busybox/editors/sed.c
<<
>>
Prefs
   1/* vi: set sw=4 ts=4: */
   2/*
   3 * sed.c - very minimalist version of sed
   4 *
   5 * Copyright (C) 1999,2000,2001 by Lineo, inc. and Mark Whitley
   6 * Copyright (C) 1999,2000,2001 by Mark Whitley <markw@codepoet.org>
   7 * Copyright (C) 2002  Matt Kraai
   8 * Copyright (C) 2003 by Glenn McGrath
   9 * Copyright (C) 2003,2004 by Rob Landley <rob@landley.net>
  10 *
  11 * MAINTAINER: Rob Landley <rob@landley.net>
  12 *
  13 * Licensed under GPLv2, see file LICENSE in this source tree.
  14 */
  15/* Code overview.
  16 *
  17 * Files are laid out to avoid unnecessary function declarations.  So for
  18 * example, every function add_cmd calls occurs before add_cmd in this file.
  19 *
  20 * add_cmd() is called on each line of sed command text (from a file or from
  21 * the command line).  It calls get_address() and parse_cmd_args().  The
  22 * resulting sed_cmd_t structures are appended to a linked list
  23 * (G.sed_cmd_head/G.sed_cmd_tail).
  24 *
  25 * process_files() does actual sedding, reading data lines from each input FILE*
  26 * (which could be stdin) and applying the sed command list (sed_cmd_head) to
  27 * each of the resulting lines.
  28 *
  29 * sed_main() is where external code calls into this, with a command line.
  30 */
  31/* Supported features and commands in this version of sed:
  32 *
  33 * - comments ('#')
  34 * - address matching: num|/matchstr/[,num|/matchstr/|$]command
  35 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
  36 * - edit commands: (a)ppend, (i)nsert, (c)hange
  37 * - file commands: (r)ead
  38 * - backreferences in substitution expressions (\0, \1, \2...\9)
  39 * - grouped commands: {cmd1;cmd2}
  40 * - transliteration (y/source-chars/dest-chars/)
  41 * - pattern space hold space storing / swapping (g, h, x)
  42 * - labels / branching (: label, b, t, T)
  43 *
  44 * (Note: Specifying an address (range) to match is *optional*; commands
  45 * default to the whole pattern space if no specific address match was
  46 * requested.)
  47 *
  48 * Todo:
  49 * - Create a wrapper around regex to make libc's regex conform with sed
  50 *
  51 * Reference
  52 * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
  53 * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html
  54 * http://sed.sourceforge.net/sedfaq3.html
  55 */
  56//config:config SED
  57//config:       bool "sed (12 kb)"
  58//config:       default y
  59//config:       help
  60//config:       sed is used to perform text transformations on a file
  61//config:       or input from a pipeline.
  62
  63//applet:IF_SED(APPLET(sed, BB_DIR_BIN, BB_SUID_DROP))
  64
  65//kbuild:lib-$(CONFIG_SED) += sed.o
  66
  67//usage:#define sed_trivial_usage
  68//usage:       "[-i[SFX]] [-nrE] [-f FILE]... [-e CMD]... [FILE]...\n"
  69//usage:       "or: sed [-i[SFX]] [-nrE] CMD [FILE]..."
  70//usage:#define sed_full_usage "\n\n"
  71//usage:       "        -e CMD  Add CMD to sed commands to be executed"
  72//usage:     "\n        -f FILE Add FILE contents to sed commands to be executed"
  73//usage:     "\n        -i[SFX] Edit files in-place (otherwise sends to stdout)"
  74//usage:     "\n                Optionally back files up, appending SFX"
  75//usage:     "\n        -n      Suppress automatic printing of pattern space"
  76//usage:     "\n        -r,-E   Use extended regex syntax"
  77//usage:     "\n"
  78//usage:     "\nIf no -e or -f, the first non-option argument is the sed command string."
  79//usage:     "\nRemaining arguments are input files (stdin if none)."
  80//usage:
  81//usage:#define sed_example_usage
  82//usage:       "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
  83//usage:       "bar\n"
  84
  85#include "libbb.h"
  86#include "common_bufsiz.h"
  87#include "xregex.h"
  88
  89#if 0
  90# define dbg(...) bb_error_msg(__VA_ARGS__)
  91#else
  92# define dbg(...) ((void)0)
  93#endif
  94
  95
  96enum {
  97        OPT_in_place = 1 << 0,
  98};
  99
 100/* Each sed command turns into one of these structures. */
 101typedef struct sed_cmd_s {
 102        /* Ordered by alignment requirements: currently 36 bytes on x86 */
 103        struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
 104
 105        /* address storage */
 106        regex_t *beg_match;     /* sed -e '/match/cmd' */
 107        regex_t *end_match;     /* sed -e '/match/,/end_match/cmd' */
 108        regex_t *sub_match;     /* For 's/sub_match/string/' */
 109        int beg_line;           /* 'sed 1p'   0 == apply commands to all lines */
 110        int beg_line_orig;      /* copy of the above, needed for -i */
 111        int end_line;           /* 'sed 1,3p' 0 == one line only. -1 = last line ($). -2-N = +N */
 112        int end_line_orig;
 113
 114        FILE *sw_file;          /* File (sw) command writes to, NULL for none. */
 115        char *string;           /* Data string for (saicytb) commands. */
 116
 117        unsigned which_match;   /* (s) Which match to replace (0 for all) */
 118
 119        /* Bitfields (gcc won't group them if we don't) */
 120        unsigned invert:1;      /* the '!' after the address */
 121        unsigned in_match:1;    /* Next line also included in match? */
 122        unsigned sub_p:1;       /* (s) print option */
 123
 124        char sw_last_char;      /* Last line written by (sw) had no '\n' */
 125
 126        /* GENERAL FIELDS */
 127        char cmd;               /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
 128} sed_cmd_t;
 129
 130static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
 131
 132struct globals {
 133        /* options */
 134        int be_quiet, regex_type;
 135
 136        FILE *nonstdout;
 137        char *outname, *hold_space;
 138        smallint exitcode;
 139
 140        /* list of input files */
 141        int current_input_file, last_input_file;
 142        char **input_file_list;
 143        FILE *current_fp;
 144
 145        regmatch_t regmatch[10];
 146        regex_t *previous_regex_ptr;
 147
 148        /* linked list of sed commands */
 149        sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
 150
 151        /* linked list of append lines */
 152        llist_t *append_head;
 153
 154        char *add_cmd_line;
 155
 156        struct pipeline {
 157                char *buf;  /* Space to hold string */
 158                int idx;    /* Space used */
 159                int len;    /* Space allocated */
 160        } pipeline;
 161} FIX_ALIASING;
 162#define G (*(struct globals*)bb_common_bufsiz1)
 163#define INIT_G() do { \
 164        setup_common_bufsiz(); \
 165        BUILD_BUG_ON(sizeof(G) > COMMON_BUFSIZE); \
 166        G.sed_cmd_tail = &G.sed_cmd_head; \
 167} while (0)
 168
 169
 170#if ENABLE_FEATURE_CLEAN_UP
 171static void sed_free_and_close_stuff(void)
 172{
 173        sed_cmd_t *sed_cmd = G.sed_cmd_head;
 174
 175        llist_free(G.append_head, free);
 176
 177        while (sed_cmd) {
 178                sed_cmd_t *sed_cmd_next = sed_cmd->next;
 179
 180                if (sed_cmd->sw_file)
 181                        fclose(sed_cmd->sw_file);
 182
 183                /* Used to free regexps, but now there is code
 184                 * in get_address() which can reuse a regexp
 185                 * for constructs as /regexp/cmd1;//cmd2
 186                 * leading to double-frees here:
 187                 */
 188                //if (sed_cmd->beg_match) {
 189                //      regfree(sed_cmd->beg_match);
 190                //      free(sed_cmd->beg_match);
 191                //}
 192                //if (sed_cmd->end_match) {
 193                //      regfree(sed_cmd->end_match);
