1/* CPP Library - lexical analysis. 2 Copyright (C) 2000-2022 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 = data == repl_nl; 395 t |= data == repl_cr; 396 t |= data == repl_bs; 397 t |= 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 + 15) & -16); 451 } 452 453 /* Main loop, processing 16 bytes at a time. */ 454#ifdef __GCC_ASM_FLAG_OUTPUTS__ 455 while (1) 456 { 457 char f; 458 459 /* By using inline assembly instead of the builtin, 460 we can use the result, as well as the flags set. */ 461 __asm ("%vpcmpestri\t$0, %2, %3" 462 : "=c"(index), "=@ccc"(f) 463 : "m"(*s), "x"(search), "a"(4), "d"(16)); 464 if (f) 465 break; 466 467 s += 16; 468 } 469#else 470 s -= 16; 471 /* By doing the whole loop in inline assembly, 472 we can make proper use of the flags set. */ 473 __asm ( ".balign 16\n" 474 "0: add $16, %1\n" 475 " %vpcmpestri\t$0, (%1), %2\n" 476 " jnc 0b" 477 : "=&c"(index), "+r"(s) 478 : "x"(search), "a"(4), "d"(16)); 479#endif 480 481 found: 482 return s + index; 483} 484 485#else 486/* Work around out-dated assemblers without sse4 support. */ 487#define search_line_sse42 search_line_sse2 488#endif 489 490/* Check the CPU capabilities. */ 491 492#include "../gcc/config/i386/cpuid.h" 493 494typedef const uchar * (*search_line_fast_type) (const uchar *, const uchar *); 495static search_line_fast_type search_line_fast; 496 497#define HAVE_init_vectorized_lexer 1 498static inline void 499init_vectorized_lexer (void) 500{ 501 unsigned dummy, ecx = 0, edx = 0; 502 search_line_fast_type impl = search_line_acc_char; 503 int minimum = 0; 504 505#if defined(__SSE4_2__) 506 minimum = 3; 507#elif defined(__SSE2__) 508 minimum = 2; 509#elif defined(__SSE__) 510 minimum = 1; 511#endif 512 513 if (minimum == 3) 514 impl = search_line_sse42; 515 else if (__get_cpuid (1, &dummy, &dummy, &ecx, &edx) || minimum == 2) 516 { 517 if (minimum == 3 || (ecx & bit_SSE4_2)) 518 impl = search_line_sse42; 519 else if (minimum == 2 || (edx & bit_SSE2)) 520 impl = search_line_sse2; 521 else if (minimum == 1 || (edx & bit_SSE)) 522 impl = search_line_mmx; 523 } 524 else if (__get_cpuid (0x80000001, &dummy, &dummy, &dummy, &edx)) 525 { 526 if (minimum == 1 527 || (edx & (bit_MMXEXT | bit_CMOV)) == (bit_MMXEXT | bit_CMOV)) 528 impl = search_line_mmx; 529 } 530 531 search_line_fast = impl; 532} 533 534#elif (GCC_VERSION >= 4005) && defined(_ARCH_PWR8) && defined(__ALTIVEC__) 535 536/* A vection of the fast scanner using AltiVec vectorized byte compares 537 and VSX unaligned loads (when VSX is available). This is otherwise 538 the same as the AltiVec version. */ 539 540ATTRIBUTE_NO_SANITIZE_UNDEFINED 541static const uchar * 542search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 543{ 544 typedef __attribute__((altivec(vector))) unsigned char vc; 545 546 const vc repl_nl = { 547 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', 548 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' 549 }; 550 const vc repl_cr = { 551 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', 552 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' 553 }; 554 const vc repl_bs = { 555 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', 556 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' 557 }; 558 const vc repl_qm = { 559 '?', '?', '?', '?', '?', '?', '?', '?', 560 '?', '?', '?', '?', '?', '?', '?', '?', 561 }; 562 const vc zero = { 0 }; 563 564 vc data, t; 565 566 /* Main loop processing 16 bytes at a time. */ 567 do 568 { 569 vc m_nl, m_cr, m_bs, m_qm; 570 571 data = __builtin_vec_vsx_ld (0, s); 572 s += 16; 573 574 m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl); 575 m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr); 576 m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs); 577 m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm); 578 t = (m_nl | m_cr) | (m_bs | m_qm); 579 580 /* T now contains 0xff in bytes for which we matched one of the relevant 581 characters. We want to exit the loop if any byte in T is non-zero. 582 Below is the expansion of vec_any_ne(t, zero). */ 583 } 584 while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero)); 585 586 /* Restore s to to point to the 16 bytes we just processed. */ 587 s -= 16; 588 589 { 590#define N (sizeof(vc) / sizeof(long)) 591 592 union { 593 vc v; 594 /* Statically assert that N is 2 or 4. */ 595 unsigned long l[(N == 2 || N == 4) ? N : -1]; 596 } u; 597 unsigned long l, i = 0; 598 599 u.v = t; 600 601 /* Find the first word of T that is non-zero. */ 602 switch (N) 603 { 604 case 4: 605 l = u.l[i++]; 606 if (l != 0) 607 break; 608 s += sizeof(unsigned long); 609 l = u.l[i++]; 610 if (l != 0) 611 break; 612 s += sizeof(unsigned long); 613 /* FALLTHRU */ 614 case 2: 615 l = u.l[i++]; 616 if (l != 0) 617 break; 618 s += sizeof(unsigned long); 619 l = u.l[i]; 620 } 621 622 /* L now contains 0xff in bytes for which we matched one of the 623 relevant characters. We can find the byte index by finding 624 its bit index and dividing by 8. */ 625#ifdef __BIG_ENDIAN__ 626 l = __builtin_clzl(l) >> 3; 627#else 628 l = __builtin_ctzl(l) >> 3; 629#endif 630 return s + l; 631 632#undef N 633 } 634} 635 636#elif (GCC_VERSION >= 4005) && defined(__ALTIVEC__) && defined (__BIG_ENDIAN__) 637 638/* A vection of the fast scanner using AltiVec vectorized byte compares. 639 This cannot be used for little endian because vec_lvsl/lvsr are 640 deprecated for little endian and the code won't work properly. */ 641/* ??? Unfortunately, attribute(target("altivec")) is not yet supported, 642 so we can't compile this function without -maltivec on the command line 643 (or implied by some other switch). */ 644 645static const uchar * 646search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 647{ 648 typedef __attribute__((altivec(vector))) unsigned char vc; 649 650 const vc repl_nl = { 651 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', 652 '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' 653 }; 654 const vc repl_cr = { 655 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', 656 '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' 657 }; 658 const vc repl_bs = { 659 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', 660 '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' 661 }; 662 const vc repl_qm = { 663 '?', '?', '?', '?', '?', '?', '?', '?', 664 '?', '?', '?', '?', '?', '?', '?', '?', 665 }; 666 const vc ones = { 667 -1, -1, -1, -1, -1, -1, -1, -1, 668 -1, -1, -1, -1, -1, -1, -1, -1, 669 }; 670 const vc zero = { 0 }; 671 672 vc data, mask, t; 673 674 /* Altivec loads automatically mask addresses with -16. This lets us 675 issue the first load as early as possible. */ 676 data = __builtin_vec_ld(0, (const vc *)s); 677 678 /* Discard bytes before the beginning of the buffer. Do this by 679 beginning with all ones and shifting in zeros according to the 680 mis-alignment. The LVSR instruction pulls the exact shift we 681 want from the address. */ 682 mask = __builtin_vec_lvsr(0, s); 683 mask = __builtin_vec_perm(zero, ones, mask); 684 data &= mask; 685 686 /* While altivec loads mask addresses, we still need to align S so 687 that the offset we compute at the end is correct. */ 688 s = (const uchar *)((uintptr_t)s & -16); 689 690 /* Main loop processing 16 bytes at a time. */ 691 goto start; 692 do 693 { 694 vc m_nl, m_cr, m_bs, m_qm; 695 696 s += 16; 697 data = __builtin_vec_ld(0, (const vc *)s); 698 699 start: 700 m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl); 701 m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr); 702 m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs); 703 m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm); 704 t = (m_nl | m_cr) | (m_bs | m_qm); 705 706 /* T now contains 0xff in bytes for which we matched one of the relevant 707 characters. We want to exit the loop if any byte in T is non-zero. 708 Below is the expansion of vec_any_ne(t, zero). */ 709 } 710 while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero)); 711 712 { 713#define N (sizeof(vc) / sizeof(long)) 714 715 union { 716 vc v; 717 /* Statically assert that N is 2 or 4. */ 718 unsigned long l[(N == 2 || N == 4) ? N : -1]; 719 } u; 720 unsigned long l, i = 0; 721 722 u.v = t; 723 724 /* Find the first word of T that is non-zero. */ 725 switch (N) 726 { 727 case 4: 728 l = u.l[i++]; 729 if (l != 0) 730 break; 731 s += sizeof(unsigned long); 732 l = u.l[i++]; 733 if (l != 0) 734 break; 735 s += sizeof(unsigned long); 736 /* FALLTHROUGH */ 737 case 2: 738 l = u.l[i++]; 739 if (l != 0) 740 break; 741 s += sizeof(unsigned long); 742 l = u.l[i]; 743 } 744 745 /* L now contains 0xff in bytes for which we matched one of the 746 relevant characters. We can find the byte index by finding 747 its bit index and dividing by 8. */ 748 l = __builtin_clzl(l) >> 3; 749 return s + l; 750 751#undef N 752 } 753} 754 755#elif defined (__ARM_NEON) && defined (__ARM_64BIT_STATE) 756#include "arm_neon.h" 757 758/* This doesn't have to be the exact page size, but no system may use 759 a size smaller than this. ARMv8 requires a minimum page size of 760 4k. The impact of being conservative here is a small number of 761 cases will take the slightly slower entry path into the main 762 loop. */ 763 764#define AARCH64_MIN_PAGE_SIZE 4096 765 766static const uchar * 767search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 768{ 769 const uint8x16_t repl_nl = vdupq_n_u8 ('\n'); 770 const uint8x16_t repl_cr = vdupq_n_u8 ('\r'); 771 const uint8x16_t repl_bs = vdupq_n_u8 ('\\'); 772 const uint8x16_t repl_qm = vdupq_n_u8 ('?'); 773 const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL); 774 775#ifdef __ARM_BIG_ENDIAN 776 const int16x8_t shift = {8, 8, 8, 8, 0, 0, 0, 0}; 777#else 778 const int16x8_t shift = {0, 0, 0, 0, 8, 8, 8, 8}; 779#endif 780 781 unsigned int found; 782 const uint8_t *p; 783 uint8x16_t data; 784 uint8x16_t t; 785 uint16x8_t m; 786 uint8x16_t u, v, w; 787 788 /* Align the source pointer. */ 789 p = (const uint8_t *)((uintptr_t)s & -16); 790 791 /* Assuming random string start positions, with a 4k page size we'll take 792 the slow path about 0.37% of the time. */ 793 if (__builtin_expect ((AARCH64_MIN_PAGE_SIZE 794 - (((uintptr_t) s) & (AARCH64_MIN_PAGE_SIZE - 1))) 795 < 16, 0)) 796 { 797 /* Slow path: the string starts near a possible page boundary. */ 798 uint32_t misalign, mask; 799 800 misalign = (uintptr_t)s & 15; 801 mask = (-1u << misalign) & 0xffff; 802 data = vld1q_u8 (p); 803 t = vceqq_u8 (data, repl_nl); 804 u = vceqq_u8 (data, repl_cr); 805 v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); 806 w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); 807 t = vorrq_u8 (v, w); 808 t = vandq_u8 (t, xmask); 809 m = vpaddlq_u8 (t); 810 m = vshlq_u16 (m, shift); 811 found = vaddvq_u16 (m); 812 found &= mask; 813 if (found) 814 return (const uchar*)p + __builtin_ctz (found); 815 } 816 else 817 { 818 data = vld1q_u8 ((const uint8_t *) s); 819 t = vceqq_u8 (data, repl_nl); 820 u = vceqq_u8 (data, repl_cr); 821 v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); 822 w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); 823 t = vorrq_u8 (v, w); 824 if (__builtin_expect (vpaddd_u64 ((uint64x2_t)t) != 0, 0)) 825 goto done; 826 } 827 828 do 829 { 830 p += 16; 831 data = vld1q_u8 (p); 832 t = vceqq_u8 (data, repl_nl); 833 u = vceqq_u8 (data, repl_cr); 834 v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); 835 w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); 836 t = vorrq_u8 (v, w); 837 } while (!vpaddd_u64 ((uint64x2_t)t)); 838 839done: 840 /* Now that we've found the terminating substring, work out precisely where 841 we need to stop. */ 842 t = vandq_u8 (t, xmask); 843 m = vpaddlq_u8 (t); 844 m = vshlq_u16 (m, shift); 845 found = vaddvq_u16 (m); 846 return (((((uintptr_t) p) < (uintptr_t) s) ? s : (const uchar *)p) 847 + __builtin_ctz (found)); 848} 849 850#elif defined (__ARM_NEON) 851#include "arm_neon.h" 852 853static const uchar * 854search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) 855{ 856 const uint8x16_t repl_nl = vdupq_n_u8 ('\n'); 857 const uint8x16_t repl_cr = vdupq_n_u8 ('\r'); 858 const uint8x16_t repl_bs = vdupq_n_u8 ('\\'); 859 const uint8x16_t repl_qm = vdupq_n_u8 ('?'); 860 const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL); 861 862 unsigned int misalign, found, mask; 863 const uint8_t *p; 864 uint8x16_t data; 865 866 /* Align the source pointer. */ 867 misalign = (uintptr_t)s & 15; 868 p = (const uint8_t *)((uintptr_t)s & -16); 869 data = vld1q_u8 (p); 870 871 /* Create a mask for the bytes that are valid within the first 872 16-byte block. The Idea here is that the AND with the mask 873 within the loop is "free", since we need some AND or TEST 874 insn in order to set the flags for the branch anyway. */ 875 mask = (-1u << misalign) & 0xffff; 876 877 /* Main loop, processing 16 bytes at a time. */ 878 goto start; 879 880 do 881 { 882 uint8x8_t l; 883 uint16x4_t m; 884 uint32x2_t n; 885 uint8x16_t t, u, v, w; 886 887 p += 16; 888 data = vld1q_u8 (p); 889 mask = 0xffff; 890 891 start: 892 t = vceqq_u8 (data, repl_nl); 893 u = vceqq_u8 (data, repl_cr); 894 v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); 895 w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); 896 t = vandq_u8 (vorrq_u8 (v, w), xmask); 897 l = vpadd_u8 (vget_low_u8 (t), vget_high_u8 (t)); 898 m = vpaddl_u8 (l); 899 n = vpaddl_u16 (m); 900 901 found = vget_lane_u32 ((uint32x2_t) vorr_u64 ((uint64x1_t) n, 902 vshr_n_u64 ((uint64x1_t) n, 24)), 0); 903 found &= mask; 904 } 905 while (!found); 906 907 /* FOUND contains 1 in bits for which we matched a relevant 908 character. Conversion to the byte index is trivial. */ 909 found = __builtin_ctz (found); 910 return (const uchar *)p + found; 911} 912 913#else 914 915/* We only have one accelerated alternative. Use a direct call so that 916 we encourage inlining. */ 917 918#define search_line_fast search_line_acc_char 919 920#endif 921 922/* Initialize the lexer if needed. */ 923 924void 925_cpp_init_lexer (void) 926{ 927#ifdef HAVE_init_vectorized_lexer 928 init_vectorized_lexer (); 929#endif 930} 931 932/* Returns with a logical line that contains no escaped newlines or 933 trigraphs. This is a time-critical inner loop. */ 934void 935_cpp_clean_line (cpp_reader *pfile) 936{ 937 cpp_buffer *buffer; 938 const uchar *s; 939 uchar c, *d, *p; 940 941 buffer = pfile->buffer; 942 buffer->cur_note = buffer->notes_used = 0; 943 buffer->cur = buffer->line_base = buffer->next_line; 944 buffer->need_line = false; 945 s = buffer->next_line; 946 947 if (!buffer->from_stage3) 948 { 949 const uchar *pbackslash = NULL; 950 951 /* Fast path. This is the common case of an un-escaped line with 952 no trigraphs. The primary win here is by not writing any 953 data back to memory until we have to. */ 954 while (1) 955 { 956 /* Perform an optimized search for \n, \r, \\, ?. */ 957 s = search_line_fast (s, buffer->rlimit); 958 959 c = *s; 960 if (c == '\\') 961 { 962 /* Record the location of the backslash and continue. */ 963 pbackslash = s++; 964 } 965 else if (__builtin_expect (c == '?', 0)) 966 { 967 if (__builtin_expect (s[1] == '?', false) 968 && _cpp_trigraph_map[s[2]]) 969 { 970 /* Have a trigraph. We may or may not have to convert 971 it. Add a line note regardless, for -Wtrigraphs. */ 972 add_line_note (buffer, s, s[2]); 973 if (CPP_OPTION (pfile, trigraphs)) 974 { 975 /* We do, and that means we have to switch to the 976 slow path. */ 977 d = (uchar *) s; 978 *d = _cpp_trigraph_map[s[2]]; 979 s += 2; 980 goto slow_path; 981 } 982 } 983 /* Not a trigraph. Continue on fast-path. */ 984 s++; 985 } 986 else 987 break; 988 } 989 990 /* This must be \r or \n. We're either done, or we'll be forced 991 to write back to the buffer and continue on the slow path. */ 992 d = (uchar *) s; 993 994 if (__builtin_expect (s == buffer->rlimit, false)) 995 goto done; 996 997 /* DOS line ending? */ 998 if (__builtin_expect (c == '\r', false) && s[1] == '\n') 999 { 1000 s++; 1001 if (s == buffer->rlimit) 1002 goto done; 1003 } 1004 1005 if (__builtin_expect (pbackslash == NULL, true)) 1006 goto done; 1007 1008 /* Check for escaped newline. */ 1009 p = d; 1010 while (is_nvspace (p[-1])) 1011 p--; 1012 if (p - 1 != pbackslash) 1013 goto done; 1014 1015 /* Have an escaped newline; process it and proceed to 1016 the slow path. */ 1017 add_line_note (buffer, p - 1, p != d ? ' ' : '\\'); 1018 d = p - 2; 1019 buffer->next_line = p - 1; 1020 1021 slow_path: 1022 while (1) 1023 { 1024 c = *++s; 1025 *++d = c; 1026 1027 if (c == '\n' || c == '\r') 1028 { 1029 /* Handle DOS line endings. */ 1030 if (c == '\r' && s != buffer->rlimit && s[1] == '\n') 1031 s++; 1032 if (s == buffer->rlimit) 1033 break; 1034 1035 /* Escaped? */ 1036 p = d; 1037 while (p != buffer->next_line && is_nvspace (p[-1])) 1038 p--; 1039 if (p == buffer->next_line || p[-1] != '\\') 1040 break; 1041 1042 add_line_note (buffer, p - 1, p != d ? ' ': '\\'); 1043 d = p - 2; 1044 buffer->next_line = p - 1; 1045 } 1046 else if (c == '?' && s[1] == '?' && _cpp_trigraph_map[s[2]]) 1047 { 1048 /* Add a note regardless, for the benefit of -Wtrigraphs. */ 1049 add_line_note (buffer, d, s[2]); 1050 if (CPP_OPTION (pfile, trigraphs)) 1051 { 1052 *d = _cpp_trigraph_map[s[2]]; 1053 s += 2; 1054 } 1055 } 1056 } 1057 } 1058 else 1059 { 1060 while (*s != '\n' && *s != '\r') 1061 s++; 1062 d = (uchar *) s; 1063 1064 /* Handle DOS line endings. */ 1065 if (*s == '\r' && s + 1 != buffer->rlimit && s[1] == '\n') 1066 s++; 1067 } 1068 1069 done: 1070 *d = '\n'; 1071 /* A sentinel note that should never be processed. */ 1072 add_line_note (buffer, d + 1, '\n'); 1073 buffer->next_line = s + 1; 1074} 1075 1076/* Return true if the trigraph indicated by NOTE should be warned 1077 about in a comment. */ 1078static bool 1079warn_in_comment (cpp_reader *pfile, _cpp_line_note *note) 1080{ 1081 const uchar *p; 1082 1083 /* Within comments we don't warn about trigraphs, unless the 1084 trigraph forms an escaped newline, as that may change 1085 behavior. */ 1086 if (note->type != '/') 1087 return false; 1088 1089 /* If -trigraphs, then this was an escaped newline iff the next note 1090 is coincident. */ 1091 if (CPP_OPTION (pfile, trigraphs)) 1092 return note[1].pos == note->pos; 1093 1094 /* Otherwise, see if this forms an escaped newline. */ 1095 p = note->pos + 3; 1096 while (is_nvspace (*p)) 1097 p++; 1098 1099 /* There might have been escaped newlines between the trigraph and the 1100 newline we found. Hence the position test. */ 1101 return (*p == '\n' && p < note[1].pos); 1102} 1103 1104/* Process the notes created by add_line_note as far as the current 1105 location. */ 1106void 1107_cpp_process_line_notes (cpp_reader *pfile, int in_comment) 1108{ 1109 cpp_buffer *buffer = pfile->buffer; 1110 1111 for (;;) 1112 { 1113 _cpp_line_note *note = &buffer->notes[buffer->cur_note]; 1114 unsigned int col; 1115 1116 if (note->pos > buffer->cur) 1117 break; 1118 1119 buffer->cur_note++; 1120 col = CPP_BUF_COLUMN (buffer, note->pos + 1); 1121 1122 if (note->type == '\\' || note->type == ' ') 1123 { 1124 if (note->type == ' ' && !in_comment) 1125 cpp_error_with_line (pfile, CPP_DL_WARNING, pfile->line_table->highest_line, col, 1126 "backslash and newline separated by space"); 1127 1128 if (buffer->next_line > buffer->rlimit) 1129 { 1130 cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, col, 1131 "backslash-newline at end of file"); 1132 /* Prevent "no newline at end of file" warning. */ 1133 buffer->next_line = buffer->rlimit; 1134 } 1135 1136 buffer->line_base = note->pos; 1137 CPP_INCREMENT_LINE (pfile, 0); 1138 } 1139 else if (_cpp_trigraph_map[note->type]) 1140 { 1141 if (CPP_OPTION (pfile, warn_trigraphs) 1142 && (!in_comment || warn_in_comment (pfile, note))) 1143 { 1144 if (CPP_OPTION (pfile, trigraphs)) 1145 cpp_warning_with_line (pfile, CPP_W_TRIGRAPHS, 1146 pfile->line_table->highest_line, col, 1147 "trigraph ??