2 * sed.c - very minimalist version of sed
4 * Copyright (C) 1999,2000,2001 by Lineo, inc. and Mark Whitley
5 * Copyright (C) 1999,2000,2001 by Mark Whitley <markw@codepoet.org>
6 * Copyright (C) 2002 Matt Kraai
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 Supported features and commands in this version of sed:
28 - address matching: num|/matchstr/[,num|/matchstr/|$]command
29 - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
30 - edit commands: (a)ppend, (i)nsert, (c)hange
31 - file commands: (r)ead
32 - backreferences in substitution expressions (\1, \2...\9)
33 - grouped commands: {cmd1;cmd2}
35 (Note: Specifying an address (range) to match is *optional*; commands
36 default to the whole pattern space if no specific address match was
41 - transliteration (y/source-chars/dest-chars/) (use 'tr')
42 - no pattern space hold space storing / swapping (x, etc.)
43 - no labels / branching (: label, b, t, and friends)
44 - and lots, lots more.
46 Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
50 #include <unistd.h> /* for getopt() */
52 #include <string.h> /* for strdup() */
54 #include <ctype.h> /* for isspace() */
59 extern void xregcomp(regex_t *preg, const char *regex, int cflags);
60 extern int optind; /* in unistd.h */
61 extern char *optarg; /* ditto */
64 static int be_quiet = 0;
67 typedef struct sed_cmd_s {
68 /* Order by alignment requirements */
71 regex_t *beg_match; /* sed -e '/match/cmd' */
72 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
74 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
76 /* sed -e 's/sub_match/replace/' */
80 /* EDIT COMMAND (a,i,c) SPECIFIC FIELDS */
83 /* FILE COMMAND (r) SPECIFIC FIELDS */
87 int beg_line; /* 'sed 1p' 0 == no begining line, apply commands to all lines */
88 int end_line; /* 'sed 1,3p' 0 == no end line, use only beginning. -1 == $ */
89 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
91 unsigned int num_backrefs:4; /* how many back references (\1..\9) */
92 /* Note: GNU/POSIX sed does not save more than nine backrefs, so
93 * we only use 4 bits to hold the number */
94 unsigned int sub_g:1; /* sed -e 's/foo/bar/g' (global) */
95 unsigned int sub_p:2; /* sed -e 's/foo/bar/p' (print substitution) */
98 char delimiter; /* The delimiter used to separate regexps */
101 char cmd; /* p,d,s (add more at your leisure :-) */
104 int invert; /* the '!' after the address */
108 static sed_cmd_t **sed_cmds = NULL; /* growable arrary holding a sequence of sed cmds */
109 static int ncmds = 0; /* number of sed commands */
111 /*static char *cur_file = NULL;*/ /* file currently being processed XXX: do I need this? */
113 const char * const semicolon_whitespace = "; \n\r\t\v\0";
115 #ifdef CONFIG_FEATURE_CLEAN_UP
116 static void destroy_cmd_strs(void)
118 if (sed_cmds == NULL)
121 /* destroy all the elements in the array */
122 while (--ncmds >= 0) {
124 if (sed_cmds[ncmds]->beg_match) {
125 regfree(sed_cmds[ncmds]->beg_match);
126 free(sed_cmds[ncmds]->beg_match);
128 if (sed_cmds[ncmds]->end_match) {
129 regfree(sed_cmds[ncmds]->end_match);
130 free(sed_cmds[ncmds]->end_match);
132 if (sed_cmds[ncmds]->sub_match) {
133 regfree(sed_cmds[ncmds]->sub_match);
134 free(sed_cmds[ncmds]->sub_match);
136 free(sed_cmds[ncmds]->replace);
139 /* destroy the array */
147 * index_of_next_unescaped_regexp_delim - walks left to right through a string
148 * beginning at a specified index and returns the index of the next regular
149 * expression delimiter (typically a forward * slash ('/')) not preceeded by
152 static int index_of_next_unescaped_regexp_delim(const char delimiter, const char *str, int idx)
158 for ( ; (ch = str[idx]); idx++) {
160 if (ch == ']' && !(bracket == idx - 1 ||
161 (bracket == idx - 2 && str[idx-1] == '^')))
169 else if (ch == delimiter)
173 /* if we make it to here, we've hit the end of the string */
178 * returns the index in the string just past where the address ends.
