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>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 Supported features and commands in this version of sed:
27 - address matching: num|/matchstr/[,num|/matchstr/|$]command
28 - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
29 - edit commands: (a)ppend, (i)nsert, (c)hange
30 - file commands: (r)ead
31 - backreferences in substitution expressions (\1, \2...\9)
33 (Note: Specifying an address (range) to match is *optional*; commands
34 default to the whole pattern space if no specific address match was
39 - transliteration (y/source-chars/dest-chars/) (use 'tr')
40 - no pattern space hold space storing / swapping (x, etc.)
41 - no labels / branching (: label, b, t, and friends)
42 - and lots, lots more.
46 #include <unistd.h> /* for getopt() */
48 #include <string.h> /* for strdup() */
50 #include <ctype.h> /* for isspace() */
55 extern void xregcomp(regex_t *preg, const char *regex, int cflags);
56 extern int optind; /* in unistd.h */
57 extern char *optarg; /* ditto */
60 static int be_quiet = 0;
64 /* Order by alignment requirements */
67 regex_t *beg_match; /* sed -e '/match/cmd' */
68 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
70 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
72 /* sed -e 's/sub_match/replace/' */
76 /* EDIT COMMAND (a,i,c) SPECIFIC FIELDS */
79 /* FILE COMMAND (r) SPECIFIC FIELDS */
83 int beg_line; /* 'sed 1p' 0 == no begining line, apply commands to all lines */
84 int end_line; /* 'sed 1,3p' 0 == no end line, use only beginning. -1 == $ */
85 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
87 unsigned int num_backrefs:4; /* how many back references (\1..\9) */
88 /* Note: GNU/POSIX sed does not save more than nine backrefs, so
89 * we only use 4 bits to hold the number */
90 unsigned int sub_g:1; /* sed -e 's/foo/bar/g' (global) */
91 unsigned int sub_p:2; /* sed -e 's/foo/bar/p' (print substitution) */
94 char delimiter; /* The delimiter used to separate regexps */
97 char cmd; /* p,d,s (add more at your leisure :-) */
101 static struct sed_cmd *sed_cmds = NULL; /* growable arrary holding a sequence of sed cmds */
102 static int ncmds = 0; /* number of sed commands */
104 /*static char *cur_file = NULL;*/ /* file currently being processed XXX: do I need this? */
106 const char * const semicolon_whitespace = "; \n\r\t\v\0";
108 #ifdef CONFIG_FEATURE_CLEAN_UP
109 static void destroy_cmd_strs(void)
111 if (sed_cmds == NULL)
114 /* destroy all the elements in the array */
115 while (--ncmds >= 0) {
117 if (sed_cmds[ncmds].beg_match) {
118 regfree(sed_cmds[ncmds].beg_match);
119 free(sed_cmds[ncmds].beg_match);
121 if (sed_cmds[ncmds].end_match) {
122 regfree(sed_cmds[ncmds].end_match);
123 free(sed_cmds[ncmds].end_match);
125 if (sed_cmds[ncmds].sub_match) {
126 regfree(sed_cmds[ncmds].sub_match);
127 free(sed_cmds[ncmds].sub_match);
129 if (sed_cmds[ncmds].replace)
130 free(sed_cmds[ncmds].replace);
133 /* destroy the array */
141 * index_of_next_unescaped_regexp_delim - walks left to right through a string
142 * beginning at a specified index and returns the index of the next regular
143 * expression delimiter (typically a forward * slash ('/')) not preceeded by
146 static int index_of_next_unescaped_regexp_delim(const struct sed_cmd * const sed_cmd, const char *str, int idx)
152 for ( ; (ch = str[idx]); idx++) {
154 if (ch == ']' && !(bracket == idx - 1 ||
155 (bracket == idx - 2 && str[idx-1] == '^')))
163 else if (ch == sed_cmd->delimiter)
167 /* if we make it to here, we've hit the end of the string */
172 * returns the index in the string just past where the address ends.
