1 /* vi: set sw=4 ts=4: */
3 * sed.c - very minimalist version of sed
5 * Copyright (C) 1999,2000,2001 by Lineo, inc. and Mark Whitley
6 * Copyright (C) 1999,2000,2001 by Mark Whitley <markw@codepoet.org>
7 * Copyright (C) 2002 Matt Kraai
8 * Copyright (C) 2003 by Glenn McGrath
9 * Copyright (C) 2003,2004 by Rob Landley <rob@landley.net>
11 * MAINTAINER: Rob Landley <rob@landley.net>
13 * Licensed under GPL version 2, see file LICENSE in this tarball for details.
18 Files are laid out to avoid unnecessary function declarations. So for
19 example, every function add_cmd calls occurs before add_cmd in this file.
21 add_cmd() is called on each line of sed command text (from a file or from
22 the command line). It calls get_address() and parse_cmd_args(). The
23 resulting sed_cmd_t structures are appended to a linked list
24 (G.sed_cmd_head/G.sed_cmd_tail).
26 add_input_file() adds a FILE * to the list of input files. We need to
27 know all input sources ahead of time to find the last line for the $ match.
29 process_files() does actual sedding, reading data lines from each input FILE *
30 (which could be stdin) and applying the sed command list (sed_cmd_head) to
31 each of the resulting lines.
33 sed_main() is where external code calls into this, with a command line.
38 Supported features and commands in this version of sed:
41 - address matching: num|/matchstr/[,num|/matchstr/|$]command
42 - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
43 - edit commands: (a)ppend, (i)nsert, (c)hange
44 - file commands: (r)ead
45 - backreferences in substitution expressions (\0, \1, \2...\9)
46 - grouped commands: {cmd1;cmd2}
47 - transliteration (y/source-chars/dest-chars/)
48 - pattern space hold space storing / swapping (g, h, x)
49 - labels / branching (: label, b, t, T)
51 (Note: Specifying an address (range) to match is *optional*; commands
52 default to the whole pattern space if no specific address match was
56 - Create a wrapper around regex to make libc's regex conform with sed
58 Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
64 /* Each sed command turns into one of these structures. */
65 typedef struct sed_cmd_s {
66 /* Ordered by alignment requirements: currently 36 bytes on x86 */
67 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
70 regex_t *beg_match; /* sed -e '/match/cmd' */
71 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
72 regex_t *sub_match; /* For 's/sub_match/string/' */
73 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
74 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
76 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
77 char *string; /* Data string for (saicytb) commands. */
79 unsigned which_match; /* (s) Which match to replace (0 for all) */
81 /* Bitfields (gcc won't group them if we don't) */
82 unsigned invert:1; /* the '!' after the address */
83 unsigned in_match:1; /* Next line also included in match? */
84 unsigned sub_p:1; /* (s) print option */
86 char sw_last_char; /* Last line written by (sw) had no '\n' */
89 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
92 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
96 int be_quiet, regex_type;
98 char *outname, *hold_space;
100 /* List of input files */
101 int input_file_count, current_input_file;
102 FILE **input_file_list;
104 regmatch_t regmatch[10];
105 regex_t *previous_regex_ptr;
107 /* linked list of sed commands */
108 sed_cmd_t sed_cmd_head, *sed_cmd_tail;
