| 1 | /* Output the generated parsing program for Bison. |
| 2 | |
| 3 | Copyright (C) 1984, 1986, 1989, 1992, 2000-2006, 2009-2015, 2018-2019 |
| 4 | Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of Bison, the GNU Compiler Compiler. |
| 7 | |
| 8 | This program 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 3 of the License, or |
| 11 | (at your option) any later version. |
| 12 | |
| 13 | This program 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 this program. If not, see <http://www.gnu.org/licenses/>. */ |
| 20 | |
| 21 | #include <config.h> |
| 22 | #include "system.h" |
| 23 | |
| 24 | #include <bitset.h> |
| 25 | #include <bitsetv.h> |
| 26 | |
| 27 | #include "complain.h" |
| 28 | #include "conflicts.h" |
| 29 | #include "files.h" |
| 30 | #include "getargs.h" |
| 31 | #include "gram.h" |
| 32 | #include "lalr.h" |
| 33 | #include "muscle-tab.h" |
| 34 | #include "reader.h" |
| 35 | #include "symtab.h" |
| 36 | #include "tables.h" |
| 37 | |
| 38 | /* Several tables are indexed both by state and nonterminal numbers. |
| 39 | We call such an index a 'vector'; i.e., a vector is either a state |
| 40 | or a nonterminal number. |
| 41 | |
| 42 | Of course vector_number_t ought to be wide enough to contain |
| 43 | state_number and symbol_number. */ |
| 44 | typedef int vector_number; |
| 45 | |
| 46 | #if 0 /* Not currently used. */ |
| 47 | static inline vector_number |
| 48 | state_number_to_vector_number (state_number s) |
| 49 | { |
| 50 | return s; |
| 51 | } |
| 52 | #endif |
| 53 | |
| 54 | static inline vector_number |
| 55 | symbol_number_to_vector_number (symbol_number sym) |
| 56 | { |
| 57 | return state_number_as_int (nstates) + sym - ntokens; |
| 58 | } |
| 59 | |
| 60 | int nvectors; |
| 61 | |
| 62 | |
| 63 | /* FROMS and TOS are indexed by vector_number. |
| 64 | |
| 65 | If VECTOR is a nonterminal, (FROMS[VECTOR], TOS[VECTOR]) form an |
| 66 | array of state numbers of the non defaulted GOTO on VECTOR. |
| 67 | |
| 68 | If VECTOR is a state, TOS[VECTOR] is the array of actions to do on |
| 69 | the (array of) symbols FROMS[VECTOR]. |
| 70 | |
| 71 | In both cases, TALLY[VECTOR] is the size of the arrays |
| 72 | FROMS[VECTOR], TOS[VECTOR]; and WIDTH[VECTOR] = |
| 73 | (FROMS[VECTOR][SIZE] - FROMS[VECTOR][0] + 1) where SIZE = |
| 74 | TALLY[VECTOR]. |
| 75 | |
| 76 | FROMS therefore contains symbol_number and action_number, |
| 77 | TOS state_number and action_number, |
| 78 | TALLY sizes, |
| 79 | WIDTH differences of FROMS. |
| 80 | |
| 81 | Let base_number be the type of FROMS, TOS, and WIDTH. */ |
| 82 | #define BASE_MAXIMUM INT_MAX |
| 83 | #define BASE_MINIMUM INT_MIN |
| 84 | |
| 85 | static base_number **froms; |
| 86 | static base_number **tos; |
| 87 | static unsigned **conflict_tos; |
| 88 | static size_t *tally; |
| 89 | static base_number *width; |
| 90 | |
| 91 | |
| 92 | /* For a given state, N = ACTROW[SYMBOL]: |
| 93 | |
| 94 | If N = 0, stands for 'run the default action'. |
| 95 | If N = MIN, stands for 'raise a syntax error'. |
| 96 | If N > 0, stands for 'shift SYMBOL and go to n'. |
| 97 | If N < 0, stands for 'reduce -N'. */ |
| 98 | typedef int action_number; |
| 99 | #define ACTION_NUMBER_MINIMUM INT_MIN |
| 100 | |
| 101 | static action_number *actrow; |
| 102 | |
| 103 | /* FROMS and TOS are reordered to be compressed. ORDER[VECTOR] is the |
| 104 | new vector number of VECTOR. We skip 'empty' vectors (i.e., |
| 105 | TALLY[VECTOR] = 0), and call these 'entries'. */ |
| 106 | static vector_number *order; |
| 107 | static int nentries; |
| 108 | |
| 109 | base_number *base = NULL; |
