* f-exp.y (yyparse): Add code to support exponentiation expression.
[binutils-gdb.git] / gdb / valarith.c
1 /* Perform arithmetic and other operations on values, for GDB.
2
3 Copyright 1986, 1988, 1989, 1990, 1991, 1992, 1993, 1994, 1995,
4 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free
5 Software Foundation, Inc.
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., 59 Temple Place - Suite 330,
22 Boston, MA 02111-1307, USA. */
23
24 #include "defs.h"
25 #include "value.h"
26 #include "symtab.h"
27 #include "gdbtypes.h"
28 #include "expression.h"
29 #include "target.h"
30 #include "language.h"
31 #include "gdb_string.h"
32 #include "doublest.h"
33 #include <math.h>
34 #include "infcall.h"
35
36 /* Define whether or not the C operator '/' truncates towards zero for
37 differently signed operands (truncation direction is undefined in C). */
38
39 #ifndef TRUNCATION_TOWARDS_ZERO
40 #define TRUNCATION_TOWARDS_ZERO ((-5 / 2) == -2)
41 #endif
42
43 static struct value *value_subscripted_rvalue (struct value *, struct value *, int);
44
45 void _initialize_valarith (void);
46 \f
47
48 /* Given a pointer, return the size of its target.
49 If the pointer type is void *, then return 1.
50 If the target type is incomplete, then error out.
51 This isn't a general purpose function, but just a
52 helper for value_sub & value_add.
53 */
54
55 static LONGEST
56 find_size_for_pointer_math (struct type *ptr_type)
57 {
58 LONGEST sz = -1;
59 struct type *ptr_target;
60
61 ptr_target = check_typedef (TYPE_TARGET_TYPE (ptr_type));
62
63 sz = TYPE_LENGTH (ptr_target);
64 if (sz == 0)
65 {
66 if (TYPE_CODE (ptr_type) == TYPE_CODE_VOID)
67 sz = 1;
68 else
69 {
70 char *name;
71
72 name = TYPE_NAME (ptr_target);
73 if (name == NULL)
74 name = TYPE_TAG_NAME (ptr_target);
75 if (name == NULL)
76 error (_("Cannot perform pointer math on incomplete types, "
77 "try casting to a known type, or void *."));
78 else
79 error (_("Cannot perform pointer math on incomplete type \"%s\", "
80 "try casting to a known type, or void *."), name);
81 }
82 }
83 return sz;
84 }
85
86 struct value *
87 value_add (struct value *arg1, struct value *arg2)
88 {
89 struct value *valint;
90 struct value *valptr;
91 LONGEST sz;
92 struct type *type1, *type2, *valptrtype;
93
94 arg1 = coerce_array (arg1);
95 arg2 = coerce_array (arg2);
96 type1 = check_typedef (value_type (arg1));
97 type2 = check_typedef (value_type (arg2));
98
99 if ((TYPE_CODE (type1) == TYPE_CODE_PTR
100 || TYPE_CODE (type2) == TYPE_CODE_PTR)
101 &&
102 (is_integral_type (type1) || is_integral_type (type2)))
103 /* Exactly one argument is a pointer, and one is an integer. */
104 {
105 struct value *retval;
106
107 if (TYPE_CODE (type1) == TYPE_CODE_PTR)
108 {
109 valptr = arg1;
110 valint = arg2;
111 valptrtype = type1;
112 }
113 else
114 {
115 valptr = arg2;
116 valint = arg1;
117 valptrtype = type2;
118 }
119
120 sz = find_size_for_pointer_math (valptrtype);
121
122 retval = value_from_pointer (valptrtype,
123 value_as_address (valptr)
124 + (sz * value_as_long (valint)));
125 return retval;
126 }
127
128 return value_binop (arg1, arg2, BINOP_ADD);
129 }
130
131 struct value *
132 value_sub (struct value *arg1, struct value *arg2)
133 {
134 struct type *type1, *type2;
135 arg1 = coerce_array (arg1);
136 arg2 = coerce_array (arg2);
137 type1 = check_typedef (value_type (arg1));
138 type2 = check_typedef (value_type (arg2));
139
140 if (TYPE_CODE (type1) == TYPE_CODE_PTR)
141 {
142 if (is_integral_type (type2))
143 {
144 /* pointer - integer. */
145 LONGEST sz = find_size_for_pointer_math (type1);
146
147 return value_from_pointer (type1,
148 (value_as_address (arg1)
149 - (sz * value_as_long (arg2))));
150 }
151 else if (TYPE_CODE (type2) == TYPE_CODE_PTR
152 && TYPE_LENGTH (check_typedef (TYPE_TARGET_TYPE (type1)))
153 == TYPE_LENGTH (check_typedef (TYPE_TARGET_TYPE (type2))))
154 {
155 /* pointer to <type x> - pointer to <type x>. */
156 LONGEST sz = TYPE_LENGTH (check_typedef (TYPE_TARGET_TYPE (type1)));
157 return value_from_longest
158 (builtin_type_long, /* FIXME -- should be ptrdiff_t */
159 (value_as_long (arg1) - value_as_long (arg2)) / sz);
160 }
161 else
162 {
163 error (_("\
164 First argument of `-' is a pointer and second argument is neither\n\
165 an integer nor a pointer of the same type."));
166 }
167 }
168
169 return value_binop (arg1, arg2, BINOP_SUB);
170 }
171
172 /* Return the value of ARRAY[IDX].
173 See comments in value_coerce_array() for rationale for reason for
174 doing lower bounds adjustment here rather than there.
175 FIXME: Perhaps we should validate that the index is valid and if
176 verbosity is set, warn about invalid indices (but still use them). */
177
178 struct value *
179 value_subscript (struct value *array, struct value *idx)
180 {
181 struct value *bound;
182 int c_style = current_language->c_style_arrays;
183 struct type *tarray;
184
185 array = coerce_ref (array);
186 tarray = check_typedef (value_type (array));
187
188 if (TYPE_CODE (tarray) == TYPE_CODE_ARRAY
189 || TYPE_CODE (tarray) == TYPE_CODE_STRING)
190 {
191 struct type *range_type = TYPE_INDEX_TYPE (tarray);
192 LONGEST lowerbound, upperbound;
193 get_discrete_bounds (range_type, &lowerbound, &upperbound);
194
195 if (VALUE_LVAL (array) != lval_memory)
196 return value_subscripted_rvalue (array, idx, lowerbound);
197
198 if (c_style == 0)
199 {
200 LONGEST index = value_as_long (idx);
201 if (index >= lowerbound && index <= upperbound)
202 return value_subscripted_rvalue (array, idx, lowerbound);
203 /* Emit warning unless we have an array of unknown size.