%c converted to %c", 1148 note->type, 1149 (int) _cpp_trigraph_map[note->type]); 1150 else 1151 { 1152 cpp_warning_with_line 1153 (pfile, CPP_W_TRIGRAPHS, 1154 pfile->line_table->highest_line, col, 1155 "trigraph ??%c ignored, use -trigraphs to enable", 1156 note->type); 1157 } 1158 } 1159 } 1160 else if (note->type == 0) 1161 /* Already processed in lex_raw_string. */; 1162 else 1163 abort (); 1164 } 1165} 1166 1167namespace bidi { 1168 enum class kind { 1169 NONE, LRE, RLE, LRO, RLO, LRI, RLI, FSI, PDF, PDI, LTR, RTL 1170 }; 1171 1172 /* All the UTF-8 encodings of bidi characters start with E2. */ 1173 constexpr uchar utf8_start = 0xe2; 1174 1175 struct context 1176 { 1177 context () {} 1178 context (location_t loc, kind k, bool pdf, bool ucn) 1179 : m_loc (loc), m_kind (k), m_pdf (pdf), m_ucn (ucn) 1180 { 1181 } 1182 1183 kind get_pop_kind () const 1184 { 1185 return m_pdf ? kind::PDF : kind::PDI; 1186 } 1187 bool ucn_p () const 1188 { 1189 return m_ucn; 1190 } 1191 1192 location_t m_loc; 1193 kind m_kind; 1194 unsigned m_pdf : 1; 1195 unsigned m_ucn : 1; 1196 }; 1197 1198 /* A vector holding currently open bidi contexts. We use a char for 1199 each context, its LSB is 1 if it represents a PDF context, 0 if it 1200 represents a PDI context. The next bit is 1 if this context was open 1201 by a bidi character written as a UCN, and 0 when it was UTF-8. */ 1202 semi_embedded_vec <context, 16> vec; 1203 1204 /* Close the whole comment/identifier/string literal/character constant 1205 context. */ 1206 void on_close () 1207 { 1208 vec.truncate (0); 1209 } 1210 1211 /* Pop the last element in the vector. */ 1212 void pop () 1213 { 1214 unsigned int len = vec.count (); 1215 gcc_checking_assert (len > 0); 1216 vec.truncate (len - 1); 1217 } 1218 1219 /* Return the pop kind of the context of the Ith element. */ 1220 kind pop_kind_at (unsigned int i) 1221 { 1222 return vec[i].get_pop_kind (); 1223 } 1224 1225 /* Return the pop kind of the context that is currently opened. */ 1226 kind current_ctx () 1227 { 1228 unsigned int len = vec.count (); 1229 if (len == 0) 1230 return kind::NONE; 1231 return vec[len - 1].get_pop_kind (); 1232 } 1233 1234 /* Return true if the current context comes from a UCN origin, that is, 1235 the bidi char which started this bidi context was written as a UCN. */ 1236 bool current_ctx_ucn_p () 1237 { 1238 unsigned int len = vec.count (); 1239 gcc_checking_assert (len > 0); 1240 return vec[len - 1].m_ucn; 1241 } 1242 1243 location_t current_ctx_loc () 1244 { 1245 unsigned int len = vec.count (); 1246 gcc_checking_assert (len > 0); 1247 return vec[len - 1].m_loc; 1248 } 1249 1250 /* We've read a bidi char, update the current vector as necessary. 1251 LOC is only valid when K is not kind::NONE. */ 1252 void on_char (kind k, bool ucn_p, location_t loc) 1253 { 1254 switch (k) 1255 { 1256 case kind::LRE: 1257 case kind::RLE: 1258 case kind::LRO: 1259 case kind::RLO: 1260 vec.push (context (loc, k, true, ucn_p)); 1261 break; 1262 case kind::LRI: 1263 case kind::RLI: 1264 case kind::FSI: 1265 vec.push (context (loc, k, false, ucn_p)); 1266 break; 1267 /* PDF terminates the scope of the last LRE, RLE, LRO, or RLO 1268 whose scope has not yet been terminated. */ 1269 case kind::PDF: 1270 if (current_ctx () == kind::PDF) 1271 pop (); 1272 break; 1273 /* PDI terminates the scope of the last LRI, RLI, or FSI whose 1274 scope has not yet been terminated, as well as the scopes of 1275 any subsequent LREs, RLEs, LROs, or RLOs whose scopes have not 1276 yet been terminated. */ 1277 case kind::PDI: 1278 for (int i = vec.count () - 1; i >= 0; --i) 1279 if (pop_kind_at (i) == kind::PDI) 1280 { 1281 vec.truncate (i); 1282 break; 1283 } 1284 break; 1285 case kind::LTR: 1286 case kind::RTL: 1287 /* These aren't popped by a PDF/PDI. */ 1288 break; 1289 ATTR_LIKELY case kind::NONE: 1290 break; 1291 default: 1292 abort (); 1293 } 1294 } 1295 1296 /* Return a descriptive string for K. */ 1297 const char *to_str (kind k) 1298 { 1299 switch (k) 1300 { 1301 case kind::LRE: 1302 return "U+202A (LEFT-TO-RIGHT EMBEDDING)"; 1303 case kind::RLE: 1304 return "U+202B (RIGHT-TO-LEFT EMBEDDING)"; 1305 case kind::LRO: 1306 return "U+202D (LEFT-TO-RIGHT OVERRIDE)"; 1307 case kind::RLO: 1308 return "U+202E (RIGHT-TO-LEFT OVERRIDE)"; 1309 case kind::LRI: 1310 return "U+2066 (LEFT-TO-RIGHT ISOLATE)"; 1311 case kind::RLI: 1312 return "U+2067 (RIGHT-TO-LEFT ISOLATE)"; 1313 case kind::FSI: 1314 return "U+2068 (FIRST STRONG ISOLATE)"; 1315 case kind::PDF: 1316 return "U+202C (POP DIRECTIONAL FORMATTING)"; 1317 case kind::PDI: 1318 return "U+2069 (POP DIRECTIONAL ISOLATE)"; 1319 case kind::LTR: 1320 return "U+200E (LEFT-TO-RIGHT MARK)"; 1321 case kind::RTL: 1322 return "U+200F (RIGHT-TO-LEFT MARK)"; 1323 default: 1324 abort (); 1325 } 1326 } 1327} 1328 1329/* Get location_t for the range of bytes [START, START + NUM_BYTES) 1330 within the current line in FILE, with the caret at START. */ 1331 1332static location_t 1333get_location_for_byte_range_in_cur_line (cpp_reader *pfile, 1334 const unsigned char *const start, 1335 size_t num_bytes) 1336{ 1337 gcc_checking_assert (num_bytes > 0); 1338 1339 /* CPP_BUF_COLUMN and linemap_position_for_column both refer 1340 to offsets in bytes, but CPP_BUF_COLUMN is 0-based, 1341 whereas linemap_position_for_column is 1-based. */ 1342 1343 /* Get 0-based offsets within the line. */ 1344 size_t start_offset = CPP_BUF_COLUMN (pfile->buffer, start); 1345 size_t end_offset = start_offset + num_bytes - 1; 1346 1347 /* Now convert to location_t, where "columns" are 1-based byte offsets. */ 1348 location_t start_loc = linemap_position_for_column (pfile->line_table, 1349 start_offset + 1); 1350 location_t end_loc = linemap_position_for_column (pfile->line_table, 1351 end_offset + 1); 1352 1353 if (start_loc == end_loc) 1354 return start_loc; 1355 1356 source_range src_range; 1357 src_range.m_start = start_loc; 1358 src_range.m_finish = end_loc; 1359 location_t combined_loc = COMBINE_LOCATION_DATA (pfile->line_table, 1360 start_loc, 1361 src_range, 1362 NULL); 1363 return combined_loc; 1364} 1365 1366/* Parse a sequence of 3 bytes starting with P and return its bidi code. */ 1367 1368static bidi::kind 1369get_bidi_utf8_1 (const unsigned char *const p) 1370{ 1371 gcc_checking_assert (p[0] == bidi::utf8_start); 1372 1373 if (p[1] == 0x80) 1374 switch (p[2]) 1375 { 1376 case 0xaa: 1377 return bidi::kind::LRE; 1378 case 0xab: 1379 return bidi::kind::RLE; 1380 case 0xac: 1381 return bidi::kind::PDF; 1382 case 0xad: 1383 return bidi::kind::LRO; 1384 case 0xae: 1385 return bidi::kind::RLO; 1386 case 0x8e: 1387 return bidi::kind::LTR; 1388 case 0x8f: 1389 return bidi::kind::RTL; 1390 default: 1391 break; 1392 } 1393 else if (p[1] == 0x81) 1394 switch (p[2]) 1395 { 1396 case 0xa6: 1397 return bidi::kind::LRI; 1398 case 0xa7: 1399 return bidi::kind::RLI; 1400 case 0xa8: 1401 return bidi::kind::FSI; 1402 case 0xa9: 1403 return bidi::kind::PDI; 1404 default: 1405 break; 1406 } 1407 1408 return bidi::kind::NONE; 1409} 1410 1411/* Parse a sequence of 3 bytes starting with P and return its bidi code. 1412 If the kind is not NONE, write the location to *OUT.*/ 1413 1414static bidi::kind 1415get_bidi_utf8 (cpp_reader *pfile, const unsigned char *const p, location_t *out) 1416{ 1417 bidi::kind result = get_bidi_utf8_1 (p); 1418 if (result != bidi::kind::NONE) 1419 { 1420 /* We have a sequence of 3 bytes starting at P. */ 1421 *out = get_location_for_byte_range_in_cur_line (pfile, p, 3); 1422 } 1423 return result; 1424} 1425 1426/* Parse a UCN where P points just past \u or \U and return its bidi code. */ 1427 1428static bidi::kind 1429get_bidi_ucn_1 (const unsigned char *p, bool is_U) 1430{ 1431 /* 6.4.3 Universal Character Names 1432 \u hex-quad 1433 \U hex-quad hex-quad 1434 where \unnnn means \U0000nnnn. */ 1435 1436 if (is_U) 1437 { 1438 if (p[0] != '0' || p[1] != '0' || p[2] != '0' || p[3] != '0') 1439 return bidi::kind::NONE; 1440 /* Skip 4B so we can treat \u and \U the same below. */ 1441 p += 4; 1442 } 1443 1444 /* All code points we are looking for start with 20xx. */ 1445 if (p[0] != '2' || p[1] != '0') 1446 return bidi::kind::NONE; 1447 else if (p[2] == '2') 1448 switch (p[3]) 1449 { 1450 case 'a': 1451 case 'A': 1452 return bidi::kind::LRE; 1453 case 'b': 1454 case 'B': 1455 return bidi::kind::RLE; 1456 case 'c': 1457 case 'C': 1458 return bidi::kind::PDF; 1459 case 'd': 1460 case 'D': 1461 return bidi::kind::LRO; 1462 case 'e': 1463 case 'E': 1464 return bidi::kind::RLO; 1465 default: 1466 break; 1467 } 1468 else if (p[2] == '6') 1469 switch (p[3]) 1470 { 1471 case '6': 1472 return bidi::kind::LRI; 1473 case '7': 1474 return bidi::kind::RLI; 1475 case '8': 1476 return bidi::kind::FSI; 1477 case '9': 1478 return bidi::kind::PDI; 1479 default: 1480 break; 1481 } 1482 else if (p[2] == '0') 1483 switch (p[3]) 1484 { 1485 case 'e': 1486 case 'E': 1487 return bidi::kind::LTR; 1488 case 'f': 1489 case 'F': 1490 return bidi::kind::RTL; 1491 default: 1492 break; 1493 } 1494 1495 return bidi::kind::NONE; 1496} 1497 1498/* Parse a UCN where P points just past \u or \U and return its bidi code. 1499 If the kind is not NONE, write the location to *OUT.*/ 1500 1501static bidi::kind 1502get_bidi_ucn (cpp_reader *pfile, const unsigned char *p, bool is_U, 1503 location_t *out) 1504{ 1505 bidi::kind result = get_bidi_ucn_1 (p, is_U); 1506 if (result != bidi::kind::NONE) 1507 { 1508 const unsigned char *start = p - 2; 1509 size_t num_bytes = 2 + (is_U ? 8 : 4); 1510 *out = get_location_for_byte_range_in_cur_line (pfile, start, num_bytes); 1511 } 1512 return result; 1513} 1514 1515/* Subclass of rich_location for reporting on unpaired UTF-8 1516 bidirectional control character(s). 1517 Escape the source lines on output, and show all unclosed 1518 bidi context, labelling everything. */ 1519 1520class unpaired_bidi_rich_location : public rich_location 1521{ 1522 public: 1523 class custom_range_label : public range_label 1524 { 1525 public: 1526 label_text get_text (unsigned range_idx) const FINAL OVERRIDE 1527 { 1528 /* range 0 is the primary location; each subsequent range i + 1 1529 is for bidi::vec[i]. */ 1530 if (range_idx > 0) 1531 { 1532 const bidi::context &ctxt (bidi::vec[range_idx - 1]); 1533 return label_text::borrow (bidi::to_str (ctxt.m_kind)); 1534 } 1535 else 1536 return label_text::borrow (_("end of bidirectional context")); 1537 } 1538 }; 1539 1540 unpaired_bidi_rich_location (cpp_reader *pfile, location_t loc) 1541 : rich_location (pfile->line_table, loc, &m_custom_label) 1542 { 1543 set_escape_on_output (true); 1544 for (unsigned i = 0; i < bidi::vec.count (); i++) 1545 add_range (bidi::vec[i].m_loc, 1546 SHOW_RANGE_WITHOUT_CARET, 1547 &m_custom_label); 1548 } 1549 1550 private: 1551 custom_range_label m_custom_label; 1552}; 1553 1554/* We're closing a bidi context, that is, we've encountered a newline, 1555 are closing a C-style comment, or are at the end of a string literal, 1556 character constant, or identifier. Warn if this context was not 1557 properly terminated by a PDI or PDF. P points to the last character 1558 in this context. */ 1559 1560static void 1561maybe_warn_bidi_on_close (cpp_reader *pfile, const uchar *p) 1562{ 1563 const auto warn_bidi = CPP_OPTION (pfile, cpp_warn_bidirectional); 1564 if (bidi::vec.count () > 0 1565 && (warn_bidi & bidirectional_unpaired 1566 && (!bidi::current_ctx_ucn_p () 1567 || (warn_bidi & bidirectional_ucn)))) 1568 { 1569 const location_t loc 1570 = linemap_position_for_column (pfile->line_table, 1571 CPP_BUF_COLUMN (pfile->buffer, p)); 1572 unpaired_bidi_rich_location rich_loc (pfile, loc); 1573 /* cpp_callbacks doesn't yet have a way to handle singular vs plural 1574 forms of a diagnostic, so fake it for now. */ 1575 if (bidi::vec.count () > 1) 1576 cpp_warning_at (pfile, CPP_W_BIDIRECTIONAL, &rich_loc, 1577 "unpaired UTF-8 bidirectional control characters " 1578 "detected"); 1579 else 1580 cpp_warning_at (pfile, CPP_W_BIDIRECTIONAL, &rich_loc, 1581 "unpaired UTF-8 bidirectional control character " 1582 "detected"); 1583 } 1584 /* We're done with this context. */ 1585 bidi::on_close (); 1586} 1587 1588/* We're at the beginning or in the middle of an identifier/comment/string 1589 literal/character constant. Warn if we've encountered a bidi character. 1590 KIND says which bidi control character it was; UCN_P is true iff this bidi 1591 control character was written as a UCN. LOC is the location of the 1592 character, but is only valid if KIND != bidi::kind::NONE. */ 1593 1594static void 1595maybe_warn_bidi_on_char (cpp_reader *pfile, bidi::kind kind, 1596 bool ucn_p, location_t loc) 1597{ 1598 if (__builtin_expect (kind == bidi::kind::NONE, 1)) 1599 return; 1600 1601 const auto warn_bidi = CPP_OPTION (pfile, cpp_warn_bidirectional); 1602 1603 if (warn_bidi & (bidirectional_unpaired|bidirectional_any)) 1604 { 1605 rich_location rich_loc (pfile->line_table, loc); 1606 rich_loc.set_escape_on_output (true); 1607 1608 /* It seems excessive to warn about a PDI/PDF that is closing 1609 an opened context because we've already warned about the 1610 opening character. Except warn when we have a UCN x UTF-8 1611 mismatch, if UCN checking is enabled. */ 1612 if (kind == bidi::current_ctx ()) 1613 { 1614 if (warn_bidi == (bidirectional_unpaired|bidirectional_ucn) 1615 && bidi::current_ctx_ucn_p () != ucn_p) 1616 { 1617 rich_loc.add_range (bidi::current_ctx_loc ()); 1618 cpp_warning_at (pfile, CPP_W_BIDIRECTIONAL, &rich_loc, 1619 "UTF-8 vs UCN mismatch when closing " 1620 "a context by \"%s\"", bidi::to_str (kind)); 1621 } 1622 } 1623 else if (warn_bidi & bidirectional_any 1624 && (!ucn_p || (warn_bidi & bidirectional_ucn))) 1625 { 1626 if (kind == bidi::kind::PDF || kind == bidi::kind::PDI) 1627 cpp_warning_at (pfile, CPP_W_BIDIRECTIONAL, &rich_loc, 1628 "\"%s\" is closing an unopened context", 1629 bidi::to_str (kind)); 1630 else 1631 cpp_warning_at (pfile, CPP_W_BIDIRECTIONAL, &rich_loc, 1632 "found problematic Unicode character \"%s\"", 1633 bidi::to_str (kind)); 1634 } 1635 } 1636 /* We're done with this context. */ 1637 bidi::on_char (kind, ucn_p, loc); 1638} 1639 1640/* Skip a C-style block comment. We find the end of the comment by 1641 seeing if an asterisk is before every '/' we encounter. Returns 1642 nonzero if comment terminated by EOF, zero otherwise. 1643 1644 Buffer->cur points to the initial asterisk of the comment. */ 1645bool 1646_cpp_skip_block_comment (cpp_reader *pfile) 1647{ 1648 cpp_buffer *buffer = pfile->buffer; 1649 const uchar *cur = buffer->cur; 1650 uchar c; 1651 const bool warn_bidi_p = pfile->warn_bidi_p (); 1652 1653 cur++; 1654 if (*cur == '/') 1655 cur++; 1656 1657 for (;;) 1658 { 1659 /* People like decorating comments with '*', so check for '/' 1660 instead for efficiency. */ 1661 c = *cur++; 1662 1663 if (c == '/') 1664 { 1665 if (cur[-2] == '*') 1666 { 1667 if (warn_bidi_p) 1668 maybe_warn_bidi_on_close (pfile, cur); 1669 break; 1670 } 1671 1672 /* Warn about potential nested comments, but not if the '/' 1673 comes immediately before the true comment delimiter. 