1 /* vi: set sw=4 ts=4: */
3 * sed.c - very minimalist version of sed
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>
11 * MAINTAINER: Rob Landley <rob@landley.net>
13 * Licensed under GPLv2, see file LICENSE in this source tree.
18 * Files are laid out to avoid unnecessary function declarations. So for
19 * example, every function add_cmd calls occurs before add_cmd in this file.
21 * add_cmd() is called on each line of sed command text (from a file or from
22 * the command line). It calls get_address() and parse_cmd_args(). The
23 * resulting sed_cmd_t structures are appended to a linked list
24 * (G.sed_cmd_head/G.sed_cmd_tail).
26 * process_files() does actual sedding, reading data lines from each input FILE*
27 * (which could be stdin) and applying the sed command list (sed_cmd_head) to
28 * each of the resulting lines.
30 * sed_main() is where external code calls into this, with a command line.
33 /* Supported features and commands in this version of sed:
36 * - address matching: num|/matchstr/[,num|/matchstr/|$]command
37 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
38 * - edit commands: (a)ppend, (i)nsert, (c)hange
39 * - file commands: (r)ead
40 * - backreferences in substitution expressions (\0, \1, \2...\9)
41 * - grouped commands: {cmd1;cmd2}
42 * - transliteration (y/source-chars/dest-chars/)
43 * - pattern space hold space storing / swapping (g, h, x)
44 * - labels / branching (: label, b, t, T)
46 * (Note: Specifying an address (range) to match is *optional*; commands
47 * default to the whole pattern space if no specific address match was
51 * - Create a wrapper around regex to make libc's regex conform with sed
54 * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
55 * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html
56 * http://sed.sourceforge.net/sedfaq3.html
63 //config: sed is used to perform text transformations on a file
64 //config: or input from a pipeline.
66 //kbuild:lib-$(CONFIG_SED) += sed.o
68 //applet:IF_SED(APPLET(sed, BB_DIR_BIN, BB_SUID_DROP))
70 //usage:#define sed_trivial_usage
71 //usage: "[-inrE] [-f FILE]... [-e CMD]... [FILE]...\n"
72 //usage: "or: sed [-inrE] CMD [FILE]..."
73 //usage:#define sed_full_usage "\n\n"
74 //usage: " -e CMD Add CMD to sed commands to be executed"
75 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
76 //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)"
77 //usage: "\n Optionally back files up, appending SFX"
78 //usage: "\n -n Suppress automatic printing of pattern space"
79 //usage: "\n -r,-E Use extended regex syntax"
81 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
82 //usage: "\nRemaining arguments are input files (stdin if none)."
84 //usage:#define sed_example_usage
85 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
89 #include "common_bufsiz.h"
93 # define dbg(...) bb_error_msg(__VA_ARGS__)
95 # define dbg(...) ((void)0)
100 OPT_in_place = 1 << 0,
103 /* Each sed command turns into one of these structures. */
104 typedef struct sed_cmd_s {
105 /* Ordered by alignment requirements: currently 36 bytes on x86 */
106 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
108 /* address storage */
109 regex_t *beg_match; /* sed -e '/match/cmd' */
110 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
111 regex_t *sub_match; /* For 's/sub_match/string/' */
112 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
113 int beg_line_orig; /* copy of the above, needed for -i */
114 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($). -2-N = +N */
117 FILE *sw_file; /* File (sw) command writes to, NULL for none. */
118 char *string; /* Data string for (saicytb) commands. */
120 unsigned which_match; /* (s) Which match to replace (0 for all) */
122 /* Bitfields (gcc won't group them if we don't) */
123 unsigned invert:1; /* the '!' after the address */
124 unsigned in_match:1; /* Next line also included in match? */
125 unsigned sub_p:1; /* (s) print option */
127 char sw_last_char; /* Last line written by (sw) had no '\n' */
130 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
133 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
137 int be_quiet, regex_type;
140 char *outname, *hold_space;
143 /* list of input files */
144 int current_input_file, last_input_file;
145 char **input_file_list;
148 regmatch_t regmatch[10];
149 regex_t *previous_regex_ptr;
151 /* linked list of sed commands */
152 sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
154 /* linked list of append lines */
155 llist_t *append_head;
160 char *buf; /* Space to hold string */
161 int idx; /* Space used */
162 int len; /* Space allocated */
165 #define G (*(struct globals*)bb_common_bufsiz1)
166 #define INIT_G() do { \
167 setup_common_bufsiz(); \
168 BUILD_BUG_ON(sizeof(G) > COMMON_BUFSIZE); \
169 G.sed_cmd_tail = &G.sed_cmd_head; \
173 #if ENABLE_FEATURE_CLEAN_UP
