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 * add_input_file() adds a FILE* to the list of input files. We need to
27 * know all input sources ahead of time to find the last line for the $ match.
29 * process_files() does actual sedding, reading data lines from each input FILE*
30 * (which could be stdin) and applying the sed command list (sed_cmd_head) to
31 * each of the resulting lines.
33 * sed_main() is where external code calls into this, with a command line.
36 /* Supported features and commands in this version of sed:
39 * - address matching: num|/matchstr/[,num|/matchstr/|$]command
40 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
41 * - edit commands: (a)ppend, (i)nsert, (c)hange
42 * - file commands: (r)ead
43 * - backreferences in substitution expressions (\0, \1, \2...\9)
44 * - grouped commands: {cmd1;cmd2}
45 * - transliteration (y/source-chars/dest-chars/)
46 * - pattern space hold space storing / swapping (g, h, x)
47 * - labels / branching (: label, b, t, T)
49 * (Note: Specifying an address (range) to match is *optional*; commands
50 * default to the whole pattern space if no specific address match was
54 * - Create a wrapper around regex to make libc's regex conform with sed
56 * Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 //usage:#define sed_trivial_usage
60 //usage: "[-inr] [-f FILE]... [-e CMD]... [FILE]...\n"
61 //usage: "or: sed [-inr] CMD [FILE]..."
62 //usage:#define sed_full_usage "\n\n"
63 //usage: " -e CMD Add CMD to sed commands to be executed"
64 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
65 //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)"
66 //usage: "\n Optionally backs files up, appending SFX"
67 //usage: "\n -n Suppress automatic printing of pattern space"
68 //usage: "\n -r Use extended regex syntax"
70 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
71 //usage: "\nRemaining arguments are input files (stdin if none)."
73 //usage:#define sed_example_usage
74 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
81 # define dbg(...) bb_error_msg(__VA_ARGS__)
83 # define dbg(...) ((void)0)
88 OPT_in_place = 1 << 0,
91 /* Each sed command turns into one of these structures. */
92 typedef struct sed_cmd_s {
93 /* Ordered by alignment requirements: currently 36 bytes on x86 */
94 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
97 regex_t *beg_match; /* sed -e '/match/cmd' */
98 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
99 regex_t *sub_match; /* For 's/sub_match/string/' */
100 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
101 int beg_line_orig; /* copy of the above, needed for -i */
102 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
104 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
105 char *string; /* Data string for (saicytb) commands. */
107 unsigned which_match; /* (s) Which match to replace (0 for all) */
109 /* Bitfields (gcc won't group them if we don't) */
110 unsigned invert:1; /* the '!' after the address */
111 unsigned in_match:1; /* Next line also included in match? */
112 unsigned sub_p:1; /* (s) print option */
114 char sw_last_char; /* Last line written by (sw) had no '\n' */
117 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
120 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
124 int be_quiet, regex_type;
126 char *outname, *hold_space;
128 /* List of input files */
129 int input_file_count, current_input_file;
130 FILE **input_file_list;
132 regmatch_t regmatch[10];
133 regex_t *previous_regex_ptr;
135 /* linked list of sed commands */
136 sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
138 /* Linked list of append lines */
139 llist_t *append_head;
144 char *buf; /* Space to hold string */
145 int idx; /* Space used */
146 int len; /* Space allocated */
149 #define G (*(struct globals*)&bb_common_bufsiz1)
150 struct BUG_G_too_big {
151 char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
153 #define INIT_G() do { \
154 G.sed_cmd_tail = &G.sed_cmd_head; \
158 #if ENABLE_FEATURE_CLEAN_UP
159 static void sed_free_and_close_stuff(void)
161 sed_cmd_t *sed_cmd = G.sed_cmd_head;
163 llist_free(G.append_head, free);
166 sed_cmd_t *sed_cmd_next = sed_cmd->next;
