| 1 | /* C preprocessor macro expansion for GDB. |
| 2 | Copyright 2002 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 2 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, write to the Free Software |
| 19 | Foundation, Inc., 59 Temple Place - Suite 330, |
| 20 | Boston, MA 02111-1307, USA. */ |
| 21 | |
| 22 | #include "defs.h" |
| 23 | #include "gdb_obstack.h" |
| 24 | #include "bcache.h" |
| 25 | #include "macrotab.h" |
| 26 | #include "macroexp.h" |
| 27 | #include "gdb_assert.h" |
| 28 | |
| 29 | |
| 30 | \f |
| 31 | /* A resizeable, substringable string type. */ |
| 32 | |
| 33 | |
| 34 | /* A string type that we can resize, quickly append to, and use to |
| 35 | refer to substrings of other strings. */ |
| 36 | struct macro_buffer |
| 37 | { |
| 38 | /* An array of characters. The first LEN bytes are the real text, |
| 39 | but there are SIZE bytes allocated to the array. If SIZE is |
| 40 | zero, then this doesn't point to a malloc'ed block. If SHARED is |
| 41 | non-zero, then this buffer is actually a pointer into some larger |
| 42 | string, and we shouldn't append characters to it, etc. Because |
| 43 | of sharing, we can't assume in general that the text is |
| 44 | null-terminated. */ |
| 45 | char *text; |
| 46 | |
| 47 | /* The number of characters in the string. */ |
| 48 | int len; |
| 49 | |
| 50 | /* The number of characters allocated to the string. If SHARED is |
| 51 | non-zero, this is meaningless; in this case, we set it to zero so |
| 52 | that any "do we have room to append something?" tests will fail, |
| 53 | so we don't always have to check SHARED before using this field. */ |
| 54 | int size; |
| 55 | |
| 56 | /* Zero if TEXT can be safely realloc'ed (i.e., it's its own malloc |
| 57 | block). Non-zero if TEXT is actually pointing into the middle of |
| 58 | some other block, and we shouldn't reallocate it. */ |
| 59 | int shared; |
| 60 | |
| 61 | /* For detecting token splicing. |
| 62 | |
| 63 | This is the index in TEXT of the first character of the token |
| 64 | that abuts the end of TEXT. If TEXT contains no tokens, then we |
| 65 | set this equal to LEN. If TEXT ends in whitespace, then there is |
| 66 | no token abutting the end of TEXT (it's just whitespace), and |
| 67 | again, we set this equal to LEN. We set this to -1 if we don't |
| 68 | know the nature of TEXT. */ |
| 69 | int last_token; |
| 70 | |
| 71 | /* If this buffer is holding the result from get_token, then this |
| 72 | is non-zero if it is an identifier token, zero otherwise. */ |
| 73 | int is_identifier; |
| 74 | }; |
| 75 | |
| 76 | |
| 77 | /* Set the macro buffer *B to the empty string, guessing that its |
| 78 | final contents will fit in N bytes. (It'll get resized if it |
| 79 | doesn't, so the guess doesn't have to be right.) Allocate the |
| 80 | initial storage with xmalloc. */ |
| 81 | static void |
| 82 | init_buffer (struct macro_buffer *b, int n) |
| 83 | { |
| 84 | /* Small value for initial testing. */ |
| 85 | n = 1; |
| 86 | |
| 87 | b->size = n; |
| 88 | if (n > 0) |
| 89 | b->text = (char *) xmalloc (n); |
| 90 | else |
| 91 | b->text = NULL; |
| 92 | b->len = 0; |
| 93 | b->shared = 0; |
| 94 | b->last_token = -1; |
| 95 | } |
| 96 | |
| 97 | |
| 98 | /* Set the macro buffer *BUF to refer to the LEN bytes at ADDR, as a |
| 99 | shared substring. */ |
| 100 | static void |
| 101 | init_shared_buffer (struct macro_buffer *buf, char *addr, int len) |
| 102 | { |
| 103 | buf->text = addr; |
| 104 | buf->len = len; |
| 105 | buf->shared = 1; |
| 106 | buf->size = 0; |
| 107 | buf->last_token = -1; |
| 108 | } |
| 109 | |
| 110 | |
| 111 | /* Free the text of the buffer B. Raise an error if B is shared. */ |
| 112 | static void |
| 113 | free_buffer (struct macro_buffer *b) |
| 114 | { |
| 115 | gdb_assert (! b->shared); |
| 116 | if (b->size) |
| 117 | xfree (b->text); |
| 118 | } |
| 119 | |
| 120 | |
| 121 | /* A cleanup function for macro buffers. */ |
| 122 | static void |
| 123 | cleanup_macro_buffer (void *untyped_buf) |
| 124 | { |
| 125 | free_buffer ((struct macro_buffer *) untyped_buf); |
| 126 | } |
| 127 | |
| 128 | |
| 129 | /* Resize the buffer B to be at least N bytes long. Raise an error if |
| 130 | B shouldn't be resized. */ |
| 131 | static void |
| 132 | resize_buffer (struct macro_buffer *b, int n) |
| 133 | { |
| 134 | /* We shouldn't be trying to resize shared strings. */ |
| 135 | gdb_assert (! b->shared); |
| 136 | |
| 137 | if (b->size == 0) |
| 138 | b->size = n; |
| 139 | else |
| 140 | while (b->size <= n) |
| 141 | b->size *= 2; |
| 142 | |
| 143 | b->text = xrealloc (b->text, b->size); |
| 144 | } |
| 145 | |
| 146 | |
| 147 | /* Append the character C to the buffer B. */ |
| 148 | static inline void |
| 149 | appendc (struct macro_buffer *b, int c) |
| 150 | { |
| 151 | int new_len = b->len + 1; |
| 152 | |
| 153 | if (new_len > b->size) |
| 154 | resize_buffer (b, new_len); |
| 155 | |
| 156 | b->text[b->len] = c; |
