Bernhard Rosenkraenzer | c83ebe5 | 2012-09-18 21:38:03 +0159 | [diff] [blame] | 1 | /* CPP Library - lexical analysis. |
| 2 | Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2007, 2008, 2009, 2010, |
| 3 | 2011, 2012 Free Software Foundation, Inc. |
| 4 | Contributed by Per Bothner, 1994-95. |
| 5 | Based on CCCP program by Paul Rubin, June 1986 |
| 6 | Adapted to ANSI C, Richard Stallman, Jan 1987 |
| 7 | Broken out to separate file, Zack Weinberg, Mar 2000 |
| 8 | |
| 9 | This program is free software; you can redistribute it and/or modify it |
| 10 | under the terms of the GNU General Public License as published by the |
| 11 | Free Software Foundation; either version 3, or (at your option) any |
| 12 | later version. |
| 13 | |
| 14 | This program is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with this program; see the file COPYING3. If not see |
| 21 | <http://www.gnu.org/licenses/>. */ |
| 22 | |
| 23 | #include "config.h" |
| 24 | #include "system.h" |
| 25 | #include "cpplib.h" |
| 26 | #include "internal.h" |
| 27 | |
| 28 | enum spell_type |
| 29 | { |
| 30 | SPELL_OPERATOR = 0, |
| 31 | SPELL_IDENT, |
| 32 | SPELL_LITERAL, |
| 33 | SPELL_NONE |
| 34 | }; |
| 35 | |
| 36 | struct token_spelling |
| 37 | { |
| 38 | enum spell_type category; |
| 39 | const unsigned char *name; |
| 40 | }; |
| 41 | |
| 42 | static const unsigned char *const digraph_spellings[] = |
| 43 | { UC"%:", UC"%:%:", UC"<:", UC":>", UC"<%", UC"%>" }; |
| 44 | |
| 45 | #define OP(e, s) { SPELL_OPERATOR, UC s }, |
| 46 | #define TK(e, s) { SPELL_ ## s, UC #e }, |
| 47 | static const struct token_spelling token_spellings[N_TTYPES] = { TTYPE_TABLE }; |
| 48 | #undef OP |
| 49 | #undef TK |
| 50 | |
| 51 | #define TOKEN_SPELL(token) (token_spellings[(token)->type].category) |
| 52 | #define TOKEN_NAME(token) (token_spellings[(token)->type].name) |
| 53 | |
| 54 | static void add_line_note (cpp_buffer *, const uchar *, unsigned int); |
| 55 | static int skip_line_comment (cpp_reader *); |
| 56 | static void skip_whitespace (cpp_reader *, cppchar_t); |
| 57 | static void lex_string (cpp_reader *, cpp_token *, const uchar *); |
| 58 | static void save_comment (cpp_reader *, cpp_token *, const uchar *, cppchar_t); |
| 59 | static void store_comment (cpp_reader *, cpp_token *); |
| 60 | static void create_literal (cpp_reader *, cpp_token *, const uchar *, |
| 61 | unsigned int, enum cpp_ttype); |
| 62 | static bool warn_in_comment (cpp_reader *, _cpp_line_note *); |
| 63 | static int name_p (cpp_reader *, const cpp_string *); |
| 64 | static tokenrun *next_tokenrun (tokenrun *); |
| 65 | |
| 66 | static _cpp_buff *new_buff (size_t); |
| 67 | |
| 68 | |
| 69 | /* Utility routine: |
| 70 | |
| 71 | Compares, the token TOKEN to the NUL-terminated string STRING. |
| 72 | TOKEN must be a CPP_NAME. Returns 1 for equal, 0 for unequal. */ |
| 73 | int |
| 74 | cpp_ideq (const cpp_token *token, const char *string) |
| 75 | { |
| 76 | if (token->type != CPP_NAME) |
| 77 | return 0; |
| 78 | |
| 79 | return !ustrcmp (NODE_NAME (token->val.node.node), (const uchar *) string); |
| 80 | } |
| 81 | |
| 82 | /* Record a note TYPE at byte POS into the current cleaned logical |
| 83 | line. */ |
| 84 | static void |
| 85 | add_line_note (cpp_buffer *buffer, const uchar *pos, unsigned int type) |
| 86 | { |
| 87 | if (buffer->notes_used == buffer->notes_cap) |
| 88 | { |
| 89 | buffer->notes_cap = buffer->notes_cap * 2 + 200; |
| 90 | buffer->notes = XRESIZEVEC (_cpp_line_note, buffer->notes, |
| 91 | buffer->notes_cap); |
| 92 | } |
| 93 | |
| 94 | buffer->notes[buffer->notes_used].pos = pos; |
| 95 | buffer->notes[buffer->notes_used].type = type; |
| 96 | buffer->notes_used++; |
| 97 | } |
| 98 | |
| 99 | |
| 100 | /* Fast path to find line special characters using optimized character |
| 101 | scanning algorithms. Anything complicated falls back to the slow |
| 102 | path below. Since this loop is very hot it's worth doing these kinds |
| 103 | of optimizations. |
| 104 | |
| 105 | One of the paths through the ifdefs should provide |
| 106 | |
| 107 | const uchar *search_line_fast (const uchar *s, const uchar *end); |
| 108 | |
| 109 | Between S and END, search for \n, \r, \\, ?. Return a pointer to |
| 110 | the found character. |
| 111 | |
| 112 | Note that the last character of the buffer is *always* a newline, |
| 113 | as forced by _cpp_convert_input. This fact can be used to avoid |
| 114 | explicitly looking for the end of the buffer. */ |
| 115 | |
| 116 | /* Configure gives us an ifdef test. */ |
| 117 | #ifndef WORDS_BIGENDIAN |
| 118 | #define WORDS_BIGENDIAN 0 |
| 119 | #endif |
| 120 | |
| 121 | /* We'd like the largest integer that fits into a register. There's nothing |
| 122 | in <stdint.h> that gives us that. For most hosts this is unsigned long, |
| 123 | but MS decided on an LLP64 model. Thankfully when building with GCC we |
| 124 | can get the "real" word size. */ |
| 125 | #ifdef __GNUC__ |
| 126 | typedef unsigned int word_type __attribute__((__mode__(__word__))); |
| 127 | #else |
| 128 | typedef unsigned long word_type; |
| 129 | #endif |
| 130 | |
| 131 | /* The code below is only expecting sizes 4 or 8. |
| 132 | Die at compile-time if this expectation is violated. */ |
| 133 | typedef char check_word_type_size |
| 134 | [(sizeof(word_type) == 8 || sizeof(word_type) == 4) * 2 - 1]; |
| 135 | |
| 136 | /* Return X with the first N bytes forced to values that won't match one |
| 137 | of the interesting characters. Note that NUL is not interesting. */ |
| 138 | |
| 139 | static inline word_type |
| 140 | acc_char_mask_misalign (word_type val, unsigned int n) |
| 141 | { |
| 142 | word_type mask = -1; |
| 143 | if (WORDS_BIGENDIAN) |
| 144 | mask >>= n * 8; |
| 145 | else |
| 146 | mask <<= n * 8; |
| 147 | return val & mask; |
| 148 | } |
| 149 | |
| 150 | /* Return X replicated to all byte positions within WORD_TYPE. */ |
| 151 | |
| 152 | static inline word_type |
| 153 | acc_char_replicate (uchar x) |
| 154 | { |
| 155 | word_type ret; |
| 156 | |
| 157 | ret = (x << 24) | (x << 16) | (x << 8) | x; |
| 158 | if (sizeof(word_type) == 8) |
| 159 | ret = (ret << 16 << 16) | ret; |
| 160 | return ret; |
| 161 | } |
| 162 | |
| 163 | /* Return non-zero if some byte of VAL is (probably) C. */ |
| 164 | |
| 165 | static inline word_type |
| 166 | acc_char_cmp (word_type val, word_type c) |
| 167 | { |
| 168 | #if defined(__GNUC__) && defined(__alpha__) |
| 169 | /* We can get exact results using a compare-bytes instruction. |
| 170 | Get (val == c) via (0 >= (val ^ c)). */ |
| 171 | return __builtin_alpha_cmpbge (0, val ^ c); |
| 172 | #else |
| 173 | word_type magic = 0x7efefefeU; |
| 174 | if (sizeof(word_type) == 8) |
| 175 | magic = (magic << 16 << 16) | 0xfefefefeU; |
| 176 | magic |= 1; |
| 177 | |
| 178 | val ^= c; |
| 179 | return ((val + magic) ^ ~val) & ~magic; |
| 180 | #endif |
| 181 | } |
| 182 | |
| 183 | /* Given the result of acc_char_cmp is non-zero, return the index of |
| 184 | the found character. If this was a false positive, return -1. */ |
| 185 | |
| 186 | static inline int |
| 187 | acc_char_index (word_type cmp ATTRIBUTE_UNUSED, |
| 188 | word_type val ATTRIBUTE_UNUSED) |
| 189 | { |
| 190 | #if defined(__GNUC__) && defined(__alpha__) && !WORDS_BIGENDIAN |
| 191 | /* The cmpbge instruction sets *bits* of the result corresponding to |
| 192 | matches in the bytes with no false positives. */ |
| 193 | return __builtin_ctzl (cmp); |
| 194 | #else |
| 195 | unsigned int i; |
| 196 | |
| 197 | /* ??? It would be nice to force unrolling here, |
| 198 | and have all of these constants folded. */ |
| 199 | for (i = 0; i < sizeof(word_type); ++i) |
| 200 | { |
| 201 | uchar c; |
| 202 | if (WORDS_BIGENDIAN) |
| 203 | c = (val >> (sizeof(word_type) - i - 1) * 8) & 0xff; |
| 204 | else |
| 205 | c = (val >> i * 8) & 0xff; |
| 206 | |
| 207 | if (c == '\n' || c == '\r' || c == '\\' || c == '?') |
| 208 | return i; |
| 209 | } |
| 210 | |
| 211 | return -1; |
| 212 | #endif |
| 213 | } |
| 214 | |
| 215 | /* A version of the fast scanner using bit fiddling techniques. |
| 216 | |
| 217 | For 32-bit words, one would normally perform 16 comparisons and |
| 218 | 16 branches. With this algorithm one performs 24 arithmetic |
| 219 | operations and one branch. Whether this is faster with a 32-bit |
| 220 | word size is going to be somewhat system dependent. |
| 221 | |
| 222 | For 64-bit words, we eliminate twice the number of comparisons |
| 223 | and branches without increasing the number of arithmetic operations. |
| 224 | It's almost certainly going to be a win with 64-bit word size. */ |
| 225 | |
| 226 | static const uchar * search_line_acc_char (const uchar *, const uchar *) |
| 227 | ATTRIBUTE_UNUSED; |
| 228 | |
| 229 | static const uchar * |
| 230 | search_line_acc_char (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) |
| 231 | { |
| 232 | const word_type repl_nl = acc_char_replicate ('\n'); |
| 233 | const word_type repl_cr = acc_char_replicate ('\r'); |
| 234 | const word_type repl_bs = acc_char_replicate ('\\'); |
| 235 | const word_type repl_qm = acc_char_replicate ('?'); |
| 236 | |
| 237 | unsigned int misalign; |
| 238 | const word_type *p; |
| 239 | word_type val, t; |
| 240 | |
| 241 | /* Align the buffer. Mask out any bytes from before the beginning. */ |
| 242 | p = (word_type *)((uintptr_t)s & -sizeof(word_type)); |
| 243 | val = *p; |
| 244 | misalign = (uintptr_t)s & (sizeof(word_type) - 1); |
| 245 | if (misalign) |
| 246 | val = acc_char_mask_misalign (val, misalign); |
| 247 | |
| 248 | /* Main loop. */ |
| 249 | while (1) |
| 250 | { |
| 251 | t = acc_char_cmp (val, repl_nl); |
| 252 | t |= acc_char_cmp (val, repl_cr); |
| 253 | t |= acc_char_cmp (val, repl_bs); |
| 254 | t |= acc_char_cmp (val, repl_qm); |
| 255 | |
| 256 | if (__builtin_expect (t != 0, 0)) |
| 257 | { |
| 258 | int i = acc_char_index (t, val); |
| 259 | if (i >= 0) |
| 260 | return (const uchar *)p + i; |
| 261 | } |
| 262 | |
| 263 | val = *++p; |
| 264 | } |
| 265 | } |
| 266 | |
| 267 | /* Disable on Solaris 2/x86 until the following problems can be properly |
| 268 | autoconfed: |
| 269 | |
| 270 | The Solaris 9 assembler cannot assemble SSE4.2 insns. |
| 271 | Before Solaris 9 Update 6, SSE insns cannot be executed. |
| 272 | The Solaris 10+ assembler tags objects with the instruction set |
| 273 | extensions used, so SSE4.2 executables cannot run on machines that |
| 274 | don't support that extension. */ |
| 275 | |
| 276 | #if (GCC_VERSION >= 4005) && (defined(__i386__) || defined(__x86_64__)) && !(defined(__sun__) && defined(__svr4__)) |
| 277 | |
| 278 | /* Replicated character data to be shared between implementations. |
| 279 | Recall that outside of a context with vector support we can't |
| 280 | define compatible vector types, therefore these are all defined |
| 281 | in terms of raw characters. */ |
| 282 | static const char repl_chars[4][16] __attribute__((aligned(16))) = { |
| 283 | { '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', |
| 284 | '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' }, |
| 285 | { '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', |
| 286 | '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' }, |
| 287 | { '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', |
| 288 | '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' }, |
| 289 | { '?', '?', '?', '?', '?', '?', '?', '?', |
| 290 | '?', '?', '?', '?', '?', '?', '?', '?' }, |
| 291 | }; |
| 292 | |
| 293 | /* A version of the fast scanner using MMX vectorized byte compare insns. |
| 294 | |
| 295 | This uses the PMOVMSKB instruction which was introduced with "MMX2", |
| 296 | which was packaged into SSE1; it is also present in the AMD MMX |
| 297 | extension. Mark the function as using "sse" so that we emit a real |
| 298 | "emms" instruction, rather than the 3dNOW "femms" instruction. */ |
| 299 | |
| 300 | static const uchar * |
| 301 | #ifndef __SSE__ |
| 302 | __attribute__((__target__("sse"))) |
| 303 | #endif |
| 304 | search_line_mmx (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) |
| 305 | { |
| 306 | typedef char v8qi __attribute__ ((__vector_size__ (8))); |
| 307 | typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__)); |
| 308 | |
| 309 | const v8qi repl_nl = *(const v8qi *)repl_chars[0]; |
| 310 | const v8qi repl_cr = *(const v8qi *)repl_chars[1]; |
| 311 | const v8qi repl_bs = *(const v8qi *)repl_chars[2]; |
| 312 | const v8qi repl_qm = *(const v8qi *)repl_chars[3]; |
| 313 | |
| 314 | unsigned int misalign, found, mask; |
| 315 | const v8qi *p; |
| 316 | v8qi data, t, c; |
| 317 | |
| 318 | /* Align the source pointer. While MMX doesn't generate unaligned data |
| 319 | faults, this allows us to safely scan to the end of the buffer without |
| 320 | reading beyond the end of the last page. */ |
| 321 | misalign = (uintptr_t)s & 7; |
| 322 | p = (const v8qi *)((uintptr_t)s & -8); |
| 323 | data = *p; |
| 324 | |
| 325 | /* Create a mask for the bytes that are valid within the first |
| 326 | 16-byte block. The Idea here is that the AND with the mask |
| 327 | within the loop is "free", since we need some AND or TEST |
| 328 | insn in order to set the flags for the branch anyway. */ |
| 329 | mask = -1u << misalign; |
| 330 | |
| 331 | /* Main loop processing 8 bytes at a time. */ |
| 332 | goto start; |
| 333 | do |
| 334 | { |
| 335 | data = *++p; |
| 336 | mask = -1; |
| 337 | |
| 338 | start: |
| 339 | t = __builtin_ia32_pcmpeqb(data, repl_nl); |
| 340 | c = __builtin_ia32_pcmpeqb(data, repl_cr); |
| 341 | t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); |
