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)
34 (Note: Specifying an address (range) to match is *optional*; commands
35 default to the whole pattern space if no specific address match was
40 - transliteration (y/source-chars/dest-chars/) (use 'tr')
41 - no pattern space hold space storing / swapping (x, etc.)
42 - no labels / branching (: label, b, t, and friends)
43 - and lots, lots more.
47 #include <unistd.h> /* for getopt() */
49 #include <string.h> /* for strdup() */
51 #include <ctype.h> /* for isspace() */
56 extern void xregcomp(regex_t *preg, const char *regex, int cflags);
57 extern int optind; /* in unistd.h */
58 extern char *optarg; /* ditto */
61 static int be_quiet = 0;
65 /* Order by alignment requirements */
68 regex_t *beg_match; /* sed -e '/match/cmd' */
69 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
71 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
73 /* sed -e 's/sub_match/replace/' */
77 /* EDIT COMMAND (a,i,c) SPECIFIC FIELDS */
80 /* FILE COMMAND (r) SPECIFIC FIELDS */
84 int beg_line; /* 'sed 1p' 0 == no begining line, apply commands to all lines */
85 int end_line; /* 'sed 1,3p' 0 == no end line, use only beginning. -1 == $ */
86 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
88 unsigned int num_backrefs:4; /* how many back references (\1..\9) */
89 /* Note: GNU/POSIX sed does not save more than nine backrefs, so
90 * we only use 4 bits to hold the number */
91 unsigned int sub_g:1; /* sed -e 's/foo/bar/g' (global) */
92 unsigned int sub_p:2; /* sed -e 's/foo/bar/p' (print substitution) */
95 char delimiter; /* The delimiter used to separate regexps */
98 char cmd; /* p,d,s (add more at your leisure :-) */
101 int invert; /* the '!' after the address */
105 static struct sed_cmd *sed_cmds = NULL; /* growable arrary holding a sequence of sed cmds */
106 static int ncmds = 0; /* number of sed commands */
108 /*static char *cur_file = NULL;*/ /* file currently being processed XXX: do I need this? */
110 const char * const semicolon_whitespace = "; \n\r\t\v\0";
112 #ifdef CONFIG_FEATURE_CLEAN_UP
113 static void destroy_cmd_strs(void)
115 if (sed_cmds == NULL)
118 /* destroy all the elements in the array */
119 while (--ncmds >= 0) {
121 if (sed_cmds[ncmds].beg_match) {
122 regfree(sed_cmds[ncmds].beg_match);
123 free(sed_cmds[ncmds].beg_match);
125 if (sed_cmds[ncmds].end_match) {
126 regfree(sed_cmds[ncmds].end_match);
127 free(sed_cmds[ncmds].end_match);
129 if (sed_cmds[ncmds].sub_match) {
130 regfree(sed_cmds[ncmds].sub_match);
131 free(sed_cmds[ncmds].sub_match);
133 if (sed_cmds[ncmds].replace)
134 free(sed_cmds[ncmds].replace);
137 /* destroy the array */
145 * index_of_next_unescaped_regexp_delim - walks left to right through a string
146 * beginning at a specified index and returns the index of the next regular
147 * expression delimiter (typically a forward * slash ('/')) not preceeded by
150 static int index_of_next_unescaped_regexp_delim(const struct sed_cmd * const sed_cmd, const char *str, int idx)
156 for ( ; (ch = str[idx]); idx++) {
158 if (ch == ']' && !(bracket == idx - 1 ||
159 (bracket == idx - 2 && str[idx-1] == '^')))
167 else if (ch == sed_cmd->delimiter)
171 /* if we make it to here, we've hit the end of the string */
176 * returns the index in the string just past where the address ends.