| 157 | b->len = new_len; |
| 158 | } |
| 159 | |
| 160 | |
| 161 | /* Append the LEN bytes at ADDR to the buffer B. */ |
| 162 | static inline void |
| 163 | appendmem (struct macro_buffer *b, char *addr, int len) |
| 164 | { |
| 165 | int new_len = b->len + len; |
| 166 | |
| 167 | if (new_len > b->size) |
| 168 | resize_buffer (b, new_len); |
| 169 | |
| 170 | memcpy (b->text + b->len, addr, len); |
| 171 | b->len = new_len; |
| 172 | } |
| 173 | |
| 174 | |
| 175 | \f |
| 176 | /* Recognizing preprocessor tokens. */ |
| 177 | |
| 178 | |
| 179 | static int |
| 180 | is_whitespace (int c) |
| 181 | { |
| 182 | return (c == ' ' |
| 183 | || c == '\t' |
| 184 | || c == '\n' |
| 185 | || c == '\v' |
| 186 | || c == '\f'); |
| 187 | } |
| 188 | |
| 189 | |
| 190 | static int |
| 191 | is_digit (int c) |
| 192 | { |
| 193 | return ('0' <= c && c <= '9'); |
| 194 | } |
| 195 | |
| 196 | |
| 197 | static int |
| 198 | is_identifier_nondigit (int c) |
| 199 | { |
| 200 | return (c == '_' |
| 201 | || ('a' <= c && c <= 'z') |
| 202 | || ('A' <= c && c <= 'Z')); |
| 203 | } |
| 204 | |
| 205 | |
| 206 | static void |
| 207 | set_token (struct macro_buffer *tok, char *start, char *end) |
| 208 | { |
| 209 | init_shared_buffer (tok, start, end - start); |
| 210 | tok->last_token = 0; |
| 211 | |
| 212 | /* Presumed; get_identifier may overwrite this. */ |
| 213 | tok->is_identifier = 0; |
| 214 | } |
| 215 | |
| 216 | |
| 217 | static int |
| 218 | get_comment (struct macro_buffer *tok, char *p, char *end) |
| 219 | { |
| 220 | if (p + 2 > end) |
| 221 | return 0; |
| 222 | else if (p[0] == '/' |
| 223 | && p[1] == '*') |
| 224 | { |
| 225 | char *tok_start = p; |
| 226 | |
| 227 | p += 2; |
| 228 | |
| 229 | for (; p < end; p++) |
| 230 | if (p + 2 <= end |
| 231 | && p[0] == '*' |
| 232 | && p[1] == '/') |
| 233 | { |
| 234 | p += 2; |
| 235 | set_token (tok, tok_start, p); |
| 236 | return 1; |
| 237 | } |
| 238 | |
| 239 | error ("Unterminated comment in macro expansion."); |
| 240 | } |
| 241 | else if (p[0] == '/' |
| 242 | && p[1] == '/') |
| 243 | { |
| 244 | char *tok_start = p; |
| 245 | |
| 246 | p += 2; |
| 247 | for (; p < end; p++) |
| 248 | if (*p == '\n') |
| 249 | break; |
| 250 | |
| 251 | set_token (tok, tok_start, p); |
| 252 | return 1; |
| 253 | } |
| 254 | else |
| 255 | return 0; |
| 256 | } |
| 257 | |
| 258 | |
| 259 | static int |
| 260 | get_identifier (struct macro_buffer *tok, char *p, char *end) |
| 261 | { |
| 262 | if (p < end |
| 263 | && is_identifier_nondigit (*p)) |
| 264 | { |
| 265 | char *tok_start = p; |
| 266 | |
| 267 | while (p < end |
| 268 | && (is_identifier_nondigit (*p) |
| 269 | || is_digit (*p))) |
| 270 | p++; |
| 271 | |
| 272 | set_token (tok, tok_start, p); |
| 273 | tok->is_identifier = 1; |
| 274 | return 1; |
| 275 | } |
| 276 | else |
| 277 | return 0; |
| 278 | } |
| 279 | |
| 280 | |
| 281 | static int |
| 282 | get_pp_number (struct macro_buffer *tok, char *p, char *end) |
| 283 | { |
| 284 | if (p < end |
| 285 | && (is_digit (*p) |
| 286 | || *p == '.')) |
| 287 | { |
| 288 | char *tok_start = p; |
| 289 | |
| 290 | while (p < end) |
| 291 | { |
| 292 | if (is_digit (*p) |
| 293 | || is_identifier_nondigit (*p) |
| 294 | || *p == '.') |
| 295 | p++; |
| 296 | else if (p + 2 <= end |
| 297 | && strchr ("eEpP.", *p) |
| 298 | && (p[1] == '+' || p[1] == '-')) |
| 299 | p += 2; |
| 300 | else |
| 301 | break; |
| 302 | } |
| 303 | |
| 304 | set_token (tok, tok_start, p); |
| 305 | return 1; |
| 306 | } |
| 307 | else |
| 308 | return 0; |
| 309 | } |
| 310 | |
| 311 | |
| 312 | |
| 313 | /* If the text starting at P going up to (but not including) END |
| 314 | starts with a character constant, set *TOK to point to that |
| 315 | character constant, and return 1. Otherwise, return zero. |
| 316 | Signal an error if it contains a malformed or incomplete character |
| 317 | constant. */ |
| 318 | static int |
| 319 | get_character_constant (struct macro_buffer *tok, char *p, char *end) |
| 320 | { |
| 321 | /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 |
| 322 | But of course, what really matters is that we handle it the same |
| 323 | way GDB's C/C++ lexer does. So we call parse_escape in utils.c |
| 324 | to handle escape sequences. */ |
| 325 | if ((p + 1 <= end && *p == '\'') |
| 326 | || (p + 2 <= end && p[0] == 'L' && p[1] == '\'')) |
| 327 | { |
| 328 | char *tok_start = p; |
| 329 | char *body_start; |
| 330 | |
| 331 | if (*p == '\'') |
| 332 | p++; |
| 333 | else if (*p == 'L') |
| 334 | p += 2; |
| 335 | else |
| 336 | gdb_assert (0); |
| 337 | |
| 338 | body_start = p; |
| 339 | for (;;) |
| 340 | { |
| 341 | if (p >= end) |
| 342 | error ("Unmatched single quote."); |
| 343 | else if (*p == '\'') |
| 344 | { |
| 345 | if (p == body_start) |
| 346 | error ("A character constant must contain at least one " |
| 347 | "character."); |
| 348 | p++; |
| 349 | break; |
| 350 | } |
| 351 | else if (*p == '\\') |