| 342 | c = __builtin_ia32_pcmpeqb(data, repl_bs); |
| 343 | t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); |
| 344 | c = __builtin_ia32_pcmpeqb(data, repl_qm); |
| 345 | t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c); |
| 346 | found = __builtin_ia32_pmovmskb (t); |
| 347 | found &= mask; |
| 348 | } |
| 349 | while (!found); |
| 350 | |
| 351 | __builtin_ia32_emms (); |
| 352 | |
| 353 | /* FOUND contains 1 in bits for which we matched a relevant |
| 354 | character. Conversion to the byte index is trivial. */ |
| 355 | found = __builtin_ctz(found); |
| 356 | return (const uchar *)p + found; |
| 357 | } |
| 358 | |
| 359 | /* A version of the fast scanner using SSE2 vectorized byte compare insns. */ |
| 360 | |
| 361 | static const uchar * |
| 362 | #ifndef __SSE2__ |
| 363 | __attribute__((__target__("sse2"))) |
| 364 | #endif |
| 365 | search_line_sse2 (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) |
| 366 | { |
| 367 | typedef char v16qi __attribute__ ((__vector_size__ (16))); |
| 368 | |
| 369 | const v16qi repl_nl = *(const v16qi *)repl_chars[0]; |
| 370 | const v16qi repl_cr = *(const v16qi *)repl_chars[1]; |
| 371 | const v16qi repl_bs = *(const v16qi *)repl_chars[2]; |
| 372 | const v16qi repl_qm = *(const v16qi *)repl_chars[3]; |
| 373 | |
| 374 | unsigned int misalign, found, mask; |
| 375 | const v16qi *p; |
| 376 | v16qi data, t; |
| 377 | |
| 378 | /* Align the source pointer. */ |
| 379 | misalign = (uintptr_t)s & 15; |
| 380 | p = (const v16qi *)((uintptr_t)s & -16); |
| 381 | data = *p; |
| 382 | |
| 383 | /* Create a mask for the bytes that are valid within the first |
| 384 | 16-byte block. The Idea here is that the AND with the mask |
| 385 | within the loop is "free", since we need some AND or TEST |
| 386 | insn in order to set the flags for the branch anyway. */ |
| 387 | mask = -1u << misalign; |
| 388 | |
| 389 | /* Main loop processing 16 bytes at a time. */ |
| 390 | goto start; |
| 391 | do |
| 392 | { |
| 393 | data = *++p; |
| 394 | mask = -1; |
| 395 | |
| 396 | start: |
| 397 | t = __builtin_ia32_pcmpeqb128(data, repl_nl); |
| 398 | t |= __builtin_ia32_pcmpeqb128(data, repl_cr); |
| 399 | t |= __builtin_ia32_pcmpeqb128(data, repl_bs); |
| 400 | t |= __builtin_ia32_pcmpeqb128(data, repl_qm); |
| 401 | found = __builtin_ia32_pmovmskb128 (t); |
| 402 | found &= mask; |
| 403 | } |
| 404 | while (!found); |
| 405 | |
| 406 | /* FOUND contains 1 in bits for which we matched a relevant |
| 407 | character. Conversion to the byte index is trivial. */ |
| 408 | found = __builtin_ctz(found); |
| 409 | return (const uchar *)p + found; |
| 410 | } |
| 411 | |
| 412 | #ifdef HAVE_SSE4 |
| 413 | /* A version of the fast scanner using SSE 4.2 vectorized string insns. */ |
| 414 | |
| 415 | static const uchar * |
| 416 | #ifndef __SSE4_2__ |
| 417 | __attribute__((__target__("sse4.2"))) |
| 418 | #endif |
| 419 | search_line_sse42 (const uchar *s, const uchar *end) |
| 420 | { |
| 421 | typedef char v16qi __attribute__ ((__vector_size__ (16))); |
| 422 | static const v16qi search = { '\n', '\r', '?', '\\' }; |
| 423 | |
| 424 | uintptr_t si = (uintptr_t)s; |
| 425 | uintptr_t index; |
| 426 | |
| 427 | /* Check for unaligned input. */ |
| 428 | if (si & 15) |
| 429 | { |
| 430 | v16qi sv; |
| 431 | |
| 432 | if (__builtin_expect (end - s < 16, 0) |
| 433 | && __builtin_expect ((si & 0xfff) > 0xff0, 0)) |
| 434 | { |
| 435 | /* There are less than 16 bytes left in the buffer, and less |
| 436 | than 16 bytes left on the page. Reading 16 bytes at this |
| 437 | point might generate a spurious page fault. Defer to the |
| 438 | SSE2 implementation, which already handles alignment. */ |
| 439 | return search_line_sse2 (s, end); |
| 440 | } |
| 441 | |
| 442 | /* ??? The builtin doesn't understand that the PCMPESTRI read from |
| 443 | memory need not be aligned. */ |
| 444 | sv = __builtin_ia32_loaddqu ((const char *) s); |
| 445 | index = __builtin_ia32_pcmpestri128 (search, 4, sv, 16, 0); |
| 446 | |
| 447 | if (__builtin_expect (index < 16, 0)) |
| 448 | goto found; |
| 449 | |
| 450 | /* Advance the pointer to an aligned address. We will re-scan a |
| 451 | few bytes, but we no longer need care for reading past the |
| 452 | end of a page, since we're guaranteed a match. */ |
| 453 | s = (const uchar *)((si + 16) & -16); |
| 454 | } |
| 455 | |
| 456 | /* Main loop, processing 16 bytes at a time. By doing the whole loop |
| 457 | in inline assembly, we can make proper use of the flags set. */ |
| 458 | __asm ( "sub $16, %1\n" |
| 459 | " .balign 16\n" |
| 460 | "0: add $16, %1\n" |
| 461 | " %vpcmpestri $0, (%1), %2\n" |
| 462 | " jnc 0b" |
| 463 | : "=&c"(index), "+r"(s) |
| 464 | : "x"(search), "a"(4), "d"(16)); |
| 465 | |
| 466 | found: |
| 467 | return s + index; |
| 468 | } |
| 469 | |
| 470 | #else |
| 471 | /* Work around out-dated assemblers without sse4 support. */ |
| 472 | #define search_line_sse42 search_line_sse2 |
| 473 | #endif |
| 474 | |
| 475 | /* Check the CPU capabilities. */ |
| 476 | |
| 477 | #include "../gcc/config/i386/cpuid.h" |
| 478 | |
| 479 | typedef const uchar * (*search_line_fast_type) (const uchar *, const uchar *); |
| 480 | static search_line_fast_type search_line_fast; |
| 481 | |
| 482 | #define HAVE_init_vectorized_lexer 1 |
| 483 | static inline void |
| 484 | init_vectorized_lexer (void) |
| 485 | { |
| 486 | unsigned dummy, ecx = 0, edx = 0; |
| 487 | search_line_fast_type impl = search_line_acc_char; |
| 488 | int minimum = 0; |
| 489 | |
| 490 | #if defined(__SSE4_2__) |
| 491 | minimum = 3; |
| 492 | #elif defined(__SSE2__) |
| 493 | minimum = 2; |
| 494 | #elif defined(__SSE__) |
| 495 | minimum = 1; |
| 496 | #endif |
| 497 | |
| 498 | if (minimum == 3) |
| 499 | impl = search_line_sse42; |
| 500 | else if (__get_cpuid (1, &dummy, &dummy, &ecx, &edx) || minimum == 2) |
| 501 | { |
| 502 | if (minimum == 3 || (ecx & bit_SSE4_2)) |
| 503 | impl = search_line_sse42; |
| 504 | else if (minimum == 2 || (edx & bit_SSE2)) |
| 505 | impl = search_line_sse2; |
| 506 | else if (minimum == 1 || (edx & bit_SSE)) |
| 507 | impl = search_line_mmx; |
| 508 | } |
| 509 | else if (__get_cpuid (0x80000001, &dummy, &dummy, &dummy, &edx)) |
| 510 | { |
| 511 | if (minimum == 1 |
| 512 | || (edx & (bit_MMXEXT | bit_CMOV)) == (bit_MMXEXT | bit_CMOV)) |
| 513 | impl = search_line_mmx; |
| 514 | } |
| 515 | |
| 516 | search_line_fast = impl; |
| 517 | } |
| 518 | |
| 519 | #elif (GCC_VERSION >= 4005) && defined(__ALTIVEC__) |
| 520 | |
| 521 | /* A vection of the fast scanner using AltiVec vectorized byte compares. */ |
| 522 | /* ??? Unfortunately, attribute(target("altivec")) is not yet supported, |
| 523 | so we can't compile this function without -maltivec on the command line |
| 524 | (or implied by some other switch). */ |
| 525 | |
| 526 | static const uchar * |
| 527 | search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) |
| 528 | { |
| 529 | typedef __attribute__((altivec(vector))) unsigned char vc; |
| 530 | |
| 531 | const vc repl_nl = { |
| 532 | '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n', |
| 533 | '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' |
| 534 | }; |
| 535 | const vc repl_cr = { |
| 536 | '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r', |
| 537 | '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' |
| 538 | }; |
| 539 | const vc repl_bs = { |
| 540 | '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\', |
| 541 | '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' |
| 542 | }; |
| 543 | const vc repl_qm = { |
| 544 | '?', '?', '?', '?', '?', '?', '?', '?', |
| 545 | '?', '?', '?', '?', '?', '?', '?', '?', |
| 546 | }; |
| 547 | const vc ones = { |
| 548 | -1, -1, -1, -1, -1, -1, -1, -1, |
| 549 | -1, -1, -1, -1, -1, -1, -1, -1, |
| 550 | }; |
| 551 | const vc zero = { 0 }; |
| 552 | |
| 553 | vc data, mask, t; |
| 554 | |
| 555 | /* Altivec loads automatically mask addresses with -16. This lets us |
| 556 | issue the first load as early as possible. */ |
| 557 | data = __builtin_vec_ld(0, (const vc *)s); |
| 558 | |
| 559 | /* Discard bytes before the beginning of the buffer. Do this by |
| 560 | beginning with all ones and shifting in zeros according to the |
| 561 | mis-alignment. The LVSR instruction pulls the exact shift we |
| 562 | want from the address. */ |
| 563 | mask = __builtin_vec_lvsr(0, s); |
| 564 | mask = __builtin_vec_perm(zero, ones, mask); |
| 565 | data &= mask; |
| 566 | |
| 567 | /* While altivec loads mask addresses, we still need to align S so |
| 568 | that the offset we compute at the end is correct. */ |
| 569 | s = (const uchar *)((uintptr_t)s & -16); |
| 570 | |
| 571 | /* Main loop processing 16 bytes at a time. */ |
| 572 | goto start; |
| 573 | do |
| 574 | { |
| 575 | vc m_nl, m_cr, m_bs, m_qm; |
| 576 | |
| 577 | s += 16; |
| 578 | data = __builtin_vec_ld(0, (const vc *)s); |
| 579 | |
| 580 | start: |
| 581 | m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl); |
| 582 | m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr); |
| 583 | m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs); |
| 584 | m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm); |
| 585 | t = (m_nl | m_cr) | (m_bs | m_qm); |
| 586 | |
| 587 | /* T now contains 0xff in bytes for which we matched one of the relevant |
| 588 | characters. We want to exit the loop if any byte in T is non-zero. |
| 589 | Below is the expansion of vec_any_ne(t, zero). */ |
| 590 | } |
| 591 | while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero)); |
| 592 | |
| 593 | { |
| 594 | #define N (sizeof(vc) / sizeof(long)) |
| 595 | |
| 596 | union { |
| 597 | vc v; |
| 598 | /* Statically assert that N is 2 or 4. */ |
| 599 | unsigned long l[(N == 2 || N == 4) ? N : -1]; |
| 600 | } u; |
| 601 | unsigned long l, i = 0; |
| 602 | |
| 603 | u.v = t; |
| 604 | |
| 605 | /* Find the first word of T that is non-zero. */ |
| 606 | switch (N) |
| 607 | { |
| 608 | case 4: |
| 609 | l = u.l[i++]; |
| 610 | if (l != 0) |
| 611 | break; |
| 612 | s += sizeof(unsigned long); |
| 613 | l = u.l[i++]; |
| 614 | if (l != 0) |
| 615 | break; |
| 616 | s += sizeof(unsigned long); |
| 617 | case 2: |
| 618 | l = u.l[i++]; |
| 619 | if (l != 0) |
| 620 | break; |
| 621 | s += sizeof(unsigned long); |
| 622 | l = u.l[i]; |
| 623 | } |
| 624 | |
| 625 | /* L now contains 0xff in bytes for which we matched one of the |
| 626 | relevant characters. We can find the byte index by finding |
| 627 | its bit index and dividing by 8. */ |
| 628 | l = __builtin_clzl(l) >> 3; |
| 629 | return s + l; |
| 630 | |
| 631 | #undef N |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | #elif defined (__ARM_NEON__) |
| 636 | #include "arm_neon.h" |
| 637 | |
| 638 | static const uchar * |
| 639 | search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED) |
| 640 | { |
| 641 | const uint8x16_t repl_nl = vdupq_n_u8 ('\n'); |
| 642 | const uint8x16_t repl_cr = vdupq_n_u8 ('\r'); |
| 643 | const uint8x16_t repl_bs = vdupq_n_u8 ('\\'); |
| 644 | const uint8x16_t repl_qm = vdupq_n_u8 ('?'); |
| 645 | const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL); |
| 646 | |
| 647 | unsigned int misalign, found, mask; |
| 648 | const uint8_t *p; |
| 649 | uint8x16_t data; |
| 650 | |
| 651 | /* Align the source pointer. */ |
| 652 | misalign = (uintptr_t)s & 15; |
| 653 | p = (const uint8_t *)((uintptr_t)s & -16); |
| 654 | data = vld1q_u8 (p); |
| 655 | |
| 656 | /* Create a mask for the bytes that are valid within the first |
| 657 | 16-byte block. The Idea here is that the AND with the mask |
| 658 | within the loop is "free", since we need some AND or TEST |
| 659 | insn in order to set the flags for the branch anyway. */ |
| 660 | mask = (-1u << misalign) & 0xffff; |
| 661 | |
| 662 | /* Main loop, processing 16 bytes at a time. */ |
| 663 | goto start; |
| 664 | |
| 665 | do |
| 666 | { |
| 667 | uint8x8_t l; |
| 668 | uint16x4_t m; |
| 669 | uint32x2_t n; |
| 670 | uint8x16_t t, u, v, w; |
| 671 | |
| 672 | p += 16; |
| 673 | data = vld1q_u8 (p); |
| 674 | mask = 0xffff; |
| 675 | |
| 676 | start: |
| 677 | t = vceqq_u8 (data, repl_nl); |
| 678 | u = vceqq_u8 (data, repl_cr); |
| 679 | v = vorrq_u8 (t, vceqq_u8 (data, repl_bs)); |
| 680 | w = vorrq_u8 (u, vceqq_u8 (data, repl_qm)); |
| 681 | t = vandq_u8 (vorrq_u8 (v, w), xmask); |
| 682 | l = vpadd_u8 (vget_low_u8 (t), vget_high_u8 (t)); |
| 683 | m = vpaddl_u8 (l); |
| 684 | n = vpaddl_u16 (m); |
| 685 | |
| 686 | found = vget_lane_u32 ((uint32x2_t) vorr_u64 ((uint64x1_t) n, |
| 687 | vshr_n_u64 ((uint64x1_t) n, 24)), 0); |
| 688 | found &= mask; |
| 689 | } |
| 690 | while (!found); |
| 691 | |
| 692 | /* FOUND contains 1 in bits for which we matched a relevant |
| 693 | character. Conversion to the byte index is trivial. */ |
| 694 | found = __builtin_ctz (found); |
| 695 | return (const uchar *)p + found; |
| 696 | } |
| 697 | |
| 698 | #else |
| 699 | |
| 700 | /* We only have one accellerated alternative. Use a direct call so that |
| 701 | we encourage inlining. */ |
| 702 | |
| 703 | #define search_line_fast search_line_acc_char |
| 704 | |
| 705 | #endif |
| 706 | |
| 707 | /* Initialize the lexer if needed. */ |
| 708 | |
| 709 | void |
| 710 | _cpp_init_lexer (void) |
| 711 | { |
| 712 | #ifdef HAVE_init_vectorized_lexer |
| 713 | init_vectorized_lexer (); |
| 714 | #endif |
| 715 | } |
| 716 | |
| 717 | /* Returns with a logical line that contains no escaped newlines or |
| 718 | trigraphs. This is a time-critical inner loop. */ |
| 719 | void |
| 720 | _cpp_clean_line (cpp_reader *pfile) |
| 721 | { |
| 722 | cpp_buffer *buffer; |
| 723 | const uchar *s; |
| 724 | uchar c, *d, *p; |
| 725 | |
