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/|$]command
40 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
41 * - edit commands: (a)ppend, (i)nsert, (c)hange
42 * - file commands: (r)ead
43 * - backreferences in substitution expressions (\0, \1, \2...\9)
44 * - grouped commands: {cmd1;cmd2}
45 * - transliteration (y/source-chars/dest-chars/)
46 * - pattern space hold space storing / swapping (g, h, x)
47 * - labels / branching (: label, b, t, T)
49 * (Note: Specifying an address (range) to match is *optional*; commands
50 * default to the whole pattern space if no specific address match was
54 * - Create a wrapper around regex to make libc's regex conform with sed
57 * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
58 * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html
61 //usage:#define sed_trivial_usage
62 //usage: "[-inr] [-f FILE]... [-e CMD]... [FILE]...\n"
63 //usage: "or: sed [-inr] CMD [FILE]..."
64 //usage:#define sed_full_usage "\n\n"
65 //usage: " -e CMD Add CMD to sed commands to be executed"
66 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
67 //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)"
68 //usage: "\n Optionally back files up, appending SFX"
69 //usage: "\n -n Suppress automatic printing of pattern space"
70 //usage: "\n -r Use extended regex syntax"
72 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
73 //usage: "\nRemaining arguments are input files (stdin if none)."
75 //usage:#define sed_example_usage
76 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
83 # define dbg(...) bb_error_msg(__VA_ARGS__)
85 # define dbg(...) ((void)0)
90 OPT_in_place = 1 << 0,
93 /* Each sed command turns into one of these structures. */
94 typedef struct sed_cmd_s {
95 /* Ordered by alignment requirements: currently 36 bytes on x86 */
96 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
99 regex_t *beg_match; /* sed -e '/match/cmd' */
100 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
101 regex_t *sub_match; /* For 's/sub_match/string/' */
102 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
103 int beg_line_orig; /* copy of the above, needed for -i */
104 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
106 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
107 char *string; /* Data string for (saicytb) commands. */
109 unsigned which_match; /* (s) Which match to replace (0 for all) */
111 /* Bitfields (gcc won't group them if we don't) */
112 unsigned invert:1; /* the '!' after the address */
113 unsigned in_match:1; /* Next line also included in match? */
114 unsigned sub_p:1; /* (s) print option */
116 char sw_last_char; /* Last line written by (sw) had no '\n' */
119 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
122 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
126 int be_quiet, regex_type;
128 char *outname, *hold_space;
130 /* List of input files */
131 int input_file_count, current_input_file;
132 FILE **input_file_list;
134 regmatch_t regmatch[10];
135 regex_t *previous_regex_ptr;
137 /* linked list of sed commands */
138 sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
140 /* Linked list of append lines */
141 llist_t *append_head;
146 char *buf; /* Space to hold string */
147 int idx; /* Space used */
148 int len; /* Space allocated */
151 #define G (*(struct globals*)&bb_common_bufsiz1)
152 struct BUG_G_too_big {
153 char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
155 #define INIT_G() do { \
156 G.sed_cmd_tail = &G.sed_cmd_head; \
160 #if ENABLE_FEATURE_CLEAN_UP
161 static void sed_free_and_close_stuff(void)
163 sed_cmd_t *sed_cmd = G.sed_cmd_head;
165 llist_free(G.append_head, free);
168 sed_cmd_t *sed_cmd_next = sed_cmd->next;
170 if (sed_cmd->sw_file)
171 xprint_and_close_file(sed_cmd->sw_file);
173 if (sed_cmd->beg_match) {
174 regfree(sed_cmd->beg_match);
175 free(sed_cmd->beg_match);
177 if (sed_cmd->end_match) {
178 regfree(sed_cmd->end_match);
179 free(sed_cmd->end_match);
181 if (sed_cmd->sub_match) {
182 regfree(sed_cmd->sub_match);
183 free(sed_cmd->sub_match);
185 free(sed_cmd->string);
187 sed_cmd = sed_cmd_next;
192 while (G.current_input_file < G.input_file_count)
193 fclose(G.input_file_list[G.current_input_file++]);
196 void sed_free_and_close_stuff(void);
199 /* If something bad happens during -i operation, delete temp file */
201 static void cleanup_outname(void)
203 if (G.outname) unlink(G.outname);
206 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
208 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
213 if (string[i] == '\\') {
214 if (!to || string[i+1] == from) {
215 *dest++ = to ? to : string[i+1];
219 *dest++ = string[i++];
221 /* TODO: is it safe wrt a string with trailing '\\' ? */
222 *dest++ = string[i++];
227 static char *copy_parsing_escapes(const char *string, int len)
230 char *dest = xmalloc(len + 1);
232 /* sed recognizes \n */
233 /* GNU sed also recognizes \t and \r */
234 for (s = "\nn\tt\rr"; *s; s += 2) {
235 parse_escapes(dest, string, len, s[1], s[0]);
244 * index_of_next_unescaped_regexp_delim - walks left to right through a string
245 * beginning at a specified index and returns the index of the next regular
246 * expression delimiter (typically a forward slash ('/')) not preceded by
247 * a backslash ('\'). A negative delimiter disables square bracket checking.
