xref: /freebsd/contrib/lua/src/ltable.c (revision a9490b81)
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 /*
72 ** When the original hash value is good, hashing by a power of 2
73 ** avoids the cost of '%'.
74 */
75 #define hashpow2(t,n)		(gnode(t, lmod((n), sizenode(t))))
76 
77 /*
78 ** for other types, it is better to avoid modulo by power of 2, as
79 ** they can have many 2 factors.
80 */
81 #define hashmod(t,n)	(gnode(t, ((n) % ((sizenode(t)-1)|1))))
82 
83 
84 #define hashstr(t,str)		hashpow2(t, (str)->hash)
85 #define hashboolean(t,p)	hashpow2(t, p)
86 
87 
88 #define hashpointer(t,p)	hashmod(t, point2uint(p))
89 
90 
91 #define dummynode		(&dummynode_)
92 
93 static const Node dummynode_ = {
94   {{NULL}, LUA_VEMPTY,  /* value's value and type */
95    LUA_VNIL, 0, {NULL}}  /* key type, next, and key value */
96 };
97 
98 
99 static const TValue absentkey = {ABSTKEYCONSTANT};
100 
101 
102 /*
103 ** Hash for integers. To allow a good hash, use the remainder operator
104 ** ('%'). If integer fits as a non-negative int, compute an int
105 ** remainder, which is faster. Otherwise, use an unsigned-integer
106 ** remainder, which uses all bits and ensures a non-negative result.
107 */
hashint(const Table * t,lua_Integer i)108 static Node *hashint (const Table *t, lua_Integer i) {
109   lua_Unsigned ui = l_castS2U(i);
110   if (ui <= cast_uint(INT_MAX))
111     return hashmod(t, cast_int(ui));
112   else
113     return hashmod(t, ui);
114 }
115 
116 
117 /*
118 ** Hash for floating-point numbers.
119 ** The main computation should be just
120 **     n = frexp(n, &i); return (n * INT_MAX) + i
121 ** but there are some numerical subtleties.
122 ** In a two-complement representation, INT_MAX does not has an exact
123 ** representation as a float, but INT_MIN does; because the absolute
124 ** value of 'frexp' is smaller than 1 (unless 'n' is inf/NaN), the
125 ** absolute value of the product 'frexp * -INT_MIN' is smaller or equal
126 ** to INT_MAX. Next, the use of 'unsigned int' avoids overflows when
127 ** adding 'i'; the use of '~u' (instead of '-u') avoids problems with
128 ** INT_MIN.
129 */
130 #if !defined(l_hashfloat)
l_hashfloat(lua_Number n)131 static int l_hashfloat (lua_Number n) {
132   int i;
133   lua_Integer ni;
134   n = l_mathop(frexp)(n, &i) * -cast_num(INT_MIN);
135   if (!lua_numbertointeger(n, &ni)) {  /* is 'n' inf/-inf/NaN? */
136     lua_assert(luai_numisnan(n) || l_mathop(fabs)(n) == cast_num(HUGE_VAL));
137     return 0;
138   }
139   else {  /* normal case */
140     unsigned int u = cast_uint(i) + cast_uint(ni);
141     return cast_int(u <= cast_uint(INT_MAX) ? u : ~u);
142   }
143 }
144 #endif
145 
146 
147 /*
148 ** returns the 'main' position of an element in a table (that is,
149 ** the index of its hash value).
150 */
mainpositionTV(const Table * t,const TValue * key)151 static Node *mainpositionTV (const Table *t, const TValue *key) {
152   switch (ttypetag(key)) {
153     case LUA_VNUMINT: {
154       lua_Integer i = ivalue(key);
155       return hashint(t, i);
156     }
157     case LUA_VNUMFLT: {
158       lua_Number n = fltvalue(key);
159       return hashmod(t, l_hashfloat(n));
160     }
161     case LUA_VSHRSTR: {
162       TString *ts = tsvalue(key);
163       return hashstr(t, ts);
164     }
165     case LUA_VLNGSTR: {
166       TString *ts = tsvalue(key);
167       return hashpow2(t, luaS_hashlongstr(ts));
168     }
169     case LUA_VFALSE:
170       return hashboolean(t, 0);
171     case LUA_VTRUE:
172       return hashboolean(t, 1);
173     case LUA_VLIGHTUSERDATA: {
174       void *p = pvalue(key);
175       return hashpointer(t, p);
176     }
177     case LUA_VLCF: {
178       lua_CFunction f = fvalue(key);
179       return hashpointer(t, f);
180     }
181     default: {
182       GCObject *o = gcvalue(key);
183       return hashpointer(t, o);
184     }
185   }
186 }
187 
188 
mainpositionfromnode(const Table * t,Node * nd)189 l_sinline Node *mainpositionfromnode (const Table *t, Node *nd) {
190   TValue key;
191   getnodekey(cast(lua_State *, NULL), &key, nd);
192   return mainpositionTV(t, &key);
193 }
194 
195 
196 /*
197 ** Check whether key 'k1' is equal to the key in node 'n2'. This
198 ** equality is raw, so there are no metamethods. Floats with integer
199 ** values have been normalized, so integers cannot be equal to
200 ** floats. It is assumed that 'eqshrstr' is simply pointer equality, so
201 ** that short strings are handled in the default case.
