* solib-svr4.c (LM_ADDR_FROM_LINK_MAP): Use builtin types of
[deliverable/binutils-gdb.git] / gdb / f-exp.y
CommitLineData
c906108c 1/* YACC parser for Fortran expressions, for GDB.
197e01b6 2 Copyright (C) 1986, 1989, 1990, 1991, 1993, 1994, 1995, 1996, 2000, 2001,
9b254dd1 3 2002, 2003, 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
4fcf66da 4
c906108c
SS
5 Contributed by Motorola. Adapted from the C parser by Farooq Butt
6 (fmbutt@engage.sps.mot.com).
7
8This file is part of GDB.
9
10This program is free software; you can redistribute it and/or modify
11it under the terms of the GNU General Public License as published by
12the Free Software Foundation; either version 2 of the License, or
13(at your option) any later version.
14
15This program is distributed in the hope that it will be useful,
16but WITHOUT ANY WARRANTY; without even the implied warranty of
17MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18GNU General Public License for more details.
19
20You should have received a copy of the GNU General Public License
21along with this program; if not, write to the Free Software
197e01b6
EZ
22Foundation, Inc., 51 Franklin Street, Fifth Floor,
23Boston, MA 02110-1301, USA. */
c906108c
SS
24
25/* This was blantantly ripped off the C expression parser, please
26 be aware of that as you look at its basic structure -FMB */
27
28/* Parse a F77 expression from text in a string,
29 and return the result as a struct expression pointer.
30 That structure contains arithmetic operations in reverse polish,
31 with constants represented by operations that are followed by special data.
32 See expression.h for the details of the format.
33 What is important here is that it can be built up sequentially
34 during the process of parsing; the lower levels of the tree always
35 come first in the result.
36
37 Note that malloc's and realloc's in this file are transformed to
38 xmalloc and xrealloc respectively by the same sed command in the
39 makefile that remaps any other malloc/realloc inserted by the parser
40 generator. Doing this with #defines and trying to control the interaction
41 with include files (<malloc.h> and <stdlib.h> for example) just became
42 too messy, particularly when such includes can be inserted at random
43 times by the parser generator. */
44
45%{
46
47#include "defs.h"
48#include "gdb_string.h"
49#include "expression.h"
50#include "value.h"
51#include "parser-defs.h"
52#include "language.h"
53#include "f-lang.h"
54#include "bfd.h" /* Required by objfiles.h. */
55#include "symfile.h" /* Required by objfiles.h. */
56#include "objfiles.h" /* For have_full_symbols and have_partial_symbols */
fe898f56 57#include "block.h"
0f6e1ba6 58#include <ctype.h>
c906108c 59
3e79cecf
UW
60#define parse_type builtin_type (parse_gdbarch)
61#define parse_f_type builtin_f_type (parse_gdbarch)
62
c906108c
SS
63/* Remap normal yacc parser interface names (yyparse, yylex, yyerror, etc),
64 as well as gratuitiously global symbol names, so we can have multiple
65 yacc generated parsers in gdb. Note that these are only the variables
66 produced by yacc. If other parser generators (bison, byacc, etc) produce
67 additional global names that conflict at link time, then those parser
68 generators need to be fixed instead of adding those names to this list. */
69
70#define yymaxdepth f_maxdepth
71#define yyparse f_parse
72#define yylex f_lex
73#define yyerror f_error
74#define yylval f_lval
75#define yychar f_char
76#define yydebug f_debug
77#define yypact f_pact
78#define yyr1 f_r1
79#define yyr2 f_r2
80#define yydef f_def
81#define yychk f_chk
82#define yypgo f_pgo
83#define yyact f_act
84#define yyexca f_exca
85#define yyerrflag f_errflag
86#define yynerrs f_nerrs
87#define yyps f_ps
88#define yypv f_pv
89#define yys f_s
90#define yy_yys f_yys
91#define yystate f_state
92#define yytmp f_tmp
93#define yyv f_v
94#define yy_yyv f_yyv
95#define yyval f_val
96#define yylloc f_lloc
97#define yyreds f_reds /* With YYDEBUG defined */
98#define yytoks f_toks /* With YYDEBUG defined */
06891d83
JT
99#define yyname f_name /* With YYDEBUG defined */
100#define yyrule f_rule /* With YYDEBUG defined */
c906108c
SS
101#define yylhs f_yylhs
102#define yylen f_yylen
103#define yydefred f_yydefred
104#define yydgoto f_yydgoto
105#define yysindex f_yysindex
106#define yyrindex f_yyrindex
107#define yygindex f_yygindex
108#define yytable f_yytable
109#define yycheck f_yycheck
110
111#ifndef YYDEBUG
f461f5cf 112#define YYDEBUG 1 /* Default to yydebug support */
c906108c
SS
113#endif
114
f461f5cf
PM
115#define YYFPRINTF parser_fprintf
116
a14ed312 117int yyparse (void);
c906108c 118
a14ed312 119static int yylex (void);
c906108c 120
a14ed312 121void yyerror (char *);
c906108c 122
a14ed312 123static void growbuf_by_size (int);
c906108c 124
a14ed312 125static int match_string_literal (void);
c906108c
SS
126
127%}
128
129/* Although the yacc "value" of an expression is not used,
130 since the result is stored in the structure being created,
131 other node types do have values. */
132
133%union
134 {
135 LONGEST lval;
136 struct {
137 LONGEST val;
138 struct type *type;
139 } typed_val;
140 DOUBLEST dval;
141 struct symbol *sym;
142 struct type *tval;
143 struct stoken sval;
144 struct ttype tsym;
145 struct symtoken ssym;
146 int voidval;
147 struct block *bval;
148 enum exp_opcode opcode;
149 struct internalvar *ivar;
150
151 struct type **tvec;
152 int *ivec;
153 }
154
155%{
156/* YYSTYPE gets defined by %union */
a14ed312 157static int parse_number (char *, int, int, YYSTYPE *);
c906108c
SS
158%}
159
160%type <voidval> exp type_exp start variable
161%type <tval> type typebase
162%type <tvec> nonempty_typelist
163/* %type <bval> block */
164
165/* Fancy type parsing. */
166%type <voidval> func_mod direct_abs_decl abs_decl
167%type <tval> ptype
168
169%token <typed_val> INT
170%token <dval> FLOAT
171
172/* Both NAME and TYPENAME tokens represent symbols in the input,
173 and both convey their data as strings.
