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.
17 * Files are laid out to avoid unnecessary function declarations. So for
18 * example, every function add_cmd calls occurs before add_cmd in this file.
20 * add_cmd() is called on each line of sed command text (from a file or from
21 * the command line). It calls get_address() and parse_cmd_args(). The
22 * resulting sed_cmd_t structures are appended to a linked list
23 * (G.sed_cmd_head/G.sed_cmd_tail).
25 * process_files() does actual sedding, reading data lines from each input FILE*
26 * (which could be stdin) and applying the sed command list (sed_cmd_head) to
27 * each of the resulting lines.
29 * sed_main() is where external code calls into this, with a command line.
31 /* Supported features and commands in this version of sed:
34 * - address matching: num|/matchstr/[,num|/matchstr/|$]command
35 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
36 * - edit commands: (a)ppend, (i)nsert, (c)hange
37 * - file commands: (r)ead
38 * - backreferences in substitution expressions (\0, \1, \2...\9)
39 * - grouped commands: {cmd1;cmd2}
40 * - transliteration (y/source-chars/dest-chars/)
41 * - pattern space hold space storing / swapping (g, h, x)
42 * - labels / branching (: label, b, t, T)
44 * (Note: Specifying an address (range) to match is *optional*; commands
45 * default to the whole pattern space if no specific address match was
49 * - Create a wrapper around regex to make libc's regex conform with sed
52 * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
53 * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html
54 * http://sed.sourceforge.net/sedfaq3.html
57 //config: bool "sed (12 kb)"
60 //config: sed is used to perform text transformations on a file
61 //config: or input from a pipeline.
63 //applet:IF_SED(APPLET(sed, BB_DIR_BIN, BB_SUID_DROP))
65 //kbuild:lib-$(CONFIG_SED) += sed.o
67 //usage:#define sed_trivial_usage
68 //usage: "[-i[SFX]] [-nrE] [-f FILE]... [-e CMD]... [FILE]...\n"
69 //usage: "or: sed [-i[SFX]] [-nrE] CMD [FILE]..."
70 //usage:#define sed_full_usage "\n\n"
71 //usage: " -e CMD Add CMD to sed commands to be executed"
72 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
73 //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)"
74 //usage: "\n Optionally back files up, appending SFX"
75 //usage: "\n -n Suppress automatic printing of pattern space"
76 //usage: "\n -r,-E Use extended regex syntax"
78 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
79 //usage: "\nRemaining arguments are input files (stdin if none)."
81 //usage:#define sed_example_usage
82 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
86 #include "common_bufsiz.h"
90 # define dbg(...) bb_error_msg(__VA_ARGS__)
92 # define dbg(...) ((void)0)
97 OPT_in_place = 1 << 0,
100 /* Each sed command turns into one of these structures. */
101 typedef struct sed_cmd_s {
102 /* Ordered by alignment requirements: currently 36 bytes on x86 */
103 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
105 /* address storage */
106 regex_t *beg_match; /* sed -e '/match/cmd' */
107 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
108 regex_t *sub_match; /* For 's/sub_match/string/' */
109 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
110 int beg_line_orig; /* copy of the above, needed for -i */
111 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($). -2-N = +N */
114 FILE *sw_file; /* File (sw) command writes to, NULL for none. */
115 char *string; /* Data string for (saicytb) commands. */
117 unsigned which_match; /* (s) Which match to replace (0 for all) */
119 /* Bitfields (gcc won't group them if we don't) */
120 unsigned invert:1; /* the '!' after the address */
121 unsigned in_match:1; /* Next line also included in match? */
122 unsigned sub_p:1; /* (s) print option */
124 char sw_last_char; /* Last line written by (sw) had no '\n' */
127 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
130 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
134 int be_quiet, regex_type;
137 char *outname, *hold_space;
140 /* list of input files */
141 int current_input_file, last_input_file;
142 char **input_file_list;
145 regmatch_t regmatch[10];
146 regex_t *previous_regex_ptr;
148 /* linked list of sed commands */
149 sed_cmd_t *sed_cmd_head, **sed_cmd_tail;
151 /* linked list of append lines */
152 llist_t *append_head;
157 char *buf; /* Space to hold string */
158 int idx; /* Space used */
159 int len; /* Space allocated */
162 #define G (*(struct globals*)bb_common_bufsiz1)
163 #define INIT_G() do { \
164 setup_common_bufsiz(); \
165 BUILD_BUG_ON(sizeof(G) > COMMON_BUFSIZE); \
166 G.sed_cmd_tail = &G.sed_cmd_head; \
170 #if ENABLE_FEATURE_CLEAN_UP
171 static void sed_free_and_close_stuff(void)
173 sed_cmd_t *sed_cmd = G.sed_cmd_head;
175 llist_free(G.append_head, free);
178 sed_cmd_t *sed_cmd_next = sed_cmd->next;
180 if (sed_cmd->sw_file)
181 fclose(sed_cmd->sw_file);