110 /* Linked list of append lines */
111 llist_t *append_head;
116 char *buf; /* Space to hold string */
117 int idx; /* Space used */
118 int len; /* Space allocated */
121 #define G (*(struct globals*)&bb_common_bufsiz1)
122 void BUG_sed_globals_too_big(void);
123 #define INIT_G() do { \
124 if (sizeof(struct globals) > COMMON_BUFSIZE) \
125 BUG_sed_globals_too_big(); \
126 G.sed_cmd_tail = &G.sed_cmd_head; \
130 #if ENABLE_FEATURE_CLEAN_UP
131 static void sed_free_and_close_stuff(void)
133 sed_cmd_t *sed_cmd = G.sed_cmd_head.next;
135 llist_free(G.append_head, free);
138 sed_cmd_t *sed_cmd_next = sed_cmd->next;
140 if (sed_cmd->sw_file)
141 xprint_and_close_file(sed_cmd->sw_file);
143 if (sed_cmd->beg_match) {
144 regfree(sed_cmd->beg_match);
145 free(sed_cmd->beg_match);
147 if (sed_cmd->end_match) {
148 regfree(sed_cmd->end_match);
149 free(sed_cmd->end_match);
151 if (sed_cmd->sub_match) {
152 regfree(sed_cmd->sub_match);
153 free(sed_cmd->sub_match);
155 free(sed_cmd->string);
157 sed_cmd = sed_cmd_next;
162 while (G.current_input_file < G.input_file_count)
163 fclose(G.input_file_list[G.current_input_file++]);
166 void sed_free_and_close_stuff(void);
169 /* If something bad happens during -i operation, delete temp file */
171 static void cleanup_outname(void)
173 if (G.outname) unlink(G.outname);
176 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
178 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
183 if (string[i] == '\\') {
184 if (!to || string[i+1] == from) {
185 *dest++ = to ? to : string[i+1];
189 *dest++ = string[i++];
191 /* TODO: is it safe wrt a string with trailing '\\' ? */
192 *dest++ = string[i++];
197 static char *copy_parsing_escapes(const char *string, int len)
199 char *dest = xmalloc(len + 1);
201 parse_escapes(dest, string, len, 'n', '\n');
202 /* GNU sed also recognizes \t */
203 parse_escapes(dest, dest, strlen(dest), 't', '\t');
209 * index_of_next_unescaped_regexp_delim - walks left to right through a string
210 * beginning at a specified index and returns the index of the next regular
211 * expression delimiter (typically a forward slash ('/')) not preceded by
212 * a backslash ('\'). A negative delimiter disables square bracket checking.
214 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
223 delimiter = -delimiter;
226 for (; (ch = str[idx]); idx++) {
228 if (ch == ']' && !(bracket == idx - 1 || (bracket == idx - 2
229 && str[idx - 1] == '^')))
235 else if (bracket == -1 && ch == '[')
237 else if (ch == delimiter)
241 /* if we make it to here, we've hit the end of the string */
242 bb_error_msg_and_die("unmatched '%c'", delimiter);
246 * Returns the index of the third delimiter
248 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
250 const char *cmdstr_ptr = cmdstr;
254 /* verify that the 's' or 'y' is followed by something. That something
255 * (typically a 'slash') is now our regexp delimiter... */
257 bb_error_msg_and_die("bad format in substitution expression");
258 delimiter = *cmdstr_ptr++;
260 /* save the match string */
261 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
262 *match = copy_parsing_escapes(cmdstr_ptr, idx);
264 /* save the replacement string */
265 cmdstr_ptr += idx + 1;
266 idx = index_of_next_unescaped_regexp_delim(-delimiter, cmdstr_ptr);
267 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
269 return ((cmdstr_ptr - cmdstr) + idx);
273 * returns the index in the string just past where the address ends.
