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 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., 51 Franklin Street, Fifth Floor,
  21    Boston, MA 02110-1301, USA.  */
  22
  23 #include <config.h>
  24 #include "system.h"
  25
  26 #include <quotearg.h>
  27
  28 #include "complain.h"
  29 #include "conflicts.h"
  30 #include "files.h"
  31 #include "getargs.h"
  32 #include "gram.h"
  33 #include "muscle_tab.h"
  34 #include "reader.h"
  35 #include "symlist.h"
  36 #include "symtab.h"
  37 #include "scan-gram.h"
  38 #include "scan-code.h"
  39
  40 static void check_and_convert_grammar (void);
  41
  42 static symbol_list *grammar = NULL;
  43 static bool start_flag = false;
  44 merger_list *merge_functions;
  45
  46 /* Was %union seen?  */
  47 bool union_seen = false;
  48
  49 /* Was a tag seen?  */
  50 bool tag_seen = false;
  51
  52 /* Should rules have a default precedence?  */
  53 bool default_prec = true;
  54 \f
  55 /*-----------------------.
  56 | Set the start symbol.  |
  57 `-----------------------*/
  58
  59 void
  60 grammar_start_symbol_set (symbol *sym, location loc)
  61 {
  62   if (start_flag)
  63     complain_at (loc, _("multiple %s declarations"), "%start");
  64   else
  65     {
  66       start_flag = true;
  67       startsymbol = sym;
  68       startsymbol_location = loc;
  69     }
  70 }
  71
  72 \f
  73
  74 /*------------------------------------------------------------------------.
  75 | Return the merger index for a merging function named NAME.  Records the |
  76 | function, if new, in MERGER_LIST.                                       |
  77 `------------------------------------------------------------------------*/
  78
  79 static int
  80 get_merge_function (uniqstr name)
  81 {
  82   merger_list *syms;
  83   merger_list head;
  84   int n;
  85
  86   if (! glr_parser)
  87     return 0;
  88
  89   head.next = merge_functions;
  90   for (syms = &head, n = 1; syms->next; syms = syms->next, n += 1)
  91     if (UNIQSTR_EQ (name, syms->next->name))
  92       break;
  93   if (syms->next == NULL)
  94     {
  95       syms->next = xmalloc (sizeof syms->next[0]);
  96       syms->next->name = uniqstr_new (name);
  97       /* After all symbol type declarations have been parsed, packgram invokes
  98          record_merge_function_type to set the type.  */
  99       syms->next->type = NULL;
 100       syms->next->next = NULL;
 101       merge_functions = head.next;
 102     }
 103   return n;
 104 }
 105
 106 /*-------------------------------------------------------------------------.
 107 | For the existing merging function with index MERGER, record the result   |
 108 | type as TYPE as required by the lhs of the rule whose %merge declaration |
 109 | is at DECLARATION_LOC.                                                   |
 110 `-------------------------------------------------------------------------*/
 111
 112 static void
 113 record_merge_function_type (int merger, uniqstr type, location declaration_loc)
 114 {
 115   int merger_find;
 116   merger_list *merge_function;
 117
 118   if (merger <= 0)
 119     return;
 120
 121   if (type == NULL)
 122     type = uniqstr_new ("");
 123
 124   merger_find = 1;
 125   for (merge_function = merge_functions;
 126        merge_function != NULL && merger_find != merger;
 127        merge_function = merge_function->next)
 128     merger_find += 1;
 129   aver (merge_function != NULL && merger_find == merger);
 130   if (merge_function->type != NULL && !UNIQSTR_EQ (merge_function->type, type))
 131     {
 132       complain_at (declaration_loc,
 133                    _("result type clash on merge function `%s': <%s> != <%s>"),
 134                    merge_function->name, type, merge_function->type);
 135       complain_at (merge_function->type_declaration_location,
 136                    _("previous declaration"));
 137     }
 138   merge_function->type = uniqstr_new (type);
 139   merge_function->type_declaration_location = declaration_loc;
 140 }
 141
 142 /*--------------------------------------.
