src/reader.c

   1 /* Input parser for Bison
   2
   3    Copyright (C) 1984, 1986, 1989, 1992, 1998, 2000, 2001, 2002, 2003,
   4    2005 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 "output.h"
  35 #include "reader.h"
  36 #include "symlist.h"
  37 #include "symtab.h"
  38
  39 static symbol_list *grammar = NULL;
  40 static bool start_flag = false;
  41 merger_list *merge_functions;
  42
  43 /* Has %union been seen?  */
  44 bool typed = false;
  45
  46 /* Should rules have a default precedence?  */
  47 bool default_prec = true;
  48 \f
  49 /*-----------------------.
  50 | Set the start symbol.  |
  51 `-----------------------*/
  52
  53 void
  54 grammar_start_symbol_set (symbol *sym, location loc)
  55 {
  56   if (start_flag)
  57     complain_at (loc, _("multiple %s declarations"), "%start");
  58   else
  59     {
  60       start_flag = true;
  61       startsymbol = sym;
  62       startsymbol_location = loc;
  63     }
  64 }
  65
  66
  67 /*----------------------------------------------------------------.
  68 | There are two prologues: one before %union, one after.  Augment |
  69 | the current one.                                                |
  70 `----------------------------------------------------------------*/
  71
  72 void
  73 prologue_augment (const char *prologue, location loc)
  74 {
  75   struct obstack *oout =
  76     !typed ? &pre_prologue_obstack : &post_prologue_obstack;
  77
  78   obstack_fgrow1 (oout, "]b4_syncline(%d, [[", loc.start.line);
  79   MUSCLE_OBSTACK_SGROW (oout,
  80                         quotearg_style (c_quoting_style, loc.start.file));
  81   obstack_sgrow (oout, "]])[\n");
  82   obstack_sgrow (oout, prologue);
  83 }
  84
  85 \f
  86
  87 /*-------------------------------------------------------------------.
  88 | Return the merger index for a merging function named NAME, whose   |
  89 | arguments have type TYPE.  Records the function, if new, in        |
  90 | MERGER_LIST.                                                       |
  91 `-------------------------------------------------------------------*/
  92
  93 static int
  94 get_merge_function (uniqstr name, uniqstr type, location loc)
  95 {
  96   merger_list *syms;
  97   merger_list head;
  98   int n;
  99
 100   if (! glr_parser)
 101     return 0;
 102
 103   if (type == NULL)
 104     type = uniqstr_new ("");
 105
 106   head.next = merge_functions;
 107   for (syms = &head, n = 1; syms->next != NULL; syms = syms->next, n += 1)
 108     if (UNIQSTR_EQ (name, syms->next->name))
 109       break;
 110   if (syms->next == NULL)
 111     {
 112       syms->next = xmalloc (sizeof syms->next[0]);
 113       syms->next->name = uniqstr_new (name);
 114       syms->next->type = uniqstr_new (type);
 115       syms->next->next = NULL;
 116       merge_functions = head.next;
 117     }
 118   else if (!UNIQSTR_EQ (type, syms->next->type))
 119     warn_at (loc, _("result type clash on merge function %s: <%s> != <%s>"),
 120              name, type, syms->next->type);
 121   return n;
 122 }
 123
 124 /*--------------------------------------.
 125 | Free all merge-function definitions.  |
 126 `--------------------------------------*/
 127
 128 void
 129 free_merger_functions (void)
 130 {
 131   merger_list *L0;
 132   if (! glr_parser)
 133     return;
 134   L0 = merge_functions;
 135   while (L0 != NULL)
 136     {
 137       merger_list *L1 = L0->next;
 138       free (L0);
 139       L0 = L1;
 140     }
 141 }
 142
 143 \f
 144 /*-------------------------------------------------------------------.
 145 | Parse the input grammar into a one symbol_list structure.  Each    |
 146 | rule is represented by a sequence of symbols: the left hand side   |
 147 | followed by the contents of the right hand side, followed by a     |
 148 | null pointer instead of a symbol to terminate the rule.  The next  |
 149 | symbol is the lhs of the following rule.                           |
 150 |                                                                    |
 151 | All actions are copied out, labelled by the rule number they apply |
 152 | to.                                                                |
 153 |                                                                    |
 154 | Bison used to allow some %directives in the rules sections, but    |
 155 | this is no longer consider appropriate: (i) the documented grammar |
 156 | doesn't claim it, (ii), it would promote bad style, (iii), error   |
 157 | recovery for %directives consists in skipping the junk until a `%' |
 158 | is seen and helrp synchronizing.  This scheme is definitely wrong  |
 159 | in the rules section.                                              |
 160 `-------------------------------------------------------------------*/
 161
 162 /* The (currently) last symbol of GRAMMAR. */
 163 static symbol_list *grammar_end = NULL;
 164
 165 /* Append SYM to the grammar.  */
 166 void
 167 grammar_symbol_append (symbol *sym, location loc)
 168 {
 169   symbol_list *p = symbol_list_new (sym, loc);
 170
 171   if (grammar_end)
 172     grammar_end->next = p;
 173   else
 174     grammar = p;
 175
 176   grammar_end = p;
 177 }
 178
 179 /* The rule currently being defined, and the previous rule.
