Magellan Linux

Contents of /trunk/mkinitrd-magellan/busybox/editors/sed.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 816 - (show annotations) (download)
Fri Apr 24 18:33:46 2009 UTC (15 years, 1 month ago) by niro
File MIME type: text/plain
File size: 35449 byte(s)
-updated to busybox-1.13.4
1 /* vi: set sw=4 ts=4: */
2 /*
3 * sed.c - very minimalist version of sed
4 *
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>
10 *
11 * MAINTAINER: Rob Landley <rob@landley.net>
12 *
13 * Licensed under GPL version 2, see file LICENSE in this tarball for details.
14 */
15
16 /* Code overview.
17
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.
20
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).
25
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.
28
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.
32
33 sed_main() is where external code calls into this, with a command line.
34 */
35
36
37 /*
38 Supported features and commands in this version of sed:
39
40 - comments ('#')
41 - address matching: num|/matchstr/[,num|/matchstr/|$]command
42 - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
43 - edit commands: (a)ppend, (i)nsert, (c)hange
44 - file commands: (r)ead
45 - backreferences in substitution expressions (\0, \1, \2...\9)
46 - grouped commands: {cmd1;cmd2}
47 - transliteration (y/source-chars/dest-chars/)
48 - pattern space hold space storing / swapping (g, h, x)
49 - labels / branching (: label, b, t, T)
50
51 (Note: Specifying an address (range) to match is *optional*; commands
52 default to the whole pattern space if no specific address match was
53 requested.)
54
55 Todo:
56 - Create a wrapper around regex to make libc's regex conform with sed
57
58 Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 */
60
61 #include "libbb.h"
62 #include "xregex.h"
63
64 /* Each sed command turns into one of these structures. */
65 typedef struct sed_cmd_s {
66 /* Ordered by alignment requirements: currently 36 bytes on x86 */
67 struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
68
69 /* address storage */
70 regex_t *beg_match; /* sed -e '/match/cmd' */
71 regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */
72 regex_t *sub_match; /* For 's/sub_match/string/' */
73 int beg_line; /* 'sed 1p' 0 == apply commands to all lines */
74 int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
75
76 FILE *sw_file; /* File (sw) command writes to, -1 for none. */
77 char *string; /* Data string for (saicytb) commands. */
78
79 unsigned which_match; /* (s) Which match to replace (0 for all) */
80
81 /* Bitfields (gcc won't group them if we don't) */
82 unsigned invert:1; /* the '!' after the address */
83 unsigned in_match:1; /* Next line also included in match? */
84 unsigned sub_p:1; /* (s) print option */
85
86 char sw_last_char; /* Last line written by (sw) had no '\n' */
87
88 /* GENERAL FIELDS */
89 char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
90 } sed_cmd_t;
91
92 static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v";
93
94 struct globals {
95 /* options */
96 int be_quiet, regex_type;
97 FILE *nonstdout;
98 char *outname, *hold_space;
99
100 /* List of input files */
101 int input_file_count, current_input_file;
102 FILE **input_file_list;
103
104 regmatch_t regmatch[10];
105 regex_t *previous_regex_ptr;
106
107 /* linked list of sed commands */
108 sed_cmd_t sed_cmd_head, *sed_cmd_tail;
109
110 /* Linked list of append lines */
111 llist_t *append_head;
112
113 char *add_cmd_line;
114
115 struct pipeline {
116 char *buf; /* Space to hold string */
117 int idx; /* Space used */
118 int len; /* Space allocated */
119 } pipeline;
120 };
121 #define G (*(struct globals*)&bb_common_bufsiz1)
122 void BUG_sed_globals_too_big(void);
123 #define INIT_G() do { \
124 if (sizeof(struct globals) > COMMON_BUFSIZE) \
125 BUG_sed_globals_too_big(); \
126 G.sed_cmd_tail = &G.sed_cmd_head; \
127 } while (0)
128
129
130 #if ENABLE_FEATURE_CLEAN_UP
131 static void sed_free_and_close_stuff(void)
132 {
133 sed_cmd_t *sed_cmd = G.sed_cmd_head.next;
134
135 llist_free(G.append_head, free);
136
137 while (sed_cmd) {
138 sed_cmd_t *sed_cmd_next = sed_cmd->next;
139
140 if (sed_cmd->sw_file)
141 xprint_and_close_file(sed_cmd->sw_file);
142
143 if (sed_cmd->beg_match) {
144 regfree(sed_cmd->beg_match);
145 free(sed_cmd->beg_match);
146 }
147 if (sed_cmd->end_match) {
148 regfree(sed_cmd->end_match);
149 free(sed_cmd->end_match);
150 }
151 if (sed_cmd->sub_match) {
152 regfree(sed_cmd->sub_match);
153 free(sed_cmd->sub_match);
154 }
155 free(sed_cmd->string);
156 free(sed_cmd);
157 sed_cmd = sed_cmd_next;
158 }
159
160 free(G.hold_space);
161
162 while (G.current_input_file < G.input_file_count)
163 fclose(G.input_file_list[G.current_input_file++]);
164 }
165 #else
166 void sed_free_and_close_stuff(void);
167 #endif
168
169 /* If something bad happens during -i operation, delete temp file */
170
171 static void cleanup_outname(void)
172 {
173 if (G.outname) unlink(G.outname);
174 }
175
176 /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */
177
178 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
179 {
180 int i = 0;
181
182 while (i < len) {
183 if (string[i] == '\\') {
184 if (!to || string[i+1] == from) {
185 *dest++ = to ? to : string[i+1];
186 i += 2;
187 continue;
188 }
189 *dest++ = string[i++];
190 }
191 /* TODO: is it safe wrt a string with trailing '\\' ? */
192 *dest++ = string[i++];
193 }
194 *dest = '\0';
195 }
196
197 static char *copy_parsing_escapes(const char *string, int len)
198 {
199 char *dest = xmalloc(len + 1);
200
201 parse_escapes(dest, string, len, 'n', '\n');
202 /* GNU sed also recognizes \t */
203 parse_escapes(dest, dest, strlen(dest), 't', '\t');
204 return dest;
205 }
206
207
208 /*
209 * index_of_next_unescaped_regexp_delim - walks left to right through a string
210 * beginning at a specified index and returns the index of the next regular
211 * expression delimiter (typically a forward slash ('/')) not preceded by
212 * a backslash ('\'). A negative delimiter disables square bracket checking.
