gcc/gengtype-parse.c

   1 /* Process source files and output type information.
   2    Copyright (C) 2006, 2007, 2010, 2012 Free Software Foundation, Inc.
   3
   4    This file is part of GCC.
   5
   6    GCC is free software; you can redistribute it and/or modify it under
   7    the terms of the GNU General Public License as published by the Free
   8    Software Foundation; either version 3, or (at your option) any later
   9    version.
  10
  11    GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  12    WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13    FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14    for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with GCC; see the file COPYING3.  If not see
  18    <http://www.gnu.org/licenses/>.  */
  19
  20 #ifdef GENERATOR_FILE
  21 #include "bconfig.h"
  22 #else
  23 #include "config.h"
  24 #endif
  25 #include "system.h"
  26 #include "gengtype.h"
  27
  28 /* This is a simple recursive-descent parser which understands a subset of
  29    the C type grammar.
  30
  31    Rule functions are suffixed _seq if they scan a sequence of items;
  32    _opt if they may consume zero tokens; _seqopt if both are true.  The
  33    "consume_" prefix indicates that a sequence of tokens is parsed for
  34    syntactic correctness and then thrown away.  */
  35
  36 /* Simple one-token lookahead mechanism.  */
  37
  38 struct token
  39 {
  40   const char *value;
  41   int code;
  42   bool valid;
  43 };
  44 static struct token T;
  45
  46 /* Retrieve the code of the current token; if there is no current token,
  47    get the next one from the lexer.  */
  48 static inline int
  49 token (void)
  50 {
  51   if (!T.valid)
  52     {
  53       T.code = yylex (&T.value);
  54       T.valid = true;
  55     }
  56   return T.code;
  57 }
  58
  59 /* Retrieve the value of the current token (if any) and mark it consumed.
  60    The next call to token() will get another token from the lexer.  */
  61 static inline const char *
  62 advance (void)
  63 {
  64   T.valid = false;
  65   return T.value;
  66 }
  67
  68 /* Diagnostics.  */
  69
  70 /* This array is indexed by the token code minus CHAR_TOKEN_OFFSET.  */
  71 static const char *const token_names[] = {
  72   "GTY",
  73   "typedef",
  74   "extern",
  75   "static",
  76   "union",
  77   "struct",
  78   "enum",
  79   "...",
  80   "ptr_alias",
  81   "nested_ptr",
  82   "a param<N>_is option",
  83   "a number",
  84   "a scalar type",
  85   "an identifier",
  86   "a string constant",
  87   "a character constant",
  88   "an array declarator",
  89   "a C++ keyword to ignore"
  90 };
  91
  92 /* This array is indexed by token code minus FIRST_TOKEN_WITH_VALUE.  */
  93 static const char *const token_value_format[] = {
  94   "%s",
  95   "'%s'",
  96   "'%s'",
  97   "'%s'",
  98   "'\"%s\"'",
  99   "\"'%s'\"",
 100   "'[%s]'",
 101   "'%s'",
 102 };
 103
 104 /* Produce a printable representation for a token defined by CODE and
 105    VALUE.  This sometimes returns pointers into malloc memory and
 106    sometimes not, therefore it is unsafe to free the pointer it
 107    returns, so that memory is leaked.  This does not matter, as this
 108    function is only used for diagnostics, and in a successful run of
 109    the program there will be none.  */
 110 static const char *
 111 print_token (int code, const char *value)
 112 {
 113   if (code < CHAR_TOKEN_OFFSET)
 114     return xasprintf ("'%c'", code);
 115   else if (code < FIRST_TOKEN_WITH_VALUE)
 116     return xasprintf ("'%s'", token_names[code - CHAR_TOKEN_OFFSET]);
 117   else if (!value)
 118     return token_names[code - CHAR_TOKEN_OFFSET];       /* don't quote these */
 119   else
 120     return xasprintf (token_value_format[code - FIRST_TOKEN_WITH_VALUE],
 121                       value);
 122 }
 123
 124 /* Convenience wrapper around print_token which produces the printable
 125    representation of the current token.  */
 126 static inline const char *
 127 print_cur_token (void)
 128 {
 129   return print_token (T.code, T.value);
 130 }
 131
 132 /* Report a parse error on the current line, with diagnostic MSG.