174 static void sed_free_and_close_stuff(void)
176 sed_cmd_t *sed_cmd = G.sed_cmd_head;
178 llist_free(G.append_head, free);
181 sed_cmd_t *sed_cmd_next = sed_cmd->next;
183 if (sed_cmd->sw_file)
184 fclose(sed_cmd->sw_file);
186 if (sed_cmd->beg_match) {
187 regfree(sed_cmd->beg_match);
188 free(sed_cmd->beg_match);
190 if (sed_cmd->end_match) {
191 regfree(sed_cmd->end_match);
192 free(sed_cmd->end_match);
194 if (sed_cmd->sub_match) {
195 regfree(sed_cmd->sub_match);
196 free(sed_cmd->sub_match);
198 free(sed_cmd->string);
200 sed_cmd = sed_cmd_next;
206 fclose(G.current_fp);
209 void sed_free_and_close_stuff(void);
212 /* If something bad happens during -i operation, delete temp file */
214 static void cleanup_outname(void)
216 if (G.outname) unlink(G.outname);
219 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
221 static unsigned parse_escapes(char *dest, const char *string, int len, char from, char to)
227 len = strlen(string);
230 if (string[i] == '\\') {
231 if (!to || string[i+1] == from) {
232 if ((*d = to ? to : string[i+1]) == '\0')
238 i++; /* skip backslash in string[] */
240 /* fall through: copy next char verbatim */
242 if ((*d = string[i++]) == '\0')
250 static char *copy_parsing_escapes(const char *string, int len)
253 char *dest = xmalloc(len + 1);
255 /* sed recognizes \n */
256 /* GNU sed also recognizes \t and \r */
257 for (s = "\nn\tt\rr"; *s; s += 2) {
258 len = parse_escapes(dest, string, len, s[1], s[0]);
266 * index_of_next_unescaped_regexp_delim - walks left to right through a string
267 * beginning at a specified index and returns the index of the next regular
268 * expression delimiter (typically a forward slash ('/')) not preceded by
269 * a backslash ('\'). A negative delimiter disables square bracket checking.
271 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
280 delimiter = -delimiter;
283 for (; (ch = str[idx]) != '\0'; idx++) {
286 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
294 else if (bracket == -1 && ch == '[')
296 else if (ch == delimiter)
300 /* if we make it to here, we've hit the end of the string */
301 bb_error_msg_and_die("unmatched '%c'", delimiter);
305 * Returns the index of the third delimiter
307 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
309 const char *cmdstr_ptr = cmdstr;
310 unsigned char delimiter;
313 /* verify that the 's' or 'y' is followed by something. That something
314 * (typically a 'slash') is now our regexp delimiter... */
316 bb_error_msg_and_die("bad format in substitution expression");
317 delimiter = *cmdstr_ptr++;
319 /* save the match string */
320 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
321 *match = copy_parsing_escapes(cmdstr_ptr, idx);
323 /* save the replacement string */
324 cmdstr_ptr += idx + 1;
325 idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr);
326 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
328 return ((cmdstr_ptr - cmdstr) + idx);
332 * returns the index in the string just past where the address ends.
334 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
336 const char *pos = my_str;
338 if (isdigit(*my_str)) {
339 *linenum = strtol(my_str, (char**)&pos, 10);
340 /* endstr shouldnt ever equal NULL */
341 } else if (*my_str == '$') {
344 } else if (*my_str == '/' || *my_str == '\\') {
352 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
353 temp = copy_parsing_escapes(pos, next);
354 *regex = xzalloc(sizeof(regex_t));
355 xregcomp(*regex, temp, G.regex_type);
357 /* Move position to next character after last delimiter */
363 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
364 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
366 int start = 0, idx, hack = 0;
368 /* Skip whitespace, then grab filename to end of line */
369 while (isspace(filecmdstr[start]))
372 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
375 /* If lines glued together, put backslash back. */
376 if (filecmdstr[idx] == '\n')
379 bb_error_msg_and_die("empty filename");
380 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
382 (*retval)[idx] = '\\';
387 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
389 int cflags = G.regex_type;
394 * A substitution command should look something like this:
395 * s/match/replace/ #giIpw
399 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
401 /* determine the number of back references in the match string */
402 /* Note: we compute this here rather than in the do_subst_command()
403 * function to save processor time, at the expense of a little more memory
404 * (4 bits) per sed_cmd */
406 /* process the flags */
408 sed_cmd->which_match = 1;
409 dbg("s flags:'%s'", substr + idx + 1);
410 while (substr[++idx]) {
411 dbg("s flag:'%c'", substr[idx]);
412 /* Parse match number */
413 if (isdigit(substr[idx])) {
414 if (match[0] != '^') {
415 /* Match 0 treated as all, multiple matches we take the last one. */