249 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
258 delimiter = -delimiter;
261 for (; (ch = str[idx]) != '\0'; idx++) {
264 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
272 else if (bracket == -1 && ch == '[')
274 else if (ch == delimiter)
278 /* if we make it to here, we've hit the end of the string */
279 bb_error_msg_and_die("unmatched '%c'", delimiter);
283 * Returns the index of the third delimiter
285 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
287 const char *cmdstr_ptr = cmdstr;
288 unsigned char delimiter;
291 /* verify that the 's' or 'y' is followed by something. That something
292 * (typically a 'slash') is now our regexp delimiter... */
294 bb_error_msg_and_die("bad format in substitution expression");
295 delimiter = *cmdstr_ptr++;
297 /* save the match string */
298 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
299 *match = copy_parsing_escapes(cmdstr_ptr, idx);
301 /* save the replacement string */
302 cmdstr_ptr += idx + 1;
303 idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr);
304 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
306 return ((cmdstr_ptr - cmdstr) + idx);
310 * returns the index in the string just past where the address ends.
312 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
314 const char *pos = my_str;
316 if (isdigit(*my_str)) {
317 *linenum = strtol(my_str, (char**)&pos, 10);
318 /* endstr shouldnt ever equal NULL */
319 } else if (*my_str == '$') {
322 } else if (*my_str == '/' || *my_str == '\\') {
330 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
331 temp = copy_parsing_escapes(pos, next);
332 *regex = xzalloc(sizeof(regex_t));
333 xregcomp(*regex, temp, G.regex_type);
335 /* Move position to next character after last delimiter */
341 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
342 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
344 int start = 0, idx, hack = 0;
346 /* Skip whitespace, then grab filename to end of line */
347 while (isspace(filecmdstr[start]))
350 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
353 /* If lines glued together, put backslash back. */
354 if (filecmdstr[idx] == '\n')
357 bb_error_msg_and_die("empty filename");
358 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
360 (*retval)[idx] = '\\';
365 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
367 int cflags = G.regex_type;
372 * A substitution command should look something like this:
373 * s/match/replace/ #gIpw
377 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
379 /* determine the number of back references in the match string */
380 /* Note: we compute this here rather than in the do_subst_command()
381 * function to save processor time, at the expense of a little more memory
382 * (4 bits) per sed_cmd */
384 /* process the flags */
386 sed_cmd->which_match = 1;
387 while (substr[++idx]) {
388 /* Parse match number */
389 if (isdigit(substr[idx])) {
390 if (match[0] != '^') {
391 /* Match 0 treated as all, multiple matches we take the last one. */
392 const char *pos = substr + idx;
393 /* FIXME: error check? */
394 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
400 if (isspace(substr[idx]))
403 switch (substr[idx]) {
404 /* Replace all occurrences */
407 sed_cmd->which_match = 0;
409 /* Print pattern space */
417 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
420 /* Ignore case (gnu exension) */
426 // while (substr[++idx]) continue;
427 idx += strlen(substr + idx); // same
434 bb_error_msg_and_die("bad option in substitution expression");
438 /* compile the match string into a regex */
439 if (*match != '\0') {
440 /* If match is empty, we use last regex used at runtime */
441 sed_cmd->sub_match = xzalloc(sizeof(regex_t));
442 dbg("xregcomp('%s',%x)", match, cflags);
443 xregcomp(sed_cmd->sub_match, match, cflags);
452 * Process the commands arguments
454 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
456 static const char cmd_letters[] = "saicrw:btTydDgGhHlnNpPqx={}";
