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.
45 Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
49 #include <unistd.h> /* for getopt() */
51 #include <string.h> /* for strdup() */
53 #include <ctype.h> /* for isspace() */
58 extern void xregcomp(regex_t *preg, const char *regex, int cflags);
59 extern int optind; /* in unistd.h */
60 extern char *optarg; /* ditto */
63 static int be_quiet = 0;
67 /* Order by alignment requirements */
70 regex_t *beg_match; /* sed -e '/match/cmd' */
71 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
73 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
75 /* sed -e 's/sub_match/replace/' */
79 /* EDIT COMMAND (a,i,c) SPECIFIC FIELDS */
82 /* FILE COMMAND (r) SPECIFIC FIELDS */
86 int beg_line; /* 'sed 1p' 0 == no begining line, apply commands to all lines */
87 int end_line; /* 'sed 1,3p' 0 == no end line, use only beginning. -1 == $ */
88 /* SUBSTITUTION COMMAND SPECIFIC FIELDS */
90 unsigned int num_backrefs:4; /* how many back references (\1..\9) */
91 /* Note: GNU/POSIX sed does not save more than nine backrefs, so
92 * we only use 4 bits to hold the number */
93 unsigned int sub_g:1; /* sed -e 's/foo/bar/g' (global) */
94 unsigned int sub_p:2; /* sed -e 's/foo/bar/p' (print substitution) */
97 char delimiter; /* The delimiter used to separate regexps */
100 char cmd; /* p,d,s (add more at your leisure :-) */
103 int invert; /* the '!' after the address */
107 static struct sed_cmd *sed_cmds = NULL; /* growable arrary holding a sequence of sed cmds */
108 static int ncmds = 0; /* number of sed commands */
110 /*static char *cur_file = NULL;*/ /* file currently being processed XXX: do I need this? */
112 const char * const semicolon_whitespace = "; \n\r\t\v\0";
114 #ifdef CONFIG_FEATURE_CLEAN_UP
115 static void destroy_cmd_strs(void)
117 if (sed_cmds == NULL)
120 /* destroy all the elements in the array */
121 while (--ncmds >= 0) {
123 if (sed_cmds[ncmds].beg_match) {
124 regfree(sed_cmds[ncmds].beg_match);
125 free(sed_cmds[ncmds].beg_match);
127 if (sed_cmds[ncmds].end_match) {
128 regfree(sed_cmds[ncmds].end_match);
129 free(sed_cmds[ncmds].end_match);
131 if (sed_cmds[ncmds].sub_match) {
132 regfree(sed_cmds[ncmds].sub_match);
133 free(sed_cmds[ncmds].sub_match);
135 free(sed_cmds[ncmds].replace);
138 /* destroy the array */
146 * index_of_next_unescaped_regexp_delim - walks left to right through a string
147 * beginning at a specified index and returns the index of the next regular
148 * expression delimiter (typically a forward * slash ('/')) not preceeded by
151 static int index_of_next_unescaped_regexp_delim(const char delimiter, const char *str, int idx)
157 for ( ; (ch = str[idx]); idx++) {
159 if (ch == ']' && !(bracket == idx - 1 ||
160 (bracket == idx - 2 && str[idx-1] == '^')))
168 else if (ch == delimiter)
172 /* if we make it to here, we've hit the end of the string */
177 * returns the index in the string just past where the address ends.
179 static int get_address(char *delimiter, char *my_str, int *linenum, regex_t **regex)
182 if (isdigit(my_str[idx])) {
184 *linenum = strtol(my_str, &endstr, 10);
185 /* endstr shouldnt ever equal NULL */
186 idx = endstr - my_str;
188 else if (my_str[idx] == '$') {
192 else if (my_str[idx] == '/' || my_str[idx] == '\\') {
195 if (my_str[idx] == '\\') {
197 *delimiter = my_str[++idx];
199 idx = index_of_next_unescaped_regexp_delim(*delimiter, my_str, ++idx);
201 error_msg_and_die("unterminated match expression");
204 *regex = (regex_t *)xmalloc(sizeof(regex_t));
205 xregcomp(*regex, my_str+idx_start, REG_NEWLINE);
206 idx++; /* so it points to the next character after the last '/' */
211 static int parse_subst_cmd(struct sed_cmd * const 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->delimiter, 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->delimiter, 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(semicolon_whitespace, 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 void move_back(char *str, int offset)
292 memmove(str, str + offset, strlen(str + offset) + 1);
295 static int parse_edit_cmd(struct sed_cmd *sed_cmd, const char *editstr)
300 * the string that gets passed to this function should look like this:
304 * | this backslash (immediately following the edit command) is mandatory
311 * as soon as we hit a TEXT line that has no trailing '\', we're done.