| 352 | { |
| 353 | p++; |
| 354 | parse_escape (&p); |
| 355 | } |
| 356 | else |
| 357 | p++; |
| 358 | } |
| 359 | |
| 360 | set_token (tok, tok_start, p); |
| 361 | return 1; |
| 362 | } |
| 363 | else |
| 364 | return 0; |
| 365 | } |
| 366 | |
| 367 | |
| 368 | /* If the text starting at P going up to (but not including) END |
| 369 | starts with a string literal, set *TOK to point to that string |
| 370 | literal, and return 1. Otherwise, return zero. Signal an error if |
| 371 | it contains a malformed or incomplete string literal. */ |
| 372 | static int |
| 373 | get_string_literal (struct macro_buffer *tok, char *p, char *end) |
| 374 | { |
| 375 | if ((p + 1 <= end |
| 376 | && *p == '\"') |
| 377 | || (p + 2 <= end |
| 378 | && p[0] == 'L' |
| 379 | && p[1] == '\"')) |
| 380 | { |
| 381 | char *tok_start = p; |
| 382 | |
| 383 | if (*p == '\"') |
| 384 | p++; |
| 385 | else if (*p == 'L') |
| 386 | p += 2; |
| 387 | else |
| 388 | gdb_assert (0); |
| 389 | |
| 390 | for (;;) |
| 391 | { |
| 392 | if (p >= end) |
| 393 | error ("Unterminated string in expression."); |
| 394 | else if (*p == '\"') |
| 395 | { |
| 396 | p++; |
| 397 | break; |
| 398 | } |
| 399 | else if (*p == '\n') |
| 400 | error ("Newline characters may not appear in string " |
| 401 | "constants."); |
| 402 | else if (*p == '\\') |
| 403 | { |
| 404 | p++; |
| 405 | parse_escape (&p); |
| 406 | } |
| 407 | else |
| 408 | p++; |
| 409 | } |
| 410 | |
| 411 | set_token (tok, tok_start, p); |
| 412 | return 1; |
| 413 | } |
| 414 | else |
| 415 | return 0; |
| 416 | } |
| 417 | |
| 418 | |
| 419 | static int |
| 420 | get_punctuator (struct macro_buffer *tok, char *p, char *end) |
| 421 | { |
| 422 | /* Here, speed is much less important than correctness and clarity. */ |
| 423 | |
| 424 | /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1 */ |
| 425 | static const char * const punctuators[] = { |
| 426 | "[", "]", "(", ")", "{", "}", ".", "->", |
| 427 | "++", "--", "&", "*", "+", "-", "~", "!", |
| 428 | "/", "%", "<<", ">>", "<", ">", "<=", ">=", "==", "!=", |
| 429 | "^", "|", "&&", "||", |
| 430 | "?", ":", ";", "...", |
| 431 | "=", "*=", "/=", "%=", "+=", "-=", "<<=", ">>=", "&=", "^=", "|=", |
| 432 | ",", "#", "##", |
| 433 | "<:", ":>", "<%", "%>", "%:", "%:%:", |
| 434 | 0 |
| 435 | }; |
| 436 | |
| 437 | int i; |
| 438 | |
| 439 | if (p + 1 <= end) |
| 440 | { |
| 441 | for (i = 0; punctuators[i]; i++) |
| 442 | { |
| 443 | const char *punctuator = punctuators[i]; |
| 444 | |
| 445 | if (p[0] == punctuator[0]) |
| 446 | { |
| 447 | int len = strlen (punctuator); |
| 448 | |
| 449 | if (p + len <= end |
| 450 | && ! memcmp (p, punctuator, len)) |
| 451 | { |
| 452 | set_token (tok, p, p + len); |
| 453 | return 1; |
| 454 | } |
| 455 | } |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | return 0; |
| 460 | } |
| 461 | |
| 462 | |
| 463 | /* Peel the next preprocessor token off of SRC, and put it in TOK. |
| 464 | Mutate TOK to refer to the first token in SRC, and mutate SRC to |
| 465 | refer to the text after that token. SRC must be a shared buffer; |
| 466 | the resulting TOK will be shared, pointing into the same string SRC |
| 467 | does. Initialize TOK's last_token field. Return non-zero if we |
| 468 | succeed, or 0 if we didn't find any more tokens in SRC. */ |
| 469 | static int |
| 470 | get_token (struct macro_buffer *tok, |
| 471 | struct macro_buffer *src) |
| 472 | { |
| 473 | char *p = src->text; |
| 474 | char *end = p + src->len; |
| 475 | |
| 476 | gdb_assert (src->shared); |
| 477 | |
| 478 | /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: |
| 479 | |
| 480 | preprocessing-token: |
| 481 | header-name |
| 482 | identifier |
| 483 | pp-number |
| 484 | character-constant |
| 485 | string-literal |
| 486 | punctuator |
| 487 | each non-white-space character that cannot be one of the above |
| 488 | |
| 489 | We don't have to deal with header-name tokens, since those can |
| 490 | only occur after a #include, which we will never see. */ |
| 491 | |
| 492 | while (p < end) |
| 493 | if (is_whitespace (*p)) |
| 494 | p++; |
| 495 | else if (get_comment (tok, p, end)) |
| 496 | p += tok->len; |
| 497 | else if (get_pp_number (tok, p, end) |
| 498 | || get_character_constant (tok, p, end) |
| 499 | || get_string_literal (tok, p, end) |
| 500 | /* Note: the grammar in the standard seems to be |
| 501 | ambiguous: L'x' can be either a wide character |
| 502 | constant, or an identifier followed by a normal |
| 503 | character constant. By trying `get_identifier' after |
| 504 | we try get_character_constant and get_string_literal, |
| 505 | we give the wide character syntax precedence. Now, |
| 506 | since GDB doesn't handle wide character constants |
| 507 | anyway, is this the right thing to do? */ |
| 508 | || get_identifier (tok, p, end) |
| 509 | || get_punctuator (tok, p, end)) |
| 510 | { |
| 511 | /* How many characters did we consume, including whitespace? */ |
