1 /* C preprocessor macro expansion for GDB. 2 Copyright (C) 2002, 2007, 2008, 2009 Free Software Foundation, Inc. 3 Contributed by Red Hat, Inc. 4 5 This file is part of GDB. 6 7 This program is free software; you can redistribute it and/or modify 8 it under the terms of the GNU General Public License as published by 9 the Free Software Foundation; either version 3 of the License, or 10 (at your option) any later version. 11 12 This program is distributed in the hope that it will be useful, 13 but WITHOUT ANY WARRANTY; without even the implied warranty of 14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 15 GNU General Public License for more details. 16 17 You should have received a copy of the GNU General Public License 18 along with this program. If not, see <http://www.gnu.org/licenses/>. */ 19 20 #include "defs.h" 21 #include "gdb_obstack.h" 22 #include "bcache.h" 23 #include "macrotab.h" 24 #include "macroexp.h" 25 #include "gdb_assert.h" 26 #include "c-lang.h" 27 28 29 30 /* A resizeable, substringable string type. */ 31 32 33 /* A string type that we can resize, quickly append to, and use to 34 refer to substrings of other strings. */ 35 struct macro_buffer 36 { 37 /* An array of characters. The first LEN bytes are the real text, 38 but there are SIZE bytes allocated to the array. If SIZE is 39 zero, then this doesn't point to a malloc'ed block. If SHARED is 40 non-zero, then this buffer is actually a pointer into some larger 41 string, and we shouldn't append characters to it, etc. Because 42 of sharing, we can't assume in general that the text is 43 null-terminated. */ 44 char *text; 45 46 /* The number of characters in the string. */ 47 int len; 48 49 /* The number of characters allocated to the string. If SHARED is 50 non-zero, this is meaningless; in this case, we set it to zero so 51 that any "do we have room to append something?" tests will fail, 52 so we don't always have to check SHARED before using this field. */ 53 int size; 54 55 /* Zero if TEXT can be safely realloc'ed (i.e., it's its own malloc 56 block). Non-zero if TEXT is actually pointing into the middle of 57 some other block, and we shouldn't reallocate it. */ 58 int shared; 59 60 /* For detecting token splicing. 61 62 This is the index in TEXT of the first character of the token 63 that abuts the end of TEXT. If TEXT contains no tokens, then we 64 set this equal to LEN. If TEXT ends in whitespace, then there is 65 no token abutting the end of TEXT (it's just whitespace), and 66 again, we set this equal to LEN. We set this to -1 if we don't 67 know the nature of TEXT. */ 68 int last_token; 69 70 /* If this buffer is holding the result from get_token, then this 71 is non-zero if it is an identifier token, zero otherwise. */ 72 int is_identifier; 73 }; 74 75 76 /* Set the macro buffer *B to the empty string, guessing that its 77 final contents will fit in N bytes. (It'll get resized if it 78 doesn't, so the guess doesn't have to be right.) Allocate the 79 initial storage with xmalloc. */ 80 static void 81 init_buffer (struct macro_buffer *b, int n) 82 { 83 b->size = n; 84 if (n > 0) 85 b->text = (char *) xmalloc (n); 86 else 87 b->text = NULL; 88 b->len = 0; 89 b->shared = 0; 90 b->last_token = -1; 91 } 92 93 94 /* Set the macro buffer *BUF to refer to the LEN bytes at ADDR, as a 95 shared substring. */ 96 static void 97 init_shared_buffer (struct macro_buffer *buf, char *addr, int len) 98 { 99 buf->text = addr; 100 buf->len = len; 101 buf->shared = 1; 102 buf->size = 0; 103 buf->last_token = -1; 104 } 105 106 107 /* Free the text of the buffer B. Raise an error if B is shared. */ 108 static void 109 free_buffer (struct macro_buffer *b) 110 { 111 gdb_assert (! b->shared); 112 if (b->size) 113 xfree (b->text); 114 } 115 116 117 /* A cleanup function for macro buffers. */ 118 static void 119 cleanup_macro_buffer (void *untyped_buf) 120 { 121 free_buffer ((struct macro_buffer *) untyped_buf); 122 } 123 124 125 /* Resize the buffer B to be at least N bytes long. Raise an error if 126 B shouldn't be resized. */ 127 static void 128 resize_buffer (struct macro_buffer *b, int n) 129 { 130 /* We shouldn't be trying to resize shared strings. */ 131 gdb_assert (! b->shared); 132 133 if (b->size == 0) 134 b->size = n; 135 else 136 while (b->size <= n) 137 b->size *= 2; 138 139 b->text = xrealloc (b->text, b->size); 140 } 141 142 143 /* Append the character C to the buffer B. */ 144 static void 145 appendc (struct macro_buffer *b, int c) 146 { 147 int new_len = b->len + 1; 148 149 if (new_len > b->size) 150 resize_buffer (b, new_len); 151 152 b->text[b->len] = c; 153 b->len = new_len; 154 } 155 156 157 /* Append the LEN bytes at ADDR to the buffer B. */ 158 static void 159 appendmem (struct macro_buffer *b, char *addr, int len) 160 { 161 int new_len = b->len + len; 162 163 if (new_len > b->size) 164 resize_buffer (b, new_len); 165 166 memcpy (b->text + b->len, addr, len); 167 b->len = new_len; 168 } 169 170 171 172 /* Recognizing preprocessor tokens. */ 173 174 175 int 176 macro_is_whitespace (int c) 177 { 178 return (c == ' ' 179 || c == '\t' 180 || c == '\n' 181 || c == '\v' 182 || c == '\f'); 183 } 184 185 186 int 187 macro_is_digit (int c) 188 { 189 return ('0' <= c && c <= '9'); 190 } 191 192 193 int 194 macro_is_identifier_nondigit (int c) 195 { 196 return (c == '_' 197 || ('a' <= c && c <= 'z') 198 || ('A' <= c && c <= 'Z')); 199 } 200 201 202 static void 203 set_token (struct macro_buffer *tok, char *start, char *end) 204 { 205 init_shared_buffer (tok, start, end - start); 206 tok->last_token = 0; 207 208 /* Presumed; get_identifier may overwrite this. */ 209 tok->is_identifier = 0; 210 } 211 212 213 static int 214 get_comment (struct macro_buffer *tok, char *p, char *end) 215 { 216 if (p + 2 > end) 217 return 0; 218 else if (p[0] == '/' 219 && p[1] == '*') 220 { 221 char *tok_start = p; 222 223 p += 2; 224 225 for (; p < end; p++) 226 if (p + 2 <= end 227 && p[0] == '*' 228 && p[1] == '/') 229 { 230 p += 2; 231 set_token (tok, tok_start, p); 232 return 1; 233 } 234 235 error (_("Unterminated comment in macro expansion.")); 236 } 237 else if (p[0] == '/' 238 && p[1] == '/') 239 { 240 char *tok_start = p; 241 242 p += 2; 243 for (; p < end; p++) 244 if (*p == '\n') 245 break; 246 247 set_token (tok, tok_start, p); 248 return 1; 249 } 250 else 251 return 0; 252 } 253 254 255 static int 256 get_identifier (struct macro_buffer *tok, char *p, char *end) 257 { 258 if (p < end 259 && macro_is_identifier_nondigit (*p)) 260 { 261 char *tok_start = p; 262 263 while (p < end 264 && (macro_is_identifier_nondigit (*p) 265 || macro_is_digit (*p))) 266 p++; 267 268 set_token (tok, tok_start, p); 269 tok->is_identifier = 1; 270 return 1; 271 } 272 else 273 return 0; 274 } 275 276 277 static int 278 get_pp_number (struct macro_buffer *tok, char *p, char *end) 279 { 280 if (p < end 281 && (macro_is_digit (*p) 282 || (*p == '.' 283 && p + 2 <= end 284 && macro_is_digit (p[1])))) 285 { 286 char *tok_start = p; 287 288 while (p < end) 289 { 290 if (p + 2 <= end 291 && strchr ("eEpP", *p) 292 && (p[1] == '+' || p[1] == '-')) 293 p += 2; 294 else if (macro_is_digit (*p) 295 || macro_is_identifier_nondigit (*p) 296 || *p == '.') 297 p++; 298 else 299 break; 300 } 301 302 set_token (tok, tok_start, p); 303 return 1; 304 } 305 else 306 return 0; 307 } 308 309 310 311 /* If the text starting at P going up to (but not including) END 312 starts with a character constant, set *TOK to point to that 313 character constant, and return 1. Otherwise, return zero. 314 Signal an error if it contains a malformed or incomplete character 315 constant. */ 316 static int 317 get_character_constant (struct macro_buffer *tok, char *p, char *end) 318 { 319 /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 320 But of course, what really matters is that we handle it the same 321 way GDB's C/C++ lexer does. So we call parse_escape in utils.c 322 to handle escape sequences. */ 323 if ((p + 1 <= end && *p == '\'') 324 || (p + 2 <= end 325 && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') 326 && p[1] == '\'')) 327 { 328 char *tok_start = p; 329 char *body_start; 330 int char_count = 0; 331 332 if (*p == '\'') 333 p++; 334 else if (*p == 'L' || *p == 'u' || *p == 'U') 335 p += 2; 336 else 337 gdb_assert (0); 338 339 body_start = p; 340 for (;;) 341 { 342 if (p >= end) 343 error (_("Unmatched single quote.")); 344 else if (*p == '\'') 345 { 346 if (!char_count) 347 error (_("A character constant must contain at least one " 348 "character.")); 349 p++; 350 break; 351 } 352 else if (*p == '\\') 353 { 354 p++; 355 char_count += c_parse_escape (&p, NULL); 356 } 357 else 358 { 359 p++; 360 char_count++; 361 } 362 } 363 364 set_token (tok, tok_start, p); 365 return 1; 366 } 367 else 368 return 0; 369 } 370 371 372 /* If the text starting at P going up to (but not including) END 373 starts with a string literal, set *TOK to point to that string 374 literal, and return 1. Otherwise, return zero. Signal an error if 375 it contains a malformed or incomplete string literal. */ 376 static int 377 get_string_literal (struct macro_buffer *tok, char *p, char *end) 378 { 379 if ((p + 1 <= end 380 && *p == '"') 381 || (p + 2 <= end 382 && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') 383 && p[1] == '"')) 384 { 385 char *tok_start = p; 386 387 if (*p == '"') 388 p++; 389 else if (*p == 'L' || *p == 'u' || *p == 'U') 390 p += 2; 391 else 392 gdb_assert (0); 393 394 for (;;) 395 { 396 if (p >= end) 397 error (_("Unterminated string in expression.")); 398 else if (*p == '"') 399 { 400 p++; 401 break; 402 } 403 else if (*p == '\n') 404 error (_("Newline characters may not appear in string " 405 "constants.")); 406 else if (*p == '\\') 407 { 408 p++; 409 c_parse_escape (&p, NULL); 410 } 411 else 412 p++; 413 } 414 415 set_token (tok, tok_start, p); 416 return 1; 417 } 418 else 419 return 0; 420 } 421 422 423 static int 424 get_punctuator (struct macro_buffer *tok, char *p, char *end) 425 { 426 /* Here, speed is much less important than correctness and clarity. */ 427 428 /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1. 429 Note that this table is ordered in a special way. A punctuator 430 which is a prefix of another punctuator must appear after its 431 "extension". Otherwise, the wrong token will be returned. */ 432 static const char * const punctuators[] = { 433 "[", "]", "(", ")", "{", "}", "?", ";", ",", "~", 434 "...", ".", 435 "->", "--", "-=", "-", 436 "++", "+=", "+", 437 "*=", "*", 438 "!