dict2.cpp revision 0:a61af66fc99e
1238384Sjkim/* 2238384Sjkim * Copyright 1998-2002 Sun Microsystems, Inc. All Rights Reserved. 3238384Sjkim * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4238384Sjkim * 5238384Sjkim * This code is free software; you can redistribute it and/or modify it 6238384Sjkim * under the terms of the GNU General Public License version 2 only, as 7238384Sjkim * published by the Free Software Foundation. 8238384Sjkim * 9238384Sjkim * This code is distributed in the hope that it will be useful, but WITHOUT 10238384Sjkim * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11238384Sjkim * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12238384Sjkim * version 2 for more details (a copy is included in the LICENSE file that 13238384Sjkim * accompanied this code). 14238384Sjkim * 15238384Sjkim * You should have received a copy of the GNU General Public License version 16238384Sjkim * 2 along with this work; if not, write to the Free Software Foundation, 17238384Sjkim * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18238384Sjkim * 19238384Sjkim * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20238384Sjkim * CA 95054 USA or visit www.sun.com if you need additional information or 21238384Sjkim * have any questions. 22238384Sjkim * 23238384Sjkim */ 24238384Sjkim 25238384Sjkim// Dictionaries - An Abstract Data Type 26238384Sjkim 27238384Sjkim#include "adlc.hpp" 28238384Sjkim 29238384Sjkim// #include "dict.hpp" 30238384Sjkim 31238384Sjkim 32238384Sjkim//------------------------------data----------------------------------------- 33238384Sjkim// String hash tables 34238384Sjkim#define MAXID 20 35238384Sjkimstatic char initflag = 0; // True after 1st initialization 36238384Sjkimstatic char shft[MAXID] = {1,2,3,4,5,6,7,1,2,3,4,5,6,7,1,2,3,4,5,6}; 37238384Sjkimstatic short xsum[MAXID + 1]; 38238384Sjkim 39238384Sjkim//------------------------------bucket--------------------------------------- 40238384Sjkimclass bucket { 41238384Sjkimpublic: 42238384Sjkim int _cnt, _max; // Size of bucket 43238384Sjkim const void **_keyvals; // Array of keys and values 44238384Sjkim}; 45238384Sjkim 46238384Sjkim//------------------------------Dict----------------------------------------- 47238384Sjkim// The dictionary is kept has a hash table. The hash table is a even power 48238384Sjkim// of two, for nice modulo operations. Each bucket in the hash table points 49238384Sjkim// to a linear list of key-value pairs; each key & value is just a (void *). 50238384Sjkim// The list starts with a count. A hash lookup finds the list head, then a 51238384Sjkim// simple linear scan finds the key. If the table gets too full, it's 52238384Sjkim// doubled in size; the total amount of EXTRA times all hash functions are 53238384Sjkim// computed for the doubling is no more than the current size - thus the 54238384Sjkim// doubling in size costs no more than a constant factor in speed. 