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/|$]command
40 * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
41 * - edit commands: (a)ppend, (i)nsert, (c)hange
42 * - file commands: (r)ead
43 * - backreferences in substitution expressions (\0, \1, \2...\9)
44 * - grouped commands: {cmd1;cmd2}
45 * - transliteration (y/source-chars/dest-chars/)
46 * - pattern space hold space storing / swapping (g, h, x)
47 * - labels / branching (: label, b, t, T)
49 * (Note: Specifying an address (range) to match is *optional*; commands
50 * default to the whole pattern space if no specific address match was
54 * - Create a wrapper around regex to make libc's regex conform with sed
56 * Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 //usage:#define sed_trivial_usage
60 //usage: "[-inr] [-f FILE]... [-e CMD]... [FILE]...\n"
61 //usage: "or: sed [-inr] CMD [FILE]..."
62 //usage:#define sed_full_usage "\n\n"
63 //usage: " -e CMD Add CMD to sed commands to be executed"
64 //usage: "\n -f FILE Add FILE contents to sed commands to be executed"
65 //usage: "\n -i Edit files in-place (else sends result to stdout)"
66 //usage: "\n -n Suppress automatic printing of pattern space"
67 //usage: "\n -r Use extended regex syntax"
69 //usage: "\nIf no -e or -f, the first non-option argument is the sed command string."
70 //usage: "\nRemaining arguments are input files (stdin if none)."
72 //usage:#define sed_example_usage
73 //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n"
80 # define dbg(...) bb_error_msg(__VA_ARGS__)
82 # define dbg(...) ((void)0)
87 OPT_in_place
= 1 << 0,
90 /* Each sed command turns into one of these structures. */
91 typedef struct sed_cmd_s
{
92 /* Ordered by alignment requirements: currently 36 bytes on x86 */
93 struct sed_cmd_s
*next
; /* Next command (linked list, NULL terminated) */
96 regex_t
*beg_match
; /* sed -e '/match/cmd' */
97 regex_t
*end_match
; /* sed -e '/match/,/end_match/cmd' */
98 regex_t
*sub_match
; /* For 's/sub_match/string/' */
99 int beg_line
; /* 'sed 1p' 0 == apply commands to all lines */
100 int beg_line_orig
; /* copy of the above, needed for -i */
101 int end_line
; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
103 FILE *sw_file
; /* File (sw) command writes to, -1 for none. */
104 char *string
; /* Data string for (saicytb) commands. */
106 unsigned which_match
; /* (s) Which match to replace (0 for all) */
108 /* Bitfields (gcc won't group them if we don't) */
109 unsigned invert
:1; /* the '!' after the address */
110 unsigned in_match
:1; /* Next line also included in match? */
111 unsigned sub_p
:1; /* (s) print option */
113 char sw_last_char
; /* Last line written by (sw) had no '\n' */
116 char cmd
; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
119 static const char semicolon_whitespace
[] ALIGN1
= "; \n\r\t\v";
123 int be_quiet
, regex_type
;
125 char *outname
, *hold_space
;
127 /* List of input files */
128 int input_file_count
, current_input_file
;
129 FILE **input_file_list
;
131 regmatch_t regmatch
[10];
132 regex_t
*previous_regex_ptr
;
134 /* linked list of sed commands */
135 sed_cmd_t
*sed_cmd_head
, **sed_cmd_tail
;
137 /* Linked list of append lines */
138 llist_t
*append_head
;
143 char *buf
; /* Space to hold string */
144 int idx
; /* Space used */
145 int len
; /* Space allocated */
148 #define G (*(struct globals*)&bb_common_bufsiz1)
149 struct BUG_G_too_big
{
150 char BUG_G_too_big
[sizeof(G
) <= COMMON_BUFSIZE
? 1 : -1];
152 #define INIT_G() do { \
153 G.sed_cmd_tail = &G.sed_cmd_head; \
157 #if ENABLE_FEATURE_CLEAN_UP
158 static void sed_free_and_close_stuff(void)
160 sed_cmd_t
*sed_cmd
= G
.sed_cmd_head
;
162 llist_free(G
.append_head
, free
);
165 sed_cmd_t
*sed_cmd_next
= sed_cmd
->next
;
167 if (sed_cmd
->sw_file
)
168 xprint_and_close_file(sed_cmd
->sw_file
);
170 if (sed_cmd
->beg_match
) {
171 regfree(sed_cmd
->beg_match
);
172 free(sed_cmd
->beg_match
);
174 if (sed_cmd
->end_match
) {
175 regfree(sed_cmd
->end_match
);
176 free(sed_cmd
->end_match
);
178 if (sed_cmd
->sub_match
) {
179 regfree(sed_cmd
->sub_match
);
180 free(sed_cmd
->sub_match
);
182 free(sed_cmd
->string
);
184 sed_cmd
= sed_cmd_next
;
189 while (G
.current_input_file
< G
.input_file_count
)
190 fclose(G
.input_file_list
[G
.current_input_file
++]);
193 void sed_free_and_close_stuff(void);
196 /* If something bad happens during -i operation, delete temp file */
198 static void cleanup_outname(void)
200 if (G
.outname
) unlink(G
.outname
);
203 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
205 static void parse_escapes(char *dest
, const char *string
, int len
, char from
, char to
)
210 if (string
[i
] == '\\') {
211 if (!to
|| string
[i
+1] == from
) {
212 *dest
++ = to
? to
: string
[i
+1];
216 *dest
++ = string
[i
++];
218 /* TODO: is it safe wrt a string with trailing '\\' ? */
219 *dest
++ = string
[i
++];
224 static char *copy_parsing_escapes(const char *string
, int len
)
227 char *dest
= xmalloc(len
+ 1);
229 /* sed recognizes \n */
230 /* GNU sed also recognizes \t and \r */
231 for (s
= "\nn\tt\rr"; *s
; s
+= 2) {
232 parse_escapes(dest
, string
, len
, s
[1], s
[0]);
241 * index_of_next_unescaped_regexp_delim - walks left to right through a string
242 * beginning at a specified index and returns the index of the next regular
243 * expression delimiter (typically a forward slash ('/')) not preceded by
244 * a backslash ('\'). A negative delimiter disables square bracket checking.
