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 GPLv2, see file LICENSE in this source tree.
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.
36 /* Supported features and commands in this version of sed:
39 * - address matching: num|/matchstr/[,num|/matchstr/|$]command
40 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
41 * - edit commands: (a)ppend, (i)nsert, (c)hange
42 * - file commands: (r)ead
43 * - backreferences in substitution expressions (\0, \1, \2...\9)
44 * - grouped commands: {cmd1;cmd2}
45 * - transliteration (y/source-chars/dest-chars/)
46 * - pattern space hold space storing / swapping (g, h, x)
47 * - labels / branching (: label, b, t, T)
49 * (Note: Specifying an address (range) to match is *optional*; commands
50 * default to the whole pattern space if no specific address match was
54 * - Create a wrapper around regex to make libc's regex conform with sed
56 * Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 //usage:#define sed_trivial_usage
60 //usage: "[-efinr] SED_CMD [FILE]..."
61 //usage:#define sed_full_usage "\n\n"
63 //usage: "\n -e CMD Add CMD to sed commands to be executed"
64 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
65 //usage: "\n -i Edit files in-place (else sends result to stdout)"
66 //usage: "\n -n Suppress automatic printing of pattern space"
67 //usage: "\n -r Use extended regex syntax"
69 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
70 //usage: "\nRemaining arguments are input files (stdin if none)."
72 //usage:#define sed_example_usage
73 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
80 OPT_in_place = 1 << 0,
83 /* Each sed command turns into one of these structures. */
84 typedef struct sed_cmd_s {
85 /* Ordered by alignment requirements: currently 36 bytes on x86 */
86 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
89 regex_t *beg_match; /* sed -e '/match/cmd' */
90 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
91 regex_t *sub_match; /* For 's/sub_match/string/' */
92 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
93 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
95 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
96 char *string; /* Data string for (saicytb) commands. */
98 unsigned which_match; /* (s) Which match to replace (0 for all) */
100 /* Bitfields (gcc won't group them if we don't) */
101 unsigned invert:1; /* the '!' after the address */
102 unsigned in_match:1; /* Next line also included in match? */
103 unsigned sub_p:1; /* (s) print option */
105 char sw_last_char; /* Last line written by (sw) had no '\n' */
108 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
111 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
115 int be_quiet, regex_type;
117 char *outname, *hold_space;
119 /* List of input files */
120 int input_file_count, current_input_file;
121 FILE **input_file_list;
123 regmatch_t regmatch[10];
124 regex_t *previous_regex_ptr;
126 /* linked list of sed commands */
127 sed_cmd_t sed_cmd_head, *sed_cmd_tail;
129 /* Linked list of append lines */
130 llist_t *append_head;
135 char *buf; /* Space to hold string */
136 int idx; /* Space used */
137 int len; /* Space allocated */
140 #define G (*(struct globals*)&bb_common_bufsiz1)
141 struct BUG_G_too_big {
142 char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
144 #define INIT_G() do { \
145 G.sed_cmd_tail = &G.sed_cmd_head; \
149 #if ENABLE_FEATURE_CLEAN_UP
150 static void sed_free_and_close_stuff(void)
152 sed_cmd_t *sed_cmd = G.sed_cmd_head.next;
154 llist_free(G.append_head, free);
157 sed_cmd_t *sed_cmd_next = sed_cmd->next;
159 if (sed_cmd->sw_file)
160 xprint_and_close_file(sed_cmd->sw_file);
162 if (sed_cmd->beg_match) {
163 regfree(sed_cmd->beg_match);
164 free(sed_cmd->beg_match);
166 if (sed_cmd->end_match) {
167 regfree(sed_cmd->end_match);
168 free(sed_cmd->end_match);
170 if (sed_cmd->sub_match) {
171 regfree(sed_cmd->sub_match);
172 free(sed_cmd->sub_match);
174 free(sed_cmd->string);
176 sed_cmd = sed_cmd_next;
181 while (G.current_input_file < G.input_file_count)
182 fclose(G.input_file_list[G.current_input_file++]);
185 void sed_free_and_close_stuff(void);
188 /* If something bad happens during -i operation, delete temp file */
190 static void cleanup_outname(void)
192 if (G.outname) unlink(G.outname);
195 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
197 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
202 if (string[i] == '\\') {
203 if (!to || string[i+1] == from) {
204 *dest++ = to ? to : string[i+1];
208 *dest++ = string[i++];
210 /* TODO: is it safe wrt a string with trailing '\\' ? */
211 *dest++ = string[i++];
216 static char *copy_parsing_escapes(const char *string, int len)
218 char *dest = xmalloc(len + 1);
220 parse_escapes(dest, string, len, 'n', '\n');
221 /* GNU sed also recognizes \t and \r */
222 parse_escapes(dest, dest, strlen(dest), 't', '\t');
223 parse_escapes(dest, dest, strlen(dest), 'r', '\r');
229 * index_of_next_unescaped_regexp_delim - walks left to right through a string
230 * beginning at a specified index and returns the index of the next regular
231 * expression delimiter (typically a forward slash ('/')) not preceded by
232 * a backslash ('\'). A negative delimiter disables square bracket checking.