183 /* Used to free regexps, but now there is code
184 * in get_address() which can reuse a regexp
185 * for constructs as /regexp/cmd1;//cmd2
186 * leading to double-frees here:
188 //if (sed_cmd->beg_match) {
189 // regfree(sed_cmd->beg_match);
190 // free(sed_cmd->beg_match);
192 //if (sed_cmd->end_match) {
193 // regfree(sed_cmd->end_match);
194 // free(sed_cmd->end_match);
196 //if (sed_cmd->sub_match) {
197 // regfree(sed_cmd->sub_match);
198 // free(sed_cmd->sub_match);
200 free(sed_cmd->string);
202 sed_cmd = sed_cmd_next;
208 fclose(G.current_fp);
211 void sed_free_and_close_stuff(void);
214 /* If something bad happens during -i operation, delete temp file */
216 static void cleanup_outname(void)
218 if (G.outname) unlink(G.outname);
221 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
223 static unsigned parse_escapes(char *dest, const char *string, int len, char from, char to)
229 len = strlen(string);
232 if (string[i] == '\\') {
233 if (!to || string[i+1] == from) {
234 if ((*d = to ? to : string[i+1]) == '\0')
240 i++; /* skip backslash in string[] */
242 /* fall through: copy next char verbatim */
244 if ((*d = string[i++]) == '\0')
252 static char *copy_parsing_escapes(const char *string, int len)
255 char *dest = xmalloc(len + 1);
257 /* sed recognizes \n */
258 /* GNU sed also recognizes \t and \r */
259 for (s = "\nn\tt\rr"; *s; s += 2) {
260 len = parse_escapes(dest, string, len, s[1], s[0]);
268 * index_of_next_unescaped_regexp_delim - walks left to right through a string
269 * beginning at a specified index and returns the index of the next regular
270 * expression delimiter (typically a forward slash ('/')) not preceded by
271 * a backslash ('\'). A negative delimiter disables square bracket checking.
273 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
282 delimiter = -delimiter;
285 for (; (ch = str[idx]) != '\0'; idx++) {
288 && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^'))
296 else if (bracket == -1 && ch == '[')
298 else if (ch == delimiter)
302 /* if we make it to here, we've hit the end of the string */
303 bb_error_msg_and_die("unmatched '%c'", delimiter);
307 * Returns the index of the third delimiter
309 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
311 const char *cmdstr_ptr = cmdstr;
312 unsigned char delimiter;
315 /* verify that the 's' or 'y' is followed by something. That something
316 * (typically a 'slash') is now our regexp delimiter... */
318 bb_error_msg_and_die("bad format in substitution expression");
319 delimiter = *cmdstr_ptr++;
321 /* save the match string */
322 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
323 *match = copy_parsing_escapes(cmdstr_ptr, idx);
325 /* save the replacement string */
326 cmdstr_ptr += idx + 1;
327 idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr);
328 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
330 return ((cmdstr_ptr - cmdstr) + idx);
334 * returns the index in the string just past where the address ends.
336 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
338 const char *pos = my_str;
340 if (isdigit(*my_str)) {
341 *linenum = strtol(my_str, (char**)&pos, 10);
342 /* endstr shouldn't ever equal NULL */
343 } else if (*my_str == '$') {
346 } else if (*my_str == '/' || *my_str == '\\') {
354 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
356 temp = copy_parsing_escapes(pos, next);
357 G.previous_regex_ptr = *regex = xzalloc(sizeof(regex_t));
358 xregcomp(*regex, temp, G.regex_type);
361 *regex = G.previous_regex_ptr;
362 if (!G.previous_regex_ptr)
363 bb_error_msg_and_die("no previous regexp");
365 /* Move position to next character after last delimiter */
371 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
372 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
377 /* Skip whitespace, then grab filename to end of line */
378 start = skip_whitespace(filecmdstr);
379 eol = strchrnul(start, '\n');
381 bb_error_msg_and_die("empty filename");
384 /* If lines glued together, put backslash back. */
385 *retval = xstrndup(start, eol-start + 1);
386 (*retval)[eol-start] = '\\';
389 *retval = xstrdup(start);
392 return eol - filecmdstr;
395 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
397 int cflags = G.regex_type;
402 * A substitution command should look something like this:
403 * s/match/replace/ #giIpw
407 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
409 /* determine the number of back references in the match string */
410 /* Note: we compute this here rather than in the do_subst_command()
411 * function to save processor time, at the expense of a little more memory
412 * (4 bits) per sed_cmd */
414 /* process the flags */
416 sed_cmd->which_match = 1;
417 dbg("s flags:'%s'", substr + idx + 1);
418 while (substr[++idx]) {
419 dbg("s flag:'%c'", substr[idx]);
420 /* Parse match number */
421 if (isdigit(substr[idx])) {
422 if (match[0] != '^') {