| 512 | int consumed = p - src->text + tok->len; |
| 513 | src->text += consumed; |
| 514 | src->len -= consumed; |
| 515 | return 1; |
| 516 | } |
| 517 | else |
| 518 | { |
| 519 | /* We have found a "non-whitespace character that cannot be |
| 520 | one of the above." Make a token out of it. */ |
| 521 | int consumed; |
| 522 | |
| 523 | set_token (tok, p, p + 1); |
| 524 | consumed = p - src->text + tok->len; |
| 525 | src->text += consumed; |
| 526 | src->len -= consumed; |
| 527 | return 1; |
| 528 | } |
| 529 | |
| 530 | return 0; |
| 531 | } |
| 532 | |
| 533 | |
| 534 | \f |
| 535 | /* Appending token strings, with and without splicing */ |
| 536 | |
| 537 | |
| 538 | /* Append the macro buffer SRC to the end of DEST, and ensure that |
| 539 | doing so doesn't splice the token at the end of SRC with the token |
| 540 | at the beginning of DEST. SRC and DEST must have their last_token |
| 541 | fields set. Upon return, DEST's last_token field is set correctly. |
| 542 | |
| 543 | For example: |
| 544 | |
| 545 | If DEST is "(" and SRC is "y", then we can return with |
| 546 | DEST set to "(y" --- we've simply appended the two buffers. |
| 547 | |
| 548 | However, if DEST is "x" and SRC is "y", then we must not return |
| 549 | with DEST set to "xy" --- that would splice the two tokens "x" and |
| 550 | "y" together to make a single token "xy". However, it would be |
| 551 | fine to return with DEST set to "x y". Similarly, "<" and "<" must |
| 552 | yield "< <", not "<<", etc. */ |
| 553 | static void |
| 554 | append_tokens_without_splicing (struct macro_buffer *dest, |
| 555 | struct macro_buffer *src) |
| 556 | { |
| 557 | int original_dest_len = dest->len; |
| 558 | struct macro_buffer dest_tail, new_token; |
| 559 | |
| 560 | gdb_assert (src->last_token != -1); |
| 561 | gdb_assert (dest->last_token != -1); |
| 562 | |
| 563 | /* First, just try appending the two, and call get_token to see if |
| 564 | we got a splice. */ |
| 565 | appendmem (dest, src->text, src->len); |
| 566 | |
| 567 | /* If DEST originally had no token abutting its end, then we can't |
| 568 | have spliced anything, so we're done. */ |
| 569 | if (dest->last_token == original_dest_len) |
| 570 | { |
| 571 | dest->last_token = original_dest_len + src->last_token; |
| 572 | return; |
| 573 | } |
| 574 | |
| 575 | /* Set DEST_TAIL to point to the last token in DEST, followed by |
| 576 | all the stuff we just appended. */ |
| 577 | init_shared_buffer (&dest_tail, |
| 578 | dest->text + dest->last_token, |
| 579 | dest->len - dest->last_token); |
| 580 | |
| 581 | /* Re-parse DEST's last token. We know that DEST used to contain |
| 582 | at least one token, so if it doesn't contain any after the |
| 583 | append, then we must have spliced "/" and "*" or "/" and "/" to |
| 584 | make a comment start. (Just for the record, I got this right |
| 585 | the first time. This is not a bug fix.) */ |
| 586 | if (get_token (&new_token, &dest_tail) |
| 587 | && (new_token.text + new_token.len |
| 588 | == dest->text + original_dest_len)) |
| 589 | { |
| 590 | /* No splice, so we're done. */ |
| 591 | dest->last_token = original_dest_len + src->last_token; |
| 592 | return; |
| 593 | } |
| 594 | |
| 595 | /* Okay, a simple append caused a splice. Let's chop dest back to |
| 596 | its original length and try again, but separate the texts with a |
| 597 | space. */ |
| 598 | dest->len = original_dest_len; |
| 599 | appendc (dest, ' '); |
| 600 | appendmem (dest, src->text, src->len); |
| 601 | |
| 602 | init_shared_buffer (&dest_tail, |
| 603 | dest->text + dest->last_token, |
| 604 | dest->len - dest->last_token); |
| 605 | |
| 606 | /* Try to re-parse DEST's last token, as above. */ |
| 607 | if (get_token (&new_token, &dest_tail) |
| 608 | && (new_token.text + new_token.len |
| 609 | == dest->text + original_dest_len)) |
| 610 | { |
| 611 | /* No splice, so we're done. */ |
| 612 | dest->last_token = original_dest_len + 1 + src->last_token; |
| 613 | return; |
| 614 | } |
| 615 | |
| 616 | /* As far as I know, there's no case where inserting a space isn't |
| 617 | enough to prevent a splice. */ |
| 618 | internal_error (__FILE__, __LINE__, |
| 619 | "unable to avoid splicing tokens during macro expansion"); |
| 620 | } |
| 621 | |
| 622 | |
| 623 | \f |
| 624 | /* Expanding macros! */ |
| 625 | |
| 626 | |
| 627 | /* A singly-linked list of the names of the macros we are currently |
| 628 | expanding --- for detecting expansion loops. */ |
| 629 | struct macro_name_list { |
| 630 | const char *name; |
| 631 | struct macro_name_list *next; |
| 632 | }; |
| 633 | |
| 634 | |
| 635 | /* Return non-zero if we are currently expanding the macro named NAME, |
| 636 | according to LIST; otherwise, return zero. |
| 637 | |
| 638 | You know, it would be possible to get rid of all the NO_LOOP |
| 639 | arguments to these functions by simply generating a new lookup |
