blob: e18e48ab5ba0fc0265ee688456e5da9569de9186
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 GPLv2, see file LICENSE in this source tree. |
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 | * process_files() does actual sedding, reading data lines from each input FILE* |
27 | * (which could be stdin) and applying the sed command list (sed_cmd_head) to |
28 | * each of the resulting lines. |
29 | * |
30 | * sed_main() is where external code calls into this, with a command line. |
31 | */ |
32 | |
33 | /* Supported features and commands in this version of sed: |
34 | * |
35 | * - comments ('#') |
36 | * - address matching: num|/matchstr/[,num|/matchstr/|$]command |
37 | * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags) |
38 | * - edit commands: (a)ppend, (i)nsert, (c)hange |
39 | * - file commands: (r)ead |
40 | * - backreferences in substitution expressions (\0, \1, \2...\9) |
41 | * - grouped commands: {cmd1;cmd2} |
42 | * - transliteration (y/source-chars/dest-chars/) |
43 | * - pattern space hold space storing / swapping (g, h, x) |
44 | * - labels / branching (: label, b, t, T) |
45 | * |
46 | * (Note: Specifying an address (range) to match is *optional*; commands |
47 | * default to the whole pattern space if no specific address match was |
48 | * requested.) |
49 | * |
50 | * Todo: |
51 | * - Create a wrapper around regex to make libc's regex conform with sed |
52 | * |
53 | * Reference |
54 | * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html |
55 | * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html |
56 | */ |
57 | |
58 | //config:config SED |
59 | //config: bool "sed" |
60 | //config: default y |
61 | //config: help |
62 | //config: sed is used to perform text transformations on a file |
63 | //config: or input from a pipeline. |
64 | |
65 | //kbuild:lib-$(CONFIG_SED) += sed.o |
66 | |
67 | //applet:IF_SED(APPLET(sed, BB_DIR_BIN, BB_SUID_DROP)) |
68 | |
69 | //usage:#define sed_trivial_usage |
70 | //usage: "[-inrE] [-f FILE]... [-e CMD]... [FILE]...\n" |
71 | //usage: "or: sed [-inrE] CMD [FILE]..." |
72 | //usage:#define sed_full_usage "\n\n" |
73 | //usage: " -e CMD Add CMD to sed commands to be executed" |
74 | //usage: "\n -f FILE Add FILE contents to sed commands to be executed" |
75 | //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)" |
76 | //usage: "\n Optionally back files up, appending SFX" |
77 | //usage: "\n -n Suppress automatic printing of pattern space" |
78 | //usage: "\n -r,-E Use extended regex syntax" |
79 | //usage: "\n" |
80 | //usage: "\nIf no -e or -f, the first non-option argument is the sed command string." |
81 | //usage: "\nRemaining arguments are input files (stdin if none)." |
82 | //usage: |
83 | //usage:#define sed_example_usage |
84 | //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n" |
85 | //usage: "bar\n" |
86 | |
87 | #include "libbb.h" |
88 | #include "xregex.h" |
89 | |
90 | #if 0 |
91 | # define dbg(...) bb_error_msg(__VA_ARGS__) |
92 | #else |
93 | # define dbg(...) ((void)0) |
94 | #endif |
95 | |
96 | |
97 | enum { |
98 | OPT_in_place = 1 << 0, |
99 | }; |
100 | |
101 | /* Each sed command turns into one of these structures. */ |
102 | typedef struct sed_cmd_s { |
103 | /* Ordered by alignment requirements: currently 36 bytes on x86 */ |
104 | struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */ |
105 | |
106 | /* address storage */ |
107 | regex_t *beg_match; /* sed -e '/match/cmd' */ |
108 | regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */ |
109 | regex_t *sub_match; /* For 's/sub_match/string/' */ |
110 | int beg_line; /* 'sed 1p' 0 == apply commands to all lines */ |
111 | int beg_line_orig; /* copy of the above, needed for -i */ |
112 | int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */ |
113 | |
114 | FILE *sw_file; /* File (sw) command writes to, -1 for none. */ |
115 | char *string; /* Data string for (saicytb) commands. */ |
116 | |
117 | unsigned which_match; /* (s) Which match to replace (0 for all) */ |
118 | |
119 | /* Bitfields (gcc won't group them if we don't) */ |
120 | unsigned invert:1; /* the '!' after the address */ |
121 | unsigned in_match:1; /* Next line also included in match? */ |
122 | unsigned sub_p:1; /* (s) print option */ |
123 | |
124 | char sw_last_char; /* Last line written by (sw) had no '\n' */ |
125 | |
126 | /* GENERAL FIELDS */ |
127 | char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */ |
128 | } sed_cmd_t; |
129 | |
130 | static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v"; |
131 | |
132 | struct globals { |
133 | /* options */ |
134 | int be_quiet, regex_type; |
135 | |
136 | FILE *nonstdout; |
137 | char *outname, *hold_space; |
138 | smallint exitcode; |
139 | |
140 | /* list of input files */ |
141 | int current_input_file, last_input_file; |
142 | char **input_file_list; |
143 | FILE *current_fp; |
144 | |
145 | regmatch_t regmatch[10]; |
146 | regex_t *previous_regex_ptr; |
147 | |
148 | /* linked list of sed commands */ |
149 | sed_cmd_t *sed_cmd_head, **sed_cmd_tail; |
150 | |
151 | /* linked list of append lines */ |
152 | llist_t *append_head; |
153 | |
154 | char *add_cmd_line; |
155 | |
156 | struct pipeline { |
157 | char *buf; /* Space to hold string */ |
158 | int idx; /* Space used */ |
159 | int len; /* Space allocated */ |
160 | } pipeline; |
161 | } FIX_ALIASING; |
162 | #define G (*(struct globals*)&bb_common_bufsiz1) |
163 | struct BUG_G_too_big { |
164 | char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1]; |
165 | }; |
166 | #define INIT_G() do { \ |
167 | G.sed_cmd_tail = &G.sed_cmd_head; \ |
168 | } while (0) |
169 | |
170 | |
171 | #if ENABLE_FEATURE_CLEAN_UP |
172 | static void sed_free_and_close_stuff(void) |
173 | { |
174 | sed_cmd_t *sed_cmd = G.sed_cmd_head; |
175 | |
176 | llist_free(G.append_head, free); |
177 | |
178 | while (sed_cmd) { |
179 | sed_cmd_t *sed_cmd_next = sed_cmd->next; |
180 | |
181 | if (sed_cmd->sw_file) |
182 | xprint_and_close_file(sed_cmd->sw_file); |
183 | |
184 | if (sed_cmd->beg_match) { |
185 | regfree(sed_cmd->beg_match); |
186 | free(sed_cmd->beg_match); |
187 | } |
188 | if (sed_cmd->end_match) { |
189 | regfree(sed_cmd->end_match); |
190 | free(sed_cmd->end_match); |
191 | } |
