| 1 | /* C preprocessor macro expansion for GDB. |
| 2 | Copyright (C) 2002-2019 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 "macrotab.h" |
| 23 | #include "macroexp.h" |
| 24 | #include "c-lang.h" |
| 25 | |
| 26 | |
| 27 | \f |
| 28 | /* A resizeable, substringable string type. */ |
| 29 | |
| 30 | |
| 31 | /* A string type that we can resize, quickly append to, and use to |
| 32 | refer to substrings of other strings. */ |
| 33 | struct macro_buffer |
| 34 | { |
| 35 | /* An array of characters. The first LEN bytes are the real text, |
| 36 | but there are SIZE bytes allocated to the array. If SIZE is |
| 37 | zero, then this doesn't point to a malloc'ed block. If SHARED is |
| 38 | non-zero, then this buffer is actually a pointer into some larger |
| 39 | string, and we shouldn't append characters to it, etc. Because |
| 40 | of sharing, we can't assume in general that the text is |
| 41 | null-terminated. */ |
| 42 | char *text; |
| 43 | |
| 44 | /* The number of characters in the string. */ |
| 45 | int len; |
| 46 | |
| 47 | /* The number of characters allocated to the string. If SHARED is |
| 48 | non-zero, this is meaningless; in this case, we set it to zero so |
| 49 | that any "do we have room to append something?" tests will fail, |
| 50 | so we don't always have to check SHARED before using this field. */ |
| 51 | int size; |
| 52 | |
| 53 | /* Zero if TEXT can be safely realloc'ed (i.e., it's its own malloc |
| 54 | block). Non-zero if TEXT is actually pointing into the middle of |
| 55 | some other block, or to a string literal, and we shouldn't |
| 56 | reallocate it. */ |
| 57 | bool shared; |
| 58 | |
| 59 | /* For detecting token splicing. |
| 60 | |
| 61 | This is the index in TEXT of the first character of the token |
| 62 | that abuts the end of TEXT. If TEXT contains no tokens, then we |
| 63 | set this equal to LEN. If TEXT ends in whitespace, then there is |
| 64 | no token abutting the end of TEXT (it's just whitespace), and |
| 65 | again, we set this equal to LEN. We set this to -1 if we don't |
| 66 | know the nature of TEXT. */ |
| 67 | int last_token = -1; |
| 68 | |
| 69 | /* If this buffer is holding the result from get_token, then this |
| 70 | is non-zero if it is an identifier token, zero otherwise. */ |
| 71 | int is_identifier = 0; |
| 72 | |
| 73 | |
| 74 | macro_buffer () |
| 75 | : text (NULL), |
| 76 | len (0), |
| 77 | size (0), |
| 78 | shared (false) |
| 79 | { |
| 80 | } |
| 81 | |
| 82 | /* Set the macro buffer to the empty string, guessing that its |
| 83 | final contents will fit in N bytes. (It'll get resized if it |
| 84 | doesn't, so the guess doesn't have to be right.) Allocate the |
| 85 | initial storage with xmalloc. */ |
| 86 | explicit macro_buffer (int n) |
| 87 | : len (0), |
| 88 | size (n), |
| 89 | shared (false) |
| 90 | { |
| 91 | if (n > 0) |
| 92 | text = (char *) xmalloc (n); |
| 93 | else |
| 94 | text = NULL; |
| 95 | } |
| 96 | |
| 97 | /* Set the macro buffer to refer to the LEN bytes at ADDR, as a |
| 98 | shared substring. */ |
| 99 | macro_buffer (const char *addr, int len) |
| 100 | { |
| 101 | set_shared (addr, len); |
| 102 | } |
| 103 | |
| 104 | /* Set the macro buffer to refer to the LEN bytes at ADDR, as a |
| 105 | shared substring. */ |
| 106 | void set_shared (const char *addr, int len_) |
| 107 | { |
| 108 | text = (char *) addr; |
| 109 | len = len_; |
| 110 | size = 0; |
| 111 | shared = true; |
| 112 | } |
| 113 | |
| 114 | macro_buffer& operator= (const macro_buffer &src) |
| 115 | { |
| 116 | gdb_assert (src.shared); |
| 117 | gdb_assert (shared); |
| 118 | set_shared (src.text, src.len); |
| 119 | last_token = src.last_token; |
| 120 | is_identifier = src.is_identifier; |
| 121 | return *this; |
| 122 | } |
| 123 | |
| 124 | ~macro_buffer () |
| 125 | { |
| 126 | if (! shared && size) |
| 127 | xfree (text); |
| 128 | } |
| 129 | |
| 130 | /* Release the text of the buffer to the caller, which is now |
| 131 | responsible for freeing it. */ |
| 132 | ATTRIBUTE_UNUSED_RESULT char *release () |
| 133 | { |
| 134 | gdb_assert (! shared); |
| 135 | gdb_assert (size); |
| 136 | char *result = text; |
| 137 | text = NULL; |
| 138 | return result; |
| 139 | } |
| 140 | |
| 141 | /* Resize the buffer to be at least N bytes long. Raise an error if |
| 142 | the buffer shouldn't be resized. */ |
| 143 | void resize_buffer (int n) |
| 144 | { |
| 145 | /* We shouldn't be trying to resize shared strings. */ |
| 146 | gdb_assert (! shared); |
| 147 | |
| 148 | if (size == 0) |
| 149 | size = n; |
| 150 | else |
| 151 | while (size <= n) |
| 152 | size *= 2; |
| 153 | |
| 154 | text = (char *) xrealloc (text, size); |
| 155 | } |
| 156 | |
| 157 | /* Append the character C to the buffer. */ |
| 158 | void appendc (int c) |
| 159 | { |
| 160 | int new_len = len + 1; |
| 161 | |
| 162 | if (new_len > size) |
| 163 | resize_buffer (new_len); |
| 164 | |
| 165 | text[len] = c; |
| 166 | len = new_len; |
| 167 | } |
| 168 | |
| 169 | /* Append the COUNT bytes at ADDR to the buffer. */ |
| 170 | void appendmem (const char *addr, int count) |
| 171 | { |
| 172 | int new_len = len + count; |
| 173 | |
| 174 | if (new_len > size) |
| 175 | resize_buffer (new_len); |
| 176 | |
| 177 | memcpy (text + len, addr, count); |
| 178 | len = new_len; |
| 179 | } |
| 180 | }; |
| 181 | |
| 182 | |
| 183 | \f |
| 184 | /* Recognizing preprocessor tokens. */ |
| 185 | |
| 186 | |
| 187 | int |
| 188 | macro_is_whitespace (int c) |
| 189 | { |
| 190 | return (c == ' ' |
| 191 | || c == '\t' |
| 192 | || c == '\n' |
| 193 | || c == '\v' |
| 194 | || c == '\f'); |
| 195 | } |
| 196 | |
| 197 | |
| 198 | int |
| 199 | macro_is_digit (int c) |
| 200 | { |
| 201 | return ('0' <= c && c <= '9'); |
| 202 | } |
| 203 | |
| 204 | |
| 205 | int |
| 206 | macro_is_identifier_nondigit (int c) |
| 207 | { |
| 208 | return (c == '_' |
| 209 | || ('a' <= c && c <= 'z') |
| 210 | || ('A' <= c && c <= 'Z')); |
| 211 | } |
| 212 | |
| 213 | |
| 214 | static void |
| 215 | set_token (struct macro_buffer *tok, char *start, char *end) |
| 216 | { |
| 217 | tok->set_shared (start, end - start); |
| 218 | tok->last_token = 0; |
| 219 | |
| 220 | /* Presumed; get_identifier may overwrite this. */ |
| 221 | tok->is_identifier = 0; |
| 222 | } |
| 223 | |
| 224 | |