| 726 | buffer = pfile->buffer; |
| 727 | buffer->cur_note = buffer->notes_used = 0; |
| 728 | buffer->cur = buffer->line_base = buffer->next_line; |
| 729 | buffer->need_line = false; |
| 730 | s = buffer->next_line; |
| 731 | |
| 732 | if (!buffer->from_stage3) |
| 733 | { |
| 734 | const uchar *pbackslash = NULL; |
| 735 | |
| 736 | /* Fast path. This is the common case of an un-escaped line with |
| 737 | no trigraphs. The primary win here is by not writing any |
| 738 | data back to memory until we have to. */ |
| 739 | while (1) |
| 740 | { |
| 741 | /* Perform an optimized search for \n, \r, \\, ?. */ |
| 742 | s = search_line_fast (s, buffer->rlimit); |
| 743 | |
| 744 | c = *s; |
| 745 | if (c == '\\') |
| 746 | { |
| 747 | /* Record the location of the backslash and continue. */ |
| 748 | pbackslash = s++; |
| 749 | } |
| 750 | else if (__builtin_expect (c == '?', 0)) |
| 751 | { |
| 752 | if (__builtin_expect (s[1] == '?', false) |
| 753 | && _cpp_trigraph_map[s[2]]) |
| 754 | { |
| 755 | /* Have a trigraph. We may or may not have to convert |
| 756 | it. Add a line note regardless, for -Wtrigraphs. */ |
| 757 | add_line_note (buffer, s, s[2]); |
| 758 | if (CPP_OPTION (pfile, trigraphs)) |
| 759 | { |
| 760 | /* We do, and that means we have to switch to the |
| 761 | slow path. */ |
| 762 | d = (uchar *) s; |
| 763 | *d = _cpp_trigraph_map[s[2]]; |
| 764 | s += 2; |
| 765 | goto slow_path; |
| 766 | } |
| 767 | } |
| 768 | /* Not a trigraph. Continue on fast-path. */ |
| 769 | s++; |
| 770 | } |
| 771 | else |
| 772 | break; |
| 773 | } |
| 774 | |
| 775 | /* This must be \r or \n. We're either done, or we'll be forced |
| 776 | to write back to the buffer and continue on the slow path. */ |
| 777 | d = (uchar *) s; |
| 778 | |
| 779 | if (__builtin_expect (s == buffer->rlimit, false)) |
| 780 | goto done; |
| 781 | |
| 782 | /* DOS line ending? */ |
| 783 | if (__builtin_expect (c == '\r', false) && s[1] == '\n') |
| 784 | { |
| 785 | s++; |
| 786 | if (s == buffer->rlimit) |
| 787 | goto done; |
| 788 | } |
| 789 | |
| 790 | if (__builtin_expect (pbackslash == NULL, true)) |
| 791 | goto done; |
| 792 | |
| 793 | /* Check for escaped newline. */ |
| 794 | p = d; |
| 795 | while (is_nvspace (p[-1])) |
| 796 | p--; |
| 797 | if (p - 1 != pbackslash) |
| 798 | goto done; |
| 799 | |
| 800 | /* Have an escaped newline; process it and proceed to |
| 801 | the slow path. */ |
| 802 | add_line_note (buffer, p - 1, p != d ? ' ' : '\\'); |
| 803 | d = p - 2; |
| 804 | buffer->next_line = p - 1; |
| 805 | |
| 806 | slow_path: |
| 807 | while (1) |
| 808 | { |
| 809 | c = *++s; |
| 810 | *++d = c; |
| 811 | |
| 812 | if (c == '\n' || c == '\r') |
| 813 | { |
| 814 | /* Handle DOS line endings. */ |
| 815 | if (c == '\r' && s != buffer->rlimit && s[1] == '\n') |
| 816 | s++; |
| 817 | if (s == buffer->rlimit) |
| 818 | break; |
| 819 | |
| 820 | /* Escaped? */ |
| 821 | p = d; |
| 822 | while (p != buffer->next_line && is_nvspace (p[-1])) |
| 823 | p--; |
| 824 | if (p == buffer->next_line || p[-1] != '\\') |
| 825 | break; |
| 826 | |
| 827 | add_line_note (buffer, p - 1, p != d ? ' ': '\\'); |
| 828 | d = p - 2; |
| 829 | buffer->next_line = p - 1; |
| 830 | } |
| 831 | else if (c == '?' && s[1] == '?' && _cpp_trigraph_map[s[2]]) |
| 832 | { |
| 833 | /* Add a note regardless, for the benefit of -Wtrigraphs. */ |
| 834 | add_line_note (buffer, d, s[2]); |
| 835 | if (CPP_OPTION (pfile, trigraphs)) |
| 836 | { |
| 837 | *d = _cpp_trigraph_map[s[2]]; |
| 838 | s += 2; |
| 839 | } |
| 840 | } |
| 841 | } |
| 842 | } |
| 843 | else |
| 844 | { |
| 845 | while (*s != '\n' && *s != '\r') |
| 846 | s++; |
| 847 | d = (uchar *) s; |
| 848 | |
| 849 | /* Handle DOS line endings. */ |
| 850 | if (*s == '\r' && s != buffer->rlimit && s[1] == '\n') |
| 851 | s++; |
| 852 | } |
| 853 | |
| 854 | done: |
| 855 | *d = '\n'; |
| 856 | /* A sentinel note that should never be processed. */ |
| 857 | add_line_note (buffer, d + 1, '\n'); |
| 858 | buffer->next_line = s + 1; |
| 859 | } |
| 860 | |
| 861 | /* Return true if the trigraph indicated by NOTE should be warned |
| 862 | about in a comment. */ |
| 863 | static bool |
| 864 | warn_in_comment (cpp_reader *pfile, _cpp_line_note *note) |
| 865 | { |
| 866 | const uchar *p; |
| 867 | |
| 868 | /* Within comments we don't warn about trigraphs, unless the |
| 869 | trigraph forms an escaped newline, as that may change |
| 870 | behavior. */ |
| 871 | if (note->type != '/') |
| 872 | return false; |
| 873 | |
| 874 | /* If -trigraphs, then this was an escaped newline iff the next note |
| 875 | is coincident. */ |
| 876 | if (CPP_OPTION (pfile, trigraphs)) |
| 877 | return note[1].pos == note->pos; |
| 878 | |
| 879 | /* Otherwise, see if this forms an escaped newline. */ |
| 880 | p = note->pos + 3; |
| 881 | while (is_nvspace (*p)) |
| 882 | p++; |
| 883 | |
| 884 | /* There might have been escaped newlines between the trigraph and the |
| 885 | newline we found. Hence the position test. */ |
| 886 | return (*p == '\n' && p < note[1].pos); |
| 887 | } |
| 888 | |
| 889 | /* Process the notes created by add_line_note as far as the current |
| 890 | location. */ |
| 891 | void |
| 892 | _cpp_process_line_notes (cpp_reader *pfile, int in_comment) |
| 893 | { |
| 894 | cpp_buffer *buffer = pfile->buffer; |
| 895 | |
| 896 | for (;;) |
| 897 | { |
| 898 | _cpp_line_note *note = &buffer->notes[buffer->cur_note]; |
| 899 | unsigned int col; |
| 900 | |
| 901 | if (note->pos > buffer->cur) |
| 902 | break; |
| 903 | |
| 904 | buffer->cur_note++; |
| 905 | col = CPP_BUF_COLUMN (buffer, note->pos + 1); |
| 906 | |
| 907 | if (note->type == '\\' || note->type == ' ') |
| 908 | { |
| 909 | if (note->type == ' ' && !in_comment) |
| 910 | cpp_error_with_line (pfile, CPP_DL_WARNING, pfile->line_table->highest_line, col, |
| 911 | "backslash and newline separated by space"); |
| 912 | |
| 913 | if (buffer->next_line > buffer->rlimit) |
| 914 | { |
| 915 | cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, col, |
| 916 | "backslash-newline at end of file"); |
| 917 | /* Prevent "no newline at end of file" warning. */ |
| 918 | buffer->next_line = buffer->rlimit; |
| 919 | } |
| 920 | |
| 921 | buffer->line_base = note->pos; |
| 922 | CPP_INCREMENT_LINE (pfile, 0); |
| 923 | } |
| 924 | else if (_cpp_trigraph_map[note->type]) |
| 925 | { |
| 926 | if (CPP_OPTION (pfile, warn_trigraphs) |
| 927 | && (!in_comment || warn_in_comment (pfile, note))) |
| 928 | { |
| 929 | if (CPP_OPTION (pfile, trigraphs)) |
| 930 | cpp_warning_with_line (pfile, CPP_W_TRIGRAPHS, |
| 931 | pfile->line_table->highest_line, col, |
| 932 | "trigraph ??%c converted to %c", |
| 933 | note->type, |
| 934 | (int) _cpp_trigraph_map[note->type]); |
| 935 | else |
| 936 | { |
| 937 | cpp_warning_with_line |
| 938 | (pfile, CPP_W_TRIGRAPHS, |
| 939 | pfile->line_table->highest_line, col, |
| 940 | "trigraph ??%c ignored, use -trigraphs to enable", |
| 941 | note->type); |
| 942 | } |
| 943 | } |
| 944 | } |
| 945 | else if (note->type == 0) |
| 946 | /* Already processed in lex_raw_string. */; |
| 947 | else |
| 948 | abort (); |
| 949 | } |
| 950 | } |
| 951 | |
| 952 | /* Skip a C-style block comment. We find the end of the comment by |
| 953 | seeing if an asterisk is before every '/' we encounter. Returns |
| 954 | nonzero if comment terminated by EOF, zero otherwise. |
| 955 | |
| 956 | Buffer->cur points to the initial asterisk of the comment. */ |
| 957 | bool |
| 958 | _cpp_skip_block_comment (cpp_reader *pfile) |
| 959 | { |
| 960 | cpp_buffer *buffer = pfile->buffer; |
| 961 | const uchar *cur = buffer->cur; |
| 962 | uchar c; |
| 963 | |
| 964 | cur++; |
| 965 | if (*cur == '/') |
| 966 | cur++; |
| 967 | |
| 968 | for (;;) |
| 969 | { |
| 970 | /* People like decorating comments with '*', so check for '/' |
| 971 | instead for efficiency. */ |
| 972 | c = *cur++; |
| 973 | |
| 974 | if (c == '/') |
| 975 | { |
| 976 | if (cur[-2] == '*') |
| 977 | break; |
| 978 | |
| 979 | /* Warn about potential nested comments, but not if the '/' |
| 980 | comes immediately before the true comment delimiter. |
| 981 | Don't bother to get it right across escaped newlines. */ |
| 982 | if (CPP_OPTION (pfile, warn_comments) |
| 983 | && cur[0] == '*' && cur[1] != '/') |
| 984 | { |
| 985 | buffer->cur = cur; |
| 986 | cpp_warning_with_line (pfile, CPP_W_COMMENTS, |
| 987 | pfile->line_table->highest_line, |
| 988 | CPP_BUF_COL (buffer), |
| 989 | "\"/*\" within comment"); |
| 990 | } |
| 991 | } |
| 992 | else if (c == '\n') |
| 993 | { |
| 994 | unsigned int cols; |
| 995 | buffer->cur = cur - 1; |
| 996 | _cpp_process_line_notes (pfile, true); |
| 997 | if (buffer->next_line >= buffer->rlimit) |
| 998 | return true; |
| 999 | _cpp_clean_line (pfile); |
| 1000 | |
| 1001 | cols = buffer->next_line - buffer->line_base; |
| 1002 | CPP_INCREMENT_LINE (pfile, cols); |
| 1003 | |
| 1004 | cur = buffer->cur; |
| 1005 | } |
| 1006 | } |
| 1007 | |
| 1008 | buffer->cur = cur; |
| 1009 | _cpp_process_line_notes (pfile, true); |
| 1010 | return false; |
| 1011 | } |
| 1012 | |
| 1013 | /* Skip a C++ line comment, leaving buffer->cur pointing to the |
| 1014 | terminating newline. Handles escaped newlines. Returns nonzero |
| 1015 | if a multiline comment. */ |
| 1016 | static int |
| 1017 | skip_line_comment (cpp_reader *pfile) |
| 1018 | { |
| 1019 | cpp_buffer *buffer = pfile->buffer; |
| 1020 | source_location orig_line = pfile->line_table->highest_line; |
| 1021 | |
| 1022 | while (*buffer->cur != '\n') |
| 1023 | buffer->cur++; |
| 1024 | |
| 1025 | _cpp_process_line_notes (pfile, true); |
| 1026 | return orig_line != pfile->line_table->highest_line; |
| 1027 | } |
| 1028 | |
| 1029 | /* Skips whitespace, saving the next non-whitespace character. */ |
| 1030 | static void |
| 1031 | skip_whitespace (cpp_reader *pfile, cppchar_t c) |
| 1032 | { |
| 1033 | cpp_buffer *buffer = pfile->buffer; |
| 1034 | bool saw_NUL = false; |
| 1035 | |
| 1036 | do |
| 1037 | { |
| 1038 | /* Horizontal space always OK. */ |
| 1039 | if (c == ' ' || c == '\t') |
| 1040 | ; |
| 1041 | /* Just \f \v or \0 left. */ |
| 1042 | else if (c == '\0') |
| 1043 | saw_NUL = true; |
| 1044 | else if (pfile->state.in_directive && CPP_PEDANTIC (pfile)) |
| 1045 | cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, |
| 1046 | CPP_BUF_COL (buffer), |
| 1047 | "%s in preprocessing directive", |
| 1048 | c == '\f' ? "form feed" : "vertical tab"); |
| 1049 | |
| 1050 | c = *buffer->cur++; |
| 1051 | } |
| 1052 | /* We only want non-vertical space, i.e. ' ' \t \f \v \0. */ |
| 1053 | while (is_nvspace (c)); |
| 1054 | |
| 1055 | if (saw_NUL) |
| 1056 | cpp_error (pfile, CPP_DL_WARNING, "null character(s) ignored"); |
| 1057 | |
| 1058 | buffer->cur--; |
| 1059 | } |
| 1060 | |
| 1061 | /* See if the characters of a number token are valid in a name (no |
| 1062 | '.', '+' or '-'). */ |
| 1063 | static int |
| 1064 | name_p (cpp_reader *pfile, const cpp_string *string) |
| 1065 | { |
| 1066 | unsigned int i; |
| 1067 | |
| 1068 | for (i = 0; i < string->len; i++) |
| 1069 | if (!is_idchar (string->text[i])) |
| 1070 | return 0; |
| 1071 | |
| 1072 | return 1; |
| 1073 | } |
| 1074 | |
| 1075 | /* After parsing an identifier or other sequence, produce a warning about |
| 1076 | sequences not in NFC/NFKC. */ |
| 1077 | static void |
| 1078 | warn_about_normalization (cpp_reader *pfile, |
| 1079 | const cpp_token *token, |
| 1080 | const struct normalize_state *s) |
| 1081 | { |
| 1082 | if (CPP_OPTION (pfile, warn_normalize) < NORMALIZE_STATE_RESULT (s) |
| 1083 | && !pfile->state.skipping) |
| 1084 | { |
| 1085 | /* Make sure that the token is printed using UCNs, even |
| 1086 | if we'd otherwise happily print UTF-8. */ |
| 1087 | unsigned char *buf = XNEWVEC (unsigned char, cpp_token_len (token)); |
| 1088 | size_t sz; |
| 1089 | |
| 1090 | sz = cpp_spell_token (pfile, token, buf, false) - buf; |
| 1091 | if (NORMALIZE_STATE_RESULT (s) == normalized_C) |
| 1092 | cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0, |
| 1093 | "`%.*s' is not in NFKC", (int) sz, buf); |
| 1094 | else |
| 1095 | cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0, |
| 1096 | "`%.*s' is not in NFC", (int) sz, buf); |
Bernhard Rosenkraenzer | 7d3ad0b | 2012-10-23 01:39:53 +0159 | [diff] [blame^] | 1097 | free (buf); |
Bernhard Rosenkraenzer | c83ebe5 | 2012-09-18 21:38:03 +0159 | [diff] [blame] | 1098 | } |
| 1099 | } |
| 1100 | |
| 1101 | /* Returns TRUE if the sequence starting at buffer->cur is invalid in |
| 1102 | an identifier. FIRST is TRUE if this starts an identifier. */ |
| 1103 | static bool |
| 1104 | forms_identifier_p (cpp_reader *pfile, int first, |
| 1105 | struct normalize_state *state) |
| 1106 | { |
| 1107 | cpp_buffer *buffer = pfile->buffer; |
| 1108 | |
| 1109 | if (*buffer->cur == '$') |
| 1110 | { |
| 1111 | if (!CPP_OPTION (pfile, dollars_in_ident)) |
| 1112 | return false; |
| 1113 | |
| 1114 | buffer->cur++; |
| 1115 | if (CPP_OPTION (pfile, warn_dollars) && !pfile->state.skipping) |
| 1116 | { |
| 1117 | CPP_OPTION (pfile, warn_dollars) = 0; |
| 1118 | cpp_error (pfile, CPP_DL_PEDWARN, "'$' in identifier or number"); |
| 1119 | } |
| 1120 | |
| 1121 | return true; |
| 1122 | } |
| 1123 | |
| 1124 | /* Is this a syntactically valid UCN? */ |
| 1125 | if (CPP_OPTION (pfile, extended_identifiers) |
| 1126 | && *buffer->cur == '\\' |
| 1127 | && (buffer->cur[1] == 'u' || buffer->cur[1] == 'U')) |
| 1128 | { |
| 1129 | buffer->cur += 2; |
| 1130 | if (_cpp_valid_ucn (pfile, &buffer->cur, buffer->rlimit, 1 + !first, |
| 1131 | state)) |
| 1132 | return true; |
| 1133 | buffer->cur -= 2; |