213 */
214 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
215 {
216 int bracket = -1;
217 int escaped = 0;
218 int idx = 0;
219 char ch;
220
221 if (delimiter < 0) {
222 bracket--;
223 delimiter = -delimiter;
224 }
225
226 for (; (ch = str[idx]); idx++) {
227 if (bracket >= 0) {
228 if (ch == ']' && !(bracket == idx - 1 || (bracket == idx - 2
229 && str[idx - 1] == '^')))
230 bracket = -1;
231 } else if (escaped)
232 escaped = 0;
233 else if (ch == '\\')
234 escaped = 1;
235 else if (bracket == -1 && ch == '[')
236 bracket = idx;
237 else if (ch == delimiter)
238 return idx;
239 }
240
241 /* if we make it to here, we've hit the end of the string */
242 bb_error_msg_and_die("unmatched '%c'", delimiter);
243 }
244
245 /*
246 * Returns the index of the third delimiter
247 */
248 static int parse_regex_delim(const char *cmdstr, char **match, char **replace)
249 {
250 const char *cmdstr_ptr = cmdstr;
251 char delimiter;
252 int idx = 0;
253
254 /* verify that the 's' or 'y' is followed by something. That something
255 * (typically a 'slash') is now our regexp delimiter... */
256 if (*cmdstr == '\0')
257 bb_error_msg_and_die("bad format in substitution expression");
258 delimiter = *cmdstr_ptr++;
259
260 /* save the match string */
261 idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
262 *match = copy_parsing_escapes(cmdstr_ptr, idx);
263
264 /* save the replacement string */
265 cmdstr_ptr += idx + 1;
266 idx = index_of_next_unescaped_regexp_delim(-delimiter, cmdstr_ptr);
267 *replace = copy_parsing_escapes(cmdstr_ptr, idx);
268
269 return ((cmdstr_ptr - cmdstr) + idx);
270 }
271
272 /*
273 * returns the index in the string just past where the address ends.
274 */
275 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
276 {
277 const char *pos = my_str;
278
279 if (isdigit(*my_str)) {
280 *linenum = strtol(my_str, (char**)&pos, 10);
281 /* endstr shouldnt ever equal NULL */
282 } else if (*my_str == '$') {
283 *linenum = -1;
284 pos++;
285 } else if (*my_str == '/' || *my_str == '\\') {
286 int next;
287 char delimiter;
288 char *temp;
289
290 delimiter = '/';
291 if (*my_str == '\\') delimiter = *++pos;
292 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
293 temp = copy_parsing_escapes(pos, next);
294 *regex = xmalloc(sizeof(regex_t));
295 xregcomp(*regex, temp, G.regex_type|REG_NEWLINE);
296 free(temp);
297 /* Move position to next character after last delimiter */
298 pos += (next+1);
299 }
300 return pos - my_str;
301 }
302
303 /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */
304 static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval)
305 {
306 int start = 0, idx, hack = 0;
307
308 /* Skip whitespace, then grab filename to end of line */
309 while (isspace(filecmdstr[start]))
310 start++;
311 idx = start;
312 while (filecmdstr[idx] && filecmdstr[idx] != '\n')
313 idx++;
314
315 /* If lines glued together, put backslash back. */
316 if (filecmdstr[idx] == '\n')
317 hack = 1;
318 if (idx == start)
319 bb_error_msg_and_die("empty filename");
320 *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
321 if (hack)
322 (*retval)[idx] = '\\';
323
324 return idx;
325 }
326
327 static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
328 {
329 int cflags = G.regex_type;
330 char *match;
331 int idx;
332
333 /*
334 * A substitution command should look something like this:
335 * s/match/replace/ #gIpw
336 * || | |||
337 * mandatory optional
338 */
339 idx = parse_regex_delim(substr, &match, &sed_cmd->string);
340
341 /* determine the number of back references in the match string */
342 /* Note: we compute this here rather than in the do_subst_command()
343 * function to save processor time, at the expense of a little more memory