| 225 | static int |
| 226 | get_comment (struct macro_buffer *tok, char *p, char *end) |
| 227 | { |
| 228 | if (p + 2 > end) |
| 229 | return 0; |
| 230 | else if (p[0] == '/' |
| 231 | && p[1] == '*') |
| 232 | { |
| 233 | char *tok_start = p; |
| 234 | |
| 235 | p += 2; |
| 236 | |
| 237 | for (; p < end; p++) |
| 238 | if (p + 2 <= end |
| 239 | && p[0] == '*' |
| 240 | && p[1] == '/') |
| 241 | { |
| 242 | p += 2; |
| 243 | set_token (tok, tok_start, p); |
| 244 | return 1; |
| 245 | } |
| 246 | |
| 247 | error (_("Unterminated comment in macro expansion.")); |
| 248 | } |
| 249 | else if (p[0] == '/' |
| 250 | && p[1] == '/') |
| 251 | { |
| 252 | char *tok_start = p; |
| 253 | |
| 254 | p += 2; |
| 255 | for (; p < end; p++) |
| 256 | if (*p == '\n') |
| 257 | break; |
| 258 | |
| 259 | set_token (tok, tok_start, p); |
| 260 | return 1; |
| 261 | } |
| 262 | else |
| 263 | return 0; |
| 264 | } |
| 265 | |
| 266 | |
| 267 | static int |
| 268 | get_identifier (struct macro_buffer *tok, char *p, char *end) |
| 269 | { |
| 270 | if (p < end |
| 271 | && macro_is_identifier_nondigit (*p)) |
| 272 | { |
| 273 | char *tok_start = p; |
| 274 | |
| 275 | while (p < end |
| 276 | && (macro_is_identifier_nondigit (*p) |
| 277 | || macro_is_digit (*p))) |
| 278 | p++; |
| 279 | |
| 280 | set_token (tok, tok_start, p); |
| 281 | tok->is_identifier = 1; |
| 282 | return 1; |
| 283 | } |
| 284 | else |
| 285 | return 0; |
| 286 | } |
| 287 | |
| 288 | |
| 289 | static int |
| 290 | get_pp_number (struct macro_buffer *tok, char *p, char *end) |
| 291 | { |
| 292 | if (p < end |
| 293 | && (macro_is_digit (*p) |
| 294 | || (*p == '.' |
| 295 | && p + 2 <= end |
| 296 | && macro_is_digit (p[1])))) |
| 297 | { |
| 298 | char *tok_start = p; |
| 299 | |
| 300 | while (p < end) |
| 301 | { |
| 302 | if (p + 2 <= end |
| 303 | && strchr ("eEpP", *p) |
| 304 | && (p[1] == '+' || p[1] == '-')) |
| 305 | p += 2; |
| 306 | else if (macro_is_digit (*p) |
| 307 | || macro_is_identifier_nondigit (*p) |
| 308 | || *p == '.') |
| 309 | p++; |
| 310 | else |
| 311 | break; |
| 312 | } |
| 313 | |
| 314 | set_token (tok, tok_start, p); |
| 315 | return 1; |
| 316 | } |
| 317 | else |
| 318 | return 0; |
| 319 | } |
| 320 | |
| 321 | |
| 322 | |
| 323 | /* If the text starting at P going up to (but not including) END |
| 324 | starts with a character constant, set *TOK to point to that |
| 325 | character constant, and return 1. Otherwise, return zero. |
| 326 | Signal an error if it contains a malformed or incomplete character |
| 327 | constant. */ |
| 328 | static int |
| 329 | get_character_constant (struct macro_buffer *tok, char *p, char *end) |
| 330 | { |
| 331 | /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 |
| 332 | But of course, what really matters is that we handle it the same |
| 333 | way GDB's C/C++ lexer does. So we call parse_escape in utils.c |
| 334 | to handle escape sequences. */ |
| 335 | if ((p + 1 <= end && *p == '\'') |
| 336 | || (p + 2 <= end |
| 337 | && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') |
| 338 | && p[1] == '\'')) |
| 339 | { |
| 340 | char *tok_start = p; |
| 341 | int char_count = 0; |
| 342 | |
| 343 | if (*p == '\'') |
| 344 | p++; |
| 345 | else if (*p == 'L' || *p == 'u' || *p == 'U') |
| 346 | p += 2; |
| 347 | else |
| 348 | gdb_assert_not_reached ("unexpected character constant"); |
| 349 | |
| 350 | for (;;) |
| 351 | { |
| 352 | if (p >= end) |
| 353 | error (_("Unmatched single quote.")); |
| 354 | else if (*p == '\'') |
| 355 | { |
| 356 | if (!char_count) |
| 357 | error (_("A character constant must contain at least one " |
| 358 | "character.")); |
| 359 | p++; |
| 360 | break; |
| 361 | } |
| 362 | else if (*p == '\\') |
| 363 | { |
| 364 | const char *s, *o; |
| 365 | |
| 366 | s = o = ++p; |
| 367 | char_count += c_parse_escape (&s, NULL); |
| 368 | p += s - o; |
| 369 | } |
| 370 | else |
| 371 | { |
| 372 | p++; |
| 373 | char_count++; |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | set_token (tok, tok_start, p); |
| 378 | return 1; |
| 379 | } |
| 380 | else |
| 381 | return 0; |
| 382 | } |
| 383 | |
| 384 | |
| 385 | /* If the text starting at P going up to (but not including) END |
| 386 | starts with a string literal, set *TOK to point to that string |
| 387 | literal, and return 1. Otherwise, return zero. Signal an error if |
| 388 | it contains a malformed or incomplete string literal. */ |
| 389 | static int |
| 390 | get_string_literal (struct macro_buffer *tok, char *p, char *end) |
| 391 | { |
| 392 | if ((p + 1 <= end |
| 393 | && *p == '"') |
| 394 | || (p + 2 <= end |
| 395 | && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') |
| 396 | && p[1] == '"')) |
| 397 | { |
| 398 | char *tok_start = p; |
| 399 | |
| 400 | if (*p == '"') |
| 401 | p++; |
| 402 | else if (*p == 'L' || *p == 'u' || *p == 'U') |
| 403 | p += 2; |
| 404 | else |
| 405 | gdb_assert_not_reached ("unexpected string literal"); |
| 406 | |
| 407 | for (;;) |
| 408 | { |
| 409 | if (p >= end) |
| 410 | error (_("Unterminated string in expression.")); |
| 411 | else if (*p == '"') |
| 412 | { |
| 413 | p++; |
| 414 | break; |
| 415 | } |
| 416 | else if (*p == '\n') |
| 417 | error (_("Newline characters may not appear in string " |
| 418 | "constants.")); |
| 419 | else if (*p == '\\') |
| 420 | { |
| 421 | const char *s, *o; |
| 422 | |
| 423 | s = o = ++p; |
| 424 | c_parse_escape (&s, NULL); |
| 425 | p += s - o; |
| 426 | } |
| 427 | else |
| 428 | p++; |
| 429 | } |
| 430 | |
| 431 | set_token (tok, tok_start, p); |
| 432 | return 1; |
| 433 | } |
| 434 | else |
| 435 | return 0; |
| 436 | } |
| 437 | |
| 438 | |
| 439 | static int |
| 440 | get_punctuator (struct macro_buffer *tok, char *p, char *end) |
| 441 | { |
| 442 | /* Here, speed is much less important than correctness and clarity. */ |
| 443 | |
| 444 | /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1. |
| 445 | Note that this table is ordered in a special way. A punctuator |
| 446 | which is a prefix of another punctuator must appear after its |
| 447 | "extension". Otherwise, the wrong token will be returned. */ |
| 448 | static const char * const punctuators[] = { |
| 449 | "[", "]", "(", ")", "{", "}", "?", ";", ",", "~", |
| 450 | "...", ".", |
| 451 | "->", "--", "-=", "-", |
