Contents of /trunk/mkinitrd-magellan/busybox/editors/sed.c
Parent Directory | Revision Log
Revision 1123 -
(show annotations)
(download)
Wed Aug 18 21:56:57 2010 UTC (13 years, 9 months ago) by niro
File MIME type: text/plain
File size: 36937 byte(s)
Wed Aug 18 21:56:57 2010 UTC (13 years, 9 months ago) by niro
File MIME type: text/plain
File size: 36937 byte(s)
-updated to busybox-1.17.1
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 | } FIX_ALIASING; |
121 | #define G (*(struct globals*)&bb_common_bufsiz1) |
122 | struct BUG_G_too_big { |
123 | char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1]; |
124 | }; |
125 | #define INIT_G() do { \ |
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])) |
363 | continue; |
364 | |
365 | switch (substr[idx]) { |
366 | /* Replace all occurrences */ |
367 | case 'g': |
368 | if (match[0] != '^') |
369 | sed_cmd->which_match = 0; |
370 | break; |
371 | /* Print pattern space */ |
372 | case 'p': |
373 | sed_cmd->sub_p = 1; |
374 | break; |
375 | /* Write to file */ |
376 | case 'w': |
377 | { |
378 | char *temp; |
379 | idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp); |
380 | break; |
381 | } |
382 | /* Ignore case (gnu exension) */ |
383 | case 'I': |
384 | cflags |= REG_ICASE; |
385 | break; |
386 | /* Comment */ |
387 | case '#': |
388 | // while (substr[++idx]) continue; |
389 | idx += strlen(substr + idx); // same |
390 | /* Fall through */ |
391 | /* End of command */ |
392 | case ';': |
393 | case '}': |
394 | goto out; |
395 | default: |
396 | bb_error_msg_and_die("bad option in substitution expression"); |
397 | } |
398 | } |
399 | out: |
400 | /* compile the match string into a regex */ |
401 | if (*match != '\0') { |
402 | /* If match is empty, we use last regex used at runtime */ |
403 | sed_cmd->sub_match = xmalloc(sizeof(regex_t)); |
404 | xregcomp(sed_cmd->sub_match, match, cflags); |
405 | } |
406 | free(match); |
407 | |
408 | return idx; |
409 | } |
410 | |
411 | /* |
412 | * Process the commands arguments |
413 | */ |
414 | static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr) |
415 | { |
416 | /* handle (s)ubstitution command */ |
417 | if (sed_cmd->cmd == 's') |
418 | cmdstr += parse_subst_cmd(sed_cmd, cmdstr); |
419 | /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */ |
420 | else if (strchr("aic", sed_cmd->cmd)) { |
421 | if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c') |
422 | bb_error_msg_and_die("only a beginning address can be specified for edit commands"); |
423 | for (;;) { |
424 | if (*cmdstr == '\n' || *cmdstr == '\\') { |
425 | cmdstr++; |
426 | break; |
427 | } |
428 | if (!isspace(*cmdstr)) |
429 | break; |
430 | cmdstr++; |
431 | } |
432 | sed_cmd->string = xstrdup(cmdstr); |
433 | /* "\anychar" -> "anychar" */ |
434 | parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0'); |
435 | cmdstr += strlen(cmdstr); |
436 | /* handle file cmds: (r)ead */ |
437 | } else if (strchr("rw", sed_cmd->cmd)) { |
438 | if (sed_cmd->end_line || sed_cmd->end_match) |
439 | bb_error_msg_and_die("command only uses one address"); |
440 | cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string); |
441 | if (sed_cmd->cmd == 'w') { |
442 | sed_cmd->sw_file = xfopen_for_write(sed_cmd->string); |
443 | sed_cmd->sw_last_char = '\n'; |
444 | } |
445 | /* handle branch commands */ |
446 | } else if (strchr(":btT", sed_cmd->cmd)) { |
447 | int length; |
448 | |
449 | cmdstr = skip_whitespace(cmdstr); |
450 | length = strcspn(cmdstr, semicolon_whitespace); |
451 | if (length) { |
452 | sed_cmd->string = xstrndup(cmdstr, length); |
453 | cmdstr += length; |
454 | } |
455 | } |
456 | /* translation command */ |
457 | else if (sed_cmd->cmd == 'y') { |
458 | char *match, *replace; |
459 | int i = cmdstr[0]; |
460 | |
461 | cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1; |
462 | /* \n already parsed, but \delimiter needs unescaping. */ |
463 | parse_escapes(match, match, strlen(match), i, i); |
464 | parse_escapes(replace, replace, strlen(replace), i, i); |
465 | |
466 | sed_cmd->string = xzalloc((strlen(match) + 1) * 2); |
467 | for (i = 0; match[i] && replace[i]; i++) { |
468 | sed_cmd->string[i*2] = match[i]; |