487 struct chk { char chk[sizeof(cmd_letters)-1 == IDX_nul ? 1 : -1]; };
489 unsigned idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
491 /* handle (s)ubstitution command */
493 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
495 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
496 else if (idx <= IDX_c) { /* a,i,c */
497 if (idx < IDX_c) { /* a,i */
498 if (sed_cmd->end_line || sed_cmd->end_match)
499 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
502 if (*cmdstr == '\n' || *cmdstr == '\\') {
506 if (!isspace(*cmdstr))
510 sed_cmd->string = xstrdup(cmdstr);
511 /* "\anychar" -> "anychar" */
512 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
513 cmdstr += strlen(cmdstr);
515 /* handle file cmds: (r)ead */
516 else if (idx <= IDX_w) { /* r,w */
517 if (idx < IDX_w) { /* r */
518 if (sed_cmd->end_line || sed_cmd->end_match)
519 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
521 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
522 if (sed_cmd->cmd == 'w') {
523 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
524 sed_cmd->sw_last_char = '\n';
527 /* handle branch commands */
528 else if (idx <= IDX_T) { /* :,b,t,T */
531 cmdstr = skip_whitespace(cmdstr);
532 length = strcspn(cmdstr, semicolon_whitespace);
534 sed_cmd->string = xstrndup(cmdstr, length);
538 /* translation command */
539 else if (idx == IDX_y) {
540 char *match, *replace;
543 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
544 /* \n already parsed, but \delimiter needs unescaping. */
545 parse_escapes(match, match, strlen(match), i, i);
546 parse_escapes(replace, replace, strlen(replace), i, i);
548 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
549 for (i = 0; match[i] && replace[i]; i++) {
550 sed_cmd->string[i*2] = match[i];
551 sed_cmd->string[i*2+1] = replace[i];
556 /* if it wasnt a single-letter command that takes no arguments
557 * then it must be an invalid command.
559 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
560 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
563 /* give back whatever's left over */
568 /* Parse address+command sets, skipping comment lines. */
570 static void add_cmd(const char *cmdstr)
575 /* Append this line to any unfinished line from last time. */
576 if (G.add_cmd_line) {
577 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
578 free(G.add_cmd_line);
579 cmdstr = G.add_cmd_line = tp;
582 /* If this line ends with unescaped backslash, request next line. */
583 n = len = strlen(cmdstr);
584 while (n && cmdstr[n-1] == '\\')
586 if ((len - n) & 1) { /* if odd number of trailing backslashes */
588 G.add_cmd_line = xstrdup(cmdstr);
589 G.add_cmd_line[len-1] = '\0';
593 /* Loop parsing all commands in this line. */
595 /* Skip leading whitespace and semicolons */
596 cmdstr += strspn(cmdstr, semicolon_whitespace);
598 /* If no more commands, exit. */
601 /* if this is a comment, jump past it and keep going */
602 if (*cmdstr == '#') {
603 /* "#n" is the same as using -n on the command line */
604 if (cmdstr[1] == 'n')
606 cmdstr = strpbrk(cmdstr, "\n\r");
612 * format is: [addr][,addr][!]cmd
617 sed_cmd = xzalloc(sizeof(sed_cmd_t));
619 /* first part (if present) is an address: either a '$', a number or a /regex/ */
620 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
621 sed_cmd->beg_line_orig = sed_cmd->beg_line;
623 /* second part (if present) will begin with a comma */
624 if (*cmdstr == ',') {
628 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
630 bb_error_msg_and_die("no address after comma");
634 /* skip whitespace before the command */
635 cmdstr = skip_whitespace(cmdstr);
637 /* Check for inversion flag */
638 if (*cmdstr == '!') {
642 /* skip whitespace before the command */
643 cmdstr = skip_whitespace(cmdstr);
646 /* last part (mandatory) will be a command */
648 bb_error_msg_and_die("missing command");
649 sed_cmd->cmd = *cmdstr++;
650 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
652 /* cmdstr now points past args.