55238384SjkimDict::Dict(CmpKey initcmp, Hash inithash) : _hash(inithash), _cmp(initcmp), _arena(NULL) { 56238384Sjkim init(); 57238384Sjkim} 58238384Sjkim 59238384SjkimDict::Dict(CmpKey initcmp, Hash inithash, Arena *arena) : _hash(inithash), _cmp(initcmp), _arena(arena) { 60238384Sjkim init(); 61238384Sjkim} 62238384Sjkim 63238384Sjkimvoid Dict::init() { 64238384Sjkim int i; 65238384Sjkim 66238384Sjkim // Precompute table of null character hashes 67238384Sjkim if( !initflag ) { // Not initializated yet? 68238384Sjkim xsum[0] = (1<<shft[0])+1; // Initialize 69238384Sjkim for( i = 1; i < MAXID + 1; i++) { 70238384Sjkim xsum[i] = (1<<shft[i])+1+xsum[i-1]; 71238384Sjkim } 72238384Sjkim initflag = 1; // Never again 73238384Sjkim } 74238384Sjkim 75238384Sjkim _size = 16; // Size is a power of 2 76238384Sjkim _cnt = 0; // Dictionary is empty 77238384Sjkim _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 78238384Sjkim memset(_bin,0,sizeof(bucket)*_size); 79238384Sjkim} 80238384Sjkim 81238384Sjkim//------------------------------~Dict------------------------------------------ 82238384Sjkim// Delete an existing dictionary. 83238384SjkimDict::~Dict() { 84238384Sjkim} 85238384Sjkim 86238384Sjkim//------------------------------Clear---------------------------------------- 87238384Sjkim// Zap to empty; ready for re-use 88238384Sjkimvoid Dict::Clear() { 89238384Sjkim _cnt = 0; // Empty contents 90238384Sjkim for( int i=0; i<_size; i++ ) 91238384Sjkim _bin[i]._cnt = 0; // Empty buckets, but leave allocated 92238384Sjkim // Leave _size & _bin alone, under the assumption that dictionary will 93238384Sjkim // grow to this size again. 94238384Sjkim} 95238384Sjkim 96238384Sjkim//------------------------------doubhash--------------------------------------- 97238384Sjkim// Double hash table size. If can't do so, just suffer. If can, then run 98238384Sjkim// thru old hash table, moving things to new table. Note that since hash 99238384Sjkim// table doubled, exactly 1 new bit is exposed in the mask - so everything 100238384Sjkim// in the old table ends up on 1 of two lists in the new table; a hi and a 101238384Sjkim// lo list depending on the value of the bit. 102238384Sjkimvoid Dict::doubhash(void) { 103238384Sjkim int oldsize = _size; 104238384Sjkim _size <<= 1; // Double in size 105238384Sjkim _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*oldsize, sizeof(bucket)*_size ); 106238384Sjkim memset( &_bin[oldsize], 0, oldsize*sizeof(bucket) ); 107238384Sjkim // Rehash things to spread into new table 108238384Sjkim for( int i=0; i < oldsize; i++) { // For complete OLD table do 109238384Sjkim bucket *b = &_bin[i]; // Handy shortcut for _bin[i] 110238384Sjkim if( !b->_keyvals ) continue; // Skip empties fast 111238384Sjkim 112238384Sjkim bucket *nb = &_bin[i+oldsize]; // New bucket shortcut 113238384Sjkim int j = b->_max; // Trim new bucket to nearest power of 2 114238384Sjkim while( j > b->_cnt ) j >>= 1; // above old bucket _cnt 115238384Sjkim if( !j ) j = 1; // Handle zero-sized buckets 116238384Sjkim nb->_max = j<<1; 117238384Sjkim // Allocate worst case space for key-value pairs 118238384Sjkim nb->_keyvals = (const void**)_arena->Amalloc_4( sizeof(void *)*nb->_max*2 ); 119238384Sjkim int nbcnt = 0; 120238384Sjkim 121238384Sjkim for( j=0; j<b->_cnt; j++ ) { // Rehash all keys in this bucket 122238384Sjkim const void *key = b->_keyvals[j+j]; 123238384Sjkim if( (_hash( key ) & (_size-1)) != i ) { // Moving to hi bucket? 124238384Sjkim nb->_keyvals[nbcnt+nbcnt] = key; 125238384Sjkim nb->_keyvals[nbcnt+nbcnt+1] = b->_keyvals[j+j+1]; 126238384Sjkim nb->_cnt = nbcnt = nbcnt+1; 127238384Sjkim b->_cnt--; // Remove key/value from lo bucket 128238384Sjkim b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 129238384Sjkim b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 130238384Sjkim j--; // Hash compacted element also 131238384Sjkim } 132238384Sjkim } // End of for all key-value pairs in bucket 133238384Sjkim } // End of for all buckets 134238384Sjkim 135238384Sjkim 136238384Sjkim} 137238384Sjkim 138238384Sjkim//------------------------------Dict----------------------------------------- 139238384Sjkim// Deep copy a dictionary. 