1/* CPP Library - lexical analysis. 2 Copyright (C) 2000-2015 Free Software Foundation, Inc. 3 Contributed by Per Bothner, 1994-95. 4 Based on CCCP program by Paul Rubin, June 1986 5 Adapted to ANSI C, Richard Stallman, Jan 1987 6 Broken out to separate file, Zack Weinberg, Mar 2000 7 8This program is free software; you can redistribute it and/or modify it 9under the terms of the GNU General Public License as published by the 10Free Software Foundation; either version 3, or (at your option) any 11later version. 12 13This program is distributed in the hope that it will be useful, 14but WITHOUT ANY WARRANTY; without even the implied warranty of 15MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16GNU General Public License for more details. 17 18You should have received a copy of the GNU General Public License 19along with this program; see the file COPYING3. If not see 20<http://www.gnu.org/licenses/>. */ 21 22#include "config.h" 23#include "system.h" 24#include "cpplib.h" 25#include "internal.h" 26 27enum spell_type 28{ 29 SPELL_OPERATOR = 0, 30 SPELL_IDENT, 31 SPELL_LITERAL, 32 SPELL_NONE 33}; 34 35struct token_spelling 36{ 37 enum spell_type category; 38 const unsigned char *name; 39}; 40 41static const unsigned char *const digraph_spellings[] = 42{ UC"%:", UC"%:%:", UC"<:", UC":>", UC"<%", UC"%>" }; 43 44#define OP(e, s) { SPELL_OPERATOR, UC s }, 45#define TK(e, s) { SPELL_ ## s, UC #e }, 46static const struct token_spelling token_spellings[N_TTYPES] = { TTYPE_TABLE }; 47#undef OP 48#undef TK 49 50#define TOKEN_SPELL(token) (token_spellings[(token)->type].category) 51#define TOKEN_NAME(token) (token_spellings[(token)->type].name) 52 53static void add_line_note (cpp_buffer *, const uchar *, unsigned int); 54static int skip_line_comment (cpp_reader *); 55static void skip_whitespace (cpp_reader *, cppchar_t); 56static void lex_string (cpp_reader *, cpp_token *, const uchar *); 57static void save_comment (cpp_reader *, cpp_token *, const uchar *, cppchar_t); 58static void store_comment (cpp_reader *, cpp_token *); 59static void create_literal (cpp_reader *, cpp_token *, const uchar *, 60 unsigned int, enum cpp_ttype); 61static bool warn_in_comment (cpp_reader *, _cpp_line_note *); 62static int name_p (cpp_reader *, const cpp_string *); 63static tokenrun *next_tokenrun (tokenrun *); 64 65static _cpp_buff *new_buff (size_t); 66 67 68/* Utility routine: 69 70 Compares, the token TOKEN to the NUL-terminated string STRING. 71 TOKEN must be a CPP_NAME. Returns 1 for equal, 0 for unequal. */ 72int 73cpp_ideq (const cpp_token *token, const char *string) 74{ 75 if (token->type != CPP_NAME) 76 return 0; 77 78 return !ustrcmp (NODE_NAME (token->val.node.node), (const uchar *) string); 79} 80 81/* Record a note TYPE at byte POS into the current cleaned logical 82 line. */ 83static void 84add_line_note (cpp_buffer *buffer, const uchar *pos, unsigned int type) 85{ 86 if (buffer->notes_used == buffer->notes_cap) 87 { 88 buffer->notes_cap = buffer->notes_cap * 2 + 200; 89 buffer->notes = XRESIZEVEC (_cpp_line_note, buffer->notes, 90 buffer->notes_cap); 91 } 92 93 buffer->notes[buffer->notes_used].pos = pos; 94 buffer->notes[buffer->notes_used].type = type; 95 buffer->notes_used++; 96} 97 98 99/* Fast path to find line special characters using optimized character 100 scanning algorithms. Anything complicated falls back to the slow 101 path below. Since this loop is very hot it's worth doing these kinds 102 of optimizations. 103 104 One of the paths through the ifdefs should provide 105 106 const uchar *search_line_fast (const uchar *s, const uchar *end); 107 108 Between S and END, search for \n, \r, \\, ?. Return a pointer to 109 the found character. 110 111 Note that the last character of the buffer is *always* a newline, 112 as forced by _cpp_convert_input. This fact can be used to avoid 113 explicitly looking for the end of the buffer. */ 114 115/* Configure gives us an ifdef test. */ 116#ifndef WORDS_BIGENDIAN 117#define WORDS_BIGENDIAN 0 118#endif 119 120/* We'd like the largest integer that fits into a register. There's nothing 121 in <stdint.h> that gives us that. For most hosts this is unsigned long, 122 but MS decided on an LLP64 model. Thankfully when building with GCC we 123 can get the "real" word size. */ 124#ifdef __GNUC__ 125typedef unsigned int word_type __attribute__((__mode__(__word__))); 126#else 127typedef unsigned long word_type; 128#endif 129 130/* The code below is only expecting sizes 4 or 8. 131 Die at compile-time if this expectation is violated. */ 132typedef char check_word_type_size 133 [(sizeof(word_type) == 8 || sizeof(word_type) == 4) * 2 - 1]; 134 135/* Return X with the first N bytes forced to values that won't match one 136 of the interesting characters. Note that NUL is not interesting. */ 137 138static inline word_type 139acc_char_mask_misalign (word_type val, unsigned int n) 140{ 141 word_type mask = -1; 142 if (WORDS_BIGENDIAN) 143 mask >>= n * 8; 144 else 145 mask <<= n * 8; 146 return val & mask; 147} 148 149/* Return X replicated to all byte positions within WORD_TYPE. */ 150 151static inline word_type 152acc_char_replicate (uchar x) 153{ 154 word_type ret; 155 156 ret = (x << 24) | (x << 16) | (x << 8) | x; 157 if (sizeof(word_type) == 8) 158 ret = (ret << 16 << 16) | ret; 159 return ret; 160} 161 162/* Return non-zero if some byte of VAL is (probably) C. */ 163 164static inline word_type 165acc_char_cmp (word_type val, word_type c) 166{ 167#if defined(__GNUC__) && defined(__alpha__) 168 /* We can get exact results using a compare-bytes instruction. 169 Get (val == c) via (0 >= (val ^ c)). */ 170 return __builtin_alpha_cmpbge (0, val ^ c); 171#else 172 word_type magic = 0x7efefefeU; 173 if (sizeof(word_type) == 8) 174 magic = (magic << 16 << 16) | 0xfefefefeU; 175 magic |= 1; 176 177 val ^= c; 178 return ((val + magic) ^ ~val) & ~magic; 179#endif 180} 181 182/* Given the result of acc_char_cmp is non-zero, return the index of 183 the found character. If this was a false positive, return -1. */ 184 185static inline int 186acc_char_index (word_type cmp ATTRIBUTE_UNUSED, 187 word_type val ATTRIBUTE_UNUSED) 188{ 189#if defined(__GNUC__) && defined(__alpha__) && !WORDS_BIGENDIAN 190 /* The cmpbge instruction sets *bits* of the result corresponding to 191 matches in the bytes with no false positives. */ 192 return __builtin_ctzl (cmp); 193#else 194 unsigned int i; 195 196 /* ??? It would be nice to force unrolling here, 197 and have all of these constants folded. */ 198 for (i = 0; i < sizeof(word_type); ++i) 199 { 200 uchar c; 201 if (WORDS_BIGENDIAN) 202 c = (val >> (sizeof(word_type) - i - 1) * 8) & 0xff; 203 else 204 c = (val >> i * 8) & 0xff; 205 206 if (c == '\n' || c == '\r' || c == '\\' || c == '?') 207 return i; 208 } 209 210 return -1; 211#endif 212} 213 214/* A version of the fast scanner using bit fiddling techniques. 215 216 For 32-bit words, one would normally perform 16 comparisons and 217 16 branches. With this algorithm one performs 24 arithmetic 218 operations and one branch. Whether this is faster with a 32-bit 219 word size is going to be somewhat system dependent. 220 221 For 64-bit words, we eliminate twice the number of comparisons 222 and branches without increasing the number of arithmetic operations. 223 It's almost certainly going to be a win with 64-bit word size. */ 224 225static const uchar * search_line_acc_char (const uchar *, const uchar *) 226 ATTRIBUTE_UNUSED; 227 228static const uchar * 229search_line_acc_char (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 230{ 231 const word_type repl_nl = acc_char_replicate ('\n'); 232 const word_type repl_cr = acc_char_replicate ('\r'); 233 const word_type repl_bs = acc_char_replicate ('\\'); 234 const word_type repl_qm = acc_char_replicate ('?'); 235 236 unsigned int misalign; 237 const word_type *p; 238 word_type val, t; 239 240 /* Align the buffer. Mask out any bytes from before the beginning. */ 241 p = (word_type *)((uintptr_t)s & -sizeof(word_type)); 242 val = *p; 243 misalign = (uintptr_t)s & (sizeof(word_type) - 1); 244 if (misalign) 245 val = acc_char_mask_misalign (val, misalign); 246 247 /* Main loop. */ 248 while (1) 249 { 250 t = acc_char_cmp (val, repl_nl); 251 t |= acc_char_cmp (val, repl_cr); 252 t |= acc_char_cmp (val, repl_bs); 253 t |= acc_char_cmp (val, repl_qm); 254 255 if (__builtin_expect (t != 0, 0)) 256 { 257 int i = acc_char_index (t, val); 258 if (i >= 0) 259 return (const uchar *)p + i; 260 } 261 262 val = *++p; 263 } 264} 265 266/* Disable on Solaris 2/x86 until the following problem can be properly 267 autoconfed: 268 269 The Solaris 10+ assembler tags objects with the instruction set 270 extensions used, so SSE4.2 executables cannot run on machines that 271 don't support that extension. */ 272 273#if (GCC_VERSION >= 4005) && (__GNUC__ >= 5 || !defined(__PIC__)) && (defined(__i386__) || defined(__x86_64__)) && !(defined(__sun__) && defined(__svr4__)) 274 275/* Replicated character data to be shared between implementations. 276 Recall that outside of a context with vector support we can't 277 define compatible vector types, therefore these are all defined 278 in terms of raw characters. */ 279static const char repl_chars[4][16] __attribute__((aligned(16))) = { 280 { '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', 281 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' }, 282 { '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', 283 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' }, 284 { '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', 285 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' }, 286 { '?', '?', '?', '?', '?', '?', '?', '?', 287 '?', '?', '?', '?', '?', '?', '?', '?' }, 288}; 289 290/* A version of the fast scanner using MMX vectorized byte compare insns. 291 292 This uses the PMOVMSKB instruction which was introduced with "MMX2", 293 which was packaged into SSE1; it is also present in the AMD MMX 294 extension. Mark the function as using "sse" so that we emit a real 295 "emms" instruction, rather than the 3dNOW "femms" instruction. */ 296 297static const uchar * 298#ifndef __SSE__ 299__attribute__((__target__("sse"))) 300#endif 301search_line_mmx (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 302{ 303 typedef char v8qi __attribute__ ((__vector_size__ (8))); 304 typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__)); 305 306 const v8qi repl_nl = *(const v8qi *)repl_chars[0]; 307 const v8qi repl_cr = *(const v8qi *)repl_chars[1]; 308 const v8qi repl_bs = *(const v8qi *)repl_chars[2]; 309 const v8qi repl_qm = *(const v8qi *)repl_chars[3]; 310 311 unsigned int misalign, found, mask; 312 const v8qi *p; 313 v8qi data, t, c; 314 315 /* Align the source pointer. While MMX doesn't generate unaligned data 316 faults, this allows us to safely scan to the end of the buffer without 317 reading beyond the end of the last page. */ 318 misalign = (uintptr_t)s & 7; 319 p = (const v8qi *)((uintptr_t)s & -8); 320 data = *p; 321 322 /* Create a mask for the bytes that are valid within the first 323 16-byte block. The Idea here is that the AND with the mask 324 within the loop is "free", since we need some AND or TEST 325 insn in order to set the flags for the branch anyway. */ 326 mask = -1u << misalign; 327 328 /* Main loop processing 8 bytes at a time. */ 329 goto start; 330 do 331 { 332 data = *++p; 333 mask = -1; 334 335 start: 336 t = __builtin_ia32_pcmpeqb(data, repl_nl); 337 c = __builtin_ia32_pcmpeqb(data, repl_cr); 338 t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); 339 c = __builtin_ia32_pcmpeqb(data, repl_bs); 340 t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); 341 c = __builtin_ia32_pcmpeqb(data, repl_qm); 342 t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); 343 found = __builtin_ia32_pmovmskb (t); 344 found &= mask; 345 } 346 while (!found); 347 348 __builtin_ia32_emms (); 349 350 /* FOUND contains 1 in bits for which we matched a relevant 351 character. Conversion to the byte index is trivial. */ 352 found = __builtin_ctz(found); 353 return (const uchar *)p + found; 354} 355 356/* A version of the fast scanner using SSE2 vectorized byte compare insns. */ 357 358static const uchar * 359#ifndef __SSE2__ 360__attribute__((__target__("sse2"))) 361#endif 362search_line_sse2 (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 363{ 364 typedef char v16qi __attribute__ ((__vector_size__ (16))); 365 366 const v16qi repl_nl = *(const v16qi *)repl_chars[0]; 367 const v16qi