312 * this means a command like:
321 if (editstr[1] != '\\' || (editstr[2] != '\n' && editstr[2] != '\r'))
322 error_msg_and_die("bad format in edit expression");
324 /* store the edit line text */
325 sed_cmd->editline = xmalloc(strlen(&editstr[3]) + 2);
326 for (i = 3, j = 0; editstr[i] != '\0' && strchr("\r\n", editstr[i]) == NULL;
328 if (editstr[i] == '\\' && strchr("\n\r", editstr[i+1]) != NULL) {
329 sed_cmd->editline[j] = '\n';
332 sed_cmd->editline[j] = editstr[i];
335 /* figure out if we need to add a newline */
336 if (sed_cmd->editline[j-1] != '\n')
337 sed_cmd->editline[j++] = '\n';
339 /* terminate string */
340 sed_cmd->editline[j] = '\0';
346 static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
352 * the string that gets passed to this function should look like this:
357 * optional whitespace
359 * re: the file to be read, the GNU manual says the following: "Note that
360 * if filename cannot be read, it is treated as if it were an empty file,
361 * without any error indication." Thus, all of the following commands are
364 * sed -e '1r noexist'
369 /* the file command may be followed by whitespace; move past it. */
370 while (isspace(filecmdstr[++idx]))
373 /* the first non-whitespace we get is a filename. the filename ends when we
374 * hit a normal sed command terminator or end of string */
375 filenamelen = strcspn(&filecmdstr[idx], semicolon_whitespace);
376 sed_cmd->filename = xmalloc(filenamelen + 1);
377 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
379 return idx + filenamelen;
383 static char *parse_cmd_str(struct sed_cmd * const sed_cmd, char *cmdstr)
388 * format is: [addr][,addr]cmd
393 /* first part (if present) is an address: either a '$', a number or a /regex/ */
394 idx = get_address(&sed_cmd->delimiter, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
396 /* second part (if present) will begin with a comma */
397 if (cmdstr[idx] == ',') {
400 tmp_idx = get_address(&sed_cmd->delimiter, &cmdstr[idx], &sed_cmd->end_line, &sed_cmd->end_match);
402 error_msg_and_die("get_address: no address found in string\n"
403 "\t(you probably didn't check the string you passed me)");
408 /* skip whitespace before the command */
409 while (isspace(cmdstr[idx]))
412 /* there my be the inversion flag between part2 and part3 */
414 if (cmdstr[idx] == '!') {
418 #ifdef SED_FEATURE_STRICT_CHECKING
419 /* According to the spec
420 * It is unspecified whether <blank>s can follow a '!' character,
421 * and conforming applications shall not follow a '!' character
424 if (isblank(cmdstr[idx]) {
425 error_msg_and_die("blank follows '!'");
428 /* skip whitespace before the command */
429 while (isspace(cmdstr[idx]))
434 /* last part (mandatory) will be a command */
435 if (cmdstr[idx] == '\0')
436 error_msg_and_die("missing command");
437 sed_cmd->cmd = cmdstr[idx];
439 /* if it was a single-letter command that takes no arguments (such as 'p'
440 * or 'd') all we need to do is increment the index past that command */
441 if (strchr("pd=", sed_cmd->cmd)) {
444 /* handle (s)ubstitution command */
445 else if (sed_cmd->cmd == 's') {
446 idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
448 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
449 else if (strchr("aic", sed_cmd->cmd)) {
450 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
451 error_msg_and_die("only a beginning address can be specified for edit commands");
452 idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
454 /* handle file cmds: (r)ead */
455 else if (sed_cmd->cmd == 'r') {
456 if (sed_cmd->end_line || sed_cmd->end_match)
457 error_msg_and_die("Command only uses one address");
458 idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
461 error_msg_and_die("Unsupported command %c", sed_cmd->cmd);
464 /* give back whatever's left over */
465 return (char *)&cmdstr[idx];
468 static void add_cmd_str(const char * const cmdstr)
470 char *mystr = (char *)cmdstr;
474 /* trim leading whitespace and semicolons */
475 move_back(mystr, strspn(mystr, semicolon_whitespace));
476 /* if we ate the whole thing, that means there was just trailing
477 * whitespace or a final / no-op semicolon. either way, get out */
478 if (strlen(mystr) == 0)
480 /* if this is a comment, jump past it and keep going */
481 if (mystr[0] == '#') {
482 mystr = strpbrk(mystr, "\n\r");
486 sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
487 /* zero new element */
488 memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
489 /* load command string into new array element, get remainder */
