1 /*
2 ** $Id: ltable.c,v 2.72.1.1 2013/04/12 18:48:47 roberto Exp $
3 ** Lua tables (hash)
4 ** See Copyright Notice in lua.h
5 */
6 
7 
8 /*
9 ** Implementation of tables (aka arrays, objects, or hash tables).
10 ** Tables keep its elements in two parts: an array part and a hash part.
11 ** Non-negative integer keys are all candidates to be kept in the array
12 ** part. The actual size of the array is the largest `n' such that at
13 ** least half the slots between 0 and n are in use.
14 ** Hash uses a mix of chained scatter table with Brent's variation.
15 ** A main invariant of these tables is that, if an element is not
16 ** in its main position (i.e. the `original' position that its hash gives
17 ** to it), then the colliding element is in its own main position.
18 ** Hence even when the load factor reaches 100%, performance remains good.
19 */
20 
21 #include <string.h>
22 
23 #define ltable_c
24 #define LUA_CORE
25 
26 #include "lua.h"
27 
28 #include "ldebug.h"
29 #include "ldo.h"
30 #include "lgc.h"
31 #include "lmem.h"
32 #include "lobject.h"
33 #include "lstate.h"
34 #include "lstring.h"
35 #include "ltable.h"
36 #include "lvm.h"
37 
38 
39 /*
40 ** max size of array part is 2^MAXBITS
41 */
42 #if LUAI_BITSINT >= 32
43 #define MAXBITS		30
44 #else
45 #define MAXBITS		(LUAI_BITSINT-2)
46 #endif
47 
48 #define MAXASIZE	(1 << MAXBITS)
49 
50 
51 #define hashpow2(t,n)		(gnode(t, lmod((n), sizenode(t))))
52 
53 #define hashstr(t,str)		hashpow2(t, (str)->tsv.hash)
54 #define hashboolean(t,p)	hashpow2(t, p)
55 
56 
57 /*
58 ** for some types, it is better to avoid modulus by power of 2, as
59 ** they tend to have many 2 factors.
60 */
61 #define hashmod(t,n)	(gnode(t, ((n) % ((sizenode(t)-1)|1))))
62 
63 
64 #define hashpointer(t,p)	hashmod(t, IntPoint(p))
65 
66 
67 #define dummynode		(&dummynode_)
68 
69 #define isdummy(n)		((n) == dummynode)
70 
71 static const Node dummynode_ = {
72   {NILCONSTANT},  /* value */
73   {{NILCONSTANT, NULL}}  /* key */
74 };
75 
76 
77 /*
78 ** hash for lua_Numbers
79 */
80 /* Taken from Lua 5.1 to avoid frexp() */
81 #define numints	cast_int(sizeof(lua_Number)/sizeof(int))
hashnum(const Table * t,lua_Number n)82 static Node *hashnum (const Table *t, lua_Number n) {
83   unsigned int a[numints];
84   int i;
85   if (luai_numeq(n, 0))  /* avoid problems with -0 */
86     return gnode(t, 0);
87   memcpy(a, &n, sizeof(a));
88   for (i = 1; i < numints; i++) a[0] += a[i];
89   return hashmod(t, a[0]);
90 }
91 
92 
93 
94 /*
95 ** returns the `main' position of an element in a table (that is, the index
96 ** of its hash value)
97 */
mainposition(const Table * t,const TValue * key)98 static Node *mainposition (const Table *t, const TValue *key) {
99   switch (ttype(key)) {
100     case LUA_TNUMBER:
101       return hashnum(t, nvalue(key));
102     case LUA_TLNGSTR: {
103       TString *s = rawtsvalue(key);
104       if (s->tsv.extra == 0) {  /* no hash? */
105         s->tsv.hash = luaS_hash(getstr(s), s->tsv.len, s->tsv.hash);
106         s->tsv.extra = 1;  /* now it has its hash */
107       }
108       return hashstr(t, rawtsvalue(key));
109     }
110     case LUA_TSHRSTR:
111       return hashstr(t, rawtsvalue(key));
112     case LUA_TBOOLEAN:
113       return hashboolean(t, bvalue(key));
114     case LUA_TLIGHTUSERDATA:
115       return hashpointer(t, pvalue(key));
116     case LUA_TLCF:
117       return hashpointer(t, fvalue(key));
118     default:
119       return hashpointer(t, gcvalue(key));
120   }
121 }
122 
123 
124 /*
125 ** returns the index for `key' if `key' is an appropriate key to live in
126 ** the array part of the table, -1 otherwise.