204 An array of unknown size has lowerbound 0 and upperbound -1. */
205 if (upperbound > -1)
206 warning (_("array or string index out of range"));
207 /* fall doing C stuff */
208 c_style = 1;
209 }
210
211 if (lowerbound != 0)
212 {
213 bound = value_from_longest (builtin_type_int, (LONGEST) lowerbound);
214 idx = value_sub (idx, bound);
215 }
216
217 array = value_coerce_array (array);
218 }
219
220 if (TYPE_CODE (tarray) == TYPE_CODE_BITSTRING)
221 {
222 struct type *range_type = TYPE_INDEX_TYPE (tarray);
223 LONGEST index = value_as_long (idx);
224 struct value *v;
225 int offset, byte, bit_index;
226 LONGEST lowerbound, upperbound;
227 get_discrete_bounds (range_type, &lowerbound, &upperbound);
228 if (index < lowerbound || index > upperbound)
229 error (_("bitstring index out of range"));
230 index -= lowerbound;
231 offset = index / TARGET_CHAR_BIT;
232 byte = *((char *) value_contents (array) + offset);
233 bit_index = index % TARGET_CHAR_BIT;
234 byte >>= (BITS_BIG_ENDIAN ? TARGET_CHAR_BIT - 1 - bit_index : bit_index);
235 v = value_from_longest (LA_BOOL_TYPE, byte & 1);
236 set_value_bitpos (v, bit_index);
237 set_value_bitsize (v, 1);
238 VALUE_LVAL (v) = VALUE_LVAL (array);
239 if (VALUE_LVAL (array) == lval_internalvar)
240 VALUE_LVAL (v) = lval_internalvar_component;
241 VALUE_ADDRESS (v) = VALUE_ADDRESS (array);
242 VALUE_FRAME_ID (v) = VALUE_FRAME_ID (array);
243 set_value_offset (v, offset + value_offset (array));
244 return v;
245 }
246
247 if (c_style)
248 return value_ind (value_add (array, idx));
249 else
250 error (_("not an array or string"));
251 }
252
253 /* Return the value of EXPR[IDX], expr an aggregate rvalue
254 (eg, a vector register). This routine used to promote floats
255 to doubles, but no longer does. */
256
257 static struct value *
258 value_subscripted_rvalue (struct value *array, struct value *idx, int lowerbound)
259 {
260 struct type *array_type = check_typedef (value_type (array));
261 struct type *elt_type = check_typedef (TYPE_TARGET_TYPE (array_type));
262 unsigned int elt_size = TYPE_LENGTH (elt_type);
263 LONGEST index = value_as_long (idx);
264 unsigned int elt_offs = elt_size * longest_to_int (index - lowerbound);
265 struct value *v;
266
267 if (index < lowerbound || elt_offs >= TYPE_LENGTH (array_type))
268 error (_("no such vector element"));
269
270 v = allocate_value (elt_type);
271 if (value_lazy (array))
272 set_value_lazy (v, 1);
273 else
274 memcpy (value_contents_writeable (v),
275 value_contents (array) + elt_offs, elt_size);
276
277 if (VALUE_LVAL (array) == lval_internalvar)
278 VALUE_LVAL (v) = lval_internalvar_component;
279 else
280 VALUE_LVAL (v) = VALUE_LVAL (array);
281 VALUE_ADDRESS (v) = VALUE_ADDRESS (array);
282 VALUE_REGNUM (v) = VALUE_REGNUM (array);
283 VALUE_FRAME_ID (v) = VALUE_FRAME_ID (array);
284 set_value_offset (v, value_offset (array) + elt_offs);
285 return v;
286 }
287 \f
288 /* Check to see if either argument is a structure. This is called so
289 we know whether to go ahead with the normal binop or look for a
290 user defined function instead.
291
292 For now, we do not overload the `=' operator. */
293
294 int
295 binop_user_defined_p (enum exp_opcode op, struct value *arg1, struct value *arg2)
296 {
297 struct type *type1, *type2;
298 if (op == BINOP_ASSIGN || op == BINOP_CONCAT)
299 return 0;
300 type1 = check_typedef (value_type (arg1));
301 type2 = check_typedef (value_type (arg2));
302 return (TYPE_CODE (type1) == TYPE_CODE_STRUCT
303 || TYPE_CODE (type2) == TYPE_CODE_STRUCT
304 || (TYPE_CODE (type1) == TYPE_CODE_REF
305 && TYPE_CODE (TYPE_TARGET_TYPE (type1)) == TYPE_CODE_STRUCT)
306 || (TYPE_CODE (type2) == TYPE_CODE_REF
307 && TYPE_CODE (TYPE_TARGET_TYPE (type2)) == TYPE_CODE_STRUCT));
308 }
309
310 /* Check to see if argument is a structure. This is called so
311 we know whether to go ahead with the normal unop or look for a
312 user defined function instead.
313
314 For now, we do not overload the `&' operator. */
315
316 int
317 unop_user_defined_p (enum exp_opcode op, struct value *arg1)
318 {
319 struct type *type1;
320 if (op == UNOP_ADDR)
321 return 0;
322 type1 = check_typedef (value_type (arg1));
323 for (;;)
324 {
325 if (TYPE_CODE (type1) == TYPE_CODE_STRUCT)
326 return 1;
327 else if (TYPE_CODE (type1) == TYPE_CODE_REF)
328 type1 = TYPE_TARGET_TYPE (type1);
329 else
330 return 0;
331 }
332 }
333
334 /* We know either arg1 or arg2 is a structure, so try to find the right
335 user defined function. Create an argument vector that calls
336 arg1.operator @ (arg1,arg2) and return that value (where '@' is any
337 binary operator which is legal for GNU C++).