 143 | Free all merge-function definitions.  |
 144 `--------------------------------------*/
 145
 146 void
 147 free_merger_functions (void)
 148 {
 149   merger_list *L0 = merge_functions;
 150   while (L0)
 151     {
 152       merger_list *L1 = L0->next;
 153       free (L0);
 154       L0 = L1;
 155     }
 156 }
 157
 158 \f
 159 /*-------------------------------------------------------------------.
 160 | Parse the input grammar into a one symbol_list structure.  Each    |
 161 | rule is represented by a sequence of symbols: the left hand side   |
 162 | followed by the contents of the right hand side, followed by a     |
 163 | null pointer instead of a symbol to terminate the rule.  The next  |
 164 | symbol is the lhs of the following rule.                           |
 165 |                                                                    |
 166 | All actions are copied out, labelled by the rule number they apply |
 167 | to.                                                                |
 168 `-------------------------------------------------------------------*/
 169
 170 /* The (currently) last symbol of GRAMMAR. */
 171 static symbol_list *grammar_end = NULL;
 172
 173 /* Append SYM to the grammar.  */
 174 static void
 175 grammar_symbol_append (symbol *sym, location loc)
 176 {
 177   symbol_list *p = symbol_list_sym_new (sym, loc);
 178
 179   if (grammar_end)
 180     grammar_end->next = p;
 181   else
 182     grammar = p;
 183
 184   grammar_end = p;
 185
 186   /* A null SYM stands for an end of rule; it is not an actual
 187      part of it.  */
 188   if (sym)
 189     ++nritems;
 190 }
 191
 192 /* The rule currently being defined, and the previous rule.
 193    CURRENT_RULE points to the first LHS of the current rule, while
 194    PREVIOUS_RULE_END points to the *end* of the previous rule (NULL).  */
 195 static symbol_list *current_rule = NULL;
 196 static symbol_list *previous_rule_end = NULL;
 197
 198
 199 /*----------------------------------------------.
 200 | Create a new rule for LHS in to the GRAMMAR.  |
 201 `----------------------------------------------*/
 202
 203 void
 204 grammar_current_rule_begin (symbol *lhs, location loc)
 205 {
 206   /* Start a new rule and record its lhs.  */
 207   ++nrules;
 208   previous_rule_end = grammar_end;
 209   grammar_symbol_append (lhs, loc);
 210   current_rule = grammar_end;
 211
 212   /* Mark the rule's lhs as a nonterminal if not already so.  */
 213   if (lhs->class == unknown_sym)
 214     {
 215       lhs->class = nterm_sym;
 216       lhs->number = nvars;
 217       ++nvars;
 218     }
 219   else if (lhs->class == token_sym)
 220     complain_at (loc, _("rule given for %s, which is a token"), lhs->tag);
 221 }
 222
 223
 224 /*----------------------------------------------------------------------.
 225 | A symbol should be used if either:                                    |
 226 |   1. It has a destructor.                                             |
 227 |   2. --warnings=midrule-values and the symbol is a mid-rule symbol    |
 228 |      (i.e., the generated LHS replacing a mid-rule action) that was   |
 229 |      assigned to or used, as in "exp: { $$ = 1; } { $$ = $1; }".      |
 230 `----------------------------------------------------------------------*/
 231
 232 static bool
 233 symbol_should_be_used (symbol_list const *s)
 234 {
 235   if (symbol_destructor_get (s->content.sym)->code)
 236     return true;
 237   if (warnings_flag & warnings_midrule_values)
 238     return ((s->midrule && s->midrule->action_props.is_value_used)
 239             || (s->midrule_parent_rule
 240                 && symbol_list_n_get (s->midrule_parent_rule,
 241                                       s->midrule_parent_rhs_index)
 242                      ->action_props.is_value_used));
 243   return false;
 244 }
 245
 246 /*----------------------------------------------------------------.