| 1134 | } |
| 1135 | |
| 1136 | return false; |
| 1137 | } |
| 1138 | |
| 1139 | /* Helper function to get the cpp_hashnode of the identifier BASE. */ |
| 1140 | static cpp_hashnode * |
| 1141 | lex_identifier_intern (cpp_reader *pfile, const uchar *base) |
| 1142 | { |
| 1143 | cpp_hashnode *result; |
| 1144 | const uchar *cur; |
| 1145 | unsigned int len; |
| 1146 | unsigned int hash = HT_HASHSTEP (0, *base); |
| 1147 | |
| 1148 | cur = base + 1; |
| 1149 | while (ISIDNUM (*cur)) |
| 1150 | { |
| 1151 | hash = HT_HASHSTEP (hash, *cur); |
| 1152 | cur++; |
| 1153 | } |
| 1154 | len = cur - base; |
| 1155 | hash = HT_HASHFINISH (hash, len); |
| 1156 | result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, |
| 1157 | base, len, hash, HT_ALLOC)); |
| 1158 | |
| 1159 | /* Rarely, identifiers require diagnostics when lexed. */ |
| 1160 | if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) |
| 1161 | && !pfile->state.skipping, 0)) |
| 1162 | { |
| 1163 | /* It is allowed to poison the same identifier twice. */ |
| 1164 | if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) |
| 1165 | cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", |
| 1166 | NODE_NAME (result)); |
| 1167 | |
| 1168 | /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the |
| 1169 | replacement list of a variadic macro. */ |
| 1170 | if (result == pfile->spec_nodes.n__VA_ARGS__ |
| 1171 | && !pfile->state.va_args_ok) |
| 1172 | cpp_error (pfile, CPP_DL_PEDWARN, |
| 1173 | "__VA_ARGS__ can only appear in the expansion" |
| 1174 | " of a C99 variadic macro"); |
| 1175 | |
| 1176 | /* For -Wc++-compat, warn about use of C++ named operators. */ |
| 1177 | if (result->flags & NODE_WARN_OPERATOR) |
| 1178 | cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, |
| 1179 | "identifier \"%s\" is a special operator name in C++", |
| 1180 | NODE_NAME (result)); |
| 1181 | } |
| 1182 | |
| 1183 | return result; |
| 1184 | } |
| 1185 | |
| 1186 | /* Get the cpp_hashnode of an identifier specified by NAME in |
| 1187 | the current cpp_reader object. If none is found, NULL is returned. */ |
| 1188 | cpp_hashnode * |
| 1189 | _cpp_lex_identifier (cpp_reader *pfile, const char *name) |
| 1190 | { |
| 1191 | cpp_hashnode *result; |
| 1192 | result = lex_identifier_intern (pfile, (uchar *) name); |
| 1193 | return result; |
| 1194 | } |
| 1195 | |
| 1196 | /* Lex an identifier starting at BUFFER->CUR - 1. */ |
| 1197 | static cpp_hashnode * |
| 1198 | lex_identifier (cpp_reader *pfile, const uchar *base, bool starts_ucn, |
| 1199 | struct normalize_state *nst) |
| 1200 | { |
| 1201 | cpp_hashnode *result; |
| 1202 | const uchar *cur; |
| 1203 | unsigned int len; |
| 1204 | unsigned int hash = HT_HASHSTEP (0, *base); |
| 1205 | |
| 1206 | cur = pfile->buffer->cur; |
| 1207 | if (! starts_ucn) |
| 1208 | while (ISIDNUM (*cur)) |
| 1209 | { |
| 1210 | hash = HT_HASHSTEP (hash, *cur); |
| 1211 | cur++; |
| 1212 | } |
| 1213 | pfile->buffer->cur = cur; |
| 1214 | if (starts_ucn || forms_identifier_p (pfile, false, nst)) |
| 1215 | { |
| 1216 | /* Slower version for identifiers containing UCNs (or $). */ |
| 1217 | do { |
| 1218 | while (ISIDNUM (*pfile->buffer->cur)) |
| 1219 | { |
| 1220 | pfile->buffer->cur++; |
| 1221 | NORMALIZE_STATE_UPDATE_IDNUM (nst); |
| 1222 | } |
| 1223 | } while (forms_identifier_p (pfile, false, nst)); |
| 1224 | result = _cpp_interpret_identifier (pfile, base, |
| 1225 | pfile->buffer->cur - base); |
| 1226 | } |
| 1227 | else |
| 1228 | { |
| 1229 | len = cur - base; |
| 1230 | hash = HT_HASHFINISH (hash, len); |
| 1231 | |
| 1232 | result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table, |
| 1233 | base, len, hash, HT_ALLOC)); |
| 1234 | } |
| 1235 | |
| 1236 | /* Rarely, identifiers require diagnostics when lexed. */ |
| 1237 | if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC) |
| 1238 | && !pfile->state.skipping, 0)) |
| 1239 | { |
| 1240 | /* It is allowed to poison the same identifier twice. */ |
| 1241 | if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok) |
| 1242 | cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"", |
| 1243 | NODE_NAME (result)); |
| 1244 | |
| 1245 | /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the |
| 1246 | replacement list of a variadic macro. */ |
| 1247 | if (result == pfile->spec_nodes.n__VA_ARGS__ |
| 1248 | && !pfile->state.va_args_ok) |
| 1249 | cpp_error (pfile, CPP_DL_PEDWARN, |
| 1250 | "__VA_ARGS__ can only appear in the expansion" |
| 1251 | " of a C99 variadic macro"); |
| 1252 | |
| 1253 | /* For -Wc++-compat, warn about use of C++ named operators. */ |
| 1254 | if (result->flags & NODE_WARN_OPERATOR) |
| 1255 | cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES, |
| 1256 | "identifier \"%s\" is a special operator name in C++", |
| 1257 | NODE_NAME (result)); |
| 1258 | } |
| 1259 | |
| 1260 | return result; |
| 1261 | } |
| 1262 | |
| 1263 | /* Lex a number to NUMBER starting at BUFFER->CUR - 1. */ |
| 1264 | static void |
| 1265 | lex_number (cpp_reader *pfile, cpp_string *number, |
| 1266 | struct normalize_state *nst) |
| 1267 | { |
| 1268 | const uchar *cur; |
| 1269 | const uchar *base; |
| 1270 | uchar *dest; |
| 1271 | |
| 1272 | base = pfile->buffer->cur - 1; |
| 1273 | do |
| 1274 | { |
| 1275 | cur = pfile->buffer->cur; |
| 1276 | |
| 1277 | /* N.B. ISIDNUM does not include $. */ |
| 1278 | while (ISIDNUM (*cur) || *cur == '.' || VALID_SIGN (*cur, cur[-1])) |
| 1279 | { |
| 1280 | cur++; |
| 1281 | NORMALIZE_STATE_UPDATE_IDNUM (nst); |
| 1282 | } |
| 1283 | |
| 1284 | pfile->buffer->cur = cur; |
| 1285 | } |
| 1286 | while (forms_identifier_p (pfile, false, nst)); |
| 1287 | |
| 1288 | number->len = cur - base; |
| 1289 | dest = _cpp_unaligned_alloc (pfile, number->len + 1); |
| 1290 | memcpy (dest, base, number->len); |
| 1291 | dest[number->len] = '\0'; |
| 1292 | number->text = dest; |
| 1293 | } |
| 1294 | |
| 1295 | /* Create a token of type TYPE with a literal spelling. */ |
| 1296 | static void |
| 1297 | create_literal (cpp_reader *pfile, cpp_token *token, const uchar *base, |
| 1298 | unsigned int len, enum cpp_ttype type) |
| 1299 | { |
| 1300 | uchar *dest = _cpp_unaligned_alloc (pfile, len + 1); |
| 1301 | |
| 1302 | memcpy (dest, base, len); |
| 1303 | dest[len] = '\0'; |
| 1304 | token->type = type; |
| 1305 | token->val.str.len = len; |
| 1306 | token->val.str.text = dest; |
| 1307 | } |
| 1308 | |
| 1309 | /* Subroutine of lex_raw_string: Append LEN chars from BASE to the buffer |
| 1310 | sequence from *FIRST_BUFF_P to LAST_BUFF_P. */ |
| 1311 | |
| 1312 | static void |
| 1313 | bufring_append (cpp_reader *pfile, const uchar *base, size_t len, |
| 1314 | _cpp_buff **first_buff_p, _cpp_buff **last_buff_p) |
| 1315 | { |
| 1316 | _cpp_buff *first_buff = *first_buff_p; |
| 1317 | _cpp_buff *last_buff = *last_buff_p; |
| 1318 | |
| 1319 | if (first_buff == NULL) |
| 1320 | first_buff = last_buff = _cpp_get_buff (pfile, len); |
| 1321 | else if (len > BUFF_ROOM (last_buff)) |
| 1322 | { |
| 1323 | size_t room = BUFF_ROOM (last_buff); |
| 1324 | memcpy (BUFF_FRONT (last_buff), base, room); |
| 1325 | BUFF_FRONT (last_buff) += room; |
| 1326 | base += room; |
| 1327 | len -= room; |
| 1328 | last_buff = _cpp_append_extend_buff (pfile, last_buff, len); |
| 1329 | } |
| 1330 | |
| 1331 | memcpy (BUFF_FRONT (last_buff), base, len); |
| 1332 | BUFF_FRONT (last_buff) += len; |
| 1333 | |
| 1334 | *first_buff_p = first_buff; |
| 1335 | *last_buff_p = last_buff; |
| 1336 | } |
| 1337 | |
| 1338 | /* Lexes a raw string. The stored string contains the spelling, including |
| 1339 | double quotes, delimiter string, '(' and ')', any leading |
| 1340 | 'L', 'u', 'U' or 'u8' and 'R' modifier. It returns the type of the |
| 1341 | literal, or CPP_OTHER if it was not properly terminated. |
| 1342 | |
| 1343 | The spelling is NUL-terminated, but it is not guaranteed that this |
| 1344 | is the first NUL since embedded NULs are preserved. */ |
| 1345 | |
| 1346 | static void |
| 1347 | lex_raw_string (cpp_reader *pfile, cpp_token *token, const uchar *base, |
| 1348 | const uchar *cur) |
| 1349 | { |
| 1350 | const uchar *raw_prefix; |
| 1351 | unsigned int raw_prefix_len = 0; |
| 1352 | enum cpp_ttype type; |
| 1353 | size_t total_len = 0; |
| 1354 | _cpp_buff *first_buff = NULL, *last_buff = NULL; |
| 1355 | _cpp_line_note *note = &pfile->buffer->notes[pfile->buffer->cur_note]; |
| 1356 | |
| 1357 | type = (*base == 'L' ? CPP_WSTRING : |
| 1358 | *base == 'U' ? CPP_STRING32 : |
| 1359 | *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16) |
| 1360 | : CPP_STRING); |
| 1361 | |
| 1362 | raw_prefix = cur + 1; |
| 1363 | while (raw_prefix_len < 16) |
| 1364 | { |
| 1365 | switch (raw_prefix[raw_prefix_len]) |
| 1366 | { |
| 1367 | case ' ': case '(': case ')': case '\\': case '\t': |
| 1368 | case '\v': case '\f': case '\n': default: |
| 1369 | break; |
| 1370 | /* Basic source charset except the above chars. */ |
| 1371 | case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': |
| 1372 | case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': |
| 1373 | case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': |
| 1374 | case 's': case 't': case 'u': case 'v': case 'w': case 'x': |
| 1375 | case 'y': case 'z': |
| 1376 | case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': |
| 1377 | case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': |
| 1378 | case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R': |
| 1379 | case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': |
| 1380 | case 'Y': case 'Z': |
| 1381 | case '0': case '1': case '2': case '3': case '4': case '5': |
| 1382 | case '6': case '7': case '8': case '9': |
| 1383 | case '_': case '{': case '}': case '#': case '[': case ']': |
| 1384 | case '<': case '>': case '%': case ':': case ';': case '.': |
| 1385 | case '?': case '*': case '+': case '-': case '/': case '^': |
| 1386 | case '&': case '|': case '~': case '!': case '=': case ',': |
| 1387 | case '"': case '\'': |
| 1388 | raw_prefix_len++; |
| 1389 | continue; |
| 1390 | } |
| 1391 | break; |
| 1392 | } |
| 1393 | |
| 1394 | if (raw_prefix[raw_prefix_len] != '(') |
| 1395 | { |
| 1396 | int col = CPP_BUF_COLUMN (pfile->buffer, raw_prefix + raw_prefix_len) |
| 1397 | + 1; |
| 1398 | if (raw_prefix_len == 16) |
| 1399 | cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, col, |
| 1400 | "raw string delimiter longer than 16 characters"); |
| 1401 | else |
| 1402 | cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, col, |
| 1403 | "invalid character '%c' in raw string delimiter", |
| 1404 | (int) raw_prefix[raw_prefix_len]); |
| 1405 | pfile->buffer->cur = raw_prefix - 1; |
| 1406 | create_literal (pfile, token, base, raw_prefix - 1 - base, CPP_OTHER); |
| 1407 | return; |
| 1408 | } |
| 1409 | |
| 1410 | cur = raw_prefix + raw_prefix_len + 1; |
| 1411 | for (;;) |
| 1412 | { |
| 1413 | #define BUF_APPEND(STR,LEN) \ |
| 1414 | do { \ |
| 1415 | bufring_append (pfile, (const uchar *)(STR), (LEN), \ |
| 1416 | &first_buff, &last_buff); \ |
| 1417 | total_len += (LEN); \ |
| 1418 | } while (0); |
| 1419 | |
| 1420 | cppchar_t c; |
| 1421 | |
| 1422 | /* If we previously performed any trigraph or line splicing |
| 1423 | transformations, undo them within the body of the raw string. */ |
| 1424 | while (note->pos < cur) |
| 1425 | ++note; |
| 1426 | for (; note->pos == cur; ++note) |
| 1427 | { |
| 1428 | switch (note->type) |
| 1429 | { |
| 1430 | case '\\': |
| 1431 | case ' ': |
| 1432 | /* Restore backslash followed by newline. */ |
| 1433 | BUF_APPEND (base, cur - base); |
| 1434 | base = cur; |
| 1435 | BUF_APPEND ("\\", 1); |
| 1436 | after_backslash: |
| 1437 | if (note->type == ' ') |
| 1438 | { |
| 1439 | /* GNU backslash whitespace newline extension. FIXME |
| 1440 | could be any sequence of non-vertical space. When we |
| 1441 | can properly restore any such sequence, we should mark |
| 1442 | this note as handled so _cpp_process_line_notes |
| 1443 | doesn't warn. */ |
| 1444 | BUF_APPEND (" ", 1); |
| 1445 | } |
| 1446 | |
| 1447 | BUF_APPEND ("\n", 1); |
| 1448 | break; |
| 1449 | |
| 1450 | case 0: |
| 1451 | /* Already handled. */ |
| 1452 | break; |
| 1453 | |
| 1454 | default: |
| 1455 | if (_cpp_trigraph_map[note->type]) |
| 1456 | { |
| 1457 | /* Don't warn about this trigraph in |
| 1458 | _cpp_process_line_notes, since trigraphs show up as |
| 1459 | trigraphs in raw strings. */ |
| 1460 | uchar type = note->type; |
| 1461 | note->type = 0; |
| 1462 | |
| 1463 | if (!CPP_OPTION (pfile, trigraphs)) |
| 1464 | /* If we didn't convert the trigraph in the first |
| 1465 | place, don't do anything now either. */ |
| 1466 | break; |
| 1467 | |
| 1468 | BUF_APPEND (base, cur - base); |
| 1469 | base = cur; |
| 1470 | BUF_APPEND ("??", 2); |
| 1471 | |
| 1472 | /* ??/ followed by newline gets two line notes, one for |
| 1473 | the trigraph and one for the backslash/newline. */ |
| 1474 | if (type == '/' && note[1].pos == cur) |
| 1475 | { |
| 1476 | if (note[1].type != '\\' |
| 1477 | && note[1].type != ' ') |
| 1478 | abort (); |
| 1479 | BUF_APPEND ("/", 1); |
| 1480 | ++note; |
| 1481 | goto after_backslash; |
| 1482 | } |
| 1483 | /* The ) from ??) could be part of the suffix. */ |
| 1484 | else if (type == ')' |
| 1485 | && strncmp ((const char *) cur+1, |
| 1486 | (const char *) raw_prefix, |
| 1487 | raw_prefix_len) == 0 |
| 1488 | && cur[raw_prefix_len+1] == '"') |
| 1489 | { |
| 1490 | BUF_APPEND (")", 1); |
| 1491 | base++; |
| 1492 | cur += raw_prefix_len + 2; |
| 1493 | goto break_outer_loop; |
| 1494 | } |
| 1495 | else |
| 1496 | { |
| 1497 | /* Skip the replacement character. */ |
