/* ** $Id: ltable.c,v 2.134 2018/02/23 13:13:31 roberto Exp roberto $ ** Lua tables (hash) ** See Copyright Notice in lua.h */ #define ltable_c #define LUA_CORE #include "lprefix.h" /* ** Implementation of tables (aka arrays, objects, or hash tables). ** Tables keep its elements in two parts: an array part and a hash part. ** Non-negative integer keys are all candidates to be kept in the array ** part. The actual size of the array is the largest 'n' such that ** more than half the slots between 1 and n are in use. ** Hash uses a mix of chained scatter table with Brent's variation. ** A main invariant of these tables is that, if an element is not ** in its main position (i.e. the 'original' position that its hash gives ** to it), then the colliding element is in its own main position. ** Hence even when the load factor reaches 100%, performance remains good. */ #include #include #include "lua.h" #include "ldebug.h" #include "ldo.h" #include "lgc.h" #include "lmem.h" #include "lobject.h" #include "lstate.h" #include "lstring.h" #include "ltable.h" #include "lvm.h" /* ** MAXABITS is the largest integer such that MAXASIZE fits in an ** unsigned int. */ #define MAXABITS cast_int(sizeof(int) * CHAR_BIT - 1) /* ** MAXASIZE is the maximum size of the array part. It is the minimum ** between 2^MAXABITS and the maximum size such that, measured in bytes, ** it fits in a 'size_t'. */ #define MAXASIZE luaM_limitN(1u << MAXABITS, TValue) /* ** MAXHBITS is the largest integer such that 2^MAXHBITS fits in a ** signed int. */ #define MAXHBITS (MAXABITS - 1) /* ** MAXHSIZE is the maximum size of the hash part. It is the minimum ** between 2^MAXHBITS and the maximum size such that, measured in bytes, ** it fits in a 'size_t'. */ #define MAXHSIZE luaM_limitN(1u << MAXHBITS, Node) #define hashpow2(t,n) (gnode(t, lmod((n), sizenode(t)))) #define hashstr(t,str) hashpow2(t, (str)->hash) #define hashboolean(t,p) hashpow2(t, p) #define hashint(t,i) hashpow2(t, i) /* ** for some types, it is better to avoid modulus by power of 2, as ** they tend to have many 2 factors. */ #define hashmod(t,n) (gnode(t, ((n) % ((sizenode(t)-1)|1)))) #define hashpointer(t,p) hashmod(t, point2uint(p)) #define dummynode (&dummynode_) static const Node dummynode_ = { {{NULL}, LUA_TEMPTY, /* value's value and type */ LUA_TNIL, 0, {NULL}} /* key type, next, and key value */ }; LUAI_DDEF const TValue luaH_emptyobject_ = {EMPTYCONSTANT}; /* ** Hash for floating-point numbers. ** The main computation should be just ** n = frexp(n, &i); return (n * INT_MAX) + i ** but there are some numerical subtleties. ** In a two-complement representation, INT_MAX does not has an exact ** representation as a float, but INT_MIN does; because the absolute ** value of 'frexp' is smaller than 1 (unless 'n' is inf/NaN), the ** absolute value of the product 'frexp * -INT_MIN' is smaller or equal ** to INT_MAX. Next, the use of 'unsigned int' avoids overflows when ** adding 'i'; the use of '~u' (instead of '-u') avoids problems with ** INT_MIN. */ #if !defined(l_hashfloat) static int l_hashfloat (lua_Number n) { int i; lua_Integer ni; n = l_mathop(frexp)(n, &i) * -cast_num(INT_MIN); if (!lua_numbertointeger(n, &ni)) { /* is 'n' inf/-inf/NaN? */ lua_assert(luai_numisnan(n) || l_mathop(fabs)(n) == cast_num(HUGE_VAL)); return 0; } else { /* normal case */ unsigned int u = cast_uint(i) + cast_uint(ni); return cast_int(u <= cast_uint(INT_MAX) ? u : ~u); } } #endif /* ** returns the 'main' position of an element in a table (that is, ** the index of its hash value). The key comes broken (tag in 'ktt' ** and value in 'vkl') so that we can call it on keys inserted into ** nodes. */ static Node *mainposition (const Table *t, int ktt, const Value *kvl) { switch (withvariant(ktt)) { case