 133    Behaves as standard printf with respect to additional arguments and
 134    format escapes.  */
 135 static void ATTRIBUTE_PRINTF_1
 136 parse_error (const char *msg, ...)
 137 {
 138   va_list ap;
 139
 140   fprintf (stderr, "%s:%d: parse error: ",
 141            get_input_file_name (lexer_line.file), lexer_line.line);
 142
 143   va_start (ap, msg);
 144   vfprintf (stderr, msg, ap);
 145   va_end (ap);
 146
 147   fputc ('\n', stderr);
 148
 149   hit_error = true;
 150 }
 151
 152 /* If the next token does not have code T, report a parse error; otherwise
 153    return the token's value.  */
 154 static const char *
 155 require (int t)
 156 {
 157   int u = token ();
 158   const char *v = advance ();
 159   if (u != t)
 160     {
 161       parse_error ("expected %s, have %s",
 162                    print_token (t, 0), print_token (u, v));
 163       return 0;
 164     }
 165   return v;
 166 }
 167
 168 /* If the next token does not have one of the codes T1 or T2, report a
 169    parse error; otherwise return the token's value.  */
 170 static const char *
 171 require2 (int t1, int t2)
 172 {
 173   int u = token ();
 174   const char *v = advance ();
 175   if (u != t1 && u != t2)
 176     {
 177       parse_error ("expected %s or %s, have %s",
 178                    print_token (t1, 0), print_token (t2, 0),
 179                    print_token (u, v));
 180       return 0;
 181     }
 182   return v;
 183 }
 184
 185 /* Near-terminals.  */
 186
 187 /* C-style string constant concatenation: STRING+
 188    Bare STRING should appear nowhere else in this file.  */
 189 static const char *
 190 string_seq (void)
 191 {
 192   const char *s1, *s2;
 193   size_t l1, l2;
 194   char *buf;
 195
 196   s1 = require (STRING);
 197   if (s1 == 0)
 198     return "";
 199   while (token () == STRING)
 200     {
 201       s2 = advance ();
 202
 203       l1 = strlen (s1);
 204       l2 = strlen (s2);
 205       buf = XRESIZEVEC (char, CONST_CAST (char *, s1), l1 + l2 + 1);
 206       memcpy (buf + l1, s2, l2 + 1);
 207       XDELETE (CONST_CAST (char *, s2));
 208       s1 = buf;
 209     }
 210   return s1;
 211 }
 212
 213
 214 /* The caller has detected a template declaration that starts
 215    with TMPL_NAME.  Parse up to the closing '>'.  This recognizes
 216    simple template declarations of the form ID<ID1,ID2,...,IDn>.
 217    It does not try to parse anything more sophisticated than that.
 218
 219    Returns the template declaration string "ID<ID1,ID2,...,IDn>".  */
 220
 221 static const char *
 222 require_template_declaration (const char *tmpl_name)
 223 {
 224   char *str;
 225
 226   /* Recognize the opening '<'.  */
 227   require ('<');
 228   str = concat (tmpl_name, "<", (char *) 0);
 229
 230   /* Read the comma-separated list of identifiers.  */
 231   while (token () != '>')
 232     {
 233       const char *id = require2 (ID, ',');
 234       if (id == NULL)
 235         id = ",";
 236       str = concat (str, id, (char *) 0);
 237     }
 238
 239   /* Recognize the closing '>'.  */
 240   require ('>');
 241   str = concat (str, ">", (char *) 0);
 242
 243   return str;
 244 }
 245
 246
 247 /* typedef_name: either an ID, or a template type
 248    specification of the form ID<t1,t2,...,tn>.  */
 249
 250 static const char *
 251 typedef_name (void)
 252 {
 253   const char *id = require (ID);
 254   if (token () == '<')
 255     return require_template_declaration (id);
 256   else
 257     return id;
 258 }
 259
 260 /* Absorb a sequence of tokens delimited by balanced ()[]{}.  */
 261 static void
 262 consume_balanced (int opener, int closer)
 263 {
 264   require (opener);
 265   for (;;)
 266     switch (token ())
 267       {
 268       default:
 269         advance ();
