* parser-defs.h (write_exp_msymbol): Remove TEXT_SYMBOL_TYPE
[deliverable/binutils-gdb.git] / gdb / m2-exp.y
1 /* YACC grammar for Modula-2 expressions, for GDB.
2 Copyright (C) 1986, 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1999,
3 2000, 2007, 2008 Free Software Foundation, Inc.
4 Generated from expread.y (now c-exp.y) and contributed by the Department
5 of Computer Science at the State University of New York at Buffalo, 1991.
6
7 This file is part of GDB.
8
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 2 of the License, or
12 (at your option) any later version.
13
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
18
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor,
22 Boston, MA 02110-1301, USA. */
23
24 /* Parse a Modula-2 expression from text in a string,
25 and return the result as a struct expression pointer.
26 That structure contains arithmetic operations in reverse polish,
27 with constants represented by operations that are followed by special data.
28 See expression.h for the details of the format.
29 What is important here is that it can be built up sequentially
30 during the process of parsing; the lower levels of the tree always
31 come first in the result.
32
33 Note that malloc's and realloc's in this file are transformed to
34 xmalloc and xrealloc respectively by the same sed command in the
35 makefile that remaps any other malloc/realloc inserted by the parser
36 generator. Doing this with #defines and trying to control the interaction
37 with include files (<malloc.h> and <stdlib.h> for example) just became
38 too messy, particularly when such includes can be inserted at random
39 times by the parser generator. */
40
41 %{
42
43 #include "defs.h"
44 #include "gdb_string.h"
45 #include "expression.h"
46 #include "language.h"
47 #include "value.h"
48 #include "parser-defs.h"
49 #include "m2-lang.h"
50 #include "bfd.h" /* Required by objfiles.h. */
51 #include "symfile.h" /* Required by objfiles.h. */
52 #include "objfiles.h" /* For have_full_symbols and have_partial_symbols */
53 #include "block.h"
54
55 /* Remap normal yacc parser interface names (yyparse, yylex, yyerror, etc),
56 as well as gratuitiously global symbol names, so we can have multiple
57 yacc generated parsers in gdb. Note that these are only the variables
58 produced by yacc. If other parser generators (bison, byacc, etc) produce
59 additional global names that conflict at link time, then those parser
60 generators need to be fixed instead of adding those names to this list. */
61
62 #define yymaxdepth m2_maxdepth
63 #define yyparse m2_parse
64 #define yylex m2_lex
65 #define yyerror m2_error
66 #define yylval m2_lval
67 #define yychar m2_char
68 #define yydebug m2_debug
69 #define yypact m2_pact
70 #define yyr1 m2_r1
71 #define yyr2 m2_r2
72 #define yydef m2_def
73 #define yychk m2_chk
74 #define yypgo m2_pgo
75 #define yyact m2_act
76 #define yyexca m2_exca
77 #define yyerrflag m2_errflag
78 #define yynerrs m2_nerrs
79 #define yyps m2_ps
80 #define yypv m2_pv
81 #define yys m2_s
82 #define yy_yys m2_yys
83 #define yystate m2_state
84 #define yytmp m2_tmp
85 #define yyv m2_v
86 #define yy_yyv m2_yyv
87 #define yyval m2_val
88 #define yylloc m2_lloc
89 #define yyreds m2_reds /* With YYDEBUG defined */
90 #define yytoks m2_toks /* With YYDEBUG defined */
91 #define yyname m2_name /* With YYDEBUG defined */
92 #define yyrule m2_rule /* With YYDEBUG defined */
93 #define yylhs m2_yylhs
94 #define yylen m2_yylen
95 #define yydefred m2_yydefred
96 #define yydgoto m2_yydgoto
97 #define yysindex m2_yysindex
98 #define yyrindex m2_yyrindex
99 #define yygindex m2_yygindex
100 #define yytable m2_yytable
101 #define yycheck m2_yycheck
102
103 #ifndef YYDEBUG
104 #define YYDEBUG 1 /* Default to yydebug support */
105 #endif
106
107 #define YYFPRINTF parser_fprintf
108
109 int yyparse (void);
110
111 static int yylex (void);
112
113 void yyerror (char *);
114
115 #if 0
116 static char *make_qualname (char *, char *);
117 #endif
118
119 static int parse_number (int);
120
121 /* The sign of the number being parsed. */
122 static int number_sign = 1;
123
124 /* The block that the module specified by the qualifer on an identifer is
125 contained in, */
126 #if 0
127 static struct block *modblock=0;
128 #endif
129
130 %}
131
132 /* Although the yacc "value" of an expression is not used,
133 since the result is stored in the structure being created,
134 other node types do have values. */
135
136 %union
137 {
138 LONGEST lval;
139 ULONGEST ulval;
140 DOUBLEST dval;
141 struct symbol *sym;
142 struct type *tval;
143 struct stoken sval;
144 int voidval;
145 struct block *bval;
146 enum exp_opcode opcode;
147 struct internalvar *ivar;
148
149 struct type **tvec;
150 int *ivec;
151 }
152
153 %type <voidval> exp type_exp start set
154 %type <voidval> variable
155 %type <tval> type
156 %type <bval> block
157 %type <sym> fblock
158
159 %token <lval> INT HEX ERROR
160 %token <ulval> UINT M2_TRUE M2_FALSE CHAR
161 %token <dval> FLOAT
162
163 /* Both NAME and TYPENAME tokens represent symbols in the input,
164 and both convey their data as strings.