246 static int index_of_next_unescaped_regexp_delim(int delimiter
, const char *str
)
255 delimiter
= -delimiter
;
258 for (; (ch
= str
[idx
]) != '\0'; idx
++) {
261 && !(bracket
== idx
- 1 || (bracket
== idx
- 2 && str
[idx
- 1] == '^'))
269 else if (bracket
== -1 && ch
== '[')
271 else if (ch
== delimiter
)
275 /* if we make it to here, we've hit the end of the string */
276 bb_error_msg_and_die("unmatched '%c'", delimiter
);
280 * Returns the index of the third delimiter
282 static int parse_regex_delim(const char *cmdstr
, char **match
, char **replace
)
284 const char *cmdstr_ptr
= cmdstr
;
285 unsigned char delimiter
;
288 /* verify that the 's' or 'y' is followed by something. That something
289 * (typically a 'slash') is now our regexp delimiter... */
291 bb_error_msg_and_die("bad format in substitution expression");
292 delimiter
= *cmdstr_ptr
++;
294 /* save the match string */
295 idx
= index_of_next_unescaped_regexp_delim(delimiter
, cmdstr_ptr
);
296 *match
= copy_parsing_escapes(cmdstr_ptr
, idx
);
298 /* save the replacement string */
299 cmdstr_ptr
+= idx
+ 1;
300 idx
= index_of_next_unescaped_regexp_delim(- (int)delimiter
, cmdstr_ptr
);
301 *replace
= copy_parsing_escapes(cmdstr_ptr
, idx
);
303 return ((cmdstr_ptr
- cmdstr
) + idx
);
307 * returns the index in the string just past where the address ends.
309 static int get_address(const char *my_str
, int *linenum
, regex_t
** regex
)
311 const char *pos
= my_str
;
313 if (isdigit(*my_str
)) {
314 *linenum
= strtol(my_str
, (char**)&pos
, 10);
315 /* endstr shouldnt ever equal NULL */
316 } else if (*my_str
== '$') {
319 } else if (*my_str
== '/' || *my_str
== '\\') {
327 next
= index_of_next_unescaped_regexp_delim(delimiter
, ++pos
);
328 temp
= copy_parsing_escapes(pos
, next
);
329 *regex
= xzalloc(sizeof(regex_t
));
330 xregcomp(*regex
, temp
, G
.regex_type
|REG_NEWLINE
);
332 /* Move position to next character after last delimiter */
338 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
339 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr
, char **retval
)
341 int start
= 0, idx
, hack
= 0;
343 /* Skip whitespace, then grab filename to end of line */
344 while (isspace(filecmdstr
[start
]))
347 while (filecmdstr
[idx
] && filecmdstr
[idx
] != '\n')
350 /* If lines glued together, put backslash back. */
351 if (filecmdstr
[idx
] == '\n')
354 bb_error_msg_and_die("empty filename");
355 *retval
= xstrndup(filecmdstr
+start
, idx
-start
+hack
+1);
357 (*retval
)[idx
] = '\\';
362 static int parse_subst_cmd(sed_cmd_t
*sed_cmd
, const char *substr
)
364 int cflags
= G
.regex_type
;
369 * A substitution command should look something like this:
370 * s/match/replace/ #gIpw
374 idx
= parse_regex_delim(substr
, &match
, &sed_cmd
->string
);
376 /* determine the number of back references in the match string */
377 /* Note: we compute this here rather than in the do_subst_command()
378 * function to save processor time, at the expense of a little more memory
379 * (4 bits) per sed_cmd */
381 /* process the flags */
383 sed_cmd
->which_match
= 1;
384 while (substr
[++idx
]) {
385 /* Parse match number */
386 if (isdigit(substr
[idx