| 110 | /* A distinguished value of BASE, negative infinite. During the |
| 111 | computation equals to BASE_MINIMUM, later mapped to BASE_NINF to |
| 112 | keep parser tables small. */ |
| 113 | base_number base_ninf = 0; |
| 114 | /* Bitset representing an integer set in the range |
| 115 | -nstates..table_size (as an upper bound) */ |
| 116 | static bitset pos_set = NULL; |
| 117 | |
| 118 | static unsigned *conflrow; |
| 119 | unsigned *conflict_table; |
| 120 | unsigned *conflict_list; |
| 121 | int conflict_list_cnt; |
| 122 | static int conflict_list_free; |
| 123 | |
| 124 | /* TABLE_SIZE is the allocated size of both TABLE and CHECK. We start |
| 125 | with more or less the original hard-coded value (which was |
| 126 | SHRT_MAX). */ |
| 127 | static int table_size = 32768; |
| 128 | base_number *table; |
| 129 | base_number *check; |
| 130 | /* The value used in TABLE to denote explicit syntax errors |
| 131 | (%nonassoc), a negative infinite. First defaults to ACTION_NUMBER_MINIMUM, |
| 132 | but in order to keep small tables, renumbered as TABLE_ERROR, which |
| 133 | is the smallest (non error) value minus 1. */ |
| 134 | base_number table_ninf = 0; |
| 135 | static int lowzero; |
| 136 | int high; |
| 137 | |
| 138 | state_number *yydefgoto; |
| 139 | rule_number *yydefact; |
| 140 | |
| 141 | /*-------------------------------------------------------------------. |
| 142 | | If TABLE, CONFLICT_TABLE, and CHECK are too small to be addressed | |
| 143 | | at DESIRED, grow them. TABLE[DESIRED] can be used, so the desired | |
| 144 | | size is at least DESIRED + 1. | |
| 145 | `-------------------------------------------------------------------*/ |
| 146 | |
| 147 | static void |
| 148 | table_grow (int desired) |
| 149 | { |
| 150 | int old_size = table_size; |
| 151 | |
| 152 | while (table_size <= desired) |
| 153 | table_size *= 2; |
| 154 | |
| 155 | if (trace_flag & trace_resource) |
| 156 | fprintf (stderr, "growing tables from %d to %d\n" , |
| 157 | old_size, table_size); |
| 158 | |
| 159 | table = xnrealloc (table, table_size, sizeof *table); |
| 160 | memset (table + old_size, 0, |
| 161 | sizeof *table * (table_size - old_size)); |
| 162 | |
| 163 | conflict_table = xnrealloc (conflict_table, table_size, |
| 164 | sizeof *conflict_table); |
| 165 | memset (conflict_table + old_size, 0, |
| 166 | sizeof *conflict_table * (table_size - old_size)); |
| 167 | |
| 168 | check = xnrealloc (check, table_size, sizeof *check); |
| 169 | for (int i = old_size; i < table_size; ++i) |
| 170 | check[i] = -1; |
| 171 | |
| 172 | bitset_resize (pos_set, table_size + nstates); |
| 173 | } |
| 174 | |
| 175 | |
| 176 | |
| 177 | |
| 178 | /*-------------------------------------------------------------------. |
| 179 | | For GLR parsers, for each conflicted token in S, as indicated | |
| 180 | | by non-zero entries in CONFLROW, create a list of possible | |
| 181 | | reductions that are alternatives to the shift or reduction | |
| 182 | | currently recorded for that token in S. Store the alternative | |
| 183 | | reductions followed by a 0 in CONFLICT_LIST, updating | |
| 184 | | CONFLICT_LIST_CNT, and storing an index to the start of the list | |
| 185 | | back into CONFLROW. | |
| 186 | `-------------------------------------------------------------------*/ |
| 187 | |
| 188 | static void |
| 189 | conflict_row (state *s) |
| 190 | { |
| 191 | if (!nondeterministic_parser) |
| 192 | return; |
| 193 | |
| 194 | const reductions *reds = s->reductions; |
| 195 | for (state_number j = 0; j < ntokens; j += 1) |
| 196 | if (conflrow[j]) |
| 197 | { |
| 198 | conflrow[j] = conflict_list_cnt; |
| 199 | |