344 * (4 bits) per sed_cmd */
345
346 /* process the flags */
347
348 sed_cmd->which_match = 1;
349 while (substr[++idx]) {
350 /* Parse match number */
351 if (isdigit(substr[idx])) {
352 if (match[0] != '^') {
353 /* Match 0 treated as all, multiple matches we take the last one. */
354 const char *pos = substr + idx;
355 /* FIXME: error check? */
356 sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10);
357 idx = pos - substr;
358 }
359 continue;
360 }
361 /* Skip spaces */
362 if (isspace(substr[idx])) continue;
363
364 switch (substr[idx]) {
365 /* Replace all occurrences */
366 case 'g':
367 if (match[0] != '^')
368 sed_cmd->which_match = 0;
369 break;
370 /* Print pattern space */
371 case 'p':
372 sed_cmd->sub_p = 1;
373 break;
374 /* Write to file */
375 case 'w':
376 {
377 char *temp;
378 idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp);
379 break;
380 }
381 /* Ignore case (gnu exension) */
382 case 'I':
383 cflags |= REG_ICASE;
384 break;
385 /* Comment */
386 case '#':
387 while (substr[++idx]) /*skip all*/;
388 /* Fall through */
389 /* End of command */
390 case ';':
391 case '}':
392 goto out;
393 default:
394 bb_error_msg_and_die("bad option in substitution expression");
395 }
396 }
397 out:
398 /* compile the match string into a regex */
399 if (*match != '\0') {
400 /* If match is empty, we use last regex used at runtime */
401 sed_cmd->sub_match = xmalloc(sizeof(regex_t));
402 xregcomp(sed_cmd->sub_match, match, cflags);
403 }
404 free(match);
405
406 return idx;
407 }
408
409 /*
410 * Process the commands arguments
411 */
412 static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr)
413 {
414 /* handle (s)ubstitution command */
415 if (sed_cmd->cmd == 's')
416 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
417 /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
418 else if (strchr("aic", sed_cmd->cmd)) {
419 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
420 bb_error_msg_and_die
421 ("only a beginning address can be specified for edit commands");
422 for (;;) {
423 if (*cmdstr == '\n' || *cmdstr == '\\') {
424 cmdstr++;
425 break;
426 } else if (isspace(*cmdstr))
427 cmdstr++;
428 else
429 break;
430 }
431 sed_cmd->string = xstrdup(cmdstr);
432 /* "\anychar" -> "anychar" */
433 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0');
434 cmdstr += strlen(cmdstr);
435 /* handle file cmds: (r)ead */
436 } else if (strchr("rw", sed_cmd->cmd)) {
437 if (sed_cmd->end_line || sed_cmd->end_match)
438 bb_error_msg_and_die("command only uses one address");
439 cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string);
440 if (sed_cmd->cmd == 'w') {
441 sed_cmd->sw_file = xfopen_for_write(sed_cmd->string);
442 sed_cmd->sw_last_char = '\n';
443 }
444 /* handle branch commands */
445 } else if (strchr(":btT", sed_cmd->cmd)) {
446 int length;
447
448 cmdstr = skip_whitespace(cmdstr);
449 length = strcspn(cmdstr, semicolon_whitespace);
450 if (length) {
451 sed_cmd->string = xstrndup(cmdstr, length);
452 cmdstr += length;
453 }
454 }
455 /* translation command */
456 else if (sed_cmd->cmd == 'y') {
457 char *match, *replace;
458 int i = cmdstr[0];
459
460 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
461 /* \n already parsed, but \delimiter needs unescaping. */
462 parse_escapes(match, match, strlen(match), i, i);
463 parse_escapes(replace, replace, strlen(replace), i, i);
464
465 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
466 for (i = 0; match[i] && replace[i]; i++) {
467 sed_cmd->string[i*2] = match[i];
468 sed_cmd->string[i*2+1] = replace[i];
469 }
470 free(match);
471 free(replace);
472 }
473 /* if it wasnt a single-letter command that takes no arguments
474 * then it must be an invalid command.
