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/|$]command
40 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
41 * - edit commands: (a)ppend, (i)nsert, (c)hange
42 * - file commands: (r)ead
43 * - backreferences in substitution expressions (\0, \1, \2...\9)
44 * - grouped commands: {cmd1;cmd2}
45 * - transliteration (y/source-chars/dest-chars/)
46 * - pattern space hold space storing / swapping (g, h, x)
47 * - labels / branching (: label, b, t, T)
49 * (Note: Specifying an address (range) to match is *optional*; commands
50 * default to the whole pattern space if no specific address match was
54 * - Create a wrapper around regex to make libc's regex conform with sed
56 * Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 //usage:#define sed_trivial_usage
60 //usage: "[-efinr] SED_CMD [FILE]..."
61 //usage:#define sed_full_usage "\n\n"
62 //usage: " -e CMD Add CMD to sed commands to be executed"
63 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
64 //usage: "\n -i Edit files in-place (else sends result to stdout)"
65 //usage: "\n -n Suppress automatic printing of pattern space"
66 //usage: "\n -r Use extended regex syntax"
68 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
69 //usage: "\nRemaining arguments are input files (stdin if none)."
71 //usage:#define sed_example_usage
72 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
79 # define dbg(...) bb_error_msg(__VA_ARGS__)
81 # define dbg(...) ((void)0)
86 OPT_in_place = 1 << 0,
89 /* Each sed command turns into one of these structures. */
90 typedef struct sed_cmd_s {
91 /* Ordered by alignment requirements: currently 36 bytes on x86 */
92 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
95 regex_t *beg_match; /* sed -e '/match/cmd' */
96 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
97 regex_t *sub_match; /* For 's/sub_match/string/' */
98 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
99 int beg_line_orig; /* copy of the above, needed for -i */
100 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
102 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
103 char *string; /* Data string for (saicytb) commands. */
105 unsigned which_match; /* (s) Which match to replace (0 for all) */
107 /* Bitfields (gcc won't group them if we don't) */
108 unsigned invert:1; /* the '!' after the address */
109 unsigned in_match:1; /* Next line also included in match? */
110 unsigned sub_p:1; /* (s) print option */
112 char sw_last_char; /* Last line written by (sw) had no '\n' */
115 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
118 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
122 int be_quiet, regex_type;
124 char *outname, *hold_space;
126 /* List of input files */
127 int input_file_count, current_input_file;
128 FILE **input_file_list;
130 regmatch_t regmatch[10];
131 regex_t *previous_regex_ptr;
133 /* linked list of sed commands */
134 sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
136 /* Linked list of append lines */
137 llist_t *append_head;
142 char *buf; /* Space to hold string */
143 int idx; /* Space used */
144 int len; /* Space allocated */
147 #define G (*(struct globals*)&bb_common_bufsiz1)
148 struct BUG_G_too_big {
149 char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
151 #define INIT_G() do { \
152 G.sed_cmd_tail = &G.sed_cmd_head; \
156 #if ENABLE_FEATURE_CLEAN_UP
157 static void sed_free_and_close_stuff(void)
159 sed_cmd_t *sed_cmd = G.sed_cmd_head;
161 llist_free(G.append_head, free);
164 sed_cmd_t *sed_cmd_next = sed_cmd->next;
166 if (sed_cmd->sw_file)
167 xprint_and_close_file(sed_cmd->sw_file);
169 if (sed_cmd->beg_match) {
170 regfree(sed_cmd->beg_match);
171 free(sed_cmd->beg_match);
173 if (sed_cmd->end_match) {
174 regfree(sed_cmd->end_match);
175 free(sed_cmd->end_match);
177 if (sed_cmd->sub_match) {
178 regfree(sed_cmd->sub_match);
179 free(sed_cmd->sub_match);
181 free(sed_cmd->string);
183 sed_cmd = sed_cmd_next;
188 while (G.current_input_file < G.input_file_count)
189 fclose(G.input_file_list[G.current_input_file++]);
192 void sed_free_and_close_stuff(void);
195 /* If something bad happens during -i operation, delete temp file */
197 static void cleanup_outname(void)
199 if (G.outname) unlink(G.outname);
202 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
204 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
209 if (string[i] == '\\') {
210 if (!to || string[i+1] == from) {
211 *dest++ = to ? to : string[i+1];
215 *dest++ = string[i++];
217 /* TODO: is it safe wrt a string with trailing '\\' ? */
218 *dest++ = string[i++];
223 static char *copy_parsing_escapes(const char *string, int len)
226 char *dest = xmalloc(len + 1);
228 /* sed recognizes \n */
229 /* GNU sed also recognizes \t and \r */
230 for (s = "\nn\tt\rr"; *s; s += 2) {
231 parse_escapes(dest, string, len, s[1], s[0]);
240 * index_of_next_unescaped_regexp_delim - walks left to right through a string
241 * beginning at a specified index and returns the index of the next regular
242 * expression delimiter (typically a forward slash ('/')) not preceded by
243 * a backslash ('\'). A negative delimiter disables square bracket checking.