168 if (sed_cmd->sw_file)
169 xprint_and_close_file(sed_cmd->sw_file);
171 if (sed_cmd->beg_match) {
172 regfree(sed_cmd->beg_match);
173 free(sed_cmd->beg_match);
175 if (sed_cmd->end_match) {
176 regfree(sed_cmd->end_match);
177 free(sed_cmd->end_match);
179 if (sed_cmd->sub_match) {
180 regfree(sed_cmd->sub_match);
181 free(sed_cmd->sub_match);
183 free(sed_cmd->string);
185 sed_cmd = sed_cmd_next;
190 while (G.current_input_file < G.input_file_count)
191 fclose(G.input_file_list[G.current_input_file++]);
194 void sed_free_and_close_stuff(void);
197 /* If something bad happens during -i operation, delete temp file */
199 static void cleanup_outname(void)
201 if (G.outname) unlink(G.outname);
204 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
206 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
211 if (string[i] == '\\') {
212 if (!to || string[i+1] == from) {
213 *dest++ = to ? to : string[i+1];
217 *dest++ = string[i++];
219 /* TODO: is it safe wrt a string with trailing '\\' ? */
220 *dest++ = string[i++];
225 static char *copy_parsing_escapes(const char *string, int len)
228 char *dest = xmalloc(len + 1);
230 /* sed recognizes \n */
231 /* GNU sed also recognizes \t and \r */
232 for (s = "\nn\tt\rr"; *s; s += 2) {
233 parse_escapes(dest, string, len, s[1], s[0]);
242 * index_of_next_unescaped_regexp_delim - walks left to right through a string
243 * beginning at a specified index and returns the index of the next regular
244 * expression delimiter (typically a forward slash ('/')) not preceded by
245 * a backslash ('\'). A negative delimiter disables square bracket checking.
247 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
256 delimiter = -delimiter;
259 for (; (ch = str[idx]) != '\0'; idx++) {
262 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
270 else if (bracket == -1 && ch == '[')
272 else if (ch == delimiter)
276 /* if we make it to here, we've hit the end of the string */
277 bb_error_msg_and_die("unmatched '%c'", delimiter);
281 * Returns the index of the third delimiter
283 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
285 const char *cmdstr_ptr = cmdstr;
286 unsigned char delimiter;
289 /* verify that the 's' or 'y' is followed by something. That something
290 * (typically a 'slash') is now our regexp delimiter... */
292 bb_error_msg_and_die("bad format in substitution expression");
293 delimiter = *cmdstr_ptr++;
295 /* save the match string */
296 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
297 *match = copy_parsing_escapes(cmdstr_ptr, idx);
299 /* save the replacement string */
300 cmdstr_ptr += idx + 1;
301 idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr);
302 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
304 return ((cmdstr_ptr - cmdstr) + idx);
308 * returns the index in the string just past where the address ends.
310 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
312 const char *pos = my_str;
314 if (isdigit(*my_str)) {
315 *linenum = strtol(my_str, (char**)&pos, 10);
316 /* endstr shouldnt ever equal NULL */
317 } else if (*my_str == '$') {
320 } else if (*my_str == '/' || *my_str == '\\') {
328 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
329 temp = copy_parsing_escapes(pos, next);
330 *regex = xzalloc(sizeof(regex_t));
331 xregcomp(*regex, temp, G.regex_type|REG_NEWLINE);
333 /* Move position to next character after last delimiter */
339 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
340 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
342 int start = 0, idx, hack = 0;
344 /* Skip whitespace, then grab filename to end of line */
345 while (isspace(filecmdstr[start]))
348 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
351 /* If lines glued together, put backslash back. */
352 if (filecmdstr[idx] == '\n')
355 bb_error_msg_and_die("empty filename");
356 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
358 (*retval)[idx] = '\\';
363 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
365 int cflags = G.regex_type;
370 * A substitution command should look something like this:
371 * s/match/replace/ #gIpw
375 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