423 /* Match 0 treated as all, multiple matches we take the last one. */
424 const char *pos = substr + idx;
425 /* FIXME: error check? */
426 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
427 idx = pos - substr - 1;
432 if (isspace(substr[idx]))
435 switch (substr[idx]) {
436 /* Replace all occurrences */
439 sed_cmd->which_match = 0;
441 /* Print pattern space */
449 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx+1, &fname);
450 sed_cmd->sw_file = xfopen_for_write(fname);
451 sed_cmd->sw_last_char = '\n';
455 /* Ignore case (gnu extension) */
462 // while (substr[++idx]) continue;
463 idx += strlen(substr + idx); // same
470 dbg("s bad flags:'%s'", substr + idx);
471 bb_error_msg_and_die("bad option in substitution expression");
475 /* compile the match string into a regex */
476 if (*match != '\0') {
477 /* If match is empty, we use last regex used at runtime */
478 sed_cmd->sub_match = xzalloc(sizeof(regex_t));
479 dbg("xregcomp('%s',%x)", match, cflags);
480 xregcomp(sed_cmd->sub_match, match, cflags);
489 * Process the commands arguments
491 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
493 static const char cmd_letters[] ALIGN1 = "saicrw:btTydDgGhHlnNpPqx={}";
526 BUILD_BUG_ON(sizeof(cmd_letters)-1 != IDX_nul);
528 idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters;
530 /* handle (s)ubstitution command */
532 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
534 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
535 else if (idx <= IDX_c) { /* a,i,c */
538 if (idx < IDX_c) { /* a,i */
539 if (sed_cmd->end_line || sed_cmd->end_match)
540 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
543 if (*cmdstr == '\n' || *cmdstr == '\\') {
547 if (!isspace(*cmdstr))
551 len = strlen(cmdstr);
552 sed_cmd->string = copy_parsing_escapes(cmdstr, len);
554 /* "\anychar" -> "anychar" */
555 parse_escapes(sed_cmd->string, sed_cmd->string, -1, '\0', '\0');
557 /* handle file cmds: (r)ead */
558 else if (idx <= IDX_w) { /* r,w */
559 if (idx < IDX_w) { /* r */
560 if (sed_cmd->end_line || sed_cmd->end_match)
561 bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd);
563 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
564 if (sed_cmd->cmd == 'w') {
565 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
566 sed_cmd->sw_last_char = '\n';
569 /* handle branch commands */
570 else if (idx <= IDX_T) { /* :,b,t,T */
573 cmdstr = skip_whitespace(cmdstr);
574 length = strcspn(cmdstr, semicolon_whitespace);
576 sed_cmd->string = xstrndup(cmdstr, length);
580 /* translation command */
581 else if (idx == IDX_y) {
582 char *match, *replace;
585 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
586 /* \n already parsed, but \delimiter needs unescaping. */
587 parse_escapes(match, match, -1, i, i);
588 parse_escapes(replace, replace, -1, i, i);
590 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
591 for (i = 0; match[i] && replace[i]; i++) {
592 sed_cmd->string[i*2] = match[i];
593 sed_cmd->string[i*2+1] = replace[i];
598 /* if it wasn't a single-letter command that takes no arguments
599 * then it must be an invalid command.
601 else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
602 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
605 /* give back whatever's left over */
610 /* Parse address+command sets, skipping comment lines. */
612 static void add_cmd(const char *cmdstr)
617 /* Append this line to any unfinished line from last time. */
618 if (G.add_cmd_line) {
619 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
620 free(G.add_cmd_line);
621 cmdstr = G.add_cmd_line = tp;
624 /* If this line ends with unescaped backslash, request next line. */
625 n = len = strlen(cmdstr);
626 while (n && cmdstr[n-1] == '\\')
628 if ((len - n) & 1) { /* if odd number of trailing backslashes */
630 G.add_cmd_line = xstrdup(cmdstr);
631 G.add_cmd_line[len-1] = '\0';
635 /* Loop parsing all commands in this line. */
637 /* Skip leading whitespace and semicolons */
638 cmdstr += strspn(cmdstr, semicolon_whitespace);
640 /* If no more commands, exit. */
643 /* if this is a comment, jump past it and keep going */
644 if (*cmdstr == '#') {
645 /* "#n" is the same as using -n on the command line */
646 if (cmdstr[1] == 'n')
648 cmdstr = strpbrk(cmdstr, "\n\r");
654 * format is: [addr][,addr][!]cmd
659 sed_cmd = xzalloc(sizeof(sed_cmd_t));
661 /* first part (if present) is an address: either a '$', a number or a /regex/ */
662 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
663 sed_cmd->beg_line_orig = sed_cmd->beg_line;
665 /* second part (if present) will begin with a comma */
666 if (*cmdstr == ',') {
670 if (*cmdstr == '+' && isdigit(cmdstr[1])) {
671 /* http://sed.sourceforge.net/sedfaq3.html#s3.3
672 * Under GNU sed 3.02+, ssed, and sed15+, <address2>
673 * may also be a notation of the form +num,
674 * indicating the next num lines after <address1> is
676 * GNU sed 4.2.1 accepts even "+" (meaning "+0").