165 But a TYPENAME is a string that happens to be defined as a typedef
166 or builtin type name (such as int or char)
167 and a NAME is any other symbol.
168
169 Contexts where this distinction is not important can use the
170 nonterminal "name", which matches either NAME or TYPENAME. */
171
172 %token <sval> STRING
173 %token <sval> NAME BLOCKNAME IDENT VARNAME
174 %token <sval> TYPENAME
175
176 %token SIZE CAP ORD HIGH ABS MIN_FUNC MAX_FUNC FLOAT_FUNC VAL CHR ODD TRUNC
177 %token TSIZE
178 %token INC DEC INCL EXCL
179
180 /* The GDB scope operator */
181 %token COLONCOLON
182
183 %token <voidval> INTERNAL_VAR
184
185 /* M2 tokens */
186 %left ','
187 %left ABOVE_COMMA
188 %nonassoc ASSIGN
189 %left '<' '>' LEQ GEQ '=' NOTEQUAL '#' IN
190 %left OROR
191 %left LOGICAL_AND '&'
192 %left '@'
193 %left '+' '-'
194 %left '*' '/' DIV MOD
195 %right UNARY
196 %right '^' DOT '[' '('
197 %right NOT '~'
198 %left COLONCOLON QID
199 /* This is not an actual token ; it is used for precedence.
200 %right QID
201 */
202
203 \f
204 %%
205
206 start : exp
207 | type_exp
208 ;
209
210 type_exp: type
211 { write_exp_elt_opcode(OP_TYPE);
212 write_exp_elt_type($1);
213 write_exp_elt_opcode(OP_TYPE);
214 }
215 ;
216
217 /* Expressions */
218
219 exp : exp '^' %prec UNARY
220 { write_exp_elt_opcode (UNOP_IND); }
221 ;
222
223 exp : '-'
224 { number_sign = -1; }
225 exp %prec UNARY
226 { number_sign = 1;
227 write_exp_elt_opcode (UNOP_NEG); }
228 ;
229
230 exp : '+' exp %prec UNARY
231 { write_exp_elt_opcode(UNOP_PLUS); }
232 ;
233
234 exp : not_exp exp %prec UNARY
235 { write_exp_elt_opcode (UNOP_LOGICAL_NOT); }
236 ;
237
238 not_exp : NOT
239 | '~'
240 ;
241
242 exp : CAP '(' exp ')'
243 { write_exp_elt_opcode (UNOP_CAP); }
244 ;
245
246 exp : ORD '(' exp ')'
247 { write_exp_elt_opcode (UNOP_ORD); }
248 ;
249
250 exp : ABS '(' exp ')'
251 { write_exp_elt_opcode (UNOP_ABS); }
252 ;
253
254 exp : HIGH '(' exp ')'
255 { write_exp_elt_opcode (UNOP_HIGH); }
256 ;
257
258 exp : MIN_FUNC '(' type ')'
259 { write_exp_elt_opcode (UNOP_MIN);
260 write_exp_elt_type ($3);
261 write_exp_elt_opcode (UNOP_MIN); }
262 ;
263
264 exp : MAX_FUNC '(' type ')'
265 { write_exp_elt_opcode (UNOP_MAX);
266 write_exp_elt_type ($3);
267 write_exp_elt_opcode (UNOP_MIN); }
268 ;
269
270 exp : FLOAT_FUNC '(' exp ')'
271 { write_exp_elt_opcode (UNOP_FLOAT); }
272 ;
273
274 exp : VAL '(' type ',' exp ')'
275 { write_exp_elt_opcode (BINOP_VAL);
276 write_exp_elt_type ($3);
277 write_exp_elt_opcode (BINOP_VAL); }
278 ;
279
280 exp : CHR '(' exp ')'
281 { write_exp_elt_opcode (UNOP_CHR); }
282 ;
283
284 exp : ODD '(' exp ')'
285 { write_exp_elt_opcode (UNOP_ODD); }
286 ;
287
288 exp : TRUNC '(' exp ')'
289 { write_exp_elt_opcode (UNOP_TRUNC); }
290 ;
291
292 exp : TSIZE '(' exp ')'
293 { write_exp_elt_opcode (UNOP_SIZEOF); }
294 ;
295
296 exp : SIZE exp %prec UNARY