490 mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
492 } while (mystr && strlen(mystr));
496 static void load_cmd_file(char *filename)
502 cmdfile = xfopen(filename, "r");
504 while ((line = get_line_from_file(cmdfile)) != NULL) {
505 /* if a line ends with '\' it needs the next line appended to it */
506 while (line[strlen(line)-2] == '\\' &&
507 (nextline = get_line_from_file(cmdfile)) != NULL) {
508 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
509 strcat(line, nextline);
512 /* eat trailing newline (if any) --if I don't do this, edit commands
513 * (aic) will print an extra newline */
526 #define PIPE_MAGIC 0x7f
529 void pipe_putc(struct pipeline *const pipeline, char c)
531 if (pipeline->buf[pipeline->idx] == PIPE_MAGIC) {
533 xrealloc(pipeline->buf, pipeline->len + PIPE_GROW);
534 memset(pipeline->buf + pipeline->len, 0, PIPE_GROW);
535 pipeline->len += PIPE_GROW;
536 pipeline->buf[pipeline->len - 1] = PIPE_MAGIC;
538 pipeline->buf[pipeline->idx++] = (c);
541 #define pipeputc(c) pipe_putc(pipeline, c)
544 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
545 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
546 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
547 pipeline_len += PIPE_GROW; \
548 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
549 pipeline[pipeline_idx++] = (c); }
552 static void print_subst_w_backrefs(const char *line, const char *replace,
553 regmatch_t *regmatch, struct pipeline *const pipeline, int matches)
557 /* go through the replacement string */
558 for (i = 0; replace[i]; i++) {
559 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
560 if (replace[i] == '\\' && isdigit(replace[i+1])) {
564 ++i; /* i now indexes the backref number, instead of the leading slash */
565 tmpstr[0] = replace[i];
567 backref = atoi(tmpstr);
568 /* print out the text held in regmatch[backref] */
569 if (backref <= matches && regmatch[backref].rm_so != -1)
570 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
574 /* if we find a backslash escaped character, print the character */
575 else if (replace[i] == '\\') {
577 pipeputc(replace[i]);
580 /* if we find an unescaped '&' print out the whole matched text.
581 * fortunately, regmatch[0] contains the indicies to the whole matched
582 * expression (kinda seems like it was designed for just such a
584 else if (replace[i] == '&' && replace[i-1] != '\\') {
586 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
589 /* nothing special, just print this char of the replacement string to stdout */
591 pipeputc(replace[i]);
595 static int do_subst_command(const struct sed_cmd *sed_cmd, char **line)
597 char *hackline = *line;
598 struct pipeline thepipe = { NULL, 0 , 0};
599 struct pipeline *const pipeline = &thepipe;
601 regmatch_t *regmatch = NULL;
603 /* we only proceed if the substitution 'search' expression matches */
604 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
607 /* whaddaya know, it matched. get the number of back references */
608 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
610 /* allocate more PIPE_GROW bytes
611 if replaced string is larger than original */
612 thepipe.len = strlen(hackline)+PIPE_GROW;
613 thepipe.buf = xcalloc(1, thepipe.len);
615 thepipe.buf[thepipe.len-1] = PIPE_MAGIC;
617 /* and now, as long as we've got a line to try matching and if we can match
618 * the search string, we make substitutions */
619 while ((*hackline || !altered) && (regexec(sed_cmd->sub_match, hackline,
620 sed_cmd->num_backrefs+1, regmatch, 0) != REG_NOMATCH) ) {
623 /* print everything before the match */
624 for (i = 0; i < regmatch[0].rm_so; i++)
625 pipeputc(hackline[i]);
627 /* then print the substitution string */
628 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
629 pipeline, sed_cmd->num_backrefs);
631 /* advance past the match */
632 hackline += regmatch[0].rm_eo;
633 /* flag that something has changed */
636 /* if we're not doing this globally, get out now */
641 for (; *hackline; hackline++) pipeputc(*hackline);
642 if (thepipe.buf[thepipe.idx] == PIPE_MAGIC) thepipe.buf[thepipe.idx] = 0;
653 static void process_file(FILE *file)
656 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
657 unsigned int still_in_range = 0;
661 line = get_line_from_file(file);
666 /* go through every line in the file */
670 /* Read one line in advance so we can act on the last line, the '$' address */
671 next_line = get_line_from_file(file);
677 /* for every line, go through all the commands */
678 for (i = 0; i < ncmds; i++) {
679 struct sed_cmd *sed_cmd = &sed_cmds[i];
683 * entry point into sedding...