174 But a TYPENAME is a string that happens to be defined as a typedef
175 or builtin type name (such as int or char)
176 and a NAME is any other symbol.
177 Contexts where this distinction is not important can use the
178 nonterminal "name", which matches either NAME or TYPENAME. */
179
180%token <sval> STRING_LITERAL
181%token <lval> BOOLEAN_LITERAL
182%token <ssym> NAME
183%token <tsym> TYPENAME
2a5e440c 184%type <sval> name
c906108c 185%type <ssym> name_not_typename
c906108c
SS
186
187/* A NAME_OR_INT is a symbol which is not known in the symbol table,
188 but which would parse as a valid number in the current input radix.
189 E.g. "c" when input_radix==16. Depending on the parse, it will be
190 turned into a name or into a number. */
191
192%token <ssym> NAME_OR_INT
193
194%token SIZEOF
195%token ERROR
196
197/* Special type cases, put in to allow the parser to distinguish different
198 legal basetypes. */
199%token INT_KEYWORD INT_S2_KEYWORD LOGICAL_S1_KEYWORD LOGICAL_S2_KEYWORD
200%token LOGICAL_KEYWORD REAL_KEYWORD REAL_S8_KEYWORD REAL_S16_KEYWORD
201%token COMPLEX_S8_KEYWORD COMPLEX_S16_KEYWORD COMPLEX_S32_KEYWORD
202%token BOOL_AND BOOL_OR BOOL_NOT
203%token <lval> CHARACTER
204
205%token <voidval> VARIABLE
206
207%token <opcode> ASSIGN_MODIFY
208
209%left ','
210%left ABOVE_COMMA
211%right '=' ASSIGN_MODIFY
212%right '?'
213%left BOOL_OR
214%right BOOL_NOT
215%left BOOL_AND
216%left '|'
217%left '^'
218%left '&'
219%left EQUAL NOTEQUAL
220%left LESSTHAN GREATERTHAN LEQ GEQ
221%left LSH RSH
222%left '@'
223%left '+' '-'
2a5e440c 224%left '*' '/'
bd49c137 225%right STARSTAR
2a5e440c 226%right '%'
c906108c
SS
227%right UNARY
228%right '('
229
230\f
231%%
232
233start : exp
234 | type_exp
235 ;
236
237type_exp: type
238 { write_exp_elt_opcode(OP_TYPE);
239 write_exp_elt_type($1);
240 write_exp_elt_opcode(OP_TYPE); }
241 ;
242
243exp : '(' exp ')'
244 { }
245 ;
246
247/* Expressions, not including the comma operator. */
248exp : '*' exp %prec UNARY
249 { write_exp_elt_opcode (UNOP_IND); }
ef944135 250 ;
c906108c
SS
251
252exp : '&' exp %prec UNARY
253 { write_exp_elt_opcode (UNOP_ADDR); }
ef944135 254 ;
c906108c
SS
255
256exp : '-' exp %prec UNARY
257 { write_exp_elt_opcode (UNOP_NEG); }
258 ;
259
260exp : BOOL_NOT exp %prec UNARY
261 { write_exp_elt_opcode (UNOP_LOGICAL_NOT); }
262 ;
263
264exp : '~' exp %prec UNARY
265 { write_exp_elt_opcode (UNOP_COMPLEMENT); }
266 ;
267
268exp : SIZEOF exp %prec UNARY
269 { write_exp_elt_opcode (UNOP_SIZEOF); }
270 ;
271
272/* No more explicit array operators, we treat everything in F77 as
273 a function call. The disambiguation as to whether we are
274 doing a subscript operation or a function call is done
275 later in eval.c. */
276
277exp : exp '('
278 { start_arglist (); }
279 arglist ')'
280 { write_exp_elt_opcode (OP_F77_UNDETERMINED_ARGLIST);
281 write_exp_elt_longcst ((LONGEST) end_arglist ());
282 write_exp_elt_opcode (OP_F77_UNDETERMINED_ARGLIST); }
283 ;
284
285arglist :
286 ;
287
288arglist : exp
289 { arglist_len = 1; }
290 ;
291
0b4e1325
WZ
292arglist : subrange
293 { arglist_len = 1; }
ef944135 294 ;
c906108c
SS
295
296arglist : arglist ',' exp %prec ABOVE_COMMA
297 { arglist_len++; }
298 ;
299
0b4e1325
WZ
300/* There are four sorts of subrange types in F90. */
301
302subrange: exp ':' exp %prec ABOVE_COMMA
303 { write_exp_elt_opcode (OP_F90_RANGE);
304 write_exp_elt_longcst (NONE_BOUND_DEFAULT);
305 write_exp_elt_opcode (OP_F90_RANGE); }
306 ;
307
308subrange: exp ':' %prec ABOVE_COMMA
309 { write_exp_elt_opcode (OP_F90_RANGE);
310 write_exp_elt_longcst (HIGH_BOUND_DEFAULT);
311 write_exp_elt_opcode (OP_F90_RANGE); }
c906108c
SS
312 ;
313
0b4e1325
WZ
314subrange: ':' exp %prec ABOVE_COMMA
315 { write_exp_elt_opcode (OP_F90_RANGE);
316 write_exp_elt_longcst (LOW_BOUND_DEFAULT);