=", "!", 439 "&&", "&=", "&", 440 "/=", "/", 441 "%>", "%:%:", "%:", "%=", "%", 442 "^=", "^", 443 "##", "#", 444 ":>", ":", 445 "||", "|=", "|", 446 "<<=", "<<", "<=", "<:", "<%", "<", 447 ">>=", ">>", ">=", ">", 448 "==", "=", 449 0 450 }; 451 452 int i; 453 454 if (p + 1 <= end) 455 { 456 for (i = 0; punctuators[i]; i++) 457 { 458 const char *punctuator = punctuators[i]; 459 460 if (p[0] == punctuator[0]) 461 { 462 int len = strlen (punctuator); 463 464 if (p + len <= end 465 && ! memcmp (p, punctuator, len)) 466 { 467 set_token (tok, p, p + len); 468 return 1; 469 } 470 } 471 } 472 } 473 474 return 0; 475 } 476 477 478 /* Peel the next preprocessor token off of SRC, and put it in TOK. 479 Mutate TOK to refer to the first token in SRC, and mutate SRC to 480 refer to the text after that token. SRC must be a shared buffer; 481 the resulting TOK will be shared, pointing into the same string SRC 482 does. Initialize TOK's last_token field. Return non-zero if we 483 succeed, or 0 if we didn't find any more tokens in SRC. */ 484 static int 485 get_token (struct macro_buffer *tok, 486 struct macro_buffer *src) 487 { 488 char *p = src->text; 489 char *end = p + src->len; 490 491 gdb_assert (src->shared); 492 493 /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: 494 495 preprocessing-token: 496 header-name 497 identifier 498 pp-number 499 character-constant 500 string-literal 501 punctuator 502 each non-white-space character that cannot be one of the above 503 504 We don't have to deal with header-name tokens, since those can 505 only occur after a #include, which we will never see. */ 506 507 while (p < end) 508 if (macro_is_whitespace (*p)) 509 p++; 510 else if (get_comment (tok, p, end)) 511 p += tok->len; 512 else if (get_pp_number (tok, p, end) 513 || get_character_constant (tok, p, end) 514 || get_string_literal (tok, p, end) 515 /* Note: the grammar in the standard seems to be 516 ambiguous: L'x' can be either a wide character 517 constant, or an identifier followed by a normal 518 character constant. By trying `get_identifier' after 519 we try get_character_constant and get_string_literal, 520 we give the wide character syntax precedence. Now, 521 since GDB doesn't handle wide character constants 522 anyway, is this the right thing to do? */ 523 || get_identifier (tok, p, end) 524 || get_punctuator (tok, p, end)) 525 { 526 /* How many characters did we consume, including whitespace? */ 527 int consumed = p - src->text + tok->len; 528 src->text += consumed; 529 src->len -= consumed; 530 return 1; 531 } 532 else 533 { 534 /* We have found a "non-whitespace character that cannot be 535 one of the above." Make a token out of it. */ 536 int consumed; 537 538 set_token (tok, p, p + 1); 539 consumed = p - src->text + tok->len; 540 src->text += consumed; 541 src->len -= consumed; 542 return 1; 543 } 544 545 return 0; 546 } 547 548 549 550 /* Appending token strings, with and without splicing */ 551 552 553 /* Append the macro buffer SRC to the end of DEST, and ensure that 554 doing so doesn't splice the token at the end of SRC with the token 555 at the beginning of DEST. SRC and DEST must have their last_token 556 fields set. Upon return, DEST's last_token field is set correctly. 557 558 For example: 559 560 If DEST is "(" and SRC is "y", then we can return with 561 DEST set to "(y" --- we've simply appended the two buffers. 562 563 However, if DEST is "x" and SRC is "y", then we must not return 564 with DEST set to "xy" --- that would splice the two tokens "x" and 565 "y" together to make a single token "xy". However, it would be 566 fine to return with DEST set to "x y". Similarly, "<" and "<" must 567 yield "< <", not "<<", etc. */ 568 static void 569 append_tokens_without_splicing (struct macro_buffer *dest, 570 struct macro_buffer *src) 571 { 572 int original_dest_len = dest->len; 573 struct macro_buffer dest_tail, new_token; 574 575 gdb_assert (src->last_token != -1); 576 gdb_assert (dest->last_token != -1); 577 578 /* First, just try appending the two, and call get_token to see if 579 we got a splice. */ 580 appendmem (dest, src->text, src->len); 581 582 /* If DEST originally had no token abutting its end, then we can't 583 have spliced anything, so we're done. */ 584 if (dest->last_token == original_dest_len) 585 { 586 dest->last_token = original_dest_len + src->last_token; 587 return; 588 } 589 590 /* Set DEST_TAIL to point to the last token in DEST, followed by 591 all the stuff we just appended. */ 592 init_shared_buffer (&dest_tail, 593 dest->text + dest->last_token, 594 dest->len - dest->last_token); 595 596 /* Re-parse DEST's last token. We know that DEST used to contain 597 at least one token, so if it doesn't contain any after the 598 append, then we must have spliced "/" and "*" or "/" and "/" to 599 make a comment start. (Just for the record, I got this right 600 the first time. This is not a bug fix.) */ 601 if (get_token (&new_token, &dest_tail) 602 && (new_token.text + new_token.len 603 == dest->text + original_dest_len)) 604 { 605 /* No splice, so we're done. */ 606 dest->last_token = original_dest_len + src->last_token; 607 return; 608 } 609 610 /* Okay, a simple append caused a splice. Let's chop dest back to 611 its original length and try again, but separate the texts with a 612 space. */ 613 dest->len = original_dest_len; 614 appendc (dest, ' '); 615 appendmem (dest, src->text, src->len); 616 617 init_shared_buffer (&dest_tail, 618 dest->text + dest->last_token, 619 dest->len - dest->last_token); 620 621 /* Try to re-parse DEST's last token, as above. */ 622 if (get_token (&new_token, &dest_tail) 623 && (new_token.text + new_token.len 624 == dest->text + original_dest_len)) 625 { 626 /* No splice, so we're done. */ 627 dest->last_token = original_dest_len + 1 + src->last_token; 628 return; 629 } 630 631 /* As far as I know, there's no case where inserting a space isn't 632 enough to prevent a splice. */ 633 internal_error (__FILE__, __LINE__, 634 _("unable to avoid splicing tokens during macro expansion")); 635 } 636 637 /* Stringify an argument, and insert it into DEST. ARG is the text to 638 stringify; it is LEN bytes long. */ 639 640 static void 641 stringify (struct macro_buffer *dest, char *arg, int len) 642 { 643 /* Trim initial whitespace from ARG. */ 644 while (len > 0 && macro_is_whitespace (*arg)) 645 { 646 ++arg; 647 --len; 648 } 649 650 /* Trim trailing whitespace from ARG. */ 651 while (len > 0 && macro_is_whitespace (arg[len - 1])) 652 --len; 653 654 /* Insert the string. */ 655 appendc (dest, '"'); 656 while (len > 0) 657 { 658 /* We could try to handle strange cases here, like control 659 characters, but there doesn't seem to be much point. */ 660 if (macro_is_whitespace (*arg)) 661 { 662 /* Replace a sequence of whitespace with a single space. */ 663 appendc (dest, ' '); 664 while (len > 1 && macro_is_whitespace (arg[1])) 665 { 666 ++arg; 667 --len; 668 } 669 } 670 else if (*arg == '\\' || *arg == '"') 671 { 672 appendc (dest, '\\'); 673 appendc (dest, *arg); 674 } 675 else 676 appendc (dest, *arg); 677 ++arg; 678 --len; 679 } 680 appendc (dest, '"'); 681 dest->last_token = dest->len; 682 } 683 684 685 /* Expanding macros! */ 686 687 688 /* A singly-linked list of the names of the macros we are currently 689 expanding --- for detecting expansion loops. */ 690 struct macro_name_list { 691 const char *name; 692 struct macro_name_list *next; 693 }; 694 695 696 /* Return non-zero if we are currently expanding the macro named NAME, 697 according to LIST; otherwise, return zero. 698 699 You know, it would be possible to get rid of all the NO_LOOP 700 arguments to these functions by simply generating a new lookup 701 function and baton which refuses to find the definition for a 702 particular macro, and otherwise delegates the decision to another 703 function/baton pair. But that makes the linked list of excluded 704 macros chained through untyped baton pointers, which will make it 705 harder to debug. :( */ 706 static int 707 currently_rescanning (struct macro_name_list *list, const char *name) 708 { 709 for (; list; list = list->next) 710 if (strcmp (name, list->name) == 0) 711 return 1; 712 713 return 0; 714 } 715 716 717 /* Gather the arguments to a macro expansion. 718 719 NAME is the name of the macro being invoked. (It's only used for 720 printing error messages.) 721 722 Assume that SRC is the text of the macro invocation immediately 723 following the macro name. For example, if we're processing the 724 text foo(bar, baz), then NAME would be foo and SRC will be (bar, 725 baz). 726 727 If SRC doesn't start with an open paren ( token at all, return 728 zero, leave SRC unchanged, and don't set *ARGC_P to anything. 729 730 If SRC doesn't contain a properly terminated argument list, then 731 raise an error. 732 733 For a variadic macro, NARGS holds the number of formal arguments to 734 the macro. For a GNU-style variadic macro, this should be the 735 number of named arguments. For a non-variadic macro, NARGS should 736 be -1. 737 738 Otherwise, return a pointer to the first element of an array of 739 macro buffers referring to the argument texts, and set *ARGC_P to 740 the number of arguments we found --- the number of elements in the 741 array. The macro buffers share their text with SRC, and their 742 last_token fields are initialized. The array is allocated with 743 xmalloc, and the caller is responsible for freeing it. 744 745 NOTE WELL: if SRC starts with a open paren ( token followed 746 immediately by a close paren ) token (e.g., the invocation looks 747 like "foo()"), we treat that as one argument, which happens to be 748 the empty list of tokens. The caller should keep in mind that such 749 a sequence of tokens is a valid way to invoke one-parameter 750 function-like macros, but also a valid way to invoke zero-parameter 751 function-like macros. Eeew. 752 753 Consume the tokens from SRC; after this call, SRC contains the text 754 following the invocation. */ 755 756 static struct macro_buffer * 757 gather_arguments (const char *name, struct macro_buffer *src, 758 int nargs, int *argc_p) 759 { 760 struct macro_buffer tok; 761 int args_len, args_size; 762 struct macro_buffer *args = NULL; 763 struct cleanup *back_to = make_cleanup (free_current_contents, &args); 764 765 /* Does SRC start