234 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
243 delimiter = -delimiter;
246 for (; (ch = str[idx]) != '\0'; idx++) {
249 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
257 else if (bracket == -1 && ch == '[')
259 else if (ch == delimiter)
263 /* if we make it to here, we've hit the end of the string */
264 bb_error_msg_and_die("unmatched '%c'", delimiter);
268 * Returns the index of the third delimiter
270 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
272 const char *cmdstr_ptr = cmdstr;
276 /* verify that the 's' or 'y' is followed by something. That something
277 * (typically a 'slash') is now our regexp delimiter... */
279 bb_error_msg_and_die("bad format in substitution expression");
280 delimiter = *cmdstr_ptr++;
282 /* save the match string */
283 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
284 *match = copy_parsing_escapes(cmdstr_ptr, idx);
286 /* save the replacement string */
287 cmdstr_ptr += idx + 1;
288 idx = index_of_next_unescaped_regexp_delim(-delimiter, cmdstr_ptr);
289 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
291 return ((cmdstr_ptr - cmdstr) + idx);
295 * returns the index in the string just past where the address ends.
297 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
299 const char *pos = my_str;
301 if (isdigit(*my_str)) {
302 *linenum = strtol(my_str, (char**)&pos, 10);
303 /* endstr shouldnt ever equal NULL */
304 } else if (*my_str == '$') {
307 } else if (*my_str == '/' || *my_str == '\\') {
313 if (*my_str == '\\') delimiter = *++pos;
314 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
315 temp = copy_parsing_escapes(pos, next);
316 *regex = xmalloc(sizeof(regex_t));
317 xregcomp(*regex, temp, G.regex_type|REG_NEWLINE);
319 /* Move position to next character after last delimiter */
325 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
326 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
328 int start = 0, idx, hack = 0;
330 /* Skip whitespace, then grab filename to end of line */
331 while (isspace(filecmdstr[start]))
334 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
337 /* If lines glued together, put backslash back. */
338 if (filecmdstr[idx] == '\n')
341 bb_error_msg_and_die("empty filename");
342 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
344 (*retval)[idx] = '\\';
349 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
351 int cflags = G.regex_type;
356 * A substitution command should look something like this:
357 * s/match/replace/ #gIpw
361 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
363 /* determine the number of back references in the match string */
364 /* Note: we compute this here rather than in the do_subst_command()
365 * function to save processor time, at the expense of a little more memory
366 * (4 bits) per sed_cmd */
368 /* process the flags */
370 sed_cmd->which_match = 1;
371 while (substr[++idx]) {
372 /* Parse match number */
373 if (isdigit(substr[idx])) {
374 if (match[0] != '^') {
375 /* Match 0 treated as all, multiple matches we take the last one. */
376 const char *pos = substr + idx;
377 /* FIXME: error check? */
378 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
384 if (isspace(substr[idx]))
387 switch (substr[idx]) {
388 /* Replace all occurrences */
391 sed_cmd->which_match = 0;
393 /* Print pattern space */
401 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
404 /* Ignore case (gnu exension) */
410 // while (substr[++idx]) continue;
411 idx += strlen(substr + idx); // same
418 bb_error_msg_and_die("bad option in substitution expression");
422 /* compile the match string into a regex */
423 if (*match != '\0') {
424 /* If match is empty, we use last regex used at runtime */
425 sed_cmd->sub_match = xmalloc(sizeof(regex_t));
426 xregcomp(sed_cmd->sub_match, match, cflags);
434 * Process the commands arguments
436 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
438 static const char cmd_letters[] = "saicrw:btTydDgGhHlnNpPqx={}";
