, t); |
605 | } |
606 | |
607 | |
608 | static Node *getfreepos (Table *t) { |
609 | if (!isdummy(t)) { |
610 | while (t->lastfree > t->node) { |
611 | t->lastfree--; |
612 | if (keyisnil(t->lastfree)) |
613 | return t->lastfree; |
614 | } |
615 | } |
616 | return NULL; /* could not find a free place */ |
617 | } |
618 | |
619 | |
620 | |
621 | /* |
622 | ** inserts a new key into a hash table; first, check whether key's main |
623 | ** position is free. If not, check whether colliding node is in its main |
624 | ** position or not: if it is not, move colliding node to an empty place and |
625 | ** put new key in its main position; otherwise (colliding node is in its main |
626 | ** position), new key goes to an empty position. |
627 | */ |
628 | TValue *luaH_newkey (lua_State *L, Table *t, const TValue *key) { |
629 | Node *mp; |
630 | TValue aux; |
631 | if (unlikely(ttisnil(key))) |
632 | luaG_runerror(L, "table index is nil" ); |
633 | else if (ttisfloat(key)) { |
634 | lua_Number f = fltvalue(key); |
635 | lua_Integer k; |
636 | if (luaV_flttointeger(f, &k, F2Ieq)) { /* does key fit in an integer? */ |
637 | setivalue(&aux, k); |
638 | key = &aux; /* insert it as an integer */ |
639 | } |
640 | else if (unlikely(luai_numisnan(f))) |
641 | luaG_runerror(L, "table index is NaN" ); |
642 | } |
643 | mp = mainpositionTV(t, key); |
644 | if (!isempty(gval(mp)) || isdummy(t)) { /* main position is taken? */ |
645 | Node *othern; |
646 | Node *f = getfreepos(t); /* get a free place */ |
647 | if (f == NULL) { /* cannot find a free place? */ |
648 | rehash(L, t, key); /* grow table */ |
649 | /* whatever called 'newkey' takes care of TM cache */ |
650 | return luaH_set(L, t, key); /* insert key into grown table */ |
651 | } |
652 | lua_assert(!isdummy(t)); |
653 | othern = mainposition(t, keytt(mp), &keyval(mp)); |
654 | if (othern != mp) { /* is colliding node out of its main position? */ |
655 | /* yes; move colliding node into free position */ |
656 | while (othern + gnext(othern) != mp) /* find previous */ |
657 | othern += gnext(othern); |
658 | gnext(othern) = cast_int(f - othern); /* rechain to point to 'f' */ |
659 | *f = *mp; /* copy colliding node into free pos. (mp->next also goes) */ |
660 | if (gnext(mp) != 0) { |
661 | gnext(f) += cast_int(mp - f); /* correct 'next' */ |
662 | gnext(mp) = 0; /* now 'mp' is free */ |
663 | } |
664 | setempty(gval(mp)); |
665 | } |
666 | else { /* colliding node is in its own main position */ |
667 | /* new node will go into free position */ |
668 | if (gnext(mp) != 0) |
669 | gnext(f) = cast_int((mp + gnext(mp)) - f); /* chain new position */ |
670 | else lua_assert(gnext(f) == 0); |
671 | gnext(mp) = cast_int(f - mp); |
672 | mp = f; |
673 | } |
674 | } |
675 | setnodekey(L, mp, key); |
676 | luaC_barrierback(L, obj2gco(t), key); |
677 | lua_assert(isempty(gval(mp))); |
678 | return gval(mp); |
679 | } |
680 | |
681 | |
682 | /* |
683 | ** Search function for integers. If integer is inside 'alimit', get it |
684 | ** directly from the array part. Otherwise, if 'alimit' is not equal to |
685 | ** the real size of the array, key still can be in the array part. In |
686 | ** this case, try to avoid a call to 'luaH_realasize' when key is just |
687 | ** one more than the limit (so that it can be incremented without |
688 | ** changing the real size of the array). |
689 | */ |
690 | const TValue *luaH_getint (Table *t, lua_Integer key) { |
691 | if (l_castS2U(key) - 1u < t->alimit) /* 'key' in [1, t->alimit]? */ |
692 | return &t->array[key - 1]; |
693 | else if (!limitequalsasize(t) && /* key still may be in the array part? */ |
694 | (l_castS2U(key) == t->alimit + 1 || |
695 | l_castS2U(key) - 1u < luaH_realasize(t))) { |
696 | t->alimit = cast_uint(key); /* probably '#t' is here now */ |
697 | return &t->array[key - 1]; |
698 | } |
699 | else { |
700 | Node *n = hashint(t, key); |
701 | for (;;) { /* check whether 'key' is somewhere in the chain */ |
702 | if (keyisinteger(n) && keyival(n) == key) |
703 | return gval(n); /* that's it */ |
704 | else { |
705 | int nx = gnext(n); |
706 | if (nx == 0) break; |
707 | n += nx; |
708 | } |
709 | } |
710 | return &absentkey; |
711 | } |
712 | } |
713 | |
714 | |
715 | /* |
716 | ** search function for short strings |
717 | */ |
718 | const TValue *luaH_getshortstr (Table *t, TString *key) { |