174 static int get_address(struct sed_cmd *sed_cmd, const char *str, int *linenum, regex_t **regex)
176 char *my_str = xstrdup(str);
179 olddelimiter = sed_cmd->delimiter;
180 sed_cmd->delimiter = '/';
182 if (isdigit(my_str[idx])) {
185 } while (isdigit(my_str[idx]));
187 *linenum = atoi(my_str);
189 else if (my_str[idx] == '$') {
193 else if (my_str[idx] == '/') {
194 idx = index_of_next_unescaped_regexp_delim(sed_cmd, my_str, ++idx);
196 error_msg_and_die("unterminated match expression");
198 *regex = (regex_t *)xmalloc(sizeof(regex_t));
199 xregcomp(*regex, my_str+1, REG_NEWLINE);
200 idx++; /* so it points to the next character after the last '/' */
203 error_msg("get_address: no address found in string\n"
204 "\t(you probably didn't check the string you passed me)");
209 sed_cmd->delimiter = olddelimiter;
213 static int parse_subst_cmd(struct sed_cmd * 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... */
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, 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, 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]) {
275 /* any whitespace or semicolon trailing after a s/// is ok */
276 if (strchr(semicolon_whitespace, substr[idx]))
279 error_msg_and_die("bad option in substitution expression");
284 /* compile the match string into a regex */
285 sed_cmd->sub_match = (regex_t *)xmalloc(sizeof(regex_t));
286 xregcomp(sed_cmd->sub_match, match, cflags);
292 static void move_back(char *str, int offset)
294 memmove(str, str + offset, strlen(str + offset) + 1);
297 static int parse_edit_cmd(struct sed_cmd *sed_cmd, const char *editstr)
300 int slashes_eaten = 0;
301 char *ptr; /* shorthand */
304 * the string that gets passed to this function should look like this:
308 * | this backslash (immediately following the edit command) is mandatory
315 * as soon as we hit a TEXT line that has no trailing '\', we're done.
316 * this means a command like:
325 if (editstr[1] != '\\' && (editstr[2] != '\n' || editstr[2] != '\r'))
326 error_msg_and_die("bad format in edit expression");
328 /* store the edit line text */
329 /* make editline big enough to accomodate the extra '\n' we will tack on
331 sed_cmd->editline = xmalloc(strlen(&editstr[3]) + 2);
332 strcpy(sed_cmd->editline, &editstr[3]);
333 ptr = sed_cmd->editline;
335 /* now we need to go through * and: s/\\[\r\n]$/\n/g on the edit line */
337 while (ptr[idx] != '\\' || (ptr[idx+1] != '\n' && ptr[idx+1] != '\r')) {
343 /* move the newline over the '\' before it (effectively eats the '\') */
344 move_back(&ptr[idx], 1);
346 /* substitue \r for \n if needed */
347 if (ptr[idx] == '\r')
352 /* figure out if we need to add a newline */
353 if (ptr[idx-1] != '\n') {
358 /* terminate string */
361 /* this accounts for discrepancies between the modified string and the
362 * original string passed in to this function */
364 /* adjust for opening 2 chars [aic]\ */
366 return idx + slashes_eaten + 2;
370 static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
376 * the string that gets passed to this function should look like this:
381 * optional whitespace
383 * re: the file to be read, the GNU manual says the following: "Note that
384 * if filename cannot be read, it is treated as if it were an empty file,
385 * without any error indication." Thus, all of the following commands are
388 * sed -e '1r noexist'
393 /* the file command may be followed by whitespace; move past it. */
394 while (isspace(filecmdstr[++idx]))
397 /* the first non-whitespace we get is a filename. the filename ends when we
398 * hit a normal sed command terminator or end of string */
399 filenamelen = strcspn(&filecmdstr[idx], semicolon_whitespace);
400 sed_cmd->filename = xmalloc(filenamelen + 1);
401 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
403 return idx + filenamelen;
407 static char *parse_cmd_str(struct sed_cmd * const sed_cmd, const char *const cmdstr)
412 * format is: [addr][,addr]cmd
417 /* first part (if present) is an address: either a number or a /regex/ */
418 if (isdigit(cmdstr[idx]) || cmdstr[idx] == '/')
419 idx = get_address(sed_cmd, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
421 /* second part (if present) will begin with a comma */
422 if (cmdstr[idx] == ',')
423 idx += get_address(sed_cmd, &cmdstr[++idx], &sed_cmd->end_line, &sed_cmd->end_match);
425 /* last part (mandatory) will be a command */
426 if (cmdstr[idx] == '\0')
427 error_msg_and_die("missing command");
428 sed_cmd->cmd = cmdstr[idx];
430 /* if it was a single-letter command that takes no arguments (such as 'p'
431 * or 'd') all we need to do is increment the index past that command */
432 if (strchr("pd", sed_cmd->cmd)) {
435 /* handle (s)ubstitution command */
436 else if (sed_cmd->cmd == 's') {
437 idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
439 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
440 else if (strchr("aic", sed_cmd->cmd)) {
441 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
442 error_msg_and_die("only a beginning address can be specified for edit commands");
443 idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
445 /* handle file cmds: (r)ead */
446 else if (sed_cmd->cmd == 'r') {
447 if (sed_cmd->end_line || sed_cmd->end_match)
448 error_msg_and_die("Command only uses one address");
449 idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