275 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
277 const char *pos = my_str;
279 if (isdigit(*my_str)) {
280 *linenum = strtol(my_str, (char**)&pos, 10);
281 /* endstr shouldnt ever equal NULL */
282 } else if (*my_str == '$') {
285 } else if (*my_str == '/' || *my_str == '\\') {
291 if (*my_str == '\\') delimiter = *++pos;
292 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
293 temp = copy_parsing_escapes(pos, next);
294 *regex = xmalloc(sizeof(regex_t));
295 xregcomp(*regex, temp, G.regex_type|REG_NEWLINE);
297 /* Move position to next character after last delimiter */
303 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
304 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
306 int start = 0, idx, hack = 0;
308 /* Skip whitespace, then grab filename to end of line */
309 while (isspace(filecmdstr[start]))
312 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
315 /* If lines glued together, put backslash back. */
316 if (filecmdstr[idx] == '\n')
319 bb_error_msg_and_die("empty filename");
320 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
322 (*retval)[idx] = '\\';
327 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
329 int cflags = G.regex_type;
334 * A substitution command should look something like this:
335 * s/match/replace/ #gIpw
339 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
341 /* determine the number of back references in the match string */
342 /* Note: we compute this here rather than in the do_subst_command()
343 * function to save processor time, at the expense of a little more memory
344 * (4 bits) per sed_cmd */
346 /* process the flags */
348 sed_cmd->which_match = 1;
349 while (substr[++idx]) {
350 /* Parse match number */
351 if (isdigit(substr[idx])) {
352 if (match[0] != '^') {
353 /* Match 0 treated as all, multiple matches we take the last one. */
354 const char *pos = substr + idx;
355 /* FIXME: error check? */
356 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
362 if (isspace(substr[idx])) continue;
364 switch (substr[idx]) {
365 /* Replace all occurrences */
368 sed_cmd->which_match = 0;
370 /* Print pattern space */
378 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
381 /* Ignore case (gnu exension) */
387 while (substr[++idx]) /*skip all*/;
394 bb_error_msg_and_die("bad option in substitution expression");
398 /* compile the match string into a regex */
399 if (*match != '\0') {
400 /* If match is empty, we use last regex used at runtime */
401 sed_cmd->sub_match = xmalloc(sizeof(regex_t));
402 xregcomp(sed_cmd->sub_match, match, cflags);
410 * Process the commands arguments
412 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
414 /* handle (s)ubstitution command */
415 if (sed_cmd->cmd == 's')
416 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
417 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
418 else if (strchr("aic", sed_cmd->cmd)) {
419 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
421 ("only a beginning address can be specified for edit commands");
423 if (*cmdstr == '\n' || *cmdstr == '\\') {
426 } else if (isspace(*cmdstr))
431 sed_cmd->string = xstrdup(cmdstr);
432 /* "\anychar" -> "anychar" */
433 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
434 cmdstr += strlen(cmdstr);
435 /* handle file cmds: (r)ead */
436 } else if (strchr("rw", sed_cmd->cmd)) {
437 if (sed_cmd->end_line || sed_cmd->end_match)
438 bb_error_msg_and_die("command only uses one address");
439 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
440 if (sed_cmd->cmd == 'w') {
441 sed_cmd->sw_file = xfopen(sed_cmd->string, "w");
442 sed_cmd->sw_last_char = '\n';
444 /* handle branch commands */
445 } else if (strchr(":btT", sed_cmd->cmd)) {
448 cmdstr = skip_whitespace(cmdstr);
449 length = strcspn(cmdstr, semicolon_whitespace);
451 sed_cmd->string = xstrndup(cmdstr, length);
455 /* translation command */
456 else if (sed_cmd->cmd == 'y') {
457 char *match, *replace;
460 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
461 /* \n already parsed, but \delimiter needs unescaping. */
462 parse_escapes(match, match, strlen(match), i, i);
463 parse_escapes(replace, replace, strlen(replace), i, i);
465 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
466 for (i = 0; match[i] && replace[i]; i++) {
467 sed_cmd->string[i*2] = match[i];
468 sed_cmd->string[i*2+1] = replace[i];
473 /* if it wasnt a single-letter command that takes no arguments
474 * then it must be an invalid command.