297 { write_exp_elt_opcode (UNOP_SIZEOF); }
298 ;
299
300
301 exp : INC '(' exp ')'
302 { write_exp_elt_opcode(UNOP_PREINCREMENT); }
303 ;
304
305 exp : INC '(' exp ',' exp ')'
306 { write_exp_elt_opcode(BINOP_ASSIGN_MODIFY);
307 write_exp_elt_opcode(BINOP_ADD);
308 write_exp_elt_opcode(BINOP_ASSIGN_MODIFY); }
309 ;
310
311 exp : DEC '(' exp ')'
312 { write_exp_elt_opcode(UNOP_PREDECREMENT);}
313 ;
314
315 exp : DEC '(' exp ',' exp ')'
316 { write_exp_elt_opcode(BINOP_ASSIGN_MODIFY);
317 write_exp_elt_opcode(BINOP_SUB);
318 write_exp_elt_opcode(BINOP_ASSIGN_MODIFY); }
319 ;
320
321 exp : exp DOT NAME
322 { write_exp_elt_opcode (STRUCTOP_STRUCT);
323 write_exp_string ($3);
324 write_exp_elt_opcode (STRUCTOP_STRUCT); }
325 ;
326
327 exp : set
328 ;
329
330 exp : exp IN set
331 { error("Sets are not implemented.");}
332 ;
333
334 exp : INCL '(' exp ',' exp ')'
335 { error("Sets are not implemented.");}
336 ;
337
338 exp : EXCL '(' exp ',' exp ')'
339 { error("Sets are not implemented.");}
340 ;
341
342 set : '{' arglist '}'
343 { error("Sets are not implemented.");}
344 | type '{' arglist '}'
345 { error("Sets are not implemented.");}
346 ;
347
348
349 /* Modula-2 array subscript notation [a,b,c...] */
350 exp : exp '['
351 /* This function just saves the number of arguments
352 that follow in the list. It is *not* specific to
353 function types */
354 { start_arglist(); }
355 non_empty_arglist ']' %prec DOT
356 { write_exp_elt_opcode (MULTI_SUBSCRIPT);
357 write_exp_elt_longcst ((LONGEST) end_arglist());
358 write_exp_elt_opcode (MULTI_SUBSCRIPT); }
359 ;
360
361 exp : exp '[' exp ']'
362 { write_exp_elt_opcode (BINOP_SUBSCRIPT); }
363 ;
364
365 exp : exp '('
366 /* This is to save the value of arglist_len
367 being accumulated by an outer function call. */
368 { start_arglist (); }
369 arglist ')' %prec DOT
370 { write_exp_elt_opcode (OP_FUNCALL);
371 write_exp_elt_longcst ((LONGEST) end_arglist ());
372 write_exp_elt_opcode (OP_FUNCALL); }
373 ;
374
375 arglist :
376 ;
377
378 arglist : exp
379 { arglist_len = 1; }
380 ;
381
382 arglist : arglist ',' exp %prec ABOVE_COMMA
383 { arglist_len++; }
384 ;
385
386 non_empty_arglist
387 : exp
388 { arglist_len = 1; }
389 ;
390
391 non_empty_arglist
392 : non_empty_arglist ',' exp %prec ABOVE_COMMA
393 { arglist_len++; }
394 ;
395
396 /* GDB construct */
397 exp : '{' type '}' exp %prec UNARY
398 { write_exp_elt_opcode (UNOP_MEMVAL);
399 write_exp_elt_type ($2);
400 write_exp_elt_opcode (UNOP_MEMVAL); }
401 ;
402
403 exp : type '(' exp ')' %prec UNARY
404 { write_exp_elt_opcode (UNOP_CAST);
405 write_exp_elt_type ($1);
406 write_exp_elt_opcode (UNOP_CAST); }
407 ;
408
409 exp : '(' exp ')'
410 { }
411 ;
412
413 /* Binary operators in order of decreasing precedence. Note that some
414 of these operators are overloaded! (ie. sets) */
415
416 /* GDB construct */
417 exp : exp '@' exp