202 ** A true 'deadok' means to accept dead keys as equal to their original
203 ** values. All dead keys are compared in the default case, by pointer
204 ** identity. (Only collectable objects can produce dead keys.) Note that
205 ** dead long strings are also compared by identity.
206 ** Once a key is dead, its corresponding value may be collected, and
207 ** then another value can be created with the same address. If this
208 ** other value is given to 'next', 'equalkey' will signal a false
209 ** positive. In a regular traversal, this situation should never happen,
210 ** as all keys given to 'next' came from the table itself, and therefore
211 ** could not have been collected. Outside a regular traversal, we
212 ** have garbage in, garbage out. What is relevant is that this false
213 ** positive does not break anything.  (In particular, 'next' will return
214 ** some other valid item on the table or nil.)
215 */
equalkey(const TValue * k1,const Node * n2,int deadok)216 static int equalkey (const TValue *k1, const Node *n2, int deadok) {
217   if ((rawtt(k1) != keytt(n2)) &&  /* not the same variants? */
218        !(deadok && keyisdead(n2) && iscollectable(k1)))
219    return 0;  /* cannot be same key */
220   switch (keytt(n2)) {
221     case LUA_VNIL: case LUA_VFALSE: case LUA_VTRUE:
222       return 1;
223     case LUA_VNUMINT:
224       return (ivalue(k1) == keyival(n2));
225     case LUA_VNUMFLT:
226       return luai_numeq(fltvalue(k1), fltvalueraw(keyval(n2)));
227     case LUA_VLIGHTUSERDATA:
228       return pvalue(k1) == pvalueraw(keyval(n2));
229     case LUA_VLCF:
230       return fvalue(k1) == fvalueraw(keyval(n2));
231     case ctb(LUA_VLNGSTR):
232       return luaS_eqlngstr(tsvalue(k1), keystrval(n2));
233     default:
234       return gcvalue(k1) == gcvalueraw(keyval(n2));
235   }
236 }
237 
238 
239 /*
240 ** True if value of 'alimit' is equal to the real size of the array
241 ** part of table 't'. (Otherwise, the array part must be larger than
242 ** 'alimit'.)
243 */
244 #define limitequalsasize(t)	(isrealasize(t) || ispow2((t)->alimit))
245 
246 
247 /*
248 ** Returns the real size of the 'array' array
249 */
luaH_realasize(const Table * t)250 LUAI_FUNC unsigned int luaH_realasize (const Table *t) {
251   if (limitequalsasize(t))
252     return t->alimit;  /* this is the size */
253   else {
254     unsigned int size = t->alimit;
255     /* compute the smallest power of 2 not smaller than 'n' */
256     size |= (size >> 1);
257     size |= (size >> 2);
258     size |= (size >> 4);
259     size |= (size >> 8);
260 #if (UINT_MAX >> 14) > 3  /* unsigned int has more than 16 bits */
261     size |= (size >> 16);
262 #if (UINT_MAX >> 30) > 3
263     size |= (size >> 32);  /* unsigned int has more than 32 bits */
264 #endif
265 #endif
266     size++;
267     lua_assert(ispow2(size) && size/2 < t->alimit && t->alimit < size);
268     return size;
269   }
270 }
271 
272 
273 /*
274 ** Check whether real size of the array is a power of 2.
275 ** (If it is not, 'alimit' cannot be changed to any other value
276 ** without changing the real size.)
277 */
ispow2realasize(const Table * t)278 static int ispow2realasize (const Table *t) {
279   return (!isrealasize(t) || ispow2(t->alimit));
280 }
281 
282 
setlimittosize(Table * t)283 static unsigned int setlimittosize (Table *t) {
284   t->alimit = luaH_realasize(t);
285   setrealasize(t);
286   return t->alimit;
287 }
288 
289 
290 #define limitasasize(t)	check_exp(isrealasize(t), t->alimit)
291 
292 
293 
294 /*
295 ** "Generic" get version. (Not that generic: not valid for integers,
296 ** which may be in array part, nor for floats with integral values.)
297 ** See explanation about 'deadok' in function 'equalkey'.