| 640 | function and baton which refuses to find the definition for a |
| 641 | particular macro, and otherwise delegates the decision to another |
| 642 | function/baton pair. But that makes the linked list of excluded |
| 643 | macros chained through untyped baton pointers, which will make it |
| 644 | harder to debug. :( */ |
| 645 | static int |
| 646 | currently_rescanning (struct macro_name_list *list, const char *name) |
| 647 | { |
| 648 | for (; list; list = list->next) |
| 649 | if (strcmp (name, list->name) == 0) |
| 650 | return 1; |
| 651 | |
| 652 | return 0; |
| 653 | } |
| 654 | |
| 655 | |
| 656 | /* Gather the arguments to a macro expansion. |
| 657 | |
| 658 | NAME is the name of the macro being invoked. (It's only used for |
| 659 | printing error messages.) |
| 660 | |
| 661 | Assume that SRC is the text of the macro invocation immediately |
| 662 | following the macro name. For example, if we're processing the |
| 663 | text foo(bar, baz), then NAME would be foo and SRC will be (bar, |
| 664 | baz). |
| 665 | |
| 666 | If SRC doesn't start with an open paren ( token at all, return |
| 667 | zero, leave SRC unchanged, and don't set *ARGC_P to anything. |
| 668 | |
| 669 | If SRC doesn't contain a properly terminated argument list, then |
| 670 | raise an error. |
| 671 | |
| 672 | Otherwise, return a pointer to the first element of an array of |
| 673 | macro buffers referring to the argument texts, and set *ARGC_P to |
| 674 | the number of arguments we found --- the number of elements in the |
| 675 | array. The macro buffers share their text with SRC, and their |
| 676 | last_token fields are initialized. The array is allocated with |
| 677 | xmalloc, and the caller is responsible for freeing it. |
| 678 | |
| 679 | NOTE WELL: if SRC starts with a open paren ( token followed |
| 680 | immediately by a close paren ) token (e.g., the invocation looks |
| 681 | like "foo()"), we treat that as one argument, which happens to be |
| 682 | the empty list of tokens. The caller should keep in mind that such |
| 683 | a sequence of tokens is a valid way to invoke one-parameter |
| 684 | function-like macros, but also a valid way to invoke zero-parameter |
| 685 | function-like macros. Eeew. |
| 686 | |
| 687 | Consume the tokens from SRC; after this call, SRC contains the text |
| 688 | following the invocation. */ |
| 689 | |
| 690 | static struct macro_buffer * |
| 691 | gather_arguments (const char *name, struct macro_buffer *src, int *argc_p) |
| 692 | { |
| 693 | struct macro_buffer tok; |
| 694 | int args_len, args_size; |
| 695 | struct macro_buffer *args = NULL; |
| 696 | struct cleanup *back_to = make_cleanup (free_current_contents, &args); |
| 697 | |
| 698 | /* Does SRC start with an opening paren token? Read from a copy of |
| 699 | SRC, so SRC itself is unaffected if we don't find an opening |
| 700 | paren. */ |
| 701 | { |
| 702 | struct macro_buffer temp; |
| 703 | init_shared_buffer (&temp, src->text, src->len); |
| 704 | |
| 705 | if (! get_token (&tok, &temp) |
| 706 | || tok.len != 1 |
| 707 | || tok.text[0] != '(') |
| 708 | { |
| 709 | discard_cleanups (back_to); |
| 710 | return 0; |
| 711 | } |
| 712 | } |
| 713 | |
| 714 | /* Consume SRC's opening paren. */ |
| 715 | get_token (&tok, src); |
| 716 | |
| 717 | args_len = 0; |
| 718 | args_size = 1; /* small for initial testing */ |
| 719 | args = (struct macro_buffer *) xmalloc (sizeof (*args) * args_size); |
| 720 | |
| 721 | for (;;) |
| 722 | { |
| 723 | struct macro_buffer *arg; |
| 724 | int depth; |
| 725 | |
| 726 | /* Make sure we have room for the next argument. */ |
| 727 | if (args_len >= args_size) |
| 728 | { |
| 729 | args_size *= 2; |
| 730 | args = xrealloc (args, sizeof (*args) * args_size); |
| 731 | } |
| 732 | |
| 733 | /* Initialize the next argument. */ |
| 734 | arg = &args[args_len++]; |
| 735 | set_token (arg, src->text, src->text); |
| 736 | |
| 737 | /* Gather the argument's tokens. */ |
| 738 | depth = 0; |
| 739 | for (;;) |
| 740 | { |
| 741 | char *start = src->text; |
| 742 | |
| 743 | if (! get_token (&tok, src)) |
| 744 | error ("Malformed argument list for macro `%s'.", name); |
| 745 | |
| 746 | /* Is tok an opening paren? */ |
| 747 | if (tok.len == 1 && tok.text[0] == '(') |
| 748 | depth++; |
| 749 | |
| 750 | /* Is tok is a closing paren? */ |
| 751 | else if (tok.len == 1 && tok.text[0] == ')') |
| 752 | { |
| 753 | /* If it's a closing paren at the top level, then that's |
| 754 | the end of the argument list. */ |
| 755 | if (depth == 0) |
| 756 | { |
| 757 | discard_cleanups (back_to); |
| 758 | *argc_p = args_len; |
| 759 | return args; |
| 760 | } |
| 761 | |
| 762 | depth--; |
| 763 | } |
| 764 | |
| 765 | /* If tok is a comma at top level, then that's the end of |
| 766 | the current argument. */ |
| 767 | else if (tok.len == 1 && tok.text[0] == ',' && depth == 0) |
| 768 | break; |
| 769 | |
| 770 | /* Extend the current argument to enclose this token. If |