338
339 OP is the operatore, and if it is BINOP_ASSIGN_MODIFY, then OTHEROP
340 is the opcode saying how to modify it. Otherwise, OTHEROP is
341 unused. */
342
343 struct value *
344 value_x_binop (struct value *arg1, struct value *arg2, enum exp_opcode op,
345 enum exp_opcode otherop, enum noside noside)
346 {
347 struct value **argvec;
348 char *ptr;
349 char tstr[13];
350 int static_memfuncp;
351
352 arg1 = coerce_ref (arg1);
353 arg2 = coerce_ref (arg2);
354 arg1 = coerce_enum (arg1);
355 arg2 = coerce_enum (arg2);
356
357 /* now we know that what we have to do is construct our
358 arg vector and find the right function to call it with. */
359
360 if (TYPE_CODE (check_typedef (value_type (arg1))) != TYPE_CODE_STRUCT)
361 error (_("Can't do that binary op on that type")); /* FIXME be explicit */
362
363 argvec = (struct value **) alloca (sizeof (struct value *) * 4);
364 argvec[1] = value_addr (arg1);
365 argvec[2] = arg2;
366 argvec[3] = 0;
367
368 /* make the right function name up */
369 strcpy (tstr, "operator__");
370 ptr = tstr + 8;
371 switch (op)
372 {
373 case BINOP_ADD:
374 strcpy (ptr, "+");
375 break;
376 case BINOP_SUB:
377 strcpy (ptr, "-");
378 break;
379 case BINOP_MUL:
380 strcpy (ptr, "*");
381 break;
382 case BINOP_DIV:
383 strcpy (ptr, "/");
384 break;
385 case BINOP_REM:
386 strcpy (ptr, "%");
387 break;
388 case BINOP_LSH:
389 strcpy (ptr, "<<");
390 break;
391 case BINOP_RSH:
392 strcpy (ptr, ">>");
393 break;
394 case BINOP_BITWISE_AND:
395 strcpy (ptr, "&");
396 break;
397 case BINOP_BITWISE_IOR:
398 strcpy (ptr, "|");
399 break;
400 case BINOP_BITWISE_XOR:
401 strcpy (ptr, "^");
402 break;
403 case BINOP_LOGICAL_AND:
404 strcpy (ptr, "&&");
405 break;
406 case BINOP_LOGICAL_OR:
407 strcpy (ptr, "||");
408 break;
409 case BINOP_MIN:
410 strcpy (ptr, "<?");
411 break;
412 case BINOP_MAX:
413 strcpy (ptr, ">?");
414 break;
415 case BINOP_ASSIGN:
416 strcpy (ptr, "=");
417 break;
418 case BINOP_ASSIGN_MODIFY:
419 switch (otherop)
420 {
421 case BINOP_ADD:
422 strcpy (ptr, "+=");
423 break;
424 case BINOP_SUB:
425 strcpy (ptr, "-=");
426 break;
427 case BINOP_MUL:
428 strcpy (ptr, "*=");
429 break;
430 case BINOP_DIV:
431 strcpy (ptr, "/=");
432 break;
433 case BINOP_REM:
434 strcpy (ptr, "%=");
435 break;
436 case BINOP_BITWISE_AND:
437 strcpy (ptr, "&=");
438 break;
439 case BINOP_BITWISE_IOR:
440 strcpy (ptr, "|=");
441 break;
442 case BINOP_BITWISE_XOR:
443 strcpy (ptr, "^=");
444 break;
445 case BINOP_MOD: /* invalid */
446 default:
447 error (_("Invalid binary operation specified."));
448 }
449 break;
450 case BINOP_SUBSCRIPT:
451 strcpy (ptr, "[]");
452 break;
453 case BINOP_EQUAL:
454 strcpy (ptr, "==");
455 break;
456 case BINOP_NOTEQUAL:
457 strcpy (ptr, "!=");
458 break;
459 case BINOP_LESS:
460 strcpy (ptr, "<");
461 break;
462 case BINOP_GTR:
463 strcpy (ptr, ">");
464 break;
465 case BINOP_GEQ:
466 strcpy (ptr, ">=");
467 break;
468 case BINOP_LEQ:
469 strcpy (ptr, "<=");
470 break;
471 case BINOP_MOD: /* invalid */
472 default:
473 error (_("Invalid binary operation specified."));
474 }
475
476 argvec[0] = value_struct_elt (&arg1, argvec + 1, tstr, &static_memfuncp, "structure");
477
478 if (argvec[0])
479 {
480 if (static_memfuncp)
481 {
482 argvec[1] = argvec[0];
483 argvec++;
484 }
485 if (noside == EVAL_AVOID_SIDE_EFFECTS)
486 {
487 struct type *return_type;
488 return_type
489 = TYPE_TARGET_TYPE (check_typedef (value_type (argvec[0])));
490 return value_zero (return_type, VALUE_LVAL (arg1));
491 }
492 return call_function_by_hand (argvec[0], 2 - static_memfuncp, argvec + 1);
493 }
494 error (_("member function %s not found"), tstr);
495 #ifdef lint
496 return call_function_by_hand (argvec[0], 2 - static_memfuncp, argvec + 1);
497 #endif
498 }
499
500 /* We know that arg1 is a structure, so try to find a unary user
501 defined operator that matches the operator in question.