 247 | Check that the rule R is properly defined.  For instance, there |
 248 | should be no type clash on the default action.                  |
 249 `----------------------------------------------------------------*/
 250
 251 static void
 252 grammar_rule_check (const symbol_list *r)
 253 {
 254   /* Type check.
 255
 256      If there is an action, then there is nothing we can do: the user
 257      is allowed to shoot herself in the foot.
 258
 259      Don't worry about the default action if $$ is untyped, since $$'s
 260      value can't be used.  */
 261   if (!r->action_props.code && r->content.sym->type_name)
 262     {
 263       symbol *first_rhs = r->next->content.sym;
 264       /* If $$ is being set in default way, report if any type mismatch.  */
 265       if (first_rhs)
 266         {
 267           char const *lhs_type = r->content.sym->type_name;
 268           const char *rhs_type =
 269             first_rhs->type_name ? first_rhs->type_name : "";
 270           if (!UNIQSTR_EQ (lhs_type, rhs_type))
 271             warn_at (r->location,
 272                      _("type clash on default action: <%s> != <%s>"),
 273                      lhs_type, rhs_type);
 274         }
 275       /* Warn if there is no default for $$ but we need one.  */
 276       else
 277         warn_at (r->location,
 278                  _("empty rule for typed nonterminal, and no action"));
 279     }
 280
 281   /* Check that symbol values that should be used are in fact used.  */
 282   {
 283     symbol_list const *l = r;
 284     int n = 0;
 285     for (; l && l->content.sym; l = l->next, ++n)
 286       if (! (l->action_props.is_value_used
 287              || !symbol_should_be_used (l)
 288              /* The default action, $$ = $1, `uses' both.  */
 289              || (!r->action_props.code && (n == 0 || n == 1))))
 290         {
 291           if (n)
 292             warn_at (r->location, _("unused value: $%d"), n);
 293           else
 294             warn_at (r->location, _("unset value: $$"));
 295         }
 296   }
 297 }
 298
 299
 300 /*-------------------------------------.
 301 | End the currently being grown rule.  |
 302 `-------------------------------------*/
 303
 304 void
 305 grammar_current_rule_end (location loc)
 306 {
 307   /* Put an empty link in the list to mark the end of this rule  */
 308   grammar_symbol_append (NULL, grammar_end->location);
 309   current_rule->location = loc;
 310 }
 311
 312
 313 /*-------------------------------------------------------------------.
 314 | The previous action turns out the be a mid-rule action.  Attach it |
 315 | to the current rule, i.e., create a dummy symbol, attach it this   |
 316 | mid-rule action, and append this dummy nonterminal to the current  |
 317 | rule.                                                              |
 318 `-------------------------------------------------------------------*/
 319
 320 void
 321 grammar_midrule_action (void)
 322 {
 323   /* Since the action was written out with this rule's number, we must
 324      give the new rule this number by inserting the new rule before
 325      it.  */
 326
 327   /* Make a DUMMY nonterminal, whose location is that of the midrule
 328      action.  Create the MIDRULE.  */
 329   location dummy_location = current_rule->action_props.location;
 330   symbol *dummy = dummy_symbol_get (dummy_location);
 331   symbol_list *midrule = symbol_list_sym_new (dummy, dummy_location);
 332
 333   /* Make a new rule, whose body is empty, before the current one, so
