src/reader.c

   1 /* Input parser for Bison
   2
   3    Copyright (C) 1984, 1986, 1989, 1992, 1998, 2000, 2001, 2002, 2003,
   4    2005, 2006, 2007, 2009 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 <quotearg.h>
  25
  26 #include "complain.h"
  27 #include "conflicts.h"
  28 #include "files.h"
  29 #include "getargs.h"
  30 #include "gram.h"
  31 #include "muscle-tab.h"
  32 #include "reader.h"
  33 #include "symlist.h"
  34 #include "symtab.h"
  35 #include "scan-gram.h"
  36 #include "scan-code.h"
  37
  38 static void check_and_convert_grammar (void);
  39
  40 static symbol_list *grammar = NULL;
  41 static bool start_flag = false;
  42 merger_list *merge_functions;
  43
  44 /* Was %union seen?  */
  45 bool union_seen = false;
  46
  47 /* Was a tag seen?  */
  48 bool tag_seen = false;
  49
  50 /* Should rules have a default precedence?  */
  51 bool default_prec = true;
  52 \f
  53 /*-----------------------.
  54 | Set the start symbol.  |
  55 `-----------------------*/
  56
  57 void
  58 grammar_start_symbol_set (symbol *sym, location loc)
  59 {
  60   if (start_flag)
  61     complain_at (loc, _("multiple %s declarations"), "%start");
  62   else
  63     {
  64       start_flag = true;
  65       startsymbol = sym;
  66       startsymbol_location = loc;
  67     }
  68 }
  69
  70 \f
  71
  72 /*------------------------------------------------------------------------.
  73 | Return the merger index for a merging function named NAME.  Records the |
  74 | function, if new, in MERGER_LIST.                                       |
  75 `------------------------------------------------------------------------*/
  76
  77 static int
  78 get_merge_function (uniqstr name)
  79 {
  80   merger_list *syms;
  81   merger_list head;
  82   int n;
  83
  84   if (! glr_parser)
  85     return 0;
  86
  87   head.next = merge_functions;
  88   for (syms = &head, n = 1; syms->next; syms = syms->next, n += 1)
  89     if (UNIQSTR_EQ (name, syms->next->name))
  90       break;
  91   if (syms->next == NULL)
  92     {
  93       syms->next = xmalloc (sizeof syms->next[0]);
  94       syms->next->name = uniqstr_new (name);
  95       /* After all symbol type declarations have been parsed, packgram invokes
  96          record_merge_function_type to set the type.  */
  97       syms->next->type = NULL;
  98       syms->next->next = NULL;
  99       merge_functions = head.next;
 100     }
 101   return n;
 102 }
 103
 104 /*-------------------------------------------------------------------------.
 105 | For the existing merging function with index MERGER, record the result   |
 106 | type as TYPE as required by the lhs of the rule whose %merge declaration |
 107 | is at DECLARATION_LOC.                                                   |
 108 `-------------------------------------------------------------------------*/
 109
 110 static void
 111 record_merge_function_type (int merger, uniqstr type, location declaration_loc)
 112 {
 113   int merger_find;
 114   merger_list *merge_function;
 115
 116   if (merger <= 0)
 117     return;
 118
 119   if (type == NULL)
 120     type = uniqstr_new ("");
 121
 122   merger_find = 1;
 123   for (merge_function = merge_functions;
 124        merge_function != NULL && merger_find != merger;
 125        merge_function = merge_function->next)
 126     merger_find += 1;
 127   aver (merge_function != NULL && merger_find == merger);
 128   if (merge_function->type != NULL && !UNIQSTR_EQ (merge_function->type, type))
 129     {
 130       complain_at (declaration_loc,
 131                    _("result type clash on merge function `%s': <%s> != <%s>"),
 132                    merge_function->name, type, merge_function->type);
 133       complain_at (merge_function->type_declaration_location,
 134                    _("previous declaration"));
 135     }
 136   merge_function->type = uniqstr_new (type);
 137   merge_function->type_declaration_location = declaration_loc;
 138 }
 139
 140 /*--------------------------------------.
