tccpp.c

   1 /*
   2  *  TCC - Tiny C Compiler
   3  *
   4  *  Copyright (c) 2001-2004 Fabrice Bellard
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20
  21 #define USING_GLOBALS
  22 #include "tcc.h"
  23
  24 /* #define to 1 to enable (see parse_pp_string()) */
  25 #define ACCEPT_LF_IN_STRINGS 0
  26
  27 /********************************************************/
  28 /* global variables */
  29
  30 ST_DATA int tok_flags;
  31 ST_DATA int parse_flags;
  32
  33 ST_DATA struct BufferedFile *file;
  34 ST_DATA int tok;
  35 ST_DATA CValue tokc;
  36 ST_DATA const int *macro_ptr;
  37 ST_DATA CString tokcstr; /* current parsed string, if any */
  38
  39 /* display benchmark infos */
  40 ST_DATA int tok_ident;
  41 ST_DATA TokenSym **table_ident;
  42
  43 /* ------------------------------------------------------------------------- */
  44
  45 static TokenSym *hash_ident[TOK_HASH_SIZE];
  46 static char token_buf[STRING_MAX_SIZE + 1];
  47 static CString cstr_buf;
  48 static TokenString tokstr_buf;
  49 static unsigned char isidnum_table[256 - CH_EOF];
  50 static int pp_debug_tok, pp_debug_symv;
  51 static int pp_once;
  52 static int pp_expr;
  53 static int pp_counter;
  54 static void tok_print(const char *msg, const int *str);
  55
  56 static struct TinyAlloc *toksym_alloc;
  57 static struct TinyAlloc *tokstr_alloc;
  58
  59 static TokenString *macro_stack;
  60
  61 static const char tcc_keywords[] =
  62 #define DEF(id, str) str "\0"
  63 #include "tcctok.h"
  64 #undef DEF
  65 ;
  66
  67 /* WARNING: the content of this string encodes token numbers */
  68 static const unsigned char tok_two_chars[] =
  69 /* outdated -- gr
  70     "<=\236>=\235!=\225&&\240||\241++\244--\242==\224<<\1>>\2+=\253"
  71     "-=\255*=\252/=\257%=\245&=\246^=\336|=\374->\313..\250##\266";
  72 */{
  73     '<','=', TOK_LE,
  74     '>','=', TOK_GE,
  75     '!','=', TOK_NE,
  76     '&','&', TOK_LAND,
  77     '|','|', TOK_LOR,
  78     '+','+', TOK_INC,
  79     '-','-', TOK_DEC,
  80     '=','=', TOK_EQ,
  81     '<','<', TOK_SHL,
  82     '>','>', TOK_SAR,
  83     '+','=', TOK_A_ADD,
  84     '-','=', TOK_A_SUB,
  85     '*','=', TOK_A_MUL,
  86     '/','=', TOK_A_DIV,
  87     '%','=', TOK_A_MOD,
  88     '&','=', TOK_A_AND,
  89     '^','=', TOK_A_XOR,
  90     '|','=', TOK_A_OR,
  91     '-','>', TOK_ARROW,
  92     '.','.', TOK_TWODOTS,
  93     '#','#', TOK_TWOSHARPS,
  94     '#','#', TOK_PPJOIN,
  95     0
  96 };
  97
  98 static void next_nomacro(void);
  99
 100 ST_FUNC void skip(int c)
 101 {
 102     if (tok != c)
 103         tcc_error("'%c' expected (got \"%s\")", c, get_tok_str(tok, &tokc));
 104     next();
 105 }
 106
 107 ST_FUNC void expect(const char *msg)
 108 {
 109     tcc_error("%s expected", msg);
 110 }
 111
 112 /* ------------------------------------------------------------------------- */
 113 /* Custom allocator for tiny objects */
 114
 115 #define USE_TAL
 116
 117 #ifndef USE_TAL
 118 #define tal_free(al, p) tcc_free(p)
 119 #define tal_realloc(al, p, size) tcc_realloc(p, size)
 120 #define tal_new(a,b,c)
 121 #define tal_delete(a)
 122 #else
 123 #if !defined(MEM_DEBUG)
 124 #define tal_free(al, p) tal_free_impl(al, p)
 125 #define tal_realloc(al, p, size) tal_realloc_impl(&al, p, size)
 126 #define TAL_DEBUG_PARAMS
 127 #else
 128 #define TAL_DEBUG 1
 129 //#define TAL_INFO 1 /* collect and dump allocators stats */
 130 #define tal_free(al, p) tal_free_impl(al, p, __FILE__, __LINE__)
 131 #define tal_realloc(al, p, size) tal_realloc_impl(&al, p, size, __FILE__, __LINE__)
 132 #define TAL_DEBUG_PARAMS , const char *file, int line
 133 #define TAL_DEBUG_FILE_LEN 40
 134 #endif
 135
 136 #define TOKSYM_TAL_SIZE     (768 * 1024) /* allocator for tiny TokenSym in table_ident */
 137 #define TOKSTR_TAL_SIZE     (768 * 1024) /* allocator for tiny TokenString instances */
 138 #define CSTR_TAL_SIZE       (256 * 1024) /* allocator for tiny CString instances */
 139 #define TOKSYM_TAL_LIMIT    256 /* prefer unique limits to distinguish allocators debug msgs */
 140 #define TOKSTR_TAL_LIMIT    128 /* 32 * sizeof(int) */
 141 #define CSTR_TAL_LIMIT      1024
 142
 143 typedef struct TinyAlloc {
 144     unsigned  limit;
 145     unsigned  size;
 146     uint8_t *buffer;
 147     uint8_t *p;
 148     unsigned  nb_allocs;
 149     struct TinyAlloc *next, *top;
 150 #ifdef TAL_INFO
 151     unsigned  nb_peak;
 152     unsigned  nb_total;
 153     unsigned  nb_missed;
 154     uint8_t *peak_p;
 155 #endif
 156 } TinyAlloc;
 157
 158 typedef struct tal_header_t {
 159     unsigned  size;
 160 #ifdef TAL_DEBUG
 161     int     line_num; /* negative line_num used for double free check */
 162     char    file_name[TAL_DEBUG_FILE_LEN + 1];
 163 #endif
 164 } tal_header_t;
 165
 166 /* ------------------------------------------------------------------------- */
 167
 168 static TinyAlloc *tal_new(TinyAlloc **pal, unsigned limit, unsigned size)
 169 {
 170     TinyAlloc *al = tcc_mallocz(sizeof(TinyAlloc));
 171     al->p = al->buffer = tcc_malloc(size);
 172     al->limit = limit;
 173     al->size = size;
 174     if (pal) *pal = al;
 175     return al;
 176 }
 177
 178 static void tal_delete(TinyAlloc *al)
 179 {
 180     TinyAlloc *next;
 181
 182 tail_call:
 183     if (!al)
 184         return;
 185 #ifdef TAL_INFO
 186     fprintf(stderr, "limit=%5d, size=%5g MB, nb_peak=%6d, nb_total=%8d, nb_missed=%6d, usage=%5.1f%%\n",
 187             al->limit, al->size / 1024.0 / 1024.0, al->nb_peak, al->nb_total, al->nb_missed,
 188             (al->peak_p - al->buffer) * 100.0 / al->size);
 189 #endif
 190 #ifdef TAL_DEBUG
 191     if (al->nb_allocs > 0) {
 192         uint8_t *p;
 193         fprintf(stderr, "TAL_DEBUG: memory leak %d chunk(s) (limit= %d)\n",
 194                 al->nb_allocs, al->limit);
 195         p = al->buffer;
 196         while (p < al->p) {
 197             tal_header_t *header = (tal_header_t *)p;
 198             if (header->line_num > 0) {
 199                 fprintf(stderr, "%s:%d: chunk of %d bytes leaked\n",
 200                         header->file_name, header->line_num, header->size);
 201             }
 202             p += header->size + sizeof(tal_header_t);
 203         }
 204 #if MEM_DEBUG-0 == 2
 205         exit(2);
 206 #endif
 207     }
 208 #endif
 209     next = al->next;
 210     tcc_free(al->buffer);
 211     tcc_free(al);
 212     al = next;
 213     goto tail_call;
 214 }
 215
 216 static void tal_free_impl(TinyAlloc *al, void *p TAL_DEBUG_PARAMS)
 217 {
 218     if (!p)
 219         return;
 220 tail_call:
 221     if (al->buffer <= (uint8_t *)p && (uint8_t *)p < al->buffer + al->size) {
 222 #ifdef TAL_DEBUG
 223         tal_header_t *header = (((tal_header_t *)p) - 1);
 224         if (header->line_num < 0) {
 225             fprintf(stderr, "%s:%d: TAL_DEBUG: double frees chunk from\n",
 226                     file, line);
 227             fprintf(stderr, "%s:%d: %d bytes\n",
 228                     header->file_name, (int)-header->line_num, (int)header->size);
 229         } else
 230             header->line_num = -header->line_num;
 231 #endif
 232         al->nb_allocs--;
 233         if (!al->nb_allocs)
 234             al->p = al->buffer;
 235     } else if (al->next) {
 236         al = al->next;
 237         goto tail_call;
 238     }
 239     else
 240         tcc_free(p);
 241 }
 242
 243 static void *tal_realloc_impl(TinyAlloc **pal, void *p, unsigned size TAL_DEBUG_PARAMS)
 244 {
 245     tal_header_t *header;
 246     void *ret;
 247     int is_own;
 248     unsigned adj_size = (size + 3) & -4;
 249     TinyAlloc *al = *pal;
 250
 251 tail_call:
 252     is_own = (al->buffer <= (uint8_t *)p && (uint8_t *)p < al->buffer + al->size);
 253     if ((!p || is_own) && size <= al->limit) {
 254         if (al->p - al->buffer + adj_size + sizeof(tal_header_t) < al->size) {
 255             header = (tal_header_t *)al->p;
 256             header->size = adj_size;
 257 #ifdef TAL_DEBUG
 258             { int ofs = strlen(file) - TAL_DEBUG_FILE_LEN;
 259             strncpy(header->file_name, file + (ofs > 0 ? ofs : 0), TAL_DEBUG_FILE_LEN);
 260             header->file_name[TAL_DEBUG_FILE_LEN] = 0;
 261             header->line_num = line; }
 262 #endif
 263             ret = al->p + sizeof(tal_header_t);
 264             al->p += adj_size + sizeof(tal_header_t);
 265             if (is_own) {
 266                 header = (((tal_header_t *)p) - 1);
 267                 if (p) memcpy(ret, p, header->size);
 268 #ifdef TAL_DEBUG
 269                 header->line_num = -header->line_num;
 270 #endif
 271             } else {
 272                 al->nb_allocs++;
 273             }
 274 #ifdef TAL_INFO
 275             if (al->nb_peak < al->nb_allocs)
 276                 al->nb_peak = al->nb_allocs;
 277             if (al->peak_p < al->p)
 278                 al->peak_p = al->p;
 279             al->nb_total++;
 280 #endif
 281             return ret;
 282         } else if (is_own) {
 283             al->nb_allocs--;
 284             ret = tal_realloc(*pal, 0, size);
 285             header = (((tal_header_t *)p) - 1);
 286             if (p) memcpy(ret, p, header->size);
 287 #ifdef TAL_DEBUG
 288             header->line_num = -header->line_num;
 289 #endif
 290             return ret;
 291         }
 292         if (al->next) {
 293             al = al->next;
 294         } else {
 295             TinyAlloc *bottom = al, *next = al->top ? al->top : al;
 296
 297             al = tal_new(pal, next->limit, next->size * 2);
 298             al->next = next;
 299             bottom->top = al;
 300         }
 301         goto tail_call;
 302     }
 303     if (is_own) {
 304         al->nb_allocs--;
 305         ret = tcc_malloc(size);
 306         header = (((tal_header_t *)p) - 1);
 307         if (p) memcpy(ret, p, header->size);
 308 #ifdef TAL_DEBUG
 309         header->line_num = -header->line_num;
 310 #endif
 311     } else if (al->next) {
 312         al = al->next;
 313         goto tail_call;
 314     } else
 315         ret = tcc_realloc(p, size);
 316 #ifdef TAL_INFO
 317     al->nb_missed++;
 318 #endif
 319     return ret;
 320 }
 321
 322 #endif /* USE_TAL */
 323
 324 /* ------------------------------------------------------------------------- */
 325 /* CString handling */
 326 static void cstr_realloc(CString *cstr, int new_size)
 327 {
 328     int size;
 329
 330     size = cstr->size_allocated;
 331     if (size < 8)
 332         size = 8; /* no need to allocate a too small first string */
 333     while (size < new_size)
 334         size = size * 2;
 335     cstr->data = tcc_realloc(cstr->data, size);
 336     cstr->size_allocated = size;
 337 }
 338
 339 /* add a byte */
 340 ST_INLN void cstr_ccat(CString *cstr, int ch)
 341 {
 342     int size;
 343     size = cstr->size + 1;
 344     if (size > cstr->size_allocated)
 345         cstr_realloc(cstr, size);
 346     ((unsigned char *)cstr->data)[size - 1] = ch;
 347     cstr->size = size;
 348 }
 349
 350 ST_INLN char *unicode_to_utf8 (char *b, uint32_t Uc)
 351 {
 352     if (Uc<0x80) *b++=Uc;
 353     else if (Uc<0x800) *b++=192+Uc/64, *b++=128+Uc%64;
 354     else if (Uc-0xd800u<0x800) goto error;
 355     else if (Uc<0x10000) *b++=224+Uc/4096, *b++=128+Uc/64%64, *b++=128+Uc%64;
 356     else if (Uc<0x110000) *b++=240+Uc/262144, *b++=128+Uc/4096%64, *b++=128+Uc/64%64, *b++=128+Uc%64;
 357     else error: tcc_error("0x%x is not a valid universal character", Uc);
 358     return b;
 359 }
 360
 361 /* add a unicode character expanded into utf8 */
 362 ST_INLN void cstr_u8cat(CString *cstr, int ch)
 363 {
 364     char buf[4], *e;
 365     e = unicode_to_utf8(buf, (uint32_t)ch);
 366     cstr_cat(cstr, buf, e - buf);
 367 }
 368
 369 ST_FUNC void cstr_cat(CString *cstr, const char *str, int len)
 370 {
 371     int size;
 372     if (len <= 0)
 373         len = strlen(str) + 1 + len;
 374     size = cstr->size + len;
 375     if (size > cstr->size_allocated)
 376         cstr_realloc(cstr, size);
 377     memmove(((unsigned char *)cstr->data) + cstr->size, str, len);
 378     cstr->size = size;
 379 }
 380
 381 /* add a wide char */
 382 ST_FUNC void cstr_wccat(CString *cstr, int ch)
 383 {
 384     int size;
 385     size = cstr->size + sizeof(nwchar_t);
 386     if (size > cstr->size_allocated)
 387         cstr_realloc(cstr, size);
 388     *(nwchar_t *)(((unsigned char *)cstr->data) + size - sizeof(nwchar_t)) = ch;
 389     cstr->size = size;
 390 }
 391
 392 ST_FUNC void cstr_new(CString *cstr)
 393 {
 394     memset(cstr, 0, sizeof(CString));
 395 }
 396
 397 /* free string and reset it to NULL */
 398 ST_FUNC void cstr_free(CString *cstr)
 399 {
 400     tcc_free(cstr->data);
 401 }
 402
 403 /* reset string to empty */
 404 ST_FUNC void cstr_reset(CString *cstr)
 405 {
 406     cstr->size = 0;
 407 }
 408
 409 ST_FUNC int cstr_vprintf(CString *cstr, const char *fmt, va_list ap)
 410 {
 411     va_list v;
 412     int len, size = 80;
 413     for (;;) {
 414         size += cstr->size;
 415         if (size > cstr->size_allocated)
 416             cstr_realloc(cstr, size);
 417         size = cstr->size_allocated - cstr->size;
 418         va_copy(v, ap);
 419         len = vsnprintf((char*)cstr->data + cstr->size, size, fmt, v);
 420         va_end(v);
 421         if (len >= 0 && len < size)
 422             break;
 423         size *= 2;
 424     }
 425     cstr->size += len;
 426     return len;