469 struct chk { char chk[sizeof(cmd_letters)-1 == IDX_nul ? 1 : -1]; };
471 unsigned idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
473 /* handle (s)ubstitution command */
475 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
477 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
478 else if (idx <= IDX_c) { /* a,i,c */
479 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
480 bb_error_msg_and_die("only a beginning address can be specified for edit commands");
482 if (*cmdstr == '\n' || *cmdstr == '\\') {
486 if (!isspace(*cmdstr))
490 sed_cmd->string = xstrdup(cmdstr);
491 /* "\anychar" -> "anychar" */
492 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
493 cmdstr += strlen(cmdstr);
495 /* handle file cmds: (r)ead */
496 else if (idx <= IDX_w) { /* r,w */
497 if (sed_cmd->end_line || sed_cmd->end_match)
498 bb_error_msg_and_die("command only uses one address");
499 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
500 if (sed_cmd->cmd == 'w') {
501 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
502 sed_cmd->sw_last_char = '\n';
505 /* handle branch commands */
506 else if (idx <= IDX_T) { /* :,b,t,T */
509 cmdstr = skip_whitespace(cmdstr);
510 length = strcspn(cmdstr, semicolon_whitespace);
512 sed_cmd->string = xstrndup(cmdstr, length);
516 /* translation command */
517 else if (idx == IDX_y) {
518 char *match, *replace;
521 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
522 /* \n already parsed, but \delimiter needs unescaping. */
523 parse_escapes(match, match, strlen(match), i, i);
524 parse_escapes(replace, replace, strlen(replace), i, i);
526 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
527 for (i = 0; match[i] && replace[i]; i++) {
528 sed_cmd->string[i*2] = match[i];
529 sed_cmd->string[i*2+1] = replace[i];
534 /* if it wasnt a single-letter command that takes no arguments
535 * then it must be an invalid command.
537 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
538 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
541 /* give back whatever's left over */
546 /* Parse address+command sets, skipping comment lines. */
548 static void add_cmd(const char *cmdstr)
553 /* Append this line to any unfinished line from last time. */
554 if (G.add_cmd_line) {
555 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
556 free(G.add_cmd_line);
557 cmdstr = G.add_cmd_line = tp;
560 /* If this line ends with unescaped backslash, request next line. */
561 n = len = strlen(cmdstr);
562 while (n && cmdstr[n-1] == '\\')
564 if ((len - n) & 1) { /* if odd number of trailing backslashes */
566 G.add_cmd_line = xstrdup(cmdstr);
567 G.add_cmd_line[len-1] = '\0';
571 /* Loop parsing all commands in this line. */
573 /* Skip leading whitespace and semicolons */
574 cmdstr += strspn(cmdstr, semicolon_whitespace);
576 /* If no more commands, exit. */
579 /* if this is a comment, jump past it and keep going */
580 if (*cmdstr == '#') {
581 /* "#n" is the same as using -n on the command line */
582 if (cmdstr[1] == 'n')
584 cmdstr = strpbrk(cmdstr, "\n\r");
590 * format is: [addr][,addr][!]cmd
595 sed_cmd = xzalloc(sizeof(sed_cmd_t));
597 /* first part (if present) is an address: either a '$', a number or a /regex/ */
598 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
600 /* second part (if present) will begin with a comma */
601 if (*cmdstr == ',') {
605 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
607 bb_error_msg_and_die("no address after comma");
611 /* skip whitespace before the command */
612 cmdstr = skip_whitespace(cmdstr);
614 /* Check for inversion flag */
615 if (*cmdstr == '!') {
619 /* skip whitespace before the command */
620 cmdstr = skip_whitespace(cmdstr);
623 /* last part (mandatory) will be a command */
625 bb_error_msg_and_die("missing command");
626 sed_cmd->cmd = *cmdstr++;
627 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
629 /* Add the command to the command array */
630 G.sed_cmd_tail->next = sed_cmd;
631 G.sed_cmd_tail = G.sed_cmd_tail->next;