LUA_TNUMINT: return hashint(t, ivalueraw(*kvl)); case LUA_TNUMFLT: return hashmod(t, l_hashfloat(fltvalueraw(*kvl))); case LUA_TSHRSTR: return hashstr(t, tsvalueraw(*kvl)); case LUA_TLNGSTR: return hashpow2(t, luaS_hashlongstr(tsvalueraw(*kvl))); case LUA_TBOOLEAN: return hashboolean(t, bvalueraw(*kvl)); case LUA_TLIGHTUSERDATA: return hashpointer(t, pvalueraw(*kvl)); case LUA_TLCF: return hashpointer(t, fvalueraw(*kvl)); default: return hashpointer(t, gcvalueraw(*kvl)); } } static Node *mainpositionTV (const Table *t, const TValue *key) { return mainposition(t, rawtt(key), valraw(key)); } /* ** Check whether key 'k1' is equal to the key in node 'n2'. ** This equality is raw, so there are no metamethods. Floats ** with integer values have been normalized, so integers cannot ** be equal to floats. It is assumed that 'eqshrstr' is simply ** pointer equality, so that short strings are handled in the ** default case. */ static int equalkey (const TValue *k1, const Node *n2) { if (rawtt(k1) != keytt(n2)) /* not the same variants? */ return 0; /* cannot be same key */ switch (ttypetag(k1)) { case LUA_TNIL: return 1; case LUA_TNUMINT: return (ivalue(k1) == keyival(n2)); case LUA_TNUMFLT: return luai_numeq(fltvalue(k1), fltvalueraw(keyval(n2))); case LUA_TBOOLEAN: return bvalue(k1) == bvalueraw(keyval(n2)); case LUA_TLIGHTUSERDATA: return pvalue(k1) == pvalueraw(keyval(n2)); case LUA_TLCF: return fvalue(k1) == fvalueraw(keyval(n2)); case LUA_TLNGSTR: return luaS_eqlngstr(tsvalue(k1), keystrval(n2)); default: return gcvalue(k1) == gcvalueraw(keyval(n2)); } } /* ** "Generic" get version. (Not that generic: not valid for integers, ** which may be in array part, nor for floats with integral values.) */ static const TValue *getgeneric (Table *t, const TValue *key) { Node *n = mainpositionTV(t, key); for (;;) { /* check whether 'key' is somewhere in the chain */ if (equalkey(key, n)) return gval(n); /* that's it */ else { int nx = gnext(n); if (nx == 0) return luaH_emptyobject; /* not found */ n += nx; } } } /* ** returns the index for 'k' if 'k' is an appropriate key to live in ** the array part of a table, 0 otherwise. */ static unsigned int arrayindex (lua_Integer k) { if (0 < k && l_castS2U(k) <= MAXASIZE) return cast_uint(k); /* 'key' is an appropriate array index */ else return 0; } /* ** returns the index of a 'key' for table traversals. First goes all ** elements in the array part, then elements in the hash part. The ** beginning of a traversal is signaled by 0. */ static unsigned int findindex (lua_State *L, Table *t, TValue *key) { unsigned int i; if (ttisnil(key)) return 0; /* first iteration */ i = ttisinteger(key) ? arrayindex(ivalue(key)) : 0; if (i != 0 && i <= t->sizearray) /* is 'key' inside array part? */ return i; /* yes; that's the index */ else { const TValue *n = getgeneric(t, key); if (n == luaH_emptyobject) luaG_runerror(L, "invalid key to 'next'"); /* key not found */ i = cast_int(nodefromval(n) - gnode(t, 0)); /* key index in hash table */ /* hash elements are numbered after array ones */ return (i + 1) + t->sizearray; } } int luaH_next (lua_State *L, Table *t, StkId key) { unsigned int i = findindex(L, t, s2v(key)); /* find original element */ for (; i < t->sizearray; i++) { /* try first array part */ if (!isempty(&t->array[i])) { /* a non-empty entry? */ setivalue(s2v(key), i + 1); setobj2s(L, key + 1, &t->array[i]); return 1; } } for (i -= t->sizearray; cast_int(i) < sizenode(t); i++) { /* hash part */ if (!isempty(gval(gnode(t, i)))) { /* a non-empty entry? */ Node *n = gnode(t, i); getnodekey(L, s2v(key), n); setobj2s(L, key + 1, gval(n)); return 1; } } return 0; /* no more elements */ } static void freehash (lua_State *L, Table *t) { if (!isdummy(t)) luaM_freearray(L, t->node, cast_sizet(sizenode(t))); } /* ** {============================================================= ** Rehash ** ============================================================== */ /* ** Compute the optimal size for the array part of table 't'. 