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 struct sed_cmd * const sed_cmd, 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 == sed_cmd->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(struct sed_cmd *sed_cmd, const char *str, int *linenum, regex_t **regex)
181 char *my_str = xstrdup(str);
182 int idx = 0, idx_start = 1;
184 olddelimiter = sed_cmd->delimiter;
185 sed_cmd->delimiter = '/';
187 if (isdigit(my_str[idx])) {
190 } while (isdigit(my_str[idx]));
192 *linenum = atoi(my_str);
194 else if (my_str[idx] == '$') {
198 else if (my_str[idx] == '/' || my_str[idx] == '\\') {
199 if (my_str[idx] == '\\') {
201 sed_cmd-> delimiter = my_str[++idx];
203 idx = index_of_next_unescaped_regexp_delim(sed_cmd, my_str, ++idx);
205 error_msg_and_die("unterminated match expression");
207 *regex = (regex_t *)xmalloc(sizeof(regex_t));
208 xregcomp(*regex, my_str+idx_start, REG_NEWLINE);
209 idx++; /* so it points to the next character after the last '/' */
212 error_msg("get_address: no address found in string\n"
213 "\t(you probably didn't check the string you passed me)");
218 sed_cmd->delimiter = olddelimiter;
222 static int parse_subst_cmd(struct sed_cmd * const sed_cmd, const char *substr)
224 int oldidx, cflags = REG_NEWLINE;
230 * the string that gets passed to this function should look like this:
231 * s/match/replace/gIp
235 * (all three of the '/' slashes are mandatory)
238 /* verify that the 's' is followed by something. That something
239 * (typically a 'slash') is now our regexp delimiter... */
241 error_msg_and_die("bad format in substitution expression");
243 sed_cmd->delimiter=substr[idx];
245 /* save the match string */
247 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
249 error_msg_and_die("bad format in substitution expression");
250 match = xstrndup(substr + oldidx, idx - oldidx);
252 /* determine the number of back references in the match string */
253 /* Note: we compute this here rather than in the do_subst_command()
254 * function to save processor time, at the expense of a little more memory
255 * (4 bits) per sed_cmd */
257 /* sed_cmd->num_backrefs = 0; */ /* XXX: not needed? --apparently not */
258 for (j = 0; match[j]; j++) {
259 /* GNU/POSIX sed does not save more than nine backrefs */
260 if (match[j] == '\\' && match[j+1] == '(' && sed_cmd->num_backrefs <= 9)
261 sed_cmd->num_backrefs++;
264 /* save the replacement string */
266 idx = index_of_next_unescaped_regexp_delim(sed_cmd, substr, ++idx);
268 error_msg_and_die("bad format in substitution expression");
269 sed_cmd->replace = xstrndup(substr + oldidx, idx - oldidx);
271 /* process the flags */
272 while (substr[++idx]) {
273 switch (substr[idx]) {
284 /* any whitespace or semicolon trailing after a s/// is ok */
285 if (strchr(semicolon_whitespace, substr[idx]))
288 error_msg_and_die("bad option in substitution expression");
293 /* compile the match string into a regex */
294 sed_cmd->sub_match = (regex_t *)xmalloc(sizeof(regex_t));
295 xregcomp(sed_cmd->sub_match, match, cflags);
301 static void move_back(char *str, int offset)
303 memmove(str, str + offset, strlen(str + offset) + 1);
306 static int parse_edit_cmd(struct sed_cmd *sed_cmd, const char *editstr)
311 * the string that gets passed to this function should look like this:
315 * | this backslash (immediately following the edit command) is mandatory
322 * as soon as we hit a TEXT line that has no trailing '\', we're done.