653 * GNU sed requires a separator, if there are more commands,
654 * else it complains "char N: extra characters after command".
655 * Example: "sed 'p;d'". We also allow "sed 'pd'".
658 /* Add the command to the command array */
659 *G.sed_cmd_tail = sed_cmd;
660 G.sed_cmd_tail = &sed_cmd->next;
663 /* If we glued multiple lines together, free the memory. */
664 free(G.add_cmd_line);
665 G.add_cmd_line = NULL;
668 /* Append to a string, reallocating memory as necessary. */
672 static void pipe_putc(char c)
674 if (G.pipeline.idx == G.pipeline.len) {
675 G.pipeline.buf = xrealloc(G.pipeline.buf,
676 G.pipeline.len + PIPE_GROW);
677 G.pipeline.len += PIPE_GROW;
679 G.pipeline.buf[G.pipeline.idx++] = c;
682 static void do_subst_w_backrefs(char *line, char *replace)
686 /* go through the replacement string */
687 for (i = 0; replace[i]; i++) {
688 /* if we find a backreference (\1, \2, etc.) print the backref'ed text */
689 if (replace[i] == '\\') {
690 unsigned backref = replace[++i] - '0';
692 /* print out the text held in G.regmatch[backref] */
693 if (G.regmatch[backref].rm_so != -1) {
694 j = G.regmatch[backref].rm_so;
695 while (j < G.regmatch[backref].rm_eo)
696 pipe_putc(line[j++]);
700 /* I _think_ it is impossible to get '\' to be
701 * the last char in replace string. Thus we dont check
702 * for replace[i] == NUL. (counterexample anyone?) */
703 /* if we find a backslash escaped character, print the character */
704 pipe_putc(replace[i]);
707 /* if we find an unescaped '&' print out the whole matched text. */
708 if (replace[i] == '&') {
709 j = G.regmatch[0].rm_so;
710 while (j < G.regmatch[0].rm_eo)
711 pipe_putc(line[j++]);
714 /* Otherwise just output the character. */
715 pipe_putc(replace[i]);
719 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
721 char *line = *line_p;
722 unsigned match_count = 0;
724 bool prev_match_empty = 1;
725 bool tried_at_eol = 0;
726 regex_t *current_regex;
728 current_regex = sed_cmd->sub_match;
729 /* Handle empty regex. */
730 if (!current_regex) {
731 current_regex = G.previous_regex_ptr;
733 bb_error_msg_and_die("no previous regexp");
735 G.previous_regex_ptr = current_regex;
737 /* Find the first match */
738 dbg("matching '%s'", line);
739 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) {
745 /* Initialize temporary output buffer. */
746 G.pipeline.buf = xmalloc(PIPE_GROW);
747 G.pipeline.len = PIPE_GROW;
750 /* Now loop through, substituting for matches */
752 int start = G.regmatch[0].rm_so;
753 int end = G.regmatch[0].rm_eo;
758 /* If we aren't interested in this match, output old line to
759 * end of match and continue */
760 if (sed_cmd->which_match
761 && (sed_cmd->which_match != match_count)
763 for (i = 0; i < end; i++)
765 /* Null match? Print one more char */
766 if (start == end && *line)
771 /* Print everything before the match */
772 for (i = 0; i < start; i++)
775 /* Then print the substitution string,
776 * unless we just matched empty string after non-empty one.
777 * Example: string "cccd", pattern "c*", repl "R":
778 * result is "RdR", not "RRdR": first match "ccc",
779 * second is "" before "d", third is "" after "d".
780 * Second match is NOT replaced!