repl_cr = *(const v16qi *)repl_chars[1]; 368 const v16qi repl_bs = *(const v16qi *)repl_chars[2]; 369 const v16qi repl_qm = *(const v16qi *)repl_chars[3]; 370 371 unsigned int misalign, found, mask; 372 const v16qi *p; 373 v16qi data, t; 374 375 /* Align the source pointer. */ 376 misalign = (uintptr_t)s & 15; 377 p = (const v16qi *)((uintptr_t)s & -16); 378 data = *p; 379 380 /* Create a mask for the bytes that are valid within the first 381 16-byte block. The Idea here is that the AND with the mask 382 within the loop is "free", since we need some AND or TEST 383 insn in order to set the flags for the branch anyway. */ 384 mask = -1u << misalign; 385 386 /* Main loop processing 16 bytes at a time. */ 387 goto start; 388 do 389 { 390 data = *++p; 391 mask = -1; 392 393 start: 394 t = __builtin_ia32_pcmpeqb128(data, repl_nl); 395 t |= __builtin_ia32_pcmpeqb128(data, repl_cr); 396 t |= __builtin_ia32_pcmpeqb128(data, repl_bs); 397 t |= __builtin_ia32_pcmpeqb128(data, repl_qm); 398 found = __builtin_ia32_pmovmskb128 (t); 399 found &= mask; 400 } 401 while (!found); 402 403 /* FOUND contains 1 in bits for which we matched a relevant 404 character. Conversion to the byte index is trivial. */ 405 found = __builtin_ctz(found); 406 return (const uchar *)p + found; 407} 408 409#ifdef HAVE_SSE4 410/* A version of the fast scanner using SSE 4.2 vectorized string insns. */ 411 412static const uchar * 413#ifndef __SSE4_2__ 414__attribute__((__target__("sse4.2"))) 415#endif 416search_line_sse42 (const uchar *s, const uchar *end) 417{ 418 typedef char v16qi __attribute__ ((__vector_size__ (16))); 419 static const v16qi search = { '\n', '\r', '?', '\\' }; 420 421 uintptr_t si = (uintptr_t)s; 422 uintptr_t index; 423 424 /* Check for unaligned input. */ 425 if (si & 15) 426 { 427 v16qi sv; 428 429 if (__builtin_expect (end - s < 16, 0) 430 && __builtin_expect ((si & 0xfff) > 0xff0, 0)) 431 { 432 /* There are less than 16 bytes left in the buffer, and less 433 than 16 bytes left on the page. Reading 16 bytes at this 434 point might generate a spurious page fault. Defer to the 435 SSE2 implementation, which already handles alignment. */ 436 return search_line_sse2 (s, end); 437 } 438 439 /* ??? The builtin doesn't understand that the PCMPESTRI read from 440 memory need not be aligned. */ 441 sv = __builtin_ia32_loaddqu ((const char *) s); 442 index = __builtin_ia32_pcmpestri128 (search, 4, sv, 16, 0); 443 444 if (__builtin_expect (index < 16, 0)) 445 goto found; 446 447 /* Advance the pointer to an aligned address. We will re-scan a 448 few bytes, but we no longer need care for reading past the 449 end of a page, since we're guaranteed a match. */ 450 s = (const uchar *)((si + 16) & -16); 451 } 452 453 /* Main loop, processing 16 bytes at a time. By doing the whole loop 454 in inline assembly, we can make proper use of the flags set. */ 455 __asm ( "sub $16, %1\n" 456 " .balign 16\n" 457 "0: add $16, %1\n" 458 " %vpcmpestri $0, (%1), %2\n" 459 " jnc 0b" 460 : "=&c"(index), "+r"(s) 461 : "x"(search), "a"(4), "d"(16)); 462 463 found: 464 return s + index; 465} 466 467#else 468/* Work around out-dated assemblers without sse4 support. */ 469#define search_line_sse42 search_line_sse2 470#endif 471 472/* Check the CPU capabilities. */ 473 474#include "../gcc/config/i386/cpuid.h" 475 476typedef const uchar * (*search_line_fast_type) (const uchar *, const uchar *); 477static search_line_fast_type search_line_fast; 478 479#define HAVE_init_vectorized_lexer 1 480static inline void 481init_vectorized_lexer (void) 482{ 483 unsigned dummy, ecx = 0, edx = 0; 484 search_line_fast_type impl = search_line_acc_char; 485 int minimum = 0; 486 487#if defined(__SSE4_2__) 488 minimum = 3; 489#elif defined(__SSE2__) 490 minimum = 2; 491#elif defined(__SSE__) 492 minimum = 1; 493#endif 494 495 if (minimum == 3) 496 impl = search_line_sse42; 497 else if (__get_cpuid (1, &dummy, &dummy, &ecx, &edx) || minimum == 2) 498 { 499 if (minimum == 3 || (ecx & bit_SSE4_2)) 500 impl = search_line_sse42; 501 else if (minimum == 2 || (edx & bit_SSE2)) 502 impl = search_line_sse2; 503 else if (minimum == 1 || (edx & bit_SSE)) 504 impl = search_line_mmx; 505 } 506 else if (__get_cpuid (0x80000001, &dummy, &dummy, &dummy, &edx)) 507 { 508 if (minimum == 1 509 || (edx & (bit_MMXEXT | bit_CMOV)) == (bit_MMXEXT | bit_CMOV)) 510 impl = search_line_mmx; 511 } 512 513 search_line_fast = impl; 514} 515 516#elif defined(_ARCH_PWR8) && defined(__ALTIVEC__) 517 518/* A vection of the fast scanner using AltiVec vectorized byte compares 519 and VSX unaligned loads (when VSX is available). This is otherwise 520 the same as the pre-GCC 5 version. */ 521 522ATTRIBUTE_NO_SANITIZE_UNDEFINED 523static const uchar * 524search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 525{ 526 typedef __attribute__((altivec(vector))) unsigned char vc; 527 528 const vc repl_nl = { 529 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', 530 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' 531 }; 532 const vc repl_cr = { 533 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', 534 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' 535 }; 536 const vc repl_bs = { 537 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', 538 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' 539 }; 540 const vc repl_qm = { 541 '?', '?', '?', '?', '?', '?', '?', '?', 542 '?', '?', '?', '?', '?', '?', '?', '?', 543 }; 544 const vc zero = { 0 }; 545 546 vc data, t; 547 548 /* Main loop processing 16 bytes at a time. */ 549 do 550 { 551 vc m_nl, m_cr, m_bs, m_qm; 552 553 data = *((const vc *)s); 554 s += 16; 555 556 m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl); 557 m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr); 558 m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs); 559 m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm); 560 t = (m_nl | m_cr) | (m_bs | m_qm); 561 562 /* T now contains 0xff in bytes for which we matched one of the relevant 563 characters. We want to exit the loop if any byte in T is non-zero. 564 Below is the expansion of vec_any_ne(t, zero). */ 565 } 566 while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero)); 567 568 /* Restore s to to point to the 16 bytes we just processed. */ 569 s -= 16; 570 571 { 572#define N (sizeof(vc) / sizeof(long)) 573 574 union { 575 vc v; 576 /* Statically assert that N is 2 or 4. */ 577 unsigned long l[(N == 2 || N == 4) ? N : -1]; 578 } u; 579 unsigned long l, i = 0; 580 581 u.v = t; 582 583 /* Find the first word of T that is non-zero. */ 584 switch (N) 585 { 586 case 4: 587 l = u.l[i++]; 588 if (l != 0) 589 break; 590 s += sizeof(unsigned long); 591 l = u.l[i++]; 592 if (l != 0) 593 break; 594 s += sizeof(unsigned long); 595 case 2: 596 l = u.l[i++]; 597 if (l != 0) 598 break; 599 s += sizeof(unsigned long); 600 l = u.l[i]; 601 } 602 603 /* L now contains 0xff in bytes for which we matched one of the 604 relevant characters. We can find the byte index by finding 605 its bit index and dividing by 8. */ 606#ifdef __BIG_ENDIAN__ 607 l = __builtin_clzl(l) >> 3; 608#else 609 l = __builtin_ctzl(l) >> 3; 610#endif 611 return s + l; 612 613#undef N 614 } 615} 616 617#elif (GCC_VERSION >= 4005) && defined(__ALTIVEC__) && defined (__BIG_ENDIAN__) 618 619/* A vection of the fast scanner using AltiVec vectorized byte compares. 620 This cannot be used for little endian because vec_lvsl/lvsr are 621 deprecated for little endian and the code won't work properly. */ 622/* ??? Unfortunately, attribute(target("altivec")) is not yet supported, 623 so we can't compile this function without -maltivec on the command line 624 (or implied by some other switch). */ 625 626static const uchar * 627search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 628{ 629 typedef __attribute__((altivec(vector))) unsigned char vc; 630 631 const vc repl_nl = { 632 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', 633 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' 634 }; 635 const vc repl_cr = { 636 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', 637 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' 638 }; 639 const vc repl_bs = { 640 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', 641 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' 642 }; 643 const vc repl_qm = { 644 '?', '?', '?', '?', '?', '?', '?', '?', 645 '?', '?', '?', '?', '?', '?', '?', '?', 646 }; 647 const vc ones = { 648 -1, -1, -1, -1, -1, -1, -1, -1, 649 -1, -1, -1, -1, -1, -1, -1, -1, 650 }; 651 const vc zero = { 0 }; 652 653 vc data, mask, t; 654 655 /* Altivec loads automatically mask addresses with -16. This lets us 656 issue the first load as early as possible. */ 657 data = __builtin_vec_ld(0, (const vc *)s); 658 659 /* Discard bytes before the beginning of the buffer. Do this by 660 beginning with all ones and shifting in zeros according to the 661 mis-alignment. The LVSR instruction pulls the exact shift we 662 want from the address. */ 663 mask = __builtin_vec_lvsr(0, s); 664 mask = __builtin_vec_perm(zero, ones, mask); 665 data &= mask; 666 667 /* While altivec loads mask addresses, we still need to align S so 668 that the offset we compute at the end is correct. */ 669 s = (const uchar *)((uintptr_t)s & -16); 670 671 /* Main loop processing 16 bytes at a time. */ 672 goto start; 673 do 674 { 675 vc m_nl, m_cr, m_bs, m_qm; 676 677 s += 16; 678 data = __builtin_vec_ld(0, (const vc *)s); 679 680 start: 681 m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl); 682 m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr); 683 m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs); 684 m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm); 685 t = (m_nl | m_cr) | (m_bs | m_qm); 686 687 /* T now contains 0xff in bytes for which we matched one of the relevant 688 characters. We want to exit the loop if any byte in T is non-zero. 689 Below is the expansion of vec_any_ne(t, zero). */ 690 } 691 while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero)); 692 693 { 694#define N (sizeof(vc) / sizeof(long)) 695 696 union { 697 vc v; 698 /* Statically assert that N is 2 or 4. */ 699 unsigned long l[(N == 2 || N == 4) ? N : -1]; 700 } u; 701 unsigned long l, i = 0; 702 703 u.v = t; 704 705 /* Find the first word of T that is non-zero. */ 706 switch (N) 707 { 708 case 4: 709 l = u.l[i++]; 710 if (l != 0) 711 break; 712 s += sizeof(unsigned long); 713 l = u.l[i++]; 714 if (l != 0) 715 break; 716 s += sizeof(unsigned long); 717 case 2: 718 l = u.l[i++]; 719 if (l != 0) 720 break; 721 s += sizeof(unsigned long); 722 l = u.l[i]; 723 } 724 725 /* L now contains 0xff in bytes for which we matched one of the 726 relevant characters. We can find the byte index by finding 727 its bit index and dividing by 8. */ 728 l = __builtin_clzl(l) >> 3; 729 return s + l; 730 731#undef N 732 } 733} 734 735#elif defined (__ARM_NEON) 736#include "arm_neon.h" 737 738static const uchar * 739search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 740{ 741 const uint8x16_t repl_nl = vdupq_n_u8 ('\n'); 742 const uint8x16_t repl_cr = vdupq_n_u8 ('\r'); 743 const uint8x16_t repl_bs = vdupq_n_u8 ('\\'); 744 const uint8x16_t repl_qm = vdupq_n_u8 ('?'); 745 const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL); 746 747 unsigned int misalign, found, mask; 748 const uint8_t *p; 749 uint8x16_t data; 750 751 /* Align the source pointer. */ 752 misalign = (uintptr_t)s & 15; 753 p = (const uint8_t *)((uintptr_t)s & -16); 754 data = vld1q_u8 (p); 755 756 /* Create a mask for the bytes that are valid within the first 757 16-byte block. The Idea here is that the AND with the mask 758 within the loop is "free", since we need some AND or TEST 759 insn in order to set the flags for the branch anyway. */ 760 mask = (-1u << misalign) & 0xffff; 761 762 /* Main loop, processing 16 bytes at a time. */ 763 goto start; 764 765 do 766 { 767 uint8x8_t l; 768 uint16x4_t m; 769 uint32x2_t n; 770 uint8x16_t t, u, v, w; 771 772 p += 16; 773 data = vld1q_u8 (p); 774 mask = 0xffff; 775 776 start: 777 t = vceqq_u8 (data, repl_nl); 778 u = vceqq_u8 (data, repl_cr); 779 v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); 780 w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); 781 t = vandq_u8 (vorrq_u8 (v, w), xmask); 782 l = vpadd_u8 (vget_low_u8 (t), vget_high_u8 (t)); 783 m = vpaddl_u8 (l); 784 n = vpaddl_u16 (m); 785 786 found = vget_lane_u32 ((uint32x2_t) vorr_u64 ((uint64x1_t) n, 787 vshr_n_u64 ((uint64x1_t) n, 24)), 0); 788 found &= mask; 789 } 790 while (!found); 791 792 /* FOUND contains 1 in bits for which we matched a relevant 793 character. Conversion to the byte index is trivial. */ 794 found = __builtin_ctz (found); 795 return (const uchar *)p + found; 796} 797 798#else 799 800/* We only have one accellerated alternative. Use a direct call so that 801 we encourage inlining. */ 802 803#define search_line_fast search_line_acc_char 804 805#endif 806 807/* Initialize the lexer if needed. */ 808 809void 810_cpp_init_lexer (void) 811{ 812#ifdef HAVE_init_vectorized_lexer 813 init_vectorized_lexer (); 814#endif 815} 816 817/* Returns with a logical line that contains no escaped newlines or 818 trigraphs. This is a time-critical inner loop. */ 819void 820_cpp_clean_line (cpp_reader *pfile) 821{ 822 cpp_buffer *buffer; 823 const uchar *s; 824 uchar c, *d, *p; 825 826 buffer = pfile->buffer; 827 buffer->cur_note = buffer->notes_used = 0; 828 buffer->cur = buffer->line_base = buffer->next_line; 829 buffer->need_line = false; 830 s = buffer->next_line; 831 832 if (!buffer->from_stage3) 833 { 834 const uchar *pbackslash = NULL; 835 836 /* Fast path. This is the common case of an un-escaped line with 837 no trigraphs. The primary win here is by not writing any 838 data back to memory until we have to. */ 839 while (1) 840 { 841 /* Perform an optimized search for \n, \r, \\, ?. */ 842 s = search_line_fast (s, buffer->rlimit); 843 844 c = *s; 845 if (c == '\\') 846 { 847 /* Record the location of the backslash and continue. */ 848 pbackslash = s++; 849 } 850 else if (__builtin_expect (c == '?', 0)) 851 { 852 if (__builtin_expect (s[1] == '?', false) 853 && _cpp_trigraph_map[s[2]]) 854 { 855 /* Have a trigraph. We may or may not have to convert 856 it. Add a line note regardless, for -Wtrigraphs. */ 857 add_line_note (buffer, s, s[2]); 858 if (CPP_OPTION (pfile, trigraphs)) 859 { 860 /* We do, and that means we have to switch to the 861 slow path. */ 862 d = (uchar *) s; 863 *d = _cpp_trigraph_map[s[2]]; 864 s += 2; 865 goto slow_path; 866 } 867 } 868 /* Not a trigraph. Continue on fast-path. */ 869 s++; 870 } 871 else 872 break; 873 } 874 875 /* This must be \r or \n. We're either done, or we'll be forced 876 to write back to the buffer and continue on the slow path. */ 877 d = (uchar *) s; 878 879 if (__builtin_expect (s == buffer->rlimit, false)) 880 goto done; 881 882 /* DOS line ending? */ 883 if (__builtin_expect (c == '\r', false) && s[1] == '\n') 884 { 885 s++; 886 if (s == buffer->rlimit) 887 goto done; 888 } 889 890 if (__builtin_expect (pbackslash == NULL, true)) 891 goto done; 892 893 /* Check for escaped newline. */ 894 p = d; 895 while (is_nvspace (p[-1])) 896 p--; 897 if (p - 1 != pbackslash) 898 goto done; 899 900 /* Have an escaped newline; process it and proceed to 901 the slow path. */ 902 add_line_note (buffer, p - 1, p != d ? ' ' : '\\'); 903 d = p - 2; 904 buffer->next_line = p - 1; 905 906 slow_path: 907 while (1) 908 { 909 c = *++s; 910 *++d = c; 911 912 if (c == '\n' || c == '\r') 913 { 914 /* Handle DOS line endings. */ 915 if (c == '\r' && s != buffer->rlimit && s[1] == '\n') 916 s++; 917 if (s == buffer->rlimit) 918 break; 919 920 /* Escaped? */ 921 p = d; 922 while (p != buffer->next_line && is_nvspace (p[-1])) 923 p--; 924 if (p == buffer->next_line || p[-1] != '\\') 925 break; 926 927 add_line_note (buffer, p - 1, p != d ? ' ': '\\'); 928 d = p - 2; 929 buffer->next_line = p - 1; 930 } 931 else if (c == '?' && s[1] == '?' && _cpp_trigraph_map[s[2]]) 932 { 933 /* Add a note regardless, for the benefit of -Wtrigraphs. */ 934 add_line_note (buffer, d, s[2]); 935 if (CPP_OPTION (pfile, trigraphs)) 936 { 937 *d = _cpp_trigraph_map[s[2]]; 938 s += 2; 939 } 940 } 941 } 942 } 943 else 944 { 945 while (*s != '\n' && *s != '\r') 946 s++; 947 d = (uchar *) s; 948 949 /* Handle DOS line endings. */ 950 if (*s == '\r' && s != buffer->rlimit && s[1] == '\n') 951 s++; 952 } 953 954 done: 955 *d = '\n'; 956 /* A sentinel note that should never be processed. */ 957 add_line_note (buffer, d + 1, '\n'); 958 buffer->next_line = s + 1; 959} 960 961/* Return true if the trigraph indicated by NOTE should be warned 962 about in a comment. */ 963static bool 964warn_in_comment (cpp_reader *pfile, _cpp_line_note *note) 965{ 966 const uchar *p; 967 968 /* Within comments we don't warn about trigraphs, unless the 969 trigraph forms an escaped newline, as that may change 970 behavior. */ 971 if (note->type != '/') 972 return false; 973 974 /* If -trigraphs, then this was an escaped newline iff the next note 975 is coincident. */ 976 if (CPP_OPTION (pfile, trigraphs)) 977 return note[1].pos == note->pos; 978 979 /* Otherwise, see if this forms an escaped newline. */ 980 p = note->pos + 3; 981 while (is_nvspace (*p)) 982 p++; 983 984 /* There might have been escaped newlines between the trigraph and the 985 newline we found. Hence the position test. */ 986 return (*p == '\n' && p < note[1].pos); 987} 988 989/* Process the notes created by add_line_note as far as the current 990 location. */ 991void 992_cpp_process_line_notes (cpp_reader *pfile, int in_comment) 993{ 994 cpp_buffer *buffer = pfile->buffer; 995 996 for (;;) 997 { 998 _cpp_line_note *note = &buffer->notes[buffer->cur_note]; 999 unsigned int col; 1000 1001 if (note->pos > buffer->cur) 1002 break; 1003 1004 buffer->cur_note++; 1005 col = CPP_BUF_COLUMN (buffer, note->pos + 1); 1006 1007 if (note->type == '\\' || note->type == ' ') 1008 { 1009 if (note->type == ' ' && !in_comment) 1010 cpp_error_with_line (pfile, CPP_DL_WARNING, pfile->line_table->highest_line, col, 1011 "backslash and newline separated by space"); 1012 1013 if (buffer->next_line > buffer->rlimit) 1014 { 1015 cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, col, 1016 "backslash-newline at end of file"); 1017 /* Prevent "no newline at end of file" warning. */ 1018 buffer->next_line = buffer->rlimit; 1019 } 1020 1021 buffer->line_base = note->pos; 1022 CPP_INCREMENT_LINE (pfile, 0); 1023 } 1024 else if (_cpp_trigraph_map[note->type]) 1025 { 1026 if (CPP_OPTION (pfile, warn_trigraphs) 1027 && (!in_comment || warn_in_comment (pfile, note))) 1028 { 1029 if (CPP_OPTION (pfile, trigraphs)) 1030 cpp_warning_with_line (pfile, CPP_W_TRIGRAPHS, 1031 pfile->line_table->highest_line, col, 1032 "trigraph ??%c converted to %c", 1033 note->type, 1034 (int) _cpp_trigraph_map[note->type]); 1035 else 1036 { 1037 cpp_warning_with_line 1038 (pfile, CPP_W_TRIGRAPHS, 1039 pfile->line_table->highest_line, col, 1040 "trigraph ??%c ignored, use -trigraphs to enable", 1041 note->type); 1042 } 1043 } 1044 } 1045 else if (note->type == 0) 1046 /* Already processed in lex_raw_string. */; 1047 else 1048 abort (); 1049 } 1050} 1051 1052/* Skip a C-style block comment. We find the end of the comment by 1053 seeing if an asterisk is before every '/' we encounter. Returns 1054 nonzero if comment terminated by EOF, zero otherwise. 1055 1056 Buffer->cur points to the initial asterisk of the comment. */ 1057bool 1058_cpp_skip_block_comment (cpp_reader *pfile) 1059{ 1060 cpp_buffer *buffer = pfile->buffer; 1061 const uchar *cur = buffer->cur; 1062 uchar c; 1063 1064 cur++; 1065 if (*cur == '/') 1066 cur++; 1067 1068 for (;;) 1069 { 1070 /* People like decorating comments with '*', so check for '/' 1071 instead for efficiency. */ 1072 c = *cur++; 1073 1074 if (c == '/') 1075 { 1076 if (cur[-2] == '*') 1077 break; 1078 1079 /* Warn about potential nested comments, but not if the '/' 1080 comes immediately before the true comment delimiter. 1081 Don't bother to get it right across escaped newlines. */ 1082 if (CPP_OPTION (pfile, warn_comments) 1083 && cur[0] == '*' && cur[1] != '/') 1084 { 1085 buffer->cur = cur; 1086 cpp_warning_with_line (pfile, CPP_W_COMMENTS, 1087 pfile->line_table->highest_line, 1088 CPP_BUF_COL (buffer), 1089 "\"/*\" within comment"); 1090 } 1091 } 1092 else if (c == '\n') 1093 { 1094 unsigned int cols; 1095 buffer->cur = cur - 1; 1096 _cpp_process_line_notes (pfile, true); 1097 if (buffer->next_line >= buffer->rlimit) 1098 return true; 1099 _cpp_clean_line (pfile); 1100 1101 cols = buffer->next_line - buffer->line_base; 1102 CPP_INCREMENT_LINE (pfile, cols); 1103 1104 cur = buffer->cur; 1105 } 1106 } 1107 1108 buffer->cur = cur; 1109 _cpp_process_line_notes (pfile, true); 1110 return false; 1111} 1112 1113/* Skip a C++ line comment, leaving buffer->cur pointing to the 1114 terminating newline. Handles escaped newlines. Returns nonzero 1115 if a multiline comment. */ 1116static int 1117skip_line_comment (cpp_reader *pfile) 1118{ 1119 cpp_buffer *buffer = pfile->buffer; 1120 source_location orig_line = pfile->line_table->highest_line; 1121 1122 while (*buffer->cur != '\n') 1123 buffer->cur++; 1124 1125 _cpp_process_line_notes (pfile, true); 1126 return orig_line != pfile->line_table->highest_line; 1127} 1128 1129/* Skips whitespace, saving the next non-whitespace character. */ 1130static void 1131skip_whitespace (cpp_reader *pfile, cppchar_t c) 1132{ 1133 cpp_buffer *buffer = pfile->buffer; 1134 bool saw_NUL = false; 1135 1136 do 1137 { 1138 /* Horizontal space always OK. */ 1139 if (c == ' ' || c == '\t') 1140 ; 1141 /* Just \f \v or \0 left. */ 1142 else if (c == '\0') 1143 saw_NUL = true; 1144 else if (pfile->state.in_directive && CPP_PEDANTIC (pfile)) 1145 cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, 1146 CPP_BUF_COL (buffer), 1147 "%s in preprocessing directive", 1148 c == '\f' ? "form feed" : "vertical tab"); 1149 1150 c = *buffer->cur++; 1151 } 1152 /* We only want non-vertical space, i.e. ' ' \t \f \v \0. */ 1153 while (is_nvspace (c)); 1154 1155 if (saw_NUL) 1156 cpp_error (pfile, CPP_DL_WARNING, "null character(s) ignored"); 1157 1158 buffer->cur--; 1159} 1160 1161/* See if the characters of a number token are valid in a name (no 1162 '.', '+' or '-'). */ 1163static int 1164name_p (cpp_reader *pfile, const cpp_string *string) 1165{ 1166 unsigned int i; 1167 1168 for (i = 0; i < string->len; i++) 1169 if (!is_idchar (string->text[i])) 1170 return 0; 1171 1172 return 1; 1173} 1174 1175/* After parsing an identifier or other sequence, produce a warning about 1176 sequences not in NFC/NFKC. */ 1177static void 1178warn_about_normalization (cpp_reader *pfile, 1179 const cpp_token *token, 1180 const struct normalize_state *s) 1181{ 1182 if (CPP_OPTION (pfile, warn_normalize) < NORMALIZE_STATE_RESULT (s) 1183 && !pfile->state.skipping) 1184 { 1185 /* Make sure that the token is printed using UCNs, even 1186 if we'd otherwise happily print UTF-8. */ 1187 unsigned char *buf = XNEWVEC (unsigned char, cpp_token_len (token)); 1188 size_t sz; 1189 1190 sz = cpp_spell_token (pfile, token, buf, false) - buf; 1191 if (NORMALIZE_STATE_RESULT (s) == normalized_C) 1192 cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0, 1193 "`%.*s' is not in NFKC", (int) sz, buf); 1194 else 1195 cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0, 1196 "`%.