| 1498 | base = ++cur; |
| 1499 | BUF_APPEND (&type, 1); |
| 1500 | } |
| 1501 | } |
| 1502 | else |
| 1503 | abort (); |
| 1504 | break; |
| 1505 | } |
| 1506 | } |
| 1507 | c = *cur++; |
| 1508 | |
| 1509 | if (c == ')' |
| 1510 | && strncmp ((const char *) cur, (const char *) raw_prefix, |
| 1511 | raw_prefix_len) == 0 |
| 1512 | && cur[raw_prefix_len] == '"') |
| 1513 | { |
| 1514 | cur += raw_prefix_len + 1; |
| 1515 | break; |
| 1516 | } |
| 1517 | else if (c == '\n') |
| 1518 | { |
| 1519 | if (pfile->state.in_directive |
| 1520 | || pfile->state.parsing_args |
| 1521 | || pfile->state.in_deferred_pragma) |
| 1522 | { |
| 1523 | cur--; |
| 1524 | type = CPP_OTHER; |
| 1525 | cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 0, |
| 1526 | "unterminated raw string"); |
| 1527 | break; |
| 1528 | } |
| 1529 | |
| 1530 | BUF_APPEND (base, cur - base); |
| 1531 | |
| 1532 | if (pfile->buffer->cur < pfile->buffer->rlimit) |
| 1533 | CPP_INCREMENT_LINE (pfile, 0); |
| 1534 | pfile->buffer->need_line = true; |
| 1535 | |
| 1536 | pfile->buffer->cur = cur-1; |
| 1537 | _cpp_process_line_notes (pfile, false); |
| 1538 | if (!_cpp_get_fresh_line (pfile)) |
| 1539 | { |
| 1540 | source_location src_loc = token->src_loc; |
| 1541 | token->type = CPP_EOF; |
| 1542 | /* Tell the compiler the line number of the EOF token. */ |
| 1543 | token->src_loc = pfile->line_table->highest_line; |
| 1544 | token->flags = BOL; |
| 1545 | if (first_buff != NULL) |
| 1546 | _cpp_release_buff (pfile, first_buff); |
| 1547 | cpp_error_with_line (pfile, CPP_DL_ERROR, src_loc, 0, |
| 1548 | "unterminated raw string"); |
| 1549 | return; |
| 1550 | } |
| 1551 | |
| 1552 | cur = base = pfile->buffer->cur; |
| 1553 | note = &pfile->buffer->notes[pfile->buffer->cur_note]; |
| 1554 | } |
| 1555 | } |
| 1556 | break_outer_loop: |
| 1557 | |
| 1558 | if (CPP_OPTION (pfile, user_literals)) |
| 1559 | { |
| 1560 | /* According to C++11 [lex.ext]p10, a ud-suffix not starting with an |
| 1561 | underscore is ill-formed. Since this breaks programs using macros |
| 1562 | from inttypes.h, we generate a warning and treat the ud-suffix as a |
| 1563 | separate preprocessing token. This approach is under discussion by |
| 1564 | the standards committee, and has been adopted as a conforming |
| 1565 | extension by other front ends such as clang. */ |
| 1566 | if (ISALPHA (*cur)) |
| 1567 | { |
| 1568 | /* Raise a warning, but do not consume subsequent tokens. */ |
| 1569 | if (CPP_OPTION (pfile, warn_literal_suffix)) |
| 1570 | cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, |
| 1571 | token->src_loc, 0, |
| 1572 | "invalid suffix on literal; C++11 requires " |
| 1573 | "a space between literal and identifier"); |
| 1574 | } |
| 1575 | /* Grab user defined literal suffix. */ |
| 1576 | else if (*cur == '_') |
| 1577 | { |
| 1578 | type = cpp_userdef_string_add_type (type); |
| 1579 | ++cur; |
| 1580 | |
| 1581 | while (ISIDNUM (*cur)) |
| 1582 | ++cur; |
| 1583 | } |
| 1584 | } |
| 1585 | |
| 1586 | pfile->buffer->cur = cur; |
| 1587 | if (first_buff == NULL) |
| 1588 | create_literal (pfile, token, base, cur - base, type); |
| 1589 | else |
| 1590 | { |
| 1591 | uchar *dest = _cpp_unaligned_alloc (pfile, total_len + (cur - base) + 1); |
| 1592 | |
| 1593 | token->type = type; |
| 1594 | token->val.str.len = total_len + (cur - base); |
| 1595 | token->val.str.text = dest; |
| 1596 | last_buff = first_buff; |
| 1597 | while (last_buff != NULL) |
| 1598 | { |
| 1599 | memcpy (dest, last_buff->base, |
| 1600 | BUFF_FRONT (last_buff) - last_buff->base); |
| 1601 | dest += BUFF_FRONT (last_buff) - last_buff->base; |
| 1602 | last_buff = last_buff->next; |
| 1603 | } |
| 1604 | _cpp_release_buff (pfile, first_buff); |
| 1605 | memcpy (dest, base, cur - base); |
| 1606 | dest[cur - base] = '\0'; |
| 1607 | } |
| 1608 | } |
| 1609 | |
| 1610 | /* Lexes a string, character constant, or angle-bracketed header file |
| 1611 | name. The stored string contains the spelling, including opening |
| 1612 | quote and any leading 'L', 'u', 'U' or 'u8' and optional |
| 1613 | 'R' modifier. It returns the type of the literal, or CPP_OTHER |
| 1614 | if it was not properly terminated, or CPP_LESS for an unterminated |
| 1615 | header name which must be relexed as normal tokens. |
| 1616 | |
| 1617 | The spelling is NUL-terminated, but it is not guaranteed that this |
| 1618 | is the first NUL since embedded NULs are preserved. */ |
| 1619 | static void |
| 1620 | lex_string (cpp_reader *pfile, cpp_token *token, const uchar *base) |
| 1621 | { |
| 1622 | bool saw_NUL = false; |
| 1623 | const uchar *cur; |
| 1624 | cppchar_t terminator; |
| 1625 | enum cpp_ttype type; |
| 1626 | |
| 1627 | cur = base; |
| 1628 | terminator = *cur++; |
| 1629 | if (terminator == 'L' || terminator == 'U') |
| 1630 | terminator = *cur++; |
| 1631 | else if (terminator == 'u') |
| 1632 | { |
| 1633 | terminator = *cur++; |
| 1634 | if (terminator == '8') |
| 1635 | terminator = *cur++; |
| 1636 | } |
| 1637 | if (terminator == 'R') |
| 1638 | { |
| 1639 | lex_raw_string (pfile, token, base, cur); |
| 1640 | return; |
| 1641 | } |
| 1642 | if (terminator == '"') |
| 1643 | type = (*base == 'L' ? CPP_WSTRING : |
| 1644 | *base == 'U' ? CPP_STRING32 : |
| 1645 | *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16) |
| 1646 | : CPP_STRING); |
| 1647 | else if (terminator == '\'') |
| 1648 | type = (*base == 'L' ? CPP_WCHAR : |
| 1649 | *base == 'U' ? CPP_CHAR32 : |
| 1650 | *base == 'u' ? CPP_CHAR16 : CPP_CHAR); |
| 1651 | else |
| 1652 | terminator = '>', type = CPP_HEADER_NAME; |
| 1653 | |
| 1654 | for (;;) |
| 1655 | { |
| 1656 | cppchar_t c = *cur++; |
| 1657 | |
| 1658 | /* In #include-style directives, terminators are not escapable. */ |
| 1659 | if (c == '\\' && !pfile->state.angled_headers && *cur != '\n') |
| 1660 | cur++; |
| 1661 | else if (c == terminator) |
| 1662 | break; |
| 1663 | else if (c == '\n') |
| 1664 | { |
| 1665 | cur--; |
| 1666 | /* Unmatched quotes always yield undefined behavior, but |
| 1667 | greedy lexing means that what appears to be an unterminated |
| 1668 | header name may actually be a legitimate sequence of tokens. */ |
| 1669 | if (terminator == '>') |
| 1670 | { |
| 1671 | token->type = CPP_LESS; |
| 1672 | return; |
| 1673 | } |
| 1674 | type = CPP_OTHER; |
| 1675 | break; |
| 1676 | } |
| 1677 | else if (c == '\0') |
| 1678 | saw_NUL = true; |
| 1679 | } |
| 1680 | |
| 1681 | if (saw_NUL && !pfile->state.skipping) |
| 1682 | cpp_error (pfile, CPP_DL_WARNING, |
| 1683 | "null character(s) preserved in literal"); |
| 1684 | |
| 1685 | if (type == CPP_OTHER && CPP_OPTION (pfile, lang) != CLK_ASM) |
| 1686 | cpp_error (pfile, CPP_DL_PEDWARN, "missing terminating %c character", |
| 1687 | (int) terminator); |
| 1688 | |
| 1689 | if (CPP_OPTION (pfile, user_literals)) |
| 1690 | { |
| 1691 | /* According to C++11 [lex.ext]p10, a ud-suffix not starting with an |
| 1692 | underscore is ill-formed. Since this breaks programs using macros |
| 1693 | from inttypes.h, we generate a warning and treat the ud-suffix as a |
| 1694 | separate preprocessing token. This approach is under discussion by |
| 1695 | the standards committee, and has been adopted as a conforming |
| 1696 | extension by other front ends such as clang. */ |
| 1697 | if (ISALPHA (*cur)) |
| 1698 | { |
| 1699 | /* Raise a warning, but do not consume subsequent tokens. */ |
| 1700 | if (CPP_OPTION (pfile, warn_literal_suffix)) |
| 1701 | cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX, |
| 1702 | token->src_loc, 0, |
| 1703 | "invalid suffix on literal; C++11 requires " |
| 1704 | "a space between literal and identifier"); |
| 1705 | } |
| 1706 | /* Grab user defined literal suffix. */ |
| 1707 | else if (*cur == '_') |
| 1708 | { |
| 1709 | type = cpp_userdef_char_add_type (type); |
| 1710 | type = cpp_userdef_string_add_type (type); |
| 1711 | ++cur; |
| 1712 | |
| 1713 | while (ISIDNUM (*cur)) |
| 1714 | ++cur; |
| 1715 | } |
| 1716 | } |
| 1717 | |
| 1718 | pfile->buffer->cur = cur; |
| 1719 | create_literal (pfile, token, base, cur - base, type); |
| 1720 | } |
| 1721 | |
| 1722 | /* Return the comment table. The client may not make any assumption |
| 1723 | about the ordering of the table. */ |
| 1724 | cpp_comment_table * |
| 1725 | cpp_get_comments (cpp_reader *pfile) |
| 1726 | { |
| 1727 | return &pfile->comments; |
| 1728 | } |
| 1729 | |
| 1730 | /* Append a comment to the end of the comment table. */ |
| 1731 | static void |
| 1732 | store_comment (cpp_reader *pfile, cpp_token *token) |
| 1733 | { |
| 1734 | int len; |
| 1735 | |
| 1736 | if (pfile->comments.allocated == 0) |
| 1737 | { |
| 1738 | pfile->comments.allocated = 256; |
| 1739 | pfile->comments.entries = (cpp_comment *) xmalloc |
| 1740 | (pfile->comments.allocated * sizeof (cpp_comment)); |
| 1741 | } |
| 1742 | |
| 1743 | if (pfile->comments.count == pfile->comments.allocated) |
| 1744 | { |
| 1745 | pfile->comments.allocated *= 2; |
| 1746 | pfile->comments.entries = (cpp_comment *) xrealloc |
| 1747 | (pfile->comments.entries, |
| 1748 | pfile->comments.allocated * sizeof (cpp_comment)); |
| 1749 | } |
| 1750 | |
| 1751 | len = token->val.str.len; |
| 1752 | |
| 1753 | /* Copy comment. Note, token may not be NULL terminated. */ |
| 1754 | pfile->comments.entries[pfile->comments.count].comment = |
| 1755 | (char *) xmalloc (sizeof (char) * (len + 1)); |
| 1756 | memcpy (pfile->comments.entries[pfile->comments.count].comment, |
| 1757 | token->val.str.text, len); |
| 1758 | pfile->comments.entries[pfile->comments.count].comment[len] = '\0'; |
| 1759 | |
| 1760 | /* Set source location. */ |
| 1761 | pfile->comments.entries[pfile->comments.count].sloc = token->src_loc; |
| 1762 | |
| 1763 | /* Increment the count of entries in the comment table. */ |
| 1764 | pfile->comments.count++; |
| 1765 | } |
| 1766 | |
| 1767 | /* The stored comment includes the comment start and any terminator. */ |
| 1768 | static void |
| 1769 | save_comment (cpp_reader *pfile, cpp_token *token, const unsigned char *from, |
| 1770 | cppchar_t type) |
| 1771 | { |
| 1772 | unsigned char *buffer; |
| 1773 | unsigned int len, clen, i; |
| 1774 | |
| 1775 | len = pfile->buffer->cur - from + 1; /* + 1 for the initial '/'. */ |
| 1776 | |
| 1777 | /* C++ comments probably (not definitely) have moved past a new |
| 1778 | line, which we don't want to save in the comment. */ |
| 1779 | if (is_vspace (pfile->buffer->cur[-1])) |
| 1780 | len--; |
| 1781 | |
| 1782 | /* If we are currently in a directive or in argument parsing, then |
| 1783 | we need to store all C++ comments as C comments internally, and |
| 1784 | so we need to allocate a little extra space in that case. |
| 1785 | |
| 1786 | Note that the only time we encounter a directive here is |
| 1787 | when we are saving comments in a "#define". */ |
| 1788 | clen = ((pfile->state.in_directive || pfile->state.parsing_args) |
| 1789 | && type == '/') ? len + 2 : len; |
| 1790 | |
| 1791 | buffer = _cpp_unaligned_alloc (pfile, clen); |
| 1792 | |
| 1793 | token->type = CPP_COMMENT; |
| 1794 | token->val.str.len = clen; |
| 1795 | token->val.str.text = buffer; |
| 1796 | |
| 1797 | buffer[0] = '/'; |
| 1798 | memcpy (buffer + 1, from, len - 1); |
| 1799 | |
| 1800 | /* Finish conversion to a C comment, if necessary. */ |
| 1801 | if ((pfile->state.in_directive || pfile->state.parsing_args) && type == '/') |
| 1802 | { |
| 1803 | buffer[1] = '*'; |
| 1804 | buffer[clen - 2] = '*'; |
| 1805 | buffer[clen - 1] = '/'; |
| 1806 | /* As there can be in a C++ comments illegal sequences for C comments |
| 1807 | we need to filter them out. */ |
| 1808 | for (i = 2; i < (clen - 2); i++) |
| 1809 | if (buffer[i] == '/' && (buffer[i - 1] == '*' || buffer[i + 1] == '*')) |
| 1810 | buffer[i] = '|'; |
| 1811 | } |
| 1812 | |
| 1813 | /* Finally store this comment for use by clients of libcpp. */ |
| 1814 | store_comment (pfile, token); |
| 1815 | } |
| 1816 | |
| 1817 | /* Allocate COUNT tokens for RUN. */ |
| 1818 | void |
| 1819 | _cpp_init_tokenrun (tokenrun *run, unsigned int count) |
| 1820 | { |
| 1821 | run->base = XNEWVEC (cpp_token, count); |
| 1822 | run->limit = run->base + count; |
| 1823 | run->next = NULL; |
| 1824 | } |
| 1825 | |
| 1826 | /* Returns the next tokenrun, or creates one if there is none. */ |
| 1827 | static tokenrun * |
| 1828 | next_tokenrun (tokenrun *run) |
| 1829 | { |
| 1830 | if (run->next == NULL) |
| 1831 | { |
| 1832 | run->next = XNEW (tokenrun); |
| 1833 | run->next->prev = run; |
| 1834 | _cpp_init_tokenrun (run->next, 250); |
| 1835 | } |
| 1836 | |
| 1837 | return run->next; |
| 1838 | } |
| 1839 | |
| 1840 | /* Return the number of not yet processed token in a given |
| 1841 | context. */ |
| 1842 | int |
| 1843 | _cpp_remaining_tokens_num_in_context (cpp_context *context) |
| 1844 | { |
| 1845 | if (context->tokens_kind == TOKENS_KIND_DIRECT) |
| 1846 | return (LAST (context).token - FIRST (context).token); |
| 1847 | else if (context->tokens_kind == TOKENS_KIND_INDIRECT |
| 1848 | || context->tokens_kind == TOKENS_KIND_EXTENDED) |
| 1849 | return (LAST (context).ptoken - FIRST (context).ptoken); |
| 1850 | else |
| 1851 | abort (); |
| 1852 | } |
| 1853 | |
| 1854 | /* Returns the token present at index INDEX in a given context. If |
| 1855 | INDEX is zero, the next token to be processed is returned. */ |
| 1856 | static const cpp_token* |
| 1857 | _cpp_token_from_context_at (cpp_context *context, int index) |
| 1858 | { |
| 1859 | if (context->tokens_kind == TOKENS_KIND_DIRECT) |
| 1860 | return &(FIRST (context).token[index]); |
| 1861 | else if (context->tokens_kind == TOKENS_KIND_INDIRECT |