'nums' is a ** "count array" where 'nums[i]' is the number of integers in the table ** between 2^(i - 1) + 1 and 2^i. 'pna' enters with the total number of ** integer keys in the table and leaves with the number of keys that ** will go to the array part; return the optimal size. (The condition ** 'twotoi > 0' in the for loop stops the loop if 'twotoi' overflows.) */ static unsigned int computesizes (unsigned int nums[], unsigned int *pna) { int i; unsigned int twotoi; /* 2^i (candidate for optimal size) */ unsigned int a = 0; /* number of elements smaller than 2^i */ unsigned int na = 0; /* number of elements to go to array part */ unsigned int optimal = 0; /* optimal size for array part */ /* loop while keys can fill more than half of total size */ for (i = 0, twotoi = 1; twotoi > 0 && *pna > twotoi / 2; i++, twotoi *= 2) { a += nums[i]; if (a > twotoi/2) { /* more than half elements present? */ optimal = twotoi; /* optimal size (till now) */ na = a; /* all elements up to 'optimal' will go to array part */ } } lua_assert((optimal == 0 || optimal / 2 < na) && na <= optimal); *pna = na; return optimal; } static int countint (lua_Integer key, unsigned int *nums) { unsigned int k = arrayindex(key); if (k != 0) { /* is 'key' an appropriate array index? */ nums[luaO_ceillog2(k)]++; /* count as such */ return 1; } else return 0; } /* ** Count keys in array part of table 't': Fill 'nums[i]' with ** number of keys that will go into corresponding slice and return ** total number of non-nil keys. */ static unsigned int numusearray (const Table *t, unsigned int *nums) { int lg; unsigned int ttlg; /* 2^lg */ unsigned int ause = 0; /* summation of 'nums' */ unsigned int i = 1; /* count to traverse all array keys */ /* traverse each slice */ for (lg = 0, ttlg = 1; lg <= MAXABITS; lg++, ttlg *= 2) { unsigned int lc = 0; /* counter */ unsigned int lim = ttlg; if (lim > t->sizearray) { lim = t->sizearray; /* adjust upper limit */ if (i > lim) break; /* no more elements to count */ } /* count elements in range (2^(lg - 1), 2^lg] */ for (; i <= lim; i++) { if (!isempty(&t->array[i-1])) lc++; } nums[lg] += lc; ause += lc; } return ause; } static int numusehash (const Table *t, unsigned int *nums, unsigned int *pna) { int totaluse = 0; /* total number of elements */ int ause = 0; /* elements added to 'nums' (can go to array part) */ int i = sizenode(t); while (i--) { Node *n = &t->node[i]; if (!isempty(gval(n))) { if (keyisinteger(n)) ause += countint(keyival(n), nums); totaluse++; } } *pna += ause; return totaluse; } /* ** Creates an array for the hash part of a table with the given ** size, or reuses the dummy node if size is zero. ** The computation for size overflow is in two steps: the first ** comparison ensures that the shift in the second one does not ** overflow. */ static void setnodevector (lua_State *L, Table *t, unsigned int size) { if (size == 0) { /* no elements to hash part? */ t->node = cast(Node *, dummynode); /* use common 'dummynode' */ t->lsizenode = 0; t->lastfree = NULL; /* signal that it is using dummy node */ } else { int i; int lsize = luaO_ceillog2(size); if (lsize > MAXHBITS || (1u << lsize) > MAXHSIZE) luaG_runerror(L, "table overflow"); size = twoto(lsize); t->node = luaM_newvector(L, size, Node); for (i = 0; i < (int)size; i++) { Node *n = gnode(t, i); gnext(n) = 0; setnilkey(n); setempty(gval(n)); } t->lsizenode = cast_byte(lsize); t->lastfree = gnode(t, size); /* all positions are free */ } } /* ** (Re)insert all elements from the hash part of 'ot' into table 