| 200 | /* Find all reductions for token J, and record all that do not |
| 201 | match ACTROW[J]. */ |
| 202 | for (int i = 0; i < reds->num; i += 1) |
| 203 | if (bitset_test (reds->lookahead_tokens[i], j) |
| 204 | && (actrow[j] |
| 205 | != rule_number_as_item_number (reds->rules[i]->number))) |
| 206 | { |
| 207 | aver (0 < conflict_list_free); |
| 208 | conflict_list[conflict_list_cnt] = reds->rules[i]->number + 1; |
| 209 | conflict_list_cnt += 1; |
| 210 | conflict_list_free -= 1; |
| 211 | } |
| 212 | |
| 213 | /* Leave a 0 at the end. */ |
| 214 | aver (0 < conflict_list_free); |
| 215 | conflict_list[conflict_list_cnt] = 0; |
| 216 | conflict_list_cnt += 1; |
| 217 | conflict_list_free -= 1; |
| 218 | } |
| 219 | } |
| 220 | |
| 221 | |
| 222 | /*------------------------------------------------------------------. |
| 223 | | Decide what to do for each type of token if seen as the | |
| 224 | | lookahead in specified state. The value returned is used as the | |
| 225 | | default action (yydefact) for the state. In addition, ACTROW is | |
| 226 | | filled with what to do for each kind of token, index by symbol | |
| 227 | | number, with zero meaning do the default action. The value | |
| 228 | | ACTION_NUMBER_MINIMUM, a very negative number, means this | |
| 229 | | situation is an error. The parser recognizes this value | |
| 230 | | specially. | |
| 231 | | | |
| 232 | | This is where conflicts are resolved. The loop over lookahead | |
| 233 | | rules considered lower-numbered rules last, and the last rule | |
| 234 | | considered that likes a token gets to handle it. | |
| 235 | | | |
| 236 | | For GLR parsers, also sets CONFLROW[SYM] to an index into | |
| 237 | | CONFLICT_LIST iff there is an unresolved conflict (s/r or r/r) | |
| 238 | | with symbol SYM. The default reduction is not used for a symbol | |
| 239 | | that has any such conflicts. | |
| 240 | `------------------------------------------------------------------*/ |
| 241 | |
| 242 | static rule * |
| 243 | action_row (state *s) |
| 244 | { |
| 245 | for (state_number i = 0; i < ntokens; i++) |
| 246 | actrow[i] = conflrow[i] = 0; |
| 247 | |
| 248 | reductions *reds = s->reductions; |
| 249 | bool conflicted = false; |
| 250 | if (reds->lookahead_tokens) |
| 251 | /* loop over all the rules available here which require |
| 252 | lookahead (in reverse order to give precedence to the first |
| 253 | rule) */ |
| 254 | for (int i = reds->num - 1; 0 <= i; --i) |
| 255 | /* and find each token which the rule finds acceptable |
| 256 | to come next */ |
| 257 | { |
| 258 | bitset_iterator biter; |
| 259 | int j; |
| 260 | BITSET_FOR_EACH (biter, reds->lookahead_tokens[i], j, 0) |
| 261 | { |
| 262 | /* and record this rule as the rule to use if that |
| 263 | token follows. */ |
| 264 | if (actrow[j] != 0) |
| 265 | { |
| 266 | conflicted = true; |
| 267 | conflrow[j] = 1; |
| 268 | } |
| 269 | actrow[j] = rule_number_as_item_number (reds->rules[i]->number); |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | /* Now see which tokens are allowed for shifts in this state. For |
| 274 | them, record the shift as the thing to do. So shift is preferred |
| 275 | to reduce. */ |
| 276 | transitions *trans = s->transitions; |
| 277 | /* Set to nonzero to inhibit having any default reduction. */ |
| 278 | bool nodefault = false; |
| 279 | { |
| 280 | int i; |
| 281 | FOR_EACH_SHIFT (trans, i) |
| 282 | { |
| 283 | symbol_number sym = TRANSITION_SYMBOL (trans, i); |
| 284 | state *shift_state = trans->states[i]; |
| 285 | |
| 286 | if (actrow[sym] != 0) |
| 287 | { |
| 288 | conflicted = true; |
| 289 | conflrow[sym] = 1; |
| 290 | } |
