2 * sed.c - very minimalist version of sed
4 * Copyright (C) 1999,2000,2001 by Lineo, inc.
5 * Written by Mark Whitley <markw@lineo.com>, <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;
67 char delimiter; /* The delimiter used to separate regexps */
70 int beg_line; /* 'sed 1p' 0 == no begining line, apply commands to all lines */
71 int end_line; /* 'sed 1,3p' 0 == no end line, use only beginning. -1 == $ */
72 regex_t *beg_match; /* sed -e '/match/cmd' */
73 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
76 char cmd; /* p,d,s (add more at your leisure :-) */
79 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
81 /* sed -e 's/sub_match/replace/' */
84 unsigned int num_backrefs:4; /* how many back references (\1..\9) */
85 /* Note: GNU/POSIX sed does not save more than nine backrefs, so
86 * we only use 4 bits to hold the number */
87 unsigned int sub_g:1; /* sed -e 's/foo/bar/g' (global) */
88 unsigned int sub_p:2; /* sed -e 's/foo/bar/p' (print substitution) */
91 /* EDIT COMMAND (a,i,c) SPEICIFIC FIELDS */
96 /* FILE COMMAND (r) SPEICIFIC FIELDS */
102 static struct sed_cmd *sed_cmds = NULL; /* growable arrary holding a sequence of sed cmds */
103 static int ncmds = 0; /* number of sed commands */
105 /*static char *cur_file = NULL;*/ /* file currently being processed XXX: do I need this? */
107 #ifdef BB_FEATURE_CLEAN_UP
108 static void destroy_cmd_strs()
110 if (sed_cmds == NULL)
113 /* destroy all the elements in the array */
114 while (--ncmds >= 0) {
116 if (sed_cmds[ncmds].beg_match) {
117 regfree(sed_cmds[ncmds].beg_match);
118 free(sed_cmds[ncmds].beg_match);
120 if (sed_cmds[ncmds].end_match) {
121 regfree(sed_cmds[ncmds].end_match);
122 free(sed_cmds[ncmds].end_match);
124 if (sed_cmds[ncmds].sub_match) {
125 regfree(sed_cmds[ncmds].sub_match);
126 free(sed_cmds[ncmds].sub_match);
128 if (sed_cmds[ncmds].replace)
129 free(sed_cmds[ncmds].replace);
132 /* destroy the array */
140 * index_of_next_unescaped_regexp_delim - walks left to right through a string
141 * beginning at a specified index and returns the index of the next regular
142 * expression delimiter (typically a forward * slash ('/')) not preceeded by
145 static int index_of_next_unescaped_regexp_delim(struct sed_cmd *sed_cmd, const char *str, int idx)
150 for ( ; str[idx]; idx++) {
152 if (str[idx] == ']' && !(bracket == idx - 1 ||
153 (bracket == idx - 2 && str[idx-1] == '^')))
157 else if (str[idx] == '\\')
159 else if (str[idx] == '[')
161 else if (str[idx] == sed_cmd->delimiter)
165 /* if we make it to here, we've hit the end of the string */
170 * returns the index in the string just past where the address ends.
172 static int get_address(struct sed_cmd *sed_cmd, const char *str, int *linenum, regex_t **regex)
174 char *my_str = strdup(str);
177 olddelimiter = sed_cmd->delimiter;
178 sed_cmd->delimiter = '/';
180 if (isdigit(my_str[idx])) {
183 } while (isdigit(my_str[idx]));
185 *linenum = atoi(my_str);
187 else if (my_str[idx] == '$') {
191 else if (my_str[idx] == '/') {
192 idx = index_of_next_unescaped_regexp_delim(sed_cmd, my_str, ++idx);
194 error_msg_and_die("unterminated match expression");
196 *regex = (regex_t *)xmalloc(sizeof(regex_t));
197 xregcomp(*regex, my_str+1, REG_NEWLINE);
198 idx++; /* so it points to the next character after the last '/' */
201 error_msg("get_address: no address found in string\n"
202 "\t(you probably didn't check the string you passed me)");
207 sed_cmd->delimiter = olddelimiter;
211 static int parse_subst_cmd(struct sed_cmd *sed_cmd, const char *substr)
213 int oldidx, cflags = REG_NEWLINE;
219 * the string that gets passed to this function should look like this:
220 * s/match/replace/gIp
224 * (all three of the '/' slashes are mandatory)
227 /* verify that the 's' is followed by something. That something
228 * (typically a 'slash') is now our regexp delimiter... */
230 error_msg_and_die("bad format in substitution expression");
232 sed_cmd->delimiter=substr[idx];
234 /* save the match string */
236 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
238 error_msg_and_die("bad format in substitution expression");
239 match = xstrndup(substr + oldidx, idx - oldidx);
241 /* determine the number of back references in the match string */
242 /* Note: we compute this here rather than in the do_subst_command()
243 * function to save processor time, at the expense of a little more memory
244 * (4 bits) per sed_cmd */
246 /* sed_cmd->num_backrefs = 0; */ /* XXX: not needed? --apparently not */
247 for (j = 0; match[j]; j++) {
248 /* GNU/POSIX sed does not save more than nine backrefs */
249 if (match[j] == '\\' && match[j+1] == '(' && sed_cmd->num_backrefs <= 9)
250 sed_cmd->num_backrefs++;
253 /* save the replacement string */
255 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