475 */
476 else if (strchr("dDgGhHlnNpPqx={}", sed_cmd->cmd) == 0) {
477 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
478 }
479
480 /* give back whatever's left over */
481 return cmdstr;
482 }
483
484
485 /* Parse address+command sets, skipping comment lines. */
486
487 static void add_cmd(const char *cmdstr)
488 {
489 sed_cmd_t *sed_cmd;
490 int temp;
491
492 /* Append this line to any unfinished line from last time. */
493 if (G.add_cmd_line) {
494 char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr);
495 free(G.add_cmd_line);
496 cmdstr = G.add_cmd_line = tp;
497 }
498
499 /* If this line ends with backslash, request next line. */
500 temp = strlen(cmdstr);
501 if (temp && cmdstr[--temp] == '\\') {
502 if (!G.add_cmd_line)
503 G.add_cmd_line = xstrdup(cmdstr);
504 G.add_cmd_line[temp] = '\0';
505 return;
506 }
507
508 /* Loop parsing all commands in this line. */
509 while (*cmdstr) {
510 /* Skip leading whitespace and semicolons */
511 cmdstr += strspn(cmdstr, semicolon_whitespace);
512
513 /* If no more commands, exit. */
514 if (!*cmdstr) break;
515
516 /* if this is a comment, jump past it and keep going */
517 if (*cmdstr == '#') {
518 /* "#n" is the same as using -n on the command line */
519 if (cmdstr[1] == 'n')
520 G.be_quiet++;
521 cmdstr = strpbrk(cmdstr, "\n\r");
522 if (!cmdstr) break;
523 continue;
524 }
525
526 /* parse the command
527 * format is: [addr][,addr][!]cmd
528 * |----||-----||-|
529 * part1 part2 part3
530 */
531
532 sed_cmd = xzalloc(sizeof(sed_cmd_t));
533
534 /* first part (if present) is an address: either a '$', a number or a /regex/ */
535 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
536
537 /* second part (if present) will begin with a comma */
538 if (*cmdstr == ',') {
539 int idx;
540
541 cmdstr++;
542 idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
543 if (!idx)
544 bb_error_msg_and_die("no address after comma");
545 cmdstr += idx;
546 }
547
548 /* skip whitespace before the command */
549 cmdstr = skip_whitespace(cmdstr);
550
551 /* Check for inversion flag */
552 if (*cmdstr == '!') {
553 sed_cmd->invert = 1;
554 cmdstr++;
555
556 /* skip whitespace before the command */
557 cmdstr = skip_whitespace(cmdstr);
558 }
559
560 /* last part (mandatory) will be a command */
561 if (!*cmdstr)
562 bb_error_msg_and_die("missing command");
563 sed_cmd->cmd = *(cmdstr++);
564 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
565
566 /* Add the command to the command array */
567 G.sed_cmd_tail->next = sed_cmd;
568 G.sed_cmd_tail = G.sed_cmd_tail->next;
569 }
570
571 /* If we glued multiple lines together, free the memory. */
572 free(G.add_cmd_line);
573 G.add_cmd_line = NULL;
574 }
575
576 /* Append to a string, reallocating memory as necessary. */
577
578 #define PIPE_GROW 64
579
580 static void pipe_putc(char c)
581 {
582 if (G.pipeline.idx == G.pipeline.len) {
583 G.pipeline.buf = xrealloc(G.pipeline.buf,
584 G.pipeline.len + PIPE_GROW);
585 G.pipeline.len += PIPE_GROW;
586 }
587 G.pipeline.buf[G.pipeline.idx++] = c;
588 }
589
590 static void do_subst_w_backrefs(char *line, char *replace)
591 {
592 int i,j;
593
594 /* go through the replacement string */
595 for (i = 0; replace[i]; i++) {
596 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
597 if (replace[i] == '\\') {
598 unsigned backref = replace[++i] - '0';
599 if (backref <= 9) {
600 /* print out the text held in G.regmatch[backref] */
601 if (G.regmatch[backref].rm_so != -1) {
602 j = G.regmatch[backref].rm_so;
603 while (j < G.regmatch[backref].rm_eo)
604 pipe_putc(line[j++]);
605 }
606 continue;
607 }
608 /* I _think_ it is impossible to get '\' to be
609 * the last char in replace string. Thus we dont check
610 * for replace[i] == NUL. (counterexample anyone?) */
611 /* if we find a backslash escaped character, print the character */
612 pipe_putc(replace[i]);
613 continue;
614 }
615 /* if we find an unescaped '&' print out the whole matched text. */
616 if (replace[i] == '&') {
617 j = G.regmatch[0].rm_so;
618 while (j < G.regmatch[0].rm_eo)
619 pipe_putc(line[j++]);
620 continue;
621 }
622 /* Otherwise just output the character. */
623 pipe_putc(replace[i]);
624 }
625 }
626
627 static int do_subst_command(sed_cmd_t *sed_cmd, char **line)
628 {
629 char *oldline = *line;
630 int altered = 0;
631 unsigned match_count = 0;
632 regex_t *current_regex;
633
634 /* Handle empty regex. */
635 if (sed_cmd->sub_match == NULL) {
636 current_regex = G.previous_regex_ptr;
637 if (!current_regex)
638 bb_error_msg_and_die("no previous regexp");
639 } else
640 G.previous_regex_ptr = current_regex = sed_cmd->sub_match;
641
642 /* Find the first match */
643 if (REG_NOMATCH == regexec(current_regex, oldline, 10, G.regmatch, 0))
644 return 0;
645
646 /* Initialize temporary output buffer. */