317 write_exp_elt_opcode (OP_F90_RANGE); }
318 ;
319
320subrange: ':' %prec ABOVE_COMMA
321 { write_exp_elt_opcode (OP_F90_RANGE);
322 write_exp_elt_longcst (BOTH_BOUND_DEFAULT);
323 write_exp_elt_opcode (OP_F90_RANGE); }
324 ;
c906108c
SS
325
326complexnum: exp ',' exp
327 { }
328 ;
329
330exp : '(' complexnum ')'
c806c55a
UW
331 { write_exp_elt_opcode(OP_COMPLEX);
332 write_exp_elt_type (parse_f_type->builtin_complex_s16);
333 write_exp_elt_opcode(OP_COMPLEX); }
c906108c
SS
334 ;
335
336exp : '(' type ')' exp %prec UNARY
337 { write_exp_elt_opcode (UNOP_CAST);
338 write_exp_elt_type ($2);
339 write_exp_elt_opcode (UNOP_CAST); }
340 ;
341
2a5e440c
WZ
342exp : exp '%' name
343 { write_exp_elt_opcode (STRUCTOP_STRUCT);
344 write_exp_string ($3);
345 write_exp_elt_opcode (STRUCTOP_STRUCT); }
346 ;
347
c906108c
SS
348/* Binary operators in order of decreasing precedence. */
349
350exp : exp '@' exp
351 { write_exp_elt_opcode (BINOP_REPEAT); }
352 ;
353
bd49c137
WZ
354exp : exp STARSTAR exp
355 { write_exp_elt_opcode (BINOP_EXP); }
356 ;
357
c906108c
SS
358exp : exp '*' exp
359 { write_exp_elt_opcode (BINOP_MUL); }
360 ;
361
362exp : exp '/' exp
363 { write_exp_elt_opcode (BINOP_DIV); }
364 ;
365
c906108c
SS
366exp : exp '+' exp
367 { write_exp_elt_opcode (BINOP_ADD); }
368 ;
369
370exp : exp '-' exp
371 { write_exp_elt_opcode (BINOP_SUB); }
372 ;
373
374exp : exp LSH exp
375 { write_exp_elt_opcode (BINOP_LSH); }
376 ;
377
378exp : exp RSH exp
379 { write_exp_elt_opcode (BINOP_RSH); }
380 ;
381
382exp : exp EQUAL exp
383 { write_exp_elt_opcode (BINOP_EQUAL); }
384 ;
385
386exp : exp NOTEQUAL exp
387 { write_exp_elt_opcode (BINOP_NOTEQUAL); }
388 ;
389
390exp : exp LEQ exp
391 { write_exp_elt_opcode (BINOP_LEQ); }
392 ;
393
394exp : exp GEQ exp
395 { write_exp_elt_opcode (BINOP_GEQ); }
396 ;
397
398exp : exp LESSTHAN exp
399 { write_exp_elt_opcode (BINOP_LESS); }
400 ;
401
402exp : exp GREATERTHAN exp
403 { write_exp_elt_opcode (BINOP_GTR); }
404 ;
405
406exp : exp '&' exp
407 { write_exp_elt_opcode (BINOP_BITWISE_AND); }
408 ;
409
410exp : exp '^' exp
411 { write_exp_elt_opcode (BINOP_BITWISE_XOR); }
412 ;
413
414exp : exp '|' exp
415 { write_exp_elt_opcode (BINOP_BITWISE_IOR); }
416 ;
417
418exp : exp BOOL_AND exp
419 { write_exp_elt_opcode (BINOP_LOGICAL_AND); }
420 ;
421
422
423exp : exp BOOL_OR exp
424 { write_exp_elt_opcode (BINOP_LOGICAL_OR); }
425 ;
426
427exp : exp '=' exp
428 { write_exp_elt_opcode (BINOP_ASSIGN); }
429 ;
430
431exp : exp ASSIGN_MODIFY exp
432 { write_exp_elt_opcode (BINOP_ASSIGN_MODIFY);
433 write_exp_elt_opcode ($2);
434 write_exp_elt_opcode (BINOP_ASSIGN_MODIFY); }
435 ;
436
437exp : INT
438 { write_exp_elt_opcode (OP_LONG);
439 write_exp_elt_type ($1.type);
440 write_exp_elt_longcst ((LONGEST)($1.val));
441 write_exp_elt_opcode (OP_LONG); }
442 ;
443
444exp : NAME_OR_INT
445 { YYSTYPE val;
446 parse_number ($1.stoken.ptr, $1.stoken.length, 0, &val);
447 write_exp_elt_opcode (OP_LONG);
448 write_exp_elt_type (val.typed_val.type);
449 write_exp_elt_longcst ((LONGEST)val.typed_val.val);
450 write_exp_elt_opcode (OP_LONG); }
451 ;
452
453exp : FLOAT
454 { write_exp_elt_opcode (OP_DOUBLE);
3e79cecf 455 write_exp_elt_type (parse_f_type->builtin_real_s8);
c906108c
SS
456 write_exp_elt_dblcst ($1);
457 write_exp_elt_opcode (OP_DOUBLE); }
458 ;
459
460exp : variable
461 ;
462
463exp : VARIABLE
464 ;
465
466exp : SIZEOF '(' type ')' %prec UNARY
467 { write_exp_elt_opcode (OP_LONG);
3e79cecf 468 write_exp_elt_type (parse_f_type->builtin_integer);
c906108c
SS
469 CHECK_TYPEDEF ($3);
470 write_exp_elt_longcst ((LONGEST) TYPE_LENGTH ($3));
471 write_exp_elt_opcode (OP_LONG); }
472 ;
473
474exp : BOOLEAN_LITERAL
475 { write_exp_elt_opcode (OP_BOOL);
476 write_exp_elt_longcst ((LONGEST) $1);