677 * We don't (we check for isdigit, see above), think
678 * about the "+-3" case.
681 /* code is smaller compared to using &cmdstr here: */
682 idx = strtol(cmdstr+1, &end, 10);
683 sed_cmd->end_line = -2 - idx;
686 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
688 idx--; /* if 0, trigger error check below */
691 bb_error_msg_and_die("no address after comma");
692 sed_cmd->end_line_orig = sed_cmd->end_line;
695 /* skip whitespace before the command */
696 cmdstr = skip_whitespace(cmdstr);
698 /* Check for inversion flag */
699 if (*cmdstr == '!') {
703 /* skip whitespace before the command */
704 cmdstr = skip_whitespace(cmdstr);
707 /* last part (mandatory) will be a command */
709 bb_error_msg_and_die("missing command");
710 sed_cmd->cmd = *cmdstr++;
711 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
713 /* cmdstr now points past args.
714 * GNU sed requires a separator, if there are more commands,
715 * else it complains "char N: extra characters after command".
716 * Example: "sed 'p;d'". We also allow "sed 'pd'".
719 /* Add the command to the command array */
720 *G.sed_cmd_tail = sed_cmd;
721 G.sed_cmd_tail = &sed_cmd->next;
724 /* If we glued multiple lines together, free the memory. */
725 free(G.add_cmd_line);
726 G.add_cmd_line = NULL;
729 /* Append to a string, reallocating memory as necessary. */
733 static void pipe_putc(char c)
735 if (G.pipeline.idx == G.pipeline.len) {
736 G.pipeline.buf = xrealloc(G.pipeline.buf,
737 G.pipeline.len + PIPE_GROW);
738 G.pipeline.len += PIPE_GROW;
740 G.pipeline.buf[G.pipeline.idx++] = c;
743 static void do_subst_w_backrefs(char *line, char *replace)
747 /* go through the replacement string */
748 for (i = 0; replace[i]; i++) {
749 /* if we find a backreference (\1, \2, etc.) print the backref'ed text */
750 if (replace[i] == '\\') {
751 unsigned backref = replace[++i] - '0';
753 /* print out the text held in G.regmatch[backref] */
754 if (G.regmatch[backref].rm_so != -1) {
755 j = G.regmatch[backref].rm_so;
756 while (j < G.regmatch[backref].rm_eo)
757 pipe_putc(line[j++]);
761 /* I _think_ it is impossible to get '\' to be
762 * the last char in replace string. Thus we don't check
763 * for replace[i] == NUL. (counterexample anyone?) */
764 /* if we find a backslash escaped character, print the character */
765 pipe_putc(replace[i]);
768 /* if we find an unescaped '&' print out the whole matched text. */
769 if (replace[i] == '&') {
770 j = G.regmatch[0].rm_so;
771 while (j < G.regmatch[0].rm_eo)
772 pipe_putc(line[j++]);
775 /* Otherwise just output the character. */
776 pipe_putc(replace[i]);
780 static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p)
782 char *line = *line_p;
783 unsigned match_count = 0;
785 bool prev_match_empty = 1;
786 bool tried_at_eol = 0;
787 regex_t *current_regex;
789 current_regex = sed_cmd->sub_match;
790 /* Handle empty regex. */
791 if (!current_regex) {
792 current_regex = G.previous_regex_ptr;
794 bb_error_msg_and_die("no previous regexp");
796 G.previous_regex_ptr = current_regex;
798 /* Find the first match */
799 dbg("matching '%s'", line);
800 if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) {
806 /* Initialize temporary output buffer. */
807 G.pipeline.buf = xmalloc(PIPE_GROW);
808 G.pipeline.len = PIPE_GROW;
811 /* Now loop through, substituting for matches */
813 int start = G.regmatch[0].rm_so;
814 int end = G.regmatch[0].rm_eo;
819 /* If we aren't interested in this match, output old line to
820 * end of match and continue */
821 if (sed_cmd->which_match
822 && (sed_cmd->which_match != match_count)
824 for (i = 0; i < end; i++)
826 /* Null match? Print one more char */
827 if (start == end && *line)
832 /* Print everything before the match */
833 for (i = 0; i < start; i++)
836 /* Then print the substitution string,
837 * unless we just matched empty string after non-empty one.
838 * Example: string "cccd", pattern "c*", repl "R":
839 * result is "RdR", not "RRdR": first match "ccc",
840 * second is "" before "d", third is "" after "d".
841 * Second match is NOT replaced!