245 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
254 delimiter = -delimiter;
257 for (; (ch = str[idx]) != '\0'; idx++) {
260 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
268 else if (bracket == -1 && ch == '[')
270 else if (ch == delimiter)
274 /* if we make it to here, we've hit the end of the string */
275 bb_error_msg_and_die("unmatched '%c'", delimiter);
279 * Returns the index of the third delimiter
281 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
283 const char *cmdstr_ptr = cmdstr;
287 /* verify that the 's' or 'y' is followed by something. That something
288 * (typically a 'slash') is now our regexp delimiter... */
290 bb_error_msg_and_die("bad format in substitution expression");
291 delimiter = *cmdstr_ptr++;
293 /* save the match string */
294 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
295 *match = copy_parsing_escapes(cmdstr_ptr, idx);
297 /* save the replacement string */
298 cmdstr_ptr += idx + 1;
299 idx = index_of_next_unescaped_regexp_delim(-delimiter, cmdstr_ptr);
300 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
302 return ((cmdstr_ptr - cmdstr) + idx);
306 * returns the index in the string just past where the address ends.
308 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
310 const char *pos = my_str;
312 if (isdigit(*my_str)) {
313 *linenum = strtol(my_str, (char**)&pos, 10);
314 /* endstr shouldnt ever equal NULL */
315 } else if (*my_str == '$') {
318 } else if (*my_str == '/' || *my_str == '\\') {
324 if (*my_str == '\\') delimiter = *++pos;
325 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
326 temp = copy_parsing_escapes(pos, next);
327 *regex = xmalloc(sizeof(regex_t));
328 xregcomp(*regex, temp, G.regex_type|REG_NEWLINE);
330 /* Move position to next character after last delimiter */
336 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
337 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
339 int start = 0, idx, hack = 0;
341 /* Skip whitespace, then grab filename to end of line */
342 while (isspace(filecmdstr[start]))
345 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
348 /* If lines glued together, put backslash back. */
349 if (filecmdstr[idx] == '\n')
352 bb_error_msg_and_die("empty filename");
353 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
355 (*retval)[idx] = '\\';
360 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
362 int cflags = G.regex_type;
367 * A substitution command should look something like this:
368 * s/match/replace/ #gIpw
372 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
374 /* determine the number of back references in the match string */
375 /* Note: we compute this here rather than in the do_subst_command()
376 * function to save processor time, at the expense of a little more memory
377 * (4 bits) per sed_cmd */
379 /* process the flags */
381 sed_cmd->which_match = 1;
382 while (substr[++idx]) {
383 /* Parse match number */
384 if (isdigit(substr[idx])) {
385 if (match[0] != '^') {
386 /* Match 0 treated as all, multiple matches we take the last one. */
387 const char *pos = substr + idx;
388 /* FIXME: error check? */
389 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
395 if (isspace(substr[idx]))
398 switch (substr[idx]) {
399 /* Replace all occurrences */
402 sed_cmd->which_match = 0;
404 /* Print pattern space */
412 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
415 /* Ignore case (gnu exension) */
421 // while (substr[++idx]) continue;
422 idx += strlen(substr + idx); // same
429 bb_error_msg_and_die("bad option in substitution expression");
433 /* compile the match string into a regex */
434 if (*match != '\0') {
435 /* If match is empty, we use last regex used at runtime */
436 sed_cmd->sub_match = xmalloc(sizeof(regex_t));
437 xregcomp(sed_cmd->sub_match, match, cflags);
445 * Process the commands arguments
447 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
449 static const char cmd_letters[] = "saicrw:btTydDgGhHlnNpPqx={}";
480 struct chk { char chk[sizeof(cmd_letters)-1 == IDX_nul ? 1 : -1]; };