477 write_exp_elt_opcode (OP_BOOL);
478 }
479 ;
480
481exp : STRING_LITERAL
482 {
483 write_exp_elt_opcode (OP_STRING);
484 write_exp_string ($1);
485 write_exp_elt_opcode (OP_STRING);
486 }
487 ;
488
489variable: name_not_typename
490 { struct symbol *sym = $1.sym;
491
492 if (sym)
493 {
494 if (symbol_read_needs_frame (sym))
495 {
496 if (innermost_block == 0 ||
497 contained_in (block_found,
498 innermost_block))
499 innermost_block = block_found;
500 }
501 write_exp_elt_opcode (OP_VAR_VALUE);
502 /* We want to use the selected frame, not
503 another more inner frame which happens to
504 be in the same block. */
505 write_exp_elt_block (NULL);
506 write_exp_elt_sym (sym);
507 write_exp_elt_opcode (OP_VAR_VALUE);
508 break;
509 }
510 else
511 {
512 struct minimal_symbol *msymbol;
710122da 513 char *arg = copy_name ($1.stoken);
c906108c
SS
514
515 msymbol =
516 lookup_minimal_symbol (arg, NULL, NULL);
517 if (msymbol != NULL)
c841afd5 518 write_exp_msymbol (msymbol);
c906108c
SS
519 else if (!have_full_symbols () && !have_partial_symbols ())
520 error ("No symbol table is loaded. Use the \"file\" command.");
521 else
522 error ("No symbol \"%s\" in current context.",
523 copy_name ($1.stoken));
524 }
525 }
526 ;
527
528
529type : ptype
530 ;
531
532ptype : typebase
533 | typebase abs_decl
534 {
535 /* This is where the interesting stuff happens. */
536 int done = 0;
537 int array_size;
538 struct type *follow_type = $1;
539 struct type *range_type;
540
541 while (!done)
542 switch (pop_type ())
543 {
544 case tp_end:
545 done = 1;
546 break;
547 case tp_pointer:
548 follow_type = lookup_pointer_type (follow_type);
549 break;
550 case tp_reference:
551 follow_type = lookup_reference_type (follow_type);
552 break;
553 case tp_array:
554 array_size = pop_type_int ();
555 if (array_size != -1)
556 {
557 range_type =
558 create_range_type ((struct type *) NULL,
3e79cecf
UW
559 parse_f_type->builtin_integer,
560 0, array_size - 1);
c906108c
SS
561 follow_type =
562 create_array_type ((struct type *) NULL,
563 follow_type, range_type);
564 }
565 else
566 follow_type = lookup_pointer_type (follow_type);
567 break;
568 case tp_function:
569 follow_type = lookup_function_type (follow_type);
570 break;
571 }
572 $$ = follow_type;
573 }
574 ;
575
576abs_decl: '*'
577 { push_type (tp_pointer); $$ = 0; }
578 | '*' abs_decl
579 { push_type (tp_pointer); $$ = $2; }
580 | '&'
581 { push_type (tp_reference); $$ = 0; }
582 | '&' abs_decl
583 { push_type (tp_reference); $$ = $2; }
584 | direct_abs_decl
585 ;
586
587direct_abs_decl: '(' abs_decl ')'
588 { $$ = $2; }
589 | direct_abs_decl func_mod
590 { push_type (tp_function); }
591 | func_mod
592 { push_type (tp_function); }
593 ;
594
595func_mod: '(' ')'
596 { $$ = 0; }
597 | '(' nonempty_typelist ')'
8dbb1c65 598 { free ($2); $$ = 0; }
c906108c
SS
599 ;
600
601typebase /* Implements (approximately): (type-qualifier)* type-specifier */
602 : TYPENAME
603 { $$ = $1.type; }
604 | INT_KEYWORD
3e79cecf 605 { $$ = parse_f_type->builtin_integer; }
c906108c 606 | INT_S2_KEYWORD
3e79cecf 607 { $$ = parse_f_type->builtin_integer_s2; }
c906108c 608 | CHARACTER
3e79cecf 609 { $$ = parse_f_type->builtin_character; }
c906108c 610 | LOGICAL_KEYWORD
3e79cecf 611 { $$ = parse_f_type->builtin_logical; }
c906108c 612 | LOGICAL_S2_KEYWORD
3e79cecf 613 { $$ = parse_f_type->builtin_logical_s2; }
c906108c 614 | LOGICAL_S1_KEYWORD
3e79cecf 615 { $$ = parse_f_type->builtin_logical_s1; }
c906108c 616 | REAL_KEYWORD
3e79cecf 617 { $$ = parse_f_type->builtin_real; }
c906108c 618 | REAL_S8_KEYWORD
3e79cecf 619 { $$ = parse_f_type->builtin_real_s8; }
c906108c 620 | REAL_S16_KEYWORD
3e79cecf 621 { $$ = parse_f_type->builtin_real_s16; }
c906108c 622 | COMPLEX_S8_KEYWORD
3e79cecf 623 { $$ = parse_f_type->builtin_complex_s8; }