298 */
getgeneric(Table * t,const TValue * key,int deadok)299 static const TValue *getgeneric (Table *t, const TValue *key, int deadok) {
300   Node *n = mainpositionTV(t, key);
301   for (;;) {  /* check whether 'key' is somewhere in the chain */
302     if (equalkey(key, n, deadok))
303       return gval(n);  /* that's it */
304     else {
305       int nx = gnext(n);
306       if (nx == 0)
307         return &absentkey;  /* not found */
308       n += nx;
309     }
310   }
311 }
312 
313 
314 /*
315 ** returns the index for 'k' if 'k' is an appropriate key to live in
316 ** the array part of a table, 0 otherwise.
317 */
arrayindex(lua_Integer k)318 static unsigned int arrayindex (lua_Integer k) {
319   if (l_castS2U(k) - 1u < MAXASIZE)  /* 'k' in [1, MAXASIZE]? */
320     return cast_uint(k);  /* 'key' is an appropriate array index */
321   else
322     return 0;
323 }
324 
325 
326 /*
327 ** returns the index of a 'key' for table traversals. First goes all
328 ** elements in the array part, then elements in the hash part. The
329 ** beginning of a traversal is signaled by 0.
330 */
findindex(lua_State * L,Table * t,TValue * key,unsigned int asize)331 static unsigned int findindex (lua_State *L, Table *t, TValue *key,
332                                unsigned int asize) {
333   unsigned int i;
334   if (ttisnil(key)) return 0;  /* first iteration */
335   i = ttisinteger(key) ? arrayindex(ivalue(key)) : 0;
336   if (i - 1u < asize)  /* is 'key' inside array part? */
337     return i;  /* yes; that's the index */
338   else {
339     const TValue *n = getgeneric(t, key, 1);
340     if (l_unlikely(isabstkey(n)))
341       luaG_runerror(L, "invalid key to 'next'");  /* key not found */
342     i = cast_int(nodefromval(n) - gnode(t, 0));  /* key index in hash table */
343     /* hash elements are numbered after array ones */
344     return (i + 1) + asize;
345   }
346 }
347 
348 
luaH_next(lua_State * L,Table * t,StkId key)349 int luaH_next (lua_State *L, Table *t, StkId key) {
350   unsigned int asize = luaH_realasize(t);
351   unsigned int i = findindex(L, t, s2v(key), asize);  /* find original key */
352   for (; i < asize; i++) {  /* try first array part */
353     if (!isempty(&t->array[i])) {  /* a non-empty entry? */
354       setivalue(s2v(key), i + 1);
355       setobj2s(L, key + 1, &t->array[i]);
356       return 1;
357     }
358   }
359   for (i -= asize; cast_int(i) < sizenode(t); i++) {  /* hash part */
360     if (!isempty(gval(gnode(t, i)))) {  /* a non-empty entry? */
361       Node *n = gnode(t, i);
362       getnodekey(L, s2v(key), n);
363       setobj2s(L, key + 1, gval(n));
364       return 1;
365     }
366   }
367   return 0;  /* no more elements */
368 }
369 
370 
freehash(lua_State * L,Table * t)371 static void freehash (lua_State *L, Table *t) {
372   if (!isdummy(t))
373     luaM_freearray(L, t->node, cast_sizet(sizenode(t)));
374 }
375 
376 
377 /*
378 ** {=============================================================
379 ** Rehash
380 ** ==============================================================
381 */
382 
383 /*
384 ** Compute the optimal size for the array part of table 't'. 'nums' is a
385 ** "count array" where 'nums[i]' is the number of integers in the table
386 ** between 2^(i - 1) + 1 and 2^i. 'pna' enters with the total number of
387 ** integer keys in the table and leaves with the number of keys that
388 ** will go to the array part; return the optimal size.  (The condition
389 ** 'twotoi > 0' in the for loop stops the loop if 'twotoi' overflows.)
390 */
computesizes(unsigned int nums[],unsigned int * pna)391 static unsigned int computesizes (unsigned int nums[], unsigned int *pna) {
392   int i;
393   unsigned int twotoi;  /* 2^i (candidate for optimal size) */
394   unsigned int a = 0;  /* number of elements smaller than 2^i */
395   unsigned int na = 0;  /* number of elements to go to array part */
396   unsigned int optimal = 0;  /* optimal size for array part */
397   /* loop while keys can fill more than half of total size */
398   for (i = 0, twotoi = 1;
399        twotoi > 0 && *pna > twotoi / 2;
400        i++, twotoi *= 2) {
401     a += nums[i];
402     if (a > twotoi/2) {  /* more than half elements present? */
403       optimal = twotoi;  /* optimal size (till now) */
404       na = a;  /* all elements up to 'optimal' will go to array part */
405     }
406   }
407   lua_assert((optimal == 0 || optimal / 2 < na) && na <= optimal);
408   *pna = na;
409   return optimal;
410 }
411 
412 
countint(lua_Integer key,unsigned int * nums)413 static int countint (lua_Integer key, unsigned int *nums) {
414   unsigned int k = arrayindex(key);
415   if (k != 0) {  /* is 'key' an appropriate array index? */
416     nums[luaO_ceillog2(k)]++;  /* count as such */
417     return 1;
418   }
419   else
420     return 0;
421 }
422 
423 
424 /*
425 ** Count keys in array part of table 't': Fill 'nums[i]' with
426 ** number of keys that will go into corresponding slice and return
427 ** total number of non-nil keys.