| 291 | actrow[sym] = state_number_as_int (shift_state->number); |
| 292 | |
| 293 | /* Do not use any default reduction if there is a shift for |
| 294 | error */ |
| 295 | if (sym == errtoken->content->number) |
| 296 | nodefault = true; |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | /* See which tokens are an explicit error in this state (due to |
| 301 | %nonassoc). For them, record ACTION_NUMBER_MINIMUM as the |
| 302 | action. */ |
| 303 | errs *errp = s->errs; |
| 304 | for (int i = 0; i < errp->num; i++) |
| 305 | { |
| 306 | symbol *sym = errp->symbols[i]; |
| 307 | actrow[sym->content->number] = ACTION_NUMBER_MINIMUM; |
| 308 | } |
| 309 | |
| 310 | /* Turn off default reductions where requested by the user. See |
| 311 | state_lookahead_tokens_count in lalr.c to understand when states are |
| 312 | labeled as consistent. */ |
| 313 | { |
| 314 | char *default_reductions = |
| 315 | muscle_percent_define_get ("lr.default-reduction" ); |
| 316 | if (STRNEQ (default_reductions, "most" ) && !s->consistent) |
| 317 | nodefault = true; |
| 318 | free (default_reductions); |
| 319 | } |
| 320 | |
| 321 | /* Now find the most common reduction and make it the default action |
| 322 | for this state. */ |
| 323 | rule *default_reduction = NULL; |
| 324 | if (reds->num >= 1 && !nodefault) |
| 325 | { |
| 326 | if (s->consistent) |
| 327 | default_reduction = reds->rules[0]; |
| 328 | else |
| 329 | { |
| 330 | int max = 0; |
| 331 | for (int i = 0; i < reds->num; i++) |
| 332 | { |
| 333 | int count = 0; |
| 334 | rule *r = reds->rules[i]; |
| 335 | for (symbol_number j = 0; j < ntokens; j++) |
| 336 | if (actrow[j] == rule_number_as_item_number (r->number)) |
| 337 | count++; |
| 338 | |
| 339 | if (count > max) |
| 340 | { |
| 341 | max = count; |
| 342 | default_reduction = r; |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | /* GLR parsers need space for conflict lists, so we can't |
| 347 | default conflicted entries. For non-conflicted entries |
| 348 | or as long as we are not building a GLR parser, |
| 349 | actions that match the default are replaced with zero, |
| 350 | which means "use the default". */ |
| 351 | |
| 352 | if (0 < max) |
| 353 | for (symbol_number j = 0; j < ntokens; j++) |
| 354 | if (actrow[j] |
| 355 | == rule_number_as_item_number (default_reduction->number) |
| 356 | && ! (nondeterministic_parser && conflrow[j])) |
| 357 | actrow[j] = 0; |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | /* If have no default reduction, the default is an error. |
| 362 | So replace any action which says "error" with "use default". */ |
| 363 | |
| 364 | if (!default_reduction) |
| 365 | for (symbol_number i = 0; i < ntokens; i++) |
| 366 | if (actrow[i] == ACTION_NUMBER_MINIMUM) |
| 367 | actrow[i] = 0; |
| 368 | |
| 369 | if (conflicted) |
| 370 | conflict_row (s); |
| 371 | |
| 372 | return default_reduction; |
| 373 | } |
| 374 | |
| 375 | |
| 376 | /*----------------------------------------. |
| 377 | | Set FROMS, TOS, TALLY and WIDTH for S. | |
| 378 | `----------------------------------------*/ |
| 379 | |
| 380 | static void |
| 381 | save_row (state_number s) |
| 382 | { |
| 383 | /* Number of non default actions in S. */ |
| 384 | size_t count = 0; |
| 385 | for (symbol_number i = 0; i < ntokens; i++) |
| 386 | if (actrow[i] != 0) |
| 387 | count++; |
| 388 | |
| 389 | if (count) |
| 390 | { |
| 391 | /* Allocate non defaulted actions. */ |
| 392 | base_number *sp1 = froms[s] = xnmalloc (count, sizeof *sp1); |
| 393 | base_number *sp2 = tos[s] = xnmalloc (count, sizeof *sp2); |
| 394 | unsigned *sp3 = conflict_tos[s] = |