469 | sed_cmd->string[i*2+1] = replace[i]; |
470 | } |
471 | free(match); |
472 | free(replace); |
473 | } |
474 | /* if it wasnt a single-letter command that takes no arguments |
475 | * then it must be an invalid command. |
476 | */ |
477 | else if (strchr("dDgGhHlnNpPqx={}", sed_cmd->cmd) == 0) { |
478 | bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd); |
479 | } |
480 | |
481 | /* give back whatever's left over */ |
482 | return cmdstr; |
483 | } |
484 | |
485 | |
486 | /* Parse address+command sets, skipping comment lines. */ |
487 | |
488 | static void add_cmd(const char *cmdstr) |
489 | { |
490 | sed_cmd_t *sed_cmd; |
491 | unsigned len, n; |
492 | |
493 | /* Append this line to any unfinished line from last time. */ |
494 | if (G.add_cmd_line) { |
495 | char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr); |
496 | free(G.add_cmd_line); |
497 | cmdstr = G.add_cmd_line = tp; |
498 | } |
499 | |
500 | /* If this line ends with unescaped backslash, request next line. */ |
501 | n = len = strlen(cmdstr); |
502 | while (n && cmdstr[n-1] == '\\') |
503 | n--; |
504 | if ((len - n) & 1) { /* if odd number of trailing backslashes */ |
505 | if (!G.add_cmd_line) |
506 | G.add_cmd_line = xstrdup(cmdstr); |
507 | G.add_cmd_line[len-1] = '\0'; |
508 | return; |
509 | } |
510 | |
511 | /* Loop parsing all commands in this line. */ |
512 | while (*cmdstr) { |
513 | /* Skip leading whitespace and semicolons */ |
514 | cmdstr += strspn(cmdstr, semicolon_whitespace); |
515 | |
516 | /* If no more commands, exit. */ |
517 | if (!*cmdstr) break; |
518 | |
519 | /* if this is a comment, jump past it and keep going */ |
520 | if (*cmdstr == '#') { |
521 | /* "#n" is the same as using -n on the command line */ |
522 | if (cmdstr[1] == 'n') |
523 | G.be_quiet++; |
524 | cmdstr = strpbrk(cmdstr, "\n\r"); |
525 | if (!cmdstr) break; |
526 | continue; |
527 | } |
528 | |
529 | /* parse the command |
530 | * format is: [addr][,addr][!]cmd |
531 | * |----||-----||-| |
532 | * part1 part2 part3 |
533 | */ |
534 | |
535 | sed_cmd = xzalloc(sizeof(sed_cmd_t)); |
536 | |
537 | /* first part (if present) is an address: either a '$', a number or a /regex/ */ |
538 | cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match); |
539 | |
540 | /* second part (if present) will begin with a comma */ |
541 | if (*cmdstr == ',') { |
542 | int idx; |
543 | |
544 | cmdstr++; |
545 | idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match); |
546 | if (!idx) |
547 | bb_error_msg_and_die("no address after comma"); |
548 | cmdstr += idx; |
549 | } |
550 | |
551 | /* skip whitespace before the command */ |
552 | cmdstr = skip_whitespace(cmdstr); |
553 | |
554 | /* Check for inversion flag */ |
555 | if (*cmdstr == '!') { |
556 | sed_cmd->invert = 1; |
557 | cmdstr++; |
558 | |
559 | /* skip whitespace before the command */ |
560 | cmdstr = skip_whitespace(cmdstr); |
561 | } |
562 | |
563 | /* last part (mandatory) will be a command */ |
564 | if (!*cmdstr) |
565 | bb_error_msg_and_die("missing command"); |
566 | sed_cmd->cmd = *cmdstr++; |
567 | cmdstr = parse_cmd_args(sed_cmd, cmdstr); |
568 | |
569 | /* Add the command to the command array */ |
570 | G.sed_cmd_tail->next = sed_cmd; |
571 | G.sed_cmd_tail = G.sed_cmd_tail->next; |
572 | } |
573 | |
574 | /* If we glued multiple lines together, free the memory. */ |
575 | free(G.add_cmd_line); |
576 | G.add_cmd_line = NULL; |
577 | } |
578 | |
579 | /* Append to a string, reallocating memory as necessary. */ |
580 | |
581 | #define PIPE_GROW 64 |
582 | |
583 | static void pipe_putc(char c) |
584 | { |
585 | if (G.pipeline.idx == G.pipeline.len) { |
586 | G.pipeline.buf = xrealloc(G.pipeline.buf, |
587 | G.pipeline.len + PIPE_GROW); |
588 | G.pipeline.len += PIPE_GROW; |
589 | } |
590 | G.pipeline.buf[G.pipeline.idx++] = c; |
591 | } |
592 | |
593 | static void do_subst_w_backrefs(char *line, char *replace) |
594 | { |
595 | int i, j; |
596 | |
597 | /* go through the replacement string */ |
598 | for (i = 0; replace[i]; i++) { |
599 | /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */ |
600 | if (replace[i] == '\\') { |
601 | unsigned backref = replace[++i] - '0'; |