428 */
numusearray(const Table * t,unsigned int * nums)429 static unsigned int numusearray (const Table *t, unsigned int *nums) {
430   int lg;
431   unsigned int ttlg;  /* 2^lg */
432   unsigned int ause = 0;  /* summation of 'nums' */
433   unsigned int i = 1;  /* count to traverse all array keys */
434   unsigned int asize = limitasasize(t);  /* real array size */
435   /* traverse each slice */
436   for (lg = 0, ttlg = 1; lg <= MAXABITS; lg++, ttlg *= 2) {
437     unsigned int lc = 0;  /* counter */
438     unsigned int lim = ttlg;
439     if (lim > asize) {
440       lim = asize;  /* adjust upper limit */
441       if (i > lim)
442         break;  /* no more elements to count */
443     }
444     /* count elements in range (2^(lg - 1), 2^lg] */
445     for (; i <= lim; i++) {
446       if (!isempty(&t->array[i-1]))
447         lc++;
448     }
449     nums[lg] += lc;
450     ause += lc;
451   }
452   return ause;
453 }
454 
455 
numusehash(const Table * t,unsigned int * nums,unsigned int * pna)456 static int numusehash (const Table *t, unsigned int *nums, unsigned int *pna) {
457   int totaluse = 0;  /* total number of elements */
458   int ause = 0;  /* elements added to 'nums' (can go to array part) */
459   int i = sizenode(t);
460   while (i--) {
461     Node *n = &t->node[i];
462     if (!isempty(gval(n))) {
463       if (keyisinteger(n))
464         ause += countint(keyival(n), nums);
465       totaluse++;
466     }
467   }
468   *pna += ause;
469   return totaluse;
470 }
471 
472 
473 /*
474 ** Creates an array for the hash part of a table with the given
475 ** size, or reuses the dummy node if size is zero.
476 ** The computation for size overflow is in two steps: the first
477 ** comparison ensures that the shift in the second one does not
478 ** overflow.
479 */
setnodevector(lua_State * L,Table * t,unsigned int size)480 static void setnodevector (lua_State *L, Table *t, unsigned int size) {
481   if (size == 0) {  /* no elements to hash part? */
482     t->node = cast(Node *, dummynode);  /* use common 'dummynode' */
483     t->lsizenode = 0;
484     t->lastfree = NULL;  /* signal that it is using dummy node */
485   }
486   else {
487     int i;
488     int lsize = luaO_ceillog2(size);
489     if (lsize > MAXHBITS || (1u << lsize) > MAXHSIZE)
490       luaG_runerror(L, "table overflow");
491     size = twoto(lsize);
492     t->node = luaM_newvector(L, size, Node);
493     for (i = 0; i < cast_int(size); i++) {
494       Node *n = gnode(t, i);
495       gnext(n) = 0;
496       setnilkey(n);
497       setempty(gval(n));
498     }
499     t->lsizenode = cast_byte(lsize);
500     t->lastfree = gnode(t, size);  /* all positions are free */
501   }
502 }
503 
504 
505 /*
506 ** (Re)insert all elements from the hash part of 'ot' into table 't'.
507 */
reinsert(lua_State * L,Table * ot,Table * t)508 static void reinsert (lua_State *L, Table *ot, Table *t) {
509   int j;
510   int size = sizenode(ot);
511   for (j = 0; j < size; j++) {
512     Node *old = gnode(ot, j);
513     if (!isempty(gval(old))) {
514       /* doesn't need barrier/invalidate cache, as entry was
515          already present in the table */
516       TValue k;
517       getnodekey(L, &k, old);
518       luaH_set(L, t, &k, gval(old));
519     }
520   }
521 }
522 
523 
524 /*
525 ** Exchange the hash part of 't1' and 't2'.
526 */
exchangehashpart(Table * t1,Table * t2)527 static void exchangehashpart (Table *t1, Table *t2) {
528   lu_byte lsizenode = t1->lsizenode;
529   Node *node = t1->node;
530   Node *lastfree = t1->lastfree;
531   t1->lsizenode = t2->lsizenode;
532   t1->node = t2->node;
533   t1->lastfree = t2->lastfree;
534   t2->lsizenode = lsizenode;
535   t2->node = node;
536   t2->lastfree = lastfree;
537 }
538 
539 
540 /*
541 ** Resize table 't' for the new given sizes. Both allocations (for
542 ** the hash part and for the array part) can fail, which creates some
543 ** subtleties. If the first allocation, for the hash part, fails, an
544 ** error is raised and that is it. Otherwise, it copies the elements from
545 ** the shrinking part of the array (if it is shrinking) into the new
546 ** hash. Then it reallocates the array part.  If that fails, the table
547 ** is in its original state; the function frees the new hash part and then
548 ** raises the allocation error. Otherwise, it sets the new hash part
549 ** into the table, initializes the new part of the array (if any) with
550 ** nils and reinserts the elements of the old hash back into the new
551 ** parts of the table.