323 * this means a command like:
332 if (editstr[1] != '\\' || (editstr[2] != '\n' && editstr[2] != '\r'))
333 error_msg_and_die("bad format in edit expression");
335 /* store the edit line text */
336 sed_cmd->editline = xmalloc(strlen(&editstr[3]) + 2);
337 for (i = 3, j = 0; editstr[i] != '\0' && strchr("\r\n", editstr[i]) == NULL;
339 if (editstr[i] == '\\' && strchr("\n\r", editstr[i+1]) != NULL) {
340 sed_cmd->editline[j] = '\n';
343 sed_cmd->editline[j] = editstr[i];
346 /* figure out if we need to add a newline */
347 if (sed_cmd->editline[j-1] != '\n')
348 sed_cmd->editline[j++] = '\n';
350 /* terminate string */
351 sed_cmd->editline[j] = '\0';
357 static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
363 * the string that gets passed to this function should look like this:
368 * optional whitespace
370 * re: the file to be read, the GNU manual says the following: "Note that
371 * if filename cannot be read, it is treated as if it were an empty file,
372 * without any error indication." Thus, all of the following commands are
375 * sed -e '1r noexist'
380 /* the file command may be followed by whitespace; move past it. */
381 while (isspace(filecmdstr[++idx]))
384 /* the first non-whitespace we get is a filename. the filename ends when we
385 * hit a normal sed command terminator or end of string */
386 filenamelen = strcspn(&filecmdstr[idx], semicolon_whitespace);
387 sed_cmd->filename = xmalloc(filenamelen + 1);
388 safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
390 return idx + filenamelen;
394 static char *parse_cmd_str(struct sed_cmd * const sed_cmd, const char *const cmdstr)
399 * format is: [addr][,addr]cmd
404 /* first part (if present) is an address: either a '$', a number or a /regex/ */
405 if ((cmdstr[idx] == '$') || (isdigit(cmdstr[idx])) || (cmdstr[idx] == '/') || ((cmdstr[idx] == '\\') && (cmdstr[idx+1] != '\\')))
406 idx = get_address(sed_cmd, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
408 /* second part (if present) will begin with a comma */
409 if (cmdstr[idx] == ',') {
411 idx += get_address(sed_cmd, &cmdstr[idx], &sed_cmd->end_line, &sed_cmd->end_match);
414 /* skip whitespace before the command */
415 while (isspace(cmdstr[idx]))
418 /* there my be the inversion flag between part2 and part3 */
420 if (cmdstr[idx] == '!') {
424 #ifdef SED_FEATURE_STRICT_CHECKING
425 /* According to the spec
426 * It is unspecified whether <blank>s can follow a '!' character,
427 * and conforming applications shall not follow a '!' character
430 if (isblank(cmdstr[idx]) {
431 error_msg_and_die("blank follows '!'");
434 /* skip whitespace before the command */
435 while (isspace(cmdstr[idx]))
440 /* last part (mandatory) will be a command */
441 if (cmdstr[idx] == '\0')
442 error_msg_and_die("missing command");
443 sed_cmd->cmd = cmdstr[idx];
445 /* if it was a single-letter command that takes no arguments (such as 'p'
446 * or 'd') all we need to do is increment the index past that command */
447 if (strchr("pd=", sed_cmd->cmd)) {
450 /* handle (s)ubstitution command */
451 else if (sed_cmd->cmd == 's') {
452 idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
454 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
455 else if (strchr("aic", sed_cmd->cmd)) {
456 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
457 error_msg_and_die("only a beginning address can be specified for edit commands");
458 idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
460 /* handle file cmds: (r)ead */
461 else if (sed_cmd->cmd == 'r') {
462 if (sed_cmd->end_line || sed_cmd->end_match)
463 error_msg_and_die("Command only uses one address");
464 idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
467 error_msg_and_die("Unsupported command %c", sed_cmd->cmd);
470 /* give back whatever's left over */
471 return (char *)&cmdstr[idx];
474 static void add_cmd_str(const char * const cmdstr)
476 char *mystr = (char *)cmdstr;
480 /* trim leading whitespace and semicolons */
481 move_back(mystr, strspn(mystr, semicolon_whitespace));
482 /* if we ate the whole thing, that means there was just trailing
483 * whitespace or a final / no-op semicolon. either way, get out */
484 if (strlen(mystr) == 0)
486 /* if this is a comment, jump past it and keep going */
487 if (mystr[0] == '#') {
488 mystr = strpbrk(mystr, "\n\r");
492 sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
493 /* zero new element */
494 memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
495 /* load command string into new array element, get remainder */
496 mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