| 771 | this is the current argument's first token, leave out any |
| 772 | leading whitespace, just for aesthetics. */ |
| 773 | if (arg->len == 0) |
| 774 | { |
| 775 | arg->text = tok.text; |
| 776 | arg->len = tok.len; |
| 777 | arg->last_token = 0; |
| 778 | } |
| 779 | else |
| 780 | { |
| 781 | arg->len = (tok.text + tok.len) - arg->text; |
| 782 | arg->last_token = tok.text - arg->text; |
| 783 | } |
| 784 | } |
| 785 | } |
| 786 | } |
| 787 | |
| 788 | |
| 789 | /* The `expand' and `substitute_args' functions both invoke `scan' |
| 790 | recursively, so we need a forward declaration somewhere. */ |
| 791 | static void scan (struct macro_buffer *dest, |
| 792 | struct macro_buffer *src, |
| 793 | struct macro_name_list *no_loop, |
| 794 | macro_lookup_ftype *lookup_func, |
| 795 | void *lookup_baton); |
| 796 | |
| 797 | |
| 798 | /* Given the macro definition DEF, being invoked with the actual |
| 799 | arguments given by ARGC and ARGV, substitute the arguments into the |
| 800 | replacement list, and store the result in DEST. |
| 801 | |
| 802 | If it is necessary to expand macro invocations in one of the |
| 803 | arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro |
| 804 | definitions, and don't expand invocations of the macros listed in |
| 805 | NO_LOOP. */ |
| 806 | static void |
| 807 | substitute_args (struct macro_buffer *dest, |
| 808 | struct macro_definition *def, |
| 809 | int argc, struct macro_buffer *argv, |
| 810 | struct macro_name_list *no_loop, |
| 811 | macro_lookup_ftype *lookup_func, |
| 812 | void *lookup_baton) |
| 813 | { |
| 814 | /* A macro buffer for the macro's replacement list. */ |
| 815 | struct macro_buffer replacement_list; |
| 816 | |
| 817 | init_shared_buffer (&replacement_list, (char *) def->replacement, |
| 818 | strlen (def->replacement)); |
| 819 | |
| 820 | gdb_assert (dest->len == 0); |
| 821 | dest->last_token = 0; |
| 822 | |
| 823 | for (;;) |
| 824 | { |
| 825 | struct macro_buffer tok; |
| 826 | char *original_rl_start = replacement_list.text; |
| 827 | int substituted = 0; |
| 828 | |
| 829 | /* Find the next token in the replacement list. */ |
| 830 | if (! get_token (&tok, &replacement_list)) |
| 831 | break; |
| 832 | |
| 833 | /* Just for aesthetics. If we skipped some whitespace, copy |
| 834 | that to DEST. */ |
| 835 | if (tok.text > original_rl_start) |
| 836 | { |
| 837 | appendmem (dest, original_rl_start, tok.text - original_rl_start); |
| 838 | dest->last_token = dest->len; |
| 839 | } |
| 840 | |
| 841 | /* Is this token the stringification operator? */ |
| 842 | if (tok.len == 1 |
| 843 | && tok.text[0] == '#') |
| 844 | error ("Stringification is not implemented yet."); |
| 845 | |
| 846 | /* Is this token the splicing operator? */ |
| 847 | if (tok.len == 2 |
| 848 | && tok.text[0] == '#' |
| 849 | && tok.text[1] == '#') |
| 850 | error ("Token splicing is not implemented yet."); |
| 851 | |
| 852 | /* Is this token an identifier? */ |
| 853 | if (tok.is_identifier) |
| 854 | { |
| 855 | int i; |
| 856 | |
| 857 | /* Is it the magic varargs parameter? */ |
| 858 | if (tok.len == 11 |
| 859 | && ! memcmp (tok.text, "__VA_ARGS__", 11)) |
| 860 | error ("Variable-arity macros not implemented yet."); |
| 861 | |
| 862 | /* Is it one of the parameters? */ |
| 863 | for (i = 0; i < def->argc; i++) |
| 864 | if (tok.len == strlen (def->argv[i]) |
| 865 | && ! memcmp (tok.text, def->argv[i], tok.len)) |
| 866 | { |
| 867 | struct macro_buffer arg_src; |
| 868 | |
| 869 | /* Expand any macro invocations in the argument text, |
| 870 | and append the result to dest. Remember that scan |
| 871 | mutates its source, so we need to scan a new buffer |
| 872 | referring to the argument's text, not the argument |
| 873 | itself. */ |
| 874 | init_shared_buffer (&arg_src, argv[i].text, argv[i].len); |
| 875 | scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); |
| 876 | substituted = 1; |
| 877 | break; |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | /* If it wasn't a parameter, then just copy it across. */ |
| 882 | if (! substituted) |
| 883 | append_tokens_without_splicing (dest, &tok); |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | |
| 888 | /* Expand a call to a macro named ID, whose definition is DEF. Append |
| 889 | its expansion to DEST. SRC is the input text following the ID |
| 890 | token. We are currently rescanning the expansions of the macros |
| 891 | named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and |
| 892 | LOOKUP_BATON to find definitions for any nested macro references. |
| 893 | |
| 894 | Return 1 if we decided to expand it, zero otherwise. (If it's a |
| 895 | function-like macro name that isn't followed by an argument list, |
| 896 | we don't expand it.) If we return zero, leave SRC unchanged. */ |
| 897 | static int |
| 898 | expand (const char *id, |
| 899 | struct macro_definition *def, |
| 900 | struct macro_buffer *dest, |
| 901 | struct macro_buffer *src, |