with an opening paren token? Read from a copy of 766 SRC, so SRC itself is unaffected if we don't find an opening 767 paren. */ 768 { 769 struct macro_buffer temp; 770 init_shared_buffer (&temp, src->text, src->len); 771 772 if (! get_token (&tok, &temp) 773 || tok.len != 1 774 || tok.text[0] != '(') 775 { 776 discard_cleanups (back_to); 777 return 0; 778 } 779 } 780 781 /* Consume SRC's opening paren. */ 782 get_token (&tok, src); 783 784 args_len = 0; 785 args_size = 6; 786 args = (struct macro_buffer *) xmalloc (sizeof (*args) * args_size); 787 788 for (;;) 789 { 790 struct macro_buffer *arg; 791 int depth; 792 793 /* Make sure we have room for the next argument. */ 794 if (args_len >= args_size) 795 { 796 args_size *= 2; 797 args = xrealloc (args, sizeof (*args) * args_size); 798 } 799 800 /* Initialize the next argument. */ 801 arg = &args[args_len++]; 802 set_token (arg, src->text, src->text); 803 804 /* Gather the argument's tokens. */ 805 depth = 0; 806 for (;;) 807 { 808 char *start = src->text; 809 810 if (! get_token (&tok, src)) 811 error (_("Malformed argument list for macro `%s'."), name); 812 813 /* Is tok an opening paren? */ 814 if (tok.len == 1 && tok.text[0] == '(') 815 depth++; 816 817 /* Is tok is a closing paren? */ 818 else if (tok.len == 1 && tok.text[0] == ')') 819 { 820 /* If it's a closing paren at the top level, then that's 821 the end of the argument list. */ 822 if (depth == 0) 823 { 824 /* In the varargs case, the last argument may be 825 missing. Add an empty argument in this case. */ 826 if (nargs != -1 && args_len == nargs - 1) 827 { 828 /* Make sure we have room for the argument. */ 829 if (args_len >= args_size) 830 { 831 args_size++; 832 args = xrealloc (args, sizeof (*args) * args_size); 833 } 834 arg = &args[args_len++]; 835 set_token (arg, src->text, src->text); 836 } 837 838 discard_cleanups (back_to); 839 *argc_p = args_len; 840 return args; 841 } 842 843 depth--; 844 } 845 846 /* If tok is a comma at top level, then that's the end of 847 the current argument. However, if we are handling a 848 variadic macro and we are computing the last argument, we 849 want to include the comma and remaining tokens. */ 850 else if (tok.len == 1 && tok.text[0] == ',' && depth == 0 851 && (nargs == -1 || args_len < nargs)) 852 break; 853 854 /* Extend the current argument to enclose this token. If 855 this is the current argument's first token, leave out any 856 leading whitespace, just for aesthetics. */ 857 if (arg->len == 0) 858 { 859 arg->text = tok.text; 860 arg->len = tok.len; 861 arg->last_token = 0; 862 } 863 else 864 { 865 arg->len = (tok.text + tok.len) - arg->text; 866 arg->last_token = tok.text - arg->text; 867 } 868 } 869 } 870 } 871 872 873 /* The `expand' and `substitute_args' functions both invoke `scan' 874 recursively, so we need a forward declaration somewhere. */ 875 static void scan (struct macro_buffer *dest, 876 struct macro_buffer *src, 877 struct macro_name_list *no_loop, 878 macro_lookup_ftype *lookup_func, 879 void *lookup_baton); 880 881 882 /* A helper function for substitute_args. 883 884 ARGV is a vector of all the arguments; ARGC is the number of 885 arguments. IS_VARARGS is true if the macro being substituted is a 886 varargs macro; in this case VA_ARG_NAME is the name of the 887 "variable" argument. VA_ARG_NAME is ignored if IS_VARARGS is 888 false. 889 890 If the token TOK is the name of a parameter, return the parameter's 891 index. If TOK is not an argument, return -1. */ 892 893 static int 894 find_parameter (const struct macro_buffer *tok, 895 int is_varargs, const struct macro_buffer *va_arg_name, 896 int argc, const char * const *argv) 897 { 898 int i; 899 900 if (! tok->is_identifier) 901 return -1; 902 903 for (i = 0; i < argc; ++i) 904 if (tok->len == strlen (argv[i]) && ! memcmp (tok->text, argv[i], tok->len)) 905 return i; 906 907 if (is_varargs && tok->len == va_arg_name->len 908 && ! memcmp (tok->text, va_arg_name->text, tok->len)) 909 return argc - 1; 910 911 return -1; 912 } 913 914 /* Given the macro definition DEF, being invoked with the actual 915 arguments given by ARGC and ARGV, substitute the arguments into the 916 replacement list, and store the result in DEST. 917 918 IS_VARARGS should be true if DEF is a varargs macro. In this case, 919 VA_ARG_NAME should be the name of the "variable" argument -- either 920 __VA_ARGS__ for c99-style varargs, or the final argument name, for 921 GNU-style varargs. If IS_VARARGS is false, this parameter is 922 ignored. 