192 | if (sed_cmd->sub_match) { |
193 | regfree(sed_cmd->sub_match); |
194 | free(sed_cmd->sub_match); |
195 | } |
196 | free(sed_cmd->string); |
197 | free(sed_cmd); |
198 | sed_cmd = sed_cmd_next; |
199 | } |
200 | |
201 | free(G.hold_space); |
202 | |
203 | if (G.current_fp) |
204 | fclose(G.current_fp); |
205 | } |
206 | #else |
207 | void sed_free_and_close_stuff(void); |
208 | #endif |
209 | |
210 | /* If something bad happens during -i operation, delete temp file */ |
211 | |
212 | static void cleanup_outname(void) |
213 | { |
214 | if (G.outname) unlink(G.outname); |
215 | } |
216 | |
217 | /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */ |
218 | |
219 | static void parse_escapes(char *dest, const char *string, int len, char from, char to) |
220 | { |
221 | int i = 0; |
222 | |
223 | while (i < len) { |
224 | if (string[i] == '\\') { |
225 | if (!to || string[i+1] == from) { |
226 | *dest++ = to ? to : string[i+1]; |
227 | i += 2; |
228 | continue; |
229 | } |
230 | *dest++ = string[i++]; |
231 | } |
232 | /* TODO: is it safe wrt a string with trailing '\\' ? */ |
233 | *dest++ = string[i++]; |
234 | } |
235 | *dest = '\0'; |
236 | } |
237 | |
238 | static char *copy_parsing_escapes(const char *string, int len) |
239 | { |
240 | const char *s; |
241 | char *dest = xmalloc(len + 1); |
242 | |
243 | /* sed recognizes \n */ |
244 | /* GNU sed also recognizes \t and \r */ |
245 | for (s = "\nn\tt\rr"; *s; s += 2) { |
246 | parse_escapes(dest, string, len, s[1], s[0]); |
247 | string = dest; |
248 | len = strlen(dest); |
249 | } |
250 | return dest; |
251 | } |
252 | |
253 | |
254 | /* |
255 | * index_of_next_unescaped_regexp_delim - walks left to right through a string |
256 | * beginning at a specified index and returns the index of the next regular |
257 | * expression delimiter (typically a forward slash ('/')) not preceded by |
258 | * a backslash ('\'). A negative delimiter disables square bracket checking. |
259 | */ |
260 | static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str) |
261 | { |
262 | int bracket = -1; |
263 | int escaped = 0; |
264 | int idx = 0; |
265 | char ch; |
266 | |
267 | if (delimiter < 0) { |
268 | bracket--; |
269 | delimiter = -delimiter; |
270 | } |
271 | |
272 | for (; (ch = str[idx]) != '\0'; idx++) { |
273 | if (bracket >= 0) { |
274 | if (ch == ']' |
275 | && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^')) |
276 | ) { |
277 | bracket = -1; |
278 | } |
279 | } else if (escaped) |
280 | escaped = 0; |
281 | else if (ch == '\\') |
282 | escaped = 1; |
283 | else if (bracket == -1 && ch == '[') |
284 | bracket = idx; |
285 | else if (ch == delimiter) |
286 | return idx; |
287 | } |
288 | |
289 | /* if we make it to here, we've hit the end of the string */ |
290 | bb_error_msg_and_die("unmatched '%c'", delimiter); |
291 | } |
292 | |
293 | /* |
294 | * Returns the index of the third delimiter |
295 | */ |
296 | static int parse_regex_delim(const char *cmdstr, char **match, char **replace) |
297 | { |
298 | const char *cmdstr_ptr = cmdstr; |
299 | unsigned char delimiter; |
300 | int idx = 0; |
301 | |
302 | /* verify that the 's' or 'y' is followed by something. That something |
303 | * (typically a 'slash') is now our regexp delimiter... */ |
304 | if (*cmdstr == '\0') |
305 | bb_error_msg_and_die("bad format in substitution expression"); |
306 | delimiter = *cmdstr_ptr++; |
307 | |
308 | /* save the match string */ |
309 | idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr); |
310 | *match = copy_parsing_escapes(cmdstr_ptr, idx); |
311 | |
312 | /* save the replacement string */ |
313 | cmdstr_ptr += idx + 1; |
314 | idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr); |
315 | *replace = copy_parsing_escapes(cmdstr_ptr, idx); |
316 | |
317 | return ((cmdstr_ptr - cmdstr) + idx); |
318 | } |
319 | |
320 | /* |
321 | * returns the index in the string just past where the address ends. |
322 | */ |
323 | static int get_address(const char *my_str, int *linenum, regex_t ** regex) |
324 | { |
325 | const char *pos = my_str; |
326 | |
327 | if (isdigit(*my_str)) { |
328 | *linenum = strtol(my_str, (char**)&pos, 10); |
329 | /* endstr shouldnt ever equal NULL */ |
330 | } else if (*my_str == '$') { |
331 | *linenum = -1; |
332 | pos++; |
333 | } else if (*my_str == '/' || *my_str == '\\') { |
334 | int next; |
335 | char delimiter; |
336 | char *temp; |
337 | |
338 | delimiter = '/'; |
339 | if (*my_str == '\\') |
340 | delimiter = *++pos; |
341 | next = index_of_next_unescaped_regexp_delim(delimiter, ++pos); |
342 | temp = copy_parsing_escapes(pos, next); |
343 | *regex = xzalloc(sizeof(regex_t)); |
344 | xregcomp(*regex, temp, G.regex_type); |
345 | free(temp); |
346 | /* Move position to next character after last delimiter */ |
347 | pos += (next+1); |
348 | } |
349 | return pos - my_str; |
350 | } |
351 | |
352 | /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */ |
353 | static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval) |
354 | { |
355 | int start = 0, idx, hack = 0; |
356 | |
357 | /* Skip whitespace, then grab filename to end of line */ |
358 | while (isspace(filecmdstr[start])) |
359 | start++; |
360 | idx = start; |
361 | while (filecmdstr[idx] && filecmdstr[idx] != '\n') |
362 | idx++; |
363 | |
364 | /* If lines glued together, put backslash back. */ |
365 | if (filecmdstr[idx] == '\n') |
366 | hack = 1; |
367 | if (idx == start) |
368 | bb_error_msg_and_die("empty filename"); |
369 | *retval = xstrndup(filecmdstr+start, idx-start+hack+1); |
370 | if (hack) |
371 | (*retval)[idx] = '\\'; |
372 | |
373 | return idx; |
374 | } |
375 | |
376 | static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr) |
377 | { |
378 | int cflags = G.regex_type; |
379 | char *match; |
380 | int idx; |
381 | |
382 | /* |
383 | * A substitution command should look something like this: |
384 | * s/match/replace/ #giIpw |
385 | * || | ||| |
386 | * mandatory optional |
387 | */ |
388 | idx = parse_regex_delim(substr, &match, &sed_cmd->string); |
389 | |
390 | /* determine the number of back references in the match string */ |
391 | /* Note: we compute this here rather than in the do_subst_command() |
392 | * function to save processor time, at the expense of a little more memory |