 141 | Free all merge-function definitions.  |
 142 `--------------------------------------*/
 143
 144 void
 145 free_merger_functions (void)
 146 {
 147   merger_list *L0 = merge_functions;
 148   while (L0)
 149     {
 150       merger_list *L1 = L0->next;
 151       free (L0);
 152       L0 = L1;
 153     }
 154 }
 155
 156 \f
 157 /*-------------------------------------------------------------------.
 158 | Parse the input grammar into a one symbol_list structure.  Each    |
 159 | rule is represented by a sequence of symbols: the left hand side   |
 160 | followed by the contents of the right hand side, followed by a     |
 161 | null pointer instead of a symbol to terminate the rule.  The next  |
 162 | symbol is the lhs of the following rule.                           |
 163 |                                                                    |
 164 | All actions are copied out, labelled by the rule number they apply |
 165 | to.                                                                |
 166 `-------------------------------------------------------------------*/
 167
 168 /* The (currently) last symbol of GRAMMAR. */
 169 static symbol_list *grammar_end = NULL;
 170
 171 /* Append SYM to the grammar.  */
 172 static void
 173 grammar_symbol_append (symbol *sym, location loc)
 174 {
 175   symbol_list *p = symbol_list_sym_new (sym, loc);
 176
 177   if (grammar_end)
 178     grammar_end->next = p;
 179   else
 180     grammar = p;
 181
 182   grammar_end = p;
 183
 184   /* A null SYM stands for an end of rule; it is not an actual
 185      part of it.  */
 186   if (sym)
 187     ++nritems;
 188 }
 189
 190 /* The rule currently being defined, and the previous rule.
 191    CURRENT_RULE points to the first LHS of the current rule, while
 192    PREVIOUS_RULE_END points to the *end* of the previous rule (NULL).  */
 193 static symbol_list *current_rule = NULL;
 194 static symbol_list *previous_rule_end = NULL;
 195
 196
 197 /*----------------------------------------------.
 198 | Create a new rule for LHS in to the GRAMMAR.  |
 199 `----------------------------------------------*/
 200
 201 void
 202 grammar_current_rule_begin (symbol *lhs, location loc)
 203 {
 204   /* Start a new rule and record its lhs.  */
 205   ++nrules;
 206   previous_rule_end = grammar_end;
 207   grammar_symbol_append (lhs, loc);
 208   current_rule = grammar_end;
 209
 210   /* Mark the rule's lhs as a nonterminal if not already so.  */
 211   if (lhs->class == unknown_sym)
 212     {
 213       lhs->class = nterm_sym;
 214       lhs->number = nvars;
 215       ++nvars;
 216     }
 217   else if (lhs->class == token_sym)
 218     complain_at (loc, _("rule given for %s, which is a token"), lhs->tag);
 219 }
 220
 221
 222 /*----------------------------------------------------------------------.
 223 | A symbol should be used if either:                                    |
 224 |   1. It has a destructor.                                             |
 225 |   2. --warnings=midrule-values and the symbol is a mid-rule symbol    |
 226 |      (i.e., the generated LHS replacing a mid-rule action) that was   |
 227 |      assigned to or used, as in "exp: { $$ = 1; } { $$ = $1; }".      |
 228 `----------------------------------------------------------------------*/
 229
 230 static bool
 231 symbol_should_be_used (symbol_list const *s)
 232 {
 233   if (symbol_destructor_get (s->content.sym)->code)
 234     return true;
 235   if (warnings_flag & warnings_midrule_values)
 236     return ((s->midrule && s->midrule->action_props.is_value_used)
 237             || (s->midrule_parent_rule
 238                 && symbol_list_n_get (s->midrule_parent_rule,
 239                                       s->midrule_parent_rhs_index)
 240                      ->action_props.is_value_used));
 241   return false;
 242 }
 243
 244 /*----------------------------------------------------------------.
 245 | Check that the rule R is properly defined.  For instance, there |
 246 | should be no type clash on the default action.                  |
 247 `----------------------------------------------------------------*/
 248
 249 static void
 250 grammar_rule_check (const symbol_list *r)
 251 {
 252   /* Type check.