| 902 | struct macro_name_list *no_loop, |
| 903 | macro_lookup_ftype *lookup_func, |
| 904 | void *lookup_baton) |
| 905 | { |
| 906 | struct macro_name_list new_no_loop; |
| 907 | |
| 908 | /* Create a new node to be added to the front of the no-expand list. |
| 909 | This list is appropriate for re-scanning replacement lists, but |
| 910 | it is *not* appropriate for scanning macro arguments; invocations |
| 911 | of the macro whose arguments we are gathering *do* get expanded |
| 912 | there. */ |
| 913 | new_no_loop.name = id; |
| 914 | new_no_loop.next = no_loop; |
| 915 | |
| 916 | /* What kind of macro are we expanding? */ |
| 917 | if (def->kind == macro_object_like) |
| 918 | { |
| 919 | struct macro_buffer replacement_list; |
| 920 | |
| 921 | init_shared_buffer (&replacement_list, (char *) def->replacement, |
| 922 | strlen (def->replacement)); |
| 923 | |
| 924 | scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); |
| 925 | return 1; |
| 926 | } |
| 927 | else if (def->kind == macro_function_like) |
| 928 | { |
| 929 | struct cleanup *back_to = make_cleanup (null_cleanup, 0); |
| 930 | int argc; |
| 931 | struct macro_buffer *argv = NULL; |
| 932 | struct macro_buffer substituted; |
| 933 | struct macro_buffer substituted_src; |
| 934 | |
| 935 | if (def->argc >= 1 |
| 936 | && strcmp (def->argv[def->argc - 1], "...") == 0) |
| 937 | error ("Varargs macros not implemented yet."); |
| 938 | |
| 939 | make_cleanup (free_current_contents, &argv); |
| 940 | argv = gather_arguments (id, src, &argc); |
| 941 | |
| 942 | /* If we couldn't find any argument list, then we don't expand |
| 943 | this macro. */ |
| 944 | if (! argv) |
| 945 | { |
| 946 | do_cleanups (back_to); |
| 947 | return 0; |
| 948 | } |
| 949 | |
| 950 | /* Check that we're passing an acceptable number of arguments for |
| 951 | this macro. */ |
| 952 | if (argc != def->argc) |
| 953 | { |
| 954 | /* Remember that a sequence of tokens like "foo()" is a |
| 955 | valid invocation of a macro expecting either zero or one |
| 956 | arguments. */ |
| 957 | if (! (argc == 1 |
| 958 | && argv[0].len == 0 |
| 959 | && def->argc == 0)) |
| 960 | error ("Wrong number of arguments to macro `%s' " |
| 961 | "(expected %d, got %d).", |
| 962 | id, def->argc, argc); |
| 963 | } |
| 964 | |
| 965 | /* Note that we don't expand macro invocations in the arguments |
| 966 | yet --- we let subst_args take care of that. Parameters that |
| 967 | appear as operands of the stringifying operator "#" or the |
| 968 | splicing operator "##" don't get macro references expanded, |
| 969 | so we can't really tell whether it's appropriate to macro- |
| 970 | expand an argument until we see how it's being used. */ |
| 971 | init_buffer (&substituted, 0); |
| 972 | make_cleanup (cleanup_macro_buffer, &substituted); |
| 973 | substitute_args (&substituted, def, argc, argv, no_loop, |
| 974 | lookup_func, lookup_baton); |
| 975 | |
| 976 | /* Now `substituted' is the macro's replacement list, with all |
| 977 | argument values substituted into it properly. Re-scan it for |
| 978 | macro references, but don't expand invocations of this macro. |
| 979 | |
| 980 | We create a new buffer, `substituted_src', which points into |
| 981 | `substituted', and scan that. We can't scan `substituted' |
| 982 | itself, since the tokenization process moves the buffer's |
| 983 | text pointer around, and we still need to be able to find |
| 984 | `substituted's original text buffer after scanning it so we |
| 985 | can free it. */ |
| 986 | init_shared_buffer (&substituted_src, substituted.text, substituted.len); |
| 987 | scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); |
| 988 | |
| 989 | do_cleanups (back_to); |
| 990 | |
| 991 | return 1; |
| 992 | } |
| 993 | else |
| 994 | internal_error (__FILE__, __LINE__, "bad macro definition kind"); |
| 995 | } |
| 996 | |
| 997 | |
| 998 | /* If the single token in SRC_FIRST followed by the tokens in SRC_REST |
| 999 | constitute a macro invokation not forbidden in NO_LOOP, append its |
| 1000 | expansion to DEST and return non-zero. Otherwise, return zero, and |
| 1001 | leave DEST unchanged. |
| 1002 | |
| 1003 | SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. |
| 1004 | SRC_FIRST must be a string built by get_token. */ |
| 1005 | static int |
| 1006 | maybe_expand (struct macro_buffer *dest, |
| 1007 | struct macro_buffer *src_first, |
| 1008 | struct macro_buffer *src_rest, |
| 1009 | struct macro_name_list *no_loop, |
| 1010 | macro_lookup_ftype *lookup_func, |
| 1011 | void *lookup_baton) |
| 1012 | { |
| 1013 | gdb_assert (src_first->shared); |
| 1014 | gdb_assert (src_rest->shared); |
| 1015 | gdb_assert (! dest->shared); |
| 1016 | |
| 1017 | /* Is this token an identifier? */ |
| 1018 | if (src_first->is_identifier) |
| 1019 | { |
| 1020 | /* Make a null-terminated copy of it, since that's what our |
| 1021 | lookup function expects. */ |