377 /* determine the number of back references in the match string */
378 /* Note: we compute this here rather than in the do_subst_command()
379 * function to save processor time, at the expense of a little more memory
380 * (4 bits) per sed_cmd */
382 /* process the flags */
384 sed_cmd->which_match = 1;
385 while (substr[++idx]) {
386 /* Parse match number */
387 if (isdigit(substr[idx])) {
388 if (match[0] != '^') {
389 /* Match 0 treated as all, multiple matches we take the last one. */
390 const char *pos = substr + idx;
391 /* FIXME: error check? */
392 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
398 if (isspace(substr[idx]))
401 switch (substr[idx]) {
402 /* Replace all occurrences */
405 sed_cmd->which_match = 0;
407 /* Print pattern space */
415 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
418 /* Ignore case (gnu exension) */
424 // while (substr[++idx]) continue;
425 idx += strlen(substr + idx); // same
432 bb_error_msg_and_die("bad option in substitution expression");
436 /* compile the match string into a regex */
437 if (*match != '\0') {
438 /* If match is empty, we use last regex used at runtime */
439 sed_cmd->sub_match = xzalloc(sizeof(regex_t));
440 dbg("xregcomp('%s',%x)", match, cflags);
441 xregcomp(sed_cmd->sub_match, match, cflags);
450 * Process the commands arguments
452 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
454 static const char cmd_letters[] = "saicrw:btTydDgGhHlnNpPqx={}";
485 struct chk { char chk[sizeof(cmd_letters)-1 == IDX_nul ? 1 : -1]; };
487 unsigned idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
489 /* handle (s)ubstitution command */
491 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
493 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
494 else if (idx <= IDX_c) { /* a,i,c */
495 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
496 bb_error_msg_and_die("only a beginning address can be specified for edit commands");
498 if (*cmdstr == '\n' || *cmdstr == '\\') {
502 if (!isspace(*cmdstr))
506 sed_cmd->string = xstrdup(cmdstr);
507 /* "\anychar" -> "anychar" */
508 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
509 cmdstr += strlen(cmdstr);
511 /* handle file cmds: (r)ead */
512 else if (idx <= IDX_w) { /* r,w */
513 if (sed_cmd->end_line || sed_cmd->end_match)
514 bb_error_msg_and_die("command only uses one address");
515 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
516 if (sed_cmd->cmd == 'w') {
517 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
518 sed_cmd->sw_last_char = '\n';
521 /* handle branch commands */
522 else if (idx <= IDX_T) { /* :,b,t,T */
525 cmdstr = skip_whitespace(cmdstr);
526 length = strcspn(cmdstr, semicolon_whitespace);
528 sed_cmd->string = xstrndup(cmdstr, length);
532 /* translation command */
533 else if (idx == IDX_y) {
534 char *match, *replace;
537 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
538 /* \n already parsed, but \delimiter needs unescaping. */
539 parse_escapes(match, match, strlen(match), i, i);
540 parse_escapes(replace, replace, strlen(replace), i, i);
542 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
543 for (i = 0; match[i] && replace[i]; i++) {
544 sed_cmd->string[i*2] = match[i];
545 sed_cmd->string[i*2+1] = replace[i];
550 /* if it wasnt a single-letter command that takes no arguments
551 * then it must be an invalid command.
553 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
554 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
557 /* give back whatever's left over */
562 /* Parse address+command sets, skipping comment lines. */
564 static void add_cmd(const char *cmdstr)
569 /* Append this line to any unfinished line from last time. */
570 if (G.add_cmd_line) {
571 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
572 free(G.add_cmd_line);
573 cmdstr = G.add_cmd_line = tp;
576 /* If this line ends with unescaped backslash, request next line. */
577 n = len = strlen(cmdstr);
578 while (n && cmdstr[n-1] == '\\')
580 if ((len - n) & 1) { /* if odd number of trailing backslashes */
582 G.add_cmd_line = xstrdup(cmdstr);
583 G.add_cmd_line[len-1] = '\0';
587 /* Loop parsing all commands in this line. */
589 /* Skip leading whitespace and semicolons */