*s' is not in NFC", (int) sz, buf); 1197 free (buf); 1198 } 1199} 1200 1201/* Returns TRUE if the sequence starting at buffer->cur is invalid in 1202 an identifier. FIRST is TRUE if this starts an identifier. */ 1203static bool 1204forms_identifier_p (cpp_reader *pfile, int first, 1205 struct normalize_state *state) 1206{ 1207 cpp_buffer *buffer = pfile->buffer; 1208 1209 if (*buffer->cur == '$') 1210 { 1211 if (!CPP_OPTION (pfile, dollars_in_ident)) 1212 return false; 1213 1214 buffer->cur++; 1215 if (CPP_OPTION (pfile, warn_dollars) && !pfile->state.skipping) 1216 { 1217 CPP_OPTION (pfile, warn_dollars) = 0; 1218 cpp_error (pfile, CPP_DL_PEDWARN, "'$' in identifier or number"); 1219 } 1220 1221 return true; 1222 } 1223 1224 /* Is this a syntactically valid UCN? */ 1225 if (CPP_OPTION (pfile, extended_identifiers) 1226 && *buffer->cur == '\\' 1227 && (buffer->cur[1] == 'u' || buffer->cur[1] == 'U')) 1228 { 1229 buffer->cur += 2; 1230 if (_cpp_valid_ucn (pfile, &buffer->cur, buffer->rlimit, 1 + !first, 1231 state)) 1232 return true; 1233 buffer->cur -= 2; 1234 } 1235 1236 return false; 1237} 1238 1239/* Helper function to get the cpp_hashnode of the identifier BASE. */ 1240static cpp_hashnode * 1241lex_identifier_intern (cpp_reader *pfile, const uchar *base) 1242{ 1243 cpp_hashnode *result; 1244 const uchar *cur; 1245 unsigned int len; 1246 unsigned int hash = HT_HASHSTEP (0, *base); 1247 1248 cur = base + 1; 1249 while (ISIDNUM (*cur)) 1250 { 1251 hash = HT_HASHSTEP (hash, *cur); 1252 cur++; 1253 } 1254 len = cur - base; 1255 hash = HT_HASHFINISH (hash, len); 1256 result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 1257 base, len, hash, HT_ALLOC)); 1258 1259 /* Rarely, identifiers require diagnostics when lexed. */ 1260 if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) 1261 && !pfile->state.skipping, 0)) 1262 { 1263 /* It is allowed to poison the same identifier twice. */ 1264 if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) 1265 cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", 1266 NODE_NAME (result)); 1267 1268 /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the 1269 replacement list of a variadic macro. */ 1270 if (result == pfile->spec_nodes.n__VA_ARGS__ 1271 && !pfile->state.va_args_ok) 1272 { 1273 if (CPP_OPTION (pfile, cplusplus)) 1274 cpp_error (pfile, CPP_DL_PEDWARN, 1275 "__VA_ARGS__ can only appear in the expansion" 1276 " of a C++11 variadic macro"); 1277 else 1278 cpp_error (pfile, CPP_DL_PEDWARN, 1279 "__VA_ARGS__ can only appear in the expansion" 1280 " of a C99 variadic macro"); 1281 } 1282 1283 /* For -Wc++-compat, warn about use of C++ named operators. */ 1284 if (result->flags & NODE_WARN_OPERATOR) 1285 cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, 1286 "identifier \"%s\" is a special operator name in C++", 1287 NODE_NAME (result)); 1288 } 1289 1290 return result; 1291} 1292 1293/* Get the cpp_hashnode of an identifier specified by NAME in 1294 the current cpp_reader object. If none is found, NULL is returned. */ 1295cpp_hashnode * 1296_cpp_lex_identifier (cpp_reader *pfile, const char *name) 1297{ 1298 cpp_hashnode *result; 1299 result = lex_identifier_intern (pfile, (uchar *) name); 1300 return result; 1301} 1302 1303/* Lex an identifier starting at BUFFER->CUR - 1. */ 1304static cpp_hashnode * 1305lex_identifier (cpp_reader *pfile, const uchar *base, bool starts_ucn, 1306 struct normalize_state *nst, cpp_hashnode **spelling) 1307{ 1308 cpp_hashnode *result; 1309 const uchar *cur; 1310 unsigned int len; 1311 unsigned int hash = HT_HASHSTEP (0, *base); 1312 1313 cur = pfile->buffer->cur; 1314 if (! starts_ucn) 1315 { 1316 while (ISIDNUM (*cur)) 1317 { 1318 hash = HT_HASHSTEP (hash, *cur); 1319 cur++; 1320 } 1321 NORMALIZE_STATE_UPDATE_IDNUM (nst, *(cur - 1)); 1322 } 1323 pfile->buffer->cur = cur; 1324 if (starts_ucn || forms_identifier_p (pfile, false, nst)) 1325 { 1326 /* Slower version for identifiers containing UCNs (or $). */ 1327 do { 1328 while (ISIDNUM (*pfile->buffer->cur)) 1329 { 1330 NORMALIZE_STATE_UPDATE_IDNUM (nst, *pfile->buffer->cur); 1331 pfile->buffer->cur++; 1332 } 1333 } while (forms_identifier_p (pfile, false, nst)); 1334 result = _cpp_interpret_identifier (pfile, base, 1335 pfile->buffer->cur - base); 1336 *spelling = cpp_lookup (pfile, base, pfile->buffer->cur - base); 1337 } 1338 else 1339 { 1340 len = cur - base; 1341 hash = HT_HASHFINISH (hash, len); 1342 1343 result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 1344 base, len, hash, HT_ALLOC)); 1345 *spelling = result; 1346 } 1347 1348 /* Rarely, identifiers require diagnostics when lexed. */ 1349 if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) 1350 && !pfile->state.skipping, 0)) 1351 { 1352 /* It is allowed to poison the same identifier twice. */ 1353 if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) 1354 cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", 1355 NODE_NAME (result)); 1356 1357 /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the 1358 replacement list of a variadic macro. */ 1359 if (result == pfile->spec_nodes.n__VA_ARGS__ 1360 && !pfile->state.va_args_ok) 1361 { 1362 if (CPP_OPTION (pfile, cplusplus)) 1363 cpp_error (pfile, CPP_DL_PEDWARN, 1364 "__VA_ARGS__ can only appear in the expansion" 1365 " of a C++11 variadic macro"); 1366 else 1367 cpp_error (pfile, CPP_DL_PEDWARN, 1368 "__VA_ARGS__ can only appear in the expansion" 1369 " of a C99 variadic macro"); 1370 } 1371 1372 /* For -Wc++-compat, warn about use of C++ named operators. */ 1373 if (result->flags & NODE_WARN_OPERATOR) 1374 cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, 1375 "identifier \"%s\" is a special operator name in C++", 1376 NODE_NAME (result)); 1377 } 1378 1379 return result; 1380} 1381 1382/* Lex a number to NUMBER starting at BUFFER->CUR - 1. */ 1383static void 1384lex_number (cpp_reader *pfile, cpp_string *number, 1385 struct normalize_state *nst) 1386{ 1387 const uchar *cur; 1388 const uchar *base; 1389 uchar *dest; 1390 1391 base = pfile->buffer->cur - 1; 1392 do 1393 { 1394 cur = pfile->buffer->cur; 1395 1396 /* N.B. ISIDNUM does not include $. */ 1397 while (ISIDNUM (*cur) || *cur == '.' || DIGIT_SEP (*cur) 1398 || VALID_SIGN (*cur, cur[-1])) 1399 { 1400 NORMALIZE_STATE_UPDATE_IDNUM (nst, *cur); 1401 cur++; 1402 } 1403 /* A number can't end with a digit separator. */ 1404 while (cur > pfile->buffer->cur && DIGIT_SEP (cur[-1])) 1405 --cur; 1406 1407 pfile->buffer->cur = cur; 1408 } 1409 while (forms_identifier_p (pfile, false, nst)); 1410 1411 number->len = cur - base; 1412 dest = _cpp_unaligned_alloc (pfile, number->len + 1); 1413 memcpy (dest, base, number->len); 1414 dest[number->len] = '\0'; 1415 number->text = dest; 1416} 1417 1418/* Create a token of type TYPE with a literal spelling. */ 1419static void 1420create_literal (cpp_reader *pfile, cpp_token *token, const uchar *base, 1421 unsigned int len, enum cpp_ttype type) 1422{ 1423 uchar *dest = _cpp_unaligned_alloc (pfile, len + 1); 1424 1425 memcpy (dest, base, len); 1426 dest[len] = '\0'; 1427 token->type = type; 1428 token->val.str.len = len; 1429 token->val.str.text = dest; 1430} 1431 1432/* Subroutine of lex_raw_string: Append LEN chars from BASE to the buffer 1433 sequence from *FIRST_BUFF_P to LAST_BUFF_P. */ 1434 1435static void 1436bufring_append (cpp_reader *pfile, const uchar *base, size_t len, 1437 _cpp_buff **first_buff_p, _cpp_buff **last_buff_p) 1438{ 1439 _cpp_buff *first_buff = *first_buff_p; 1440 _cpp_buff *last_buff = *last_buff_p; 1441 1442 if (first_buff == NULL) 1443 first_buff = last_buff = _cpp_get_buff (pfile, len); 1444 else if (len > BUFF_ROOM (last_buff)) 1445 { 1446 size_t room = BUFF_ROOM (last_buff); 1447 memcpy (BUFF_FRONT (last_buff), base, room); 1448 BUFF_FRONT (last_buff) += room; 1449 base += room; 1450 len -= room; 1451 last_buff = _cpp_append_extend_buff (pfile, last_buff, len); 1452 } 1453 1454 memcpy (BUFF_FRONT (last_buff), base, len); 1455 BUFF_FRONT (last_buff) += len; 1456 1457 *first_buff_p = first_buff; 1458 *last_buff_p = last_buff; 1459} 1460 1461 1462/* Returns true if a macro has been defined. 1463 This might not work if compile with -save-temps, 1464 or preprocess separately from compilation. */ 1465 1466static bool 1467is_macro(cpp_reader *pfile, const uchar *base) 1468{ 1469 const uchar *cur = base; 1470 if (! ISIDST (*cur)) 1471 return false; 1472 unsigned int hash = HT_HASHSTEP (0, *cur); 1473 ++cur; 1474 while (ISIDNUM (*cur)) 1475 { 1476 hash = HT_HASHSTEP (hash, *cur); 1477 ++cur; 1478 } 1479 hash = HT_HASHFINISH (hash, cur - base); 1480 1481 cpp_hashnode *result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 1482 base, cur - base, hash, HT_NO_INSERT)); 1483 1484 return !result ? false : (result->type == NT_MACRO); 1485} 1486 1487 1488/* Lexes a raw string. The stored string contains the spelling, including 1489 double quotes, delimiter string, '(' and ')', any leading 1490 'L', 'u', 'U' or 'u8' and 'R' modifier. It returns the type of the 1491 literal, or CPP_OTHER if it was not properly terminated. 1492 1493 The spelling is NUL-terminated, but it is not guaranteed that this 1494 is the first NUL since embedded NULs are preserved. */ 1495 1496static void 1497lex_raw_string (cpp_reader *pfile, cpp_token *token, const uchar *base, 1498 const uchar *cur) 1499{ 1500 uchar raw_prefix[17]; 1501 uchar temp_buffer[18]; 1502 const uchar *orig_base; 1503 unsigned int raw_prefix_len = 0, raw_suffix_len = 0; 1504 enum raw_str_phase { RAW_STR_PREFIX, RAW_STR, RAW_STR_SUFFIX }; 1505 raw_str_phase phase = RAW_STR_PREFIX; 1506 enum cpp_ttype type; 1507 size_t total_len = 0; 1508 /* Index into temp_buffer during phases other than RAW_STR, 1509 during RAW_STR phase 17 to tell BUF_APPEND that nothing should 1510 be appended to temp_buffer. */ 1511 size_t temp_buffer_len = 0; 1512 _cpp_buff *first_buff = NULL, *last_buff = NULL; 1513 size_t raw_prefix_start; 1514 _cpp_line_note *note = &pfile->buffer->notes[pfile->buffer->cur_note]; 1515 1516 type = (*base == 'L' ? CPP_WSTRING : 1517 *base == 'U' ? CPP_STRING32 : 1518 *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16) 1519 : CPP_STRING); 1520 1521#define BUF_APPEND(STR,LEN) \ 1522 do { \ 1523 bufring_append (pfile, (const uchar *)(STR), (LEN), \ 1524 &first_buff, &last_buff); \ 1525 total_len += (LEN); \ 1526 if (__builtin_expect (temp_buffer_len < 17, 0) \ 1527 && (const uchar *)(STR) != base \ 1528 && (LEN) <= 2) \ 1529 { \ 1530 memcpy (temp_buffer + temp_buffer_len, \ 1531 (const uchar *)(STR), (LEN)); \ 1532 temp_buffer_len += (LEN); \ 1533 } \ 1534 } while (0); 1535 1536 orig_base = base; 1537 ++cur; 1538 raw_prefix_start = cur - base; 1539 for (;;) 1540 { 1541 cppchar_t c; 1542 1543 /* If we previously performed any trigraph or line splicing 1544 transformations, undo them in between the opening and closing 1545 double quote. */ 1546 while (note->pos < cur) 1547 ++note; 1548 for (; note->pos == cur; ++note) 1549 { 1550 switch (note->type) 1551 { 1552 case '\\': 1553 case ' ': 1554 /* Restore backslash followed by newline. */ 1555 BUF_APPEND (base, cur - base); 1556 base = cur; 1557 BUF_APPEND ("\\", 1); 1558 after_backslash: 1559 if (note->type == ' ') 1560 { 1561 /* GNU backslash whitespace newline extension. FIXME 1562 could be any sequence of non-vertical space. When we 1563 can properly restore any such sequence, we should mark 1564 this note as handled so _cpp_process_line_notes 1565 doesn't warn. */ 1566 BUF_APPEND (" ", 1); 1567 } 1568 1569 BUF_APPEND ("\n", 1); 1570 break; 1571 1572 case 0: 1573 /* Already handled. */ 1574 break; 1575 1576 default: 1577 if (_cpp_trigraph_map[note->type]) 1578 { 1579 /* Don't warn about this trigraph in 1580 _cpp_process_line_notes, since trigraphs show up as 1581 trigraphs in raw strings. */ 1582 uchar type = note->type; 1583 note->type = 0; 1584 1585 if (!CPP_OPTION (pfile, trigraphs)) 1586 /* If we didn't convert the trigraph in the first 1587 place, don't do anything now either. */ 1588 break; 1589 1590 BUF_APPEND (base, cur - base); 1591 base = cur; 1592 BUF_APPEND ("??", 2); 1593 1594 /* ??/ followed by newline gets two line notes, one for 1595 the trigraph and one for the backslash/newline. */ 1596 if (type == '/' && note[1].pos == cur) 1597 { 1598 if (note[1].type != '\\' 1599 && note[1].type != ' ') 1600 abort (); 1601 BUF_APPEND ("/", 1); 1602 ++note; 1603 goto after_backslash; 1604 } 1605 else 1606 { 1607 /* Skip the replacement character. */ 1608 base = ++cur; 1609 BUF_APPEND (&type, 1); 1610 c = type; 1611 goto check_c; 1612 } 1613 } 1614 else 1615 abort (); 1616 break; 1617 } 1618 } 1619 c = *cur++; 1620 if (__builtin_expect (temp_buffer_len < 17, 0)) 1621 temp_buffer[temp_buffer_len++] = c; 1622 1623 check_c: 1624 if (phase == RAW_STR_PREFIX) 1625 { 1626 while (raw_prefix_len < temp_buffer_len) 1627 { 1628 raw_prefix[raw_prefix_len] = temp_buffer[raw_prefix_len]; 1629 switch (raw_prefix[raw_prefix_len]) 1630 { 1631 case ' ': case '(': case ')': case '\\': case '\t': 1632 case '\v': case '\f': case '\n': default: 1633 break; 1634 /* Basic source charset except the above chars. */ 1635 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 1636 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 1637 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': 1638 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 1639 case 'y': case 'z': 1640 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 1641 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 1642 case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R': 1643 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 1644 case 'Y': case 'Z': 1645 case '0': case '1': case '2': case '3': case '4': case '5': 1646 case '6': case '7': case '8': case '9': 1647 case '_': case '{': case '}': case '#': case '[': case ']': 1648 case '<': case '>': case '%': case ':': case ';': case '.': 1649 case '?': case '*': case '+': case '-': case '/': case '^': 1650 case '&': case '|': case '~': case '!': case '=': case ',': 1651 case '"': case '\'': 1652 if (raw_prefix_len < 16) 1653 { 