502 Create an argument vector that calls arg1.operator @ (arg1)
503 and return that value (where '@' is (almost) any unary operator which
504 is legal for GNU C++). */
505
506 struct value *
507 value_x_unop (struct value *arg1, enum exp_opcode op, enum noside noside)
508 {
509 struct value **argvec;
510 char *ptr, *mangle_ptr;
511 char tstr[13], mangle_tstr[13];
512 int static_memfuncp, nargs;
513
514 arg1 = coerce_ref (arg1);
515 arg1 = coerce_enum (arg1);
516
517 /* now we know that what we have to do is construct our
518 arg vector and find the right function to call it with. */
519
520 if (TYPE_CODE (check_typedef (value_type (arg1))) != TYPE_CODE_STRUCT)
521 error (_("Can't do that unary op on that type")); /* FIXME be explicit */
522
523 argvec = (struct value **) alloca (sizeof (struct value *) * 4);
524 argvec[1] = value_addr (arg1);
525 argvec[2] = 0;
526
527 nargs = 1;
528
529 /* make the right function name up */
530 strcpy (tstr, "operator__");
531 ptr = tstr + 8;
532 strcpy (mangle_tstr, "__");
533 mangle_ptr = mangle_tstr + 2;
534 switch (op)
535 {
536 case UNOP_PREINCREMENT:
537 strcpy (ptr, "++");
538 break;
539 case UNOP_PREDECREMENT:
540 strcpy (ptr, "--");
541 break;
542 case UNOP_POSTINCREMENT:
543 strcpy (ptr, "++");
544 argvec[2] = value_from_longest (builtin_type_int, 0);
545 argvec[3] = 0;
546 nargs ++;
547 break;
548 case UNOP_POSTDECREMENT:
549 strcpy (ptr, "--");
550 argvec[2] = value_from_longest (builtin_type_int, 0);
551 argvec[3] = 0;
552 nargs ++;
553 break;
554 case UNOP_LOGICAL_NOT:
555 strcpy (ptr, "!");
556 break;
557 case UNOP_COMPLEMENT:
558 strcpy (ptr, "~");
559 break;
560 case UNOP_NEG:
561 strcpy (ptr, "-");
562 break;
563 case UNOP_PLUS:
564 strcpy (ptr, "+");
565 break;
566 case UNOP_IND:
567 strcpy (ptr, "*");
568 break;
569 default:
570 error (_("Invalid unary operation specified."));
571 }
572
573 argvec[0] = value_struct_elt (&arg1, argvec + 1, tstr, &static_memfuncp, "structure");
574
575 if (argvec[0])
576 {
577 if (static_memfuncp)
578 {
579 argvec[1] = argvec[0];
580 nargs --;
581 argvec++;
582 }
583 if (noside == EVAL_AVOID_SIDE_EFFECTS)
584 {
585 struct type *return_type;
586 return_type
587 = TYPE_TARGET_TYPE (check_typedef (value_type (argvec[0])));
588 return value_zero (return_type, VALUE_LVAL (arg1));
589 }
590 return call_function_by_hand (argvec[0], nargs, argvec + 1);
591 }
592 error (_("member function %s not found"), tstr);
593 return 0; /* For lint -- never reached */
594 }
595 \f
596
597 /* Concatenate two values with the following conditions:
598
599 (1) Both values must be either bitstring values or character string
600 values and the resulting value consists of the concatenation of
601 ARG1 followed by ARG2.
602
603 or
604
605 One value must be an integer value and the other value must be
606 either a bitstring value or character string value, which is
607 to be repeated by the number of times specified by the integer
608 value.
609
610
611 (2) Boolean values are also allowed and are treated as bit string
612 values of length 1.
613
614 (3) Character values are also allowed and are treated as character
615 string values of length 1.
616 */
617
618 struct value *
619 value_concat (struct value *arg1, struct value *arg2)
620 {
621 struct value *inval1;
622 struct value *inval2;
623 struct value *outval = NULL;
624 int inval1len, inval2len;
625 int count, idx;
626 char *ptr;
627 char inchar;
628 struct type *type1 = check_typedef (value_type (arg1));
629 struct type *type2 = check_typedef (value_type (arg2));
630
631 /* First figure out if we are dealing with two values to be concatenated
632 or a repeat count and a value to be repeated. INVAL1 is set to the
633 first of two concatenated values, or the repeat count. INVAL2 is set
634 to the second of the two concatenated values or the value to be
635 repeated. */
636
637 if (TYPE_CODE (type2) == TYPE_CODE_INT)
638 {
639 struct type *tmp = type1;
640 type1 = tmp;
641 tmp = type2;
642 inval1 = arg2;
643 inval2 = arg1;
644 }
645 else
646 {
647 inval1 = arg1;
648 inval2 = arg2;
649 }
650
651 /* Now process the input values. */
652
653 if (TYPE_CODE (type1) == TYPE_CODE_INT)
654 {
655 /* We have a repeat count. Validate the second value and then
656 construct a value repeated that many times. */
657 if (TYPE_CODE (type2) == TYPE_CODE_STRING
658 || TYPE_CODE (type2) == TYPE_CODE_CHAR)
659 {
660 count = longest_to_int (value_as_long (inval1));
661 inval2len = TYPE_LENGTH (type2);
662 ptr = (char *) alloca (count * inval2len);
663 if (TYPE_CODE (type2) == TYPE_CODE_CHAR)
664 {
665 inchar = (char) unpack_long (type2,
666 value_contents (inval2));
667 for (idx = 0; idx < count; idx++)
668 {
669 *(ptr + idx) = inchar;
670 }
671 }
672 else
673 {
674 for (idx = 0; idx < count; idx++)
675 {
676 memcpy (ptr + (idx * inval2len), value_contents (inval2),
677 inval2len);
678 }
679 }
680 outval = value_string (ptr, count * inval2len);
681 }
682 else if (TYPE_CODE (type2) == TYPE_CODE_BITSTRING
683 || TYPE_CODE (type2) == TYPE_CODE_BOOL)
684 {
685 error (_("unimplemented support for bitstring/boolean repeats"));
686 }
687 else
688 {
689 error (_("can't repeat values of that type"));
690 }
691 }
692 else if (TYPE_CODE (type1) == TYPE_CODE_STRING
693 || TYPE_CODE (type1) == TYPE_CODE_CHAR)
694 {
695 /* We have two character strings to concatenate. */
696 if (TYPE_CODE (type2) != TYPE_CODE_STRING
697 && TYPE_CODE (type2) != TYPE_CODE_CHAR)
698 {
699 error (_("Strings can only be concatenated with other strings."));
700 }
701 inval1len = TYPE_LENGTH (type1);
702 inval2len = TYPE_LENGTH (type2);
703 ptr = (char *) alloca (inval1len + inval2len);
704 if (TYPE_CODE (type1) == TYPE_CODE_CHAR)
705 {
706 *ptr = (char) unpack_long (type1, value_contents (inval1));
707 }
708 else
709 {
710 memcpy (ptr, value_contents (inval1), inval1len);
711 }
712 if (TYPE_CODE (type2) == TYPE_CODE_CHAR)
713 {
714 *(ptr + inval1len) =
715 (char) unpack_long (type2, value_contents (inval2));
716 }
717 else
718 {
719 memcpy (ptr + inval1len, value_contents (inval2), inval2len);
720 }
721 outval = value_string (ptr, inval1len + inval2len);
722 }
723 else if (TYPE_CODE (type1) == TYPE_CODE_BITSTRING
724 || TYPE_CODE (type1) == TYPE_CODE_BOOL)
725 {
726 /* We have two bitstrings to concatenate. */
727 if (TYPE_CODE (type2) != TYPE_CODE_BITSTRING
728 && TYPE_CODE (type2) != TYPE_CODE_BOOL)
729 {
730 error (_("Bitstrings or booleans can only be concatenated with other bitstrings or booleans."));
731 }
732 error (_("unimplemented support for bitstring/boolean concatenation."));
733 }
734 else
735 {
736 /* We don't know how to concatenate these operands. */
737 error (_("illegal operands for concatenation."));
738 }
739 return (outval);
740 }
741 \f
742
743
744 /* Perform a binary operation on two operands which have reasonable
745 representations as integers or floats. This includes booleans,
746 characters, integers, or floats.