| 1862 | || context->tokens_kind == TOKENS_KIND_EXTENDED) |
| 1863 | return FIRST (context).ptoken[index]; |
| 1864 | else |
| 1865 | abort (); |
| 1866 | } |
| 1867 | |
| 1868 | /* Look ahead in the input stream. */ |
| 1869 | const cpp_token * |
| 1870 | cpp_peek_token (cpp_reader *pfile, int index) |
| 1871 | { |
| 1872 | cpp_context *context = pfile->context; |
| 1873 | const cpp_token *peektok; |
| 1874 | int count; |
| 1875 | |
| 1876 | /* First, scan through any pending cpp_context objects. */ |
| 1877 | while (context->prev) |
| 1878 | { |
| 1879 | ptrdiff_t sz = _cpp_remaining_tokens_num_in_context (context); |
| 1880 | |
| 1881 | if (index < (int) sz) |
| 1882 | return _cpp_token_from_context_at (context, index); |
| 1883 | index -= (int) sz; |
| 1884 | context = context->prev; |
| 1885 | } |
| 1886 | |
| 1887 | /* We will have to read some new tokens after all (and do so |
| 1888 | without invalidating preceding tokens). */ |
| 1889 | count = index; |
| 1890 | pfile->keep_tokens++; |
| 1891 | |
| 1892 | do |
| 1893 | { |
| 1894 | peektok = _cpp_lex_token (pfile); |
| 1895 | if (peektok->type == CPP_EOF) |
| 1896 | return peektok; |
| 1897 | } |
| 1898 | while (index--); |
| 1899 | |
| 1900 | _cpp_backup_tokens_direct (pfile, count + 1); |
| 1901 | pfile->keep_tokens--; |
| 1902 | |
| 1903 | return peektok; |
| 1904 | } |
| 1905 | |
| 1906 | /* Allocate a single token that is invalidated at the same time as the |
| 1907 | rest of the tokens on the line. Has its line and col set to the |
| 1908 | same as the last lexed token, so that diagnostics appear in the |
| 1909 | right place. */ |
| 1910 | cpp_token * |
| 1911 | _cpp_temp_token (cpp_reader *pfile) |
| 1912 | { |
| 1913 | cpp_token *old, *result; |
| 1914 | ptrdiff_t sz = pfile->cur_run->limit - pfile->cur_token; |
| 1915 | ptrdiff_t la = (ptrdiff_t) pfile->lookaheads; |
| 1916 | |
| 1917 | old = pfile->cur_token - 1; |
| 1918 | /* Any pre-existing lookaheads must not be clobbered. */ |
| 1919 | if (la) |
| 1920 | { |
| 1921 | if (sz <= la) |
| 1922 | { |
| 1923 | tokenrun *next = next_tokenrun (pfile->cur_run); |
| 1924 | |
| 1925 | if (sz < la) |
| 1926 | memmove (next->base + 1, next->base, |
| 1927 | (la - sz) * sizeof (cpp_token)); |
| 1928 | |
| 1929 | next->base[0] = pfile->cur_run->limit[-1]; |
| 1930 | } |
| 1931 | |
| 1932 | if (sz > 1) |
| 1933 | memmove (pfile->cur_token + 1, pfile->cur_token, |
| 1934 | MIN (la, sz - 1) * sizeof (cpp_token)); |
| 1935 | } |
| 1936 | |
| 1937 | if (!sz && pfile->cur_token == pfile->cur_run->limit) |
| 1938 | { |
| 1939 | pfile->cur_run = next_tokenrun (pfile->cur_run); |
| 1940 | pfile->cur_token = pfile->cur_run->base; |
| 1941 | } |
| 1942 | |
| 1943 | result = pfile->cur_token++; |
| 1944 | result->src_loc = old->src_loc; |
| 1945 | return result; |
| 1946 | } |
| 1947 | |
| 1948 | /* Lex a token into RESULT (external interface). Takes care of issues |
| 1949 | like directive handling, token lookahead, multiple include |
| 1950 | optimization and skipping. */ |
| 1951 | const cpp_token * |
| 1952 | _cpp_lex_token (cpp_reader *pfile) |
| 1953 | { |
| 1954 | cpp_token *result; |
| 1955 | |
| 1956 | for (;;) |
| 1957 | { |
| 1958 | if (pfile->cur_token == pfile->cur_run->limit) |
| 1959 | { |
| 1960 | pfile->cur_run = next_tokenrun (pfile->cur_run); |
| 1961 | pfile->cur_token = pfile->cur_run->base; |
| 1962 | } |
| 1963 | /* We assume that the current token is somewhere in the current |
| 1964 | run. */ |
| 1965 | if (pfile->cur_token < pfile->cur_run->base |
| 1966 | || pfile->cur_token >= pfile->cur_run->limit) |
| 1967 | abort (); |
| 1968 | |
| 1969 | if (pfile->lookaheads) |
| 1970 | { |
| 1971 | pfile->lookaheads--; |
| 1972 | result = pfile->cur_token++; |
| 1973 | } |
| 1974 | else |
| 1975 | result = _cpp_lex_direct (pfile); |
| 1976 | |
| 1977 | if (result->flags & BOL) |
| 1978 | { |
| 1979 | /* Is this a directive. If _cpp_handle_directive returns |
| 1980 | false, it is an assembler #. */ |
| 1981 | if (result->type == CPP_HASH |
| 1982 | /* 6.10.3 p 11: Directives in a list of macro arguments |
| 1983 | gives undefined behavior. This implementation |
| 1984 | handles the directive as normal. */ |
| 1985 | && pfile->state.parsing_args != 1) |
| 1986 | { |
| 1987 | if (_cpp_handle_directive (pfile, result->flags & PREV_WHITE)) |
| 1988 | { |
| 1989 | if (pfile->directive_result.type == CPP_PADDING) |
| 1990 | continue; |
| 1991 | result = &pfile->directive_result; |
| 1992 | } |
| 1993 | } |
| 1994 | else if (pfile->state.in_deferred_pragma) |
| 1995 | result = &pfile->directive_result; |
| 1996 | |
| 1997 | if (pfile->cb.line_change && !pfile->state.skipping) |
| 1998 | pfile->cb.line_change (pfile, result, pfile->state.parsing_args); |
| 1999 | } |
| 2000 | |
| 2001 | /* We don't skip tokens in directives. */ |
| 2002 | if (pfile->state.in_directive || pfile->state.in_deferred_pragma) |
| 2003 | break; |
| 2004 | |
| 2005 | /* Outside a directive, invalidate controlling macros. At file |
| 2006 | EOF, _cpp_lex_direct takes care of popping the buffer, so we never |
| 2007 | get here and MI optimization works. */ |
| 2008 | pfile->mi_valid = false; |
| 2009 | |
| 2010 | if (!pfile->state.skipping || result->type == CPP_EOF) |
| 2011 | break; |
| 2012 | } |
| 2013 | |
| 2014 | return result; |
| 2015 | } |
| 2016 | |
| 2017 | /* Returns true if a fresh line has been loaded. */ |
| 2018 | bool |
| 2019 | _cpp_get_fresh_line (cpp_reader *pfile) |
| 2020 | { |
| 2021 | int return_at_eof; |
| 2022 | |
| 2023 | /* We can't get a new line until we leave the current directive. */ |
| 2024 | if (pfile->state.in_directive) |
| 2025 | return false; |
| 2026 | |
| 2027 | for (;;) |
| 2028 | { |
| 2029 | cpp_buffer *buffer = pfile->buffer; |
| 2030 | |
| 2031 | if (!buffer->need_line) |
| 2032 | return true; |
| 2033 | |
| 2034 | if (buffer->next_line < buffer->rlimit) |
| 2035 | { |
| 2036 | _cpp_clean_line (pfile); |
| 2037 | return true; |
| 2038 | } |
| 2039 | |
| 2040 | /* First, get out of parsing arguments state. */ |
| 2041 | if (pfile->state.parsing_args) |
| 2042 | return false; |
| 2043 | |
| 2044 | /* End of buffer. Non-empty files should end in a newline. */ |
| 2045 | if (buffer->buf != buffer->rlimit |
| 2046 | && buffer->next_line > buffer->rlimit |
| 2047 | && !buffer->from_stage3) |
| 2048 | { |
| 2049 | /* Clip to buffer size. */ |
| 2050 | buffer->next_line = buffer->rlimit; |
| 2051 | } |
| 2052 | |
| 2053 | return_at_eof = buffer->return_at_eof; |
| 2054 | _cpp_pop_buffer (pfile); |
| 2055 | if (pfile->buffer == NULL || return_at_eof) |
| 2056 | return false; |
| 2057 | } |
| 2058 | } |
| 2059 | |
| 2060 | #define IF_NEXT_IS(CHAR, THEN_TYPE, ELSE_TYPE) \ |
| 2061 | do \ |
| 2062 | { \ |
| 2063 | result->type = ELSE_TYPE; \ |
| 2064 | if (*buffer->cur == CHAR) \ |
| 2065 | buffer->cur++, result->type = THEN_TYPE; \ |
| 2066 | } \ |
| 2067 | while (0) |
| 2068 | |
| 2069 | /* Lex a token into pfile->cur_token, which is also incremented, to |
| 2070 | get diagnostics pointing to the correct location. |
| 2071 | |
| 2072 | Does not handle issues such as token lookahead, multiple-include |
| 2073 | optimization, directives, skipping etc. This function is only |
| 2074 | suitable for use by _cpp_lex_token, and in special cases like |
| 2075 | lex_expansion_token which doesn't care for any of these issues. |
| 2076 | |
| 2077 | When meeting a newline, returns CPP_EOF if parsing a directive, |
| 2078 | otherwise returns to the start of the token buffer if permissible. |
| 2079 | Returns the location of the lexed token. */ |
| 2080 | cpp_token * |
| 2081 | _cpp_lex_direct (cpp_reader *pfile) |
| 2082 | { |
| 2083 | cppchar_t c; |
| 2084 | cpp_buffer *buffer; |
| 2085 | const unsigned char *comment_start; |
| 2086 | cpp_token *result = pfile->cur_token++; |
| 2087 | |
| 2088 | fresh_line: |
| 2089 | result->flags = 0; |
| 2090 | buffer = pfile->buffer; |
| 2091 | if (buffer->need_line) |
| 2092 | { |
| 2093 | if (pfile->state.in_deferred_pragma) |
| 2094 | { |
| 2095 | result->type = CPP_PRAGMA_EOL; |
| 2096 | pfile->state.in_deferred_pragma = false; |
| 2097 | if (!pfile->state.pragma_allow_expansion) |
| 2098 | pfile->state.prevent_expansion--; |
| 2099 | return result; |
| 2100 | } |
| 2101 | if (!_cpp_get_fresh_line (pfile)) |
| 2102 | { |
| 2103 | result->type = CPP_EOF; |
| 2104 | if (!pfile->state.in_directive) |
| 2105 | { |
| 2106 | /* Tell the compiler the line number of the EOF token. */ |
| 2107 | result->src_loc = pfile->line_table->highest_line; |
| 2108 | result->flags = BOL; |
| 2109 | } |
| 2110 | return result; |
| 2111 | } |
| 2112 | if (!pfile->keep_tokens) |
| 2113 | { |
| 2114 | pfile->cur_run = &pfile->base_run; |
| 2115 | result = pfile->base_run.base; |
| 2116 | pfile->cur_token = result + 1; |
| 2117 | } |
| 2118 | result->flags = BOL; |
| 2119 | if (pfile->state.parsing_args == 2) |
| 2120 | result->flags |= PREV_WHITE; |
| 2121 | } |
| 2122 | buffer = pfile->buffer; |
| 2123 | update_tokens_line: |
| 2124 | result->src_loc = pfile->line_table->highest_line; |
| 2125 | |
| 2126 | skipped_white: |
| 2127 | if (buffer->cur >= buffer->notes[buffer->cur_note].pos |
| 2128 | && !pfile->overlaid_buffer) |
| 2129 | { |
| 2130 | _cpp_process_line_notes (pfile, false); |
| 2131 | result->src_loc = pfile->line_table->highest_line; |
| 2132 | } |
| 2133 | c = *buffer->cur++; |
| 2134 | |
| 2135 | if (pfile->forced_token_location_p) |
| 2136 | result->src_loc = *pfile->forced_token_location_p; |
| 2137 | else |
| 2138 | result->src_loc = linemap_position_for_column (pfile->line_table, |
| 2139 | CPP_BUF_COLUMN (buffer, buffer->cur)); |
| 2140 | |
| 2141 | switch (c) |
| 2142 | { |
| 2143 | case ' ': case '\t': case '\f': case '\v': case '\0': |
| 2144 | result->flags |= PREV_WHITE; |
| 2145 | skip_whitespace (pfile, c); |
| 2146 | goto skipped_white; |
| 2147 | |
| 2148 | case '\n': |
| 2149 | if (buffer->cur < buffer->rlimit) |
| 2150 | CPP_INCREMENT_LINE (pfile, 0); |
| 2151 | buffer->need_line = true; |
| 2152 | goto fresh_line; |
| 2153 | |
| 2154 | case '0': case '1': case '2': case '3': case '4': |
| 2155 | case '5': case '6': case '7': case '8': case '9': |
| 2156 | { |
| 2157 | struct normalize_state nst = INITIAL_NORMALIZE_STATE; |
| 2158 | result->type = CPP_NUMBER; |
| 2159 | lex_number (pfile, &result->val.str, &nst); |
| 2160 | warn_about_normalization (pfile, result, &nst); |
| 2161 | break; |
| 2162 | } |
| 2163 | |
| 2164 | case 'L': |
| 2165 | case 'u': |
| 2166 | case 'U': |
| 2167 | case 'R': |
| 2168 | /* 'L', 'u', 'U', 'u8' or 'R' may introduce wide characters, |
| 2169 | wide strings or raw strings. */ |
| 2170 | if (c == 'L' || CPP_OPTION (pfile, rliterals) |
| 2171 | || (c != 'R' && CPP_OPTION (pfile, uliterals))) |
| 2172 | { |
| 2173 | if ((*buffer->cur == '\'' && c != 'R') |
| 2174 | || *buffer->cur == '"' |
| 2175 | || (*buffer->cur == 'R' |
| 2176 | && c != 'R' |
| 2177 | && buffer->cur[1] == '"' |
| 2178 | && CPP_OPTION (pfile, rliterals)) |
| 2179 | || (*buffer->cur == '8' |
| 2180 | && c == 'u' |
| 2181 | && (buffer->cur[1] == '"' |
| 2182 | || (buffer->cur[1] == 'R' && buffer->cur[2] == '"' |
| 2183 | && CPP_OPTION (pfile, rliterals))))) |
| 2184 | { |
| 2185 | lex_string (pfile, result, buffer->cur - 1); |
| 2186 | break; |
| 2187 | } |
| 2188 | } |
| 2189 | /* Fall through. */ |
| 2190 | |
| 2191 | case '_': |
| 2192 | case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': |
| 2193 | case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': |
| 2194 | case 'm': case 'n': case 'o': case 'p': case 'q': case 'r': |
| 2195 | case 's': case 't': case 'v': case 'w': case 'x': |
| 2196 | case 'y': case 'z': |
| 2197 | case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': |
| 2198 | case 'G': case 'H': case 'I': case 'J': case 'K': |
| 2199 | case 'M': case 'N': case 'O': case 'P': case 'Q': |
| 2200 | case 'S': case 'T': case 'V': case 'W': case 'X': |
| 2201 | case 'Y': case 'Z': |
| 2202 | result->type = CPP_NAME; |
| 2203 | { |
| 2204 | struct normalize_state nst = INITIAL_NORMALIZE_STATE; |
| 2205 | result->val.node.node = lex_identifier (pfile, buffer->cur - 1, false, |
| 2206 | &nst); |
| 2207 | warn_about_normalization (pfile, result, &nst); |
| 2208 | } |
| 2209 | |
| 2210 | /* Convert named operators to their proper types. */ |
| 2211 | if (result->val.node.node->flags & NODE_OPERATOR) |
| 2212 | { |
| 2213 | result->flags |= NAMED_OP; |
| 2214 | result->type = (enum cpp_ttype) result->val.node.node->directive_index; |
| 2215 | } |
| 2216 | break; |
| 2217 | |
| 2218 | case '\'': |
| 2219 | case '"': |
| 2220 | lex_string (pfile, result, buffer->cur - 1); |
| 2221 | break; |
| 2222 | |
| 2223 | case '/': |
| 2224 | /* A potential block or line comment. */ |
| 2225 | comment_start = buffer->cur; |
| 2226 | c = *buffer->cur; |
| 2227 | |
| 2228 | if (c == '*') |
| 2229 | { |
| 2230 | if (_cpp_skip_block_comment (pfile)) |
| 2231 | cpp_error (pfile, CPP_DL_ERROR, "unterminated comment"); |
| 2232 | } |
| 2233 | else if (c == '/' && (CPP_OPTION (pfile, cplusplus_comments) |
| 2234 | || cpp_in_system_header (pfile))) |
| 2235 | { |
| 2236 | /* Warn about comments only if pedantically GNUC89, and not |
| 2237 | in system headers. */ |
| 2238 | if (CPP_OPTION (pfile, lang) == CLK_GNUC89 && CPP_PEDANTIC (pfile) |
| 2239 | && ! buffer->warned_cplusplus_comments) |
| 2240 | { |
| 2241 | cpp_error (pfile, CPP_DL_PEDWARN, |
| 2242 | "C++ style comments are not allowed in ISO C90"); |
| 2243 | cpp_error (pfile, CPP_DL_PEDWARN, |
| 2244 | "(this will be reported only once per input file)"); |
| 2245 | buffer->warned_cplusplus_comments = 1; |
| 2246 | } |