590 cmdstr += strspn(cmdstr, semicolon_whitespace);
592 /* If no more commands, exit. */
595 /* if this is a comment, jump past it and keep going */
596 if (*cmdstr == '#') {
597 /* "#n" is the same as using -n on the command line */
598 if (cmdstr[1] == 'n')
600 cmdstr = strpbrk(cmdstr, "\n\r");
606 * format is: [addr][,addr][!]cmd
611 sed_cmd = xzalloc(sizeof(sed_cmd_t));
613 /* first part (if present) is an address: either a '$', a number or a /regex/ */
614 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
615 sed_cmd->beg_line_orig = sed_cmd->beg_line;
617 /* second part (if present) will begin with a comma */
618 if (*cmdstr == ',') {
622 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
624 bb_error_msg_and_die("no address after comma");
628 /* skip whitespace before the command */
629 cmdstr = skip_whitespace(cmdstr);
631 /* Check for inversion flag */
632 if (*cmdstr == '!') {
636 /* skip whitespace before the command */
637 cmdstr = skip_whitespace(cmdstr);
640 /* last part (mandatory) will be a command */
642 bb_error_msg_and_die("missing command");
643 sed_cmd->cmd = *cmdstr++;
644 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
646 /* Add the command to the command array */
647 *G.sed_cmd_tail = sed_cmd;
648 G.sed_cmd_tail = &sed_cmd->next;
651 /* If we glued multiple lines together, free the memory. */
652 free(G.add_cmd_line);
653 G.add_cmd_line = NULL;
656 /* Append to a string, reallocating memory as necessary. */
660 static void pipe_putc(char c)
662 if (G.pipeline.idx == G.pipeline.len) {
663 G.pipeline.buf = xrealloc(G.pipeline.buf,
664 G.pipeline.len + PIPE_GROW);
665 G.pipeline.len += PIPE_GROW;
667 G.pipeline.buf[G.pipeline.idx++] = c;
670 static void do_subst_w_backrefs(char *line, char *replace)
674 /* go through the replacement string */
675 for (i = 0; replace[i]; i++) {
676 /* if we find a backreference (\1, \2, etc.) print the backref'ed text */
677 if (replace[i] == '\\') {
678 unsigned backref = replace[++i] - '0';
680 /* print out the text held in G.regmatch[backref] */
681 if (G.regmatch[backref].rm_so != -1) {
682 j = G.regmatch[backref].rm_so;
683 while (j < G.regmatch[backref].rm_eo)
684 pipe_putc(line[j++]);
688 /* I _think_ it is impossible to get '\' to be
689 * the last char in replace string. Thus we dont check
690 * for replace[i] == NUL. (counterexample anyone?) */
691 /* if we find a backslash escaped character, print the character */
692 pipe_putc(replace[i]);
695 /* if we find an unescaped '&' print out the whole matched text. */
696 if (replace[i] == '&') {
697 j = G.regmatch[0].rm_so;
698 while (j < G.regmatch[0].rm_eo)
699 pipe_putc(line[j++]);
702 /* Otherwise just output the character. */
703 pipe_putc(replace[i]);
707 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
709 char *line = *line_p;
710 unsigned match_count = 0;
712 bool prev_match_empty = 1;
713 bool tried_at_eol = 0;
714 regex_t *current_regex;
716 current_regex = sed_cmd->sub_match;
717 /* Handle empty regex. */
718 if (!current_regex) {
719 current_regex = G.previous_regex_ptr;
721 bb_error_msg_and_die("no previous regexp");
723 G.previous_regex_ptr = current_regex;
725 /* Find the first match */
726 dbg("matching '%s'", line);
727 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) {
733 /* Initialize temporary output buffer. */
734 G.pipeline.buf = xmalloc(PIPE_GROW);
735 G.pipeline.len = PIPE_GROW;
738 /* Now loop through, substituting for matches */
740 int start = G.regmatch[0].rm_so;
741 int end = G.regmatch[0].rm_eo;
746 /* If we aren't interested in this match, output old line to
747 * end of match and continue */
748 if (sed_cmd->which_match
749 && (sed_cmd->which_match != match_count)
751 for (i = 0; i < end; i++)
753 /* Null match? Print one more char */
754 if (start == end && *line)
759 /* Print everything before the match */
760 for (i = 0; i < start; i++)
763 /* Then print the substitution string,
764 * unless we just matched empty string after non-empty one.
765 * Example: string "cccd", pattern "c*", repl "R":
766 * result is "RdR", not "RRdR": first match "ccc",
767 * second is "" before "d", third is "" after "d".
768 * Second match is NOT replaced!