140238384SjkimDict::Dict( const Dict &d ) : _size(d._size), _cnt(d._cnt), _hash(d._hash),_cmp(d._cmp), _arena(d._arena) { 141238384Sjkim _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 142238384Sjkim memcpy( _bin, d._bin, sizeof(bucket)*_size ); 143238384Sjkim for( int i=0; i<_size; i++ ) { 144238384Sjkim if( !_bin[i]._keyvals ) continue; 145238384Sjkim _bin[i]._keyvals=(const void**)_arena->Amalloc_4( sizeof(void *)*_bin[i]._max*2); 146238384Sjkim memcpy( _bin[i]._keyvals, d._bin[i]._keyvals,_bin[i]._cnt*2*sizeof(void*)); 147238384Sjkim } 148238384Sjkim} 149238384Sjkim 150238384Sjkim//------------------------------Dict----------------------------------------- 151238384Sjkim// Deep copy a dictionary. 152238384SjkimDict &Dict::operator =( const Dict &d ) { 153238384Sjkim if( _size < d._size ) { // If must have more buckets 154238384Sjkim _arena = d._arena; 155238384Sjkim _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*_size, sizeof(bucket)*d._size ); 156238384Sjkim memset( &_bin[_size], 0, (d._size-_size)*sizeof(bucket) ); 157238384Sjkim _size = d._size; 158238384Sjkim } 159238384Sjkim for( int i=0; i<_size; i++ ) // All buckets are empty 160238384Sjkim _bin[i]._cnt = 0; // But leave bucket allocations alone 161238384Sjkim _cnt = d._cnt; 162238384Sjkim *(Hash*)(&_hash) = d._hash; 163238384Sjkim *(CmpKey*)(&_cmp) = d._cmp; 164238384Sjkim for(int k=0; k<_size; k++ ) { 165238384Sjkim bucket *b = &d._bin[k]; // Shortcut to source bucket 166238384Sjkim for( int j=0; j<b->_cnt; j++ ) 167238384Sjkim Insert( b->_keyvals[j+j], b->_keyvals[j+j+1] ); 168238384Sjkim } 169238384Sjkim return *this; 170238384Sjkim} 171238384Sjkim 172238384Sjkim//------------------------------Insert--------------------------------------- 173238384Sjkim// Insert or replace a key/value pair in the given dictionary. If the 174238384Sjkim// dictionary is too full, it's size is doubled. The prior value being 175238384Sjkim// replaced is returned (NULL if this is a 1st insertion of that key). If 176238384Sjkim// an old value is found, it's swapped with the prior key-value pair on the 177238384Sjkim// list. This moves a commonly searched-for value towards the list head. 178238384Sjkimconst void *Dict::Insert(const void *key, const void *val) { 179238384Sjkim int hash = _hash( key ); // Get hash key 180238384Sjkim int i = hash & (_size-1); // Get hash key, corrected for size 181238384Sjkim bucket *b = &_bin[i]; // Handy shortcut 182238384Sjkim for( int j=0; j<b->_cnt; j++ ) 183238384Sjkim if( !_cmp(key,b->_keyvals[j+j]) ) { 184238384Sjkim const void *prior = b->_keyvals[j+j+1]; 185238384Sjkim b->_keyvals[j+j ] = key; // Insert current key-value 186238384Sjkim b->_keyvals[j+j+1] = val; 187238384Sjkim return prior; // Return prior 188238384Sjkim } 189238384Sjkim 190238384Sjkim if( ++_cnt > _size ) { // Hash table is full 191238384Sjkim doubhash(); // Grow whole table if too full 192238384Sjkim i = hash & (_size-1); // Rehash 193238384Sjkim b = &_bin[i]; // Handy shortcut 194238384Sjkim } 195238384Sjkim if( b->_cnt == b->_max ) { // Must grow bucket? 