476 else if (strchr("dDgGhHlnNpPqx={}", sed_cmd->cmd) == 0) {
477 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
480 /* give back whatever's left over */
485 /* Parse address+command sets, skipping comment lines. */
487 static void add_cmd(const char *cmdstr)
492 /* Append this line to any unfinished line from last time. */
493 if (G.add_cmd_line) {
494 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
495 free(G.add_cmd_line);
496 cmdstr = G.add_cmd_line = tp;
499 /* If this line ends with backslash, request next line. */
500 temp = strlen(cmdstr);
501 if (temp && cmdstr[--temp] == '\\') {
503 G.add_cmd_line = xstrdup(cmdstr);
504 G.add_cmd_line[temp] = '\0';
508 /* Loop parsing all commands in this line. */
510 /* Skip leading whitespace and semicolons */
511 cmdstr += strspn(cmdstr, semicolon_whitespace);
513 /* If no more commands, exit. */
516 /* if this is a comment, jump past it and keep going */
517 if (*cmdstr == '#') {
518 /* "#n" is the same as using -n on the command line */
519 if (cmdstr[1] == 'n')
521 cmdstr = strpbrk(cmdstr, "\n\r");
527 * format is: [addr][,addr][!]cmd
532 sed_cmd = xzalloc(sizeof(sed_cmd_t));
534 /* first part (if present) is an address: either a '$', a number or a /regex/ */
535 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
537 /* second part (if present) will begin with a comma */
538 if (*cmdstr == ',') {
542 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
544 bb_error_msg_and_die("no address after comma");
548 /* skip whitespace before the command */
549 cmdstr = skip_whitespace(cmdstr);
551 /* Check for inversion flag */
552 if (*cmdstr == '!') {
556 /* skip whitespace before the command */
557 cmdstr = skip_whitespace(cmdstr);
560 /* last part (mandatory) will be a command */
562 bb_error_msg_and_die("missing command");
563 sed_cmd->cmd = *(cmdstr++);
564 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
566 /* Add the command to the command array */
567 G.sed_cmd_tail->next = sed_cmd;
568 G.sed_cmd_tail = G.sed_cmd_tail->next;
571 /* If we glued multiple lines together, free the memory. */
572 free(G.add_cmd_line);
573 G.add_cmd_line = NULL;
576 /* Append to a string, reallocating memory as necessary. */
580 static void pipe_putc(char c)
582 if (G.pipeline.idx == G.pipeline.len) {
583 G.pipeline.buf = xrealloc(G.pipeline.buf,
584 G.pipeline.len + PIPE_GROW);
585 G.pipeline.len += PIPE_GROW;
587 G.pipeline.buf[G.pipeline.idx++] = c;
590 static void do_subst_w_backrefs(char *line, char *replace)
594 /* go through the replacement string */
595 for (i = 0; replace[i]; i++) {
596 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
597 if (replace[i] == '\\') {
598 unsigned backref = replace[++i] - '0';
600 /* print out the text held in G.regmatch[backref] */
601 if (G.regmatch[backref].rm_so != -1) {
602 j = G.regmatch[backref].rm_so;
603 while (j < G.regmatch[backref].rm_eo)
604 pipe_putc(line[j++]);
608 /* I _think_ it is impossible to get '\' to be
609 * the last char in replace string. Thus we dont check
610 * for replace[i] == NUL. (counterexample anyone?) */
611 /* if we find a backslash escaped character, print the character */
612 pipe_putc(replace[i]);
615 /* if we find an unescaped '&' print out the whole matched text. */
616 if (replace[i] == '&') {
617 j = G.regmatch[0].rm_so;
618 while (j < G.regmatch[0].rm_eo)
619 pipe_putc(line[j++]);
622 /* Otherwise just output the character. */
623 pipe_putc(replace[i]);
627 static int do_subst_command(sed_cmd_t *sed_cmd, char **line)
629 char *oldline = *line;
631 unsigned match_count = 0;
632 regex_t *current_regex;
634 /* Handle empty regex. */
635 if (sed_cmd->sub_match == NULL) {
636 current_regex = G.previous_regex_ptr;
638 bb_error_msg_and_die("no previous regexp");