843 if (prev_match_empty || start != 0 || start != end) {
844 //dbg("%d %d %d", prev_match_empty, start, end);
845 dbg("inserting replacement at %d in '%s'", start, line);
846 do_subst_w_backrefs(line, sed_cmd->string);
847 /* Flag that something has changed */
850 dbg("NOT inserting replacement at %d in '%s'", start, line);
853 /* If matched string is empty (f.e. "c*" pattern),
854 * copy verbatim one char after it before attempting more matches
856 prev_match_empty = (start == end);
857 if (prev_match_empty) {
861 pipe_putc(line[end]);
866 /* Advance past the match */
867 dbg("line += %d", end);
870 /* if we're not doing this globally, get out now */
871 if (sed_cmd->which_match != 0)
874 /* Exit if we are at EOL and already tried matching at it */
881 //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
882 } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH);
884 /* Copy rest of string into output pipeline */
893 *line_p = G.pipeline.buf;
897 /* Set command pointer to point to this label. (Does not handle null label.) */
898 static sed_cmd_t *branch_to(char *label)
902 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
903 if (sed_cmd->cmd == ':'
905 && strcmp(sed_cmd->string, label) == 0
910 bb_error_msg_and_die("can't find label for jump to '%s'", label);
913 static void append(char *s)
915 llist_add_to_end(&G.append_head, s);
918 /* Output line of text. */
920 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
921 * Without them, we had this:
925 * sed "s/i/z/" z1 z2 znull | hexdump -vC
928 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
930 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
936 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
938 char lpc = *last_puts_char;
940 /* Need to insert a '\n' between two files because first file's
941 * last line wasn't terminated? */
942 if (lpc != '\n' && lpc != '\0') {
948 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
952 /* had trailing '\0' and it was last char of file? */
953 if (last_gets_char == LAST_IS_NUL) {
957 /* had trailing '\n' or '\0'? */
958 if (last_gets_char != NO_EOL_CHAR) {
959 fputc(last_gets_char, file);
960 lpc = last_gets_char;
964 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
965 bb_error_msg_and_die(bb_msg_write_error);
967 *last_puts_char = lpc;
970 static void flush_append(char *last_puts_char)
974 /* Output appended lines. */
975 while ((data = (char *)llist_pop(&G.append_head)) != NULL) {
976 /* Append command does not respect "nonterminated-ness"
977 * of last line. Try this:
978 * $ echo -n "woot" | sed -e '/woot/a woo' -
981 * (both lines are terminated with \n)
982 * Therefore we do not propagate "last_gets_char" here,
985 puts_maybe_newline(data, G.nonstdout, last_puts_char, '\n');
990 /* Get next line of input from G.input_file_list, flushing append buffer and
991 * noting if we ran out of files without a newline on the last line we read.
993 static char *get_next_line(char *gets_char, char *last_puts_char)
999 flush_append(last_puts_char);
1001 /* will be returned if last line in the file
1002 * doesn't end with either '\n' or '\0' */
1004 for (; G.current_input_file <= G.last_input_file; G.current_input_file++) {
1005 FILE *fp = G.current_fp;
1007 const char *path = G.input_file_list[G.current_input_file];
1009 if (path != bb_msg_standard_input) {
1010 fp = fopen_or_warn(path, "r");
1012 G.exitcode = EXIT_FAILURE;
1018 /* Read line up to a newline or NUL byte, inclusive,
1019 * return malloc'ed char[]. length of the chunk read
1020 * is stored in len. NULL if EOF/error */
1021 temp = bb_get_chunk_from_file(fp, &len);
1023 /* len > 0 here, it's ok to do temp[len-1] */
1024 char c = temp[len-1];
1025 if (c == '\n' || c == '\0') {
1036 /* else we put NO_EOL_CHAR into *gets_char */
1039 /* NB: I had the idea of peeking next file(s) and returning
1040 * NO_EOL_CHAR only if it is the *last* non-empty
1041 * input file. But there is a case where this won't work:
1042 * file1: "a woo\nb woo"
1043 * file2: "c no\nd no"
1044 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
1045 * (note: *no* newline after "b bang"!) */
1047 /* Close this file and advance to next one */
1048 fclose_if_not_stdin(fp);
1049 G.current_fp = NULL;
1055 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
1057 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
1059 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
1061 G.previous_regex_ptr = sed_cmd->beg_match;
1065 /* Process all the lines in all the files */
