unifdef.c revision 205089
1/* 2 * Copyright (c) 2002 - 2010 Tony Finch <dot@dotat.at> 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 */ 25 26/* 27 * unifdef - remove ifdef'ed lines 28 * 29 * This code was derived from software contributed to Berkeley by Dave Yost. 30 * It was rewritten to support ANSI C by Tony Finch. The original version 31 * of unifdef carried the 4-clause BSD copyright licence. None of its code 32 * remains in this version (though some of the names remain) so it now 33 * carries a more liberal licence. 34 * 35 * Wishlist: 36 * provide an option which will append the name of the 37 * appropriate symbol after #else's and #endif's 38 * provide an option which will check symbols after 39 * #else's and #endif's to see that they match their 40 * corresponding #ifdef or #ifndef 41 * 42 * These require better buffer handling, which would also make 43 * it possible to handle all "dodgy" directives correctly. 44 */ 45 46#include <sys/types.h> 47#include <sys/stat.h> 48 49#include <ctype.h> 50#include <err.h> 51#include <errno.h> 52#include <stdarg.h> 53#include <stdbool.h> 54#include <stdio.h> 55#include <stdlib.h> 56#include <string.h> 57#include <unistd.h> 58 59const char copyright[] = 60 "@(#) $Version: unifdef-2.3 $\n" 61 "@(#) $FreeBSD: head/usr.bin/unifdef/unifdef.c 205089 2010-03-12 17:55:29Z fanf $\n" 62 "@(#) $Author: Tony Finch (dot@dotat.at) $\n" 63 "@(#) $URL: http://dotat.at/prog/unifdef $\n" 64; 65 66/* types of input lines: */ 67typedef enum { 68 LT_TRUEI, /* a true #if with ignore flag */ 69 LT_FALSEI, /* a false #if with ignore flag */ 70 LT_IF, /* an unknown #if */ 71 LT_TRUE, /* a true #if */ 72 LT_FALSE, /* a false #if */ 73 LT_ELIF, /* an unknown #elif */ 74 LT_ELTRUE, /* a true #elif */ 75 LT_ELFALSE, /* a false #elif */ 76 LT_ELSE, /* #else */ 77 LT_ENDIF, /* #endif */ 78 LT_DODGY, /* flag: directive is not on one line */ 79 LT_DODGY_LAST = LT_DODGY + LT_ENDIF, 80 LT_PLAIN, /* ordinary line */ 81 LT_EOF, /* end of file */ 82 LT_ERROR, /* unevaluable #if */ 83 LT_COUNT 84} Linetype; 85 86static char const * const linetype_name[] = { 87 "TRUEI", "FALSEI", "IF", "TRUE", "FALSE", 88 "ELIF", "ELTRUE", "ELFALSE", "ELSE", "ENDIF", 89 "DODGY TRUEI", "DODGY FALSEI", 90 "DODGY IF", "DODGY TRUE", "DODGY FALSE", 91 "DODGY ELIF", "DODGY ELTRUE", "DODGY ELFALSE", 92 "DODGY ELSE", "DODGY ENDIF", 93 "PLAIN", "EOF", "ERROR" 94}; 95 96/* state of #if processing */ 97typedef enum { 98 IS_OUTSIDE, 99 IS_FALSE_PREFIX, /* false #if followed by false #elifs */ 100 IS_TRUE_PREFIX, /* first non-false #(el)if is true */ 101 IS_PASS_MIDDLE, /* first non-false #(el)if is unknown */ 102 IS_FALSE_MIDDLE, /* a false #elif after a pass state */ 103 IS_TRUE_MIDDLE, /* a true #elif after a pass state */ 104 IS_PASS_ELSE, /* an else after a pass state */ 105 IS_FALSE_ELSE, /* an else after a true state */ 106 IS_TRUE_ELSE, /* an else after only false states */ 107 IS_FALSE_TRAILER, /* #elifs after a true are false */ 108 IS_COUNT 109} Ifstate; 110 111static char const * const ifstate_name[] = { 112 "OUTSIDE", "FALSE_PREFIX", "TRUE_PREFIX", 113 "PASS_MIDDLE", "FALSE_MIDDLE", "TRUE_MIDDLE", 114 "PASS_ELSE", "FALSE_ELSE", "TRUE_ELSE", 115 "FALSE_TRAILER" 116}; 117 118/* state of comment parser */ 119typedef enum { 120 NO_COMMENT = false, /* outside a comment */ 121 C_COMMENT, /* in a comment like this one */ 122 CXX_COMMENT, /* between // and end of line */ 123 STARTING_COMMENT, /* just after slash-backslash-newline */ 124 FINISHING_COMMENT, /* star-backslash-newline in a C comment */ 125 CHAR_LITERAL, /* inside '' */ 126 STRING_LITERAL /* inside "" */ 127} Comment_state; 128 129static char const * const comment_name[] = { 130 "NO", "C", "CXX", "STARTING", "FINISHING", "CHAR", "STRING" 131}; 132 133/* state of preprocessor line parser */ 134typedef enum { 135 LS_START, /* only space and comments on this line */ 136 LS_HASH, /* only space, comments, and a hash */ 137 LS_DIRTY /* this line can't be a preprocessor line */ 138} Line_state; 139 140static char const * const linestate_name[] = { 141 "START", "HASH", "DIRTY" 142}; 143 144/* 145 * Minimum translation limits from ISO/IEC 9899:1999 5.2.4.1 146 */ 147#define MAXDEPTH 64 /* maximum #if nesting */ 148#define MAXLINE 4096 /* maximum length of line */ 149#define MAXSYMS 4096 /* maximum number of symbols */ 150 151/* 152 * Sometimes when editing a keyword the replacement text is longer, so 153 * we leave some space at the end of the tline buffer to accommodate this. 154 */ 155#define EDITSLOP 10 156 157/* 158 * For temporary filenames 159 */ 160#define TEMPLATE "unifdef.XXXXXX" 161 162/* 163 * Globals. 