1674 Don't bother to get it right across escaped newlines. */ 1675 if (CPP_OPTION (pfile, warn_comments) 1676 && cur[0] == '*' && cur[1] != '/') 1677 { 1678 buffer->cur = cur; 1679 cpp_warning_with_line (pfile, CPP_W_COMMENTS, 1680 pfile->line_table->highest_line, 1681 CPP_BUF_COL (buffer), 1682 "\"/*\" within comment"); 1683 } 1684 } 1685 else if (c == '\n') 1686 { 1687 unsigned int cols; 1688 buffer->cur = cur - 1; 1689 if (warn_bidi_p) 1690 maybe_warn_bidi_on_close (pfile, cur); 1691 _cpp_process_line_notes (pfile, true); 1692 if (buffer->next_line >= buffer->rlimit) 1693 return true; 1694 _cpp_clean_line (pfile); 1695 1696 cols = buffer->next_line - buffer->line_base; 1697 CPP_INCREMENT_LINE (pfile, cols); 1698 1699 cur = buffer->cur; 1700 } 1701 /* If this is a beginning of a UTF-8 encoding, it might be 1702 a bidirectional control character. */ 1703 else if (__builtin_expect (c == bidi::utf8_start, 0) && warn_bidi_p) 1704 { 1705 location_t loc; 1706 bidi::kind kind = get_bidi_utf8 (pfile, cur - 1, &loc); 1707 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/false, loc); 1708 } 1709 } 1710 1711 buffer->cur = cur; 1712 _cpp_process_line_notes (pfile, true); 1713 return false; 1714} 1715 1716/* Skip a C++ line comment, leaving buffer->cur pointing to the 1717 terminating newline. Handles escaped newlines. Returns nonzero 1718 if a multiline comment. */ 1719static int 1720skip_line_comment (cpp_reader *pfile) 1721{ 1722 cpp_buffer *buffer = pfile->buffer; 1723 location_t orig_line = pfile->line_table->highest_line; 1724 const bool warn_bidi_p = pfile->warn_bidi_p (); 1725 1726 if (!warn_bidi_p) 1727 while (*buffer->cur != '\n') 1728 buffer->cur++; 1729 else 1730 { 1731 while (*buffer->cur != '\n' 1732 && *buffer->cur != bidi::utf8_start) 1733 buffer->cur++; 1734 if (__builtin_expect (*buffer->cur == bidi::utf8_start, 0)) 1735 { 1736 while (*buffer->cur != '\n') 1737 { 1738 if (__builtin_expect (*buffer->cur == bidi::utf8_start, 0)) 1739 { 1740 location_t loc; 1741 bidi::kind kind = get_bidi_utf8 (pfile, buffer->cur, &loc); 1742 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/false, loc); 1743 } 1744 buffer->cur++; 1745 } 1746 maybe_warn_bidi_on_close (pfile, buffer->cur); 1747 } 1748 } 1749 1750 _cpp_process_line_notes (pfile, true); 1751 return orig_line != pfile->line_table->highest_line; 1752} 1753 1754/* Skips whitespace, saving the next non-whitespace character. */ 1755static void 1756skip_whitespace (cpp_reader *pfile, cppchar_t c) 1757{ 1758 cpp_buffer *buffer = pfile->buffer; 1759 bool saw_NUL = false; 1760 1761 do 1762 { 1763 /* Horizontal space always OK. */ 1764 if (c == ' ' || c == '\t') 1765 ; 1766 /* Just \f \v or \0 left. */ 1767 else if (c == '\0') 1768 saw_NUL = true; 1769 else if (pfile->state.in_directive && CPP_PEDANTIC (pfile)) 1770 cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, 1771 CPP_BUF_COL (buffer), 1772 "%s in preprocessing directive", 1773 c == '\f' ? "form feed" : "vertical tab"); 1774 1775 c = *buffer->cur++; 1776 } 1777 /* We only want non-vertical space, i.e. ' ' \t \f \v \0. */ 1778 while (is_nvspace (c)); 1779 1780 if (saw_NUL) 1781 { 1782 encoding_rich_location rich_loc (pfile); 1783 cpp_error_at (pfile, CPP_DL_WARNING, &rich_loc, 1784 "null character(s) ignored"); 1785 } 1786 1787 buffer->cur--; 1788} 1789 1790/* See if the characters of a number token are valid in a name (no 1791 '.', '+' or '-'). */ 1792static int 1793name_p (cpp_reader *pfile, const cpp_string *string) 1794{ 1795 unsigned int i; 1796 1797 for (i = 0; i < string->len; i++) 1798 if (!is_idchar (string->text[i])) 1799 return 0; 1800 1801 return 1; 1802} 1803 1804/* After parsing an identifier or other sequence, produce a warning about 1805 sequences not in NFC/NFKC. */ 1806static void 1807warn_about_normalization (cpp_reader *pfile, 1808 const cpp_token *token, 1809 const struct normalize_state *s) 1810{ 1811 if (CPP_OPTION (pfile, warn_normalize) < NORMALIZE_STATE_RESULT (s) 1812 && !pfile->state.skipping) 1813 { 1814 location_t loc = token->src_loc; 1815 1816 /* If possible, create a location range for the token. */ 1817 if (loc >= RESERVED_LOCATION_COUNT 1818 && token->type != CPP_EOF 1819 /* There must be no line notes to process. */ 1820 && (!(pfile->buffer->cur 1821 >= pfile->buffer->notes[pfile->buffer->cur_note].pos 1822 && !pfile->overlaid_buffer))) 1823 { 1824 source_range tok_range; 1825 tok_range.m_start = loc; 1826 tok_range.m_finish 1827 = linemap_position_for_column (pfile->line_table, 1828 CPP_BUF_COLUMN (pfile->buffer, 1829 pfile->buffer->cur)); 1830 loc = COMBINE_LOCATION_DATA (pfile->line_table, 1831 loc, tok_range, NULL); 1832 } 1833 1834 encoding_rich_location rich_loc (pfile, loc); 1835 1836 /* Make sure that the token is printed using UCNs, even 1837 if we'd otherwise happily print UTF-8. */ 1838 unsigned char *buf = XNEWVEC (unsigned char, cpp_token_len (token)); 1839 size_t sz; 1840 1841 sz = cpp_spell_token (pfile, token, buf, false) - buf; 1842 if (NORMALIZE_STATE_RESULT (s) == normalized_C) 1843 cpp_warning_at (pfile, CPP_W_NORMALIZE, &rich_loc, 1844 "`%.*s' is not in NFKC", (int) sz, buf); 1845 else if (CPP_OPTION (pfile, cplusplus)) 1846 cpp_pedwarning_at (pfile, CPP_W_NORMALIZE, &rich_loc, 1847 "`%.*s' is not in NFC", (int) sz, buf); 1848 else 1849 cpp_warning_at (pfile, CPP_W_NORMALIZE, &rich_loc, 1850 "`%.*s' is not in NFC", (int) sz, buf); 1851 free (buf); 1852 } 1853} 1854 1855static const cppchar_t utf8_signifier = 0xC0; 1856 1857/* Returns TRUE if the sequence starting at buffer->cur is valid in 1858 an identifier. FIRST is TRUE if this starts an identifier. */ 1859 1860static bool 1861forms_identifier_p (cpp_reader *pfile, int first, 1862 struct normalize_state *state) 1863{ 1864 cpp_buffer *buffer = pfile->buffer; 1865 const bool warn_bidi_p = pfile->warn_bidi_p (); 1866 1867 if (*buffer->cur == '$') 1868 { 1869 if (!CPP_OPTION (pfile, dollars_in_ident)) 1870 return false; 1871 1872 buffer->cur++; 1873 if (CPP_OPTION (pfile, warn_dollars) && !pfile->state.skipping) 1874 { 1875 CPP_OPTION (pfile, warn_dollars) = 0; 1876 cpp_error (pfile, CPP_DL_PEDWARN, "'$' in identifier or number"); 1877 } 1878 1879 return true; 1880 } 1881 1882 /* Is this a syntactically valid UCN or a valid UTF-8 char? */ 1883 if (CPP_OPTION (pfile, extended_identifiers)) 1884 { 1885 cppchar_t s; 1886 if (*buffer->cur >= utf8_signifier) 1887 { 1888 if (__builtin_expect (*buffer->cur == bidi::utf8_start, 0) 1889 && warn_bidi_p) 1890 { 1891 location_t loc; 1892 bidi::kind kind = get_bidi_utf8 (pfile, buffer->cur, &loc); 1893 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/false, loc); 1894 } 1895 if (_cpp_valid_utf8 (pfile, &buffer->cur, buffer->rlimit, 1 + !first, 1896 state, &s)) 1897 return true; 1898 } 1899 else if (*buffer->cur == '\\' 1900 && (buffer->cur[1] == 'u' || buffer->cur[1] == 'U')) 1901 { 1902 buffer->cur += 2; 1903 if (warn_bidi_p) 1904 { 1905 location_t loc; 1906 bidi::kind kind = get_bidi_ucn (pfile, 1907 buffer->cur, 1908 buffer->cur[-1] == 'U', 1909 &loc); 1910 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/true, loc); 1911 } 1912 if (_cpp_valid_ucn (pfile, &buffer->cur, buffer->rlimit, 1 + !first, 1913 state, &s, NULL, NULL)) 1914 return true; 1915 buffer->cur -= 2; 1916 } 1917 } 1918 1919 return false; 1920} 1921 1922/* Helper function to issue error about improper __VA_OPT__ use. */ 1923static void 1924maybe_va_opt_error (cpp_reader *pfile) 1925{ 1926 if (CPP_PEDANTIC (pfile) && !CPP_OPTION (pfile, va_opt)) 1927 { 1928 /* __VA_OPT__ should not be accepted at all, but allow it in 1929 system headers. */ 1930 if (!_cpp_in_system_header (pfile)) 1931 cpp_error (pfile, CPP_DL_PEDWARN, 1932 "__VA_OPT__ is not available until C++20"); 1933 } 1934 else if (!pfile->state.va_args_ok) 1935 { 1936 /* __VA_OPT__ should only appear in the replacement list of a 1937 variadic macro. */ 1938 cpp_error (pfile, CPP_DL_PEDWARN, 1939 "__VA_OPT__ can only appear in the expansion" 1940 " of a C++20 variadic macro"); 1941 } 1942} 1943 1944/* Helper function to get the cpp_hashnode of the identifier BASE. */ 1945static cpp_hashnode * 1946lex_identifier_intern (cpp_reader *pfile, const uchar *base) 1947{ 1948 cpp_hashnode *result; 1949 const uchar *cur; 1950 unsigned int len; 1951 unsigned int hash = HT_HASHSTEP (0, *base); 1952 1953 cur = base + 1; 1954 while (ISIDNUM (*cur)) 1955 { 1956 hash = HT_HASHSTEP (hash, *cur); 1957 cur++; 1958 } 1959 len = cur - base; 1960 hash = HT_HASHFINISH (hash, len); 1961 result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 1962 base, len, hash, HT_ALLOC)); 1963 1964 /* Rarely, identifiers require diagnostics when lexed. */ 1965 if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) 1966 && !pfile->state.skipping, 0)) 1967 { 1968 /* It is allowed to poison the same identifier twice. */ 1969 if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) 1970 cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", 1971 NODE_NAME (result)); 1972 1973 /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the 1974 replacement list of a variadic macro. */ 1975 if (result == pfile->spec_nodes.n__VA_ARGS__ 1976 && !pfile->state.va_args_ok) 1977 { 1978 if (CPP_OPTION (pfile, cplusplus)) 1979 cpp_error (pfile, CPP_DL_PEDWARN, 1980 "__VA_ARGS__ can only appear in the expansion" 1981 " of a C++11 variadic macro"); 1982 else 1983 cpp_error (pfile, CPP_DL_PEDWARN, 1984 "__VA_ARGS__ can only appear in the expansion" 1985 " of a C99 variadic macro"); 1986 } 1987 1988 if (result == pfile->spec_nodes.n__VA_OPT__) 1989 maybe_va_opt_error (pfile); 1990 1991 /* For -Wc++-compat, warn about use of C++ named operators. */ 1992 if (result->flags & NODE_WARN_OPERATOR) 1993 cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, 1994 "identifier \"%s\" is a special operator name in C++", 1995 NODE_NAME (result)); 1996 } 1997 1998 return result; 1999} 2000 2001/* Get the cpp_hashnode of an identifier specified by NAME in 2002 the current cpp_reader object. If none is found, NULL is returned. */ 2003cpp_hashnode * 2004_cpp_lex_identifier (cpp_reader *pfile, const char *name) 2005{ 2006 cpp_hashnode *result; 2007 result = lex_identifier_intern (pfile, (uchar *) name); 2008 return result; 2009} 2010 2011/* Lex an identifier starting at BUFFER->CUR - 1. */ 2012static cpp_hashnode * 2013lex_identifier (cpp_reader *pfile, const uchar *base, bool starts_ucn, 2014 struct normalize_state *nst, cpp_hashnode **spelling) 2015{ 2016 cpp_hashnode *result; 2017 const uchar *cur; 2018 unsigned int len; 2019 unsigned int hash = HT_HASHSTEP (0, *base); 2020 const bool warn_bidi_p = pfile->warn_bidi_p (); 2021 2022 cur = pfile->buffer->cur; 2023 if (! starts_ucn) 2024 { 2025 while (ISIDNUM (*cur)) 2026 { 2027 hash = HT_HASHSTEP (hash, *cur); 2028 cur++; 2029 } 2030 NORMALIZE_STATE_UPDATE_IDNUM (nst, *(cur - 1)); 2031 } 2032 pfile->buffer->cur = cur; 2033 if (starts_ucn || forms_identifier_p (pfile, false, nst)) 2034 { 2035 /* Slower version for identifiers containing UCNs 2036 or extended chars (including $). */ 2037 do { 2038 while (ISIDNUM (*pfile->buffer->cur)) 2039 { 2040 NORMALIZE_STATE_UPDATE_IDNUM (nst, *pfile->buffer->cur); 2041 pfile->buffer->cur++; 2042 } 2043 } while (forms_identifier_p (pfile, false, nst)); 2044 if (warn_bidi_p) 2045 maybe_warn_bidi_on_close (pfile, pfile->buffer->cur); 2046 result = _cpp_interpret_identifier (pfile, base, 2047 pfile->buffer->cur - base); 2048 *spelling = cpp_lookup (pfile, base, pfile->buffer->cur - base); 2049 } 2050 else 2051 { 2052 len = cur - base; 2053 hash = HT_HASHFINISH (hash, len); 2054 2055 result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 2056 base, len, hash, HT_ALLOC)); 2057 *spelling = result; 2058 } 2059 2060 /* Rarely, identifiers require diagnostics when lexed. */ 2061 if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) 2062 && !pfile->state.skipping, 0)) 2063 { 2064 /* It is allowed to poison the same identifier twice. */ 2065 if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) 2066 cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", 2067 NODE_NAME (result)); 2068 2069 /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the 2070 replacement list of a variadic macro. */ 2071 if (result == pfile->spec_nodes.n__VA_ARGS__ 2072 && !pfile->state.va_args_ok) 2073 { 2074 if (CPP_OPTION (pfile, cplusplus)) 2075 cpp_error (pfile, CPP_DL_PEDWARN, 2076 "__VA_ARGS__ can only appear in the expansion" 2077 " of a C++11 variadic macro"); 2078 else 2079 cpp_error (pfile, CPP_DL_PEDWARN, 2080 "__VA_ARGS__ can only appear in the expansion" 2081 " of a C99 variadic macro"); 2082 } 2083 2084 /* __VA_OPT__ should only appear in the replacement list of a 2085 variadic macro. */ 2086 if (result == pfile->spec_nodes.n__VA_OPT__) 2087 maybe_va_opt_error (pfile); 2088 2089 /* For -Wc++-compat, warn about use of C++ named operators. */ 2090 if (result->flags & NODE_WARN_OPERATOR) 2091 cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, 2092 "identifier \"%s\" is a special operator name in C++", 2093 NODE_NAME (result)); 2094 } 2095 2096 return result; 2097} 2098 2099/* Lex a number to NUMBER starting at BUFFER->CUR - 1. */ 2100static void 2101lex_number (cpp_reader *pfile, cpp_string *number, 2102 struct normalize_state *nst) 2103{ 2104 const uchar *cur; 2105 const uchar *base; 2106 uchar *dest; 2107 2108 base = pfile->buffer->cur - 1; 2109 do 2110 { 2111 const uchar *adj_digit_sep = NULL; 2112 cur = pfile->buffer->cur; 2113 2114 /* N.B. ISIDNUM does not include $. */ 2115 while (ISIDNUM (*cur) 2116 || (*cur == '.' && !DIGIT_SEP (cur[-1])) 2117 || DIGIT_SEP (*cur) 2118 || (VALID_SIGN (*cur, cur[-1]) && !DIGIT_SEP (cur[-2]))) 2119 { 2120 NORMALIZE_STATE_UPDATE_IDNUM (nst, *cur); 2121 /* Adjacent digit separators do not form part of the pp-number syntax. 2122 However, they can safely be diagnosed here as an error, since '' is 2123 not a valid preprocessing token. */ 2124 if (DIGIT_SEP (*cur) && DIGIT_SEP (cur[-1]) && !adj_digit_sep) 2125 adj_digit_sep = cur; 2126 cur++; 2127 } 2128 /* A number can't end with a digit separator. */ 2129 while (cur > pfile->buffer->cur && DIGIT_SEP (cur[-1])) 2130 --cur; 2131 if (adj_digit_sep && adj_digit_sep < cur) 2132 cpp_error (pfile, CPP_DL_ERROR, "adjacent digit separators"); 2133 2134 pfile->buffer->cur = cur; 2135 } 2136 while (forms_identifier_p (pfile, false, nst)); 2137 2138 number->len = cur - base; 2139 dest = _cpp_unaligned_alloc (pfile, number->len + 1); 2140 memcpy (dest, base, number->len); 2141 dest[number->len] = '\0'; 2142 number->text = dest; 2143} 2144 2145/* Create a token of type TYPE with a literal spelling. */ 2146static void 2147create_literal (cpp_reader *pfile, cpp_token *token, const uchar *base, 2148 unsigned int len, enum cpp_ttype type) 2149{ 2150 token->type = type; 2151 token->val.str.len = len; 2152 token->val.str.text = cpp_alloc_token_string (pfile, base, len); 2153} 2154 2155const uchar * 2156cpp_alloc_token_string (cpp_reader *pfile, 2157 const unsigned char *ptr, unsigned len) 2158{ 2159 uchar *dest = _cpp_unaligned_alloc (pfile, len + 1); 2160 2161 dest[len] = 0; 2162 memcpy (dest, ptr, len); 2163 return dest; 2164} 2165 2166/* A pair of raw buffer pointers. The currently open one is [1], the 2167 first one is [0]. Used for string literal lexing. */ 2168struct lit_accum { 2169 _cpp_buff *first; 2170 _cpp_buff *last; 2171 const uchar *rpos; 2172 size_t accum; 2173 2174 lit_accum () 2175 : first (NULL), last (NULL), rpos (0), accum (0) 2176 { 2177 } 2178 2179 void append (cpp_reader *, const uchar *, size_t); 2180 2181 void read_begin (cpp_reader *); 2182 bool reading_p () const 2183 { 2184 return rpos != NULL; 2185 } 2186 char read_char () 2187 { 2188 char c = *rpos++; 2189 if (rpos == BUFF_FRONT (last)) 2190 rpos = NULL; 2191 return c; 2192 } 2193}; 2194 2195/* Subroutine of lex_raw_string: Append LEN chars from BASE to the buffer 2196 sequence from *FIRST_BUFF_P to LAST_BUFF_P. */ 2197 2198void 2199lit_accum::append (cpp_reader *pfile, const uchar *base, size_t len) 2200{ 2201 if (!last) 2202 /* Starting. */ 2203 first = last = _cpp_get_buff (pfile, len); 2204 else if (len > BUFF_ROOM (last)) 2205 { 2206 /* There is insufficient room in the buffer. Copy what we can, 2207 and then either extend or create a new one. */ 2208 size_t room = BUFF_ROOM (last); 2209 memcpy (BUFF_FRONT (last), base, room); 2210 BUFF_FRONT (last) += room; 2211 base += room; 2212 len -= room; 2213 accum += room; 2214 2215 gcc_checking_assert (!rpos); 2216 2217 last = _cpp_append_extend_buff (pfile, last, len); 2218 } 2219 2220 memcpy (BUFF_FRONT (last), base, len); 2221 BUFF_FRONT (last) += len; 2222 accum += len; 2223} 2224 2225void 2226lit_accum::read_begin (cpp_reader *pfile) 2227{ 2228 /* We never accumulate more than 4 chars to read. */ 2229 if (BUFF_ROOM (last) < 4) 2230 2231 last = _cpp_append_extend_buff (pfile, last, 4); 2232 rpos = BUFF_FRONT (last); 2233} 2234 2235/* Returns true if a macro has been defined. 2236 This might not work if compile with -save-temps, 2237 or preprocess separately from compilation. */ 2238 2239static bool 2240is_macro(cpp_reader *pfile, const uchar *base) 2241{ 2242 const uchar *cur = base; 2243 if (! ISIDST (*cur)) 2244 return false; 2245 unsigned int hash = HT_HASHSTEP (0, *cur); 2246 ++cur; 2247 while (ISIDNUM (*cur)) 2248 { 2249 hash = HT_HASHSTEP (hash, *cur); 2250 ++cur; 2251 } 2252 hash = HT_HASHFINISH (hash, cur - base); 2253 2254 cpp_hashnode *result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, 2255 base, cur - base, hash, HT_NO_INSERT)); 2256 2257 return result && cpp_macro_p (result); 2258} 2259 2260/* Returns true if a literal suffix does not have the expected form 2261 and is defined as a macro. */ 2262 2263static bool 2264is_macro_not_literal_suffix(cpp_reader *pfile, const uchar *base) 2265{ 2266 /* User-defined literals outside of namespace std must start with a single 2267 underscore, so assume anything of that form really is a UDL suffix. 2268 We don't need to worry about UDLs defined inside namespace std because 2269 their names are reserved, so cannot be used as macro names in valid 2270 programs. */ 2271 if (base[0] == '_' && base[1] != '_') 2272 return false; 2273 return is_macro (pfile, base); 2274} 2275 2276/* Lexes a raw string. The stored string contains the spelling, 2277 including double quotes, delimiter string, '(' and ')', any leading 2278 'L', 'u', 'U' or 'u8' and 'R' modifier. The created token contains 2279 the type of the literal, or CPP_OTHER if it was not properly 2280 terminated. 2281 2282 BASE is the start of the token. Updates pfile->buffer->cur to just 2283 after the lexed string. 