640 G.previous_regex_ptr = current_regex = sed_cmd->sub_match;
642 /* Find the first match */
643 if (REG_NOMATCH == regexec(current_regex, oldline, 10, G.regmatch, 0))
646 /* Initialize temporary output buffer. */
647 G.pipeline.buf = xmalloc(PIPE_GROW);
648 G.pipeline.len = PIPE_GROW;
651 /* Now loop through, substituting for matches */
655 /* Work around bug in glibc regexec, demonstrated by:
656 echo " a.b" | busybox sed 's [^ .]* x g'
657 The match_count check is so not to break
658 echo "hi" | busybox sed 's/^/!/g' */
659 if (!G.regmatch[0].rm_so && !G.regmatch[0].rm_eo && match_count) {
660 pipe_putc(*oldline++);
666 /* If we aren't interested in this match, output old line to
667 end of match and continue */
668 if (sed_cmd->which_match
669 && (sed_cmd->which_match != match_count)
671 for (i = 0; i < G.regmatch[0].rm_eo; i++)
672 pipe_putc(*oldline++);
676 /* print everything before the match */
677 for (i = 0; i < G.regmatch[0].rm_so; i++)
678 pipe_putc(oldline[i]);
680 /* then print the substitution string */
681 do_subst_w_backrefs(oldline, sed_cmd->string);
683 /* advance past the match */
684 oldline += G.regmatch[0].rm_eo;
685 /* flag that something has changed */
688 /* if we're not doing this globally, get out now */
689 if (sed_cmd->which_match)
691 } while (*oldline && (regexec(current_regex, oldline, 10, G.regmatch, 0) != REG_NOMATCH));
693 /* Copy rest of string into output pipeline */
696 pipe_putc(*oldline++);
700 *line = G.pipeline.buf;
704 /* Set command pointer to point to this label. (Does not handle null label.) */
705 static sed_cmd_t *branch_to(char *label)
709 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
710 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
714 bb_error_msg_and_die("can't find label for jump to '%s'", label);
717 static void append(char *s)
719 llist_add_to_end(&G.append_head, xstrdup(s));
722 static void flush_append(void)
726 /* Output appended lines. */
727 while ((data = (char *)llist_pop(&G.append_head))) {
728 fprintf(G.nonstdout, "%s\n", data);
733 static void add_input_file(FILE *file)
735 G.input_file_list = xrealloc(G.input_file_list,
736 (G.input_file_count + 1) * sizeof(FILE *));
737 G.input_file_list[G.input_file_count++] = file;
740 /* Get next line of input from G.input_file_list, flushing append buffer and
741 * noting if we ran out of files without a newline on the last line we read.
747 static char *get_next_line(char *gets_char)
755 /* will be returned if last line in the file
756 * doesn't end with either '\n' or '\0' */
758 while (G.current_input_file < G.input_file_count) {
759 FILE *fp = G.input_file_list[G.current_input_file];
760 /* Read line up to a newline or NUL byte, inclusive,
761 * return malloc'ed char[]. length of the chunk read
762 * is stored in len. NULL if EOF/error */
763 temp = bb_get_chunk_from_file(fp, &len);
765 /* len > 0 here, it's ok to do temp[len-1] */
766 char c = temp[len-1];
767 if (c == '\n' || c == '\0') {
778 /* else we put NO_EOL_CHAR into *gets_char */
781 /* NB: I had the idea of peeking next file(s) and returning
782 * NO_EOL_CHAR only if it is the *last* non-empty
783 * input file. But there is a case where this won't work:
784 * file1: "a woo\nb woo"
785 * file2: "c no\nd no"
786 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
787 * (note: *no* newline after "b bang"!) */
789 /* Close this file and advance to next one */
791 G.current_input_file++;
797 /* Output line of text. */
799 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
800 * Without them, we had this:
804 * sed "s/i/z/" z1 z2 znull | hexdump -vC
807 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
809 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