 253
 254      If there is an action, then there is nothing we can do: the user
 255      is allowed to shoot herself in the foot.
 256
 257      Don't worry about the default action if $$ is untyped, since $$'s
 258      value can't be used.  */
 259   if (!r->action_props.code && r->content.sym->type_name)
 260     {
 261       symbol *first_rhs = r->next->content.sym;
 262       /* If $$ is being set in default way, report if any type mismatch.  */
 263       if (first_rhs)
 264         {
 265           char const *lhs_type = r->content.sym->type_name;
 266           const char *rhs_type =
 267             first_rhs->type_name ? first_rhs->type_name : "";
 268           if (!UNIQSTR_EQ (lhs_type, rhs_type))
 269             warn_at (r->location,
 270                      _("type clash on default action: <%s> != <%s>"),
 271                      lhs_type, rhs_type);
 272         }
 273       /* Warn if there is no default for $$ but we need one.  */
 274       else
 275         warn_at (r->location,
 276                  _("empty rule for typed nonterminal, and no action"));
 277     }
 278
 279   /* Check that symbol values that should be used are in fact used.  */
 280   {
 281     symbol_list const *l = r;
 282     int n = 0;
 283     for (; l && l->content.sym; l = l->next, ++n)
 284       if (! (l->action_props.is_value_used
 285              || !symbol_should_be_used (l)
 286              /* The default action, $$ = $1, `uses' both.  */
 287              || (!r->action_props.code && (n == 0 || n == 1))))
 288         {
 289           if (n)
 290             warn_at (r->location, _("unused value: $%d"), n);
 291           else
 292             warn_at (r->location, _("unset value: $$"));
 293         }
 294   }
 295 }
 296
 297
 298 /*-------------------------------------.
 299 | End the currently being grown rule.  |
 300 `-------------------------------------*/
 301
 302 void
 303 grammar_current_rule_end (location loc)
 304 {
 305   /* Put an empty link in the list to mark the end of this rule  */
 306   grammar_symbol_append (NULL, grammar_end->location);
 307   current_rule->location = loc;
 308 }
 309
 310
 311 /*-------------------------------------------------------------------.
 312 | The previous action turns out the be a mid-rule action.  Attach it |
 313 | to the current rule, i.e., create a dummy symbol, attach it this   |
 314 | mid-rule action, and append this dummy nonterminal to the current  |
 315 | rule.                                                              |
 316 `-------------------------------------------------------------------*/
 317
 318 void
 319 grammar_midrule_action (void)
 320 {
 321   /* Since the action was written out with this rule's number, we must
 322      give the new rule this number by inserting the new rule before
 323      it.  */
 324
 325   /* Make a DUMMY nonterminal, whose location is that of the midrule
 326      action.  Create the MIDRULE.  */
 327   location dummy_location = current_rule->action_props.location;
 328   symbol *dummy = dummy_symbol_get (dummy_location);
 329   symbol_list *midrule = symbol_list_sym_new (dummy, dummy_location);
 330
 331   /* Make a new rule, whose body is empty, before the current one, so
 332      that the action just read can belong to it.  */
 333   ++nrules;
 334   ++nritems;
 335   /* Attach its location and actions to that of the DUMMY.  */
 336   midrule->location = dummy_location;
 337   code_props_rule_action_init (&midrule->action_props,
 338                                current_rule->action_props.code,
 339                                current_rule->action_props.location,
 340                                midrule);
 341   code_props_none_init (&current_rule->action_props);
 342
 343   if (previous_rule_end)
 344     previous_rule_end->next = midrule;
 345   else
 346     grammar = midrule;
 347
 348   /* End the dummy's rule.  */
 349   midrule->next = symbol_list_sym_new (NULL, dummy_location);
 350   midrule->next->next = current_rule;
 351
 352   previous_rule_end = midrule->next;
 353
 354   /* Insert the dummy nonterminal replacing the midrule action into
 355      the current rule.  Bind it to its dedicated rule.  */
 356   grammar_current_rule_symbol_append (dummy, dummy_location);
 357   grammar_end->midrule = midrule;
 358   midrule->midrule_parent_rule = current_rule;
 359   midrule->midrule_parent_rhs_index = symbol_list_length (current_rule->next);
 360 }
 361
 362 /* Set the precedence symbol of the current rule to PRECSYM. */
 363
 364 void
 365 grammar_current_rule_prec_set (symbol *precsym, location loc)
 366 {
 367   symbol_class_set (precsym, token_sym, loc, false);
 368   if (current_rule->ruleprec)