164 */ 165 166static bool compblank; /* -B: compress blank lines */ 167static bool lnblank; /* -b: blank deleted lines */ 168static bool complement; /* -c: do the complement */ 169static bool debugging; /* -d: debugging reports */ 170static bool iocccok; /* -e: fewer IOCCC errors */ 171static bool strictlogic; /* -K: keep ambiguous #ifs */ 172static bool killconsts; /* -k: eval constant #ifs */ 173static bool lnnum; /* -n: add #line directives */ 174static bool symlist; /* -s: output symbol list */ 175static bool symdepth; /* -S: output symbol depth */ 176static bool text; /* -t: this is a text file */ 177 178static const char *symname[MAXSYMS]; /* symbol name */ 179static const char *value[MAXSYMS]; /* -Dsym=value */ 180static bool ignore[MAXSYMS]; /* -iDsym or -iUsym */ 181static int nsyms; /* number of symbols */ 182 183static FILE *input; /* input file pointer */ 184static const char *filename; /* input file name */ 185static int linenum; /* current line number */ 186static FILE *output; /* output file pointer */ 187static const char *ofilename; /* output file name */ 188static bool overwriting; /* output overwrites input */ 189static char tempname[FILENAME_MAX]; /* used when overwriting */ 190 191static char tline[MAXLINE+EDITSLOP];/* input buffer plus space */ 192static char *keyword; /* used for editing #elif's */ 193 194static const char *newline; /* input file format */ 195static const char newline_unix[] = "\n"; 196static const char newline_crlf[] = "\r\n"; 197 198static Comment_state incomment; /* comment parser state */ 199static Line_state linestate; /* #if line parser state */ 200static Ifstate ifstate[MAXDEPTH]; /* #if processor state */ 201static bool ignoring[MAXDEPTH]; /* ignore comments state */ 202static int stifline[MAXDEPTH]; /* start of current #if */ 203static int depth; /* current #if nesting */ 204static int delcount; /* count of deleted lines */ 205static unsigned blankcount; /* count of blank lines */ 206static unsigned blankmax; /* maximum recent blankcount */ 207static bool constexpr; /* constant #if expression */ 208static bool zerosyms = true; /* to format symdepth output */ 209static bool firstsym; /* ditto */ 210 211static int exitstat; /* program exit status */ 212 213static void addsym(bool, bool, char *); 214static void closeout(void); 215static void debug(const char *, ...); 216static void done(void); 217static void error(const char *); 218static int findsym(const char *); 219static void flushline(bool); 220static Linetype parseline(void); 221static Linetype ifeval(const char **); 222static void ignoreoff(void); 223static void ignoreon(void); 224static void keywordedit(const char *); 225static void nest(void); 226static void process(void); 227static const char *skipargs(const char *); 228static const char *skipcomment(const char *); 229static const char *skipsym(const char *); 230static void state(Ifstate); 231static int strlcmp(const char *, const char *, size_t); 232static void unnest(void); 233static void usage(void); 234static void version(void); 235 236#define endsym(c) (!isalnum((unsigned char)c) && c != '_') 237 238/* 239 * The main program. 240 */ 241int 242main(int argc, char *argv[]) 243{ 244 int opt; 245 246 while ((opt = getopt(argc, argv, "i:D:U:I:o:bBcdeKklnsStV")) != -1) 247 switch (opt) { 248 case 'i': /* treat stuff controlled by these symbols as text */ 249 /* 250 * For strict backwards-compatibility the U or D 251 * should be immediately after the -i but it doesn't 252 * matter much if we relax that requirement. 253 */ 254 opt = *optarg++; 255 if (opt == 'D') 256 addsym(true, true, optarg); 257 else if (opt == 'U') 258 addsym(true, false, optarg); 259 else 260 usage(); 261 break; 262 case 'D': /* define a symbol */ 263 addsym(false, true, optarg); 264 break; 265 case 'U': /* undef a symbol */ 266 addsym(false, false, optarg); 267 break; 268 case 'I': /* no-op for compatibility with cpp */ 269 break; 270 case 'b': /* blank deleted lines instead of omitting them */ 271 case 'l': /* backwards compatibility */ 272 lnblank = true; 273 break; 274 case 'B': /* compress blank lines around removed section */ 275 compblank = true; 276 break; 277 case 'c': /* treat -D as -U and vice versa */ 278 complement = true; 279 break; 280 case 'd': 281 debugging = true; 282 break; 283 case 'e': /* fewer errors from dodgy lines */ 284 iocccok = true; 285 break; 286 case 'K': /* keep ambiguous #ifs */ 287 strictlogic = true; 288 break; 289 case 'k': /* process constant #ifs */ 290 killconsts = true; 