393 | * (4 bits) per sed_cmd */ |
394 | |
395 | /* process the flags */ |
396 | |
397 | sed_cmd->which_match = 1; |
398 | while (substr[++idx]) { |
399 | /* Parse match number */ |
400 | if (isdigit(substr[idx])) { |
401 | if (match[0] != '^') { |
402 | /* Match 0 treated as all, multiple matches we take the last one. */ |
403 | const char *pos = substr + idx; |
404 | /* FIXME: error check? */ |
405 | sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10); |
406 | idx = pos - substr; |
407 | } |
408 | continue; |
409 | } |
410 | /* Skip spaces */ |
411 | if (isspace(substr[idx])) |
412 | continue; |
413 | |
414 | switch (substr[idx]) { |
415 | /* Replace all occurrences */ |
416 | case 'g': |
417 | if (match[0] != '^') |
418 | sed_cmd->which_match = 0; |
419 | break; |
420 | /* Print pattern space */ |
421 | case 'p': |
422 | sed_cmd->sub_p = 1; |
423 | break; |
424 | /* Write to file */ |
425 | case 'w': |
426 | { |
427 | char *temp; |
428 | idx += parse_file_cmd(/*sed_cmd,*/ substr+idx, &temp); |
429 | break; |
430 | } |
431 | /* Ignore case (gnu exension) */ |
432 | case 'i': |
433 | case 'I': |
434 | cflags |= REG_ICASE; |
435 | break; |
436 | /* Comment */ |
437 | case '#': |
438 | // while (substr[++idx]) continue; |
439 | idx += strlen(substr + idx); // same |
440 | /* Fall through */ |
441 | /* End of command */ |
442 | case ';': |
443 | case '}': |
444 | goto out; |
445 | default: |
446 | bb_error_msg_and_die("bad option in substitution expression"); |
447 | } |
448 | } |
449 | out: |
450 | /* compile the match string into a regex */ |
451 | if (*match != '\0') { |
452 | /* If match is empty, we use last regex used at runtime */ |
453 | sed_cmd->sub_match = xzalloc(sizeof(regex_t)); |
454 | dbg("xregcomp('%s',%x)", match, cflags); |
455 | xregcomp(sed_cmd->sub_match, match, cflags); |
456 | dbg("regcomp ok"); |
457 | } |
458 | free(match); |
459 | |
460 | return idx; |
461 | } |
462 | |
463 | /* |
464 | * Process the commands arguments |
465 | */ |
466 | static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr) |
467 | { |
468 | static const char cmd_letters[] = "saicrw:btTydDgGhHlnNpPqx={}"; |
469 | enum { |
470 | IDX_s = 0, |
471 | IDX_a, |
472 | IDX_i, |
473 | IDX_c, |
474 | IDX_r, |
475 | IDX_w, |
476 | IDX_colon, |
477 | IDX_b, |
478 | IDX_t, |
479 | IDX_T, |
480 | IDX_y, |
481 | IDX_d, |
482 | IDX_D, |
483 | IDX_g, |
484 | IDX_G, |
485 | IDX_h, |
486 | IDX_H, |
487 | IDX_l, |
488 | IDX_n, |
489 | IDX_N, |
490 | IDX_p, |
491 | IDX_P, |
492 | IDX_q, |
493 | IDX_x, |
494 | IDX_equal, |
495 | IDX_lbrace, |
496 | IDX_rbrace, |
497 | IDX_nul |
498 | }; |
499 | struct chk { char chk[sizeof(cmd_letters)-1 == IDX_nul ? 1 : -1]; }; |
500 | |
501 | unsigned idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters; |
502 | |
503 | /* handle (s)ubstitution command */ |
504 | if (idx == IDX_s) { |
505 | cmdstr += parse_subst_cmd(sed_cmd, cmdstr); |
506 | } |
507 | /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */ |
508 | else if (idx <= IDX_c) { /* a,i,c */ |
509 | if (idx < IDX_c) { /* a,i */ |
510 | if (sed_cmd->end_line || sed_cmd->end_match) |
511 | bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd); |
512 | } |
513 | for (;;) { |
514 | if (*cmdstr == '\n' || *cmdstr == '\\') { |
515 | cmdstr++; |
516 | break; |
517 | } |
518 | if (!isspace(*cmdstr)) |
519 | break; |
520 | cmdstr++; |
521 | } |
522 | sed_cmd->string = xstrdup(cmdstr); |
523 | /* "\anychar" -> "anychar" */ |
524 | parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), '\0', '\0'); |
525 | cmdstr += strlen(cmdstr); |
526 | } |
527 | /* handle file cmds: (r)ead */ |
528 | else if (idx <= IDX_w) { /* r,w */ |
529 | if (idx < IDX_w) { /* r */ |
530 | if (sed_cmd->end_line || sed_cmd->end_match) |
531 | bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd); |
532 | } |
533 | cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string); |
534 | if (sed_cmd->cmd == 'w') { |
535 | sed_cmd->sw_file = xfopen_for_write(sed_cmd->string); |
536 | sed_cmd->sw_last_char = '\n'; |
537 | } |
538 | } |
539 | /* handle branch commands */ |
540 | else if (idx <= IDX_T) { /* :,b,t,T */ |
541 | int length; |
542 | |
543 | cmdstr = skip_whitespace(cmdstr); |
544 | length = strcspn(cmdstr, semicolon_whitespace); |
545 | if (length) { |
546 | sed_cmd->string = xstrndup(cmdstr, length); |
547 | cmdstr += length; |
548 | } |
549 | } |
550 | /* translation command */ |
551 | else if (idx == IDX_y) { |
552 | char *match, *replace; |
553 | int i = cmdstr[0]; |
554 | |
555 | cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1; |
556 | /* \n already parsed, but \delimiter needs unescaping. */ |
557 | parse_escapes(match, match, strlen(match), i, i); |
558 | parse_escapes(replace, replace, strlen(replace), i, i); |
559 | |
560 | sed_cmd->string = xzalloc((strlen(match) + 1) * 2); |
561 | for (i = 0; match[i] && replace[i]; i++) { |
562 | sed_cmd->string[i*2] = match[i]; |
563 | sed_cmd->string[i*2+1] = replace[i]; |
564 | } |
565 | free(match); |
566 | free(replace); |
567 | } |
568 | /* if it wasnt a single-letter command that takes no arguments |
569 | * then it must be an invalid command. |
570 | */ |
571 | else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */ |
572 | bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd); |
573 | } |
574 | |
575 | /* give back whatever's left over */ |
576 | return cmdstr; |
577 | } |
578 | |
579 | |
580 | /* Parse address+command sets, skipping comment lines. */ |
581 | |
582 | static void add_cmd(const char *cmdstr) |
583 | { |
584 | sed_cmd_t *sed_cmd; |
585 | unsigned len, n; |
586 | |
587 | /* Append this line to any unfinished line from last time. */ |
588 | if (G.add_cmd_line) { |
589 | char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr); |
590 | free(G.add_cmd_line); |
591 | cmdstr = G.add_cmd_line = tp; |
592 | } |
593 | |
594 | /* If this line ends with unescaped backslash, request next line. */ |
595 | n = len = strlen(cmdstr); |
596 | while (n && cmdstr[n-1] == '\\') |
597 | n--; |
598 | if ((len - n) & 1) { /* if odd number of trailing backslashes */ |
599 | if (!G.add_cmd_line) |
600 | G.add_cmd_line = xstrdup(cmdstr); |
601 | G.add_cmd_line[len-1] = '\0'; |
602 | return; |
603 | } |