811 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
813 char lpc = *last_puts_char;
815 /* Need to insert a '\n' between two files because first file's
816 * last line wasn't terminated? */
817 if (lpc != '\n' && lpc != '\0') {
823 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
827 /* had trailing '\0' and it was last char of file? */
828 if (last_gets_char == LAST_IS_NUL) {
832 /* had trailing '\n' or '\0'? */
833 if (last_gets_char != NO_EOL_CHAR) {
834 fputc(last_gets_char, file);
835 lpc = last_gets_char;
839 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
840 bb_error_msg_and_die(bb_msg_write_error);
842 *last_puts_char = lpc;
845 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
847 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
849 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
851 G.previous_regex_ptr = sed_cmd->beg_match;
855 /* Process all the lines in all the files */
857 static void process_files(void)
859 char *pattern_space, *next_line;
861 char last_puts_char = '\n';
862 char last_gets_char, next_gets_char;
867 next_line = get_next_line(&next_gets_char);
869 /* go through every line in each file */
873 /* Advance to next line. Stop if out of lines. */
874 pattern_space = next_line;
875 if (!pattern_space) return;
876 last_gets_char = next_gets_char;
878 /* Read one line in advance so we can act on the last line,
880 next_line = get_next_line(&next_gets_char);
883 /* for every line, go through all the commands */
884 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
885 int old_matched, matched;
887 old_matched = sed_cmd->in_match;
889 /* Determine if this command matches this line: */
891 /* Are we continuing a previous multi-line match? */
892 sed_cmd->in_match = sed_cmd->in_match
893 /* Or is no range necessary? */
894 || (!sed_cmd->beg_line && !sed_cmd->end_line
895 && !sed_cmd->beg_match && !sed_cmd->end_match)
896 /* Or did we match the start of a numerical range? */
897 || (sed_cmd->beg_line > 0 && (sed_cmd->beg_line == linenum))
898 /* Or does this line match our begin address regex? */
899 || (beg_match(sed_cmd, pattern_space))
900 /* Or did we match last line of input? */
901 || (sed_cmd->beg_line == -1 && next_line == NULL);
903 /* Snapshot the value */
905 matched = sed_cmd->in_match;
907 /* Is this line the end of the current match? */
910 sed_cmd->in_match = !(
911 /* has the ending line come, or is this a single address command? */
913 sed_cmd->end_line == -1 ?
915 : (sed_cmd->end_line <= linenum)
916 : !sed_cmd->end_match
918 /* or does this line matches our last address regex */
919 || (sed_cmd->end_match && old_matched
920 && (regexec(sed_cmd->end_match,
921 pattern_space, 0, NULL, 0) == 0))
925 /* Skip blocks of commands we didn't match. */
926 if (sed_cmd->cmd == '{') {
927 if (sed_cmd->invert ? matched : !matched) {
928 while (sed_cmd->cmd != '}') {
929 sed_cmd = sed_cmd->next;
931 bb_error_msg_and_die("unterminated {");
937 /* Okay, so did this line match? */
938 if (sed_cmd->invert ? !matched : matched) {
939 /* Update last used regex in case a blank substitute BRE is found */
940 if (sed_cmd->beg_match) {
941 G.previous_regex_ptr = sed_cmd->beg_match;
945 switch (sed_cmd->cmd) {
947 /* Print line number */
949 fprintf(G.nonstdout, "%d\n", linenum);
952 /* Write the current pattern space up to the first newline */
955 char *tmp = strchr(pattern_space, '\n');
959 /* TODO: explain why '\n' below */
960 sed_puts(pattern_space, '\n');
967 /* Write the current pattern space to output */
969 /* NB: we print this _before_ the last line
970 * (of current file) is printed. Even if
971 * that line is nonterminated, we print
972 * '\n' here (gnu sed does the same) */
973 sed_puts(pattern_space, '\n');