1654 raw_prefix_len++; 1655 continue; 1656 } 1657 break; 1658 } 1659 1660 if (raw_prefix[raw_prefix_len] != '(') 1661 { 1662 int col = CPP_BUF_COLUMN (pfile->buffer, cur) + 1; 1663 if (raw_prefix_len == 16) 1664 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 1665 col, "raw string delimiter longer " 1666 "than 16 characters"); 1667 else if (raw_prefix[raw_prefix_len] == '\n') 1668 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 1669 col, "invalid new-line in raw " 1670 "string delimiter"); 1671 else 1672 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 1673 col, "invalid character '%c' in " 1674 "raw string delimiter", 1675 (int) raw_prefix[raw_prefix_len]); 1676 pfile->buffer->cur = orig_base + raw_prefix_start - 1; 1677 create_literal (pfile, token, orig_base, 1678 raw_prefix_start - 1, CPP_OTHER); 1679 if (first_buff) 1680 _cpp_release_buff (pfile, first_buff); 1681 return; 1682 } 1683 raw_prefix[raw_prefix_len] = '"'; 1684 phase = RAW_STR; 1685 /* Nothing should be appended to temp_buffer during 1686 RAW_STR phase. */ 1687 temp_buffer_len = 17; 1688 break; 1689 } 1690 continue; 1691 } 1692 else if (phase == RAW_STR_SUFFIX) 1693 { 1694 while (raw_suffix_len <= raw_prefix_len 1695 && raw_suffix_len < temp_buffer_len 1696 && temp_buffer[raw_suffix_len] == raw_prefix[raw_suffix_len]) 1697 raw_suffix_len++; 1698 if (raw_suffix_len > raw_prefix_len) 1699 break; 1700 if (raw_suffix_len == temp_buffer_len) 1701 continue; 1702 phase = RAW_STR; 1703 /* Nothing should be appended to temp_buffer during 1704 RAW_STR phase. */ 1705 temp_buffer_len = 17; 1706 } 1707 if (c == ')') 1708 { 1709 phase = RAW_STR_SUFFIX; 1710 raw_suffix_len = 0; 1711 temp_buffer_len = 0; 1712 } 1713 else if (c == '\n') 1714 { 1715 if (pfile->state.in_directive 1716 || (pfile->state.parsing_args 1717 && pfile->buffer->next_line >= pfile->buffer->rlimit)) 1718 { 1719 cur--; 1720 type = CPP_OTHER; 1721 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 0, 1722 "unterminated raw string"); 1723 break; 1724 } 1725 1726 BUF_APPEND (base, cur - base); 1727 1728 if (pfile->buffer->cur < pfile->buffer->rlimit) 1729 CPP_INCREMENT_LINE (pfile, 0); 1730 pfile->buffer->need_line = true; 1731 1732 pfile->buffer->cur = cur-1; 1733 _cpp_process_line_notes (pfile, false); 1734 if (!_cpp_get_fresh_line (pfile)) 1735 { 1736 source_location src_loc = token->src_loc; 1737 token->type = CPP_EOF; 1738 /* Tell the compiler the line number of the EOF token. */ 1739 token->src_loc = pfile->line_table->highest_line; 1740 token->flags = BOL; 1741 if (first_buff != NULL) 1742 _cpp_release_buff (pfile, first_buff); 1743 cpp_error_with_line (pfile, CPP_DL_ERROR, src_loc, 0, 1744 "unterminated raw string"); 1745 return; 1746 } 1747 1748 cur = base = pfile->buffer->cur; 1749 note = &pfile->buffer->notes[pfile->buffer->cur_note]; 1750 } 1751 } 1752 1753 if (CPP_OPTION (pfile, user_literals)) 1754 { 1755 /* If a string format macro, say from inttypes.h, is placed touching 1756 a string literal it could be parsed as a C++11 user-defined string 1757 literal thus breaking the program. 1758 Try to identify macros with is_macro. A warning is issued. */ 1759 if (is_macro (pfile, cur)) 1760 { 1761 /* Raise a warning, but do not consume subsequent tokens. */ 1762 if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping) 1763 cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, 1764 token->src_loc, 0, 1765 "invalid suffix on literal; C++11 requires " 1766 "a space between literal and string macro"); 1767 } 1768 /* Grab user defined literal suffix. */ 1769 else if (ISIDST (*cur)) 1770 { 1771 type = cpp_userdef_string_add_type (type); 1772 ++cur; 1773 1774 while (ISIDNUM (*cur)) 1775 ++cur; 1776 } 1777 } 1778 1779 pfile->buffer->cur = cur; 1780 if (first_buff == NULL) 1781 create_literal (pfile, token, base, cur - base, type); 1782 else 1783 { 1784 uchar *dest = _cpp_unaligned_alloc (pfile, total_len + (cur - base) + 1); 1785 1786 token->type = type; 1787 token->val.str.len = total_len + (cur - base); 1788 token->val.str.text = dest; 1789 last_buff = first_buff; 1790 while (last_buff != NULL) 1791 { 1792 memcpy (dest, last_buff->base, 1793 BUFF_FRONT (last_buff) - last_buff->base); 1794 dest += BUFF_FRONT (last_buff) - last_buff->base; 1795 last_buff = last_buff->next; 1796 } 1797 _cpp_release_buff (pfile, first_buff); 1798 memcpy (dest, base, cur - base); 1799 dest[cur - base] = '\0'; 1800 } 1801} 1802 1803/* Lexes a string, character constant, or angle-bracketed header file 1804 name. The stored string contains the spelling, including opening 1805 quote and any leading 'L', 'u', 'U' or 'u8' and optional 1806 'R' modifier. It returns the type of the literal, or CPP_OTHER 1807 if it was not properly terminated, or CPP_LESS for an unterminated 1808 header name which must be relexed as normal tokens. 1809 1810 The spelling is NUL-terminated, but it is not guaranteed that this 1811 is the first NUL since embedded NULs are preserved. */ 1812static void 1813lex_string (cpp_reader *pfile, cpp_token *token, const uchar *base) 1814{ 1815 bool saw_NUL = false; 1816 const uchar *cur; 1817 cppchar_t terminator; 1818 enum cpp_ttype type; 1819 1820 cur = base; 1821 terminator = *cur++; 1822 if (terminator == 'L' || terminator == 'U') 1823 terminator = *cur++; 1824 else if (terminator == 'u') 1825 { 1826 terminator = *cur++; 1827 if (terminator == '8') 1828 terminator = *cur++; 1829 } 1830 if (terminator == 'R') 1831 { 1832 lex_raw_string (pfile, token, base, cur); 1833 return; 1834 } 1835 if (terminator == '"') 1836 type = (*base == 'L' ? CPP_WSTRING : 1837 *base == 'U' ? CPP_STRING32 : 1838 *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16) 1839 : CPP_STRING); 1840 else if (terminator == '\'') 1841 type = (*base == 'L' ? CPP_WCHAR : 1842 *base == 'U' ? CPP_CHAR32 : 1843 *base == 'u' ? CPP_CHAR16 : CPP_CHAR); 1844 else 1845 terminator = '>', type = CPP_HEADER_NAME; 1846 1847 for (;;) 1848 { 1849 cppchar_t c = *cur++; 1850 1851 /* In #include-style directives, terminators are not escapable. */ 1852 if (c == '\\' && !pfile->state.angled_headers && *cur != '\n') 1853 cur++; 1854 else if (c == terminator) 1855 break; 1856 else if (c == '\n') 1857 { 1858 cur--; 1859 /* Unmatched quotes always yield undefined behavior, but 1860 greedy lexing means that what appears to be an unterminated 1861 header name may actually be a legitimate sequence of tokens. */ 1862 if (terminator == '>') 1863 { 1864 token->type = CPP_LESS; 1865 return; 1866 } 1867 type = CPP_OTHER; 1868 break; 1869 } 1870 else if (c == '\0') 1871 saw_NUL = true; 1872 } 1873 1874 if (saw_NUL && !pfile->state.skipping) 1875 cpp_error (pfile, CPP_DL_WARNING, 1876 "null character(s) preserved in literal"); 1877 1878 if (type == CPP_OTHER && CPP_OPTION (pfile, lang) != CLK_ASM) 1879 cpp_error (pfile, CPP_DL_PEDWARN, "missing terminating %c character", 1880 (int) terminator); 1881 1882 if (CPP_OPTION (pfile, user_literals)) 1883 { 1884 /* If a string format macro, say from inttypes.h, is placed touching 1885 a string literal it could be parsed as a C++11 user-defined string 1886 literal thus breaking the program. 1887 Try to identify macros with is_macro. A warning is issued. */ 1888 if (is_macro (pfile, cur)) 1889 { 1890 /* Raise a warning, but do not consume subsequent tokens. */ 1891 if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping) 1892 cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, 1893 token->src_loc, 0, 1894 "invalid suffix on literal; C++11 requires " 1895 "a space between literal and string macro"); 1896 } 1897 /* Grab user defined literal suffix. */ 1898 else if (ISIDST (*cur)) 1899 { 1900 type = cpp_userdef_char_add_type (type); 1901 type = cpp_userdef_string_add_type (type); 1902 ++cur; 1903 1904 while (ISIDNUM (*cur)) 1905 ++cur; 1906 } 1907 } 1908 1909 pfile->buffer->cur = cur; 1910 create_literal (pfile, token, base, cur - base, type); 1911} 1912 1913/* Return the comment table. The client may not make any assumption 1914 about the ordering of the table. */ 1915cpp_comment_table * 1916cpp_get_comments (cpp_reader *pfile) 1917{ 1918 return &pfile->comments; 1919} 1920 1921/* Append a comment to the end of the comment table. */ 1922static void 1923store_comment (cpp_reader *pfile, cpp_token *token) 1924{ 1925 int len; 1926 1927 if (pfile->comments.allocated == 0) 1928 { 1929 pfile->comments.allocated = 256; 1930 pfile->comments.entries = (cpp_comment *) xmalloc 1931 (pfile->comments.allocated * sizeof (cpp_comment)); 1932 } 1933 1934 if (pfile->comments.count == pfile->comments.allocated) 1935 { 1936 pfile->comments.allocated *= 2; 1937 pfile->comments.entries = (cpp_comment *) xrealloc 1938 (pfile->comments.entries, 1939 pfile->comments.allocated * sizeof (cpp_comment)); 1940 } 1941 1942 len = token->val.str.len; 1943 1944 /* Copy comment. Note, token may not be NULL terminated. */ 1945 pfile->comments.entries[pfile->comments.count].comment = 1946 (char *) xmalloc (sizeof (char) * (len + 1)); 1947 memcpy (pfile->comments.entries[pfile->comments.count].comment, 1948 token->val.str.text, len); 1949 pfile->comments.entries[pfile->comments.count].comment[len] = '\0'; 1950 1951 /* Set source location. */ 1952 pfile->comments.entries[pfile->comments.count].sloc = token->src_loc; 1953 1954 /* Increment the count of entries in the comment table. */ 1955 pfile->comments.count++; 1956} 1957 1958/* The stored comment includes the comment start and any terminator. */ 1959static void 1960save_comment (cpp_reader *pfile, cpp_token *token, const unsigned char *from, 1961 cppchar_t type) 1962{ 1963 unsigned char *buffer; 1964 unsigned int len, clen, i; 1965 1966 len = pfile->buffer->cur - from + 1; /* + 1 for the initial '/'. */ 1967 1968 /* C++ comments probably (not definitely) have moved past a new 1969 line, which we don't want to save in the comment. */ 1970 if (is_vspace (pfile->buffer->cur[-1])) 1971 len--; 1972 1973 /* If we are currently in a directive or in argument parsing, then 1974 we need to store all C++ comments as C comments internally, and 1975 so we need to allocate a little extra space in that case. 1976 1977 Note that the only time we encounter a directive here is 1978 when we are saving comments in a "#define". */ 1979 clen = ((pfile->state.in_directive || pfile->state.parsing_args) 1980 && type == '/') ? len + 2 : len; 1981 1982 buffer = _cpp_unaligned_alloc (pfile, clen); 1983 1984 token->type = CPP_COMMENT; 1985 token->val.str.len = clen; 1986 token->val.str.text = buffer; 1987 1988 buffer[0] = '/'; 1989 memcpy (buffer + 1, from, len - 1); 1990 1991 /* Finish conversion to a C comment, if necessary. */ 1992 if ((pfile->state.in_directive || pfile->state.parsing_args) && type == '/') 1993 { 1994 buffer[1] = '*'; 1995 buffer[clen - 2] = '*'; 1996 buffer[clen - 1] = '/'; 1997 /* As there can be in a C++ comments illegal sequences for C comments 1998 we need to filter them out. */ 1999 for (i = 2; i < (clen - 2); i++) 2000 if (buffer[i] == '/' && (buffer[i - 1] == '*' || buffer[i + 1] == '*')) 2001 buffer[i] = '|'; 2002 } 2003 2004 /* Finally store this comment for use by clients of libcpp. */ 2005 store_comment (pfile, token); 2006} 2007 2008/* Allocate COUNT tokens for RUN. */ 2009void 2010_cpp_init_tokenrun (tokenrun *run, unsigned int count) 2011{ 2012 run->base = XNEWVEC (cpp_token, count); 2013 run->limit = run->base + count; 2014 run->next = NULL; 2015} 2016 2017/* Returns the next tokenrun, or creates one if there is none. */ 2018static tokenrun * 2019next_tokenrun (tokenrun *run) 2020{ 2021 if (run->next == NULL) 2022 { 2023 run->next = XNEW (tokenrun); 2024 run->next->prev = run; 2025 _cpp_init_tokenrun (run->next, 250); 2026 } 2027 2028 return run->next; 2029} 2030 2031/* Return the number of not yet processed token in a given 2032 context. */ 2033int 2034_cpp_remaining_tokens_num_in_context (cpp_context *context) 2035{ 2036 if (context->tokens_kind == TOKENS_KIND_DIRECT) 2037 return (LAST (context).token - FIRST (context).token); 2038 else if (context->tokens_kind == TOKENS_KIND_INDIRECT 2039 || context->tokens_kind == TOKENS_KIND_EXTENDED) 2040 return (LAST (context).ptoken - FIRST (context).ptoken); 2041 else 2042 abort (); 2043} 2044 2045/* Returns the token present at index INDEX in a given context. If 2046 INDEX is zero, the next token to be processed is returned. */ 2047static const cpp_token* 2048_cpp_token_from_context_at (cpp_context *context, int index) 2049{ 2050 if (context->tokens_kind == TOKENS_KIND_DIRECT) 2051 return &(FIRST (context).token[index]); 2052 else if (context->tokens_kind == TOKENS_KIND_INDIRECT 2053 || context->tokens_kind == TOKENS_KIND_EXTENDED) 2054 return FIRST (context).ptoken[index]; 2055 else 2056 abort (); 2057} 2058 2059/* Look ahead in the input stream. */ 2060const cpp_token * 2061cpp_peek_token (cpp_reader *pfile, int index) 2062{ 2063 cpp_context *context = pfile->context; 2064 const cpp_token *peektok; 2065 int count; 2066 2067 /* First, scan through any pending cpp_context objects. */ 2068 while (context->prev) 2069 { 2070 ptrdiff_t sz = _cpp_remaining_tokens_num_in_context (context); 2071 2072 if (index < (int) sz) 2073 return _cpp_token_from_context_at (context, index); 2074 index -= (int) sz; 2075 context = context->prev; 2076 } 2077 2078 /* We will have to read some new tokens after all (and do so 2079 without invalidating preceding tokens). */ 2080 count = index; 2081 pfile->keep_tokens++; 2082 2083 /* For peeked tokens temporarily disable line_change reporting, 2084 until the tokens are parsed for real. */ 2085 void (*line_change) (cpp_reader *, const cpp_token *, int) 2086 = pfile->cb.line_change; 2087 pfile->cb.line_change = NULL; 2088 2089 do 2090 { 2091 peektok = _cpp_lex_token (pfile); 2092 if (peektok->type == CPP_EOF) 2093 { 2094 index--; 2095 break; 2096 } 2097 } 2098 while (index--); 2099 2100 _cpp_backup_tokens_direct (pfile, count - index); 2101 pfile->keep_tokens--; 2102 pfile->cb.line_change = line_change; 2103 2104 return peektok; 2105} 2106 2107/* Allocate a single token that is invalidated at the same time as the 2108 rest of the tokens on the line. Has its line and col set to the 2109 same as the last lexed token, so that diagnostics appear in the 2110 right place. */ 2111cpp_token * 2112_cpp_temp_token (cpp_reader *pfile) 2113{ 2114 cpp_token *old, *result; 2115 ptrdiff_t sz = pfile->cur_run->limit - pfile->cur_token; 2116 ptrdiff_t la = (ptrdiff_t) pfile->lookaheads; 2117 2118 old = pfile->cur_token - 1; 2119 /* Any pre-existing lookaheads must not be clobbered. */ 2120 if (la) 2121 { 2122 if (sz <= la) 2123 { 2124 tokenrun *next = next_tokenrun (pfile->cur_run); 2125 2126 if (sz < la) 2127 memmove (next->base + 1, next->base, 2128 (la - sz) * sizeof (cpp_token)); 2129 2130 next->base[0] = pfile->cur_run->limit[-1]; 2131 } 2132 2133 if (sz > 1) 2134 memmove (pfile->cur_token + 1, pfile->cur_token, 2135 MIN (la, sz - 1) * sizeof (cpp_token)); 2136 } 2137 2138 if (!sz && pfile->cur_token == pfile->cur_run->limit) 2139 { 2140 pfile->cur_run = next_tokenrun (pfile->cur_run); 2141 pfile->cur_token = pfile->cur_run->base; 2142 } 2143 2144 result = pfile->cur_token++; 2145 result->src_loc = old->src_loc; 2146 return result; 2147} 2148 2149/* Lex a token into RESULT (external interface). Takes care of issues 2150 like directive handling, token lookahead, multiple include 2151 optimization and skipping. */ 2152const cpp_token * 2153_cpp_lex_token (cpp_reader *pfile) 2154{ 2155 cpp_token *result; 2156 2157 for (;;) 2158 { 2159 if (pfile->cur_token == pfile->cur_run->limit) 2160 { 2161 pfile->cur_run = next_tokenrun (pfile->cur_run); 2162 pfile->cur_token = pfile->cur_run->base; 2163 } 2164 /* We assume that the current token is somewhere in the current 2165 run. */ 2166 if (pfile->cur_token < pfile->cur_run->base 2167 || pfile->cur_token >= pfile->cur_run->limit) 2168 abort (); 2169 2170 if (pfile->lookaheads) 2171 { 2172 pfile->lookaheads--; 2173 result = pfile->cur_token++; 2174 } 2175 else 2176 result = _cpp_lex_direct (pfile); 2177 2178 if (result->flags & BOL) 2179 { 2180 /* Is this a directive. If _cpp_handle_directive returns 2181 false, it is an assembler #. */ 2182 if (result->type == CPP_HASH 2183 /* 6.10.3 p 11: Directives in a list of macro arguments 2184 gives undefined behavior. This implementation 2185 handles the directive as normal. */ 2186 && pfile->state.parsing_args != 1) 2187 { 2188 if (_cpp_handle_directive (pfile, result->flags & PREV_WHITE)) 2189 { 2190 if (pfile->directive_result.type == CPP_PADDING) 2191 continue; 2192 result = &pfile->directive_result; 2193 } 2194 } 2195 else if (pfile->state.in_deferred_pragma) 2196 result = &pfile->directive_result; 2197 2198 if (pfile->cb.line_change && !pfile->state.skipping) 2199 pfile->cb.line_change (pfile, result, pfile->state.parsing_args); 2200 } 2201 2202 /* We don't skip tokens in directives. */ 2203 if (pfile->state.in_directive || pfile->state.in_deferred_pragma) 2204 break; 2205 2206 /* Outside a directive, invalidate controlling macros. At file 2207 EOF, _cpp_lex_direct takes care of popping the buffer, so we never 2208 get here and MI optimization works. */ 2209 pfile->mi_valid = false; 2210 2211 if (!pfile->state.skipping || result->type == CPP_EOF) 2212 break; 2213 } 2214 2215 return result; 2216} 2217 2218/* Returns true if a fresh line has been loaded. */ 2219bool 2220_cpp_get_fresh_line (cpp_reader *pfile) 2221{ 2222 int return_at_eof; 2223 2224 /* We can't get a new line until we leave the current directive. */ 2225 if (pfile->state.in_directive) 2226 return false; 2227 2228 for (;;) 2229 { 2230 cpp_buffer *buffer = pfile->buffer; 2231 2232 if (!buffer->need_line) 2233 return true; 2234 2235 if (buffer->next_line < buffer->rlimit) 2236 { 2237 _cpp_clean_line (pfile); 2238 return true; 2239 } 2240 2241 /* First, get out of parsing arguments state. */ 2242 if (pfile->state.parsing_args) 2243 return false; 2244 2245 /* End of buffer. Non-empty files should end in a newline. */ 2246 if (buffer->buf != buffer->rlimit 2247 && buffer->next_line > buffer->rlimit 2248 && !buffer->from_stage3) 2249 { 2250 /* Clip to buffer size. */ 2251 buffer->next_line = buffer->rlimit; 2252 } 2253 2254 return_at_eof = buffer->return_at_eof; 2255 _cpp_pop_buffer (pfile); 2256 if (pfile->buffer == NULL || return_at_eof) 2257 return false; 2258 } 2259} 2260 2261#define IF_NEXT_IS(CHAR, THEN_TYPE, ELSE_TYPE) \ 2262 do \ 2263 { \ 2264 result->type = ELSE_TYPE; \ 2265 if (*buffer->cur == CHAR) \ 2266 buffer->cur++, result->type = THEN_TYPE; \ 2267 } \ 2268 while (0) 2269 2270/* Lex a token into pfile->cur_token, which is also incremented, to 2271 get diagnostics pointing to the correct location. 2272 2273 Does not handle issues such as token lookahead, multiple-include 2274 optimization, directives, skipping etc. This function is only 2275 suitable for use by _cpp_lex_token, and in special cases like 2276 lex_expansion_token which doesn't care for any of these issues. 2277 2278 When meeting a newline, returns CPP_EOF if parsing a directive, 2279 otherwise returns to the start of the token buffer if permissible. 2280 Returns the location of the lexed token. */ 2281cpp_token * 2282_cpp_lex_direct (cpp_reader *pfile) 2283{ 2284 cppchar_t c; 2285 cpp_buffer *buffer; 2286 const unsigned char *comment_start; 2287 cpp_token *result = pfile->cur_token++; 2288 2289 fresh_line: 2290 result->flags = 0; 2291 buffer = pfile->buffer; 2292 if (buffer->need_line) 2293 { 2294 if (pfile->state.in_deferred_pragma) 2295 { 2296 result->type = CPP_PRAGMA_EOL; 2297 pfile->state.in_deferred_pragma = false; 2298 if (!pfile->state.pragma_allow_expansion) 2299 pfile->state.prevent_expansion--; 2300 return result; 2301 } 2302 if (!_cpp_get_fresh_line (pfile)) 2303 { 2304 result->type = CPP_EOF; 2305 if (!pfile->state.in_directive) 2306 { 2307 /* Tell the compiler the line number of the EOF token. */ 2308 result->src_loc = pfile->line_table->highest_line; 2309 result->flags = BOL; 2310 } 2311 return result; 2312 } 2313 if (!pfile->keep_tokens) 2314 { 2315 pfile->cur_run = &pfile->base_run; 2316 result = pfile->base_run.base; 2317 pfile->cur_token = result + 1; 2318 } 2319 result->flags = BOL; 2320 if (pfile->state.parsing_args == 2) 2321 result->flags |= PREV_WHITE; 2322 } 2323 buffer = pfile->buffer; 2324 update_tokens_line: 2325 result->src_loc = pfile->line_table->highest_line; 2326 2327 skipped_white: 2328 if (buffer->cur >= buffer->notes[buffer->cur_note].pos 2329 && !pfile->overlaid_buffer) 2330 { 2331 _cpp_process_line_notes (pfile, false); 2332 result->src_loc = pfile->line_table->highest_line; 2333 } 2334 c = *buffer->cur++; 2335 2336 if (pfile->forced_token_location_p) 2337 result->src_loc = *pfile->forced_token_location_p; 2338 else 2339 result->src_loc = linemap_position_for_column (pfile->line_table, 2340 CPP_BUF_COLUMN (buffer, buffer->cur)); 2341 2342 switch (c) 2343 { 2344 case ' ': case '\t': case '\f': case '\v': case '\0': 2345 result->flags |= PREV_WHITE; 2346 skip_whitespace (pfile, c); 2347 goto skipped_white; 2348 2349 case '\n': 2350 if (buffer->cur < buffer->rlimit) 2351 CPP_INCREMENT_LINE (pfile, 0); 2352 buffer->need_line = true; 2353 goto fresh_line; 2354 2355 case '0': case '1': case '2': case '3': case '4': 2356 case '5': case '6': case '7': case '8': case '9': 2357 { 2358 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 2359 result->type = CPP_NUMBER; 2360 lex_number (pfile, &result->val.str, &nst); 2361 warn_about_normalization (pfile, result, &nst); 2362 break; 2363 } 2364 2365 case 'L': 2366 case 'u': 2367 case 'U': 2368 case 'R': 2369 /* 'L', 'u', 'U', 'u8' or 'R' may introduce wide characters, 2370 wide strings or raw strings. */ 2371 if (c == 'L' || CPP_OPTION (pfile, rliterals) 2372 || (c != 'R' && CPP_OPTION (pfile, uliterals))) 2373 { 2374 if ((*buffer->cur == '\'' && c != 'R') 2375 || *buffer->cur == '"' 2376 || (*buffer->cur == 'R' 2377 && c != 'R' 2378 && buffer->cur[1] == '"' 2379 && CPP_OPTION (pfile, rliterals)) 2380 || (*buffer->cur == '8' 2381 && c == 'u' 2382 && (buffer->cur[1] == '"' 2383 || (buffer->cur[1] == 'R' && buffer->cur[2] == '"' 2384 && CPP_OPTION (pfile, rliterals))))) 2385 { 2386 lex_string (pfile, result, buffer->cur - 1); 2387 break; 2388 } 2389 } 2390 /* Fall through. */ 2391 2392 case '_': 2393 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 2394 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 2395 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': 2396 case 's': case 't': case 'v': case 'w': case 'x': 2397 case 'y': case 'z': 2398 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 2399 case 'G': case 'H': case 'I': case 'J': case 'K': 2400 case 'M': case 'N': case 'O': case 'P': case 'Q': 2401 case 'S': case 'T': case 'V': case 'W': case 'X': 2402 case 'Y': case 'Z': 2403 result->type = CPP_NAME; 2404 { 2405 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 2406 result->val.node.node = lex_identifier (pfile, buffer->cur - 1, false, 2407 &nst, 2408 &result->val.node.spelling); 2409 warn_about_normalization (pfile, result, &nst); 2410 } 2411 2412 /* Convert named operators to their proper types. */ 2413 if (result->val.node.node->flags & NODE_OPERATOR) 2414 { 2415 result->flags |= NAMED_OP; 2416 result->type = (enum cpp_ttype) result->val.node.node->directive_index; 2417 } 2418 break; 2419 2420 case '\'': 2421 case '"': 2422 lex_string (pfile, result, buffer->cur - 1); 2423 break; 2424 2425 case '/': 2426 /* A potential block or line comment. */ 2427 comment_start = buffer->cur; 2428 c = *buffer->cur; 2429 2430 if (c == '*') 2431 { 2432 if (_cpp_skip_block_comment (pfile)) 2433 cpp_error (pfile, CPP_DL_ERROR, "unterminated comment"); 2434 } 2435 else if (c == '/' && ! CPP_OPTION (pfile, traditional)) 2436 { 2437 /* Don't warn for system headers. */ 2438 if (cpp_in_system_header (pfile)) 2439 ; 2440 /* Warn about comments if pedantically GNUC89, and not 2441 in system headers. */ 2442 else if (CPP_OPTION (pfile, lang) == CLK_GNUC89 2443 && CPP_PEDANTIC (pfile) 2444 && ! buffer->warned_cplusplus_comments) 2445 { 2446 cpp_error (pfile, CPP_DL_PEDWARN, 2447 "C++ style comments are not allowed in ISO C90"); 2448 cpp_error (pfile, CPP_DL_PEDWARN, 2449 "(this will be reported only once per input file)"); 2450 buffer->warned_cplusplus_comments = 1; 2451 } 2452 /* Or if specifically desired via -Wc90-c99-compat. */ 2453 else if (CPP_OPTION (pfile, cpp_warn_c90_c99_compat) > 0 2454 && ! CPP_OPTION (pfile, cplusplus) 2455 && ! buffer->warned_cplusplus_comments) 2456 { 2457 cpp_error (pfile, CPP_DL_WARNING, 2458 "C++ style comments are incompatible with C90"); 2459 cpp_error (pfile, CPP_DL_WARNING, 2460 "(this will be reported only once per input file)"); 2461 buffer->warned_cplusplus_comments = 1; 2462 } 2463 /* In C89/C94, C++ style comments are forbidden. */ 2464 else if ((CPP_OPTION (pfile, lang) == CLK_STDC89 2465 || CPP_OPTION (pfile, lang) == CLK_STDC94)) 2466 { 2467 /* But don't be confused about valid code such as 2468 - // immediately