647 G.pipeline.buf = xmalloc(PIPE_GROW);
648 G.pipeline.len = PIPE_GROW;
649 G.pipeline.idx = 0;
650
651 /* Now loop through, substituting for matches */
652 do {
653 int i;
654
655 /* Work around bug in glibc regexec, demonstrated by:
656 echo " a.b" | busybox sed 's [^ .]* x g'
657 The match_count check is so not to break
658 echo "hi" | busybox sed 's/^/!/g' */
659 if (!G.regmatch[0].rm_so && !G.regmatch[0].rm_eo && match_count) {
660 pipe_putc(*oldline++);
661 continue;
662 }
663
664 match_count++;
665
666 /* If we aren't interested in this match, output old line to
667 end of match and continue */
668 if (sed_cmd->which_match
669 && (sed_cmd->which_match != match_count)
670 ) {
671 for (i = 0; i < G.regmatch[0].rm_eo; i++)
672 pipe_putc(*oldline++);
673 continue;
674 }
675
676 /* print everything before the match */
677 for (i = 0; i < G.regmatch[0].rm_so; i++)
678 pipe_putc(oldline[i]);
679
680 /* then print the substitution string */
681 do_subst_w_backrefs(oldline, sed_cmd->string);
682
683 /* advance past the match */
684 oldline += G.regmatch[0].rm_eo;
685 /* flag that something has changed */
686 altered++;
687
688 /* if we're not doing this globally, get out now */
689 if (sed_cmd->which_match)
690 break;
691 } while (*oldline && (regexec(current_regex, oldline, 10, G.regmatch, 0) != REG_NOMATCH));
692
693 /* Copy rest of string into output pipeline */
694
695 while (*oldline)
696 pipe_putc(*oldline++);
697 pipe_putc(0);
698
699 free(*line);
700 *line = G.pipeline.buf;
701 return altered;
702 }
703
704 /* Set command pointer to point to this label. (Does not handle null label.) */
705 static sed_cmd_t *branch_to(char *label)
706 {
707 sed_cmd_t *sed_cmd;
708
709 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
710 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
711 return sed_cmd;
712 }
713 }
714 bb_error_msg_and_die("can't find label for jump to '%s'", label);
715 }
716
717 static void append(char *s)
718 {
719 llist_add_to_end(&G.append_head, xstrdup(s));
720 }
721
722 static void flush_append(void)
723 {
724 char *data;
725
726 /* Output appended lines. */
727 while ((data = (char *)llist_pop(&G.append_head))) {
728 fprintf(G.nonstdout, "%s\n", data);
729 free(data);
730 }
731 }
732
733 static void add_input_file(FILE *file)
734 {
735 G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count);
736 G.input_file_list[G.input_file_count++] = file;
737 }
738
739 /* Get next line of input from G.input_file_list, flushing append buffer and
740 * noting if we ran out of files without a newline on the last line we read.
741 */
742 enum {
743 NO_EOL_CHAR = 1,
744 LAST_IS_NUL = 2,
745 };
746 static char *get_next_line(char *gets_char)
747 {
748 char *temp = NULL;
749 int len;
750 char gc;
751
752 flush_append();
753
754 /* will be returned if last line in the file
755 * doesn't end with either '\n' or '\0' */
756 gc = NO_EOL_CHAR;
757 while (G.current_input_file < G.input_file_count) {
758 FILE *fp = G.input_file_list[G.current_input_file];
759 /* Read line up to a newline or NUL byte, inclusive,
760 * return malloc'ed char[]. length of the chunk read
761 * is stored in len. NULL if EOF/error */
762 temp = bb_get_chunk_from_file(fp, &len);
763 if (temp) {
764 /* len > 0 here, it's ok to do temp[len-1] */
765 char c = temp[len-1];
766 if (c == '\n' || c == '\0') {
767 temp[len-1] = '\0';
768 gc = c;
769 if (c == '\0') {
770 int ch = fgetc(fp);
771 if (ch != EOF)
772 ungetc(ch, fp);
773 else
774 gc = LAST_IS_NUL;
775 }
776 }
777 /* else we put NO_EOL_CHAR into *gets_char */
778 break;
779
780 /* NB: I had the idea of peeking next file(s) and returning
781 * NO_EOL_CHAR only if it is the *last* non-empty
782 * input file. But there is a case where this won't work:
783 * file1: "a woo\nb woo"
784 * file2: "c no\nd no"
785 * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
786 * (note: *no* newline after "b bang"!) */
787 }
788 /* Close this file and advance to next one */
789 fclose(fp);
790 G.current_input_file++;
791 }
792 *gets_char = gc;
793 return temp;
794 }
795
796 /* Output line of text. */
797 /* Note:
798 * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
799 * Without them, we had this:
800 * echo -n thingy >z1
801 * echo -n again >z2
802 * >znull
803 * sed "s/i/z/" z1 z2 znull | hexdump -vC
804 * output:
805 * gnu sed 4.1.5:
806 * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn|
807 * bbox:
808 * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn|
809 */
810 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
811 {
812 char lpc = *last_puts_char;
813
814 /* Need to insert a '\n' between two files because first file's
815 * last line wasn't terminated? */
816 if (lpc != '\n' && lpc != '\0') {
817 fputc('\n', file);
818 lpc = '\n';
819 }
820 fputs(s, file);
821