180 static int get_address(char *delimiter, char *my_str, int *linenum, regex_t **regex)
183 if (isdigit(my_str[idx])) {
185 *linenum = strtol(my_str, &endstr, 10);
186 /* endstr shouldnt ever equal NULL */
187 idx = endstr - my_str;
189 else if (my_str[idx] == '$') {
193 else if (my_str[idx] == '/' || my_str[idx] == '\\') {
197 if (my_str[idx] == '\\') {
199 *delimiter = my_str[++idx];
201 idx = index_of_next_unescaped_regexp_delim(*delimiter, my_str, ++idx);
203 error_msg_and_die("unterminated match expression");
206 *regex = (regex_t *)xmalloc(sizeof(regex_t));
207 xregcomp(*regex, my_str+idx_start, REG_NEWLINE);
208 idx++; /* so it points to the next character after the last '/' */
213 static int parse_subst_cmd(sed_cmd_t * const sed_cmd, const char *substr)
215 int oldidx, cflags = REG_NEWLINE;
221 * the string that gets passed to this function should look like this:
222 * s/match/replace/gIp
226 * (all three of the '/' slashes are mandatory)
229 /* verify that the 's' is followed by something. That something
230 * (typically a 'slash') is now our regexp delimiter... */
231 if (substr[idx] == '\0')
232 error_msg_and_die("bad format in substitution expression");
234 sed_cmd->delimiter=substr[idx];
236 /* save the match string */
238 idx = index_of_next_unescaped_regexp_delim(sed_cmd->delimiter, substr, ++idx);
240 error_msg_and_die("bad format in substitution expression");
241 match = xstrndup(substr + oldidx, idx - oldidx);
243 /* determine the number of back references in the match string */
244 /* Note: we compute this here rather than in the do_subst_command()
245 * function to save processor time, at the expense of a little more memory
246 * (4 bits) per sed_cmd */
248 /* sed_cmd->num_backrefs = 0; */ /* XXX: not needed? --apparently not */
249 for (j = 0; match[j]; j++) {
250 /* GNU/POSIX sed does not save more than nine backrefs */
251 if (match[j] == '\\' && match[j+1] == '(' && sed_cmd->num_backrefs <= 9)
252 sed_cmd->num_backrefs++;
255 /* save the replacement string */
257 idx = index_of_next_unescaped_regexp_delim(sed_cmd->delimiter, substr, ++idx);
259 error_msg_and_die("bad format in substitution expression");
260 sed_cmd->replace = xstrndup(substr + oldidx, idx - oldidx);
262 /* process the flags */
263 while (substr[++idx]) {
264 switch (substr[idx]) {
268 /* Hmm, i dont see the I option mentioned in the standard */
276 /* any whitespace or semicolon trailing after a s/// is ok */
277 if (strchr(semicolon_whitespace, substr[idx]))
280 error_msg_and_die("bad option in substitution expression");
285 /* compile the match string into a regex */
286 sed_cmd->sub_match = (regex_t *)xmalloc(sizeof(regex_t));
287 xregcomp(sed_cmd->sub_match, match, cflags);
293 static int parse_edit_cmd(sed_cmd_t *sed_cmd, const char *editstr)
298 * the string that gets passed to this function should look like this:
302 * | this backslash (immediately following the edit command) is mandatory
309 * as soon as we hit a TEXT line that has no trailing '\', we're done.