 180    CURRENT_RULE points to the first LHS of the current rule, while
 181    PREVIOUS_RULE_END points to the *end* of the previous rule (NULL).  */
 182 symbol_list *current_rule = NULL;
 183 static symbol_list *previous_rule_end = NULL;
 184
 185
 186 /*----------------------------------------------.
 187 | Create a new rule for LHS in to the GRAMMAR.  |
 188 `----------------------------------------------*/
 189
 190 void
 191 grammar_rule_begin (symbol *lhs, location loc)
 192 {
 193   if (!start_flag)
 194     {
 195       startsymbol = lhs;
 196       startsymbol_location = loc;
 197       start_flag = true;
 198     }
 199
 200   /* Start a new rule and record its lhs.  */
 201   ++nrules;
 202   ++nritems;
 203
 204   previous_rule_end = grammar_end;
 205   grammar_symbol_append (lhs, loc);
 206   current_rule = grammar_end;
 207
 208   /* Mark the rule's lhs as a nonterminal if not already so.  */
 209
 210   if (lhs->class == unknown_sym)
 211     {
 212       lhs->class = nterm_sym;
 213       lhs->number = nvars;
 214       ++nvars;
 215     }
 216   else if (lhs->class == token_sym)
 217     complain_at (loc, _("rule given for %s, which is a token"), lhs->tag);
 218 }
 219
 220 /* Check that the last rule (CURRENT_RULE) is properly defined.  For
 221    instance, there should be no type clash on the default action.  */
 222
 223 static void
 224 grammar_current_rule_check (void)
 225 {
 226   symbol *lhs = current_rule->sym;
 227   char const *lhs_type = lhs->type_name;
 228   symbol *first_rhs = current_rule->next->sym;
 229
 230   /* If there is an action, then there is nothing we can do: the user
 231      is allowed to shoot herself in the foot.  */
 232   if (current_rule->action)
 233     return;
 234
 235   /* Don't worry about the default action if $$ is untyped, since $$'s
 236      value can't be used.  */
 237   if (! lhs_type)
 238     return;
 239
 240   /* If $$ is being set in default way, report if any type mismatch.  */
 241   if (first_rhs)
 242     {
 243       const char *rhs_type = first_rhs->type_name ? first_rhs->type_name : "";
 244       if (!UNIQSTR_EQ (lhs_type, rhs_type))
 245         warn_at (current_rule->location,
 246                  _("type clash on default action: <%s> != <%s>"),
 247                  lhs_type, rhs_type);
 248     }
 249   /* Warn if there is no default for $$ but we need one.  */
 250   else
 251     warn_at (current_rule->location,
 252              _("empty rule for typed nonterminal, and no action"));
 253 }
 254
 255
 256 /*-------------------------------------.
 257 | End the currently being grown rule.  |
 258 `-------------------------------------*/
 259
 260 void
 261 grammar_rule_end (location loc)
 262 {
 263   /* Put an empty link in the list to mark the end of this rule  */
 264   grammar_symbol_append (NULL, grammar_end->location);
 265   current_rule->location = loc;
 266   grammar_current_rule_check ();
 267 }
 268
 269
 270 /*-------------------------------------------------------------------.