822 /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
823 if (s[0])
824 lpc = 'x';
825
826 /* had trailing '\0' and it was last char of file? */
827 if (last_gets_char == LAST_IS_NUL) {
828 fputc('\0', file);
829 lpc = 'x'; /* */
830 } else
831 /* had trailing '\n' or '\0'? */
832 if (last_gets_char != NO_EOL_CHAR) {
833 fputc(last_gets_char, file);
834 lpc = last_gets_char;
835 }
836
837 if (ferror(file)) {
838 xfunc_error_retval = 4; /* It's what gnu sed exits with... */
839 bb_error_msg_and_die(bb_msg_write_error);
840 }
841 *last_puts_char = lpc;
842 }
843
844 #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n))
845
846 static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space)
847 {
848 int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0);
849 if (retval)
850 G.previous_regex_ptr = sed_cmd->beg_match;
851 return retval;
852 }
853
854 /* Process all the lines in all the files */
855
856 static void process_files(void)
857 {
858 char *pattern_space, *next_line;
859 int linenum = 0;
860 char last_puts_char = '\n';
861 char last_gets_char, next_gets_char;
862 sed_cmd_t *sed_cmd;
863 int substituted;
864
865 /* Prime the pump */
866 next_line = get_next_line(&next_gets_char);
867
868 /* go through every line in each file */
869 again:
870 substituted = 0;
871
872 /* Advance to next line. Stop if out of lines. */
873 pattern_space = next_line;
874 if (!pattern_space) return;
875 last_gets_char = next_gets_char;
876
877 /* Read one line in advance so we can act on the last line,
878 * the '$' address */
879 next_line = get_next_line(&next_gets_char);
880 linenum++;
881 restart:
882 /* for every line, go through all the commands */
883 for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
884 int old_matched, matched;
885
886 old_matched = sed_cmd->in_match;
887
888 /* Determine if this command matches this line: */
889
890 /* Are we continuing a previous multi-line match? */
891 sed_cmd->in_match = sed_cmd->in_match
892 /* Or is no range necessary? */
893 || (!sed_cmd->beg_line && !sed_cmd->end_line
894 && !sed_cmd->beg_match && !sed_cmd->end_match)
895 /* Or did we match the start of a numerical range? */
896 || (sed_cmd->beg_line > 0 && (sed_cmd->beg_line == linenum))
897 /* Or does this line match our begin address regex? */
898 || (beg_match(sed_cmd, pattern_space))
899 /* Or did we match last line of input? */
900 || (sed_cmd->beg_line == -1 && next_line == NULL);
901
902 /* Snapshot the value */
903
904 matched = sed_cmd->in_match;
905
906 /* Is this line the end of the current match? */
907
908 if (matched) {
909 sed_cmd->in_match = !(
910 /* has the ending line come, or is this a single address command? */
911 (sed_cmd->end_line ?
912 sed_cmd->end_line == -1 ?
913 !next_line
914 : (sed_cmd->end_line <= linenum)
915 : !sed_cmd->end_match
916 )
917 /* or does this line matches our last address regex */
918 || (sed_cmd->end_match && old_matched
919 && (regexec(sed_cmd->end_match,
920 pattern_space, 0, NULL, 0) == 0))
921 );
922 }
923
924 /* Skip blocks of commands we didn't match. */
925 if (sed_cmd->cmd == '{') {
926 if (sed_cmd->invert ? matched : !matched) {
927 while (sed_cmd->cmd != '}') {
928 sed_cmd = sed_cmd->next;
929 if (!sed_cmd)
930 bb_error_msg_and_die("unterminated {");
931 }
932 }
933 continue;
934 }
935
936 /* Okay, so did this line match? */
937 if (sed_cmd->invert ? !matched : matched) {
938 /* Update last used regex in case a blank substitute BRE is found */
939 if (sed_cmd->beg_match) {
940 G.previous_regex_ptr = sed_cmd->beg_match;
941 }
942
943 /* actual sedding */
944 switch (sed_cmd->cmd) {
945
946 /* Print line number */
947 case '=':
948 fprintf(G.nonstdout, "%d\n", linenum);
949 break;
950
951 /* Write the current pattern space up to the first newline */
952 case 'P':
953 {
954 char *tmp = strchr(pattern_space, '\n');
955
956 if (tmp) {
957 *tmp = '\0';
958 /* TODO: explain why '\n' below */
959 sed_puts(pattern_space, '\n');
960 *tmp = '\n';
961 break;
962 }
963 /* Fall Through */
964 }
965
966 /* Write the current pattern space to output */
967 case 'p':
968 /* NB: we print this _before_ the last line
969 * (of current file) is printed. Even if
970 * that line is nonterminated, we print
971 * '\n' here (gnu sed does the same) */
972 sed_puts(pattern_space, '\n');
973 break;
974 /* Delete up through first newline */
975 case 'D':
976 {
977 char *tmp = strchr(pattern_space, '\n');
978
979 if (tmp) {
980 tmp = xstrdup(tmp+1);
981 free(pattern_space);
982 pattern_space = tmp;
983 goto restart;
984 }
985 }
986 /* discard this line. */
987 case 'd':
988 goto discard_line;
989
990 /* Substitute with regex */
991 case 's':
992 if (!do_subst_command(sed_cmd, &pattern_space))
993 break;
994 substituted |= 1;
995
996 /* handle p option */
997 if (sed_cmd->sub_p)
998 sed_puts(pattern_space, last_gets_char);