310 * this means a command like:
318 if ((*editstr != '\\') || ((editstr[1] != '\n') && (editstr[1] != '\r'))) {
319 error_msg_and_die("bad format in edit expression");
322 /* store the edit line text */
323 sed_cmd->editline = xmalloc(strlen(&editstr[2]) + 2);
324 for (i = 2, j = 0; editstr[i] != '\0' && strchr("\r\n", editstr[i]) == NULL;
326 if ((editstr[i] == '\\') && strchr("\n\r", editstr[i+1]) != NULL) {
327 sed_cmd->editline[j] = '\n';
330 sed_cmd->editline[j] = editstr[i];
333 /* figure out if we need to add a newline */
334 if (sed_cmd->editline[j-1] != '\n')
335 sed_cmd->editline[j++] = '\n';
337 /* terminate string */
338 sed_cmd->editline[j] = '\0';
344 static int parse_file_cmd(sed_cmd_t *sed_cmd, const char *filecmdstr)
350 * the string that gets passed to this function should look like this:
355 * optional whitespace
357 * re: the file to be read, the GNU manual says the following: "Note that
358 * if filename cannot be read, it is treated as if it were an empty file,
359 * without any error indication." Thus, all of the following commands are
362 * sed -e '1r noexist'
367 /* the file command may be followed by whitespace; move past it. */
368 while (isspace(filecmdstr[++idx]))
371 /* the first non-whitespace we get is a filename. the filename ends when we
372 * hit a normal sed command terminator or end of string */
373 filenamelen = strcspn(&filecmdstr[idx], semicolon_whitespace);
374 sed_cmd->filename = xmalloc(filenamelen + 1);
375 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
377 return idx + filenamelen;
381 * Process the commands arguments
383 static char *parse_cmd_str(sed_cmd_t * const sed_cmd, char *cmdstr)
385 /* handle (s)ubstitution command */
386 if (sed_cmd->cmd == 's') {
387 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
389 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
390 else if (strchr("aic", sed_cmd->cmd)) {
391 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
392 error_msg_and_die("only a beginning address can be specified for edit commands");
393 cmdstr += parse_edit_cmd(sed_cmd, cmdstr);
395 /* handle file cmds: (r)ead */
396 else if (sed_cmd->cmd == 'r') {
397 if (sed_cmd->end_line || sed_cmd->end_match)
398 error_msg_and_die("Command only uses one address");
399 cmdstr += parse_file_cmd(sed_cmd, cmdstr);
401 /* if it wasnt a single-letter command that takes no arguments
402 * then it must be an invalid command.
404 else if (strchr("nNpPqd=", sed_cmd->cmd) == 0) {
405 error_msg_and_die("Unsupported command %c", sed_cmd->cmd);
408 /* give back whatever's left over */
412 static char *add_cmd(sed_cmd_t *sed_cmd, char *cmdstr)
415 /* Skip over leading whitespace and semicolons */
416 cmdstr += strspn(cmdstr, semicolon_whitespace);
418 /* if we ate the whole thing, that means there was just trailing
419 * whitespace or a final / no-op semicolon. either way, get out */
420 if (*cmdstr == '\0') {
424 /* if this is a comment, jump past it and keep going */
425 if (*cmdstr == '#') {
426 return(strpbrk(cmdstr, "\n\r"));
430 * format is: [addr][,addr]cmd
435 /* first part (if present) is an address: either a '$', a number or a /regex/ */
436 cmdstr += get_address(&(sed_cmd->delimiter), cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
438 /* second part (if present) will begin with a comma */
439 if (*cmdstr == ',') {
442 idx = get_address(&(sed_cmd->delimiter), cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
444 error_msg_and_die("get_address: no address found in string\n"
445 "\t(you probably didn't check the string you passed me)");
450 /* skip whitespace before the command */
451 while (isspace(*cmdstr)) {
455 /* there my be the inversion flag between part2 and part3 */
456 if (*cmdstr == '!') {
460 #ifdef SED_FEATURE_STRICT_CHECKING
461 /* According to the spec
462 * It is unspecified whether <blank>s can follow a '!' character,
463 * and conforming applications shall not follow a '!' character
466 if (isblank(cmdstr[idx]) {
467 error_msg_and_die("blank follows '!'");
470 /* skip whitespace before the command */
471 while (isspace(*cmdstr)) {
478 /* last part (mandatory) will be a command */
480 error_msg_and_die("missing command");
482 sed_cmd->cmd = *cmdstr;
485 if (sed_cmd->cmd == '{') {
487 char *end_ptr = strpbrk(cmdstr, ";}");
489 add_cmd(sed_cmd, cmdstr);
490 cmdstr = end_ptr + 1;
491 } while (*cmdstr != '\0');
494 cmdstr = parse_cmd_str(sed_cmd, cmdstr);
496 /* Add the command to the command array */
497 sed_cmds = xrealloc(sed_cmds, sizeof(sed_cmd_t) * (++ncmds));
498 sed_cmds[ncmds-1] = xmalloc(sizeof(sed_cmd_t));
499 memcpy(sed_cmds[ncmds-1], sed_cmd, sizeof(sed_cmd_t));