])) {
387 if (match
[0] != '^') {
388 /* Match 0 treated as all, multiple matches we take the last one. */
389 const char *pos
= substr
+ idx
;
390 /* FIXME: error check? */
391 sed_cmd
->which_match
= (unsigned)strtol(substr
+idx
, (char**) &pos
, 10);
397 if (isspace(substr
[idx
]))
400 switch (substr
[idx
]) {
401 /* Replace all occurrences */
404 sed_cmd
->which_match
= 0;
406 /* Print pattern space */
414 idx
+= parse_file_cmd(/*sed_cmd,*/ substr
+idx
, &temp
);
417 /* Ignore case (gnu exension) */
423 // while (substr[++idx]) continue;
424 idx
+= strlen(substr
+ idx
); // same
431 bb_error_msg_and_die("bad option in substitution expression");
435 /* compile the match string into a regex */
436 if (*match
!= '\0') {
437 /* If match is empty, we use last regex used at runtime */
438 sed_cmd
->sub_match
= xzalloc(sizeof(regex_t
));
439 dbg("xregcomp('%s',%x)", match
, cflags
);
440 xregcomp(sed_cmd
->sub_match
, match
, cflags
);
449 * Process the commands arguments
451 static const char *parse_cmd_args(sed_cmd_t
*sed_cmd
, const char *cmdstr
)
453 static const char cmd_letters
[] = "saicrw:btTydDgGhHlnNpPqx={}";
484 struct chk
{ char chk
[sizeof(cmd_letters
)-1 == IDX_nul
? 1 : -1]; };
486 unsigned idx
= strchrnul(cmd_letters
, sed_cmd
->cmd
) - cmd_letters
;
488 /* handle (s)ubstitution command */
490 cmdstr
+= parse_subst_cmd(sed_cmd
, cmdstr
);
492 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
493 else if (idx
<= IDX_c
) { /* a,i,c */
494 if ((sed_cmd
->end_line
|| sed_cmd
->end_match
) && sed_cmd
->cmd
!= 'c')
495 bb_error_msg_and_die("only a beginning address can be specified for edit commands");
497 if (*cmdstr
== '\n' || *cmdstr
== '\\') {
501 if (!isspace(*cmdstr
))
505 sed_cmd
->string
= xstrdup(cmdstr
);
506 /* "\anychar" -> "anychar" */
507 parse_escapes(sed_cmd
->string
, sed_cmd
->string
, strlen(cmdstr
), '\0', '\0');
508 cmdstr
+= strlen(cmdstr
);
510 /* handle file cmds: (r)ead */
511 else if (idx
<= IDX_w
) { /* r,w */
512 if (sed_cmd
->end_line
|| sed_cmd
->end_match
)
513 bb_error_msg_and_die("command only uses one address");
514 cmdstr
+= parse_file_cmd(/*sed_cmd,*/ cmdstr
, &sed_cmd
->string
);
515 if (sed_cmd
->cmd
== 'w') {
516 sed_cmd
->sw_file
= xfopen_for_write(sed_cmd
->string
);
517 sed_cmd
->sw_last_char
= '\n';
520 /* handle branch commands */
521 else if (idx
<= IDX_T
) { /* :,b,t,T */
524 cmdstr
= skip_whitespace(cmdstr
);
525 length
= strcspn(cmdstr
, semicolon_whitespace
);
527 sed_cmd
->string
= xstrndup(cmdstr
, length
);
531 /* translation command */
532 else if (idx
== IDX_y
) {
533 char *match
, *replace
;
536 cmdstr
+= parse_regex_delim(cmdstr
, &match
, &replace
)+1;
537 /* \n already parsed, but \delimiter needs unescaping. */
538 parse_escapes(match
, match
, strlen(match
), i
, i
);
539 parse_escapes(replace
, replace
, strlen(replace
), i
, i
);
541 sed_cmd
->string
= xzalloc((strlen(match
) + 1) * 2);
542 for (i
= 0; match
[i
] && replace
[i
]; i
++) {
543 sed_cmd
->string
[i
*2] = match
[i
];
544 sed_cmd
->string
[i
*2+1] = replace
[i
];
549 /* if it wasnt a single-letter command that takes no arguments
550 * then it must be an invalid command.