 270         break;
 271       case '(':
 272         consume_balanced ('(', ')');
 273         break;
 274       case '[':
 275         consume_balanced ('[', ']');
 276         break;
 277       case '{':
 278         consume_balanced ('{', '}');
 279         break;
 280
 281       case '}':
 282       case ']':
 283       case ')':
 284         if (token () != closer)
 285           parse_error ("unbalanced delimiters - expected '%c', have '%c'",
 286                        closer, token ());
 287       advance ();
 288       return;
 289
 290       case EOF_TOKEN:
 291         parse_error ("unexpected end of file within %c%c-delimited construct",
 292                      opener, closer);
 293         return;
 294       }
 295 }
 296
 297 /* Absorb a sequence of tokens, possibly including ()[]{}-delimited
 298    expressions, until we encounter an end-of-statement marker (a ';' or
 299    a '}') outside any such delimiters; absorb that too.  */
 300
 301 static void
 302 consume_until_eos (void)
 303 {
 304   for (;;)
 305     switch (token ())
 306       {
 307       case ';':
 308         advance ();
 309         return;
 310
 311       case '{':
 312         consume_balanced ('{', '}');
 313         return;
 314
 315       case '(':
 316         consume_balanced ('(', ')');
 317         break;
 318
 319       case '[':
 320         consume_balanced ('[', ']');
 321         break;
 322
 323       case '}':
 324       case ']':
 325       case ')':
 326         parse_error ("unmatched '%c' while scanning for ';'", token ());
 327         return;
 328
 329       case EOF_TOKEN:
 330         parse_error ("unexpected end of file while scanning for ';'");
 331         return;
 332
 333       default:
 334         advance ();
 335         break;
 336       }
 337 }
 338
 339 /* Absorb a sequence of tokens, possibly including ()[]{}-delimited
 340    expressions, until we encounter a comma or semicolon outside any
 341    such delimiters; absorb that too.  Returns true if the loop ended
 342    with a comma.  */
 343
 344 static bool
 345 consume_until_comma_or_eos ()
 346 {
 347   for (;;)
 348     switch (token ())
 349       {
 350       case ',':
 351         advance ();
 352         return true;
 353
 354       case ';':
 355         advance ();
 356         return false;
 357
 358       case '{':
 359         consume_balanced ('{', '}');
 360         return false;
 361
 362       case '(':
 363         consume_balanced ('(', ')');
 364         break;
 365
 366       case '[':
 367         consume_balanced ('[', ']');
 368         break;
 369
 370       case '}':
 371       case ']':
 372       case ')':
 373         parse_error ("unmatched '%s' while scanning for ',' or ';'",
 374                      print_cur_token ());
 375       return false;
 376
 377       case EOF_TOKEN:
 378         parse_error ("unexpected end of file while scanning for ',' or ';'");
 379         return false;
 380
 381       default:
 382         advance ();
 383         break;
 384       }
 385 }
 386 \f
 387
 388 /* GTY(()) option handling.  */
 389 static type_p type (options_p *optsp, bool nested);
 390
 391 /* Optional parenthesized string: ('(' string_seq ')')? */
 392 static options_p
 393 str_optvalue_opt (options_p prev)
 394 {
 395   const char *name = advance ();
 396   const char *value = "";
 397   if (token () == '(')
 398     {
 399       advance ();
 400       value = string_seq ();
 401       require (')');
 402     }
 403   return create_string_option (prev, name, value);
 404 }
 405
 406 /* absdecl: type '*'*
 407    -- a vague approximation to what the C standard calls an abstract
 408    declarator.  The only kinds that are actually used are those that
 409    are just a bare type and those that have trailing pointer-stars.