 194                //      free(sed_cmd->end_match);
 195                //}
 196                //if (sed_cmd->sub_match) {
 197                //      regfree(sed_cmd->sub_match);
 198                //      free(sed_cmd->sub_match);
 199                //}
 200                free(sed_cmd->string);
 201                free(sed_cmd);
 202                sed_cmd = sed_cmd_next;
 203        }
 204
 205        free(G.hold_space);
 206
 207        if (G.current_fp)
 208                fclose(G.current_fp);
 209}
 210#else
 211void sed_free_and_close_stuff(void);
 212#endif
 213
 214/* If something bad happens during -i operation, delete temp file */
 215
 216static void cleanup_outname(void)
 217{
 218        if (G.outname) unlink(G.outname);
 219}
 220
 221/* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
 222
 223static unsigned parse_escapes(char *dest, const char *string, int len, char from, char to)
 224{
 225        char *d = dest;
 226        int i = 0;
 227
 228        if (len == -1)
 229                len = strlen(string);
 230
 231        while (i < len) {
 232                if (string[i] == '\\') {
 233                        if (!to || string[i+1] == from) {
 234                                if ((*d = to ? to : string[i+1]) == '\0')
 235                                        return d - dest;
 236                                i += 2;
 237                                d++;
 238                                continue;
 239                        }
 240                        i++; /* skip backslash in string[] */
 241                        *d++ = '\\';
 242                        /* fall through: copy next char verbatim */
 243                }
 244                if ((*d = string[i++]) == '\0')
 245                        return d - dest;
 246                d++;
 247        }
 248        *d = '\0';
 249        return d - dest;
 250}
 251
 252static char *copy_parsing_escapes(const char *string, int len)
 253{
 254        const char *s;
 255        char *dest = xmalloc(len + 1);
 256
 257        /* sed recognizes \n */
 258        /* GNU sed also recognizes \t and \r */
 259        for (s = "\nn\tt\rr"; *s; s += 2) {
 260                len = parse_escapes(dest, string, len, s[1], s[0]);
 261                string = dest;
 262        }
 263        return dest;
 264}
 265
 266
 267/*
 268 * index_of_next_unescaped_regexp_delim - walks left to right through a string
 269 * beginning at a specified index and returns the index of the next regular
 270 * expression delimiter (typically a forward slash ('/')) not preceded by
 271 * a backslash ('\').  A negative delimiter disables square bracket checking.
 272 */
 273static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
 274{
 275        int bracket = -1;
 276        int escaped = 0;
 277        int idx = 0;
 278        char ch;
 279
 280        if (delimiter < 0) {
 281                bracket--;
 282                delimiter = -delimiter;
 283        }
 284
 285        for (; (ch = str[idx]) != '\0'; idx++) {
 286                if (bracket >= 0) {
 287                        if (ch == ']'
 288                         && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
 289                        ) {
 290                                bracket = -1;
 291                        }
 292                } else if (escaped)
 293                        escaped = 0;
 294                else if (ch == '\\')
 295                        escaped = 1;
 296                else if (bracket == -1 && ch == '[')
 297                        bracket = idx;
 298                else if (ch == delimiter)
 299                        return idx;
 300        }
 301
 302        /* if we make it to here, we've hit the end of the string */
 303        bb_error_msg_and_die("unmatched '%c'", delimiter);
 304}
 305
 306/*
 307 *  Returns the index of the third delimiter
 308 */
 309static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
 310{
 311        const char *cmdstr_ptr = cmdstr;
 312        unsigned char delimiter;
 313        int idx = 0;
 314
 315        /* verify that the 's' or 'y' is followed by something.  That something
 316         * (typically a 'slash') is now our regexp delimiter... */
 317        if (*cmdstr == '\0')
 318                bb_error_msg_and_die("bad format in substitution expression");
 319        delimiter = *cmdstr_ptr++;
 320
 321        /* save the match string */
 322        idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
 323        *match = copy_parsing_escapes(cmdstr_ptr, idx);
 324
 325        /* save the replacement string */
 326        cmdstr_ptr += idx + 1;
 327        idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr);
 328        *replace = copy_parsing_escapes(cmdstr_ptr, idx);
 329
 330        return ((cmdstr_ptr - cmdstr) + idx);
 331}
 332
 333/*
 334 * returns the index in the string just past where the address ends.
 335 */
 336static int get_address(const char *my_str, int *linenum, regex_t ** regex)
 337{
 338        const char *pos = my_str;
 339
 340        if (isdigit(*my_str)) {
 341                *linenum = strtol(my_str, (char**)&pos, 10);
 342                /* endstr shouldn't ever equal NULL */
 343        } else if (*my_str == '$') {
 344                *linenum = -1;
 345                pos++;
 346        } else if (*my_str == '/' || *my_str == '\\') {
 347                int next;
 348                char delimiter;
 349                char *temp;
 350
 351                delimiter = '/';
 352                if (*my_str == '\\')
 353                        delimiter = *++pos;
 354                next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
 355                if (next != 0) {
 356                        temp = copy_parsing_escapes(pos, next);
 357                        G.previous_regex_ptr = *regex = xzalloc(sizeof(regex_t));
 358                        xregcomp(*regex, temp, G.regex_type);
 359                        free(temp);
 360                } else {
 361                        *regex = G.previous_regex_ptr;
 362                        if (!G.previous_regex_ptr)
 363                                bb_error_msg_and_die("no previous regexp");
 364                }
 365                /* Move position to next character after last delimiter */
 366                pos += (next+1);
 367        }
 368        return pos - my_str;
 369}
 370
 371/* Grab a filename.  Whitespace at start is skipped, then goes to EOL. */
 372static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
 373{
 374        const char *start;
 375        const char *eol;
 376
 377        /* Skip whitespace, then grab filename to end of line */
 378        start = skip_whitespace(filecmdstr);
 379        eol = strchrnul(start, '\n');
 380        if (eol == start)
 381                bb_error_msg_and_die("empty filename");
 382
 383        if (*eol) {
 384                /* If lines glued together, put backslash back. */
 385                *retval = xstrndup(start, eol-start + 1);
 386                (*retval)[eol-start] = '\\';
 387        } else {
 388                /* eol is NUL */
 389                *retval = xstrdup(start);
 390        }
 391
 392        return eol - filecmdstr;
 393}
 394
 395static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
 396{
 397        int cflags = G.regex_type;
 398        char *match;
 399        int idx;
 400
 401        /*
 402         * A substitution command should look something like this:
 403         *    s/match/replace/ #giIpw
 404         *    ||     |        |||
 405         *    mandatory       optional
 406         */
 407        idx = parse_regex_delim(substr, &match, &sed_cmd->string);
 408
 409        /* determine the number of back references in the match string */