 427 }
 428
 429 ST_FUNC int cstr_printf(CString *cstr, const char *fmt, ...)
 430 {
 431     va_list ap; int len;
 432     va_start(ap, fmt);
 433     len = cstr_vprintf(cstr, fmt, ap);
 434     va_end(ap);
 435     return len;
 436 }
 437
 438 /* XXX: unicode ? */
 439 static void add_char(CString *cstr, int c)
 440 {
 441     if (c == '\'' || c == '\"' || c == '\\') {
 442         /* XXX: could be more precise if char or string */
 443         cstr_ccat(cstr, '\\');
 444     }
 445     if (c >= 32 && c <= 126) {
 446         cstr_ccat(cstr, c);
 447     } else {
 448         cstr_ccat(cstr, '\\');
 449         if (c == '\n') {
 450             cstr_ccat(cstr, 'n');
 451         } else {
 452             cstr_ccat(cstr, '0' + ((c >> 6) & 7));
 453             cstr_ccat(cstr, '0' + ((c >> 3) & 7));
 454             cstr_ccat(cstr, '0' + (c & 7));
 455         }
 456     }
 457 }
 458
 459 /* ------------------------------------------------------------------------- */
 460 /* allocate a new token */
 461 static TokenSym *tok_alloc_new(TokenSym **pts, const char *str, int len)
 462 {
 463     TokenSym *ts, **ptable;
 464     int i;
 465
 466     if (tok_ident >= SYM_FIRST_ANOM)
 467         tcc_error("memory full (symbols)");
 468
 469     /* expand token table if needed */
 470     i = tok_ident - TOK_IDENT;
 471     if ((i % TOK_ALLOC_INCR) == 0) {
 472         ptable = tcc_realloc(table_ident, (i + TOK_ALLOC_INCR) * sizeof(TokenSym *));
 473         table_ident = ptable;
 474     }
 475
 476     ts = tal_realloc(toksym_alloc, 0, sizeof(TokenSym) + len);
 477     table_ident[i] = ts;
 478     ts->tok = tok_ident++;
 479     ts->sym_define = NULL;
 480     ts->sym_label = NULL;
 481     ts->sym_struct = NULL;
 482     ts->sym_identifier = NULL;
 483     ts->len = len;
 484     ts->hash_next = NULL;
 485     memcpy(ts->str, str, len);
 486     ts->str[len] = '\0';
 487     *pts = ts;
 488     return ts;
 489 }
 490
 491 #define TOK_HASH_INIT 1
 492 #define TOK_HASH_FUNC(h, c) ((h) + ((h) << 5) + ((h) >> 27) + (c))
 493
 494
 495 /* find a token and add it if not found */
 496 ST_FUNC TokenSym *tok_alloc(const char *str, int len)
 497 {
 498     TokenSym *ts, **pts;
 499     int i;
 500     unsigned int h;
 501
 502     h = TOK_HASH_INIT;
 503     for(i=0;i<len;i++)
 504         h = TOK_HASH_FUNC(h, ((unsigned char *)str)[i]);
 505     h &= (TOK_HASH_SIZE - 1);
 506
 507     pts = &hash_ident[h];
 508     for(;;) {
 509         ts = *pts;
 510         if (!ts)
 511             break;
 512         if (ts->len == len && !memcmp(ts->str, str, len))
 513             return ts;
 514         pts = &(ts->hash_next);
 515     }
 516     return tok_alloc_new(pts, str, len);
 517 }
 518
 519 ST_FUNC int tok_alloc_const(const char *str)
 520 {
 521     return tok_alloc(str, strlen(str))->tok;
 522 }
 523
 524
 525 /* XXX: buffer overflow */
 526 /* XXX: float tokens */
 527 ST_FUNC const char *get_tok_str(int v, CValue *cv)
 528 {
 529     char *p;
 530     int i, len;
 531
 532     cstr_reset(&cstr_buf);
 533     p = cstr_buf.data;
 534
 535     switch(v) {
 536     case TOK_CINT:
 537     case TOK_CUINT:
 538     case TOK_CLONG:
 539     case TOK_CULONG:
 540     case TOK_CLLONG:
 541     case TOK_CULLONG:
 542         /* XXX: not quite exact, but only useful for testing  */
 543 #ifdef _WIN32
 544         sprintf(p, "%u", (unsigned)cv->i);
 545 #else
 546         sprintf(p, "%llu", (unsigned long long)cv->i);
 547 #endif
 548         break;
 549     case TOK_LCHAR:
 550         cstr_ccat(&cstr_buf, 'L');
 551     case TOK_CCHAR:
 552         cstr_ccat(&cstr_buf, '\'');
 553         add_char(&cstr_buf, cv->i);
 554         cstr_ccat(&cstr_buf, '\'');
 555         cstr_ccat(&cstr_buf, '\0');
 556         break;
 557     case TOK_PPNUM:
 558     case TOK_PPSTR:
 559         return (char*)cv->str.data;
 560     case TOK_LSTR:
 561         cstr_ccat(&cstr_buf, 'L');
 562     case TOK_STR:
 563         cstr_ccat(&cstr_buf, '\"');
 564         if (v == TOK_STR) {
 565             len = cv->str.size - 1;
 566             for(i=0;i<len;i++)
 567                 add_char(&cstr_buf, ((unsigned char *)cv->str.data)[i]);
 568         } else {
 569             len = (cv->str.size / sizeof(nwchar_t)) - 1;
 570             for(i=0;i<len;i++)
 571                 add_char(&cstr_buf, ((nwchar_t *)cv->str.data)[i]);
 572         }
 573         cstr_ccat(&cstr_buf, '\"');
 574         cstr_ccat(&cstr_buf, '\0');
 575         break;
 576
 577     case TOK_CFLOAT:
 578         return strcpy(p, "<float>");
 579     case TOK_CDOUBLE:
 580         return strcpy(p, "<double>");
 581     case TOK_CLDOUBLE:
 582         return strcpy(p, "<long double>");
 583     case TOK_LINENUM:
 584         return strcpy(p, "<linenumber");
 585
 586     /* above tokens have value, the ones below don't */
 587     case TOK_LT:
 588         v = '<';
 589         goto addv;
 590     case TOK_GT:
 591         v = '>';
 592         goto addv;
 593     case TOK_DOTS:
 594         return strcpy(p, "...");
 595     case TOK_A_SHL:
 596         return strcpy(p, "<<=");
 597     case TOK_A_SAR:
 598         return strcpy(p, ">>=");
 599     case TOK_EOF:
 600         return strcpy(p, "<eof>");
 601     case 0: /* anonymous nameless symbols */
 602         return strcpy(p, "<no name>");
 603     default:
 604         if (v < TOK_IDENT) {
 605             /* search in two bytes table */
 606             const unsigned char *q = tok_two_chars;
 607             while (*q) {
 608                 if (q[2] == v) {
 609                     *p++ = q[0];
 610                     *p++ = q[1];
 611                     *p = '\0';
 612                     return cstr_buf.data;
 613                 }
 614                 q += 3;
 615             }
 616             if (v >= 127 || (v < 32 && !is_space(v) && v != '\n')) {
 617                 sprintf(p, "<\\x%02x>", v);
 618                 break;
 619             }
 620     addv:
 621             *p++ = v;
 622             *p = '\0';
 623         } else if (v < tok_ident) {
 624             return table_ident[v - TOK_IDENT]->str;
 625         } else if (v >= SYM_FIRST_ANOM) {
 626             /* special name for anonymous symbol */
 627             sprintf(p, "L.%u", v - SYM_FIRST_ANOM);
 628         } else {
 629             /* should never happen */
 630             return NULL;
 631         }
 632         break;
 633     }
 634     return cstr_buf.data;
 635 }
 636
 637 static inline int check_space(int t, int *spc)
 638 {
 639     if (t < 256 && (isidnum_table[t - CH_EOF] & IS_SPC)) {
 640         if (*spc)
 641             return 1;
 642         *spc = 1;
 643     } else
 644         *spc = 0;
 645     return 0;
 646 }
 647
 648 /* return the current character, handling end of block if necessary
 649    (but not stray) */
 650 static int handle_eob(void)
 651 {
 652     BufferedFile *bf = file;
 653     int len;
 654
 655     /* only tries to read if really end of buffer */
 656     if (bf->buf_ptr >= bf->buf_end) {
 657         if (bf->fd >= 0) {
 658 #if defined(PARSE_DEBUG)
 659             len = 1;
 660 #else
 661             len = IO_BUF_SIZE;
 662 #endif
 663             len = read(bf->fd, bf->buffer, len);
 664             if (len < 0)
 665                 len = 0;
 666         } else {
 667             len = 0;
 668         }
 669         total_bytes += len;
 670         bf->buf_ptr = bf->buffer;
 671         bf->buf_end = bf->buffer + len;
 672         *bf->buf_end = CH_EOB;
 673     }
 674     if (bf->buf_ptr < bf->buf_end) {
 675         return bf->buf_ptr[0];
 676     } else {
 677         bf->buf_ptr = bf->buf_end;
 678         return CH_EOF;
 679     }
 680 }
 681
 682 /* read next char from current input file and handle end of input buffer */
 683 static int next_c(void)
 684 {
 685     int ch = *++file->buf_ptr;
 686     /* end of buffer/file handling */
 687     if (ch == CH_EOB && file->buf_ptr >= file->buf_end)
 688         ch = handle_eob();
 689     return ch;
 690 }
 691
 692 /* input with '\[\r]\n' handling. */
 693 static int handle_stray_noerror(int err)
 694 {
 695     int ch;
 696     while ((ch = next_c()) == '\\') {
 697         ch = next_c();
 698         if (ch == '\n') {
 699     newl:
 700             file->line_num++;
 701         } else {
 702             if (ch == '\r') {
 703                 ch = next_c();
 704                 if (ch == '\n')
 705                     goto newl;
 706                 *--file->buf_ptr = '\r';
 707             }
 708             if (err)
 709                 tcc_error("stray '\\' in program");
 710             /* may take advantage of 'BufferedFile.unget[4}' */
 711             return *--file->buf_ptr = '\\';
 712         }
 713     }
 714     return ch;
 715 }
 716
 717 #define ninp() handle_stray_noerror(0)
 718
 719 /* handle '\\' in strings, comments and skipped regions */
 720 static int handle_bs(uint8_t **p)
 721 {
 722     int c;
 723     file->buf_ptr = *p - 1;
 724     c = ninp();
 725     *p = file->buf_ptr;
 726     return c;
 727 }
 728
 729 /* skip the stray and handle the \\n case. Output an error if
 730    incorrect char after the stray */
 731 static int handle_stray(uint8_t **p)
 732 {
 733     int c;
 734     file->buf_ptr = *p - 1;
 735     c = handle_stray_noerror(!(parse_flags & PARSE_FLAG_ACCEPT_STRAYS));
 736     *p = file->buf_ptr;
 737     return c;
 738 }
 739
 740 /* handle the complicated stray case */
 741 #define PEEKC(c, p)\
 742 {\
 743     c = *++p;\
 744     if (c == '\\')\
 745         c = handle_stray(&p); \
 746 }
 747
 748 static int skip_spaces(void)
 749 {
 750     int ch;
 751     --file->buf_ptr;
 752     do {
 753         ch = ninp();
 754     } while (isidnum_table[ch - CH_EOF] & IS_SPC);
 755     return ch;
 756 }
 757
 758 /* single line C++ comments */
 759 static uint8_t *parse_line_comment(uint8_t *p)
 760 {
 761     int c;
 762     for(;;) {
 763         for (;;) {
 764             c = *++p;
 765     redo:
 766             if (c == '\n' || c == '\\')
 767                 break;
 768             c = *++p;
 769             if (c == '\n' || c == '\\')
 770                 break;
 771         }
 772         if (c == '\n')
 773             break;
 774         c = handle_bs(&p);
 775         if (c == CH_EOF)
 776             break;
 777         if (c != '\\')
 778             goto redo;
 779     }
 780     return p;
 781 }
 782
 783 /* C comments */
 784 static uint8_t *parse_comment(uint8_t *p)
 785 {
 786     int c;
 787     for(;;) {
 788         /* fast skip loop */
 789         for(;;) {
 790             c = *++p;
 791         redo:
 792             if (c == '\n' || c == '*' || c == '\\')
 793                 break;
 794             c = *++p;
 795             if (c == '\n' || c == '*' || c == '\\')
 796                 break;
 797         }
 798         /* now we can handle all the cases */
 799         if (c == '\n') {
 800             file->line_num++;
 801         } else if (c == '*') {
 802             do {
 803                 c = *++p;
 804             } while (c == '*');
 805             if (c == '\\')
 806                 c = handle_bs(&p);
 807             if (c == '/')
 808                 break;
 809             goto check_eof;
 810         } else {
 811             c = handle_bs(&p);
 812         check_eof:
 813             if (c == CH_EOF)
 814                 tcc_error("unexpected end of file in comment");
 815             if (c != '\\')
 816                 goto redo;
 817         }
 818     }
 819     return p + 1;
 820 }
 821
 822 /* parse a string without interpreting escapes */
 823 static uint8_t *parse_pp_string(uint8_t *p, int sep, CString *str)
 824 {
 825     int c;
 826     for(;;) {
 827         c = *++p;
 828     redo:
 829         if (c == sep) {
 830             break;
 831         } else if (c == '\\') {
 832             c = handle_bs(&p);
 833             if (c == CH_EOF) {
 834         unterminated_string:
 835                 /* XXX: indicate line number of start of string */
 836                 tok_flags &= ~TOK_FLAG_BOL;
 837                 tcc_error("missing terminating %c character", sep);
 838             } else if (c == '\\') {
 839                 if (str)
 840                     cstr_ccat(str, c);
 841                 c = *++p;
 842                 /* add char after '\\' unconditionally */
 843                 if (c == '\\') {
 844                     c = handle_bs(&p);
 845                     if (c == CH_EOF)
 846                         goto unterminated_string;
 847                 }
 848                 goto add_char;
 849             } else {
 850                 goto redo;
 851             }
 852         } else if (c == '\n') {
 853         add_lf:
 854             if (ACCEPT_LF_IN_STRINGS) {
 855                 file->line_num++;
 856                 goto add_char;
 857             } else if (str) { /* not skipping */
 858                 goto unterminated_string;
 859             } else {
 860                 //tcc_warning("missing terminating %c character", sep);
 861                 return p;
 862             }
 863         } else if (c == '\r') {
 864             c = *++p;
 865             if (c == '\\')
 866                 c = handle_bs(&p);
 867             if (c == '\n')
 868                 goto add_lf;
 869             if (c == CH_EOF)
 870                 goto unterminated_string;
 871             if (str)
 872                 cstr_ccat(str, '\r');
 873             goto redo;
 874         } else {
 875         add_char:
 876             if (str)
 877                 cstr_ccat(str, c);
 878         }
 879     }
 880     p++;
 881     return p;
 882 }
 883
 884 /* skip block of text until #else, #elif or #endif. skip also pairs of
 885    #if/#endif */
 886 static void preprocess_skip(void)
 887 {
 888     int a, start_of_line, c, in_warn_or_error;
 889     uint8_t *p;
 890
 891     p = file->buf_ptr;
 892     a = 0;
 893 redo_start:
 894     start_of_line = 1;
 895     in_warn_or_error = 0;
 896     for(;;) {
 897     redo_no_start:
 898         c = *p;
 899         switch(c) {
 900         case ' ':
 901         case '\t':
 902         case '\f':
 903         case '\v':
 904         case '\r':
 905             p++;
 906             goto redo_no_start;
 907         case '\n':
 908             file->line_num++;
 909             p++;
 910             goto redo_start;
 911         case '\\':
 912             c = handle_bs(&p);
 913             if (c == CH_EOF)
 914                 expect("#endif");
 915             if (c == '\\')
 916                 ++p;
 917             goto redo_no_start;
 918         /* skip strings */
 919         case '\"':
 920         case '\'':
 921             if (in_warn_or_error)
 922                 goto _default;
 923             tok_flags &= ~TOK_FLAG_BOL;
 924             p = parse_pp_string(p, c, NULL);
 925             break;
 926         /* skip comments */
 927         case '/':
 928             if (in_warn_or_error)
 929                 goto _default;
 930             ++p;
 931             c = handle_bs(&p);
 932             if (c == '*') {
 933                 p = parse_comment(p);
 934             } else if (c == '/') {
 935                 p = parse_line_comment(p);
 936             }
 937             break;
 938         case '#':
 939             p++;
 940             if (start_of_line) {
 941                 file->buf_ptr = p;
 942                 next_nomacro();
 943                 p = file->buf_ptr;
 944                 if (a == 0 &&
 945                     (tok == TOK_ELSE || tok == TOK_ELIF || tok == TOK_ENDIF))
 946                     goto the_end;
 947                 if (tok == TOK_IF || tok == TOK_IFDEF || tok == TOK_IFNDEF)
 948                     a++;
 949                 else if (tok == TOK_ENDIF)
 950                     a--;
 951                 else if( tok == TOK_ERROR || tok == TOK_WARNING)
 952                     in_warn_or_error = 1;
 953                 else if (tok == TOK_LINEFEED)
 954                     goto redo_start;
 955                 else if (parse_flags & PARSE_FLAG_ASM_FILE)
 956                     p = parse_line_comment(p - 1);
 957             }
 958 #if !defined(TCC_TARGET_ARM)
 959             else if (parse_flags & PARSE_FLAG_ASM_FILE)
 960                 p = parse_line_comment(p - 1);
 961 #else
 962             /* ARM assembly uses '#' for constants */
 963 #endif
 964             break;
 965 _default:
 966         default:
 967             p++;
 968             break;
 969         }
 970         start_of_line = 0;
 971     }
 972  the_end: ;
 973     file->buf_ptr = p;
 974 }
 975
 976 #if 0
 977 /* return the number of additional 'ints' necessary to store the
 978    token */
 979 static inline int tok_size(const int *p)
 980 {
 981     switch(*p) {
 982         /* 4 bytes */
 983     case TOK_CINT:
 984     case TOK_CUINT:
 985     case TOK_CCHAR:
 986     case TOK_LCHAR:
 987     case TOK_CFLOAT:
 988     case TOK_LINENUM:
 989         return 1 + 1;
 990     case TOK_STR:
 991     case TOK_LSTR:
 992     case TOK_PPNUM:
 993     case TOK_PPSTR:
 994         return 1 + ((sizeof(CString) + ((CString *)(p+1))->size + 3) >> 2);
 995     case TOK_CLONG:
 996     case TOK_CULONG:
 997         return 1 + LONG_SIZE / 4;
 998     case TOK_CDOUBLE:
 999     case TOK_CLLONG:
1000     case TOK_CULLONG:
1001         return 1 + 2;
1002     case TOK_CLDOUBLE:
1003 #ifdef TCC_USING_DOUBLE_FOR_LDOUBLE
1004         return 1 + 8 / 4;
1005 #else
1006         return 1 + LDOUBLE_SIZE / 4;
1007 #endif
1008     default:
1009         return 1 + 0;
1010     }
1011 }
1012 #endif
1013
1014 /* token string handling */
1015 ST_INLN void tok_str_new(TokenString *s)
1016 {
1017     s->str = NULL;
1018     s->len = s->lastlen = 0;
1019     s->allocated_len = 0;
1020     s->last_line_num = -1;
1021 }
1022
1023 ST_FUNC TokenString *tok_str_alloc(void)
1024 {
1025     TokenString *str = tal_realloc(tokstr_alloc, 0, sizeof *str);
1026     tok_str_new(str);
1027     return str;
1028 }
1029
1030 ST_FUNC int *tok_str_dup(TokenString *s)
1031 {
1032     int *str;
1033
1034     str = tal_realloc(tokstr_alloc, 0, s->len * sizeof(int));
1035     memcpy(str, s->str, s->len * sizeof(int));
1036     return str;
1037 }
1038
1039 ST_FUNC void tok_str_free_str(int *str)
1040 {
1041     tal_free(tokstr_alloc, str);
1042 }
1043
1044 ST_FUNC void tok_str_free(TokenString *str)
1045 {
1046     tok_str_free_str(str->str);
1047     tal_free(tokstr_alloc, str);
1048 }
1049
1050 ST_FUNC int *tok_str_realloc(TokenString *s, int new_size)
1051 {
1052     int *str, size;
1053
1054     size = s->allocated_len;
1055     if (size < 16)
1056         size = 16;
1057     while (size < new_size)
1058         size = size * 2;
1059     if (size > s->allocated_len) {
1060         str = tal_realloc(tokstr_alloc, s->str, size * sizeof(int));
1061         s->allocated_len = size;
1062         s->str = str;
1063     }
1064     return s->str;