634 /* If we glued multiple lines together, free the memory. */
635 free(G.add_cmd_line);
636 G.add_cmd_line = NULL;
639 /* Append to a string, reallocating memory as necessary. */
643 static void pipe_putc(char c)
645 if (G.pipeline.idx == G.pipeline.len) {
646 G.pipeline.buf = xrealloc(G.pipeline.buf,
647 G.pipeline.len + PIPE_GROW);
648 G.pipeline.len += PIPE_GROW;
650 G.pipeline.buf[G.pipeline.idx++] = c;
653 static void do_subst_w_backrefs(char *line, char *replace)
657 /* go through the replacement string */
658 for (i = 0; replace[i]; i++) {
659 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
660 if (replace[i] == '\\') {
661 unsigned backref = replace[++i] - '0';
663 /* print out the text held in G.regmatch[backref] */
664 if (G.regmatch[backref].rm_so != -1) {
665 j = G.regmatch[backref].rm_so;
666 while (j < G.regmatch[backref].rm_eo)
667 pipe_putc(line[j++]);
671 /* I _think_ it is impossible to get '\' to be
672 * the last char in replace string. Thus we dont check
673 * for replace[i] == NUL. (counterexample anyone?) */
674 /* if we find a backslash escaped character, print the character */
675 pipe_putc(replace[i]);
678 /* if we find an unescaped '&' print out the whole matched text. */
679 if (replace[i] == '&') {
680 j = G.regmatch[0].rm_so;
681 while (j < G.regmatch[0].rm_eo)
682 pipe_putc(line[j++]);
685 /* Otherwise just output the character. */
686 pipe_putc(replace[i]);
690 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
692 char *line = *line_p;
694 unsigned match_count = 0;
695 regex_t *current_regex;
697 current_regex = sed_cmd->sub_match;
698 /* Handle empty regex. */
699 if (!current_regex) {
700 current_regex = G.previous_regex_ptr;
702 bb_error_msg_and_die("no previous regexp");
704 G.previous_regex_ptr = current_regex;
706 /* Find the first match */
707 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0))
710 /* Initialize temporary output buffer. */
711 G.pipeline.buf = xmalloc(PIPE_GROW);
712 G.pipeline.len = PIPE_GROW;
715 /* Now loop through, substituting for matches */
719 /* Work around bug in glibc regexec, demonstrated by:
720 echo " a.b" | busybox sed 's [^ .]* x g'
721 The match_count check is so not to break
722 echo "hi" | busybox sed 's/^/!/g' */
723 if (!G.regmatch[0].rm_so && !G.regmatch[0].rm_eo && match_count) {
730 /* If we aren't interested in this match, output old line to
731 end of match and continue */
732 if (sed_cmd->which_match
733 && (sed_cmd->which_match != match_count)
735 for (i = 0; i < G.regmatch[0].rm_eo; i++)
740 /* print everything before the match */
741 for (i = 0; i < G.regmatch[0].rm_so; i++)
744 /* then print the substitution string */
745 do_subst_w_backrefs(line, sed_cmd->string);
747 /* advance past the match */
748 line += G.regmatch[0].rm_eo;
749 /* flag that something has changed */
752 /* if we're not doing this globally, get out now */
753 if (sed_cmd->which_match)
756 //maybe (G.regmatch[0].rm_eo ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
757 } while (*line && regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
759 /* Copy rest of string into output pipeline */
768 *line_p = G.pipeline.buf;
772 /* Set command pointer to point to this label. (Does not handle null label.) */
773 static sed_cmd_t *branch_to(char *label)
777 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
778 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
782 bb_error_msg_and_die("can't find label for jump to '%s'", label);
785 static void append(char *s)
787 llist_add_to_end(&G.append_head, xstrdup(s));
790 static void flush_append(void)
794 /* Output appended lines. */
795 while ((data = (char *)llist_pop(&G.append_head))) {
796 fprintf(G.nonstdout, "%s\n", data);
801 static void add_input_file(FILE *file)
803 G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count);
804 G.input_file_list[G.input_file_count++] = file;
807 /* Get next line of input from G.input_file_list, flushing append buffer and
808 * noting if we ran out of files without a newline on the last line we read.