418 { write_exp_elt_opcode (BINOP_REPEAT); }
419 ;
420
421 exp : exp '*' exp
422 { write_exp_elt_opcode (BINOP_MUL); }
423 ;
424
425 exp : exp '/' exp
426 { write_exp_elt_opcode (BINOP_DIV); }
427 ;
428
429 exp : exp DIV exp
430 { write_exp_elt_opcode (BINOP_INTDIV); }
431 ;
432
433 exp : exp MOD exp
434 { write_exp_elt_opcode (BINOP_REM); }
435 ;
436
437 exp : exp '+' exp
438 { write_exp_elt_opcode (BINOP_ADD); }
439 ;
440
441 exp : exp '-' exp
442 { write_exp_elt_opcode (BINOP_SUB); }
443 ;
444
445 exp : exp '=' exp
446 { write_exp_elt_opcode (BINOP_EQUAL); }
447 ;
448
449 exp : exp NOTEQUAL exp
450 { write_exp_elt_opcode (BINOP_NOTEQUAL); }
451 | exp '#' exp
452 { write_exp_elt_opcode (BINOP_NOTEQUAL); }
453 ;
454
455 exp : exp LEQ exp
456 { write_exp_elt_opcode (BINOP_LEQ); }
457 ;
458
459 exp : exp GEQ exp
460 { write_exp_elt_opcode (BINOP_GEQ); }
461 ;
462
463 exp : exp '<' exp
464 { write_exp_elt_opcode (BINOP_LESS); }
465 ;
466
467 exp : exp '>' exp
468 { write_exp_elt_opcode (BINOP_GTR); }
469 ;
470
471 exp : exp LOGICAL_AND exp
472 { write_exp_elt_opcode (BINOP_LOGICAL_AND); }
473 ;
474
475 exp : exp OROR exp
476 { write_exp_elt_opcode (BINOP_LOGICAL_OR); }
477 ;
478
479 exp : exp ASSIGN exp
480 { write_exp_elt_opcode (BINOP_ASSIGN); }
481 ;
482
483
484 /* Constants */
485
486 exp : M2_TRUE
487 { write_exp_elt_opcode (OP_BOOL);
488 write_exp_elt_longcst ((LONGEST) $1);
489 write_exp_elt_opcode (OP_BOOL); }
490 ;
491
492 exp : M2_FALSE
493 { write_exp_elt_opcode (OP_BOOL);
494 write_exp_elt_longcst ((LONGEST) $1);
495 write_exp_elt_opcode (OP_BOOL); }
496 ;
497
498 exp : INT
499 { write_exp_elt_opcode (OP_LONG);
500 write_exp_elt_type (builtin_type_m2_int);
501 write_exp_elt_longcst ((LONGEST) $1);
502 write_exp_elt_opcode (OP_LONG); }
503 ;
504
505 exp : UINT
506 {
507 write_exp_elt_opcode (OP_LONG);
508 write_exp_elt_type (builtin_type_m2_card);
509 write_exp_elt_longcst ((LONGEST) $1);
510 write_exp_elt_opcode (OP_LONG);
511 }
512 ;
513
514 exp : CHAR
515 { write_exp_elt_opcode (OP_LONG);
516 write_exp_elt_type (builtin_type_m2_char);
517 write_exp_elt_longcst ((LONGEST) $1);
518 write_exp_elt_opcode (OP_LONG); }
519 ;
520
521
522 exp : FLOAT
523 { write_exp_elt_opcode (OP_DOUBLE);
524 write_exp_elt_type (builtin_type_m2_real);
525 write_exp_elt_dblcst ($1);
526 write_exp_elt_opcode (OP_DOUBLE); }
527 ;
528
529 exp : variable
530 ;
531
532 exp : SIZE '(' type ')' %prec UNARY
533 { write_exp_elt_opcode (OP_LONG);
534 write_exp_elt_type (builtin_type_int);
535 write_exp_elt_longcst ((LONGEST) TYPE_LENGTH ($3));
536 write_exp_elt_opcode (OP_LONG); }
537 ;
538
539 exp : STRING
540 { write_exp_elt_opcode (OP_M2_STRING);
541 write_exp_string ($1);
542 write_exp_elt_opcode (OP_M2_STRING); }
543 ;
544
545 /* This will be used for extensions later. Like adding modules. */
546 block : fblock
547 { $$ = SYMBOL_BLOCK_VALUE($1); }