| 1022 | char *id = xmalloc (src_first->len + 1); |
| 1023 | struct cleanup *back_to = make_cleanup (xfree, id); |
| 1024 | memcpy (id, src_first->text, src_first->len); |
| 1025 | id[src_first->len] = 0; |
| 1026 | |
| 1027 | /* If we're currently re-scanning the result of expanding |
| 1028 | this macro, don't expand it again. */ |
| 1029 | if (! currently_rescanning (no_loop, id)) |
| 1030 | { |
| 1031 | /* Does this identifier have a macro definition in scope? */ |
| 1032 | struct macro_definition *def = lookup_func (id, lookup_baton); |
| 1033 | |
| 1034 | if (def && expand (id, def, dest, src_rest, no_loop, |
| 1035 | lookup_func, lookup_baton)) |
| 1036 | { |
| 1037 | do_cleanups (back_to); |
| 1038 | return 1; |
| 1039 | } |
| 1040 | } |
| 1041 | |
| 1042 | do_cleanups (back_to); |
| 1043 | } |
| 1044 | |
| 1045 | return 0; |
| 1046 | } |
| 1047 | |
| 1048 | |
| 1049 | /* Expand macro references in SRC, appending the results to DEST. |
| 1050 | Assume we are re-scanning the result of expanding the macros named |
| 1051 | in NO_LOOP, and don't try to re-expand references to them. |
| 1052 | |
| 1053 | SRC must be a shared buffer; DEST must not be one. */ |
| 1054 | static void |
| 1055 | scan (struct macro_buffer *dest, |
| 1056 | struct macro_buffer *src, |
| 1057 | struct macro_name_list *no_loop, |
| 1058 | macro_lookup_ftype *lookup_func, |
| 1059 | void *lookup_baton) |
| 1060 | { |
| 1061 | gdb_assert (src->shared); |
| 1062 | gdb_assert (! dest->shared); |
| 1063 | |
| 1064 | for (;;) |
| 1065 | { |
| 1066 | struct macro_buffer tok; |
| 1067 | char *original_src_start = src->text; |
| 1068 | |
| 1069 | /* Find the next token in SRC. */ |
| 1070 | if (! get_token (&tok, src)) |
| 1071 | break; |
| 1072 | |
| 1073 | /* Just for aesthetics. If we skipped some whitespace, copy |
| 1074 | that to DEST. */ |
| 1075 | if (tok.text > original_src_start) |
| 1076 | { |
| 1077 | appendmem (dest, original_src_start, tok.text - original_src_start); |
| 1078 | dest->last_token = dest->len; |
| 1079 | } |
| 1080 | |
| 1081 | if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) |
| 1082 | /* We didn't end up expanding tok as a macro reference, so |
| 1083 | simply append it to dest. */ |
| 1084 | append_tokens_without_splicing (dest, &tok); |
| 1085 | } |
| 1086 | |
| 1087 | /* Just for aesthetics. If there was any trailing whitespace in |
| 1088 | src, copy it to dest. */ |
| 1089 | if (src->len) |
| 1090 | { |
| 1091 | appendmem (dest, src->text, src->len); |
| 1092 | dest->last_token = dest->len; |
| 1093 | } |
| 1094 | } |
| 1095 | |
| 1096 | |
| 1097 | char * |
| 1098 | macro_expand (const char *source, |
| 1099 | macro_lookup_ftype *lookup_func, |
| 1100 | void *lookup_func_baton) |
| 1101 | { |
| 1102 | struct macro_buffer src, dest; |
| 1103 | struct cleanup *back_to; |
| 1104 | |
| 1105 | init_shared_buffer (&src, (char *) source, strlen (source)); |
| 1106 | |
| 1107 | init_buffer (&dest, 0); |
| 1108 | dest.last_token = 0; |
| 1109 | back_to = make_cleanup (cleanup_macro_buffer, &dest); |
| 1110 | |
| 1111 | scan (&dest, &src, 0, lookup_func, lookup_func_baton); |
| 1112 | |
| 1113 | appendc (&dest, '\0'); |
| 1114 | |
| 1115 | discard_cleanups (back_to); |
| 1116 | return dest.text; |
| 1117 | } |
| 1118 | |
| 1119 | |
| 1120 | char * |
| 1121 | macro_expand_once (const char *source, |
| 1122 | macro_lookup_ftype *lookup_func, |
| 1123 | void *lookup_func_baton) |
| 1124 | { |
| 1125 | error ("Expand-once not implemented yet."); |
| 1126 | } |
| 1127 | |
| 1128 | |
| 1129 | char * |
| 1130 | macro_expand_next (char **lexptr, |
| 1131 | macro_lookup_ftype *lookup_func, |
| 1132 | void *lookup_baton) |
| 1133 | { |
| 1134 | struct macro_buffer src, dest, tok; |
| 1135 | struct cleanup *back_to; |
| 1136 | |
| 1137 | /* Set up SRC to refer to the input text, pointed to by *lexptr. */ |
| 1138 | init_shared_buffer (&src, *lexptr, strlen (*lexptr)); |
| 1139 | |
| 1140 | /* Set up DEST to receive the expansion, if there is one. */ |
| 1141 | init_buffer (&dest, 0); |
| 1142 | dest.last_token = 0; |
| 1143 | back_to = make_cleanup (cleanup_macro_buffer, &dest); |
| 1144 | |
| 1145 | /* Get the text's first preprocessing token. */ |
| 1146 | if (! get_token (&tok, &src)) |
| 1147 | { |
| 1148 | do_cleanups (back_to); |
| 1149 | return 0; |
| 1150 | } |
| 1151 | |
| 1152 | /* If it's a macro invocation, expand it. */ |
| 1153 | if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) |
| 1154 | { |
| 1155 | /* It was a macro invocation! Package up the expansion as a |
| 1156 | null-terminated string and return it. Set *lexptr to the |
| 1157 | start of the next token in the input. */ |
| 1158 | appendc (&dest, '\0'); |
| 1159 | discard_cleanups (back_to); |
| 1160 | *lexptr = src.text; |
| 1161 | return dest.text; |
| 1162 | } |
| 1163 | else |
| 1164 | { |
| 1165 | /* It wasn't a macro invocation. */ |
| 1166 | do_cleanups (back_to); |
| 1167 | return 0; |
| 1168 | } |
| 1169 | } |