1065 }
1066
1067 ST_FUNC void tok_str_add(TokenString *s, int t)
1068 {
1069     int len, *str;
1070
1071     len = s->len;
1072     str = s->str;
1073     if (len >= s->allocated_len)
1074         str = tok_str_realloc(s, len + 1);
1075     str[len++] = t;
1076     s->len = len;
1077 }
1078
1079 ST_FUNC void begin_macro(TokenString *str, int alloc)
1080 {
1081     str->alloc = alloc;
1082     str->prev = macro_stack;
1083     str->prev_ptr = macro_ptr;
1084     str->save_line_num = file->line_num;
1085     macro_ptr = str->str;
1086     macro_stack = str;
1087 }
1088
1089 ST_FUNC void end_macro(void)
1090 {
1091     TokenString *str = macro_stack;
1092     macro_stack = str->prev;
1093     macro_ptr = str->prev_ptr;
1094     file->line_num = str->save_line_num;
1095     str->len = 0; /* matters if str not alloced, may be tokstr_buf */
1096     if (str->alloc != 0) {
1097         if (str->alloc == 2)
1098             str->str = NULL; /* don't free */
1099         tok_str_free(str);
1100     }
1101 }
1102
1103 static void tok_str_add2(TokenString *s, int t, CValue *cv)
1104 {
1105     int len, *str;
1106
1107     len = s->lastlen = s->len;
1108     str = s->str;
1109
1110     /* allocate space for worst case */
1111     if (len + TOK_MAX_SIZE >= s->allocated_len)
1112         str = tok_str_realloc(s, len + TOK_MAX_SIZE + 1);
1113     str[len++] = t;
1114     switch(t) {
1115     case TOK_CINT:
1116     case TOK_CUINT:
1117     case TOK_CCHAR:
1118     case TOK_LCHAR:
1119     case TOK_CFLOAT:
1120     case TOK_LINENUM:
1121 #if LONG_SIZE == 4
1122     case TOK_CLONG:
1123     case TOK_CULONG:
1124 #endif
1125         str[len++] = cv->tab[0];
1126         break;
1127     case TOK_PPNUM:
1128     case TOK_PPSTR:
1129     case TOK_STR:
1130     case TOK_LSTR:
1131         {
1132             /* Insert the string into the int array. */
1133             size_t nb_words =
1134                 1 + (cv->str.size + sizeof(int) - 1) / sizeof(int);
1135             if (len + nb_words >= s->allocated_len)
1136                 str = tok_str_realloc(s, len + nb_words + 1);
1137             str[len] = cv->str.size;
1138             memcpy(&str[len + 1], cv->str.data, cv->str.size);
1139             len += nb_words;
1140         }
1141         break;
1142     case TOK_CDOUBLE:
1143     case TOK_CLLONG:
1144     case TOK_CULLONG:
1145 #if LONG_SIZE == 8
1146     case TOK_CLONG:
1147     case TOK_CULONG:
1148 #endif
1149         str[len++] = cv->tab[0];
1150         str[len++] = cv->tab[1];
1151         break;
1152     case TOK_CLDOUBLE:
1153 #if LDOUBLE_SIZE == 8 || defined TCC_USING_DOUBLE_FOR_LDOUBLE
1154         str[len++] = cv->tab[0];
1155         str[len++] = cv->tab[1];
1156 #elif LDOUBLE_SIZE == 12
1157         str[len++] = cv->tab[0];
1158         str[len++] = cv->tab[1];
1159         str[len++] = cv->tab[2];
1160 #elif LDOUBLE_SIZE == 16
1161         str[len++] = cv->tab[0];
1162         str[len++] = cv->tab[1];
1163         str[len++] = cv->tab[2];
1164         str[len++] = cv->tab[3];
1165 #else
1166 #error add long double size support
1167 #endif
1168         break;
1169     default:
1170         break;
1171     }
1172     s->len = len;
1173 }
1174
1175 /* add the current parse token in token string 's' */
1176 ST_FUNC void tok_str_add_tok(TokenString *s)
1177 {
1178     CValue cval;
1179
1180     /* save line number info */
1181     if (file->line_num != s->last_line_num) {
1182         s->last_line_num = file->line_num;
1183         cval.i = s->last_line_num;
1184         tok_str_add2(s, TOK_LINENUM, &cval);
1185     }
1186     tok_str_add2(s, tok, &tokc);
1187 }
1188
1189 /* get a token from an integer array and increment pointer. */
1190 static inline void tok_get(int *t, const int **pp, CValue *cv)
1191 {
1192     const int *p = *pp;
1193     int n, *tab;
1194
1195     tab = cv->tab;
1196     switch(*t = *p++) {
1197 #if LONG_SIZE == 4
1198     case TOK_CLONG:
1199 #endif
1200     case TOK_CINT:
1201     case TOK_CCHAR:
1202     case TOK_LCHAR:
1203     case TOK_LINENUM:
1204         cv->i = *p++;
1205         break;
1206 #if LONG_SIZE == 4
1207     case TOK_CULONG:
1208 #endif
1209     case TOK_CUINT:
1210         cv->i = (unsigned)*p++;
1211         break;
1212     case TOK_CFLOAT:
1213         tab[0] = *p++;
1214         break;
1215     case TOK_STR:
1216     case TOK_LSTR:
1217     case TOK_PPNUM:
1218     case TOK_PPSTR:
1219         cv->str.size = *p++;
1220         cv->str.data = p;
1221         p += (cv->str.size + sizeof(int) - 1) / sizeof(int);
1222         break;
1223     case TOK_CDOUBLE:
1224     case TOK_CLLONG:
1225     case TOK_CULLONG:
1226 #if LONG_SIZE == 8
1227     case TOK_CLONG:
1228     case TOK_CULONG:
1229 #endif
1230         n = 2;
1231         goto copy;
1232     case TOK_CLDOUBLE:
1233 #if LDOUBLE_SIZE == 8 || defined TCC_USING_DOUBLE_FOR_LDOUBLE
1234         n = 2;
1235 #elif LDOUBLE_SIZE == 12
1236         n = 3;
1237 #elif LDOUBLE_SIZE == 16
1238         n = 4;
1239 #else
1240 # error add long double size support
1241 #endif
1242     copy:
1243         do
1244             *tab++ = *p++;
1245         while (--n);
1246         break;
1247     default:
1248         break;
1249     }
1250     *pp = p;
1251 }
1252
1253 #if 0
1254 # define TOK_GET(t,p,c) tok_get(t,p,c)
1255 #else
1256 # define TOK_GET(t,p,c) do { \
1257     int _t = **(p); \
1258     if (TOK_HAS_VALUE(_t)) \
1259         tok_get(t, p, c); \
1260     else \
1261         *(t) = _t, ++*(p); \
1262     } while (0)
1263 #endif
1264
1265 static int macro_is_equal(const int *a, const int *b)
1266 {
1267     CValue cv;
1268     int t;
1269
1270     if (!a || !b)
1271         return 1;
1272
1273     while (*a && *b) {
1274         cstr_reset(&tokcstr);
1275         TOK_GET(&t, &a, &cv);
1276         cstr_cat(&tokcstr, get_tok_str(t, &cv), 0);
1277         TOK_GET(&t, &b, &cv);
1278         if (strcmp(tokcstr.data, get_tok_str(t, &cv)))
1279             return 0;
1280     }
1281     return !(*a || *b);
1282 }
1283
1284 /* defines handling */
1285 ST_INLN void define_push(int v, int macro_type, int *str, Sym *first_arg)
1286 {
1287     Sym *s, *o;
1288
1289     o = define_find(v);
1290     s = sym_push2(&define_stack, v, macro_type, 0);
1291     s->d = str;
1292     s->next = first_arg;
1293     table_ident[v - TOK_IDENT]->sym_define = s;
1294
1295     if (o && !macro_is_equal(o->d, s->d))
1296         tcc_warning("%s redefined", get_tok_str(v, NULL));
1297 }
1298
1299 /* undefined a define symbol. Its name is just set to zero */
1300 ST_FUNC void define_undef(Sym *s)
1301 {
1302     int v = s->v;
1303     if (v >= TOK_IDENT && v < tok_ident)
1304         table_ident[v - TOK_IDENT]->sym_define = NULL;
1305 }
1306
1307 ST_INLN Sym *define_find(int v)
1308 {
1309     v -= TOK_IDENT;
1310     if ((unsigned)v >= (unsigned)(tok_ident - TOK_IDENT))
1311         return NULL;
1312     return table_ident[v]->sym_define;
1313 }
1314
1315 /* free define stack until top reaches 'b' */
1316 ST_FUNC void free_defines(Sym *b)
1317 {
1318     while (define_stack != b) {
1319         Sym *top = define_stack;
1320         define_stack = top->prev;
1321         tok_str_free_str(top->d);
1322         define_undef(top);
1323         sym_free(top);
1324     }
1325 }
1326
1327 /* fake the nth "#if defined test_..." for tcc -dt -run */
1328 static void maybe_run_test(TCCState *s)
1329 {
1330     const char *p;
1331     if (s->include_stack_ptr != s->include_stack)
1332         return;
1333     p = get_tok_str(tok, NULL);
1334     if (0 != memcmp(p, "test_", 5))
1335         return;
1336     if (0 != --s->run_test)
1337         return;
1338     fprintf(s->ppfp, &"\n[%s]\n"[!(s->dflag & 32)], p), fflush(s->ppfp);
1339     define_push(tok, MACRO_OBJ, NULL, NULL);
1340 }
1341
1342 static CachedInclude *
1343 search_cached_include(TCCState *s1, const char *filename, int add);
1344
1345 static int parse_include(TCCState *s1, int do_next, int test)
1346 {
1347     int c, i;
1348     char name[1024], buf[1024], *p;
1349     CachedInclude *e;
1350
1351     c = skip_spaces();
1352     if (c == '<' || c == '\"') {
1353         cstr_reset(&tokcstr);
1354         file->buf_ptr = parse_pp_string(file->buf_ptr, c == '<' ? '>' : c, &tokcstr);
1355         i = tokcstr.size;
1356         pstrncpy(name, tokcstr.data, i >= sizeof name ? sizeof name - 1 : i);
1357         next_nomacro();
1358     } else {
1359         /* computed #include : concatenate tokens until result is one of
1360            the two accepted forms.  Don't convert pp-tokens to tokens here. */
1361         parse_flags = PARSE_FLAG_PREPROCESS
1362                     | PARSE_FLAG_LINEFEED
1363                     | (parse_flags & PARSE_FLAG_ASM_FILE);
1364         name[0] = 0;
1365         for (;;) {
1366             next();
1367             p = name, i = strlen(p) - 1;
1368             if (i > 0
1369                 && ((p[0] == '"' && p[i] == '"')
1370                  || (p[0] == '<' && p[i] == '>')))
1371                 break;
1372             if (tok == TOK_LINEFEED)
1373                 tcc_error("'#include' expects \"FILENAME\" or <FILENAME>");
1374             pstrcat(name, sizeof name, get_tok_str(tok, &tokc));
1375         }
1376         c = p[0];
1377         /* remove '<>|""' */
1378         memmove(p, p + 1, i - 1), p[i - 1] = 0;
1379     }
1380
1381     i = do_next ? file->include_next_index : -1;
1382     for (;;) {
1383         ++i;
1384         if (i == 0) {
1385             /* check absolute include path */
1386             if (!IS_ABSPATH(name))
1387                 continue;
1388             buf[0] = '\0';
1389         } else if (i == 1) {
1390             /* search in file's dir if "header.h" */
1391             if (c != '\"')
1392                 continue;
1393             p = file->true_filename;
1394             pstrncpy(buf, p, tcc_basename(p) - p);
1395         } else {
1396             int j = i - 2, k = j - s1->nb_include_paths;
1397             if (k < 0)
1398                 p = s1->include_paths[j];
1399             else if (k < s1->nb_sysinclude_paths)
1400                 p = s1->sysinclude_paths[k];
1401             else if (test)
1402                 return 0;
1403             else
1404                 tcc_error("include file '%s' not found", name);
1405             pstrcpy(buf, sizeof buf, p);
1406             pstrcat(buf, sizeof buf, "/");
1407         }
1408         pstrcat(buf, sizeof buf, name);
1409         e = search_cached_include(s1, buf, 0);
1410         if (e && (define_find(e->ifndef_macro) || e->once == pp_once)) {
1411             /* no need to parse the include because the 'ifndef macro'
1412                is defined (or had #pragma once) */
1413 #ifdef INC_DEBUG
1414             printf("%s: skipping cached %s\n", file->filename, buf);
1415 #endif
1416             return 1;
1417         }
1418         if (tcc_open(s1, buf) >= 0)
1419             break;
1420     }
1421
1422     if (test) {
1423         tcc_close();
1424     } else {
1425         if (s1->include_stack_ptr >= s1->include_stack + INCLUDE_STACK_SIZE)
1426             tcc_error("#include recursion too deep");
1427         /* push previous file on stack */
1428         *s1->include_stack_ptr++ = file->prev;
1429         file->include_next_index = i;
1430 #ifdef INC_DEBUG
1431         printf("%s: including %s\n", file->prev->filename, file->filename);
1432 #endif
1433         /* update target deps */
1434         if (s1->gen_deps) {
1435             BufferedFile *bf = file;
1436             while (i == 1 && (bf = bf->prev))
1437                 i = bf->include_next_index;
1438             /* skip system include files */
1439             if (s1->include_sys_deps || i - 2 < s1->nb_include_paths)
1440                 dynarray_add(&s1->target_deps, &s1->nb_target_deps,
1441                     tcc_strdup(buf));
1442         }
1443         /* add include file debug info */
1444         tcc_debug_bincl(s1);
1445         tok_flags |= TOK_FLAG_BOF | TOK_FLAG_BOL;
1446     }
1447     return 1;
1448 }
1449
1450 /* eval an expression for #if/#elif */
1451 static int expr_preprocess(TCCState *s1)
1452 {
1453     int c, t;
1454     TokenString *str;
1455
1456     str = tok_str_alloc();
1457     pp_expr = 1;
1458     while (tok != TOK_LINEFEED && tok != TOK_EOF) {
1459         next(); /* do macro subst */
1460       redo:
1461         if (tok == TOK_DEFINED) {
1462             next_nomacro();
1463             t = tok;
1464             if (t == '(')
1465                 next_nomacro();
1466             if (tok < TOK_IDENT)
1467                 expect("identifier");
1468             if (s1->run_test)
1469                 maybe_run_test(s1);
1470             c = 0;
1471             if (define_find(tok)
1472                 || tok == TOK___HAS_INCLUDE
1473                 || tok == TOK___HAS_INCLUDE_NEXT)
1474                 c = 1;
1475             if (t == '(') {
1476                 next_nomacro();
1477                 if (tok != ')')
1478                     expect("')'");
1479             }
1480             tok = TOK_CINT;
1481             tokc.i = c;
1482         } else if (tok == TOK___HAS_INCLUDE ||
1483                    tok == TOK___HAS_INCLUDE_NEXT) {
1484             t = tok;
1485             next_nomacro();
1486             if (tok != '(')
1487                 expect("(");
1488             c = parse_include(s1, t - TOK___HAS_INCLUDE, 1);
1489             if (tok != ')')
1490                 expect("')'");
1491             tok = TOK_CINT;
1492             tokc.i = c;
1493         } else if (tok >= TOK_IDENT) {
1494             /* if undefined macro, replace with zero, check for func-like */
1495             t = tok;
1496             tok = TOK_CINT;
1497             tokc.i = 0;
1498             tok_str_add_tok(str);
1499             next();
1500             if (tok == '(')
1501                 tcc_error("function-like macro '%s' is not defined",
1502                           get_tok_str(t, NULL));
1503             goto redo;
1504         }
1505         tok_str_add_tok(str);
1506     }
1507     pp_expr = 0;
1508     tok_str_add(str, -1); /* simulate end of file */
1509     tok_str_add(str, 0);
1510     /* now evaluate C constant expression */
1511     begin_macro(str, 1);
1512     next();
1513     c = expr_const();
1514     end_macro();
1515     return c != 0;
1516 }
1517
1518
1519 /* parse after #define */
1520 ST_FUNC void parse_define(void)
1521 {
1522     Sym *s, *first, **ps;
1523     int v, t, varg, is_vaargs, spc;
1524     int saved_parse_flags = parse_flags;
1525
1526     v = tok;
1527     if (v < TOK_IDENT || v == TOK_DEFINED)
1528         tcc_error("invalid macro name '%s'", get_tok_str(tok, &tokc));
1529     /* XXX: should check if same macro (ANSI) */
1530     first = NULL;
1531     t = MACRO_OBJ;
1532     /* We have to parse the whole define as if not in asm mode, in particular
1533        no line comment with '#' must be ignored.  Also for function
1534        macros the argument list must be parsed without '.' being an ID
1535        character.  */
1536     parse_flags = ((parse_flags & ~PARSE_FLAG_ASM_FILE) | PARSE_FLAG_SPACES);
1537     /* '(' must be just after macro definition for MACRO_FUNC */
1538     next_nomacro();
1539     parse_flags &= ~PARSE_FLAG_SPACES;
1540     if (tok == '(') {
1541         int dotid = set_idnum('.', 0);
1542         next_nomacro();
1543         ps = &first;
1544         if (tok != ')') for (;;) {
1545             varg = tok;
1546             next_nomacro();
1547             is_vaargs = 0;
1548             if (varg == TOK_DOTS) {
1549                 varg = TOK___VA_ARGS__;
1550                 is_vaargs = 1;
1551             } else if (tok == TOK_DOTS && gnu_ext) {
1552                 is_vaargs = 1;
1553                 next_nomacro();
1554             }
1555             if (varg < TOK_IDENT)
1556         bad_list:
1557                 tcc_error("bad macro parameter list");
1558             s = sym_push2(&define_stack, varg | SYM_FIELD, is_vaargs, 0);
1559             *ps = s;
1560             ps = &s->next;
1561             if (tok == ')')
1562                 break;
1563             if (tok != ',' || is_vaargs)
1564                 goto bad_list;
1565             next_nomacro();
1566         }
1567         parse_flags |= PARSE_FLAG_SPACES;
1568         next_nomacro();
1569         t = MACRO_FUNC;
1570         set_idnum('.', dotid);
1571     }
1572
1573     tokstr_buf.len = 0;
1574     spc = 2;
1575     parse_flags |= PARSE_FLAG_ACCEPT_STRAYS | PARSE_FLAG_SPACES | PARSE_FLAG_LINEFEED;
1576     /* The body of a macro definition should be parsed such that identifiers
1577        are parsed like the file mode determines (i.e. with '.' being an
1578        ID character in asm mode).  But '#' should be retained instead of
1579        regarded as line comment leader, so still don't set ASM_FILE
1580        in parse_flags. */
1581     while (tok != TOK_LINEFEED && tok != TOK_EOF) {
1582         /* remove spaces around ## and after '#' */
1583         if (TOK_TWOSHARPS == tok) {
1584             if (2 == spc)
1585                 goto bad_twosharp;
1586             if (1 == spc)
1587                 --tokstr_buf.len;
1588             spc = 3;
1589             tok = TOK_PPJOIN;
1590         } else if ('#' == tok) {
1591             spc = 4;
1592         } else if (check_space(tok, &spc)) {
1593             goto skip;
1594         }
1595         tok_str_add2(&tokstr_buf, tok, &tokc);
1596     skip:
1597         next_nomacro();
1598     }
1599
1600     parse_flags = saved_parse_flags;
1601     if (spc == 1)
1602         --tokstr_buf.len; /* remove trailing space */
1603     tok_str_add(&tokstr_buf, 0);
1604     if (3 == spc)
1605 bad_twosharp:
1606         tcc_error("'##' cannot appear at either end of macro");
1607     define_push(v, t, tok_str_dup(&tokstr_buf), first);
1608 }
1609
1610 #ifdef _WIN32
1611 static unsigned long long calc_file_hash(const char *filename)
1612 {
1613     unsigned long long hash = 14695981039346656037ull; // FNV_offset_basis;
1614     FILE *fp = fopen (filename, "rb");
1615
1616     if (fp == NULL)
1617         return 0;
1618     for (;;) {
1619         unsigned char temp[IO_BUF_SIZE];
1620         int i, n = fread(temp, 1, sizeof(temp), fp);
1621
1622         if (n <= 0)
1623             break;
1624         for (i = 0; i < n; i++)
1625             hash = hash * 1099511628211ull ^ temp[i]; // FNV_prime
1626     }
1627     fclose(fp);
1628     return hash ? hash : 1ull;
1629 }
1630 #endif
1631
1632 static CachedInclude *search_cached_include(TCCState *s1, const char *filename, int add)
1633 {
1634     unsigned int h = 0;
1635     CachedInclude *e;
1636     int i;
1637     struct stat st;
1638 #ifdef _WIN32
1639     unsigned long long hash = 0;
1640 #endif
1641
1642     /* This is needed for #pragmae once
1643      * We cannot use stat on windows because st_ino is not set correctly
1644      * so we calculate a hash of file contents.