552 else if (idx
>= IDX_nul
) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */
553 bb_error_msg_and_die("unsupported command %c", sed_cmd
->cmd
);
556 /* give back whatever's left over */
561 /* Parse address+command sets, skipping comment lines. */
563 static void add_cmd(const char *cmdstr
)
568 /* Append this line to any unfinished line from last time. */
569 if (G
.add_cmd_line
) {
570 char *tp
= xasprintf("%s\n%s", G
.add_cmd_line
, cmdstr
);
571 free(G
.add_cmd_line
);
572 cmdstr
= G
.add_cmd_line
= tp
;
575 /* If this line ends with unescaped backslash, request next line. */
576 n
= len
= strlen(cmdstr
);
577 while (n
&& cmdstr
[n
-1] == '\\')
579 if ((len
- n
) & 1) { /* if odd number of trailing backslashes */
581 G
.add_cmd_line
= xstrdup(cmdstr
);
582 G
.add_cmd_line
[len
-1] = '\0';
586 /* Loop parsing all commands in this line. */
588 /* Skip leading whitespace and semicolons */
589 cmdstr
+= strspn(cmdstr
, semicolon_whitespace
);
591 /* If no more commands, exit. */
594 /* if this is a comment, jump past it and keep going */
595 if (*cmdstr
== '#') {
596 /* "#n" is the same as using -n on the command line */
597 if (cmdstr
[1] == 'n')
599 cmdstr
= strpbrk(cmdstr
, "\n\r");
605 * format is: [addr][,addr][!]cmd
610 sed_cmd
= xzalloc(sizeof(sed_cmd_t
));
612 /* first part (if present) is an address: either a '$', a number or a /regex/ */
613 cmdstr
+= get_address(cmdstr
, &sed_cmd
->beg_line
, &sed_cmd
->beg_match
);
614 sed_cmd
->beg_line_orig
= sed_cmd
->beg_line
;
616 /* second part (if present) will begin with a comma */
617 if (*cmdstr
== ',') {
621 idx
= get_address(cmdstr
, &sed_cmd
->end_line
, &sed_cmd
->end_match
);
623 bb_error_msg_and_die("no address after comma");
627 /* skip whitespace before the command */
628 cmdstr
= skip_whitespace(cmdstr
);
630 /* Check for inversion flag */
631 if (*cmdstr
== '!') {
635 /* skip whitespace before the command */
636 cmdstr
= skip_whitespace(cmdstr
);
639 /* last part (mandatory) will be a command */
641 bb_error_msg_and_die("missing command");
642 sed_cmd
->cmd
= *cmdstr
++;
643 cmdstr
= parse_cmd_args(sed_cmd
, cmdstr
);
645 /* Add the command to the command array */
646 *G
.sed_cmd_tail
= sed_cmd
;
647 G
.sed_cmd_tail
= &sed_cmd
->next
;
650 /* If we glued multiple lines together, free the memory. */
651 free(G
.add_cmd_line
);
652 G
.add_cmd_line
= NULL
;
655 /* Append to a string, reallocating memory as necessary. */
659 static void pipe_putc(char c
)
661 if (G
.pipeline
.idx
== G
.pipeline
.len
) {
662 G
.pipeline
.buf
= xrealloc(G
.pipeline
.buf
,
663 G
.pipeline
.len
+ PIPE_GROW
);
664 G
.pipeline
.len
+= PIPE_GROW
;
666 G
.pipeline
.buf
[G
.pipeline
.idx
++] = c
;
669 static void do_subst_w_backrefs(char *line
, char *replace
)
673 /* go through the replacement string */
674 for (i
= 0; replace
[i
]; i
++) {
675 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
676 if (replace
[i
] == '\\') {
677 unsigned backref
= replace
[++i
] - '0';
679 /* print out the text held in G.regmatch[backref] */
680 if (G
.regmatch
[backref
].rm_so
!= -1) {
681 j
= G
.regmatch
[backref
].rm_so
;
682 while (j
< G
.regmatch
[backref
].rm_eo
)
683 pipe_putc(line
[j
++]);
687 /* I _think_ it is impossible to get '\' to be
688 * the last char in replace string. Thus we dont check
689 * for replace[i] == NUL. (counterexample anyone?) */
690 /* if we find a backslash escaped character, print the character */
691 pipe_putc(replace
[i
]);
694 /* if we find an unescaped '&' print out the whole matched text. */
695 if (replace
[i
] == '&') {
696 j
= G
.regmatch
[0].rm_so
;
697 while (j
< G
.regmatch
[0].rm_eo
)
698 pipe_putc(line
[j
++]);
701 /* Otherwise just output the character. */
702 pipe_putc(replace
[i
]);
706 static int do_subst_command(sed_cmd_t
*sed_cmd
, char **line_p
)
708 char *line
= *line_p
;
710 unsigned match_count
= 0;
711 regex_t
*current_regex
;
713 current_regex
= sed_cmd
->sub_match
;
714 /* Handle empty regex. */
715 if (!current_regex
) {
716 current_regex
= G
.previous_regex_ptr
;
718 bb_error_msg_and_die("no previous regexp");
720 G
.previous_regex_ptr
= current_regex
;
722 /* Find the first match */
723 dbg("matching '%s'", line
);
724 if (REG_NOMATCH
== regexec(current_regex
, line
, 10, G
.regmatch
, 0)) {
730 /* Initialize temporary output buffer. */
731 G
.pipeline
.buf
= xmalloc(PIPE_GROW
);
732 G
.pipeline
.len
= PIPE_GROW
;
735 /* Now loop through, substituting for matches */
739 /* Work around bug in glibc regexec, demonstrated by:
740 * echo " a.b" | busybox sed 's [^ .]* x g'
741 * The match_count check is so not to break
742 * echo "hi" | busybox sed 's/^/!/g'
744 if (!G
.regmatch
[0].rm_so
&& !G
.regmatch
[0].rm_eo
&& match_count
) {
751 /* If we aren't interested in this match, output old line to
752 end of match and continue */
753 if (sed_cmd
->which_match
754 && (sed_cmd
->which_match
!= match_count
)
756 for (i
= 0; i
< G
.regmatch
[0].rm_eo
; i
++)
761 /* print everything before the match */
762 for (i
= 0; i
< G
.regmatch
[0].rm_so
; i
++)
765 /* then print the substitution string */
766 do_subst_w_backrefs(line
, sed_cmd
->string
);
768 /* advance past the match */
769 line
+= G
.regmatch
[0].rm_eo
;
770 /* flag that something has changed */
773 /* if we're not doing this globally, get out now */
774 if (sed_cmd
->which_match
!= 0)
780 //maybe (G.regmatch[0].rm_eo ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL?
