| 1 | /* Compute look-ahead criteria for Bison. |
| 2 | |
| 3 | Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002 |
| 4 | Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of Bison, the GNU Compiler Compiler. |
| 7 | |
| 8 | Bison is free software; you can redistribute it and/or modify |
| 9 | it under the terms of the GNU General Public License as published by |
| 10 | the Free Software Foundation; either version 2, or (at your option) |
| 11 | any later version. |
| 12 | |
| 13 | Bison is distributed in the hope that it will be useful, |
| 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | GNU General Public License for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with Bison; see the file COPYING. If not, write to |
| 20 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 21 | Boston, MA 02111-1307, USA. */ |
| 22 | |
| 23 | |
| 24 | /* Compute how to make the finite state machine deterministic; find |
| 25 | which rules need lookahead in each state, and which lookahead |
| 26 | tokens they accept. */ |
| 27 | |
| 28 | #include "system.h" |
| 29 | |
| 30 | #include <bitset.h> |
| 31 | #include <bitsetv.h> |
| 32 | #include <quotearg.h> |
| 33 | |
| 34 | #include "LR0.h" |
| 35 | #include "complain.h" |
| 36 | #include "derives.h" |
| 37 | #include "getargs.h" |
| 38 | #include "gram.h" |
| 39 | #include "lalr.h" |
| 40 | #include "nullable.h" |
| 41 | #include "reader.h" |
| 42 | #include "relation.h" |
| 43 | #include "symtab.h" |
| 44 | |
| 45 | goto_number *goto_map = NULL; |
| 46 | static goto_number ngotos = 0; |
| 47 | state_number *from_state = NULL; |
| 48 | state_number *to_state = NULL; |
| 49 | |
| 50 | /* Linked list of goto numbers. */ |
| 51 | typedef struct goto_list |
| 52 | { |
| 53 | struct goto_list *next; |
| 54 | goto_number value; |
| 55 | } goto_list; |
| 56 | |
| 57 | |
| 58 | /* LA is a LR by NTOKENS matrix of bits. LA[l, i] is 1 if the rule |
| 59 | LArule[l] is applicable in the appropriate state when the next |
| 60 | token is symbol i. If LA[l, i] and LA[l, j] are both 1 for i != j, |
| 61 | it is a conflict. */ |
| 62 | |
| 63 | static bitsetv LA = NULL; |
| 64 | size_t nLA; |
| 65 | |
| 66 | |
| 67 | /* And for the famous F variable, which name is so descriptive that a |
| 68 | comment is hardly needed. <grin>. */ |
| 69 | static bitsetv F = NULL; |
| 70 | |
| 71 | static goto_number **includes; |
| 72 | static goto_list **lookback; |
| 73 | |
| 74 | |
| 75 | |
| 76 | |
| 77 | static void |
| 78 | set_goto_map (void) |
| 79 | { |
| 80 | state_number s; |
| 81 | goto_number *temp_map; |
| 82 | |
| 83 | CALLOC (goto_map, nvars + 1); |
| 84 | CALLOC (temp_map, nvars + 1); |
| 85 | |
| 86 | ngotos = 0; |
| 87 | for (s = 0; s < nstates; ++s) |
| 88 | { |
| 89 | transitions *sp = states[s]->transitions; |
| 90 | int i; |
| 91 | for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) |
| 92 | { |
| 93 | if (ngotos >= GOTO_NUMBER_MAXIMUM) |
| 94 | abort (); |
| 95 | ngotos++; |
| 96 | goto_map[TRANSITION_SYMBOL (sp, i) - ntokens]++; |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | { |
| 101 | int k = 0; |
| 102 | int i; |
| 103 | for (i = ntokens; i < nsyms; i++) |
| 104 | { |
| 105 | temp_map[i - ntokens] = k; |
| 106 | k += goto_map[i - ntokens]; |
| 107 | } |
| 108 | |
| 109 | for (i = ntokens; i < nsyms; i++) |