770 if (prev_match_empty || start != 0) {
771 dbg("inserting replacement at %d in '%s'", start, line);
772 do_subst_w_backrefs(line, sed_cmd->string);
774 dbg("NOT inserting replacement at %d in '%s'", start, line);
777 /* If matched string is empty (f.e. "c*" pattern),
778 * copy verbatim one char after it before attempting more matches
780 prev_match_empty = (start == end);
781 if (prev_match_empty && line[end]) {
782 pipe_putc(line[end]);
786 /* Advance past the match */
787 dbg("line += %d", end);
789 /* Flag that something has changed */
792 /* if we're not doing this globally, get out now */
793 if (sed_cmd->which_match != 0)
796 /* Exit if we are at EOL and already tried matching at it */
803 //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
804 } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
806 /* Copy rest of string into output pipeline */
815 *line_p = G.pipeline.buf;
819 /* Set command pointer to point to this label. (Does not handle null label.) */
820 static sed_cmd_t *branch_to(char *label)
824 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
825 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
829 bb_error_msg_and_die("can't find label for jump to '%s'", label);
832 static void append(char *s)
834 llist_add_to_end(&G.append_head, xstrdup(s));
837 static void flush_append(void)
841 /* Output appended lines. */
842 while ((data = (char *)llist_pop(&G.append_head))) {
843 fprintf(G.nonstdout, "%s\n", data);
848 static void add_input_file(FILE *file)
850 G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count);
851 G.input_file_list[G.input_file_count++] = file;
854 /* Get next line of input from G.input_file_list, flushing append buffer and
855 * noting if we ran out of files without a newline on the last line we read.
861 static char *get_next_line(char *gets_char)
869 /* will be returned if last line in the file
870 * doesn't end with either '\n' or '\0' */
872 while (G.current_input_file < G.input_file_count) {
873 FILE *fp = G.input_file_list[G.current_input_file];
874 /* Read line up to a newline or NUL byte, inclusive,
875 * return malloc'ed char[]. length of the chunk read
876 * is stored in len. NULL if EOF/error */
877 temp = bb_get_chunk_from_file(fp, &len);
879 /* len > 0 here, it's ok to do temp[len-1] */
880 char c = temp[len-1];
881 if (c == '\n' || c == '\0') {
892 /* else we put NO_EOL_CHAR into *gets_char */
895 /* NB: I had the idea of peeking next file(s) and returning
896 * NO_EOL_CHAR only if it is the *last* non-empty
897 * input file. But there is a case where this won't work:
898 * file1: "a woo\nb woo"
899 * file2: "c no\nd no"
900 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
901 * (note: *no* newline after "b bang"!) */
903 /* Close this file and advance to next one */
905 G.current_input_file++;
911 /* Output line of text. */
913 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
914 * Without them, we had this:
918 * sed "s/i/z/" z1 z2 znull | hexdump -vC
921 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
923 * 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 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
961 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
963 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
965 G.previous_regex_ptr = sed_cmd->beg_match;
969 /* Process all the lines in all the files */
971 static void process_files(void)
973 char *pattern_space, *next_line;
975 char last_puts_char = '\n';
976 char last_gets_char, next_gets_char;
981 next_line = get_next_line(&next_gets_char);
983 /* Go through every line in each file */
987 /* Advance to next line. Stop if out of lines. */
988 pattern_space = next_line;
991 last_gets_char = next_gets_char;
993 /* Read one line in advance so we can act on the last line,
995 next_line = get_next_line(&next_gets_char);
998 /* For every line, go through all the commands */
1000 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1001 int old_matched, matched;
1003 old_matched = sed_cmd->in_match;
1005 /* Determine if this command matches this line: */
1007 dbg("match1:%d", sed_cmd->in_match);
1008 dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
1009 && !sed_cmd->beg_match && !sed_cmd->end_match));
1010 dbg("match3:%d", (sed_cmd->beg_line > 0
1011 && (sed_cmd->end_line || sed_cmd->end_match
1012 ? (sed_cmd->beg_line <= linenum)
1013 : (sed_cmd->beg_line == linenum)
1016 dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
1017 dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
1019 /* Are we continuing a previous multi-line match? */
1020 sed_cmd->in_match = sed_cmd->in_match
1021 /* Or is no range necessary? */
1022 || (!sed_cmd->beg_line && !sed_cmd->end_line
1023 && !sed_cmd->beg_match && !sed_cmd->end_match)
1024 /* Or did we match the start of a numerical range? */
1025 || (sed_cmd->beg_line > 0
1026 && (sed_cmd->end_line || sed_cmd->end_match
1027 /* note: even if end is numeric and is < linenum too,
1028 * GNU sed matches! We match too, therefore we don't
1029 * check here that linenum <= end.