814 static char *get_next_line(char *gets_char)
822 /* will be returned if last line in the file
823 * doesn't end with either '\n' or '\0' */
825 while (G.current_input_file < G.input_file_count) {
826 FILE *fp = G.input_file_list[G.current_input_file];
827 /* Read line up to a newline or NUL byte, inclusive,
828 * return malloc'ed char[]. length of the chunk read
829 * is stored in len. NULL if EOF/error */
830 temp = bb_get_chunk_from_file(fp, &len);
832 /* len > 0 here, it's ok to do temp[len-1] */
833 char c = temp[len-1];
834 if (c == '\n' || c == '\0') {
845 /* else we put NO_EOL_CHAR into *gets_char */
848 /* NB: I had the idea of peeking next file(s) and returning
849 * NO_EOL_CHAR only if it is the *last* non-empty
850 * input file. But there is a case where this won't work:
851 * file1: "a woo\nb woo"
852 * file2: "c no\nd no"
853 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
854 * (note: *no* newline after "b bang"!) */
856 /* Close this file and advance to next one */
858 G.current_input_file++;
864 /* Output line of text. */
866 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
867 * Without them, we had this:
871 * sed "s/i/z/" z1 z2 znull | hexdump -vC
874 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
876 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
878 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
880 char lpc = *last_puts_char;
882 /* Need to insert a '\n' between two files because first file's
883 * last line wasn't terminated? */
884 if (lpc != '\n' && lpc != '\0') {
890 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
894 /* had trailing '\0' and it was last char of file? */
895 if (last_gets_char == LAST_IS_NUL) {
899 /* had trailing '\n' or '\0'? */
900 if (last_gets_char != NO_EOL_CHAR) {
901 fputc(last_gets_char, file);
902 lpc = last_gets_char;
906 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
907 bb_error_msg_and_die(bb_msg_write_error);
909 *last_puts_char = lpc;
912 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
914 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
916 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
918 G.previous_regex_ptr = sed_cmd->beg_match;
922 /* Process all the lines in all the files */
924 static void process_files(void)
926 char *pattern_space, *next_line;
928 char last_puts_char = '\n';
929 char last_gets_char, next_gets_char;
934 next_line = get_next_line(&next_gets_char);
936 /* Go through every line in each file */
940 /* Advance to next line. Stop if out of lines. */
941 pattern_space = next_line;
944 last_gets_char = next_gets_char;
946 /* Read one line in advance so we can act on the last line,
948 next_line = get_next_line(&next_gets_char);
951 /* For every line, go through all the commands */
953 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
954 int old_matched, matched;
956 old_matched = sed_cmd->in_match;
958 /* Determine if this command matches this line: */
960 //bb_error_msg("match1:%d", sed_cmd->in_match);
961 //bb_error_msg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
962 // && !sed_cmd->beg_match && !sed_cmd->end_match));
963 //bb_error_msg("match3:%d", (sed_cmd->beg_line > 0
964 // && (sed_cmd->end_line || sed_cmd->end_match
965 // ? (sed_cmd->beg_line <= linenum)
966 // : (sed_cmd->beg_line == linenum)
969 //bb_error_msg("match4:%d", (beg_match(sed_cmd, pattern_space)));
970 //bb_error_msg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