548 ;
549
550 fblock : BLOCKNAME
551 { struct symbol *sym
552 = lookup_symbol (copy_name ($1), expression_context_block,
553 VAR_DOMAIN, 0);
554 $$ = sym;}
555 ;
556
557
558 /* GDB scope operator */
559 fblock : block COLONCOLON BLOCKNAME
560 { struct symbol *tem
561 = lookup_symbol (copy_name ($3), $1,
562 VAR_DOMAIN, 0);
563 if (!tem || SYMBOL_CLASS (tem) != LOC_BLOCK)
564 error ("No function \"%s\" in specified context.",
565 copy_name ($3));
566 $$ = tem;
567 }
568 ;
569
570 /* Useful for assigning to PROCEDURE variables */
571 variable: fblock
572 { write_exp_elt_opcode(OP_VAR_VALUE);
573 write_exp_elt_block (NULL);
574 write_exp_elt_sym ($1);
575 write_exp_elt_opcode (OP_VAR_VALUE); }
576 ;
577
578 /* GDB internal ($foo) variable */
579 variable: INTERNAL_VAR
580 ;
581
582 /* GDB scope operator */
583 variable: block COLONCOLON NAME
584 { struct symbol *sym;
585 sym = lookup_symbol (copy_name ($3), $1,
586 VAR_DOMAIN, 0);
587 if (sym == 0)
588 error ("No symbol \"%s\" in specified context.",
589 copy_name ($3));
590
591 write_exp_elt_opcode (OP_VAR_VALUE);
592 /* block_found is set by lookup_symbol. */
593 write_exp_elt_block (block_found);
594 write_exp_elt_sym (sym);
595 write_exp_elt_opcode (OP_VAR_VALUE); }
596 ;
597
598 /* Base case for variables. */
599 variable: NAME
600 { struct symbol *sym;
601 int is_a_field_of_this;
602
603 sym = lookup_symbol (copy_name ($1),
604 expression_context_block,
605 VAR_DOMAIN,
606 &is_a_field_of_this);
607 if (sym)
608 {
609 if (symbol_read_needs_frame (sym))
610 {
611 if (innermost_block == 0 ||
612 contained_in (block_found,
613 innermost_block))
614 innermost_block = block_found;
615 }
616
617 write_exp_elt_opcode (OP_VAR_VALUE);
618 /* We want to use the selected frame, not
619 another more inner frame which happens to
620 be in the same block. */
621 write_exp_elt_block (NULL);
622 write_exp_elt_sym (sym);
623 write_exp_elt_opcode (OP_VAR_VALUE);
624 }
625 else
626 {
627 struct minimal_symbol *msymbol;
628 char *arg = copy_name ($1);
629
630 msymbol =
631 lookup_minimal_symbol (arg, NULL, NULL);
632 if (msymbol != NULL)
633 write_exp_msymbol (msymbol);
634 else if (!have_full_symbols () && !have_partial_symbols ())
635 error ("No symbol table is loaded. Use the \"symbol-file\" command.");
636 else
637 error ("No symbol \"%s\" in current context.",
638 copy_name ($1));
639 }
640 }
641 ;
642
643 type
644 : TYPENAME
645 { $$ = lookup_typename (copy_name ($1),
646 expression_context_block, 0); }
647
648 ;
649
650 %%
651
652 #if 0 /* FIXME! */
653 int
654 overflow(a,b)
655 long a,b;
656 {
657 return (MAX_OF_TYPE(builtin_type_m2_int) - b) < a;
658 }
659
660 int
661 uoverflow(a,b)
662 unsigned long a,b;
663 {
664 return (MAX_OF_TYPE(builtin_type_m2_card) - b) < a;
665 }
666 #endif /* FIXME */
667
668 /* Take care of parsing a number (anything that starts with a digit).