291 break; 292 case 'n': /* add #line directive after deleted lines */ 293 lnnum = true; 294 break; 295 case 'o': /* output to a file */ 296 ofilename = optarg; 297 break; 298 case 's': /* only output list of symbols that control #ifs */ 299 symlist = true; 300 break; 301 case 'S': /* list symbols with their nesting depth */ 302 symlist = symdepth = true; 303 break; 304 case 't': /* don't parse C comments */ 305 text = true; 306 break; 307 case 'V': /* print version */ 308 version(); 309 default: 310 usage(); 311 } 312 argc -= optind; 313 argv += optind; 314 if (compblank && lnblank) 315 errx(2, "-B and -b are mutually exclusive"); 316 if (argc > 1) { 317 errx(2, "can only do one file"); 318 } else if (argc == 1 && strcmp(*argv, "-") != 0) { 319 filename = *argv; 320 input = fopen(filename, "rb"); 321 if (input == NULL) 322 err(2, "can't open %s", filename); 323 } else { 324 filename = "[stdin]"; 325 input = stdin; 326 } 327 if (ofilename == NULL) { 328 ofilename = "[stdout]"; 329 output = stdout; 330 } else { 331 struct stat ist, ost; 332 memset(&ist, 0, sizeof(ist)); 333 memset(&ost, 0, sizeof(ost)); 334 335 if (fstat(fileno(input), &ist) != 0) 336 err(2, "can't fstat %s", filename); 337 if (stat(ofilename, &ost) != 0 && errno != ENOENT) 338 warn("can't stat %s", ofilename); 339 340 overwriting = (ist.st_dev == ost.st_dev 341 && ist.st_ino == ost.st_ino); 342 if (overwriting) { 343 const char *dirsep; 344 int ofd; 345 346 dirsep = strrchr(ofilename, '/'); 347 if (dirsep != NULL) 348 snprintf(tempname, sizeof(tempname), 349 "%.*s/" TEMPLATE, 350 (int)(dirsep - ofilename), ofilename); 351 else 352 snprintf(tempname, sizeof(tempname), 353 TEMPLATE); 354 ofd = mkstemp(tempname); 355 if (ofd != -1) 356 output = fdopen(ofd, "wb+"); 357 if (output == NULL) 358 err(2, "can't create temporary file"); 359 fchmod(ofd, ist.st_mode & (S_IRWXU|S_IRWXG|S_IRWXO)); 360 } else { 361 output = fopen(ofilename, "wb"); 362 if (output == NULL) 363 err(2, "can't open %s", ofilename); 364 } 365 } 366 process(); 367 abort(); /* bug */ 368} 369 370static void 371version(void) 372{ 373 const char *c = copyright; 374 for (;;) { 375 while (*++c != '$') 376 if (*c == '\0') 377 exit(0); 378 while (*++c != '$') 379 putc(*c, stderr); 380 putc('\n', stderr); 381 } 382} 383 384static void 385usage(void) 386{ 387 fprintf(stderr, "usage: unifdef [-bBcdeKknsStV] [-Ipath]" 388 " [-Dsym[=val]] [-Usym] [-iDsym[=val]] [-iUsym] ... [file]\n"); 389 exit(2); 390} 391 392/* 393 * A state transition function alters the global #if processing state 394 * in a particular way. The table below is indexed by the current 395 * processing state and the type of the current line. 396 * 397 * Nesting is handled by keeping a stack of states; some transition 398 * functions increase or decrease the depth. They also maintain the 399 * ignore state on a stack. In some complicated cases they have to 400 * alter the preprocessor directive, as follows. 401 * 402 * When we have processed a group that starts off with a known-false 403 * #if/#elif sequence (which has therefore been deleted) followed by a 404 * #elif that we don't understand and therefore must keep, we edit the 405 * latter into a #if to keep the nesting correct. 406 * 407 * When we find a true #elif in a group, the following block will 408 * always be kept and the rest of the sequence after the next #elif or 409 * #else will be discarded. We edit the #elif into a #else and the 410 * following directive to #endif since this has the desired behaviour. 411 * 412 * "Dodgy" directives are split across multiple lines, the most common 413 * example being a multi-line comment hanging off the right of the 414 * directive. We can handle them correctly only if there is no change 415 * from printing to dropping (or vice versa) caused by that directive. 416 * If the directive is the first of a group we have a choice between 417 * failing with an error, or passing it through unchanged instead of 418 * evaluating it. The latter is not the default to avoid questions from 419 * users about unifdef unexpectedly leaving behind preprocessor directives. 