c906108c 624 | COMPLEX_S16_KEYWORD
3e79cecf 625 { $$ = parse_f_type->builtin_complex_s16; }
c906108c 626 | COMPLEX_S32_KEYWORD
3e79cecf 627 { $$ = parse_f_type->builtin_complex_s32; }
c906108c
SS
628 ;
629
c906108c
SS
630nonempty_typelist
631 : type
632 { $$ = (struct type **) malloc (sizeof (struct type *) * 2);
633 $<ivec>$[0] = 1; /* Number of types in vector */
634 $$[1] = $1;
635 }
636 | nonempty_typelist ',' type
637 { int len = sizeof (struct type *) * (++($<ivec>1[0]) + 1);
638 $$ = (struct type **) realloc ((char *) $1, len);
639 $$[$<ivec>$[0]] = $3;
640 }
641 ;
642
2a5e440c
WZ
643name : NAME
644 { $$ = $1.stoken; }
645 ;
646
c906108c
SS
647name_not_typename : NAME
648/* These would be useful if name_not_typename was useful, but it is just
649 a fake for "variable", so these cause reduce/reduce conflicts because
650 the parser can't tell whether NAME_OR_INT is a name_not_typename (=variable,
651 =exp) or just an exp. If name_not_typename was ever used in an lvalue
652 context where only a name could occur, this might be useful.
653 | NAME_OR_INT
654 */
655 ;
656
657%%
658
659/* Take care of parsing a number (anything that starts with a digit).
660 Set yylval and return the token type; update lexptr.
661 LEN is the number of characters in it. */
662
663/*** Needs some error checking for the float case ***/
664
665static int
666parse_number (p, len, parsed_float, putithere)
710122da
DC
667 char *p;
668 int len;
c906108c
SS
669 int parsed_float;
670 YYSTYPE *putithere;
671{
710122da
DC
672 LONGEST n = 0;
673 LONGEST prevn = 0;
674 int c;
675 int base = input_radix;
c906108c
SS
676 int unsigned_p = 0;
677 int long_p = 0;
678 ULONGEST high_bit;
679 struct type *signed_type;
680 struct type *unsigned_type;
681
682 if (parsed_float)
683 {
684 /* It's a float since it contains a point or an exponent. */
685 /* [dD] is not understood as an exponent by atof, change it to 'e'. */
686 char *tmp, *tmp2;
687
4fcf66da 688 tmp = xstrdup (p);
c906108c
SS
689 for (tmp2 = tmp; *tmp2; ++tmp2)
690 if (*tmp2 == 'd' || *tmp2 == 'D')
691 *tmp2 = 'e';
692 putithere->dval = atof (tmp);
693 free (tmp);
694 return FLOAT;
695 }
696
697 /* Handle base-switching prefixes 0x, 0t, 0d, 0 */
698 if (p[0] == '0')
699 switch (p[1])
700 {
701 case 'x':
702 case 'X':
703 if (len >= 3)
704 {
705 p += 2;
706 base = 16;
707 len -= 2;
708 }
709 break;
710
711 case 't':
712 case 'T':
713 case 'd':
714 case 'D':
715 if (len >= 3)
716 {
717 p += 2;
718 base = 10;
719 len -= 2;
720 }
721 break;
722
723 default:
724 base = 8;
725 break;
726 }
727
728 while (len-- > 0)
729 {
730 c = *p++;
0f6e1ba6
AC
731 if (isupper (c))
732 c = tolower (c);
733 if (len == 0 && c == 'l')
734 long_p = 1;
735 else if (len == 0 && c == 'u')
736 unsigned_p = 1;
c906108c
SS
737 else
738 {
0f6e1ba6
AC
739 int i;
740 if (c >= '0' && c <= '9')
741 i = c - '0';
742 else if (c >= 'a' && c <= 'f')
743 i = c - 'a' + 10;
c906108c
SS
744 else
745 return ERROR; /* Char not a digit */
0f6e1ba6
AC
746 if (i >= base)
747 return ERROR; /* Invalid digit in this base */
748 n *= base;
749 n += i;
c906108c 750 }
c906108c
SS
751 /* Portably test for overflow (only works for nonzero values, so make
752 a second check for zero). */
753 if ((prevn >= n) && n != 0)
754 unsigned_p=1; /* Try something unsigned */
755 /* If range checking enabled, portably test for unsigned overflow. */
756 if (RANGE_CHECK && n != 0)
757 {
758 if ((unsigned_p && (unsigned)prevn >= (unsigned)n))
759 range_error("Overflow on numeric constant.");
760 }
761 prevn = n;
762 }
763
764 /* If the number is too big to be an int, or it's got an l suffix
765 then it's a long. Work out if this has to be a long by
766 shifting right and and seeing if anything remains, and the
767 target int size is different to the target long size.