669 Set yylval and return the token type; update lexptr.
670 LEN is the number of characters in it. */
671
672 /*** Needs some error checking for the float case ***/
673
674 static int
675 parse_number (olen)
676 int olen;
677 {
678 char *p = lexptr;
679 LONGEST n = 0;
680 LONGEST prevn = 0;
681 int c,i,ischar=0;
682 int base = input_radix;
683 int len = olen;
684 int unsigned_p = number_sign == 1 ? 1 : 0;
685
686 if(p[len-1] == 'H')
687 {
688 base = 16;
689 len--;
690 }
691 else if(p[len-1] == 'C' || p[len-1] == 'B')
692 {
693 base = 8;
694 ischar = p[len-1] == 'C';
695 len--;
696 }
697
698 /* Scan the number */
699 for (c = 0; c < len; c++)
700 {
701 if (p[c] == '.' && base == 10)
702 {
703 /* It's a float since it contains a point. */
704 yylval.dval = atof (p);
705 lexptr += len;
706 return FLOAT;
707 }
708 if (p[c] == '.' && base != 10)
709 error("Floating point numbers must be base 10.");
710 if (base == 10 && (p[c] < '0' || p[c] > '9'))
711 error("Invalid digit \'%c\' in number.",p[c]);
712 }
713
714 while (len-- > 0)
715 {
716 c = *p++;
717 n *= base;
718 if( base == 8 && (c == '8' || c == '9'))
719 error("Invalid digit \'%c\' in octal number.",c);
720 if (c >= '0' && c <= '9')
721 i = c - '0';
722 else
723 {
724 if (base == 16 && c >= 'A' && c <= 'F')
725 i = c - 'A' + 10;
726 else
727 return ERROR;
728 }
729 n+=i;
730 if(i >= base)
731 return ERROR;
732 if(!unsigned_p && number_sign == 1 && (prevn >= n))
733 unsigned_p=1; /* Try something unsigned */
734 /* Don't do the range check if n==i and i==0, since that special
735 case will give an overflow error. */
736 if(RANGE_CHECK && n!=i && i)
737 {
738 if((unsigned_p && (unsigned)prevn >= (unsigned)n) ||
739 ((!unsigned_p && number_sign==-1) && -prevn <= -n))
740 range_error("Overflow on numeric constant.");
741 }
742 prevn=n;
743 }
744
745 lexptr = p;
746 if(*p == 'B' || *p == 'C' || *p == 'H')
747 lexptr++; /* Advance past B,C or H */
748
749 if (ischar)
750 {
751 yylval.ulval = n;
752 return CHAR;
753 }
754 else if ( unsigned_p && number_sign == 1)
755 {
756 yylval.ulval = n;
757 return UINT;
758 }
759 else if((unsigned_p && (n<0))) {
760 range_error("Overflow on numeric constant -- number too large.");
761 /* But, this can return if range_check == range_warn. */
762 }
763 yylval.lval = n;
764 return INT;
765 }
766
767
768 /* Some tokens */
769
770 static struct
771 {
772 char name[2];
773 int token;
774 } tokentab2[] =
775 {
776 { {'<', '>'}, NOTEQUAL },
777 { {':', '='}, ASSIGN },
778 { {'<', '='}, LEQ },
779 { {'>', '='}, GEQ },
780 { {':', ':'}, COLONCOLON },
781
782 };
783
784 /* Some specific keywords */
785
786 struct keyword {
787 char keyw[10];
788 int token;
789 };
790
791 static struct keyword keytab[] =
792 {
793 {"OR" , OROR },
794 {"IN", IN },/* Note space after IN */
795 {"AND", LOGICAL_AND},
796 {"ABS", ABS },
797 {"CHR", CHR },
798 {"DEC", DEC },
799 {"NOT", NOT },
800 {"DIV", DIV },