| 452 | "++", "+=", "+", |
| 453 | "*=", "*", |
| 454 | "!=", "!", |
| 455 | "&&", "&=", "&", |
| 456 | "/=", "/", |
| 457 | "%>", "%:%:", "%:", "%=", "%", |
| 458 | "^=", "^", |
| 459 | "##", "#", |
| 460 | ":>", ":", |
| 461 | "||", "|=", "|", |
| 462 | "<<=", "<<", "<=", "<:", "<%", "<", |
| 463 | ">>=", ">>", ">=", ">", |
| 464 | "==", "=", |
| 465 | 0 |
| 466 | }; |
| 467 | |
| 468 | int i; |
| 469 | |
| 470 | if (p + 1 <= end) |
| 471 | { |
| 472 | for (i = 0; punctuators[i]; i++) |
| 473 | { |
| 474 | const char *punctuator = punctuators[i]; |
| 475 | |
| 476 | if (p[0] == punctuator[0]) |
| 477 | { |
| 478 | int len = strlen (punctuator); |
| 479 | |
| 480 | if (p + len <= end |
| 481 | && ! memcmp (p, punctuator, len)) |
| 482 | { |
| 483 | set_token (tok, p, p + len); |
| 484 | return 1; |
| 485 | } |
| 486 | } |
| 487 | } |
| 488 | } |
| 489 | |
| 490 | return 0; |
| 491 | } |
| 492 | |
| 493 | |
| 494 | /* Peel the next preprocessor token off of SRC, and put it in TOK. |
| 495 | Mutate TOK to refer to the first token in SRC, and mutate SRC to |
| 496 | refer to the text after that token. SRC must be a shared buffer; |
| 497 | the resulting TOK will be shared, pointing into the same string SRC |
| 498 | does. Initialize TOK's last_token field. Return non-zero if we |
| 499 | succeed, or 0 if we didn't find any more tokens in SRC. */ |
| 500 | static int |
| 501 | get_token (struct macro_buffer *tok, |
| 502 | struct macro_buffer *src) |
| 503 | { |
| 504 | char *p = src->text; |
| 505 | char *end = p + src->len; |
| 506 | |
| 507 | gdb_assert (src->shared); |
| 508 | |
| 509 | /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: |
| 510 | |
| 511 | preprocessing-token: |
| 512 | header-name |
| 513 | identifier |
| 514 | pp-number |
| 515 | character-constant |
| 516 | string-literal |
| 517 | punctuator |
| 518 | each non-white-space character that cannot be one of the above |
| 519 | |
| 520 | We don't have to deal with header-name tokens, since those can |
| 521 | only occur after a #include, which we will never see. */ |
| 522 | |
| 523 | while (p < end) |
| 524 | if (macro_is_whitespace (*p)) |
| 525 | p++; |
| 526 | else if (get_comment (tok, p, end)) |
| 527 | p += tok->len; |
| 528 | else if (get_pp_number (tok, p, end) |
| 529 | || get_character_constant (tok, p, end) |
| 530 | || get_string_literal (tok, p, end) |
| 531 | /* Note: the grammar in the standard seems to be |
| 532 | ambiguous: L'x' can be either a wide character |
| 533 | constant, or an identifier followed by a normal |
| 534 | character constant. By trying `get_identifier' after |
| 535 | we try get_character_constant and get_string_literal, |
| 536 | we give the wide character syntax precedence. Now, |
| 537 | since GDB doesn't handle wide character constants |
| 538 | anyway, is this the right thing to do? */ |
| 539 | || get_identifier (tok, p, end) |
| 540 | || get_punctuator (tok, p, end)) |
| 541 | { |
| 542 | /* How many characters did we consume, including whitespace? */ |
| 543 | int consumed = p - src->text + tok->len; |
| 544 | |
| 545 | src->text += consumed; |
| 546 | src->len -= consumed; |
| 547 | return 1; |
| 548 | } |
| 549 | else |
| 550 | { |
| 551 | /* We have found a "non-whitespace character that cannot be |
| 552 | one of the above." Make a token out of it. */ |
| 553 | int consumed; |
| 554 | |
| 555 | set_token (tok, p, p + 1); |
| 556 | consumed = p - src->text + tok->len; |
| 557 | src->text += consumed; |
| 558 | src->len -= consumed; |
| 559 | return 1; |
| 560 | } |
| 561 | |
| 562 | return 0; |
| 563 | } |
| 564 | |
| 565 | |
| 566 | \f |
| 567 | /* Appending token strings, with and without splicing */ |
| 568 | |
| 569 | |
| 570 | /* Append the macro buffer SRC to the end of DEST, and ensure that |
| 571 | doing so doesn't splice the token at the end of SRC with the token |
| 572 | at the beginning of DEST. SRC and DEST must have their last_token |
| 573 | fields set. Upon return, DEST's last_token field is set correctly. |
| 574 | |
| 575 | For example: |
| 576 | |
| 577 | If DEST is "(" and SRC is "y", then we can return with |
| 578 | DEST set to "(y" --- we've simply appended the two buffers. |
| 579 | |
| 580 | However, if DEST is "x" and SRC is "y", then we must not return |
| 581 | with DEST set to "xy" --- that would splice the two tokens "x" and |
| 582 | "y" together to make a single token "xy". However, it would be |
| 583 | fine to return with DEST set to "x y". Similarly, "<" and "<" must |
| 584 | yield "< <", not "<<", etc. */ |
| 585 | static void |
| 586 | append_tokens_without_splicing (struct macro_buffer *dest, |
| 587 | struct macro_buffer *src) |
| 588 | { |
| 589 | int original_dest_len = dest->len; |
| 590 | struct macro_buffer dest_tail, new_token; |
| 591 | |
| 592 | gdb_assert (src->last_token != -1); |
| 593 | gdb_assert (dest->last_token != -1); |
| 594 | |
| 595 | /* First, just try appending the two, and call get_token to see if |
| 596 | we got a splice. */ |
| 597 | dest->appendmem (src->text, src->len); |
| 598 | |
| 599 | /* If DEST originally had no token abutting its end, then we can't |
| 600 | have spliced anything, so we're done. */ |
| 601 | if (dest->last_token == original_dest_len) |
| 602 | { |
| 603 | dest->last_token = original_dest_len + src->last_token; |
| 604 | return; |
| 605 | } |
| 606 | |
| 607 | /* Set DEST_TAIL to point to the last token in DEST, followed by |
| 608 | all the stuff we just appended. */ |
| 609 | dest_tail.set_shared (dest->text + dest->last_token, |
| 610 | dest->len - dest->last_token); |
| 611 | |
| 612 | /* Re-parse DEST's last token. We know that DEST used to contain |
| 613 | at least one token, so if it doesn't contain any after the |
| 614 | append, then we must have spliced "/" and "*" or "/" and "/" to |
| 615 | make a comment start. (Just for the record, I got this right |
| 616 | the first time. This is not a bug fix.) */ |
| 617 | if (get_token (&new_token, &dest_tail) |
| 618 | && (new_token.text + new_token.len |
| 619 | == dest->text + original_dest_len)) |
| 620 | { |
| 621 | /* No splice, so we're done. */ |
| 622 | dest->last_token = original_dest_len + src->last_token; |
| 623 | return; |
| 624 | } |
| 625 | |
| 626 | /* Okay, a simple append caused a splice. Let's chop dest back to |