452 error_msg_and_die("invalid command");
455 /* give back whatever's left over */
456 return (char *)&cmdstr[idx];
459 static void add_cmd_str(const char * const cmdstr)
461 char *mystr = (char *)cmdstr;
465 /* trim leading whitespace and semicolons */
466 move_back(mystr, strspn(mystr, semicolon_whitespace));
467 /* if we ate the whole thing, that means there was just trailing
468 * whitespace or a final / no-op semicolon. either way, get out */
469 if (strlen(mystr) == 0)
471 /* if this is a comment, jump past it and keep going */
472 if (mystr[0] == '#') {
473 mystr = strpbrk(mystr, "\n\r");
477 sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
478 /* zero new element */
479 memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
480 /* load command string into new array element, get remainder */
481 mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
483 } while (mystr && strlen(mystr));
487 static void load_cmd_file(char *filename)
493 cmdfile = xfopen(filename, "r");
495 while ((line = get_line_from_file(cmdfile)) != NULL) {
496 /* if a line ends with '\' it needs the next line appended to it */
497 while (line[strlen(line)-2] == '\\' &&
498 (nextline = get_line_from_file(cmdfile)) != NULL) {
499 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
500 strcat(line, nextline);
503 /* eat trailing newline (if any) --if I don't do this, edit commands
504 * (aic) will print an extra newline */
517 #define PIPE_MAGIC 0x7f
520 void pipe_putc(struct pipeline *const pipeline, char c)
522 if (pipeline->buf[pipeline->idx] == PIPE_MAGIC) {
524 xrealloc(pipeline->buf, pipeline->len + PIPE_GROW);
525 memset(pipeline->buf + pipeline->len, 0, PIPE_GROW);
526 pipeline->len += PIPE_GROW;
527 pipeline->buf[pipeline->len - 1] = PIPE_MAGIC;
529 pipeline->buf[pipeline->idx++] = (c);
532 #define pipeputc(c) pipe_putc(pipeline, c)
535 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
536 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
537 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
538 pipeline_len += PIPE_GROW; \
539 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
540 pipeline[pipeline_idx++] = (c); }
543 static void print_subst_w_backrefs(const char *line, const char *replace,
544 regmatch_t *regmatch, struct pipeline *const pipeline, int matches)
548 /* go through the replacement string */
549 for (i = 0; replace[i]; i++) {
550 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
551 if (replace[i] == '\\' && isdigit(replace[i+1])) {
555 ++i; /* i now indexes the backref number, instead of the leading slash */
556 tmpstr[0] = replace[i];
558 backref = atoi(tmpstr);
559 /* print out the text held in regmatch[backref] */
560 if (backref <= matches && regmatch[backref].rm_so != -1)
561 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
565 /* if we find a backslash escaped character, print the character */
566 else if (replace[i] == '\\') {
568 pipeputc(replace[i]);
571 /* if we find an unescaped '&' print out the whole matched text.
572 * fortunately, regmatch[0] contains the indicies to the whole matched
573 * expression (kinda seems like it was designed for just such a
575 else if (replace[i] == '&' && replace[i-1] != '\\') {
577 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
580 /* nothing special, just print this char of the replacement string to stdout */
582 pipeputc(replace[i]);
586 static int do_subst_command(const struct sed_cmd *sed_cmd, char **line)
588 char *hackline = *line;
589 struct pipeline thepipe = { NULL, 0 , 0};
590 struct pipeline *const pipeline = &thepipe;
592 regmatch_t *regmatch = NULL;
594 /* we only proceed if the substitution 'search' expression matches */
595 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
598 /* whaddaya know, it matched. get the number of back references */
599 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
601 /* allocate more PIPE_GROW bytes
602 if replaced string is larger than original */
603 thepipe.len = strlen(hackline)+PIPE_GROW;
604 thepipe.buf = xcalloc(1, thepipe.len);
606 thepipe.buf[thepipe.len-1] = PIPE_MAGIC;
608 /* and now, as long as we've got a line to try matching and if we can match
609 * the search string, we make substitutions */
610 while ((*hackline || !altered) && (regexec(sed_cmd->sub_match, hackline,
611 sed_cmd->num_backrefs+1, regmatch, 0) != REG_NOMATCH) ) {
614 /* print everything before the match */
615 for (i = 0; i < regmatch[0].rm_so; i++)
616 pipeputc(hackline[i]);
618 /* then print the substitution string */
619 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
620 pipeline, sed_cmd->num_backrefs);
622 /* advance past the match */
623 hackline += regmatch[0].rm_eo;
624 /* flag that something has changed */
627 /* if we're not doing this globally, get out now */
632 for (; *hackline; hackline++) pipeputc(*hackline);
633 if (thepipe.buf[thepipe.idx] == PIPE_MAGIC) thepipe.buf[thepipe.idx] = 0;
644 static void process_file(FILE *file)
647 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
648 unsigned int still_in_range = 0;
652 /* go through every line in the file */
653 while ((line = get_line_from_file(file)) != NULL) {
659 /* for every line, go through all the commands */
660 for (i = 0; i < ncmds; i++) {
661 struct sed_cmd *sed_cmd = &sed_cmds[i];
665 * entry point into sedding...