178 static int get_address(struct sed_cmd *sed_cmd, const char *str, int *linenum, regex_t **regex)
180 char *my_str = xstrdup(str);
183 olddelimiter = sed_cmd->delimiter;
184 sed_cmd->delimiter = '/';
186 if (isdigit(my_str[idx])) {
189 } while (isdigit(my_str[idx]));
191 *linenum = atoi(my_str);
193 else if (my_str[idx] == '$') {
197 else if (my_str[idx] == '/' || my_str[idx] == '\\') {
198 if (my_str[idx] == '\\') {
200 sed_cmd-> delimiter = my_str[++idx];
202 idx = index_of_next_unescaped_regexp_delim(sed_cmd, my_str, ++idx);
204 error_msg_and_die("unterminated match expression");
206 *regex = (regex_t *)xmalloc(sizeof(regex_t));
207 xregcomp(*regex, my_str+1, REG_NEWLINE);
208 idx++; /* so it points to the next character after the last '/' */
211 error_msg("get_address: no address found in string\n"
212 "\t(you probably didn't check the string you passed me)");
217 sed_cmd->delimiter = olddelimiter;
221 static int parse_subst_cmd(struct sed_cmd * const sed_cmd, const char *substr)
223 int oldidx, cflags = REG_NEWLINE;
229 * the string that gets passed to this function should look like this:
230 * s/match/replace/gIp
234 * (all three of the '/' slashes are mandatory)
237 /* verify that the 's' is followed by something. That something
238 * (typically a 'slash') is now our regexp delimiter... */
240 error_msg_and_die("bad format in substitution expression");
242 sed_cmd->delimiter=substr[idx];
244 /* save the match string */
246 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
248 error_msg_and_die("bad format in substitution expression");
249 match = xstrndup(substr + oldidx, idx - oldidx);
251 /* determine the number of back references in the match string */
252 /* Note: we compute this here rather than in the do_subst_command()
253 * function to save processor time, at the expense of a little more memory
254 * (4 bits) per sed_cmd */
256 /* sed_cmd->num_backrefs = 0; */ /* XXX: not needed? --apparently not */
257 for (j = 0; match[j]; j++) {
258 /* GNU/POSIX sed does not save more than nine backrefs */
259 if (match[j] == '\\' && match[j+1] == '(' && sed_cmd->num_backrefs <= 9)
260 sed_cmd->num_backrefs++;
263 /* save the replacement string */
265 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
267 error_msg_and_die("bad format in substitution expression");
268 sed_cmd->replace = xstrndup(substr + oldidx, idx - oldidx);
270 /* process the flags */
271 while (substr[++idx]) {
272 switch (substr[idx]) {
283 /* any whitespace or semicolon trailing after a s/// is ok */
284 if (strchr(semicolon_whitespace, substr[idx]))
287 error_msg_and_die("bad option in substitution expression");
292 /* compile the match string into a regex */
293 sed_cmd->sub_match = (regex_t *)xmalloc(sizeof(regex_t));
294 xregcomp(sed_cmd->sub_match, match, cflags);
300 static void move_back(char *str, int offset)
302 memmove(str, str + offset, strlen(str + offset) + 1);
305 static int parse_edit_cmd(struct sed_cmd *sed_cmd, const char *editstr)
310 * the string that gets passed to this function should look like this:
314 * | this backslash (immediately following the edit command) is mandatory
321 * as soon as we hit a TEXT line that has no trailing '\', we're done.