| 627 | its original length and try again, but separate the texts with a |
| 628 | space. */ |
| 629 | dest->len = original_dest_len; |
| 630 | dest->appendc (' '); |
| 631 | dest->appendmem (src->text, src->len); |
| 632 | |
| 633 | dest_tail.set_shared (dest->text + dest->last_token, |
| 634 | dest->len - dest->last_token); |
| 635 | |
| 636 | /* Try to re-parse DEST's last token, as above. */ |
| 637 | if (get_token (&new_token, &dest_tail) |
| 638 | && (new_token.text + new_token.len |
| 639 | == dest->text + original_dest_len)) |
| 640 | { |
| 641 | /* No splice, so we're done. */ |
| 642 | dest->last_token = original_dest_len + 1 + src->last_token; |
| 643 | return; |
| 644 | } |
| 645 | |
| 646 | /* As far as I know, there's no case where inserting a space isn't |
| 647 | enough to prevent a splice. */ |
| 648 | internal_error (__FILE__, __LINE__, |
| 649 | _("unable to avoid splicing tokens during macro expansion")); |
| 650 | } |
| 651 | |
| 652 | /* Stringify an argument, and insert it into DEST. ARG is the text to |
| 653 | stringify; it is LEN bytes long. */ |
| 654 | |
| 655 | static void |
| 656 | stringify (struct macro_buffer *dest, const char *arg, int len) |
| 657 | { |
| 658 | /* Trim initial whitespace from ARG. */ |
| 659 | while (len > 0 && macro_is_whitespace (*arg)) |
| 660 | { |
| 661 | ++arg; |
| 662 | --len; |
| 663 | } |
| 664 | |
| 665 | /* Trim trailing whitespace from ARG. */ |
| 666 | while (len > 0 && macro_is_whitespace (arg[len - 1])) |
| 667 | --len; |
| 668 | |
| 669 | /* Insert the string. */ |
| 670 | dest->appendc ('"'); |
| 671 | while (len > 0) |
| 672 | { |
| 673 | /* We could try to handle strange cases here, like control |
| 674 | characters, but there doesn't seem to be much point. */ |
| 675 | if (macro_is_whitespace (*arg)) |
| 676 | { |
| 677 | /* Replace a sequence of whitespace with a single space. */ |
| 678 | dest->appendc (' '); |
| 679 | while (len > 1 && macro_is_whitespace (arg[1])) |
| 680 | { |
| 681 | ++arg; |
| 682 | --len; |
| 683 | } |
| 684 | } |
| 685 | else if (*arg == '\\' || *arg == '"') |
| 686 | { |
| 687 | dest->appendc ('\\'); |
| 688 | dest->appendc (*arg); |
| 689 | } |
| 690 | else |
| 691 | dest->appendc (*arg); |
| 692 | ++arg; |
| 693 | --len; |
| 694 | } |
| 695 | dest->appendc ('"'); |
| 696 | dest->last_token = dest->len; |
| 697 | } |
| 698 | |
| 699 | /* See macroexp.h. */ |
| 700 | |
| 701 | char * |
| 702 | macro_stringify (const char *str) |
| 703 | { |
| 704 | int len = strlen (str); |
| 705 | struct macro_buffer buffer (len); |
| 706 | |
| 707 | stringify (&buffer, str, len); |
| 708 | buffer.appendc ('\0'); |
| 709 | |
| 710 | return buffer.release (); |
| 711 | } |
| 712 | |
| 713 | \f |
| 714 | /* Expanding macros! */ |
| 715 | |
| 716 | |
| 717 | /* A singly-linked list of the names of the macros we are currently |
| 718 | expanding --- for detecting expansion loops. */ |
| 719 | struct macro_name_list { |
| 720 | const char *name; |
| 721 | struct macro_name_list *next; |
| 722 | }; |
| 723 | |
| 724 | |
| 725 | /* Return non-zero if we are currently expanding the macro named NAME, |
| 726 | according to LIST; otherwise, return zero. |
| 727 | |
| 728 | You know, it would be possible to get rid of all the NO_LOOP |
| 729 | arguments to these functions by simply generating a new lookup |
| 730 | function and baton which refuses to find the definition for a |
| 731 | particular macro, and otherwise delegates the decision to another |
| 732 | function/baton pair. But that makes the linked list of excluded |
| 733 | macros chained through untyped baton pointers, which will make it |
| 734 | harder to debug. :( */ |
| 735 | static int |
| 736 | currently_rescanning (struct macro_name_list *list, const char *name) |
| 737 | { |
| 738 | for (; list; list = list->next) |
| 739 | if (strcmp (name, list->name) == 0) |
| 740 | return 1; |
| 741 | |
| 742 | return 0; |
| 743 | } |
| 744 | |
| 745 | |
| 746 | /* Gather the arguments to a macro expansion. |
| 747 | |
| 748 | NAME is the name of the macro being invoked. (It's only used for |
| 749 | printing error messages.) |
| 750 | |
| 751 | Assume that SRC is the text of the macro invocation immediately |
| 752 | following the macro name. For example, if we're processing the |
| 753 | text foo(bar, baz), then NAME would be foo and SRC will be (bar, |
| 754 | baz). |
| 755 | |
| 756 | If SRC doesn't start with an open paren ( token at all, return |
| 757 | false, leave SRC unchanged, and don't set *ARGS_PTR to anything. |
| 758 | |
| 759 | If SRC doesn't contain a properly terminated argument list, then |
| 760 | raise an error. |
| 761 | |
| 762 | For a variadic macro, NARGS holds the number of formal arguments to |
| 763 | the macro. For a GNU-style variadic macro, this should be the |
| 764 | number of named arguments. For a non-variadic macro, NARGS should |
| 765 | be -1. |
| 766 | |
| 767 | Otherwise, return true and set *ARGS_PTR to a vector of macro |
| 768 | buffers referring to the argument texts. The macro buffers share |
| 769 | their text with SRC, and their last_token fields are initialized. |
| 770 | |
| 771 | NOTE WELL: if SRC starts with a open paren ( token followed |
| 772 | immediately by a close paren ) token (e.g., the invocation looks |
| 773 | like "foo()"), we treat that as one argument, which happens to be |
| 774 | the empty list of tokens. The caller should keep in mind that such |
| 775 | a sequence of tokens is a valid way to invoke one-parameter |
| 776 | function-like macros, but also a valid way to invoke zero-parameter |
| 777 | function-like macros. Eeew. |
| 778 | |
| 779 | Consume the tokens from SRC; after this call, SRC contains the text |
| 780 | following the invocation. */ |
| 781 | |
| 782 | static bool |
| 783 | gather_arguments (const char *name, struct macro_buffer *src, int nargs, |
| 784 | std::vector<struct macro_buffer> *args_ptr) |
| 785 | { |
| 786 | struct macro_buffer tok; |
| 787 | std::vector<struct macro_buffer> args; |
| 788 | |
| 789 | /* Does SRC start with an opening paren token? Read from a copy of |
| 790 | SRC, so SRC itself is unaffected if we don't find an opening |
| 791 | paren. */ |
| 792 | { |
| 793 | struct macro_buffer temp (src->text, src->len); |
| 794 | |
| 795 | if (! get_token (&tok, &temp) |