972 /* Are we continuing a previous multi-line match? */
973 sed_cmd->in_match = sed_cmd->in_match
974 /* Or is no range necessary? */
975 || (!sed_cmd->beg_line && !sed_cmd->end_line
976 && !sed_cmd->beg_match && !sed_cmd->end_match)
977 /* Or did we match the start of a numerical range? */
978 || (sed_cmd->beg_line > 0
979 && (sed_cmd->end_line || sed_cmd->end_match
980 /* note: even if end is numeric and is < linenum too,
981 * GNU sed matches! We match too */
982 ? (sed_cmd->beg_line <= linenum) /* N,end */
983 : (sed_cmd->beg_line == linenum) /* N */
986 /* Or does this line match our begin address regex? */
987 || (beg_match(sed_cmd, pattern_space))
988 /* Or did we match last line of input? */
989 || (sed_cmd->beg_line == -1 && next_line == NULL);
991 /* Snapshot the value */
992 matched = sed_cmd->in_match;
994 //bb_error_msg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
995 //sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
997 /* Is this line the end of the current match? */
1000 /* once matched, "n,xxx" range is dead, disabling it */
1001 if (sed_cmd->beg_line > 0
1002 && !(option_mask32 & OPT_in_place) /* but not for -i */
1004 sed_cmd->beg_line = -2;
1006 sed_cmd->in_match = !(
1007 /* has the ending line come, or is this a single address command? */
1009 ? sed_cmd->end_line == -1
1011 : (sed_cmd->end_line <= linenum)
1012 : !sed_cmd->end_match
1014 /* or does this line matches our last address regex */
1015 || (sed_cmd->end_match && old_matched
1016 && (regexec(sed_cmd->end_match,
1017 pattern_space, 0, NULL, 0) == 0))
1021 /* Skip blocks of commands we didn't match */
1022 if (sed_cmd->cmd == '{') {
1023 if (sed_cmd->invert ? matched : !matched) {
1024 unsigned nest_cnt = 0;
1026 if (sed_cmd->cmd == '{')
1028 if (sed_cmd->cmd == '}') {
1033 sed_cmd = sed_cmd->next;
1035 bb_error_msg_and_die("unterminated {");
1041 /* Okay, so did this line match? */
1042 if (sed_cmd->invert ? matched : !matched)
1045 /* Update last used regex in case a blank substitute BRE is found */
1046 if (sed_cmd->beg_match) {
1047 G.previous_regex_ptr = sed_cmd->beg_match;
1050 /* actual sedding */
1051 //bb_error_msg("pattern_space:'%s' next_line:'%s' cmd:%c",
1052 //pattern_space, next_line, sed_cmd->cmd);
1053 switch (sed_cmd->cmd) {
1055 /* Print line number */
1057 fprintf(G.nonstdout, "%d\n", linenum);
1060 /* Write the current pattern space up to the first newline */
1063 char *tmp = strchr(pattern_space, '\n');
1066 /* TODO: explain why '\n' below */
1067 sed_puts(pattern_space, '\n');
1074 /* Write the current pattern space to output */
1076 /* NB: we print this _before_ the last line
1077 * (of current file) is printed. Even if
1078 * that line is nonterminated, we print
1079 * '\n' here (gnu sed does the same) */
1080 sed_puts(pattern_space, '\n');
1082 /* Delete up through first newline */
1085 char *tmp = strchr(pattern_space, '\n');
1087 overlapping_strcpy(pattern_space, tmp + 1);
1091 /* discard this line. */
1095 /* Substitute with regex */
1097 if (!do_subst_command(sed_cmd, &pattern_space))
1101 /* handle p option */
1103 sed_puts(pattern_space, last_gets_char);
1104 /* handle w option */
1105 if (sed_cmd->sw_file)
1107 pattern_space, sed_cmd->sw_file,
1108 &sed_cmd->sw_last_char, last_gets_char);
1111 /* Append line to linked list to be printed later */
1113 append(sed_cmd->string);
1116 /* Insert text before this line */
1118 sed_puts(sed_cmd->string, '\n');
1121 /* Cut and paste text (replace) */