2284 2285 The spelling is NUL-terminated, but it is not guaranteed that this 2286 is the first NUL since embedded NULs are preserved. */ 2287 2288static void 2289lex_raw_string (cpp_reader *pfile, cpp_token *token, const uchar *base) 2290{ 2291 const uchar *pos = base; 2292 const bool warn_bidi_p = pfile->warn_bidi_p (); 2293 2294 /* 'tis a pity this information isn't passed down from the lexer's 2295 initial categorization of the token. */ 2296 enum cpp_ttype type = CPP_STRING; 2297 2298 if (*pos == 'L') 2299 { 2300 type = CPP_WSTRING; 2301 pos++; 2302 } 2303 else if (*pos == 'U') 2304 { 2305 type = CPP_STRING32; 2306 pos++; 2307 } 2308 else if (*pos == 'u') 2309 { 2310 if (pos[1] == '8') 2311 { 2312 type = CPP_UTF8STRING; 2313 pos++; 2314 } 2315 else 2316 type = CPP_STRING16; 2317 pos++; 2318 } 2319 2320 gcc_checking_assert (pos[0] == 'R' && pos[1] == '"'); 2321 pos += 2; 2322 2323 _cpp_line_note *note = &pfile->buffer->notes[pfile->buffer->cur_note]; 2324 2325 /* Skip notes before the ". */ 2326 while (note->pos < pos) 2327 ++note; 2328 2329 lit_accum accum; 2330 2331 uchar prefix[17]; 2332 unsigned prefix_len = 0; 2333 enum Phase 2334 { 2335 PHASE_PREFIX = -2, 2336 PHASE_NONE = -1, 2337 PHASE_SUFFIX = 0 2338 } phase = PHASE_PREFIX; 2339 2340 for (;;) 2341 { 2342 gcc_checking_assert (note->pos >= pos); 2343 2344 /* Undo any escaped newlines and trigraphs. */ 2345 if (!accum.reading_p () && note->pos == pos) 2346 switch (note->type) 2347 { 2348 case '\\': 2349 case ' ': 2350 /* Restore backslash followed by newline. */ 2351 accum.append (pfile, base, pos - base); 2352 base = pos; 2353 accum.read_begin (pfile); 2354 accum.append (pfile, UC"\\", 1); 2355 2356 after_backslash: 2357 if (note->type == ' ') 2358 /* GNU backslash whitespace newline extension. FIXME 2359 could be any sequence of non-vertical space. When we 2360 can properly restore any such sequence, we should 2361 mark this note as handled so _cpp_process_line_notes 2362 doesn't warn. */ 2363 accum.append (pfile, UC" ", 1); 2364 2365 accum.append (pfile, UC"\n", 1); 2366 note++; 2367 break; 2368 2369 case '\n': 2370 /* This can happen for ??/<NEWLINE> when trigraphs are not 2371 being interpretted. */ 2372 gcc_checking_assert (!CPP_OPTION (pfile, trigraphs)); 2373 note->type = 0; 2374 note++; 2375 break; 2376 2377 default: 2378 gcc_checking_assert (_cpp_trigraph_map[note->type]); 2379 2380 /* Don't warn about this trigraph in 2381 _cpp_process_line_notes, since trigraphs show up as 2382 trigraphs in raw strings. */ 2383 uchar type = note->type; 2384 note->type = 0; 2385 2386 if (CPP_OPTION (pfile, trigraphs)) 2387 { 2388 accum.append (pfile, base, pos - base); 2389 base = pos; 2390 accum.read_begin (pfile); 2391 accum.append (pfile, UC"??", 2); 2392 accum.append (pfile, &type, 1); 2393 2394 /* ??/ followed by newline gets two line notes, one for 2395 the trigraph and one for the backslash/newline. */ 2396 if (type == '/' && note[1].pos == pos) 2397 { 2398 note++; 2399 gcc_assert (note->type == '\\' || note->type == ' '); 2400 goto after_backslash; 2401 } 2402 /* Skip the replacement character. */ 2403 base = ++pos; 2404 } 2405 2406 note++; 2407 break; 2408 } 2409 2410 /* Now get a char to process. Either from an expanded note, or 2411 from the line buffer. */ 2412 bool read_note = accum.reading_p (); 2413 char c = read_note ? accum.read_char () : *pos++; 2414 2415 if (phase == PHASE_PREFIX) 2416 { 2417 if (c == '(') 2418 { 2419 /* Done. */ 2420 phase = PHASE_NONE; 2421 prefix[prefix_len++] = '"'; 2422 } 2423 else if (prefix_len < 16 2424 /* Prefix chars are any of the basic character set, 2425 [lex.charset] except for ' 2426 ()\\\t\v\f\n'. Optimized for a contiguous 2427 alphabet. */ 2428 /* Unlike a switch, this collapses down to one or 2429 two shift and bitmask operations on an ASCII 2430 system, with an outlier or two. */ 2431 && (('Z' - 'A' == 25 2432 ? ((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')) 2433 : ISIDST (c)) 2434 || (c >= '0' && c <= '9') 2435 || c == '_' || c == '{' || c == '}' 2436 || c == '[' || c == ']' || c == '#' 2437 || c == '<' || c == '>' || c == '%' 2438 || c == ':' || c == ';' || c == '.' || c == '?' 2439 || c == '*' || c == '+' || c == '-' || c == '/' 2440 || c == '^' || c == '&' || c == '|' || c == '~' 2441 || c == '!' || c == '=' || c == ',' 2442 || c == '"' || c == '\'')) 2443 prefix[prefix_len++] = c; 2444 else 2445 { 2446 /* Something is wrong. */ 2447 int col = CPP_BUF_COLUMN (pfile->buffer, pos) + read_note; 2448 if (prefix_len == 16) 2449 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 2450 col, "raw string delimiter longer " 2451 "than 16 characters"); 2452 else if (c == '\n') 2453 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 2454 col, "invalid new-line in raw " 2455 "string delimiter"); 2456 else 2457 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 2458 col, "invalid character '%c' in " 2459 "raw string delimiter", c); 2460 type = CPP_OTHER; 2461 phase = PHASE_NONE; 2462 /* Continue until we get a close quote, that's probably 2463 the best failure mode. */ 2464 prefix_len = 0; 2465 } 2466 if (c != '\n') 2467 continue; 2468 } 2469 2470 if (phase != PHASE_NONE) 2471 { 2472 if (prefix[phase] != c) 2473 phase = PHASE_NONE; 2474 else if (unsigned (phase + 1) == prefix_len) 2475 break; 2476 else 2477 { 2478 phase = Phase (phase + 1); 2479 continue; 2480 } 2481 } 2482 2483 if (!prefix_len && c == '"') 2484 /* Failure mode lexing. */ 2485 goto out; 2486 else if (prefix_len && c == ')') 2487 phase = PHASE_SUFFIX; 2488 else if (!read_note && c == '\n') 2489 { 2490 pos--; 2491 pfile->buffer->cur = pos; 2492 if (pfile->state.in_directive 2493 || (pfile->state.parsing_args 2494 && pfile->buffer->next_line >= pfile->buffer->rlimit)) 2495 { 2496 cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 0, 2497 "unterminated raw string"); 2498 type = CPP_OTHER; 2499 goto out; 2500 } 2501 2502 accum.append (pfile, base, pos - base + 1); 2503 _cpp_process_line_notes (pfile, false); 2504 2505 if (pfile->buffer->next_line < pfile->buffer->rlimit) 2506 CPP_INCREMENT_LINE (pfile, 0); 2507 pfile->buffer->need_line = true; 2508 2509 if (!_cpp_get_fresh_line (pfile)) 2510 { 2511 /* We ran out of file and failed to get a line. */ 2512 location_t src_loc = token->src_loc; 2513 token->type = CPP_EOF; 2514 /* Tell the compiler the line number of the EOF token. */ 2515 token->src_loc = pfile->line_table->highest_line; 2516 token->flags = BOL; 2517 if (accum.first) 2518 _cpp_release_buff (pfile, accum.first); 2519 cpp_error_with_line (pfile, CPP_DL_ERROR, src_loc, 0, 2520 "unterminated raw string"); 2521 /* Now pop the buffer that _cpp_get_fresh_line did not. */ 2522 _cpp_pop_buffer (pfile); 2523 return; 2524 } 2525 2526 pos = base = pfile->buffer->cur; 2527 note = &pfile->buffer->notes[pfile->buffer->cur_note]; 2528 } 2529 else if (__builtin_expect ((unsigned char) c == bidi::utf8_start, 0) 2530 && warn_bidi_p) 2531 { 2532 location_t loc; 2533 bidi::kind kind = get_bidi_utf8 (pfile, pos - 1, &loc); 2534 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/false, loc); 2535 } 2536 } 2537 2538 if (warn_bidi_p) 2539 maybe_warn_bidi_on_close (pfile, pos); 2540 2541 if (CPP_OPTION (pfile, user_literals)) 2542 { 2543 /* If a string format macro, say from inttypes.h, is placed touching 2544 a string literal it could be parsed as a C++11 user-defined string 2545 literal thus breaking the program. */ 2546 if (is_macro_not_literal_suffix (pfile, pos)) 2547 { 2548 /* Raise a warning, but do not consume subsequent tokens. */ 2549 if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping) 2550 cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, 2551 token->src_loc, 0, 2552 "invalid suffix on literal; C++11 requires " 2553 "a space between literal and string macro"); 2554 } 2555 /* Grab user defined literal suffix. */ 2556 else if (ISIDST (*pos)) 2557 { 2558 type = cpp_userdef_string_add_type (type); 2559 ++pos; 2560 2561 while (ISIDNUM (*pos)) 2562 ++pos; 2563 } 2564 } 2565 2566 out: 2567 pfile->buffer->cur = pos; 2568 if (!accum.accum) 2569 create_literal (pfile, token, base, pos - base, type); 2570 else 2571 { 2572 size_t extra_len = pos - base; 2573 uchar *dest = _cpp_unaligned_alloc (pfile, accum.accum + extra_len + 1); 2574 2575 token->type = type; 2576 token->val.str.len = accum.accum + extra_len; 2577 token->val.str.text = dest; 2578 for (_cpp_buff *buf = accum.first; buf; buf = buf->next) 2579 { 2580 size_t len = BUFF_FRONT (buf) - buf->base; 2581 memcpy (dest, buf->base, len); 2582 dest += len; 2583 } 2584 _cpp_release_buff (pfile, accum.first); 2585 memcpy (dest, base, extra_len); 2586 dest[extra_len] = '\0'; 2587 } 2588} 2589 2590/* Lexes a string, character constant, or angle-bracketed header file 2591 name. The stored string contains the spelling, including opening 2592 quote and any leading 'L', 'u', 'U' or 'u8' and optional 2593 'R' modifier. It returns the type of the literal, or CPP_OTHER 2594 if it was not properly terminated, or CPP_LESS for an unterminated 2595 header name which must be relexed as normal tokens. 2596 2597 The spelling is NUL-terminated, but it is not guaranteed that this 2598 is the first NUL since embedded NULs are preserved. */ 2599static void 2600lex_string (cpp_reader *pfile, cpp_token *token, const uchar *base) 2601{ 2602 bool saw_NUL = false; 2603 const uchar *cur; 2604 cppchar_t terminator; 2605 enum cpp_ttype type; 2606 2607 cur = base; 2608 terminator = *cur++; 2609 if (terminator == 'L' || terminator == 'U') 2610 terminator = *cur++; 2611 else if (terminator == 'u') 2612 { 2613 terminator = *cur++; 2614 if (terminator == '8') 2615 terminator = *cur++; 2616 } 2617 if (terminator == 'R') 2618 { 2619 lex_raw_string (pfile, token, base); 2620 return; 2621 } 2622 if (terminator == '"') 2623 type = (*base == 'L' ? CPP_WSTRING : 2624 *base == 'U' ? CPP_STRING32 : 2625 *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16) 2626 : CPP_STRING); 2627 else if (terminator == '\'') 2628 type = (*base == 'L' ? CPP_WCHAR : 2629 *base == 'U' ? CPP_CHAR32 : 2630 *base == 'u' ? (base[1] == '8' ? CPP_UTF8CHAR : CPP_CHAR16) 2631 : CPP_CHAR); 2632 else 2633 terminator = '>', type = CPP_HEADER_NAME; 2634 2635 const bool warn_bidi_p = pfile->warn_bidi_p (); 2636 for (;;) 2637 { 2638 cppchar_t c = *cur++; 2639 2640 /* In #include-style directives, terminators are not escapable. */ 2641 if (c == '\\' && !pfile->state.angled_headers && *cur != '\n') 2642 { 2643 if ((cur[0] == 'u' || cur[0] == 'U') && warn_bidi_p) 2644 { 2645 location_t loc; 2646 bidi::kind kind = get_bidi_ucn (pfile, cur + 1, cur[0] == 'U', 2647 &loc); 2648 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/true, loc); 2649 } 2650 cur++; 2651 } 2652 else if (c == terminator) 2653 { 2654 if (warn_bidi_p) 2655 maybe_warn_bidi_on_close (pfile, cur - 1); 2656 break; 2657 } 2658 else if (c == '\n') 2659 { 2660 cur--; 2661 /* Unmatched quotes always yield undefined behavior, but 2662 greedy lexing means that what appears to be an unterminated 2663 header name may actually be a legitimate sequence of tokens. */ 2664 if (terminator == '>') 2665 { 2666 token->type = CPP_LESS; 2667 return; 2668 } 2669 type = CPP_OTHER; 2670 break; 2671 } 2672 else if (c == '\0') 2673 saw_NUL = true; 2674 else if (__builtin_expect (c == bidi::utf8_start, 0) && warn_bidi_p) 2675 { 2676 location_t loc; 2677 bidi::kind kind = get_bidi_utf8 (pfile, cur - 1, &loc); 2678 maybe_warn_bidi_on_char (pfile, kind, /*ucn_p=*/false, loc); 2679 } 2680 } 2681 2682 if (saw_NUL && !pfile->state.skipping) 2683 cpp_error (pfile, CPP_DL_WARNING, 2684 "null character(s) preserved in literal"); 2685 2686 if (type == CPP_OTHER && CPP_OPTION (pfile, lang) != CLK_ASM) 2687 cpp_error (pfile, CPP_DL_PEDWARN, "missing terminating %c character", 2688 (int) terminator); 2689 2690 if (CPP_OPTION (pfile, user_literals)) 2691 { 2692 /* If a string format macro, say from inttypes.h, is placed touching 2693 a string literal it could be parsed as a C++11 user-defined string 2694 literal thus breaking the program. */ 2695 if (is_macro_not_literal_suffix (pfile, cur)) 2696 { 2697 /* Raise a warning, but do not consume subsequent tokens. */ 2698 if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping) 2699 cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, 2700 token->src_loc, 0, 2701 "invalid suffix on literal; C++11 requires " 2702 "a space between literal and string macro"); 2703 } 2704 /* Grab user defined literal suffix. */ 2705 else if (ISIDST (*cur)) 2706 { 2707 type = cpp_userdef_char_add_type (type); 2708 type = cpp_userdef_string_add_type (type); 2709 ++cur; 2710 2711 while (ISIDNUM (*cur)) 2712 ++cur; 2713 } 2714 } 2715 else if (CPP_OPTION (pfile, cpp_warn_cxx11_compat) 2716 && is_macro (pfile, cur) 2717 && !pfile->state.skipping) 2718 cpp_warning_with_line (pfile, CPP_W_CXX11_COMPAT, 2719 token->src_loc, 0, "C++11 requires a space " 2720 "between string literal and macro"); 2721 2722 pfile->buffer->cur = cur; 2723 create_literal (pfile, token, base, cur - base, type); 2724} 2725 2726/* Return the comment table. The client may not make any assumption 2727 about the ordering of the table. */ 2728cpp_comment_table * 2729cpp_get_comments (cpp_reader *pfile) 2730{ 2731 return &pfile->comments; 2732} 2733 2734/* Append a comment to the end of the comment table. */ 2735static void 2736store_comment (cpp_reader *pfile, cpp_token *token) 2737{ 2738 int len; 2739 2740 if (pfile->comments.allocated == 0) 2741 { 2742 pfile->comments.allocated = 256; 2743 pfile->comments.entries = (cpp_comment *) xmalloc 2744 (pfile->comments.allocated * sizeof (cpp_comment)); 2745 } 2746 2747 if (pfile->comments.count == pfile->comments.allocated) 2748 { 2749 pfile->comments.allocated *= 2; 2750 pfile->comments.entries = (cpp_comment *) xrealloc 2751 (pfile->comments.entries, 2752 pfile->comments.allocated * sizeof (cpp_comment)); 2753 } 2754 2755 len = token->val.str.len; 2756 2757 /* Copy comment. Note, token may not be NULL terminated. */ 2758 pfile->comments.entries[pfile->comments.count].comment = 2759 (char *) xmalloc (sizeof (char) * (len + 1)); 2760 memcpy (pfile->comments.entries[pfile->comments.count].comment, 2761 token->val.str.text, len); 2762 pfile->comments.entries[pfile->comments.count].comment[len] = '\0'; 2763 2764 /* Set source location. */ 2765 pfile->comments.entries[pfile->comments.count].sloc = token->src_loc; 2766 2767 /* Increment the count of entries in the comment table. */ 2768 pfile->comments.count++; 2769} 2770 2771/* The stored comment includes the comment start and any terminator. */ 2772static void 2773save_comment (cpp_reader *pfile, cpp_token *token, const unsigned char *from, 2774 cppchar_t type) 2775{ 2776 unsigned char *buffer; 2777 unsigned int len, clen, i; 2778 int convert_to_c = (pfile->state.in_directive || pfile->state.parsing_args) 2779 && type == '/'; 2780 2781 len = pfile->buffer->cur - from + 1; /* + 1 for the initial '/'. */ 2782 2783 /* C++ comments probably (not definitely) have moved past a new 2784 line, which we don't want to save in the comment. */ 2785 if (is_vspace (pfile->buffer->cur[-1])) 2786 len--; 2787 2788 /* If we are currently in a directive or in argument parsing, then 2789 we need to store all C++ comments as C comments internally, and 2790 so we need to allocate a little extra space in that case. 2791 2792 Note that the only time we encounter a directive here is 2793 when we are saving comments in a "#define". */ 2794 clen = convert_to_c ? len + 2 : len; 2795 2796 buffer = _cpp_unaligned_alloc (pfile, clen); 2797 2798 token->type = CPP_COMMENT; 2799 token->val.str.len = clen; 2800 token->val.str.text = buffer; 2801 2802 buffer[0] = '/'; 2803 memcpy (buffer + 1, from, len - 1); 2804 2805 /* Finish conversion to a C comment, if necessary. */ 2806 if (convert_to_c) 2807 { 2808 buffer[1] = '*'; 2809 buffer[clen - 2] = '*'; 2810 buffer[clen - 1] = '/'; 2811 /* As there can be in a C++ comments illegal sequences for C comments 2812 we need to filter them out. */ 2813 for (i = 2; i < (clen - 2); i++) 2814 if (buffer[i] == '/' && (buffer[i - 1] == '*' || buffer[i + 1] == '*')) 2815 buffer[i] = '|'; 2816 } 2817 2818 /* Finally store this comment for use by clients of libcpp. */ 2819 store_comment (pfile, token); 2820} 2821 2822/* Returns true if comment at COMMENT_START is a recognized FALLTHROUGH 2823 comment. */ 2824 2825static bool 2826fallthrough_comment_p (cpp_reader *pfile, const unsigned char *comment_start) 2827{ 2828 const unsigned char *from = comment_start + 1; 2829 2830 switch (CPP_OPTION (pfile, cpp_warn_implicit_fallthrough)) 2831 { 2832 /* For both -Wimplicit-fallthrough=0 and -Wimplicit-fallthrough=5 we 2833 don't recognize any comments. The latter only checks attributes, 2834 the former doesn't warn. */ 2835 case 0: 2836 default: 2837 return false; 2838 /* -Wimplicit-fallthrough=1 considers any comment, no matter what 2839 content it has. */ 2840 case 1: 2841 return true; 2842 case 2: 2843 /* -Wimplicit-fallthrough=2 looks for (case insensitive) 2844 .*falls?[ \t-]*thr(u|ough).