| 796 | || tok.len != 1 |
| 797 | || tok.text[0] != '(') |
| 798 | return false; |
| 799 | } |
| 800 | |
| 801 | /* Consume SRC's opening paren. */ |
| 802 | get_token (&tok, src); |
| 803 | |
| 804 | for (;;) |
| 805 | { |
| 806 | struct macro_buffer *arg; |
| 807 | int depth; |
| 808 | |
| 809 | /* Initialize the next argument. */ |
| 810 | args.emplace_back (); |
| 811 | arg = &args.back (); |
| 812 | set_token (arg, src->text, src->text); |
| 813 | |
| 814 | /* Gather the argument's tokens. */ |
| 815 | depth = 0; |
| 816 | for (;;) |
| 817 | { |
| 818 | if (! get_token (&tok, src)) |
| 819 | error (_("Malformed argument list for macro `%s'."), name); |
| 820 | |
| 821 | /* Is tok an opening paren? */ |
| 822 | if (tok.len == 1 && tok.text[0] == '(') |
| 823 | depth++; |
| 824 | |
| 825 | /* Is tok is a closing paren? */ |
| 826 | else if (tok.len == 1 && tok.text[0] == ')') |
| 827 | { |
| 828 | /* If it's a closing paren at the top level, then that's |
| 829 | the end of the argument list. */ |
| 830 | if (depth == 0) |
| 831 | { |
| 832 | /* In the varargs case, the last argument may be |
| 833 | missing. Add an empty argument in this case. */ |
| 834 | if (nargs != -1 && args.size () == nargs - 1) |
| 835 | { |
| 836 | args.emplace_back (); |
| 837 | arg = &args.back (); |
| 838 | set_token (arg, src->text, src->text); |
| 839 | } |
| 840 | |
| 841 | *args_ptr = std::move (args); |
| 842 | return true; |
| 843 | } |
| 844 | |
| 845 | depth--; |
| 846 | } |
| 847 | |
| 848 | /* If tok is a comma at top level, then that's the end of |
| 849 | the current argument. However, if we are handling a |
| 850 | variadic macro and we are computing the last argument, we |
| 851 | want to include the comma and remaining tokens. */ |
| 852 | else if (tok.len == 1 && tok.text[0] == ',' && depth == 0 |
| 853 | && (nargs == -1 || args.size () < nargs)) |
| 854 | break; |
| 855 | |
| 856 | /* Extend the current argument to enclose this token. If |
| 857 | this is the current argument's first token, leave out any |
| 858 | leading whitespace, just for aesthetics. */ |
| 859 | if (arg->len == 0) |
| 860 | { |
| 861 | arg->text = tok.text; |
| 862 | arg->len = tok.len; |
| 863 | arg->last_token = 0; |
| 864 | } |
| 865 | else |
| 866 | { |
| 867 | arg->len = (tok.text + tok.len) - arg->text; |
| 868 | arg->last_token = tok.text - arg->text; |
| 869 | } |
| 870 | } |
| 871 | } |
| 872 | } |
| 873 | |
| 874 | |
| 875 | /* The `expand' and `substitute_args' functions both invoke `scan' |
| 876 | recursively, so we need a forward declaration somewhere. */ |
| 877 | static void scan (struct macro_buffer *dest, |
| 878 | struct macro_buffer *src, |
| 879 | struct macro_name_list *no_loop, |
| 880 | macro_lookup_ftype *lookup_func, |
| 881 | void *lookup_baton); |
| 882 | |
| 883 | |
| 884 | /* A helper function for substitute_args. |
| 885 | |
| 886 | ARGV is a vector of all the arguments; ARGC is the number of |
| 887 | arguments. IS_VARARGS is true if the macro being substituted is a |
| 888 | varargs macro; in this case VA_ARG_NAME is the name of the |
| 889 | "variable" argument. VA_ARG_NAME is ignored if IS_VARARGS is |
| 890 | false. |
| 891 | |
| 892 | If the token TOK is the name of a parameter, return the parameter's |
| 893 | index. If TOK is not an argument, return -1. */ |
| 894 | |
| 895 | static int |
| 896 | find_parameter (const struct macro_buffer *tok, |
| 897 | int is_varargs, const struct macro_buffer *va_arg_name, |
| 898 | int argc, const char * const *argv) |
| 899 | { |
| 900 | int i; |
| 901 | |
| 902 | if (! tok->is_identifier) |
| 903 | return -1; |
| 904 | |
| 905 | for (i = 0; i < argc; ++i) |
| 906 | if (tok->len == strlen (argv[i]) |
| 907 | && !memcmp (tok->text, argv[i], tok->len)) |
| 908 | return i; |
| 909 | |
| 910 | if (is_varargs && tok->len == va_arg_name->len |
| 911 | && ! memcmp (tok->text, va_arg_name->text, tok->len)) |
| 912 | return argc - 1; |
| 913 | |
| 914 | return -1; |
| 915 | } |
| 916 | |
| 917 | /* Helper function for substitute_args that gets the next token and |
| 918 | updates the passed-in state variables. */ |
| 919 | |
| 920 | static void |
| 921 | get_next_token_for_substitution (struct macro_buffer *replacement_list, |
| 922 | struct macro_buffer *token, |
| 923 | char **start, |
| 924 | struct macro_buffer *lookahead, |
| 925 | char **lookahead_start, |
| 926 | int *lookahead_valid, |
| 927 | bool *keep_going) |
| 928 | { |
| 929 | if (!*lookahead_valid) |
| 930 | *keep_going = false; |
| 931 | else |
| 932 | { |
| 933 | *keep_going = true; |
| 934 | *token = *lookahead; |
| 935 | *start = *lookahead_start; |
| 936 | *lookahead_start = replacement_list->text; |
| 937 | *lookahead_valid = get_token (lookahead, replacement_list); |
| 938 | } |
| 939 | } |
| 940 | |
| 941 | /* Given the macro definition DEF, being invoked with the actual |
| 942 | arguments given by ARGV, substitute the arguments into the |
| 943 | replacement list, and store the result in DEST. |
| 944 | |
| 945 | IS_VARARGS should be true if DEF is a varargs macro. In this case, |
| 946 | VA_ARG_NAME should be the name of the "variable" argument -- either |
| 947 | __VA_ARGS__ for c99-style varargs, or the final argument name, for |
| 948 | GNU-style varargs. If IS_VARARGS is false, this parameter is |
| 949 | ignored. |
| 950 | |
| 951 | If it is necessary to expand macro invocations in one of the |
| 952 | arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro |
| 953 | definitions, and don't expand invocations of the macros listed in |
| 954 | NO_LOOP. */ |
| 955 | |
| 956 | static void |
| 957 | substitute_args (struct macro_buffer *dest, |
| 958 | struct macro_definition *def, |
| 959 | int is_varargs, const struct macro_buffer *va_arg_name, |
| 960 | const std::vector<struct macro_buffer> &argv, |
| 961 | struct macro_name_list *no_loop, |
| 962 | macro_lookup_ftype *lookup_func, |
| 963 | void *lookup_baton) |
| 964 | { |
| 965 | /* The token we are currently considering. */ |
| 966 | struct macro_buffer tok; |
| 967 | /* The replacement list's pointer from just before TOK was lexed. */ |
| 968 | char *original_rl_start; |
| 969 | /* We have a single lookahead token to handle token splicing. */ |
| 970 | struct macro_buffer lookahead; |
| 971 | /* The lookahead token might not be valid. */ |