't'. */ static void reinsert (lua_State *L, Table *ot, Table *t) { int j; int size = sizenode(ot); for (j = 0; j < size; j++) { Node *old = gnode(ot, j); if (!isempty(gval(old))) { /* doesn't need barrier/invalidate cache, as entry was already present in the table */ TValue k; getnodekey(L, &k, old); setobjt2t(L, luaH_set(L, t, &k), gval(old)); } } } /* ** Exchange the hash part of 't1' and 't2'. */ static void exchangehashpart (Table *t1, Table *t2) { lu_byte lsizenode = t1->lsizenode; Node *node = t1->node; Node *lastfree = t1->lastfree; t1->lsizenode = t2->lsizenode; t1->node = t2->node; t1->lastfree = t2->lastfree; t2->lsizenode = lsizenode; t2->node = node; t2->lastfree = lastfree; } /* ** Resize table 't' for the new given sizes. Both allocations (for ** the hash part and for the array part) can fail, which creates some ** subtleties. If the first allocation, for the hash part, fails, an ** error is raised and that is it. Otherwise, it copies the elements from ** the shrinking part of the array (if it is shrinking) into the new ** hash. Then it reallocates the array part. If that fails, the table ** is in its original state; the function frees the new hash part and then ** raises the allocation error. Otherwise, it sets the new hash part ** into the table, initializes the new part of the array (if any) with ** nils and reinserts the elements of the old hash back into the new ** parts of the table. */ void luaH_resize (lua_State *L, Table *t, unsigned int newasize, unsigned int nhsize) { unsigned int i; Table newt; /* to keep the new hash part */ unsigned int oldasize = t->sizearray; TValue *newarray; /* create new hash part with appropriate size into 'newt' */ setnodevector(L, &newt, nhsize); if (newasize < oldasize) { /* will array shrink? */ t->sizearray = newasize; /* pretend array has new size... */ exchangehashpart(t, &newt); /* and new hash */ /* re-insert into the new hash the elements from vanishing slice */ for (i = newasize; i < oldasize; i++) { if (!isempty(&t->array[i])) luaH_setint(L, t, i + 1, &t->array[i]); } t->sizearray = oldasize; /* restore current size... */ exchangehashpart(t, &newt); /* and hash (in case of errors) */ } /* allocate new array */ newarray = luaM_reallocvector(L, t->array, oldasize, newasize, TValue); if (newarray == NULL && newasize > 0) { /* allocation failed? */ freehash(L, &newt); /* release new hash part */ luaM_error(L); /* raise error (with array unchanged) */ } /* allocation ok; initialize new part of the array */ exchangehashpart(t, &newt); /* 't' has the new hash ('newt' has the old) */ t->array = newarray; /* set new array part */ t->sizearray = newasize; for (i = oldasize; i < newasize; i++) /* clear new slice of the array */ setempty(&t->array[i]); /* re-insert elements from old hash part into new parts */ reinsert(L, &newt, t); /* 'newt' now has the old hash */ freehash(L, &newt); /* free old hash part */ } void luaH_resizearray (lua_State *L, Table *t, unsigned int nasize) { int nsize = allocsizenode(t); luaH_resize(L, t, nasize, nsize); } /* ** nums[i] = number of keys 'k' where 2^(i - 1) < k <= 2^i */ static void rehash (lua_State *L, Table *t, const TValue *ek) { unsigned int asize; /* optimal size for array part */ unsigned int na; /* number of keys in the array part */ unsigned int nums[MAXABITS + 1]; int i; int totaluse; for (i = 0; i <= MAXABITS; i++) nums[i] = 0; /* reset counts */ na = numusearray(t, nums); /* count keys in array part */ totaluse = na; /* all those keys are integer keys */ totaluse += numusehash(t, nums, &na); /* count keys in hash part */ /* count extra key */ if (ttisinteger(ek)) na += countint(ivalue(ek), nums); totaluse++; /* compute new size for array part */ asize = computesizes(nums, &na); /* resize the table