604 | |
605 | /* Loop parsing all commands in this line. */ |
606 | while (*cmdstr) { |
607 | /* Skip leading whitespace and semicolons */ |
608 | cmdstr += strspn(cmdstr, semicolon_whitespace); |
609 | |
610 | /* If no more commands, exit. */ |
611 | if (!*cmdstr) break; |
612 | |
613 | /* if this is a comment, jump past it and keep going */ |
614 | if (*cmdstr == '#') { |
615 | /* "#n" is the same as using -n on the command line */ |
616 | if (cmdstr[1] == 'n') |
617 | G.be_quiet++; |
618 | cmdstr = strpbrk(cmdstr, "\n\r"); |
619 | if (!cmdstr) break; |
620 | continue; |
621 | } |
622 | |
623 | /* parse the command |
624 | * format is: [addr][,addr][!]cmd |
625 | * |----||-----||-| |
626 | * part1 part2 part3 |
627 | */ |
628 | |
629 | sed_cmd = xzalloc(sizeof(sed_cmd_t)); |
630 | |
631 | /* first part (if present) is an address: either a '$', a number or a /regex/ */ |
632 | cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match); |
633 | sed_cmd->beg_line_orig = sed_cmd->beg_line; |
634 | |
635 | /* second part (if present) will begin with a comma */ |
636 | if (*cmdstr == ',') { |
637 | int idx; |
638 | |
639 | cmdstr++; |
640 | idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match); |
641 | if (!idx) |
642 | bb_error_msg_and_die("no address after comma"); |
643 | cmdstr += idx; |
644 | } |
645 | |
646 | /* skip whitespace before the command */ |
647 | cmdstr = skip_whitespace(cmdstr); |
648 | |
649 | /* Check for inversion flag */ |
650 | if (*cmdstr == '!') { |
651 | sed_cmd->invert = 1; |
652 | cmdstr++; |
653 | |
654 | /* skip whitespace before the command */ |
655 | cmdstr = skip_whitespace(cmdstr); |
656 | } |
657 | |
658 | /* last part (mandatory) will be a command */ |
659 | if (!*cmdstr) |
660 | bb_error_msg_and_die("missing command"); |
661 | sed_cmd->cmd = *cmdstr++; |
662 | cmdstr = parse_cmd_args(sed_cmd, cmdstr); |
663 | |
664 | /* cmdstr now points past args. |
665 | * GNU sed requires a separator, if there are more commands, |
666 | * else it complains "char N: extra characters after command". |
667 | * Example: "sed 'p;d'". We also allow "sed 'pd'". |
668 | */ |
669 | |
670 | /* Add the command to the command array */ |
671 | *G.sed_cmd_tail = sed_cmd; |
672 | G.sed_cmd_tail = &sed_cmd->next; |
673 | } |
674 | |
675 | /* If we glued multiple lines together, free the memory. */ |
676 | free(G.add_cmd_line); |
677 | G.add_cmd_line = NULL; |
678 | } |
679 | |
680 | /* Append to a string, reallocating memory as necessary. */ |
681 | |
682 | #define PIPE_GROW 64 |
683 | |
684 | static void pipe_putc(char c) |
685 | { |
686 | if (G.pipeline.idx == G.pipeline.len) { |
687 | G.pipeline.buf = xrealloc(G.pipeline.buf, |
688 | G.pipeline.len + PIPE_GROW); |
689 | G.pipeline.len += PIPE_GROW; |
690 | } |
691 | G.pipeline.buf[G.pipeline.idx++] = c; |
692 | } |
693 | |
694 | static void do_subst_w_backrefs(char *line, char *replace) |
695 | { |
696 | int i, j; |
697 | |
698 | /* go through the replacement string */ |
699 | for (i = 0; replace[i]; i++) { |
700 | /* if we find a backreference (\1, \2, etc.) print the backref'ed text */ |
701 | if (replace[i] == '\\') { |
702 | unsigned backref = replace[++i] - '0'; |
703 | if (backref <= 9) { |
704 | /* print out the text held in G.regmatch[backref] */ |
705 | if (G.regmatch[backref].rm_so != -1) { |
706 | j = G.regmatch[backref].rm_so; |
707 | while (j < G.regmatch[backref].rm_eo) |
708 | pipe_putc(line[j++]); |
709 | } |
710 | continue; |
711 | } |
712 | /* I _think_ it is impossible to get '\' to be |
713 | * the last char in replace string. Thus we dont check |
714 | * for replace[i] == NUL. (counterexample anyone?) */ |
715 | /* if we find a backslash escaped character, print the character */ |
716 | pipe_putc(replace[i]); |
717 | continue; |
718 | } |
719 | /* if we find an unescaped '&' print out the whole matched text. */ |
720 | if (replace[i] == '&') { |
721 | j = G.regmatch[0].rm_so; |
722 | while (j < G.regmatch[0].rm_eo) |
723 | pipe_putc(line[j++]); |
724 | continue; |
725 | } |
726 | /* Otherwise just output the character. */ |
727 | pipe_putc(replace[i]); |
728 | } |
729 | } |
730 | |
731 | static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p) |
732 | { |
733 | char *line = *line_p; |
734 | unsigned match_count = 0; |
735 | bool altered = 0; |
736 | bool prev_match_empty = 1; |
737 | bool tried_at_eol = 0; |
738 | regex_t *current_regex; |
739 | |
740 | current_regex = sed_cmd->sub_match; |
741 | /* Handle empty regex. */ |
742 | if (!current_regex) { |
743 | current_regex = G.previous_regex_ptr; |
744 | if (!current_regex) |
745 | bb_error_msg_and_die("no previous regexp"); |
746 | } |
747 | G.previous_regex_ptr = current_regex; |
748 | |
749 | /* Find the first match */ |
750 | dbg("matching '%s'", line); |
751 | if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) { |
752 | dbg("no match"); |
753 | return 0; |
754 | } |
755 | dbg("match"); |
756 | |
757 | /* Initialize temporary output buffer. */ |
758 | G.pipeline.buf = xmalloc(PIPE_GROW); |
759 | G.pipeline.len = PIPE_GROW; |
760 | G.pipeline.idx = 0; |
761 | |
762 | /* Now loop through, substituting for matches */ |
763 | do { |
764 | int start = G.regmatch[0].rm_so; |
765 | int end = G.regmatch[0].rm_eo; |
766 | int i; |
767 | |
768 | match_count++; |
769 | |
770 | /* If we aren't interested in this match, output old line to |
771 | * end of match and continue */ |
772 | if (sed_cmd->which_match |
773 | && (sed_cmd->which_match != match_count) |
774 | ) { |
775 | for (i = 0; i < end; i++) |
776 | pipe_putc(*line++); |
777 | /* Null match? Print one more char */ |
778 | if (start == end && *line) |
779 | pipe_putc(*line++); |
780 | goto next; |
781 | } |
782 | |
783 | /* Print everything before the match */ |
784 | for (i = 0; i < start; i++) |
785 | pipe_putc(line[i]); |
786 | |
787 | /* Then print the substitution string, |
788 | * unless we just matched empty string after non-empty one. |
789 | * Example: string "cccd", pattern "c*", repl "R": |
790 | * result is "RdR", not "RRdR": first match "ccc", |
791 | * second is "" before "d", third is "" after "d". |
792 | * Second match is NOT replaced! |
793 | */ |
794 | if (prev_match_empty || start != 0 || start != end) { |