498 } while (mystr && strlen(mystr));
502 static void load_cmd_file(char *filename)
508 cmdfile = xfopen(filename, "r");
510 while ((line = get_line_from_file(cmdfile)) != NULL) {
511 /* if a line ends with '\' it needs the next line appended to it */
512 while (line[strlen(line)-2] == '\\' &&
513 (nextline = get_line_from_file(cmdfile)) != NULL) {
514 line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
515 strcat(line, nextline);
518 /* eat trailing newline (if any) --if I don't do this, edit commands
519 * (aic) will print an extra newline */
532 #define PIPE_MAGIC 0x7f
535 void pipe_putc(struct pipeline *const pipeline, char c)
537 if (pipeline->buf[pipeline->idx] == PIPE_MAGIC) {
539 xrealloc(pipeline->buf, pipeline->len + PIPE_GROW);
540 memset(pipeline->buf + pipeline->len, 0, PIPE_GROW);
541 pipeline->len += PIPE_GROW;
542 pipeline->buf[pipeline->len - 1] = PIPE_MAGIC;
544 pipeline->buf[pipeline->idx++] = (c);
547 #define pipeputc(c) pipe_putc(pipeline, c)
550 { if (pipeline[pipeline_idx] == PIPE_MAGIC) { \
551 pipeline = xrealloc(pipeline, pipeline_len+PIPE_GROW); \
552 memset(pipeline+pipeline_len, 0, PIPE_GROW); \
553 pipeline_len += PIPE_GROW; \
554 pipeline[pipeline_len-1] = PIPE_MAGIC; } \
555 pipeline[pipeline_idx++] = (c); }
558 static void print_subst_w_backrefs(const char *line, const char *replace,
559 regmatch_t *regmatch, struct pipeline *const pipeline, int matches)
563 /* go through the replacement string */
564 for (i = 0; replace[i]; i++) {
565 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
566 if (replace[i] == '\\' && isdigit(replace[i+1])) {
570 ++i; /* i now indexes the backref number, instead of the leading slash */
571 tmpstr[0] = replace[i];
573 backref = atoi(tmpstr);
574 /* print out the text held in regmatch[backref] */
575 if (backref <= matches && regmatch[backref].rm_so != -1)
576 for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
580 /* if we find a backslash escaped character, print the character */
581 else if (replace[i] == '\\') {
583 pipeputc(replace[i]);
586 /* if we find an unescaped '&' print out the whole matched text.
587 * fortunately, regmatch[0] contains the indicies to the whole matched
588 * expression (kinda seems like it was designed for just such a
590 else if (replace[i] == '&' && replace[i-1] != '\\') {
592 for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
595 /* nothing special, just print this char of the replacement string to stdout */
597 pipeputc(replace[i]);
601 static int do_subst_command(const struct sed_cmd *sed_cmd, char **line)
603 char *hackline = *line;
604 struct pipeline thepipe = { NULL, 0 , 0};
605 struct pipeline *const pipeline = &thepipe;
607 regmatch_t *regmatch = NULL;
609 /* we only proceed if the substitution 'search' expression matches */
610 if (regexec(sed_cmd->sub_match, hackline, 0, NULL, 0) == REG_NOMATCH)
613 /* whaddaya know, it matched. get the number of back references */
614 regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
616 /* allocate more PIPE_GROW bytes
617 if replaced string is larger than original */
618 thepipe.len = strlen(hackline)+PIPE_GROW;
619 thepipe.buf = xcalloc(1, thepipe.len);
621 thepipe.buf[thepipe.len-1] = PIPE_MAGIC;
623 /* and now, as long as we've got a line to try matching and if we can match
624 * the search string, we make substitutions */
625 while ((*hackline || !altered) && (regexec(sed_cmd->sub_match, hackline,
626 sed_cmd->num_backrefs+1, regmatch, 0) != REG_NOMATCH) ) {
629 /* print everything before the match */
630 for (i = 0; i < regmatch[0].rm_so; i++)
631 pipeputc(hackline[i]);
633 /* then print the substitution string */
634 print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch,
635 pipeline, sed_cmd->num_backrefs);
637 /* advance past the match */
638 hackline += regmatch[0].rm_eo;
639 /* flag that something has changed */
642 /* if we're not doing this globally, get out now */
647 for (; *hackline; hackline++) pipeputc(*hackline);
648 if (thepipe.buf[thepipe.idx] == PIPE_MAGIC) thepipe.buf[thepipe.idx] = 0;
659 static void process_file(FILE *file)
662 static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
663 unsigned int still_in_range = 0;
667 line = get_line_from_file(file);
672 /* go through every line in the file */
676 /* Read one line in advance so we can act on the last line, the '$' address */
677 next_line = get_line_from_file(file);
683 /* for every line, go through all the commands */
684 for (i = 0; i < ncmds; i++) {
685 struct sed_cmd *sed_cmd = &sed_cmds[i];
689 * entry point into sedding...