416 const char *pos = substr + idx;
417 /* FIXME: error check? */
418 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
419 idx = pos - substr - 1;
424 if (isspace(substr[idx]))
427 switch (substr[idx]) {
428 /* Replace all occurrences */
431 sed_cmd->which_match = 0;
433 /* Print pattern space */
441 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx+1, &fname);
442 sed_cmd->sw_file = xfopen_for_write(fname);
443 sed_cmd->sw_last_char = '\n';
447 /* Ignore case (gnu exension) */
454 // while (substr[++idx]) continue;
455 idx += strlen(substr + idx); // same
462 dbg("s bad flags:'%s'", substr + idx);
463 bb_error_msg_and_die("bad option in substitution expression");
467 /* compile the match string into a regex */
468 if (*match != '\0') {
469 /* If match is empty, we use last regex used at runtime */
470 sed_cmd->sub_match = xzalloc(sizeof(regex_t));
471 dbg("xregcomp('%s',%x)", match, cflags);
472 xregcomp(sed_cmd->sub_match, match, cflags);
481 * Process the commands arguments
483 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
485 static const char cmd_letters[] ALIGN1 = "saicrw:btTydDgGhHlnNpPqx={}";
518 BUILD_BUG_ON(sizeof(cmd_letters)-1 != IDX_nul);
520 idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
522 /* handle (s)ubstitution command */
524 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
526 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
527 else if (idx <= IDX_c) { /* a,i,c */
530 if (idx < IDX_c) { /* a,i */
531 if (sed_cmd->end_line || sed_cmd->end_match)
532 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
535 if (*cmdstr == '\n' || *cmdstr == '\\') {
539 if (!isspace(*cmdstr))
543 len = strlen(cmdstr);
544 sed_cmd->string = copy_parsing_escapes(cmdstr, len);
546 /* "\anychar" -> "anychar" */
547 parse_escapes(sed_cmd->string, sed_cmd->string, -1, '\0', '\0');
549 /* handle file cmds: (r)ead */
550 else if (idx <= IDX_w) { /* r,w */
551 if (idx < IDX_w) { /* r */
552 if (sed_cmd->end_line || sed_cmd->end_match)
553 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
555 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
556 if (sed_cmd->cmd == 'w') {
557 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
558 sed_cmd->sw_last_char = '\n';
561 /* handle branch commands */
562 else if (idx <= IDX_T) { /* :,b,t,T */
565 cmdstr = skip_whitespace(cmdstr);
566 length = strcspn(cmdstr, semicolon_whitespace);
568 sed_cmd->string = xstrndup(cmdstr, length);
572 /* translation command */
573 else if (idx == IDX_y) {
574 char *match, *replace;
577 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
578 /* \n already parsed, but \delimiter needs unescaping. */
579 parse_escapes(match, match, -1, i, i);
580 parse_escapes(replace, replace, -1, i, i);
582 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
583 for (i = 0; match[i] && replace[i]; i++) {
584 sed_cmd->string[i*2] = match[i];
585 sed_cmd->string[i*2+1] = replace[i];
590 /* if it wasnt a single-letter command that takes no arguments
591 * then it must be an invalid command.
593 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
594 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
597 /* give back whatever's left over */
602 /* Parse address+command sets, skipping comment lines. */
604 static void add_cmd(const char *cmdstr)
609 /* Append this line to any unfinished line from last time. */
610 if (G.add_cmd_line) {
611 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
612 free(G.add_cmd_line);
613 cmdstr = G.add_cmd_line = tp;
616 /* If this line ends with unescaped backslash, request next line. */
617 n = len = strlen(cmdstr);
618 while (n && cmdstr[n-1] == '\\')
620 if ((len - n) & 1) { /* if odd number of trailing backslashes */
622 G.add_cmd_line = xstrdup(cmdstr);
623 G.add_cmd_line[len-1] = '\0';
627 /* Loop parsing all commands in this line. */
629 /* Skip leading whitespace and semicolons */
630 cmdstr += strspn(cmdstr, semicolon_whitespace);
632 /* If no more commands, exit. */
635 /* if this is a comment, jump past it and keep going */
636 if (*cmdstr == '#') {
637 /* "#n" is the same as using -n on the command line */
638 if (cmdstr[1] == 'n')
640 cmdstr = strpbrk(cmdstr, "\n\r");
646 * format is: [addr][,addr][!]cmd
651 sed_cmd = xzalloc(sizeof(sed_cmd_t));
653 /* first part (if present) is an address: either a '$', a number or a /regex/ */
654 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
655 sed_cmd->beg_line_orig = sed_cmd->beg_line;
657 /* second part (if present) will begin with a comma */
658 if (*cmdstr == ',') {
662 if (*cmdstr == '+' && isdigit(cmdstr[1])) {
663 /* http://sed.sourceforge.net/sedfaq3.html#s3.3
664 * Under GNU sed 3.02+, ssed, and sed15+, <address2>
665 * may also be a notation of the form +num,
666 * indicating the next num lines after <address1> is
668 * GNU sed 4.2.1 accepts even "+" (meaning "+0").