719 | Node *n = hashstr(t, key); |
720 | lua_assert(key->tt == LUA_VSHRSTR); |
721 | for (;;) { /* check whether 'key' is somewhere in the chain */ |
722 | if (keyisshrstr(n) && eqshrstr(keystrval(n), key)) |
723 | return gval(n); /* that's it */ |
724 | else { |
725 | int nx = gnext(n); |
726 | if (nx == 0) |
727 | return &absentkey; /* not found */ |
728 | n += nx; |
729 | } |
730 | } |
731 | } |
732 | |
733 | |
734 | const TValue *luaH_getstr (Table *t, TString *key) { |
735 | if (key->tt == LUA_VSHRSTR) |
736 | return luaH_getshortstr(t, key); |
737 | else { /* for long strings, use generic case */ |
738 | TValue ko; |
739 | setsvalue(cast(lua_State *, NULL), &ko, key); |
740 | return getgeneric(t, &ko, 0); |
741 | } |
742 | } |
743 | |
744 | |
745 | /* |
746 | ** main search function |
747 | */ |
748 | const TValue *luaH_get (Table *t, const TValue *key) { |
749 | switch (ttypetag(key)) { |
750 | case LUA_VSHRSTR: return luaH_getshortstr(t, tsvalue(key)); |
751 | case LUA_VNUMINT: return luaH_getint(t, ivalue(key)); |
752 | case LUA_VNIL: return &absentkey; |
753 | case LUA_VNUMFLT: { |
754 | lua_Integer k; |
755 | if (luaV_flttointeger(fltvalue(key), &k, F2Ieq)) /* integral index? */ |
756 | return luaH_getint(t, k); /* use specialized version */ |
757 | /* else... */ |
758 | } /* FALLTHROUGH */ |
759 | default: |
760 | return getgeneric(t, key, 0); |
761 | } |
762 | } |
763 | |
764 | |
765 | /* |
766 | ** beware: when using this function you probably need to check a GC |
767 | ** barrier and invalidate the TM cache. |
768 | */ |
769 | TValue *luaH_set (lua_State *L, Table *t, const TValue *key) { |
770 | const TValue *p = luaH_get(t, key); |
771 | if (!isabstkey(p)) |
772 | return cast(TValue *, p); |
773 | else return luaH_newkey(L, t, key); |
774 | } |
775 | |
776 | |
777 | void luaH_setint (lua_State *L, Table *t, lua_Integer key, TValue *value) { |
778 | const TValue *p = luaH_getint(t, key); |
779 | TValue *cell; |
780 | if (!isabstkey(p)) |
781 | cell = cast(TValue *, p); |
782 | else { |
783 | TValue k; |
784 | setivalue(&k, key); |
785 | cell = luaH_newkey(L, t, &k); |
786 | } |
787 | setobj2t(L, cell, value); |
788 | } |
789 | |
790 | |
791 | /* |
792 | ** Try to find a boundary in the hash part of table 't'. From the |
793 | ** caller, we know that 'j' is zero or present and that 'j + 1' is |
794 | ** present. We want to find a larger key that is absent from the |
795 | ** table, so that we can do a binary search between the two keys to |
796 | ** find a boundary. We keep doubling 'j' until we get an absent index. |
797 | ** If the doubling would overflow, we try LUA_MAXINTEGER. If it is |
798 | ** absent, we are ready for the binary search. ('j', being max integer, |
799 | ** is larger or equal to 'i', but it cannot be equal because it is |
800 | ** absent while 'i' is present; so 'j > i'.) Otherwise, 'j' is a |
801 | ** boundary. ('j + 1' cannot be a present integer key because it is |
802 | ** not a valid integer in Lua.) |
803 | */ |
804 | static lua_Unsigned hash_search (Table *t, lua_Unsigned j) { |
805 | lua_Unsigned i; |
806 | if (j == 0) j++; /* the caller ensures 'j + 1' is present */ |
807 | do { |
808 | i = j; /* 'i' is a present index */ |
809 | if (j <= l_castS2U(LUA_MAXINTEGER) / 2) |
810 | j *= 2; |
811 | else { |
812 | j = LUA_MAXINTEGER; |
813 | if (isempty(luaH_getint(t, j))) /* t[j] not present? */ |
814 | break; /* 'j' now is an absent index */ |
815 | else /* weird case */ |
816 | return j; /* well, max integer is a boundary... */ |
817 | } |
818 | } while (!isempty(luaH_getint(t, j))); /* repeat until an absent t[j] */ |
819 | /* i < j && t[i] present && t[j] absent */ |
820 | while (j - i > 1u) { /* do a binary search between them */ |
821 | lua_Unsigned m = (i + j) / 2; |
822 | if (isempty(luaH_getint(t, m))) j = m; |
823 | else i = m; |
824 | } |
825 | return i; |
826 | } |
827 | |
828 | |
829 | static unsigned int binsearch (const TValue *array, unsigned int i, |
830 | unsigned int j) { |
831 | while (j - i > 1u) { /* binary search */ |
832 | unsigned int m = (i + j) / 2; |
833 | if (isempty(&array[m - 1])) j = m; |
834 | else i = m; |
835 | } |
836 | return i; |
837 | } |
838 | |
839 | |
840 | /* |
841 | ** Try to find a boundary in table 't'. (A 'boundary' is an integer index |