1645      * This also works for hard/soft links as in gcc/clang.
1646      */
1647     memset (&st, 0, sizeof(st));
1648     if (stat (filename, &st))
1649         goto  skip;
1650     h = st.st_size & (CACHED_INCLUDES_HASH_SIZE - 1);
1651 #ifdef _WIN32
1652     /* Only calculate file hash if file size same. */
1653     i = s1->cached_includes_hash[h];
1654     for(;;) {
1655         if (i == 0)
1656             break;
1657         e = s1->cached_includes[i - 1];
1658         if (e->st.st_size == st.st_size) {
1659             if (0 == PATHCMP(e->filename, filename)) {
1660                 hash = e->hash;
1661                 break;
1662             }
1663             if (e->hash == 0)
1664                 e->hash = calc_file_hash(e->filename);
1665             if (hash == 0)
1666                 hash = calc_file_hash(filename);
1667         }
1668         i = e->hash_next;
1669     }
1670 #endif
1671
1672     i = s1->cached_includes_hash[h];
1673     for(;;) {
1674         if (i == 0)
1675             break;
1676         e = s1->cached_includes[i - 1];
1677 #ifdef _WIN32
1678         if (e->st.st_size == st.st_size && e->hash == hash)
1679 #else
1680         if (st.st_dev == e->st.st_dev && st.st_ino == e->st.st_ino)
1681 #endif
1682             return e;
1683         i = e->hash_next;
1684     }
1685 skip:
1686     if (!add)
1687         return NULL;
1688
1689     e = tcc_malloc(sizeof(CachedInclude) + strlen(filename));
1690     e->st = st;
1691 #ifdef _WIN32
1692     e->hash = hash;
1693 #endif
1694     strcpy(e->filename, filename);
1695     e->ifndef_macro = e->once = 0;
1696     dynarray_add(&s1->cached_includes, &s1->nb_cached_includes, e);
1697     /* add in hash table */
1698     e->hash_next = s1->cached_includes_hash[h];
1699     s1->cached_includes_hash[h] = s1->nb_cached_includes;
1700 #ifdef INC_DEBUG
1701     printf("adding cached '%s'\n", filename);
1702 #endif
1703     return e;
1704 }
1705
1706 static void pragma_parse(TCCState *s1)
1707 {
1708     next_nomacro();
1709     if (tok == TOK_push_macro || tok == TOK_pop_macro) {
1710         int t = tok, v;
1711         Sym *s;
1712
1713         if (next(), tok != '(')
1714             goto pragma_err;
1715         if (next(), tok != TOK_STR)
1716             goto pragma_err;
1717         v = tok_alloc(tokc.str.data, tokc.str.size - 1)->tok;
1718         if (next(), tok != ')')
1719             goto pragma_err;
1720         if (t == TOK_push_macro) {
1721             while (NULL == (s = define_find(v)))
1722                 define_push(v, 0, NULL, NULL);
1723             s->type.ref = s; /* set push boundary */
1724         } else {
1725             for (s = define_stack; s; s = s->prev)
1726                 if (s->v == v && s->type.ref == s) {
1727                     s->type.ref = NULL;
1728                     break;
1729                 }
1730         }
1731         if (s)
1732             table_ident[v - TOK_IDENT]->sym_define = s->d ? s : NULL;
1733         else
1734             tcc_warning("unbalanced #pragma pop_macro");
1735         pp_debug_tok = t, pp_debug_symv = v;
1736
1737     } else if (tok == TOK_once) {
1738         search_cached_include(s1, file->filename, 1)->once = pp_once;
1739
1740     } else if (s1->output_type == TCC_OUTPUT_PREPROCESS) {
1741         /* tcc -E: keep pragmas below unchanged */
1742         unget_tok(' ');
1743         unget_tok(TOK_PRAGMA);
1744         unget_tok('#');
1745         unget_tok(TOK_LINEFEED);
1746
1747     } else if (tok == TOK_pack) {
1748         /* This may be:
1749            #pragma pack(1) // set
1750            #pragma pack() // reset to default
1751            #pragma pack(push) // push current
1752            #pragma pack(push,1) // push & set
1753            #pragma pack(pop) // restore previous */
1754         next();
1755         skip('(');
1756         if (tok == TOK_ASM_pop) {
1757             next();
1758             if (s1->pack_stack_ptr <= s1->pack_stack) {
1759             stk_error:
1760                 tcc_error("out of pack stack");
1761             }
1762             s1->pack_stack_ptr--;
1763         } else {
1764             int val = 0;
1765             if (tok != ')') {
1766                 if (tok == TOK_ASM_push) {
1767                     next();
1768                     if (s1->pack_stack_ptr >= s1->pack_stack + PACK_STACK_SIZE - 1)
1769                         goto stk_error;
1770                     val = *s1->pack_stack_ptr++;
1771                     if (tok != ',')
1772                         goto pack_set;
1773                     next();
1774                 }
1775                 if (tok != TOK_CINT)
1776                     goto pragma_err;
1777                 val = tokc.i;
1778                 if (val < 1 || val > 16 || (val & (val - 1)) != 0)
1779                     goto pragma_err;
1780                 next();
1781             }
1782         pack_set:
1783             *s1->pack_stack_ptr = val;
1784         }
1785         if (tok != ')')
1786             goto pragma_err;
1787
1788     } else if (tok == TOK_comment) {
1789         char *p; int t;
1790         next();
1791         skip('(');
1792         t = tok;
1793         next();
1794         skip(',');
1795         if (tok != TOK_STR)
1796             goto pragma_err;
1797         p = tcc_strdup((char *)tokc.str.data);
1798         next();
1799         if (tok != ')')
1800             goto pragma_err;
1801         if (t == TOK_lib) {
1802             dynarray_add(&s1->pragma_libs, &s1->nb_pragma_libs, p);
1803         } else {
1804             if (t == TOK_option)
1805                 tcc_set_options(s1, p);
1806             tcc_free(p);
1807         }
1808
1809     } else
1810         tcc_warning_c(warn_unsupported)("#pragma %s ignored", get_tok_str(tok, &tokc));
1811     return;
1812
1813 pragma_err:
1814     tcc_error("malformed #pragma directive");
1815     return;
1816 }
1817
1818 /* is_bof is true if first non space token at beginning of file */
1819 ST_FUNC void preprocess(int is_bof)
1820 {
1821     TCCState *s1 = tcc_state;
1822     int c, n, saved_parse_flags;
1823     char buf[1024], *q;
1824     Sym *s;
1825
1826     saved_parse_flags = parse_flags;
1827     parse_flags = PARSE_FLAG_PREPROCESS
1828         | PARSE_FLAG_TOK_NUM
1829         | PARSE_FLAG_TOK_STR
1830         | PARSE_FLAG_LINEFEED
1831         | (parse_flags & PARSE_FLAG_ASM_FILE)
1832         ;
1833
1834     next_nomacro();
1835  redo:
1836     switch(tok) {
1837     case TOK_DEFINE:
1838         pp_debug_tok = tok;
1839         next_nomacro();
1840         pp_debug_symv = tok;
1841         parse_define();
1842         break;
1843     case TOK_UNDEF:
1844         pp_debug_tok = tok;
1845         next_nomacro();
1846         pp_debug_symv = tok;
1847         s = define_find(tok);
1848         /* undefine symbol by putting an invalid name */
1849         if (s)
1850             define_undef(s);
1851         break;
1852     case TOK_INCLUDE:
1853     case TOK_INCLUDE_NEXT:
1854         parse_include(s1, tok - TOK_INCLUDE, 0);
1855         break;
1856     case TOK_IFNDEF:
1857         c = 1;
1858         goto do_ifdef;
1859     case TOK_IF:
1860         c = expr_preprocess(s1);
1861         goto do_if;
1862     case TOK_IFDEF:
1863         c = 0;
1864     do_ifdef:
1865         next_nomacro();
1866         if (tok < TOK_IDENT)
1867             tcc_error("invalid argument for '#if%sdef'", c ? "n" : "");
1868         if (is_bof) {
1869             if (c) {
1870 #ifdef INC_DEBUG
1871                 printf("#ifndef %s\n", get_tok_str(tok, NULL));
1872 #endif
1873                 file->ifndef_macro = tok;
1874             }
1875         }
1876         if (define_find(tok)
1877             || tok == TOK___HAS_INCLUDE
1878             || tok == TOK___HAS_INCLUDE_NEXT)
1879             c ^= 1;
1880     do_if:
1881         if (s1->ifdef_stack_ptr >= s1->ifdef_stack + IFDEF_STACK_SIZE)
1882             tcc_error("memory full (ifdef)");
1883         *s1->ifdef_stack_ptr++ = c;
1884         goto test_skip;
1885     case TOK_ELSE:
1886         if (s1->ifdef_stack_ptr == s1->ifdef_stack)
1887             tcc_error("#else without matching #if");
1888         if (s1->ifdef_stack_ptr[-1] & 2)
1889             tcc_error("#else after #else");
1890         c = (s1->ifdef_stack_ptr[-1] ^= 3);
1891         goto test_else;
1892     case TOK_ELIF:
1893         if (s1->ifdef_stack_ptr == s1->ifdef_stack)
1894             tcc_error("#elif without matching #if");
1895         c = s1->ifdef_stack_ptr[-1];
1896         if (c > 1)
1897             tcc_error("#elif after #else");
1898         /* last #if/#elif expression was true: we skip */
1899         if (c == 1) {
1900             c = 0;
1901         } else {
1902             c = expr_preprocess(s1);
1903             s1->ifdef_stack_ptr[-1] = c;
1904         }
1905     test_else:
1906         if (s1->ifdef_stack_ptr == file->ifdef_stack_ptr + 1)
1907             file->ifndef_macro = 0;
1908     test_skip:
1909         if (!(c & 1)) {
1910             preprocess_skip();
1911             is_bof = 0;
1912             goto redo;
1913         }
1914         break;
1915     case TOK_ENDIF:
1916         if (s1->ifdef_stack_ptr <= file->ifdef_stack_ptr)
1917             tcc_error("#endif without matching #if");
1918         s1->ifdef_stack_ptr--;
1919         /* '#ifndef macro' was at the start of file. Now we check if
1920            an '#endif' is exactly at the end of file */
1921         if (file->ifndef_macro &&
1922             s1->ifdef_stack_ptr == file->ifdef_stack_ptr) {
1923             file->ifndef_macro_saved = file->ifndef_macro;
1924             /* need to set to zero to avoid false matches if another
1925                #ifndef at middle of file */
1926             file->ifndef_macro = 0;
1927             while (tok != TOK_LINEFEED)
1928                 next_nomacro();
1929             tok_flags |= TOK_FLAG_ENDIF;
1930             goto the_end;
1931         }
1932         break;
1933     case TOK_PPNUM:
1934         n = strtoul((char*)tokc.str.data, &q, 10);
1935         goto _line_num;
1936     case TOK_LINE:
1937         next();
1938         if (tok != TOK_CINT)
1939     _line_err:
1940             tcc_error("wrong #line format");
1941         n = tokc.i;
1942     _line_num:
1943         next();
1944         if (tok != TOK_LINEFEED) {
1945             if (tok == TOK_STR) {
1946                 if (file->true_filename == file->filename)
1947                     file->true_filename = tcc_strdup(file->filename);
1948                 q = (char *)tokc.str.data;
1949                 buf[0] = 0;
1950                 if (!IS_ABSPATH(q)) {
1951                     /* prepend directory from real file */
1952                     pstrcpy(buf, sizeof buf, file->true_filename);
1953                     *tcc_basename(buf) = 0;
1954                 }
1955                 pstrcat(buf, sizeof buf, q);
1956                 tcc_debug_putfile(s1, buf);
1957             } else if (parse_flags & PARSE_FLAG_ASM_FILE)
1958                 break;
1959             else
1960                 goto _line_err;
1961             --n;
1962         }
1963         if (file->fd > 0)
1964             total_lines += file->line_num - n;
1965         file->line_num = n;
1966         break;
1967     case TOK_ERROR:
1968     case TOK_WARNING:
1969         q = buf;
1970         c = skip_spaces();
1971         while (c != '\n' && c != CH_EOF) {
1972             if ((q - buf) < sizeof(buf) - 1)
1973                 *q++ = c;
1974             c = ninp();
1975         }
1976         *q = '\0';
1977         if (tok == TOK_ERROR)
1978             tcc_error("#error %s", buf);
1979         else
1980             tcc_warning("#warning %s", buf);
1981         break;
1982     case TOK_PRAGMA:
1983         pragma_parse(s1);
1984         break;
1985     case TOK_LINEFEED:
1986         goto the_end;
1987     default:
1988         /* ignore gas line comment in an 'S' file. */
1989         if (saved_parse_flags & PARSE_FLAG_ASM_FILE)
1990             goto ignore;
1991         if (tok == '!' && is_bof)
1992             /* '!' is ignored at beginning to allow C scripts. */
1993             goto ignore;
1994         tcc_warning("Ignoring unknown preprocessing directive #%s", get_tok_str(tok, &tokc));
1995     ignore:
1996         file->buf_ptr = parse_line_comment(file->buf_ptr - 1);
1997         goto the_end;
1998     }
1999     /* ignore other preprocess commands or #! for C scripts */
2000     while (tok != TOK_LINEFEED)
2001         next_nomacro();
2002  the_end:
2003     parse_flags = saved_parse_flags;
2004 }
2005
2006 /* evaluate escape codes in a string. */
2007 static void parse_escape_string(CString *outstr, const uint8_t *buf, int is_long)
2008 {
2009     int c, n, i;
2010     const uint8_t *p;
2011
2012     p = buf;
2013     for(;;) {
2014         c = *p;
2015         if (c == '\0')
2016             break;
2017         if (c == '\\') {
2018             p++;
2019             /* escape */
2020             c = *p;
2021             switch(c) {
2022             case '0': case '1': case '2': case '3':
2023             case '4': case '5': case '6': case '7':
2024                 /* at most three octal digits */
2025                 n = c - '0';
2026                 p++;
2027                 c = *p;
2028                 if (isoct(c)) {
2029                     n = n * 8 + c - '0';
2030                     p++;
2031                     c = *p;
2032                     if (isoct(c)) {
2033                         n = n * 8 + c - '0';
2034                         p++;
2035                     }
2036                 }
2037                 c = n;
2038                 goto add_char_nonext;
2039             case 'x': i = 0; goto parse_hex_or_ucn;
2040             case 'u': i = 4; goto parse_hex_or_ucn;
2041             case 'U': i = 8; goto parse_hex_or_ucn;
2042     parse_hex_or_ucn:
2043                 p++;
2044                 n = 0;
2045                 do {
2046                     c = *p;
2047                     if (c >= 'a' && c <= 'f')
2048                         c = c - 'a' + 10;
2049                     else if (c >= 'A' && c <= 'F')
2050                         c = c - 'A' + 10;
2051                     else if (isnum(c))
2052                         c = c - '0';
2053                     else if (i > 0)
2054                         expect("more hex digits in universal-character-name");
2055                     else
2056                         goto add_hex_or_ucn;
2057                     n = n * 16 + c;
2058                     p++;
2059                 } while (--i);
2060                 if (is_long) {
2061     add_hex_or_ucn:
2062                     c = n;
2063                     goto add_char_nonext;
2064                 }
2065                 cstr_u8cat(outstr, n);
2066                 continue;
2067             case 'a':
2068                 c = '\a';
2069                 break;
2070             case 'b':
2071                 c = '\b';
2072                 break;
2073             case 'f':
2074                 c = '\f';
2075                 break;
2076             case 'n':
2077                 c = '\n';
2078                 break;
2079             case 'r':
2080                 c = '\r';
2081                 break;
2082             case 't':
2083                 c = '\t';
2084                 break;
2085             case 'v':
2086                 c = '\v';
2087                 break;
2088             case 'e':
2089                 if (!gnu_ext)
2090                     goto invalid_escape;
2091                 c = 27;
2092                 break;
2093             case '\'':
2094             case '\"':
2095             case '\\':
2096             case '?':
2097                 break;
2098             default:
2099             invalid_escape:
2100                 if (c >= '!' && c <= '~')
2101                     tcc_warning("unknown escape sequence: \'\\%c\'", c);
2102                 else
2103                     tcc_warning("unknown escape sequence: \'\\x%x\'", c);
2104                 break;
2105             }
2106         } else if (is_long && c >= 0x80) {
2107             /* assume we are processing UTF-8 sequence */