669 * We don't (we check for isdigit, see above), think
670 * about the "+-3" case.
673 /* code is smaller compared to using &cmdstr here: */
674 idx = strtol(cmdstr+1, &end, 10);
675 sed_cmd->end_line = -2 - idx;
678 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
680 idx--; /* if 0, trigger error check below */
683 bb_error_msg_and_die("no address after comma");
684 sed_cmd->end_line_orig = sed_cmd->end_line;
687 /* skip whitespace before the command */
688 cmdstr = skip_whitespace(cmdstr);
690 /* Check for inversion flag */
691 if (*cmdstr == '!') {
695 /* skip whitespace before the command */
696 cmdstr = skip_whitespace(cmdstr);
699 /* last part (mandatory) will be a command */
701 bb_error_msg_and_die("missing command");
702 sed_cmd->cmd = *cmdstr++;
703 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
705 /* cmdstr now points past args.
706 * GNU sed requires a separator, if there are more commands,
707 * else it complains "char N: extra characters after command".
708 * Example: "sed 'p;d'". We also allow "sed 'pd'".
711 /* Add the command to the command array */
712 *G.sed_cmd_tail = sed_cmd;
713 G.sed_cmd_tail = &sed_cmd->next;
716 /* If we glued multiple lines together, free the memory. */
717 free(G.add_cmd_line);
718 G.add_cmd_line = NULL;
721 /* Append to a string, reallocating memory as necessary. */
725 static void pipe_putc(char c)
727 if (G.pipeline.idx == G.pipeline.len) {
728 G.pipeline.buf = xrealloc(G.pipeline.buf,
729 G.pipeline.len + PIPE_GROW);
730 G.pipeline.len += PIPE_GROW;
732 G.pipeline.buf[G.pipeline.idx++] = c;
735 static void do_subst_w_backrefs(char *line, char *replace)
739 /* go through the replacement string */
740 for (i = 0; replace[i]; i++) {
741 /* if we find a backreference (\1, \2, etc.) print the backref'ed text */
742 if (replace[i] == '\\') {
743 unsigned backref = replace[++i] - '0';
745 /* print out the text held in G.regmatch[backref] */
746 if (G.regmatch[backref].rm_so != -1) {
747 j = G.regmatch[backref].rm_so;
748 while (j < G.regmatch[backref].rm_eo)
749 pipe_putc(line[j++]);
753 /* I _think_ it is impossible to get '\' to be
754 * the last char in replace string. Thus we dont check
755 * for replace[i] == NUL. (counterexample anyone?) */
756 /* if we find a backslash escaped character, print the character */
757 pipe_putc(replace[i]);
760 /* if we find an unescaped '&' print out the whole matched text. */
761 if (replace[i] == '&') {
762 j = G.regmatch[0].rm_so;
763 while (j < G.regmatch[0].rm_eo)
764 pipe_putc(line[j++]);
767 /* Otherwise just output the character. */
768 pipe_putc(replace[i]);
772 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
774 char *line = *line_p;
775 unsigned match_count = 0;
777 bool prev_match_empty = 1;
778 bool tried_at_eol = 0;
779 regex_t *current_regex;
781 current_regex = sed_cmd->sub_match;
782 /* Handle empty regex. */
783 if (!current_regex) {
784 current_regex = G.previous_regex_ptr;
786 bb_error_msg_and_die("no previous regexp");
788 G.previous_regex_ptr = current_regex;
790 /* Find the first match */
791 dbg("matching '%s'", line);
792 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) {
798 /* Initialize temporary output buffer. */
799 G.pipeline.buf = xmalloc(PIPE_GROW);
800 G.pipeline.len = PIPE_GROW;
803 /* Now loop through, substituting for matches */
805 int start = G.regmatch[0].rm_so;
806 int end = G.regmatch[0].rm_eo;
811 /* If we aren't interested in this match, output old line to
812 * end of match and continue */
813 if (sed_cmd->which_match
814 && (sed_cmd->which_match != match_count)
816 for (i = 0; i < end; i++)
818 /* Null match? Print one more char */
819 if (start == end && *line)
824 /* Print everything before the match */
825 for (i = 0; i < start; i++)
828 /* Then print the substitution string,
829 * unless we just matched empty string after non-empty one.
830 * Example: string "cccd", pattern "c*", repl "R":
831 * result is "RdR", not "RRdR": first match "ccc",
832 * second is "" before "d", third is "" after "d".
833 * Second match is NOT replaced!