 410        /* Note: we compute this here rather than in the do_subst_command()
 411         * function to save processor time, at the expense of a little more memory
 412         * (4 bits) per sed_cmd */
 413
 414        /* process the flags */
 415
 416        sed_cmd->which_match = 1;
 417        dbg("s flags:'%s'", substr + idx + 1);
 418        while (substr[++idx]) {
 419                dbg("s flag:'%c'", substr[idx]);
 420                /* Parse match number */
 421                if (isdigit(substr[idx])) {
 422                        if (match[0] != '^') {
 423                                /* Match 0 treated as all, multiple matches we take the last one. */
 424                                const char *pos = substr + idx;
 425/* FIXME: error check? */
 426                                sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
 427                                idx = pos - substr - 1;
 428                        }
 429                        continue;
 430                }
 431                /* Skip spaces */
 432                if (isspace(substr[idx]))
 433                        continue;
 434
 435                switch (substr[idx]) {
 436                /* Replace all occurrences */
 437                case 'g':
 438                        if (match[0] != '^')
 439                                sed_cmd->which_match = 0;
 440                        break;
 441                /* Print pattern space */
 442                case 'p':
 443                        sed_cmd->sub_p = 1;
 444                        break;
 445                /* Write to file */
 446                case 'w':
 447                {
 448                        char *fname;
 449                        idx += parse_file_cmd(/*sed_cmd,*/ substr+idx+1, &fname);
 450                        sed_cmd->sw_file = xfopen_for_write(fname);
 451                        sed_cmd->sw_last_char = '\n';
 452                        free(fname);
 453                        break;
 454                }
 455                /* Ignore case (gnu extension) */
 456                case 'i':
 457                case 'I':
 458                        cflags |= REG_ICASE;
 459                        break;
 460                /* Comment */
 461                case '#':
 462                        // while (substr[++idx]) continue;
 463                        idx += strlen(substr + idx); // same
 464                        /* Fall through */
 465                /* End of command */
 466                case ';':
 467                case '}':
 468                        goto out;
 469                default:
 470                        dbg("s bad flags:'%s'", substr + idx);
 471                        bb_error_msg_and_die("bad option in substitution expression");
 472                }
 473        }
 474 out:
 475        /* compile the match string into a regex */
 476        if (*match != '\0') {
 477                /* If match is empty, we use last regex used at runtime */
 478                sed_cmd->sub_match = xzalloc(sizeof(regex_t));
 479                dbg("xregcomp('%s',%x)", match, cflags);
 480                xregcomp(sed_cmd->sub_match, match, cflags);
 481                dbg("regcomp ok");
 482        }
 483        free(match);
 484
 485        return idx;
 486}
 487
 488/*
 489 *  Process the commands arguments
 490 */
 491static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
 492{
 493        static const char cmd_letters[] ALIGN1 = "saicrw:btTydDgGhHlnNpPqx={}";
 494        enum {
 495                IDX_s = 0,
 496                IDX_a,
 497                IDX_i,
 498                IDX_c,
 499                IDX_r,
 500                IDX_w,
 501                IDX_colon,
 502                IDX_b,
 503                IDX_t,
 504                IDX_T,
 505                IDX_y,
 506                IDX_d,
 507                IDX_D,
 508                IDX_g,
 509                IDX_G,
 510                IDX_h,
 511                IDX_H,
 512                IDX_l,
 513                IDX_n,
 514                IDX_N,
 515                IDX_p,
 516                IDX_P,
 517                IDX_q,
 518                IDX_x,
 519                IDX_equal,
 520                IDX_lbrace,
 521                IDX_rbrace,
 522                IDX_nul
 523        };
 524        unsigned idx;
 525
 526        BUILD_BUG_ON(sizeof(cmd_letters)-1 != IDX_nul);
 527
 528        idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
 529
 530        /* handle (s)ubstitution command */
 531        if (idx == IDX_s) {
 532                cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
 533        }
 534        /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
 535        else if (idx <= IDX_c) { /* a,i,c */
 536                unsigned len;
 537
 538                if (idx < IDX_c) { /* a,i */
 539                        if (sed_cmd->end_line || sed_cmd->end_match)
 540                                bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
 541                }
 542                for (;;) {
 543                        if (*cmdstr == '\n' || *cmdstr == '\\') {
 544                                cmdstr++;
 545                                break;
 546                        }
 547                        if (!isspace(*cmdstr))
 548                                break;
 549                        cmdstr++;
 550                }
 551                len = strlen(cmdstr);
 552                sed_cmd->string = copy_parsing_escapes(cmdstr, len);
 553                cmdstr += len;
 554                /* "\anychar" -> "anychar" */
 555                parse_escapes(sed_cmd->string, sed_cmd->string, -1, '\0', '\0');
 556        }
 557        /* handle file cmds: (r)ead */
 558        else if (idx <= IDX_w) { /* r,w */
 559                if (idx < IDX_w) { /* r */
 560                        if (sed_cmd->end_line || sed_cmd->end_match)
 561                                bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
 562                }
 563                cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
 564                if (sed_cmd->cmd == 'w') {
 565                        sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
 566                        sed_cmd->sw_last_char = '\n';
 567                }
 568        }
 569        /* handle branch commands */
 570        else if (idx <= IDX_T) { /* :,b,t,T */
 571                int length;
 572
 573                cmdstr = skip_whitespace(cmdstr);
 574                length = strcspn(cmdstr, semicolon_whitespace);
 575                if (length) {
 576                        sed_cmd->string = xstrndup(cmdstr, length);
 577                        cmdstr += length;
 578                }
 579        }
 580        /* translation command */
 581        else if (idx == IDX_y) {
 582                char *match, *replace;
 583                int i = cmdstr[0];
 584
 585                cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
 586                /* \n already parsed, but \delimiter needs unescaping. */
 587                parse_escapes(match,   match,   -1, i, i);
 588                parse_escapes(replace, replace, -1, i, i);
 589
 590                sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
 591                for (i = 0; match[i] && replace[i]; i++) {
 592                        sed_cmd->string[i*2] = match[i];
 593                        sed_cmd->string[i*2+1] = replace[i];
 594                }
 595                free(match);
 596                free(replace);
 597        }
 598        /* if it wasn't a single-letter command that takes no arguments
 599         * then it must be an invalid command.