1123 /* Only triggers on last line of a matching range. */
1124 if (!sed_cmd->in_match)
1125 sed_puts(sed_cmd->string, '\n');
1128 /* Read file, append contents to output */
1132 rfile = fopen_for_read(sed_cmd->string);
1136 while ((line = xmalloc_fgetline(rfile))
1139 xprint_and_close_file(rfile);
1145 /* Write pattern space to file. */
1148 pattern_space, sed_cmd->sw_file,
1149 &sed_cmd->sw_last_char, last_gets_char);
1152 /* Read next line from input */
1155 sed_puts(pattern_space, last_gets_char);
1157 free(pattern_space);
1158 pattern_space = next_line;
1159 last_gets_char = next_gets_char;
1160 next_line = get_next_line(&next_gets_char);
1167 /* Quit. End of script, end of input. */
1169 /* Exit the outer while loop */
1172 goto discard_commands;
1174 /* Append the next line to the current line */
1178 /* If no next line, jump to end of script and exit. */
1179 /* http://www.gnu.org/software/sed/manual/sed.html:
1180 * "Most versions of sed exit without printing anything
1181 * when the N command is issued on the last line of
1182 * a file. GNU sed prints pattern space before exiting
1183 * unless of course the -n command switch has been
1184 * specified. This choice is by design."
1186 if (next_line == NULL) {
1187 //goto discard_line;
1188 goto discard_commands; /* GNU behavior */
1190 /* Append next_line, read new next_line. */
1191 len = strlen(pattern_space);
1192 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1193 pattern_space[len] = '\n';
1194 strcpy(pattern_space + len+1, next_line);
1195 last_gets_char = next_gets_char;
1196 next_line = get_next_line(&next_gets_char);
1201 /* Test/branch if substitution occurred */
1203 if (!substituted) break;
1206 /* Test/branch if substitution didn't occur */
1208 if (substituted) break;
1210 /* Branch to label */
1212 if (!sed_cmd->string) goto discard_commands;
1213 else sed_cmd = branch_to(sed_cmd->string);
1215 /* Transliterate characters */
1219 for (i = 0; pattern_space[i]; i++) {
1220 for (j = 0; sed_cmd->string[j]; j += 2) {
1221 if (pattern_space[i] == sed_cmd->string[j]) {
1222 pattern_space[i] = sed_cmd->string[j + 1];
1230 case 'g': /* Replace pattern space with hold space */
1231 free(pattern_space);
1232 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1234 case 'G': /* Append newline and hold space to pattern space */
1236 int pattern_space_size = 2;
1237 int hold_space_size = 0;
1240 pattern_space_size += strlen(pattern_space);
1242 hold_space_size = strlen(G.hold_space);
1243 pattern_space = xrealloc(pattern_space,
1244 pattern_space_size + hold_space_size);
1245 if (pattern_space_size == 2)
1246 pattern_space[0] = 0;
1247 strcat(pattern_space, "\n");
1249 strcat(pattern_space, G.hold_space);
1250 last_gets_char = '\n';
1254 case 'h': /* Replace hold space with pattern space */
1256 G.hold_space = xstrdup(pattern_space);
1258 case 'H': /* Append newline and pattern space to hold space */
1260 int hold_space_size = 2;
1261 int pattern_space_size = 0;
1264 hold_space_size += strlen(G.hold_space);
1266 pattern_space_size = strlen(pattern_space);
1267 G.hold_space = xrealloc(G.hold_space,
1268 hold_space_size + pattern_space_size);
1270 if (hold_space_size == 2)
1272 strcat(G.hold_space, "\n");
1274 strcat(G.hold_space, pattern_space);
1278 case 'x': /* Exchange hold and pattern space */
1280 char *tmp = pattern_space;
1281 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1282 last_gets_char = '\n';
1287 } /* for each cmd */
1290 * Exit point from sedding...