| 110 | goto_map[i - ntokens] = temp_map[i - ntokens]; |
| 111 | |
| 112 | goto_map[nsyms - ntokens] = ngotos; |
| 113 | temp_map[nsyms - ntokens] = ngotos; |
| 114 | } |
| 115 | |
| 116 | CALLOC (from_state, ngotos); |
| 117 | CALLOC (to_state, ngotos); |
| 118 | |
| 119 | for (s = 0; s < nstates; ++s) |
| 120 | { |
| 121 | transitions *sp = states[s]->transitions; |
| 122 | int i; |
| 123 | for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) |
| 124 | { |
| 125 | int k = temp_map[TRANSITION_SYMBOL (sp, i) - ntokens]++; |
| 126 | from_state[k] = s; |
| 127 | to_state[k] = sp->states[i]->number; |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | XFREE (temp_map); |
| 132 | } |
| 133 | |
| 134 | |
| 135 | |
| 136 | /*----------------------------------------------------------. |
| 137 | | Map a state/symbol pair into its numeric representation. | |
| 138 | `----------------------------------------------------------*/ |
| 139 | |
| 140 | static int |
| 141 | map_goto (state_number s0, symbol_number sym) |
| 142 | { |
| 143 | int high; |
| 144 | int low; |
| 145 | int middle; |
| 146 | state_number s; |
| 147 | |
| 148 | low = goto_map[sym - ntokens]; |
| 149 | high = goto_map[sym - ntokens + 1] - 1; |
| 150 | |
| 151 | for (;;) |
| 152 | { |
| 153 | if (high < low) |
| 154 | abort (); |
| 155 | middle = (low + high) / 2; |
| 156 | s = from_state[middle]; |
| 157 | if (s == s0) |
| 158 | return middle; |
| 159 | else if (s < s0) |
| 160 | low = middle + 1; |
| 161 | else |
| 162 | high = middle - 1; |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | |
| 167 | static void |
| 168 | initialize_F (void) |
| 169 | { |
| 170 | goto_number **reads = CALLOC (reads, ngotos); |
| 171 | goto_number *edge = CALLOC (edge, ngotos + 1); |
| 172 | int nedges = 0; |
| 173 | |
| 174 | int i; |
| 175 | |
| 176 | F = bitsetv_create (ngotos, ntokens, BITSET_FIXED); |
| 177 | |
| 178 | for (i = 0; i < ngotos; i++) |
| 179 | { |
| 180 | state_number stateno = to_state[i]; |
| 181 | transitions *sp = states[stateno]->transitions; |
| 182 | |
| 183 | int j; |
| 184 | FOR_EACH_SHIFT (sp, j) |
| 185 | bitset_set (F[i], TRANSITION_SYMBOL (sp, j)); |
| 186 | |
| 187 | for (; j < sp->num; j++) |
| 188 | { |
| 189 | symbol_number sym = TRANSITION_SYMBOL (sp, j); |
| 190 | if (nullable[sym - ntokens]) |
| 191 | edge[nedges++] = map_goto (stateno, sym); |
| 192 | } |
| 193 | |
| 194 | if (nedges) |
| 195 | { |
| 196 | CALLOC (reads[i], nedges + 1); |
| 197 | memcpy (reads[i], edge, nedges * sizeof (edge[0])); |
| 198 | reads[i][nedges] = -1; |
| 199 | nedges = 0; |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | relation_digraph (reads, ngotos, &F); |
| 204 | |
| 205 | for (i = 0; i < ngotos; i++) |
| 206 | XFREE (reads[i]); |
| 207 | |
| 208 | XFREE (reads); |
| 209 | XFREE (edge); |
| 210 | } |
| 211 | |
| 212 | |
| 213 | static void |
| 214 | add_lookback_edge (state *s, rule *r, int gotono) |
| 215 | { |
| 216 | int ri = state_reduction_find (s, r); |
| 217 | goto_list *sp = MALLOC (sp, 1); |
| 218 | sp->next = lookback[(s->reductions->lookaheads - LA) + ri]; |
| 219 | sp->value = gotono; |
| 220 | lookback[(s->reductions->lookaheads - LA) + ri] = sp; |
| 221 | } |
| 222 | |
| 223 | |
| 224 | |
| 225 | static void |
| 226 | build_relations (void) |