to new computed sizes */ luaH_resize(L, t, asize, totaluse - na); } /* ** }============================================================= */ Table *luaH_new (lua_State *L) { GCObject *o = luaC_newobj(L, LUA_TTABLE, sizeof(Table)); Table *t = gco2t(o); t->metatable = NULL; t->flags = cast_byte(~0); t->array = NULL; t->sizearray = 0; setnodevector(L, t, 0); return t; } void luaH_free (lua_State *L, Table *t) { freehash(L, t); luaM_freearray(L, t->array, t->sizearray); luaM_free(L, t); } static Node *getfreepos (Table *t) { if (!isdummy(t)) { while (t->lastfree > t->node) { t->lastfree--; if (keyisnil(t->lastfree)) return t->lastfree; } } return NULL; /* could not find a free place */ } /* ** inserts a new key into a hash table; first, check whether key's main ** position is free. If not, check whether colliding node is in its main ** position or not: if it is not, move colliding node to an empty place and ** put new key in its main position; otherwise (colliding node is in its main ** position), new key goes to an empty position. */ TValue *luaH_newkey (lua_State *L, Table *t, const TValue *key) { Node *mp; TValue aux; if (ttisnil(key)) luaG_runerror(L, "table index is nil"); else if (ttisfloat(key)) { lua_Number f = fltvalue(key); lua_Integer k; if (luaV_flttointeger(f, &k, 0)) { /* does key fit in an integer? */ setivalue(&aux, k); key = &aux; /* insert it as an integer */ } else if (luai_numisnan(f)) luaG_runerror(L, "table index is NaN"); } mp = mainpositionTV(t, key); if (!isempty(gval(mp)) || isdummy(t)) { /* main position is taken? */ Node *othern; Node *f = getfreepos(t); /* get a free place */ if (f == NULL) { /* cannot find a free place? */ rehash(L, t, key); /* grow table */ /* whatever called 'newkey' takes care of TM cache */ return luaH_set(L, t, key); /* insert key into grown table */ } lua_assert(!isdummy(t)); othern = mainposition(t, keytt(mp), &keyval(mp)); if (othern != mp) { /* is colliding node out of its main position? */ /* yes; move colliding node into free position */ while (othern + gnext(othern) != mp) /* find previous */ othern += gnext(othern); gnext(othern) = cast_int(f - othern); /* rechain to point to 'f' */ *f = *mp; /* copy colliding node into free pos. (mp->next also goes) */ if (gnext(mp) != 0) { gnext(f) += cast_int(mp - f); /* correct 'next' */ gnext(mp) = 0; /* now 'mp' is free */ } setempty(gval(mp)); } else { /* colliding node is in its own main position */ /* new node will go into free position */ if (gnext(mp) != 0) gnext(f) = cast_int((mp + gnext(mp)) - f); /* chain new position */ else lua_assert(gnext(f) == 0); gnext(mp) = cast_int(f - mp); mp = f; } } setnodekey(L, mp, key); luaC_barrierback(L, obj2gco(t), key); lua_assert(isempty(gval(mp))); return gval(mp); } /* ** search function for integers */ const TValue *luaH_getint (Table *t, lua_Integer key) { /* (1 <= key && key <= t->sizearray) */ if (l_castS2U(key) - 1u < t->sizearray) return &t->array[key - 1]; else { Node *n = hashint(t, key); for (;;) { /* check whether 'key' is somewhere in the chain */ if (keyisinteger(n) && keyival(n) == key) return gval(n); /* that's it */ else { int nx = gnext(n); if (nx == 0) break; n += nx; } } return luaH_emptyobject; } } /* ** search function for short strings */ const TValue *luaH_getshortstr (Table *t, TString *key) { Node *n = hashstr(t, key); lua_assert(key->tt == LUA_TSHRSTR); for (;;) { /* check whether 'key' is somewhere in the chain */ if (keyisshrstr(n) && eqshrstr(keystrval(n), key)) return gval(n); /* that's it */ else { int nx = gnext(n); if (nx == 0) return luaH_emptyobject; /* not found */ n += nx; } } } const TValue *luaH_getstr (Table *t, TString *key) { if (key->tt == LUA_TSHRSTR) return luaH_getshortstr(t, key); else { /* for long strings, use generic