923 924 If it is necessary to expand macro invocations in one of the 925 arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro 926 definitions, and don't expand invocations of the macros listed in 927 NO_LOOP. */ 928 929 static void 930 substitute_args (struct macro_buffer *dest, 931 struct macro_definition *def, 932 int is_varargs, const struct macro_buffer *va_arg_name, 933 int argc, struct macro_buffer *argv, 934 struct macro_name_list *no_loop, 935 macro_lookup_ftype *lookup_func, 936 void *lookup_baton) 937 { 938 /* A macro buffer for the macro's replacement list. */ 939 struct macro_buffer replacement_list; 940 /* The token we are currently considering. */ 941 struct macro_buffer tok; 942 /* The replacement list's pointer from just before TOK was lexed. */ 943 char *original_rl_start; 944 /* We have a single lookahead token to handle token splicing. */ 945 struct macro_buffer lookahead; 946 /* The lookahead token might not be valid. */ 947 int lookahead_valid; 948 /* The replacement list's pointer from just before LOOKAHEAD was 949 lexed. */ 950 char *lookahead_rl_start; 951 952 init_shared_buffer (&replacement_list, (char *) def->replacement, 953 strlen (def->replacement)); 954 955 gdb_assert (dest->len == 0); 956 dest->last_token = 0; 957 958 original_rl_start = replacement_list.text; 959 if (! get_token (&tok, &replacement_list)) 960 return; 961 lookahead_rl_start = replacement_list.text; 962 lookahead_valid = get_token (&lookahead, &replacement_list); 963 964 for (;;) 965 { 966 /* Just for aesthetics. If we skipped some whitespace, copy 967 that to DEST. */ 968 if (tok.text > original_rl_start) 969 { 970 appendmem (dest, original_rl_start, tok.text - original_rl_start); 971 dest->last_token = dest->len; 972 } 973 974 /* Is this token the stringification operator? */ 975 if (tok.len == 1 976 && tok.text[0] == '#') 977 { 978 int arg; 979 980 if (!lookahead_valid) 981 error (_("Stringification operator requires an argument.")); 982 983 arg = find_parameter (&lookahead, is_varargs, va_arg_name, 984 def->argc, def->argv); 985 if (arg == -1) 986 error (_("Argument to stringification operator must name " 987 "a macro parameter.")); 988 989 stringify (dest, argv[arg].text, argv[arg].len); 990 991 /* Read one token and let the loop iteration code handle the 992 rest. */ 993 lookahead_rl_start = replacement_list.text; 994 lookahead_valid = get_token (&lookahead, &replacement_list); 995 } 996 /* Is this token the splicing operator? */ 997 else if (tok.len == 2 998 && tok.text[0] == '#' 999 && tok.text[1] == '#') 1000 error (_("Stray splicing operator")); 1001 /* Is the next token the splicing operator? */ 1002 else if (lookahead_valid 1003 && lookahead.len == 2 1004 && lookahead.text[0] == '#' 1005 && lookahead.text[1] == '#') 1006 { 1007 int arg, finished = 0; 1008 int prev_was_comma = 0; 1009 1010 /* Note that GCC warns if the result of splicing is not a 1011 token. In the debugger there doesn't seem to be much 1012 benefit from doing this. */ 1013 1014 /* Insert the first token. */ 1015 if (tok.len == 1 && tok.text[0] == ',') 1016 prev_was_comma = 1; 1017 else 1018 { 1019 int arg = find_parameter (&tok, is_varargs, va_arg_name, 1020 def->argc, def->argv); 1021 if (arg != -1) 1022 appendmem (dest, argv[arg].text, argv[arg].len); 1023 else 1024 appendmem (dest, tok.text, tok.len); 1025 } 1026 1027 /* Apply a possible sequence of ## operators. */ 1028 for (;;) 1029 { 1030 if (! get_token (&tok, &replacement_list)) 1031 error (_("Splicing operator at end of macro")); 1032 1033 /* Handle a comma before a ##. If we are handling 1034 varargs, and the token on the right hand side is the 1035 varargs marker, and the final argument is empty or 1036 missing, then drop the comma. This is a GNU 1037 extension. There is one ambiguous case here, 1038 involving pedantic behavior with an empty argument, 1039 but we settle that in favor of GNU-style (GCC uses an 1040 option). If we aren't dealing with varargs, we 1041 simply insert the comma. */ 1042 if (prev_was_comma) 1043 { 1044 if (! (is_varargs 1045 && tok.len == va_arg_name->len 1046 && !memcmp (tok.text, va_arg_name->text, tok.len) 1047 && argv[argc - 1].len == 0)) 1048 appendmem (dest, ",", 1); 1049 prev_was_comma = 0; 1050 } 1051 1052 /* Insert the token. If it is a parameter, insert the 1053 argument. If it is a comma, treat it specially. */ 1054 if (tok.len == 1 && tok.text[0] == ',') 1055 prev_was_comma = 1; 1056 else 1057 { 1058 int arg = find_parameter (&tok, is_varargs, va_arg_name, 1059 def->argc, def->argv); 1060 if (arg != -1) 1061 appendmem (dest, argv[arg].text, argv[arg].len); 1062 else 1063 appendmem (dest, tok.text, tok.len); 1064 } 1065 1066 /* Now read another token. If it is another splice, we 1067 loop. */ 1068 original_rl_start = replacement_list.text; 1069 if (! get_token (&tok, &replacement_list)) 1070 { 1071 finished = 1; 1072 break; 1073 } 1074 1075 if (! (tok.len == 2 1076 && tok.text[0] == '#' 1077 && tok.text[1] == '#')) 1078 break; 1079 } 1080 1081 if (prev_was_comma) 1082 { 1083 /* We saw a comma. Insert it now. */ 1084 appendmem (dest, ",", 1); 1085 } 1086 1087 dest->last_token = dest->len; 1088 if (finished) 1089 lookahead_valid = 0; 1090 else 1091 { 1092 /* Set up for the loop iterator. */ 1093 lookahead = tok; 1094 lookahead_rl_start = original_rl_start; 1095 lookahead_valid = 1; 1096 } 1097 } 1098 else 1099 { 1100 /* Is this token an identifier? */ 1101 int substituted = 0; 1102 int arg = find_parameter (&tok, is_varargs, va_arg_name, 1103 def->argc, def->argv); 1104 1105 if (arg != -1) 1106 { 1107 struct macro_buffer arg_src; 1108 1109 /* Expand any macro invocations in the argument text, 1110 and append the result to dest. Remember that scan 1111 mutates its source, so we need to scan a new buffer 1112 referring to the argument's text, not the argument 1113 itself. */ 1114 init_shared_buffer (&arg_src, argv[arg].text, argv[arg].len); 