 410    Further kinds should be implemented if and when they become
 411    necessary.  Used only within GTY(()) option values, therefore
 412    further GTY(()) tags within the type are invalid.  Note that the
 413    return value has already been run through adjust_field_type.  */
 414 static type_p
 415 absdecl (void)
 416 {
 417   type_p ty;
 418   options_p opts;
 419
 420   ty = type (&opts, true);
 421   while (token () == '*')
 422     {
 423       ty = create_pointer (ty);
 424       advance ();
 425     }
 426
 427   if (opts)
 428     parse_error ("nested GTY(()) options are invalid");
 429
 430   return adjust_field_type (ty, 0);
 431 }
 432
 433 /* Type-option: '(' absdecl ')' */
 434 static options_p
 435 type_optvalue (options_p prev, const char *name)
 436 {
 437   type_p ty;
 438   require ('(');
 439   ty = absdecl ();
 440   require (')');
 441   return create_type_option (prev, name, ty);
 442 }
 443
 444 /* Nested pointer data: '(' type '*'* ',' string_seq ',' string_seq ')' */
 445 static options_p
 446 nestedptr_optvalue (options_p prev)
 447 {
 448   type_p ty;
 449   const char *from, *to;
 450
 451   require ('(');
 452   ty = absdecl ();
 453   require (',');
 454   to = string_seq ();
 455   require (',');
 456   from = string_seq ();
 457   require (')');
 458
 459   return create_nested_ptr_option (prev, ty, to, from);
 460 }
 461
 462 /* One GTY(()) option:
 463    ID str_optvalue_opt
 464    | PTR_ALIAS type_optvalue
 465    | PARAM_IS type_optvalue
 466    | NESTED_PTR nestedptr_optvalue
 467 */
 468 static options_p
 469 option (options_p prev)
 470 {
 471   switch (token ())
 472     {
 473     case ID:
 474       return str_optvalue_opt (prev);
 475
 476     case PTR_ALIAS:
 477       advance ();
 478       return type_optvalue (prev, "ptr_alias");
 479
 480     case PARAM_IS:
 481       return type_optvalue (prev, advance ());
 482
 483     case NESTED_PTR:
 484       advance ();
 485       return nestedptr_optvalue (prev);
 486
 487     case USER_GTY:
 488       advance ();
 489       return create_string_option (prev, "user", "");
 490
 491     default:
 492       parse_error ("expected an option keyword, have %s", print_cur_token ());
 493       advance ();
 494       return create_string_option (prev, "", "");
 495     }
 496 }
 497
 498 /* One comma-separated list of options.  */
 499 static options_p
 500 option_seq (void)
 501 {
 502   options_p o;
 503
 504   o = option (0);
 505   while (token () == ',')
 506     {
 507       advance ();
 508       o = option (o);
 509     }
 510   return o;
 511 }
 512
 513 /* GTY marker: 'GTY' '(' '(' option_seq? ')' ')' */
 514 static options_p
 515 gtymarker (void)
 516 {
 517   options_p result = 0;
 518   require (GTY_TOKEN);
 519   require ('(');
 520   require ('(');
 521   if (token () != ')')
 522     result = option_seq ();
 523   require (')');
 524   require (')');
 525   return result;
 526 }
 527
 528 /* Optional GTY marker.  */
 529 static options_p
 530 gtymarker_opt (void)
 531 {
 532   if (token () != GTY_TOKEN)
 533     return 0;
 534   return gtymarker ();
 535 }
 536
 537
 538 \f
 539 /* Declarators. The logic here is largely lifted from c-parser.c.
 540    Note that we do not have to process abstract declarators, which can
 541    appear only in parameter type lists or casts (but see absdecl,
 542    above).  Also, type qualifiers are thrown out in gengtype-lex.l so
 543    we don't have to do it.  */
 544
 545 /* array_and_function_declarators_opt:
 546    \epsilon
 547    array_and_function_declarators_opt ARRAY
 548    array_and_function_declarators_opt '(' ... ')'
 549
 550    where '...' indicates stuff we ignore except insofar as grouping
 551    symbols ()[]{} must balance.