781 } while (regexec(current_regex
, line
, 10, G
.regmatch
, REG_NOTBOL
) != REG_NOMATCH
);
783 /* Copy rest of string into output pipeline */
792 *line_p
= G
.pipeline
.buf
;
796 /* Set command pointer to point to this label. (Does not handle null label.) */
797 static sed_cmd_t
*branch_to(char *label
)
801 for (sed_cmd
= G
.sed_cmd_head
; sed_cmd
; sed_cmd
= sed_cmd
->next
) {
802 if (sed_cmd
->cmd
== ':' && sed_cmd
->string
&& !strcmp(sed_cmd
->string
, label
)) {
806 bb_error_msg_and_die("can't find label for jump to '%s'", label
);
809 static void append(char *s
)
811 llist_add_to_end(&G
.append_head
, xstrdup(s
));
814 static void flush_append(void)
818 /* Output appended lines. */
819 while ((data
= (char *)llist_pop(&G
.append_head
))) {
820 fprintf(G
.nonstdout
, "%s\n", data
);
825 static void add_input_file(FILE *file
)
827 G
.input_file_list
= xrealloc_vector(G
.input_file_list
, 2, G
.input_file_count
);
828 G
.input_file_list
[G
.input_file_count
++] = file
;
831 /* Get next line of input from G.input_file_list, flushing append buffer and
832 * noting if we ran out of files without a newline on the last line we read.
838 static char *get_next_line(char *gets_char
)
846 /* will be returned if last line in the file
847 * doesn't end with either '\n' or '\0' */
849 while (G
.current_input_file
< G
.input_file_count
) {
850 FILE *fp
= G
.input_file_list
[G
.current_input_file
];
851 /* Read line up to a newline or NUL byte, inclusive,
852 * return malloc'ed char[]. length of the chunk read
853 * is stored in len. NULL if EOF/error */
854 temp
= bb_get_chunk_from_file(fp
, &len
);
856 /* len > 0 here, it's ok to do temp[len-1] */
857 char c
= temp
[len
-1];
858 if (c
== '\n' || c
== '\0') {
869 /* else we put NO_EOL_CHAR into *gets_char */
872 /* NB: I had the idea of peeking next file(s) and returning
873 * NO_EOL_CHAR only if it is the *last* non-empty
874 * input file. But there is a case where this won't work:
875 * file1: "a woo\nb woo"
876 * file2: "c no\nd no"
877 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
878 * (note: *no* newline after "b bang"!) */
880 /* Close this file and advance to next one */
882 G
.current_input_file
++;
888 /* Output line of text. */
890 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
891 * Without them, we had this:
895 * sed "s/i/z/" z1 z2 znull | hexdump -vC
898 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
900 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
902 static void puts_maybe_newline(char *s
, FILE *file
, char *last_puts_char
, char last_gets_char
)
904 char lpc
= *last_puts_char
;
906 /* Need to insert a '\n' between two files because first file's
907 * last line wasn't terminated? */
908 if (lpc
!= '\n' && lpc
!= '\0') {
914 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
918 /* had trailing '\0' and it was last char of file? */
919 if (last_gets_char
== LAST_IS_NUL
) {
923 /* had trailing '\n' or '\0'? */
924 if (last_gets_char
!= NO_EOL_CHAR
) {
925 fputc(last_gets_char
, file
);
926 lpc
= last_gets_char
;
930 xfunc_error_retval
= 4; /* It's what gnu sed exits with... */
931 bb_error_msg_and_die(bb_msg_write_error
);
933 *last_puts_char
= lpc
;
936 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
938 static int beg_match(sed_cmd_t
*sed_cmd
, const char *pattern_space
)
940 int retval
= sed_cmd
->beg_match
&& !regexec(sed_cmd
->beg_match
, pattern_space
, 0, NULL
, 0);
942 G
.previous_regex_ptr
= sed_cmd
->beg_match
;
946 /* Process all the lines in all the files */
948 static void process_files(void)
950 char *pattern_space
, *next_line
;
952 char last_puts_char
= '\n';
953 char last_gets_char
, next_gets_char
;
958 next_line
= get_next_line(&next_gets_char
);
960 /* Go through every line in each file */
964 /* Advance to next line. Stop if out of lines. */
965 pattern_space
= next_line
;
968 last_gets_char
= next_gets_char
;
970 /* Read one line in advance so we can act on the last line,
972 next_line
= get_next_line(&next_gets_char
);
975 /* For every line, go through all the commands */