801 {"INC", INC },
802 {"MAX", MAX_FUNC },
803 {"MIN", MIN_FUNC },
804 {"MOD", MOD },
805 {"ODD", ODD },
806 {"CAP", CAP },
807 {"ORD", ORD },
808 {"VAL", VAL },
809 {"EXCL", EXCL },
810 {"HIGH", HIGH },
811 {"INCL", INCL },
812 {"SIZE", SIZE },
813 {"FLOAT", FLOAT_FUNC },
814 {"TRUNC", TRUNC },
815 {"TSIZE", SIZE },
816 };
817
818
819 /* Read one token, getting characters through lexptr. */
820
821 /* This is where we will check to make sure that the language and the operators used are
822 compatible */
823
824 static int
825 yylex ()
826 {
827 int c;
828 int namelen;
829 int i;
830 char *tokstart;
831 char quote;
832
833 retry:
834
835 prev_lexptr = lexptr;
836
837 tokstart = lexptr;
838
839
840 /* See if it is a special token of length 2 */
841 for( i = 0 ; i < (int) (sizeof tokentab2 / sizeof tokentab2[0]) ; i++)
842 if (strncmp (tokentab2[i].name, tokstart, 2) == 0)
843 {
844 lexptr += 2;
845 return tokentab2[i].token;
846 }
847
848 switch (c = *tokstart)
849 {
850 case 0:
851 return 0;
852
853 case ' ':
854 case '\t':
855 case '\n':
856 lexptr++;
857 goto retry;
858
859 case '(':
860 paren_depth++;
861 lexptr++;
862 return c;
863
864 case ')':
865 if (paren_depth == 0)
866 return 0;
867 paren_depth--;
868 lexptr++;
869 return c;
870
871 case ',':
872 if (comma_terminates && paren_depth == 0)
873 return 0;
874 lexptr++;
875 return c;
876
877 case '.':
878 /* Might be a floating point number. */
879 if (lexptr[1] >= '0' && lexptr[1] <= '9')
880 break; /* Falls into number code. */
881 else
882 {
883 lexptr++;
884 return DOT;
885 }
886
887 /* These are character tokens that appear as-is in the YACC grammar */
888 case '+':
889 case '-':
890 case '*':
891 case '/':
892 case '^':
893 case '<':
894 case '>':
895 case '[':
896 case ']':
897 case '=':
898 case '{':
899 case '}':
900 case '#':
901 case '@':
902 case '~':
903 case '&':
904 lexptr++;
905 return c;
906
907 case '\'' :
908 case '"':
909 quote = c;
910 for (namelen = 1; (c = tokstart[namelen]) != quote && c != '\0'; namelen++)
911 if (c == '\\')
912 {
913 c = tokstart[++namelen];
914 if (c >= '0' && c <= '9')
915 {
916 c = tokstart[++namelen];
917 if (c >= '0' && c <= '9')
918 c = tokstart[++namelen];
919 }
920 }
921 if(c != quote)
922 error("Unterminated string or character constant.");
923 yylval.sval.ptr = tokstart + 1;
924 yylval.sval.length = namelen - 1;
925 lexptr += namelen + 1;
926
927 if(namelen == 2) /* Single character */
928 {
929 yylval.ulval = tokstart[1];
930 return CHAR;
931 }
932 else
933 return STRING;
934 }
935
936 /* Is it a number? */
937 /* Note: We have already dealt with the case of the token '.'.
938 See case '.' above. */
939 if ((c >= '0' && c <= '9'))
940 {
941 /* It's a number. */
942 int got_dot = 0, got_e = 0;
943 char *p = tokstart;
944 int toktype;
945
946 for (++p ;; ++p)
947 {
948 if (!got_e && (*p == 'e' || *p == 'E'))
949 got_dot = got_e = 1;
950 else if (!got_dot && *p == '.')