747 Does not support addition and subtraction on pointers;
748 use value_add or value_sub if you want to handle those possibilities. */
749
750 struct value *
751 value_binop (struct value *arg1, struct value *arg2, enum exp_opcode op)
752 {
753 struct value *val;
754 struct type *type1, *type2;
755
756 arg1 = coerce_ref (arg1);
757 arg2 = coerce_ref (arg2);
758 type1 = check_typedef (value_type (arg1));
759 type2 = check_typedef (value_type (arg2));
760
761 if ((TYPE_CODE (type1) != TYPE_CODE_FLT && !is_integral_type (type1))
762 ||
763 (TYPE_CODE (type2) != TYPE_CODE_FLT && !is_integral_type (type2)))
764 error (_("Argument to arithmetic operation not a number or boolean."));
765
766 if (TYPE_CODE (type1) == TYPE_CODE_FLT
767 ||
768 TYPE_CODE (type2) == TYPE_CODE_FLT)
769 {
770 /* FIXME-if-picky-about-floating-accuracy: Should be doing this
771 in target format. real.c in GCC probably has the necessary
772 code. */
773 DOUBLEST v1, v2, v = 0;
774 v1 = value_as_double (arg1);
775 v2 = value_as_double (arg2);
776 switch (op)
777 {
778 case BINOP_ADD:
779 v = v1 + v2;
780 break;
781
782 case BINOP_SUB:
783 v = v1 - v2;
784 break;
785
786 case BINOP_MUL:
787 v = v1 * v2;
788 break;
789
790 case BINOP_DIV:
791 v = v1 / v2;
792 break;
793
794 case BINOP_EXP:
795 errno = 0;
796 v = pow (v1, v2);
797 if (errno)
798 error (_("Cannot perform exponentiation: %s"), safe_strerror (errno));
799 break;
800
801 default:
802 error (_("Integer-only operation on floating point number."));
803 }
804
805 /* If either arg was long double, make sure that value is also long
806 double. */
807
808 if (TYPE_LENGTH (type1) * 8 > TARGET_DOUBLE_BIT
809 || TYPE_LENGTH (type2) * 8 > TARGET_DOUBLE_BIT)
810 val = allocate_value (builtin_type_long_double);
811 else
812 val = allocate_value (builtin_type_double);
813
814 store_typed_floating (value_contents_raw (val), value_type (val), v);
815 }
816 else if (TYPE_CODE (type1) == TYPE_CODE_BOOL
817 &&
818 TYPE_CODE (type2) == TYPE_CODE_BOOL)
819 {
820 LONGEST v1, v2, v = 0;
821 v1 = value_as_long (arg1);
822 v2 = value_as_long (arg2);
823
824 switch (op)
825 {
826 case BINOP_BITWISE_AND:
827 v = v1 & v2;
828 break;
829
830 case BINOP_BITWISE_IOR:
831 v = v1 | v2;
832 break;
833
834 case BINOP_BITWISE_XOR:
835 v = v1 ^ v2;
836 break;
837
838 case BINOP_EQUAL:
839 v = v1 == v2;
840 break;
841
842 case BINOP_NOTEQUAL:
843 v = v1 != v2;
844 break;
845
846 default:
847 error (_("Invalid operation on booleans."));
848 }
849
850 val = allocate_value (type1);
851 store_signed_integer (value_contents_raw (val),
852 TYPE_LENGTH (type1),
853 v);
854 }
855 else
856 /* Integral operations here. */
857 /* FIXME: Also mixed integral/booleans, with result an integer. */
858 /* FIXME: This implements ANSI C rules (also correct for C++).
859 What about FORTRAN and (the deleted) chill ? */
860 {
861 unsigned int promoted_len1 = TYPE_LENGTH (type1);
862 unsigned int promoted_len2 = TYPE_LENGTH (type2);
863 int is_unsigned1 = TYPE_UNSIGNED (type1);
864 int is_unsigned2 = TYPE_UNSIGNED (type2);
865 unsigned int result_len;
866 int unsigned_operation;
867
868 /* Determine type length and signedness after promotion for
869 both operands. */
870 if (promoted_len1 < TYPE_LENGTH (builtin_type_int))
871 {
872 is_unsigned1 = 0;
873 promoted_len1 = TYPE_LENGTH (builtin_type_int);
874 }
875 if (promoted_len2 < TYPE_LENGTH (builtin_type_int))
876 {
877 is_unsigned2 = 0;
878 promoted_len2 = TYPE_LENGTH (builtin_type_int);
879 }
880
881 /* Determine type length of the result, and if the operation should
882 be done unsigned.