602 | if (backref <= 9) { |
603 | /* print out the text held in G.regmatch[backref] */ |
604 | if (G.regmatch[backref].rm_so != -1) { |
605 | j = G.regmatch[backref].rm_so; |
606 | while (j < G.regmatch[backref].rm_eo) |
607 | pipe_putc(line[j++]); |
608 | } |
609 | continue; |
610 | } |
611 | /* I _think_ it is impossible to get '\' to be |
612 | * the last char in replace string. Thus we dont check |
613 | * for replace[i] == NUL. (counterexample anyone?) */ |
614 | /* if we find a backslash escaped character, print the character */ |
615 | pipe_putc(replace[i]); |
616 | continue; |
617 | } |
618 | /* if we find an unescaped '&' print out the whole matched text. */ |
619 | if (replace[i] == '&') { |
620 | j = G.regmatch[0].rm_so; |
621 | while (j < G.regmatch[0].rm_eo) |
622 | pipe_putc(line[j++]); |
623 | continue; |
624 | } |
625 | /* Otherwise just output the character. */ |
626 | pipe_putc(replace[i]); |
627 | } |
628 | } |
629 | |
630 | static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p) |
631 | { |
632 | char *line = *line_p; |
633 | int altered = 0; |
634 | unsigned match_count = 0; |
635 | regex_t *current_regex; |
636 | |
637 | current_regex = sed_cmd->sub_match; |
638 | /* Handle empty regex. */ |
639 | if (!current_regex) { |
640 | current_regex = G.previous_regex_ptr; |
641 | if (!current_regex) |
642 | bb_error_msg_and_die("no previous regexp"); |
643 | } |
644 | G.previous_regex_ptr = current_regex; |
645 | |
646 | /* Find the first match */ |
647 | if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) |
648 | return 0; |
649 | |
650 | /* Initialize temporary output buffer. */ |
651 | G.pipeline.buf = xmalloc(PIPE_GROW); |
652 | G.pipeline.len = PIPE_GROW; |
653 | G.pipeline.idx = 0; |
654 | |
655 | /* Now loop through, substituting for matches */ |
656 | do { |
657 | int i; |
658 | |
659 | /* Work around bug in glibc regexec, demonstrated by: |
660 | echo " a.b" | busybox sed 's [^ .]* x g' |
661 | The match_count check is so not to break |
662 | echo "hi" | busybox sed 's/^/!/g' */ |
663 | if (!G.regmatch[0].rm_so && !G.regmatch[0].rm_eo && match_count) { |
664 | pipe_putc(*line++); |
665 | continue; |
666 | } |
667 | |
668 | match_count++; |
669 | |
670 | /* If we aren't interested in this match, output old line to |
671 | end of match and continue */ |
672 | if (sed_cmd->which_match |
673 | && (sed_cmd->which_match != match_count) |
674 | ) { |
675 | for (i = 0; i < G.regmatch[0].rm_eo; i++) |
676 | pipe_putc(*line++); |
677 | continue; |
678 | } |
679 | |
680 | /* print everything before the match */ |
681 | for (i = 0; i < G.regmatch[0].rm_so; i++) |
682 | pipe_putc(line[i]); |
683 | |
684 | /* then print the substitution string */ |
685 | do_subst_w_backrefs(line, sed_cmd->string); |
686 | |
687 | /* advance past the match */ |
688 | line += G.regmatch[0].rm_eo; |
689 | /* flag that something has changed */ |
690 | altered++; |
691 | |
692 | /* if we're not doing this globally, get out now */ |
693 | if (sed_cmd->which_match) |
694 | break; |
695 | |
696 | //maybe (G.regmatch[0].rm_eo ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL? |
697 | } while (*line && regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH); |
698 | |
699 | /* Copy rest of string into output pipeline */ |
700 | while (1) { |
701 | char c = *line++; |
702 | pipe_putc(c); |
703 | if (c == '\0') |
704 | break; |
705 | } |
706 | |
707 | free(*line_p); |
708 | *line_p = G.pipeline.buf; |
709 | return altered; |
710 | } |
711 | |
712 | /* Set command pointer to point to this label. (Does not handle null label.) */ |
713 | static sed_cmd_t *branch_to(char *label) |
714 | { |
715 | sed_cmd_t *sed_cmd; |
716 | |
717 | for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) { |
718 | if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) { |
719 | return sed_cmd; |
720 | } |
721 | } |
722 | bb_error_msg_and_die("can't find label for jump to '%s'", label); |
723 | } |
724 | |
725 | static void append(char *s) |
726 | { |
727 | llist_add_to_end(&G.append_head, xstrdup(s)); |
728 | } |
729 | |
730 | static void flush_append(void) |
731 | { |
732 | char *data; |
733 | |
734 | /* Output appended lines. */ |