1031 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1032 * first three input lines are deleted;
1033 * 4th line is matched and printed
1034 * by "2,3" (!) and by "4" ranges
1036 ? (sed_cmd->beg_line <= linenum) /* N,end */
1037 : (sed_cmd->beg_line == linenum) /* N */
1040 /* Or does this line match our begin address regex? */
1041 || (beg_match(sed_cmd, pattern_space))
1042 /* Or did we match last line of input? */
1043 || (sed_cmd->beg_line == -1 && next_line == NULL);
1045 /* Snapshot the value */
1046 matched = sed_cmd->in_match;
1048 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1049 sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1051 /* Is this line the end of the current match? */
1054 /* once matched, "n,xxx" range is dead, disabling it */
1055 if (sed_cmd->beg_line > 0) {
1056 sed_cmd->beg_line = -2;
1058 sed_cmd->in_match = !(
1059 /* has the ending line come, or is this a single address command? */
1061 ? sed_cmd->end_line == -1
1063 : (sed_cmd->end_line <= linenum)
1064 : !sed_cmd->end_match
1066 /* or does this line matches our last address regex */
1067 || (sed_cmd->end_match && old_matched
1068 && (regexec(sed_cmd->end_match,
1069 pattern_space, 0, NULL, 0) == 0)
1074 /* Skip blocks of commands we didn't match */
1075 if (sed_cmd->cmd == '{') {
1076 if (sed_cmd->invert ? matched : !matched) {
1077 unsigned nest_cnt = 0;
1079 if (sed_cmd->cmd == '{')
1081 if (sed_cmd->cmd == '}') {
1086 sed_cmd = sed_cmd->next;
1088 bb_error_msg_and_die("unterminated {");
1094 /* Okay, so did this line match? */
1095 if (sed_cmd->invert ? matched : !matched)
1098 /* Update last used regex in case a blank substitute BRE is found */
1099 if (sed_cmd->beg_match) {
1100 G.previous_regex_ptr = sed_cmd->beg_match;
1103 /* actual sedding */
1104 dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1105 pattern_space, next_line, sed_cmd->cmd);
1106 switch (sed_cmd->cmd) {
1108 /* Print line number */
1110 fprintf(G.nonstdout, "%d\n", linenum);
1113 /* Write the current pattern space up to the first newline */
1116 char *tmp = strchr(pattern_space, '\n');
1119 /* TODO: explain why '\n' below */
1120 sed_puts(pattern_space, '\n');
1127 /* Write the current pattern space to output */
1129 /* NB: we print this _before_ the last line
1130 * (of current file) is printed. Even if
1131 * that line is nonterminated, we print
1132 * '\n' here (gnu sed does the same) */
1133 sed_puts(pattern_space, '\n');
1135 /* Delete up through first newline */
1138 char *tmp = strchr(pattern_space, '\n');
1140 overlapping_strcpy(pattern_space, tmp + 1);
1144 /* discard this line. */
1148 /* Substitute with regex */
1150 if (!do_subst_command(sed_cmd, &pattern_space))
1152 dbg("do_subst_command succeeded:'%s'", pattern_space);
1155 /* handle p option */
1157 sed_puts(pattern_space, last_gets_char);
1158 /* handle w option */
1159 if (sed_cmd->sw_file)
1161 pattern_space, sed_cmd->sw_file,
1162 &sed_cmd->sw_last_char, last_gets_char);
1165 /* Append line to linked list to be printed later */
1167 append(sed_cmd->string);
1170 /* Insert text before this line */
1172 sed_puts(sed_cmd->string, '\n');
1175 /* Cut and paste text (replace) */
1177 /* Only triggers on last line of a matching range. */
1178 if (!sed_cmd->in_match)
1179 sed_puts(sed_cmd->string, '\n');
1182 /* Read file, append contents to output */
1186 rfile = fopen_for_read(sed_cmd->string);
1190 while ((line = xmalloc_fgetline(rfile))
1193 xprint_and_close_file(rfile);
1199 /* Write pattern space to file. */
1202 pattern_space, sed_cmd->sw_file,
1203 &sed_cmd->sw_last_char, last_gets_char);
1206 /* Read next line from input */
1209 sed_puts(pattern_space, last_gets_char);
1211 free(pattern_space);
1212 pattern_space = next_line;
1213 last_gets_char = next_gets_char;
1214 next_line = get_next_line(&next_gets_char);
1221 /* Quit. End of script, end of input. */
1223 /* Exit the outer while loop */
1226 goto discard_commands;
1228 /* Append the next line to the current line */
1232 /* If no next line, jump to end of script and exit. */
1233 /* http://www.gnu.org/software/sed/manual/sed.html:
1234 * "Most versions of sed exit without printing anything
1235 * when the N command is issued on the last line of
1236 * a file. GNU sed prints pattern space before exiting
1237 * unless of course the -n command switch has been
1238 * specified. This choice is by design."