 271 | The previous action turns out the be a mid-rule action.  Attach it |
 272 | to the current rule, i.e., create a dummy symbol, attach it this   |
 273 | mid-rule action, and append this dummy nonterminal to the current  |
 274 | rule.                                                              |
 275 `-------------------------------------------------------------------*/
 276
 277 void
 278 grammar_midrule_action (void)
 279 {
 280   /* Since the action was written out with this rule's number, we must
 281      give the new rule this number by inserting the new rule before
 282      it.  */
 283
 284   /* Make a DUMMY nonterminal, whose location is that of the midrule
 285      action.  Create the MIDRULE.  */
 286   location dummy_location = current_rule->action_location;
 287   symbol *dummy = dummy_symbol_get (dummy_location);
 288   symbol_list *midrule = symbol_list_new (dummy, dummy_location);
 289
 290   /* Make a new rule, whose body is empty, before the current one, so
 291      that the action just read can belong to it.  */
 292   ++nrules;
 293   ++nritems;
 294   /* Attach its location and actions to that of the DUMMY.  */
 295   midrule->location = dummy_location;
 296   midrule->action = current_rule->action;
 297   midrule->action_location = dummy_location;
 298   current_rule->action = NULL;
 299
 300   if (previous_rule_end)
 301     previous_rule_end->next = midrule;
 302   else
 303     grammar = midrule;
 304
 305   /* End the dummy's rule.  */
 306   previous_rule_end = symbol_list_new (NULL, dummy_location);
 307   previous_rule_end->next = current_rule;
 308
 309   midrule->next = previous_rule_end;
 310
 311   /* Insert the dummy nonterminal replacing the midrule action into
 312      the current rule.  */
 313   grammar_current_rule_symbol_append (dummy, dummy_location);
 314 }
 315
 316 /* Set the precedence symbol of the current rule to PRECSYM. */
 317
 318 void
 319 grammar_current_rule_prec_set (symbol *precsym, location loc)
 320 {
 321   if (current_rule->ruleprec)
 322     complain_at (loc, _("only one %s allowed per rule"), "%prec");
 323   current_rule->ruleprec = precsym;
 324 }
 325
 326 /* Attach dynamic precedence DPREC to the current rule. */
 327
 328 void
 329 grammar_current_rule_dprec_set (int dprec, location loc)
 330 {
 331   if (! glr_parser)
 332     warn_at (loc, _("%s affects only GLR parsers"), "%dprec");
 333   if (dprec <= 0)
 334     complain_at (loc, _("%s must be followed by positive number"), "%dprec");
 335   else if (current_rule->dprec != 0)
 336     complain_at (loc, _("only one %s allowed per rule"), "%dprec");
 337   current_rule->dprec = dprec;
 338 }
 339
 340 /* Attach a merge function NAME with argument type TYPE to current
 341    rule. */
 342
 343 void
 344 grammar_current_rule_merge_set (uniqstr name, location loc)
 345 {
 346   if (! glr_parser)
 347     warn_at (loc, _("%s affects only GLR parsers"), "%merge");
 348   if (current_rule->merger != 0)
 349     complain_at (loc, _("only one %s allowed per rule"), "%merge");
 350   current_rule->merger =
 351     get_merge_function (name, current_rule->sym->type_name, loc);
 352 }
 353
 354 /* Attach SYM to the current rule.  If needed, move the previous
 355    action as a mid-rule action.  */
 356
 357 void
 358 grammar_current_rule_symbol_append (symbol *sym, location loc)
 359 {
 360   if (current_rule->action)
 361     grammar_midrule_action ();
 362   ++nritems;
 363   grammar_symbol_append (sym, loc);
 364 }
 365
 366 /* Attach an ACTION to the current rule.  If needed, move the previous
 367    action as a mid-rule action.  */
 368
 369 void
 370 grammar_current_rule_action_append (const char *action, location loc)
 371 {
 372   if (current_rule->action)
 373     grammar_midrule_action ();
 374   current_rule->action = action;
 375   current_rule->action_location = loc;
 376 }
 377
 378 \f
 379 /*---------------------------------------------------------------.
 380 | Convert the rules into the representation using RRHS, RLHS and |
 381 | RITEM.                                                         |
 382 `---------------------------------------------------------------*/
 383
 384 static void
 385 packgram (void)
 386 {
 387   unsigned int itemno = 0;
 388   rule_number ruleno = 0;
 389   symbol_list *p = grammar;
 390
 391   ritem = xnmalloc (nritems, sizeof *ritem);
 392   rules = xnmalloc (nrules, sizeof *rules);
 393
 394   while (p)
 395     {
 396       symbol *ruleprec = p->ruleprec;
 397       rules[ruleno].user_number = ruleno;
 398       rules[ruleno].number = ruleno;
 399       rules[ruleno].lhs = p->sym;
 400       rules[ruleno].rhs = ritem + itemno;
 401       rules[ruleno].prec = NULL;
 402       rules[ruleno].dprec = p->dprec;
 403       rules[ruleno].merger = p->merger;
 404       rules[ruleno].precsym = NULL;
 405       rules[ruleno].location = p->location;
 406       rules[ruleno].useful = true;
 407       rules[ruleno].action = p->action;
 408       rules[ruleno].action_location = p->action_location;
 409
 410       p = p->next;
 411       while (p && p->sym)
 412         {
 413           /* item_number = symbol_number.