552 */
luaH_resize(lua_State * L,Table * t,unsigned int newasize,unsigned int nhsize)553 void luaH_resize (lua_State *L, Table *t, unsigned int newasize,
554                                           unsigned int nhsize) {
555   unsigned int i;
556   Table newt;  /* to keep the new hash part */
557   unsigned int oldasize = setlimittosize(t);
558   TValue *newarray;
559   /* create new hash part with appropriate size into 'newt' */
560   setnodevector(L, &newt, nhsize);
561   if (newasize < oldasize) {  /* will array shrink? */
562     t->alimit = newasize;  /* pretend array has new size... */
563     exchangehashpart(t, &newt);  /* and new hash */
564     /* re-insert into the new hash the elements from vanishing slice */
565     for (i = newasize; i < oldasize; i++) {
566       if (!isempty(&t->array[i]))
567         luaH_setint(L, t, i + 1, &t->array[i]);
568     }
569     t->alimit = oldasize;  /* restore current size... */
570     exchangehashpart(t, &newt);  /* and hash (in case of errors) */
571   }
572   /* allocate new array */
573   newarray = luaM_reallocvector(L, t->array, oldasize, newasize, TValue);
574   if (l_unlikely(newarray == NULL && newasize > 0)) {  /* allocation failed? */
575     freehash(L, &newt);  /* release new hash part */
576     luaM_error(L);  /* raise error (with array unchanged) */
577   }
578   /* allocation ok; initialize new part of the array */
579   exchangehashpart(t, &newt);  /* 't' has the new hash ('newt' has the old) */
580   t->array = newarray;  /* set new array part */
581   t->alimit = newasize;
582   for (i = oldasize; i < newasize; i++)  /* clear new slice of the array */
583      setempty(&t->array[i]);
584   /* re-insert elements from old hash part into new parts */
585   reinsert(L, &newt, t);  /* 'newt' now has the old hash */
586   freehash(L, &newt);  /* free old hash part */
587 }
588 
589 
luaH_resizearray(lua_State * L,Table * t,unsigned int nasize)590 void luaH_resizearray (lua_State *L, Table *t, unsigned int nasize) {
591   int nsize = allocsizenode(t);
592   luaH_resize(L, t, nasize, nsize);
593 }
594 
595 /*
596 ** nums[i] = number of keys 'k' where 2^(i - 1) < k <= 2^i
597 */
rehash(lua_State * L,Table * t,const TValue * ek)598 static void rehash (lua_State *L, Table *t, const TValue *ek) {
599   unsigned int asize;  /* optimal size for array part */
600   unsigned int na;  /* number of keys in the array part */
601   unsigned int nums[MAXABITS + 1];
602   int i;
603   int totaluse;
604   for (i = 0; i <= MAXABITS; i++) nums[i] = 0;  /* reset counts */
605   setlimittosize(t);
606   na = numusearray(t, nums);  /* count keys in array part */
607   totaluse = na;  /* all those keys are integer keys */
608   totaluse += numusehash(t, nums, &na);  /* count keys in hash part */
609   /* count extra key */
610   if (ttisinteger(ek))
611     na += countint(ivalue(ek), nums);
612   totaluse++;
613   /* compute new size for array part */
614   asize = computesizes(nums, &na);
615   /* resize the table to new computed sizes */
616   luaH_resize(L, t, asize, totaluse - na);
617 }
618 
619 
620 
621 /*
622 ** }=============================================================
623 */
624 
625 
luaH_new(lua_State * L)626 Table *luaH_new (lua_State *L) {
627   GCObject *o = luaC_newobj(L, LUA_VTABLE, sizeof(Table));
628   Table *t = gco2t(o);
629   t->metatable = NULL;
630   t->flags = cast_byte(maskflags);  /* table has no metamethod fields */
631   t->array = NULL;
632   t->alimit = 0;
633   setnodevector(L, t, 0);
634   return t;
635 }
636 
637 
luaH_free(lua_State * L,Table * t)638 void luaH_free (lua_State *L, Table *t) {
639   freehash(L, t);
640   luaM_freearray(L, t->array, luaH_realasize(t));
641   luaM_free(L, t);
642 }
643 
644 
getfreepos(Table * t)645 static Node *getfreepos (Table *t) {
646   if (!isdummy(t)) {
647     while (t->lastfree > t->node) {
648       t->lastfree--;
649       if (keyisnil(t->lastfree))
650         return t->lastfree;
651     }
652   }
653   return NULL;  /* could not find a free place */
654 }
655 
656 
657 
658 /*
659 ** inserts a new key into a hash table; first, check whether key's main
660 ** position is free. If not, check whether colliding node is in its main
661 ** position or not: if it is not, move colliding node to an empty place and
662 ** put new key in its main position; otherwise (colliding node is in its main
663 ** position), new key goes to an empty position.