782 if (prev_match_empty || start != 0 || start != end) {
783 //dbg("%d %d %d", prev_match_empty, start, end);
784 dbg("inserting replacement at %d in '%s'", start, line);
785 do_subst_w_backrefs(line, sed_cmd->string);
786 /* Flag that something has changed */
789 dbg("NOT inserting replacement at %d in '%s'", start, line);
792 /* If matched string is empty (f.e. "c*" pattern),
793 * copy verbatim one char after it before attempting more matches
795 prev_match_empty = (start == end);
796 if (prev_match_empty) {
800 pipe_putc(line[end]);
805 /* Advance past the match */
806 dbg("line += %d", end);
809 /* if we're not doing this globally, get out now */
810 if (sed_cmd->which_match != 0)
813 /* Exit if we are at EOL and already tried matching at it */
820 //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
821 } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
823 /* Copy rest of string into output pipeline */
832 *line_p = G.pipeline.buf;
836 /* Set command pointer to point to this label. (Does not handle null label.) */
837 static sed_cmd_t *branch_to(char *label)
841 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
842 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
846 bb_error_msg_and_die("can't find label for jump to '%s'", label);
849 static void append(char *s)
851 llist_add_to_end(&G.append_head, s);
854 static void flush_append(void)
858 /* Output appended lines. */
859 while ((data = (char *)llist_pop(&G.append_head))) {
860 fprintf(G.nonstdout, "%s\n", data);
865 static void add_input_file(FILE *file)
867 G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count);
868 G.input_file_list[G.input_file_count++] = file;
871 /* Get next line of input from G.input_file_list, flushing append buffer and
872 * noting if we ran out of files without a newline on the last line we read.
878 static char *get_next_line(char *gets_char)
886 /* will be returned if last line in the file
887 * doesn't end with either '\n' or '\0' */
889 while (G.current_input_file < G.input_file_count) {
890 FILE *fp = G.input_file_list[G.current_input_file];
891 /* Read line up to a newline or NUL byte, inclusive,
892 * return malloc'ed char[]. length of the chunk read
893 * is stored in len. NULL if EOF/error */
894 temp = bb_get_chunk_from_file(fp, &len);
896 /* len > 0 here, it's ok to do temp[len-1] */
897 char c = temp[len-1];
898 if (c == '\n' || c == '\0') {
909 /* else we put NO_EOL_CHAR into *gets_char */
912 /* NB: I had the idea of peeking next file(s) and returning
913 * NO_EOL_CHAR only if it is the *last* non-empty
914 * input file. But there is a case where this won't work:
915 * file1: "a woo\nb woo"
916 * file2: "c no\nd no"
917 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
918 * (note: *no* newline after "b bang"!) */
920 /* Close this file and advance to next one */
922 G.current_input_file++;
928 /* Output line of text. */
930 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
931 * Without them, we had this:
935 * sed "s/i/z/" z1 z2 znull | hexdump -vC
938 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
940 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
942 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
944 char lpc = *last_puts_char;
946 /* Need to insert a '\n' between two files because first file's
947 * last line wasn't terminated? */
948 if (lpc != '\n' && lpc != '\0') {
954 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
958 /* had trailing '\0' and it was last char of file? */
959 if (last_gets_char == LAST_IS_NUL) {
963 /* had trailing '\n' or '\0'? */
964 if (last_gets_char != NO_EOL_CHAR) {
965 fputc(last_gets_char, file);
966 lpc = last_gets_char;
970 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
971 bb_error_msg_and_die(bb_msg_write_error);
973 *last_puts_char = lpc;
976 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
978 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
980 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
982 G.previous_regex_ptr = sed_cmd->beg_match;
986 /* Process all the lines in all the files */
988 static void process_files(void)
990 char *pattern_space, *next_line;
992 char last_puts_char = '\n';
993 char last_gets_char, next_gets_char;
998 next_line = get_next_line(&next_gets_char);
1000 /* Go through every line in each file */
1004 /* Advance to next line. Stop if out of lines. */
1005 pattern_space = next_line;
1008 last_gets_char = next_gets_char;
1010 /* Read one line in advance so we can act on the last line,
1011 * the '$' address */
1012 next_line = get_next_line(&next_gets_char);
1015 /* For every line, go through all the commands */
1017 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1018 int old_matched, matched;
1020 old_matched = sed_cmd->in_match;
1022 /* Determine if this command matches this line: */
1024 dbg("match1:%d", sed_cmd->in_match);
1025 dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
1026 && !sed_cmd->beg_match && !sed_cmd->end_match));
1027 dbg("match3:%d", (sed_cmd->beg_line > 0
1028 && (sed_cmd->end_line || sed_cmd->end_match
1029 ? (sed_cmd->beg_line <= linenum)
1030 : (sed_cmd->beg_line == linenum)
1033 dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
1034 dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
1036 /* Are we continuing a previous multi-line match? */
1037 sed_cmd->in_match = sed_cmd->in_match
1038 /* Or is no range necessary? */
1039 || (!sed_cmd->beg_line && !sed_cmd->end_line
1040 && !sed_cmd->beg_match && !sed_cmd->end_match)
1041 /* Or did we match the start of a numerical range? */
1042 || (sed_cmd->beg_line > 0
1043 && (sed_cmd->end_line || sed_cmd->end_match
1044 /* note: even if end is numeric and is < linenum too,
1045 * GNU sed matches! We match too, therefore we don't
1046 * check here that linenum <= end.