795 | //dbg("%d %d %d", prev_match_empty, start, end); |
796 | dbg("inserting replacement at %d in '%s'", start, line); |
797 | do_subst_w_backrefs(line, sed_cmd->string); |
798 | /* Flag that something has changed */ |
799 | altered = 1; |
800 | } else { |
801 | dbg("NOT inserting replacement at %d in '%s'", start, line); |
802 | } |
803 | |
804 | /* If matched string is empty (f.e. "c*" pattern), |
805 | * copy verbatim one char after it before attempting more matches |
806 | */ |
807 | prev_match_empty = (start == end); |
808 | if (prev_match_empty) { |
809 | if (!line[end]) { |
810 | tried_at_eol = 1; |
811 | } else { |
812 | pipe_putc(line[end]); |
813 | end++; |
814 | } |
815 | } |
816 | |
817 | /* Advance past the match */ |
818 | dbg("line += %d", end); |
819 | line += end; |
820 | |
821 | /* if we're not doing this globally, get out now */ |
822 | if (sed_cmd->which_match != 0) |
823 | break; |
824 | next: |
825 | /* Exit if we are at EOL and already tried matching at it */ |
826 | if (*line == '\0') { |
827 | if (tried_at_eol) |
828 | break; |
829 | tried_at_eol = 1; |
830 | } |
831 | |
832 | //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL? |
833 | } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH); |
834 | |
835 | /* Copy rest of string into output pipeline */ |
836 | while (1) { |
837 | char c = *line++; |
838 | pipe_putc(c); |
839 | if (c == '\0') |
840 | break; |
841 | } |
842 | |
843 | free(*line_p); |
844 | *line_p = G.pipeline.buf; |
845 | return altered; |
846 | } |
847 | |
848 | /* Set command pointer to point to this label. (Does not handle null label.) */ |
849 | static sed_cmd_t *branch_to(char *label) |
850 | { |
851 | sed_cmd_t *sed_cmd; |
852 | |
853 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
854 | if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) { |
855 | return sed_cmd; |
856 | } |
857 | } |
858 | bb_error_msg_and_die("can't find label for jump to '%s'", label); |
859 | } |
860 | |
861 | static void append(char *s) |
862 | { |
863 | llist_add_to_end(&G.append_head, s); |
864 | } |
865 | |
866 | /* Output line of text. */ |
867 | /* Note: |
868 | * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed. |
869 | * Without them, we had this: |
870 | * echo -n thingy >z1 |
871 | * echo -n again >z2 |
872 | * >znull |
873 | * sed "s/i/z/" z1 z2 znull | hexdump -vC |
874 | * output: |
875 | * gnu sed 4.1.5: |
876 | * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn| |
877 | * bbox: |
878 | * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn| |
879 | */ |
880 | enum { |
881 | NO_EOL_CHAR = 1, |
882 | LAST_IS_NUL = 2, |
883 | }; |
884 | static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char) |
885 | { |
886 | char lpc = *last_puts_char; |
887 | |
888 | /* Need to insert a '\n' between two files because first file's |
889 | * last line wasn't terminated? */ |
890 | if (lpc != '\n' && lpc != '\0') { |
891 | fputc('\n', file); |
892 | lpc = '\n'; |
893 | } |
894 | fputs(s, file); |
895 | |
896 | /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */ |
897 | if (s[0]) |
898 | lpc = 'x'; |
899 | |
900 | /* had trailing '\0' and it was last char of file? */ |
901 | if (last_gets_char == LAST_IS_NUL) { |
902 | fputc('\0', file); |
903 | lpc = 'x'; /* */ |
904 | } else |
905 | /* had trailing '\n' or '\0'? */ |
906 | if (last_gets_char != NO_EOL_CHAR) { |
907 | fputc(last_gets_char, file); |
908 | lpc = last_gets_char; |
909 | } |
910 | |
911 | if (ferror(file)) { |
912 | xfunc_error_retval = 4; /* It's what gnu sed exits with... */ |
913 | bb_error_msg_and_die(bb_msg_write_error); |
914 | } |
915 | *last_puts_char = lpc; |
916 | } |
917 | |
918 | static void flush_append(char *last_puts_char, char last_gets_char) |
919 | { |
920 | char *data; |
921 | |
922 | /* Output appended lines. */ |
923 | while ((data = (char *)llist_pop(&G.append_head))) { |
924 | puts_maybe_newline(data, G.nonstdout, last_puts_char, last_gets_char); |
925 | free(data); |
926 | } |
927 | } |
928 | |
929 | /* Get next line of input from G.input_file_list, flushing append buffer and |
930 | * noting if we ran out of files without a newline on the last line we read. |
931 | */ |
932 | static char *get_next_line(char *gets_char, char *last_puts_char, char last_gets_char) |
933 | { |
934 | char *temp = NULL; |
935 | int len; |
936 | char gc; |
937 | |
938 | flush_append(last_puts_char, last_gets_char); |
939 | |
940 | /* will be returned if last line in the file |
941 | * doesn't end with either '\n' or '\0' */ |
942 | gc = NO_EOL_CHAR; |
943 | for (; G.current_input_file <= G.last_input_file; G.current_input_file++) { |
944 | FILE *fp = G.current_fp; |
945 | if (!fp) { |
946 | const char *path = G.input_file_list[G.current_input_file]; |
947 | fp = stdin; |
948 | if (path != bb_msg_standard_input) { |
949 | fp = fopen_or_warn(path, "r"); |
950 | if (!fp) { |
951 | G.exitcode = EXIT_FAILURE; |
952 | continue; |
953 | } |
954 | } |
955 | G.current_fp = fp; |
956 | } |
957 | /* Read line up to a newline or NUL byte, inclusive, |
958 | * return malloc'ed char[]. length of the chunk read |
959 | * is stored in len. NULL if EOF/error */ |
960 | temp = bb_get_chunk_from_file(fp, &len); |
961 | if (temp) { |
962 | /* len > 0 here, it's ok to do temp[len-1] */ |
963 | char c = temp[len-1]; |
964 | if (c == '\n' || c == '\0') { |
965 | temp[len-1] = '\0'; |
966 | gc = c; |
967 | if (c == '\0') { |
968 | int ch = fgetc(fp); |
969 | if (ch != EOF) |
970 | ungetc(ch, fp); |
971 | else |
972 | gc = LAST_IS_NUL; |
973 | } |
974 | } |
975 | /* else we put NO_EOL_CHAR into *gets_char */ |
976 | break; |
977 | |
978 | /* NB: I had the idea of peeking next file(s) and returning |
979 | * NO_EOL_CHAR only if it is the *last* non-empty |
980 | * input file. But there is a case where this won't work: |
981 | * file1: "a woo\nb woo" |
982 | * file2: "c no\nd no" |
983 | * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang" |
984 | * (note: *no* newline after "b bang"!) */ |
985 | } |
986 | /* Close this file and advance to next one */ |
987 | fclose_if_not_stdin(fp); |
988 | G.current_fp = NULL; |
989 | } |
990 | *gets_char = gc; |
991 | return temp; |
992 | } |
993 | |
994 | #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n)) |