* regex. */ 2845 for (; (size_t) (pfile->buffer->cur - from) >= sizeof "fallthru" - 1; 2846 from++) 2847 { 2848 /* Is there anything like strpbrk with upper boundary, or 2849 memchr looking for 2 characters rather than just one? */ 2850 if (from[0] != 'f' && from[0] != 'F') 2851 continue; 2852 if (from[1] != 'a' && from[1] != 'A') 2853 continue; 2854 if (from[2] != 'l' && from[2] != 'L') 2855 continue; 2856 if (from[3] != 'l' && from[3] != 'L') 2857 continue; 2858 from += sizeof "fall" - 1; 2859 if (from[0] == 's' || from[0] == 'S') 2860 from++; 2861 while (*from == ' ' || *from == '\t' || *from == '-') 2862 from++; 2863 if (from[0] != 't' && from[0] != 'T') 2864 continue; 2865 if (from[1] != 'h' && from[1] != 'H') 2866 continue; 2867 if (from[2] != 'r' && from[2] != 'R') 2868 continue; 2869 if (from[3] == 'u' || from[3] == 'U') 2870 return true; 2871 if (from[3] != 'o' && from[3] != 'O') 2872 continue; 2873 if (from[4] != 'u' && from[4] != 'U') 2874 continue; 2875 if (from[5] != 'g' && from[5] != 'G') 2876 continue; 2877 if (from[6] != 'h' && from[6] != 'H') 2878 continue; 2879 return true; 2880 } 2881 return false; 2882 case 3: 2883 case 4: 2884 break; 2885 } 2886 2887 /* Whole comment contents: 2888 -fallthrough 2889 @fallthrough@ 2890 */ 2891 if (*from == '-' || *from == '@') 2892 { 2893 size_t len = sizeof "fallthrough" - 1; 2894 if ((size_t) (pfile->buffer->cur - from - 1) < len) 2895 return false; 2896 if (memcmp (from + 1, "fallthrough", len)) 2897 return false; 2898 if (*from == '@') 2899 { 2900 if (from[len + 1] != '@') 2901 return false; 2902 len++; 2903 } 2904 from += 1 + len; 2905 } 2906 /* Whole comment contents (regex): 2907 lint -fallthrough[ \t]* 2908 */ 2909 else if (*from == 'l') 2910 { 2911 size_t len = sizeof "int -fallthrough" - 1; 2912 if ((size_t) (pfile->buffer->cur - from - 1) < len) 2913 return false; 2914 if (memcmp (from + 1, "int -fallthrough", len)) 2915 return false; 2916 from += 1 + len; 2917 while (*from == ' ' || *from == '\t') 2918 from++; 2919 } 2920 /* Whole comment contents (regex): 2921 [ \t]*FALLTHR(U|OUGH)[ \t]* 2922 */ 2923 else if (CPP_OPTION (pfile, cpp_warn_implicit_fallthrough) == 4) 2924 { 2925 while (*from == ' ' || *from == '\t') 2926 from++; 2927 if ((size_t) (pfile->buffer->cur - from) < sizeof "FALLTHRU" - 1) 2928 return false; 2929 if (memcmp (from, "FALLTHR", sizeof "FALLTHR" - 1)) 2930 return false; 2931 from += sizeof "FALLTHR" - 1; 2932 if (*from == 'U') 2933 from++; 2934 else if ((size_t) (pfile->buffer->cur - from) < sizeof "OUGH" - 1) 2935 return false; 2936 else if (memcmp (from, "OUGH", sizeof "OUGH" - 1)) 2937 return false; 2938 else 2939 from += sizeof "OUGH" - 1; 2940 while (*from == ' ' || *from == '\t') 2941 from++; 2942 } 2943 /* Whole comment contents (regex): 2944 [ \t.!]*(ELSE,? |INTENTIONAL(LY)? )?FALL(S | |-)?THR(OUGH|U)[ \t.!]*(-[^\n\r]*)? 2945 [ \t.!]*(Else,? |Intentional(ly)? )?Fall((s | |-)[Tt]|t)hr(ough|u)[ \t.!]*(-[^\n\r]*)? 2946 [ \t.!]*([Ee]lse,? |[Ii]ntentional(ly)? )?fall(s | |-)?thr(ough|u)[ \t.!]*(-[^\n\r]*)? 2947 */ 2948 else 2949 { 2950 while (*from == ' ' || *from == '\t' || *from == '.' || *from == '!') 2951 from++; 2952 unsigned char f = *from; 2953 bool all_upper = false; 2954 if (f == 'E' || f == 'e') 2955 { 2956 if ((size_t) (pfile->buffer->cur - from) 2957 < sizeof "else fallthru" - 1) 2958 return false; 2959 if (f == 'E' && memcmp (from + 1, "LSE", sizeof "LSE" - 1) == 0) 2960 all_upper = true; 2961 else if (memcmp (from + 1, "lse", sizeof "lse" - 1)) 2962 return false; 2963 from += sizeof "else" - 1; 2964 if (*from == ',') 2965 from++; 2966 if (*from != ' ') 2967 return false; 2968 from++; 2969 if (all_upper && *from == 'f') 2970 return false; 2971 if (f == 'e' && *from == 'F') 2972 return false; 2973 f = *from; 2974 } 2975 else if (f == 'I' || f == 'i') 2976 { 2977 if ((size_t) (pfile->buffer->cur - from) 2978 < sizeof "intentional fallthru" - 1) 2979 return false; 2980 if (f == 'I' && memcmp (from + 1, "NTENTIONAL", 2981 sizeof "NTENTIONAL" - 1) == 0) 2982 all_upper = true; 2983 else if (memcmp (from + 1, "ntentional", 2984 sizeof "ntentional" - 1)) 2985 return false; 2986 from += sizeof "intentional" - 1; 2987 if (*from == ' ') 2988 { 2989 from++; 2990 if (all_upper && *from == 'f') 2991 return false; 2992 } 2993 else if (all_upper) 2994 { 2995 if (memcmp (from, "LY F", sizeof "LY F" - 1)) 2996 return false; 2997 from += sizeof "LY " - 1; 2998 } 2999 else 3000 { 3001 if (memcmp (from, "ly ", sizeof "ly " - 1)) 3002 return false; 3003 from += sizeof "ly " - 1; 3004 } 3005 if (f == 'i' && *from == 'F') 3006 return false; 3007 f = *from; 3008 } 3009 if (f != 'F' && f != 'f') 3010 return false; 3011 if ((size_t) (pfile->buffer->cur - from) < sizeof "fallthru" - 1) 3012 return false; 3013 if (f == 'F' && memcmp (from + 1, "ALL", sizeof "ALL" - 1) == 0) 3014 all_upper = true; 3015 else if (all_upper) 3016 return false; 3017 else if (memcmp (from + 1, "all", sizeof "all" - 1)) 3018 return false; 3019 from += sizeof "fall" - 1; 3020 if (*from == (all_upper ? 'S' : 's') && from[1] == ' ') 3021 from += 2; 3022 else if (*from == ' ' || *from == '-') 3023 from++; 3024 else if (*from != (all_upper ? 'T' : 't')) 3025 return false; 3026 if ((f == 'f' || *from != 'T') && (all_upper || *from != 't')) 3027 return false; 3028 if ((size_t) (pfile->buffer->cur - from) < sizeof "thru" - 1) 3029 return false; 3030 if (memcmp (from + 1, all_upper ? "HRU" : "hru", sizeof "hru" - 1)) 3031 { 3032 if ((size_t) (pfile->buffer->cur - from) < sizeof "through" - 1) 3033 return false; 3034 if (memcmp (from + 1, all_upper ? "HROUGH" : "hrough", 3035 sizeof "hrough" - 1)) 3036 return false; 3037 from += sizeof "through" - 1; 3038 } 3039 else 3040 from += sizeof "thru" - 1; 3041 while (*from == ' ' || *from == '\t' || *from == '.' || *from == '!') 3042 from++; 3043 if (*from == '-') 3044 { 3045 from++; 3046 if (*comment_start == '*') 3047 { 3048 do 3049 { 3050 while (*from && *from != '*' 3051 && *from != '\n' && *from != '\r') 3052 from++; 3053 if (*from != '*' || from[1] == '/') 3054 break; 3055 from++; 3056 } 3057 while (1); 3058 } 3059 else 3060 while (*from && *from != '\n' && *from != '\r') 3061 from++; 3062 } 3063 } 3064 /* C block comment. */ 3065 if (*comment_start == '*') 3066 { 3067 if (*from != '*' || from[1] != '/') 3068 return false; 3069 } 3070 /* C++ line comment. */ 3071 else if (*from != '\n') 3072 return false; 3073 3074 return true; 3075} 3076 3077/* Allocate COUNT tokens for RUN. */ 3078void 3079_cpp_init_tokenrun (tokenrun *run, unsigned int count) 3080{ 3081 run->base = XNEWVEC (cpp_token, count); 3082 run->limit = run->base + count; 3083 run->next = NULL; 3084} 3085 3086/* Returns the next tokenrun, or creates one if there is none. */ 3087static tokenrun * 3088next_tokenrun (tokenrun *run) 3089{ 3090 if (run->next == NULL) 3091 { 3092 run->next = XNEW (tokenrun); 3093 run->next->prev = run; 3094 _cpp_init_tokenrun (run->next, 250); 3095 } 3096 3097 return run->next; 3098} 3099 3100/* Return the number of not yet processed token in a given 3101 context. */ 3102int 3103_cpp_remaining_tokens_num_in_context (cpp_context *context) 3104{ 3105 if (context->tokens_kind == TOKENS_KIND_DIRECT) 3106 return (LAST (context).token - FIRST (context).token); 3107 else if (context->tokens_kind == TOKENS_KIND_INDIRECT 3108 || context->tokens_kind == TOKENS_KIND_EXTENDED) 3109 return (LAST (context).ptoken - FIRST (context).ptoken); 3110 else 3111 abort (); 3112} 3113 3114/* Returns the token present at index INDEX in a given context. If 3115 INDEX is zero, the next token to be processed is returned. */ 3116static const cpp_token* 3117_cpp_token_from_context_at (cpp_context *context, int index) 3118{ 3119 if (context->tokens_kind == TOKENS_KIND_DIRECT) 3120 return &(FIRST (context).token[index]); 3121 else if (context->tokens_kind == TOKENS_KIND_INDIRECT 3122 || context->tokens_kind == TOKENS_KIND_EXTENDED) 3123 return FIRST (context).ptoken[index]; 3124 else 3125 abort (); 3126} 3127 3128/* Look ahead in the input stream. */ 3129const cpp_token * 3130cpp_peek_token (cpp_reader *pfile, int index) 3131{ 3132 cpp_context *context = pfile->context; 3133 const cpp_token *peektok; 3134 int count; 3135 3136 /* First, scan through any pending cpp_context objects. */ 3137 while (context->prev) 3138 { 3139 ptrdiff_t sz = _cpp_remaining_tokens_num_in_context (context); 3140 3141 if (index < (int) sz) 3142 return _cpp_token_from_context_at (context, index); 3143 index -= (int) sz; 3144 context = context->prev; 3145 } 3146 3147 /* We will have to read some new tokens after all (and do so 3148 without invalidating preceding tokens). */ 3149 count = index; 3150 pfile->keep_tokens++; 3151 3152 /* For peeked tokens temporarily disable line_change reporting, 3153 until the tokens are parsed for real. */ 3154 void (*line_change) (cpp_reader *, const cpp_token *, int) 3155 = pfile->cb.line_change; 3156 pfile->cb.line_change = NULL; 3157 3158 do 3159 { 3160 peektok = _cpp_lex_token (pfile); 3161 if (peektok->type == CPP_EOF) 3162 { 3163 index--; 3164 break; 3165 } 3166 else if (peektok->type == CPP_PRAGMA) 3167 { 3168 /* Don't peek past a pragma. */ 3169 if (peektok == &pfile->directive_result) 3170 /* Save the pragma in the buffer. */ 3171 *pfile->cur_token++ = *peektok; 3172 index--; 3173 break; 3174 } 3175 } 3176 while (index--); 3177 3178 _cpp_backup_tokens_direct (pfile, count - index); 3179 pfile->keep_tokens--; 3180 pfile->cb.line_change = line_change; 3181 3182 return peektok; 3183} 3184 3185/* Allocate a single token that is invalidated at the same time as the 3186 rest of the tokens on the line. Has its line and col set to the 3187 same as the last lexed token, so that diagnostics appear in the 3188 right place. */ 3189cpp_token * 3190_cpp_temp_token (cpp_reader *pfile) 3191{ 3192 cpp_token *old, *result; 3193 ptrdiff_t sz = pfile->cur_run->limit - pfile->cur_token; 3194 ptrdiff_t la = (ptrdiff_t) pfile->lookaheads; 3195 3196 old = pfile->cur_token - 1; 3197 /* Any pre-existing lookaheads must not be clobbered. */ 3198 if (la) 3199 { 3200 if (sz <= la) 3201 { 3202 tokenrun *next = next_tokenrun (pfile->cur_run); 3203 3204 if (sz < la) 3205 memmove (next->base + 1, next->base, 3206 (la - sz) * sizeof (cpp_token)); 3207 3208 next->base[0] = pfile->cur_run->limit[-1]; 3209 } 3210 3211 if (sz > 1) 3212 memmove (pfile->cur_token + 1, pfile->cur_token, 3213 MIN (la, sz - 1) * sizeof (cpp_token)); 3214 } 3215 3216 if (!sz && pfile->cur_token == pfile->cur_run->limit) 3217 { 3218 pfile->cur_run = next_tokenrun (pfile->cur_run); 3219 pfile->cur_token = pfile->cur_run->base; 3220 } 3221 3222 result = pfile->cur_token++; 3223 result->src_loc = old->src_loc; 3224 return result; 3225} 3226 3227/* We're at the beginning of a logical line (so not in 3228 directives-mode) and RESULT is a CPP_NAME with NODE_MODULE set. See 3229 if we should enter deferred_pragma mode to tokenize the rest of the 3230 line as a module control-line. */ 3231 3232static void 3233cpp_maybe_module_directive (cpp_reader *pfile, cpp_token *result) 3234{ 3235 unsigned backup = 0; /* Tokens we peeked. */ 3236 cpp_hashnode *node = result->val.node.node; 3237 cpp_token *peek = result; 3238 cpp_token *keyword = peek; 3239 cpp_hashnode *(&n_modules)[spec_nodes::M_HWM][2] = pfile->spec_nodes.n_modules; 3240 int header_count = 0; 3241 3242 /* Make sure the incoming state is as we expect it. This way we 3243 can restore it using constants. */ 3244 gcc_checking_assert (!pfile->state.in_deferred_pragma 3245 && !pfile->state.skipping 3246 && !pfile->state.parsing_args 3247 && !pfile->state.angled_headers 3248 && (pfile->state.save_comments 3249 == !CPP_OPTION (pfile, discard_comments))); 3250 3251 /* Enter directives mode sufficiently for peeking. We don't have 3252 to actually set in_directive. */ 3253 pfile->state.in_deferred_pragma = true; 3254 3255 /* These two fields are needed to process tokenization in deferred 3256 pragma mode. They are not used outside deferred pragma mode or 3257 directives mode. */ 3258 pfile->state.pragma_allow_expansion = true; 3259 pfile->directive_line = result->src_loc; 3260 3261 /* Saving comments is incompatible with directives mode. */ 3262 pfile->state.save_comments = 0; 3263 3264 if (node == n_modules[spec_nodes::M_EXPORT][0]) 3265 { 3266 peek = _cpp_lex_direct (pfile); 3267 keyword = peek; 3268 backup++; 3269 if (keyword->type != CPP_NAME) 3270 goto not_module; 3271 node = keyword->val.node.node; 3272 if (!(node->flags & NODE_MODULE)) 3273 goto not_module; 3274 } 3275 3276 if (node == n_modules[spec_nodes::M__IMPORT][0]) 3277 /* __import */ 3278 header_count = backup + 2 + 16; 3279 else if (node == n_modules[spec_nodes::M_IMPORT][0]) 3280 /* import */ 3281 header_count = backup + 2 + (CPP_OPTION (pfile, preprocessed) ? 16 : 0); 3282 else if (node == n_modules[spec_nodes::M_MODULE][0]) 3283 ; /* module */ 3284 else 3285 goto not_module; 3286 3287 /* We've seen [export] {module|import|__import}. Check the next token. */ 3288 if (header_count) 3289 /* After '{,__}import' a header name may appear. */ 3290 pfile->state.angled_headers = true; 3291 peek = _cpp_lex_direct (pfile); 3292 backup++; 3293 3294 /* ... import followed by identifier, ':', '<' or 3295 header-name preprocessing tokens, or module 3296 followed by cpp-identifier, ':' or ';' preprocessing 3297 tokens. C++ keywords are not yet relevant. */ 3298 if (peek->type == CPP_NAME 3299 || peek->type == CPP_COLON 3300 || (header_count 3301 ? (peek->type == CPP_LESS 3302 || (peek->type == CPP_STRING && peek->val.str.text[0] != 'R') 3303 || peek->type == CPP_HEADER_NAME) 3304 : peek->type == CPP_SEMICOLON)) 3305 { 3306 pfile->state.pragma_allow_expansion = !CPP_OPTION (pfile, preprocessed); 3307 if (!pfile->state.pragma_allow_expansion) 3308 pfile->state.prevent_expansion++; 3309 3310 if (!header_count && linemap_included_from 3311 (LINEMAPS_LAST_ORDINARY_MAP (pfile->line_table))) 3312 cpp_error_with_line (pfile, CPP_DL_ERROR, keyword->src_loc, 0, 3313 "module control-line cannot be in included file"); 3314 3315 /* The first one or two tokens cannot be macro names. */ 3316 for (int ix = backup; ix--;) 3317 { 3318 cpp_token *tok = ix ? keyword : result; 3319 cpp_hashnode *node = tok->val.node.node; 3320 3321 /* Don't attempt to expand the token. */ 3322 tok->flags |= NO_EXPAND; 3323 if (_cpp_defined_macro_p (node) 3324 && _cpp_maybe_notify_macro_use (pfile, node, tok->src_loc) 3325 && !cpp_fun_like_macro_p (node)) 3326 cpp_error_with_line (pfile, CPP_DL_ERROR, tok->src_loc, 0, 3327 "module control-line \"%s\" cannot be" 3328 " an object-like macro", 3329 NODE_NAME (node)); 3330 } 3331 3332 /* Map to underbar variants. */ 3333 keyword->val.node.node = n_modules[header_count 3334 ? spec_nodes::M_IMPORT 3335 : spec_nodes::M_MODULE][1]; 3336 if (backup != 1) 3337 result->val.node.node = n_modules[spec_nodes::M_EXPORT][1]; 3338 3339 /* Maybe tell the tokenizer we expect a header-name down the 3340 road. */ 3341 pfile->state.directive_file_token = header_count; 3342 } 3343 else 3344 { 3345 not_module: 3346 /* Drop out of directive mode. */ 3347 /* We aaserted save_comments had this value upon entry. */ 3348 pfile->state.save_comments 3349 = !CPP_OPTION (pfile, discard_comments); 3350 pfile->state.in_deferred_pragma = false; 3351 /* Do not let this remain on. */ 3352 pfile->state.angled_headers = false; 3353 } 3354 3355 /* In either case we want to backup the peeked tokens. */ 3356 if (backup) 3357 { 3358 /* If we saw EOL, we should drop it, because this isn't a module 3359 control-line after all. */ 3360 bool eol = peek->type == CPP_PRAGMA_EOL; 3361 if (!eol || backup > 1) 3362 { 3363 /* Put put the peeked tokens back */ 3364 _cpp_backup_tokens_direct (pfile, backup); 3365 /* But if the last one was an EOL, forget it. */ 3366 if (eol) 3367 pfile->lookaheads--; 3368 } 3369 } 3370} 3371 3372/* Lex a token into RESULT (external interface). Takes care of issues 3373 like directive handling, token lookahead, multiple include 3374 optimization and skipping. */ 3375const cpp_token * 3376_cpp_lex_token (cpp_reader *pfile) 3377{ 3378 cpp_token *result; 3379 3380 for (;;) 3381 { 3382 if (pfile->cur_token == pfile->cur_run->limit) 3383 { 3384 pfile->cur_run = next_tokenrun (pfile->cur_run); 3385 pfile->cur_token = pfile->cur_run->base; 3386 } 3387 /* We assume that the current token is somewhere in the current 3388 run. */ 3389 if (pfile->cur_token < pfile->cur_run->base 3390 || pfile->cur_token >= pfile->cur_run->limit) 3391 abort (); 3392 3393 if (pfile->lookaheads) 3394 { 3395 pfile->lookaheads--; 3396 result = pfile->cur_token++; 3397 } 3398 else 3399 result = _cpp_lex_direct (pfile); 3400 3401 if (result->flags & BOL) 3402 { 3403 /* Is this a directive. If _cpp_handle_directive returns 3404 false, it is an assembler #. */ 3405 if (result->type == CPP_HASH 3406 /* 6.10.3 p 11: Directives in a list of macro arguments 3407 gives undefined behavior. This implementation 3408 handles the directive as normal. */ 3409 && pfile->state.parsing_args != 1) 3410 { 3411 if (_cpp_handle_directive (pfile, result->flags & PREV_WHITE)) 3412 { 3413 if (pfile->directive_result.type == CPP_PADDING) 3414 continue; 3415 result = &pfile->directive_result; 3416 } 3417 } 3418 else if (pfile->state.in_deferred_pragma) 3419 result = &pfile->directive_result; 3420 else if (result->type == CPP_NAME 3421 && (result->val.node.node->flags & NODE_MODULE) 3422 && !pfile->state.skipping 3423 /* Unlike regular directives, we do not deal with 3424 tokenizing module directives as macro arguments. 3425 That's not permitted. */ 3426 && !pfile->state.parsing_args) 3427 { 3428 /* P1857. Before macro expansion, At start of logical 3429 line ... */ 3430 /* We don't have to consider lookaheads at this point. */ 3431 gcc_checking_assert (!pfile->lookaheads); 3432 3433 cpp_maybe_module_directive (pfile, result); 3434 } 3435 3436 if (pfile->cb.line_change && !pfile->state.skipping) 3437 pfile->cb.line_change (pfile, result, pfile->state.parsing_args); 3438 } 3439 3440 /* We don't skip tokens in directives. */ 3441 if (pfile->state.in_directive || pfile->state.in_deferred_pragma) 3442 break; 3443 3444 /* Outside a directive, invalidate controlling macros. At file 3445 EOF, _cpp_lex_direct takes care of popping the buffer, so we never 3446 get here and MI optimization works. */ 3447 pfile->mi_valid = false; 3448 3449 if (!pfile->state.skipping || result->type == CPP_EOF) 3450 break; 3451 } 3452 3453 return result; 3454} 3455 3456/* Returns true if a fresh line has been loaded. */ 3457bool 3458_cpp_get_fresh_line (cpp_reader *pfile) 3459{ 3460 /* We can't get a new line until we leave the current directive. */ 3461 if (pfile->state.in_directive) 3462 return false; 3463 3464 for (;;) 3465 { 3466 cpp_buffer *buffer = pfile->buffer; 3467 3468 if (!buffer->need_line) 3469 return true; 3470 3471 if (buffer->next_line < buffer->rlimit) 3472 { 3473 _cpp_clean_line (pfile); 3474 return true; 3475 } 3476 3477 /* First, get out of parsing arguments state. */ 3478 if (pfile->state.parsing_args) 3479 return false; 3480 3481 /* End of buffer. Non-empty files should end in a newline. */ 3482 if (buffer->buf != buffer->rlimit 3483 && buffer->next_line > buffer->rlimit 3484 && !buffer->from_stage3) 3485 { 3486 /* Clip to buffer size. */ 3487 buffer->next_line = buffer->rlimit; 3488 } 3489 3490 if (buffer->prev && !buffer->return_at_eof) 3491 _cpp_pop_buffer (pfile); 3492 else 3493 { 3494 /* End of translation. Do not pop the buffer yet. Increment 3495 line number so that the EOF token is on a line of its own 3496 (_cpp_lex_direct doesn't increment in that case, because 3497 it's hard for it to distinguish this special case). */ 3498 CPP_INCREMENT_LINE (pfile, 0); 3499 return false; 3500 } 3501 } 3502} 3503 3504#define IF_NEXT_IS(CHAR, THEN_TYPE, ELSE_TYPE) \ 3505 do \ 3506 { \ 3507 result->type = ELSE_TYPE; \ 3508 if (*buffer->cur == CHAR) \ 3509 buffer->cur++, result->type = THEN_TYPE; \ 3510 } \ 3511 while (0) 3512 3513/* Lex a token into pfile->cur_token, which is also incremented, to 3514 get diagnostics pointing to the correct location. 