| 2247 | |
| 2248 | if (skip_line_comment (pfile) && CPP_OPTION (pfile, warn_comments)) |
| 2249 | cpp_warning (pfile, CPP_W_COMMENTS, "multi-line comment"); |
| 2250 | } |
| 2251 | else if (c == '=') |
| 2252 | { |
| 2253 | buffer->cur++; |
| 2254 | result->type = CPP_DIV_EQ; |
| 2255 | break; |
| 2256 | } |
| 2257 | else |
| 2258 | { |
| 2259 | result->type = CPP_DIV; |
| 2260 | break; |
| 2261 | } |
| 2262 | |
| 2263 | if (!pfile->state.save_comments) |
| 2264 | { |
| 2265 | result->flags |= PREV_WHITE; |
| 2266 | goto update_tokens_line; |
| 2267 | } |
| 2268 | |
| 2269 | /* Save the comment as a token in its own right. */ |
| 2270 | save_comment (pfile, result, comment_start, c); |
| 2271 | break; |
| 2272 | |
| 2273 | case '<': |
| 2274 | if (pfile->state.angled_headers) |
| 2275 | { |
| 2276 | lex_string (pfile, result, buffer->cur - 1); |
| 2277 | if (result->type != CPP_LESS) |
| 2278 | break; |
| 2279 | } |
| 2280 | |
| 2281 | result->type = CPP_LESS; |
| 2282 | if (*buffer->cur == '=') |
| 2283 | buffer->cur++, result->type = CPP_LESS_EQ; |
| 2284 | else if (*buffer->cur == '<') |
| 2285 | { |
| 2286 | buffer->cur++; |
| 2287 | IF_NEXT_IS ('=', CPP_LSHIFT_EQ, CPP_LSHIFT); |
| 2288 | } |
| 2289 | else if (CPP_OPTION (pfile, digraphs)) |
| 2290 | { |
| 2291 | if (*buffer->cur == ':') |
| 2292 | { |
| 2293 | buffer->cur++; |
| 2294 | result->flags |= DIGRAPH; |
| 2295 | result->type = CPP_OPEN_SQUARE; |
| 2296 | } |
| 2297 | else if (*buffer->cur == '%') |
| 2298 | { |
| 2299 | buffer->cur++; |
| 2300 | result->flags |= DIGRAPH; |
| 2301 | result->type = CPP_OPEN_BRACE; |
| 2302 | } |
| 2303 | } |
| 2304 | break; |
| 2305 | |
| 2306 | case '>': |
| 2307 | result->type = CPP_GREATER; |
| 2308 | if (*buffer->cur == '=') |
| 2309 | buffer->cur++, result->type = CPP_GREATER_EQ; |
| 2310 | else if (*buffer->cur == '>') |
| 2311 | { |
| 2312 | buffer->cur++; |
| 2313 | IF_NEXT_IS ('=', CPP_RSHIFT_EQ, CPP_RSHIFT); |
| 2314 | } |
| 2315 | break; |
| 2316 | |
| 2317 | case '%': |
| 2318 | result->type = CPP_MOD; |
| 2319 | if (*buffer->cur == '=') |
| 2320 | buffer->cur++, result->type = CPP_MOD_EQ; |
| 2321 | else if (CPP_OPTION (pfile, digraphs)) |
| 2322 | { |
| 2323 | if (*buffer->cur == ':') |
| 2324 | { |
| 2325 | buffer->cur++; |
| 2326 | result->flags |= DIGRAPH; |
| 2327 | result->type = CPP_HASH; |
| 2328 | if (*buffer->cur == '%' && buffer->cur[1] == ':') |
| 2329 | buffer->cur += 2, result->type = CPP_PASTE, result->val.token_no = 0; |
| 2330 | } |
| 2331 | else if (*buffer->cur == '>') |
| 2332 | { |
| 2333 | buffer->cur++; |
| 2334 | result->flags |= DIGRAPH; |
| 2335 | result->type = CPP_CLOSE_BRACE; |
| 2336 | } |
| 2337 | } |
| 2338 | break; |
| 2339 | |
| 2340 | case '.': |
| 2341 | result->type = CPP_DOT; |
| 2342 | if (ISDIGIT (*buffer->cur)) |
| 2343 | { |
| 2344 | struct normalize_state nst = INITIAL_NORMALIZE_STATE; |
| 2345 | result->type = CPP_NUMBER; |
| 2346 | lex_number (pfile, &result->val.str, &nst); |
| 2347 | warn_about_normalization (pfile, result, &nst); |
| 2348 | } |
| 2349 | else if (*buffer->cur == '.' && buffer->cur[1] == '.') |
| 2350 | buffer->cur += 2, result->type = CPP_ELLIPSIS; |
| 2351 | else if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) |
| 2352 | buffer->cur++, result->type = CPP_DOT_STAR; |
| 2353 | break; |
| 2354 | |
| 2355 | case '+': |
| 2356 | result->type = CPP_PLUS; |
| 2357 | if (*buffer->cur == '+') |
| 2358 | buffer->cur++, result->type = CPP_PLUS_PLUS; |
| 2359 | else if (*buffer->cur == '=') |
| 2360 | buffer->cur++, result->type = CPP_PLUS_EQ; |
| 2361 | break; |
| 2362 | |
| 2363 | case '-': |
| 2364 | result->type = CPP_MINUS; |
| 2365 | if (*buffer->cur == '>') |
| 2366 | { |
| 2367 | buffer->cur++; |
| 2368 | result->type = CPP_DEREF; |
| 2369 | if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus)) |
| 2370 | buffer->cur++, result->type = CPP_DEREF_STAR; |
| 2371 | } |
| 2372 | else if (*buffer->cur == '-') |
| 2373 | buffer->cur++, result->type = CPP_MINUS_MINUS; |
| 2374 | else if (*buffer->cur == '=') |
| 2375 | buffer->cur++, result->type = CPP_MINUS_EQ; |
| 2376 | break; |
| 2377 | |
| 2378 | case '&': |
| 2379 | result->type = CPP_AND; |
| 2380 | if (*buffer->cur == '&') |
| 2381 | buffer->cur++, result->type = CPP_AND_AND; |
| 2382 | else if (*buffer->cur == '=') |
| 2383 | buffer->cur++, result->type = CPP_AND_EQ; |
| 2384 | break; |
| 2385 | |
| 2386 | case '|': |
| 2387 | result->type = CPP_OR; |
| 2388 | if (*buffer->cur == '|') |
| 2389 | buffer->cur++, result->type = CPP_OR_OR; |
| 2390 | else if (*buffer->cur == '=') |
| 2391 | buffer->cur++, result->type = CPP_OR_EQ; |
| 2392 | break; |
| 2393 | |
| 2394 | case ':': |
| 2395 | result->type = CPP_COLON; |
| 2396 | if (*buffer->cur == ':' && CPP_OPTION (pfile, cplusplus)) |
| 2397 | buffer->cur++, result->type = CPP_SCOPE; |
| 2398 | else if (*buffer->cur == '>' && CPP_OPTION (pfile, digraphs)) |
| 2399 | { |
| 2400 | buffer->cur++; |
| 2401 | result->flags |= DIGRAPH; |
| 2402 | result->type = CPP_CLOSE_SQUARE; |
| 2403 | } |
| 2404 | break; |
| 2405 | |
| 2406 | case '*': IF_NEXT_IS ('=', CPP_MULT_EQ, CPP_MULT); break; |
| 2407 | case '=': IF_NEXT_IS ('=', CPP_EQ_EQ, CPP_EQ); break; |
| 2408 | case '!': IF_NEXT_IS ('=', CPP_NOT_EQ, CPP_NOT); break; |
| 2409 | case '^': IF_NEXT_IS ('=', CPP_XOR_EQ, CPP_XOR); break; |
| 2410 | case '#': IF_NEXT_IS ('#', CPP_PASTE, CPP_HASH); result->val.token_no = 0; break; |
| 2411 | |
| 2412 | case '?': result->type = CPP_QUERY; break; |
| 2413 | case '~': result->type = CPP_COMPL; break; |
| 2414 | case ',': result->type = CPP_COMMA; break; |
| 2415 | case '(': result->type = CPP_OPEN_PAREN; break; |
| 2416 | case ')': result->type = CPP_CLOSE_PAREN; break; |
| 2417 | case '[': result->type = CPP_OPEN_SQUARE; break; |
| 2418 | case ']': result->type = CPP_CLOSE_SQUARE; break; |
| 2419 | case '{': result->type = CPP_OPEN_BRACE; break; |
| 2420 | case '}': result->type = CPP_CLOSE_BRACE; break; |
| 2421 | case ';': result->type = CPP_SEMICOLON; break; |
| 2422 | |
| 2423 | /* @ is a punctuator in Objective-C. */ |
| 2424 | case '@': result->type = CPP_ATSIGN; break; |
| 2425 | |
| 2426 | case '$': |
| 2427 | case '\\': |
| 2428 | { |
| 2429 | const uchar *base = --buffer->cur; |
| 2430 | struct normalize_state nst = INITIAL_NORMALIZE_STATE; |
| 2431 | |
| 2432 | if (forms_identifier_p (pfile, true, &nst)) |
| 2433 | { |
| 2434 | result->type = CPP_NAME; |
| 2435 | result->val.node.node = lex_identifier (pfile, base, true, &nst); |
| 2436 | warn_about_normalization (pfile, result, &nst); |
| 2437 | break; |
| 2438 | } |
| 2439 | buffer->cur++; |
| 2440 | } |
| 2441 | |
| 2442 | default: |
| 2443 | create_literal (pfile, result, buffer->cur - 1, 1, CPP_OTHER); |
| 2444 | break; |
| 2445 | } |
| 2446 | |
| 2447 | return result; |
| 2448 | } |
| 2449 | |
| 2450 | /* An upper bound on the number of bytes needed to spell TOKEN. |
| 2451 | Does not include preceding whitespace. */ |
| 2452 | unsigned int |
| 2453 | cpp_token_len (const cpp_token *token) |
| 2454 | { |
| 2455 | unsigned int len; |
| 2456 | |
| 2457 | switch (TOKEN_SPELL (token)) |
| 2458 | { |
| 2459 | default: len = 6; break; |
| 2460 | case SPELL_LITERAL: len = token->val.str.len; break; |
| 2461 | case SPELL_IDENT: len = NODE_LEN (token->val.node.node) * 10; break; |
| 2462 | } |
| 2463 | |
| 2464 | return len; |
| 2465 | } |
| 2466 | |
| 2467 | /* Parse UTF-8 out of NAMEP and place a \U escape in BUFFER. |
| 2468 | Return the number of bytes read out of NAME. (There are always |
| 2469 | 10 bytes written to BUFFER.) */ |
| 2470 | |
| 2471 | static size_t |
| 2472 | utf8_to_ucn (unsigned char *buffer, const unsigned char *name) |
| 2473 | { |
| 2474 | int j; |
| 2475 | int ucn_len = 0; |
| 2476 | int ucn_len_c; |
| 2477 | unsigned t; |
| 2478 | unsigned long utf32; |
| 2479 | |
| 2480 | /* Compute the length of the UTF-8 sequence. */ |
| 2481 | for (t = *name; t & 0x80; t <<= 1) |
| 2482 | ucn_len++; |
| 2483 | |
| 2484 | utf32 = *name & (0x7F >> ucn_len); |
| 2485 | for (ucn_len_c = 1; ucn_len_c < ucn_len; ucn_len_c++) |
| 2486 | { |
| 2487 | utf32 = (utf32 << 6) | (*++name & 0x3F); |
| 2488 | |
| 2489 | /* Ill-formed UTF-8. */ |
| 2490 | if ((*name & ~0x3F) != 0x80) |
| 2491 | abort (); |
| 2492 | } |
| 2493 | |
| 2494 | *buffer++ = '\\'; |
| 2495 | *buffer++ = 'U'; |
| 2496 | for (j = 7; j >= 0; j--) |
| 2497 | *buffer++ = "0123456789abcdef"[(utf32 >> (4 * j)) & 0xF]; |
| 2498 | return ucn_len; |
| 2499 | } |
| 2500 | |
| 2501 | /* Given a token TYPE corresponding to a digraph, return a pointer to |
| 2502 | the spelling of the digraph. */ |
| 2503 | static const unsigned char * |
| 2504 | cpp_digraph2name (enum cpp_ttype type) |
| 2505 | { |
| 2506 | return digraph_spellings[(int) type - (int) CPP_FIRST_DIGRAPH]; |
| 2507 | } |
| 2508 | |
| 2509 | /* Write the spelling of a token TOKEN to BUFFER. The buffer must |
| 2510 | already contain the enough space to hold the token's spelling. |
| 2511 | Returns a pointer to the character after the last character written. |
| 2512 | FORSTRING is true if this is to be the spelling after translation |
| 2513 | phase 1 (this is different for UCNs). |
| 2514 | FIXME: Would be nice if we didn't need the PFILE argument. */ |
| 2515 | unsigned char * |
| 2516 | cpp_spell_token (cpp_reader *pfile, const cpp_token *token, |
| 2517 | unsigned char *buffer, bool forstring) |
| 2518 | { |
| 2519 | switch (TOKEN_SPELL (token)) |
| 2520 | { |
| 2521 | case SPELL_OPERATOR: |
| 2522 | { |
| 2523 | const unsigned char *spelling; |
| 2524 | unsigned char c; |
| 2525 | |
| 2526 | if (token->flags & DIGRAPH) |
| 2527 | spelling = cpp_digraph2name (token->type); |
| 2528 | else if (token->flags & NAMED_OP) |
| 2529 | goto spell_ident; |
| 2530 | else |
| 2531 | spelling = TOKEN_NAME (token); |
| 2532 | |
| 2533 | while ((c = *spelling++) != '\0') |
| 2534 | *buffer++ = c; |
| 2535 | } |
| 2536 | break; |
| 2537 | |
| 2538 | spell_ident: |
| 2539 | case SPELL_IDENT: |
| 2540 | if (forstring) |
| 2541 | { |
| 2542 | memcpy (buffer, NODE_NAME (token->val.node.node), |
| 2543 | NODE_LEN (token->val.node.node)); |
| 2544 | buffer += NODE_LEN (token->val.node.node); |
| 2545 | } |
| 2546 | else |
| 2547 | { |
| 2548 | size_t i; |
| 2549 | const unsigned char * name = NODE_NAME (token->val.node.node); |
| 2550 | |
| 2551 | for (i = 0; i < NODE_LEN (token->val.node.node); i++) |
| 2552 | if (name[i] & ~0x7F) |
| 2553 | { |
| 2554 | i += utf8_to_ucn (buffer, name + i) - 1; |
| 2555 | buffer += 10; |
| 2556 | } |
| 2557 | else |
| 2558 | *buffer++ = NODE_NAME (token->val.node.node)[i]; |
| 2559 | } |
| 2560 | break; |
| 2561 | |
| 2562 | case SPELL_LITERAL: |
| 2563 | memcpy (buffer, token->val.str.text, token->val.str.len); |
| 2564 | buffer += token->val.str.len; |
| 2565 | break; |
| 2566 | |
| 2567 | case SPELL_NONE: |
| 2568 | cpp_error (pfile, CPP_DL_ICE, |
| 2569 | "unspellable token %s", TOKEN_NAME (token)); |
| 2570 | break; |
| 2571 | } |
| 2572 | |
| 2573 | return buffer; |
| 2574 | } |
| 2575 | |
| 2576 | /* Returns TOKEN spelt as a null-terminated string. The string is |
| 2577 | freed when the reader is destroyed. Useful for diagnostics. */ |
| 2578 | unsigned char * |
| 2579 | cpp_token_as_text (cpp_reader *pfile, const cpp_token *token) |
| 2580 | { |
| 2581 | unsigned int len = cpp_token_len (token) + 1; |
| 2582 | unsigned char *start = _cpp_unaligned_alloc (pfile, len), *end; |
| 2583 | |
| 2584 | end = cpp_spell_token (pfile, token, start, false); |
| 2585 | end[0] = '\0'; |
| 2586 | |
| 2587 | return start; |
| 2588 | } |
| 2589 | |
| 2590 | /* Returns a pointer to a string which spells the token defined by |
| 2591 | TYPE and FLAGS. Used by C front ends, which really should move to |
| 2592 | using cpp_token_as_text. */ |
| 2593 | const char * |
| 2594 | cpp_type2name (enum cpp_ttype type, unsigned char flags) |
| 2595 | { |
| 2596 | if (flags & DIGRAPH) |
| 2597 | return (const char *) cpp_digraph2name (type); |
| 2598 | else if (flags & NAMED_OP) |
| 2599 | return cpp_named_operator2name (type); |
| 2600 | |
| 2601 | return (const char *) token_spellings[type].name; |
| 2602 | } |
| 2603 | |
| 2604 | /* Writes the spelling of token to FP, without any preceding space. |
| 2605 | Separated from cpp_spell_token for efficiency - to avoid stdio |
| 2606 | double-buffering. */ |
| 2607 | void |
| 2608 | cpp_output_token (const cpp_token *token, FILE *fp) |
| 2609 | { |
| 2610 | switch (TOKEN_SPELL (token)) |
| 2611 | { |
| 2612 | case SPELL_OPERATOR: |
| 2613 | { |
| 2614 | const unsigned char *spelling; |
| 2615 | int c; |
| 2616 | |
| 2617 | if (token->flags & DIGRAPH) |
| 2618 | spelling = cpp_digraph2name (token->type); |
| 2619 | else if (token->flags & NAMED_OP) |
| 2620 | goto spell_ident; |
| 2621 | else |
| 2622 | spelling = TOKEN_NAME (token); |
| 2623 | |
| 2624 | c = *spelling; |
| 2625 | do |
| 2626 | putc (c, fp); |
| 2627 | while ((c = *++spelling) != '\0'); |
| 2628 | } |
| 2629 | break; |
| 2630 | |
| 2631 | spell_ident: |
| 2632 | case SPELL_IDENT: |
| 2633 | { |
| 2634 | size_t i; |
| 2635 | const unsigned char * name = NODE_NAME (token->val.node.node); |
| 2636 | |
| 2637 | for (i = 0; i < NODE_LEN (token->val.node.node); i++) |
| 2638 | if (name[i] & ~0x7F) |
| 2639 | { |
| 2640 | unsigned char buffer[10]; |
| 2641 | i += utf8_to_ucn (buffer, name + i) - 1; |
| 2642 | fwrite (buffer, 1, 10, fp); |
| 2643 | } |
| 2644 | else |
| 2645 | fputc (NODE_NAME (token->val.node.node)[i], fp); |
| 2646 | } |
| 2647 | break; |
| 2648 | |
| 2649 | case SPELL_LITERAL: |
| 2650 | fwrite (token->val.str.text, 1, token->val.str.len, fp); |
| 2651 | break; |
| 2652 | |