 600         */
 601        else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
 602                bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
 603        }
 604
 605        /* give back whatever's left over */
 606        return cmdstr;
 607}
 608
 609
 610/* Parse address+command sets, skipping comment lines. */
 611
 612static void add_cmd(const char *cmdstr)
 613{
 614        sed_cmd_t *sed_cmd;
 615        unsigned len, n;
 616
 617        /* Append this line to any unfinished line from last time. */
 618        if (G.add_cmd_line) {
 619                char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
 620                free(G.add_cmd_line);
 621                cmdstr = G.add_cmd_line = tp;
 622        }
 623
 624        /* If this line ends with unescaped backslash, request next line. */
 625        n = len = strlen(cmdstr);
 626        while (n && cmdstr[n-1] == '\\')
 627                n--;
 628        if ((len - n) & 1) { /* if odd number of trailing backslashes */
 629                if (!G.add_cmd_line)
 630                        G.add_cmd_line = xstrdup(cmdstr);
 631                G.add_cmd_line[len-1] = '\0';
 632                return;
 633        }
 634
 635        /* Loop parsing all commands in this line. */
 636        while (*cmdstr) {
 637                /* Skip leading whitespace and semicolons */
 638                cmdstr += strspn(cmdstr, semicolon_whitespace);
 639
 640                /* If no more commands, exit. */
 641                if (!*cmdstr) break;
 642
 643                /* if this is a comment, jump past it and keep going */
 644                if (*cmdstr == '#') {
 645                        /* "#n" is the same as using -n on the command line */
 646                        if (cmdstr[1] == 'n')
 647                                G.be_quiet++;
 648                        cmdstr = strpbrk(cmdstr, "\n\r");
 649                        if (!cmdstr) break;
 650                        continue;
 651                }
 652
 653                /* parse the command
 654                 * format is: [addr][,addr][!]cmd
 655                 *            |----||-----||-|
 656                 *            part1 part2  part3
 657                 */
 658
 659                sed_cmd = xzalloc(sizeof(sed_cmd_t));
 660
 661                /* first part (if present) is an address: either a '$', a number or a /regex/ */
 662                cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
 663                sed_cmd->beg_line_orig = sed_cmd->beg_line;
 664
 665                /* second part (if present) will begin with a comma */
 666                if (*cmdstr == ',') {
 667                        int idx;
 668
 669                        cmdstr++;
 670                        if (*cmdstr == '+' && isdigit(cmdstr[1])) {
 671                                /* http://sed.sourceforge.net/sedfaq3.html#s3.3
 672                                 * Under GNU sed 3.02+, ssed, and sed15+, <address2>
 673                                 * may also be a notation of the form +num,
 674                                 * indicating the next num lines after <address1> is
 675                                 * matched.
 676                                 * GNU sed 4.2.1 accepts even "+" (meaning "+0").
 677                                 * We don't (we check for isdigit, see above), think
 678                                 * about the "+-3" case.
 679                                 */
 680                                char *end;
 681                                /* code is smaller compared to using &cmdstr here: */
 682                                idx = strtol(cmdstr+1, &end, 10);
 683                                sed_cmd->end_line = -2 - idx;
 684                                cmdstr = end;
 685                        } else {
 686                                idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
 687                                cmdstr += idx;
 688                                idx--; /* if 0, trigger error check below */
 689                        }
 690                        if (idx < 0)
 691                                bb_error_msg_and_die("no address after comma");
 692                        sed_cmd->end_line_orig = sed_cmd->end_line;
 693                }
 694
 695                /* skip whitespace before the command */
 696                cmdstr = skip_whitespace(cmdstr);
 697
 698                /* Check for inversion flag */
 699                if (*cmdstr == '!') {
 700                        sed_cmd->invert = 1;
 701                        cmdstr++;
 702
 703                        /* skip whitespace before the command */
 704                        cmdstr = skip_whitespace(cmdstr);
 705                }
 706
 707                /* last part (mandatory) will be a command */
 708                if (!*cmdstr)
 709                        bb_error_msg_and_die("missing command");
 710                sed_cmd->cmd = *cmdstr++;
 711                cmdstr = parse_cmd_args(sed_cmd, cmdstr);
 712
 713                /* cmdstr now points past args.
 714                 * GNU sed requires a separator, if there are more commands,
 715                 * else it complains "char N: extra characters after command".
 716                 * Example: "sed 'p;d'". We also allow "sed 'pd'".
 717                 */
 718
 719                /* Add the command to the command array */
 720                *G.sed_cmd_tail = sed_cmd;
 721                G.sed_cmd_tail = &sed_cmd->next;
 722        }
 723
 724        /* If we glued multiple lines together, free the memory. */
 725        free(G.add_cmd_line);
 726        G.add_cmd_line = NULL;
 727}
 728
 729/* Append to a string, reallocating memory as necessary. */
 730
 731#define PIPE_GROW 64
 732
 733static void pipe_putc(char c)
 734{
 735        if (G.pipeline.idx == G.pipeline.len) {
 736                G.pipeline.buf = xrealloc(G.pipeline.buf,
 737                                G.pipeline.len + PIPE_GROW);
 738                G.pipeline.len += PIPE_GROW;
 739        }
 740        G.pipeline.buf[G.pipeline.idx++] = c;
 741}
 742
 743static void do_subst_w_backrefs(char *line, char *replace)
 744{
 745        int i, j;
 746
 747        /* go through the replacement string */
 748        for (i = 0; replace[i]; i++) {
 749                /* if we find a backreference (\1, \2, etc.) print the backref'ed text */
 750                if (replace[i] == '\\') {
 751                        unsigned backref = replace[++i] - '0';
 752                        if (backref <= 9) {
 753                                /* print out the text held in G.regmatch[backref] */
 754                                if (G.regmatch[backref].rm_so != -1) {
 755                                        j = G.regmatch[backref].rm_so;
 756                                        while (j < G.regmatch[backref].rm_eo)
 757                                                pipe_putc(line[j++]);
 758                                }
 759                                continue;
 760                        }
 761                        /* I _think_ it is impossible to get '\' to be
 762                         * the last char in replace string. Thus we don't check
 763                         * for replace[i] == NUL. (counterexample anyone?) */
 764                        /* if we find a backslash escaped character, print the character */
 765                        pipe_putc(replace[i]);
 766                        continue;
 767                }
 768                /* if we find an unescaped '&' print out the whole matched text. */
 769                if (replace[i] == '&') {
 770                        j = G.regmatch[0].rm_so;
 771                        while (j < G.regmatch[0].rm_eo)
 772                                pipe_putc(line[j++]);
 773                        continue;
 774                }
 775                /* Otherwise just output the character. */
 776                pipe_putc(replace[i]);
 777        }
 778}
 779
 780static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
 781{
 782        char *line = *line_p;
 783        unsigned match_count = 0;
 784        bool altered = 0;
 785        bool prev_match_empty = 1;
 786        bool tried_at_eol = 0;
 787        regex_t *current_regex;
 788
 789        current_regex = sed_cmd->sub_match;
 790        /* Handle empty regex. */
 791        if (!current_regex) {
 792                current_regex = G.previous_regex_ptr;
 793                if (!current_regex)
 794                        bb_error_msg_and_die("no previous regexp");
 795        }
 796        G.previous_regex_ptr = current_regex;
 797
 798        /* Find the first match */
 799        dbg("matching '%s'", line);
 800        if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) {
 801                dbg("no match");
 802                return 0;
 803        }
 804        dbg("match");
 805
 806        /* Initialize temporary output buffer. */
 807        G.pipeline.buf = xmalloc(PIPE_GROW);
 808        G.pipeline.len = PIPE_GROW;
 809        G.pipeline.idx = 0;
 810
 811        /* Now loop through, substituting for matches */
 812        do {
 813                int start = G.regmatch[0].rm_so;
 814                int end = G.regmatch[0].rm_eo;
 815                int i;
 816
 817                match_count++;
 818
 819                /* If we aren't interested in this match, output old line to
 820                 * end of match and continue */
 821                if (sed_cmd->which_match
 822                 && (sed_cmd->which_match != match_count)
 823                ) {
 824                        for (i = 0; i < end; i++)
 825                                pipe_putc(*line++);
 826                        /* Null match? Print one more char */
 827                        if (start == end && *line)
 828                                pipe_putc(*line++);
 829                        goto next;
 830                }
 831
 832                /* Print everything before the match */
 833                for (i = 0; i < start; i++)
 834                        pipe_putc(line[i]);
 835
 836                /* Then print the substitution string,
 837                 * unless we just matched empty string after non-empty one.