995 | |
996 | static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space) |
997 | { |
998 | int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0); |
999 | if (retval) |
1000 | G.previous_regex_ptr = sed_cmd->beg_match; |
1001 | return retval; |
1002 | } |
1003 | |
1004 | /* Process all the lines in all the files */ |
1005 | |
1006 | static void process_files(void) |
1007 | { |
1008 | char *pattern_space, *next_line; |
1009 | int linenum = 0; |
1010 | char last_puts_char = '\n'; |
1011 | char last_gets_char, next_gets_char; |
1012 | sed_cmd_t *sed_cmd; |
1013 | int substituted; |
1014 | |
1015 | /* Prime the pump */ |
1016 | next_line = get_next_line(&next_gets_char, &last_puts_char, '\n' /*last_gets_char*/); |
1017 | |
1018 | /* Go through every line in each file */ |
1019 | again: |
1020 | substituted = 0; |
1021 | |
1022 | /* Advance to next line. Stop if out of lines. */ |
1023 | pattern_space = next_line; |
1024 | if (!pattern_space) |
1025 | return; |
1026 | last_gets_char = next_gets_char; |
1027 | |
1028 | /* Read one line in advance so we can act on the last line, |
1029 | * the '$' address */ |
1030 | next_line = get_next_line(&next_gets_char, &last_puts_char, last_gets_char); |
1031 | linenum++; |
1032 | |
1033 | /* For every line, go through all the commands */ |
1034 | restart: |
1035 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
1036 | int old_matched, matched; |
1037 | |
1038 | old_matched = sed_cmd->in_match; |
1039 | |
1040 | /* Determine if this command matches this line: */ |
1041 | |
1042 | dbg("match1:%d", sed_cmd->in_match); |
1043 | dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line |
1044 | && !sed_cmd->beg_match && !sed_cmd->end_match)); |
1045 | dbg("match3:%d", (sed_cmd->beg_line > 0 |
1046 | && (sed_cmd->end_line || sed_cmd->end_match |
1047 | ? (sed_cmd->beg_line <= linenum) |
1048 | : (sed_cmd->beg_line == linenum) |
1049 | ) |
1050 | )); |
1051 | dbg("match4:%d", (beg_match(sed_cmd, pattern_space))); |
1052 | dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL)); |
1053 | |
1054 | /* Are we continuing a previous multi-line match? */ |
1055 | sed_cmd->in_match = sed_cmd->in_match |
1056 | /* Or is no range necessary? */ |
1057 | || (!sed_cmd->beg_line && !sed_cmd->end_line |
1058 | && !sed_cmd->beg_match && !sed_cmd->end_match) |
1059 | /* Or did we match the start of a numerical range? */ |
1060 | || (sed_cmd->beg_line > 0 |
1061 | && (sed_cmd->end_line || sed_cmd->end_match |
1062 | /* note: even if end is numeric and is < linenum too, |
1063 | * GNU sed matches! We match too, therefore we don't |
1064 | * check here that linenum <= end. |
1065 | * Example: |
1066 | * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p' |
1067 | * first three input lines are deleted; |
1068 | * 4th line is matched and printed |
1069 | * by "2,3" (!) and by "4" ranges |
1070 | */ |
1071 | ? (sed_cmd->beg_line <= linenum) /* N,end */ |
1072 | : (sed_cmd->beg_line == linenum) /* N */ |
1073 | ) |
1074 | ) |
1075 | /* Or does this line match our begin address regex? */ |
1076 | || (beg_match(sed_cmd, pattern_space)) |
1077 | /* Or did we match last line of input? */ |
1078 | || (sed_cmd->beg_line == -1 && next_line == NULL); |
1079 | |
1080 | /* Snapshot the value */ |
1081 | matched = sed_cmd->in_match; |
1082 | |
1083 | dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d", |
1084 | sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum); |
1085 | |
1086 | /* Is this line the end of the current match? */ |
1087 | |
1088 | if (matched) { |
1089 | /* once matched, "n,xxx" range is dead, disabling it */ |
1090 | if (sed_cmd->beg_line > 0) { |
1091 | sed_cmd->beg_line = -2; |
1092 | } |
1093 | sed_cmd->in_match = !( |
1094 | /* has the ending line come, or is this a single address command? */ |
1095 | (sed_cmd->end_line |
1096 | ? sed_cmd->end_line == -1 |
1097 | ? !next_line |
1098 | : (sed_cmd->end_line <= linenum) |
1099 | : !sed_cmd->end_match |
1100 | ) |
1101 | /* or does this line matches our last address regex */ |
1102 | || (sed_cmd->end_match && old_matched |
1103 | && (regexec(sed_cmd->end_match, |
1104 | pattern_space, 0, NULL, 0) == 0) |
1105 | ) |
1106 | ); |
1107 | } |
1108 | |
1109 | /* Skip blocks of commands we didn't match */ |
1110 | if (sed_cmd->cmd == '{') { |
1111 | if (sed_cmd->invert ? matched : !matched) { |
1112 | unsigned nest_cnt = 0; |
1113 | while (1) { |
1114 | if (sed_cmd->cmd == '{') |
1115 | nest_cnt++; |
1116 | if (sed_cmd->cmd == '}') { |
1117 | nest_cnt--; |
1118 | if (nest_cnt == 0) |
1119 | break; |
1120 | } |
1121 | sed_cmd = sed_cmd->next; |
1122 | if (!sed_cmd) |
1123 | bb_error_msg_and_die("unterminated {"); |
1124 | } |
1125 | } |
1126 | continue; |
1127 | } |
1128 | |
1129 | /* Okay, so did this line match? */ |
1130 | if (sed_cmd->invert ? matched : !matched) |
1131 | continue; /* no */ |
1132 | |
1133 | /* Update last used regex in case a blank substitute BRE is found */ |
1134 | if (sed_cmd->beg_match) { |
1135 | G.previous_regex_ptr = sed_cmd->beg_match; |
1136 | } |
1137 | |
1138 | /* actual sedding */ |
1139 | dbg("pattern_space:'%s' next_line:'%s' cmd:%c", |
1140 | pattern_space, next_line, sed_cmd->cmd); |
1141 | switch (sed_cmd->cmd) { |
1142 | |
1143 | /* Print line number */ |
1144 | case '=': |
1145 | fprintf(G.nonstdout, "%d\n", linenum); |
1146 | break; |
1147 | |
1148 | /* Write the current pattern space up to the first newline */ |
1149 | case 'P': |
1150 | { |
1151 | char *tmp = strchr(pattern_space, '\n'); |
1152 | if (tmp) { |
1153 | *tmp = '\0'; |
1154 | /* TODO: explain why '\n' below */ |
1155 | sed_puts(pattern_space, '\n'); |
1156 | *tmp = '\n'; |
1157 | break; |
1158 | } |
1159 | /* Fall Through */ |
1160 | } |
1161 | |
1162 | /* Write the current pattern space to output */ |
1163 | case 'p': |
1164 | /* NB: we print this _before_ the last line |
1165 | * (of current file) is printed. Even if |
1166 | * that line is nonterminated, we print |
1167 | * '\n' here (gnu sed does the same) */ |
1168 | sed_puts(pattern_space, '\n'); |
1169 | break; |
1170 | /* Delete up through first newline */ |
1171 | case 'D': |
1172 | { |
1173 | char *tmp = strchr(pattern_space, '\n'); |
1174 | if (tmp) { |