504 static void add_cmd_str(char *cmdstr)
508 sed_cmd = xcalloc(1, sizeof(sed_cmd_t));
509 cmdstr = add_cmd(sed_cmd, cmdstr);
510 } while (cmdstr && strlen(cmdstr));
514 static void load_cmd_file(char *filename)
520 cmdfile = xfopen(filename, "r");
522 while ((line = get_line_from_file(cmdfile)) != NULL) {
523 /* if a line ends with '\' it needs the next line appended to it */
524 while (line[strlen(line)-2] == '\\' &&
525 (nextline = get_line_from_file(cmdfile)) != NULL) {
526 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
527 strcat(line, nextline);
530 /* eat trailing newline (if any) --if I don't do this, edit commands
531 * (aic) will print an extra newline */
544 #define PIPE_MAGIC 0x7f
547 void pipe_putc(struct pipeline *const pipeline, char c)
549 if (pipeline->buf[pipeline->idx] == PIPE_MAGIC) {
551 xrealloc(pipeline->buf, pipeline->len + PIPE_GROW);
552 memset(pipeline->buf + pipeline->len, 0, PIPE_GROW);
553 pipeline->len += PIPE_GROW;
554 pipeline->buf[pipeline->len - 1] = PIPE_MAGIC;
556 pipeline->buf[pipeline->idx++] = (c);
559 #define pipeputc(c) pipe_putc(pipeline, c)
562 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
563 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
564 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
565 pipeline_len += PIPE_GROW; \
566 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
567 pipeline[pipeline_idx++] = (c); }
570 static void print_subst_w_backrefs(const char *line, const char *replace,
571 regmatch_t *regmatch, struct pipeline *const pipeline, int matches)
575 /* go through the replacement string */
576 for (i = 0; replace[i]; i++) {
577 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
578 if (replace[i] == '\\' && isdigit(replace[i+1])) {
582 ++i; /* i now indexes the backref number, instead of the leading slash */
583 tmpstr[0] = replace[i];
585 backref = atoi(tmpstr);
586 /* print out the text held in regmatch[backref] */
587 if (backref <= matches && regmatch[backref].rm_so != -1)
588 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
592 /* if we find a backslash escaped character, print the character */
593 else if (replace[i] == '\\') {
595 pipeputc(replace[i]);
598 /* if we find an unescaped '&' print out the whole matched text.
599 * fortunately, regmatch[0] contains the indicies to the whole matched
600 * expression (kinda seems like it was designed for just such a
602 else if (replace[i] == '&' && replace[i-1] != '\\') {
604 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
607 /* nothing special, just print this char of the replacement string to stdout */
609 pipeputc(replace[i]);
613 static int do_subst_command(const sed_cmd_t *sed_cmd, char **line)
615 char *hackline = *line;
616 struct pipeline thepipe = { NULL, 0 , 0};
617 struct pipeline *const pipeline = &thepipe;
619 regmatch_t *regmatch = NULL;
621 /* we only proceed if the substitution 'search' expression matches */
622 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
625 /* whaddaya know, it matched. get the number of back references */
626 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
628 /* allocate more PIPE_GROW bytes
629 if replaced string is larger than original */
630 thepipe.len = strlen(hackline)+PIPE_GROW;
631 thepipe.buf = xcalloc(1, thepipe.len);
633 thepipe.buf[thepipe.len-1] = PIPE_MAGIC;
635 /* and now, as long as we've got a line to try matching and if we can match
636 * the search string, we make substitutions */
637 while ((*hackline || !altered) && (regexec(sed_cmd->sub_match, hackline,
638 sed_cmd->num_backrefs+1, regmatch, 0) != REG_NOMATCH) ) {
641 /* print everything before the match */
642 for (i = 0; i < regmatch[0].rm_so; i++)
643 pipeputc(hackline[i]);
645 /* then print the substitution string */
646 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
647 pipeline, sed_cmd->num_backrefs);
649 /* advance past the match */
650 hackline += regmatch[0].rm_eo;
651 /* flag that something has changed */
654 /* if we're not doing this globally, get out now */
659 for (; *hackline; hackline++) pipeputc(*hackline);
660 if (thepipe.buf[thepipe.idx] == PIPE_MAGIC) thepipe.buf[thepipe.idx] = 0;
671 static void process_file(FILE *file)
674 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
675 unsigned int still_in_range = 0;
679 line = get_line_from_file(file);
685 /* go through every line in the file */
689 /* Read one line in advance so we can act on the last line, the '$' address */
690 next_line = get_line_from_file(file);
696 /* for every line, go through all the commands */
697 for (i = 0; i < ncmds; i++) {
698 sed_cmd_t *sed_cmd = sed_cmds[i];
702 * entry point into sedding...