1115 scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); 1116 substituted = 1; 1117 } 1118 1119 /* If it wasn't a parameter, then just copy it across. */ 1120 if (! substituted) 1121 append_tokens_without_splicing (dest, &tok); 1122 } 1123 1124 if (! lookahead_valid) 1125 break; 1126 1127 tok = lookahead; 1128 original_rl_start = lookahead_rl_start; 1129 1130 lookahead_rl_start = replacement_list.text; 1131 lookahead_valid = get_token (&lookahead, &replacement_list); 1132 } 1133 } 1134 1135 1136 /* Expand a call to a macro named ID, whose definition is DEF. Append 1137 its expansion to DEST. SRC is the input text following the ID 1138 token. We are currently rescanning the expansions of the macros 1139 named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and 1140 LOOKUP_BATON to find definitions for any nested macro references. 1141 1142 Return 1 if we decided to expand it, zero otherwise. (If it's a 1143 function-like macro name that isn't followed by an argument list, 1144 we don't expand it.) If we return zero, leave SRC unchanged. */ 1145 static int 1146 expand (const char *id, 1147 struct macro_definition *def, 1148 struct macro_buffer *dest, 1149 struct macro_buffer *src, 1150 struct macro_name_list *no_loop, 1151 macro_lookup_ftype *lookup_func, 1152 void *lookup_baton) 1153 { 1154 struct macro_name_list new_no_loop; 1155 1156 /* Create a new node to be added to the front of the no-expand list. 1157 This list is appropriate for re-scanning replacement lists, but 1158 it is *not* appropriate for scanning macro arguments; invocations 1159 of the macro whose arguments we are gathering *do* get expanded 1160 there. */ 1161 new_no_loop.name = id; 1162 new_no_loop.next = no_loop; 1163 1164 /* What kind of macro are we expanding? */ 1165 if (def->kind == macro_object_like) 1166 { 1167 struct macro_buffer replacement_list; 1168 1169 init_shared_buffer (&replacement_list, (char *) def->replacement, 1170 strlen (def->replacement)); 1171 1172 scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); 1173 return 1; 1174 } 1175 else if (def->kind == macro_function_like) 1176 { 1177 struct cleanup *back_to = make_cleanup (null_cleanup, 0); 1178 int argc = 0; 1179 struct macro_buffer *argv = NULL; 1180 struct macro_buffer substituted; 1181 struct macro_buffer substituted_src; 1182 struct macro_buffer va_arg_name; 1183 int is_varargs = 0; 1184 1185 if (def->argc >= 1) 1186 { 1187 if (strcmp (def->argv[def->argc - 1], "...") == 0) 1188 { 1189 /* In C99-style varargs, substitution is done using 1190 __VA_ARGS__. */ 1191 init_shared_buffer (&va_arg_name, "__VA_ARGS__", 1192 strlen ("__VA_ARGS__")); 1193 is_varargs = 1; 1194 } 1195 else 1196 { 1197 int len = strlen (def->argv[def->argc - 1]); 1198 if (len > 3 1199 && strcmp (def->argv[def->argc - 1] + len - 3, "...") == 0) 1200 { 1201 /* In GNU-style varargs, the name of the 1202 substitution parameter is the name of the formal 1203 argument without the "...". */ 1204 init_shared_buffer (&va_arg_name, 1205 (char *) def->argv[def->argc - 1], 1206 len - 3); 1207 is_varargs = 1; 1208 } 1209 } 1210 } 1211 1212 make_cleanup (free_current_contents, &argv); 1213 argv = gather_arguments (id, src, is_varargs ? def->argc : -1, 1214 &argc); 1215 1216 /* If we couldn't find any argument list, then we don't expand 1217 this macro. */ 1218 if (! argv) 1219 { 1220 do_cleanups (back_to); 1221 return 0; 1222 } 1223 1224 /* Check that we're passing an acceptable number of arguments for 1225 this macro. */ 1226 if (argc != def->argc) 1227 { 1228 if (is_varargs && argc >= def->argc - 1) 1229 { 1230 /* Ok. */ 1231 } 1232 /* Remember that a sequence of tokens like "foo()" is a 1233 valid invocation of a macro expecting either zero or one 1234 arguments. */ 1235 else if (! (argc == 1 1236 && argv[0].len == 0 1237 && def->argc == 0)) 1238 error (_("Wrong number of arguments to macro `%s' " 1239 "(expected %d, got %d)."), 1240 id, def->argc, argc); 1241 } 1242 1243 /* Note that we don't expand macro invocations in the arguments 1244 yet --- we let subst_args take care of that. Parameters that 1245 appear as operands of the stringifying operator "#" or the 1246 splicing operator "##" don't get macro references expanded, 1247 so we can't really tell whether it's appropriate to macro- 1248 expand an argument until we see how it's being used. */ 1249 init_buffer (&substituted, 0); 1250 make_cleanup (cleanup_macro_buffer, &substituted); 1251 substitute_args (&substituted, def, is_varargs, &va_arg_name, 1252 argc, argv, no_loop, lookup_func, lookup_baton); 1253 1254 /* Now `substituted' is the macro's replacement list, with all 1255 argument values substituted into it properly. Re-scan it for 1256 macro references, but don't expand invocations of this macro. 