| 972 | int lookahead_valid; |
| 973 | /* The replacement list's pointer from just before LOOKAHEAD was |
| 974 | lexed. */ |
| 975 | char *lookahead_rl_start; |
| 976 | |
| 977 | /* A macro buffer for the macro's replacement list. */ |
| 978 | struct macro_buffer replacement_list (def->replacement, |
| 979 | strlen (def->replacement)); |
| 980 | |
| 981 | gdb_assert (dest->len == 0); |
| 982 | dest->last_token = 0; |
| 983 | |
| 984 | original_rl_start = replacement_list.text; |
| 985 | if (! get_token (&tok, &replacement_list)) |
| 986 | return; |
| 987 | lookahead_rl_start = replacement_list.text; |
| 988 | lookahead_valid = get_token (&lookahead, &replacement_list); |
| 989 | |
| 990 | /* __VA_OPT__ state variable. The states are: |
| 991 | 0 - nothing happening |
| 992 | 1 - saw __VA_OPT__ |
| 993 | >= 2 in __VA_OPT__, the value encodes the parenthesis depth. */ |
| 994 | unsigned vaopt_state = 0; |
| 995 | |
| 996 | for (bool keep_going = true; |
| 997 | keep_going; |
| 998 | get_next_token_for_substitution (&replacement_list, |
| 999 | &tok, |
| 1000 | &original_rl_start, |
| 1001 | &lookahead, |
| 1002 | &lookahead_rl_start, |
| 1003 | &lookahead_valid, |
| 1004 | &keep_going)) |
| 1005 | { |
| 1006 | bool token_is_vaopt = (tok.len == 10 |
| 1007 | && strncmp (tok.text, "__VA_OPT__", 10) == 0); |
| 1008 | |
| 1009 | if (vaopt_state > 0) |
| 1010 | { |
| 1011 | if (token_is_vaopt) |
| 1012 | error (_("__VA_OPT__ cannot appear inside __VA_OPT__")); |
| 1013 | else if (tok.len == 1 && tok.text[0] == '(') |
| 1014 | { |
| 1015 | ++vaopt_state; |
| 1016 | /* We just entered __VA_OPT__, so don't emit this |
| 1017 | token. */ |
| 1018 | continue; |
| 1019 | } |
| 1020 | else if (vaopt_state == 1) |
| 1021 | error (_("__VA_OPT__ must be followed by an open parenthesis")); |
| 1022 | else if (tok.len == 1 && tok.text[0] == ')') |
| 1023 | { |
| 1024 | --vaopt_state; |
| 1025 | if (vaopt_state == 1) |
| 1026 | { |
| 1027 | /* Done with __VA_OPT__. */ |
| 1028 | vaopt_state = 0; |
| 1029 | /* Don't emit. */ |
| 1030 | continue; |
| 1031 | } |
| 1032 | } |
| 1033 | |
| 1034 | /* If __VA_ARGS__ is empty, then drop the contents of |
| 1035 | __VA_OPT__. */ |
| 1036 | if (argv.back ().len == 0) |
| 1037 | continue; |
| 1038 | } |
| 1039 | else if (token_is_vaopt) |
| 1040 | { |
| 1041 | if (!is_varargs) |
| 1042 | error (_("__VA_OPT__ is only valid in a variadic macro")); |
| 1043 | vaopt_state = 1; |
| 1044 | /* Don't emit this token. */ |
| 1045 | continue; |
| 1046 | } |
| 1047 | |
| 1048 | /* Just for aesthetics. If we skipped some whitespace, copy |
| 1049 | that to DEST. */ |
| 1050 | if (tok.text > original_rl_start) |
| 1051 | { |
| 1052 | dest->appendmem (original_rl_start, tok.text - original_rl_start); |
| 1053 | dest->last_token = dest->len; |
| 1054 | } |
| 1055 | |
| 1056 | /* Is this token the stringification operator? */ |
| 1057 | if (tok.len == 1 |
| 1058 | && tok.text[0] == '#') |
| 1059 | { |
| 1060 | int arg; |
| 1061 | |
| 1062 | if (!lookahead_valid) |
| 1063 | error (_("Stringification operator requires an argument.")); |
| 1064 | |
| 1065 | arg = find_parameter (&lookahead, is_varargs, va_arg_name, |
| 1066 | def->argc, def->argv); |
| 1067 | if (arg == -1) |
| 1068 | error (_("Argument to stringification operator must name " |
| 1069 | "a macro parameter.")); |
| 1070 | |
| 1071 | stringify (dest, argv[arg].text, argv[arg].len); |
| 1072 | |
| 1073 | /* Read one token and let the loop iteration code handle the |
| 1074 | rest. */ |
| 1075 | lookahead_rl_start = replacement_list.text; |
| 1076 | lookahead_valid = get_token (&lookahead, &replacement_list); |
| 1077 | } |
| 1078 | /* Is this token the splicing operator? */ |
| 1079 | else if (tok.len == 2 |
| 1080 | && tok.text[0] == '#' |
| 1081 | && tok.text[1] == '#') |
| 1082 | error (_("Stray splicing operator")); |
| 1083 | /* Is the next token the splicing operator? */ |
| 1084 | else if (lookahead_valid |
| 1085 | && lookahead.len == 2 |
| 1086 | && lookahead.text[0] == '#' |
| 1087 | && lookahead.text[1] == '#') |
| 1088 | { |
| 1089 | int finished = 0; |
| 1090 | int prev_was_comma = 0; |
| 1091 | |
| 1092 | /* Note that GCC warns if the result of splicing is not a |
| 1093 | token. In the debugger there doesn't seem to be much |
| 1094 | benefit from doing this. */ |
| 1095 | |
| 1096 | /* Insert the first token. */ |
| 1097 | if (tok.len == 1 && tok.text[0] == ',') |
| 1098 | prev_was_comma = 1; |
| 1099 | else |
| 1100 | { |
| 1101 | int arg = find_parameter (&tok, is_varargs, va_arg_name, |
| 1102 | def->argc, def->argv); |
| 1103 | |
| 1104 | if (arg != -1) |
| 1105 | dest->appendmem (argv[arg].text, argv[arg].len); |
| 1106 | else |
| 1107 | dest->appendmem (tok.text, tok.len); |
| 1108 | } |
| 1109 | |
| 1110 | /* Apply a possible sequence of ## operators. */ |
| 1111 | for (;;) |
| 1112 | { |
| 1113 | if (! get_token (&tok, &replacement_list)) |
| 1114 | error (_("Splicing operator at end of macro")); |
| 1115 | |
| 1116 | /* Handle a comma before a ##. If we are handling |
| 1117 | varargs, and the token on the right hand side is the |
| 1118 | varargs marker, and the final argument is empty or |
| 1119 | missing, then drop the comma. This is a GNU |
| 1120 | extension. There is one ambiguous case here, |
| 1121 | involving pedantic behavior with an empty argument, |
| 1122 | but we settle that in favor of GNU-style (GCC uses an |
| 1123 | option). If we aren't dealing with varargs, we |
| 1124 | simply insert the comma. */ |
| 1125 | if (prev_was_comma) |
| 1126 | { |
| 1127 | if (! (is_varargs |
| 1128 | && tok.len == va_arg_name->len |
| 1129 | && !memcmp (tok.text, va_arg_name->text, tok.len) |
| 1130 | && argv.back ().len == 0)) |
| 1131 | dest->appendmem (",", 1); |
| 1132 | prev_was_comma = 0; |
| 1133 | } |
| 1134 | |
| 1135 | /* Insert the token. If it is a parameter, insert the |
| 1136 | argument. If it is a comma, treat it specially. */ |
| 1137 | if (tok.len == 1 && tok.text[0] == ',') |
| 1138 | prev_was_comma = 1; |
| 1139 | else |
| 1140 | { |
| 1141 | int arg = find_parameter (&tok, is_varargs, va_arg_name, |
| 1142 | def->argc, def->argv); |
| 1143 | |
| 1144 | if (arg != -1) |
| 1145 | dest->appendmem (argv[arg].text, argv[arg].len); |
| 1146 | else |
| 1147 | dest->appendmem (tok.text, tok.len); |