 838                 * Example: string "cccd", pattern "c*", repl "R":
 839                 * result is "RdR", not "RRdR": first match "ccc",
 840                 * second is "" before "d", third is "" after "d".
 841                 * Second match is NOT replaced!
 842                 */
 843                if (prev_match_empty || start != 0 || start != end) {
 844                        //dbg("%d %d %d", prev_match_empty, start, end);
 845                        dbg("inserting replacement at %d in '%s'", start, line);
 846                        do_subst_w_backrefs(line, sed_cmd->string);
 847                        /* Flag that something has changed */
 848                        altered = 1;
 849                } else {
 850                        dbg("NOT inserting replacement at %d in '%s'", start, line);
 851                }
 852
 853                /* If matched string is empty (f.e. "c*" pattern),
 854                 * copy verbatim one char after it before attempting more matches
 855                 */
 856                prev_match_empty = (start == end);
 857                if (prev_match_empty) {
 858                        if (!line[end]) {
 859                                tried_at_eol = 1;
 860                        } else {
 861                                pipe_putc(line[end]);
 862                                end++;
 863                        }
 864                }
 865
 866                /* Advance past the match */
 867                dbg("line += %d", end);
 868                line += end;
 869
 870                /* if we're not doing this globally, get out now */
 871                if (sed_cmd->which_match != 0)
 872                        break;
 873 next:
 874                /* Exit if we are at EOL and already tried matching at it */
 875                if (*line == '\0') {
 876                        if (tried_at_eol)
 877                                break;
 878                        tried_at_eol = 1;
 879                }
 880
 881//maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
 882        } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
 883
 884        /* Copy rest of string into output pipeline */
 885        while (1) {
 886                char c = *line++;
 887                pipe_putc(c);
 888                if (c == '\0')
 889                        break;
 890        }
 891
 892        free(*line_p);
 893        *line_p = G.pipeline.buf;
 894        return altered;
 895}
 896
 897/* Set command pointer to point to this label.  (Does not handle null label.) */
 898static sed_cmd_t *branch_to(char *label)
 899{
 900        sed_cmd_t *sed_cmd;
 901
 902        for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
 903                if (sed_cmd->cmd == ':'
 904                 && sed_cmd->string
 905                 && strcmp(sed_cmd->string, label) == 0
 906                ) {
 907                        return sed_cmd;
 908                }
 909        }
 910        bb_error_msg_and_die("can't find label for jump to '%s'", label);
 911}
 912
 913static void append(char *s)
 914{
 915        llist_add_to_end(&G.append_head, s);
 916}
 917
 918/* Output line of text. */
 919/* Note:
 920 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
 921 * Without them, we had this:
 922 * echo -n thingy >z1
 923 * echo -n again >z2
 924 * >znull
 925 * sed "s/i/z/" z1 z2 znull | hexdump -vC
 926 * output:
 927 * gnu sed 4.1.5:
 928 * 00000000  74 68 7a 6e 67 79 0a 61  67 61 7a 6e              |thzngy.agazn|
 929 * bbox:
 930 * 00000000  74 68 7a 6e 67 79 61 67  61 7a 6e                 |thzngyagazn|
 931 */
 932enum {
 933        NO_EOL_CHAR = 1,
 934        LAST_IS_NUL = 2,
 935};
 936static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
 937{
 938        char lpc = *last_puts_char;
 939
 940        /* Need to insert a '\n' between two files because first file's
 941         * last line wasn't terminated? */
 942        if (lpc != '\n' && lpc != '\0') {
 943                fputc('\n', file);
 944                lpc = '\n';
 945        }
 946        fputs(s, file);
 947
 948        /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
 949        if (s[0])
 950                lpc = 'x';
 951
 952        /* had trailing '\0' and it was last char of file? */
 953        if (last_gets_char == LAST_IS_NUL) {
 954                fputc('\0', file);
 955                lpc = 'x'; /* */
 956        } else
 957        /* had trailing '\n' or '\0'? */
 958        if (last_gets_char != NO_EOL_CHAR) {
 959                fputc(last_gets_char, file);
 960                lpc = last_gets_char;
 961        }
 962
 963        if (ferror(file)) {
 964                xfunc_error_retval = 4;  /* It's what gnu sed exits with... */
 965                bb_error_msg_and_die(bb_msg_write_error);
 966        }
 967        *last_puts_char = lpc;
 968}
 969
 970static void flush_append(char *last_puts_char)
 971{
 972        char *data;
 973
 974        /* Output appended lines. */
 975        while ((data = (char *)llist_pop(&G.append_head)) != NULL) {
 976                /* Append command does not respect "nonterminated-ness"
 977                 * of last line. Try this:
 978                 * $ echo -n "woot" | sed -e '/woot/a woo' -
 979                 * woot
 980                 * woo
 981                 * (both lines are terminated with \n)
 982                 * Therefore we do not propagate "last_gets_char" here,
 983                 * pass '\n' instead:
 984                 */
 985                puts_maybe_newline(data, G.nonstdout, last_puts_char, '\n');
 986                free(data);
 987        }
 988}
 989
 990/* Get next line of input from G.input_file_list, flushing append buffer and
 991 * noting if we ran out of files without a newline on the last line we read.