 369     complain_at (loc, _("only one %s allowed per rule"), "%prec");
 370   current_rule->ruleprec = precsym;
 371 }
 372
 373 /* Attach dynamic precedence DPREC to the current rule. */
 374
 375 void
 376 grammar_current_rule_dprec_set (int dprec, location loc)
 377 {
 378   if (! glr_parser)
 379     warn_at (loc, _("%s affects only GLR parsers"), "%dprec");
 380   if (dprec <= 0)
 381     complain_at (loc, _("%s must be followed by positive number"), "%dprec");
 382   else if (current_rule->dprec != 0)
 383     complain_at (loc, _("only one %s allowed per rule"), "%dprec");
 384   current_rule->dprec = dprec;
 385 }
 386
 387 /* Attach a merge function NAME with argument type TYPE to current
 388    rule. */
 389
 390 void
 391 grammar_current_rule_merge_set (uniqstr name, location loc)
 392 {
 393   if (! glr_parser)
 394     warn_at (loc, _("%s affects only GLR parsers"), "%merge");
 395   if (current_rule->merger != 0)
 396     complain_at (loc, _("only one %s allowed per rule"), "%merge");
 397   current_rule->merger = get_merge_function (name);
 398   current_rule->merger_declaration_location = loc;
 399 }
 400
 401 /* Attach SYM to the current rule.  If needed, move the previous
 402    action as a mid-rule action.  */
 403
 404 void
 405 grammar_current_rule_symbol_append (symbol *sym, location loc)
 406 {
 407   if (current_rule->action_props.code)
 408     grammar_midrule_action ();
 409   grammar_symbol_append (sym, loc);
 410 }
 411
 412 /* Attach an ACTION to the current rule.  */
 413
 414 void
 415 grammar_current_rule_action_append (const char *action, location loc)
 416 {
 417   if (current_rule->action_props.code)
 418     grammar_midrule_action ();
 419   /* After all symbol declarations have been parsed, packgram invokes
 420      code_props_translate_code.  */
 421   code_props_rule_action_init (&current_rule->action_props, action, loc,
 422                                current_rule);
 423 }
 424
 425 \f
 426 /*---------------------------------------------------------------.
 427 | Convert the rules into the representation using RRHS, RLHS and |
 428 | RITEM.                                                         |
 429 `---------------------------------------------------------------*/
 430
 431 static void
 432 packgram (void)
 433 {
 434   unsigned int itemno = 0;
 435   rule_number ruleno = 0;
 436   symbol_list *p = grammar;
 437
 438   ritem = xnmalloc (nritems + 1, sizeof *ritem);
 439
 440   /* This sentinel is used by build_relations in gram.c.  */
 441   *ritem++ = 0;
 442
 443   rules = xnmalloc (nrules, sizeof *rules);
 444
 445   while (p)
 446     {
 447       int rule_length = 0;
 448       symbol *ruleprec = p->ruleprec;
 449       record_merge_function_type (p->merger, p->content.sym->type_name,
 450                                   p->merger_declaration_location);
 451       rules[ruleno].user_number = ruleno;
 452       rules[ruleno].number = ruleno;
 453       rules[ruleno].lhs = p->content.sym;
 454       rules[ruleno].rhs = ritem + itemno;
 455       rules[ruleno].prec = NULL;
 456       rules[ruleno].dprec = p->dprec;
 457       rules[ruleno].merger = p->merger;
 458       rules[ruleno].precsym = NULL;
 459       rules[ruleno].location = p->location;
 460       rules[ruleno].useful = true;
 461       rules[ruleno].action = p->action_props.code;
 462       rules[ruleno].action_location = p->action_props.location;
 463
 464       /* If the midrule's $$ is set or its $n is used, remove the `$' from the
 465          symbol name so that it's a user-defined symbol so that the default
 466          %destructor and %printer apply.  */
 467       if (p->midrule_parent_rule
 468           && (p->action_props.is_value_used
 469               || symbol_list_n_get (p->midrule_parent_rule,
 470                                     p->midrule_parent_rhs_index)
 471                    ->action_props.is_value_used))
 472         p->content.sym->tag += 1;
 473
 474       /* Don't check the generated rule 0.  It has no action, so some rhs
 475          symbols may appear unused, but the parsing algorithm ensures that
 476          %destructor's are invoked appropriately.  */
 477       if (p != grammar)
 478         grammar_rule_check (p);
 479
 480       for (p = p->next; p && p->content.sym; p = p->next)
 481         {
 482           ++rule_length;
 483
 484           /* Don't allow rule_length == INT_MAX, since that might
 485              cause confusion with strtol if INT_MAX == LONG_MAX.  */
 486           if (rule_length == INT_MAX)
 487               fatal_at (rules[ruleno].location, _("rule is too long"));
 488
 489           /* item_number = symbol_number.