705 /* no range necessary */
706 (sed_cmd->beg_line == 0 && sed_cmd->end_line == 0 &&
707 sed_cmd->beg_match == NULL &&
708 sed_cmd->end_match == NULL) ||
709 /* this line number is the first address we're looking for */
710 (sed_cmd->beg_line && (sed_cmd->beg_line == linenum)) ||
711 /* this line matches our first address regex */
712 (sed_cmd->beg_match && (regexec(sed_cmd->beg_match, line, 0, NULL, 0) == 0)) ||
713 /* we are currently within the beginning & ending address range */
714 still_in_range || ((sed_cmd->beg_line == -1) && (next_line == NULL))
717 if (sed_cmd->invert ^ matched) {
722 switch (sed_cmd->cmd) {
724 printf("%d\n", linenum);
726 case 'P': { /* Write the current pattern space upto the first newline */
727 char *tmp = strchr(line, '\n');
732 case 'p': /* Write the current pattern space to output */
743 * Some special cases for 's' printing to make it compliant with
744 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
746 * -n ONLY = never print anything regardless of any successful
749 * s///p ONLY = always print successful substitutions, even if
750 * the line is going to be printed anyway (line will be printed
753 * -n AND s///p = print ONLY a successful substitution ONE TIME;
754 * no other lines are printed - this is the reason why the 'p'
755 * flag exists in the first place.
758 /* if the user specified that they didn't want anything printed (i.e., a -n
759 * flag and no 'p' flag after the s///), then there's really no point doing
761 if (be_quiet && !sed_cmd->sub_p)
764 /* we print the line once, unless we were told to be quiet */
766 altered |= do_subst_command(sed_cmd, &line);
768 /* we also print the line if we were given the 'p' flag
769 * (this is quite possibly the second printing) */
771 altered |= do_subst_command(sed_cmd, &line);
772 if (altered && (i+1 >= ncmds || sed_cmds[i+1]->cmd != 's'))
779 fputs(sed_cmd->editline, stdout);
784 fputs(sed_cmd->editline, stdout);
788 /* single-address case */
789 if ((sed_cmd->end_match == NULL && sed_cmd->end_line == 0)
790 /* multi-address case */
791 /* - matching text */
792 || (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
793 /* - matching line numbers */
794 || (sed_cmd->end_line > 0 && sed_cmd->end_line == linenum))
796 fputs(sed_cmd->editline, stdout);
805 outfile = fopen(sed_cmd->filename, "r");
808 /* else if we couldn't open the output file,
809 * no biggie, just don't print anything */
813 case 'q': /* Branch to end of script and quit */
816 case 'n': /* Read next line from input */
819 next_line = get_line_from_file(file);
823 case 'N': /* Append the next line to the current line */
824 line = realloc(line, strlen(line) + strlen(next_line) + 2);
826 strcat(line, next_line);
827 next_line = get_line_from_file(file);
834 * exit point from sedding...
838 /* this is a single-address command or... */
839 (sed_cmd->end_line == 0 && sed_cmd->end_match == NULL) || (
840 /* we were in the middle of our address range (this
841 * isn't the first time through) and.. */
842 (still_in_range == 1) && (
843 /* this line number is the last address we're looking for or... */
844 (sed_cmd->end_line && (sed_cmd->end_line == linenum)) ||
845 /* this line matches our last address regex */
846 (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
850 /* we're out of our address range */
854 /* didn't hit the exit? then we're still in the middle of an address range */
864 /* we will print the line unless we were told to be quiet or if the
865 * line was altered (via a 'd'elete or 's'ubstitution), in which case
866 * the altered line was already printed */
867 if (!be_quiet && !altered)
875 extern int sed_main(int argc, char **argv)
877 int opt, status = EXIT_SUCCESS;
879 #ifdef CONFIG_FEATURE_CLEAN_UP
880 /* destroy command strings on exit */
881 if (atexit(destroy_cmd_strs) == -1)
882 perror_msg_and_die("atexit");
885 /* do normal option parsing */
886 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
895 load_cmd_file(optarg);
902 /* if we didn't get a pattern from a -e and no command file was specified,
903 * argv[optind] should be the pattern. no pattern, no worky */
905 if (argv[optind] == NULL)
908 add_cmd_str(argv[optind]);
913 /* argv[(optind)..(argc-1)] should be names of file to process. If no
914 * files were specified or '-' was specified, take input from stdin.
915 * Otherwise, we process all the files specified. */
916 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
922 for (i = optind; i < argc; i++) {
923 file = wfopen(argv[i], "r");
928 status = EXIT_FAILURE;