followed by *, 2469 - // in a preprocessing directive, 2470 - // in an #if 0 block. */ 2471 if (buffer->cur[1] == '*' 2472 || pfile->state.in_directive 2473 || pfile->state.skipping) 2474 { 2475 result->type = CPP_DIV; 2476 break; 2477 } 2478 else if (! buffer->warned_cplusplus_comments) 2479 { 2480 cpp_error (pfile, CPP_DL_ERROR, 2481 "C++ style comments are not allowed in ISO C90"); 2482 cpp_error (pfile, CPP_DL_ERROR, 2483 "(this will be reported only once per input " 2484 "file)"); 2485 buffer->warned_cplusplus_comments = 1; 2486 } 2487 } 2488 if (skip_line_comment (pfile) && CPP_OPTION (pfile, warn_comments)) 2489 cpp_warning (pfile, CPP_W_COMMENTS, "multi-line comment"); 2490 } 2491 else if (c == '=') 2492 { 2493 buffer->cur++; 2494 result->type = CPP_DIV_EQ; 2495 break; 2496 } 2497 else 2498 { 2499 result->type = CPP_DIV; 2500 break; 2501 } 2502 2503 if (!pfile->state.save_comments) 2504 { 2505 result->flags |= PREV_WHITE; 2506 goto update_tokens_line; 2507 } 2508 2509 /* Save the comment as a token in its own right. */ 2510 save_comment (pfile, result, comment_start, c); 2511 break; 2512 2513 case '<': 2514 if (pfile->state.angled_headers) 2515 { 2516 lex_string (pfile, result, buffer->cur - 1); 2517 if (result->type != CPP_LESS) 2518 break; 2519 } 2520 2521 result->type = CPP_LESS; 2522 if (*buffer->cur == '=') 2523 buffer->cur++, result->type = CPP_LESS_EQ; 2524 else if (*buffer->cur == '<') 2525 { 2526 buffer->cur++; 2527 IF_NEXT_IS ('=', CPP_LSHIFT_EQ, CPP_LSHIFT); 2528 } 2529 else if (CPP_OPTION (pfile, digraphs)) 2530 { 2531 if (*buffer->cur == ':') 2532 { 2533 /* C++11 [2.5/3 lex.pptoken], "Otherwise, if the next 2534 three characters are <:: and the subsequent character 2535 is neither : nor >, the < is treated as a preprocessor 2536 token by itself". */ 2537 if (CPP_OPTION (pfile, cplusplus) 2538 && CPP_OPTION (pfile, lang) != CLK_CXX98 2539 && CPP_OPTION (pfile, lang) != CLK_GNUCXX 2540 && buffer->cur[1] == ':' 2541 && buffer->cur[2] != ':' && buffer->cur[2] != '>') 2542 break; 2543 2544 buffer->cur++; 2545 result->flags |= DIGRAPH; 2546 result->type = CPP_OPEN_SQUARE; 2547 } 2548 else if (*buffer->cur == '%') 2549 { 2550 buffer->cur++; 2551 result->flags |= DIGRAPH; 2552 result->type = CPP_OPEN_BRACE; 2553 } 2554 } 2555 break; 2556 2557 case '>': 2558 result->type = CPP_GREATER; 2559 if (*buffer->cur == '=') 2560 buffer->cur++, result->type = CPP_GREATER_EQ; 2561 else if (*buffer->cur == '>') 2562 { 2563 buffer->cur++; 2564 IF_NEXT_IS ('=', CPP_RSHIFT_EQ, CPP_RSHIFT); 2565 } 2566 break; 2567 2568 case '%': 2569 result->type = CPP_MOD; 2570 if (*buffer->cur == '=') 2571 buffer->cur++, result->type = CPP_MOD_EQ; 2572 else if (CPP_OPTION (pfile, digraphs)) 2573 { 2574 if (*buffer->cur == ':') 2575 { 2576 buffer->cur++; 2577 result->flags |= DIGRAPH; 2578 result->type = CPP_HASH; 2579 if (*buffer->cur == '%' && buffer->cur[1] == ':') 2580 buffer->cur += 2, result->type = CPP_PASTE, result->val.token_no = 0; 2581 } 2582 else if (*buffer->cur == '>') 2583 { 2584 buffer->cur++; 2585 result->flags |= DIGRAPH; 2586 result->type = CPP_CLOSE_BRACE; 2587 } 2588 } 2589 break; 2590 2591 case '.': 2592 result->type = CPP_DOT; 2593 if (ISDIGIT (*buffer->cur)) 2594 { 2595 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 2596 result->type = CPP_NUMBER; 2597 lex_number (pfile, &result->val.str, &nst); 2598 warn_about_normalization (pfile, result, &nst); 2599 } 2600 else if (*buffer->cur == '.' && buffer->cur[1] == '.') 2601 buffer->cur += 2, result->type = CPP_ELLIPSIS; 2602 else if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) 2603 buffer->cur++, result->type = CPP_DOT_STAR; 2604 break; 2605 2606 case '+': 2607 result->type = CPP_PLUS; 2608 if (*buffer->cur == '+') 2609 buffer->cur++, result->type = CPP_PLUS_PLUS; 2610 else if (*buffer->cur == '=') 2611 buffer->cur++, result->type = CPP_PLUS_EQ; 2612 break; 2613 2614 case '-': 2615 result->type = CPP_MINUS; 2616 if (*buffer->cur == '>') 2617 { 2618 buffer->cur++; 2619 result->type = CPP_DEREF; 2620 if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) 2621 buffer->cur++, result->type = CPP_DEREF_STAR; 2622 } 2623 else if (*buffer->cur == '-') 2624 buffer->cur++, result->type = CPP_MINUS_MINUS; 2625 else if (*buffer->cur == '=') 2626 buffer->cur++, result->type = CPP_MINUS_EQ; 2627 break; 2628 2629 case '&': 2630 result->type = CPP_AND; 2631 if (*buffer->cur == '&') 2632 buffer->cur++, result->type = CPP_AND_AND; 2633 else if (*buffer->cur == '=') 2634 buffer->cur++, result->type = CPP_AND_EQ; 2635 break; 2636 2637 case '|': 2638 result->type = CPP_OR; 2639 if (*buffer->cur == '|') 2640 buffer->cur++, result->type = CPP_OR_OR; 2641 else if (*buffer->cur == '=') 2642 buffer->cur++, result->type = CPP_OR_EQ; 2643 break; 2644 2645 case ':': 2646 result->type = CPP_COLON; 2647 if (*buffer->cur == ':' && CPP_OPTION (pfile, cplusplus)) 2648 buffer->cur++, result->type = CPP_SCOPE; 2649 else if (*buffer->cur == '>' && CPP_OPTION (pfile, digraphs)) 2650 { 2651 buffer->cur++; 2652 result->flags |= DIGRAPH; 2653 result->type = CPP_CLOSE_SQUARE; 2654 } 2655 break; 2656 2657 case '*': IF_NEXT_IS ('=', CPP_MULT_EQ, CPP_MULT); break; 2658 case '=': IF_NEXT_IS ('=', CPP_EQ_EQ, CPP_EQ); break; 2659 case '!': IF_NEXT_IS ('=', CPP_NOT_EQ, CPP_NOT); break; 2660 case '^': IF_NEXT_IS ('=', CPP_XOR_EQ, CPP_XOR); break; 2661 case '#': IF_NEXT_IS ('#', CPP_PASTE, CPP_HASH); result->val.token_no = 0; break; 2662 2663 case '?': result->type = CPP_QUERY; break; 2664 case '~': result->type = CPP_COMPL; break; 2665 case ',': result->type = CPP_COMMA; break; 2666 case '(': result->type = CPP_OPEN_PAREN; break; 2667 case ')': result->type = CPP_CLOSE_PAREN; break; 2668 case '[': result->type = CPP_OPEN_SQUARE; break; 2669 case ']': result->type = CPP_CLOSE_SQUARE; break; 2670 case '{': result->type = CPP_OPEN_BRACE; break; 2671 case '}': result->type = CPP_CLOSE_BRACE; break; 2672 case ';': result->type = CPP_SEMICOLON; break; 2673 2674 /* @ is a punctuator in Objective-C. */ 2675 case '@': result->type = CPP_ATSIGN; break; 2676 2677 case '$': 2678 case '\\': 2679 { 2680 const uchar *base = --buffer->cur; 2681 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 2682 2683 if (forms_identifier_p (pfile, true, &nst)) 2684 { 2685 result->type = CPP_NAME; 2686 result->val.node.node = lex_identifier (pfile, base, true, &nst, 2687 &result->val.node.spelling); 2688 warn_about_normalization (pfile, result, &nst); 2689 break; 2690 } 2691 buffer->cur++; 2692 } 2693 2694 default: 2695 create_literal (pfile, result, buffer->cur - 1, 1, CPP_OTHER); 2696 break; 2697 } 2698 2699 return result; 2700} 2701 2702/* An upper bound on the number of bytes needed to spell TOKEN. 2703 Does not include preceding whitespace. */ 2704unsigned int 2705cpp_token_len (const cpp_token *token) 2706{ 2707 unsigned int len; 2708 2709 switch (TOKEN_SPELL (token)) 2710 { 2711 default: len = 6; break; 2712 case SPELL_LITERAL: len = token->val.str.len; break; 2713 case SPELL_IDENT: len = NODE_LEN (token->val.node.node) * 10; break; 2714 } 2715 2716 return len; 2717} 2718 2719/* Parse UTF-8 out of NAMEP and place a \U escape in BUFFER. 2720 Return the number of bytes read out of NAME. (There are always 2721 10 bytes written to BUFFER.) */ 2722 2723static size_t 2724utf8_to_ucn (unsigned char *buffer, const unsigned char *name) 2725{ 2726 int j; 2727 int ucn_len = 0; 2728 int ucn_len_c; 2729 unsigned t; 2730 unsigned long utf32; 2731 2732 /* Compute the length of the UTF-8 sequence. */ 2733 for (t = *name; t & 0x80; t <<= 1) 2734 ucn_len++; 2735 2736 utf32 = *name & (0x7F >> ucn_len); 2737 for (ucn_len_c = 1; ucn_len_c < ucn_len; ucn_len_c++) 2738 { 2739 utf32 = (utf32 << 6) | (*++name & 0x3F); 2740 2741 /* Ill-formed UTF-8. */ 2742 if ((*name & ~0x3F) != 0x80) 2743 abort (); 2744 } 2745 2746 *buffer++ = '\\'; 2747 *buffer++ = 'U'; 2748 for (j = 7; j >= 0; j--) 2749 *buffer++ = "0123456789abcdef"[(utf32 >> (4 * j)) & 0xF]; 2750 return ucn_len; 2751} 2752 2753/* Given a token TYPE corresponding to a digraph, return a pointer to 2754 the spelling of the digraph. */ 2755static const unsigned char * 2756cpp_digraph2name (enum cpp_ttype type) 2757{ 2758 return digraph_spellings[(int) type - (int) CPP_FIRST_DIGRAPH]; 2759} 2760 2761/* Write the spelling of an identifier IDENT, using UCNs, to BUFFER. 2762 The buffer must already contain the enough space to hold the 2763 token's spelling. Returns a pointer to the character after the 2764 last character written. */ 2765unsigned char * 2766_cpp_spell_ident_ucns (unsigned char *buffer, cpp_hashnode *ident) 2767{ 2768 size_t i; 2769 const unsigned char *name = NODE_NAME (ident); 2770 2771 for (i = 0; i < NODE_LEN (ident); i++) 2772 if (name[i] & ~0x7F) 2773 { 2774 i += utf8_to_ucn (buffer, name + i) - 1; 2775 buffer += 10; 2776 } 2777 else 2778 *buffer++ = name[i]; 2779 2780 return buffer; 2781} 2782 2783/* Write the spelling of a token TOKEN to BUFFER. The buffer must 2784 already contain the enough space to hold the token's spelling. 2785 Returns a pointer to the character after the last character written. 2786 FORSTRING is true if this is to be the spelling after translation 2787 phase 1 (with the original spelling of extended identifiers), false 2788 if extended identifiers should always be written using UCNs (there is 2789 no option for always writing them in the internal UTF-8 form). 2790 FIXME: Would be nice if we didn't need the PFILE argument. */ 2791unsigned char * 2792cpp_spell_token (cpp_reader *pfile, const cpp_token *token, 2793 unsigned char *buffer, bool forstring) 2794{ 2795 switch (TOKEN_SPELL (token)) 2796 { 2797 case SPELL_OPERATOR: 2798 { 2799 const unsigned char *spelling; 2800 unsigned char c; 2801 2802 if (token->flags & DIGRAPH) 2803 spelling = cpp_digraph2name (token->type); 2804 else if (token->flags & NAMED_OP) 2805 goto spell_ident; 2806 else 2807 spelling = TOKEN_NAME (token); 2808 2809 while ((c = *spelling++) != '\0') 2810 *buffer++ = c; 2811 } 2812 break; 2813 2814 spell_ident: 2815 case SPELL_IDENT: 2816 if (forstring) 2817 { 2818 memcpy (buffer, NODE_NAME (token->val.node.spelling), 2819 NODE_LEN (token->val.node.spelling)); 2820 buffer += NODE_LEN (token->val.node.spelling); 2821 } 2822 else 2823 buffer = _cpp_spell_ident_ucns (buffer, token->val.node.node); 2824 break; 2825 2826 case SPELL_LITERAL: 2827 memcpy (buffer, token->val.str.text, token->val.str.len); 2828 buffer += token->val.str.len; 2829 break; 2830 2831 case SPELL_NONE: 2832 cpp_error (pfile, CPP_DL_ICE, 2833 "unspellable token %s", TOKEN_NAME (token)); 2834 break; 2835 } 2836 2837 return buffer; 2838} 2839 2840/* Returns TOKEN spelt as a null-terminated string. The string is 2841 freed when the reader is destroyed. Useful for diagnostics. */ 2842unsigned char * 2843cpp_token_as_text (cpp_reader *pfile, const cpp_token *token) 2844{ 2845 unsigned int len = cpp_token_len (token) + 1; 2846 unsigned char *start = _cpp_unaligned_alloc (pfile, len), *end; 2847 2848 end = cpp_spell_token (pfile, token, start, false); 2849 end[0] = '\0'; 2850 2851 return start; 2852} 2853 2854/* Returns a pointer to a string which spells the token defined by 2855 TYPE and FLAGS. Used by C front ends, which really should move to 2856 using cpp_token_as_text. */ 2857const char * 2858cpp_type2name (enum cpp_ttype type, unsigned char flags) 2859{ 2860 if (flags & DIGRAPH) 2861 return (const char *) cpp_digraph2name (type); 2862 else if (flags & NAMED_OP) 2863 return cpp_named_operator2name (type); 2864 2865 return (const char *) token_spellings[type].name; 2866} 2867 2868/* Writes the spelling of token to FP, without any preceding space. 