975 /* Delete up through first newline */
978 char *tmp = strchr(pattern_space, '\n');
981 tmp = xstrdup(tmp+1);
987 /* discard this line. */
991 /* Substitute with regex */
993 if (!do_subst_command(sed_cmd, &pattern_space))
997 /* handle p option */
999 sed_puts(pattern_space, last_gets_char);
1000 /* handle w option */
1001 if (sed_cmd->sw_file)
1003 pattern_space, sed_cmd->sw_file,
1004 &sed_cmd->sw_last_char, last_gets_char);
1007 /* Append line to linked list to be printed later */
1009 append(sed_cmd->string);
1012 /* Insert text before this line */
1014 sed_puts(sed_cmd->string, '\n');
1017 /* Cut and paste text (replace) */
1019 /* Only triggers on last line of a matching range. */
1020 if (!sed_cmd->in_match)
1021 sed_puts(sed_cmd->string, NO_EOL_CHAR);
1024 /* Read file, append contents to output */
1029 rfile = fopen(sed_cmd->string, "r");
1033 while ((line = xmalloc_fgetline(rfile))
1036 xprint_and_close_file(rfile);
1042 /* Write pattern space to file. */
1045 pattern_space, sed_cmd->sw_file,
1046 &sed_cmd->sw_last_char, last_gets_char);
1049 /* Read next line from input */
1052 sed_puts(pattern_space, last_gets_char);
1054 free(pattern_space);
1055 pattern_space = next_line;
1056 last_gets_char = next_gets_char;
1057 next_line = get_next_line(&next_gets_char);
1064 /* Quit. End of script, end of input. */
1066 /* Exit the outer while loop */
1069 goto discard_commands;
1071 /* Append the next line to the current line */
1075 /* If no next line, jump to end of script and exit. */
1076 if (next_line == NULL) {
1077 /* Jump to end of script and exit */
1081 /* append next_line, read new next_line. */
1083 len = strlen(pattern_space);
1084 pattern_space = realloc(pattern_space, len + strlen(next_line) + 2);
1085 pattern_space[len] = '\n';
1086 strcpy(pattern_space + len+1, next_line);
1087 last_gets_char = next_gets_char;
1088 next_line = get_next_line(&next_gets_char);
1093 /* Test/branch if substitution occurred */
1095 if (!substituted) break;
1098 /* Test/branch if substitution didn't occur */
1100 if (substituted) break;
1102 /* Branch to label */
1104 if (!sed_cmd->string) goto discard_commands;
1105 else sed_cmd = branch_to(sed_cmd->string);
1107 /* Transliterate characters */
1112 for (i = 0; pattern_space[i]; i++) {
1113 for (j = 0; sed_cmd->string[j]; j += 2) {
1114 if (pattern_space[i] == sed_cmd->string[j]) {
1115 pattern_space[i] = sed_cmd->string[j + 1];
1123 case 'g': /* Replace pattern space with hold space */
1124 free(pattern_space);
1125 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1127 case 'G': /* Append newline and hold space to pattern space */
1129 int pattern_space_size = 2;
1130 int hold_space_size = 0;
1133 pattern_space_size += strlen(pattern_space);
1135 hold_space_size = strlen(G.hold_space);
1136 pattern_space = xrealloc(pattern_space,
1137 pattern_space_size + hold_space_size);
1138 if (pattern_space_size == 2)
1139 pattern_space[0] = 0;
1140 strcat(pattern_space, "\n");
1142 strcat(pattern_space, G.hold_space);
1143 last_gets_char = '\n';
1147 case 'h': /* Replace hold space with pattern space */
1149 G.hold_space = xstrdup(pattern_space);
1151 case 'H': /* Append newline and pattern space to hold space */
1153 int hold_space_size = 2;
1154 int pattern_space_size = 0;
1157 hold_space_size += strlen(G.hold_space);
1159 pattern_space_size = strlen(pattern_space);
1160 G.hold_space = xrealloc(G.hold_space,
1161 hold_space_size + pattern_space_size);
1163 if (hold_space_size == 2)
1165 strcat(G.hold_space, "\n");
1167 strcat(G.hold_space, pattern_space);
1171 case 'x': /* Exchange hold and pattern space */
1173 char *tmp = pattern_space;
1174 pattern_space = G.hold_space ? : xzalloc(1);
1175 last_gets_char = '\n';
1184 * exit point from sedding...