1048 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1049 * first three input lines are deleted;
1050 * 4th line is matched and printed
1051 * by "2,3" (!) and by "4" ranges
1053 ? (sed_cmd->beg_line <= linenum) /* N,end */
1054 : (sed_cmd->beg_line == linenum) /* N */
1057 /* Or does this line match our begin address regex? */
1058 || (beg_match(sed_cmd, pattern_space))
1059 /* Or did we match last line of input? */
1060 || (sed_cmd->beg_line == -1 && next_line == NULL);
1062 /* Snapshot the value */
1063 matched = sed_cmd->in_match;
1065 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1066 sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1068 /* Is this line the end of the current match? */
1071 /* once matched, "n,xxx" range is dead, disabling it */
1072 if (sed_cmd->beg_line > 0) {
1073 sed_cmd->beg_line = -2;
1075 sed_cmd->in_match = !(
1076 /* has the ending line come, or is this a single address command? */
1078 ? sed_cmd->end_line == -1
1080 : (sed_cmd->end_line <= linenum)
1081 : !sed_cmd->end_match
1083 /* or does this line matches our last address regex */
1084 || (sed_cmd->end_match && old_matched
1085 && (regexec(sed_cmd->end_match,
1086 pattern_space, 0, NULL, 0) == 0)
1091 /* Skip blocks of commands we didn't match */
1092 if (sed_cmd->cmd == '{') {
1093 if (sed_cmd->invert ? matched : !matched) {
1094 unsigned nest_cnt = 0;
1096 if (sed_cmd->cmd == '{')
1098 if (sed_cmd->cmd == '}') {
1103 sed_cmd = sed_cmd->next;
1105 bb_error_msg_and_die("unterminated {");
1111 /* Okay, so did this line match? */
1112 if (sed_cmd->invert ? matched : !matched)
1115 /* Update last used regex in case a blank substitute BRE is found */
1116 if (sed_cmd->beg_match) {
1117 G.previous_regex_ptr = sed_cmd->beg_match;
1120 /* actual sedding */
1121 dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1122 pattern_space, next_line, sed_cmd->cmd);
1123 switch (sed_cmd->cmd) {
1125 /* Print line number */
1127 fprintf(G.nonstdout, "%d\n", linenum);
1130 /* Write the current pattern space up to the first newline */
1133 char *tmp = strchr(pattern_space, '\n');
1136 /* TODO: explain why '\n' below */
1137 sed_puts(pattern_space, '\n');
1144 /* Write the current pattern space to output */
1146 /* NB: we print this _before_ the last line
1147 * (of current file) is printed. Even if
1148 * that line is nonterminated, we print
1149 * '\n' here (gnu sed does the same) */
1150 sed_puts(pattern_space, '\n');
1152 /* Delete up through first newline */
1155 char *tmp = strchr(pattern_space, '\n');
1157 overlapping_strcpy(pattern_space, tmp + 1);
1161 /* discard this line. */
1165 /* Substitute with regex */
1167 if (!do_subst_command(sed_cmd, &pattern_space))
1169 dbg("do_subst_command succeeded:'%s'", pattern_space);
1172 /* handle p option */
1174 sed_puts(pattern_space, last_gets_char);
1175 /* handle w option */
1176 if (sed_cmd->sw_file)
1178 pattern_space, sed_cmd->sw_file,
1179 &sed_cmd->sw_last_char, last_gets_char);
1182 /* Append line to linked list to be printed later */
1184 append(xstrdup(sed_cmd->string));
1187 /* Insert text before this line */
1189 sed_puts(sed_cmd->string, '\n');
1192 /* Cut and paste text (replace) */
1194 /* Only triggers on last line of a matching range. */
1195 if (!sed_cmd->in_match)
1196 sed_puts(sed_cmd->string, '\n');
1199 /* Read file, append contents to output */
1203 rfile = fopen_for_read(sed_cmd->string);
1206 while ((line = xmalloc_fgetline(rfile))
1215 /* Write pattern space to file. */