2869 Separated from cpp_spell_token for efficiency - to avoid stdio 2870 double-buffering. */ 2871void 2872cpp_output_token (const cpp_token *token, FILE *fp) 2873{ 2874 switch (TOKEN_SPELL (token)) 2875 { 2876 case SPELL_OPERATOR: 2877 { 2878 const unsigned char *spelling; 2879 int c; 2880 2881 if (token->flags & DIGRAPH) 2882 spelling = cpp_digraph2name (token->type); 2883 else if (token->flags & NAMED_OP) 2884 goto spell_ident; 2885 else 2886 spelling = TOKEN_NAME (token); 2887 2888 c = *spelling; 2889 do 2890 putc (c, fp); 2891 while ((c = *++spelling) != '\0'); 2892 } 2893 break; 2894 2895 spell_ident: 2896 case SPELL_IDENT: 2897 { 2898 size_t i; 2899 const unsigned char * name = NODE_NAME (token->val.node.node); 2900 2901 for (i = 0; i < NODE_LEN (token->val.node.node); i++) 2902 if (name[i] & ~0x7F) 2903 { 2904 unsigned char buffer[10]; 2905 i += utf8_to_ucn (buffer, name + i) - 1; 2906 fwrite (buffer, 1, 10, fp); 2907 } 2908 else 2909 fputc (NODE_NAME (token->val.node.node)[i], fp); 2910 } 2911 break; 2912 2913 case SPELL_LITERAL: 2914 fwrite (token->val.str.text, 1, token->val.str.len, fp); 2915 break; 2916 2917 case SPELL_NONE: 2918 /* An error, most probably. */ 2919 break; 2920 } 2921} 2922 2923/* Compare two tokens. */ 2924int 2925_cpp_equiv_tokens (const cpp_token *a, const cpp_token *b) 2926{ 2927 if (a->type == b->type && a->flags == b->flags) 2928 switch (TOKEN_SPELL (a)) 2929 { 2930 default: /* Keep compiler happy. */ 2931 case SPELL_OPERATOR: 2932 /* token_no is used to track where multiple consecutive ## 2933 tokens were originally located. */ 2934 return (a->type != CPP_PASTE || a->val.token_no == b->val.token_no); 2935 case SPELL_NONE: 2936 return (a->type != CPP_MACRO_ARG 2937 || (a->val.macro_arg.arg_no == b->val.macro_arg.arg_no 2938 && a->val.macro_arg.spelling == b->val.macro_arg.spelling)); 2939 case SPELL_IDENT: 2940 return (a->val.node.node == b->val.node.node 2941 && a->val.node.spelling == b->val.node.spelling); 2942 case SPELL_LITERAL: 2943 return (a->val.str.len == b->val.str.len 2944 && !memcmp (a->val.str.text, b->val.str.text, 2945 a->val.str.len)); 2946 } 2947 2948 return 0; 2949} 2950 2951/* Returns nonzero if a space should be inserted to avoid an 2952 accidental token paste for output. For simplicity, it is 2953 conservative, and occasionally advises a space where one is not 2954 needed, e.g. "." and ".2". */ 2955int 2956cpp_avoid_paste (cpp_reader *pfile, const cpp_token *token1, 2957 const cpp_token *token2) 2958{ 2959 enum cpp_ttype a = token1->type, b = token2->type; 2960 cppchar_t c; 2961 2962 if (token1->flags & NAMED_OP) 2963 a = CPP_NAME; 2964 if (token2->flags & NAMED_OP) 2965 b = CPP_NAME; 2966 2967 c = EOF; 2968 if (token2->flags & DIGRAPH) 2969 c = digraph_spellings[(int) b - (int) CPP_FIRST_DIGRAPH][0]; 2970 else if (token_spellings[b].category == SPELL_OPERATOR) 2971 c = token_spellings[b].name[0]; 2972 2973 /* Quickly get everything that can paste with an '='. */ 2974 if ((int) a <= (int) CPP_LAST_EQ && c == '=') 2975 return 1; 2976 2977 switch (a) 2978 { 2979 case CPP_GREATER: return c == '>'; 2980 case CPP_LESS: return c == '<' || c == '%' || c == ':'; 2981 case CPP_PLUS: return c == '+'; 2982 case CPP_MINUS: return c == '-' || c == '>'; 2983 case CPP_DIV: return c == '/' || c == '*'; /* Comments. */ 2984 case CPP_MOD: return c == ':' || c == '>'; 2985 case CPP_AND: return c == '&'; 2986 case CPP_OR: return c == '|'; 2987 case CPP_COLON: return c == ':' || c == '>'; 2988 case CPP_DEREF: return c == '*'; 2989 case CPP_DOT: return c == '.' || c == '%' || b == CPP_NUMBER; 2990 case CPP_HASH: return c == '#' || c == '%'; /* Digraph form. */ 2991 case CPP_NAME: return ((b == CPP_NUMBER 2992 && name_p (pfile, &token2->val.str)) 2993 || b == CPP_NAME 2994 || b == CPP_CHAR || b == CPP_STRING); /* L */ 2995 case CPP_NUMBER: return (b == CPP_NUMBER || b == CPP_NAME 2996 || c == '.' || c == '+' || c == '-'); 2997 /* UCNs */ 2998 case CPP_OTHER: return ((token1->val.str.text[0] == '\\' 2999 && b == CPP_NAME) 3000 || (CPP_OPTION (pfile, objc) 3001 && token1->val.str.text[0] == '@' 3002 && (b == CPP_NAME || b == CPP_STRING))); 3003 case CPP_STRING: 3004 case CPP_WSTRING: 3005 case CPP_UTF8STRING: 3006 case CPP_STRING16: 3007 case CPP_STRING32: return (CPP_OPTION (pfile, user_literals) 3008 && (b == CPP_NAME 3009 || (TOKEN_SPELL (token2) == SPELL_LITERAL 3010 && ISIDST (token2->val.str.text[0])))); 3011 3012 default: break; 3013 } 3014 3015 return 0; 3016} 3017 3018/* Output all the remaining tokens on the current line, and a newline 3019 character, to FP. Leading whitespace is removed. If there are 3020 macros, special token padding is not performed. */ 3021void 3022cpp_output_line (cpp_reader *pfile, FILE *fp) 3023{ 3024 const cpp_token *token; 3025 3026 token = cpp_get_token (pfile); 3027 while (token->type != CPP_EOF) 3028 { 3029 cpp_output_token (token, fp); 3030 token = cpp_get_token (pfile); 3031 if (token->flags & PREV_WHITE) 3032 putc (' ', fp); 3033 } 3034 3035 putc ('\n', fp); 3036} 3037 3038/* Return a string representation of all the remaining tokens on the 3039 current line. The result is allocated using xmalloc and must be 3040 freed by the caller. */ 3041unsigned char * 3042cpp_output_line_to_string (cpp_reader *pfile, const unsigned char *dir_name) 3043{ 3044 const cpp_token *token; 3045 unsigned int out = dir_name ? ustrlen (dir_name) : 0; 3046 unsigned int alloced = 120 + out; 3047 unsigned char *result = (unsigned char *) xmalloc (alloced); 3048 3049 /* If DIR_NAME is empty, there are no initial contents. */ 3050 if (dir_name) 3051 { 3052 sprintf ((char *) result, "#%s ", dir_name); 3053 out += 2; 3054 } 3055 3056 token = cpp_get_token (pfile); 3057 while (token->type != CPP_EOF) 3058 { 3059 unsigned char *last; 3060 /* Include room for a possible space and the terminating nul. */ 3061 unsigned int len = cpp_token_len (token) + 2; 3062 3063 if (out + len > alloced) 3064 { 3065 alloced *= 2; 3066 if (out + len > alloced) 3067 alloced = out + len; 3068 result = (unsigned char *) xrealloc (result, alloced); 3069 } 3070 3071 last = cpp_spell_token (pfile, token, &result[out], 0); 3072 out = last - result; 3073 3074 token = cpp_get_token (pfile); 3075 if (token->flags & PREV_WHITE) 3076 result[out++] = ' '; 3077 } 3078 3079 result[out] = '\0'; 3080 return result; 3081} 3082 3083/* Memory buffers. Changing these three constants can have a dramatic 3084 effect on performance. The values here are reasonable defaults, 3085 but might be tuned. If you adjust them, be sure to test across a 3086 range of uses of cpplib, including heavy nested function-like macro 3087 expansion. Also check the change in peak memory usage (NJAMD is a 3088 good tool for this). */ 3089#define MIN_BUFF_SIZE 8000 3090#define BUFF_SIZE_UPPER_BOUND(MIN_SIZE) (MIN_BUFF_SIZE + (MIN_SIZE) * 3 / 2) 3091#define EXTENDED_BUFF_SIZE(BUFF, MIN_EXTRA) \ 3092 (MIN_EXTRA + ((BUFF)->limit - (BUFF)->cur) * 2) 3093 3094#if MIN_BUFF_SIZE > BUFF_SIZE_UPPER_BOUND (0) 3095 #error BUFF_SIZE_UPPER_BOUND must be at least as large as MIN_BUFF_SIZE! 3096#endif 3097 3098/* Create a new allocation buffer. Place the control block at the end 3099 of the buffer, so that buffer overflows will cause immediate chaos. */ 3100static _cpp_buff * 3101new_buff (size_t len) 3102{ 3103 _cpp_buff *result; 3104 unsigned char *base; 3105 3106 if (len < MIN_BUFF_SIZE) 3107 len = MIN_BUFF_SIZE; 3108 len = CPP_ALIGN (len); 3109 3110#ifdef ENABLE_VALGRIND_CHECKING 3111 /* Valgrind warns about uses of interior pointers, so put _cpp_buff 3112 struct first. */ 3113 size_t slen = CPP_ALIGN2 (sizeof (_cpp_buff), 2 * DEFAULT_ALIGNMENT); 3114 base = XNEWVEC (unsigned char, len + slen); 3115 result = (_cpp_buff *) base; 3116 base += slen; 3117#else 3118 base = XNEWVEC (unsigned char, len + sizeof (_cpp_buff)); 3119 result = (_cpp_buff *) (base + len); 3120#endif 3121 result->base = base; 3122 result->cur = base; 3123 result->limit = base + len; 3124 result->next = NULL; 3125 return result; 3126} 3127 3128/* Place a chain of unwanted allocation buffers on the free list. */ 3129void 3130_cpp_release_buff (cpp_reader *pfile, _cpp_buff *buff) 3131{ 3132 _cpp_buff *end = buff; 3133 3134 while (end->next) 3135 end = end->next; 3136 end->next = pfile->free_buffs; 3137 pfile->free_buffs = buff; 3138} 3139 3140/* Return a free buffer of size at least MIN_SIZE. */ 3141_cpp_buff * 3142_cpp_get_buff (cpp_reader *pfile, size_t min_size) 3143{ 3144 _cpp_buff *result, **p; 3145 3146 for (p = &pfile->free_buffs;; p = &(*p)->next) 3147 { 3148 size_t size; 3149 3150 if (*p == NULL) 3151 return new_buff (min_size); 3152 result = *p; 3153 size = result->limit - result->base; 3154 /* Return a buffer that's big enough, but don't waste one that's 3155 way too big. */ 3156 if (size >= min_size && size <= BUFF_SIZE_UPPER_BOUND (min_size)) 3157 break; 3158 } 3159 3160 *p = result->next; 3161 result->next = NULL; 3162 result->cur = result->base; 3163 return result; 3164} 3165 3166/* Creates a new buffer with enough space to hold the uncommitted 3167 remaining bytes of BUFF, and at least MIN_EXTRA more bytes. Copies 3168 the excess bytes to the new buffer. Chains the new buffer after 3169 BUFF, and returns the new buffer. */ 3170_cpp_buff * 3171_cpp_append_extend_buff (cpp_reader *pfile, _cpp_buff *buff, size_t min_extra) 3172{ 3173 size_t size = EXTENDED_BUFF_SIZE (buff, min_extra); 3174 _cpp_buff *new_buff = _cpp_get_buff (pfile, size); 3175 3176 buff->next = new_buff; 3177 memcpy (new_buff->base, buff->cur, BUFF_ROOM (buff)); 3178 return new_buff; 3179} 3180 3181/* Creates a new buffer with enough space to hold the uncommitted 3182 remaining bytes of the buffer pointed to by BUFF, and at least 3183 MIN_EXTRA more bytes. Copies the excess bytes to the new buffer. 3184 Chains the new buffer before the buffer pointed to by BUFF, and 3185 updates the pointer to point to the new buffer. */ 3186void 3187_cpp_extend_buff (cpp_reader *pfile, _cpp_buff **pbuff, size_t min_extra) 3188{ 3189 _cpp_buff *new_buff, *old_buff = *pbuff; 3190 size_t size = EXTENDED_BUFF_SIZE (old_buff, min_extra); 3191 3192 new_buff = _cpp_get_buff (pfile, size); 3193 memcpy (new_buff->base, old_buff->cur, BUFF_ROOM (old_buff)); 3194 new_buff->next = old_buff; 3195 *pbuff = new_buff; 3196} 3197 3198/* Free a chain of buffers starting at BUFF. */ 3199void 3200_cpp_free_buff (_cpp_buff *buff) 3201{ 3202 _cpp_buff *next; 3203 3204 for (; buff; buff = next) 3205 { 3206 next = buff->next; 3207#ifdef ENABLE_VALGRIND_CHECKING 3208 free (buff); 3209#else 3210 free (buff->base); 3211#endif 3212 } 3213} 3214 3215/* Allocate permanent, unaligned storage of length LEN. */ 3216unsigned char * 3217_cpp_unaligned_alloc (cpp_reader *pfile, size_t len) 3218{ 3219 _cpp_buff *buff = pfile->u_buff; 3220 unsigned char *result = buff->cur; 3221 3222 if (len > (size_t) (buff->limit - result)) 3223 { 3224 buff = _cpp_get_buff (pfile, len); 3225 buff->next = pfile->u_buff; 3226 pfile->u_buff = buff; 3227 result = buff->cur; 3228 } 3229 3230 buff->cur = result + len; 3231 return result; 3232} 3233 3234/* Allocate permanent, unaligned storage of length LEN from a_buff. 3235 That buffer is used for growing allocations when saving macro 3236 replacement lists in a #define, and when parsing an answer to an 3237 assertion in #assert, #unassert or #if (and therefore possibly 3238 whilst expanding macros). It therefore must not be used by any 3239 code that they might call: specifically the lexer and the guts of 3240 the macro expander. 3241 3242 All existing other uses clearly fit this restriction: storing 3243 registered pragmas during initialization. */ 3244unsigned char * 3245_cpp_aligned_alloc (cpp_reader *pfile, size_t len) 3246{ 3247 _cpp_buff *buff = pfile->a_buff; 3248 unsigned char *result = buff->cur; 3249 3250 if (len > (size_t) (buff->limit - result)) 3251 { 3252 buff = _cpp_get_buff (pfile, len); 3253 buff->next = pfile->a_buff; 3254 pfile->a_buff = buff; 3255 result = buff->cur; 3256 } 3257 3258 buff->cur = result + len; 3259 return result; 3260} 3261 3262/* Say which field of TOK is in use. */ 3263 3264enum cpp_token_fld_kind 3265cpp_token_val_index (const cpp_token *tok) 3266{ 3267 switch (TOKEN_SPELL (tok)) 3268 { 3269 case SPELL_IDENT: 3270 return CPP_TOKEN_FLD_NODE; 3271 case SPELL_LITERAL: 3272 return CPP_TOKEN_FLD_STR; 3273 case SPELL_OPERATOR: 3274 if (tok->type == CPP_PASTE) 3275 return CPP_TOKEN_FLD_TOKEN_NO; 3276 else 3277 return CPP_TOKEN_FLD_NONE; 3278 case SPELL_NONE: 3279 if (tok->type == CPP_MACRO_ARG) 3280 return CPP_TOKEN_FLD_ARG_NO; 3281 else if (tok->type == CPP_PADDING) 3282 return CPP_TOKEN_FLD_SOURCE; 3283 else if (tok->type == CPP_PRAGMA) 3284 return CPP_TOKEN_FLD_PRAGMA; 3285 /* else fall through */ 3286 default: 3287 return CPP_TOKEN_FLD_NONE; 3288 } 3289} 3290 3291/* All tokens lexed in R after calling this function will be forced to have 3292 their source_location the same as the location referenced by P, until 3293 cpp_stop_forcing_token_locations is called for R. */ 3294 3295void 3296cpp_force_token_locations (cpp_reader *r, source_location *p) 3297{ 3298 r->forced_token_location_p = p; 3299} 3300 3301/* Go back to assigning locations naturally for lexed tokens. */ 3302 3303void 3304cpp_stop_forcing_token_locations (cpp_reader *r) 3305{ 3306 r->forced_token_location_p = NULL; 3307} 3308