883 Use the signedness of the operand with the greater length.
884 If both operands are of equal length, use unsigned operation
885 if one of the operands is unsigned. */
886 if (promoted_len1 > promoted_len2)
887 {
888 unsigned_operation = is_unsigned1;
889 result_len = promoted_len1;
890 }
891 else if (promoted_len2 > promoted_len1)
892 {
893 unsigned_operation = is_unsigned2;
894 result_len = promoted_len2;
895 }
896 else
897 {
898 unsigned_operation = is_unsigned1 || is_unsigned2;
899 result_len = promoted_len1;
900 }
901
902 if (unsigned_operation)
903 {
904 ULONGEST v1, v2, v = 0;
905 v1 = (ULONGEST) value_as_long (arg1);
906 v2 = (ULONGEST) value_as_long (arg2);
907
908 /* Truncate values to the type length of the result. */
909 if (result_len < sizeof (ULONGEST))
910 {
911 v1 &= ((LONGEST) 1 << HOST_CHAR_BIT * result_len) - 1;
912 v2 &= ((LONGEST) 1 << HOST_CHAR_BIT * result_len) - 1;
913 }
914
915 switch (op)
916 {
917 case BINOP_ADD:
918 v = v1 + v2;
919 break;
920
921 case BINOP_SUB:
922 v = v1 - v2;
923 break;
924
925 case BINOP_MUL:
926 v = v1 * v2;
927 break;
928
929 case BINOP_DIV:
930 v = v1 / v2;
931 break;
932
933 case BINOP_EXP:
934 errno = 0;
935 v = pow (v1, v2);
936 if (errno)
937 error (_("Cannot perform exponentiation: %s"), safe_strerror (errno));
938 break;
939
940 case BINOP_REM:
941 v = v1 % v2;
942 break;
943
944 case BINOP_MOD:
945 /* Knuth 1.2.4, integer only. Note that unlike the C '%' op,
946 v1 mod 0 has a defined value, v1. */
947 if (v2 == 0)
948 {
949 v = v1;
950 }
951 else
952 {
953 v = v1 / v2;
954 /* Note floor(v1/v2) == v1/v2 for unsigned. */
955 v = v1 - (v2 * v);
956 }
957 break;
958
959 case BINOP_LSH:
960 v = v1 << v2;
961 break;
962
963 case BINOP_RSH:
964 v = v1 >> v2;
965 break;
966
967 case BINOP_BITWISE_AND:
968 v = v1 & v2;
969 break;
970
971 case BINOP_BITWISE_IOR:
972 v = v1 | v2;
973 break;
974
975 case BINOP_BITWISE_XOR:
976 v = v1 ^ v2;
977 break;
978
979 case BINOP_LOGICAL_AND:
980 v = v1 && v2;
981 break;
982
983 case BINOP_LOGICAL_OR:
984 v = v1 || v2;
985 break;
986
987 case BINOP_MIN:
988 v = v1 < v2 ? v1 : v2;
989 break;
990
991 case BINOP_MAX:
992 v = v1 > v2 ? v1 : v2;
993 break;
994
995 case BINOP_EQUAL:
996 v = v1 == v2;
997 break;
998
999 case BINOP_NOTEQUAL:
1000 v = v1 != v2;
1001 break;
1002
1003 case BINOP_LESS:
1004 v = v1 < v2;
1005 break;
1006
1007 default:
1008 error (_("Invalid binary operation on numbers."));
1009 }
1010
1011 /* This is a kludge to get around the fact that we don't
1012 know how to determine the result type from the types of
1013 the operands. (I'm not really sure how much we feel the
1014 need to duplicate the exact rules of the current
1015 language. They can get really hairy. But not to do so
1016 makes it hard to document just what we *do* do). */
1017
1018 /* Can't just call init_type because we wouldn't know what
1019 name to give the type. */
1020 val = allocate_value
1021 (result_len > TARGET_LONG_BIT / HOST_CHAR_BIT
1022 ? builtin_type_unsigned_long_long
1023 : builtin_type_unsigned_long);
1024 store_unsigned_integer (value_contents_raw (val),
1025 TYPE_LENGTH (value_type (val)),
1026 v);
1027 }
1028 else
1029 {
1030 LONGEST v1, v2, v = 0;
1031 v1 = value_as_long (arg1);
1032 v2 = value_as_long (arg2);
1033
1034 switch (op)
1035 {
1036 case BINOP_ADD:
1037 v = v1 + v2;
1038 break;
1039
1040 case BINOP_SUB:
1041 v = v1 - v2;
1042 break;
1043
1044 case BINOP_MUL:
1045 v = v1 * v2;
1046 break;
1047
1048 case BINOP_DIV:
1049 if (v2 != 0)
1050 v = v1 / v2;
1051 else
1052 error (_("Division by zero"));
1053 break;
1054
1055 case BINOP_EXP:
1056 errno = 0;
1057 v = pow (v1, v2);
1058 if (errno)
1059 error (_("Cannot perform exponentiation: %s"), safe_strerror (errno));
1060 break;
1061
1062 case BINOP_REM:
1063 if (v2 != 0)
1064 v = v1 % v2;
1065 else
1066 error (_("Division by zero"));
1067 break;
1068
1069 case BINOP_MOD:
1070 /* Knuth 1.2.4, integer only. Note that unlike the C '%' op,
1071 X mod 0 has a defined value, X. */
1072 if (v2 == 0)
1073 {
1074 v = v1;
1075 }
1076 else
1077 {
1078 v = v1 / v2;
1079 /* Compute floor. */
1080 if (TRUNCATION_TOWARDS_ZERO && (v < 0) && ((v1 % v2) != 0))
1081 {
1082 v--;
1083 }
1084 v = v1 - (v2 * v);
1085 }
1086 break;
1087
1088 case BINOP_LSH:
1089 v = v1 << v2;
1090 break;
1091
1092 case BINOP_RSH:
1093 v = v1 >> v2;
1094 break;
1095
1096 case BINOP_BITWISE_AND:
1097 v = v1 & v2;
1098 break;
1099
1100 case BINOP_BITWISE_IOR:
1101 v = v1 | v2;
1102 break;
1103
1104 case BINOP_BITWISE_XOR:
1105 v = v1 ^ v2;
1106 break;
1107
1108 case BINOP_LOGICAL_AND:
1109 v = v1 && v2;
1110 break;
1111
1112 case BINOP_LOGICAL_OR:
1113 v = v1 || v2;
1114 break;
1115
1116 case BINOP_MIN:
1117 v = v1 < v2 ? v1 : v2;
1118 break;
1119
1120 case BINOP_MAX:
1121 v = v1 > v2 ? v1 : v2;
1122 break;
1123
1124 case BINOP_EQUAL:
1125 v = v1 == v2;