842 | ** such that t[i] is present and t[i+1] is absent, or 0 if t[1] is absent |
843 | ** and 'maxinteger' if t[maxinteger] is present.) |
844 | ** (In the next explanation, we use Lua indices, that is, with base 1. |
845 | ** The code itself uses base 0 when indexing the array part of the table.) |
846 | ** The code starts with 'limit = t->alimit', a position in the array |
847 | ** part that may be a boundary. |
848 | ** |
849 | ** (1) If 't[limit]' is empty, there must be a boundary before it. |
850 | ** As a common case (e.g., after 't[#t]=nil'), check whether 'limit-1' |
851 | ** is present. If so, it is a boundary. Otherwise, do a binary search |
852 | ** between 0 and limit to find a boundary. In both cases, try to |
853 | ** use this boundary as the new 'alimit', as a hint for the next call. |
854 | ** |
855 | ** (2) If 't[limit]' is not empty and the array has more elements |
856 | ** after 'limit', try to find a boundary there. Again, try first |
857 | ** the special case (which should be quite frequent) where 'limit+1' |
858 | ** is empty, so that 'limit' is a boundary. Otherwise, check the |
859 | ** last element of the array part. If it is empty, there must be a |
860 | ** boundary between the old limit (present) and the last element |
861 | ** (absent), which is found with a binary search. (This boundary always |
862 | ** can be a new limit.) |
863 | ** |
864 | ** (3) The last case is when there are no elements in the array part |
865 | ** (limit == 0) or its last element (the new limit) is present. |
866 | ** In this case, must check the hash part. If there is no hash part |
867 | ** or 'limit+1' is absent, 'limit' is a boundary. Otherwise, call |
868 | ** 'hash_search' to find a boundary in the hash part of the table. |
869 | ** (In those cases, the boundary is not inside the array part, and |
870 | ** therefore cannot be used as a new limit.) |
871 | */ |
872 | lua_Unsigned luaH_getn (Table *t) { |
873 | unsigned int limit = t->alimit; |
874 | if (limit > 0 && isempty(&t->array[limit - 1])) { /* (1)? */ |
875 | /* there must be a boundary before 'limit' */ |
876 | if (limit >= 2 && !isempty(&t->array[limit - 2])) { |
877 | /* 'limit - 1' is a boundary; can it be a new limit? */ |
878 | if (ispow2realasize(t) && !ispow2(limit - 1)) { |
879 | t->alimit = limit - 1; |
880 | setnorealasize(t); /* now 'alimit' is not the real size */ |
881 | } |
882 | return limit - 1; |
883 | } |
884 | else { /* must search for a boundary in [0, limit] */ |
885 | unsigned int boundary = binsearch(t->array, 0, limit); |
886 | /* can this boundary represent the real size of the array? */ |
887 | if (ispow2realasize(t) && boundary > luaH_realasize(t) / 2) { |
888 | t->alimit = boundary; /* use it as the new limit */ |
889 | setnorealasize(t); |
890 | } |
891 | return boundary; |
892 | } |
893 | } |
894 | /* 'limit' is zero or present in table */ |
895 | if (!limitequalsasize(t)) { /* (2)? */ |
896 | /* 'limit' > 0 and array has more elements after 'limit' */ |
897 | if (isempty(&t->array[limit])) /* 'limit + 1' is empty? */ |
898 | return limit; /* this is the boundary */ |
899 | /* else, try last element in the array */ |
900 | limit = luaH_realasize(t); |
901 | if (isempty(&t->array[limit - 1])) { /* empty? */ |
902 | /* there must be a boundary in the array after old limit, |
903 | and it must be a valid new limit */ |
904 | unsigned int boundary = binsearch(t->array, t->alimit, limit); |
905 | t->alimit = boundary; |
906 | return boundary; |
907 | } |
908 | /* else, new limit is present in the table; check the hash part */ |
909 | } |
910 | /* (3) 'limit' is the last element and either is zero or present in table */ |
911 | lua_assert(limit == luaH_realasize(t) && |
912 | (limit == 0 || !isempty(&t->array[limit - 1]))); |
913 | if (isdummy(t) || isempty(luaH_getint(t, cast(lua_Integer, limit + 1)))) |
914 | return limit; /* 'limit + 1' is absent */ |
915 | else /* 'limit + 1' is also present */ |
916 | return hash_search(t, limit); |
917 | } |
918 | |
919 | |
920 | |
921 | #if defined(LUA_DEBUG) |
922 | |
923 | /* export these functions for the test library */ |
924 | |
925 | Node *luaH_mainposition (const Table *t, const TValue *key) { |
926 | return mainpositionTV(t, key); |
927 | } |
928 | |
929 | int luaH_isdummy (const Table *t) { return isdummy(t); } |
930 | |
931 | #endif |
932 | |