482 unsigned idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
484 /* handle (s)ubstitution command */
486 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
488 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
489 else if (idx <= IDX_c) { /* a,i,c */
490 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
491 bb_error_msg_and_die("only a beginning address can be specified for edit commands");
493 if (*cmdstr == '\n' || *cmdstr == '\\') {
497 if (!isspace(*cmdstr))
501 sed_cmd->string = xstrdup(cmdstr);
502 /* "\anychar" -> "anychar" */
503 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
504 cmdstr += strlen(cmdstr);
506 /* handle file cmds: (r)ead */
507 else if (idx <= IDX_w) { /* r,w */
508 if (sed_cmd->end_line || sed_cmd->end_match)
509 bb_error_msg_and_die("command only uses one address");
510 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
511 if (sed_cmd->cmd == 'w') {
512 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
513 sed_cmd->sw_last_char = '\n';
516 /* handle branch commands */
517 else if (idx <= IDX_T) { /* :,b,t,T */
520 cmdstr = skip_whitespace(cmdstr);
521 length = strcspn(cmdstr, semicolon_whitespace);
523 sed_cmd->string = xstrndup(cmdstr, length);
527 /* translation command */
528 else if (idx == IDX_y) {
529 char *match, *replace;
532 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
533 /* \n already parsed, but \delimiter needs unescaping. */
534 parse_escapes(match, match, strlen(match), i, i);
535 parse_escapes(replace, replace, strlen(replace), i, i);
537 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
538 for (i = 0; match[i] && replace[i]; i++) {
539 sed_cmd->string[i*2] = match[i];
540 sed_cmd->string[i*2+1] = replace[i];
545 /* if it wasnt a single-letter command that takes no arguments
546 * then it must be an invalid command.
548 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
549 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
552 /* give back whatever's left over */
557 /* Parse address+command sets, skipping comment lines. */
559 static void add_cmd(const char *cmdstr)
564 /* Append this line to any unfinished line from last time. */
565 if (G.add_cmd_line) {
566 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
567 free(G.add_cmd_line);
568 cmdstr = G.add_cmd_line = tp;
571 /* If this line ends with unescaped backslash, request next line. */
572 n = len = strlen(cmdstr);
573 while (n && cmdstr[n-1] == '\\')
575 if ((len - n) & 1) { /* if odd number of trailing backslashes */
577 G.add_cmd_line = xstrdup(cmdstr);
578 G.add_cmd_line[len-1] = '\0';
582 /* Loop parsing all commands in this line. */
584 /* Skip leading whitespace and semicolons */
585 cmdstr += strspn(cmdstr, semicolon_whitespace);
587 /* If no more commands, exit. */
590 /* if this is a comment, jump past it and keep going */
591 if (*cmdstr == '#') {
592 /* "#n" is the same as using -n on the command line */
593 if (cmdstr[1] == 'n')
595 cmdstr = strpbrk(cmdstr, "\n\r");
601 * format is: [addr][,addr][!]cmd
606 sed_cmd = xzalloc(sizeof(sed_cmd_t));
608 /* first part (if present) is an address: either a '$', a number or a /regex/ */
609 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
610 sed_cmd->beg_line_orig = sed_cmd->beg_line;
612 /* second part (if present) will begin with a comma */
613 if (*cmdstr == ',') {
617 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
619 bb_error_msg_and_die("no address after comma");
623 /* skip whitespace before the command */
624 cmdstr = skip_whitespace(cmdstr);
626 /* Check for inversion flag */
627 if (*cmdstr == '!') {
631 /* skip whitespace before the command */
632 cmdstr = skip_whitespace(cmdstr);
635 /* last part (mandatory) will be a command */
637 bb_error_msg_and_die("missing command");
638 sed_cmd->cmd = *cmdstr++;
639 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
641 /* Add the command to the command array */
642 *G.sed_cmd_tail = sed_cmd;
643 G.sed_cmd_tail = &sed_cmd->next;