1240 if (next_line == NULL) {
1241 //goto discard_line;
1242 goto discard_commands; /* GNU behavior */
1244 /* Append next_line, read new next_line. */
1245 len = strlen(pattern_space);
1246 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1247 pattern_space[len] = '\n';
1248 strcpy(pattern_space + len+1, next_line);
1249 last_gets_char = next_gets_char;
1250 next_line = get_next_line(&next_gets_char);
1255 /* Test/branch if substitution occurred */
1257 if (!substituted) break;
1260 /* Test/branch if substitution didn't occur */
1262 if (substituted) break;
1264 /* Branch to label */
1266 if (!sed_cmd->string) goto discard_commands;
1267 else sed_cmd = branch_to(sed_cmd->string);
1269 /* Transliterate characters */
1273 for (i = 0; pattern_space[i]; i++) {
1274 for (j = 0; sed_cmd->string[j]; j += 2) {
1275 if (pattern_space[i] == sed_cmd->string[j]) {
1276 pattern_space[i] = sed_cmd->string[j + 1];
1284 case 'g': /* Replace pattern space with hold space */
1285 free(pattern_space);
1286 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1288 case 'G': /* Append newline and hold space to pattern space */
1290 int pattern_space_size = 2;
1291 int hold_space_size = 0;
1294 pattern_space_size += strlen(pattern_space);
1296 hold_space_size = strlen(G.hold_space);
1297 pattern_space = xrealloc(pattern_space,
1298 pattern_space_size + hold_space_size);
1299 if (pattern_space_size == 2)
1300 pattern_space[0] = 0;
1301 strcat(pattern_space, "\n");
1303 strcat(pattern_space, G.hold_space);
1304 last_gets_char = '\n';
1308 case 'h': /* Replace hold space with pattern space */
1310 G.hold_space = xstrdup(pattern_space);
1312 case 'H': /* Append newline and pattern space to hold space */
1314 int hold_space_size = 2;
1315 int pattern_space_size = 0;
1318 hold_space_size += strlen(G.hold_space);
1320 pattern_space_size = strlen(pattern_space);
1321 G.hold_space = xrealloc(G.hold_space,
1322 hold_space_size + pattern_space_size);
1324 if (hold_space_size == 2)
1326 strcat(G.hold_space, "\n");
1328 strcat(G.hold_space, pattern_space);
1332 case 'x': /* Exchange hold and pattern space */
1334 char *tmp = pattern_space;
1335 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1336 last_gets_char = '\n';
1341 } /* for each cmd */
1344 * Exit point from sedding...
1347 /* we will print the line unless we were told to be quiet ('-n')
1348 or if the line was suppressed (ala 'd'elete) */
1350 sed_puts(pattern_space, last_gets_char);
1352 /* Delete and such jump here. */
1355 free(pattern_space);
1360 /* It is possible to have a command line argument with embedded
1361 * newlines. This counts as multiple command lines.
1362 * However, newline can be escaped: 's/e/z\<newline>z/'
1363 * We check for this.