2108             /* reference: The Unicode Standard, Version 10.0, ch3.9 */
2109
2110             int cont; /* count of continuation bytes */
2111             int skip; /* how many bytes should skip when error occurred */
2112             int i;
2113
2114             /* decode leading byte */
2115             if (c < 0xC2) {
2116                     skip = 1; goto invalid_utf8_sequence;
2117             } else if (c <= 0xDF) {
2118                     cont = 1; n = c & 0x1f;
2119             } else if (c <= 0xEF) {
2120                     cont = 2; n = c & 0xf;
2121             } else if (c <= 0xF4) {
2122                     cont = 3; n = c & 0x7;
2123             } else {
2124                     skip = 1; goto invalid_utf8_sequence;
2125             }
2126
2127             /* decode continuation bytes */
2128             for (i = 1; i <= cont; i++) {
2129                 int l = 0x80, h = 0xBF;
2130
2131                 /* adjust limit for second byte */
2132                 if (i == 1) {
2133                     switch (c) {
2134                     case 0xE0: l = 0xA0; break;
2135                     case 0xED: h = 0x9F; break;
2136                     case 0xF0: l = 0x90; break;
2137                     case 0xF4: h = 0x8F; break;
2138                     }
2139                 }
2140
2141                 if (p[i] < l || p[i] > h) {
2142                     skip = i; goto invalid_utf8_sequence;
2143                 }
2144
2145                 n = (n << 6) | (p[i] & 0x3f);
2146             }
2147
2148             /* advance pointer */
2149             p += 1 + cont;
2150             c = n;
2151             goto add_char_nonext;
2152
2153             /* error handling */
2154         invalid_utf8_sequence:
2155             tcc_warning("ill-formed UTF-8 subsequence starting with: \'\\x%x\'", c);
2156             c = 0xFFFD;
2157             p += skip;
2158             goto add_char_nonext;
2159
2160         }
2161         p++;
2162     add_char_nonext:
2163         if (!is_long)
2164             cstr_ccat(outstr, c);
2165         else {
2166 #ifdef TCC_TARGET_PE
2167             /* store as UTF-16 */
2168             if (c < 0x10000) {
2169                 cstr_wccat(outstr, c);
2170             } else {
2171                 c -= 0x10000;
2172                 cstr_wccat(outstr, (c >> 10) + 0xD800);
2173                 cstr_wccat(outstr, (c & 0x3FF) + 0xDC00);
2174             }
2175 #else
2176             cstr_wccat(outstr, c);
2177 #endif
2178         }
2179     }
2180     /* add a trailing '\0' */
2181     if (!is_long)
2182         cstr_ccat(outstr, '\0');
2183     else
2184         cstr_wccat(outstr, '\0');
2185 }
2186
2187 static void parse_string(const char *s, int len)
2188 {
2189     uint8_t buf[1000], *p = buf;
2190     int is_long, sep;
2191
2192     if ((is_long = *s == 'L'))
2193         ++s, --len;
2194     sep = *s++;
2195     len -= 2;
2196     if (len >= sizeof buf)
2197         p = tcc_malloc(len + 1);
2198     memcpy(p, s, len);
2199     p[len] = 0;
2200
2201     cstr_reset(&tokcstr);
2202     parse_escape_string(&tokcstr, p, is_long);
2203     if (p != buf)
2204         tcc_free(p);
2205
2206     if (sep == '\'') {
2207         int char_size, i, n, c;
2208         /* XXX: make it portable */
2209         if (!is_long)
2210             tok = TOK_CCHAR, char_size = 1;
2211         else
2212             tok = TOK_LCHAR, char_size = sizeof(nwchar_t);
2213         n = tokcstr.size / char_size - 1;
2214         if (n < 1)
2215             tcc_error("empty character constant");
2216         if (n > 1)
2217             tcc_warning_c(warn_all)("multi-character character constant");
2218         for (c = i = 0; i < n; ++i) {
2219             if (is_long)
2220                 c = ((nwchar_t *)tokcstr.data)[i];
2221             else
2222                 c = (c << 8) | ((char *)tokcstr.data)[i];
2223         }
2224         tokc.i = c;
2225     } else {
2226         tokc.str.size = tokcstr.size;
2227         tokc.str.data = tokcstr.data;
2228         if (!is_long)
2229             tok = TOK_STR;
2230         else
2231             tok = TOK_LSTR;
2232     }
2233 }
2234
2235 /* we use 64 bit numbers */
2236 #define BN_SIZE 2
2237
2238 /* bn = (bn << shift) | or_val */
2239 static void bn_lshift(unsigned int *bn, int shift, int or_val)
2240 {
2241     int i;
2242     unsigned int v;
2243     for(i=0;i<BN_SIZE;i++) {
2244         v = bn[i];
2245         bn[i] = (v << shift) | or_val;
2246         or_val = v >> (32 - shift);
2247     }
2248 }
2249
2250 static void bn_zero(unsigned int *bn)
2251 {
2252     int i;
2253     for(i=0;i<BN_SIZE;i++) {
2254         bn[i] = 0;
2255     }
2256 }
2257
2258 /* parse number in null terminated string 'p' and return it in the
2259    current token */
2260 static void parse_number(const char *p)
2261 {
2262     int b, t, shift, frac_bits, s, exp_val, ch;
2263     char *q;
2264     unsigned int bn[BN_SIZE];
2265     double d;
2266
2267     /* number */
2268     q = token_buf;
2269     ch = *p++;
2270     t = ch;
2271     ch = *p++;
2272     *q++ = t;
2273     b = 10;
2274     if (t == '.') {
2275         goto float_frac_parse;
2276     } else if (t == '0') {
2277         if (ch == 'x' || ch == 'X') {
2278             q--;
2279             ch = *p++;
2280             b = 16;
2281         } else if (tcc_state->tcc_ext && (ch == 'b' || ch == 'B')) {
2282             q--;
2283             ch = *p++;
2284             b = 2;
2285         }
2286     }
2287     /* parse all digits. cannot check octal numbers at this stage
2288        because of floating point constants */
2289     while (1) {
2290         if (ch >= 'a' && ch <= 'f')
2291             t = ch - 'a' + 10;
2292         else if (ch >= 'A' && ch <= 'F')
2293             t = ch - 'A' + 10;
2294         else if (isnum(ch))
2295             t = ch - '0';
2296         else
2297             break;
2298         if (t >= b)
2299             break;
2300         if (q >= token_buf + STRING_MAX_SIZE) {
2301         num_too_long:
2302             tcc_error("number too long");
2303         }
2304         *q++ = ch;
2305         ch = *p++;
2306     }
2307     if (ch == '.' ||
2308         ((ch == 'e' || ch == 'E') && b == 10) ||
2309         ((ch == 'p' || ch == 'P') && (b == 16 || b == 2))) {
2310         if (b != 10) {
2311             /* NOTE: strtox should support that for hexa numbers, but
2312                non ISOC99 libcs do not support it, so we prefer to do
2313                it by hand */
2314             /* hexadecimal or binary floats */
2315             /* XXX: handle overflows */
2316             *q = '\0';
2317             if (b == 16)
2318                 shift = 4;
2319             else
2320                 shift = 1;
2321             bn_zero(bn);
2322             q = token_buf;
2323             while (1) {
2324                 t = *q++;
2325                 if (t == '\0') {
2326                     break;
2327                 } else if (t >= 'a') {
2328                     t = t - 'a' + 10;
2329                 } else if (t >= 'A') {
2330                     t = t - 'A' + 10;
2331                 } else {
2332                     t = t - '0';
2333                 }
2334                 bn_lshift(bn, shift, t);
2335             }
2336             frac_bits = 0;
2337             if (ch == '.') {
2338                 ch = *p++;
2339                 while (1) {
2340                     t = ch;
2341                     if (t >= 'a' && t <= 'f') {
2342                         t = t - 'a' + 10;
2343                     } else if (t >= 'A' && t <= 'F') {
2344                         t = t - 'A' + 10;
2345                     } else if (t >= '0' && t <= '9') {
2346                         t = t - '0';
2347                     } else {
2348                         break;
2349                     }
2350                     if (t >= b)
2351                         tcc_error("invalid digit");
2352                     bn_lshift(bn, shift, t);
2353                     frac_bits += shift;
2354                     ch = *p++;
2355                 }
2356             }
2357             if (ch != 'p' && ch != 'P')
2358                 expect("exponent");
2359             ch = *p++;
2360             s = 1;
2361             exp_val = 0;
2362             if (ch == '+') {
2363                 ch = *p++;
2364             } else if (ch == '-') {
2365                 s = -1;
2366                 ch = *p++;
2367             }
2368             if (ch < '0' || ch > '9')
2369                 expect("exponent digits");
2370             while (ch >= '0' && ch <= '9') {
2371                 exp_val = exp_val * 10 + ch - '0';
2372                 ch = *p++;
2373             }
2374             exp_val = exp_val * s;
2375
2376             /* now we can generate the number */
2377             /* XXX: should patch directly float number */
2378             d = (double)bn[1] * 4294967296.0 + (double)bn[0];
2379             d = ldexp(d, exp_val - frac_bits);
2380             t = toup(ch);
2381             if (t == 'F') {
2382                 ch = *p++;
2383                 tok = TOK_CFLOAT;
2384                 /* float : should handle overflow */
2385                 tokc.f = (float)d;
2386             } else if (t == 'L') {
2387                 ch = *p++;
2388                 tok = TOK_CLDOUBLE;
2389 #ifdef TCC_USING_DOUBLE_FOR_LDOUBLE
2390                 tokc.d = d;
2391 #else
2392                 /* XXX: not large enough */
2393                 tokc.ld = (long double)d;
2394 #endif
2395             } else {
2396                 tok = TOK_CDOUBLE;
2397                 tokc.d = d;
2398             }
2399         } else {
2400             /* decimal floats */
2401             if (ch == '.') {
2402                 if (q >= token_buf + STRING_MAX_SIZE)
2403                     goto num_too_long;
2404                 *q++ = ch;
2405                 ch = *p++;
2406             float_frac_parse:
2407                 while (ch >= '0' && ch <= '9') {
2408                     if (q >= token_buf + STRING_MAX_SIZE)
2409                         goto num_too_long;
2410                     *q++ = ch;
2411                     ch = *p++;
2412                 }
2413             }
2414             if (ch == 'e' || ch == 'E') {
2415                 if (q >= token_buf + STRING_MAX_SIZE)
2416                     goto num_too_long;
2417                 *q++ = ch;
2418                 ch = *p++;
2419                 if (ch == '-' || ch == '+') {
2420                     if (q >= token_buf + STRING_MAX_SIZE)
2421                         goto num_too_long;
2422                     *q++ = ch;
2423                     ch = *p++;
2424                 }
2425                 if (ch < '0' || ch > '9')
2426                     expect("exponent digits");
2427                 while (ch >= '0' && ch <= '9') {
2428                     if (q >= token_buf + STRING_MAX_SIZE)
2429                         goto num_too_long;
2430                     *q++ = ch;
2431                     ch = *p++;
2432                 }
2433             }
2434             *q = '\0';
2435             t = toup(ch);
2436             errno = 0;
2437             if (t == 'F') {
2438                 ch = *p++;
2439                 tok = TOK_CFLOAT;
2440                 tokc.f = strtof(token_buf, NULL);
2441             } else if (t == 'L') {
2442                 ch = *p++;
2443                 tok = TOK_CLDOUBLE;
2444 #ifdef TCC_USING_DOUBLE_FOR_LDOUBLE
2445                 tokc.d = strtod(token_buf, NULL);
2446 #else
2447                 tokc.ld = strtold(token_buf, NULL);
2448 #endif
2449             } else {
2450                 tok = TOK_CDOUBLE;
2451                 tokc.d = strtod(token_buf, NULL);
2452             }
2453         }
2454     } else {
2455         unsigned long long n, n1;
2456         int lcount, ucount, ov = 0;
2457         const char *p1;
2458
2459         /* integer number */
2460         *q = '\0';
2461         q = token_buf;
2462         if (b == 10 && *q == '0') {
2463             b = 8;
2464             q++;
2465         }
2466         n = 0;
2467         while(1) {
2468             t = *q++;
2469             /* no need for checks except for base 10 / 8 errors */
2470             if (t == '\0')
2471                 break;
2472             else if (t >= 'a')
2473                 t = t - 'a' + 10;
2474             else if (t >= 'A')
2475                 t = t - 'A' + 10;
2476             else
2477                 t = t - '0';
2478             if (t >= b)
2479                 tcc_error("invalid digit");
2480             n1 = n;
2481             n = n * b + t;
2482             /* detect overflow */
2483             if (n1 >= 0x1000000000000000ULL && n / b != n1)
2484                 ov = 1;
2485         }
2486
2487         /* Determine the characteristics (unsigned and/or 64bit) the type of
2488            the constant must have according to the constant suffix(es) */
2489         lcount = ucount = 0;
2490         p1 = p;
2491         for(;;) {
2492             t = toup(ch);
2493             if (t == 'L') {
2494                 if (lcount >= 2)
2495                     tcc_error("three 'l's in integer constant");
2496                 if (lcount && *(p - 1) != ch)
2497                     tcc_error("incorrect integer suffix: %s", p1);
2498                 lcount++;
2499                 ch = *p++;
2500             } else if (t == 'U') {
2501                 if (ucount >= 1)
2502                     tcc_error("two 'u's in integer constant");
2503                 ucount++;
2504                 ch = *p++;
2505             } else {
2506                 break;
2507             }
2508         }
2509
2510         /* Determine if it needs 64 bits and/or unsigned in order to fit */
2511         if (ucount == 0 && b == 10) {
2512             if (lcount <= (LONG_SIZE == 4)) {
2513                 if (n >= 0x80000000U)
2514                     lcount = (LONG_SIZE == 4) + 1;
2515             }
2516             if (n >= 0x8000000000000000ULL)
2517                 ov = 1, ucount = 1;
2518         } else {
2519             if (lcount <= (LONG_SIZE == 4)) {
2520                 if (n >= 0x100000000ULL)
2521                     lcount = (LONG_SIZE == 4) + 1;
2522                 else if (n >= 0x80000000U)
2523                     ucount = 1;
2524             }
2525             if (n >= 0x8000000000000000ULL)
2526                 ucount = 1;
2527         }
2528
2529         if (ov)
2530             tcc_warning("integer constant overflow");
2531
2532         tok = TOK_CINT;
2533         if (lcount) {
2534             tok = TOK_CLONG;
2535             if (lcount == 2)
2536                 tok = TOK_CLLONG;
2537         }
2538         if (ucount)
2539             ++tok; /* TOK_CU... */
2540         tokc.i = n;
2541     }
2542     if (ch)
2543         tcc_error("invalid number");
2544 }
2545
2546
2547 #define PARSE2(c1, tok1, c2, tok2)              \
2548     case c1:                                    \
2549         PEEKC(c, p);                            \
2550         if (c == c2) {                          \
2551             p++;                                \
2552             tok = tok2;                         \
2553         } else {                                \
2554             tok = tok1;                         \
2555         }                                       \
2556         break;
2557
2558 /* return next token without macro substitution */
2559 static inline void next_nomacro1(void)
2560 {
2561     int t, c, is_long, len;
2562     TokenSym *ts;
2563     uint8_t *p, *p1;
2564     unsigned int h;
2565
2566     p = file->buf_ptr;
2567  redo_no_start:
2568     c = *p;
2569     switch(c) {
2570     case ' ':
2571     case '\t':
2572         tok = c;
2573         p++;
2574  maybe_space:
2575         if (parse_flags & PARSE_FLAG_SPACES)
2576             goto keep_tok_flags;
2577         while (isidnum_table[*p - CH_EOF] & IS_SPC)
2578             ++p;
2579         goto redo_no_start;
2580     case '\f':
2581     case '\v':
2582     case '\r':
2583         p++;
2584         goto redo_no_start;
2585     case '\\':
2586         /* first look if it is in fact an end of buffer */
2587         c = handle_stray(&p);