835 if (prev_match_empty || start != 0 || start != end) {
836 //dbg("%d %d %d", prev_match_empty, start, end);
837 dbg("inserting replacement at %d in '%s'", start, line);
838 do_subst_w_backrefs(line, sed_cmd->string);
839 /* Flag that something has changed */
842 dbg("NOT inserting replacement at %d in '%s'", start, line);
845 /* If matched string is empty (f.e. "c*" pattern),
846 * copy verbatim one char after it before attempting more matches
848 prev_match_empty = (start == end);
849 if (prev_match_empty) {
853 pipe_putc(line[end]);
858 /* Advance past the match */
859 dbg("line += %d", end);
862 /* if we're not doing this globally, get out now */
863 if (sed_cmd->which_match != 0)
866 /* Exit if we are at EOL and already tried matching at it */
873 //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
874 } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
876 /* Copy rest of string into output pipeline */
885 *line_p = G.pipeline.buf;
889 /* Set command pointer to point to this label. (Does not handle null label.) */
890 static sed_cmd_t *branch_to(char *label)
894 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
895 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
899 bb_error_msg_and_die("can't find label for jump to '%s'", label);
902 static void append(char *s)
904 llist_add_to_end(&G.append_head, s);
907 /* Output line of text. */
909 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
910 * Without them, we had this:
914 * sed "s/i/z/" z1 z2 znull | hexdump -vC
917 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
919 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
925 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
927 char lpc = *last_puts_char;
929 /* Need to insert a '\n' between two files because first file's
930 * last line wasn't terminated? */
931 if (lpc != '\n' && lpc != '\0') {
937 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
941 /* had trailing '\0' and it was last char of file? */
942 if (last_gets_char == LAST_IS_NUL) {
946 /* had trailing '\n' or '\0'? */
947 if (last_gets_char != NO_EOL_CHAR) {
948 fputc(last_gets_char, file);
949 lpc = last_gets_char;
953 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
954 bb_error_msg_and_die(bb_msg_write_error);
956 *last_puts_char = lpc;
959 static void flush_append(char *last_puts_char)
963 /* Output appended lines. */
964 while ((data = (char *)llist_pop(&G.append_head)) != NULL) {
965 /* Append command does not respect "nonterminated-ness"
966 * of last line. Try this:
967 * $ echo -n "woot" | sed -e '/woot/a woo' -
970 * (both lines are terminated with \n)
971 * Therefore we do not propagate "last_gets_char" here,
974 puts_maybe_newline(data, G.nonstdout, last_puts_char, '\n');
979 /* Get next line of input from G.input_file_list, flushing append buffer and
980 * noting if we ran out of files without a newline on the last line we read.
982 static char *get_next_line(char *gets_char, char *last_puts_char)
988 flush_append(last_puts_char);
990 /* will be returned if last line in the file
991 * doesn't end with either '\n' or '\0' */
993 for (; G.current_input_file <= G.last_input_file; G.current_input_file++) {
994 FILE *fp = G.current_fp;
996 const char *path = G.input_file_list[G.current_input_file];
998 if (path != bb_msg_standard_input) {
999 fp = fopen_or_warn(path, "r");
1001 G.exitcode = EXIT_FAILURE;
1007 /* Read line up to a newline or NUL byte, inclusive,
1008 * return malloc'ed char[]. length of the chunk read
1009 * is stored in len. NULL if EOF/error */
1010 temp = bb_get_chunk_from_file(fp, &len);
1012 /* len > 0 here, it's ok to do temp[len-1] */
1013 char c = temp[len-1];
1014 if (c == '\n' || c == '\0') {
1025 /* else we put NO_EOL_CHAR into *gets_char */
1028 /* NB: I had the idea of peeking next file(s) and returning
1029 * NO_EOL_CHAR only if it is the *last* non-empty
1030 * input file. But there is a case where this won't work:
1031 * file1: "a woo\nb woo"
1032 * file2: "c no\nd no"
1033 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
1034 * (note: *no* newline after "b bang"!) */
1036 /* Close this file and advance to next one */
1037 fclose_if_not_stdin(fp);
1038 G.current_fp = NULL;
1044 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
1046 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
1048 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
1050 G.previous_regex_ptr = sed_cmd->beg_match;
1054 /* Process all the lines in all the files */
1056 static void process_files(void)
1058 char *pattern_space, *next_line;
1060 char last_puts_char = '\n';
1061 char last_gets_char, next_gets_char;
1065 /* Prime the pump */
1066 next_line = get_next_line(&next_gets_char, &last_puts_char);
1068 /* Go through every line in each file */
1072 /* Advance to next line. Stop if out of lines. */
1073 pattern_space = next_line;
1076 last_gets_char = next_gets_char;
1078 /* Read one line in advance so we can act on the last line,
1079 * the '$' address */
1080 next_line = get_next_line(&next_gets_char, &last_puts_char);
1083 /* For every line, go through all the commands */
1085 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1086 int old_matched, matched;
1088 old_matched = sed_cmd->in_match;
1090 /* Determine if this command matches this line: */
1092 dbg("match1:%d", sed_cmd->in_match);
1093 dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
1094 && !sed_cmd->beg_match && !sed_cmd->end_match));
1095 dbg("match3:%d", (sed_cmd->beg_line > 0
1096 && (sed_cmd->end_line || sed_cmd->end_match
1097 ? (sed_cmd->beg_line <= linenum)
1098 : (sed_cmd->beg_line == linenum)
1101 dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
1102 dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
1104 /* Are we continuing a previous multi-line match? */
1105 sed_cmd->in_match = sed_cmd->in_match
1106 /* Or is no range necessary? */
1107 || (!sed_cmd->beg_line && !sed_cmd->end_line
1108 && !sed_cmd->beg_match && !sed_cmd->end_match)
1109 /* Or did we match the start of a numerical range? */
1110 || (sed_cmd->beg_line > 0
1111 && (sed_cmd->end_line || sed_cmd->end_match
1112 /* note: even if end is numeric and is < linenum too,
1113 * GNU sed matches! We match too, therefore we don't
1114 * check here that linenum <= end.