1126 break;
1127
1128 case BINOP_LESS:
1129 v = v1 < v2;
1130 break;
1131
1132 default:
1133 error (_("Invalid binary operation on numbers."));
1134 }
1135
1136 /* This is a kludge to get around the fact that we don't
1137 know how to determine the result type from the types of
1138 the operands. (I'm not really sure how much we feel the
1139 need to duplicate the exact rules of the current
1140 language. They can get really hairy. But not to do so
1141 makes it hard to document just what we *do* do). */
1142
1143 /* Can't just call init_type because we wouldn't know what
1144 name to give the type. */
1145 val = allocate_value
1146 (result_len > TARGET_LONG_BIT / HOST_CHAR_BIT
1147 ? builtin_type_long_long
1148 : builtin_type_long);
1149 store_signed_integer (value_contents_raw (val),
1150 TYPE_LENGTH (value_type (val)),
1151 v);
1152 }
1153 }
1154
1155 return val;
1156 }
1157 \f
1158 /* Simulate the C operator ! -- return 1 if ARG1 contains zero. */
1159
1160 int
1161 value_logical_not (struct value *arg1)
1162 {
1163 int len;
1164 const gdb_byte *p;
1165 struct type *type1;
1166
1167 arg1 = coerce_number (arg1);
1168 type1 = check_typedef (value_type (arg1));
1169
1170 if (TYPE_CODE (type1) == TYPE_CODE_FLT)
1171 return 0 == value_as_double (arg1);
1172
1173 len = TYPE_LENGTH (type1);
1174 p = value_contents (arg1);
1175
1176 while (--len >= 0)
1177 {
1178 if (*p++)
1179 break;
1180 }
1181
1182 return len < 0;
1183 }
1184
1185 /* Perform a comparison on two string values (whose content are not
1186 necessarily null terminated) based on their length */
1187
1188 static int
1189 value_strcmp (struct value *arg1, struct value *arg2)
1190 {
1191 int len1 = TYPE_LENGTH (value_type (arg1));
1192 int len2 = TYPE_LENGTH (value_type (arg2));
1193 const gdb_byte *s1 = value_contents (arg1);
1194 const gdb_byte *s2 = value_contents (arg2);
1195 int i, len = len1 < len2 ? len1 : len2;
1196
1197 for (i = 0; i < len; i++)
1198 {
1199 if (s1[i] < s2[i])
1200 return -1;
1201 else if (s1[i] > s2[i])
1202 return 1;
1203 else
1204 continue;
1205 }
1206
1207 if (len1 < len2)
1208 return -1;
1209 else if (len1 > len2)
1210 return 1;
1211 else
1212 return 0;
1213 }
1214
1215 /* Simulate the C operator == by returning a 1
1216 iff ARG1 and ARG2 have equal contents. */
1217
1218 int
1219 value_equal (struct value *arg1, struct value *arg2)
1220 {
1221 int len;
1222 const gdb_byte *p1;
1223 const gdb_byte *p2;
1224 struct type *type1, *type2;
1225 enum type_code code1;
1226 enum type_code code2;
1227 int is_int1, is_int2;
1228
1229 arg1 = coerce_array (arg1);
1230 arg2 = coerce_array (arg2);
1231
1232 type1 = check_typedef (value_type (arg1));
1233 type2 = check_typedef (value_type (arg2));
1234 code1 = TYPE_CODE (type1);
1235 code2 = TYPE_CODE (type2);
1236 is_int1 = is_integral_type (type1);
1237 is_int2 = is_integral_type (type2);
1238
1239 if (is_int1 && is_int2)
1240 return longest_to_int (value_as_long (value_binop (arg1, arg2,
1241 BINOP_EQUAL)));
1242 else if ((code1 == TYPE_CODE_FLT || is_int1)
1243 && (code2 == TYPE_CODE_FLT || is_int2))
1244 return value_as_double (arg1) == value_as_double (arg2);
1245
1246 /* FIXME: Need to promote to either CORE_ADDR or LONGEST, whichever
1247 is bigger. */
1248 else if (code1 == TYPE_CODE_PTR && is_int2)
1249 return value_as_address (arg1) == (CORE_ADDR) value_as_long (arg2);
1250 else if (code2 == TYPE_CODE_PTR && is_int1)
1251 return (CORE_ADDR) value_as_long (arg1) == value_as_address (arg2);
1252
1253 else if (code1 == code2
1254 && ((len = (int) TYPE_LENGTH (type1))
1255 == (int) TYPE_LENGTH (type2)))
1256 {
1257 p1 = value_contents (arg1);
1258 p2 = value_contents (arg2);
1259 while (--len >= 0)
1260 {
1261 if (*p1++ != *p2++)
1262 break;
1263 }
1264 return len < 0;
1265 }
1266 else if (code1 == TYPE_CODE_STRING && code2 == TYPE_CODE_STRING)
1267 {
1268 return value_strcmp (arg1, arg2) == 0;
1269 }
1270 else
1271 {
1272 error (_("Invalid type combination in equality test."));
1273 return 0; /* For lint -- never reached */
1274 }
1275 }
1276
1277 /* Simulate the C operator < by returning 1
1278 iff ARG1's contents are less than ARG2's. */
1279
1280 int
1281 value_less (struct value *arg1, struct value *arg2)
1282 {
1283 enum type_code code1;
1284 enum type_code code2;
1285 struct type *type1, *type2;
1286 int is_int1, is_int2;
1287
1288 arg1 = coerce_array (arg1);
1289 arg2 = coerce_array (arg2);
1290
1291 type1 = check_typedef (value_type (arg1));
1292 type2 = check_typedef (value_type (arg2));
1293 code1 = TYPE_CODE (type1);
1294 code2 = TYPE_CODE (type2);
1295 is_int1 = is_integral_type (type1);
1296 is_int2 = is_integral_type (type2);
1297
1298 if (is_int1 && is_int2)
1299 return longest_to_int (value_as_long (value_binop (arg1, arg2,
1300 BINOP_LESS)));
1301 else if ((code1 == TYPE_CODE_FLT || is_int1)
1302 && (code2 == TYPE_CODE_FLT || is_int2))