999 /* handle w option */
1000 if (sed_cmd->sw_file)
1001 puts_maybe_newline(
1002 pattern_space, sed_cmd->sw_file,
1003 &sed_cmd->sw_last_char, last_gets_char);
1004 break;
1005
1006 /* Append line to linked list to be printed later */
1007 case 'a':
1008 append(sed_cmd->string);
1009 break;
1010
1011 /* Insert text before this line */
1012 case 'i':
1013 sed_puts(sed_cmd->string, '\n');
1014 break;
1015
1016 /* Cut and paste text (replace) */
1017 case 'c':
1018 /* Only triggers on last line of a matching range. */
1019 if (!sed_cmd->in_match)
1020 sed_puts(sed_cmd->string, NO_EOL_CHAR);
1021 goto discard_line;
1022
1023 /* Read file, append contents to output */
1024 case 'r':
1025 {
1026 FILE *rfile;
1027
1028 rfile = fopen_for_read(sed_cmd->string);
1029 if (rfile) {
1030 char *line;
1031
1032 while ((line = xmalloc_fgetline(rfile))
1033 != NULL)
1034 append(line);
1035 xprint_and_close_file(rfile);
1036 }
1037
1038 break;
1039 }
1040
1041 /* Write pattern space to file. */
1042 case 'w':
1043 puts_maybe_newline(
1044 pattern_space, sed_cmd->sw_file,
1045 &sed_cmd->sw_last_char, last_gets_char);
1046 break;
1047
1048 /* Read next line from input */
1049 case 'n':
1050 if (!G.be_quiet)
1051 sed_puts(pattern_space, last_gets_char);
1052 if (next_line) {
1053 free(pattern_space);
1054 pattern_space = next_line;
1055 last_gets_char = next_gets_char;
1056 next_line = get_next_line(&next_gets_char);
1057 substituted = 0;
1058 linenum++;
1059 break;
1060 }
1061 /* fall through */
1062
1063 /* Quit. End of script, end of input. */
1064 case 'q':
1065 /* Exit the outer while loop */
1066 free(next_line);
1067 next_line = NULL;
1068 goto discard_commands;
1069
1070 /* Append the next line to the current line */
1071 case 'N':
1072 {
1073 int len;
1074 /* If no next line, jump to end of script and exit. */
1075 if (next_line == NULL) {
1076 /* Jump to end of script and exit */
1077 free(next_line);
1078 next_line = NULL;
1079 goto discard_line;
1080 /* append next_line, read new next_line. */
1081 }
1082 len = strlen(pattern_space);
1083 pattern_space = realloc(pattern_space, len + strlen(next_line) + 2);
1084 pattern_space[len] = '\n';
1085 strcpy(pattern_space + len+1, next_line);
1086 last_gets_char = next_gets_char;
1087 next_line = get_next_line(&next_gets_char);
1088 linenum++;
1089 break;
1090 }
1091
1092 /* Test/branch if substitution occurred */
1093 case 't':
1094 if (!substituted) break;
1095 substituted = 0;
1096 /* Fall through */
1097 /* Test/branch if substitution didn't occur */
1098 case 'T':
1099 if (substituted) break;
1100 /* Fall through */
1101 /* Branch to label */
1102 case 'b':
1103 if (!sed_cmd->string) goto discard_commands;
1104 else sed_cmd = branch_to(sed_cmd->string);
1105 break;
1106 /* Transliterate characters */
1107 case 'y':
1108 {
1109 int i, j;
1110
1111 for (i = 0; pattern_space[i]; i++) {
1112 for (j = 0; sed_cmd->string[j]; j += 2) {
1113 if (pattern_space[i] == sed_cmd->string[j]) {
1114 pattern_space[i] = sed_cmd->string[j + 1];
1115 break;
1116 }
1117 }
1118 }
1119
1120 break;
1121 }
1122 case 'g': /* Replace pattern space with hold space */
1123 free(pattern_space);
1124 pattern_space = xstrdup(G.hold_space ? G.hold_space : "");
1125 break;
1126 case 'G': /* Append newline and hold space to pattern space */
1127 {
1128 int pattern_space_size = 2;
1129 int hold_space_size = 0;
1130
1131 if (pattern_space)
1132 pattern_space_size += strlen(pattern_space);
1133 if (G.hold_space)
1134 hold_space_size = strlen(G.hold_space);
1135 pattern_space = xrealloc(pattern_space,
1136 pattern_space_size + hold_space_size);
1137 if (pattern_space_size == 2)
1138 pattern_space[0] = 0;
1139 strcat(pattern_space, "\n");
1140 if (G.hold_space)
1141 strcat(pattern_space, G.hold_space);
1142 last_gets_char = '\n';
1143
1144 break;
1145 }
1146 case 'h': /* Replace hold space with pattern space */
1147 free(G.hold_space);
1148 G.hold_space = xstrdup(pattern_space);
1149 break;
1150 case 'H': /* Append newline and pattern space to hold space */
1151 {
1152 int hold_space_size = 2;
1153 int pattern_space_size = 0;
1154
1155 if (G.hold_space)
1156 hold_space_size += strlen(G.hold_space);
1157 if (pattern_space)
1158 pattern_space_size = strlen(pattern_space);
1159 G.hold_space = xrealloc(G.hold_space,
1160 hold_space_size + pattern_space_size);
1161
1162 if (hold_space_size == 2)
1163 *G.hold_space = 0;
1164 strcat(G.hold_space, "\n");
1165 if (pattern_space)
1166 strcat(G.hold_space, pattern_space);
1167
1168 break;
1169 }
1170 case 'x': /* Exchange hold and pattern space */
1171 {
1172 char *tmp = pattern_space;
1173 pattern_space = G.hold_space ? : xzalloc(1);
1174 last_gets_char = '\n';
1175 G.hold_space = tmp;
1176 break;
1177 }
1178 }
1179 }
1180 }
1181
1182 /*
1183 * exit point from sedding...