1175 | overlapping_strcpy(pattern_space, tmp + 1); |
1176 | goto restart; |
1177 | } |
1178 | } |
1179 | /* discard this line. */ |
1180 | case 'd': |
1181 | goto discard_line; |
1182 | |
1183 | /* Substitute with regex */ |
1184 | case 's': |
1185 | if (!do_subst_command(sed_cmd, &pattern_space)) |
1186 | break; |
1187 | dbg("do_subst_command succeeded:'%s'", pattern_space); |
1188 | substituted |= 1; |
1189 | |
1190 | /* handle p option */ |
1191 | if (sed_cmd->sub_p) |
1192 | sed_puts(pattern_space, last_gets_char); |
1193 | /* handle w option */ |
1194 | if (sed_cmd->sw_file) |
1195 | puts_maybe_newline( |
1196 | pattern_space, sed_cmd->sw_file, |
1197 | &sed_cmd->sw_last_char, last_gets_char); |
1198 | break; |
1199 | |
1200 | /* Append line to linked list to be printed later */ |
1201 | case 'a': |
1202 | append(xstrdup(sed_cmd->string)); |
1203 | break; |
1204 | |
1205 | /* Insert text before this line */ |
1206 | case 'i': |
1207 | sed_puts(sed_cmd->string, '\n'); |
1208 | break; |
1209 | |
1210 | /* Cut and paste text (replace) */ |
1211 | case 'c': |
1212 | /* Only triggers on last line of a matching range. */ |
1213 | if (!sed_cmd->in_match) |
1214 | sed_puts(sed_cmd->string, '\n'); |
1215 | goto discard_line; |
1216 | |
1217 | /* Read file, append contents to output */ |
1218 | case 'r': |
1219 | { |
1220 | FILE *rfile; |
1221 | rfile = fopen_for_read(sed_cmd->string); |
1222 | if (rfile) { |
1223 | char *line; |
1224 | while ((line = xmalloc_fgetline(rfile)) |
1225 | != NULL) |
1226 | append(line); |
1227 | fclose(rfile); |
1228 | } |
1229 | |
1230 | break; |
1231 | } |
1232 | |
1233 | /* Write pattern space to file. */ |
1234 | case 'w': |
1235 | puts_maybe_newline( |
1236 | pattern_space, sed_cmd->sw_file, |
1237 | &sed_cmd->sw_last_char, last_gets_char); |
1238 | break; |
1239 | |
1240 | /* Read next line from input */ |
1241 | case 'n': |
1242 | if (!G.be_quiet) |
1243 | sed_puts(pattern_space, last_gets_char); |
1244 | if (next_line) { |
1245 | free(pattern_space); |
1246 | pattern_space = next_line; |
1247 | last_gets_char = next_gets_char; |
1248 | next_line = get_next_line(&next_gets_char, &last_puts_char, last_gets_char); |
1249 | substituted = 0; |
1250 | linenum++; |
1251 | break; |
1252 | } |
1253 | /* fall through */ |
1254 | |
1255 | /* Quit. End of script, end of input. */ |
1256 | case 'q': |
1257 | /* Exit the outer while loop */ |
1258 | free(next_line); |
1259 | next_line = NULL; |
1260 | goto discard_commands; |
1261 | |
1262 | /* Append the next line to the current line */ |
1263 | case 'N': |
1264 | { |
1265 | int len; |
1266 | /* If no next line, jump to end of script and exit. */ |
1267 | /* http://www.gnu.org/software/sed/manual/sed.html: |
1268 | * "Most versions of sed exit without printing anything |
1269 | * when the N command is issued on the last line of |
1270 | * a file. GNU sed prints pattern space before exiting |
1271 | * unless of course the -n command switch has been |
1272 | * specified. This choice is by design." |
1273 | */ |
1274 | if (next_line == NULL) { |
1275 | //goto discard_line; |
1276 | goto discard_commands; /* GNU behavior */ |
1277 | } |
1278 | /* Append next_line, read new next_line. */ |
1279 | len = strlen(pattern_space); |
1280 | pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2); |
1281 | pattern_space[len] = '\n'; |
1282 | strcpy(pattern_space + len+1, next_line); |
1283 | last_gets_char = next_gets_char; |
1284 | next_line = get_next_line(&next_gets_char, &last_puts_char, last_gets_char); |
1285 | linenum++; |
1286 | break; |
1287 | } |
1288 | |
1289 | /* Test/branch if substitution occurred */ |
1290 | case 't': |
1291 | if (!substituted) break; |
1292 | substituted = 0; |
1293 | /* Fall through */ |
1294 | /* Test/branch if substitution didn't occur */ |
1295 | case 'T': |
1296 | if (substituted) break; |
1297 | /* Fall through */ |
1298 | /* Branch to label */ |
1299 | case 'b': |
1300 | if (!sed_cmd->string) goto discard_commands; |
1301 | else sed_cmd = branch_to(sed_cmd->string); |
1302 | break; |
1303 | /* Transliterate characters */ |
1304 | case 'y': |
1305 | { |
1306 | int i, j; |
1307 | for (i = 0; pattern_space[i]; i++) { |
1308 | for (j = 0; sed_cmd->string[j]; j += 2) { |
1309 | if (pattern_space[i] == sed_cmd->string[j]) { |
1310 | pattern_space[i] = sed_cmd->string[j + 1]; |
1311 | break; |
1312 | } |
1313 | } |
1314 | } |
1315 | |
1316 | break; |
1317 | } |
1318 | case 'g': /* Replace pattern space with hold space */ |
1319 | free(pattern_space); |
1320 | pattern_space = xstrdup(G.hold_space ? G.hold_space : ""); |
1321 | break; |
1322 | case 'G': /* Append newline and hold space to pattern space */ |
1323 | { |
1324 | int pattern_space_size = 2; |
1325 | int hold_space_size = 0; |
1326 | |
1327 | if (pattern_space) |
1328 | pattern_space_size += strlen(pattern_space); |
1329 | if (G.hold_space) |
1330 | hold_space_size = strlen(G.hold_space); |
1331 | pattern_space = xrealloc(pattern_space, |
1332 | pattern_space_size + hold_space_size); |
1333 | if (pattern_space_size == 2) |
1334 | pattern_space[0] = 0; |
1335 | strcat(pattern_space, "\n"); |
1336 | if (G.hold_space) |
1337 | strcat(pattern_space, G.hold_space); |
1338 | last_gets_char = '\n'; |
1339 | |
1340 | break; |
1341 | } |
1342 | case 'h': /* Replace hold space with pattern space */ |
1343 | free(G.hold_space); |
1344 | G.hold_space = xstrdup(pattern_space); |
1345 | break; |
1346 | case 'H': /* Append newline and pattern space to hold space */ |
1347 | { |
1348 | int hold_space_size = 2; |
1349 | int pattern_space_size = 0; |
1350 | |
1351 | if (G.hold_space) |
1352 | hold_space_size += strlen(G.hold_space); |
1353 | if (pattern_space) |
1354 | pattern_space_size = strlen(pattern_space); |
1355 | G.hold_space = xrealloc(G.hold_space, |
1356 | hold_space_size + pattern_space_size); |
1357 | |
1358 | if (hold_space_size == 2) |
1359 | *G.hold_space = 0; |
1360 | strcat(G.hold_space, "\n"); |
1361 | if (pattern_space) |
1362 | strcat(G.hold_space, pattern_space); |
1363 | |
1364 | break; |
1365 | } |
1366 | case 'x': /* Exchange hold and pattern space */ |
1367 | { |
1368 | char *tmp = pattern_space; |
1369 | pattern_space = G.hold_space ? G.hold_space : xzalloc(1); |
1370 | last_gets_char = '\n'; |