 490              But the former needs to contain more: negative rule numbers. */
 491           ritem[itemno++] =
 492             symbol_number_as_item_number (p->content.sym->number);
 493           /* A rule gets by default the precedence and associativity
 494              of its last token.  */
 495           if (p->content.sym->class == token_sym && default_prec)
 496             rules[ruleno].prec = p->content.sym;
 497         }
 498
 499       /* If this rule has a %prec,
 500          the specified symbol's precedence replaces the default.  */
 501       if (ruleprec)
 502         {
 503           rules[ruleno].precsym = ruleprec;
 504           rules[ruleno].prec = ruleprec;
 505         }
 506       /* An item ends by the rule number (negated).  */
 507       ritem[itemno++] = rule_number_as_item_number (ruleno);
 508       aver (itemno < ITEM_NUMBER_MAX);
 509       ++ruleno;
 510       aver (ruleno < RULE_NUMBER_MAX);
 511
 512       if (p)
 513         p = p->next;
 514     }
 515
 516   aver (itemno == nritems);
 517
 518   if (trace_flag & trace_sets)
 519     ritem_print (stderr);
 520 }
 521 \f
 522 /*------------------------------------------------------------------.
 523 | Read in the grammar specification and record it in the format     |
 524 | described in gram.h.  All actions are copied into ACTION_OBSTACK, |
 525 | in each case forming the body of a C function (YYACTION) which    |
 526 | contains a switch statement to decide which action to execute.    |
 527 `------------------------------------------------------------------*/
 528
 529 void
 530 reader (void)
 531 {
 532   /* Initialize the symbol table.  */
 533   symbols_new ();
 534
 535   /* Construct the accept symbol. */
 536   accept = symbol_get ("$accept", empty_location);
 537   accept->class = nterm_sym;
 538   accept->number = nvars++;
 539
 540   /* Construct the error token */
 541   errtoken = symbol_get ("error", empty_location);
 542   errtoken->class = token_sym;
 543   errtoken->number = ntokens++;
 544
 545   /* Construct a token that represents all undefined literal tokens.
 546      It is always token number 2.  */
 547   undeftoken = symbol_get ("$undefined", empty_location);
 548   undeftoken->class = token_sym;
 549   undeftoken->number = ntokens++;
 550
 551   gram_in = xfopen (grammar_file, "r");
 552
 553   gram__flex_debug = trace_flag & trace_scan;
 554   gram_debug = trace_flag & trace_parse;
 555   gram_scanner_initialize ();
 556   gram_parse ();
 557
 558   /* Set front-end %define variable defaults.  */
 559   muscle_percent_define_default ("lr.keep_unreachable_states", "false");
 560   {
 561     char *lr_type;
 562     /* IELR would be a better default, but LALR is historically the
 563        default.  */
 564     muscle_percent_define_default ("lr.type", "LALR");
 565     lr_type = muscle_percent_define_get ("lr.type");
 566     if (0 != strcmp (lr_type, "canonical LR"))
 567       muscle_percent_define_default ("lr.default-reductions", "all");
 568     else
 569       muscle_percent_define_default ("lr.default-reductions", "accepting");
 570     free (lr_type);
 571   }
 572
 573   /* Check front-end %define variables.  */
 574   {
 575     static char const * const values[] = {
 576       "lr.type", "LALR", "IELR", "canonical LR", NULL,
 577       "lr.default-reductions", "all", "consistent", "accepting", NULL,
 578       NULL
 579     };
 580     muscle_percent_define_check_values (values);
 581   }
 582
 583   if (! complaint_issued)
 584     check_and_convert_grammar ();
 585
 586   xfclose (gram_in);
 587 }
 588
 589
 590 /*-------------------------------------------------------------.