3515 3516 Does not handle issues such as token lookahead, multiple-include 3517 optimization, directives, skipping etc. This function is only 3518 suitable for use by _cpp_lex_token, and in special cases like 3519 lex_expansion_token which doesn't care for any of these issues. 3520 3521 When meeting a newline, returns CPP_EOF if parsing a directive, 3522 otherwise returns to the start of the token buffer if permissible. 3523 Returns the location of the lexed token. */ 3524cpp_token * 3525_cpp_lex_direct (cpp_reader *pfile) 3526{ 3527 cppchar_t c; 3528 cpp_buffer *buffer; 3529 const unsigned char *comment_start; 3530 bool fallthrough_comment = false; 3531 cpp_token *result = pfile->cur_token++; 3532 3533 fresh_line: 3534 result->flags = 0; 3535 buffer = pfile->buffer; 3536 if (buffer->need_line) 3537 { 3538 if (pfile->state.in_deferred_pragma) 3539 { 3540 /* This can happen in cases like: 3541 #define loop(x) whatever 3542 #pragma omp loop 3543 where when trying to expand loop we need to peek 3544 next token after loop, but aren't still in_deferred_pragma 3545 mode but are in in_directive mode, so buffer->need_line 3546 is set, a CPP_EOF is peeked. */ 3547 result->type = CPP_PRAGMA_EOL; 3548 pfile->state.in_deferred_pragma = false; 3549 if (!pfile->state.pragma_allow_expansion) 3550 pfile->state.prevent_expansion--; 3551 return result; 3552 } 3553 if (!_cpp_get_fresh_line (pfile)) 3554 { 3555 result->type = CPP_EOF; 3556 /* Not a real EOF in a directive or arg parsing -- we refuse 3557 to advance to the next file now, and will once we're out 3558 of those modes. */ 3559 if (!pfile->state.in_directive && !pfile->state.parsing_args) 3560 { 3561 /* Tell the compiler the line number of the EOF token. */ 3562 result->src_loc = pfile->line_table->highest_line; 3563 result->flags = BOL; 3564 /* Now pop the buffer that _cpp_get_fresh_line did not. */ 3565 _cpp_pop_buffer (pfile); 3566 } 3567 return result; 3568 } 3569 if (buffer != pfile->buffer) 3570 fallthrough_comment = false; 3571 if (!pfile->keep_tokens) 3572 { 3573 pfile->cur_run = &pfile->base_run; 3574 result = pfile->base_run.base; 3575 pfile->cur_token = result + 1; 3576 } 3577 result->flags = BOL; 3578 if (pfile->state.parsing_args == 2) 3579 result->flags |= PREV_WHITE; 3580 } 3581 buffer = pfile->buffer; 3582 update_tokens_line: 3583 result->src_loc = pfile->line_table->highest_line; 3584 3585 skipped_white: 3586 if (buffer->cur >= buffer->notes[buffer->cur_note].pos 3587 && !pfile->overlaid_buffer) 3588 { 3589 _cpp_process_line_notes (pfile, false); 3590 result->src_loc = pfile->line_table->highest_line; 3591 } 3592 c = *buffer->cur++; 3593 3594 if (pfile->forced_token_location) 3595 result->src_loc = pfile->forced_token_location; 3596 else 3597 result->src_loc = linemap_position_for_column (pfile->line_table, 3598 CPP_BUF_COLUMN (buffer, buffer->cur)); 3599 3600 switch (c) 3601 { 3602 case ' ': case '\t': case '\f': case '\v': case '\0': 3603 result->flags |= PREV_WHITE; 3604 skip_whitespace (pfile, c); 3605 goto skipped_white; 3606 3607 case '\n': 3608 /* Increment the line, unless this is the last line ... */ 3609 if (buffer->cur < buffer->rlimit 3610 /* ... or this is a #include, (where _cpp_stack_file needs to 3611 unwind by one line) ... */ 3612 || (pfile->state.in_directive > 1 3613 /* ... except traditional-cpp increments this elsewhere. */ 3614 && !CPP_OPTION (pfile, traditional))) 3615 CPP_INCREMENT_LINE (pfile, 0); 3616 buffer->need_line = true; 3617 if (pfile->state.in_deferred_pragma) 3618 { 3619 /* Produce the PRAGMA_EOL on this line. File reading 3620 ensures there is always a \n at end of the buffer, thus 3621 in a deferred pragma we always see CPP_PRAGMA_EOL before 3622 any CPP_EOF. */ 3623 result->type = CPP_PRAGMA_EOL; 3624 result->flags &= ~PREV_WHITE; 3625 pfile->state.in_deferred_pragma = false; 3626 if (!pfile->state.pragma_allow_expansion) 3627 pfile->state.prevent_expansion--; 3628 return result; 3629 } 3630 goto fresh_line; 3631 3632 case '0': case '1': case '2': case '3': case '4': 3633 case '5': case '6': case '7': case '8': case '9': 3634 { 3635 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 3636 result->type = CPP_NUMBER; 3637 lex_number (pfile, &result->val.str, &nst); 3638 warn_about_normalization (pfile, result, &nst); 3639 break; 3640 } 3641 3642 case 'L': 3643 case 'u': 3644 case 'U': 3645 case 'R': 3646 /* 'L', 'u', 'U', 'u8' or 'R' may introduce wide characters, 3647 wide strings or raw strings. */ 3648 if (c == 'L' || CPP_OPTION (pfile, rliterals) 3649 || (c != 'R' && CPP_OPTION (pfile, uliterals))) 3650 { 3651 if ((*buffer->cur == '\'' && c != 'R') 3652 || *buffer->cur == '"' 3653 || (*buffer->cur == 'R' 3654 && c != 'R' 3655 && buffer->cur[1] == '"' 3656 && CPP_OPTION (pfile, rliterals)) 3657 || (*buffer->cur == '8' 3658 && c == 'u' 3659 && ((buffer->cur[1] == '"' || (buffer->cur[1] == '\'' 3660 && CPP_OPTION (pfile, utf8_char_literals))) 3661 || (buffer->cur[1] == 'R' && buffer->cur[2] == '"' 3662 && CPP_OPTION (pfile, rliterals))))) 3663 { 3664 lex_string (pfile, result, buffer->cur - 1); 3665 break; 3666 } 3667 } 3668 /* Fall through. */ 3669 3670 case '_': 3671 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 3672 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 3673 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': 3674 case 's': case 't': case 'v': case 'w': case 'x': 3675 case 'y': case 'z': 3676 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 3677 case 'G': case 'H': case 'I': case 'J': case 'K': 3678 case 'M': case 'N': case 'O': case 'P': case 'Q': 3679 case 'S': case 'T': case 'V': case 'W': case 'X': 3680 case 'Y': case 'Z': 3681 result->type = CPP_NAME; 3682 { 3683 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 3684 result->val.node.node = lex_identifier (pfile, buffer->cur - 1, false, 3685 &nst, 3686 &result->val.node.spelling); 3687 warn_about_normalization (pfile, result, &nst); 3688 } 3689 3690 /* Convert named operators to their proper types. */ 3691 if (result->val.node.node->flags & NODE_OPERATOR) 3692 { 3693 result->flags |= NAMED_OP; 3694 result->type = (enum cpp_ttype) result->val.node.node->directive_index; 3695 } 3696 3697 /* Signal FALLTHROUGH comment followed by another token. */ 3698 if (fallthrough_comment) 3699 result->flags |= PREV_FALLTHROUGH; 3700 break; 3701 3702 case '\'': 3703 case '"': 3704 lex_string (pfile, result, buffer->cur - 1); 3705 break; 3706 3707 case '/': 3708 /* A potential block or line comment. */ 3709 comment_start = buffer->cur; 3710 c = *buffer->cur; 3711 3712 if (c == '*') 3713 { 3714 if (_cpp_skip_block_comment (pfile)) 3715 cpp_error (pfile, CPP_DL_ERROR, "unterminated comment"); 3716 } 3717 else if (c == '/' && ! CPP_OPTION (pfile, traditional)) 3718 { 3719 /* Don't warn for system headers. */ 3720 if (_cpp_in_system_header (pfile)) 3721 ; 3722 /* Warn about comments if pedantically GNUC89, and not 3723 in system headers. */ 3724 else if (CPP_OPTION (pfile, lang) == CLK_GNUC89 3725 && CPP_PEDANTIC (pfile) 3726 && ! buffer->warned_cplusplus_comments) 3727 { 3728 if (cpp_error (pfile, CPP_DL_PEDWARN, 3729 "C++ style comments are not allowed in ISO C90")) 3730 cpp_error (pfile, CPP_DL_NOTE, 3731 "(this will be reported only once per input file)"); 3732 buffer->warned_cplusplus_comments = 1; 3733 } 3734 /* Or if specifically desired via -Wc90-c99-compat. */ 3735 else if (CPP_OPTION (pfile, cpp_warn_c90_c99_compat) > 0 3736 && ! CPP_OPTION (pfile, cplusplus) 3737 && ! buffer->warned_cplusplus_comments) 3738 { 3739 if (cpp_error (pfile, CPP_DL_WARNING, 3740 "C++ style comments are incompatible with C90")) 3741 cpp_error (pfile, CPP_DL_NOTE, 3742 "(this will be reported only once per input file)"); 3743 buffer->warned_cplusplus_comments = 1; 3744 } 3745 /* In C89/C94, C++ style comments are forbidden. */ 3746 else if ((CPP_OPTION (pfile, lang) == CLK_STDC89 3747 || CPP_OPTION (pfile, lang) == CLK_STDC94)) 3748 { 3749 /* But don't be confused about valid code such as 3750 - // immediately followed by *, 3751 - // in a preprocessing directive, 3752 - // in an #if 0 block. */ 3753 if (buffer->cur[1] == '*' 3754 || pfile->state.in_directive 3755 || pfile->state.skipping) 3756 { 3757 result->type = CPP_DIV; 3758 break; 3759 } 3760 else if (! buffer->warned_cplusplus_comments) 3761 { 3762 if (cpp_error (pfile, CPP_DL_ERROR, 3763 "C++ style comments are not allowed in " 3764 "ISO C90")) 3765 cpp_error (pfile, CPP_DL_NOTE, 3766 "(this will be reported only once per input " 3767 "file)"); 3768 buffer->warned_cplusplus_comments = 1; 3769 } 3770 } 3771 if (skip_line_comment (pfile) && CPP_OPTION (pfile, warn_comments)) 3772 cpp_warning (pfile, CPP_W_COMMENTS, "multi-line comment"); 3773 } 3774 else if (c == '=') 3775 { 3776 buffer->cur++; 3777 result->type = CPP_DIV_EQ; 3778 break; 3779 } 3780 else 3781 { 3782 result->type = CPP_DIV; 3783 break; 3784 } 3785 3786 if (fallthrough_comment_p (pfile, comment_start)) 3787 fallthrough_comment = true; 3788 3789 if (pfile->cb.comment) 3790 { 3791 size_t len = pfile->buffer->cur - comment_start; 3792 pfile->cb.comment (pfile, result->src_loc, comment_start - 1, 3793 len + 1); 3794 } 3795 3796 if (!pfile->state.save_comments) 3797 { 3798 result->flags |= PREV_WHITE; 3799 goto update_tokens_line; 3800 } 3801 3802 if (fallthrough_comment) 3803 result->flags |= PREV_FALLTHROUGH; 3804 3805 /* Save the comment as a token in its own right. */ 3806 save_comment (pfile, result, comment_start, c); 3807 break; 3808 3809 case '<': 3810 if (pfile->state.angled_headers) 3811 { 3812 lex_string (pfile, result, buffer->cur - 1); 3813 if (result->type != CPP_LESS) 3814 break; 3815 } 3816 3817 result->type = CPP_LESS; 3818 if (*buffer->cur == '=') 3819 { 3820 buffer->cur++, result->type = CPP_LESS_EQ; 3821 if (*buffer->cur == '>' 3822 && CPP_OPTION (pfile, cplusplus) 3823 && CPP_OPTION (pfile, lang) >= CLK_GNUCXX20) 3824 buffer->cur++, result->type = CPP_SPACESHIP; 3825 } 3826 else if (*buffer->cur == '<') 3827 { 3828 buffer->cur++; 3829 IF_NEXT_IS ('=', CPP_LSHIFT_EQ, CPP_LSHIFT); 3830 } 3831 else if (CPP_OPTION (pfile, digraphs)) 3832 { 3833 if (*buffer->cur == ':') 3834 { 3835 /* C++11 [2.5/3 lex.pptoken], "Otherwise, if the next 3836 three characters are <:: and the subsequent character 3837 is neither : nor >, the < is treated as a preprocessor 3838 token by itself". */ 3839 if (CPP_OPTION (pfile, cplusplus) 3840 && CPP_OPTION (pfile, lang) != CLK_CXX98 3841 && CPP_OPTION (pfile, lang) != CLK_GNUCXX 3842 && buffer->cur[1] == ':' 3843 && buffer->cur[2] != ':' && buffer->cur[2] != '>') 3844 break; 3845 3846 buffer->cur++; 3847 result->flags |= DIGRAPH; 3848 result->type = CPP_OPEN_SQUARE; 3849 } 3850 else if (*buffer->cur == '%') 3851 { 3852 buffer->cur++; 3853 result->flags |= DIGRAPH; 3854 result->type = CPP_OPEN_BRACE; 3855 } 3856 } 3857 break; 3858 3859 case '>': 3860 result->type = CPP_GREATER; 3861 if (*buffer->cur == '=') 3862 buffer->cur++, result->type = CPP_GREATER_EQ; 3863 else if (*buffer->cur == '>') 3864 { 3865 buffer->cur++; 3866 IF_NEXT_IS ('=', CPP_RSHIFT_EQ, CPP_RSHIFT); 3867 } 3868 break; 3869 3870 case '%': 3871 result->type = CPP_MOD; 3872 if (*buffer->cur == '=') 3873 buffer->cur++, result->type = CPP_MOD_EQ; 3874 else if (CPP_OPTION (pfile, digraphs)) 3875 { 3876 if (*buffer->cur == ':') 3877 { 3878 buffer->cur++; 3879 result->flags |= DIGRAPH; 3880 result->type = CPP_HASH; 3881 if (*buffer->cur == '%' && buffer->cur[1] == ':') 3882 buffer->cur += 2, result->type = CPP_PASTE, result->val.token_no = 0; 3883 } 3884 else if (*buffer->cur == '>') 3885 { 3886 buffer->cur++; 3887 result->flags |= DIGRAPH; 3888 result->type = CPP_CLOSE_BRACE; 3889 } 3890 } 3891 break; 3892 3893 case '.': 3894 result->type = CPP_DOT; 3895 if (ISDIGIT (*buffer->cur)) 3896 { 3897 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 3898 result->type = CPP_NUMBER; 3899 lex_number (pfile, &result->val.str, &nst); 3900 warn_about_normalization (pfile, result, &nst); 3901 } 3902 else if (*buffer->cur == '.' && buffer->cur[1] == '.') 3903 buffer->cur += 2, result->type = CPP_ELLIPSIS; 3904 else if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) 3905 buffer->cur++, result->type = CPP_DOT_STAR; 3906 break; 3907 3908 case '+': 3909 result->type = CPP_PLUS; 3910 if (*buffer->cur == '+') 3911 buffer->cur++, result->type = CPP_PLUS_PLUS; 3912 else if (*buffer->cur == '=') 3913 buffer->cur++, result->type = CPP_PLUS_EQ; 3914 break; 3915 3916 case '-': 3917 result->type = CPP_MINUS; 3918 if (*buffer->cur == '>') 3919 { 3920 buffer->cur++; 3921 result->type = CPP_DEREF; 3922 if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) 3923 buffer->cur++, result->type = CPP_DEREF_STAR; 3924 } 3925 else if (*buffer->cur == '-') 3926 buffer->cur++, result->type = CPP_MINUS_MINUS; 3927 else if (*buffer->cur == '=') 3928 buffer->cur++, result->type = CPP_MINUS_EQ; 3929 break; 3930 3931 case '&': 3932 result->type = CPP_AND; 3933 if (*buffer->cur == '&') 3934 buffer->cur++, result->type = CPP_AND_AND; 3935 else if (*buffer->cur == '=') 3936 buffer->cur++, result->type = CPP_AND_EQ; 3937 break; 3938 3939 case '|': 3940 result->type = CPP_OR; 3941 if (*buffer->cur == '|') 3942 buffer->cur++, result->type = CPP_OR_OR; 3943 else if (*buffer->cur == '=') 3944 buffer->cur++, result->type = CPP_OR_EQ; 3945 break; 3946 3947 case ':': 3948 result->type = CPP_COLON; 3949 if (*buffer->cur == ':' && CPP_OPTION (pfile, scope)) 3950 buffer->cur++, result->type = CPP_SCOPE; 3951 else if (*buffer->cur == '>' && CPP_OPTION (pfile, digraphs)) 3952 { 3953 buffer->cur++; 3954 result->flags |= DIGRAPH; 3955 result->type = CPP_CLOSE_SQUARE; 3956 } 3957 break; 3958 3959 case '*': IF_NEXT_IS ('=', CPP_MULT_EQ, CPP_MULT); break; 3960 case '=': IF_NEXT_IS ('=', CPP_EQ_EQ, CPP_EQ); break; 3961 case '!': IF_NEXT_IS ('=', CPP_NOT_EQ, CPP_NOT); break; 3962 case '^': IF_NEXT_IS ('=', CPP_XOR_EQ, CPP_XOR); break; 3963 case '#': IF_NEXT_IS ('#', CPP_PASTE, CPP_HASH); result->val.token_no = 0; break; 3964 3965 case '?': result->type = CPP_QUERY; break; 3966 case '~': result->type = CPP_COMPL; break; 3967 case ',': result->type = CPP_COMMA; break; 3968 case '(': result->type = CPP_OPEN_PAREN; break; 3969 case ')': result->type = CPP_CLOSE_PAREN; break; 3970 case '[': result->type = CPP_OPEN_SQUARE; break; 3971 case ']': result->type = CPP_CLOSE_SQUARE; break; 3972 case '{': result->type = CPP_OPEN_BRACE; break; 3973 case '}': result->type = CPP_CLOSE_BRACE; break; 3974 case ';': result->type = CPP_SEMICOLON; break; 3975 3976 /* @ is a punctuator in Objective-C. */ 3977 case '@': result->type = CPP_ATSIGN; break; 3978 3979 default: 3980 { 3981 const uchar *base = --buffer->cur; 3982 3983 /* Check for an extended identifier ($ or UCN or UTF-8). */ 3984 struct normalize_state nst = INITIAL_NORMALIZE_STATE; 3985 if (forms_identifier_p (pfile, true, &nst)) 3986 { 3987 result->type = CPP_NAME; 3988 result->val.node.node = lex_identifier (pfile, base, true, &nst, 3989 &result->val.node.spelling); 3990 warn_about_normalization (pfile, result, &nst); 3991 break; 3992 } 3993 3994 /* Otherwise this will form a CPP_OTHER token. Parse valid UTF-8 as a 3995 single token. */ 3996 buffer->cur++; 3997 if (c >= utf8_signifier) 3998 { 3999 const uchar *pstr = base; 4000 cppchar_t s; 4001 if (_cpp_valid_utf8 (pfile, &pstr, buffer->rlimit, 0, NULL, &s)) 4002 buffer->cur = pstr; 4003 } 4004 create_literal (pfile, result, base, buffer->cur - base, CPP_OTHER); 4005 break; 4006 } 4007 4008 } 4009 4010 /* Potentially convert the location of the token to a range. */ 4011 if (result->src_loc >= RESERVED_LOCATION_COUNT 4012 && result->type != CPP_EOF) 4013 { 4014 /* Ensure that any line notes are processed, so that we have the 4015 correct physical line/column for the end-point of the token even 4016 when a logical line is split via one or more backslashes. */ 4017 if (buffer->cur >= buffer->notes[buffer->cur_note].pos 4018 && !pfile->overlaid_buffer) 4019 _cpp_process_line_notes (pfile, false); 4020 4021 source_range tok_range; 4022 tok_range.m_start = result->src_loc; 4023 tok_range.m_finish 4024 = linemap_position_for_column (pfile->line_table, 4025 CPP_BUF_COLUMN (buffer, buffer->cur)); 4026 4027 result->src_loc = COMBINE_LOCATION_DATA (pfile->line_table, 4028 result->src_loc, 4029 tok_range, NULL); 4030 } 4031 4032 return result; 4033} 4034 4035/* An upper bound on the number of bytes needed to spell TOKEN. 4036 Does not include preceding whitespace. */ 4037unsigned int 4038cpp_token_len (const cpp_token *token) 4039{ 4040 unsigned int len; 4041 4042 switch (TOKEN_SPELL (token)) 4043 { 4044 default: len = 6; break; 4045 case SPELL_LITERAL: len = token->val.str.len; break; 4046 case SPELL_IDENT: len = NODE_LEN (token->val.node.node) * 10; break; 4047 } 4048 4049 return len; 4050} 4051 4052/* Parse UTF-8 out of NAMEP and place a \U escape in BUFFER. 