 334      that the action just read can belong to it.  */
 335   ++nrules;
 336   ++nritems;
 337   /* Attach its location and actions to that of the DUMMY.  */
 338   midrule->location = dummy_location;
 339   code_props_rule_action_init (&midrule->action_props,
 340                                current_rule->action_props.code,
 341                                current_rule->action_props.location,
 342                                midrule);
 343   code_props_none_init (&current_rule->action_props);
 344
 345   if (previous_rule_end)
 346     previous_rule_end->next = midrule;
 347   else
 348     grammar = midrule;
 349
 350   /* End the dummy's rule.  */
 351   midrule->next = symbol_list_sym_new (NULL, dummy_location);
 352   midrule->next->next = current_rule;
 353
 354   previous_rule_end = midrule->next;
 355
 356   /* Insert the dummy nonterminal replacing the midrule action into
 357      the current rule.  Bind it to its dedicated rule.  */
 358   grammar_current_rule_symbol_append (dummy, dummy_location);
 359   grammar_end->midrule = midrule;
 360   midrule->midrule_parent_rule = current_rule;
 361   midrule->midrule_parent_rhs_index = symbol_list_length (current_rule->next);
 362 }
 363
 364 /* Set the precedence symbol of the current rule to PRECSYM. */
 365
 366 void
 367 grammar_current_rule_prec_set (symbol *precsym, location loc)
 368 {
 369   symbol_class_set (precsym, token_sym, loc, false);
 370   if (current_rule->ruleprec)
 371     complain_at (loc, _("only one %s allowed per rule"), "%prec");
 372   current_rule->ruleprec = precsym;
 373 }
 374
 375 /* Attach dynamic precedence DPREC to the current rule. */
 376
 377 void
 378 grammar_current_rule_dprec_set (int dprec, location loc)
 379 {
 380   if (! glr_parser)
 381     warn_at (loc, _("%s affects only GLR parsers"), "%dprec");
 382   if (dprec <= 0)
 383     complain_at (loc, _("%s must be followed by positive number"), "%dprec");
 384   else if (current_rule->dprec != 0)
 385     complain_at (loc, _("only one %s allowed per rule"), "%dprec");
 386   current_rule->dprec = dprec;
 387 }
 388
 389 /* Attach a merge function NAME with argument type TYPE to current
 390    rule. */
 391
 392 void
 393 grammar_current_rule_merge_set (uniqstr name, location loc)
 394 {
 395   if (! glr_parser)
 396     warn_at (loc, _("%s affects only GLR parsers"), "%merge");
 397   if (current_rule->merger != 0)
 398     complain_at (loc, _("only one %s allowed per rule"), "%merge");
 399   current_rule->merger = get_merge_function (name);
 400   current_rule->merger_declaration_location = loc;
 401 }
 402
 403 /* Attach SYM to the current rule.  If needed, move the previous
 404    action as a mid-rule action.  */
 405
 406 void
 407 grammar_current_rule_symbol_append (symbol *sym, location loc)
 408 {
 409   if (current_rule->action_props.code)
 410     grammar_midrule_action ();
 411   grammar_symbol_append (sym, loc);
 412 }
 413
 414 /* Attach an ACTION to the current rule.  */
 415
 416 void
 417 grammar_current_rule_action_append (const char *action, location loc)
 418 {
 419   if (current_rule->action_props.code)
 420     grammar_midrule_action ();
 421   /* After all symbol declarations have been parsed, packgram invokes
 422      code_props_translate_code.  */
 423   code_props_rule_action_init (&current_rule->action_props, action, loc,
 424                                current_rule);
 425 }
 426
 427 \f
 428 /*---------------------------------------------------------------.