2588         if (c == '\\')
2589             goto parse_simple;
2590         if (c == CH_EOF) {
2591             TCCState *s1 = tcc_state;
2592             if ((parse_flags & PARSE_FLAG_LINEFEED)
2593                 && !(tok_flags & TOK_FLAG_EOF)) {
2594                 tok_flags |= TOK_FLAG_EOF;
2595                 tok = TOK_LINEFEED;
2596                 goto keep_tok_flags;
2597             } else if (!(parse_flags & PARSE_FLAG_PREPROCESS)) {
2598                 tok = TOK_EOF;
2599             } else if (s1->ifdef_stack_ptr != file->ifdef_stack_ptr) {
2600                 tcc_error("missing #endif");
2601             } else if (s1->include_stack_ptr == s1->include_stack) {
2602                 /* no include left : end of file. */
2603                 tok = TOK_EOF;
2604             } else {
2605                 tok_flags &= ~TOK_FLAG_EOF;
2606                 /* pop include file */
2607
2608                 /* test if previous '#endif' was after a #ifdef at
2609                    start of file */
2610                 if (tok_flags & TOK_FLAG_ENDIF) {
2611 #ifdef INC_DEBUG
2612                     printf("#endif %s\n", get_tok_str(file->ifndef_macro_saved, NULL));
2613 #endif
2614                     search_cached_include(s1, file->filename, 1)
2615                         ->ifndef_macro = file->ifndef_macro_saved;
2616                     tok_flags &= ~TOK_FLAG_ENDIF;
2617                 }
2618
2619                 /* add end of include file debug info */
2620                 tcc_debug_eincl(tcc_state);
2621                 /* pop include stack */
2622                 tcc_close();
2623                 s1->include_stack_ptr--;
2624                 p = file->buf_ptr;
2625                 if (p == file->buffer)
2626                     tok_flags = TOK_FLAG_BOF;
2627                 tok_flags |= TOK_FLAG_BOL;
2628                 goto redo_no_start;
2629             }
2630         } else {
2631             goto redo_no_start;
2632         }
2633         break;
2634
2635     case '\n':
2636         file->line_num++;
2637         tok_flags |= TOK_FLAG_BOL;
2638         p++;
2639 maybe_newline:
2640         if (0 == (parse_flags & PARSE_FLAG_LINEFEED))
2641             goto redo_no_start;
2642         tok = TOK_LINEFEED;
2643         goto keep_tok_flags;
2644
2645     case '#':
2646         /* XXX: simplify */
2647         PEEKC(c, p);
2648         if ((tok_flags & TOK_FLAG_BOL) &&
2649             (parse_flags & PARSE_FLAG_PREPROCESS)) {
2650             file->buf_ptr = p;
2651             preprocess(tok_flags & TOK_FLAG_BOF);
2652             p = file->buf_ptr;
2653             goto maybe_newline;
2654         } else {
2655             if (c == '#') {
2656                 p++;
2657                 tok = TOK_TWOSHARPS;
2658             } else {
2659 #if !defined(TCC_TARGET_ARM)
2660                 if (parse_flags & PARSE_FLAG_ASM_FILE) {
2661                     p = parse_line_comment(p - 1);
2662                     goto redo_no_start;
2663                 } else
2664 #endif
2665                 {
2666                     tok = '#';
2667                 }
2668             }
2669         }
2670         break;
2671
2672     /* dollar is allowed to start identifiers when not parsing asm */
2673     case '$':
2674         if (!(isidnum_table[c - CH_EOF] & IS_ID)
2675          || (parse_flags & PARSE_FLAG_ASM_FILE))
2676             goto parse_simple;
2677
2678     case 'a': case 'b': case 'c': case 'd':
2679     case 'e': case 'f': case 'g': case 'h':
2680     case 'i': case 'j': case 'k': case 'l':
2681     case 'm': case 'n': case 'o': case 'p':
2682     case 'q': case 'r': case 's': case 't':
2683     case 'u': case 'v': case 'w': case 'x':
2684     case 'y': case 'z':
2685     case 'A': case 'B': case 'C': case 'D':
2686     case 'E': case 'F': case 'G': case 'H':
2687     case 'I': case 'J': case 'K':
2688     case 'M': case 'N': case 'O': case 'P':
2689     case 'Q': case 'R': case 'S': case 'T':
2690     case 'U': case 'V': case 'W': case 'X':
2691     case 'Y': case 'Z':
2692     case '_':
2693     parse_ident_fast:
2694         p1 = p;
2695         h = TOK_HASH_INIT;
2696         h = TOK_HASH_FUNC(h, c);
2697         while (c = *++p, isidnum_table[c - CH_EOF] & (IS_ID|IS_NUM))
2698             h = TOK_HASH_FUNC(h, c);
2699         len = p - p1;
2700         if (c != '\\') {
2701             TokenSym **pts;
2702
2703             /* fast case : no stray found, so we have the full token
2704                and we have already hashed it */
2705             h &= (TOK_HASH_SIZE - 1);
2706             pts = &hash_ident[h];
2707             for(;;) {
2708                 ts = *pts;
2709                 if (!ts)
2710                     break;
2711                 if (ts->len == len && !memcmp(ts->str, p1, len))
2712                     goto token_found;
2713                 pts = &(ts->hash_next);
2714             }
2715             ts = tok_alloc_new(pts, (char *) p1, len);
2716         token_found: ;
2717         } else {
2718             /* slower case */
2719             cstr_reset(&tokcstr);
2720             cstr_cat(&tokcstr, (char *) p1, len);
2721             p--;
2722             PEEKC(c, p);
2723         parse_ident_slow:
2724             while (isidnum_table[c - CH_EOF] & (IS_ID|IS_NUM))
2725             {
2726                 cstr_ccat(&tokcstr, c);
2727                 PEEKC(c, p);
2728             }
2729             ts = tok_alloc(tokcstr.data, tokcstr.size);
2730         }
2731         tok = ts->tok;
2732         break;
2733     case 'L':
2734         t = p[1];
2735         if (t != '\\' && t != '\'' && t != '\"') {
2736             /* fast case */
2737             goto parse_ident_fast;
2738         } else {
2739             PEEKC(c, p);
2740             if (c == '\'' || c == '\"') {
2741                 is_long = 1;
2742                 goto str_const;
2743             } else {
2744                 cstr_reset(&tokcstr);
2745                 cstr_ccat(&tokcstr, 'L');
2746                 goto parse_ident_slow;
2747             }
2748         }
2749         break;
2750
2751     case '0': case '1': case '2': case '3':
2752     case '4': case '5': case '6': case '7':
2753     case '8': case '9':
2754         t = c;
2755         PEEKC(c, p);
2756         /* after the first digit, accept digits, alpha, '.' or sign if
2757            prefixed by 'eEpP' */
2758     parse_num:
2759         cstr_reset(&tokcstr);
2760         for(;;) {
2761             cstr_ccat(&tokcstr, t);
2762             if (!((isidnum_table[c - CH_EOF] & (IS_ID|IS_NUM))
2763                   || c == '.'
2764                   || ((c == '+' || c == '-')
2765                       && (((t == 'e' || t == 'E')
2766                             && !(parse_flags & PARSE_FLAG_ASM_FILE
2767                                 /* 0xe+1 is 3 tokens in asm */
2768                                 && ((char*)tokcstr.data)[0] == '0'
2769                                 && toup(((char*)tokcstr.data)[1]) == 'X'))
2770                           || t == 'p' || t == 'P'))))
2771                 break;
2772             t = c;
2773             PEEKC(c, p);
2774         }
2775         /* We add a trailing '\0' to ease parsing */
2776         cstr_ccat(&tokcstr, '\0');
2777         tokc.str.size = tokcstr.size;
2778         tokc.str.data = tokcstr.data;
2779         tok = TOK_PPNUM;
2780         break;
2781
2782     case '.':
2783         /* special dot handling because it can also start a number */
2784         PEEKC(c, p);
2785         if (isnum(c)) {
2786             t = '.';
2787             goto parse_num;
2788         } else if ((isidnum_table['.' - CH_EOF] & IS_ID)
2789                    && (isidnum_table[c - CH_EOF] & (IS_ID|IS_NUM))) {
2790             *--p = c = '.';
2791             goto parse_ident_fast;
2792         } else if (c == '.') {
2793             PEEKC(c, p);
2794             if (c == '.') {
2795                 p++;
2796                 tok = TOK_DOTS;
2797             } else {
2798                 *--p = '.'; /* may underflow into file->unget[] */
2799                 tok = '.';
2800             }
2801         } else {
2802             tok = '.';
2803         }
2804         break;
2805     case '\'':
2806     case '\"':
2807         is_long = 0;
2808     str_const:
2809         cstr_reset(&tokcstr);
2810         if (is_long)
2811             cstr_ccat(&tokcstr, 'L');
2812         cstr_ccat(&tokcstr, c);
2813         p = parse_pp_string(p, c, &tokcstr);
2814         cstr_ccat(&tokcstr, c);
2815         cstr_ccat(&tokcstr, '\0');
2816         tokc.str.size = tokcstr.size;
2817         tokc.str.data = tokcstr.data;
2818         tok = TOK_PPSTR;
2819         break;
2820
2821     case '<':
2822         PEEKC(c, p);
2823         if (c == '=') {
2824             p++;
2825             tok = TOK_LE;
2826         } else if (c == '<') {
2827             PEEKC(c, p);
2828             if (c == '=') {
2829                 p++;
2830                 tok = TOK_A_SHL;
2831             } else {
2832                 tok = TOK_SHL;
2833             }
2834         } else {
2835             tok = TOK_LT;
2836         }
2837         break;
2838     case '>':
2839         PEEKC(c, p);
2840         if (c == '=') {
2841             p++;
2842             tok = TOK_GE;
2843         } else if (c == '>') {
2844             PEEKC(c, p);
2845             if (c == '=') {
2846                 p++;
2847                 tok = TOK_A_SAR;
2848             } else {
2849                 tok = TOK_SAR;
2850             }
2851         } else {
2852             tok = TOK_GT;
2853         }
2854         break;
2855
2856     case '&':
2857         PEEKC(c, p);
2858         if (c == '&') {
2859             p++;
2860             tok = TOK_LAND;
2861         } else if (c == '=') {
2862             p++;
2863             tok = TOK_A_AND;
2864         } else {
2865             tok = '&';
2866         }
2867         break;
2868
2869     case '|':
2870         PEEKC(c, p);
2871         if (c == '|') {
2872             p++;
2873             tok = TOK_LOR;
2874         } else if (c == '=') {
2875             p++;
2876             tok = TOK_A_OR;
2877         } else {
2878             tok = '|';
2879         }
2880         break;
2881
2882     case '+':
2883         PEEKC(c, p);
2884         if (c == '+') {
2885             p++;
2886             tok = TOK_INC;
2887         } else if (c == '=') {
2888             p++;
2889             tok = TOK_A_ADD;
2890         } else {
2891             tok = '+';
2892         }
2893         break;
2894
2895     case '-':
2896         PEEKC(c, p);
2897         if (c == '-') {
2898             p++;
2899             tok = TOK_DEC;
2900         } else if (c == '=') {
2901             p++;
2902             tok = TOK_A_SUB;
2903         } else if (c == '>') {
2904             p++;
2905             tok = TOK_ARROW;
2906         } else {
2907             tok = '-';
2908         }
2909         break;
2910
2911     PARSE2('!', '!', '=', TOK_NE)
2912     PARSE2('=', '=', '=', TOK_EQ)
2913     PARSE2('*', '*', '=', TOK_A_MUL)
2914     PARSE2('%', '%', '=', TOK_A_MOD)
2915     PARSE2('^', '^', '=', TOK_A_XOR)
2916
2917         /* comments or operator */
2918     case '/':
2919         PEEKC(c, p);
2920         if (c == '*') {
2921             p = parse_comment(p);
2922             /* comments replaced by a blank */
2923             tok = ' ';
2924             goto maybe_space;
2925         } else if (c == '/') {
2926             p = parse_line_comment(p);
2927             tok = ' ';
2928             goto maybe_space;
2929         } else if (c == '=') {
2930             p++;
2931             tok = TOK_A_DIV;
2932         } else {
2933             tok = '/';
2934         }
2935         break;
2936
2937         /* simple tokens */
2938     case '(':
2939     case ')':
2940     case '[':
2941     case ']':
2942     case '{':
2943     case '}':
2944     case ',':
2945     case ';':
2946     case ':':
2947     case '?':
2948     case '~':
2949     case '@': /* only used in assembler */
2950     parse_simple:
2951         tok = c;
2952         p++;
2953         break;
2954     default:
2955         if (c >= 0x80 && c <= 0xFF) /* utf8 identifiers */
2956             goto parse_ident_fast;
2957         if (parse_flags & PARSE_FLAG_ASM_FILE)
2958             goto parse_simple;
2959         tcc_error("unrecognized character \\x%02x", c);
2960         break;
2961     }
2962     tok_flags = 0;
2963 keep_tok_flags:
2964     file->buf_ptr = p;
2965 #if defined(PARSE_DEBUG)
2966     printf("token = %d %s\n", tok, get_tok_str(tok, &tokc));
2967 #endif
2968 }
2969
2970 static void macro_subst(
2971     TokenString *tok_str,
2972     Sym **nested_list,
2973     const int *macro_str
2974     );
2975
2976 /* substitute arguments in replacement lists in macro_str by the values in
2977    args (field d) and return allocated string */
2978 static int *macro_arg_subst(Sym **nested_list, const int *macro_str, Sym *args)
2979 {
2980     int t, t0, t1, spc;
2981     const int *st;
2982     Sym *s;
2983     CValue cval;
2984     TokenString str;
2985
2986     tok_str_new(&str);
2987     t0 = t1 = 0;
2988     while(1) {
2989         TOK_GET(&t, &macro_str, &cval);
2990         if (!t)
2991             break;
2992         if (t == '#') {
2993             /* stringize */
2994             TOK_GET(&t, &macro_str, &cval);
2995             if (!t)
2996                 goto bad_stringy;
2997             s = sym_find2(args, t);
2998             if (s) {
2999                 cstr_reset(&tokcstr);
3000                 cstr_ccat(&tokcstr, '\"');
3001                 st = s->d;
3002                 spc = 0;
3003                 while (*st >= 0) {
3004                     TOK_GET(&t, &st, &cval);
3005                     if (t != TOK_PLCHLDR
3006                      && t != TOK_NOSUBST
3007                      && 0 == check_space(t, &spc)) {
3008                         const char *s = get_tok_str(t, &cval);
3009                         while (*s) {
3010                             if (t == TOK_PPSTR && *s != '\'')
3011                                 add_char(&tokcstr, *s);
3012                             else
3013                                 cstr_ccat(&tokcstr, *s);
3014                             ++s;
3015                         }
3016                     }
3017                 }
3018                 tokcstr.size -= spc;
3019                 cstr_ccat(&tokcstr, '\"');
3020                 cstr_ccat(&tokcstr, '\0');
3021 #ifdef PP_DEBUG
3022                 printf("\nstringize: <%s>\n", (char *)tokcstr.data);
3023 #endif
3024                 /* add string */
3025                 cval.str.size = tokcstr.size;
3026                 cval.str.data = tokcstr.data;
3027                 tok_str_add2(&str, TOK_PPSTR, &cval);
3028             } else {
3029         bad_stringy:
3030                 expect("macro parameter after '#'");
3031             }
3032         } else if (t >= TOK_IDENT) {
3033             s = sym_find2(args, t);
3034             if (s) {
3035                 st = s->d;
3036                 /* if '##' is present before or after, no arg substitution */
3037                 if (*macro_str == TOK_PPJOIN || t1 == TOK_PPJOIN) {
3038                     /* special case for var arg macros : ## eats the ','
3039                        if empty VA_ARGS variable. */
3040                     if (t1 == TOK_PPJOIN && t0 == ',' && gnu_ext && s->type.t) {
3041                         if (*st <= 0) {
3042                             /* suppress ',' '##' */
3043                             str.len -= 2;
3044                         } else {
3045                             /* suppress '##' and add variable */
3046                             str.len--;
3047                             goto add_var;
3048                         }
3049                     }
3050                 } else {
3051             add_var:
3052                     if (!s->next) {
3053                         /* Expand arguments tokens and store them.  In most
3054                            cases we could also re-expand each argument if
3055                            used multiple times, but not if the argument