1184 */
1185 discard_commands:
1186 /* we will print the line unless we were told to be quiet ('-n')
1187 or if the line was suppressed (ala 'd'elete) */
1188 if (!G.be_quiet)
1189 sed_puts(pattern_space, last_gets_char);
1190
1191 /* Delete and such jump here. */
1192 discard_line:
1193 flush_append();
1194 free(pattern_space);
1195
1196 goto again;
1197 }
1198
1199 /* It is possible to have a command line argument with embedded
1200 * newlines. This counts as multiple command lines.
1201 * However, newline can be escaped: 's/e/z\<newline>z/'
1202 * We check for this.
1203 */
1204
1205 static void add_cmd_block(char *cmdstr)
1206 {
1207 char *sv, *eol;
1208
1209 cmdstr = sv = xstrdup(cmdstr);
1210 do {
1211 eol = strchr(cmdstr, '\n');
1212 next:
1213 if (eol) {
1214 /* Count preceding slashes */
1215 int slashes = 0;
1216 char *sl = eol;
1217
1218 while (sl != cmdstr && *--sl == '\\')
1219 slashes++;
1220 /* Odd number of preceding slashes - newline is escaped */
1221 if (slashes & 1) {
1222 overlapping_strcpy(eol - 1, eol);
1223 eol = strchr(eol, '\n');
1224 goto next;
1225 }
1226 *eol = '\0';
1227 }
1228 add_cmd(cmdstr);
1229 cmdstr = eol + 1;
1230 } while (eol);
1231 free(sv);
1232 }
1233
1234 int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
1235 int sed_main(int argc UNUSED_PARAM, char **argv)
1236 {
1237 enum {
1238 OPT_in_place = 1 << 0,
1239 };
1240 unsigned opt;
1241 llist_t *opt_e, *opt_f;
1242 int status = EXIT_SUCCESS;
1243
1244 INIT_G();
1245
1246 /* destroy command strings on exit */
1247 if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1248
1249 /* Lie to autoconf when it starts asking stupid questions. */
1250 if (argv[1] && !strcmp(argv[1], "--version")) {
1251 puts("This is not GNU sed version 4.0");
1252 return 0;
1253 }
1254
1255 /* do normal option parsing */
1256 opt_e = opt_f = NULL;
1257 opt_complementary = "e::f::" /* can occur multiple times */
1258 "nn"; /* count -n */
1259 opt = getopt32(argv, "irne:f:", &opt_e, &opt_f,
1260 &G.be_quiet); /* counter for -n */
1261 //argc -= optind;
1262 argv += optind;
1263 if (opt & OPT_in_place) { // -i
1264 atexit(cleanup_outname);
1265 }
1266 if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r
1267 //if (opt & 0x4) G.be_quiet++; // -n
1268 while (opt_e) { // -e
1269 add_cmd_block(llist_pop(&opt_e));
1270 }
1271 while (opt_f) { // -f
1272 char *line;
1273 FILE *cmdfile;
1274 cmdfile = xfopen_for_read(llist_pop(&opt_f));
1275 while ((line = xmalloc_fgetline(cmdfile)) != NULL) {
1276 add_cmd(line);
1277 free(line);
1278 }
1279 fclose(cmdfile);
1280 }
1281 /* if we didn't get a pattern from -e or -f, use argv[0] */
1282 if (!(opt & 0x18)) {
1283 if (!*argv)
1284 bb_show_usage();
1285 add_cmd_block(*argv++);
1286 }
1287 /* Flush any unfinished commands. */
1288 add_cmd("");
1289
1290 /* By default, we write to stdout */
1291 G.nonstdout = stdout;
1292
1293 /* argv[0..(argc-1)] should be names of file to process. If no
1294 * files were specified or '-' was specified, take input from stdin.
1295 * Otherwise, we process all the files specified. */
1296 if (argv[0] == NULL) {
1297 if (opt & OPT_in_place)
1298 bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1299 add_input_file(stdin);
1300 process_files();
1301 } else {
1302 int i;
1303 FILE *file;
1304
1305 for (i = 0; argv[i]; i++) {
1306 struct stat statbuf;
1307 int nonstdoutfd;
1308
1309 if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1310 add_input_file(stdin);
1311 process_files();
1312 continue;
1313 }
1314 file = fopen_or_warn(argv[i], "r");
1315 if (!file) {
1316 status = EXIT_FAILURE;
1317 continue;
1318 }
1319 if (!(opt & OPT_in_place)) {
1320 add_input_file(file);
1321 continue;
1322 }
1323
1324 G.outname = xasprintf("%sXXXXXX", argv[i]);
1325 nonstdoutfd = mkstemp(G.outname);
1326 if (-1 == nonstdoutfd)
1327 bb_perror_msg_and_die("cannot create temp file %s", G.outname);
1328 G.nonstdout = fdopen(nonstdoutfd, "w");
1329
1330 /* Set permissions of output file */
1331
1332 fstat(fileno(file), &statbuf);
1333 fchmod(nonstdoutfd, statbuf.st_mode);
1334 add_input_file(file);
1335 process_files();
1336 fclose(G.nonstdout);
1337
1338 G.nonstdout = stdout;
1339 /* unlink(argv[i]); */
1340 xrename(G.outname, argv[i]);
1341 free(G.outname);
1342 G.outname = NULL;
1343 }
1344 if (G.input_file_count > G.current_input_file)
1345 process_files();
1346 }
1347
1348 return status;
1349 }