127 */
arrayindex(const TValue * key)128 static int arrayindex (const TValue *key) {
129   if (ttisnumber(key)) {
130     lua_Number n = nvalue(key);
131     int k;
132     lua_number2int(k, n);
133     if (luai_numeq(cast_num(k), n))
134       return k;
135   }
136   return -1;  /* `key' did not match some condition */
137 }
138 
139 
140 /*
141 ** returns the index of a `key' for table traversals. First goes all
142 ** elements in the array part, then elements in the hash part. The
143 ** beginning of a traversal is signaled by -1.
144 */
findindex(lua_State * L,Table * t,StkId key)145 static int findindex (lua_State *L, Table *t, StkId key) {
146   int i;
147   if (ttisnil(key)) return -1;  /* first iteration */
148   i = arrayindex(key);
149   if (0 < i && i <= t->sizearray)  /* is `key' inside array part? */
150     return i-1;  /* yes; that's the index (corrected to C) */
151   else {
152     Node *n = mainposition(t, key);
153     for (;;) {  /* check whether `key' is somewhere in the chain */
154       /* key may be dead already, but it is ok to use it in `next' */
155       if (luaV_rawequalobj(gkey(n), key) ||
156             (ttisdeadkey(gkey(n)) && iscollectable(key) &&
157              deadvalue(gkey(n)) == gcvalue(key))) {
158         i = cast_int(n - gnode(t, 0));  /* key index in hash table */
159         /* hash elements are numbered after array ones */
160         return i + t->sizearray;
161       }
162       else n = gnext(n);
163       if (n == NULL)
164         luaG_runerror(L, "invalid key to " LUA_QL("next"));  /* key not found */
165     }
166   }
167 }
168 
169 
luaH_next(lua_State * L,Table * t,StkId key)170 int luaH_next (lua_State *L, Table *t, StkId key) {
171   int i = findindex(L, t, key);  /* find original element */
172   for (i++; i < t->sizearray; i++) {  /* try first array part */
173     if (!ttisnil(&t->array[i])) {  /* a non-nil value? */
174       setnvalue(key, cast_num(i+1));
175       setobj2s(L, key+1, &t->array[i]);
176       return 1;
177     }
178   }
179   for (i -= t->sizearray; i < sizenode(t); i++) {  /* then hash part */
180     if (!ttisnil(gval(gnode(t, i)))) {  /* a non-nil value? */
181       setobj2s(L, key, gkey(gnode(t, i)));
182       setobj2s(L, key+1, gval(gnode(t, i)));
183       return 1;
184     }
185   }
186   return 0;  /* no more elements */
187 }
188 
189 
190 /*
191 ** {=============================================================
192 ** Rehash
193 ** ==============================================================
194 */
195 
196 
computesizes(int nums[],int * narray)197 static int computesizes (int nums[], int *narray) {
198   int i;
199   int twotoi;  /* 2^i */
200   int a = 0;  /* number of elements smaller than 2^i */
201   int na = 0;  /* number of elements to go to array part */
202   int n = 0;  /* optimal size for array part */
203   for (i = 0, twotoi = 1; twotoi/2 < *narray; i++, twotoi *= 2) {
204     if (nums[i] > 0) {
205       a += nums[i];
206       if (a > twotoi/2) {  /* more than half elements present? */
207         n = twotoi;  /* optimal size (till now) */
208         na = a;  /* all elements smaller than n will go to array part */
209       }
210     }
211     if (a == *narray) break;  /* all elements already counted */
212   }
213   *narray = n;
214   lua_assert(*narray/2 <= na && na <= *narray);
215   return na;
216 }
217 
218 
countint(const TValue * key,int * nums)219 static int countint (const TValue *key, int *nums) {
220   int k = arrayindex(key);
221   if (0 < k && k <= MAXASIZE) {  /* is `key' an appropriate array index? */
222     nums[luaO_ceillog2(k)]++;  /* count as such */
223     return 1;
224   }
225   else
226     return 0;
227 }
228 
229 
numusearray(const Table * t,int * nums)230 static int numusearray (const Table *t, int *nums) {
231   int lg;
232   int ttlg;  /* 2^lg */
233   int ause = 0;  /* summation of `nums' */
234   int i = 1;  /* count to traverse all array keys */
235   for (lg=0, ttlg=1; lg<=MAXBITS; lg++, ttlg*=2) {  /* for each slice */
236     int lc = 0;  /* counter */