977 for (sed_cmd
= G
.sed_cmd_head
; sed_cmd
; sed_cmd
= sed_cmd
->next
) {
978 int old_matched
, matched
;
980 old_matched
= sed_cmd
->in_match
;
982 /* Determine if this command matches this line: */
984 dbg("match1:%d", sed_cmd
->in_match
);
985 dbg("match2:%d", (!sed_cmd
->beg_line
&& !sed_cmd
->end_line
986 && !sed_cmd
->beg_match
&& !sed_cmd
->end_match
));
987 dbg("match3:%d", (sed_cmd
->beg_line
> 0
988 && (sed_cmd
->end_line
|| sed_cmd
->end_match
989 ? (sed_cmd
->beg_line
<= linenum
)
990 : (sed_cmd
->beg_line
== linenum
)
993 dbg("match4:%d", (beg_match(sed_cmd
, pattern_space
)));
994 dbg("match5:%d", (sed_cmd
->beg_line
== -1 && next_line
== NULL
));
996 /* Are we continuing a previous multi-line match? */
997 sed_cmd
->in_match
= sed_cmd
->in_match
998 /* Or is no range necessary? */
999 || (!sed_cmd
->beg_line
&& !sed_cmd
->end_line
1000 && !sed_cmd
->beg_match
&& !sed_cmd
->end_match
)
1001 /* Or did we match the start of a numerical range? */
1002 || (sed_cmd
->beg_line
> 0
1003 && (sed_cmd
->end_line
|| sed_cmd
->end_match
1004 /* note: even if end is numeric and is < linenum too,
1005 * GNU sed matches! We match too, therefore we don't
1006 * check here that linenum <= end.
1008 * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p'
1009 * first three input lines are deleted;
1010 * 4th line is matched and printed
1011 * by "2,3" (!) and by "4" ranges
1013 ? (sed_cmd
->beg_line
<= linenum
) /* N,end */
1014 : (sed_cmd
->beg_line
== linenum
) /* N */
1017 /* Or does this line match our begin address regex? */
1018 || (beg_match(sed_cmd
, pattern_space
))
1019 /* Or did we match last line of input? */
1020 || (sed_cmd
->beg_line
== -1 && next_line
== NULL
);
1022 /* Snapshot the value */
1023 matched
= sed_cmd
->in_match
;
1025 dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d",
1026 sed_cmd
->cmd
, matched
, sed_cmd
->beg_line
, sed_cmd
->end_line
, linenum
);
1028 /* Is this line the end of the current match? */
1031 /* once matched, "n,xxx" range is dead, disabling it */
1032 if (sed_cmd
->beg_line
> 0) {
1033 sed_cmd
->beg_line
= -2;
1035 sed_cmd
->in_match
= !(
1036 /* has the ending line come, or is this a single address command? */
1038 ? sed_cmd
->end_line
== -1
1040 : (sed_cmd
->end_line
<= linenum
)
1041 : !sed_cmd
->end_match
1043 /* or does this line matches our last address regex */
1044 || (sed_cmd
->end_match
&& old_matched
1045 && (regexec(sed_cmd
->end_match
,
1046 pattern_space
, 0, NULL
, 0) == 0)
1051 /* Skip blocks of commands we didn't match */
1052 if (sed_cmd
->cmd
== '{') {
1053 if (sed_cmd
->invert
? matched
: !matched
) {
1054 unsigned nest_cnt
= 0;
1056 if (sed_cmd
->cmd
== '{')
1058 if (sed_cmd
->cmd
== '}') {
1063 sed_cmd
= sed_cmd
->next
;
1065 bb_error_msg_and_die("unterminated {");
1071 /* Okay, so did this line match? */
1072 if (sed_cmd
->invert
? matched
: !matched
)
1075 /* Update last used regex in case a blank substitute BRE is found */
1076 if (sed_cmd
->beg_match
) {
1077 G
.previous_regex_ptr
= sed_cmd
->beg_match
;
1080 /* actual sedding */
1081 dbg("pattern_space:'%s' next_line:'%s' cmd:%c",
1082 pattern_space
, next_line
, sed_cmd
->cmd
);
1083 switch (sed_cmd
->cmd
) {
1085 /* Print line number */
1087 fprintf(G
.nonstdout
, "%d\n", linenum
);
1090 /* Write the current pattern space up to the first newline */
1093 char *tmp
= strchr(pattern_space
, '\n');
1096 /* TODO: explain why '\n' below */
1097 sed_puts(pattern_space
, '\n');
1104 /* Write the current pattern space to output */
1106 /* NB: we print this _before_ the last line
1107 * (of current file) is printed. Even if
1108 * that line is nonterminated, we print
1109 * '\n' here (gnu sed does the same) */
1110 sed_puts(pattern_space
, '\n');
1112 /* Delete up through first newline */
1115 char *tmp
= strchr(pattern_space
, '\n');
1117 overlapping_strcpy(pattern_space
, tmp
+ 1);
1121 /* discard this line. */
1125 /* Substitute with regex */
1127 if (!do_subst_command(sed_cmd
, &pattern_space
))
1129 dbg("do_subst_command succeeeded:'%s'", pattern_space