4053 Return the number of bytes read out of NAME. (There are always 4054 10 bytes written to BUFFER.) */ 4055 4056static size_t 4057utf8_to_ucn (unsigned char *buffer, const unsigned char *name) 4058{ 4059 int j; 4060 int ucn_len = 0; 4061 int ucn_len_c; 4062 unsigned t; 4063 unsigned long utf32; 4064 4065 /* Compute the length of the UTF-8 sequence. */ 4066 for (t = *name; t & 0x80; t <<= 1) 4067 ucn_len++; 4068 4069 utf32 = *name & (0x7F >> ucn_len); 4070 for (ucn_len_c = 1; ucn_len_c < ucn_len; ucn_len_c++) 4071 { 4072 utf32 = (utf32 << 6) | (*++name & 0x3F); 4073 4074 /* Ill-formed UTF-8. */ 4075 if ((*name & ~0x3F) != 0x80) 4076 abort (); 4077 } 4078 4079 *buffer++ = '\\'; 4080 *buffer++ = 'U'; 4081 for (j = 7; j >= 0; j--) 4082 *buffer++ = "0123456789abcdef"[(utf32 >> (4 * j)) & 0xF]; 4083 return ucn_len; 4084} 4085 4086/* Given a token TYPE corresponding to a digraph, return a pointer to 4087 the spelling of the digraph. */ 4088static const unsigned char * 4089cpp_digraph2name (enum cpp_ttype type) 4090{ 4091 return digraph_spellings[(int) type - (int) CPP_FIRST_DIGRAPH]; 4092} 4093 4094/* Write the spelling of an identifier IDENT, using UCNs, to BUFFER. 4095 The buffer must already contain the enough space to hold the 4096 token's spelling. Returns a pointer to the character after the 4097 last character written. */ 4098unsigned char * 4099_cpp_spell_ident_ucns (unsigned char *buffer, cpp_hashnode *ident) 4100{ 4101 size_t i; 4102 const unsigned char *name = NODE_NAME (ident); 4103 4104 for (i = 0; i < NODE_LEN (ident); i++) 4105 if (name[i] & ~0x7F) 4106 { 4107 i += utf8_to_ucn (buffer, name + i) - 1; 4108 buffer += 10; 4109 } 4110 else 4111 *buffer++ = name[i]; 4112 4113 return buffer; 4114} 4115 4116/* Write the spelling of a token TOKEN to BUFFER. The buffer must 4117 already contain the enough space to hold the token's spelling. 4118 Returns a pointer to the character after the last character written. 4119 FORSTRING is true if this is to be the spelling after translation 4120 phase 1 (with the original spelling of extended identifiers), false 4121 if extended identifiers should always be written using UCNs (there is 4122 no option for always writing them in the internal UTF-8 form). 4123 FIXME: Would be nice if we didn't need the PFILE argument. */ 4124unsigned char * 4125cpp_spell_token (cpp_reader *pfile, const cpp_token *token, 4126 unsigned char *buffer, bool forstring) 4127{ 4128 switch (TOKEN_SPELL (token)) 4129 { 4130 case SPELL_OPERATOR: 4131 { 4132 const unsigned char *spelling; 4133 unsigned char c; 4134 4135 if (token->flags & DIGRAPH) 4136 spelling = cpp_digraph2name (token->type); 4137 else if (token->flags & NAMED_OP) 4138 goto spell_ident; 4139 else 4140 spelling = TOKEN_NAME (token); 4141 4142 while ((c = *spelling++) != '\0') 4143 *buffer++ = c; 4144 } 4145 break; 4146 4147 spell_ident: 4148 case SPELL_IDENT: 4149 if (forstring) 4150 { 4151 memcpy (buffer, NODE_NAME (token->val.node.spelling), 4152 NODE_LEN (token->val.node.spelling)); 4153 buffer += NODE_LEN (token->val.node.spelling); 4154 } 4155 else 4156 buffer = _cpp_spell_ident_ucns (buffer, token->val.node.node); 4157 break; 4158 4159 case SPELL_LITERAL: 4160 memcpy (buffer, token->val.str.text, token->val.str.len); 4161 buffer += token->val.str.len; 4162 break; 4163 4164 case SPELL_NONE: 4165 cpp_error (pfile, CPP_DL_ICE, 4166 "unspellable token %s", TOKEN_NAME (token)); 4167 break; 4168 } 4169 4170 return buffer; 4171} 4172 4173/* Returns TOKEN spelt as a null-terminated string. The string is 4174 freed when the reader is destroyed. Useful for diagnostics. */ 4175unsigned char * 4176cpp_token_as_text (cpp_reader *pfile, const cpp_token *token) 4177{ 4178 unsigned int len = cpp_token_len (token) + 1; 4179 unsigned char *start = _cpp_unaligned_alloc (pfile, len), *end; 4180 4181 end = cpp_spell_token (pfile, token, start, false); 4182 end[0] = '\0'; 4183 4184 return start; 4185} 4186 4187/* Returns a pointer to a string which spells the token defined by 4188 TYPE and FLAGS. Used by C front ends, which really should move to 4189 using cpp_token_as_text. */ 4190const char * 4191cpp_type2name (enum cpp_ttype type, unsigned char flags) 4192{ 4193 if (flags & DIGRAPH) 4194 return (const char *) cpp_digraph2name (type); 4195 else if (flags & NAMED_OP) 4196 return cpp_named_operator2name (type); 4197 4198 return (const char *) token_spellings[type].name; 4199} 4200 4201/* Writes the spelling of token to FP, without any preceding space. 4202 Separated from cpp_spell_token for efficiency - to avoid stdio 4203 double-buffering. */ 4204void 4205cpp_output_token (const cpp_token *token, FILE *fp) 4206{ 4207 switch (TOKEN_SPELL (token)) 4208 { 4209 case SPELL_OPERATOR: 4210 { 4211 const unsigned char *spelling; 4212 int c; 4213 4214 if (token->flags & DIGRAPH) 4215 spelling = cpp_digraph2name (token->type); 4216 else if (token->flags & NAMED_OP) 4217 goto spell_ident; 4218 else 4219 spelling = TOKEN_NAME (token); 4220 4221 c = *spelling; 4222 do 4223 putc (c, fp); 4224 while ((c = *++spelling) != '\0'); 4225 } 4226 break; 4227 4228 spell_ident: 4229 case SPELL_IDENT: 4230 { 4231 size_t i; 4232 const unsigned char * name = NODE_NAME (token->val.node.node); 4233 4234 for (i = 0; i < NODE_LEN (token->val.node.node); i++) 4235 if (name[i] & ~0x7F) 4236 { 4237 unsigned char buffer[10]; 4238 i += utf8_to_ucn (buffer, name + i) - 1; 4239 fwrite (buffer, 1, 10, fp); 4240 } 4241 else 4242 fputc (NODE_NAME (token->val.node.node)[i], fp); 4243 } 4244 break; 4245 4246 case SPELL_LITERAL: 4247 if (token->type == CPP_HEADER_NAME) 4248 fputc ('"', fp); 4249 fwrite (token->val.str.text, 1, token->val.str.len, fp); 4250 if (token->type == CPP_HEADER_NAME) 4251 fputc ('"', fp); 4252 break; 4253 4254 case SPELL_NONE: 4255 /* An error, most probably. */ 4256 break; 4257 } 4258} 4259 4260/* Compare two tokens. */ 4261int 4262_cpp_equiv_tokens (const cpp_token *a, const cpp_token *b) 4263{ 4264 if (a->type == b->type && a->flags == b->flags) 4265 switch (TOKEN_SPELL (a)) 4266 { 4267 default: /* Keep compiler happy. */ 4268 case SPELL_OPERATOR: 4269 /* token_no is used to track where multiple consecutive ## 4270 tokens were originally located. */ 4271 return (a->type != CPP_PASTE || a->val.token_no == b->val.token_no); 4272 case SPELL_NONE: 4273 return (a->type != CPP_MACRO_ARG 4274 || (a->val.macro_arg.arg_no == b->val.macro_arg.arg_no 4275 && a->val.macro_arg.spelling == b->val.macro_arg.spelling)); 4276 case SPELL_IDENT: 4277 return (a->val.node.node == b->val.node.node 4278 && a->val.node.spelling == b->val.node.spelling); 4279 case SPELL_LITERAL: 4280 return (a->val.str.len == b->val.str.len 4281 && !memcmp (a->val.str.text, b->val.str.text, 4282 a->val.str.len)); 4283 } 4284 4285 return 0; 4286} 4287 4288/* Returns nonzero if a space should be inserted to avoid an 4289 accidental token paste for output. For simplicity, it is 4290 conservative, and occasionally advises a space where one is not 4291 needed, e.g. "." and ".2". */ 4292int 4293cpp_avoid_paste (cpp_reader *pfile, const cpp_token *token1, 4294 const cpp_token *token2) 4295{ 4296 enum cpp_ttype a = token1->type, b = token2->type; 4297 cppchar_t c; 4298 4299 if (token1->flags & NAMED_OP) 4300 a = CPP_NAME; 4301 if (token2->flags & NAMED_OP) 4302 b = CPP_NAME; 4303 4304 c = EOF; 4305 if (token2->flags & DIGRAPH) 4306 c = digraph_spellings[(int) b - (int) CPP_FIRST_DIGRAPH][0]; 4307 else if (token_spellings[b].category == SPELL_OPERATOR) 4308 c = token_spellings[b].name[0]; 4309 4310 /* Quickly get everything that can paste with an '='. */ 4311 if ((int) a <= (int) CPP_LAST_EQ && c == '=') 4312 return 1; 4313 4314 switch (a) 4315 { 4316 case CPP_GREATER: return c == '>'; 4317 case CPP_LESS: return c == '<' || c == '%' || c == ':'; 4318 case CPP_PLUS: return c == '+'; 4319 case CPP_MINUS: return c == '-' || c == '>'; 4320 case CPP_DIV: return c == '/' || c == '*'; /* Comments. */ 4321 case CPP_MOD: return c == ':' || c == '>'; 4322 case CPP_AND: return c == '&'; 4323 case CPP_OR: return c == '|'; 4324 case CPP_COLON: return c == ':' || c == '>'; 4325 case CPP_DEREF: return c == '*'; 4326 case CPP_DOT: return c == '.' || c == '%' || b == CPP_NUMBER; 4327 case CPP_HASH: return c == '#' || c == '%'; /* Digraph form. */ 4328 case CPP_PRAGMA: 4329 case CPP_NAME: return ((b == CPP_NUMBER 4330 && name_p (pfile, &token2->val.str)) 4331 || b == CPP_NAME 4332 || b == CPP_CHAR || b == CPP_STRING); /* L */ 4333 case CPP_NUMBER: return (b == CPP_NUMBER || b == CPP_NAME 4334 || b == CPP_CHAR 4335 || c == '.' || c == '+' || c == '-'); 4336 /* UCNs */ 4337 case CPP_OTHER: return ((token1->val.str.text[0] == '\\' 4338 && b == CPP_NAME) 4339 || (CPP_OPTION (pfile, objc) 4340 && token1->val.str.text[0] == '@' 4341 && (b == CPP_NAME || b == CPP_STRING))); 4342 case CPP_LESS_EQ: return c == '>'; 4343 case CPP_STRING: 4344 case CPP_WSTRING: 4345 case CPP_UTF8STRING: 4346 case CPP_STRING16: 4347 case CPP_STRING32: return (CPP_OPTION (pfile, user_literals) 4348 && (b == CPP_NAME 4349 || (TOKEN_SPELL (token2) == SPELL_LITERAL 4350 && ISIDST (token2->val.str.text[0])))); 4351 4352 default: break; 4353 } 4354 4355 return 0; 4356} 4357 4358/* Output all the remaining tokens on the current line, and a newline 4359 character, to FP. Leading whitespace is removed. If there are 4360 macros, special token padding is not performed. */ 4361void 4362cpp_output_line (cpp_reader *pfile, FILE *fp) 4363{ 4364 const cpp_token *token; 4365 4366 token = cpp_get_token (pfile); 4367 while (token->type != CPP_EOF) 4368 { 4369 cpp_output_token (token, fp); 4370 token = cpp_get_token (pfile); 4371 if (token->flags & PREV_WHITE) 4372 putc (' ', fp); 4373 } 4374 4375 putc ('\n', fp); 4376} 4377 4378/* Return a string representation of all the remaining tokens on the 4379 current line. The result is allocated using xmalloc and must be 4380 freed by the caller. */ 4381unsigned char * 4382cpp_output_line_to_string (cpp_reader *pfile, const unsigned char *dir_name) 4383{ 4384 const cpp_token *token; 4385 unsigned int out = dir_name ? ustrlen (dir_name) : 0; 4386 unsigned int alloced = 120 + out; 4387 unsigned char *result = (unsigned char *) xmalloc (alloced); 4388 4389 /* If DIR_NAME is empty, there are no initial contents. */ 4390 if (dir_name) 4391 { 4392 sprintf ((char *) result, "#%s ", dir_name); 4393 out += 2; 4394 } 4395 4396 token = cpp_get_token (pfile); 4397 while (token->type != CPP_EOF) 4398 { 4399 unsigned char *last; 4400 /* Include room for a possible space and the terminating nul. */ 4401 unsigned int len = cpp_token_len (token) + 2; 4402 4403 if (out + len > alloced) 4404 { 4405 alloced *= 2; 4406 if (out + len > alloced) 4407 alloced = out + len; 4408 result = (unsigned char *) xrealloc (result, alloced); 4409 } 4410 4411 last = cpp_spell_token (pfile, token, &result[out], 0); 4412 out = last - result; 4413 4414 token = cpp_get_token (pfile); 4415 if (token->flags & PREV_WHITE) 4416 result[out++] = ' '; 4417 } 4418 4419 result[out] = '\0'; 4420 return result; 4421} 4422 4423/* Memory buffers. Changing these three constants can have a dramatic 4424 effect on performance. The values here are reasonable defaults, 4425 but might be tuned. If you adjust them, be sure to test across a 4426 range of uses of cpplib, including heavy nested function-like macro 4427 expansion. Also check the change in peak memory usage (NJAMD is a 4428 good tool for this). */ 4429#define MIN_BUFF_SIZE 8000 4430#define BUFF_SIZE_UPPER_BOUND(MIN_SIZE) (MIN_BUFF_SIZE + (MIN_SIZE) * 3 / 2) 4431#define EXTENDED_BUFF_SIZE(BUFF, MIN_EXTRA) \ 4432 (MIN_EXTRA + ((BUFF)->limit - (BUFF)->cur) * 2) 4433 4434#if MIN_BUFF_SIZE > BUFF_SIZE_UPPER_BOUND (0) 4435 #error BUFF_SIZE_UPPER_BOUND must be at least as large as MIN_BUFF_SIZE! 4436#endif 4437 4438/* Create a new allocation buffer. Place the control block at the end 4439 of the buffer, so that buffer overflows will cause immediate chaos. */ 4440static _cpp_buff * 4441new_buff (size_t len) 4442{ 4443 _cpp_buff *result; 4444 unsigned char *base; 4445 4446 if (len < MIN_BUFF_SIZE) 4447 len = MIN_BUFF_SIZE; 4448 len = CPP_ALIGN (len); 4449 4450#ifdef ENABLE_VALGRIND_ANNOTATIONS 4451 /* Valgrind warns about uses of interior pointers, so put _cpp_buff 4452 struct first. */ 4453 size_t slen = CPP_ALIGN2 (sizeof (_cpp_buff), 2 * DEFAULT_ALIGNMENT); 4454 base = XNEWVEC (unsigned char, len + slen); 4455 result = (_cpp_buff *) base; 4456 base += slen; 4457#else 4458 base = XNEWVEC (unsigned char, len + sizeof (_cpp_buff)); 4459 result = (_cpp_buff *) (base + len); 4460#endif 4461 result->base = base; 4462 result->cur = base; 4463 result->limit = base + len; 4464 result->next = NULL; 4465 return result; 4466} 4467 4468/* Place a chain of unwanted allocation buffers on the free list. */ 4469void 4470_cpp_release_buff (cpp_reader *pfile, _cpp_buff *buff) 4471{ 4472 _cpp_buff *end = buff; 4473 4474 while (end->next) 4475 end = end->next; 4476 end->next = pfile->free_buffs; 4477 pfile->free_buffs = buff; 4478} 4479 4480/* Return a free buffer of size at least MIN_SIZE. */ 4481_cpp_buff * 4482_cpp_get_buff (cpp_reader *pfile, size_t min_size) 4483{ 4484 _cpp_buff *result, **p; 4485 4486 for (p = &pfile->free_buffs;; p = &(*p)->next) 4487 { 4488 size_t size; 4489 4490 if (*p == NULL) 4491 return new_buff (min_size); 4492 result = *p; 4493 size = result->limit - result->base; 4494 /* Return a buffer that's big enough, but don't waste one that's 4495 way too big. */ 4496 if (size >= min_size && size <= BUFF_SIZE_UPPER_BOUND (min_size)) 4497 break; 4498 } 4499 4500 *p = result->next; 4501 result->next = NULL; 4502 result->cur = result->base; 4503 return result; 4504} 4505 4506/* Creates a new buffer with enough space to hold the uncommitted 4507 remaining bytes of BUFF, and at least MIN_EXTRA more bytes. Copies 4508 the excess bytes to the new buffer. Chains the new buffer after 4509 BUFF, and returns the new buffer. */ 4510_cpp_buff * 4511_cpp_append_extend_buff (cpp_reader *pfile, _cpp_buff *buff, size_t min_extra) 4512{ 4513 size_t size = EXTENDED_BUFF_SIZE (buff, min_extra); 4514 _cpp_buff *new_buff = _cpp_get_buff (pfile, size); 4515 4516 buff->next = new_buff; 4517 memcpy (new_buff->base, buff->cur, BUFF_ROOM (buff)); 4518 return new_buff; 4519} 4520 4521/* Creates a new buffer with enough space to hold the uncommitted 4522 remaining bytes of the buffer pointed to by BUFF, and at least 4523 MIN_EXTRA more bytes. Copies the excess bytes to the new buffer. 4524 Chains the new buffer before the buffer pointed to by BUFF, and 4525 updates the pointer to point to the new buffer. */ 4526void 4527_cpp_extend_buff (cpp_reader *pfile, _cpp_buff **pbuff, size_t min_extra) 4528{ 4529 _cpp_buff *new_buff, *old_buff = *pbuff; 4530 size_t size = EXTENDED_BUFF_SIZE (old_buff, min_extra); 4531 4532 new_buff = _cpp_get_buff (pfile, size); 4533 memcpy (new_buff->base, old_buff->cur, BUFF_ROOM (old_buff)); 4534 new_buff->next = old_buff; 4535 *pbuff = new_buff; 4536} 4537 4538/* Free a chain of buffers starting at BUFF. */ 4539void 4540_cpp_free_buff (_cpp_buff *buff) 4541{ 4542 _cpp_buff *next; 4543 4544 for (; buff; buff = next) 4545 { 4546 next = buff->next; 4547#ifdef ENABLE_VALGRIND_ANNOTATIONS 4548 free (buff); 4549#else 4550 free (buff->base); 4551#endif 4552 } 4553} 4554 4555/* Allocate permanent, unaligned storage of length LEN. */ 4556unsigned char * 4557_cpp_unaligned_alloc (cpp_reader *pfile, size_t len) 4558{ 4559 _cpp_buff *buff = pfile->u_buff; 4560 unsigned char *result = buff->cur; 4561 4562 if (len > (size_t) (buff->limit - result)) 4563 { 4564 buff = _cpp_get_buff (pfile, len); 4565 buff->next = pfile->u_buff; 4566 pfile->u_buff = buff; 4567 result = buff->cur; 4568 } 4569 4570 buff->cur = result + len; 4571 return result; 4572} 4573 4574/* Allocate permanent, unaligned storage of length LEN from a_buff. 4575 That buffer is used for growing allocations when saving macro 4576 replacement lists in a #define, and when parsing an answer to an 4577 assertion in #assert, #unassert or #if (and therefore possibly 4578 whilst expanding macros). It therefore must not be used by any 4579 code that they might call: specifically the lexer and the guts of 4580 the macro expander. 