735 | while ((data = (char *)llist_pop(&G.append_head))) { |
736 | fprintf(G.nonstdout, "%s\n", data); |
737 | free(data); |
738 | } |
739 | } |
740 | |
741 | static void add_input_file(FILE *file) |
742 | { |
743 | G.input_file_list = xrealloc_vector(G.input_file_list, 2, G.input_file_count); |
744 | G.input_file_list[G.input_file_count++] = file; |
745 | } |
746 | |
747 | /* Get next line of input from G.input_file_list, flushing append buffer and |
748 | * noting if we ran out of files without a newline on the last line we read. |
749 | */ |
750 | enum { |
751 | NO_EOL_CHAR = 1, |
752 | LAST_IS_NUL = 2, |
753 | }; |
754 | static char *get_next_line(char *gets_char) |
755 | { |
756 | char *temp = NULL; |
757 | int len; |
758 | char gc; |
759 | |
760 | flush_append(); |
761 | |
762 | /* will be returned if last line in the file |
763 | * doesn't end with either '\n' or '\0' */ |
764 | gc = NO_EOL_CHAR; |
765 | while (G.current_input_file < G.input_file_count) { |
766 | FILE *fp = G.input_file_list[G.current_input_file]; |
767 | /* Read line up to a newline or NUL byte, inclusive, |
768 | * return malloc'ed char[]. length of the chunk read |
769 | * is stored in len. NULL if EOF/error */ |
770 | temp = bb_get_chunk_from_file(fp, &len); |
771 | if (temp) { |
772 | /* len > 0 here, it's ok to do temp[len-1] */ |
773 | char c = temp[len-1]; |
774 | if (c == '\n' || c == '\0') { |
775 | temp[len-1] = '\0'; |
776 | gc = c; |
777 | if (c == '\0') { |
778 | int ch = fgetc(fp); |
779 | if (ch != EOF) |
780 | ungetc(ch, fp); |
781 | else |
782 | gc = LAST_IS_NUL; |
783 | } |
784 | } |
785 | /* else we put NO_EOL_CHAR into *gets_char */ |
786 | break; |
787 | |
788 | /* NB: I had the idea of peeking next file(s) and returning |
789 | * NO_EOL_CHAR only if it is the *last* non-empty |
790 | * input file. But there is a case where this won't work: |
791 | * file1: "a woo\nb woo" |
792 | * file2: "c no\nd no" |
793 | * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang" |
794 | * (note: *no* newline after "b bang"!) */ |
795 | } |
796 | /* Close this file and advance to next one */ |
797 | fclose(fp); |
798 | G.current_input_file++; |
799 | } |
800 | *gets_char = gc; |
801 | return temp; |
802 | } |
803 | |
804 | /* Output line of text. */ |
805 | /* Note: |
806 | * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed. |
807 | * Without them, we had this: |
808 | * echo -n thingy >z1 |
809 | * echo -n again >z2 |
810 | * >znull |
811 | * sed "s/i/z/" z1 z2 znull | hexdump -vC |
812 | * output: |
813 | * gnu sed 4.1.5: |
814 | * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn| |
815 | * bbox: |
816 | * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn| |
817 | */ |
818 | static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char) |
819 | { |
820 | char lpc = *last_puts_char; |
821 | |
822 | /* Need to insert a '\n' between two files because first file's |
823 | * last line wasn't terminated? */ |
824 | if (lpc != '\n' && lpc != '\0') { |
825 | fputc('\n', file); |
826 | lpc = '\n'; |
827 | } |
828 | fputs(s, file); |
829 | |
830 | /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */ |
831 | if (s[0]) |
832 | lpc = 'x'; |
833 | |
834 | /* had trailing '\0' and it was last char of file? */ |
835 | if (last_gets_char == LAST_IS_NUL) { |
836 | fputc('\0', file); |
837 | lpc = 'x'; /* */ |
838 | } else |
839 | /* had trailing '\n' or '\0'? */ |
840 | if (last_gets_char != NO_EOL_CHAR) { |
841 | fputc(last_gets_char, file); |
842 | lpc = last_gets_char; |
843 | } |
844 | |
845 | if (ferror(file)) { |
846 | xfunc_error_retval = 4; /* It's what gnu sed exits with... */ |
847 | bb_error_msg_and_die(bb_msg_write_error); |
848 | } |
849 | *last_puts_char = lpc; |
850 | } |
851 | |
852 | #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n)) |
853 | |
854 | static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space) |
855 | { |
856 | int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0); |
857 | if (retval) |
858 | G.previous_regex_ptr = sed_cmd->beg_match; |
859 | return retval; |
860 | } |
861 | |
862 | /* Process all the lines in all the files */ |
863 | |