257 error_msg_and_die("bad format in substitution expression");
258 sed_cmd->replace = xstrndup(substr + oldidx, idx - oldidx);
260 /* process the flags */
261 while (substr[++idx]) {
262 switch (substr[idx]) {
273 /* any whitespace or semicolon trailing after a s/// is ok */
274 if (strchr("; \t\v\n\r", substr[idx]))
277 error_msg_and_die("bad option in substitution expression");
282 /* compile the match string into a regex */
283 sed_cmd->sub_match = (regex_t *)xmalloc(sizeof(regex_t));
284 xregcomp(sed_cmd->sub_match, match, cflags);
290 static int parse_edit_cmd(struct sed_cmd *sed_cmd, const char *editstr)
293 int slashes_eaten = 0;
294 char *ptr; /* shorthand */
297 * the string that gets passed to this function should look like this:
301 * | this backslash (immediately following the edit command) is mandatory
308 * as soon as we hit a TEXT line that has no trailing '\', we're done.
309 * this means a command like:
318 if (editstr[1] != '\\' && (editstr[2] != '\n' || editstr[2] != '\r'))
319 error_msg_and_die("bad format in edit expression");
321 /* store the edit line text */
322 /* make editline big enough to accomodate the extra '\n' we will tack on
324 sed_cmd->editline = xmalloc(strlen(&editstr[3]) + 2);
325 strcpy(sed_cmd->editline, &editstr[3]);
326 ptr = sed_cmd->editline;
328 /* now we need to go through * and: s/\\[\r\n]$/\n/g on the edit line */
330 while (ptr[idx] != '\\' || (ptr[idx+1] != '\n' && ptr[idx+1] != '\r')) {
336 /* move the newline over the '\' before it (effectively eats the '\') */
337 memmove(&ptr[idx], &ptr[idx+1], strlen(&ptr[idx+1]));
338 ptr[strlen(ptr)-1] = 0;
340 /* substitue \r for \n if needed */
341 if (ptr[idx] == '\r')
346 /* this accounts for discrepancies between the modified string and the
347 * original string passed in to this function */
348 idx += slashes_eaten;
350 /* figure out if we need to add a newline */
351 if (ptr[idx-1] != '\n') {
356 /* terminate string */
358 /* adjust for opening 2 chars [aic]\ */
365 static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
371 * the string that gets passed to this function should look like this:
376 * optional whitespace
378 * re: the file to be read, the GNU manual says the following: "Note that
379 * if filename cannot be read, it is treated as if it were an empty file,
380 * without any error indication." Thus, all of the following commands are
383 * sed -e '1r noexist'
388 /* the file command may be followed by whitespace; move past it. */
389 while (isspace(filecmdstr[++idx]))
392 /* the first non-whitespace we get is a filename. the filename ends when we
393 * hit a normal sed command terminator or end of string */
394 filenamelen = strcspn(&filecmdstr[idx], "; \n\r\t\v\0");
395 sed_cmd->filename = xmalloc(filenamelen + 1);
396 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
398 return idx + filenamelen;
402 static char *parse_cmd_str(struct sed_cmd *sed_cmd, const char *cmdstr)
407 * format is: [addr][,addr]cmd
412 /* first part (if present) is an address: either a number or a /regex/ */
413 if (isdigit(cmdstr[idx]) || cmdstr[idx] == '/')
414 idx = get_address(sed_cmd, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
416 /* second part (if present) will begin with a comma */
417 if (cmdstr[idx] == ',')
418 idx += get_address(sed_cmd, &cmdstr[++idx], &sed_cmd->end_line, &sed_cmd->end_match);
420 /* last part (mandatory) will be a command */
421 if (cmdstr[idx] == '\0')
422 error_msg_and_die("missing command");
423 sed_cmd->cmd = cmdstr[idx];
425 /* if it was a single-letter command that takes no arguments (such as 'p'
426 * or 'd') all we need to do is increment the index past that command */
427 if (strchr("pd", cmdstr[idx])) {
430 /* handle (s)ubstitution command */
431 else if (sed_cmd->cmd == 's') {
432 idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
434 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
435 else if (strchr("aic", sed_cmd->cmd)) {
436 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
437 error_msg_and_die("only a beginning address can be specified for edit commands");
438 idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
440 /* handle file cmds: (r)ead */
441 else if (sed_cmd->cmd == 'r') {
442 if (sed_cmd->end_line || sed_cmd->end_match)
443 error_msg_and_die("Command only uses one address");
444 idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
447 error_msg_and_die("invalid command");
450 /* give back whatever's left over */
451 return (char *)&cmdstr[idx];
454 static void add_cmd_str(const char *cmdstr)
456 char *mystr = (char *)cmdstr;
460 /* trim leading whitespace and semicolons */
461 memmove(mystr, &mystr[strspn(mystr, "; \n\r\t\v")], strlen(mystr));
462 /* if we ate the whole thing, that means there was just trailing