420 */ 421typedef void state_fn(void); 422 423/* report an error */ 424static void Eelif (void) { error("Inappropriate #elif"); } 425static void Eelse (void) { error("Inappropriate #else"); } 426static void Eendif(void) { error("Inappropriate #endif"); } 427static void Eeof (void) { error("Premature EOF"); } 428static void Eioccc(void) { error("Obfuscated preprocessor control line"); } 429/* plain line handling */ 430static void print (void) { flushline(true); } 431static void drop (void) { flushline(false); } 432/* output lacks group's start line */ 433static void Strue (void) { drop(); ignoreoff(); state(IS_TRUE_PREFIX); } 434static void Sfalse(void) { drop(); ignoreoff(); state(IS_FALSE_PREFIX); } 435static void Selse (void) { drop(); state(IS_TRUE_ELSE); } 436/* print/pass this block */ 437static void Pelif (void) { print(); ignoreoff(); state(IS_PASS_MIDDLE); } 438static void Pelse (void) { print(); state(IS_PASS_ELSE); } 439static void Pendif(void) { print(); unnest(); } 440/* discard this block */ 441static void Dfalse(void) { drop(); ignoreoff(); state(IS_FALSE_TRAILER); } 442static void Delif (void) { drop(); ignoreoff(); state(IS_FALSE_MIDDLE); } 443static void Delse (void) { drop(); state(IS_FALSE_ELSE); } 444static void Dendif(void) { drop(); unnest(); } 445/* first line of group */ 446static void Fdrop (void) { nest(); Dfalse(); } 447static void Fpass (void) { nest(); Pelif(); } 448static void Ftrue (void) { nest(); Strue(); } 449static void Ffalse(void) { nest(); Sfalse(); } 450/* variable pedantry for obfuscated lines */ 451static void Oiffy (void) { if (!iocccok) Eioccc(); Fpass(); ignoreon(); } 452static void Oif (void) { if (!iocccok) Eioccc(); Fpass(); } 453static void Oelif (void) { if (!iocccok) Eioccc(); Pelif(); } 454/* ignore comments in this block */ 455static void Idrop (void) { Fdrop(); ignoreon(); } 456static void Itrue (void) { Ftrue(); ignoreon(); } 457static void Ifalse(void) { Ffalse(); ignoreon(); } 458/* edit this line */ 459static void Mpass (void) { strncpy(keyword, "if ", 4); Pelif(); } 460static void Mtrue (void) { keywordedit("else"); state(IS_TRUE_MIDDLE); } 461static void Melif (void) { keywordedit("endif"); state(IS_FALSE_TRAILER); } 462static void Melse (void) { keywordedit("endif"); state(IS_FALSE_ELSE); } 463 464static state_fn * const trans_table[IS_COUNT][LT_COUNT] = { 465/* IS_OUTSIDE */ 466{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Eendif, 467 Oiffy, Oiffy, Fpass, Oif, Oif, Eelif, Eelif, Eelif, Eelse, Eendif, 468 print, done, abort }, 469/* IS_FALSE_PREFIX */ 470{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Mpass, Strue, Sfalse,Selse, Dendif, 471 Idrop, Idrop, Fdrop, Fdrop, Fdrop, Mpass, Eioccc,Eioccc,Eioccc,Eioccc, 472 drop, Eeof, abort }, 473/* IS_TRUE_PREFIX */ 474{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Dfalse,Dfalse,Dfalse,Delse, Dendif, 475 Oiffy, Oiffy, Fpass, Oif, Oif, Eioccc,Eioccc,Eioccc,Eioccc,Eioccc, 476 print, Eeof, abort }, 477/* IS_PASS_MIDDLE */ 478{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Pelif, Mtrue, Delif, Pelse, Pendif, 479 Oiffy, Oiffy, Fpass, Oif, Oif, Pelif, Oelif, Oelif, Pelse, Pendif, 480 print, Eeof, abort }, 481/* IS_FALSE_MIDDLE */ 482{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Pelif, Mtrue, Delif, Pelse, Pendif, 483 Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eioccc,Eioccc,Eioccc,Eioccc,Eioccc, 484 drop, Eeof, abort }, 485/* IS_TRUE_MIDDLE */ 486{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Melif, Melif, Melif, Melse, Pendif, 487 Oiffy, Oiffy, Fpass, Oif, Oif, Eioccc,Eioccc,Eioccc,Eioccc,Pendif, 488 print, Eeof, abort }, 489/* IS_PASS_ELSE */ 490{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Pendif, 491 Oiffy, Oiffy, Fpass, Oif, Oif, Eelif, Eelif, Eelif, Eelse, Pendif, 492 print, Eeof, abort }, 493/* IS_FALSE_ELSE */ 494{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eelif, Eelif, Eelif, Eelse, Dendif, 495 Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eelif, Eelif, Eelif, Eelse, Eioccc, 496 drop, Eeof, abort }, 497/* IS_TRUE_ELSE */ 498{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Dendif, 499 Oiffy, Oiffy, Fpass, Oif, Oif, Eelif, Eelif, Eelif, Eelse, Eioccc, 500 print, Eeof, abort }, 501/* IS_FALSE_TRAILER */ 502{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Dfalse,Dfalse,Dfalse,Delse, Dendif, 503 Idrop, Idrop, Fdrop, Fdrop, Fdrop, Dfalse,Dfalse,Dfalse,Delse, Eioccc, 504 drop, Eeof, abort } 505/*TRUEI FALSEI IF TRUE FALSE ELIF ELTRUE ELFALSE ELSE ENDIF 506 TRUEI FALSEI IF TRUE FALSE ELIF ELTRUE ELFALSE ELSE ENDIF (DODGY) 507 PLAIN EOF ERROR */ 508}; 509 510/* 511 * State machine utility functions 512 */ 513static void 514ignoreoff(void) 515{ 516 if (depth == 0) 517 abort(); /* bug */ 518 ignoring[depth] = ignoring[depth-1]; 519} 520static void 521ignoreon(void) 522{ 523 ignoring[depth] = true; 524} 525static void 526keywordedit(const char *replacement) 527{ 528 snprintf(keyword, tline + sizeof(tline) - keyword, 529 "%s%s", replacement, newline); 530 print(); 531} 532static void 533nest(void) 534{ 535 if (depth > MAXDEPTH-1) 536 abort(); /* bug */ 537 if (depth == MAXDEPTH-1) 538 error("Too many levels of nesting"); 539 depth += 1; 540 stifline[depth] = linenum; 541} 542static void 543unnest(void) 544{ 545 if (depth == 0) 546 abort(); /* bug */ 547 depth -= 1; 548} 549static void 550state(Ifstate is) 551{ 552 ifstate[depth] = is; 553} 554 555/* 556 * Write a line to the output or not, according to command line options. 