768
769 In the expression below, we could have tested
3e79cecf 770 (n >> gdbarch_int_bit (parse_gdbarch))
c906108c
SS
771 to see if it was zero,
772 but too many compilers warn about that, when ints and longs
773 are the same size. So we shift it twice, with fewer bits
774 each time, for the same result. */
775
3e79cecf 776 if ((gdbarch_int_bit (parse_gdbarch) != gdbarch_long_bit (parse_gdbarch)
9a76efb6 777 && ((n >> 2)
3e79cecf 778 >> (gdbarch_int_bit (parse_gdbarch)-2))) /* Avoid shift warning */
c906108c
SS
779 || long_p)
780 {
3e79cecf
UW
781 high_bit = ((ULONGEST)1) << (gdbarch_long_bit (parse_gdbarch)-1);
782 unsigned_type = parse_type->builtin_unsigned_long;
783 signed_type = parse_type->builtin_long;
c906108c
SS
784 }
785 else
786 {
3e79cecf
UW
787 high_bit = ((ULONGEST)1) << (gdbarch_int_bit (parse_gdbarch)-1);
788 unsigned_type = parse_type->builtin_unsigned_int;
789 signed_type = parse_type->builtin_int;
c906108c
SS
790 }
791
792 putithere->typed_val.val = n;
793
794 /* If the high bit of the worked out type is set then this number
795 has to be unsigned. */
796
797 if (unsigned_p || (n & high_bit))
798 putithere->typed_val.type = unsigned_type;
799 else
800 putithere->typed_val.type = signed_type;
801
802 return INT;
803}
804
805struct token
806{
807 char *operator;
808 int token;
809 enum exp_opcode opcode;
810};
811
812static const struct token dot_ops[] =
813{
814 { ".and.", BOOL_AND, BINOP_END },
815 { ".AND.", BOOL_AND, BINOP_END },
816 { ".or.", BOOL_OR, BINOP_END },
817 { ".OR.", BOOL_OR, BINOP_END },
818 { ".not.", BOOL_NOT, BINOP_END },
819 { ".NOT.", BOOL_NOT, BINOP_END },
820 { ".eq.", EQUAL, BINOP_END },
821 { ".EQ.", EQUAL, BINOP_END },
822 { ".eqv.", EQUAL, BINOP_END },
823 { ".NEQV.", NOTEQUAL, BINOP_END },
824 { ".neqv.", NOTEQUAL, BINOP_END },
825 { ".EQV.", EQUAL, BINOP_END },
826 { ".ne.", NOTEQUAL, BINOP_END },
827 { ".NE.", NOTEQUAL, BINOP_END },
828 { ".le.", LEQ, BINOP_END },
829 { ".LE.", LEQ, BINOP_END },
830 { ".ge.", GEQ, BINOP_END },
831 { ".GE.", GEQ, BINOP_END },
832 { ".gt.", GREATERTHAN, BINOP_END },
833 { ".GT.", GREATERTHAN, BINOP_END },
834 { ".lt.", LESSTHAN, BINOP_END },
835 { ".LT.", LESSTHAN, BINOP_END },
836 { NULL, 0, 0 }
837};
838
839struct f77_boolean_val
840{
841 char *name;
842 int value;
843};
844
845static const struct f77_boolean_val boolean_values[] =
846{
847 { ".true.", 1 },
848 { ".TRUE.", 1 },
849 { ".false.", 0 },
850 { ".FALSE.", 0 },
851 { NULL, 0 }
852};
853
854static const struct token f77_keywords[] =
855{
856 { "complex_16", COMPLEX_S16_KEYWORD, BINOP_END },
857 { "complex_32", COMPLEX_S32_KEYWORD, BINOP_END },
858 { "character", CHARACTER, BINOP_END },
859 { "integer_2", INT_S2_KEYWORD, BINOP_END },
860 { "logical_1", LOGICAL_S1_KEYWORD, BINOP_END },
861 { "logical_2", LOGICAL_S2_KEYWORD, BINOP_END },
862 { "complex_8", COMPLEX_S8_KEYWORD, BINOP_END },
863 { "integer", INT_KEYWORD, BINOP_END },
864 { "logical", LOGICAL_KEYWORD, BINOP_END },
865 { "real_16", REAL_S16_KEYWORD, BINOP_END },
866 { "complex", COMPLEX_S8_KEYWORD, BINOP_END },
867 { "sizeof", SIZEOF, BINOP_END },
868 { "real_8", REAL_S8_KEYWORD, BINOP_END },
869 { "real", REAL_KEYWORD, BINOP_END },
870 { NULL, 0, 0 }
871};
872
873/* Implementation of a dynamically expandable buffer for processing input
874 characters acquired through lexptr and building a value to return in
875 yylval. Ripped off from ch-exp.y */
876
877static char *tempbuf; /* Current buffer contents */
878static int tempbufsize; /* Size of allocated buffer */
879static int tempbufindex; /* Current index into buffer */
880
881#define GROWBY_MIN_SIZE 64 /* Minimum amount to grow buffer by */
882
883#define CHECKBUF(size) \
884 do { \
885 if (tempbufindex + (size) >= tempbufsize) \
886 { \
887 growbuf_by_size (size); \
888 } \
889 } while (0);
890
891
892/* Grow the static temp buffer if necessary, including allocating the first one
893 on demand. */
894
895static void
896growbuf_by_size (count)
897 int count;
898{
899 int growby;
900
901 growby = max (count, GROWBY_MIN_SIZE);
902 tempbufsize += growby;
903 if (tempbuf == NULL)
904 tempbuf = (char *) malloc (tempbufsize);
905 else
906 tempbuf = (char *) realloc (tempbuf, tempbufsize);
907}
908
909/* Blatantly ripped off from ch-exp.y. This routine recognizes F77
910 string-literals.