664 */
luaH_newkey(lua_State * L,Table * t,const TValue * key,TValue * value)665 void luaH_newkey (lua_State *L, Table *t, const TValue *key, TValue *value) {
666   Node *mp;
667   TValue aux;
668   if (l_unlikely(ttisnil(key)))
669     luaG_runerror(L, "table index is nil");
670   else if (ttisfloat(key)) {
671     lua_Number f = fltvalue(key);
672     lua_Integer k;
673     if (luaV_flttointeger(f, &k, F2Ieq)) {  /* does key fit in an integer? */
674       setivalue(&aux, k);
675       key = &aux;  /* insert it as an integer */
676     }
677     else if (l_unlikely(luai_numisnan(f)))
678       luaG_runerror(L, "table index is NaN");
679   }
680   if (ttisnil(value))
681     return;  /* do not insert nil values */
682   mp = mainpositionTV(t, key);
683   if (!isempty(gval(mp)) || isdummy(t)) {  /* main position is taken? */
684     Node *othern;
685     Node *f = getfreepos(t);  /* get a free place */
686     if (f == NULL) {  /* cannot find a free place? */
687       rehash(L, t, key);  /* grow table */
688       /* whatever called 'newkey' takes care of TM cache */
689       luaH_set(L, t, key, value);  /* insert key into grown table */
690       return;
691     }
692     lua_assert(!isdummy(t));
693     othern = mainpositionfromnode(t, mp);
694     if (othern != mp) {  /* is colliding node out of its main position? */
695       /* yes; move colliding node into free position */
696       while (othern + gnext(othern) != mp)  /* find previous */
697         othern += gnext(othern);
698       gnext(othern) = cast_int(f - othern);  /* rechain to point to 'f' */
699       *f = *mp;  /* copy colliding node into free pos. (mp->next also goes) */
700       if (gnext(mp) != 0) {
701         gnext(f) += cast_int(mp - f);  /* correct 'next' */
702         gnext(mp) = 0;  /* now 'mp' is free */
703       }
704       setempty(gval(mp));
705     }
706     else {  /* colliding node is in its own main position */
707       /* new node will go into free position */
708       if (gnext(mp) != 0)
709         gnext(f) = cast_int((mp + gnext(mp)) - f);  /* chain new position */
710       else lua_assert(gnext(f) == 0);
711       gnext(mp) = cast_int(f - mp);
712       mp = f;
713     }
714   }
715   setnodekey(L, mp, key);
716   luaC_barrierback(L, obj2gco(t), key);
717   lua_assert(isempty(gval(mp)));
718   setobj2t(L, gval(mp), value);
719 }
720 
721 
722 /*
723 ** Search function for integers. If integer is inside 'alimit', get it
724 ** directly from the array part. Otherwise, if 'alimit' is not equal to
725 ** the real size of the array, key still can be in the array part. In
726 ** this case, try to avoid a call to 'luaH_realasize' when key is just
727 ** one more than the limit (so that it can be incremented without
728 ** changing the real size of the array).