557 */ 558static void 559flushline(bool keep) 560{ 561 if (symlist) 562 return; 563 if (keep ^ complement) { 564 bool blankline = tline[strspn(tline, " \t\r\n")] == '\0'; 565 if (blankline && compblank && blankcount != blankmax) { 566 delcount += 1; 567 blankcount += 1; 568 } else { 569 if (lnnum && delcount > 0) 570 printf("#line %d%s", linenum, newline); 571 fputs(tline, output); 572 delcount = 0; 573 blankmax = blankcount = blankline ? blankcount + 1 : 0; 574 } 575 } else { 576 if (lnblank) 577 fputs(newline, output); 578 exitstat = 1; 579 delcount += 1; 580 blankcount = 0; 581 } 582 if (debugging) 583 fflush(output); 584} 585 586/* 587 * The driver for the state machine. 588 */ 589static void 590process(void) 591{ 592 /* When compressing blank lines, act as if the file 593 is preceded by a large number of blank lines. */ 594 blankmax = blankcount = 1000; 595 for (;;) { 596 Linetype lineval = parseline(); 597 trans_table[ifstate[depth]][lineval](); 598 debug("process line %d %s -> %s depth %d", 599 linenum, linetype_name[lineval], 600 ifstate_name[ifstate[depth]], depth); 601 } 602} 603 604/* 605 * Flush the output and handle errors. 606 */ 607static void 608closeout(void) 609{ 610 if (symdepth && !zerosyms) 611 printf("\n"); 612 if (fclose(output) == EOF) { 613 warn("couldn't write to %s", ofilename); 614 if (overwriting) { 615 unlink(tempname); 616 errx(2, "%s unchanged", filename); 617 } else { 618 exit(2); 619 } 620 } 621} 622 623/* 624 * Clean up and exit. 625 */ 626static void 627done(void) 628{ 629 if (incomment) 630 error("EOF in comment"); 631 closeout(); 632 if (overwriting && rename(tempname, filename) == -1) { 633 warn("couldn't rename temporary file"); 634 unlink(tempname); 635 errx(2, "%s unchanged", filename); 636 } 637 exit(exitstat); 638} 639 640/* 641 * Parse a line and determine its type. We keep the preprocessor line 642 * parser state between calls in the global variable linestate, with 643 * help from skipcomment(). 644 */ 645static Linetype 646parseline(void) 647{ 648 const char *cp; 649 int cursym; 650 int kwlen; 651 Linetype retval; 652 Comment_state wascomment; 653 654 linenum++; 655 if (fgets(tline, MAXLINE, input) == NULL) 656 return (LT_EOF); 657 if (newline == NULL) { 658 if (strrchr(tline, '\n') == strrchr(tline, '\r') + 1) 659 newline = newline_crlf; 660 else 661 newline = newline_unix; 662 } 663 retval = LT_PLAIN; 664 wascomment = incomment; 665 cp = skipcomment(tline); 666 if (linestate == LS_START) { 667 if (*cp == '#') { 668 linestate = LS_HASH; 669 firstsym = true; 670 cp = skipcomment(cp + 1); 671 } else if (*cp != '\0') 672 linestate = LS_DIRTY; 673 } 674 if (!incomment && linestate == LS_HASH) { 675 keyword = tline + (cp - tline); 676 cp = skipsym(cp); 677 kwlen = cp - keyword; 678 /* no way can we deal with a continuation inside a keyword */ 679 if (strncmp(cp, "\\\r\n", 3) == 0 || 680 strncmp(cp, "\\\n", 2) == 0) 681 Eioccc(); 682 if (strlcmp("ifdef", keyword, kwlen) == 0 || 683 strlcmp("ifndef", keyword, kwlen) == 0) { 684 cp = skipcomment(cp); 685 if ((cursym = findsym(cp)) < 0) 686 retval = LT_IF; 687 else { 688 retval = (keyword[2] == 'n') 689 ? LT_FALSE : LT_TRUE; 690 if (value[cursym] == NULL) 691 retval = (retval == LT_TRUE) 692 ? LT_FALSE : LT_TRUE; 693 if (ignore[cursym]) 694 retval = (retval == LT_TRUE) 695 ? LT_TRUEI : LT_FALSEI; 696 } 697 cp = skipsym(cp); 698 } else if (strlcmp("if", keyword, kwlen) == 0) 699 retval = ifeval(&cp); 700 else if (strlcmp("elif", keyword, kwlen) == 0) 701 retval = ifeval(&cp) - LT_IF + LT_ELIF; 702 else if (strlcmp("else", keyword, kwlen) == 0) 703 retval = LT_ELSE; 704 else if (strlcmp("endif", keyword, kwlen) == 0) 705 retval = LT_ENDIF; 706 else { 707 linestate = LS_DIRTY; 708 retval = LT_PLAIN; 709 } 710 cp = skipcomment(cp); 711 if (*cp != '\0') { 712 linestate = LS_DIRTY; 713 if (retval == LT_TRUE || retval == LT_FALSE || 714 retval == LT_TRUEI || retval == LT_FALSEI) 715 retval = LT_IF; 716 if (retval == LT_ELTRUE || retval == LT_ELFALSE) 717 retval = LT_ELIF; 718 } 719 if (retval != LT_PLAIN && (wascomment || incomment)) { 720 retval += LT_DODGY; 721 if (incomment) 722 linestate = LS_DIRTY; 723 } 724 /* skipcomment normally changes the state, except 725 if the last line of the file lacks a newline, or 726 if there is too much whitespace in a directive */ 727 if (linestate == LS_HASH) { 728 size_t len = cp - tline; 729 if (fgets(tline + len, MAXLINE - len, input) == NULL) { 730 /* append the missing newline */ 731 strcpy(tline + len, newline); 732 cp += strlen(newline); 733 linestate = LS_START; 734 } else { 735 linestate = LS_DIRTY; 736 } 737 } 738 } 739 if (linestate == LS_DIRTY) { 740 while (*cp != '\0') 741 cp = skipcomment(cp + 1); 742 } 743 debug("parser line %d state %s comment %s line", linenum, 744 comment_name[incomment], linestate_name[linestate]); 745 return (retval); 746} 747 748/* 749 * These are the binary operators that are supported by the expression 750 * evaluator. 751 */ 752static Linetype op_strict(int *p, int v, Linetype at, Linetype bt) { 753 if(at == LT_IF || bt == LT_IF) return (LT_IF); 754 return (*p = v, v ? LT_TRUE : LT_FALSE); 755} 756static Linetype op_lt(int *p, Linetype at, int a, Linetype bt, int b) { 757 return op_strict(p, a < b, at, bt); 758} 759static Linetype op_gt(int *p, Linetype at, int a, Linetype bt, int b) { 760 return op_strict(p, a > b, at, bt); 761} 762static Linetype op_le(int *p, Linetype at, int a, Linetype bt, int b) { 763 return op_strict(p, a <= b, at, bt); 764} 765static Linetype op_ge(int *p, Linetype at, int a, Linetype bt, int b) { 766 return op_strict(p, a >= b, at, bt); 767} 768static Linetype op_eq(int *p, Linetype at, int a, Linetype bt, int b) { 769 return op_strict(p, a == b, at, bt); 770} 771static Linetype op_ne(int *p, Linetype at, int a, Linetype bt, int b) { 772 return op_strict(p, a != b, at, bt); 773} 774static Linetype op_or(int *p, Linetype at, int a, Linetype bt, int b) { 775 if (!strictlogic && (at == LT_TRUE || bt == LT_TRUE)) 776 return (*p = 1, LT_TRUE); 777 return op_strict(p, a || b, at, bt); 778} 779static Linetype op_and(int *p, Linetype at, int a, Linetype bt, int b) { 780 if (!strictlogic && (at == LT_FALSE || bt == LT_FALSE)) 781 return (*p = 0, LT_FALSE); 782 return op_strict(p, a && b, at, bt); 783} 784 785/* 786 * An evaluation function takes three arguments, as follows: (1) a pointer to 787 * an element of the precedence table which lists the operators at the current 788 * level of precedence; (2) a pointer to an integer which will receive the 789 * value of the expression; and (3) a pointer to a char* that points to the 790 * expression to be evaluated and that is updated to the end of the expression 791 * when evaluation is complete. The function returns LT_FALSE if the value of 792 * the expression is zero, LT_TRUE if it is non-zero, LT_IF if the expression 793 * depends on an unknown symbol, or LT_ERROR if there is a parse failure. 794 */ 795struct ops; 796 797typedef Linetype eval_fn(const struct ops *, int *, const char **); 798 799static eval_fn eval_table, eval_unary; 800 801/* 802 * The precedence table. Expressions involving binary operators are evaluated 803 * in a table-driven way by eval_table. When it evaluates a subexpression it 804 * calls the inner function with its first argument pointing to the next 805 * element of the table. Innermost expressions have special non-table-driven 806 * handling. 807 */ 808static const struct ops { 809 eval_fn *inner; 810 struct op { 811 const char *str; 812 Linetype (*fn)(int *, Linetype, int, Linetype, int); 813 } op[5]; 814} eval_ops[] = { 815 { eval_table, { { "||", op_or } } }, 816 { eval_table, { { "&&", op_and } } }, 817 { eval_table, { { "==", op_eq }, 818 { "!=", op_ne } } }, 819 { eval_unary, { { "<=", op_le }, 820 { ">=", op_ge }, 821 { "<", op_lt }, 822 { ">", op_gt } } } 823}; 824 825/* 826 * Function for evaluating the innermost parts of expressions, 827 * viz. !expr (expr) number defined(symbol) symbol 828 * We reset the constexpr flag in the last two cases. 829 */ 830static Linetype 831eval_unary(const struct ops *ops, int *valp, const char **cpp) 832{ 833 const char *cp; 834 char *ep; 835 int sym; 836 bool defparen; 837 Linetype lt; 838 839 cp = skipcomment(*cpp); 840 if (*cp == '!') { 841 debug("eval%d !", ops - eval_ops); 842 cp++; 843 lt = eval_unary(ops, valp, &cp); 844 if (lt == LT_ERROR) 845 return (LT_ERROR); 846 if (lt != LT_IF) { 847 *valp = !