646 /* If we glued multiple lines together, free the memory. */
647 free(G.add_cmd_line);
648 G.add_cmd_line = NULL;
651 /* Append to a string, reallocating memory as necessary. */
655 static void pipe_putc(char c)
657 if (G.pipeline.idx == G.pipeline.len) {
658 G.pipeline.buf = xrealloc(G.pipeline.buf,
659 G.pipeline.len + PIPE_GROW);
660 G.pipeline.len += PIPE_GROW;
662 G.pipeline.buf[G.pipeline.idx++] = c;
665 static void do_subst_w_backrefs(char *line, char *replace)
669 /* go through the replacement string */
670 for (i = 0; replace[i]; i++) {
671 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
672 if (replace[i] == '\\') {
673 unsigned backref = replace[++i] - '0';
675 /* print out the text held in G.regmatch[backref] */
676 if (G.regmatch[backref].rm_so != -1) {
677 j = G.regmatch[backref].rm_so;
678 while (j < G.regmatch[backref].rm_eo)
679 pipe_putc(line[j++]);
683 /* I _think_ it is impossible to get '\' to be
684 * the last char in replace string. Thus we dont check
685 * for replace[i] == NUL. (counterexample anyone?) */
686 /* if we find a backslash escaped character, print the character */
687 pipe_putc(replace[i]);
690 /* if we find an unescaped '&' print out the whole matched text. */
691 if (replace[i] == '&') {
692 j = G.regmatch[0].rm_so;
693 while (j < G.regmatch[0].rm_eo)
694 pipe_putc(line[j++]);
697 /* Otherwise just output the character. */
698 pipe_putc(replace[i]);
702 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
704 char *line = *line_p;
706 unsigned match_count = 0;
707 regex_t *current_regex;
709 current_regex = sed_cmd->sub_match;
710 /* Handle empty regex. */
711 if (!current_regex) {
712 current_regex = G.previous_regex_ptr;
714 bb_error_msg_and_die("no previous regexp");
716 G.previous_regex_ptr = current_regex;
718 /* Find the first match */
719 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0))
722 /* Initialize temporary output buffer. */
723 G.pipeline.buf = xmalloc(PIPE_GROW);
724 G.pipeline.len = PIPE_GROW;
727 /* Now loop through, substituting for matches */
731 /* Work around bug in glibc regexec, demonstrated by:
732 echo " a.b" | busybox sed 's [^ .]* x g'
733 The match_count check is so not to break
734 echo "hi" | busybox sed 's/^/!/g' */
735 if (!G.regmatch[0].rm_so && !G.regmatch[0].rm_eo && match_count) {
742 /* If we aren't interested in this match, output old line to
743 end of match and continue */
744 if (sed_cmd->which_match
745 && (sed_cmd->which_match != match_count)
747 for (i = 0; i < G.regmatch[0].rm_eo; i++)
752 /* print everything before the match */
753 for (i = 0; i < G.regmatch[0].rm_so; i++)
756 /* then print the substitution string */
757 do_subst_w_backrefs(line, sed_cmd->string);
759 /* advance past the match */
760 line += G.regmatch[0].rm_eo;
761 /* flag that something has changed */
764 /* if we're not doing this globally, get out now */
765 if (sed_cmd->which_match)
768 //maybe (G.regmatch[0].rm_eo ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
769 } while (*line && regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
771 /* Copy rest of string into output pipeline */
780 *line_p = G.pipeline.buf;
784 /* Set command pointer to point to this label. (Does not handle null label.) */
785 static sed_cmd_t *branch_to(char *label)
789 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
790 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
794 bb_error_msg_and_die("can't find label for jump to '%s'", label);
797 static void append(char *s)
799 llist_add_to_end(&G.append_head, xstrdup(s));
802 static void flush_append(void)
806 /* Output appended lines. */
807 while ((data = (char *)llist_pop(&G.append_head))) {
808 fprintf(G.nonstdout, "%s\n", data);
813 static void add_input_file(FILE *file)
815 G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count);
816 G.input_file_list[G.input_file_count++] = file;
819 /* Get next line of input from G.input_file_list, flushing append buffer and
820 * noting if we ran out of files without a newline on the last line we read.