322 * this means a command like:
331 if (editstr[1] != '\\' || (editstr[2] != '\n' && editstr[2] != '\r'))
332 error_msg_and_die("bad format in edit expression");
334 /* store the edit line text */
335 sed_cmd->editline = xmalloc(strlen(&editstr[3]) + 2);
336 for (i = 3, j = 0; editstr[i] != '\0' && strchr("\r\n", editstr[i]) == NULL;
338 if (editstr[i] == '\\' && strchr("\n\r", editstr[i+1]) != NULL) {
339 sed_cmd->editline[j] = '\n';
342 sed_cmd->editline[j] = editstr[i];
345 /* figure out if we need to add a newline */
346 if (sed_cmd->editline[j-1] != '\n')
347 sed_cmd->editline[j++] = '\n';
349 /* terminate string */
350 sed_cmd->editline[j] = '\0';
356 static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
362 * the string that gets passed to this function should look like this:
367 * optional whitespace
369 * re: the file to be read, the GNU manual says the following: "Note that
370 * if filename cannot be read, it is treated as if it were an empty file,
371 * without any error indication." Thus, all of the following commands are
374 * sed -e '1r noexist'
379 /* the file command may be followed by whitespace; move past it. */
380 while (isspace(filecmdstr[++idx]))
383 /* the first non-whitespace we get is a filename. the filename ends when we
384 * hit a normal sed command terminator or end of string */
385 filenamelen = strcspn(&filecmdstr[idx], semicolon_whitespace);
386 sed_cmd->filename = xmalloc(filenamelen + 1);
387 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
389 return idx + filenamelen;
393 static char *parse_cmd_str(struct sed_cmd * const sed_cmd, const char *const cmdstr)
398 * format is: [addr][,addr]cmd
403 /* first part (if present) is an address: either a number or a /regex/ */
404 if (isdigit(cmdstr[idx]) || cmdstr[idx] == '/')
405 idx = get_address(sed_cmd, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
407 /* second part (if present) will begin with a comma */
408 if (cmdstr[idx] == ',') {
410 idx += get_address(sed_cmd, &cmdstr[idx], &sed_cmd->end_line, &sed_cmd->end_match);
413 /* skip whitespace before the command */
414 while (isspace(cmdstr[idx]))
417 /* there my be the inversion flag between part2 and part3 */
419 if (cmdstr[idx] == '!') {
423 /* skip whitespace before the command */
424 while (isspace(cmdstr[idx]))
428 /* last part (mandatory) will be a command */
429 if (cmdstr[idx] == '\0')
430 error_msg_and_die("missing command");
431 sed_cmd->cmd = cmdstr[idx];
433 /* if it was a single-letter command that takes no arguments (such as 'p'
434 * or 'd') all we need to do is increment the index past that command */
435 if (strchr("pd", sed_cmd->cmd)) {
438 /* handle (s)ubstitution command */
439 else if (sed_cmd->cmd == 's') {
440 idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
442 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
443 else if (strchr("aic", sed_cmd->cmd)) {
444 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
445 error_msg_and_die("only a beginning address can be specified for edit commands");
446 idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
448 /* handle file cmds: (r)ead */
449 else if (sed_cmd->cmd == 'r') {
450 if (sed_cmd->end_line || sed_cmd->end_match)
451 error_msg_and_die("Command only uses one address");
452 idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
455 error_msg_and_die("invalid command");
458 /* give back whatever's left over */
459 return (char *)&cmdstr[idx];
462 static void add_cmd_str(const char * const cmdstr)
464 char *mystr = (char *)cmdstr;
468 /* trim leading whitespace and semicolons */
469 move_back(mystr, strspn(mystr, semicolon_whitespace));
470 /* if we ate the whole thing, that means there was just trailing
471 * whitespace or a final / no-op semicolon. either way, get out */
472 if (strlen(mystr) == 0)
474 /* if this is a comment, jump past it and keep going */
475 if (mystr[0] == '#') {
476 mystr = strpbrk(mystr, "\n\r");
480 sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
481 /* zero new element */
482 memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
483 /* load command string into new array element, get remainder */
484 mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
486 } while (mystr && strlen(mystr));
490 static void load_cmd_file(char *filename)
496 cmdfile = xfopen(filename, "r");