*valp; 848 lt = *valp ? LT_TRUE : LT_FALSE; 849 } 850 } else if (*cp == '(') { 851 cp++; 852 debug("eval%d (", ops - eval_ops); 853 lt = eval_table(eval_ops, valp, &cp); 854 if (lt == LT_ERROR) 855 return (LT_ERROR); 856 cp = skipcomment(cp); 857 if (*cp++ != ')') 858 return (LT_ERROR); 859 } else if (isdigit((unsigned char)*cp)) { 860 debug("eval%d number", ops - eval_ops); 861 *valp = strtol(cp, &ep, 0); 862 if (ep == cp) 863 return (LT_ERROR); 864 lt = *valp ? LT_TRUE : LT_FALSE; 865 cp = skipsym(cp); 866 } else if (strncmp(cp, "defined", 7) == 0 && endsym(cp[7])) { 867 cp = skipcomment(cp+7); 868 debug("eval%d defined", ops - eval_ops); 869 if (*cp == '(') { 870 cp = skipcomment(cp+1); 871 defparen = true; 872 } else { 873 defparen = false; 874 } 875 sym = findsym(cp); 876 if (sym < 0) { 877 lt = LT_IF; 878 } else { 879 *valp = (value[sym] != NULL); 880 lt = *valp ? LT_TRUE : LT_FALSE; 881 } 882 cp = skipsym(cp); 883 cp = skipcomment(cp); 884 if (defparen && *cp++ != ')') 885 return (LT_ERROR); 886 constexpr = false; 887 } else if (!endsym(*cp)) { 888 debug("eval%d symbol", ops - eval_ops); 889 sym = findsym(cp); 890 cp = skipsym(cp); 891 if (sym < 0) { 892 lt = LT_IF; 893 cp = skipargs(cp); 894 } else if (value[sym] == NULL) { 895 *valp = 0; 896 lt = LT_FALSE; 897 } else { 898 *valp = strtol(value[sym], &ep, 0); 899 if (*ep != '\0' || ep == value[sym]) 900 return (LT_ERROR); 901 lt = *valp ? LT_TRUE : LT_FALSE; 902 cp = skipargs(cp); 903 } 904 constexpr = false; 905 } else { 906 debug("eval%d bad expr", ops - eval_ops); 907 return (LT_ERROR); 908 } 909 910 *cpp = cp; 911 debug("eval%d = %d", ops - eval_ops, *valp); 912 return (lt); 913} 914 915/* 916 * Table-driven evaluation of binary operators. 917 */ 918static Linetype 919eval_table(const struct ops *ops, int *valp, const char **cpp) 920{ 921 const struct op *op; 922 const char *cp; 923 int val; 924 Linetype lt, rt; 925 926 debug("eval%d", ops - eval_ops); 927 cp = *cpp; 928 lt = ops->inner(ops+1, valp, &cp); 929 if (lt == LT_ERROR) 930 return (LT_ERROR); 931 for (;;) { 932 cp = skipcomment(cp); 933 for (op = ops->op; op->str != NULL; op++) 934 if (strncmp(cp, op->str, strlen(op->str)) == 0) 935 break; 936 if (op->str == NULL) 937 break; 938 cp += strlen(op->str); 939 debug("eval%d %s", ops - eval_ops, op->str); 940 rt = ops->inner(ops+1, &val, &cp); 941 if (rt == LT_ERROR) 942 return (LT_ERROR); 943 lt = op->fn(valp, lt, *valp, rt, val); 944 } 945 946 *cpp = cp; 947 debug("eval%d = %d", ops - eval_ops, *valp); 948 debug("eval%d lt = %s", ops - eval_ops, linetype_name[lt]); 949 return (lt); 950} 951 952/* 953 * Evaluate the expression on a #if or #elif line. If we can work out 954 * the result we return LT_TRUE or LT_FALSE accordingly, otherwise we 955 * return just a generic LT_IF. 956 */ 957static Linetype 958ifeval(const char **cpp) 959{ 960 int ret; 961 int val = 0; 962 963 debug("eval %s", *cpp); 964 constexpr = killconsts ? false : true; 965 ret = eval_table(eval_ops, &val, cpp); 966 debug("eval = %d", val); 967 return (constexpr ? LT_IF : ret == LT_ERROR ? LT_IF : ret); 968} 969 970/* 971 * Skip over comments, strings, and character literals and stop at the 972 * next character position that is not whitespace. Between calls we keep 973 * the comment state in the global variable incomment, and we also adjust 974 * the global variable linestate when we see a newline. 975 * XXX: doesn't cope with the buffer splitting inside a state transition. 976 */ 977static const char * 978skipcomment(const char *cp) 979{ 980 if (text || ignoring[depth]) { 981 for (; isspace((unsigned char)*cp); cp++) 982 if (*cp == '\n') 983 linestate = LS_START; 984 return (cp); 985 } 986 while (*cp != '\0') 987 /* don't reset to LS_START after a line continuation */ 988 if (strncmp(cp, "\\\r\n", 3) == 0) 989 cp += 3; 990 else if (strncmp(cp, "\\\n", 2) == 0) 991 cp += 2; 992 else switch (incomment) { 993 case NO_COMMENT: 994 if (strncmp(cp, "/\\\r\n", 4) == 0) { 995 incomment = STARTING_COMMENT; 996 cp += 4; 997 } else if (strncmp(cp, "/\\\n", 3) == 0) { 998 incomment = STARTING_COMMENT; 999 cp += 3; 1000 } else if (strncmp(cp, "/*", 2) == 0) { 1001 incomment = C_COMMENT; 1002 cp += 2; 1003 } else if (strncmp(cp, "//", 2) == 0) { 1004 incomment = CXX_COMMENT; 1005 cp += 2; 1006 } else if (strncmp(cp, "\'", 1) == 0) { 1007 incomment = CHAR_LITERAL; 1008 linestate = LS_DIRTY; 1009 cp += 1; 1010 } else if (strncmp(cp, "\"", 1) == 0) { 1011 incomment = STRING_LITERAL; 1012 linestate = LS_DIRTY; 1013 cp += 1; 1014 } else if (strncmp(cp, "\n", 1) == 0) { 1015 linestate = LS_START; 1016 cp += 1; 1017 } else if (strchr(" \r\t", *cp) != NULL) { 1018 cp += 1; 1019 } else 1020 return (cp); 1021 continue; 1022 case CXX_COMMENT: 1023 if (strncmp(cp, "\n", 1) == 0) { 1024 incomment = NO_COMMENT; 1025 linestate = LS_START; 