826 static char *get_next_line(char *gets_char)
834 /* will be returned if last line in the file
835 * doesn't end with either '\n' or '\0' */
837 while (G.current_input_file < G.input_file_count) {
838 FILE *fp = G.input_file_list[G.current_input_file];
839 /* Read line up to a newline or NUL byte, inclusive,
840 * return malloc'ed char[]. length of the chunk read
841 * is stored in len. NULL if EOF/error */
842 temp = bb_get_chunk_from_file(fp, &len);
844 /* len > 0 here, it's ok to do temp[len-1] */
845 char c = temp[len-1];
846 if (c == '\n' || c == '\0') {
857 /* else we put NO_EOL_CHAR into *gets_char */
860 /* NB: I had the idea of peeking next file(s) and returning
861 * NO_EOL_CHAR only if it is the *last* non-empty
862 * input file. But there is a case where this won't work:
863 * file1: "a woo\nb woo"
864 * file2: "c no\nd no"
865 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
866 * (note: *no* newline after "b bang"!) */
868 /* Close this file and advance to next one */
870 G.current_input_file++;
876 /* Output line of text. */
878 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
879 * Without them, we had this:
883 * sed "s/i/z/" z1 z2 znull | hexdump -vC
886 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
888 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
890 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
892 char lpc = *last_puts_char;
894 /* Need to insert a '\n' between two files because first file's
895 * last line wasn't terminated? */
896 if (lpc != '\n' && lpc != '\0') {
902 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
906 /* had trailing '\0' and it was last char of file? */
907 if (last_gets_char == LAST_IS_NUL) {
911 /* had trailing '\n' or '\0'? */
912 if (last_gets_char != NO_EOL_CHAR) {
913 fputc(last_gets_char, file);
914 lpc = last_gets_char;
918 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
919 bb_error_msg_and_die(bb_msg_write_error);
921 *last_puts_char = lpc;
924 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
926 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
928 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
930 G.previous_regex_ptr = sed_cmd->beg_match;
934 /* Process all the lines in all the files */
936 static void process_files(void)
938 char *pattern_space, *next_line;
940 char last_puts_char = '\n';
941 char last_gets_char, next_gets_char;
946 next_line = get_next_line(&next_gets_char);
948 /* Go through every line in each file */
952 /* Advance to next line. Stop if out of lines. */
953 pattern_space = next_line;
956 last_gets_char = next_gets_char;
958 /* Read one line in advance so we can act on the last line,
960 next_line = get_next_line(&next_gets_char);
963 /* For every line, go through all the commands */
965 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
966 int old_matched, matched;
968 old_matched = sed_cmd->in_match;
970 /* Determine if this command matches this line: */
972 dbg("match1:%d", sed_cmd->in_match);
973 dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
974 && !sed_cmd->beg_match && !sed_cmd->end_match));
975 dbg("match3:%d", (sed_cmd->beg_line > 0
976 && (sed_cmd->end_line || sed_cmd->end_match
977 ? (sed_cmd->beg_line <= linenum)
978 : (sed_cmd->beg_line == linenum)
981 dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
982 dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
984 /* Are we continuing a previous multi-line match? */
985 sed_cmd->in_match = sed_cmd->in_match
986 /* Or is no range necessary? */
987 || (!sed_cmd->beg_line && !sed_cmd->end_line
988 && !sed_cmd->beg_match && !sed_cmd->end_match)
989 /* Or did we match the start of a numerical range? */
990 || (sed_cmd->beg_line > 0
991 && (sed_cmd->end_line || sed_cmd->end_match
992 /* note: even if end is numeric and is < linenum too,
993 * GNU sed matches! We match too, therefore we don't
994 * check here that linenum <= end.