| 1148 | } |
| 1149 | |
| 1150 | /* Now read another token. If it is another splice, we |
| 1151 | loop. */ |
| 1152 | original_rl_start = replacement_list.text; |
| 1153 | if (! get_token (&tok, &replacement_list)) |
| 1154 | { |
| 1155 | finished = 1; |
| 1156 | break; |
| 1157 | } |
| 1158 | |
| 1159 | if (! (tok.len == 2 |
| 1160 | && tok.text[0] == '#' |
| 1161 | && tok.text[1] == '#')) |
| 1162 | break; |
| 1163 | } |
| 1164 | |
| 1165 | if (prev_was_comma) |
| 1166 | { |
| 1167 | /* We saw a comma. Insert it now. */ |
| 1168 | dest->appendmem (",", 1); |
| 1169 | } |
| 1170 | |
| 1171 | dest->last_token = dest->len; |
| 1172 | if (finished) |
| 1173 | lookahead_valid = 0; |
| 1174 | else |
| 1175 | { |
| 1176 | /* Set up for the loop iterator. */ |
| 1177 | lookahead = tok; |
| 1178 | lookahead_rl_start = original_rl_start; |
| 1179 | lookahead_valid = 1; |
| 1180 | } |
| 1181 | } |
| 1182 | else |
| 1183 | { |
| 1184 | /* Is this token an identifier? */ |
| 1185 | int substituted = 0; |
| 1186 | int arg = find_parameter (&tok, is_varargs, va_arg_name, |
| 1187 | def->argc, def->argv); |
| 1188 | |
| 1189 | if (arg != -1) |
| 1190 | { |
| 1191 | /* Expand any macro invocations in the argument text, |
| 1192 | and append the result to dest. Remember that scan |
| 1193 | mutates its source, so we need to scan a new buffer |
| 1194 | referring to the argument's text, not the argument |
| 1195 | itself. */ |
| 1196 | struct macro_buffer arg_src (argv[arg].text, argv[arg].len); |
| 1197 | scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); |
| 1198 | substituted = 1; |
| 1199 | } |
| 1200 | |
| 1201 | /* If it wasn't a parameter, then just copy it across. */ |
| 1202 | if (! substituted) |
| 1203 | append_tokens_without_splicing (dest, &tok); |
| 1204 | } |
| 1205 | } |
| 1206 | |
| 1207 | if (vaopt_state > 0) |
| 1208 | error (_("Unterminated __VA_OPT__")); |
| 1209 | } |
| 1210 | |
| 1211 | |
| 1212 | /* Expand a call to a macro named ID, whose definition is DEF. Append |
| 1213 | its expansion to DEST. SRC is the input text following the ID |
| 1214 | token. We are currently rescanning the expansions of the macros |
| 1215 | named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and |
| 1216 | LOOKUP_BATON to find definitions for any nested macro references. |
| 1217 | |
| 1218 | Return 1 if we decided to expand it, zero otherwise. (If it's a |
| 1219 | function-like macro name that isn't followed by an argument list, |
| 1220 | we don't expand it.) If we return zero, leave SRC unchanged. */ |
| 1221 | static int |
| 1222 | expand (const char *id, |
| 1223 | struct macro_definition *def, |
| 1224 | struct macro_buffer *dest, |
| 1225 | struct macro_buffer *src, |
| 1226 | struct macro_name_list *no_loop, |
| 1227 | macro_lookup_ftype *lookup_func, |
| 1228 | void *lookup_baton) |
| 1229 | { |
| 1230 | struct macro_name_list new_no_loop; |
| 1231 | |
| 1232 | /* Create a new node to be added to the front of the no-expand list. |
| 1233 | This list is appropriate for re-scanning replacement lists, but |
| 1234 | it is *not* appropriate for scanning macro arguments; invocations |
| 1235 | of the macro whose arguments we are gathering *do* get expanded |
| 1236 | there. */ |
| 1237 | new_no_loop.name = id; |
| 1238 | new_no_loop.next = no_loop; |
| 1239 | |
| 1240 | /* What kind of macro are we expanding? */ |
| 1241 | if (def->kind == macro_object_like) |
| 1242 | { |
| 1243 | struct macro_buffer replacement_list (def->replacement, |
| 1244 | strlen (def->replacement)); |
| 1245 | |
| 1246 | scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); |
| 1247 | return 1; |
| 1248 | } |
| 1249 | else if (def->kind == macro_function_like) |
| 1250 | { |
| 1251 | struct macro_buffer va_arg_name; |
| 1252 | int is_varargs = 0; |
| 1253 | |
| 1254 | if (def->argc >= 1) |
| 1255 | { |
| 1256 | if (strcmp (def->argv[def->argc - 1], "...") == 0) |
| 1257 | { |
| 1258 | /* In C99-style varargs, substitution is done using |
| 1259 | __VA_ARGS__. */ |
| 1260 | va_arg_name.set_shared ("__VA_ARGS__", strlen ("__VA_ARGS__")); |
| 1261 | is_varargs = 1; |
| 1262 | } |
| 1263 | else |
| 1264 | { |
| 1265 | int len = strlen (def->argv[def->argc - 1]); |
| 1266 | |
| 1267 | if (len > 3 |
| 1268 | && strcmp (def->argv[def->argc - 1] + len - 3, "...") == 0) |
| 1269 | { |
| 1270 | /* In GNU-style varargs, the name of the |
| 1271 | substitution parameter is the name of the formal |
| 1272 | argument without the "...". */ |
| 1273 | va_arg_name.set_shared (def->argv[def->argc - 1], len - 3); |
| 1274 | is_varargs = 1; |
| 1275 | } |
| 1276 | } |
| 1277 | } |
| 1278 | |
| 1279 | std::vector<struct macro_buffer> argv; |
| 1280 | /* If we couldn't find any argument list, then we don't expand |
| 1281 | this macro. */ |
| 1282 | if (!gather_arguments (id, src, is_varargs ? def->argc : -1, |
| 1283 | &argv)) |
| 1284 | return 0; |
| 1285 | |
| 1286 | /* Check that we're passing an acceptable number of arguments for |
| 1287 | this macro. */ |
| 1288 | if (argv.size () != def->argc) |
| 1289 | { |
| 1290 | if (is_varargs && argv.size () >= def->argc - 1) |
| 1291 | { |
| 1292 | /* Ok. */ |
| 1293 | } |
| 1294 | /* Remember that a sequence of tokens like "foo()" is a |
| 1295 | valid invocation of a macro expecting either zero or one |
| 1296 | arguments. */ |
| 1297 | else if (! (argv.size () == 1 |
| 1298 | && argv[0].len == 0 |
| 1299 | && def->argc == 0)) |
| 1300 | error (_("Wrong number of arguments to macro `%s' " |
| 1301 | "(expected %d, got %d)."), |
| 1302 | id, def->argc, int (argv.size ())); |
| 1303 | } |
| 1304 | |
| 1305 | /* Note that we don't expand macro invocations in the arguments |
| 1306 | yet --- we let subst_args take care of that. Parameters that |
| 1307 | appear as operands of the stringifying operator "#" or the |
| 1308 | splicing operator "##" don't get macro references expanded, |
| 1309 | so we can't really tell whether it's appropriate to macro- |
| 1310 | expand an argument until we see how it's being used. */ |
| 1311 | struct macro_buffer substituted (0); |
| 1312 | substitute_args (&substituted, def, is_varargs, &va_arg_name, |
| 1313 | argv, no_loop, lookup_func, lookup_baton); |
| 1314 | |
| 1315 | /* Now `substituted' is the macro's replacement list, with all |