| 395 | nondeterministic_parser ? xnmalloc (count, sizeof *sp3) : NULL; |
| 396 | |
| 397 | /* Store non defaulted actions. */ |
| 398 | for (symbol_number i = 0; i < ntokens; i++) |
| 399 | if (actrow[i] != 0) |
| 400 | { |
| 401 | *sp1++ = i; |
| 402 | *sp2++ = actrow[i]; |
| 403 | if (nondeterministic_parser) |
| 404 | *sp3++ = conflrow[i]; |
| 405 | } |
| 406 | |
| 407 | tally[s] = count; |
| 408 | width[s] = sp1[-1] - froms[s][0] + 1; |
| 409 | } |
| 410 | } |
| 411 | |
| 412 | |
| 413 | /*------------------------------------------------------------------. |
| 414 | | Figure out the actions for the specified state, indexed by | |
| 415 | | lookahead token type. | |
| 416 | | | |
| 417 | | The YYDEFACT table is output now. The detailed info is saved for | |
| 418 | | putting into YYTABLE later. | |
| 419 | `------------------------------------------------------------------*/ |
| 420 | |
| 421 | static void |
| 422 | token_actions (void) |
| 423 | { |
| 424 | int nconflict = nondeterministic_parser ? conflicts_total_count () : 0; |
| 425 | |
| 426 | yydefact = xnmalloc (nstates, sizeof *yydefact); |
| 427 | |
| 428 | actrow = xnmalloc (ntokens, sizeof *actrow); |
| 429 | conflrow = xnmalloc (ntokens, sizeof *conflrow); |
| 430 | |
| 431 | conflict_list = xnmalloc (1 + 2 * nconflict, sizeof *conflict_list); |
| 432 | conflict_list_free = 2 * nconflict; |
| 433 | conflict_list_cnt = 1; |
| 434 | |
| 435 | /* Find the rules which are reduced. */ |
| 436 | if (!nondeterministic_parser) |
| 437 | for (rule_number r = 0; r < nrules; ++r) |
| 438 | rules[r].useful = false; |
| 439 | |
| 440 | for (state_number i = 0; i < nstates; ++i) |
| 441 | { |
| 442 | rule *default_reduction = action_row (states[i]); |
| 443 | yydefact[i] = default_reduction ? default_reduction->number + 1 : 0; |
| 444 | save_row (i); |
| 445 | |
| 446 | /* Now that the parser was computed, we can find which rules are |
| 447 | really reduced, and which are not because of SR or RR |
| 448 | conflicts. */ |
| 449 | if (!nondeterministic_parser) |
| 450 | { |
| 451 | for (symbol_number j = 0; j < ntokens; ++j) |
| 452 | if (actrow[j] < 0 && actrow[j] != ACTION_NUMBER_MINIMUM) |
| 453 | rules[item_number_as_rule_number (actrow[j])].useful = true; |
| 454 | if (yydefact[i]) |
| 455 | rules[yydefact[i] - 1].useful = true; |
| 456 | } |
| 457 | } |
| 458 | free (actrow); |
| 459 | free (conflrow); |
| 460 | } |
| 461 | |
| 462 | |
| 463 | /*------------------------------------------------------------------. |
| 464 | | Compute FROMS[VECTOR], TOS[VECTOR], TALLY[VECTOR], WIDTH[VECTOR], | |
| 465 | | i.e., the information related to non defaulted GOTO on the nterm | |
| 466 | | SYM. | |
| 467 | | | |
| 468 | | DEFAULT_STATE is the principal destination on SYM, i.e., the | |
| 469 | | default GOTO destination on SYM. | |
| 470 | `------------------------------------------------------------------*/ |
| 471 | |
| 472 | static void |
| 473 | save_column (symbol_number sym, state_number default_state) |
| 474 | { |
| 475 | const goto_number begin = goto_map[sym - ntokens]; |
| 476 | const goto_number end = goto_map[sym - ntokens + 1]; |
| 477 | |
| 478 | /* Number of non default GOTO. */ |
| 479 | size_t count = 0; |
| 480 | for (goto_number i = begin; i < end; i++) |
| 481 | if (to_state[i] != default_state) |
| 482 | count++; |
| 483 | |
| 484 | if (count) |
| 485 | { |
| 486 | /* Allocate room for non defaulted gotos. */ |
| 487 | vector_number symno = symbol_number_to_vector_number (sym); |
| 488 | base_number *sp1 = froms[symno] = xnmalloc (count, sizeof *sp1); |
| 489 | base_number *sp2 = tos[symno] = xnmalloc (count, sizeof *sp2); |
| 490 | |