 552
 553    Subroutine of direct_declarator - do not use elsewhere. */
 554
 555 static type_p
 556 array_and_function_declarators_opt (type_p ty)
 557 {
 558   if (token () == ARRAY)
 559     {
 560       const char *array = advance ();
 561       return create_array (array_and_function_declarators_opt (ty), array);
 562     }
 563   else if (token () == '(')
 564     {
 565       /* We don't need exact types for functions.  */
 566       consume_balanced ('(', ')');
 567       array_and_function_declarators_opt (ty);
 568       return create_scalar_type ("function type");
 569     }
 570   else
 571     return ty;
 572 }
 573
 574 static type_p inner_declarator (type_p, const char **, options_p *, bool);
 575
 576 /* direct_declarator:
 577    '(' inner_declarator ')'
 578    '(' \epsilon ')'     <-- C++ ctors/dtors
 579    gtymarker_opt ID array_and_function_declarators_opt
 580
 581    Subroutine of declarator, mutually recursive with inner_declarator;
 582    do not use elsewhere.
 583
 584    IN_STRUCT is true if we are called while parsing structures or classes.  */
 585
 586 static type_p
 587 direct_declarator (type_p ty, const char **namep, options_p *optsp,
 588                    bool in_struct)
 589 {
 590   /* The first token in a direct-declarator must be an ID, a
 591      GTY marker, or an open parenthesis.  */
 592   switch (token ())
 593     {
 594     case GTY_TOKEN:
 595       *optsp = gtymarker ();
 596       /* fall through */
 597
 598     case ID:
 599       *namep = require (ID);
 600       /* If the next token is '(', we are parsing a function declaration.
 601          Functions are ignored by gengtype, so we return NULL.  */
 602       if (token () == '(')
 603         return NULL;
 604       break;
 605
 606     case '(':
 607       /* If the declarator starts with a '(', we have three options.  We
 608          are either parsing 'TYPE (*ID)' (i.e., a function pointer)
 609          or 'TYPE(...)'.
 610
 611          The latter will be a constructor iff we are inside a
 612          structure or class.  Otherwise, it could be a typedef, but
 613          since we explicitly reject typedefs inside structures, we can
 614          assume that we found a ctor and return NULL.  */
 615       advance ();
 616       if (in_struct && token () != '*')
 617         {
 618           /* Found a constructor.  Find and consume the closing ')'.  */
 619           while (token () != ')')
 620             advance ();
 621           advance ();
 622           /* Tell the caller to ignore this.  */
 623           return NULL;
 624         }
 625       ty = inner_declarator (ty, namep, optsp, in_struct);
 626       require (')');
 627       break;
 628
 629     case IGNORABLE_CXX_KEYWORD:
 630       /* Any C++ keyword like 'operator' means that we are not looking
 631          at a regular data declarator.  */
 632       return NULL;
 633
 634     default:
 635       parse_error ("expected '(', ')', 'GTY', or an identifier, have %s",
 636                    print_cur_token ());
 637       /* Do _not_ advance if what we have is a close squiggle brace, as
 638          we will get much better error recovery that way.  */
 639       if (token () != '}')
 640         advance ();
 641       return 0;
 642     }
 643   return array_and_function_declarators_opt (ty);
 644 }
 645
 646 /* The difference between inner_declarator and declarator is in the
 647    handling of stars.  Consider this declaration:
 648
 649    char * (*pfc) (void)
 650
 651    It declares a pointer to a function that takes no arguments and
 652    returns a char*.  To construct the correct type for this
 653    declaration, the star outside the parentheses must be processed
 654    _before_ the function type, the star inside the parentheses must
 655    be processed _after_ the function type.  To accomplish this,
 656    declarator() creates pointers before recursing (it is actually
 657    coded as a while loop), whereas inner_declarator() recurses before
 658    creating pointers.  */
 659
 660 /* inner_declarator:
 661    '*' inner_declarator
 662    direct_declarator
 663
 664    Mutually recursive subroutine of direct_declarator; do not use
 665    elsewhere.
 666
 667    IN_STRUCT is true if we are called while parsing structures or classes.  */
 668
 669 static type_p
 670 inner_declarator (type_p ty, const char **namep, options_p *optsp,
 671                   bool in_struct)
 672 {
 673   if (token () == '*')
 674     {
 675       type_p inner;
 676       advance ();
 677       inner = inner_declarator (ty, namep, optsp, in_struct);
 678       if (inner == 0)
 679         return 0;
 680       else
 681         return create_pointer (ty);
 682     }
 683   else
 684     return direct_declarator (ty, namep, optsp, in_struct);
 685 }
 686
 687 /* declarator: '*'+ direct_declarator
 688
 689    This is the sole public interface to this part of the grammar.
 690    Arguments are the type known so far, a pointer to where the name
 691    may be stored, and a pointer to where GTY options may be stored.