3056                            contains the __COUNTER__ macro.  */
3057                         TokenString str2;
3058                         sym_push2(&s->next, s->v, s->type.t, 0);
3059                         tok_str_new(&str2);
3060                         macro_subst(&str2, nested_list, st);
3061                         tok_str_add(&str2, 0);
3062                         s->next->d = str2.str;
3063                     }
3064                     st = s->next->d;
3065                 }
3066                 if (*st <= 0) {
3067                     /* expanded to empty string */
3068                     tok_str_add(&str, TOK_PLCHLDR);
3069                 } else for (;;) {
3070                     int t2;
3071                     TOK_GET(&t2, &st, &cval);
3072                     if (t2 <= 0)
3073                         break;
3074                     tok_str_add2(&str, t2, &cval);
3075                 }
3076             } else {
3077                 tok_str_add(&str, t);
3078             }
3079         } else {
3080             tok_str_add2(&str, t, &cval);
3081         }
3082         t0 = t1, t1 = t;
3083     }
3084     tok_str_add(&str, 0);
3085     return str.str;
3086 }
3087
3088 static char const ab_month_name[12][4] =
3089 {
3090     "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3091     "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3092 };
3093
3094 static int paste_tokens(int t1, CValue *v1, int t2, CValue *v2)
3095 {
3096     int n, ret = 1;
3097
3098     cstr_reset(&tokcstr);
3099     if (t1 != TOK_PLCHLDR)
3100         cstr_cat(&tokcstr, get_tok_str(t1, v1), -1);
3101     n = tokcstr.size;
3102     if (t2 != TOK_PLCHLDR)
3103         cstr_cat(&tokcstr, get_tok_str(t2, v2), -1);
3104     cstr_ccat(&tokcstr, '\0');
3105     //printf("paste <%s>\n", (char*)tokcstr.data);
3106
3107     tcc_open_bf(tcc_state, ":paste:", tokcstr.size);
3108     memcpy(file->buffer, tokcstr.data, tokcstr.size);
3109     tok_flags = 0;
3110     for (;;) {
3111         next_nomacro1();
3112         if (0 == *file->buf_ptr)
3113             break;
3114         if (is_space(tok))
3115             continue;
3116         tcc_warning("pasting \"%.*s\" and \"%s\" does not give a valid"
3117             " preprocessing token", n, file->buffer, file->buffer + n);
3118         ret = 0;
3119         break;
3120     }
3121     tcc_close();
3122     return ret;
3123 }
3124
3125 /* handle the '##' operator. Return NULL if no '##' seen. Otherwise
3126    return the resulting string (which must be freed). */
3127 static inline int *macro_twosharps(const int *ptr0)
3128 {
3129     int t;
3130     CValue cval;
3131     TokenString macro_str1;
3132     int start_of_nosubsts = -1;
3133     const int *ptr;
3134
3135     /* we search the first '##' */
3136     for (ptr = ptr0;;) {
3137         TOK_GET(&t, &ptr, &cval);
3138         if (t == TOK_PPJOIN)
3139             break;
3140         if (t == 0)
3141             return NULL;
3142     }
3143
3144     tok_str_new(&macro_str1);
3145
3146     //tok_print(" $$$", ptr0);
3147     for (ptr = ptr0;;) {
3148         TOK_GET(&t, &ptr, &cval);
3149         if (t == 0)
3150             break;
3151         if (t == TOK_PPJOIN)
3152             continue;
3153         while (*ptr == TOK_PPJOIN) {
3154             int t1; CValue cv1;
3155             /* given 'a##b', remove nosubsts preceding 'a' */
3156             if (start_of_nosubsts >= 0)
3157                 macro_str1.len = start_of_nosubsts;
3158             /* given 'a##b', remove nosubsts preceding 'b' */
3159             while ((t1 = *++ptr) == TOK_NOSUBST)
3160                 ;
3161             if (t1 && t1 != TOK_PPJOIN) {
3162                 TOK_GET(&t1, &ptr, &cv1);
3163                 if (t != TOK_PLCHLDR || t1 != TOK_PLCHLDR) {
3164                     if (paste_tokens(t, &cval, t1, &cv1)) {
3165                         t = tok, cval = tokc;
3166                     } else {
3167                         tok_str_add2(&macro_str1, t, &cval);
3168                         t = t1, cval = cv1;
3169                     }
3170                 }
3171             }
3172         }
3173         if (t == TOK_NOSUBST) {
3174             if (start_of_nosubsts < 0)
3175                 start_of_nosubsts = macro_str1.len;
3176         } else {
3177             start_of_nosubsts = -1;
3178         }
3179         tok_str_add2(&macro_str1, t, &cval);
3180     }
3181     tok_str_add(&macro_str1, 0);
3182     //tok_print(" ###", macro_str1.str);
3183     return macro_str1.str;
3184 }
3185
3186 /* peek or read [ws_str == NULL] next token from function macro call,
3187    walking up macro levels up to the file if necessary */
3188 static int next_argstream(Sym **nested_list, TokenString *ws_str)
3189 {
3190     int t;
3191     const int *p;
3192     Sym *sa;
3193
3194     for (;;) {
3195         if (macro_ptr) {
3196             p = macro_ptr, t = *p;
3197             if (ws_str) {
3198                 while (is_space(t) || TOK_LINEFEED == t || TOK_PLCHLDR == t)
3199                     tok_str_add(ws_str, t), t = *++p;
3200             }
3201             if (t == 0) {
3202                 end_macro();
3203                 /* also, end of scope for nested defined symbol */
3204                 sa = *nested_list;
3205                 while (sa && sa->v == 0)
3206                     sa = sa->prev;
3207                 if (sa)
3208                     sa->v = 0;
3209                 continue;
3210             }
3211         } else {
3212             uint8_t *p = file->buf_ptr;
3213             int ch = handle_bs(&p);
3214             if (ws_str) {
3215                 while (is_space(ch) || ch == '\n' || ch == '/') {
3216                     if (ch == '/') {
3217                         int c;
3218                         PEEKC(c, p);
3219                         if (c == '*') {
3220                             p = parse_comment(p) - 1;
3221                         } else if (c == '/') {
3222                             p = parse_line_comment(p) - 1;
3223                         } else {
3224                             *--p = ch;
3225                             break;
3226                         }
3227                         ch = ' ';
3228                     }
3229                     if (ch == '\n')
3230                         file->line_num++;
3231                     if (!(ch == '\f' || ch == '\v' || ch == '\r'))
3232                         tok_str_add(ws_str, ch);
3233                     PEEKC(ch, p);
3234                 }
3235             }
3236             file->buf_ptr = p;
3237             t = ch;
3238         }
3239
3240         if (ws_str)
3241             return t;
3242         next_nomacro();
3243         return tok;
3244     }
3245 }
3246
3247 /* do macro substitution of current token with macro 's' and add
3248    result to (tok_str,tok_len). 'nested_list' is the list of all
3249    macros we got inside to avoid recursing. Return non zero if no
3250    substitution needs to be done */
3251 static int macro_subst_tok(
3252     TokenString *tok_str,
3253     Sym **nested_list,
3254     Sym *s)
3255 {
3256     Sym *args, *sa, *sa1;
3257     int parlevel, t, t1, spc;
3258     TokenString str;
3259     char *cstrval;
3260     CValue cval;
3261     char buf[32];
3262
3263     /* if symbol is a macro, prepare substitution */
3264     /* special macros */
3265     if (tok == TOK___LINE__ || tok == TOK___COUNTER__) {
3266         t = tok == TOK___LINE__ ? file->line_num : pp_counter++;
3267         snprintf(buf, sizeof(buf), "%d", t);
3268         cstrval = buf;
3269         t1 = TOK_PPNUM;
3270         goto add_cstr1;
3271     } else if (tok == TOK___FILE__) {
3272         cstrval = file->filename;
3273         goto add_cstr;
3274     } else if (tok == TOK___DATE__ || tok == TOK___TIME__) {
3275         time_t ti;
3276         struct tm *tm;
3277
3278         time(&ti);
3279         tm = localtime(&ti);
3280         if (tok == TOK___DATE__) {
3281             snprintf(buf, sizeof(buf), "%s %2d %d",
3282                      ab_month_name[tm->tm_mon], tm->tm_mday, tm->tm_year + 1900);
3283         } else {
3284             snprintf(buf, sizeof(buf), "%02d:%02d:%02d",
3285                      tm->tm_hour, tm->tm_min, tm->tm_sec);
3286         }
3287         cstrval = buf;
3288     add_cstr:
3289         t1 = TOK_STR;
3290     add_cstr1:
3291         cstr_reset(&tokcstr);
3292         cstr_cat(&tokcstr, cstrval, 0);
3293         cval.str.size = tokcstr.size;
3294         cval.str.data = tokcstr.data;
3295         tok_str_add2(tok_str, t1, &cval);
3296     } else if (s->d) {
3297         int saved_parse_flags = parse_flags;
3298         int *joined_str = NULL;
3299         int *mstr = s->d;
3300
3301         if (s->type.t == MACRO_FUNC) {
3302             /* whitespace between macro name and argument list */
3303             TokenString ws_str;
3304             tok_str_new(&ws_str);
3305
3306             spc = 0;
3307             parse_flags |= PARSE_FLAG_SPACES | PARSE_FLAG_LINEFEED
3308                 | PARSE_FLAG_ACCEPT_STRAYS;
3309
3310             /* get next token from argument stream */
3311             t = next_argstream(nested_list, &ws_str);
3312             if (t != '(') {
3313                 /* not a macro substitution after all, restore the
3314                  * macro token plus all whitespace we've read.
3315                  * whitespace is intentionally not merged to preserve
3316                  * newlines. */
3317                 parse_flags = saved_parse_flags;
3318                 tok_str_add(tok_str, tok);
3319                 if (parse_flags & PARSE_FLAG_SPACES) {
3320                     int i;
3321                     for (i = 0; i < ws_str.len; i++)
3322                         tok_str_add(tok_str, ws_str.str[i]);
3323                 }
3324                 if (ws_str.len && ws_str.str[ws_str.len - 1] == '\n')
3325                     tok_flags |= TOK_FLAG_BOL;
3326                 tok_str_free_str(ws_str.str);
3327                 return 0;
3328             } else {
3329                 tok_str_free_str(ws_str.str);
3330             }
3331             do {
3332                 next_nomacro(); /* eat '(' */
3333             } while (tok == TOK_PLCHLDR || is_space(tok));
3334
3335             /* argument macro */
3336             args = NULL;
3337             sa = s->next;
3338             /* NOTE: empty args are allowed, except if no args */
3339             for(;;) {
3340                 do {
3341                     next_argstream(nested_list, NULL);
3342                 } while (tok == TOK_PLCHLDR || is_space(tok) ||
3343                          TOK_LINEFEED == tok);
3344     empty_arg:
3345                 /* handle '()' case */
3346                 if (!args && !sa && tok == ')')
3347                     break;
3348                 if (!sa)
3349                     tcc_error("macro '%s' used with too many args",
3350                           get_tok_str(s->v, 0));
3351                 tok_str_new(&str);
3352                 parlevel = spc = 0;
3353                 /* NOTE: non zero sa->t indicates VA_ARGS */
3354                 while ((parlevel > 0 ||
3355                         (tok != ')' &&
3356                          (tok != ',' || sa->type.t)))) {
3357                     if (tok == TOK_EOF || tok == 0)
3358                         break;
3359                     if (tok == '(')
3360                         parlevel++;
3361                     else if (tok == ')')
3362                         parlevel--;
3363                     if (tok == TOK_LINEFEED)
3364                         tok = ' ';
3365                     if (!check_space(tok, &spc))
3366                         tok_str_add2(&str, tok, &tokc);
3367                     next_argstream(nested_list, NULL);
3368                 }
3369                 if (parlevel)
3370                     expect(")");
3371                 str.len -= spc;
3372                 tok_str_add(&str, -1);
3373                 tok_str_add(&str, 0);
3374                 sa1 = sym_push2(&args, sa->v & ~SYM_FIELD, sa->type.t, 0);
3375                 sa1->d = str.str;
3376                 sa = sa->next;
3377                 if (tok == ')') {
3378                     /* special case for gcc var args: add an empty
3379                        var arg argument if it is omitted */
3380                     if (sa && sa->type.t && gnu_ext)
3381                         goto empty_arg;
3382                     break;
3383                 }
3384                 if (tok != ',')
3385                     expect(",");
3386             }
3387             if (sa) {
3388                 tcc_error("macro '%s' used with too few args",
3389                       get_tok_str(s->v, 0));
3390             }
3391
3392             /* now subst each arg */
3393             mstr = macro_arg_subst(nested_list, mstr, args);
3394             /* free memory */
3395             sa = args;
3396             while (sa) {
3397                 sa1 = sa->prev;
3398                 tok_str_free_str(sa->d);
3399                 if (sa->next) {
3400                     tok_str_free_str(sa->next->d);
3401                     sym_free(sa->next);
3402                 }
3403                 sym_free(sa);
3404                 sa = sa1;
3405             }
3406             parse_flags = saved_parse_flags;
3407         }
3408
3409         sym_push2(nested_list, s->v, 0, 0);
3410         parse_flags = saved_parse_flags;
3411         joined_str = macro_twosharps(mstr);
3412         macro_subst(tok_str, nested_list, joined_str ? joined_str : mstr);
3413
3414         /* pop nested defined symbol */
3415         sa1 = *nested_list;
3416         *nested_list = sa1->prev;
3417         sym_free(sa1);
3418         if (joined_str)
3419             tok_str_free_str(joined_str);
3420         if (mstr != s->d)
3421             tok_str_free_str(mstr);
3422     }
3423     return 0;
3424 }
3425
3426 /* do macro substitution of macro_str and add result to
3427    (tok_str,tok_len). 'nested_list' is the list of all macros we got
3428    inside to avoid recursing. */
3429 static void macro_subst(
3430     TokenString *tok_str,
3431     Sym **nested_list,
3432     const int *macro_str
3433     )
3434 {
3435     Sym *s;
3436     int t, spc, nosubst;
3437     CValue cval;
3438
3439     spc = nosubst = 0;
3440
3441     while (1) {
3442         TOK_GET(&t, &macro_str, &cval);
3443         if (t <= 0)
3444             break;
3445
3446         if (t >= TOK_IDENT && 0 == nosubst) {
3447             s = define_find(t);
3448             if (s == NULL)
3449                 goto no_subst;
3450
3451             /* if nested substitution, do nothing */
3452             if (sym_find2(*nested_list, t)) {
3453                 /* and mark it as TOK_NOSUBST, so it doesn't get subst'd again */
3454                 tok_str_add2(tok_str, TOK_NOSUBST, NULL);
3455                 goto no_subst;
3456             }
3457
3458             {
3459                 TokenString *str = tok_str_alloc();
3460                 str->str = (int*)macro_str;
3461                 begin_macro(str, 2);
3462
3463                 tok = t;
3464                 macro_subst_tok(tok_str, nested_list, s);
3465
3466                 if (macro_stack != str) {
3467                     /* already finished by reading function macro arguments */
3468                     break;
3469                 }
3470
3471                 macro_str = macro_ptr;
3472                 end_macro ();
3473             }
3474             if (tok_str->len)
3475                 spc = is_space(t = tok_str->str[tok_str->lastlen]);
3476         } else {
3477 no_subst:
3478             if (!check_space(t, &spc))
3479                 tok_str_add2(tok_str, t, &cval);
3480
3481             if (nosubst) {
3482                 if (nosubst > 1 && (spc || (++nosubst == 3 && t == '(')))
3483                     continue;
3484                 nosubst = 0;
3485             }
3486             if (t == TOK_NOSUBST)
3487                 nosubst = 1;
3488         }
3489         /* GCC supports 'defined' as result of a macro substitution */
3490         if (t == TOK_DEFINED && pp_expr)
3491             nosubst = 2;
3492     }
3493 }
3494