996 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
997 * first three input lines are deleted;
998 * 4th line is matched and printed
999 * by "2,3" (!) and by "4" ranges
1001 ? (sed_cmd->beg_line <= linenum) /* N,end */
1002 : (sed_cmd->beg_line == linenum) /* N */
1005 /* Or does this line match our begin address regex? */
1006 || (beg_match(sed_cmd, pattern_space))
1007 /* Or did we match last line of input? */
1008 || (sed_cmd->beg_line == -1 && next_line == NULL);
1010 /* Snapshot the value */
1011 matched = sed_cmd->in_match;
1013 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1014 sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1016 /* Is this line the end of the current match? */
1019 /* once matched, "n,xxx" range is dead, disabling it */
1020 if (sed_cmd->beg_line > 0) {
1021 sed_cmd->beg_line = -2;
1023 sed_cmd->in_match = !(
1024 /* has the ending line come, or is this a single address command? */
1026 ? sed_cmd->end_line == -1
1028 : (sed_cmd->end_line <= linenum)
1029 : !sed_cmd->end_match
1031 /* or does this line matches our last address regex */
1032 || (sed_cmd->end_match && old_matched
1033 && (regexec(sed_cmd->end_match,
1034 pattern_space, 0, NULL, 0) == 0)
1039 /* Skip blocks of commands we didn't match */
1040 if (sed_cmd->cmd == '{') {
1041 if (sed_cmd->invert ? matched : !matched) {
1042 unsigned nest_cnt = 0;
1044 if (sed_cmd->cmd == '{')
1046 if (sed_cmd->cmd == '}') {
1051 sed_cmd = sed_cmd->next;
1053 bb_error_msg_and_die("unterminated {");
1059 /* Okay, so did this line match? */
1060 if (sed_cmd->invert ? matched : !matched)
1063 /* Update last used regex in case a blank substitute BRE is found */
1064 if (sed_cmd->beg_match) {
1065 G.previous_regex_ptr = sed_cmd->beg_match;
1068 /* actual sedding */
1069 //bb_error_msg("pattern_space:'%s' next_line:'%s' cmd:%c",
1070 //pattern_space, next_line, sed_cmd->cmd);
1071 switch (sed_cmd->cmd) {
1073 /* Print line number */
1075 fprintf(G.nonstdout, "%d\n", linenum);
1078 /* Write the current pattern space up to the first newline */
1081 char *tmp = strchr(pattern_space, '\n');
1084 /* TODO: explain why '\n' below */
1085 sed_puts(pattern_space, '\n');
1092 /* Write the current pattern space to output */
1094 /* NB: we print this _before_ the last line
1095 * (of current file) is printed. Even if
1096 * that line is nonterminated, we print
1097 * '\n' here (gnu sed does the same) */
1098 sed_puts(pattern_space, '\n');
1100 /* Delete up through first newline */
1103 char *tmp = strchr(pattern_space, '\n');
1105 overlapping_strcpy(pattern_space, tmp + 1);
1109 /* discard this line. */
1113 /* Substitute with regex */
1115 if (!do_subst_command(sed_cmd, &pattern_space))
1119 /* handle p option */
1121 sed_puts(pattern_space, last_gets_char);
1122 /* handle w option */
1123 if (sed_cmd->sw_file)
1125 pattern_space, sed_cmd->sw_file,
1126 &sed_cmd->sw_last_char, last_gets_char);
1129 /* Append line to linked list to be printed later */
1131 append(sed_cmd->string);
1134 /* Insert text before this line */
1136 sed_puts(sed_cmd->string, '\n');
1139 /* Cut and paste text (replace) */
1141 /* Only triggers on last line of a matching range. */
1142 if (!sed_cmd->in_match)
1143 sed_puts(sed_cmd->string, '\n');
1146 /* Read file, append contents to output */