1067 static void process_files(void)
1069 char *pattern_space, *next_line;
1071 char last_puts_char = '\n';
1072 char last_gets_char, next_gets_char;
1076 /* Prime the pump */
1077 next_line = get_next_line(&next_gets_char, &last_puts_char);
1079 /* Go through every line in each file */
1083 /* Advance to next line. Stop if out of lines. */
1084 pattern_space = next_line;
1087 last_gets_char = next_gets_char;
1089 /* Read one line in advance so we can act on the last line,
1090 * the '$' address */
1091 next_line = get_next_line(&next_gets_char, &last_puts_char);
1094 /* For every line, go through all the commands */
1096 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1097 int old_matched, matched;
1099 old_matched = sed_cmd->in_match;
1101 /* Determine if this command matches this line: */
1103 dbg("match1:%d", sed_cmd->in_match);
1104 dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line
1105 && !sed_cmd->beg_match && !sed_cmd->end_match));
1106 dbg("match3:%d", (sed_cmd->beg_line > 0
1107 && (sed_cmd->end_line || sed_cmd->end_match
1108 ? (sed_cmd->beg_line <= linenum)
1109 : (sed_cmd->beg_line == linenum)
1112 dbg("match4:%d", (beg_match(sed_cmd, pattern_space)));
1113 dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL));
1115 /* Are we continuing a previous multi-line match? */
1116 sed_cmd->in_match = sed_cmd->in_match
1117 /* Or is no range necessary? */
1118 || (!sed_cmd->beg_line && !sed_cmd->end_line
1119 && !sed_cmd->beg_match && !sed_cmd->end_match)
1120 /* Or did we match the start of a numerical range? */
1121 || (sed_cmd->beg_line > 0
1122 && (sed_cmd->end_line || sed_cmd->end_match
1123 /* note: even if end is numeric and is < linenum too,
1124 * GNU sed matches! We match too, therefore we don't
1125 * check here that linenum <= end.
1127 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1128 * first three input lines are deleted;
1129 * 4th line is matched and printed
1130 * by "2,3" (!) and by "4" ranges
1132 ? (sed_cmd->beg_line <= linenum) /* N,end */
1133 : (sed_cmd->beg_line == linenum) /* N */
1136 /* Or does this line match our begin address regex? */
1137 || (beg_match(sed_cmd, pattern_space))
1138 /* Or did we match last line of input? */
1139 || (sed_cmd->beg_line == -1 && next_line == NULL);
1141 /* Snapshot the value */
1142 matched = sed_cmd->in_match;
1144 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1145 sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum);
1147 /* Is this line the end of the current match? */
1150 if (sed_cmd->end_line <= -2) {
1151 /* address2 is +N, i.e. N lines from beg_line */
1152 sed_cmd->end_line = linenum + (-sed_cmd->end_line - 2);
1154 /* once matched, "n,xxx" range is dead, disabling it */
1155 if (sed_cmd->beg_line > 0) {
1156 sed_cmd->beg_line = -2;
1158 dbg("end1:%d", sed_cmd->end_line ? sed_cmd->end_line == -1
1159 ? !next_line : (sed_cmd->end_line <= linenum)
1160 : !sed_cmd->end_match);
1161 dbg("end2:%d", sed_cmd->end_match && old_matched
1162 && !regexec(sed_cmd->end_match,pattern_space, 0, NULL, 0));
1163 sed_cmd->in_match = !(
1164 /* has the ending line come, or is this a single address command? */
1166 ? sed_cmd->end_line == -1
1168 : (sed_cmd->end_line <= linenum)
1169 : !sed_cmd->end_match
1171 /* or does this line matches our last address regex */
1172 || (sed_cmd->end_match && old_matched
1173 && (regexec(sed_cmd->end_match,
1174 pattern_space, 0, NULL, 0) == 0)
1179 /* Skip blocks of commands we didn't match */
1180 if (sed_cmd->cmd == '{') {
1181 if (sed_cmd->invert ? matched : !matched) {
1182 unsigned nest_cnt = 0;
1184 if (sed_cmd->cmd == '{')
1186 if (sed_cmd->cmd == '}') {
1191 sed_cmd = sed_cmd->next;
1193 bb_error_msg_and_die("unterminated {");
1199 /* Okay, so did this line match? */
1200 if (sed_cmd->invert ? matched : !matched)
1203 /* Update last used regex in case a blank substitute BRE is found */
1204 if (sed_cmd->beg_match) {
1205 G.previous_regex_ptr = sed_cmd->beg_match;
1208 /* actual sedding */
1209 dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1210 pattern_space, next_line, sed_cmd->cmd);
1211 switch (sed_cmd->cmd) {
1213 /* Print line number */
1215 fprintf(G.nonstdout, "%d\n", linenum);
1218 /* Write the current pattern space up to the first newline */
1221 char *tmp = strchr(pattern_space, '\n');
1224 /* TODO: explain why '\n' below */
1225 sed_puts(pattern_space, '\n');
1232 /* Write the current pattern space to output */
1234 /* NB: we print this _before_ the last line