1116 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1117 * first three input lines are deleted;
1118 * 4th line is matched and printed
1119 * by "2,3" (!) and by "4" ranges
1121 ? (sed_cmd->beg_line <= linenum) /* N,end */
1122 : (sed_cmd->beg_line == linenum) /* N */
1125 /* Or does this line match our begin address regex? */
1126 || (beg_match(sed_cmd, pattern_space))
1127 /* Or did we match last line of input? */
1128 || (sed_cmd->beg_line == -1 && next_line == NULL);
1130 /* Snapshot the value */
1131 matched = sed_cmd->in_match;
1133 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1134 sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1136 /* Is this line the end of the current match? */
1139 if (sed_cmd->end_line <= -2) {
1140 /* address2 is +N, i.e. N lines from beg_line */
1141 sed_cmd->end_line = linenum + (-sed_cmd->end_line - 2);
1143 /* once matched, "n,xxx" range is dead, disabling it */
1144 if (sed_cmd->beg_line > 0) {
1145 sed_cmd->beg_line = -2;
1147 dbg("end1:%d", sed_cmd->end_line ? sed_cmd->end_line == -1
1148 ? !next_line : (sed_cmd->end_line <= linenum)
1149 : !sed_cmd->end_match);
1150 dbg("end2:%d", sed_cmd->end_match && old_matched
1151 && !regexec(sed_cmd->end_match,pattern_space, 0, NULL, 0));
1152 sed_cmd->in_match = !(
1153 /* has the ending line come, or is this a single address command? */
1155 ? sed_cmd->end_line == -1
1157 : (sed_cmd->end_line <= linenum)
1158 : !sed_cmd->end_match
1160 /* or does this line matches our last address regex */
1161 || (sed_cmd->end_match && old_matched
1162 && (regexec(sed_cmd->end_match,
1163 pattern_space, 0, NULL, 0) == 0)
1168 /* Skip blocks of commands we didn't match */
1169 if (sed_cmd->cmd == '{') {
1170 if (sed_cmd->invert ? matched : !matched) {
1171 unsigned nest_cnt = 0;
1173 if (sed_cmd->cmd == '{')
1175 if (sed_cmd->cmd == '}') {
1180 sed_cmd = sed_cmd->next;
1182 bb_error_msg_and_die("unterminated {");
1188 /* Okay, so did this line match? */
1189 if (sed_cmd->invert ? matched : !matched)
1192 /* Update last used regex in case a blank substitute BRE is found */
1193 if (sed_cmd->beg_match) {
1194 G.previous_regex_ptr = sed_cmd->beg_match;
1197 /* actual sedding */
1198 dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1199 pattern_space, next_line, sed_cmd->cmd);
1200 switch (sed_cmd->cmd) {
1202 /* Print line number */
1204 fprintf(G.nonstdout, "%d\n", linenum);
1207 /* Write the current pattern space up to the first newline */
1210 char *tmp = strchr(pattern_space, '\n');
1213 /* TODO: explain why '\n' below */
1214 sed_puts(pattern_space, '\n');
1221 /* Write the current pattern space to output */
1223 /* NB: we print this _before_ the last line
1224 * (of current file) is printed. Even if
1225 * that line is nonterminated, we print
1226 * '\n' here (gnu sed does the same) */
1227 sed_puts(pattern_space, '\n');
1229 /* Delete up through first newline */
1232 char *tmp = strchr(pattern_space, '\n');
1234 overlapping_strcpy(pattern_space, tmp + 1);
1238 /* discard this line. */
1242 /* Substitute with regex */
1244 if (!do_subst_command(sed_cmd, &pattern_space))
1246 dbg("do_subst_command succeeded:'%s'", pattern_space);
1249 /* handle p option */
1251 sed_puts(pattern_space, last_gets_char);
1252 /* handle w option */
1253 if (sed_cmd->sw_file)
1255 pattern_space, sed_cmd->sw_file,
1256 &sed_cmd->sw_last_char, last_gets_char);
1259 /* Append line to linked list to be printed later */
1261 append(xstrdup(sed_cmd->string));
1264 /* Insert text before this line */
1266 sed_puts(sed_cmd->string, '\n');
1269 /* Cut and paste text (replace) */