 692
 693    IN_STRUCT is true when we are called to parse declarators inside
 694    a structure or class.
 695
 696    Returns the final type.  */
 697
 698 static type_p
 699 declarator (type_p ty, const char **namep, options_p *optsp,
 700             bool in_struct = false)
 701 {
 702   *namep = 0;
 703   *optsp = 0;
 704   while (token () == '*')
 705     {
 706       advance ();
 707       ty = create_pointer (ty);
 708     }
 709   return direct_declarator (ty, namep, optsp, in_struct);
 710 }
 711 \f
 712 /* Types and declarations.  */
 713
 714 /* Structure field(s) declaration:
 715    (
 716    type bitfield ';'
 717    | type declarator bitfield? ( ',' declarator bitfield? )+ ';'
 718    )+
 719
 720    Knows that such declarations must end with a close brace (or,
 721    erroneously, at EOF).
 722 */
 723 static pair_p
 724 struct_field_seq (void)
 725 {
 726   pair_p f = 0;
 727   type_p ty, dty;
 728   options_p opts, dopts;
 729   const char *name;
 730   bool another;
 731
 732   do
 733     {
 734       ty = type (&opts, true);
 735
 736       if (!ty || token () == ':')
 737         {
 738           consume_until_eos ();
 739           continue;
 740         }
 741
 742       do
 743         {
 744           dty = declarator (ty, &name, &dopts, true);
 745
 746           /* There could be any number of weird things after the declarator,
 747              notably bitfield declarations and __attribute__s.  If this
 748              function returns true, the last thing was a comma, so we have
 749              more than one declarator paired with the current type.  */
 750           another = consume_until_comma_or_eos ();
 751
 752           if (!dty)
 753             continue;
 754
 755           if (opts && dopts)
 756             parse_error ("two GTY(()) options for field %s", name);
 757           if (opts && !dopts)
 758             dopts = opts;
 759
 760           f = create_field_at (f, dty, name, dopts, &lexer_line);
 761         }
 762       while (another);
 763     }
 764   while (token () != '}' && token () != EOF_TOKEN);
 765   return nreverse_pairs (f);
 766 }
 767
 768 /* Return true if OPTS contain the option named STR.  */
 769
 770 static bool
 771 opts_have (options_p opts, const char *str)
 772 {
 773   for (options_p opt = opts; opt; opt = opt->next)
 774     if (strcmp (opt->name, str) == 0)
 775       return true;
 776   return false;
 777 }
 778
 779
 780 /* This is called type(), but what it parses (sort of) is what C calls
 781    declaration-specifiers and specifier-qualifier-list:
 782
 783    SCALAR
 784    | ID     // typedef
 785    | (STRUCT|UNION) ID? gtymarker? ( '{' gtymarker? struct_field_seq '}' )?
 786    | ENUM ID ( '{' ... '}' )?
 787
 788    Returns a partial type; under some conditions (notably
 789    "struct foo GTY((...)) thing;") it may write an options
 790    structure to *OPTSP.