498 while ((line = get_line_from_file(cmdfile)) != NULL) {
499 /* if a line ends with '\' it needs the next line appended to it */
500 while (line[strlen(line)-2] == '\\' &&
501 (nextline = get_line_from_file(cmdfile)) != NULL) {
502 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
503 strcat(line, nextline);
506 /* eat trailing newline (if any) --if I don't do this, edit commands
507 * (aic) will print an extra newline */
520 #define PIPE_MAGIC 0x7f
523 void pipe_putc(struct pipeline *const pipeline, char c)
525 if (pipeline->buf[pipeline->idx] == PIPE_MAGIC) {
527 xrealloc(pipeline->buf, pipeline->len + PIPE_GROW);
528 memset(pipeline->buf + pipeline->len, 0, PIPE_GROW);
529 pipeline->len += PIPE_GROW;
530 pipeline->buf[pipeline->len - 1] = PIPE_MAGIC;
532 pipeline->buf[pipeline->idx++] = (c);
535 #define pipeputc(c) pipe_putc(pipeline, c)
538 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
539 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
540 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
541 pipeline_len += PIPE_GROW; \
542 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
543 pipeline[pipeline_idx++] = (c); }
546 static void print_subst_w_backrefs(const char *line, const char *replace,
547 regmatch_t *regmatch, struct pipeline *const pipeline, int matches)
551 /* go through the replacement string */
552 for (i = 0; replace[i]; i++) {
553 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
554 if (replace[i] == '\\' && isdigit(replace[i+1])) {
558 ++i; /* i now indexes the backref number, instead of the leading slash */
559 tmpstr[0] = replace[i];
561 backref = atoi(tmpstr);
562 /* print out the text held in regmatch[backref] */
563 if (backref <= matches && regmatch[backref].rm_so != -1)
564 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
568 /* if we find a backslash escaped character, print the character */
569 else if (replace[i] == '\\') {
571 pipeputc(replace[i]);
574 /* if we find an unescaped '&' print out the whole matched text.
575 * fortunately, regmatch[0] contains the indicies to the whole matched
576 * expression (kinda seems like it was designed for just such a
578 else if (replace[i] == '&' && replace[i-1] != '\\') {
580 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
583 /* nothing special, just print this char of the replacement string to stdout */
585 pipeputc(replace[i]);
589 static int do_subst_command(const struct sed_cmd *sed_cmd, char **line)
591 char *hackline = *line;
592 struct pipeline thepipe = { NULL, 0 , 0};
593 struct pipeline *const pipeline = &thepipe;
595 regmatch_t *regmatch = NULL;
597 /* we only proceed if the substitution 'search' expression matches */
598 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
601 /* whaddaya know, it matched. get the number of back references */
602 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
604 /* allocate more PIPE_GROW bytes
605 if replaced string is larger than original */
606 thepipe.len = strlen(hackline)+PIPE_GROW;
607 thepipe.buf = xcalloc(1, thepipe.len);
609 thepipe.buf[thepipe.len-1] = PIPE_MAGIC;
611 /* and now, as long as we've got a line to try matching and if we can match
612 * the search string, we make substitutions */
613 while ((*hackline || !altered) && (regexec(sed_cmd->sub_match, hackline,
614 sed_cmd->num_backrefs+1, regmatch, 0) != REG_NOMATCH) ) {
617 /* print everything before the match */
618 for (i = 0; i < regmatch[0].rm_so; i++)
619 pipeputc(hackline[i]);
621 /* then print the substitution string */
622 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
623 pipeline, sed_cmd->num_backrefs);
625 /* advance past the match */
626 hackline += regmatch[0].rm_eo;
627 /* flag that something has changed */
630 /* if we're not doing this globally, get out now */
635 for (; *hackline; hackline++) pipeputc(*hackline);
636 if (thepipe.buf[thepipe.idx] == PIPE_MAGIC) thepipe.buf[thepipe.idx] = 0;
647 static void process_file(FILE *file)
650 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
651 unsigned int still_in_range = 0;
655 /* go through every line in the file */
656 while ((line = get_line_from_file(file)) != NULL) {
662 /* for every line, go through all the commands */
663 for (i = 0; i < ncmds; i++) {
664 struct sed_cmd *sed_cmd = &sed_cmds[i];
668 * entry point into sedding...