196238384Sjkim if( !b->_keyvals ) { 197238384Sjkim b->_max = 2; // Initial bucket size 198238384Sjkim b->_keyvals = (const void**)_arena->Amalloc_4( sizeof(void *)*b->_max*2 ); 199238384Sjkim } else { 200238384Sjkim b->_keyvals = (const void**)_arena->Arealloc( b->_keyvals, sizeof(void *)*b->_max*2, sizeof(void *)*b->_max*4 ); 201238384Sjkim b->_max <<= 1; // Double bucket 202238384Sjkim } 203238384Sjkim } 204238384Sjkim b->_keyvals[b->_cnt+b->_cnt ] = key; 205238384Sjkim b->_keyvals[b->_cnt+b->_cnt+1] = val; 206238384Sjkim b->_cnt++; 207238384Sjkim return NULL; // Nothing found prior 208238384Sjkim} 209238384Sjkim 210238384Sjkim//------------------------------Delete--------------------------------------- 211238384Sjkim// Find & remove a value from dictionary. Return old value. 212238384Sjkimconst void *Dict::Delete(void *key) { 213238384Sjkim int i = _hash( key ) & (_size-1); // Get hash key, corrected for size 214238384Sjkim bucket *b = &_bin[i]; // Handy shortcut 215238384Sjkim for( int j=0; j<b->_cnt; j++ ) 216238384Sjkim if( !_cmp(key,b->_keyvals[j+j]) ) { 217238384Sjkim const void *prior = b->_keyvals[j+j+1]; 218238384Sjkim b->_cnt--; // Remove key/value from lo bucket 219238384Sjkim b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 220238384Sjkim b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 221238384Sjkim _cnt--; // One less thing in table 222238384Sjkim return prior; 223238384Sjkim } 224238384Sjkim return NULL; 225238384Sjkim} 226238384Sjkim 227238384Sjkim//------------------------------FindDict------------------------------------- 228238384Sjkim// Find a key-value pair in the given dictionary. If not found, return NULL. 229238384Sjkim// If found, move key-value pair towards head of list. 230238384Sjkimconst void *Dict::operator [](const void *key) const { 231238384Sjkim int i = _hash( key ) & (_size-1); // Get hash key, corrected for size 232238384Sjkim bucket *b = &_bin[i]; // Handy shortcut 233238384Sjkim for( int j=0; j<b->_cnt; j++ ) 234238384Sjkim if( !_cmp(key,b->_keyvals[j+j]) ) 235238384Sjkim return b->_keyvals[j+j+1]; 236238384Sjkim return NULL; 237238384Sjkim} 238238384Sjkim 239238384Sjkim//------------------------------CmpDict-------------------------------------- 240238384Sjkim// CmpDict compares two dictionaries; they must have the same keys (their 241238384Sjkim// keys must match using CmpKey) and they must have the same values (pointer 242238384Sjkim// comparison). If so 1 is returned, if not 0 is returned. 243238384Sjkimint Dict::operator ==(const Dict &d2) const { 244238384Sjkim if( _cnt != d2._cnt ) return 0; 245238384Sjkim if( _hash != d2._hash ) return 0; 246238384Sjkim if( _cmp != d2._cmp ) return 0; 247238384Sjkim for( int i=0; i < _size; i++) { // For complete hash table do 248238384Sjkim bucket *b = &_bin[i]; // Handy shortcut 249238384Sjkim if( b->_cnt != d2._bin[i]._cnt ) return 0; 250238384Sjkim if( memcmp(b->_keyvals, d2._bin[i]._keyvals, b->_cnt*2*sizeof(void*) ) ) 251238384Sjkim return 0; // Key-value pairs must match 252238384Sjkim } 253 return 1; // All match, is OK 254} 255 256 257//------------------------------print---------------------------------------- 258static void printvoid(const void* x) { printf("%p", x); } 259void Dict::print() { 260 print(printvoid, printvoid); 261} 262void Dict::print(PrintKeyOrValue print_key, PrintKeyOrValue print_value) { 263 for( int i=0; i < _size; i++) { // For complete hash table do 264 bucket *b = &_bin[i]; // Handy shortcut 265 