 992 */
 993static char *get_next_line(char *gets_char, char *last_puts_char)
 994{
 995        char *temp = NULL;
 996        size_t len;
 997        char gc;
 998
 999        flush_append(last_puts_char);
1000
1001        /* will be returned if last line in the file
1002         * doesn't end with either '\n' or '\0' */
1003        gc = NO_EOL_CHAR;
1004        for (; G.current_input_file <= G.last_input_file; G.current_input_file++) {
1005                FILE *fp = G.current_fp;
1006                if (!fp) {
1007                        const char *path = G.input_file_list[G.current_input_file];
1008                        fp = stdin;
1009                        if (path != bb_msg_standard_input) {
1010                                fp = fopen_or_warn(path, "r");
1011                                if (!fp) {
1012                                        G.exitcode = EXIT_FAILURE;
1013                                        continue;
1014                                }
1015                        }
1016                        G.current_fp = fp;
1017                }
1018                /* Read line up to a newline or NUL byte, inclusive,
1019                 * return malloc'ed char[]. length of the chunk read
1020                 * is stored in len. NULL if EOF/error */
1021                temp = bb_get_chunk_from_file(fp, &len);
1022                if (temp) {
1023                        /* len > 0 here, it's ok to do temp[len-1] */
1024                        char c = temp[len-1];
1025                        if (c == '\n' || c == '\0') {
1026                                temp[len-1] = '\0';
1027                                gc = c;
1028                                if (c == '\0') {
1029                                        int ch = fgetc(fp);
1030                                        if (ch != EOF)
1031                                                ungetc(ch, fp);
1032                                        else
1033                                                gc = LAST_IS_NUL;
1034                                }
1035                        }
1036                        /* else we put NO_EOL_CHAR into *gets_char */
1037                        break;
1038
1039                /* NB: I had the idea of peeking next file(s) and returning
1040                 * NO_EOL_CHAR only if it is the *last* non-empty
1041                 * input file. But there is a case where this won't work:
1042                 * file1: "a woo\nb woo"
1043                 * file2: "c no\nd no"
1044                 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
1045                 * (note: *no* newline after "b bang"!) */
1046                }
1047                /* Close this file and advance to next one */
1048                fclose_if_not_stdin(fp);
1049                G.current_fp = NULL;
1050        }
1051        *gets_char = gc;
1052        return temp;
1053}
1054
1055#define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
1056
1057static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
1058{
1059        int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
1060        if (retval)
1061                G.previous_regex_ptr = sed_cmd->beg_match;
1062        return retval;
1063}
1064
1065/* Process all the lines in all the files */
1066
1067static void process_files(void)
1068{
1069        char *pattern_space, *next_line;
1070        int linenum = 0;
1071        char last_puts_char = '\n';
1072        char last_gets_char, next_gets_char;
1073        sed_cmd_t *sed_cmd;
1074        int substituted;
1075
1076        /* Prime the pump */
1077        next_line = get_next_line(&next_gets_char, &last_puts_char);
1078
1079        /* Go through every line in each file */
1080 again:
1081        substituted = 0;
1082
1083        /* Advance to next line.  Stop if out of lines. */
1084        pattern_space = next_line;
1085        if (!pattern_space)
1086                return;
1087        last_gets_char = next_gets_char;
1088
1089        /* Read one line in advance so we can act on the last line,
1090         * the '$' address */
1091        next_line = get_next_line(&next_gets_char, &last_puts_char);
1092        linenum++;
1093
1094        /* For every line, go through all the commands */
1095 restart:
1096        for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1097                int old_matched, matched;
1098
1099                old_matched = sed_cmd->in_match;
1100                if (!old_matched)
1101                        sed_cmd->end_line = sed_cmd->end_line_orig;
1102
1103                /* Determine if this command matches this line: */
1104
1105                dbg("match1:%d", sed_cmd->in_match);
1106                dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
1107                                && !sed_cmd->beg_match && !sed_cmd->end_match));
1108                dbg("match3:%d", (sed_cmd->beg_line > 0
1109                        && (sed_cmd->end_line || sed_cmd->end_match
1110                            ? (sed_cmd->beg_line <= linenum)
1111                            : (sed_cmd->beg_line == linenum)
1112                            )
1113                        ));
1114                dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
1115                dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
1116
1117                /* Are we continuing a previous multi-line match? */
1118                sed_cmd->in_match = sed_cmd->in_match
1119                        /* Or is no range necessary? */
1120                        || (!sed_cmd->beg_line && !sed_cmd->end_line
1121                                && !sed_cmd->beg_match && !sed_cmd->end_match)
1122                        /* Or did we match the start of a numerical range? */
1123                        || (sed_cmd->beg_line > 0
1124                            && (sed_cmd->end_line || sed_cmd->end_match
1125                                  /* note: even if end is numeric and is < linenum too,
1126                                   * GNU sed matches! We match too, therefore we don't
1127                                   * check here that linenum <= end.
1128                                   * Example:
1129                                   * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1130                                   * first three input lines are deleted;
1131                                   * 4th line is matched and printed
1132                                   * by "2,3" (!) and by "4" ranges
1133                                   */
1134                                ? (sed_cmd->beg_line <= linenum)    /* N,end */
1135                                : (sed_cmd->beg_line == linenum)    /* N */
1136                                )
1137                            )
1138                        /* Or does this line match our begin address regex? */
1139                        || (beg_match(sed_cmd, pattern_space))
1140                        /* Or did we match last line of input? */
1141                        || (sed_cmd->beg_line == -1 && next_line == NULL);
1142
1143                /* Snapshot the value */
1144                matched = sed_cmd->in_match;
1145
1146                dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1147                        sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1148
1149                /* Is this line the end of the current match? */
1150
1151                if (matched) {
1152                        if (sed_cmd->end_line <= -2) {
1153                                /* address2 is +N, i.e. N lines from beg_line */
1154                                sed_cmd->end_line = linenum + (-sed_cmd->end_line - 2);
1155                        }
1156                        /* once matched, "n,xxx" range is dead, disabling it */
1157                        if (sed_cmd->beg_line > 0) {
1158                                sed_cmd->beg_line = -2;
1159                        }
1160                        dbg("end1:%d", sed_cmd->end_line ? sed_cmd->end_line == -1
1161                                                ? !next_line : (sed_cmd->end_line <= linenum)
1162                                        : !sed_cmd->end_match);
1163                        dbg("end2:%d", sed_cmd->end_match && old_matched
1164                                        && !regexec(sed_cmd->end_match,pattern_space, 0, NULL, 0));
1165                        sed_cmd->in_match = !(
1166                                /* has the ending line come, or is this a single address command? */
1167                                (sed_cmd->end_line
1168                                        ? sed_cmd->end_line == -1
1169                                                ? !next_line
1170                                                : (sed_cmd->end_line <= linenum)
1171                                        : !sed_cmd->end_match
1172                                )
1173                                /* or does this line matches our last address regex */
1174                                || (sed_cmd->end_match && old_matched
1175                                     && (regexec(sed_cmd->end_match,
1176                                                pattern_space, 0, NULL, 0) == 0)
1177                                )
1178                        );
1179                }
1180
1181                /* Skip blocks of commands we didn't match */
1182                if (sed_cmd->cmd == '{') {
1183                        if (sed_cmd->invert ? matched : !matched) {
1184                                unsigned nest_cnt = 0;
1185                                while (1) {
1186                                        if (sed_cmd->cmd == '{')
1187                                                nest_cnt++;
1188                                        if (sed_cmd->cmd == '}') {
1189                                                nest_cnt--;
1190                                                if (nest_cnt == 0)
1191                                                        break;
1192                                        }
1193                                        sed_cmd = sed_cmd->next;
1194                                        if (!sed_cmd)
1195                                                bb_error_msg_and_die("unterminated {");
1196                                }
1197                        }
1198                        continue;
1199                }
1200
1201                /* Okay, so did this line match? */
1202                if (sed_cmd->invert ? matched : !matched)
1203                        continue; /* no */
1204
1205                /* Update last used regex in case a blank substitute BRE is found */
1206                if (sed_cmd->beg_match) {
1207                        G.previous_regex_ptr = sed_cmd->beg_match;
1208                }
1209
1210                /* actual sedding */
1211                dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1212                                pattern_space, next_line, sed_cmd->cmd);
1213                switch (sed_cmd->cmd) {
1214
1215                /* Print line number */
1216                case '=':
1217                        fprintf(G.nonstdout, "%d\n", linenum);