 791
 792    NESTED is true when parsing a declaration already known to have a
 793    GTY marker. In these cases, typedef and enum declarations are not
 794    allowed because gengtype only understands types at the global
 795    scope.  */
 796
 797 static type_p
 798 type (options_p *optsp, bool nested)
 799 {
 800   const char *s;
 801   *optsp = 0;
 802   switch (token ())
 803     {
 804     case SCALAR:
 805       s = advance ();
 806       return create_scalar_type (s);
 807
 808     case ID:
 809       s = typedef_name ();
 810       return resolve_typedef (s, &lexer_line);
 811
 812     case IGNORABLE_CXX_KEYWORD:
 813       /* By returning NULL here, we indicate to the caller that they
 814          should ignore everything following this keyword up to the
 815          next ';' or '}'.  */
 816       return NULL;
 817
 818     case STRUCT:
 819     case UNION:
 820       {
 821         options_p opts = 0;
 822         /* GTY annotations follow attribute syntax
 823            GTY_BEFORE_ID is for union/struct declarations
 824            GTY_AFTER_ID is for variable declarations.  */
 825         enum
 826         {
 827           NO_GTY,
 828           GTY_BEFORE_ID,
 829           GTY_AFTER_ID
 830         } is_gty = NO_GTY;
 831         enum typekind kind = (token () == UNION) ? TYPE_UNION : TYPE_STRUCT;
 832         advance ();
 833
 834         /* Top-level structures that are not explicitly tagged GTY(())
 835            are treated as mere forward declarations.  This is because
 836            there are a lot of structures that we don't need to know
 837            about, and some of those have C++ and macro constructs that
 838            we cannot handle.  */
 839         if (nested || token () == GTY_TOKEN)
 840           {
 841             is_gty = GTY_BEFORE_ID;
 842             opts = gtymarker_opt ();
 843           }
 844
 845         if (token () == ID)
 846           s = advance ();
 847         else
 848           s = xasprintf ("anonymous:%s:%d",
 849                          get_input_file_name (lexer_line.file),
 850                          lexer_line.line);
 851
 852         /* Unfortunately above GTY_TOKEN check does not capture the
 853            typedef struct_type GTY case.  */
 854         if (token () == GTY_TOKEN)
 855           {
 856             is_gty = GTY_AFTER_ID;
 857             opts = gtymarker_opt ();
 858           }
 859
 860         if (token () == ':')
 861           {
 862             /* Skip over C++ inheritance specification.  */
 863             while (token () != '{')
 864               advance ();
 865           }
 866
 867         if (is_gty)
 868           {
 869             bool is_user_gty = opts_have (opts, "user");
 870             if (token () == '{')
 871               {
 872                 pair_p fields;
 873
 874                 if (is_gty == GTY_AFTER_ID)
 875                   parse_error ("GTY must be specified before identifier");
 876
 877                 if (!is_user_gty)
 878                   {
 879                     advance ();
 880                     fields = struct_field_seq ();
 881                     require ('}');
 882                   }
 883                 else
 884                   {
 885                     /* Do not look inside user defined structures.  */
 886                     fields = NULL;
 887                     kind = TYPE_USER_STRUCT;
 888                     consume_balanced ('{', '}');
 889                     return create_user_defined_type (s, &lexer_line);
 890                   }
 891
 892                 return new_structure (s, kind, &lexer_line, fields, opts);
 893               }
 894           }
 895         else if (token () == '{')
 896           consume_balanced ('{', '}');
 897         if (opts)
 898           *optsp = opts;
 899         return find_structure (s, kind);
 900       }
 901
 902     case TYPEDEF:
 903       /* In C++, a typedef inside a struct/class/union defines a new
 904          type for that inner scope.  We cannot support this in
 905          gengtype because we have no concept of scoping.
 906
 907          We handle typedefs in the global scope separately (see
 908          parse_file), so if we find a 'typedef', we must be inside
 909          a struct.  */
 910       gcc_assert (nested);
 911       parse_error ("typedefs not supported in structures marked with "
 912                    "automatic GTY markers.  Use GTY((user)) to mark "
 913                    "this structure.");
 914       advance ();
 915       return NULL;
 916
 917     case ENUM:
 918       advance ();
 919       if (token () == ID)
 920         s = advance ();
 921       else
 922         s = xasprintf ("anonymous:%s:%d",
 923                        get_input_file_name (lexer_line.file),
 924                        lexer_line.line);
 925
 926       if (token () == '{')
 927         consume_balanced ('{', '}');
 928
 929       /* If after parsing the enum we are at the end of the statement,
 930          and we are currently inside a structure, then this was an
 931          enum declaration inside this scope.
 932
 933          We cannot support this for the same reason we cannot support
 934          'typedef' inside structures (see the TYPEDEF handler above).