1235 * (of current file) is printed. Even if
1236 * that line is nonterminated, we print
1237 * '\n' here (gnu sed does the same) */
1238 sed_puts(pattern_space, '\n');
1240 /* Delete up through first newline */
1243 char *tmp = strchr(pattern_space, '\n');
1245 overlapping_strcpy(pattern_space, tmp + 1);
1249 /* discard this line. */
1253 /* Substitute with regex */
1255 if (!do_subst_command(sed_cmd, &pattern_space))
1257 dbg("do_subst_command succeeded:'%s'", pattern_space);
1260 /* handle p option */
1262 sed_puts(pattern_space, last_gets_char);
1263 /* handle w option */
1264 if (sed_cmd->sw_file)
1266 pattern_space, sed_cmd->sw_file,
1267 &sed_cmd->sw_last_char, last_gets_char);
1270 /* Append line to linked list to be printed later */
1272 append(xstrdup(sed_cmd->string));
1275 /* Insert text before this line */
1277 sed_puts(sed_cmd->string, '\n');
1280 /* Cut and paste text (replace) */
1282 /* Only triggers on last line of a matching range. */
1283 if (!sed_cmd->in_match)
1284 sed_puts(sed_cmd->string, '\n');
1287 /* Read file, append contents to output */
1291 rfile = fopen_for_read(sed_cmd->string);
1294 while ((line = xmalloc_fgetline(rfile))
1303 /* Write pattern space to file. */
1306 pattern_space, sed_cmd->sw_file,
1307 &sed_cmd->sw_last_char, last_gets_char);
1310 /* Read next line from input */
1313 sed_puts(pattern_space, last_gets_char);
1314 if (next_line == NULL) {
1315 /* If no next line, jump to end of script and exit. */
1318 free(pattern_space);
1319 pattern_space = next_line;
1320 last_gets_char = next_gets_char;
1321 next_line = get_next_line(&next_gets_char, &last_puts_char);
1326 /* Quit. End of script, end of input. */
1328 /* Exit the outer while loop */
1331 goto discard_commands;
1333 /* Append the next line to the current line */
1337 /* If no next line, jump to end of script and exit. */
1338 /* http://www.gnu.org/software/sed/manual/sed.html:
1339 * "Most versions of sed exit without printing anything
1340 * when the N command is issued on the last line of
1341 * a file. GNU sed prints pattern space before exiting
1342 * unless of course the -n command switch has been
1343 * specified. This choice is by design."
1345 if (next_line == NULL) {
1346 //goto discard_line;
1347 goto discard_commands; /* GNU behavior */
1349 /* Append next_line, read new next_line. */
1350 len = strlen(pattern_space);
1351 pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2);
1352 pattern_space[len] = '\n';
1353 strcpy(pattern_space + len+1, next_line);
1354 last_gets_char = next_gets_char;
1355 next_line = get_next_line(&next_gets_char, &last_puts_char);
1360 /* Test/branch if substitution occurred */
1362 if (!substituted) break;
1365 /* Test/branch if substitution didn't occur */
1367 if (substituted) break;
1369 /* Branch to label */
1371 if (!sed_cmd->string) goto discard_commands;
1372 else sed_cmd = branch_to(sed_cmd->string);
1374 /* Transliterate characters */
1378 for (i = 0; pattern_space[i]; i++) {
1379 for (j = 0; sed_cmd->string[j]; j += 2) {
1380 if (pattern_space[i] == sed_cmd->string[j]) {
1381 pattern_space[i] = sed_cmd->string[j + 1];
1389 case 'g': /* Replace pattern space with hold space */
1390 free(pattern_space);
1391 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1393 case 'G': /* Append newline and hold space to pattern space */
1395 int pattern_space_size = 2;
1396 int hold_space_size = 0;
1399 pattern_space_size += strlen(pattern_space);
1401 hold_space_size = strlen(G.hold_space);
1402 pattern_space = xrealloc(pattern_space,
1403 pattern_space_size + hold_space_size);
1404 if (pattern_space_size == 2)
1405 pattern_space[0] = 0;
1406 strcat(pattern_space, "\n");
1408 strcat(pattern_space, G.hold_space);
1409 last_gets_char = '\n';
1413 case 'h': /* Replace hold space with pattern space */
1415 G.hold_space = xstrdup(pattern_space);
1417 case 'H': /* Append newline and pattern space to hold space */
1419 int hold_space_size = 2;
1420 int pattern_space_size = 0;
1423 hold_space_size += strlen(G.hold_space);
1425 pattern_space_size = strlen(pattern_space);
1426 G.hold_space = xrealloc(G.hold_space,
1427 hold_space_size + pattern_space_size);
1429 if (hold_space_size == 2)
1431 strcat(G.hold_space, "\n");
1433 strcat(G.hold_space, pattern_space);
1437 case 'x': /* Exchange hold and pattern space */
1439 char *tmp = pattern_space;
1440 pattern_space = G.hold_space ? G.hold_space : xzalloc(1);
1441 last_gets_char = '\n';
1446 } /* for each cmd */
1449 * Exit point from sedding...