1187 /* we will print the line unless we were told to be quiet ('-n')
1188 or if the line was suppressed (ala 'd'elete) */
1190 sed_puts(pattern_space, last_gets_char);
1192 /* Delete and such jump here. */
1195 free(pattern_space);
1200 /* It is possible to have a command line argument with embedded
1201 * newlines. This counts as multiple command lines.
1202 * However, newline can be escaped: 's/e/z\<newline>z/'
1203 * We check for this.
1206 static void add_cmd_block(char *cmdstr)
1210 cmdstr = sv = xstrdup(cmdstr);
1212 eol = strchr(cmdstr, '\n');
1215 /* Count preceding slashes */
1219 while (sl != cmdstr && *--sl == '\\')
1221 /* Odd number of preceding slashes - newline is escaped */
1224 eol = strchr(eol, '\n');
1235 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1236 int sed_main(int argc ATTRIBUTE_UNUSED, char **argv)
1239 OPT_in_place = 1 << 0,
1242 llist_t *opt_e, *opt_f;
1243 int status = EXIT_SUCCESS;
1247 /* destroy command strings on exit */
1248 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1250 /* Lie to autoconf when it starts asking stupid questions. */
1251 if (argv[1] && !strcmp(argv[1], "--version")) {
1252 puts("This is not GNU sed version 4.0");
1256 /* do normal option parsing */
1257 opt_e = opt_f = NULL;
1258 opt_complementary = "e::f::" /* can occur multiple times */
1259 "nn"; /* count -n */
1260 opt = getopt32(argv, "irne:f:", &opt_e, &opt_f,
1261 &G.be_quiet); /* counter for -n */
1264 if (opt & OPT_in_place) { // -i
1265 atexit(cleanup_outname);
1267 if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r
1268 //if (opt & 0x4) G.be_quiet++; // -n
1269 while (opt_e) { // -e
1270 add_cmd_block(llist_pop(&opt_e));
1272 while (opt_f) { // -f
1275 cmdfile = xfopen(llist_pop(&opt_f), "r");
1276 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1282 /* if we didn't get a pattern from -e or -f, use argv[0] */
1283 if (!(opt & 0x18)) {
1286 add_cmd_block(*argv++);
1288 /* Flush any unfinished commands. */
1291 /* By default, we write to stdout */
1292 G.nonstdout = stdout;
1294 /* argv[0..(argc-1)] should be names of file to process. If no
1295 * files were specified or '-' was specified, take input from stdin.
1296 * Otherwise, we process all the files specified. */
1297 if (argv[0] == NULL) {
1298 if (opt & OPT_in_place)
1299 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1300 add_input_file(stdin);
1306 for (i = 0; argv[i]; i++) {
1307 struct stat statbuf;
1310 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1311 add_input_file(stdin);
1315 file = fopen_or_warn(argv[i], "r");
1317 status = EXIT_FAILURE;
1320 if (!(opt & OPT_in_place)) {
1321 add_input_file(file);
1325 G.outname = xasprintf("%sXXXXXX", argv[i]);
1326 nonstdoutfd = mkstemp(G.outname);
1327 if (-1 == nonstdoutfd)
1328 bb_perror_msg_and_die("cannot create temp file %s", G.outname);
1329 G.nonstdout = fdopen(nonstdoutfd, "w");
1331 /* Set permissions of output file */
1333 fstat(fileno(file), &statbuf);
1334 fchmod(nonstdoutfd, statbuf.st_mode);
1335 add_input_file(file);
1337 fclose(G.nonstdout);
1339 G.nonstdout = stdout;
1340 /* unlink(argv[i]); */
1341 xrename(G.outname, argv[i]);
1345 if (G.input_file_count > G.current_input_file)