671 /* no range necessary */
672 (sed_cmd->beg_line == 0 && sed_cmd->end_line == 0 &&
673 sed_cmd->beg_match == NULL &&
674 sed_cmd->end_match == NULL) ||
675 /* this line number is the first address we're looking for */
676 (sed_cmd->beg_line && (sed_cmd->beg_line == linenum)) ||
677 /* this line matches our first address regex */
678 (sed_cmd->beg_match && (regexec(sed_cmd->beg_match, line, 0, NULL, 0) == 0)) ||
679 /* we are currently within the beginning & ending address range */
683 if (sed_cmd->invert ^ matched) {
688 switch (sed_cmd->cmd) {
702 * Some special cases for 's' printing to make it compliant with
703 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
705 * -n ONLY = never print anything regardless of any successful
708 * s///p ONLY = always print successful substitutions, even if
709 * the line is going to be printed anyway (line will be printed
712 * -n AND s///p = print ONLY a successful substitution ONE TIME;
713 * no other lines are printed - this is the reason why the 'p'
714 * flag exists in the first place.
717 /* if the user specified that they didn't want anything printed (i.e., a -n
718 * flag and no 'p' flag after the s///), then there's really no point doing
720 if (be_quiet && !sed_cmd->sub_p)
723 /* we print the line once, unless we were told to be quiet */
725 altered |= do_subst_command(sed_cmd, &line);
727 /* we also print the line if we were given the 'p' flag
728 * (this is quite possibly the second printing) */
730 altered |= do_subst_command(sed_cmd, &line);
731 if (altered && (i+1 >= ncmds || sed_cmds[i+1].cmd != 's'))
738 fputs(sed_cmd->editline, stdout);
743 fputs(sed_cmd->editline, stdout);
747 /* single-address case */
748 if ((sed_cmd->end_match == NULL && sed_cmd->end_line == 0)
749 /* multi-address case */
750 /* - matching text */
751 || (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
752 /* - matching line numbers */
753 || (sed_cmd->end_line > 0 && sed_cmd->end_line == linenum))
755 fputs(sed_cmd->editline, stdout);
764 outfile = fopen(sed_cmd->filename, "r");
767 /* else if we couldn't open the output file,
768 * no biggie, just don't print anything */
776 * exit point from sedding...
780 /* this is a single-address command or... */
781 (sed_cmd->end_line == 0 && sed_cmd->end_match == NULL) || (
782 /* we were in the middle of our address range (this
783 * isn't the first time through) and.. */
784 (still_in_range == 1) && (
785 /* this line number is the last address we're looking for or... */
786 (sed_cmd->end_line && (sed_cmd->end_line == linenum)) ||
787 /* this line matches our last address regex */
788 (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
792 /* we're out of our address range */
796 /* didn't hit the exit? then we're still in the middle of an address range */
806 /* we will print the line unless we were told to be quiet or if the
807 * line was altered (via a 'd'elete or 's'ubstitution), in which case
808 * the altered line was already printed */
809 if (!be_quiet && !altered)
816 extern int sed_main(int argc, char **argv)
818 int opt, status = EXIT_SUCCESS;
820 #ifdef CONFIG_FEATURE_CLEAN_UP
821 /* destroy command strings on exit */
822 if (atexit(destroy_cmd_strs) == -1)
823 perror_msg_and_die("atexit");
826 /* do normal option parsing */
827 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
836 load_cmd_file(optarg);
843 /* if we didn't get a pattern from a -e and no command file was specified,
844 * argv[optind] should be the pattern. no pattern, no worky */
846 if (argv[optind] == NULL)
849 add_cmd_str(argv[optind]);
855 /* argv[(optind)..(argc-1)] should be names of file to process. If no
856 * files were specified or '-' was specified, take input from stdin.
857 * Otherwise, we process all the files specified. */
858 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
864 for (i = optind; i < argc; i++) {
865 file = wfopen(argv[i], "r");
870 status = EXIT_FAILURE;