| 2653 | case SPELL_NONE: |
| 2654 | /* An error, most probably. */ |
| 2655 | break; |
| 2656 | } |
| 2657 | } |
| 2658 | |
| 2659 | /* Compare two tokens. */ |
| 2660 | int |
| 2661 | _cpp_equiv_tokens (const cpp_token *a, const cpp_token *b) |
| 2662 | { |
| 2663 | if (a->type == b->type && a->flags == b->flags) |
| 2664 | switch (TOKEN_SPELL (a)) |
| 2665 | { |
| 2666 | default: /* Keep compiler happy. */ |
| 2667 | case SPELL_OPERATOR: |
| 2668 | /* token_no is used to track where multiple consecutive ## |
| 2669 | tokens were originally located. */ |
| 2670 | return (a->type != CPP_PASTE || a->val.token_no == b->val.token_no); |
| 2671 | case SPELL_NONE: |
| 2672 | return (a->type != CPP_MACRO_ARG |
| 2673 | || a->val.macro_arg.arg_no == b->val.macro_arg.arg_no); |
| 2674 | case SPELL_IDENT: |
| 2675 | return a->val.node.node == b->val.node.node; |
| 2676 | case SPELL_LITERAL: |
| 2677 | return (a->val.str.len == b->val.str.len |
| 2678 | && !memcmp (a->val.str.text, b->val.str.text, |
| 2679 | a->val.str.len)); |
| 2680 | } |
| 2681 | |
| 2682 | return 0; |
| 2683 | } |
| 2684 | |
| 2685 | /* Returns nonzero if a space should be inserted to avoid an |
| 2686 | accidental token paste for output. For simplicity, it is |
| 2687 | conservative, and occasionally advises a space where one is not |
| 2688 | needed, e.g. "." and ".2". */ |
| 2689 | int |
| 2690 | cpp_avoid_paste (cpp_reader *pfile, const cpp_token *token1, |
| 2691 | const cpp_token *token2) |
| 2692 | { |
| 2693 | enum cpp_ttype a = token1->type, b = token2->type; |
| 2694 | cppchar_t c; |
| 2695 | |
| 2696 | if (token1->flags & NAMED_OP) |
| 2697 | a = CPP_NAME; |
| 2698 | if (token2->flags & NAMED_OP) |
| 2699 | b = CPP_NAME; |
| 2700 | |
| 2701 | c = EOF; |
| 2702 | if (token2->flags & DIGRAPH) |
| 2703 | c = digraph_spellings[(int) b - (int) CPP_FIRST_DIGRAPH][0]; |
| 2704 | else if (token_spellings[b].category == SPELL_OPERATOR) |
| 2705 | c = token_spellings[b].name[0]; |
| 2706 | |
| 2707 | /* Quickly get everything that can paste with an '='. */ |
| 2708 | if ((int) a <= (int) CPP_LAST_EQ && c == '=') |
| 2709 | return 1; |
| 2710 | |
| 2711 | switch (a) |
| 2712 | { |
| 2713 | case CPP_GREATER: return c == '>'; |
| 2714 | case CPP_LESS: return c == '<' || c == '%' || c == ':'; |
| 2715 | case CPP_PLUS: return c == '+'; |
| 2716 | case CPP_MINUS: return c == '-' || c == '>'; |
| 2717 | case CPP_DIV: return c == '/' || c == '*'; /* Comments. */ |
| 2718 | case CPP_MOD: return c == ':' || c == '>'; |
| 2719 | case CPP_AND: return c == '&'; |
| 2720 | case CPP_OR: return c == '|'; |
| 2721 | case CPP_COLON: return c == ':' || c == '>'; |
| 2722 | case CPP_DEREF: return c == '*'; |
| 2723 | case CPP_DOT: return c == '.' || c == '%' || b == CPP_NUMBER; |
| 2724 | case CPP_HASH: return c == '#' || c == '%'; /* Digraph form. */ |
| 2725 | case CPP_NAME: return ((b == CPP_NUMBER |
| 2726 | && name_p (pfile, &token2->val.str)) |
| 2727 | || b == CPP_NAME |
| 2728 | || b == CPP_CHAR || b == CPP_STRING); /* L */ |
| 2729 | case CPP_NUMBER: return (b == CPP_NUMBER || b == CPP_NAME |
| 2730 | || c == '.' || c == '+' || c == '-'); |
| 2731 | /* UCNs */ |
| 2732 | case CPP_OTHER: return ((token1->val.str.text[0] == '\\' |
| 2733 | && b == CPP_NAME) |
| 2734 | || (CPP_OPTION (pfile, objc) |
| 2735 | && token1->val.str.text[0] == '@' |
| 2736 | && (b == CPP_NAME || b == CPP_STRING))); |
| 2737 | default: break; |
| 2738 | } |
| 2739 | |
| 2740 | return 0; |
| 2741 | } |
| 2742 | |
| 2743 | /* Output all the remaining tokens on the current line, and a newline |
| 2744 | character, to FP. Leading whitespace is removed. If there are |
| 2745 | macros, special token padding is not performed. */ |
| 2746 | void |
| 2747 | cpp_output_line (cpp_reader *pfile, FILE *fp) |
| 2748 | { |
| 2749 | const cpp_token *token; |
| 2750 | |
| 2751 | token = cpp_get_token (pfile); |
| 2752 | while (token->type != CPP_EOF) |
| 2753 | { |
| 2754 | cpp_output_token (token, fp); |
| 2755 | token = cpp_get_token (pfile); |
| 2756 | if (token->flags & PREV_WHITE) |
| 2757 | putc (' ', fp); |
| 2758 | } |
| 2759 | |
| 2760 | putc ('\n', fp); |
| 2761 | } |
| 2762 | |
| 2763 | /* Return a string representation of all the remaining tokens on the |
| 2764 | current line. The result is allocated using xmalloc and must be |
| 2765 | freed by the caller. */ |
| 2766 | unsigned char * |
| 2767 | cpp_output_line_to_string (cpp_reader *pfile, const unsigned char *dir_name) |
| 2768 | { |
| 2769 | const cpp_token *token; |
| 2770 | unsigned int out = dir_name ? ustrlen (dir_name) : 0; |
| 2771 | unsigned int alloced = 120 + out; |
| 2772 | unsigned char *result = (unsigned char *) xmalloc (alloced); |
| 2773 | |
| 2774 | /* If DIR_NAME is empty, there are no initial contents. */ |
| 2775 | if (dir_name) |
| 2776 | { |
| 2777 | sprintf ((char *) result, "#%s ", dir_name); |
| 2778 | out += 2; |
| 2779 | } |
| 2780 | |
| 2781 | token = cpp_get_token (pfile); |
| 2782 | while (token->type != CPP_EOF) |
| 2783 | { |
| 2784 | unsigned char *last; |
| 2785 | /* Include room for a possible space and the terminating nul. */ |
| 2786 | unsigned int len = cpp_token_len (token) + 2; |
| 2787 | |
| 2788 | if (out + len > alloced) |
| 2789 | { |
| 2790 | alloced *= 2; |
| 2791 | if (out + len > alloced) |
| 2792 | alloced = out + len; |
| 2793 | result = (unsigned char *) xrealloc (result, alloced); |
| 2794 | } |
| 2795 | |
| 2796 | last = cpp_spell_token (pfile, token, &result[out], 0); |
| 2797 | out = last - result; |
| 2798 | |
| 2799 | token = cpp_get_token (pfile); |
| 2800 | if (token->flags & PREV_WHITE) |
| 2801 | result[out++] = ' '; |
| 2802 | } |
| 2803 | |
| 2804 | result[out] = '\0'; |
| 2805 | return result; |
| 2806 | } |
| 2807 | |
| 2808 | /* Memory buffers. Changing these three constants can have a dramatic |
| 2809 | effect on performance. The values here are reasonable defaults, |
| 2810 | but might be tuned. If you adjust them, be sure to test across a |
| 2811 | range of uses of cpplib, including heavy nested function-like macro |
| 2812 | expansion. Also check the change in peak memory usage (NJAMD is a |
| 2813 | good tool for this). */ |
| 2814 | #define MIN_BUFF_SIZE 8000 |
| 2815 | #define BUFF_SIZE_UPPER_BOUND(MIN_SIZE) (MIN_BUFF_SIZE + (MIN_SIZE) * 3 / 2) |
| 2816 | #define EXTENDED_BUFF_SIZE(BUFF, MIN_EXTRA) \ |
| 2817 | (MIN_EXTRA + ((BUFF)->limit - (BUFF)->cur) * 2) |
| 2818 | |
| 2819 | #if MIN_BUFF_SIZE > BUFF_SIZE_UPPER_BOUND (0) |
| 2820 | #error BUFF_SIZE_UPPER_BOUND must be at least as large as MIN_BUFF_SIZE! |
| 2821 | #endif |
| 2822 | |
| 2823 | /* Create a new allocation buffer. Place the control block at the end |
| 2824 | of the buffer, so that buffer overflows will cause immediate chaos. */ |
| 2825 | static _cpp_buff * |
| 2826 | new_buff (size_t len) |
| 2827 | { |
| 2828 | _cpp_buff *result; |
| 2829 | unsigned char *base; |
| 2830 | |
| 2831 | if (len < MIN_BUFF_SIZE) |
| 2832 | len = MIN_BUFF_SIZE; |
| 2833 | len = CPP_ALIGN (len); |
| 2834 | |
| 2835 | base = XNEWVEC (unsigned char, len + sizeof (_cpp_buff)); |
| 2836 | result = (_cpp_buff *) (base + len); |
| 2837 | result->base = base; |
| 2838 | result->cur = base; |
| 2839 | result->limit = base + len; |
| 2840 | result->next = NULL; |
| 2841 | return result; |
| 2842 | } |
| 2843 | |
| 2844 | /* Place a chain of unwanted allocation buffers on the free list. */ |
| 2845 | void |
| 2846 | _cpp_release_buff (cpp_reader *pfile, _cpp_buff *buff) |
| 2847 | { |
| 2848 | _cpp_buff *end = buff; |
| 2849 | |
| 2850 | while (end->next) |
| 2851 | end = end->next; |
| 2852 | end->next = pfile->free_buffs; |
| 2853 | pfile->free_buffs = buff; |
| 2854 | } |
| 2855 | |
| 2856 | /* Return a free buffer of size at least MIN_SIZE. */ |
| 2857 | _cpp_buff * |
| 2858 | _cpp_get_buff (cpp_reader *pfile, size_t min_size) |
| 2859 | { |
| 2860 | _cpp_buff *result, **p; |
| 2861 | |
| 2862 | for (p = &pfile->free_buffs;; p = &(*p)->next) |
| 2863 | { |
| 2864 | size_t size; |
| 2865 | |
| 2866 | if (*p == NULL) |
| 2867 | return new_buff (min_size); |
| 2868 | result = *p; |
| 2869 | size = result->limit - result->base; |
| 2870 | /* Return a buffer that's big enough, but don't waste one that's |
| 2871 | way too big. */ |
| 2872 | if (size >= min_size && size <= BUFF_SIZE_UPPER_BOUND (min_size)) |
| 2873 | break; |
| 2874 | } |
| 2875 | |
| 2876 | *p = result->next; |
| 2877 | result->next = NULL; |
| 2878 | result->cur = result->base; |
| 2879 | return result; |
| 2880 | } |
| 2881 | |
| 2882 | /* Creates a new buffer with enough space to hold the uncommitted |
| 2883 | remaining bytes of BUFF, and at least MIN_EXTRA more bytes. Copies |
| 2884 | the excess bytes to the new buffer. Chains the new buffer after |
| 2885 | BUFF, and returns the new buffer. */ |
| 2886 | _cpp_buff * |
| 2887 | _cpp_append_extend_buff (cpp_reader *pfile, _cpp_buff *buff, size_t min_extra) |
| 2888 | { |
| 2889 | size_t size = EXTENDED_BUFF_SIZE (buff, min_extra); |
| 2890 | _cpp_buff *new_buff = _cpp_get_buff (pfile, size); |
| 2891 | |
| 2892 | buff->next = new_buff; |
| 2893 | memcpy (new_buff->base, buff->cur, BUFF_ROOM (buff)); |
| 2894 | return new_buff; |
| 2895 | } |
| 2896 | |
| 2897 | /* Creates a new buffer with enough space to hold the uncommitted |
| 2898 | remaining bytes of the buffer pointed to by BUFF, and at least |
| 2899 | MIN_EXTRA more bytes. Copies the excess bytes to the new buffer. |
| 2900 | Chains the new buffer before the buffer pointed to by BUFF, and |
| 2901 | updates the pointer to point to the new buffer. */ |
| 2902 | void |
| 2903 | _cpp_extend_buff (cpp_reader *pfile, _cpp_buff **pbuff, size_t min_extra) |
| 2904 | { |
| 2905 | _cpp_buff *new_buff, *old_buff = *pbuff; |
| 2906 | size_t size = EXTENDED_BUFF_SIZE (old_buff, min_extra); |
| 2907 | |
| 2908 | new_buff = _cpp_get_buff (pfile, size); |
| 2909 | memcpy (new_buff->base, old_buff->cur, BUFF_ROOM (old_buff)); |
| 2910 | new_buff->next = old_buff; |
| 2911 | *pbuff = new_buff; |
| 2912 | } |
| 2913 | |
| 2914 | /* Free a chain of buffers starting at BUFF. */ |
| 2915 | void |
| 2916 | _cpp_free_buff (_cpp_buff *buff) |
| 2917 | { |
| 2918 | _cpp_buff *next; |
| 2919 | |
| 2920 | for (; buff; buff = next) |
| 2921 | { |
| 2922 | next = buff->next; |
| 2923 | free (buff->base); |
| 2924 | } |
| 2925 | } |
| 2926 | |
| 2927 | /* Allocate permanent, unaligned storage of length LEN. */ |
| 2928 | unsigned char * |
| 2929 | _cpp_unaligned_alloc (cpp_reader *pfile, size_t len) |
| 2930 | { |
| 2931 | _cpp_buff *buff = pfile->u_buff; |
| 2932 | unsigned char *result = buff->cur; |
| 2933 | |
| 2934 | if (len > (size_t) (buff->limit - result)) |
| 2935 | { |
| 2936 | buff = _cpp_get_buff (pfile, len); |
| 2937 | buff->next = pfile->u_buff; |
| 2938 | pfile->u_buff = buff; |
| 2939 | result = buff->cur; |
| 2940 | } |
| 2941 | |
| 2942 | buff->cur = result + len; |
| 2943 | return result; |
| 2944 | } |
| 2945 | |
| 2946 | /* Allocate permanent, unaligned storage of length LEN from a_buff. |
| 2947 | That buffer is used for growing allocations when saving macro |
| 2948 | replacement lists in a #define, and when parsing an answer to an |
| 2949 | assertion in #assert, #unassert or #if (and therefore possibly |
| 2950 | whilst expanding macros). It therefore must not be used by any |
| 2951 | code that they might call: specifically the lexer and the guts of |
| 2952 | the macro expander. |
| 2953 | |
| 2954 | All existing other uses clearly fit this restriction: storing |
| 2955 | registered pragmas during initialization. */ |
| 2956 | unsigned char * |
| 2957 | _cpp_aligned_alloc (cpp_reader *pfile, size_t len) |
| 2958 | { |
| 2959 | _cpp_buff *buff = pfile->a_buff; |
| 2960 | unsigned char *result = buff->cur; |
| 2961 | |
| 2962 | if (len > (size_t) (buff->limit - result)) |
| 2963 | { |
| 2964 | buff = _cpp_get_buff (pfile, len); |
| 2965 | buff->next = pfile->a_buff; |
| 2966 | pfile->a_buff = buff; |
| 2967 | result = buff->cur; |
| 2968 | } |
| 2969 | |
| 2970 | buff->cur = result + len; |
| 2971 | return result; |
| 2972 | } |
| 2973 | |
| 2974 | /* Say which field of TOK is in use. */ |
| 2975 | |
| 2976 | enum cpp_token_fld_kind |
| 2977 | cpp_token_val_index (cpp_token *tok) |
| 2978 | { |
| 2979 | switch (TOKEN_SPELL (tok)) |
| 2980 | { |
| 2981 | case SPELL_IDENT: |
| 2982 | return CPP_TOKEN_FLD_NODE; |
| 2983 | case SPELL_LITERAL: |
| 2984 | return CPP_TOKEN_FLD_STR; |
| 2985 | case SPELL_OPERATOR: |
| 2986 | if (tok->type == CPP_PASTE) |
| 2987 | return CPP_TOKEN_FLD_TOKEN_NO; |
| 2988 | else |
| 2989 | return CPP_TOKEN_FLD_NONE; |
| 2990 | case SPELL_NONE: |
| 2991 | if (tok->type == CPP_MACRO_ARG) |
| 2992 | return CPP_TOKEN_FLD_ARG_NO; |
| 2993 | else if (tok->type == CPP_PADDING) |
| 2994 | return CPP_TOKEN_FLD_SOURCE; |
| 2995 | else if (tok->type == CPP_PRAGMA) |
| 2996 | return CPP_TOKEN_FLD_PRAGMA; |
| 2997 | /* else fall through */ |
| 2998 | default: |
| 2999 | return CPP_TOKEN_FLD_NONE; |
| 3000 | } |
| 3001 | } |
| 3002 | |
| 3003 | /* All tokens lexed in R after calling this function will be forced to have |
| 3004 | their source_location the same as the location referenced by P, until |
| 3005 | cpp_stop_forcing_token_locations is called for R. */ |
| 3006 | |
| 3007 | void |
| 3008 | cpp_force_token_locations (cpp_reader *r, source_location *p) |
| 3009 | { |
| 3010 | r->forced_token_location_p = p; |
| 3011 | } |
| 3012 | |
| 3013 | /* Go back to assigning locations naturally for lexed tokens. */ |
| 3014 | |
| 3015 | void |
| 3016 | cpp_stop_forcing_token_locations (cpp_reader *r) |
| 3017 | { |
| 3018 | r->forced_token_location_p = NULL; |
| 3019 | } |