1218 pattern_space, sed_cmd->sw_file,
1219 &sed_cmd->sw_last_char, last_gets_char);
1222 /* Read next line from input */
1225 sed_puts(pattern_space, last_gets_char);
1227 free(pattern_space);
1228 pattern_space = next_line;
1229 last_gets_char = next_gets_char;
1230 next_line = get_next_line(&next_gets_char);
1237 /* Quit. End of script, end of input. */
1239 /* Exit the outer while loop */
1242 goto discard_commands;
1244 /* Append the next line to the current line */
1248 /* If no next line, jump to end of script and exit. */
1249 /* http://www.gnu.org/software/sed/manual/sed.html:
1250 * "Most versions of sed exit without printing anything
1251 * when the N command is issued on the last line of
1252 * a file. GNU sed prints pattern space before exiting
1253 * unless of course the -n command switch has been
1254 * specified. This choice is by design."
1256 if (next_line == NULL) {
1257 //goto discard_line;
1258 goto discard_commands; /* GNU behavior */
1260 /* Append next_line, read new next_line. */
1261 len = strlen(pattern_space);
1262 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1263 pattern_space[len] = '\n';
1264 strcpy(pattern_space + len+1, next_line);
1265 last_gets_char = next_gets_char;
1266 next_line = get_next_line(&next_gets_char);
1271 /* Test/branch if substitution occurred */
1273 if (!substituted) break;
1276 /* Test/branch if substitution didn't occur */
1278 if (substituted) break;
1280 /* Branch to label */
1282 if (!sed_cmd->string) goto discard_commands;
1283 else sed_cmd = branch_to(sed_cmd->string);
1285 /* Transliterate characters */
1289 for (i = 0; pattern_space[i]; i++) {
1290 for (j = 0; sed_cmd->string[j]; j += 2) {
1291 if (pattern_space[i] == sed_cmd->string[j]) {
1292 pattern_space[i] = sed_cmd->string[j + 1];
1300 case 'g': /* Replace pattern space with hold space */
1301 free(pattern_space);
1302 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1304 case 'G': /* Append newline and hold space to pattern space */
1306 int pattern_space_size = 2;
1307 int hold_space_size = 0;
1310 pattern_space_size += strlen(pattern_space);
1312 hold_space_size = strlen(G.hold_space);
1313 pattern_space = xrealloc(pattern_space,
1314 pattern_space_size + hold_space_size);
1315 if (pattern_space_size == 2)
1316 pattern_space[0] = 0;
1317 strcat(pattern_space, "\n");
1319 strcat(pattern_space, G.hold_space);
1320 last_gets_char = '\n';
1324 case 'h': /* Replace hold space with pattern space */
1326 G.hold_space = xstrdup(pattern_space);
1328 case 'H': /* Append newline and pattern space to hold space */
1330 int hold_space_size = 2;
1331 int pattern_space_size = 0;
1334 hold_space_size += strlen(G.hold_space);
1336 pattern_space_size = strlen(pattern_space);
1337 G.hold_space = xrealloc(G.hold_space,
1338 hold_space_size + pattern_space_size);
1340 if (hold_space_size == 2)
1342 strcat(G.hold_space, "\n");
1344 strcat(G.hold_space, pattern_space);
1348 case 'x': /* Exchange hold and pattern space */
1350 char *tmp = pattern_space;
1351 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1352 last_gets_char = '\n';
1357 } /* for each cmd */
1360 * Exit point from sedding...
1363 /* we will print the line unless we were told to be quiet ('-n')
1364 or if the line was suppressed (ala 'd'elete) */
1366 sed_puts(pattern_space, last_gets_char);
1368 /* Delete and such jump here. */
1371 free(pattern_space);
1376 /* It is possible to have a command line argument with embedded
1377 * newlines. This counts as multiple command lines.