729 */
luaH_getint(Table * t,lua_Integer key)730 const TValue *luaH_getint (Table *t, lua_Integer key) {
731   if (l_castS2U(key) - 1u < t->alimit)  /* 'key' in [1, t->alimit]? */
732     return &t->array[key - 1];
733   else if (!limitequalsasize(t) &&  /* key still may be in the array part? */
734            (l_castS2U(key) == t->alimit + 1 ||
735             l_castS2U(key) - 1u < luaH_realasize(t))) {
736     t->alimit = cast_uint(key);  /* probably '#t' is here now */
737     return &t->array[key - 1];
738   }
739   else {
740     Node *n = hashint(t, key);
741     for (;;) {  /* check whether 'key' is somewhere in the chain */
742       if (keyisinteger(n) && keyival(n) == key)
743         return gval(n);  /* that's it */
744       else {
745         int nx = gnext(n);
746         if (nx == 0) break;
747         n += nx;
748       }
749     }
750     return &absentkey;
751   }
752 }
753 
754 
755 /*
756 ** search function for short strings
757 */
luaH_getshortstr(Table * t,TString * key)758 const TValue *luaH_getshortstr (Table *t, TString *key) {
759   Node *n = hashstr(t, key);
760   lua_assert(key->tt == LUA_VSHRSTR);
761   for (;;) {  /* check whether 'key' is somewhere in the chain */
762     if (keyisshrstr(n) && eqshrstr(keystrval(n), key))
763       return gval(n);  /* that's it */
764     else {
765       int nx = gnext(n);
766       if (nx == 0)
767         return &absentkey;  /* not found */
768       n += nx;
769     }
770   }
771 }
772 
773 
luaH_getstr(Table * t,TString * key)774 const TValue *luaH_getstr (Table *t, TString *key) {
775   if (key->tt == LUA_VSHRSTR)
776     return luaH_getshortstr(t, key);
777   else {  /* for long strings, use generic case */
778     TValue ko;
779     setsvalue(cast(lua_State *, NULL), &ko, key);
780     return getgeneric(t, &ko, 0);
781   }
782 }
783 
784 
785 /*
786 ** main search function
787 */
luaH_get(Table * t,const TValue * key)788 const TValue *luaH_get (Table *t, const TValue *key) {
789   switch (ttypetag(key)) {
790     case LUA_VSHRSTR: return luaH_getshortstr(t, tsvalue(key));
791     case LUA_VNUMINT: return luaH_getint(t, ivalue(key));
792     case LUA_VNIL: return &absentkey;
793     case LUA_VNUMFLT: {
794       lua_Integer k;
795       if (luaV_flttointeger(fltvalue(key), &k, F2Ieq)) /* integral index? */
796         return luaH_getint(t, k);  /* use specialized version */
797       /* else... */
798     }  /* FALLTHROUGH */
799     default:
800       return getgeneric(t, key, 0);
801   }
802 }
803 
804 
805 /*
806 ** Finish a raw "set table" operation, where 'slot' is where the value
807 ** should have been (the result of a previous "get table").
808 ** Beware: when using this function you probably need to check a GC
809 ** barrier and invalidate the TM cache.
810 */
luaH_finishset(lua_State * L,Table * t,const TValue * key,const TValue * slot,TValue * value)811 void luaH_finishset (lua_State *L, Table *t, const TValue *key,
812                                    const TValue *slot, TValue *value) {
813   if (isabstkey(slot))
814     luaH_newkey(L, t, key, value);
815   else
816     setobj2t(L, cast(TValue *, slot), value);
817 }
818 
819 
820 /*
821 ** beware: when using this function you probably need to check a GC
822 ** barrier and invalidate the TM cache.
823 */
luaH_set(lua_State * L,Table * t,const TValue * key,TValue * value)824 void luaH_set (lua_State *L, Table *t, const TValue *key, TValue *value) {
825   const TValue *slot = luaH_get(t, key);
826   luaH_finishset(L, t, key, slot, value);
827 }
828 
829 
luaH_setint(lua_State * L,Table * t,lua_Integer key,TValue * value)830 void luaH_setint (lua_State *L, Table *t, lua_Integer key, TValue *value) {
831   const TValue *p = luaH_getint(t, key);
832   if (isabstkey(p)) {
833     TValue k;
834     setivalue(&k, key);
835     luaH_newkey(L, t, &k, value);
836   }
837   else
838     setobj2t(L, cast(TValue *, p), value);
839 }
840 
841 
842 /*
843 ** Try to find a boundary in the hash part of table 't'. From the
844 ** caller, we know that 'j' is zero or present and that 'j + 1' is
845 ** present. We want to find a larger key that is absent from the
846 ** table, so that we can do a binary search between the two keys to
847 ** find a boundary. We keep doubling 'j' until we get an absent index.
848 ** If the doubling would overflow, we try LUA_MAXINTEGER. If it is
849 ** absent, we are ready for the binary search. ('j', being max integer,
850 ** is larger or equal to 'i', but it cannot be equal because it is
851 ** absent while 'i' is present; so 'j > i'.) Otherwise, 'j' is a
852 ** boundary. ('j + 1' cannot be a present integer key because it is
853 ** not a valid integer in Lua.)
854 */
hash_search(Table * t,lua_Unsigned j)855 static lua_Unsigned hash_search (Table *t, lua_Unsigned j) {
856   lua_Unsigned i;
857   if (j == 0) j++;  /* the caller ensures 'j + 1' is present */
858   do {
859     i = j;  /* 'i' is a present index */
860     if (j <= l_castS2U(LUA_MAXINTEGER) / 2)
861       j *= 2;
862     else {
863       j = LUA_MAXINTEGER;
864       if (isempty(luaH_getint(t, j)))  /* t[j] not present? */
865         break;  /* 'j' now is an absent index */
866       else  /* weird case */
867         return j;  /* well, max integer is a boundary... */
868     }
869   } while (!isempty(luaH_getint(t, j)));  /* repeat until an absent t[j] */
870   /* i < j  &&  t[i] present  &&  t[j] absent */
871   while (j - i > 1u) {  /* do a binary search between them */
872     lua_Unsigned m = (i + j) / 2;
873     if (isempty(luaH_getint(t, m))) j = m;
874     else i = m;
875   }
876   return i;
877 }
878 
879 
binsearch(const TValue * array,unsigned int i,unsigned int j)880 static unsigned int binsearch (const TValue *array, unsigned int i,
881                                                     unsigned int j) {
882   while (j - i > 1u) {  /* binary search */
883     unsigned int m = (i + j) / 2;
884     if (isempty(&array[m - 1])) j = m;
885     else i = m;
886   }
887   return i;
888 }
889 
890 
891 /*
892 ** Try to find a boundary in table 't'. (A 'boundary' is an integer index
893 ** such that t[i] is present and t[i+1] is absent, or 0 if t[1] is absent
894 ** and 'maxinteger' if t[maxinteger] is present.)