237     int lim = ttlg;
238     if (lim > t->sizearray) {
239       lim = t->sizearray;  /* adjust upper limit */
240       if (i > lim)
241         break;  /* no more elements to count */
242     }
243     /* count elements in range (2^(lg-1), 2^lg] */
244     for (; i <= lim; i++) {
245       if (!ttisnil(&t->array[i-1]))
246         lc++;
247     }
248     nums[lg] += lc;
249     ause += lc;
250   }
251   return ause;
252 }
253 
254 
numusehash(const Table * t,int * nums,int * pnasize)255 static int numusehash (const Table *t, int *nums, int *pnasize) {
256   int totaluse = 0;  /* total number of elements */
257   int ause = 0;  /* summation of `nums' */
258   int i = sizenode(t);
259   while (i--) {
260     Node *n = &t->node[i];
261     if (!ttisnil(gval(n))) {
262       ause += countint(gkey(n), nums);
263       totaluse++;
264     }
265   }
266   *pnasize += ause;
267   return totaluse;
268 }
269 
270 
setarrayvector(lua_State * L,Table * t,int size)271 static void setarrayvector (lua_State *L, Table *t, int size) {
272   int i;
273   luaM_reallocvector(L, t->array, t->sizearray, size, TValue);
274   for (i=t->sizearray; i<size; i++)
275      setnilvalue(&t->array[i]);
276   t->sizearray = size;
277 }
278 
279 
setnodevector(lua_State * L,Table * t,int size)280 static void setnodevector (lua_State *L, Table *t, int size) {
281   int lsize;
282   if (size == 0) {  /* no elements to hash part? */
283     t->node = cast(Node *, dummynode);  /* use common `dummynode' */
284     lsize = 0;
285   }
286   else {
287     int i;
288     lsize = luaO_ceillog2(size);
289     if (lsize > MAXBITS)
290       luaG_runerror(L, "table overflow");
291     size = twoto(lsize);
292     t->node = luaM_newvector(L, size, Node);
293     for (i=0; i<size; i++) {
294       Node *n = gnode(t, i);
295       gnext(n) = NULL;
296       setnilvalue(gkey(n));
297       setnilvalue(gval(n));
298     }
299   }
300   t->lsizenode = cast_byte(lsize);
301   t->lastfree = gnode(t, size);  /* all positions are free */
302 }
303 
304 
luaH_resize(lua_State * L,Table * t,int nasize,int nhsize)305 void luaH_resize (lua_State *L, Table *t, int nasize, int nhsize) {
306   int i;
307   int oldasize = t->sizearray;
308   int oldhsize = t->lsizenode;
309   Node *nold = t->node;  /* save old hash ... */
310   if (nasize > oldasize)  /* array part must grow? */
311     setarrayvector(L, t, nasize);
312   /* create new hash part with appropriate size */
313   setnodevector(L, t, nhsize);
314   if (nasize < oldasize) {  /* array part must shrink? */
315     t->sizearray = nasize;
316     /* re-insert elements from vanishing slice */
317     for (i=nasize; i<oldasize; i++) {
318       if (!ttisnil(&t->array[i]))
319         luaH_setint(L, t, i + 1, &t->array[i]);
320     }
321     /* shrink array */
322     luaM_reallocvector(L, t->array, oldasize, nasize, TValue);
323   }
324   /* re-insert elements from hash part */
325   for (i = twoto(oldhsize) - 1; i >= 0; i--) {
326     Node *old = nold+i;
327     if (!ttisnil(gval(old))) {
328       /* doesn't need barrier/invalidate cache, as entry was
329          already present in the table */
330       setobjt2t(L, luaH_set(L, t, gkey(old)), gval(old));
331     }
332   }
333   if (!isdummy(nold))
334     luaM_freearray(L, nold, cast(size_t, twoto(oldhsize))); /* free old array */
335 }
336 
337 
luaH_resizearray(lua_State * L,Table * t,int nasize)338 void luaH_resizearray (lua_State *L, Table *t, int nasize) {
339   int nsize = isdummy(t->node) ? 0 : sizenode(t);
340   luaH_resize(L, t, nasize, nsize);
341 }
342 
343 
rehash(lua_State * L,Table * t,const TValue * ek)344 static void rehash (lua_State *L, Table *t, const TValue *ek) {
345   int nasize, na;
346   int nums[MAXBITS+1];  /* nums[i] = number of keys with 2^(i-1) < k <= 2^i */
347   int i;
348   int totaluse;
349   for (i=0; i<=MAXBITS; i++) nums[i] = 0;  /* reset counts */
350   nasize = numusearray(t, nums);  /* count keys in array part */
351   totaluse = nasize;  /* all those keys are integer keys */