911
912 Recognize a string literal. A string literal is a nonzero sequence
913 of characters enclosed in matching single quotes, except that
914 a single character inside single quotes is a character literal, which
915 we reject as a string literal. To embed the terminator character inside
916 a string, it is simply doubled (I.E. 'this''is''one''string') */
917
918static int
919match_string_literal ()
920{
921 char *tokptr = lexptr;
922
923 for (tempbufindex = 0, tokptr++; *tokptr != '\0'; tokptr++)
924 {
925 CHECKBUF (1);
926 if (*tokptr == *lexptr)
927 {
928 if (*(tokptr + 1) == *lexptr)
929 tokptr++;
930 else
931 break;
932 }
933 tempbuf[tempbufindex++] = *tokptr;
934 }
935 if (*tokptr == '\0' /* no terminator */
936 || tempbufindex == 0) /* no string */
937 return 0;
938 else
939 {
940 tempbuf[tempbufindex] = '\0';
941 yylval.sval.ptr = tempbuf;
942 yylval.sval.length = tempbufindex;
943 lexptr = ++tokptr;
944 return STRING_LITERAL;
945 }
946}
947
948/* Read one token, getting characters through lexptr. */
949
950static int
951yylex ()
952{
953 int c;
954 int namelen;
955 unsigned int i,token;
956 char *tokstart;
957
958 retry:
065432a8
PM
959
960 prev_lexptr = lexptr;
961
c906108c
SS
962 tokstart = lexptr;
963
964 /* First of all, let us make sure we are not dealing with the
965 special tokens .true. and .false. which evaluate to 1 and 0. */
966
967 if (*lexptr == '.')
968 {
969 for (i = 0; boolean_values[i].name != NULL; i++)
970 {
bf896cb0
AC
971 if (strncmp (tokstart, boolean_values[i].name,
972 strlen (boolean_values[i].name)) == 0)
c906108c
SS
973 {
974 lexptr += strlen (boolean_values[i].name);
975 yylval.lval = boolean_values[i].value;
976 return BOOLEAN_LITERAL;
977 }
978 }
979 }
980
bd49c137 981 /* See if it is a special .foo. operator. */
c906108c
SS
982
983 for (i = 0; dot_ops[i].operator != NULL; i++)
bf896cb0 984 if (strncmp (tokstart, dot_ops[i].operator, strlen (dot_ops[i].operator)) == 0)
c906108c
SS
985 {
986 lexptr += strlen (dot_ops[i].operator);
987 yylval.opcode = dot_ops[i].opcode;
988 return dot_ops[i].token;
989 }
990
bd49c137
WZ
991 /* See if it is an exponentiation operator. */
992
993 if (strncmp (tokstart, "**", 2) == 0)
994 {
995 lexptr += 2;
996 yylval.opcode = BINOP_EXP;
997 return STARSTAR;
998 }
999
c906108c
SS
1000 switch (c = *tokstart)
1001 {
1002 case 0:
1003 return 0;
1004
1005 case ' ':
1006 case '\t':
1007 case '\n':
1008 lexptr++;
1009 goto retry;
1010
1011 case '\'':
1012 token = match_string_literal ();
1013 if (token != 0)
1014 return (token);
1015 break;
1016
1017 case '(':
1018 paren_depth++;
1019 lexptr++;
1020 return c;
1021
1022 case ')':
1023 if (paren_depth == 0)
1024 return 0;
1025 paren_depth--;
1026 lexptr++;
1027 return c;
1028
1029 case ',':
1030 if (comma_terminates && paren_depth == 0)
1031 return 0;
1032 lexptr++;
1033 return c;
1034
1035 case '.':
1036 /* Might be a floating point number. */
1037 if (lexptr[1] < '0' || lexptr[1] > '9')
1038 goto symbol; /* Nope, must be a symbol. */
1039 /* FALL THRU into number case. */
1040
1041 case '0':
1042 case '1':
1043 case '2':
1044 case '3':
1045 case '4':
1046 case '5':
1047 case '6':
1048 case '7':
1049 case '8':
1050 case '9':
1051 {
1052 /* It's a number. */
1053 int got_dot = 0, got_e = 0, got_d = 0, toktype;
710122da 1054 char *p = tokstart;
c906108c
SS
1055 int hex = input_radix > 10;
1056
1057 if (c == '0' && (p[1] == 'x' || p[1] == 'X'))
1058 {
1059 p += 2;
1060 hex = 1;
1061 }
1062 else if (c == '0' && (p[1]=='t' || p[1]=='T' || p[1]=='d' || p[1]=='D'))
1063 {
1064 p += 2;
1065 hex = 0;
1066 }
1067
1068 for (;; ++p)
1069 {
1070 if (!hex && !got_e && (*p == 'e' || *p == 'E'))
1071 got_dot = got_e = 1;
1072 else if (!hex && !got_d && (*p == 'd' || *p == 'D'))
1073 got_dot = got_d = 1;
1074 else if (!hex && !got_dot && *p == '.')