| 491 | /* Store the state numbers of the non defaulted gotos. */ |
| 492 | for (goto_number i = begin; i < end; i++) |
| 493 | if (to_state[i] != default_state) |
| 494 | { |
| 495 | *sp1++ = from_state[i]; |
| 496 | *sp2++ = to_state[i]; |
| 497 | } |
| 498 | |
| 499 | tally[symno] = count; |
| 500 | width[symno] = sp1[-1] - froms[symno][0] + 1; |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | |
| 505 | /*----------------------------------------------------------------. |
| 506 | | The default state for SYM: the state which is 'the' most common | |
| 507 | | GOTO destination on SYM (an nterm). | |
| 508 | `----------------------------------------------------------------*/ |
| 509 | |
| 510 | static state_number |
| 511 | default_goto (symbol_number sym, size_t state_count[]) |
| 512 | { |
| 513 | const goto_number begin = goto_map[sym - ntokens]; |
| 514 | const goto_number end = goto_map[sym - ntokens + 1]; |
| 515 | state_number res = -1; |
| 516 | |
| 517 | if (begin != end) |
| 518 | { |
| 519 | for (state_number s = 0; s < nstates; s++) |
| 520 | state_count[s] = 0; |
| 521 | |
| 522 | for (goto_number i = begin; i < end; i++) |
| 523 | state_count[to_state[i]]++; |
| 524 | |
| 525 | size_t max = 0; |
| 526 | for (state_number s = 0; s < nstates; s++) |
| 527 | if (max < state_count[s]) |
| 528 | { |
| 529 | max = state_count[s]; |
| 530 | res = s; |
| 531 | } |
| 532 | } |
| 533 | return res; |
| 534 | } |
| 535 | |
| 536 | |
| 537 | /*-------------------------------------------------------------------. |
| 538 | | Figure out what to do after reducing with each rule, depending on | |
| 539 | | the saved state from before the beginning of parsing the data that | |
| 540 | | matched this rule. | |
| 541 | | | |
| 542 | | The YYDEFGOTO table is output now. The detailed info is saved for | |
| 543 | | putting into YYTABLE later. | |
| 544 | `-------------------------------------------------------------------*/ |
| 545 | |
| 546 | static void |
| 547 | goto_actions (void) |
| 548 | { |
| 549 | size_t *state_count = xnmalloc (nstates, sizeof *state_count); |
| 550 | yydefgoto = xnmalloc (nvars, sizeof *yydefgoto); |
| 551 | |
| 552 | /* For a given nterm I, STATE_COUNT[S] is the number of times there |
| 553 | is a GOTO to S on I. */ |
| 554 | for (symbol_number i = ntokens; i < nsyms; ++i) |
| 555 | { |
| 556 | state_number default_state = default_goto (i, state_count); |
| 557 | save_column (i, default_state); |
| 558 | yydefgoto[i - ntokens] = default_state; |
| 559 | } |
| 560 | free (state_count); |
| 561 | } |
| 562 | |
| 563 | |
| 564 | /*------------------------------------------------------------------. |
| 565 | | Compute ORDER, a reordering of vectors, in order to decide how to | |
| 566 | | pack the actions and gotos information into yytable. | |
| 567 | `------------------------------------------------------------------*/ |
| 568 | |
| 569 | static void |
| 570 | sort_actions (void) |
| 571 | { |
| 572 | nentries = 0; |
| 573 | |
| 574 | for (int i = 0; i < nvectors; i++) |
| 575 | if (0 < tally[i]) |
| 576 | { |
| 577 | const size_t t = tally[i]; |
| 578 | const int w = width[i]; |
| 579 | int j = nentries - 1; |
| 580 | |
| 581 | while (0 <= j && width[order[j]] < w) |
| 582 | j--; |
| 583 | |
| 584 | while (0 <= j && width[order[j]] == w && tally[order[j]] < t) |
| 585 | j--; |
| 586 | |
| 587 | for (int k = nentries - 1; k > j; k--) |
| 588 | order[k + 1] = order[k]; |
| 589 | |
| 590 | order[j + 1] = i; |
| 591 | nentries++; |
| 592 | } |
| 593 | } |
| 594 | |
| 595 | |
| 596 | /* If VECTOR is a state whose actions (reflected by FROMS, TOS, TALLY |