686 /* no range necessary */
687 (sed_cmd->beg_line == 0 && sed_cmd->end_line == 0 &&
688 sed_cmd->beg_match == NULL &&
689 sed_cmd->end_match == NULL) ||
690 /* this line number is the first address we're looking for */
691 (sed_cmd->beg_line && (sed_cmd->beg_line == linenum)) ||
692 /* this line matches our first address regex */
693 (sed_cmd->beg_match && (regexec(sed_cmd->beg_match, line, 0, NULL, 0) == 0)) ||
694 /* we are currently within the beginning & ending address range */
695 still_in_range || ((sed_cmd->beg_line == -1) && (next_line == NULL))
698 if (sed_cmd->invert ^ matched) {
703 switch (sed_cmd->cmd) {
705 printf("%d\n", linenum);
719 * Some special cases for 's' printing to make it compliant with
720 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
722 * -n ONLY = never print anything regardless of any successful
725 * s///p ONLY = always print successful substitutions, even if
726 * the line is going to be printed anyway (line will be printed
729 * -n AND s///p = print ONLY a successful substitution ONE TIME;
730 * no other lines are printed - this is the reason why the 'p'
731 * flag exists in the first place.
734 /* if the user specified that they didn't want anything printed (i.e., a -n
735 * flag and no 'p' flag after the s///), then there's really no point doing
737 if (be_quiet && !sed_cmd->sub_p)
740 /* we print the line once, unless we were told to be quiet */
742 altered |= do_subst_command(sed_cmd, &line);
744 /* we also print the line if we were given the 'p' flag
745 * (this is quite possibly the second printing) */
747 altered |= do_subst_command(sed_cmd, &line);
748 if (altered && (i+1 >= ncmds || sed_cmds[i+1].cmd != 's'))
755 fputs(sed_cmd->editline, stdout);
760 fputs(sed_cmd->editline, stdout);
764 /* single-address case */
765 if ((sed_cmd->end_match == NULL && sed_cmd->end_line == 0)
766 /* multi-address case */
767 /* - matching text */
768 || (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
769 /* - matching line numbers */
770 || (sed_cmd->end_line > 0 && sed_cmd->end_line == linenum))
772 fputs(sed_cmd->editline, stdout);
781 outfile = fopen(sed_cmd->filename, "r");
784 /* else if we couldn't open the output file,
785 * no biggie, just don't print anything */
793 * exit point from sedding...
797 /* this is a single-address command or... */
798 (sed_cmd->end_line == 0 && sed_cmd->end_match == NULL) || (
799 /* we were in the middle of our address range (this
800 * isn't the first time through) and.. */
801 (still_in_range == 1) && (
802 /* this line number is the last address we're looking for or... */
803 (sed_cmd->end_line && (sed_cmd->end_line == linenum)) ||
804 /* this line matches our last address regex */
805 (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
809 /* we're out of our address range */
813 /* didn't hit the exit? then we're still in the middle of an address range */
823 /* we will print the line unless we were told to be quiet or if the
824 * line was altered (via a 'd'elete or 's'ubstitution), in which case
825 * the altered line was already printed */
826 if (!be_quiet && !altered)
834 extern int sed_main(int argc, char **argv)
836 int opt, status = EXIT_SUCCESS;
838 #ifdef CONFIG_FEATURE_CLEAN_UP
839 /* destroy command strings on exit */
840 if (atexit(destroy_cmd_strs) == -1)
841 perror_msg_and_die("atexit");
844 /* do normal option parsing */
845 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
854 load_cmd_file(optarg);
861 /* if we didn't get a pattern from a -e and no command file was specified,
862 * argv[optind] should be the pattern. no pattern, no worky */
864 if (argv[optind] == NULL)
867 add_cmd_str(argv[optind]);
873 /* argv[(optind)..(argc-1)] should be names of file to process. If no
874 * files were specified or '-' was specified, take input from stdin.
875 * Otherwise, we process all the files specified. */
876 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
882 for (i = optind; i < argc; i++) {
883 file = wfopen(argv[i], "r");
888 status = EXIT_FAILURE;