1257 1258 We create a new buffer, `substituted_src', which points into 1259 `substituted', and scan that. We can't scan `substituted' 1260 itself, since the tokenization process moves the buffer's 1261 text pointer around, and we still need to be able to find 1262 `substituted's original text buffer after scanning it so we 1263 can free it. */ 1264 init_shared_buffer (&substituted_src, substituted.text, substituted.len); 1265 scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); 1266 1267 do_cleanups (back_to); 1268 1269 return 1; 1270 } 1271 else 1272 internal_error (__FILE__, __LINE__, _("bad macro definition kind")); 1273 } 1274 1275 1276 /* If the single token in SRC_FIRST followed by the tokens in SRC_REST 1277 constitute a macro invokation not forbidden in NO_LOOP, append its 1278 expansion to DEST and return non-zero. Otherwise, return zero, and 1279 leave DEST unchanged. 1280 1281 SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. 1282 SRC_FIRST must be a string built by get_token. */ 1283 static int 1284 maybe_expand (struct macro_buffer *dest, 1285 struct macro_buffer *src_first, 1286 struct macro_buffer *src_rest, 1287 struct macro_name_list *no_loop, 1288 macro_lookup_ftype *lookup_func, 1289 void *lookup_baton) 1290 { 1291 gdb_assert (src_first->shared); 1292 gdb_assert (src_rest->shared); 1293 gdb_assert (! dest->shared); 1294 1295 /* Is this token an identifier? */ 1296 if (src_first->is_identifier) 1297 { 1298 /* Make a null-terminated copy of it, since that's what our 1299 lookup function expects. */ 1300 char *id = xmalloc (src_first->len + 1); 1301 struct cleanup *back_to = make_cleanup (xfree, id); 1302 memcpy (id, src_first->text, src_first->len); 1303 id[src_first->len] = 0; 1304 1305 /* If we're currently re-scanning the result of expanding 1306 this macro, don't expand it again. */ 1307 if (! currently_rescanning (no_loop, id)) 1308 { 1309 /* Does this identifier have a macro definition in scope? */ 1310 struct macro_definition *def = lookup_func (id, lookup_baton); 1311 1312 if (def && expand (id, def, dest, src_rest, no_loop, 1313 lookup_func, lookup_baton)) 1314 { 1315 do_cleanups (back_to); 1316 return 1; 1317 } 1318 } 1319 1320 do_cleanups (back_to); 1321 } 1322 1323 return 0; 1324 } 1325 1326 1327 /* Expand macro references in SRC, appending the results to DEST. 1328 Assume we are re-scanning the result of expanding the macros named 1329 in NO_LOOP, and don't try to re-expand references to them. 1330 1331 SRC must be a shared buffer; DEST must not be one. */ 1332 static void 1333 scan (struct macro_buffer *dest, 1334 struct macro_buffer *src, 1335 struct macro_name_list *no_loop, 1336 macro_lookup_ftype *lookup_func, 1337 void *lookup_baton) 1338 { 1339 gdb_assert (src->shared); 1340 gdb_assert (! dest->shared); 1341 1342 for (;;) 1343 { 1344 struct macro_buffer tok; 1345 char *original_src_start = src->text; 1346 1347 /* Find the next token in SRC. */ 1348 if (! get_token (&tok, src)) 1349 break; 1350 1351 /* Just for aesthetics. If we skipped some whitespace, copy 1352 that to DEST. */ 1353 if (tok.text > original_src_start) 1354 { 1355 appendmem (dest, original_src_start, tok.text - original_src_start); 1356 dest->last_token = dest->len; 1357 } 1358 1359 if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) 1360 /* We didn't end up expanding tok as a macro reference, so 1361 simply append it to dest. */ 1362 append_tokens_without_splicing (dest, &tok); 1363 } 1364 1365 /* Just for aesthetics. If there was any trailing whitespace in 1366 src, copy it to dest. */ 1367 if (src->len) 1368 { 1369 appendmem (dest, src->text, src->len); 1370 dest->last_token = dest->len; 1371 } 1372 } 1373 1374 1375 char * 1376 macro_expand (const char *source, 1377 macro_lookup_ftype *lookup_func, 1378 void *lookup_func_baton) 1379 { 1380 struct macro_buffer src, dest; 1381 struct cleanup *back_to; 1382 1383 init_shared_buffer (&src, (char *) source, strlen (source)); 1384 1385 init_buffer (&dest, 0); 1386 dest.last_token = 0; 1387 back_to = make_cleanup (cleanup_macro_buffer, &dest); 1388 1389 scan (&dest, &src, 0, lookup_func, lookup_func_baton); 1390 1391 appendc (&dest, '\0'); 1392 1393 discard_cleanups (back_to); 1394 return dest.text; 1395 } 1396 1397 1398 char * 1399 macro_expand_once (const char *source, 1400 macro_lookup_ftype *lookup_func, 1401 void *lookup_func_baton) 1402 { 1403 error (_("Expand-once not implemented yet.")); 1404 } 1405 1406 1407 char * 1408 macro_expand_next (char **lexptr, 1409 macro_lookup_ftype *lookup_func, 1410 void *lookup_baton) 1411 { 1412 struct macro_buffer src, dest, tok; 1413 struct cleanup *back_to; 1414 1415 /* Set up SRC to refer to the input text, pointed to by *lexptr. */ 1416 init_shared_buffer (&src, *lexptr, strlen (*lexptr)); 1417 1418 /* Set up DEST to receive the expansion, if there is one. */ 1419 init_buffer (&dest, 0); 1420 dest.last_token = 0; 1421 back_to = make_cleanup (cleanup_macro_buffer, &dest); 1422 1423 /* Get the text's first preprocessing token. */ 1424 if (! get_token (&tok, &src)) 1425 { 1426 do_cleanups (back_to); 1427 return 0; 1428 } 1429 1430 /* If it's a macro invocation, expand it. */ 1431 if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) 1432 { 1433 /* It was a macro invocation! Package up the expansion as a 1434 null-terminated string and return it. Set *lexptr to the 1435 start of the next token in the input. */ 1436 appendc (&dest, '\0'); 1437 discard_cleanups (back_to); 1438 *lexptr = src.text; 1439 return dest.text; 1440 } 1441 else 1442 { 1443 /* It wasn't a macro invocation. */ 1444 do_cleanups (back_to); 1445 return 0; 1446 } 1447 } 1448