case */ TValue ko; setsvalue(cast(lua_State *, NULL), &ko, key); return getgeneric(t, &ko); } } /* ** main search function */ const TValue *luaH_get (Table *t, const TValue *key) { switch (ttypetag(key)) { case LUA_TSHRSTR: return luaH_getshortstr(t, tsvalue(key)); case LUA_TNUMINT: return luaH_getint(t, ivalue(key)); case LUA_TNIL: return luaH_emptyobject; case LUA_TNUMFLT: { lua_Integer k; if (luaV_flttointeger(fltvalue(key), &k, 0)) /* index is an integral? */ return luaH_getint(t, k); /* use specialized version */ /* else... */ } /* FALLTHROUGH */ default: return getgeneric(t, key); } } /* ** beware: when using this function you probably need to check a GC ** barrier and invalidate the TM cache. */ TValue *luaH_set (lua_State *L, Table *t, const TValue *key) { const TValue *p = luaH_get(t, key); if (p != luaH_emptyobject) return cast(TValue *, p); else return luaH_newkey(L, t, key); } void luaH_setint (lua_State *L, Table *t, lua_Integer key, TValue *value) { const TValue *p = luaH_getint(t, key); TValue *cell; if (p != luaH_emptyobject) cell = cast(TValue *, p); else { TValue k; setivalue(&k, key); cell = luaH_newkey(L, t, &k); } setobj2t(L, cell, value); } /* ** Try to find a boundary in the hash part of table 't'. From the ** caller, we know that 'j' is zero or present and that 'j + 1' is ** present. We want to find a larger key that is absent from the ** table, so that we can do a binary search between the two keys to ** find a boundary. We keep doubling 'j' until we get an absent index. ** If the doubling would overflow, we try LUA_MAXINTEGER. If it is ** absent, we are ready for the binary search. ('j', being max integer, ** is larger or equal to 'i', but it cannot be equal because it is ** absent while 'i' is present; so 'j > i'.) Otherwise, 'j' is a ** boundary. ('j + 1' cannot be a present integer key because it is ** not a valid integer in Lua.) */ static lua_Unsigned hash_search (Table *t, lua_Unsigned j) { lua_Unsigned i; if (j == 0) j++; /* the caller ensures 'j + 1' is present */ do { i = j; /* 'i' is a present index */ if (j <= l_castS2U(LUA_MAXINTEGER) / 2) j *= 2; else { j = LUA_MAXINTEGER; if (isempty(luaH_getint(t, j))) /* t[j] not present? */ break; /* 'j' now is an absent index */ else /* weird case */ return j; /* well, max integer is a boundary... */ } } while (!isempty(luaH_getint(t, j))); /* repeat until an absent t[j] */ /* i < j && t[i] present && t[j] absent */ while (j - i > 1u) { /* do a binary search between them */ lua_Unsigned m = (i + j) / 2; if (isempty(luaH_getint(t, m))) j = m; else i = m; } return i; } /* ** Try to find a boundary in table 't'. (A 'boundary' is an integer index ** such that t[i] is present and t[i+1] is absent, or 0 if t[1] is absent ** and 'maxinteger' if t[maxinteger] is present.) ** First, try the array part: if there is an array part and its last ** element is absent, there must be a boundary there; a binary search ** finds that boundary. Otherwise, if the hash part is empty or does not ** contain 'j + 1', 'j' is a boundary. Otherwize, call 'hash_search' ** to find a boundary in the hash part. */ lua_Unsigned luaH_getn (Table *t) { unsigned int j = t->sizearray; if (j > 0 && isempty(&t->array[j - 1])) { unsigned int i = 0; while (j - i > 1u) { /* binary search */ unsigned int m = (i + j) / 2; if (isempty(&t->array[m - 1])) j = m; else i = m; } return i; } else { /* 'j' is zero or present in table */ if (isdummy(t) || isempty(luaH_getint(t, l_castU2S(j + 1)))) return j; /* 'j + 1' is absent... */ else /* 'j + 1' is also present */ return hash_search(t, j); } } #if defined(LUA_DEBUG) Node *luaH_mainposition (const Table *t, const TValue *key) { return mainpositionTV(t, key); } int luaH_isdummy (const Table *t) { return isdummy(t); } #endif