864 | static void process_files(void) |
865 | { |
866 | char *pattern_space, *next_line; |
867 | int linenum = 0; |
868 | char last_puts_char = '\n'; |
869 | char last_gets_char, next_gets_char; |
870 | sed_cmd_t *sed_cmd; |
871 | int substituted; |
872 | |
873 | /* Prime the pump */ |
874 | next_line = get_next_line(&next_gets_char); |
875 | |
876 | /* Go through every line in each file */ |
877 | again: |
878 | substituted = 0; |
879 | |
880 | /* Advance to next line. Stop if out of lines. */ |
881 | pattern_space = next_line; |
882 | if (!pattern_space) |
883 | return; |
884 | last_gets_char = next_gets_char; |
885 | |
886 | /* Read one line in advance so we can act on the last line, |
887 | * the '$' address */ |
888 | next_line = get_next_line(&next_gets_char); |
889 | linenum++; |
890 | |
891 | /* For every line, go through all the commands */ |
892 | restart: |
893 | for (sed_cmd = G.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) { |
894 | int old_matched, matched; |
895 | |
896 | old_matched = sed_cmd->in_match; |
897 | |
898 | /* Determine if this command matches this line: */ |
899 | |
900 | //bb_error_msg("match1:%d", sed_cmd->in_match); |
901 | //bb_error_msg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line |
902 | // && !sed_cmd->beg_match && !sed_cmd->end_match)); |
903 | //bb_error_msg("match3:%d", (sed_cmd->beg_line > 0 |
904 | // && (sed_cmd->end_line || sed_cmd->end_match |
905 | // ? (sed_cmd->beg_line <= linenum) |
906 | // : (sed_cmd->beg_line == linenum) |
907 | // ) |
908 | // ) |
909 | //bb_error_msg("match4:%d", (beg_match(sed_cmd, pattern_space))); |
910 | //bb_error_msg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL)); |
911 | |
912 | /* Are we continuing a previous multi-line match? */ |
913 | sed_cmd->in_match = sed_cmd->in_match |
914 | /* Or is no range necessary? */ |
915 | || (!sed_cmd->beg_line && !sed_cmd->end_line |
916 | && !sed_cmd->beg_match && !sed_cmd->end_match) |
917 | /* Or did we match the start of a numerical range? */ |
918 | || (sed_cmd->beg_line > 0 |
919 | && (sed_cmd->end_line || sed_cmd->end_match |
920 | /* note: even if end is numeric and is < linenum too, |
921 | * GNU sed matches! We match too */ |
922 | ? (sed_cmd->beg_line <= linenum) /* N,end */ |
923 | : (sed_cmd->beg_line == linenum) /* N */ |
924 | ) |
925 | ) |
926 | /* Or does this line match our begin address regex? */ |
927 | || (beg_match(sed_cmd, pattern_space)) |
928 | /* Or did we match last line of input? */ |
929 | || (sed_cmd->beg_line == -1 && next_line == NULL); |
930 | |
931 | /* Snapshot the value */ |
932 | matched = sed_cmd->in_match; |
933 | |
934 | //bb_error_msg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d", |
935 | //sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum); |
936 | |
937 | /* Is this line the end of the current match? */ |
938 | |
939 | if (matched) { |
940 | /* once matched, "n,xxx" range is dead, disabling it */ |
941 | if (sed_cmd->beg_line > 0) |
942 | sed_cmd->beg_line = -2; |
943 | sed_cmd->in_match = !( |
944 | /* has the ending line come, or is this a single address command? */ |
945 | (sed_cmd->end_line ? |
946 | sed_cmd->end_line == -1 ? |
947 | !next_line |
948 | : (sed_cmd->end_line <= linenum) |
949 | : !sed_cmd->end_match |
950 | ) |
951 | /* or does this line matches our last address regex */ |
952 | || (sed_cmd->end_match && old_matched |
953 | && (regexec(sed_cmd->end_match, |
954 | pattern_space, 0, NULL, 0) == 0)) |
955 | ); |
956 | } |
957 | |
958 | /* Skip blocks of commands we didn't match */ |
959 | if (sed_cmd->cmd == '{') { |
960 | if (sed_cmd->invert ? matched : !matched) { |
961 | unsigned nest_cnt = 0; |
962 | while (1) { |
963 | if (sed_cmd->cmd == '{') |
964 | nest_cnt++; |
965 | if (sed_cmd->cmd == '}') { |
966 | nest_cnt--; |
967 | if (nest_cnt == 0) |
968 | break; |
969 | } |
970 | sed_cmd = sed_cmd->next; |
971 | if (!sed_cmd) |
972 | bb_error_msg_and_die("unterminated {"); |
973 | } |
974 | } |
975 | continue; |
976 | } |
977 | |
978 | /* Okay, so did this line match? */ |
979 | if (sed_cmd->invert ? matched : !matched) |
980 | continue; /* no */ |
981 | |
982 | /* Update last used regex in case a blank substitute BRE is found */ |