 429 | Convert the rules into the representation using RRHS, RLHS and |
 430 | RITEM.                                                         |
 431 `---------------------------------------------------------------*/
 432
 433 static void
 434 packgram (void)
 435 {
 436   unsigned int itemno = 0;
 437   rule_number ruleno = 0;
 438   symbol_list *p = grammar;
 439
 440   ritem = xnmalloc (nritems + 1, sizeof *ritem);
 441
 442   /* This sentinel is used by build_relations in gram.c.  */
 443   *ritem++ = 0;
 444
 445   rules = xnmalloc (nrules, sizeof *rules);
 446
 447   while (p)
 448     {
 449       int rule_length = 0;
 450       symbol *ruleprec = p->ruleprec;
 451       record_merge_function_type (p->merger, p->content.sym->type_name,
 452                                   p->merger_declaration_location);
 453       rules[ruleno].user_number = ruleno;
 454       rules[ruleno].number = ruleno;
 455       rules[ruleno].lhs = p->content.sym;
 456       rules[ruleno].rhs = ritem + itemno;
 457       rules[ruleno].prec = NULL;
 458       rules[ruleno].dprec = p->dprec;
 459       rules[ruleno].merger = p->merger;
 460       rules[ruleno].precsym = NULL;
 461       rules[ruleno].location = p->location;
 462       rules[ruleno].useful = true;
 463       rules[ruleno].action = p->action_props.code;
 464       rules[ruleno].action_location = p->action_props.location;
 465
 466       /* If the midrule's $$ is set or its $n is used, remove the `$' from the
 467          symbol name so that it's a user-defined symbol so that the default
 468          %destructor and %printer apply.  */
 469       if (p->midrule_parent_rule
 470           && (p->action_props.is_value_used
 471               || symbol_list_n_get (p->midrule_parent_rule,
 472                                     p->midrule_parent_rhs_index)
 473                    ->action_props.is_value_used))
 474         p->content.sym->tag += 1;
 475
 476       /* Don't check the generated rule 0.  It has no action, so some rhs
 477          symbols may appear unused, but the parsing algorithm ensures that
 478          %destructor's are invoked appropriately.  */
 479       if (p != grammar)
 480         grammar_rule_check (p);
 481
 482       for (p = p->next; p && p->content.sym; p = p->next)
 483         {
 484           ++rule_length;
 485
 486           /* Don't allow rule_length == INT_MAX, since that might
 487              cause confusion with strtol if INT_MAX == LONG_MAX.  */
 488           if (rule_length == INT_MAX)
 489               fatal_at (rules[ruleno].location, _("rule is too long"));
 490
 491           /* item_number = symbol_number.
 492              But the former needs to contain more: negative rule numbers. */
 493           ritem[itemno++] =
 494             symbol_number_as_item_number (p->content.sym->number);
 495           /* A rule gets by default the precedence and associativity
 496              of its last token.  */
 497           if (p->content.sym->class == token_sym && default_prec)
 498             rules[ruleno].prec = p->content.sym;
 499         }
 500
 501       /* If this rule has a %prec,
 502          the specified symbol's precedence replaces the default.  */
 503       if (ruleprec)
 504         {
 505           rules[ruleno].precsym = ruleprec;
 506           rules[ruleno].prec = ruleprec;
 507         }
 508       /* An item ends by the rule number (negated).  */
 509       ritem[itemno++] = rule_number_as_item_number (ruleno);
 510       aver (itemno < ITEM_NUMBER_MAX);
 511       ++ruleno;
 512       aver (ruleno < RULE_NUMBER_MAX);
 513
 514       if (p)
 515         p = p->next;
 516     }
 517
 518   aver (itemno == nritems);
 519
 520   if (trace_flag & trace_sets)
 521     ritem_print (stderr);
 522 }
 523 \f
 524 /*------------------------------------------------------------------.
 525 | Read in the grammar specification and record it in the format     |
 526 | described in gram.h.  All actions are copied into ACTION_OBSTACK, |
 527 | in each case forming the body of a C function (YYACTION) which    |
 528 | contains a switch statement to decide which action to execute.    |
 529 `------------------------------------------------------------------*/
 530
 531 void
 532 reader (void)
 533 {
 534   /* Initialize the symbol table.  */
 535   symbols_new ();
 536
 537   /* Construct the accept symbol. */
 538   accept = symbol_get ("$accept", empty_location);
 539   accept->class = nterm_sym;
 540   accept->number = nvars++;
 541
 542   /* Construct the error token */
 543   errtoken = symbol_get ("error", empty_location);
 544   errtoken->class = token_sym;
 545   errtoken->number = ntokens++;
 546
 547   /* Construct a token that represents all undefined literal tokens.