1271 /* Only triggers on last line of a matching range. */
1272 if (!sed_cmd->in_match)
1273 sed_puts(sed_cmd->string, '\n');
1276 /* Read file, append contents to output */
1280 rfile = fopen_for_read(sed_cmd->string);
1283 while ((line = xmalloc_fgetline(rfile))
1292 /* Write pattern space to file. */
1295 pattern_space, sed_cmd->sw_file,
1296 &sed_cmd->sw_last_char, last_gets_char);
1299 /* Read next line from input */
1302 sed_puts(pattern_space, last_gets_char);
1303 if (next_line == NULL) {
1304 /* If no next line, jump to end of script and exit. */
1307 free(pattern_space);
1308 pattern_space = next_line;
1309 last_gets_char = next_gets_char;
1310 next_line = get_next_line(&next_gets_char, &last_puts_char);
1315 /* Quit. End of script, end of input. */
1317 /* Exit the outer while loop */
1320 goto discard_commands;
1322 /* Append the next line to the current line */
1326 /* If no next line, jump to end of script and exit. */
1327 /* http://www.gnu.org/software/sed/manual/sed.html:
1328 * "Most versions of sed exit without printing anything
1329 * when the N command is issued on the last line of
1330 * a file. GNU sed prints pattern space before exiting
1331 * unless of course the -n command switch has been
1332 * specified. This choice is by design."
1334 if (next_line == NULL) {
1335 //goto discard_line;
1336 goto discard_commands; /* GNU behavior */
1338 /* Append next_line, read new next_line. */
1339 len = strlen(pattern_space);
1340 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1341 pattern_space[len] = '\n';
1342 strcpy(pattern_space + len+1, next_line);
1343 last_gets_char = next_gets_char;
1344 next_line = get_next_line(&next_gets_char, &last_puts_char);
1349 /* Test/branch if substitution occurred */
1351 if (!substituted) break;
1354 /* Test/branch if substitution didn't occur */
1356 if (substituted) break;
1358 /* Branch to label */
1360 if (!sed_cmd->string) goto discard_commands;
1361 else sed_cmd = branch_to(sed_cmd->string);
1363 /* Transliterate characters */
1367 for (i = 0; pattern_space[i]; i++) {
1368 for (j = 0; sed_cmd->string[j]; j += 2) {
1369 if (pattern_space[i] == sed_cmd->string[j]) {
1370 pattern_space[i] = sed_cmd->string[j + 1];
1378 case 'g': /* Replace pattern space with hold space */
1379 free(pattern_space);
1380 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1382 case 'G': /* Append newline and hold space to pattern space */
1384 int pattern_space_size = 2;
1385 int hold_space_size = 0;
1388 pattern_space_size += strlen(pattern_space);
1390 hold_space_size = strlen(G.hold_space);
1391 pattern_space = xrealloc(pattern_space,
1392 pattern_space_size + hold_space_size);
1393 if (pattern_space_size == 2)
1394 pattern_space[0] = 0;
1395 strcat(pattern_space, "\n");
1397 strcat(pattern_space, G.hold_space);
1398 last_gets_char = '\n';
1402 case 'h': /* Replace hold space with pattern space */
1404 G.hold_space = xstrdup(pattern_space);
1406 case 'H': /* Append newline and pattern space to hold space */
1408 int hold_space_size = 2;
1409 int pattern_space_size = 0;
1412 hold_space_size += strlen(G.hold_space);
1414 pattern_space_size = strlen(pattern_space);
1415 G.hold_space = xrealloc(G.hold_space,
1416 hold_space_size + pattern_space_size);
1418 if (hold_space_size == 2)
1420 strcat(G.hold_space, "\n");
1422 strcat(G.hold_space, pattern_space);
1426 case 'x': /* Exchange hold and pattern space */
1428 char *tmp = pattern_space;
1429 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1430 last_gets_char = '\n';
1435 } /* for each cmd */
1438 * Exit point from sedding...