1075 got_dot = 1;
1076 else if (((got_e && (p[-1] == 'e' || p[-1] == 'E'))
1077 || (got_d && (p[-1] == 'd' || p[-1] == 'D')))
1078 && (*p == '-' || *p == '+'))
1079 /* This is the sign of the exponent, not the end of the
1080 number. */
1081 continue;
1082 /* We will take any letters or digits. parse_number will
1083 complain if past the radix, or if L or U are not final. */
1084 else if ((*p < '0' || *p > '9')
1085 && ((*p < 'a' || *p > 'z')
1086 && (*p < 'A' || *p > 'Z')))
1087 break;
1088 }
1089 toktype = parse_number (tokstart, p - tokstart, got_dot|got_e|got_d,
1090 &yylval);
1091 if (toktype == ERROR)
1092 {
1093 char *err_copy = (char *) alloca (p - tokstart + 1);
1094
1095 memcpy (err_copy, tokstart, p - tokstart);
1096 err_copy[p - tokstart] = 0;
1097 error ("Invalid number \"%s\".", err_copy);
1098 }
1099 lexptr = p;
1100 return toktype;
1101 }
1102
1103 case '+':
1104 case '-':
1105 case '*':
1106 case '/':
1107 case '%':
1108 case '|':
1109 case '&':
1110 case '^':
1111 case '~':
1112 case '!':
1113 case '@':
1114 case '<':
1115 case '>':
1116 case '[':
1117 case ']':
1118 case '?':
1119 case ':':
1120 case '=':
1121 case '{':
1122 case '}':
1123 symbol:
1124 lexptr++;
1125 return c;
1126 }
1127
1128 if (!(c == '_' || c == '$'
1129 || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')))
1130 /* We must have come across a bad character (e.g. ';'). */
1131 error ("Invalid character '%c' in expression.", c);
1132
1133 namelen = 0;
1134 for (c = tokstart[namelen];
1135 (c == '_' || c == '$' || (c >= '0' && c <= '9')
1136 || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z'));
1137 c = tokstart[++namelen]);
1138
1139 /* The token "if" terminates the expression and is NOT
1140 removed from the input stream. */
1141
1142 if (namelen == 2 && tokstart[0] == 'i' && tokstart[1] == 'f')
1143 return 0;
1144
1145 lexptr += namelen;
1146
1147 /* Catch specific keywords. */
1148
1149 for (i = 0; f77_keywords[i].operator != NULL; i++)
bf896cb0
AC
1150 if (strncmp (tokstart, f77_keywords[i].operator,
1151 strlen(f77_keywords[i].operator)) == 0)
c906108c
SS
1152 {
1153 /* lexptr += strlen(f77_keywords[i].operator); */
1154 yylval.opcode = f77_keywords[i].opcode;
1155 return f77_keywords[i].token;
1156 }
1157
1158 yylval.sval.ptr = tokstart;
1159 yylval.sval.length = namelen;
1160
1161 if (*tokstart == '$')
1162 {
1163 write_dollar_variable (yylval.sval);
1164 return VARIABLE;
1165 }
1166
1167 /* Use token-type TYPENAME for symbols that happen to be defined
1168 currently as names of types; NAME for other symbols.
1169 The caller is not constrained to care about the distinction. */
1170 {
1171 char *tmp = copy_name (yylval.sval);
1172 struct symbol *sym;
1173 int is_a_field_of_this = 0;
1174 int hextype;
1175
1176 sym = lookup_symbol (tmp, expression_context_block,
176620f1 1177 VAR_DOMAIN,
3e79cecf 1178 parse_language->la_language == language_cplus
2570f2b7 1179 ? &is_a_field_of_this : NULL);
c906108c
SS
1180 if (sym && SYMBOL_CLASS (sym) == LOC_TYPEDEF)
1181 {
1182 yylval.tsym.type = SYMBOL_TYPE (sym);
1183 return TYPENAME;
1184 }
54a5b07d 1185 yylval.tsym.type
3e79cecf
UW
1186 = language_lookup_primitive_type_by_name (parse_language,
1187 parse_gdbarch, tmp);
54a5b07d 1188 if (yylval.tsym.type != NULL)
c906108c
SS
1189 return TYPENAME;
1190
1191 /* Input names that aren't symbols but ARE valid hex numbers,
1192 when the input radix permits them, can be names or numbers
1193 depending on the parse. Note we support radixes > 16 here. */
1194 if (!sym
1195 && ((tokstart[0] >= 'a' && tokstart[0] < 'a' + input_radix - 10)
1196 || (tokstart[0] >= 'A' && tokstart[0] < 'A' + input_radix - 10)))
1197 {
1198 YYSTYPE newlval; /* Its value is ignored. */
1199 hextype = parse_number (tokstart, namelen, 0, &newlval);
1200 if (hextype == INT)
1201 {
1202 yylval.ssym.sym = sym;
1203 yylval.ssym.is_a_field_of_this = is_a_field_of_this;
1204 return NAME_OR_INT;
1205 }
1206 }
1207
1208 /* Any other kind of symbol */
1209 yylval.ssym.sym = sym;
1210 yylval.ssym.is_a_field_of_this = is_a_field_of_this;
1211 return NAME;
1212 }
1213}
1214
1215void
1216yyerror (msg)
1217 char *msg;
1218{
065432a8
PM
1219 if (prev_lexptr)
1220 lexptr = prev_lexptr;
1221
c906108c
SS
1222 error ("A %s in expression, near `%s'.", (msg ? msg : "error"), lexptr);
1223}
This page took 0.642125 seconds and 4 git commands to generate.