1366 static void add_cmd_block(char *cmdstr)
1370 cmdstr = sv = xstrdup(cmdstr);
1372 eol = strchr(cmdstr, '\n');
1375 /* Count preceding slashes */
1379 while (sl != cmdstr && *--sl == '\\')
1381 /* Odd number of preceding slashes - newline is escaped */
1383 overlapping_strcpy(eol - 1, eol);
1384 eol = strchr(eol, '\n');
1395 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1396 int sed_main(int argc UNUSED_PARAM, char **argv)
1399 llist_t *opt_e, *opt_f;
1402 #if ENABLE_LONG_OPTS
1403 static const char sed_longopts[] ALIGN1 =
1404 /* name has_arg short */
1405 "in-place\0" Optional_argument "i"
1406 "regexp-extended\0" No_argument "r"
1407 "quiet\0" No_argument "n"
1408 "silent\0" No_argument "n"
1409 "expression\0" Required_argument "e"
1410 "file\0" Required_argument "f";
1413 int status = EXIT_SUCCESS;
1417 /* destroy command strings on exit */
1418 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1420 /* Lie to autoconf when it starts asking stupid questions. */
1421 if (argv[1] && strcmp(argv[1], "--version") == 0) {
1422 puts("This is not GNU sed version 4.0");
1426 /* do normal option parsing */
1427 opt_e = opt_f = NULL;
1429 opt_complementary = "e::f::" /* can occur multiple times */
1430 "nn"; /* count -n */
1432 IF_LONG_OPTS(applet_long_options = sed_longopts);
1434 /* -i must be first, to match OPT_in_place definition */
1435 opt = getopt32(argv, "i::rne:f:", &opt_i, &opt_e, &opt_f,
1436 &G.be_quiet); /* counter for -n */
1439 if (opt & OPT_in_place) { // -i
1440 atexit(cleanup_outname);
1442 if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r
1443 //if (opt & 0x4) G.be_quiet++; // -n
1444 while (opt_e) { // -e
1445 add_cmd_block(llist_pop(&opt_e));
1447 while (opt_f) { // -f
1450 cmdfile = xfopen_for_read(llist_pop(&opt_f));
1451 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1457 /* if we didn't get a pattern from -e or -f, use argv[0] */
1458 if (!(opt & 0x18)) {
1461 add_cmd_block(*argv++);
1463 /* Flush any unfinished commands. */
1466 /* By default, we write to stdout */
1467 G.nonstdout = stdout;
1469 /* argv[0..(argc-1)] should be names of file to process. If no
1470 * files were specified or '-' was specified, take input from stdin.
1471 * Otherwise, we process all the files specified. */
1472 if (argv[0] == NULL) {
1473 if (opt & OPT_in_place)
1474 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1475 add_input_file(stdin);
1479 for (i = 0; argv[i]; i++) {
1480 struct stat statbuf;
1485 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1486 add_input_file(stdin);
1490 file = fopen_or_warn(argv[i], "r");
1492 status = EXIT_FAILURE;
1495 add_input_file(file);
1496 if (!(opt & OPT_in_place)) {
1500 /* -i: process each FILE separately: */
1502 G.outname = xasprintf("%sXXXXXX", argv[i]);
1503 nonstdoutfd = xmkstemp(G.outname);
1504 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1506 /* Set permissions/owner of output file */
1507 fstat(fileno(file), &statbuf);
1508 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1509 * but GNU sed 4.2.1 does not preserve them either */
1510 fchmod(nonstdoutfd, statbuf.st_mode);
1511 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1514 fclose(G.nonstdout);
1515 G.nonstdout = stdout;
1518 char *backupname = xasprintf("%s%s", argv[i], opt_i);
1519 xrename(argv[i], backupname);
1522 /* else unlink(argv[i]); - rename below does this */
1523 xrename(G.outname, argv[i]); //TODO: rollback backup on error?
1527 /* Re-enable disabled range matches */
1528 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1529 sed_cmd->beg_line = sed_cmd->beg_line_orig;
1532 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1533 * if (G.current_input_file >= G.input_file_count)
1535 * but it's not needed since process_files() works correctly
1536 * in this case too. */