352   totaluse += numusehash(t, nums, &nasize);  /* count keys in hash part */
353   /* count extra key */
354   nasize += countint(ek, nums);
355   totaluse++;
356   /* compute new size for array part */
357   na = computesizes(nums, &nasize);
358   /* resize the table to new computed sizes */
359   luaH_resize(L, t, nasize, totaluse - na);
360 }
361 
362 
363 
364 /*
365 ** }=============================================================
366 */
367 
368 
luaH_new(lua_State * L)369 Table *luaH_new (lua_State *L) {
370   Table *t = &luaC_newobj(L, LUA_TTABLE, sizeof(Table), NULL, 0)->h;
371   t->metatable = NULL;
372   t->flags = cast_byte(~0);
373   t->array = NULL;
374   t->sizearray = 0;
375   setnodevector(L, t, 0);
376   return t;
377 }
378 
379 
luaH_free(lua_State * L,Table * t)380 void luaH_free (lua_State *L, Table *t) {
381   if (!isdummy(t->node))
382     luaM_freearray(L, t->node, cast(size_t, sizenode(t)));
383   luaM_freearray(L, t->array, t->sizearray);
384   luaM_free(L, t);
385 }
386 
387 
getfreepos(Table * t)388 static Node *getfreepos (Table *t) {
389   while (t->lastfree > t->node) {
390     t->lastfree--;
391     if (ttisnil(gkey(t->lastfree)))
392       return t->lastfree;
393   }
394   return NULL;  /* could not find a free place */
395 }
396 
397 
398 
399 /*
400 ** inserts a new key into a hash table; first, check whether key's main
401 ** position is free. If not, check whether colliding node is in its main
402 ** position or not: if it is not, move colliding node to an empty place and
403 ** put new key in its main position; otherwise (colliding node is in its main
404 ** position), new key goes to an empty position.
405 */
luaH_newkey(lua_State * L,Table * t,const TValue * key)406 TValue *luaH_newkey (lua_State *L, Table *t, const TValue *key) {
407   Node *mp;
408   if (ttisnil(key)) luaG_runerror(L, "table index is nil");
409   else if (ttisnumber(key) && luai_numisnan(L, nvalue(key)))
410     luaG_runerror(L, "table index is NaN");
411   mp = mainposition(t, key);
412   if (!ttisnil(gval(mp)) || isdummy(mp)) {  /* main position is taken? */
413     Node *othern;
414     Node *n = getfreepos(t);  /* get a free place */
415     if (n == NULL) {  /* cannot find a free place? */
416       rehash(L, t, key);  /* grow table */
417       /* whatever called 'newkey' take care of TM cache and GC barrier */
418       return luaH_set(L, t, key);  /* insert key into grown table */
419     }
420     lua_assert(!isdummy(n));
421     othern = mainposition(t, gkey(mp));
422     if (othern != mp) {  /* is colliding node out of its main position? */
423       /* yes; move colliding node into free position */
424       while (gnext(othern) != mp) othern = gnext(othern);  /* find previous */
425       gnext(othern) = n;  /* redo the chain with `n' in place of `mp' */
426       *n = *mp;  /* copy colliding node into free pos. (mp->next also goes) */
427       gnext(mp) = NULL;  /* now `mp' is free */
428       setnilvalue(gval(mp));
429     }
430     else {  /* colliding node is in its own main position */
431       /* new node will go into free position */
432       gnext(n) = gnext(mp);  /* chain new position */
433       gnext(mp) = n;
434       mp = n;
435     }
436   }
437   setobj2t(L, gkey(mp), key);
438   luaC_barrierback(L, obj2gco(t), key);
439   lua_assert(ttisnil(gval(mp)));
440   return gval(mp);
441 }
442 
443 
444 /*
445 ** search function for integers
446 */
luaH_getint(Table * t,int key)447 const TValue *luaH_getint (Table *t, int key) {
448   /* (1 <= key && key <= t->sizearray) */
449   if (cast(unsigned int, key-1) < cast(unsigned int, t->sizearray))
450     return &t->array[key-1];
451   else {
452     lua_Number nk = cast_num(key);
453     Node *n = hashnum(t, nk);
454     do {  /* check whether `key' is somewhere in the chain */
455       if (ttisnumber(gkey(n)) && luai_numeq(nvalue(gkey(n)), nk))
456         return gval(n);  /* that's it */
457       else n = gnext(n);
458     } while (n);
459     return luaO_nilobject;
460   }
461 }
462 
463 
464 /*
465 ** search function for short strings