1026 } 1027 cp += 1; 1028 continue; 1029 case CHAR_LITERAL: 1030 case STRING_LITERAL: 1031 if ((incomment == CHAR_LITERAL && cp[0] == '\'') || 1032 (incomment == STRING_LITERAL && cp[0] == '\"')) { 1033 incomment = NO_COMMENT; 1034 cp += 1; 1035 } else if (cp[0] == '\\') { 1036 if (cp[1] == '\0') 1037 cp += 1; 1038 else 1039 cp += 2; 1040 } else if (strncmp(cp, "\n", 1) == 0) { 1041 if (incomment == CHAR_LITERAL) 1042 error("unterminated char literal"); 1043 else 1044 error("unterminated string literal"); 1045 } else 1046 cp += 1; 1047 continue; 1048 case C_COMMENT: 1049 if (strncmp(cp, "*\\\r\n", 4) == 0) { 1050 incomment = FINISHING_COMMENT; 1051 cp += 4; 1052 } else if (strncmp(cp, "*\\\n", 3) == 0) { 1053 incomment = FINISHING_COMMENT; 1054 cp += 3; 1055 } else if (strncmp(cp, "*/", 2) == 0) { 1056 incomment = NO_COMMENT; 1057 cp += 2; 1058 } else 1059 cp += 1; 1060 continue; 1061 case STARTING_COMMENT: 1062 if (*cp == '*') { 1063 incomment = C_COMMENT; 1064 cp += 1; 1065 } else if (*cp == '/') { 1066 incomment = CXX_COMMENT; 1067 cp += 1; 1068 } else { 1069 incomment = NO_COMMENT; 1070 linestate = LS_DIRTY; 1071 } 1072 continue; 1073 case FINISHING_COMMENT: 1074 if (*cp == '/') { 1075 incomment = NO_COMMENT; 1076 cp += 1; 1077 } else 1078 incomment = C_COMMENT; 1079 continue; 1080 default: 1081 abort(); /* bug */ 1082 } 1083 return (cp); 1084} 1085 1086/* 1087 * Skip macro arguments. 1088 */ 1089static const char * 1090skipargs(const char *cp) 1091{ 1092 const char *ocp = cp; 1093 int level = 0; 1094 cp = skipcomment(cp); 1095 if (*cp != '(') 1096 return (cp); 1097 do { 1098 if (*cp == '(') 1099 level++; 1100 if (*cp == ')') 1101 level--; 1102 cp = skipcomment(cp+1); 1103 } while (level != 0 && *cp != '\0'); 1104 if (level == 0) 1105 return (cp); 1106 else 1107 /* Rewind and re-detect the syntax error later. */ 1108 return (ocp); 1109} 1110 1111/* 1112 * Skip over an identifier. 1113 */ 1114static const char * 1115skipsym(const char *cp) 1116{ 1117 while (!endsym(*cp)) 1118 ++cp; 1119 return (cp); 1120} 1121 1122/* 1123 * Look for the symbol in the symbol table. If it is found, we return 1124 * the symbol table index, else we return -1. 1125 */ 1126static int 1127findsym(const char *str) 1128{ 1129 const char *cp; 1130 int symind; 1131 1132 cp = skipsym(str); 1133 if (cp == str) 1134 return (-1); 1135 if (symlist) { 1136 if (symdepth && firstsym) 1137 printf("%s%3d", zerosyms ? "" : "\n", depth); 1138 firstsym = zerosyms = false; 1139 printf("%s%.*s%s", 1140 symdepth ? " " : "", 1141 (int)(cp-str), str, 1142 symdepth ? "" : "\n"); 1143 /* we don't care about the value of the symbol */ 1144 return (0); 1145 } 1146 for (symind = 0; symind < nsyms; ++symind) { 1147 if (strlcmp(symname[symind], str, cp-str) == 0) { 1148 debug("findsym %s %s", symname[symind], 1149 value[symind] ? value[symind] : ""); 1150 return (symind); 1151 } 1152 } 1153 return (-1); 1154} 1155 1156/* 1157 * Add a symbol to the symbol table. 1158 */ 1159static void 1160addsym(bool ignorethis, bool definethis, char *sym) 1161{ 1162 int symind; 1163 char *val; 1164 1165 symind = findsym(sym); 1166 if (symind < 0) { 1167 if (nsyms >= MAXSYMS) 1168 errx(2, "too many symbols"); 1169 symind = nsyms++; 1170 } 1171 symname[symind] = sym; 1172 ignore[symind] = ignorethis; 1173 val = sym + (skipsym(sym) - sym); 1174 if (definethis) { 1175 if (*val == '=') { 1176 value[symind] = val+1; 1177 *val = '\0'; 1178 } else if (*val == '\0') 1179 value[symind] = "1"; 1180 else 1181 usage(); 1182 } else { 1183 if (*val != '\0') 1184 usage(); 1185 value[symind] = NULL; 1186 } 1187 debug("addsym %s=%s", symname[symind], 1188 value[symind] ? value[symind] : "undef"); 1189} 1190 1191/* 1192 * Compare s with n characters of t. 1193 * The same as strncmp() except that it checks that s[n] == '\0'. 1194 */ 1195static int 1196strlcmp(const char *s, const char *t, size_t n) 1197{ 1198 while (n-- && *t != '\0') 1199 if (*s != *t) 1200 return ((unsigned char)*s - (unsigned char)*t); 1201 else 1202 ++s, ++t; 1203 return ((unsigned char)*s); 1204} 1205 1206/* 1207 * Diagnostics. 1208 */ 1209static void 1210debug(const char *msg, ...) 1211{ 1212 va_list ap; 1213 1214 if (debugging) { 1215 va_start(ap, msg); 1216 vwarnx(msg, ap); 1217 va_end(ap); 1218 } 1219} 1220 1221static void 1222error(const char *msg) 1223{ 1224 if (depth == 0) 1225 warnx("%s: %d: %s", filename, linenum, msg); 1226 else 1227 warnx("%s: %d: %s (#if line %d depth %d)", 1228 filename, linenum, msg, stifline[depth], depth); 1229 closeout(); 1230 errx(2, "output may be truncated"); 1231} 1232