1293 /* we will print the line unless we were told to be quiet ('-n')
1294 or if the line was suppressed (ala 'd'elete) */
1296 sed_puts(pattern_space, last_gets_char);
1298 /* Delete and such jump here. */
1301 free(pattern_space);
1306 /* It is possible to have a command line argument with embedded
1307 * newlines. This counts as multiple command lines.
1308 * However, newline can be escaped: 's/e/z\<newline>z/'
1309 * We check for this.
1312 static void add_cmd_block(char *cmdstr)
1316 cmdstr = sv = xstrdup(cmdstr);
1318 eol = strchr(cmdstr, '\n');
1321 /* Count preceding slashes */
1325 while (sl != cmdstr && *--sl == '\\')
1327 /* Odd number of preceding slashes - newline is escaped */
1329 overlapping_strcpy(eol - 1, eol);
1330 eol = strchr(eol, '\n');
1341 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1342 int sed_main(int argc UNUSED_PARAM, char **argv)
1345 llist_t *opt_e, *opt_f;
1346 int status = EXIT_SUCCESS;
1350 /* destroy command strings on exit */
1351 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1353 /* Lie to autoconf when it starts asking stupid questions. */
1354 if (argv[1] && !strcmp(argv[1], "--version")) {
1355 puts("This is not GNU sed version 4.0");
1359 /* do normal option parsing */
1360 opt_e = opt_f = NULL;
1361 opt_complementary = "e::f::" /* can occur multiple times */
1362 "nn"; /* count -n */
1363 /* -i must be first, to match OPT_in_place definition */
1364 opt = getopt32(argv, "irne:f:", &opt_e, &opt_f,
1365 &G.be_quiet); /* counter for -n */
1368 if (opt & OPT_in_place) { // -i
1369 atexit(cleanup_outname);
1371 if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r
1372 //if (opt & 0x4) G.be_quiet++; // -n
1373 while (opt_e) { // -e
1374 add_cmd_block(llist_pop(&opt_e));
1376 while (opt_f) { // -f
1379 cmdfile = xfopen_for_read(llist_pop(&opt_f));
1380 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1386 /* if we didn't get a pattern from -e or -f, use argv[0] */
1387 if (!(opt & 0x18)) {
1390 add_cmd_block(*argv++);
1392 /* Flush any unfinished commands. */
1395 /* By default, we write to stdout */
1396 G.nonstdout = stdout;
1398 /* argv[0..(argc-1)] should be names of file to process. If no
1399 * files were specified or '-' was specified, take input from stdin.
1400 * Otherwise, we process all the files specified. */
1401 if (argv[0] == NULL) {
1402 if (opt & OPT_in_place)
1403 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1404 add_input_file(stdin);
1409 for (i = 0; argv[i]; i++) {
1410 struct stat statbuf;
1413 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1414 add_input_file(stdin);
1418 file = fopen_or_warn(argv[i], "r");
1420 status = EXIT_FAILURE;
1423 if (!(opt & OPT_in_place)) {
1424 add_input_file(file);
1428 G.outname = xasprintf("%sXXXXXX", argv[i]);
1429 nonstdoutfd = xmkstemp(G.outname);
1430 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1432 /* Set permissions/owner of output file */
1433 fstat(fileno(file), &statbuf);
1434 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1435 * but GNU sed 4.2.1 does not preserve them either */
1436 fchmod(nonstdoutfd, statbuf.st_mode);
1437 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1438 add_input_file(file);
1440 fclose(G.nonstdout);
1442 G.nonstdout = stdout;
1443 /* unlink(argv[i]); */
1444 xrename(G.outname, argv[i]);
1448 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1449 * if (G.current_input_file >= G.input_file_count)
1451 * but it's not needed since process_files() works correctly
1452 * in this case too. */