1303 return value_as_double (arg1) < value_as_double (arg2);
1304 else if (code1 == TYPE_CODE_PTR && code2 == TYPE_CODE_PTR)
1305 return value_as_address (arg1) < value_as_address (arg2);
1306
1307 /* FIXME: Need to promote to either CORE_ADDR or LONGEST, whichever
1308 is bigger. */
1309 else if (code1 == TYPE_CODE_PTR && is_int2)
1310 return value_as_address (arg1) < (CORE_ADDR) value_as_long (arg2);
1311 else if (code2 == TYPE_CODE_PTR && is_int1)
1312 return (CORE_ADDR) value_as_long (arg1) < value_as_address (arg2);
1313 else if (code1 == TYPE_CODE_STRING && code2 == TYPE_CODE_STRING)
1314 return value_strcmp (arg1, arg2) < 0;
1315 else
1316 {
1317 error (_("Invalid type combination in ordering comparison."));
1318 return 0;
1319 }
1320 }
1321 \f
1322 /* The unary operators +, - and ~. They free the argument ARG1. */
1323
1324 struct value *
1325 value_pos (struct value *arg1)
1326 {
1327 struct type *type;
1328
1329 arg1 = coerce_ref (arg1);
1330
1331 type = check_typedef (value_type (arg1));
1332
1333 if (TYPE_CODE (type) == TYPE_CODE_FLT)
1334 return value_from_double (type, value_as_double (arg1));
1335 else if (is_integral_type (type))
1336 {
1337 /* Perform integral promotion for ANSI C/C++. FIXME: What about
1338 FORTRAN and (the deleted) chill ? */
1339 if (TYPE_LENGTH (type) < TYPE_LENGTH (builtin_type_int))
1340 type = builtin_type_int;
1341
1342 return value_from_longest (type, value_as_long (arg1));
1343 }
1344 else
1345 {
1346 error ("Argument to positive operation not a number.");
1347 return 0; /* For lint -- never reached */
1348 }
1349 }
1350
1351 struct value *
1352 value_neg (struct value *arg1)
1353 {
1354 struct type *type;
1355 struct type *result_type = value_type (arg1);
1356
1357 arg1 = coerce_ref (arg1);
1358
1359 type = check_typedef (value_type (arg1));
1360
1361 if (TYPE_CODE (type) == TYPE_CODE_FLT)
1362 return value_from_double (result_type, -value_as_double (arg1));
1363 else if (is_integral_type (type))
1364 {
1365 /* Perform integral promotion for ANSI C/C++. FIXME: What about
1366 FORTRAN and (the deleted) chill ? */
1367 if (TYPE_LENGTH (type) < TYPE_LENGTH (builtin_type_int))
1368 result_type = builtin_type_int;
1369
1370 return value_from_longest (result_type, -value_as_long (arg1));
1371 }
1372 else
1373 {
1374 error (_("Argument to negate operation not a number."));
1375 return 0; /* For lint -- never reached */
1376 }
1377 }
1378
1379 struct value *
1380 value_complement (struct value *arg1)
1381 {
1382 struct type *type;
1383 struct type *result_type = value_type (arg1);
1384
1385 arg1 = coerce_ref (arg1);
1386
1387 type = check_typedef (value_type (arg1));
1388
1389 if (!is_integral_type (type))
1390 error (_("Argument to complement operation not an integer or boolean."));
1391
1392 /* Perform integral promotion for ANSI C/C++.
1393 FIXME: What about FORTRAN ? */
1394 if (TYPE_LENGTH (type) < TYPE_LENGTH (builtin_type_int))
1395 result_type = builtin_type_int;
1396
1397 return value_from_longest (result_type, ~value_as_long (arg1));
1398 }
1399 \f
1400 /* The INDEX'th bit of SET value whose value_type is TYPE,
1401 and whose value_contents is valaddr.
1402 Return -1 if out of range, -2 other error. */
1403
1404 int
1405 value_bit_index (struct type *type, const gdb_byte *valaddr, int index)
1406 {
1407 LONGEST low_bound, high_bound;
1408 LONGEST word;
1409 unsigned rel_index;
1410 struct type *range = TYPE_FIELD_TYPE (type, 0);
1411 if (get_discrete_bounds (range, &low_bound, &high_bound) < 0)
1412 return -2;
1413 if (index < low_bound || index > high_bound)
1414 return -1;
1415 rel_index = index - low_bound;
1416 word = unpack_long (builtin_type_unsigned_char,
1417 valaddr + (rel_index / TARGET_CHAR_BIT));
1418 rel_index %= TARGET_CHAR_BIT;
1419 if (BITS_BIG_ENDIAN)
1420 rel_index = TARGET_CHAR_BIT - 1 - rel_index;
1421 return (word >> rel_index) & 1;
1422 }
1423
1424 struct value *
1425 value_in (struct value *element, struct value *set)
1426 {
1427 int member;
1428 struct type *settype = check_typedef (value_type (set));
1429 struct type *eltype = check_typedef (value_type (element));
1430 if (TYPE_CODE (eltype) == TYPE_CODE_RANGE)
1431 eltype = TYPE_TARGET_TYPE (eltype);
1432 if (TYPE_CODE (settype) != TYPE_CODE_SET)
1433 error (_("Second argument of 'IN' has wrong type"));
1434 if (TYPE_CODE (eltype) != TYPE_CODE_INT
1435 && TYPE_CODE (eltype) != TYPE_CODE_CHAR
1436 && TYPE_CODE (eltype) != TYPE_CODE_ENUM
1437 && TYPE_CODE (eltype) != TYPE_CODE_BOOL)
1438 error (_("First argument of 'IN' has wrong type"));
1439 member = value_bit_index (settype, value_contents (set),
1440 value_as_long (element));
1441 if (member < 0)
1442 error (_("First argument of 'IN' not in range"));
1443 return value_from_longest (LA_BOOL_TYPE, member);
1444 }
1445
1446 void
1447 _initialize_valarith (void)
1448 {
1449 }