1218                        break;
1219
1220                /* Write the current pattern space up to the first newline */
1221                case 'P':
1222                {
1223                        char *tmp = strchr(pattern_space, '\n');
1224                        if (tmp) {
1225                                *tmp = '\0';
1226                                /* TODO: explain why '\n' below */
1227                                sed_puts(pattern_space, '\n');
1228                                *tmp = '\n';
1229                                break;
1230                        }
1231                        /* Fall Through */
1232                }
1233
1234                /* Write the current pattern space to output */
1235                case 'p':
1236                        /* NB: we print this _before_ the last line
1237                         * (of current file) is printed. Even if
1238                         * that line is nonterminated, we print
1239                         * '\n' here (gnu sed does the same) */
1240                        sed_puts(pattern_space, '\n');
1241                        break;
1242                /* Delete up through first newline */
1243                case 'D':
1244                {
1245                        char *tmp = strchr(pattern_space, '\n');
1246                        if (tmp) {
1247                                overlapping_strcpy(pattern_space, tmp + 1);
1248                                goto restart;
1249                        }
1250                }
1251                /* discard this line. */
1252                case 'd':
1253                        goto discard_line;
1254
1255                /* Substitute with regex */
1256                case 's':
1257                        if (!do_subst_command(sed_cmd, &pattern_space))
1258                                break;
1259                        dbg("do_subst_command succeeded:'%s'", pattern_space);
1260                        substituted |= 1;
1261
1262                        /* handle p option */
1263                        if (sed_cmd->sub_p)
1264                                sed_puts(pattern_space, last_gets_char);
1265                        /* handle w option */
1266                        if (sed_cmd->sw_file)
1267                                puts_maybe_newline(
1268                                        pattern_space, sed_cmd->sw_file,
1269                                        &sed_cmd->sw_last_char, last_gets_char);
1270                        break;
1271
1272                /* Append line to linked list to be printed later */
1273                case 'a':
1274                        append(xstrdup(sed_cmd->string));
1275                        break;
1276
1277                /* Insert text before this line */
1278                case 'i':
1279                        sed_puts(sed_cmd->string, '\n');
1280                        break;
1281
1282                /* Cut and paste text (replace) */
1283                case 'c':
1284                        /* Only triggers on last line of a matching range. */
1285                        if (!sed_cmd->in_match)
1286                                sed_puts(sed_cmd->string, '\n');
1287                        goto discard_line;
1288
1289                /* Read file, append contents to output */
1290                case 'r':
1291                {
1292                        FILE *rfile;
1293                        rfile = fopen_for_read(sed_cmd->string);
1294                        if (rfile) {
1295                                char *line;
1296                                while ((line = xmalloc_fgetline(rfile))
1297                                                != NULL)
1298                                        append(line);
1299                                fclose(rfile);
1300                        }
1301
1302                        break;
1303                }
1304
1305                /* Write pattern space to file. */
1306                case 'w':
1307                        puts_maybe_newline(
1308                                pattern_space, sed_cmd->sw_file,
1309                                &sed_cmd->sw_last_char, last_gets_char);
1310                        break;
1311
1312                /* Read next line from input */
1313                case 'n':
1314                        if (!G.be_quiet)
1315                                sed_puts(pattern_space, last_gets_char);
1316                        if (next_line == NULL) {
1317                                /* If no next line, jump to end of script and exit. */
1318                                goto discard_line;
1319                        }
1320                        free(pattern_space);
1321                        pattern_space = next_line;
1322                        last_gets_char = next_gets_char;
1323                        next_line = get_next_line(&next_gets_char, &last_puts_char);
1324                        substituted = 0;
1325                        linenum++;
1326                        break;
1327
1328                /* Quit.  End of script, end of input. */
1329                case 'q':
1330                        /* Exit the outer while loop */
1331                        free(next_line);
1332                        next_line = NULL;
1333                        goto discard_commands;
1334
1335                /* Append the next line to the current line */
1336                case 'N':
1337                {
1338                        int len;
1339                        /* If no next line, jump to end of script and exit. */
1340                        /* http://www.gnu.org/software/sed/manual/sed.html:
1341                         * "Most versions of sed exit without printing anything
1342                         * when the N command is issued on the last line of
1343                         * a file. GNU sed prints pattern space before exiting
1344                         * unless of course the -n command switch has been
1345                         * specified. This choice is by design."
1346                         */
1347                        if (next_line == NULL) {
1348                                //goto discard_line;
1349                                goto discard_commands; /* GNU behavior */
1350                        }
1351                        /* Append next_line, read new next_line. */
1352                        len = strlen(pattern_space);
1353                        pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1354                        pattern_space[len] = '\n';
1355                        strcpy(pattern_space + len+1, next_line);
1356                        last_gets_char = next_gets_char;
1357                        next_line = get_next_line(&next_gets_char, &last_puts_char);
1358                        linenum++;
1359                        break;
1360                }
1361
1362                /* Test/branch if substitution occurred */
1363                case 't':
1364                        if (!substituted) break;
1365                        substituted = 0;
1366                        /* Fall through */
1367                /* Test/branch if substitution didn't occur */
1368                case 'T':
1369                        if (substituted) break;
1370                        /* Fall through */
1371                /* Branch to label */
1372                case 'b':
1373                        if (!sed_cmd->string) goto discard_commands;
1374                        else sed_cmd = branch_to(sed_cmd->string);
1375                        break;
1376                /* Transliterate characters */
1377                case 'y':
1378                {
1379                        int i, j;
1380                        for (i = 0; pattern_space[i]; i++) {
1381                                for (j = 0; sed_cmd->string[j]; j += 2) {
1382                                        if (pattern_space[i] == sed_cmd->string[j]) {
1383                                                pattern_space[i] = sed_cmd->string[j + 1];
1384                                                break;
1385                                        }
1386                                }
1387                        }
1388
1389                        break;
1390                }
1391                case 'g':       /* Replace pattern space with hold space */
1392                        free(pattern_space);
1393                        pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1394                        break;
1395                case 'G':       /* Append newline and hold space to pattern space */
1396                {
1397                        int pattern_space_size = 2;
1398                        int hold_space_size = 0;
1399
1400                        if (pattern_space)
1401                                pattern_space_size += strlen(pattern_space);
1402                        if (G.hold_space)
1403                                hold_space_size = strlen(G.hold_space);
1404                        pattern_space = xrealloc(pattern_space,
1405                                        pattern_space_size + hold_space_size);
1406                        if (pattern_space_size == 2)
1407                                pattern_space[0] = 0;
1408                        strcat(pattern_space, "\n");
1409                        if (G.hold_space)
1410                                strcat(pattern_space, G.hold_space);
1411                        last_gets_char = '\n';
1412
1413                        break;
1414                }
1415                case 'h':       /* Replace hold space with pattern space */
1416                        free(G.hold_space);
1417                        G.hold_space = xstrdup(pattern_space);
1418                        break;
1419                case 'H':       /* Append newline and pattern space to hold space */
1420                {
1421                        int hold_space_size = 2;
1422                        int pattern_space_size = 0;
1423
1424                        if (G.hold_space)
1425                                hold_space_size += strlen(G.hold_space);
1426                        if (pattern_space)
1427                                pattern_space_size = strlen(pattern_space);
1428                        G.hold_space = xrealloc(G.hold_space,
1429                                        hold_space_size + pattern_space_size);
1430
1431                        if (hold_space_size == 2)
1432                                *G.hold_space = 0;
1433                        strcat(G.hold_space, "\n");
1434                        if (pattern_space)
1435                                strcat(G.hold_space, pattern_space);
1436
1437                        break;
1438                }
1439                case 'x': /* Exchange hold and pattern space */
1440                {
1441                        char *tmp = pattern_space;
1442                        pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1443                        last_gets_char = '\n';
1444                        G.hold_space = tmp;
1445                        break;
1446                }
1447                } /* switch */
1448        } /* for each cmd */
1449
1450        /*
1451         * Exit point from sedding...