| 597 | and WIDTH of VECTOR) are common to a previous state, return this |
| 598 | state number. |
| 599 | |
| 600 | In any other case, return -1. */ |
| 601 | |
| 602 | static state_number |
| 603 | matching_state (vector_number vector) |
| 604 | { |
| 605 | vector_number i = order[vector]; |
| 606 | /* If VECTOR is a nterm, return -1. */ |
| 607 | if (i < nstates) |
| 608 | { |
| 609 | size_t t = tally[i]; |
| 610 | int w = width[i]; |
| 611 | |
| 612 | /* If VECTOR has GLR conflicts, return -1 */ |
| 613 | if (conflict_tos[i] != NULL) |
| 614 | for (int j = 0; j < t; j += 1) |
| 615 | if (conflict_tos[i][j] != 0) |
| 616 | return -1; |
| 617 | |
| 618 | for (int prev = vector - 1; 0 <= prev; prev--) |
| 619 | { |
| 620 | vector_number j = order[prev]; |
| 621 | /* Given how ORDER was computed, if the WIDTH or TALLY is |
| 622 | different, there cannot be a matching state. */ |
| 623 | if (width[j] != w || tally[j] != t) |
| 624 | return -1; |
| 625 | else |
| 626 | { |
| 627 | bool match = true; |
| 628 | for (int k = 0; match && k < t; k++) |
| 629 | if (tos[j][k] != tos[i][k] |
| 630 | || froms[j][k] != froms[i][k] |
| 631 | || (conflict_tos[j] != NULL && conflict_tos[j][k] != 0)) |
| 632 | match = false; |
| 633 | if (match) |
| 634 | return j; |
| 635 | } |
| 636 | } |
| 637 | } |
| 638 | return -1; |
| 639 | } |
| 640 | |
| 641 | |
| 642 | static base_number |
| 643 | pack_vector (vector_number vector) |
| 644 | { |
| 645 | vector_number i = order[vector]; |
| 646 | size_t t = tally[i]; |
| 647 | base_number *from = froms[i]; |
| 648 | base_number *to = tos[i]; |
| 649 | unsigned *conflict_to = conflict_tos[i]; |
| 650 | |
| 651 | aver (t != 0); |
| 652 | |
| 653 | for (base_number res = lowzero - from[0]; ; res++) |
| 654 | { |
| 655 | bool ok = true; |
| 656 | aver (res < table_size); |
| 657 | { |
| 658 | for (int k = 0; ok && k < t; k++) |
| 659 | { |
| 660 | int loc = res + state_number_as_int (from[k]); |
| 661 | if (table_size <= loc) |
| 662 | table_grow (loc); |
| 663 | |
| 664 | if (table[loc] != 0) |
| 665 | ok = false; |
| 666 | } |
| 667 | |
| 668 | if (ok && bitset_test (pos_set, nstates + res)) |
| 669 | ok = false; |
| 670 | } |
| 671 | |
| 672 | if (ok) |
| 673 | { |
| 674 | int loc PACIFY_CC (= -1); |
| 675 | for (int k = 0; k < t; k++) |
| 676 | { |
| 677 | loc = res + state_number_as_int (from[k]); |
| 678 | table[loc] = to[k]; |
| 679 | if (nondeterministic_parser && conflict_to != NULL) |
| 680 | conflict_table[loc] = conflict_to[k]; |
| 681 | check[loc] = from[k]; |
| 682 | } |
| 683 | |
| 684 | while (table[lowzero] != 0) |
| 685 | lowzero++; |
| 686 | |
| 687 | if (high < loc) |
| 688 | high = loc; |
| 689 | |
| 690 | aver (BASE_MINIMUM <= res && res <= BASE_MAXIMUM); |
| 691 | return res; |
| 692 | } |
| 693 | } |
| 694 | } |
| 695 | |
| 696 | |
| 697 | /*-------------------------------------------------------------. |
| 698 | | Remap the negative infinite in TAB from NINF to the greatest | |
| 699 | | possible smallest value. Return it. | |
| 700 | | | |
| 701 | | In most case this allows us to use shorts instead of ints in | |
| 702 | | parsers. | |
| 703 | `-------------------------------------------------------------*/ |
| 704 | |
| 705 | static base_number |
| 706 | table_ninf_remap (base_number tab[], int size, base_number ninf) |
| 707 | { |
| 708 | base_number res = 0; |
| 709 | |
| 710 | for (int i = 0; i < size; i++) |
| 711 | if (tab[i] < res && tab[i] != ninf) |
| 712 | res = tab[i]; |
| 713 | |
| 714 | --res; |
| 715 | |
| 716 | for (int i = 0; i < size; i++) |
| 717 | if (tab[i] == ninf) |
| 718 | tab[i] = res; |
| 719 | |
| 720 | return res; |
| 721 | } |