);
1132 /* handle p option */
1134 sed_puts(pattern_space
, last_gets_char
);
1135 /* handle w option */
1136 if (sed_cmd
->sw_file
)
1138 pattern_space
, sed_cmd
->sw_file
,
1139 &sed_cmd
->sw_last_char
, last_gets_char
);
1142 /* Append line to linked list to be printed later */
1144 append(sed_cmd
->string
);
1147 /* Insert text before this line */
1149 sed_puts(sed_cmd
->string
, '\n');
1152 /* Cut and paste text (replace) */
1154 /* Only triggers on last line of a matching range. */
1155 if (!sed_cmd
->in_match
)
1156 sed_puts(sed_cmd
->string
, '\n');
1159 /* Read file, append contents to output */
1163 rfile
= fopen_for_read(sed_cmd
->string
);
1167 while ((line
= xmalloc_fgetline(rfile
))
1170 xprint_and_close_file(rfile
);
1176 /* Write pattern space to file. */
1179 pattern_space
, sed_cmd
->sw_file
,
1180 &sed_cmd
->sw_last_char
, last_gets_char
);
1183 /* Read next line from input */
1186 sed_puts(pattern_space
, last_gets_char
);
1188 free(pattern_space
);
1189 pattern_space
= next_line
;
1190 last_gets_char
= next_gets_char
;
1191 next_line
= get_next_line(&next_gets_char
);
1198 /* Quit. End of script, end of input. */
1200 /* Exit the outer while loop */
1203 goto discard_commands
;
1205 /* Append the next line to the current line */
1209 /* If no next line, jump to end of script and exit. */
1210 /* http://www.gnu.org/software/sed/manual/sed.html:
1211 * "Most versions of sed exit without printing anything
1212 * when the N command is issued on the last line of
1213 * a file. GNU sed prints pattern space before exiting
1214 * unless of course the -n command switch has been
1215 * specified. This choice is by design."
1217 if (next_line
== NULL
) {
1218 //goto discard_line;
1219 goto discard_commands
; /* GNU behavior */
1221 /* Append next_line, read new next_line. */
1222 len
= strlen(pattern_space
);
1223 pattern_space
= xrealloc(pattern_space
, len
+ strlen(next_line
) + 2);
1224 pattern_space
[len
] = '\n';
1225 strcpy(pattern_space
+ len
+1, next_line
);
1226 last_gets_char
= next_gets_char
;
1227 next_line
= get_next_line(&next_gets_char
);
1232 /* Test/branch if substitution occurred */
1234 if (!substituted
) break;
1237 /* Test/branch if substitution didn't occur */
1239 if (substituted
) break;
1241 /* Branch to label */
1243 if (!sed_cmd
->string
) goto discard_commands
;
1244 else sed_cmd
= branch_to(sed_cmd
->string
);
1246 /* Transliterate characters */
1250 for (i
= 0; pattern_space
[i
]; i
++) {
1251 for (j
= 0; sed_cmd
->string
[j
]; j
+= 2) {
1252 if (pattern_space
[i
] == sed_cmd
->string
[j
]) {
1253 pattern_space
[i
] = sed_cmd
->string
[j
+ 1];
1261 case 'g': /* Replace pattern space with hold space */
1262 free(pattern_space
);
1263 pattern_space
= xstrdup(G
.hold_space
? G
.hold_space
: "");
1265 case 'G': /* Append newline and hold space to pattern space */
1267 int pattern_space_size
= 2;
1268 int hold_space_size
= 0;
1271 pattern_space_size
+= strlen(pattern_space
);
1273 hold_space_size
= strlen(G
.hold_space
);
1274 pattern_space
= xrealloc(pattern_space
,
1275 pattern_space_size
+ hold_space_size
);
1276 if (pattern_space_size
== 2)
1277 pattern_space
[0] = 0;
1278 strcat(pattern_space
, "\n");
1280 strcat(pattern_space
, G
.hold_space
);
1281 last_gets_char
= '\n';
1285 case 'h': /* Replace hold space with pattern space */
1287 G
.hold_space
= xstrdup(pattern_space
);
1289 case 'H': /* Append newline and pattern space to hold space */
1291 int hold_space_size
= 2;
1292 int pattern_space_size
= 0;
1295 hold_space_size
+= strlen(G
.hold_space
);
1297 pattern_space_size
= strlen(pattern_space
);
1298 G
.hold_space
= xrealloc(G
.hold_space
,
1299 hold_space_size
+ pattern_space_size
);
1301 if (hold_space_size
== 2)
1303 strcat(G
.hold_space
, "\n");
1305 strcat(G
.hold_space
, pattern_space
);
1309 case 'x': /* Exchange hold and pattern space */
1311 char *tmp
= pattern_space
;
1312 pattern_space
= G
.hold_space
? G
.hold_space
: xzalloc(1);
1313 last_gets_char
= '\n';
1318 } /* for each cmd */
1321 * Exit point from sedding...