| 227 | { |
| 228 | goto_number *edge = CALLOC (edge, ngotos + 1); |
| 229 | state_number *states1 = CALLOC (states1, ritem_longest_rhs () + 1); |
| 230 | int i; |
| 231 | |
| 232 | CALLOC (includes, ngotos); |
| 233 | |
| 234 | for (i = 0; i < ngotos; i++) |
| 235 | { |
| 236 | int nedges = 0; |
| 237 | symbol_number symbol1 = states[to_state[i]]->accessing_symbol; |
| 238 | rule **rulep; |
| 239 | |
| 240 | for (rulep = derives[symbol1 - ntokens]; *rulep; rulep++) |
| 241 | { |
| 242 | int done; |
| 243 | int length = 1; |
| 244 | item_number *rp; |
| 245 | state *s = states[from_state[i]]; |
| 246 | states1[0] = s->number; |
| 247 | |
| 248 | for (rp = (*rulep)->rhs; *rp >= 0; rp++) |
| 249 | { |
| 250 | s = transitions_to (s->transitions, |
| 251 | item_number_as_symbol_number (*rp)); |
| 252 | states1[length++] = s->number; |
| 253 | } |
| 254 | |
| 255 | if (!s->consistent) |
| 256 | add_lookback_edge (s, *rulep, i); |
| 257 | |
| 258 | length--; |
| 259 | done = 0; |
| 260 | while (!done) |
| 261 | { |
| 262 | done = 1; |
| 263 | rp--; |
| 264 | /* JF added rp>=ritem && I hope to god its right! */ |
| 265 | if (rp >= ritem && ISVAR (*rp)) |
| 266 | { |
| 267 | /* Downcasting from item_number to symbol_number. */ |
| 268 | edge[nedges++] = map_goto (states1[--length], |
| 269 | item_number_as_symbol_number (*rp)); |
| 270 | if (nullable[*rp - ntokens]) |
| 271 | done = 0; |
| 272 | } |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | if (nedges) |
| 277 | { |
| 278 | int j; |
| 279 | CALLOC (includes[i], nedges + 1); |
| 280 | for (j = 0; j < nedges; j++) |
| 281 | includes[i][j] = edge[j]; |
| 282 | includes[i][nedges] = -1; |
| 283 | } |
| 284 | } |
| 285 | |
| 286 | XFREE (edge); |
| 287 | XFREE (states1); |
| 288 | |
| 289 | relation_transpose (&includes, ngotos); |
| 290 | } |
| 291 | |
| 292 | |
| 293 | |
| 294 | static void |
| 295 | compute_FOLLOWS (void) |
| 296 | { |
| 297 | int i; |
| 298 | |
| 299 | relation_digraph (includes, ngotos, &F); |
| 300 | |
| 301 | for (i = 0; i < ngotos; i++) |
| 302 | XFREE (includes[i]); |
| 303 | |
| 304 | XFREE (includes); |
| 305 | } |
| 306 | |
| 307 | |
| 308 | static void |
| 309 | compute_lookaheads (void) |
| 310 | { |
| 311 | size_t i; |
| 312 | goto_list *sp; |
| 313 | |
| 314 | for (i = 0; i < nLA; i++) |
| 315 | for (sp = lookback[i]; sp; sp = sp->next) |
| 316 | bitset_or (LA[i], LA[i], F[sp->value]); |
| 317 | |
| 318 | /* Free LOOKBACK. */ |
| 319 | for (i = 0; i < nLA; i++) |
| 320 | LIST_FREE (goto_list, lookback[i]); |
| 321 | |
| 322 | XFREE (lookback); |
| 323 | bitsetv_free (F); |
| 324 | } |
| 325 | |
| 326 | |
| 327 | /*-----------------------------------------------------------. |
| 328 | | Count the number of lookaheads required for S (NLOOKAHEADS | |
| 329 | | member). | |
| 330 | `-----------------------------------------------------------*/ |
| 331 | |
| 332 | static int |
| 333 | state_lookaheads_count (state *s) |
| 334 | { |
| 335 | int k; |
| 336 | int nlookaheads = 0; |
| 337 | reductions *rp = s->reductions; |
| 338 | transitions *sp = s->transitions; |
| 339 | |
| 340 | /* We need a lookahead either to distinguish different |
| 341 | reductions (i.e., there are two or more), or to distinguish a |
| 342 | reduction from a shift. Otherwise, it is straightforward, |