1150 rfile = fopen_for_read(sed_cmd->string);
1154 while ((line = xmalloc_fgetline(rfile))
1157 xprint_and_close_file(rfile);
1163 /* Write pattern space to file. */
1166 pattern_space, sed_cmd->sw_file,
1167 &sed_cmd->sw_last_char, last_gets_char);
1170 /* Read next line from input */
1173 sed_puts(pattern_space, last_gets_char);
1175 free(pattern_space);
1176 pattern_space = next_line;
1177 last_gets_char = next_gets_char;
1178 next_line = get_next_line(&next_gets_char);
1185 /* Quit. End of script, end of input. */
1187 /* Exit the outer while loop */
1190 goto discard_commands;
1192 /* Append the next line to the current line */
1196 /* If no next line, jump to end of script and exit. */
1197 /* http://www.gnu.org/software/sed/manual/sed.html:
1198 * "Most versions of sed exit without printing anything
1199 * when the N command is issued on the last line of
1200 * a file. GNU sed prints pattern space before exiting
1201 * unless of course the -n command switch has been
1202 * specified. This choice is by design."
1204 if (next_line == NULL) {
1205 //goto discard_line;
1206 goto discard_commands; /* GNU behavior */
1208 /* Append next_line, read new next_line. */
1209 len = strlen(pattern_space);
1210 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1211 pattern_space[len] = '\n';
1212 strcpy(pattern_space + len+1, next_line);
1213 last_gets_char = next_gets_char;
1214 next_line = get_next_line(&next_gets_char);
1219 /* Test/branch if substitution occurred */
1221 if (!substituted) break;
1224 /* Test/branch if substitution didn't occur */
1226 if (substituted) break;
1228 /* Branch to label */
1230 if (!sed_cmd->string) goto discard_commands;
1231 else sed_cmd = branch_to(sed_cmd->string);
1233 /* Transliterate characters */
1237 for (i = 0; pattern_space[i]; i++) {
1238 for (j = 0; sed_cmd->string[j]; j += 2) {
1239 if (pattern_space[i] == sed_cmd->string[j]) {
1240 pattern_space[i] = sed_cmd->string[j + 1];
1248 case 'g': /* Replace pattern space with hold space */
1249 free(pattern_space);
1250 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1252 case 'G': /* Append newline and hold space to pattern space */
1254 int pattern_space_size = 2;
1255 int hold_space_size = 0;
1258 pattern_space_size += strlen(pattern_space);
1260 hold_space_size = strlen(G.hold_space);
1261 pattern_space = xrealloc(pattern_space,
1262 pattern_space_size + hold_space_size);
1263 if (pattern_space_size == 2)
1264 pattern_space[0] = 0;
1265 strcat(pattern_space, "\n");
1267 strcat(pattern_space, G.hold_space);
1268 last_gets_char = '\n';
1272 case 'h': /* Replace hold space with pattern space */
1274 G.hold_space = xstrdup(pattern_space);
1276 case 'H': /* Append newline and pattern space to hold space */
1278 int hold_space_size = 2;
1279 int pattern_space_size = 0;
1282 hold_space_size += strlen(G.hold_space);
1284 pattern_space_size = strlen(pattern_space);
1285 G.hold_space = xrealloc(G.hold_space,
1286 hold_space_size + pattern_space_size);
1288 if (hold_space_size == 2)
1290 strcat(G.hold_space, "\n");
1292 strcat(G.hold_space, pattern_space);
1296 case 'x': /* Exchange hold and pattern space */
1298 char *tmp = pattern_space;
1299 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1300 last_gets_char = '\n';
1305 } /* for each cmd */
1308 * Exit point from sedding...