| 722 | |
| 723 | static void |
| 724 | pack_table (void) |
| 725 | { |
| 726 | base = xnmalloc (nvectors, sizeof *base); |
| 727 | pos_set = bitset_create (table_size + nstates, BITSET_FRUGAL); |
| 728 | table = xcalloc (table_size, sizeof *table); |
| 729 | conflict_table = xcalloc (table_size, sizeof *conflict_table); |
| 730 | check = xnmalloc (table_size, sizeof *check); |
| 731 | |
| 732 | lowzero = 0; |
| 733 | high = 0; |
| 734 | |
| 735 | for (int i = 0; i < nvectors; i++) |
| 736 | base[i] = BASE_MINIMUM; |
| 737 | |
| 738 | for (int i = 0; i < table_size; i++) |
| 739 | check[i] = -1; |
| 740 | |
| 741 | for (int i = 0; i < nentries; i++) |
| 742 | { |
| 743 | state_number s = matching_state (i); |
| 744 | base_number place; |
| 745 | |
| 746 | if (s < 0) |
| 747 | /* A new set of state actions, or a nonterminal. */ |
| 748 | place = pack_vector (i); |
| 749 | else |
| 750 | /* Action of I were already coded for S. */ |
| 751 | place = base[s]; |
| 752 | |
| 753 | /* Store PLACE into POS_SET. PLACE might not belong to the set |
| 754 | of possible values for instance with useless tokens. It |
| 755 | would be more satisfying to eliminate the need for this |
| 756 | 'if'. */ |
| 757 | if (0 <= nstates + place) |
| 758 | bitset_set (pos_set, nstates + place); |
| 759 | base[order[i]] = place; |
| 760 | } |
| 761 | |
| 762 | /* Use the greatest possible negative infinites. */ |
| 763 | base_ninf = table_ninf_remap (base, nvectors, BASE_MINIMUM); |
| 764 | table_ninf = table_ninf_remap (table, high + 1, ACTION_NUMBER_MINIMUM); |
| 765 | |
| 766 | bitset_free (pos_set); |
| 767 | } |
| 768 | |
| 769 | |
| 770 | |
| 771 | /*-----------------------------------------------------------------. |
| 772 | | Compute and output yydefact, yydefgoto, yypact, yypgoto, yytable | |
| 773 | | and yycheck. | |
| 774 | `-----------------------------------------------------------------*/ |
| 775 | |
| 776 | void |
| 777 | tables_generate (void) |
| 778 | { |
| 779 | /* This is a poor way to make sure the sizes are properly |
| 780 | correlated. In particular the signedness is not taken into |
| 781 | account. But it's not useless. */ |
| 782 | verify (sizeof nstates <= sizeof nvectors); |
| 783 | verify (sizeof nvars <= sizeof nvectors); |
| 784 | |
| 785 | nvectors = state_number_as_int (nstates) + nvars; |
| 786 | |
| 787 | froms = xcalloc (nvectors, sizeof *froms); |
| 788 | tos = xcalloc (nvectors, sizeof *tos); |
| 789 | conflict_tos = xcalloc (nvectors, sizeof *conflict_tos); |
| 790 | tally = xcalloc (nvectors, sizeof *tally); |
| 791 | width = xnmalloc (nvectors, sizeof *width); |
| 792 | |
| 793 | token_actions (); |
| 794 | |
| 795 | goto_actions (); |
| 796 | free (goto_map); |
| 797 | free (from_state); |
| 798 | free (to_state); |
| 799 | |
| 800 | order = xcalloc (nvectors, sizeof *order); |
| 801 | sort_actions (); |
| 802 | pack_table (); |
| 803 | free (order); |
| 804 | |
| 805 | free (tally); |
| 806 | free (width); |
| 807 | |
| 808 | for (int i = 0; i < nvectors; i++) |
| 809 | { |
| 810 | free (froms[i]); |
| 811 | free (tos[i]); |
| 812 | free (conflict_tos[i]); |
| 813 | } |
| 814 | |
| 815 | free (froms); |
| 816 | free (tos); |
| 817 | free (conflict_tos); |
| 818 | } |
| 819 | |
| 820 | |
| 821 | /*-------------------------. |
| 822 | | Free the parser tables. | |
| 823 | `-------------------------*/ |
| 824 | |
| 825 | void |
| 826 | tables_free (void) |
| 827 | { |
| 828 | free (base); |
| 829 | free (conflict_table); |
| 830 | free (conflict_list); |
| 831 | free (table); |
| 832 | free (check); |
| 833 | free (yydefgoto); |
| 834 | free (yydefact); |
| 835 | } |
| 836 | |