463 * whitespace or a final / no-op semicolon. either way, get out */
464 if (strlen(mystr) == 0)
466 /* if this is a comment, jump past it and keep going */
467 if (mystr[0] == '#') {
468 mystr = strpbrk(mystr, ";\n\r");
472 sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
473 /* zero new element */
474 memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
475 /* load command string into new array element, get remainder */
476 mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
478 } while (mystr && strlen(mystr));
482 static void load_cmd_file(char *filename)
488 cmdfile = xfopen(filename, "r");
490 while ((line = get_line_from_file(cmdfile)) != NULL) {
491 /* if a line ends with '\' it needs the next line appended to it */
492 while (line[strlen(line)-2] == '\\' &&
493 (nextline = get_line_from_file(cmdfile)) != NULL) {
494 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
495 strcat(line, nextline);
498 /* eat trailing newline (if any) --if I don't do this, edit commands
499 * (aic) will print an extra newline */
506 #define PIPE_MAGIC 0x7f
508 #define pipeputc(c) \
509 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
510 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
511 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
512 pipeline_len += PIPE_GROW; \
513 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
514 pipeline[pipeline_idx++] = (c); }
516 static void print_subst_w_backrefs(const char *line, const char *replace,
517 regmatch_t *regmatch, char **pipeline_p, int *pipeline_idx_p,
518 int *pipeline_len_p, int matches)
520 char *pipeline = *pipeline_p;
521 int pipeline_idx = *pipeline_idx_p;
522 int pipeline_len = *pipeline_len_p;
525 /* go through the replacement string */
526 for (i = 0; replace[i]; i++) {
527 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
528 if (replace[i] == '\\' && isdigit(replace[i+1])) {
532 ++i; /* i now indexes the backref number, instead of the leading slash */
533 tmpstr[0] = replace[i];
535 backref = atoi(tmpstr);
536 /* print out the text held in regmatch[backref] */
537 if (backref <= matches && regmatch[backref].rm_so != -1)
538 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
542 /* if we find a backslash escaped character, print the character */
543 else if (replace[i] == '\\') {
545 pipeputc(replace[i]);
548 /* if we find an unescaped '&' print out the whole matched text.
549 * fortunately, regmatch[0] contains the indicies to the whole matched
550 * expression (kinda seems like it was designed for just such a
552 else if (replace[i] == '&' && replace[i-1] != '\\') {
554 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
557 /* nothing special, just print this char of the replacement string to stdout */
559 pipeputc(replace[i]);
561 *pipeline_p = pipeline;
562 *pipeline_idx_p = pipeline_idx;
563 *pipeline_len_p = pipeline_len;
566 static int do_subst_command(const struct sed_cmd *sed_cmd, char **line)
568 char *hackline = *line;
570 int pipeline_idx = 0;
571 int pipeline_len = 0;
573 regmatch_t *regmatch = NULL;
575 /* we only proceed if the substitution 'search' expression matches */
576 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
579 /* whaddaya know, it matched. get the number of back references */
580 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
582 /* allocate more PIPE_GROW bytes
583 if replaced string is larger than original */
584 pipeline_len = strlen(hackline)+PIPE_GROW;
585 pipeline = xmalloc(pipeline_len);
586 memset(pipeline, 0, pipeline_len);
588 pipeline[pipeline_len-1] = PIPE_MAGIC;
590 /* and now, as long as we've got a line to try matching and if we can match
591 * the search string, we make substitutions */
592 while (*hackline && (regexec(sed_cmd->sub_match, hackline,
593 sed_cmd->num_backrefs+1, regmatch, 0) == 0) ) {
596 /* print everything before the match */
597 for (i = 0; i < regmatch[0].rm_so; i++)
598 pipeputc(hackline[i]);
600 /* then print the substitution string */
601 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
602 &pipeline, &pipeline_idx, &pipeline_len,
603 sed_cmd->num_backrefs);
605 /* advance past the match */
606 hackline += regmatch[0].rm_eo;
607 /* flag that something has changed */
610 /* if we're not doing this globally, get out now */
615 for (; *hackline; hackline++) pipeputc(*hackline);
616 if (pipeline[pipeline_idx] == PIPE_MAGIC) pipeline[pipeline_idx] = 0;
627 static void process_file(FILE *file)
630 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
631 unsigned int still_in_range = 0;
635 /* go through every line in the file */
636 while ((line = get_line_from_file(file)) != NULL) {
642 /* for every line, go through all the commands */
643 for (i = 0; i < ncmds; i++) {
647 * entry point into sedding...