 414              But the former needs to contain more: negative rule numbers. */
 415           ritem[itemno++] = symbol_number_as_item_number (p->sym->number);
 416           /* A rule gets by default the precedence and associativity
 417              of the last token in it.  */
 418           if (p->sym->class == token_sym && default_prec)
 419             rules[ruleno].prec = p->sym;
 420           if (p)
 421             p = p->next;
 422         }
 423
 424       /* If this rule has a %prec,
 425          the specified symbol's precedence replaces the default.  */
 426       if (ruleprec)
 427         {
 428           rules[ruleno].precsym = ruleprec;
 429           rules[ruleno].prec = ruleprec;
 430         }
 431       ritem[itemno++] = rule_number_as_item_number (ruleno);
 432       ++ruleno;
 433
 434       if (p)
 435         p = p->next;
 436     }
 437
 438   if (itemno != nritems)
 439     abort ();
 440
 441   if (trace_flag & trace_sets)
 442     ritem_print (stderr);
 443 }
 444 \f
 445 /*------------------------------------------------------------------.
 446 | Read in the grammar specification and record it in the format     |
 447 | described in gram.h.  All actions are copied into ACTION_OBSTACK, |
 448 | in each case forming the body of a C function (YYACTION) which    |
 449 | contains a switch statement to decide which action to execute.    |
 450 `------------------------------------------------------------------*/
 451
 452 void
 453 reader (void)
 454 {
 455   /* Initialize the symbol table.  */
 456   symbols_new ();
 457
 458   /* Construct the accept symbol. */
 459   accept = symbol_get ("$accept", empty_location);
 460   accept->class = nterm_sym;
 461   accept->number = nvars++;
 462
 463   /* Construct the error token */
 464   errtoken = symbol_get ("error", empty_location);
 465   errtoken->class = token_sym;
 466   errtoken->number = ntokens++;
 467
 468   /* Construct a token that represents all undefined literal tokens.
 469      It is always token number 2.  */
 470   undeftoken = symbol_get ("$undefined", empty_location);
 471   undeftoken->class = token_sym;
 472   undeftoken->number = ntokens++;
 473
 474   /* Initialize the obstacks. */
 475   obstack_init (&pre_prologue_obstack);
 476   obstack_init (&post_prologue_obstack);
 477
 478   gram_in = xfopen (grammar_file, "r");
 479
 480   gram__flex_debug = trace_flag & trace_scan;
 481   gram_debug = trace_flag & trace_parse;
 482   scanner_initialize ();
 483   gram_parse ();
 484
 485   /* If something went wrong during the parsing, don't try to
 486      continue.  */
 487   if (complaint_issued)
 488     return;
 489
 490   /* Grammar has been read.  Do some checking */
 491   if (nrules == 0)
 492     fatal (_("no rules in the input grammar"));
 493
 494   /* Report any undefined symbols and consider them nonterminals.  */
 495   symbols_check_defined ();
 496
 497   /* If the user did not define her ENDTOKEN, do it now. */
 498   if (!endtoken)
 499     {
 500       endtoken = symbol_get ("$end", empty_location);
 501       endtoken->class = token_sym;
 502       endtoken->number = 0;
 503       /* Value specified by POSIX.  */
 504       endtoken->user_token_number = 0;
 505     }
 506
 507   /* Insert the initial rule, which line is that of the first rule
 508      (not that of the start symbol):
 509
 510      accept: %start EOF.  */
 511   {
 512     symbol_list *p = symbol_list_new (accept, empty_location);
 513     p->location = grammar->location;
 514     p->next = symbol_list_new (startsymbol, empty_location);
 515     p->next->next = symbol_list_new (endtoken, empty_location);
 516     p->next->next->next = symbol_list_new (NULL, empty_location);
 517     p->next->next->next->next = grammar;
 518     nrules += 1;
 519     nritems += 3;
 520     grammar = p;
 521   }
 522
 523   if (! (nsyms <= SYMBOL_NUMBER_MAXIMUM && nsyms == ntokens + nvars))
 524     abort ();
 525
 526   xfclose (gram_in);
 527
 528   /* Assign the symbols their symbol numbers.  Write #defines for the
 529      token symbols into FDEFINES if requested.  */
 530   symbols_pack ();
 531
 532   /* Convert the grammar into the format described in gram.h.  */
 533   packgram ();
 534
 535   /* The grammar as a symbol_list is no longer needed. */
 536   LIST_FREE (symbol_list, grammar);
 537 }