983 | if (sed_cmd->beg_match) { |
984 | G.previous_regex_ptr = sed_cmd->beg_match; |
985 | } |
986 | |
987 | /* actual sedding */ |
988 | switch (sed_cmd->cmd) { |
989 | |
990 | /* Print line number */ |
991 | case '=': |
992 | fprintf(G.nonstdout, "%d\n", linenum); |
993 | break; |
994 | |
995 | /* Write the current pattern space up to the first newline */ |
996 | case 'P': |
997 | { |
998 | char *tmp = strchr(pattern_space, '\n'); |
999 | if (tmp) { |
1000 | *tmp = '\0'; |
1001 | /* TODO: explain why '\n' below */ |
1002 | sed_puts(pattern_space, '\n'); |
1003 | *tmp = '\n'; |
1004 | break; |
1005 | } |
1006 | /* Fall Through */ |
1007 | } |
1008 | |
1009 | /* Write the current pattern space to output */ |
1010 | case 'p': |
1011 | /* NB: we print this _before_ the last line |
1012 | * (of current file) is printed. Even if |
1013 | * that line is nonterminated, we print |
1014 | * '\n' here (gnu sed does the same) */ |
1015 | sed_puts(pattern_space, '\n'); |
1016 | break; |
1017 | /* Delete up through first newline */ |
1018 | case 'D': |
1019 | { |
1020 | char *tmp = strchr(pattern_space, '\n'); |
1021 | if (tmp) { |
1022 | overlapping_strcpy(pattern_space, tmp + 1); |
1023 | goto restart; |
1024 | } |
1025 | } |
1026 | /* discard this line. */ |
1027 | case 'd': |
1028 | goto discard_line; |
1029 | |
1030 | /* Substitute with regex */ |
1031 | case 's': |
1032 | if (!do_subst_command(sed_cmd, &pattern_space)) |
1033 | break; |
1034 | substituted |= 1; |
1035 | |
1036 | /* handle p option */ |
1037 | if (sed_cmd->sub_p) |
1038 | sed_puts(pattern_space, last_gets_char); |
1039 | /* handle w option */ |
1040 | if (sed_cmd->sw_file) |
1041 | puts_maybe_newline( |
1042 | pattern_space, sed_cmd->sw_file, |
1043 | &sed_cmd->sw_last_char, last_gets_char); |
1044 | break; |
1045 | |
1046 | /* Append line to linked list to be printed later */ |
1047 | case 'a': |
1048 | append(sed_cmd->string); |
1049 | break; |
1050 | |
1051 | /* Insert text before this line */ |
1052 | case 'i': |
1053 | sed_puts(sed_cmd->string, '\n'); |
1054 | break; |
1055 | |
1056 | /* Cut and paste text (replace) */ |
1057 | case 'c': |
1058 | /* Only triggers on last line of a matching range. */ |
1059 | if (!sed_cmd->in_match) |
1060 | sed_puts(sed_cmd->string, '\n'); |
1061 | goto discard_line; |
1062 | |
1063 | /* Read file, append contents to output */ |
1064 | case 'r': |
1065 | { |
1066 | FILE *rfile; |
1067 | rfile = fopen_for_read(sed_cmd->string); |
1068 | if (rfile) { |
1069 | char *line; |
1070 | |
1071 | while ((line = xmalloc_fgetline(rfile)) |
1072 | != NULL) |
1073 | append(line); |
1074 | xprint_and_close_file(rfile); |
1075 | } |
1076 | |
1077 | break; |
1078 | } |
1079 | |
1080 | /* Write pattern space to file. */ |
1081 | case 'w': |
1082 | puts_maybe_newline( |
1083 | pattern_space, sed_cmd->sw_file, |
1084 | &sed_cmd->sw_last_char, last_gets_char); |
1085 | break; |
1086 | |
1087 | /* Read next line from input */ |
1088 | case 'n': |
1089 | if (!G.be_quiet) |
1090 | sed_puts(pattern_space, last_gets_char); |
1091 | if (next_line) { |
1092 | free(pattern_space); |
1093 | pattern_space = next_line; |
1094 | last_gets_char = next_gets_char; |
1095 | next_line = get_next_line(&next_gets_char); |
1096 | substituted = 0; |
1097 | linenum++; |
1098 | break; |
1099 | } |
1100 | /* fall through */ |
1101 | |
1102 | /* Quit. End of script, end of input. */ |
1103 | case 'q': |
1104 | /* Exit the outer while loop */ |
1105 | free(next_line); |
1106 | next_line = NULL; |
1107 | goto discard_commands; |
1108 | |
1109 | /* Append the next line to the current line */ |
1110 | case 'N': |
1111 | { |
1112 | int len; |
1113 | /* If no next line, jump to end of script and exit. */ |
1114 | if (next_line == NULL) { |
1115 | free(next_line); |
1116 | next_line = NULL; |
1117 | goto discard_line; |
1118 | } |
1119 | /* Append next_line, read new next_line. */ |
1120 | len = strlen(pattern_space); |
1121 | pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2); |
1122 | pattern_space[len] = '\n'; |
1123 | strcpy(pattern_space + len+1, next_line); |
1124 | last_gets_char = next_gets_char; |