951 got_dot = 1;
952 else if (got_e && (p[-1] == 'e' || p[-1] == 'E')
953 && (*p == '-' || *p == '+'))
954 /* This is the sign of the exponent, not the end of the
955 number. */
956 continue;
957 else if ((*p < '0' || *p > '9') &&
958 (*p < 'A' || *p > 'F') &&
959 (*p != 'H')) /* Modula-2 hexadecimal number */
960 break;
961 }
962 toktype = parse_number (p - tokstart);
963 if (toktype == ERROR)
964 {
965 char *err_copy = (char *) alloca (p - tokstart + 1);
966
967 memcpy (err_copy, tokstart, p - tokstart);
968 err_copy[p - tokstart] = 0;
969 error ("Invalid number \"%s\".", err_copy);
970 }
971 lexptr = p;
972 return toktype;
973 }
974
975 if (!(c == '_' || c == '$'
976 || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')))
977 /* We must have come across a bad character (e.g. ';'). */
978 error ("Invalid character '%c' in expression.", c);
979
980 /* It's a name. See how long it is. */
981 namelen = 0;
982 for (c = tokstart[namelen];
983 (c == '_' || c == '$' || (c >= '0' && c <= '9')
984 || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z'));
985 c = tokstart[++namelen])
986 ;
987
988 /* The token "if" terminates the expression and is NOT
989 removed from the input stream. */
990 if (namelen == 2 && tokstart[0] == 'i' && tokstart[1] == 'f')
991 {
992 return 0;
993 }
994
995 lexptr += namelen;
996
997 /* Lookup special keywords */
998 for(i = 0 ; i < (int) (sizeof(keytab) / sizeof(keytab[0])) ; i++)
999 if (namelen == strlen (keytab[i].keyw)
1000 && strncmp (tokstart, keytab[i].keyw, namelen) == 0)
1001 return keytab[i].token;
1002
1003 yylval.sval.ptr = tokstart;
1004 yylval.sval.length = namelen;
1005
1006 if (*tokstart == '$')
1007 {
1008 write_dollar_variable (yylval.sval);
1009 return INTERNAL_VAR;
1010 }
1011
1012 /* Use token-type BLOCKNAME for symbols that happen to be defined as
1013 functions. If this is not so, then ...
1014 Use token-type TYPENAME for symbols that happen to be defined
1015 currently as names of types; NAME for other symbols.
1016 The caller is not constrained to care about the distinction. */
1017 {
1018
1019
1020 char *tmp = copy_name (yylval.sval);
1021 struct symbol *sym;
1022
1023 if (lookup_partial_symtab (tmp))
1024 return BLOCKNAME;
1025 sym = lookup_symbol (tmp, expression_context_block, VAR_DOMAIN, 0);
1026 if (sym && SYMBOL_CLASS (sym) == LOC_BLOCK)
1027 return BLOCKNAME;
1028 if (lookup_typename (copy_name (yylval.sval), expression_context_block, 1))
1029 return TYPENAME;
1030
1031 if(sym)
1032 {
1033 switch(sym->aclass)
1034 {
1035 case LOC_STATIC:
1036 case LOC_REGISTER:
1037 case LOC_ARG:
1038 case LOC_REF_ARG:
1039 case LOC_REGPARM_ADDR:
1040 case LOC_LOCAL:
1041 case LOC_CONST:
1042 case LOC_CONST_BYTES:
1043 case LOC_OPTIMIZED_OUT:
1044 case LOC_COMPUTED:
1045 return NAME;
1046
1047 case LOC_TYPEDEF:
1048 return TYPENAME;
1049
1050 case LOC_BLOCK:
1051 return BLOCKNAME;
1052
1053 case LOC_UNDEF:
1054 error("internal: Undefined class in m2lex()");
1055
1056 case LOC_LABEL:
1057 case LOC_UNRESOLVED:
1058 error("internal: Unforseen case in m2lex()");
1059
1060 default:
1061 error ("unhandled token in m2lex()");
1062 break;
1063 }
1064 }
1065 else
1066 {
1067 /* Built-in BOOLEAN type. This is sort of a hack. */
1068 if (strncmp (tokstart, "TRUE", 4) == 0)
1069 {
1070 yylval.ulval = 1;
1071 return M2_TRUE;
1072 }
1073 else if (strncmp (tokstart, "FALSE", 5) == 0)
1074 {
1075 yylval.ulval = 0;
1076 return M2_FALSE;
1077 }
1078 }
1079
1080 /* Must be another type of name... */
1081 return NAME;
1082 }
1083 }
1084
1085 #if 0 /* Unused */
1086 static char *
1087 make_qualname(mod,ident)
1088 char *mod, *ident;
1089 {
1090 char *new = malloc(strlen(mod)+strlen(ident)+2);
1091
1092 strcpy(new,mod);
1093 strcat(new,".");
1094 strcat(new,ident);
1095 return new;
1096 }
1097 #endif /* 0 */
1098
1099 void
1100 yyerror (msg)
1101 char *msg;
1102 {
1103 if (prev_lexptr)
1104 lexptr = prev_lexptr;
1105
1106 error ("A %s in expression, near `%s'.", (msg ? msg : "error"), lexptr);
1107 }
This page took 0.056341 seconds and 5 git commands to generate.