1452 /* we will print the line unless we were told to be quiet ('-n')
1453 or if the line was suppressed (ala 'd'elete) */
1455 sed_puts(pattern_space, last_gets_char);
1457 /* Delete and such jump here. */
1459 flush_append(&last_puts_char /*,last_gets_char*/);
1460 free(pattern_space);
1465 /* It is possible to have a command line argument with embedded
1466 * newlines. This counts as multiple command lines.
1467 * However, newline can be escaped: 's/e/z\<newline>z/'
1468 * add_cmd() handles this.
1471 static void add_cmd_block(char *cmdstr)
1475 cmdstr = sv = xstrdup(cmdstr);
1477 eol = strchr(cmdstr, '\n');
1486 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1487 int sed_main(int argc UNUSED_PARAM, char **argv)
1490 llist_t *opt_e, *opt_f;
1493 #if ENABLE_LONG_OPTS
1494 static const char sed_longopts[] ALIGN1 =
1495 /* name has_arg short */
1496 "in-place\0" Optional_argument "i"
1497 "regexp-extended\0" No_argument "r"
1498 "quiet\0" No_argument "n"
1499 "silent\0" No_argument "n"
1500 "expression\0" Required_argument "e"
1501 "file\0" Required_argument "f";
1506 /* destroy command strings on exit */
1507 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1509 /* Lie to autoconf when it starts asking stupid questions. */
1510 if (argv[1] && strcmp(argv[1], "--version") == 0) {
1511 puts("This is not GNU sed version 4.0");
1515 /* do normal option parsing */
1516 opt_e = opt_f = NULL;
1518 /* -i must be first, to match OPT_in_place definition */
1519 /* -E is a synonym of -r:
1520 * GNU sed 4.2.1 mentions it in neither --help
1521 * nor manpage, but does recognize it.
1523 opt = getopt32long(argv, "^"
1525 "\0" "nn"/*count -n*/,
1527 &opt_i, &opt_e, &opt_f,
1528 &G.be_quiet); /* counter for -n */
1531 if (opt & OPT_in_place) { // -i
1532 die_func = cleanup_outname;
1535 G.regex_type |= REG_EXTENDED; // -r or -E
1537 // G.be_quiet++; // -n (implemented with a counter instead)
1538 while (opt_e) { // -e
1539 add_cmd_block(llist_pop(&opt_e));
1541 while (opt_f) { // -f
1544 cmdfile = xfopen_stdin(llist_pop(&opt_f));
1545 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1549 fclose_if_not_stdin(cmdfile);
1551 /* if we didn't get a pattern from -e or -f, use argv[0] */
1552 if (!(opt & 0x30)) {
1555 add_cmd_block(*argv++);
1557 /* Flush any unfinished commands. */
1560 /* By default, we write to stdout */
1561 G.nonstdout = stdout;
1563 /* argv[0..(argc-1)] should be names of file to process. If no
1564 * files were specified or '-' was specified, take input from stdin.
1565 * Otherwise, we process all the files specified. */
1566 G.input_file_list = argv;
1568 if (opt & OPT_in_place)
1569 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1570 argv[0] = (char*)bb_msg_standard_input;
1571 /* G.last_input_file = 0; - already is */
1575 for (; *argv; argv++) {
1576 struct stat statbuf;
1580 G.last_input_file++;
1582 if (!(opt & OPT_in_place)) {
1583 if (LONE_DASH(*argv)) {
1584 *argv = (char*)bb_msg_standard_input;
1590 /* -i: process each FILE separately: */
1592 if (stat(*argv, &statbuf) != 0) {
1593 bb_simple_perror_msg(*argv);
1594 G.exitcode = EXIT_FAILURE;
1595 G.current_input_file++;
1598 G.outname = xasprintf("%sXXXXXX", *argv);
1599 nonstdoutfd = xmkstemp(G.outname);
1600 G.nonstdout = xfdopen_for_write(nonstdoutfd);
1601 /* Set permissions/owner of output file */
1602 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1603 * but GNU sed 4.2.1 does not preserve them either */
1604 fchmod(nonstdoutfd, statbuf.st_mode);
1605 fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid);
1608 fclose(G.nonstdout);
1609 G.nonstdout = stdout;
1612 char *backupname = xasprintf("%s%s", *argv, opt_i);
1613 xrename(*argv, backupname);
1616 /* else unlink(*argv); - rename below does this */
1617 xrename(G.outname, *argv); //TODO: rollback backup on error?
1621 /* Fix disabled range matches and mangled ",+N" ranges */
1622 for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) {
1623 sed_cmd->beg_line = sed_cmd->beg_line_orig;
1624 sed_cmd->end_line = sed_cmd->end_line_orig;
1627 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1628 * if (G.current_input_file[G.current_input_file] == NULL)
1629 * return G.exitcode;
1630 * but it's not needed since process_files() works correctly
1631 * in this case too. */