1378 * However, newline can be escaped: 's/e/z\<newline>z/'
1379 * add_cmd() handles this.
1382 static void add_cmd_block(char *cmdstr)
1386 cmdstr = sv = xstrdup(cmdstr);
1388 eol = strchr(cmdstr, '\n');
1397 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1398 int sed_main(int argc UNUSED_PARAM, char **argv)
1401 llist_t *opt_e, *opt_f;
1404 #if ENABLE_LONG_OPTS
1405 static const char sed_longopts[] ALIGN1 =
1406 /* name has_arg short */
1407 "in-place\0" Optional_argument "i"
1408 "regexp-extended\0" No_argument "r"
1409 "quiet\0" No_argument "n"
1410 "silent\0" No_argument "n"
1411 "expression\0" Required_argument "e"
1412 "file\0" Required_argument "f";
1415 int status = EXIT_SUCCESS;
1419 /* destroy command strings on exit */
1420 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1422 /* Lie to autoconf when it starts asking stupid questions. */
1423 if (argv[1] && strcmp(argv[1], "--version") == 0) {
1424 puts("This is not GNU sed version 4.0");
1428 /* do normal option parsing */
1429 opt_e = opt_f = NULL;
1431 opt_complementary = "e::f::" /* can occur multiple times */
1432 "nn"; /* count -n */
1434 IF_LONG_OPTS(applet_long_options = sed_longopts);
1436 /* -i must be first, to match OPT_in_place definition */
1437 /* -E is a synonym of -r:
1438 * GNU sed 4.2.1 mentions it in neither --help
1439 * nor manpage, but does recognize it.
1441 opt = getopt32(argv, "i::rEne:f:", &opt_i, &opt_e, &opt_f,
1442 &G.be_quiet); /* counter for -n */
1445 if (opt & OPT_in_place) { // -i
1446 atexit(cleanup_outname);
1449 G.regex_type |= REG_EXTENDED; // -r or -E
1451 // G.be_quiet++; // -n (implemented with a counter instead)
1452 while (opt_e) { // -e
1453 add_cmd_block(llist_pop(&opt_e));
1455 while (opt_f) { // -f
1458 cmdfile = xfopen_for_read(llist_pop(&opt_f));
1459 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1465 /* if we didn't get a pattern from -e or -f, use argv[0] */
1466 if (!(opt & 0x30)) {
1469 add_cmd_block(*argv++);
1471 /* Flush any unfinished commands. */
1474 /* By default, we write to stdout */
1475 G.nonstdout = stdout;
1477 /* argv[0..(argc-1)] should be names of file to process. If no
1478 * files were specified or '-' was specified, take input from stdin.
1479 * Otherwise, we process all the files specified. */
1480 if (argv[0] == NULL) {
1481 if (opt & OPT_in_place)
1482 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1483 add_input_file(stdin);
1487 for (i = 0; argv[i]; i++) {
1488 struct stat statbuf;
1493 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1494 add_input_file(stdin);
1498 file = fopen_or_warn(argv[i], "r");
1500 status = EXIT_FAILURE;
1503 add_input_file(file);
1504 if (!(opt & OPT_in_place)) {
1508 /* -i: process each FILE separately: */
1510 G.outname = xasprintf("%sXXXXXX", argv[i]);
1511 nonstdoutfd = xmkstemp(G.outname);
1512 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1514 /* Set permissions/owner of output file */
1515 fstat(fileno(file), &statbuf);
1516 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1517 * but GNU sed 4.2.1 does not preserve them either */
1518 fchmod(nonstdoutfd, statbuf.st_mode);
1519 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1522 fclose(G.nonstdout);
1523 G.nonstdout = stdout;
1526 char *backupname = xasprintf("%s%s", argv[i], opt_i);
1527 xrename(argv[i], backupname);
1530 /* else unlink(argv[i]); - rename below does this */
1531 xrename(G.outname, argv[i]); //TODO: rollback backup on error?
1535 /* Re-enable disabled range matches */
1536 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1537 sed_cmd->beg_line = sed_cmd->beg_line_orig;
1540 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1541 * if (G.current_input_file >= G.input_file_count)
1543 * but it's not needed since process_files() works correctly
1544 * in this case too. */