| 343 | and the state is `consistent'. */ |
| 344 | if (rp->num > 1 |
| 345 | || (rp->num == 1 && sp->num && |
| 346 | !TRANSITION_IS_DISABLED (sp, 0) && TRANSITION_IS_SHIFT (sp, 0))) |
| 347 | nlookaheads += rp->num; |
| 348 | else |
| 349 | s->consistent = 1; |
| 350 | |
| 351 | for (k = 0; k < sp->num; k++) |
| 352 | if (!TRANSITION_IS_DISABLED (sp, k) && TRANSITION_IS_ERROR (sp, k)) |
| 353 | { |
| 354 | s->consistent = 0; |
| 355 | break; |
| 356 | } |
| 357 | |
| 358 | return nlookaheads; |
| 359 | } |
| 360 | |
| 361 | |
| 362 | /*----------------------------------------------. |
| 363 | | Compute LA, NLA, and the lookaheads members. | |
| 364 | `----------------------------------------------*/ |
| 365 | |
| 366 | static void |
| 367 | initialize_LA (void) |
| 368 | { |
| 369 | state_number i; |
| 370 | bitsetv pLA; |
| 371 | |
| 372 | /* Compute the total number of reductions requiring a lookahead. */ |
| 373 | nLA = 0; |
| 374 | for (i = 0; i < nstates; i++) |
| 375 | nLA += state_lookaheads_count (states[i]); |
| 376 | /* Avoid having to special case 0. */ |
| 377 | if (!nLA) |
| 378 | nLA = 1; |
| 379 | |
| 380 | pLA = LA = bitsetv_create (nLA, ntokens, BITSET_FIXED); |
| 381 | CALLOC (lookback, nLA); |
| 382 | |
| 383 | /* Initialize the members LOOKAHEADS for each state which reductions |
| 384 | require lookaheads. */ |
| 385 | for (i = 0; i < nstates; i++) |
| 386 | { |
| 387 | int count = state_lookaheads_count (states[i]); |
| 388 | if (count) |
| 389 | { |
| 390 | states[i]->reductions->lookaheads = pLA; |
| 391 | pLA += count; |
| 392 | } |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | |
| 397 | /*---------------------------------------. |
| 398 | | Output the lookaheads for each state. | |
| 399 | `---------------------------------------*/ |
| 400 | |
| 401 | static void |
| 402 | lookaheads_print (FILE *out) |
| 403 | { |
| 404 | state_number i; |
| 405 | int j, k; |
| 406 | fprintf (out, "Lookaheads: BEGIN\n"); |
| 407 | for (i = 0; i < nstates; ++i) |
| 408 | { |
| 409 | reductions *reds = states[i]->reductions; |
| 410 | bitset_iterator iter; |
| 411 | int nlookaheads = 0; |
| 412 | |
| 413 | if (reds->lookaheads) |
| 414 | for (k = 0; k < reds->num; ++k) |
| 415 | if (reds->lookaheads[k]) |
| 416 | ++nlookaheads; |
| 417 | |
| 418 | fprintf (out, "State %d: %d lookaheads\n", |
| 419 | i, nlookaheads); |
| 420 | |
| 421 | if (reds->lookaheads) |
| 422 | for (j = 0; j < reds->num; ++j) |
| 423 | BITSET_FOR_EACH (iter, reds->lookaheads[j], k, 0) |
| 424 | { |
| 425 | fprintf (out, " on %d (%s) -> rule %d\n", |
| 426 | k, symbols[k]->tag, |
| 427 | reds->rules[j]->number); |
| 428 | }; |
| 429 | } |
| 430 | fprintf (out, "Lookaheads: END\n"); |
| 431 | } |
| 432 | |
| 433 | void |
| 434 | lalr (void) |
| 435 | { |
| 436 | initialize_LA (); |
| 437 | set_goto_map (); |
| 438 | initialize_F (); |
| 439 | build_relations (); |
| 440 | compute_FOLLOWS (); |
| 441 | compute_lookaheads (); |
| 442 | |
| 443 | if (trace_flag & trace_sets) |
| 444 | lookaheads_print (stderr); |
| 445 | } |
| 446 | |
| 447 | |
| 448 | void |
| 449 | lalr_free (void) |
| 450 | { |
| 451 | state_number s; |
| 452 | for (s = 0; s < nstates; ++s) |
| 453 | states[s]->reductions->lookaheads = NULL; |
| 454 | bitsetv_free (LA); |
| 455 | } |