692 /* no range necessary */
693 (sed_cmd->beg_line == 0 && sed_cmd->end_line == 0 &&
694 sed_cmd->beg_match == NULL &&
695 sed_cmd->end_match == NULL) ||
696 /* this line number is the first address we're looking for */
697 (sed_cmd->beg_line && (sed_cmd->beg_line == linenum)) ||
698 /* this line matches our first address regex */
699 (sed_cmd->beg_match && (regexec(sed_cmd->beg_match, line, 0, NULL, 0) == 0)) ||
700 /* we are currently within the beginning & ending address range */
701 still_in_range || ((sed_cmd->beg_line == -1) && (next_line == NULL))
704 if (sed_cmd->invert ^ matched) {
709 switch (sed_cmd->cmd) {
711 printf("%d\n", linenum);
725 * Some special cases for 's' printing to make it compliant with
726 * GNU sed printing behavior (aka "The -n | s///p Matrix"):
728 * -n ONLY = never print anything regardless of any successful
731 * s///p ONLY = always print successful substitutions, even if
732 * the line is going to be printed anyway (line will be printed
735 * -n AND s///p = print ONLY a successful substitution ONE TIME;
736 * no other lines are printed - this is the reason why the 'p'
737 * flag exists in the first place.
740 /* if the user specified that they didn't want anything printed (i.e., a -n
741 * flag and no 'p' flag after the s///), then there's really no point doing
743 if (be_quiet && !sed_cmd->sub_p)
746 /* we print the line once, unless we were told to be quiet */
748 altered |= do_subst_command(sed_cmd, &line);
750 /* we also print the line if we were given the 'p' flag
751 * (this is quite possibly the second printing) */
753 altered |= do_subst_command(sed_cmd, &line);
754 if (altered && (i+1 >= ncmds || sed_cmds[i+1].cmd != 's'))
761 fputs(sed_cmd->editline, stdout);
766 fputs(sed_cmd->editline, stdout);
770 /* single-address case */
771 if ((sed_cmd->end_match == NULL && sed_cmd->end_line == 0)
772 /* multi-address case */
773 /* - matching text */
774 || (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
775 /* - matching line numbers */
776 || (sed_cmd->end_line > 0 && sed_cmd->end_line == linenum))
778 fputs(sed_cmd->editline, stdout);
787 outfile = fopen(sed_cmd->filename, "r");
790 /* else if we couldn't open the output file,
791 * no biggie, just don't print anything */
799 * exit point from sedding...
803 /* this is a single-address command or... */
804 (sed_cmd->end_line == 0 && sed_cmd->end_match == NULL) || (
805 /* we were in the middle of our address range (this
806 * isn't the first time through) and.. */
807 (still_in_range == 1) && (
808 /* this line number is the last address we're looking for or... */
809 (sed_cmd->end_line && (sed_cmd->end_line == linenum)) ||
810 /* this line matches our last address regex */
811 (sed_cmd->end_match && (regexec(sed_cmd->end_match, line, 0, NULL, 0) == 0))
815 /* we're out of our address range */
819 /* didn't hit the exit? then we're still in the middle of an address range */
829 /* we will print the line unless we were told to be quiet or if the
830 * line was altered (via a 'd'elete or 's'ubstitution), in which case
831 * the altered line was already printed */
832 if (!be_quiet && !altered)
840 extern int sed_main(int argc, char **argv)
842 int opt, status = EXIT_SUCCESS;
844 #ifdef CONFIG_FEATURE_CLEAN_UP
845 /* destroy command strings on exit */
846 if (atexit(destroy_cmd_strs) == -1)
847 perror_msg_and_die("atexit");
850 /* do normal option parsing */
851 while ((opt = getopt(argc, argv, "ne:f:")) > 0) {
860 load_cmd_file(optarg);
867 /* if we didn't get a pattern from a -e and no command file was specified,
868 * argv[optind] should be the pattern. no pattern, no worky */
870 if (argv[optind] == NULL)
873 add_cmd_str(argv[optind]);
879 /* argv[(optind)..(argc-1)] should be names of file to process. If no
880 * files were specified or '-' was specified, take input from stdin.
881 * Otherwise, we process all the files specified. */
882 if (argv[optind] == NULL || (strcmp(argv[optind], "-") == 0)) {
888 for (i = optind; i < argc; i++) {
889 file = wfopen(argv[i], "r");
894 status = EXIT_FAILURE;