1125 | next_line = get_next_line(&next_gets_char); |
1126 | linenum++; |
1127 | break; |
1128 | } |
1129 | |
1130 | /* Test/branch if substitution occurred */ |
1131 | case 't': |
1132 | if (!substituted) break; |
1133 | substituted = 0; |
1134 | /* Fall through */ |
1135 | /* Test/branch if substitution didn't occur */ |
1136 | case 'T': |
1137 | if (substituted) break; |
1138 | /* Fall through */ |
1139 | /* Branch to label */ |
1140 | case 'b': |
1141 | if (!sed_cmd->string) goto discard_commands; |
1142 | else sed_cmd = branch_to(sed_cmd->string); |
1143 | break; |
1144 | /* Transliterate characters */ |
1145 | case 'y': |
1146 | { |
1147 | int i, j; |
1148 | for (i = 0; pattern_space[i]; i++) { |
1149 | for (j = 0; sed_cmd->string[j]; j += 2) { |
1150 | if (pattern_space[i] == sed_cmd->string[j]) { |
1151 | pattern_space[i] = sed_cmd->string[j + 1]; |
1152 | break; |
1153 | } |
1154 | } |
1155 | } |
1156 | |
1157 | break; |
1158 | } |
1159 | case 'g': /* Replace pattern space with hold space */ |
1160 | free(pattern_space); |
1161 | pattern_space = xstrdup(G.hold_space ? G.hold_space : ""); |
1162 | break; |
1163 | case 'G': /* Append newline and hold space to pattern space */ |
1164 | { |
1165 | int pattern_space_size = 2; |
1166 | int hold_space_size = 0; |
1167 | |
1168 | if (pattern_space) |
1169 | pattern_space_size += strlen(pattern_space); |
1170 | if (G.hold_space) |
1171 | hold_space_size = strlen(G.hold_space); |
1172 | pattern_space = xrealloc(pattern_space, |
1173 | pattern_space_size + hold_space_size); |
1174 | if (pattern_space_size == 2) |
1175 | pattern_space[0] = 0; |
1176 | strcat(pattern_space, "\n"); |
1177 | if (G.hold_space) |
1178 | strcat(pattern_space, G.hold_space); |
1179 | last_gets_char = '\n'; |
1180 | |
1181 | break; |
1182 | } |
1183 | case 'h': /* Replace hold space with pattern space */ |
1184 | free(G.hold_space); |
1185 | G.hold_space = xstrdup(pattern_space); |
1186 | break; |
1187 | case 'H': /* Append newline and pattern space to hold space */ |
1188 | { |
1189 | int hold_space_size = 2; |
1190 | int pattern_space_size = 0; |
1191 | |
1192 | if (G.hold_space) |
1193 | hold_space_size += strlen(G.hold_space); |
1194 | if (pattern_space) |
1195 | pattern_space_size = strlen(pattern_space); |
1196 | G.hold_space = xrealloc(G.hold_space, |
1197 | hold_space_size + pattern_space_size); |
1198 | |
1199 | if (hold_space_size == 2) |
1200 | *G.hold_space = 0; |
1201 | strcat(G.hold_space, "\n"); |
1202 | if (pattern_space) |
1203 | strcat(G.hold_space, pattern_space); |
1204 | |
1205 | break; |
1206 | } |
1207 | case 'x': /* Exchange hold and pattern space */ |
1208 | { |
1209 | char *tmp = pattern_space; |
1210 | pattern_space = G.hold_space ? G.hold_space : xzalloc(1); |
1211 | last_gets_char = '\n'; |
1212 | G.hold_space = tmp; |
1213 | break; |
1214 | } |
1215 | } /* switch */ |
1216 | } /* for each cmd */ |
1217 | |
1218 | /* |
1219 | * Exit point from sedding... |
1220 | */ |
1221 | discard_commands: |
1222 | /* we will print the line unless we were told to be quiet ('-n') |
1223 | or if the line was suppressed (ala 'd'elete) */ |
1224 | if (!G.be_quiet) |
1225 | sed_puts(pattern_space, last_gets_char); |
1226 | |
1227 | /* Delete and such jump here. */ |
1228 | discard_line: |
1229 | flush_append(); |
1230 | free(pattern_space); |
1231 | |
1232 | goto again; |
1233 | } |
1234 | |
1235 | /* It is possible to have a command line argument with embedded |
1236 | * newlines. This counts as multiple command lines. |
1237 | * However, newline can be escaped: 's/e/z\<newline>z/' |
1238 | * We check for this. |
1239 | */ |
1240 | |
1241 | static void add_cmd_block(char *cmdstr) |
1242 | { |
1243 | char *sv, *eol; |
1244 | |
1245 | cmdstr = sv = xstrdup(cmdstr); |
1246 | do { |
1247 | eol = strchr(cmdstr, '\n'); |
1248 | next: |
1249 | if (eol) { |
1250 | /* Count preceding slashes */ |
1251 | int slashes = 0; |
1252 | char *sl = eol; |
1253 | |
1254 | while (sl != cmdstr && *--sl == '\\') |
1255 | slashes++; |
1256 | /* Odd number of preceding slashes - newline is escaped */ |
1257 | if (slashes & 1) { |
1258 | overlapping_strcpy(eol - 1, eol); |
1259 | eol = strchr(eol, '\n'); |
1260 | goto next; |