1324 /* we will print the line unless we were told to be quiet ('-n')
1325 or if the line was suppressed (ala 'd'elete) */
1327 sed_puts(pattern_space
, last_gets_char
);
1329 /* Delete and such jump here. */
1332 free(pattern_space
);
1337 /* It is possible to have a command line argument with embedded
1338 * newlines. This counts as multiple command lines.
1339 * However, newline can be escaped: 's/e/z\<newline>z/'
1340 * We check for this.
1343 static void add_cmd_block(char *cmdstr
)
1347 cmdstr
= sv
= xstrdup(cmdstr
);
1349 eol
= strchr(cmdstr
, '\n');
1352 /* Count preceding slashes */
1356 while (sl
!= cmdstr
&& *--sl
== '\\')
1358 /* Odd number of preceding slashes - newline is escaped */
1360 overlapping_strcpy(eol
- 1, eol
);
1361 eol
= strchr(eol
, '\n');
1372 int sed_main(int argc
, char **argv
) MAIN_EXTERNALLY_VISIBLE
;
1373 int sed_main(int argc UNUSED_PARAM
, char **argv
)
1376 llist_t
*opt_e
, *opt_f
;
1377 int status
= EXIT_SUCCESS
;
1381 /* destroy command strings on exit */
1382 if (ENABLE_FEATURE_CLEAN_UP
) atexit(sed_free_and_close_stuff
);
1384 /* Lie to autoconf when it starts asking stupid questions. */
1385 if (argv
[1] && !strcmp(argv
[1], "--version")) {
1386 puts("This is not GNU sed version 4.0");
1390 /* do normal option parsing */
1391 opt_e
= opt_f
= NULL
;
1392 opt_complementary
= "e::f::" /* can occur multiple times */
1393 "nn"; /* count -n */
1394 /* -i must be first, to match OPT_in_place definition */
1395 opt
= getopt32(argv
, "irne:f:", &opt_e
, &opt_f
,
1396 &G
.be_quiet
); /* counter for -n */
1399 if (opt
& OPT_in_place
) { // -i
1400 atexit(cleanup_outname
);
1402 if (opt
& 0x2) G
.regex_type
|= REG_EXTENDED
; // -r
1403 //if (opt & 0x4) G.be_quiet++; // -n
1404 while (opt_e
) { // -e
1405 add_cmd_block(llist_pop(&opt_e
));
1407 while (opt_f
) { // -f
1410 cmdfile
= xfopen_for_read(llist_pop(&opt_f
));
1411 while ((line
= xmalloc_fgetline(cmdfile
)) != NULL
) {
1417 /* if we didn't get a pattern from -e or -f, use argv[0] */
1418 if (!(opt
& 0x18)) {
1421 add_cmd_block(*argv
++);
1423 /* Flush any unfinished commands. */
1426 /* By default, we write to stdout */
1427 G
.nonstdout
= stdout
;
1429 /* argv[0..(argc-1)] should be names of file to process. If no
1430 * files were specified or '-' was specified, take input from stdin.
1431 * Otherwise, we process all the files specified. */
1432 if (argv
[0] == NULL
) {
1433 if (opt
& OPT_in_place
)
1434 bb_error_msg_and_die(bb_msg_requires_arg
, "-i");
1435 add_input_file(stdin
);
1439 for (i
= 0; argv
[i
]; i
++) {
1440 struct stat statbuf
;
1445 if (LONE_DASH(argv
[i
]) && !(opt
& OPT_in_place
)) {
1446 add_input_file(stdin
);
1450 file
= fopen_or_warn(argv
[i
], "r");
1452 status
= EXIT_FAILURE
;
1455 add_input_file(file
);
1456 if (!(opt
& OPT_in_place
)) {
1460 /* -i: process each FILE separately: */
1462 G
.outname
= xasprintf("%sXXXXXX", argv
[i
]);
1463 nonstdoutfd
= xmkstemp(G
.outname
);
1464 G
.nonstdout
= xfdopen_for_write(nonstdoutfd
);
1466 /* Set permissions/owner of output file */
1467 fstat(fileno(file
), &statbuf
);
1468 /* chmod'ing AFTER chown would preserve suid/sgid bits,
1469 * but GNU sed 4.2.1 does not preserve them either */
1470 fchmod(nonstdoutfd
, statbuf
.st_mode
);
1471 fchown(nonstdoutfd
, statbuf
.st_uid
, statbuf
.st_gid
);
1474 fclose(G
.nonstdout
);
1475 G
.nonstdout
= stdout
;
1477 /* unlink(argv[i]); */
1478 xrename(G
.outname
, argv
[i
]);
1482 /* Re-enable disabled range matches */
1483 for (sed_cmd
= G
.sed_cmd_head
; sed_cmd
; sed_cmd
= sed_cmd
->next
) {
1484 sed_cmd
->beg_line
= sed_cmd
->beg_line_orig
;
1487 /* Here, to handle "sed 'cmds' nonexistent_file" case we did:
1488 * if (G.current_input_file >= G.input_file_count)
1490 * but it's not needed since process_files() works correctly
1491 * in this case too. */