895 ** (In the next explanation, we use Lua indices, that is, with base 1.
896 ** The code itself uses base 0 when indexing the array part of the table.)
897 ** The code starts with 'limit = t->alimit', a position in the array
898 ** part that may be a boundary.
899 **
900 ** (1) If 't[limit]' is empty, there must be a boundary before it.
901 ** As a common case (e.g., after 't[#t]=nil'), check whether 'limit-1'
902 ** is present. If so, it is a boundary. Otherwise, do a binary search
903 ** between 0 and limit to find a boundary. In both cases, try to
904 ** use this boundary as the new 'alimit', as a hint for the next call.
905 **
906 ** (2) If 't[limit]' is not empty and the array has more elements
907 ** after 'limit', try to find a boundary there. Again, try first
908 ** the special case (which should be quite frequent) where 'limit+1'
909 ** is empty, so that 'limit' is a boundary. Otherwise, check the
910 ** last element of the array part. If it is empty, there must be a
911 ** boundary between the old limit (present) and the last element
912 ** (absent), which is found with a binary search. (This boundary always
913 ** can be a new limit.)
914 **
915 ** (3) The last case is when there are no elements in the array part
916 ** (limit == 0) or its last element (the new limit) is present.
917 ** In this case, must check the hash part. If there is no hash part
918 ** or 'limit+1' is absent, 'limit' is a boundary.  Otherwise, call
919 ** 'hash_search' to find a boundary in the hash part of the table.
920 ** (In those cases, the boundary is not inside the array part, and
921 ** therefore cannot be used as a new limit.)
922 */
luaH_getn(Table * t)923 lua_Unsigned luaH_getn (Table *t) {
924   unsigned int limit = t->alimit;
925   if (limit > 0 && isempty(&t->array[limit - 1])) {  /* (1)? */
926     /* there must be a boundary before 'limit' */
927     if (limit >= 2 && !isempty(&t->array[limit - 2])) {
928       /* 'limit - 1' is a boundary; can it be a new limit? */
929       if (ispow2realasize(t) && !ispow2(limit - 1)) {
930         t->alimit = limit - 1;
931         setnorealasize(t);  /* now 'alimit' is not the real size */
932       }
933       return limit - 1;
934     }
935     else {  /* must search for a boundary in [0, limit] */
936       unsigned int boundary = binsearch(t->array, 0, limit);
937       /* can this boundary represent the real size of the array? */
938       if (ispow2realasize(t) && boundary > luaH_realasize(t) / 2) {
939         t->alimit = boundary;  /* use it as the new limit */
940         setnorealasize(t);
941       }
942       return boundary;
943     }
944   }
945   /* 'limit' is zero or present in table */
946   if (!limitequalsasize(t)) {  /* (2)? */
947     /* 'limit' > 0 and array has more elements after 'limit' */
948     if (isempty(&t->array[limit]))  /* 'limit + 1' is empty? */
949       return limit;  /* this is the boundary */
950     /* else, try last element in the array */
951     limit = luaH_realasize(t);
952     if (isempty(&t->array[limit - 1])) {  /* empty? */
953       /* there must be a boundary in the array after old limit,
954          and it must be a valid new limit */
955       unsigned int boundary = binsearch(t->array, t->alimit, limit);
956       t->alimit = boundary;
957       return boundary;
958     }
959     /* else, new limit is present in the table; check the hash part */
960   }
961   /* (3) 'limit' is the last element and either is zero or present in table */
962   lua_assert(limit == luaH_realasize(t) &&
963              (limit == 0 || !isempty(&t->array[limit - 1])));
964   if (isdummy(t) || isempty(luaH_getint(t, cast(lua_Integer, limit + 1))))
965     return limit;  /* 'limit + 1' is absent */
966   else  /* 'limit + 1' is also present */
967     return hash_search(t, limit);
968 }
969 
970 
971 
972 #if defined(LUA_DEBUG)
973 
974 /* export these functions for the test library */
975 
luaH_mainposition(const Table * t,const TValue * key)976 Node *luaH_mainposition (const Table *t, const TValue *key) {
977   return mainpositionTV(t, key);
978 }
979 
980 #endif
981