650 /* no range necessary */
651 (sed_cmds[i].beg_line == 0 && sed_cmds[i].end_line == 0 &&
652 sed_cmds[i].beg_match == NULL &&
653 sed_cmds[i].end_match == NULL) ||
654 /* this line number is the first address we're looking for */
655 (sed_cmds[i].beg_line && (sed_cmds[i].beg_line == linenum)) ||
656 /* this line matches our first address regex */
657 (sed_cmds[i].beg_match && (regexec(sed_cmds[i].beg_match, line, 0, NULL, 0) == 0)) ||
658 /* we are currently within the beginning & ending address range */
665 switch (sed_cmds[i].cmd) {
678 * Some special cases for 's' printing to make it compliant with
679 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
681 * -n ONLY = never print anything regardless of any successful
684 * s///p ONLY = always print successful substitutions, even if
685 * the line is going to be printed anyway (line will be printed
688 * -n AND s///p = print ONLY a successful substitution ONE TIME;
689 * no other lines are printed - this is the reason why the 'p'
690 * flag exists in the first place.
693 /* if the user specified that they didn't want anything printed (i.e., a -n
694 * flag and no 'p' flag after the s///), then there's really no point doing
696 if (be_quiet && !sed_cmds[i].sub_p)
699 /* we print the line once, unless we were told to be quiet */
701 altered |= do_subst_command(&sed_cmds[i], &line);
703 /* we also print the line if we were given the 'p' flag
704 * (this is quite possibly the second printing) */
705 if (sed_cmds[i].sub_p)
706 altered |= do_subst_command(&sed_cmds[i], &line);
707 if (altered && (i+1 >= ncmds || sed_cmds[i+1].cmd != 's'))
714 fputs(sed_cmds[i].editline, stdout);
719 fputs(sed_cmds[i].editline, stdout);
723 /* single-address case */
724 if (sed_cmds[i].end_match == NULL && sed_cmds[i].end_line == 0) {
725 fputs(sed_cmds[i].editline, stdout);
727 /* multi-address case */
730 if (sed_cmds[i].end_match && (regexec(sed_cmds[i].end_match, line, 0, NULL, 0) == 0))
731 fputs(sed_cmds[i].editline, stdout);
732 /* matching line numbers */
733 if (sed_cmds[i].end_line > 0 && sed_cmds[i].end_line == linenum)
734 fputs(sed_cmds[i].editline, stdout);
743 outfile = fopen(sed_cmds[i].filename, "r");
746 /* else if we couldn't open the output file,
747 * no biggie, just don't print anything */
754 * exit point from sedding...
757 /* this is a single-address command or... */
758 (sed_cmds[i].end_line == 0 && sed_cmds[i].end_match == NULL) || (
759 /* we were in the middle of our address range (this
760 * isn't the first time through) and.. */
761 (still_in_range == 1) && (
762 /* this line number is the last address we're looking for or... */
763 (sed_cmds[i].end_line && (sed_cmds[i].end_line == linenum)) ||
764 /* this line matches our last address regex */
765 (sed_cmds[i].end_match && (regexec(sed_cmds[i].end_match, line, 0, NULL, 0) == 0))
769 /* we're out of our address range */
773 /* didn't hit the exit? then we're still in the middle of an address range */
780 /* we will print the line unless we were told to be quiet or if the
781 * line was altered (via a 'd'elete or 's'ubstitution), in which case
782 * the altered line was already printed */
783 if (!be_quiet && !altered)
790 extern int sed_main(int argc, char **argv)
794 #ifdef BB_FEATURE_CLEAN_UP
795 /* destroy command strings on exit */
796 if (atexit(destroy_cmd_strs) == -1)
797 perror_msg_and_die("atexit");
800 /* do normal option parsing */
801 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
810 load_cmd_file(optarg);
817 /* if we didn't get a pattern from a -e and no command file was specified,
818 * argv[optind] should be the pattern. no pattern, no worky */
820 if (argv[optind] == NULL)
823 add_cmd_str(argv[optind]);
829 /* argv[(optind)..(argc-1)] should be names of file to process. If no
830 * files were specified or '-' was specified, take input from stdin.
831 * Otherwise, we process all the files specified. */
832 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
838 for (i = optind; i < argc; i++) {
839 file = fopen(argv[i], "r");
841 perror_msg("%s", argv[i]);