1452         */
1453 discard_commands:
1454        /* we will print the line unless we were told to be quiet ('-n')
1455           or if the line was suppressed (ala 'd'elete) */
1456        if (!G.be_quiet)
1457                sed_puts(pattern_space, last_gets_char);
1458
1459        /* Delete and such jump here. */
1460 discard_line:
1461        flush_append(&last_puts_char /*,last_gets_char*/);
1462        free(pattern_space);
1463
1464        goto again;
1465}
1466
1467/* It is possible to have a command line argument with embedded
1468 * newlines.  This counts as multiple command lines.
1469 * However, newline can be escaped: 's/e/z\<newline>z/'
1470 * add_cmd() handles this.
1471 */
1472
1473static void add_cmd_block(char *cmdstr)
1474{
1475        char *sv, *eol;
1476
1477        cmdstr = sv = xstrdup(cmdstr);
1478        do {
1479                eol = strchr(cmdstr, '\n');
1480                if (eol)
1481                        *eol = '\0';
1482                add_cmd(cmdstr);
1483                cmdstr = eol + 1;
1484        } while (eol);
1485        free(sv);
1486}
1487
1488int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1489int sed_main(int argc UNUSED_PARAM, char **argv)
1490{
1491        unsigned opt;
1492        llist_t *opt_e, *opt_f;
1493        char *opt_i;
1494
1495#if ENABLE_LONG_OPTS
1496        static const char sed_longopts[] ALIGN1 =
1497                /* name             has_arg             short */
1498                "in-place\0"        Optional_argument   "i"
1499                "regexp-extended\0" No_argument         "r"
1500                "quiet\0"           No_argument         "n"
1501                "silent\0"          No_argument         "n"
1502                "expression\0"      Required_argument   "e"
1503                "file\0"            Required_argument   "f";
1504#endif
1505
1506        INIT_G();
1507
1508        /* destroy command strings on exit */
1509        if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1510
1511        /* Lie to autoconf when it starts asking stupid questions. */
1512        if (argv[1] && strcmp(argv[1], "--version") == 0) {
1513                puts("This is not GNU sed version 4.0");
1514                return 0;
1515        }
1516
1517        /* do normal option parsing */
1518        opt_e = opt_f = NULL;
1519        opt_i = NULL;
1520        /* -i must be first, to match OPT_in_place definition */
1521        /* -E is a synonym of -r:
1522         * GNU sed 4.2.1 mentions it in neither --help
1523         * nor manpage, but does recognize it.
1524         */
1525        opt = getopt32long(argv, "^"
1526                        "i::rEne:*f:*"
1527                        "\0" "nn"/*count -n*/,
1528                        sed_longopts,
1529                        &opt_i, &opt_e, &opt_f,
1530                        &G.be_quiet); /* counter for -n */
1531        //argc -= optind;
1532        argv += optind;
1533        if (opt & OPT_in_place) { // -i
1534                die_func = cleanup_outname;
1535        }
1536        if (opt & (2|4))
1537                G.regex_type |= REG_EXTENDED; // -r or -E
1538        //if (opt & 8)
1539        //      G.be_quiet++; // -n (implemented with a counter instead)
1540        while (opt_e) { // -e
1541                add_cmd_block(llist_pop(&opt_e));
1542        }
1543        while (opt_f) { // -f
1544                char *line;
1545                FILE *cmdfile;
1546                cmdfile = xfopen_stdin(llist_pop(&opt_f));
1547                while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1548                        add_cmd(line);
1549                        free(line);
1550                }
1551                fclose_if_not_stdin(cmdfile);
1552        }
1553        /* if we didn't get a pattern from -e or -f, use argv[0] */
1554        if (!(opt & 0x30)) {
1555                if (!*argv)
1556                        bb_show_usage();
1557                add_cmd_block(*argv++);
1558        }
1559        /* Flush any unfinished commands. */
1560        add_cmd("");
1561
1562        /* By default, we write to stdout */
1563        G.nonstdout = stdout;
1564
1565        /* argv[0..(argc-1)] should be names of file to process. If no
1566         * files were specified or '-' was specified, take input from stdin.
1567         * Otherwise, we process all the files specified. */
1568        G.input_file_list = argv;
1569        if (!argv[0]) {
1570                if (opt & OPT_in_place)
1571                        bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1572                argv[0] = (char*)bb_msg_standard_input;
1573                /* G.last_input_file = 0; - already is */
1574        } else {
1575                goto start;
1576
1577                for (; *argv; argv++) {
1578                        struct stat statbuf;
1579                        int nonstdoutfd;
1580                        sed_cmd_t *sed_cmd;
1581
1582                        G.last_input_file++;
1583 start:
1584                        if (!(opt & OPT_in_place)) {
1585                                if (LONE_DASH(*argv)) {
1586                                        *argv = (char*)bb_msg_standard_input;
1587                                        process_files();
1588                                }
1589                                continue;
1590                        }
1591
1592                        /* -i: process each FILE separately: */
1593
1594                        if (stat(*argv, &statbuf) != 0) {
1595                                bb_simple_perror_msg(*argv);
1596                                G.exitcode = EXIT_FAILURE;
1597                                G.current_input_file++;
1598                                continue;
1599                        }
1600                        G.outname = xasprintf("%sXXXXXX", *argv);
1601                        nonstdoutfd = xmkstemp(G.outname);
1602                        G.nonstdout = xfdopen_for_write(nonstdoutfd);
1603                        /* Set permissions/owner of output file */
1604                        /* chmod'ing AFTER chown would preserve suid/sgid bits,
1605                         * but GNU sed 4.2.1 does not preserve them either */
1606                        fchmod(nonstdoutfd, statbuf.st_mode);
1607                        fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1608
1609                        process_files();
1610                        fclose(G.nonstdout);
1611                        G.nonstdout = stdout;
1612
1613                        if (opt_i) {
1614                                char *backupname = xasprintf("%s%s", *argv, opt_i);
1615                                xrename(*argv, backupname);
1616                                free(backupname);
1617                        }
1618                        /* else unlink(*argv); - rename below does this */
1619                        xrename(G.outname, *argv); //TODO: rollback backup on error?
1620                        free(G.outname);
1621                        G.outname = NULL;
1622
1623                        /* Fix disabled range matches and mangled ",+N" ranges */
1624                        for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1625                                sed_cmd->beg_line = sed_cmd->beg_line_orig;
1626                                sed_cmd->end_line = sed_cmd->end_line_orig;
1627                        }
1628                }
1629                /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1630                 * if (G.current_input_file[G.current_input_file] == NULL)
1631                 *      return G.exitcode;
1632                 * but it's not needed since process_files() works correctly
1633                 * in this case too. */
1634        }
1635
1636        process_files();
1637
1638        return G.exitcode;
1639}
1640