 935          If this happens, emit an error and return NULL.  */
 936       if (nested && token () == ';')
 937         {
 938           parse_error ("enum definitions not supported in structures marked "
 939                        "with automatic GTY markers.  Use GTY((user)) to mark "
 940                        "this structure.");
 941           advance ();
 942           return NULL;
 943         }
 944
 945       return create_scalar_type (s);
 946
 947     default:
 948       parse_error ("expected a type specifier, have %s", print_cur_token ());
 949       advance ();
 950       return create_scalar_type ("erroneous type");
 951     }
 952 }
 953 \f
 954 /* Top level constructs.  */
 955
 956 /* Dispatch declarations beginning with 'typedef'.  */
 957
 958 static void
 959 typedef_decl (void)
 960 {
 961   type_p ty, dty;
 962   const char *name;
 963   options_p opts;
 964   bool another;
 965
 966   gcc_assert (token () == TYPEDEF);
 967   advance ();
 968
 969   ty = type (&opts, false);
 970   if (!ty)
 971     return;
 972   if (opts)
 973     parse_error ("GTY((...)) cannot be applied to a typedef");
 974   do
 975     {
 976       dty = declarator (ty, &name, &opts);
 977       if (opts)
 978         parse_error ("GTY((...)) cannot be applied to a typedef");
 979
 980       /* Yet another place where we could have junk (notably attributes)
 981          after the declarator.  */
 982       another = consume_until_comma_or_eos ();
 983       if (dty)
 984         do_typedef (name, dty, &lexer_line);
 985     }
 986   while (another);
 987 }
 988
 989 /* Structure definition: type() does all the work.  */
 990
 991 static void
 992 struct_or_union (void)
 993 {
 994   options_p dummy;
 995   type (&dummy, false);
 996   /* There may be junk after the type: notably, we cannot currently
 997      distinguish 'struct foo *function(prototype);' from 'struct foo;'
 998      ...  we could call declarator(), but it's a waste of time at
 999      present.  Instead, just eat whatever token is currently lookahead
1000      and go back to lexical skipping mode. */
1001   advance ();
1002 }
1003
1004 /* GC root declaration:
1005    (extern|static) gtymarker? type ID array_declarators_opt (';'|'=')
1006    If the gtymarker is not present, we ignore the rest of the declaration.  */
1007 static void
1008 extern_or_static (void)
1009 {
1010   options_p opts, opts2, dopts;
1011   type_p ty, dty;
1012   const char *name;
1013   require2 (EXTERN, STATIC);
1014
1015   if (token () != GTY_TOKEN)
1016     {
1017       advance ();
1018       return;
1019     }
1020
1021   opts = gtymarker ();
1022   ty = type (&opts2, true);     /* if we get here, it's got a GTY(()) */
1023   dty = declarator (ty, &name, &dopts);
1024
1025   if ((opts && dopts) || (opts && opts2) || (opts2 && dopts))
1026     parse_error ("GTY((...)) specified more than once for %s", name);
1027   else if (opts2)
1028     opts = opts2;
1029   else if (dopts)
1030     opts = dopts;
1031
1032   if (dty)
1033     {
1034       note_variable (name, adjust_field_type (dty, opts), opts, &lexer_line);
1035       require2 (';', '=');
1036     }
1037 }
1038
1039 /* Parse the file FNAME for GC-relevant declarations and definitions.
1040    This is the only entry point to this file.  */
1041 void
1042 parse_file (const char *fname)
1043 {
1044   yybegin (fname);
1045   for (;;)
1046     {
1047       switch (token ())
1048         {
1049         case EXTERN:
1050         case STATIC:
1051           extern_or_static ();
1052           break;
1053
1054         case STRUCT:
1055         case UNION:
1056           struct_or_union ();
1057           break;
1058
1059         case TYPEDEF:
1060           typedef_decl ();
1061           break;
1062
1063         case EOF_TOKEN:
1064           goto eof;
1065
1066         default:
1067           parse_error ("unexpected top level token, %s", print_cur_token ());
1068           goto eof;
1069         }
1070       lexer_toplevel_done = 1;
1071     }
1072
1073  eof:
1074   advance ();
1075   yyend ();
1076 }