1441 /* we will print the line unless we were told to be quiet ('-n')
1442 or if the line was suppressed (ala 'd'elete) */
1444 sed_puts(pattern_space, last_gets_char);
1446 /* Delete and such jump here. */
1448 flush_append(&last_puts_char /*,last_gets_char*/);
1449 free(pattern_space);
1454 /* It is possible to have a command line argument with embedded
1455 * newlines. This counts as multiple command lines.
1456 * However, newline can be escaped: 's/e/z\<newline>z/'
1457 * add_cmd() handles this.
1460 static void add_cmd_block(char *cmdstr)
1464 cmdstr = sv = xstrdup(cmdstr);
1466 eol = strchr(cmdstr, '\n');
1475 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1476 int sed_main(int argc UNUSED_PARAM, char **argv)
1479 llist_t *opt_e, *opt_f;
1482 #if ENABLE_LONG_OPTS
1483 static const char sed_longopts[] ALIGN1 =
1484 /* name has_arg short */
1485 "in-place\0" Optional_argument "i"
1486 "regexp-extended\0" No_argument "r"
1487 "quiet\0" No_argument "n"
1488 "silent\0" No_argument "n"
1489 "expression\0" Required_argument "e"
1490 "file\0" Required_argument "f";
1495 /* destroy command strings on exit */
1496 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1498 /* Lie to autoconf when it starts asking stupid questions. */
1499 if (argv[1] && strcmp(argv[1], "--version") == 0) {
1500 puts("This is not GNU sed version 4.0");
1504 /* do normal option parsing */
1505 opt_e = opt_f = NULL;
1507 opt_complementary = "nn"; /* count -n */
1509 IF_LONG_OPTS(applet_long_options = sed_longopts);
1511 /* -i must be first, to match OPT_in_place definition */
1512 /* -E is a synonym of -r:
1513 * GNU sed 4.2.1 mentions it in neither --help
1514 * nor manpage, but does recognize it.
1516 opt = getopt32(argv, "i::rEne:*f:*", &opt_i, &opt_e, &opt_f,
1517 &G.be_quiet); /* counter for -n */
1520 if (opt & OPT_in_place) { // -i
1521 atexit(cleanup_outname);
1524 G.regex_type |= REG_EXTENDED; // -r or -E
1526 // G.be_quiet++; // -n (implemented with a counter instead)
1527 while (opt_e) { // -e
1528 add_cmd_block(llist_pop(&opt_e));
1530 while (opt_f) { // -f
1533 cmdfile = xfopen_stdin(llist_pop(&opt_f));
1534 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1538 fclose_if_not_stdin(cmdfile);
1540 /* if we didn't get a pattern from -e or -f, use argv[0] */
1541 if (!(opt & 0x30)) {
1544 add_cmd_block(*argv++);
1546 /* Flush any unfinished commands. */
1549 /* By default, we write to stdout */
1550 G.nonstdout = stdout;
1552 /* argv[0..(argc-1)] should be names of file to process. If no
1553 * files were specified or '-' was specified, take input from stdin.
1554 * Otherwise, we process all the files specified. */
1555 G.input_file_list = argv;
1557 if (opt & OPT_in_place)
1558 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1559 argv[0] = (char*)bb_msg_standard_input;
1560 /* G.last_input_file = 0; - already is */
1564 for (; *argv; argv++) {
1565 struct stat statbuf;
1569 G.last_input_file++;
1571 if (!(opt & OPT_in_place)) {
1572 if (LONE_DASH(*argv)) {
1573 *argv = (char*)bb_msg_standard_input;
1579 /* -i: process each FILE separately: */
1581 if (stat(*argv, &statbuf) != 0) {
1582 bb_simple_perror_msg(*argv);
1583 G.exitcode = EXIT_FAILURE;
1584 G.current_input_file++;
1587 G.outname = xasprintf("%sXXXXXX", *argv);
1588 nonstdoutfd = xmkstemp(G.outname);
1589 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1590 /* Set permissions/owner of output file */
1591 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1592 * but GNU sed 4.2.1 does not preserve them either */
1593 fchmod(nonstdoutfd, statbuf.st_mode);
1594 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1597 fclose(G.nonstdout);
1598 G.nonstdout = stdout;
1601 char *backupname = xasprintf("%s%s", *argv, opt_i);
1602 xrename(*argv, backupname);
1605 /* else unlink(*argv); - rename below does this */
1606 xrename(G.outname, *argv); //TODO: rollback backup on error?
1610 /* Fix disabled range matches and mangled ",+N" ranges */
1611 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1612 sed_cmd->beg_line = sed_cmd->beg_line_orig;
1613 sed_cmd->end_line = sed_cmd->end_line_orig;
1616 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1617 * if (G.current_input_file[G.current_input_file] == NULL)
1618 * return G.exitcode;
1619 * but it's not needed since process_files() works correctly
1620 * in this case too. */