1261 | } |
1262 | *eol = '\0'; |
1263 | } |
1264 | add_cmd(cmdstr); |
1265 | cmdstr = eol + 1; |
1266 | } while (eol); |
1267 | free(sv); |
1268 | } |
1269 | |
1270 | int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
1271 | int sed_main(int argc UNUSED_PARAM, char **argv) |
1272 | { |
1273 | enum { |
1274 | OPT_in_place = 1 << 0, |
1275 | }; |
1276 | unsigned opt; |
1277 | llist_t *opt_e, *opt_f; |
1278 | int status = EXIT_SUCCESS; |
1279 | |
1280 | INIT_G(); |
1281 | |
1282 | /* destroy command strings on exit */ |
1283 | if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff); |
1284 | |
1285 | /* Lie to autoconf when it starts asking stupid questions. */ |
1286 | if (argv[1] && !strcmp(argv[1], "--version")) { |
1287 | puts("This is not GNU sed version 4.0"); |
1288 | return 0; |
1289 | } |
1290 | |
1291 | /* do normal option parsing */ |
1292 | opt_e = opt_f = NULL; |
1293 | opt_complementary = "e::f::" /* can occur multiple times */ |
1294 | "nn"; /* count -n */ |
1295 | opt = getopt32(argv, "irne:f:", &opt_e, &opt_f, |
1296 | &G.be_quiet); /* counter for -n */ |
1297 | //argc -= optind; |
1298 | argv += optind; |
1299 | if (opt & OPT_in_place) { // -i |
1300 | atexit(cleanup_outname); |
1301 | } |
1302 | if (opt & 0x2) G.regex_type |= REG_EXTENDED; // -r |
1303 | //if (opt & 0x4) G.be_quiet++; // -n |
1304 | while (opt_e) { // -e |
1305 | add_cmd_block(llist_pop(&opt_e)); |
1306 | } |
1307 | while (opt_f) { // -f |
1308 | char *line; |
1309 | FILE *cmdfile; |
1310 | cmdfile = xfopen_for_read(llist_pop(&opt_f)); |
1311 | while ((line = xmalloc_fgetline(cmdfile)) != NULL) { |
1312 | add_cmd(line); |
1313 | free(line); |
1314 | } |
1315 | fclose(cmdfile); |
1316 | } |
1317 | /* if we didn't get a pattern from -e or -f, use argv[0] */ |
1318 | if (!(opt & 0x18)) { |
1319 | if (!*argv) |
1320 | bb_show_usage(); |
1321 | add_cmd_block(*argv++); |
1322 | } |
1323 | /* Flush any unfinished commands. */ |
1324 | add_cmd(""); |
1325 | |
1326 | /* By default, we write to stdout */ |
1327 | G.nonstdout = stdout; |
1328 | |
1329 | /* argv[0..(argc-1)] should be names of file to process. If no |
1330 | * files were specified or '-' was specified, take input from stdin. |
1331 | * Otherwise, we process all the files specified. */ |
1332 | if (argv[0] == NULL) { |
1333 | if (opt & OPT_in_place) |
1334 | bb_error_msg_and_die(bb_msg_requires_arg, "-i"); |
1335 | add_input_file(stdin); |
1336 | } else { |
1337 | int i; |
1338 | FILE *file; |
1339 | |
1340 | for (i = 0; argv[i]; i++) { |
1341 | struct stat statbuf; |
1342 | int nonstdoutfd; |
1343 | |
1344 | if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) { |
1345 | add_input_file(stdin); |
1346 | process_files(); |
1347 | continue; |
1348 | } |
1349 | file = fopen_or_warn(argv[i], "r"); |
1350 | if (!file) { |
1351 | status = EXIT_FAILURE; |
1352 | continue; |
1353 | } |
1354 | if (!(opt & OPT_in_place)) { |
1355 | add_input_file(file); |
1356 | continue; |
1357 | } |
1358 | |
1359 | G.outname = xasprintf("%sXXXXXX", argv[i]); |
1360 | nonstdoutfd = mkstemp(G.outname); |
1361 | if (-1 == nonstdoutfd) |
1362 | bb_perror_msg_and_die("can't create temp file %s", G.outname); |
1363 | G.nonstdout = xfdopen_for_write(nonstdoutfd); |
1364 | |
1365 | /* Set permissions/owner of output file */ |
1366 | fstat(fileno(file), &statbuf); |
1367 | /* chmod'ing AFTER chown would preserve suid/sgid bits, |
1368 | * but GNU sed 4.2.1 does not preserve them either */ |
1369 | fchmod(nonstdoutfd, statbuf.st_mode); |
1370 | fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid); |
1371 | add_input_file(file); |
1372 | process_files(); |
1373 | fclose(G.nonstdout); |
1374 | |
1375 | G.nonstdout = stdout; |
1376 | /* unlink(argv[i]); */ |
1377 | xrename(G.outname, argv[i]); |
1378 | free(G.outname); |
1379 | G.outname = NULL; |
1380 | } |
1381 | /* Here, to handle "sed 'cmds' nonexistent_file" case we did: |
1382 | * if (G.current_input_file >= G.input_file_count) |
1383 | * return status; |
1384 | * but it's not needed since process_files() works correctly |
1385 | * in this case too. */ |
1386 | } |
1387 | process_files(); |
1388 | |
1389 | return status; |
1390 | } |