| 1316 | argument values substituted into it properly. Re-scan it for |
| 1317 | macro references, but don't expand invocations of this macro. |
| 1318 | |
| 1319 | We create a new buffer, `substituted_src', which points into |
| 1320 | `substituted', and scan that. We can't scan `substituted' |
| 1321 | itself, since the tokenization process moves the buffer's |
| 1322 | text pointer around, and we still need to be able to find |
| 1323 | `substituted's original text buffer after scanning it so we |
| 1324 | can free it. */ |
| 1325 | struct macro_buffer substituted_src (substituted.text, substituted.len); |
| 1326 | scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); |
| 1327 | |
| 1328 | return 1; |
| 1329 | } |
| 1330 | else |
| 1331 | internal_error (__FILE__, __LINE__, _("bad macro definition kind")); |
| 1332 | } |
| 1333 | |
| 1334 | |
| 1335 | /* If the single token in SRC_FIRST followed by the tokens in SRC_REST |
| 1336 | constitute a macro invokation not forbidden in NO_LOOP, append its |
| 1337 | expansion to DEST and return non-zero. Otherwise, return zero, and |
| 1338 | leave DEST unchanged. |
| 1339 | |
| 1340 | SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. |
| 1341 | SRC_FIRST must be a string built by get_token. */ |
| 1342 | static int |
| 1343 | maybe_expand (struct macro_buffer *dest, |
| 1344 | struct macro_buffer *src_first, |
| 1345 | struct macro_buffer *src_rest, |
| 1346 | struct macro_name_list *no_loop, |
| 1347 | macro_lookup_ftype *lookup_func, |
| 1348 | void *lookup_baton) |
| 1349 | { |
| 1350 | gdb_assert (src_first->shared); |
| 1351 | gdb_assert (src_rest->shared); |
| 1352 | gdb_assert (! dest->shared); |
| 1353 | |
| 1354 | /* Is this token an identifier? */ |
| 1355 | if (src_first->is_identifier) |
| 1356 | { |
| 1357 | /* Make a null-terminated copy of it, since that's what our |
| 1358 | lookup function expects. */ |
| 1359 | std::string id (src_first->text, src_first->len); |
| 1360 | |
| 1361 | /* If we're currently re-scanning the result of expanding |
| 1362 | this macro, don't expand it again. */ |
| 1363 | if (! currently_rescanning (no_loop, id.c_str ())) |
| 1364 | { |
| 1365 | /* Does this identifier have a macro definition in scope? */ |
| 1366 | struct macro_definition *def = lookup_func (id.c_str (), |
| 1367 | lookup_baton); |
| 1368 | |
| 1369 | if (def && expand (id.c_str (), def, dest, src_rest, no_loop, |
| 1370 | lookup_func, lookup_baton)) |
| 1371 | return 1; |
| 1372 | } |
| 1373 | } |
| 1374 | |
| 1375 | return 0; |
| 1376 | } |
| 1377 | |
| 1378 | |
| 1379 | /* Expand macro references in SRC, appending the results to DEST. |
| 1380 | Assume we are re-scanning the result of expanding the macros named |
| 1381 | in NO_LOOP, and don't try to re-expand references to them. |
| 1382 | |
| 1383 | SRC must be a shared buffer; DEST must not be one. */ |
| 1384 | static void |
| 1385 | scan (struct macro_buffer *dest, |
| 1386 | struct macro_buffer *src, |
| 1387 | struct macro_name_list *no_loop, |
| 1388 | macro_lookup_ftype *lookup_func, |
| 1389 | void *lookup_baton) |
| 1390 | { |
| 1391 | gdb_assert (src->shared); |
| 1392 | gdb_assert (! dest->shared); |
| 1393 | |
| 1394 | for (;;) |
| 1395 | { |
| 1396 | struct macro_buffer tok; |
| 1397 | char *original_src_start = src->text; |
| 1398 | |
| 1399 | /* Find the next token in SRC. */ |
| 1400 | if (! get_token (&tok, src)) |
| 1401 | break; |
| 1402 | |
| 1403 | /* Just for aesthetics. If we skipped some whitespace, copy |
| 1404 | that to DEST. */ |
| 1405 | if (tok.text > original_src_start) |
| 1406 | { |
| 1407 | dest->appendmem (original_src_start, tok.text - original_src_start); |
| 1408 | dest->last_token = dest->len; |
| 1409 | } |
| 1410 | |
| 1411 | if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) |
| 1412 | /* We didn't end up expanding tok as a macro reference, so |
| 1413 | simply append it to dest. */ |
| 1414 | append_tokens_without_splicing (dest, &tok); |
| 1415 | } |
| 1416 | |
| 1417 | /* Just for aesthetics. If there was any trailing whitespace in |
| 1418 | src, copy it to dest. */ |
| 1419 | if (src->len) |
| 1420 | { |
| 1421 | dest->appendmem (src->text, src->len); |
| 1422 | dest->last_token = dest->len; |
| 1423 | } |
| 1424 | } |
| 1425 | |
| 1426 | |
| 1427 | gdb::unique_xmalloc_ptr<char> |
| 1428 | macro_expand (const char *source, |
| 1429 | macro_lookup_ftype *lookup_func, |
| 1430 | void *lookup_func_baton) |
| 1431 | { |
| 1432 | struct macro_buffer src (source, strlen (source)); |
| 1433 | |
| 1434 | struct macro_buffer dest (0); |
| 1435 | dest.last_token = 0; |
| 1436 | |
| 1437 | scan (&dest, &src, 0, lookup_func, lookup_func_baton); |
| 1438 | |
| 1439 | dest.appendc ('\0'); |
| 1440 | |
| 1441 | return gdb::unique_xmalloc_ptr<char> (dest.release ()); |
| 1442 | } |
| 1443 | |
| 1444 | |
| 1445 | gdb::unique_xmalloc_ptr<char> |
| 1446 | macro_expand_once (const char *source, |
| 1447 | macro_lookup_ftype *lookup_func, |
| 1448 | void *lookup_func_baton) |
| 1449 | { |
| 1450 | error (_("Expand-once not implemented yet.")); |
| 1451 | } |
| 1452 | |
| 1453 | |
| 1454 | char * |
| 1455 | macro_expand_next (const char **lexptr, |
| 1456 | macro_lookup_ftype *lookup_func, |
| 1457 | void *lookup_baton) |
| 1458 | { |
| 1459 | struct macro_buffer tok; |
| 1460 | |
| 1461 | /* Set up SRC to refer to the input text, pointed to by *lexptr. */ |
| 1462 | struct macro_buffer src (*lexptr, strlen (*lexptr)); |
| 1463 | |
| 1464 | /* Set up DEST to receive the expansion, if there is one. */ |
| 1465 | struct macro_buffer dest (0); |
| 1466 | dest.last_token = 0; |
| 1467 | |
| 1468 | /* Get the text's first preprocessing token. */ |
| 1469 | if (! get_token (&tok, &src)) |
| 1470 | return 0; |
| 1471 | |
| 1472 | /* If it's a macro invocation, expand it. */ |
| 1473 | if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) |
| 1474 | { |
| 1475 | /* It was a macro invocation! Package up the expansion as a |
| 1476 | null-terminated string and return it. Set *lexptr to the |
| 1477 | start of the next token in the input. */ |
| 1478 | dest.appendc ('\0'); |
| 1479 | *lexptr = src.text; |
| 1480 | return dest.release (); |
| 1481 | } |
| 1482 | else |
| 1483 | { |
| 1484 | /* It wasn't a macro invocation. */ |
| 1485 | return 0; |
| 1486 | } |
| 1487 | } |