466 */
luaH_getstr(Table * t,TString * key)467 const TValue *luaH_getstr (Table *t, TString *key) {
468   Node *n = hashstr(t, key);
469   lua_assert(key->tsv.tt == LUA_TSHRSTR);
470   do {  /* check whether `key' is somewhere in the chain */
471     if (ttisshrstring(gkey(n)) && eqshrstr(rawtsvalue(gkey(n)), key))
472       return gval(n);  /* that's it */
473     else n = gnext(n);
474   } while (n);
475   return luaO_nilobject;
476 }
477 
478 
479 /*
480 ** main search function
481 */
luaH_get(Table * t,const TValue * key)482 const TValue *luaH_get (Table *t, const TValue *key) {
483   switch (ttype(key)) {
484     case LUA_TSHRSTR: return luaH_getstr(t, rawtsvalue(key));
485     case LUA_TNIL: return luaO_nilobject;
486     case LUA_TNUMBER: {
487       int k;
488       lua_Number n = nvalue(key);
489       lua_number2int(k, n);
490       if (luai_numeq(cast_num(k), n)) /* index is int? */
491         return luaH_getint(t, k);  /* use specialized version */
492       /* else go through */
493     }
494     default: {
495       Node *n = mainposition(t, key);
496       do {  /* check whether `key' is somewhere in the chain */
497         if (luaV_rawequalobj(gkey(n), key))
498           return gval(n);  /* that's it */
499         else n = gnext(n);
500       } while (n);
501       return luaO_nilobject;
502     }
503   }
504 }
505 
506 
507 /*
508 ** beware: when using this function you probably need to check a GC
509 ** barrier and invalidate the TM cache.
510 */
luaH_set(lua_State * L,Table * t,const TValue * key)511 TValue *luaH_set (lua_State *L, Table *t, const TValue *key) {
512   const TValue *p = luaH_get(t, key);
513   if (p != luaO_nilobject)
514     return cast(TValue *, p);
515   else return luaH_newkey(L, t, key);
516 }
517 
518 
luaH_setint(lua_State * L,Table * t,int key,TValue * value)519 void luaH_setint (lua_State *L, Table *t, int key, TValue *value) {
520   const TValue *p = luaH_getint(t, key);
521   TValue *cell;
522   if (p != luaO_nilobject)
523     cell = cast(TValue *, p);
524   else {
525     TValue k;
526     setnvalue(&k, cast_num(key));
527     cell = luaH_newkey(L, t, &k);
528   }
529   setobj2t(L, cell, value);
530 }
531 
532 
unbound_search(Table * t,unsigned int j)533 static int unbound_search (Table *t, unsigned int j) {
534   unsigned int i = j;  /* i is zero or a present index */
535   j++;
536   /* find `i' and `j' such that i is present and j is not */
537   while (!ttisnil(luaH_getint(t, j))) {
538     i = j;
539     j *= 2;
540     if (j > cast(unsigned int, MAX_INT)) {  /* overflow? */
541       /* table was built with bad purposes: resort to linear search */
542       i = 1;
543       while (!ttisnil(luaH_getint(t, i))) i++;
544       return i - 1;
545     }
546   }
547   /* now do a binary search between them */
548   while (j - i > 1) {
549     unsigned int m = (i+j)/2;
550     if (ttisnil(luaH_getint(t, m))) j = m;
551     else i = m;
552   }
553   return i;
554 }
555 
556 
557 /*
558 ** Try to find a boundary in table `t'. A `boundary' is an integer index
559 ** such that t[i] is non-nil and t[i+1] is nil (and 0 if t[1] is nil).
560 */
luaH_getn(Table * t)561 int luaH_getn (Table *t) {
562   unsigned int j = t->sizearray;
563   if (j > 0 && ttisnil(&t->array[j - 1])) {
564     /* there is a boundary in the array part: (binary) search for it */
565     unsigned int i = 0;
566     while (j - i > 1) {
567       unsigned int m = (i+j)/2;
568       if (ttisnil(&t->array[m - 1])) j = m;
569       else i = m;
570     }
571     return i;
572   }
573   /* else must find a boundary in hash part */
574   else if (isdummy(t->node))  /* hash part is empty? */
575     return j;  /* that is easy... */
576   else return unbound_search(t, j);
577 }
578 
579 
580 
581 #if defined(LUA_DEBUG)
582 
luaH_mainposition(const Table * t,const TValue * key)583 Node *luaH_mainposition (const Table *t, const TValue *key) {
584   return mainposition(t, key);
585 }
586 
luaH_isdummy(Node * n)587 int luaH_isdummy (Node *n) { return isdummy(n); }
588 
589 #endif
590