for( int j=0; j<b->_cnt; j++ ) { 266 print_key( b->_keyvals[j+j ]); 267 printf(" -> "); 268 print_value(b->_keyvals[j+j+1]); 269 printf("\n"); 270 } 271 } 272} 273 274//------------------------------Hashing Functions---------------------------- 275// Convert string to hash key. This algorithm implements a universal hash 276// function with the multipliers frozen (ok, so it's not universal). The 277// multipliers (and allowable characters) are all odd, so the resultant sum 278// is odd - guarenteed not divisible by any power of two, so the hash tables 279// can be any power of two with good results. Also, I choose multipliers 280// that have only 2 bits set (the low is always set to be odd) so 281// multiplication requires only shifts and adds. Characters are required to 282// be in the range 0-127 (I double & add 1 to force oddness). Keys are 283// limited to MAXID characters in length. Experimental evidence on 150K of 284// C text shows excellent spreading of values for any size hash table. 285int hashstr(const void *t) { 286 register char c, k = 0; 287 register int sum = 0; 288 register const char *s = (const char *)t; 289 290 while( ((c = s[k]) != '\0') && (k < MAXID-1) ) { // Get characters till nul 291 c = (c<<1)+1; // Characters are always odd! 292 sum += c + (c<<shft[k++]); // Universal hash function 293 } 294 assert( k < (MAXID + 1), "Exceeded maximum name length"); 295 return (int)((sum+xsum[k]) >> 1); // Hash key, un-modulo'd table size 296} 297 298//------------------------------hashptr-------------------------------------- 299// Slimey cheap hash function; no guarenteed performance. Better than the 300// default for pointers, especially on MS-DOS machines. 301int hashptr(const void *key) { 302#ifdef __TURBOC__ 303 return (int)((intptr_t)key >> 16); 304#else // __TURBOC__ 305 return (int)((intptr_t)key >> 2); 306#endif 307} 308 309// Slimey cheap hash function; no guarenteed performance. 310int hashkey(const void *key) { 311 return (int)((intptr_t)key); 312} 313 314//------------------------------Key Comparator Functions--------------------- 315int cmpstr(const void *k1, const void *k2) { 316 return strcmp((const char *)k1,(const char *)k2); 317} 318 319// Slimey cheap key comparator. 320int cmpkey(const void *key1, const void *key2) { 321 return (int)((intptr_t)key1 - (intptr_t)key2); 322} 323 324//============================================================================= 325//------------------------------reset------------------------------------------ 326// Create an iterator and initialize the first variables. 327void DictI::reset( const Dict *dict ) { 328 _d = dict; // The dictionary 329 _i = (int)-1; // Before the first bin 330 _j = 0; // Nothing left in the current bin 331 ++(*this); // Step to first real value 332} 333 334//------------------------------next------------------------------------------- 335// Find the next key-value pair in the dictionary, or return a NULL key and 336// value. 337void DictI::operator ++(void) { 338 if( _j-- ) { // Still working in current bin? 339 _key = _d->_bin[_i]._keyvals[_j+_j]; 340 _value = _d->_bin[_i]._keyvals[_j+_j+1]; 341 return; 342 } 343 344 while( ++_i < _d->_size ) { // Else scan for non-zero bucket 345 _j = _d->_bin[_i]._cnt; 346 if( !_j ) continue; 347 _j--; 348 _key = _d->_bin[_i]._keyvals[_j+_j]; 349 _value = _d->_bin[_i]._keyvals[_j+_j+1]; 350 return; 351 } 352 _key = _value = NULL; 353} 354