668 /* no range necessary */
669 (sed_cmd->beg_line == 0 && sed_cmd->end_line == 0 &&
670 sed_cmd->beg_match == NULL &&
671 sed_cmd->end_match == NULL) ||
672 /* this line number is the first address we're looking for */
673 (sed_cmd->beg_line && (sed_cmd->beg_line == linenum)) ||
674 /* this line matches our first address regex */
675 (sed_cmd->beg_match && (regexec(sed_cmd->beg_match, line, 0, NULL, 0) == 0)) ||
676 /* we are currently within the beginning & ending address range */
683 switch (sed_cmd->cmd) {
696 * Some special cases for 's' printing to make it compliant with
697 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
699 * -n ONLY = never print anything regardless of any successful
702 * s///p ONLY = always print successful substitutions, even if
703 * the line is going to be printed anyway (line will be printed
706 * -n AND s///p = print ONLY a successful substitution ONE TIME;
707 * no other lines are printed - this is the reason why the 'p'
708 * flag exists in the first place.
711 /* if the user specified that they didn't want anything printed (i.e., a -n
712 * flag and no 'p' flag after the s///), then there's really no point doing
714 if (be_quiet && !sed_cmd->sub_p)
717 /* we print the line once, unless we were told to be quiet */
719 altered |= do_subst_command(sed_cmd, &line);
721 /* we also print the line if we were given the 'p' flag
722 * (this is quite possibly the second printing) */
724 altered |= do_subst_command(sed_cmd, &line);
725 if (altered && (i+1 >= ncmds || sed_cmds[i+1].cmd != 's'))
732 fputs(sed_cmd->editline, stdout);
737 fputs(sed_cmd->editline, stdout);
741 /* single-address case */
742 if ((sed_cmd->end_match == NULL && sed_cmd->end_line == 0)
743 /* multi-address case */
744 /* - matching text */
745 || (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
746 /* - matching line numbers */
747 || (sed_cmd->end_line > 0 && sed_cmd->end_line == linenum))
749 fputs(sed_cmd->editline, stdout);
758 outfile = fopen(sed_cmd->filename, "r");
761 /* else if we couldn't open the output file,
762 * no biggie, just don't print anything */
769 * exit point from sedding...
772 /* this is a single-address command or... */
773 (sed_cmd->end_line == 0 && sed_cmd->end_match == NULL) || (
774 /* we were in the middle of our address range (this
775 * isn't the first time through) and.. */
776 (still_in_range == 1) && (
777 /* this line number is the last address we're looking for or... */
778 (sed_cmd->end_line && (sed_cmd->end_line == linenum)) ||
779 /* this line matches our last address regex */
780 (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
784 /* we're out of our address range */
788 /* didn't hit the exit? then we're still in the middle of an address range */
795 /* we will print the line unless we were told to be quiet or if the
796 * line was altered (via a 'd'elete or 's'ubstitution), in which case
797 * the altered line was already printed */
798 if (!be_quiet && !altered)
805 extern int sed_main(int argc, char **argv)
807 int opt, status = EXIT_SUCCESS;
809 #ifdef CONFIG_FEATURE_CLEAN_UP
810 /* destroy command strings on exit */
811 if (atexit(destroy_cmd_strs) == -1)
812 perror_msg_and_die("atexit");
815 /* do normal option parsing */
816 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
825 load_cmd_file(optarg);
832 /* if we didn't get a pattern from a -e and no command file was specified,
833 * argv[optind] should be the pattern. no pattern, no worky */
835 if (argv[optind] == NULL)
838 add_cmd_str(argv[optind]);
844 /* argv[(optind)..(argc-1)] should be names of file to process. If no
845 * files were specified or '-' was specified, take input from stdin.
846 * Otherwise, we process all the files specified. */
847 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
853 for (i = optind; i < argc; i++) {
854 file = wfopen(argv[i], "r");
859 status = EXIT_FAILURE;