3495 /* return next token without macro substitution. Can read input from
3496    macro_ptr buffer */
3497 static void next_nomacro(void)
3498 {
3499     int t;
3500     if (macro_ptr) {
3501  redo:
3502         t = *macro_ptr;
3503         if (TOK_HAS_VALUE(t)) {
3504             tok_get(&tok, &macro_ptr, &tokc);
3505             if (t == TOK_LINENUM) {
3506                 file->line_num = tokc.i;
3507                 goto redo;
3508             }
3509         } else {
3510             macro_ptr++;
3511             if (t < TOK_IDENT) {
3512                 if (!(parse_flags & PARSE_FLAG_SPACES)
3513                     && (isidnum_table[t - CH_EOF] & IS_SPC))
3514                     goto redo;
3515             }
3516             tok = t;
3517         }
3518     } else {
3519         next_nomacro1();
3520     }
3521 }
3522
3523 /* return next token with macro substitution */
3524 ST_FUNC void next(void)
3525 {
3526     int t;
3527  redo:
3528     next_nomacro();
3529     t = tok;
3530     if (macro_ptr) {
3531         if (!TOK_HAS_VALUE(t)) {
3532             if (t == TOK_NOSUBST || t == TOK_PLCHLDR) {
3533                 /* discard preprocessor markers */
3534                 goto redo;
3535             } else if (t == 0) {
3536                 /* end of macro or unget token string */
3537                 end_macro();
3538                 goto redo;
3539             } else if (t == '\\') {
3540                 if (!(parse_flags & PARSE_FLAG_ACCEPT_STRAYS))
3541                     tcc_error("stray '\\' in program");
3542             }
3543             return;
3544         }
3545     } else if (t >= TOK_IDENT && (parse_flags & PARSE_FLAG_PREPROCESS)) {
3546         /* if reading from file, try to substitute macros */
3547         Sym *s = define_find(t);
3548         if (s) {
3549             Sym *nested_list = NULL;
3550             tokstr_buf.len = 0;
3551             macro_subst_tok(&tokstr_buf, &nested_list, s);
3552             tok_str_add(&tokstr_buf, 0);
3553             begin_macro(&tokstr_buf, 0);
3554             goto redo;
3555         }
3556         return;
3557     }
3558     /* convert preprocessor tokens into C tokens */
3559     if (t == TOK_PPNUM) {
3560         if  (parse_flags & PARSE_FLAG_TOK_NUM)
3561             parse_number((char *)tokc.str.data);
3562     } else if (t == TOK_PPSTR) {
3563         if (parse_flags & PARSE_FLAG_TOK_STR)
3564             parse_string((char *)tokc.str.data, tokc.str.size - 1);
3565     }
3566 }
3567
3568 /* push back current token and set current token to 'last_tok'. Only
3569    identifier case handled for labels. */
3570 ST_INLN void unget_tok(int last_tok)
3571 {
3572
3573     TokenString *str = tok_str_alloc();
3574     tok_str_add2(str, tok, &tokc);
3575     tok_str_add(str, 0);
3576     begin_macro(str, 1);
3577     tok = last_tok;
3578 }
3579
3580 /* ------------------------------------------------------------------------- */
3581 /* init preprocessor */
3582
3583 static const char * const target_os_defs =
3584 #ifdef TCC_TARGET_PE
3585     "_WIN32\0"
3586 # if PTR_SIZE == 8
3587     "_WIN64\0"
3588 # endif
3589 #else
3590 # if defined TCC_TARGET_MACHO
3591     "__APPLE__\0"
3592 # elif TARGETOS_FreeBSD
3593     "__FreeBSD__ 12\0"
3594 # elif TARGETOS_FreeBSD_kernel
3595     "__FreeBSD_kernel__\0"
3596 # elif TARGETOS_NetBSD
3597     "__NetBSD__\0"
3598 # elif TARGETOS_OpenBSD
3599     "__OpenBSD__\0"
3600 # else
3601     "__linux__\0"
3602     "__linux\0"
3603 #  if TARGETOS_ANDROID
3604     "__ANDROID__\0"
3605 #  endif
3606 # endif
3607     "__unix__\0"
3608     "__unix\0"
3609 #endif
3610     ;
3611
3612 static void putdef(CString *cs, const char *p)
3613 {
3614     cstr_printf(cs, "#define %s%s\n", p, &" 1"[!!strchr(p, ' ')*2]);
3615 }
3616
3617 static void putdefs(CString *cs, const char *p)
3618 {
3619     while (*p)
3620         putdef(cs, p), p = strchr(p, 0) + 1;
3621 }
3622
3623 static void tcc_predefs(TCCState *s1, CString *cs, int is_asm)
3624 {
3625     int a, b, c;
3626
3627     sscanf(TCC_VERSION, "%d.%d.%d", &a, &b, &c);
3628     cstr_printf(cs, "#define __TINYC__ %d\n", a*10000 + b*100 + c);
3629
3630     putdefs(cs, target_machine_defs);
3631     putdefs(cs, target_os_defs);
3632
3633 #ifdef TCC_TARGET_ARM
3634     if (s1->float_abi == ARM_HARD_FLOAT)
3635       putdef(cs, "__ARM_PCS_VFP");
3636 #endif
3637     if (is_asm)
3638       putdef(cs, "__ASSEMBLER__");
3639     if (s1->output_type == TCC_OUTPUT_PREPROCESS)
3640       putdef(cs, "__TCC_PP__");
3641     if (s1->output_type == TCC_OUTPUT_MEMORY)
3642       putdef(cs, "__TCC_RUN__");
3643 #ifdef CONFIG_TCC_BACKTRACE
3644     if (s1->do_backtrace)
3645       putdef(cs, "__TCC_BACKTRACE__");
3646 #endif
3647 #ifdef CONFIG_TCC_BCHECK
3648     if (s1->do_bounds_check)
3649       putdef(cs, "__TCC_BCHECK__");
3650 #endif
3651     if (s1->char_is_unsigned)
3652       putdef(cs, "__CHAR_UNSIGNED__");
3653     if (s1->optimize > 0)
3654       putdef(cs, "__OPTIMIZE__");
3655     if (s1->option_pthread)
3656       putdef(cs, "_REENTRANT");
3657     if (s1->leading_underscore)
3658       putdef(cs, "__leading_underscore");
3659     cstr_printf(cs, "#define __SIZEOF_POINTER__ %d\n", PTR_SIZE);
3660     cstr_printf(cs, "#define __SIZEOF_LONG__ %d\n", LONG_SIZE);
3661     if (!is_asm) {
3662       putdef(cs, "__STDC__");
3663       cstr_printf(cs, "#define __STDC_VERSION__ %dL\n", s1->cversion);
3664       cstr_cat(cs,
3665         /* load more predefs and __builtins */
3666 #if CONFIG_TCC_PREDEFS
3667         #include "tccdefs_.h" /* include as strings */
3668 #else
3669         "#include <tccdefs.h>\n" /* load at runtime */
3670 #endif
3671         , -1);
3672     }
3673     cstr_printf(cs, "#define __BASE_FILE__ \"%s\"\n", file->filename);
3674 }
3675
3676 ST_FUNC void preprocess_start(TCCState *s1, int filetype)
3677 {
3678     int is_asm = !!(filetype & (AFF_TYPE_ASM|AFF_TYPE_ASMPP));
3679
3680     tccpp_new(s1);
3681
3682     s1->include_stack_ptr = s1->include_stack;
3683     s1->ifdef_stack_ptr = s1->ifdef_stack;
3684     file->ifdef_stack_ptr = s1->ifdef_stack_ptr;
3685     pp_expr = 0;
3686     pp_counter = 0;
3687     pp_debug_tok = pp_debug_symv = 0;
3688     pp_once++;
3689     s1->pack_stack[0] = 0;
3690     s1->pack_stack_ptr = s1->pack_stack;
3691
3692     set_idnum('$', !is_asm && s1->dollars_in_identifiers ? IS_ID : 0);
3693     set_idnum('.', is_asm ? IS_ID : 0);
3694
3695     if (!(filetype & AFF_TYPE_ASM)) {
3696         CString cstr;
3697         cstr_new(&cstr);
3698         tcc_predefs(s1, &cstr, is_asm);
3699         if (s1->cmdline_defs.size)
3700           cstr_cat(&cstr, s1->cmdline_defs.data, s1->cmdline_defs.size);
3701         if (s1->cmdline_incl.size)
3702           cstr_cat(&cstr, s1->cmdline_incl.data, s1->cmdline_incl.size);
3703         //printf("%s\n", (char*)cstr.data);
3704         *s1->include_stack_ptr++ = file;
3705         tcc_open_bf(s1, "<command line>", cstr.size);
3706         memcpy(file->buffer, cstr.data, cstr.size);
3707         cstr_free(&cstr);
3708     }
3709
3710     parse_flags = is_asm ? PARSE_FLAG_ASM_FILE : 0;
3711     tok_flags = TOK_FLAG_BOL | TOK_FLAG_BOF;
3712 }
3713
3714 /* cleanup from error/setjmp */
3715 ST_FUNC void preprocess_end(TCCState *s1)
3716 {
3717     while (macro_stack)
3718         end_macro();
3719     macro_ptr = NULL;
3720     while (file)
3721         tcc_close();
3722     tccpp_delete(s1);
3723 }
3724
3725 ST_FUNC int set_idnum(int c, int val)
3726 {
3727     int prev = isidnum_table[c - CH_EOF];
3728     isidnum_table[c - CH_EOF] = val;
3729     return prev;
3730 }
3731
3732 ST_FUNC void tccpp_new(TCCState *s)
3733 {
3734     int i, c;
3735     const char *p, *r;
3736
3737     /* init isid table */
3738     for(i = CH_EOF; i<128; i++)
3739         set_idnum(i,
3740             is_space(i) ? IS_SPC
3741             : isid(i) ? IS_ID
3742             : isnum(i) ? IS_NUM
3743             : 0);
3744
3745     for(i = 128; i<256; i++)
3746         set_idnum(i, IS_ID);
3747
3748     /* init allocators */
3749     tal_new(&toksym_alloc, TOKSYM_TAL_LIMIT, TOKSYM_TAL_SIZE);
3750     tal_new(&tokstr_alloc, TOKSTR_TAL_LIMIT, TOKSTR_TAL_SIZE);
3751
3752     memset(hash_ident, 0, TOK_HASH_SIZE * sizeof(TokenSym *));
3753     memset(s->cached_includes_hash, 0, sizeof s->cached_includes_hash);
3754
3755     cstr_new(&tokcstr);
3756     cstr_new(&cstr_buf);
3757     cstr_realloc(&cstr_buf, STRING_MAX_SIZE);
3758     tok_str_new(&tokstr_buf);
3759     tok_str_realloc(&tokstr_buf, TOKSTR_MAX_SIZE);
3760
3761     tok_ident = TOK_IDENT;
3762     p = tcc_keywords;
3763     while (*p) {
3764         r = p;
3765         for(;;) {
3766             c = *r++;
3767             if (c == '\0')
3768                 break;
3769         }
3770         tok_alloc(p, r - p - 1);
3771         p = r;
3772     }
3773
3774     /* we add dummy defines for some special macros to speed up tests
3775        and to have working defined() */
3776     define_push(TOK___LINE__, MACRO_OBJ, NULL, NULL);
3777     define_push(TOK___FILE__, MACRO_OBJ, NULL, NULL);
3778     define_push(TOK___DATE__, MACRO_OBJ, NULL, NULL);
3779     define_push(TOK___TIME__, MACRO_OBJ, NULL, NULL);
3780     define_push(TOK___COUNTER__, MACRO_OBJ, NULL, NULL);
3781 }
3782
3783 ST_FUNC void tccpp_delete(TCCState *s)
3784 {
3785     int i, n;
3786
3787     dynarray_reset(&s->cached_includes, &s->nb_cached_includes);
3788
3789     /* free tokens */
3790     n = tok_ident - TOK_IDENT;
3791     if (n > total_idents)
3792         total_idents = n;
3793     for(i = 0; i < n; i++)
3794         tal_free(toksym_alloc, table_ident[i]);
3795     tcc_free(table_ident);
3796     table_ident = NULL;
3797
3798     /* free static buffers */
3799     cstr_free(&tokcstr);
3800     cstr_free(&cstr_buf);
3801     tok_str_free_str(tokstr_buf.str);
3802
3803     /* free allocators */
3804     tal_delete(toksym_alloc);
3805     toksym_alloc = NULL;
3806     tal_delete(tokstr_alloc);
3807     tokstr_alloc = NULL;
3808 }
3809
3810 /* ------------------------------------------------------------------------- */
3811 /* tcc -E [-P[1]] [-dD} support */
3812
3813 static void tok_print(const char *msg, const int *str)
3814 {
3815     FILE *fp;
3816     int t, s = 0;
3817     CValue cval;
3818
3819     fp = tcc_state->ppfp;
3820     fprintf(fp, "%s", msg);
3821     while (str) {
3822         TOK_GET(&t, &str, &cval);
3823         if (!t)
3824             break;
3825         fprintf(fp, &" %s"[s], get_tok_str(t, &cval)), s = 1;
3826     }
3827     fprintf(fp, "\n");
3828 }
3829
3830 static void pp_line(TCCState *s1, BufferedFile *f, int level)
3831 {
3832     int d = f->line_num - f->line_ref;
3833
3834     if (s1->dflag & 4)
3835         return;
3836
3837     if (s1->Pflag == LINE_MACRO_OUTPUT_FORMAT_NONE) {
3838         ;
3839     } else if (level == 0 && f->line_ref && d < 8) {
3840         while (d > 0)
3841             fputs("\n", s1->ppfp), --d;
3842     } else if (s1->Pflag == LINE_MACRO_OUTPUT_FORMAT_STD) {
3843         fprintf(s1->ppfp, "#line %d \"%s\"\n", f->line_num, f->filename);
3844     } else {
3845         fprintf(s1->ppfp, "# %d \"%s\"%s\n", f->line_num, f->filename,
3846             level > 0 ? " 1" : level < 0 ? " 2" : "");
3847     }
3848     f->line_ref = f->line_num;
3849 }
3850
3851 static void define_print(TCCState *s1, int v)
3852 {
3853     FILE *fp;
3854     Sym *s;
3855
3856     s = define_find(v);
3857     if (NULL == s || NULL == s->d)
3858         return;
3859
3860     fp = s1->ppfp;
3861     fprintf(fp, "#define %s", get_tok_str(v, NULL));
3862     if (s->type.t == MACRO_FUNC) {
3863         Sym *a = s->next;
3864         fprintf(fp,"(");
3865         if (a)
3866             for (;;) {
3867                 fprintf(fp,"%s", get_tok_str(a->v & ~SYM_FIELD, NULL));
3868                 if (!(a = a->next))
3869                     break;
3870                 fprintf(fp,",");
3871             }
3872         fprintf(fp,")");
3873     }
3874     tok_print("", s->d);
3875 }
3876
3877 static void pp_debug_defines(TCCState *s1)
3878 {
3879     int v, t;
3880     const char *vs;
3881     FILE *fp;
3882
3883     t = pp_debug_tok;
3884     if (t == 0)
3885         return;
3886
3887     file->line_num--;
3888     pp_line(s1, file, 0);
3889     file->line_ref = ++file->line_num;
3890
3891     fp = s1->ppfp;
3892     v = pp_debug_symv;
3893     vs = get_tok_str(v, NULL);
3894     if (t == TOK_DEFINE) {
3895         define_print(s1, v);
3896     } else if (t == TOK_UNDEF) {
3897         fprintf(fp, "#undef %s\n", vs);
3898     } else if (t == TOK_push_macro) {
3899         fprintf(fp, "#pragma push_macro(\"%s\")\n", vs);
3900     } else if (t == TOK_pop_macro) {
3901         fprintf(fp, "#pragma pop_macro(\"%s\")\n", vs);
3902     }
3903     pp_debug_tok = 0;
3904 }
3905
3906 static void pp_debug_builtins(TCCState *s1)
3907 {
3908     int v;
3909     for (v = TOK_IDENT; v < tok_ident; ++v)
3910         define_print(s1, v);
3911 }
3912
3913 /* Add a space between tokens a and b to avoid unwanted textual pasting */
3914 static int pp_need_space(int a, int b)
3915 {
3916     return 'E' == a ? '+' == b || '-' == b
3917         : '+' == a ? TOK_INC == b || '+' == b
3918         : '-' == a ? TOK_DEC == b || '-' == b
3919         : a >= TOK_IDENT ? b >= TOK_IDENT
3920         : a == TOK_PPNUM ? b >= TOK_IDENT
3921         : 0;
3922 }
3923
3924 /* maybe hex like 0x1e */
3925 static int pp_check_he0xE(int t, const char *p)
3926 {
3927     if (t == TOK_PPNUM && toup(strchr(p, 0)[-1]) == 'E')
3928         return 'E';
3929     return t;
3930 }
3931
3932 /* Preprocess the current file */
3933 ST_FUNC int tcc_preprocess(TCCState *s1)
3934 {
3935     BufferedFile **iptr;
3936     int token_seen, spcs, level;
3937     const char *p;
3938     char white[400];
3939
3940     parse_flags = PARSE_FLAG_PREPROCESS
3941                 | (parse_flags & PARSE_FLAG_ASM_FILE)
3942                 | PARSE_FLAG_LINEFEED
3943                 | PARSE_FLAG_SPACES
3944                 | PARSE_FLAG_ACCEPT_STRAYS
3945                 ;
3946     /* Credits to Fabrice Bellard's initial revision to demonstrate its
3947        capability to compile and run itself, provided all numbers are
3948        given as decimals. tcc -E -P10 will do. */
3949     if (s1->Pflag == LINE_MACRO_OUTPUT_FORMAT_P10)
3950         parse_flags |= PARSE_FLAG_TOK_NUM, s1->Pflag = 1;
3951
3952     if (s1->do_bench) {
3953         /* for PP benchmarks */
3954         do next(); while (tok != TOK_EOF);
3955         return 0;
3956     }
3957
3958     if (s1->dflag & 1) {
3959         pp_debug_builtins(s1);
3960         s1->dflag &= ~1;
3961     }
3962
3963     token_seen = TOK_LINEFEED, spcs = 0, level = 0;
3964     if (file->prev)
3965         pp_line(s1, file->prev, level++);
3966     pp_line(s1, file, level);
3967     for (;;) {
3968         iptr = s1->include_stack_ptr;
3969         next();
3970         if (tok == TOK_EOF)
3971             break;
3972
3973         level = s1->include_stack_ptr - iptr;
3974         if (level) {
3975             if (level > 0)
3976                 pp_line(s1, *iptr, 0);
3977             pp_line(s1, file, level);
3978         }
3979         if (s1->dflag & 7) {
3980             pp_debug_defines(s1);
3981             if (s1->dflag & 4)
3982                 continue;
3983         }
3984
3985         if (is_space(tok)) {
3986             if (spcs < sizeof white - 1)
3987                 white[spcs++] = tok;
3988             continue;
3989         } else if (tok == TOK_LINEFEED) {
3990             spcs = 0;
3991             if (token_seen == TOK_LINEFEED)
3992                 continue;
3993             ++file->line_ref;
3994         } else if (token_seen == TOK_LINEFEED) {
3995             pp_line(s1, file, 0);
3996         } else if (spcs == 0 && pp_need_space(token_seen, tok)) {
3997             white[spcs++] = ' ';
3998         }
3999
4000         white[spcs] = 0, fputs(white, s1->ppfp), spcs = 0;
4001         fputs(p = get_tok_str(tok, &tokc), s1->ppfp);
4002         token_seen = pp_check_he0xE(tok, p);
4003     }
4004     return 0;
4005 }
4006
4007 /* ------------------------------------------------------------------------- */