 548      It is always token number 2.  */
 549   undeftoken = symbol_get ("$undefined", empty_location);
 550   undeftoken->class = token_sym;
 551   undeftoken->number = ntokens++;
 552
 553   gram_in = xfopen (grammar_file, "r");
 554
 555   gram__flex_debug = trace_flag & trace_scan;
 556   gram_debug = trace_flag & trace_parse;
 557   gram_scanner_initialize ();
 558   gram_parse ();
 559
 560   if (! complaint_issued)
 561     check_and_convert_grammar ();
 562
 563   xfclose (gram_in);
 564 }
 565
 566
 567 /*-------------------------------------------------------------.
 568 | Check the grammar that has just been read, and convert it to |
 569 | internal form.                                               |
 570 `-------------------------------------------------------------*/
 571
 572 static void
 573 check_and_convert_grammar (void)
 574 {
 575   /* Grammar has been read.  Do some checking.  */
 576   if (nrules == 0)
 577     fatal (_("no rules in the input grammar"));
 578
 579   /* Report any undefined symbols and consider them nonterminals.  */
 580   symbols_check_defined ();
 581
 582   /* If the user did not define her ENDTOKEN, do it now. */
 583   if (!endtoken)
 584     {
 585       endtoken = symbol_get ("$end", empty_location);
 586       endtoken->class = token_sym;
 587       endtoken->number = 0;
 588       /* Value specified by POSIX.  */
 589       endtoken->user_token_number = 0;
 590     }
 591
 592   /* Find the start symbol if no %start.  */
 593   if (!start_flag)
 594     {
 595       symbol_list *node;
 596       for (node = grammar;
 597            node != NULL && symbol_is_dummy (node->content.sym);
 598            node = node->next)
 599         {
 600           for (node = node->next;
 601                node != NULL && node->content.sym != NULL;
 602                node = node->next)
 603             ;
 604         }
 605       aver (node != NULL);
 606       grammar_start_symbol_set (node->content.sym,
 607                                 node->content.sym->location);
 608     }
 609
 610   /* Insert the initial rule, whose line is that of the first rule
 611      (not that of the start symbol):
 612
 613      accept: %start EOF.  */
 614   {
 615     symbol_list *p = symbol_list_sym_new (accept, empty_location);
 616     p->location = grammar->location;
 617     p->next = symbol_list_sym_new (startsymbol, empty_location);
 618     p->next->next = symbol_list_sym_new (endtoken, empty_location);
 619     p->next->next->next = symbol_list_sym_new (NULL, empty_location);
 620     p->next->next->next->next = grammar;
 621     nrules += 1;
 622     nritems += 3;
 623     grammar = p;
 624   }
 625
 626   aver (nsyms <= SYMBOL_NUMBER_MAXIMUM && nsyms == ntokens + nvars);
 627
 628   /* Assign the symbols their symbol numbers.  Write #defines for the
 629      token symbols into FDEFINES if requested.  */
 630   symbols_pack ();
 631
 632   /* Scan rule actions after invoking symbol_check_alias_consistency (in
 633      symbols_pack above) so that token types are set correctly before the rule
 634      action type checking.
 635
 636      Before invoking grammar_rule_check (in packgram below) on any rule, make
 637      sure all actions have already been scanned in order to set `used' flags.
 638      Otherwise, checking that a midrule's $$ should be set will not always work
 639      properly because the check must forward-reference the midrule's parent
 640      rule.  For the same reason, all the `used' flags must be set before
 641      checking whether to remove `$' from any midrule symbol name (also in
 642      packgram).  */
 643   for (symbol_list *sym = grammar; sym; sym = sym->next)
 644     code_props_translate_code (&sym->action_props);
 645
 646   /* Convert the grammar into the format described in gram.h.  */
 647   packgram ();
 648
 649   /* The grammar as a symbol_list is no longer needed. */
 650   symbol_list_free (grammar);
 651 }