4581 4582 All existing other uses clearly fit this restriction: storing 4583 registered pragmas during initialization. */ 4584unsigned char * 4585_cpp_aligned_alloc (cpp_reader *pfile, size_t len) 4586{ 4587 _cpp_buff *buff = pfile->a_buff; 4588 unsigned char *result = buff->cur; 4589 4590 if (len > (size_t) (buff->limit - result)) 4591 { 4592 buff = _cpp_get_buff (pfile, len); 4593 buff->next = pfile->a_buff; 4594 pfile->a_buff = buff; 4595 result = buff->cur; 4596 } 4597 4598 buff->cur = result + len; 4599 return result; 4600} 4601 4602/* Commit or allocate storage from a buffer. */ 4603 4604void * 4605_cpp_commit_buff (cpp_reader *pfile, size_t size) 4606{ 4607 void *ptr = BUFF_FRONT (pfile->a_buff); 4608 4609 if (pfile->hash_table->alloc_subobject) 4610 { 4611 void *copy = pfile->hash_table->alloc_subobject (size); 4612 memcpy (copy, ptr, size); 4613 ptr = copy; 4614 } 4615 else 4616 BUFF_FRONT (pfile->a_buff) += size; 4617 4618 return ptr; 4619} 4620 4621/* Say which field of TOK is in use. */ 4622 4623enum cpp_token_fld_kind 4624cpp_token_val_index (const cpp_token *tok) 4625{ 4626 switch (TOKEN_SPELL (tok)) 4627 { 4628 case SPELL_IDENT: 4629 return CPP_TOKEN_FLD_NODE; 4630 case SPELL_LITERAL: 4631 return CPP_TOKEN_FLD_STR; 4632 case SPELL_OPERATOR: 4633 /* Operands which were originally spelled as ident keep around 4634 the node for the exact spelling. */ 4635 if (tok->flags & NAMED_OP) 4636 return CPP_TOKEN_FLD_NODE; 4637 else if (tok->type == CPP_PASTE) 4638 return CPP_TOKEN_FLD_TOKEN_NO; 4639 else 4640 return CPP_TOKEN_FLD_NONE; 4641 case SPELL_NONE: 4642 if (tok->type == CPP_MACRO_ARG) 4643 return CPP_TOKEN_FLD_ARG_NO; 4644 else if (tok->type == CPP_PADDING) 4645 return CPP_TOKEN_FLD_SOURCE; 4646 else if (tok->type == CPP_PRAGMA) 4647 return CPP_TOKEN_FLD_PRAGMA; 4648 /* fall through */ 4649 default: 4650 return CPP_TOKEN_FLD_NONE; 4651 } 4652} 4653 4654/* All tokens lexed in R after calling this function will be forced to 4655 have their location_t to be P, until 4656 cpp_stop_forcing_token_locations is called for R. */ 4657 4658void 4659cpp_force_token_locations (cpp_reader *r, location_t loc) 4660{ 4661 r->forced_token_location = loc; 4662} 4663 4664/* Go back to assigning locations naturally for lexed tokens. */ 4665 4666void 4667cpp_stop_forcing_token_locations (cpp_reader *r) 4668{ 4669 r->forced_token_location = 0; 4670} 4671 4672/* We're looking at \, if it's escaping EOL, look past it. If at 4673 LIMIT, don't advance. */ 4674 4675static const unsigned char * 4676do_peek_backslash (const unsigned char *peek, const unsigned char *limit) 4677{ 4678 const unsigned char *probe = peek; 4679 4680 if (__builtin_expect (peek[1] == '\n', true)) 4681 { 4682 eol: 4683 probe += 2; 4684 if (__builtin_expect (probe < limit, true)) 4685 { 4686 peek = probe; 4687 if (*peek == '\\') 4688 /* The user might be perverse. */ 4689 return do_peek_backslash (peek, limit); 4690 } 4691 } 4692 else if (__builtin_expect (peek[1] == '\r', false)) 4693 { 4694 if (probe[2] == '\n') 4695 probe++; 4696 goto eol; 4697 } 4698 4699 return peek; 4700} 4701 4702static const unsigned char * 4703do_peek_next (const unsigned char *peek, const unsigned char *limit) 4704{ 4705 if (__builtin_expect (*peek == '\\', false)) 4706 peek = do_peek_backslash (peek, limit); 4707 return peek; 4708} 4709 4710static const unsigned char * 4711do_peek_prev (const unsigned char *peek, const unsigned char *bound) 4712{ 4713 if (peek == bound) 4714 return NULL; 4715 4716 unsigned char c = *--peek; 4717 if (__builtin_expect (c == '\n', false) 4718 || __builtin_expect (c == 'r', false)) 4719 { 4720 if (peek == bound) 4721 return peek; 4722 int ix = -1; 4723 if (c == '\n' && peek[ix] == '\r') 4724 { 4725 if (peek + ix == bound) 4726 return peek; 4727 ix--; 4728 } 4729 4730 if (peek[ix] == '\\') 4731 return do_peek_prev (peek + ix, bound); 4732 4733 return peek; 4734 } 4735 else 4736 return peek; 4737} 4738 4739/* If PEEK[-1] is identifier MATCH, scan past it and trailing white 4740 space. Otherwise return NULL. */ 4741 4742static const unsigned char * 4743do_peek_ident (const char *match, const unsigned char *peek, 4744 const unsigned char *limit) 4745{ 4746 for (; *++match; peek++) 4747 if (*peek != *match) 4748 { 4749 peek = do_peek_next (peek, limit); 4750 if (*peek != *match) 4751 return NULL; 4752 } 4753 4754 /* Must now not be looking at an identifier char. */ 4755 peek = do_peek_next (peek, limit); 4756 if (ISIDNUM (*peek)) 4757 return NULL; 4758 4759 /* Skip control-line whitespace. */ 4760 ws: 4761 while (*peek == ' ' || *peek == '\t') 4762 peek++; 4763 if (__builtin_expect (*peek == '\\', false)) 4764 { 4765 peek = do_peek_backslash (peek, limit); 4766 if (*peek != '\\') 4767 goto ws; 4768 } 4769 4770 return peek; 4771} 4772 4773/* Are we looking at a module control line starting as PEEK - 1? */ 4774 4775static bool 4776do_peek_module (cpp_reader *pfile, unsigned char c, 4777 const unsigned char *peek, const unsigned char *limit) 4778{ 4779 bool import = false; 4780 4781 if (__builtin_expect (c == 'e', false)) 4782 { 4783 if (!((peek[0] == 'x' || peek[0] == '\\') 4784 && (peek = do_peek_ident ("export", peek, limit)))) 4785 return false; 4786 4787 /* export, peek for import or module. No need to peek __import 4788 here. */ 4789 if (peek[0] == 'i') 4790 { 4791 if (!((peek[1] == 'm' || peek[1] == '\\') 4792 && (peek = do_peek_ident ("import", peek + 1, limit)))) 4793 return false; 4794 import = true; 4795 } 4796 else if (peek[0] == 'm') 4797 { 4798 if (!((peek[1] == 'o' || peek[1] == '\\') 4799 && (peek = do_peek_ident ("module", peek + 1, limit)))) 4800 return false; 4801 } 4802 else 4803 return false; 4804 } 4805 else if (__builtin_expect (c == 'i', false)) 4806 { 4807 if (!((peek[0] == 'm' || peek[0] == '\\') 4808 && (peek = do_peek_ident ("import", peek, limit)))) 4809 return false; 4810 import = true; 4811 } 4812 else if (__builtin_expect (c == '_', false)) 4813 { 4814 /* Needed for translated includes. */ 4815 if (!((peek[0] == '_' || peek[0] == '\\') 4816 && (peek = do_peek_ident ("__import", peek, limit)))) 4817 return false; 4818 import = true; 4819 } 4820 else if (__builtin_expect (c == 'm', false)) 4821 { 4822 if (!((peek[0] == 'o' || peek[0] == '\\') 4823 && (peek = do_peek_ident ("module", peek, limit)))) 4824 return false; 4825 } 4826 else 4827 return false; 4828 4829 /* Peek the next character to see if it's good enough. We'll be at 4830 the first non-whitespace char, including skipping an escaped 4831 newline. */ 4832 /* ... import followed by identifier, ':', '<' or header-name 4833 preprocessing tokens, or module followed by identifier, ':' or 4834 ';' preprocessing tokens. */ 4835 unsigned char p = *peek++; 4836 4837 /* A character literal is ... single quotes, ... optionally preceded 4838 by u8, u, U, or L */ 4839 /* A string-literal is a ... double quotes, optionally prefixed by 4840 R, u8, u8R, u, uR, U, UR, L, or LR */ 4841 if (p == 'u') 4842 { 4843 peek = do_peek_next (peek, limit); 4844 if (*peek == '8') 4845 { 4846 peek++; 4847 goto peek_u8; 4848 } 4849 goto peek_u; 4850 } 4851 else if (p == 'U' || p == 'L') 4852 { 4853 peek_u8: 4854 peek = do_peek_next (peek, limit); 4855 peek_u: 4856 if (*peek == '\"' || *peek == '\'') 4857 return false; 4858 4859 if (*peek == 'R') 4860 goto peek_R; 4861 /* Identifier. Ok. */ 4862 } 4863 else if (p == 'R') 4864 { 4865 peek_R: 4866 if (CPP_OPTION (pfile, rliterals)) 4867 { 4868 peek = do_peek_next (peek, limit); 4869 if (*peek == '\"') 4870 return false; 4871 } 4872 /* Identifier. Ok. */ 4873 } 4874 else if ('Z' - 'A' == 25 4875 ? ((p >= 'A' && p <= 'Z') || (p >= 'a' && p <= 'z') || p == '_') 4876 : ISIDST (p)) 4877 { 4878 /* Identifier. Ok. */ 4879 } 4880 else if (p == '<') 4881 { 4882 /* Maybe angle header, ok for import. Reject 4883 '<=', '<<' digraph:'<:'. */ 4884 if (!import) 4885 return false; 4886 peek = do_peek_next (peek, limit); 4887 if (*peek == '=' || *peek == '<' 4888 || (*peek == ':' && CPP_OPTION (pfile, digraphs))) 4889 return false; 4890 } 4891 else if (p == ';') 4892 { 4893 /* SEMICOLON, ok for module. */ 4894 if (import) 4895 return false; 4896 } 4897 else if (p == '"') 4898 { 4899 /* STRING, ok for import. */ 4900 if (!import) 4901 return false; 4902 } 4903 else if (p == ':') 4904 { 4905 /* Maybe COLON, ok. Reject '::', digraph:':>'. */ 4906 peek = do_peek_next (peek, limit); 4907 if (*peek == ':' || (*peek == '>' && CPP_OPTION (pfile, digraphs))) 4908 return false; 4909 } 4910 else 4911 /* FIXME: Detect a unicode character, excluding those not 4912 permitted as the initial character. [lex.name]/1. I presume 4913 we need to check the \[uU] spellings, and directly using 4914 Unicode in say UTF8 form? Or perhaps we do the phase-1 4915 conversion of UTF8 to universal-character-names? */ 4916 return false; 4917 4918 return true; 4919} 4920 4921/* Directives-only scanning. Somewhat more relaxed than correct 4922 parsing -- some ill-formed programs will not be rejected. */ 4923 4924void 4925cpp_directive_only_process (cpp_reader *pfile, 4926 void *data, 4927 void (*cb) (cpp_reader *, CPP_DO_task, void *, ...)) 4928{ 4929 bool module_p = CPP_OPTION (pfile, module_directives); 4930 4931 do 4932 { 4933 restart: 4934 /* Buffer initialization, but no line cleaning. */ 4935 cpp_buffer *buffer = pfile->buffer; 4936 buffer->cur_note = buffer->notes_used = 0; 4937 buffer->cur = buffer->line_base = buffer->next_line; 4938 buffer->need_line = false; 4939 /* Files always end in a newline or carriage return. We rely on this for 4940 character peeking safety. */ 4941 gcc_assert (buffer->rlimit[0] == '\n' || buffer->rlimit[0] == '\r'); 4942 4943 const unsigned char *base = buffer->cur; 4944 unsigned line_count = 0; 4945 const unsigned char *line_start = base; 4946 4947 bool bol = true; 4948 bool raw = false; 4949 4950 const unsigned char *lwm = base; 4951 for (const unsigned char *pos = base, *limit = buffer->rlimit; 4952 pos < limit;) 4953 { 4954 unsigned char c = *pos++; 4955 /* This matches the switch in _cpp_lex_direct. */ 4956 switch (c) 4957 { 4958 case ' ': case '\t': case '\f': case '\v': 4959 /* Whitespace, do nothing. */ 4960 break; 4961 4962 case '\r': /* MAC line ending, or Windows \r\n */ 4963 if (*pos == '\n') 4964 pos++; 4965 /* FALLTHROUGH */ 4966 4967 case '\n': 4968 bol = true; 4969 4970 next_line: 4971 CPP_INCREMENT_LINE (pfile, 0); 4972 line_count++; 4973 line_start = pos; 4974 break; 4975 4976 case '\\': 4977 /* <backslash><newline> is removed, and doesn't undo any 4978 preceeding escape or whatnot. */ 4979 if (*pos == '\n') 4980 { 4981 pos++; 4982 goto next_line; 4983 } 4984 else if (*pos == '\r') 4985 { 4986 if (pos[1] == '\n') 4987 pos++; 4988 pos++; 4989 goto next_line; 4990 } 4991 goto dflt; 4992 4993 case '#': 4994 if (bol) 4995 { 4996 /* Line directive. */ 4997 if (pos - 1 > base && !pfile->state.skipping) 4998 cb (pfile, CPP_DO_print, data, 4999 line_count, base, pos - 1 - base); 5000 5001 /* Prep things for directive handling. */ 5002 buffer->next_line = pos; 5003 buffer->need_line = true; 5004 bool ok = _cpp_get_fresh_line (pfile); 5005 gcc_checking_assert (ok); 5006 5007 /* Ensure proper column numbering for generated 5008 error messages. */ 5009 buffer->line_base -= pos - line_start; 5010 5011 _cpp_handle_directive (pfile, line_start + 1 != pos); 5012 5013 /* Sanitize the line settings. Duplicate #include's can 5014 mess things up. */ 5015 // FIXME: Necessary? 5016 pfile->line_table->highest_location 5017 = pfile->line_table->highest_line; 5018 5019 if (!pfile->state.skipping 5020 && pfile->buffer->next_line < pfile->buffer->rlimit) 5021 cb (pfile, CPP_DO_location, data, 5022 pfile->line_table->highest_line); 5023 5024 goto restart; 5025 } 5026 goto dflt; 5027 5028 case '/': 5029 { 5030 const unsigned char *peek = do_peek_next (pos, limit); 5031 if (!(*peek == '/' || *peek == '*')) 5032 goto dflt; 5033 5034 /* Line or block comment */ 5035 bool is_block = *peek == '*'; 5036 bool star = false; 5037 bool esc = false; 5038 location_t sloc 5039 = linemap_position_for_column (pfile->line_table, 5040 pos - line_start); 5041 5042 while (pos < limit) 5043 { 5044 char c = *pos++; 5045 switch (c) 5046 { 5047 case '\\': 5048 esc = true; 5049 break; 5050 5051 case '\r': 5052 if (*pos == '\n') 5053 pos++; 5054 /* FALLTHROUGH */ 5055 5056 case '\n': 5057 { 5058 CPP_INCREMENT_LINE (pfile, 0); 5059 line_count++; 5060 line_start = pos; 5061 if (!esc && !is_block) 5062 { 5063 bol = true; 5064 goto done_comment; 5065 } 5066 } 5067 if (!esc) 5068 star = false; 5069 esc = false; 5070 break; 5071 5072 case '*': 5073 if (pos > peek) 5074 star = is_block; 5075 esc = false; 5076 break; 5077 5078 case '/': 5079 if (star) 5080 goto done_comment; 5081 /* FALLTHROUGH */ 5082 5083 default: 5084 star = false; 5085 esc = false; 5086 break; 5087 } 5088 } 5089 if (pos < limit || is_block) 5090 cpp_error_with_line (pfile, CPP_DL_ERROR, sloc, 0, 5091 "unterminated comment"); 5092 done_comment: 5093 lwm = pos; 5094 break; 5095 } 5096 5097 case '\'': 5098 if (!CPP_OPTION (pfile, digit_separators)) 5099 goto delimited_string; 5100 5101 /* Possibly a number punctuator. */ 5102 if (!ISIDNUM (*do_peek_next (pos, limit))) 5103 goto delimited_string; 5104 5105 goto quote_peek; 5106 5107 case '\"': 5108 if (!CPP_OPTION (pfile, rliterals)) 5109 goto delimited_string; 5110 5111 quote_peek: 5112 { 5113 /* For ' see if it's a number punctuator 5114 \.?<digit>(<digit>|<identifier-nondigit> 5115 |'<digit>|'<nondigit>|[eEpP]<sign>|\.)* */ 5116 /* For " see if it's a raw string 5117 {U,L,u,u8}R. This includes CPP_NUMBER detection, 5118 because that could be 0e+R. */ 5119 const unsigned char *peek = pos - 1; 5120 bool quote_first = c == '"'; 5121 bool quote_eight = false; 5122 bool maybe_number_start = false; 5123 bool want_number = false; 5124 5125 while ((peek = do_peek_prev (peek, lwm))) 5126 { 5127 unsigned char p = *peek; 5128 if (quote_first) 5129 { 5130 if (!raw) 5131 { 5132 if (p != 'R') 5133 break; 5134 raw = true; 5135 continue; 5136 } 5137 5138 quote_first = false; 5139 if (p == 'L' || p == 'U' || p == 'u') 5140 ; 5141 else if (p == '8') 5142 quote_eight = true; 5143 else 5144 goto second_raw; 5145 } 5146 else if (quote_eight) 5147 { 5148 if (p != 'u') 5149 { 5150 raw = false; 5151 break; 5152 } 5153 quote_eight = false; 5154 } 5155 else if (c == '"') 5156 { 5157 second_raw:; 5158 if (!want_number && ISIDNUM (p)) 5159 { 5160 raw = false; 5161 break; 5162 } 5163 } 5164 5165 if (ISDIGIT (p)) 5166 maybe_number_start = true; 5167 else if (p == '.') 5168 want_number = true; 5169 else if (ISIDNUM (p)) 5170 maybe_number_start = false; 5171 else if (p == '+' || p == '-') 5172 { 5173 if (const unsigned char *peek_prev 5174 = do_peek_prev (peek, lwm)) 5175 { 5176 p = *peek_prev; 5177 if (p == 'e' || p == 'E' 5178 || p == 'p' || p == 'P') 5179 { 5180 want_number = true; 5181 maybe_number_start = false; 5182 } 5183 else 5184 break; 5185 } 5186 else 5187 break; 5188 } 5189 else if (p == '\'' || p == '\"') 5190 { 5191 /* If this is lwm, this must be the end of a 5192 previous string. So this is a trailing 5193 literal type, (a) if those are allowed, 5194 and (b) maybe_start is false. Otherwise 5195 this must be a CPP_NUMBER because we've 5196 met another ', and we'd have checked that 5197 in its own right. */ 5198 if (peek == lwm && CPP_OPTION (pfile, uliterals)) 5199 { 5200 if (!maybe_number_start && !want_number) 5201 /* Must be a literal type. */ 5202 raw = false; 5203 } 5204 else if (p == '\'' 5205 && CPP_OPTION (pfile, digit_separators)) 5206 maybe_number_start = true; 5207 break; 5208 } 5209 else if (c == '\'') 5210 break; 5211 else if (!quote_first && !quote_eight) 5212 break; 5213 } 5214 5215 if (maybe_number_start) 5216 { 5217 if (c == '\'') 5218 /* A CPP NUMBER. */ 5219 goto dflt; 5220 raw = false; 5221 } 5222 5223 goto delimited_string; 5224 } 5225 5226 delimited_string: 5227 { 5228 /* (Possibly raw) string or char literal. */ 5229 unsigned char end = c; 5230 int delim_len = -1; 5231 const unsigned char *delim = NULL; 5232 location_t sloc = linemap_position_for_column (pfile->line_table, 5233 pos - line_start); 5234 int esc = 0; 5235 5236 if (raw) 5237 { 5238 /* There can be no line breaks in the delimiter. */ 5239 delim = pos; 5240 for (delim_len = 0; (c = *pos++) != '('; delim_len++) 5241 { 5242 if (delim_len == 16) 5243 { 5244 cpp_error_with_line (pfile, CPP_DL_ERROR, 5245 sloc, 0, 5246 "raw string delimiter" 5247 " longer than %d" 5248 " characters", 5249 delim_len); 5250 raw = false; 5251 pos = delim; 5252 break; 5253 } 5254 if (strchr (") \\\t\v\f\n", c)) 5255 { 5256 cpp_error_with_line (pfile, CPP_DL_ERROR, 5257 sloc, 0, 5258 "invalid character '%c'" 5259 " in raw string" 5260 " delimiter", c); 5261 raw = false; 5262 pos = delim; 5263 break; 5264 } 5265 if (pos >= limit) 5266 goto bad_string; 5267 } 5268 } 5269 5270 while (pos < limit) 5271 { 5272 char c = *pos++; 5273 switch (c) 5274 { 5275 case '\\': 5276 if (!raw) 5277 esc++; 5278 break; 5279 5280 case '\r': 5281 if (*pos == '\n') 5282 pos++; 5283 /* FALLTHROUGH */ 5284 5285 case '\n': 5286 { 5287 CPP_INCREMENT_LINE (pfile, 0); 5288 line_count++; 5289 line_start = pos; 5290 } 5291 if (esc) 5292 esc--; 5293 break; 5294 5295 case ')': 5296 if (raw 5297 && pos + delim_len + 1 < limit 5298 && pos[delim_len] == end 5299 && !memcmp (delim, pos, delim_len)) 5300 { 5301 pos += delim_len + 1; 5302 raw = false; 5303 goto done_string; 5304 } 5305 break; 5306 5307 default: 5308 if (!raw && !(esc & 1) && c == end) 5309 goto done_string; 5310 esc = 0; 5311 break; 5312 } 5313 } 5314 bad_string: 5315 cpp_error_with_line (pfile, CPP_DL_ERROR, sloc, 0, 5316 "unterminated literal"); 5317 5318 done_string: 5319 raw = false; 5320 lwm = pos - 1; 5321 } 5322 goto dflt; 5323 5324 case '_': 5325 case 'e': 5326 case 'i': 5327 case 'm': 5328 if (bol && module_p && !pfile->state.skipping 5329 && do_peek_module (pfile, c, pos, limit)) 5330 { 5331 /* We've seen the start of a module control line. 5332 Start up the tokenizer. */ 5333 pos--; /* Backup over the first character. */ 5334 5335 /* Backup over whitespace to start of line. */ 5336 while (pos > line_start 5337 && (pos[-1] == ' ' || pos[-1] == '\t')) 5338 pos--; 5339 5340 if (pos > base) 5341 cb (pfile, CPP_DO_print, data, line_count, base, pos - base); 5342 5343 /* Prep things for directive handling. */ 5344 buffer->next_line = pos; 5345 buffer->need_line = true; 5346 5347 /* Now get tokens until the PRAGMA_EOL. */ 5348 do 5349 { 5350 location_t spelling; 5351 const cpp_token *tok 5352 = cpp_get_token_with_location (pfile, &spelling); 5353 5354 gcc_assert (pfile->state.in_deferred_pragma 5355 || tok->type == CPP_PRAGMA_EOL); 5356 cb (pfile, CPP_DO_token, data, tok, spelling); 5357 } 5358 while (pfile->state.in_deferred_pragma); 5359 5360 if (pfile->buffer->next_line < pfile->buffer->rlimit) 5361 cb (pfile, CPP_DO_location, data, 5362 pfile->line_table->highest_line); 5363 5364 pfile->mi_valid = false; 5365 goto restart; 5366 } 5367 goto dflt; 5368 5369 default: 5370 dflt: 5371 bol = false; 5372 pfile->mi_valid = false; 5373 break; 5374 } 5375 } 5376 5377 if (buffer->rlimit > base && !pfile->state.skipping) 5378 { 5379 const unsigned char *limit = buffer->rlimit; 5380 /* If the file was not newline terminated, add rlimit, which is 5381 guaranteed to point to a newline, to the end of our range. */ 5382 if (limit[-1] != '\n') 5383 { 5384 limit++; 5385 CPP_INCREMENT_LINE (pfile, 0); 5386 line_count++; 5387 } 5388 cb (pfile, CPP_DO_print, data, line_count, base, limit - base); 5389 } 5390 5391 _cpp_pop_buffer (pfile); 5392 } 5393 while (pfile->buffer); 5394} 5395