1371 | G.hold_space = tmp; |
1372 | break; |
1373 | } |
1374 | } /* switch */ |
1375 | } /* for each cmd */ |
1376 | |
1377 | /* |
1378 | * Exit point from sedding... |
1379 | */ |
1380 | discard_commands: |
1381 | /* we will print the line unless we were told to be quiet ('-n') |
1382 | or if the line was suppressed (ala 'd'elete) */ |
1383 | if (!G.be_quiet) |
1384 | sed_puts(pattern_space, last_gets_char); |
1385 | |
1386 | /* Delete and such jump here. */ |
1387 | discard_line: |
1388 | flush_append(&last_puts_char, last_gets_char); |
1389 | free(pattern_space); |
1390 | |
1391 | goto again; |
1392 | } |
1393 | |
1394 | /* It is possible to have a command line argument with embedded |
1395 | * newlines. This counts as multiple command lines. |
1396 | * However, newline can be escaped: 's/e/z\<newline>z/' |
1397 | * add_cmd() handles this. |
1398 | */ |
1399 | |
1400 | static void add_cmd_block(char *cmdstr) |
1401 | { |
1402 | char *sv, *eol; |
1403 | |
1404 | cmdstr = sv = xstrdup(cmdstr); |
1405 | do { |
1406 | eol = strchr(cmdstr, '\n'); |
1407 | if (eol) |
1408 | *eol = '\0'; |
1409 | add_cmd(cmdstr); |
1410 | cmdstr = eol + 1; |
1411 | } while (eol); |
1412 | free(sv); |
1413 | } |
1414 | |
1415 | int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
1416 | int sed_main(int argc UNUSED_PARAM, char **argv) |
1417 | { |
1418 | unsigned opt; |
1419 | llist_t *opt_e, *opt_f; |
1420 | char *opt_i; |
1421 | |
1422 | #if ENABLE_LONG_OPTS |
1423 | static const char sed_longopts[] ALIGN1 = |
1424 | /* name has_arg short */ |
1425 | "in-place\0" Optional_argument "i" |
1426 | "regexp-extended\0" No_argument "r" |
1427 | "quiet\0" No_argument "n" |
1428 | "silent\0" No_argument "n" |
1429 | "expression\0" Required_argument "e" |
1430 | "file\0" Required_argument "f"; |
1431 | #endif |
1432 | |
1433 | INIT_G(); |
1434 | |
1435 | /* destroy command strings on exit */ |
1436 | if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff); |
1437 | |
1438 | /* Lie to autoconf when it starts asking stupid questions. */ |
1439 | if (argv[1] && strcmp(argv[1], "--version") == 0) { |
1440 | puts("This is not GNU sed version 4.0"); |
1441 | return 0; |
1442 | } |
1443 | |
1444 | /* do normal option parsing */ |
1445 | opt_e = opt_f = NULL; |
1446 | opt_i = NULL; |
1447 | opt_complementary = "e::f::" /* can occur multiple times */ |
1448 | "nn"; /* count -n */ |
1449 | |
1450 | IF_LONG_OPTS(applet_long_options = sed_longopts); |
1451 | |
1452 | /* -i must be first, to match OPT_in_place definition */ |
1453 | /* -E is a synonym of -r: |
1454 | * GNU sed 4.2.1 mentions it in neither --help |
1455 | * nor manpage, but does recognize it. |
1456 | */ |
1457 | opt = getopt32(argv, "i::rEne:f:", &opt_i, &opt_e, &opt_f, |
1458 | &G.be_quiet); /* counter for -n */ |
1459 | //argc -= optind; |
1460 | argv += optind; |
1461 | if (opt & OPT_in_place) { // -i |
1462 | atexit(cleanup_outname); |
1463 | } |
1464 | if (opt & (2|4)) |
1465 | G.regex_type |= REG_EXTENDED; // -r or -E |
1466 | //if (opt & 8) |
1467 | // G.be_quiet++; // -n (implemented with a counter instead) |
1468 | while (opt_e) { // -e |
1469 | add_cmd_block(llist_pop(&opt_e)); |
1470 | } |
1471 | while (opt_f) { // -f |
1472 | char *line; |
1473 | FILE *cmdfile; |
1474 | cmdfile = xfopen_for_read(llist_pop(&opt_f)); |
1475 | while ((line = xmalloc_fgetline(cmdfile)) != NULL) { |
1476 | add_cmd(line); |
1477 | free(line); |
1478 | } |
1479 | fclose(cmdfile); |
1480 | } |
1481 | /* if we didn't get a pattern from -e or -f, use argv[0] */ |
1482 | if (!(opt & 0x30)) { |
1483 | if (!*argv) |
1484 | bb_show_usage(); |
1485 | add_cmd_block(*argv++); |
1486 | } |
1487 | /* Flush any unfinished commands. */ |
1488 | add_cmd(""); |
1489 | |
1490 | /* By default, we write to stdout */ |
1491 | G.nonstdout = stdout; |
1492 | |
1493 | /* argv[0..(argc-1)] should be names of file to process. If no |
1494 | * files were specified or '-' was specified, take input from stdin. |
1495 | * Otherwise, we process all the files specified. */ |
1496 | G.input_file_list = argv; |
1497 | if (!argv[0]) { |
1498 | if (opt & OPT_in_place) |
1499 | bb_error_msg_and_die(bb_msg_requires_arg, "-i"); |
1500 | argv[0] = (char*)bb_msg_standard_input; |
1501 | /* G.last_input_file = 0; - already is */ |
1502 | } else { |
1503 | goto start; |
1504 | |
1505 | for (; *argv; argv++) { |
1506 | struct stat statbuf; |
1507 | int nonstdoutfd; |
1508 | sed_cmd_t *sed_cmd; |
1509 | |
1510 | G.last_input_file++; |
1511 | start: |
1512 | if (!(opt & OPT_in_place)) { |
1513 | if (LONE_DASH(*argv)) { |
1514 | *argv = (char*)bb_msg_standard_input; |
1515 | process_files(); |
1516 | } |
1517 | continue; |
1518 | } |
1519 | |
1520 | /* -i: process each FILE separately: */ |
1521 | |
1522 | G.outname = xasprintf("%sXXXXXX", *argv); |
1523 | nonstdoutfd = xmkstemp(G.outname); |
1524 | G.nonstdout = xfdopen_for_write(nonstdoutfd); |
1525 | |
1526 | /* Set permissions/owner of output file */ |
1527 | stat(*argv, &statbuf); |
1528 | /* chmod'ing AFTER chown would preserve suid/sgid bits, |
1529 | * but GNU sed 4.2.1 does not preserve them either */ |
1530 | fchmod(nonstdoutfd, statbuf.st_mode); |
1531 | fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid); |
1532 | |
1533 | process_files(); |
1534 | fclose(G.nonstdout); |
1535 | G.nonstdout = stdout; |
1536 | |
1537 | if (opt_i) { |
1538 | char *backupname = xasprintf("%s%s", *argv, opt_i); |
1539 | xrename(*argv, backupname); |
1540 | free(backupname); |
1541 | } |
1542 | /* else unlink(*argv); - rename below does this */ |
1543 | xrename(G.outname, *argv); //TODO: rollback backup on error? |
1544 | free(G.outname); |
1545 | G.outname = NULL; |
1546 | |
1547 | /* Re-enable disabled range matches */ |
1548 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
1549 | sed_cmd->beg_line = sed_cmd->beg_line_orig; |
1550 | } |
1551 | } |
1552 | /* Here, to handle "sed 'cmds' nonexistent_file" case we did: |
1553 | * if (G.current_input_file[G.current_input_file] == NULL) |
1554 | * return G.exitcode; |
1555 | * but it's not needed since process_files() works correctly |
1556 | * in this case too. */ |
1557 | } |
1558 | |
1559 | process_files(); |
1560 | |
1561 | return G.exitcode; |
1562 | } |
1563 |