1311 /* we will print the line unless we were told to be quiet ('-n')
1312 or if the line was suppressed (ala 'd'elete) */
1314 sed_puts(pattern_space, last_gets_char);
1316 /* Delete and such jump here. */
1319 free(pattern_space);
1324 /* It is possible to have a command line argument with embedded
1325 * newlines. This counts as multiple command lines.
1326 * However, newline can be escaped: 's/e/z\<newline>z/'
1327 * We check for this.
1330 static void add_cmd_block(char *cmdstr)
1334 cmdstr = sv = xstrdup(cmdstr);
1336 eol = strchr(cmdstr, '\n');
1339 /* Count preceding slashes */
1343 while (sl != cmdstr && *--sl == '\\')
1345 /* Odd number of preceding slashes - newline is escaped */
1347 overlapping_strcpy(eol - 1, eol);
1348 eol = strchr(eol, '\n');
1359 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1360 int sed_main(int argc UNUSED_PARAM, char **argv)
1363 llist_t *opt_e, *opt_f;
1364 int status = EXIT_SUCCESS;
1368 /* destroy command strings on exit */
1369 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1371 /* Lie to autoconf when it starts asking stupid questions. */
1372 if (argv[1] && !strcmp(argv[1], "--version")) {
1373 puts("This is not GNU sed version 4.0");
1377 /* do normal option parsing */
1378 opt_e = opt_f = NULL;
1379 opt_complementary = "e::f::" /* can occur multiple times */
1380 "nn"; /* count -n */
1381 /* -i must be first, to match OPT_in_place definition */
1382 opt = getopt32(argv, "irne:f:", &opt_e, &opt_f,
1383 &G.be_quiet); /* counter for -n */
1386 if (opt & OPT_in_place) { // -i
1387 atexit(cleanup_outname);
1389 if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r
1390 //if (opt & 0x4) G.be_quiet++; // -n
1391 while (opt_e) { // -e
1392 add_cmd_block(llist_pop(&opt_e));
1394 while (opt_f) { // -f
1397 cmdfile = xfopen_for_read(llist_pop(&opt_f));
1398 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1404 /* if we didn't get a pattern from -e or -f, use argv[0] */
1405 if (!(opt & 0x18)) {
1408 add_cmd_block(*argv++);
1410 /* Flush any unfinished commands. */
1413 /* By default, we write to stdout */
1414 G.nonstdout = stdout;
1416 /* argv[0..(argc-1)] should be names of file to process. If no
1417 * files were specified or '-' was specified, take input from stdin.
1418 * Otherwise, we process all the files specified. */
1419 if (argv[0] == NULL) {
1420 if (opt & OPT_in_place)
1421 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1422 add_input_file(stdin);
1426 for (i = 0; argv[i]; i++) {
1427 struct stat statbuf;
1432 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1433 add_input_file(stdin);
1437 file = fopen_or_warn(argv[i], "r");
1439 status = EXIT_FAILURE;
1442 add_input_file(file);
1443 if (!(opt & OPT_in_place)) {
1447 /* -i: process each FILE separately: */
1449 G.outname = xasprintf("%sXXXXXX", argv[i]);
1450 nonstdoutfd = xmkstemp(G.outname);
1451 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1453 /* Set permissions/owner of output file */
1454 fstat(fileno(file), &statbuf);
1455 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1456 * but GNU sed 4.2.1 does not preserve them either */
1457 fchmod(nonstdoutfd, statbuf.st_mode);
1458 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1461 fclose(G.nonstdout);
1462 G.nonstdout = stdout;
1464 /* unlink(argv[i]); */
1465 xrename(G.outname, argv[i]);
1469 /* Re-enable disabled range matches */
1470 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1471 sed_cmd->beg_line = sed_cmd->beg_line_orig;
1474 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1475 * if (G.current_input_file >= G.input_file_count)
1477 * but it's not needed since process_files() works correctly
1478 * in this case too. */