 591 | Check the grammar that has just been read, and convert it to |
 592 | internal form.                                               |
 593 `-------------------------------------------------------------*/
 594
 595 static void
 596 check_and_convert_grammar (void)
 597 {
 598   /* Grammar has been read.  Do some checking.  */
 599   if (nrules == 0)
 600     fatal (_("no rules in the input grammar"));
 601
 602   /* Report any undefined symbols and consider them nonterminals.  */
 603   symbols_check_defined ();
 604
 605   /* If the user did not define her ENDTOKEN, do it now. */
 606   if (!endtoken)
 607     {
 608       endtoken = symbol_get ("$end", empty_location);
 609       endtoken->class = token_sym;
 610       endtoken->number = 0;
 611       /* Value specified by POSIX.  */
 612       endtoken->user_token_number = 0;
 613     }
 614
 615   /* Find the start symbol if no %start.  */
 616   if (!start_flag)
 617     {
 618       symbol_list *node;
 619       for (node = grammar;
 620            node != NULL && symbol_is_dummy (node->content.sym);
 621            node = node->next)
 622         {
 623           for (node = node->next;
 624                node != NULL && node->content.sym != NULL;
 625                node = node->next)
 626             ;
 627         }
 628       aver (node != NULL);
 629       grammar_start_symbol_set (node->content.sym,
 630                                 node->content.sym->location);
 631     }
 632
 633   /* Insert the initial rule, whose line is that of the first rule
 634      (not that of the start symbol):
 635
 636      accept: %start EOF.  */
 637   {
 638     symbol_list *p = symbol_list_sym_new (accept, empty_location);
 639     p->location = grammar->location;
 640     p->next = symbol_list_sym_new (startsymbol, empty_location);
 641     p->next->next = symbol_list_sym_new (endtoken, empty_location);
 642     p->next->next->next = symbol_list_sym_new (NULL, empty_location);
 643     p->next->next->next->next = grammar;
 644     nrules += 1;
 645     nritems += 3;
 646     grammar = p;
 647   }
 648
 649   aver (nsyms <= SYMBOL_NUMBER_MAXIMUM && nsyms == ntokens + nvars);
 650
 651   /* Assign the symbols their symbol numbers.  Write #defines for the
 652      token symbols into FDEFINES if requested.  */
 653   symbols_pack ();
 654
 655   /* Scan rule actions after invoking symbol_check_alias_consistency (in
 656      symbols_pack above) so that token types are set correctly before the rule
 657      action type checking.
 658
 659      Before invoking grammar_rule_check (in packgram below) on any rule, make
 660      sure all actions have already been scanned in order to set `used' flags.
 661      Otherwise, checking that a midrule's $$ should be set will not always work
 662      properly because the check must forward-reference the midrule's parent
 663      rule.  For the same reason, all the `used' flags must be set before
 664      checking whether to remove `$' from any midrule symbol name (also in
 665      packgram).  */
 666   {
 667     symbol_list *sym;
 668     for (sym = grammar; sym; sym = sym->next)
 669       code_props_translate_code (&sym->action_props);
 670   }
 671
 672   /* Convert the grammar into the format described in gram.h.  */
 673   packgram ();
 674
 675   /* The grammar as a symbol_list is no longer needed. */
 676   symbol_list_free (grammar);
 677 }