blob: 637a6851bcbd726c44c2b2786a162930b87caf77
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/|$]command |
37 | * - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags) |
38 | * - edit commands: (a)ppend, (i)nsert, (c)hange |
39 | * - file commands: (r)ead |
40 | * - backreferences in substitution expressions (\0, \1, \2...\9) |
41 | * - grouped commands: {cmd1;cmd2} |
42 | * - transliteration (y/source-chars/dest-chars/) |
43 | * - pattern space hold space storing / swapping (g, h, x) |
44 | * - labels / branching (: label, b, t, T) |
45 | * |
46 | * (Note: Specifying an address (range) to match is *optional*; commands |
47 | * default to the whole pattern space if no specific address match was |
48 | * requested.) |
49 | * |
50 | * Todo: |
51 | * - Create a wrapper around regex to make libc's regex conform with sed |
52 | * |
53 | * Reference |
54 | * http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html |
55 | * http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html |
56 | * http://sed.sourceforge.net/sedfaq3.html |
57 | */ |
58 | |
59 | //config:config SED |
60 | //config: bool "sed" |
61 | //config: default y |
62 | //config: help |
63 | //config: sed is used to perform text transformations on a file |
64 | //config: or input from a pipeline. |
65 | |
66 | //kbuild:lib-$(CONFIG_SED) += sed.o |
67 | |
68 | //applet:IF_SED(APPLET(sed, BB_DIR_BIN, BB_SUID_DROP)) |
69 | |
70 | //usage:#define sed_trivial_usage |
71 | //usage: "[-inrE] [-f FILE]... [-e CMD]... [FILE]...\n" |
72 | //usage: "or: sed [-inrE] CMD [FILE]..." |
73 | //usage:#define sed_full_usage "\n\n" |
74 | //usage: " -e CMD Add CMD to sed commands to be executed" |
75 | //usage: "\n -f FILE Add FILE contents to sed commands to be executed" |
76 | //usage: "\n -i[SFX] Edit files in-place (otherwise sends to stdout)" |
77 | //usage: "\n Optionally back files up, appending SFX" |
78 | //usage: "\n -n Suppress automatic printing of pattern space" |
79 | //usage: "\n -r,-E Use extended regex syntax" |
80 | //usage: "\n" |
81 | //usage: "\nIf no -e or -f, the first non-option argument is the sed command string." |
82 | //usage: "\nRemaining arguments are input files (stdin if none)." |
83 | //usage: |
84 | //usage:#define sed_example_usage |
85 | //usage: "$ echo \"foo\" | sed -e 's/f[a-zA-Z]o/bar/g'\n" |
86 | //usage: "bar\n" |
87 | |
88 | #include "libbb.h" |
89 | #include "common_bufsiz.h" |
90 | #include "xregex.h" |
91 | |
92 | #if 0 |
93 | # define dbg(...) bb_error_msg(__VA_ARGS__) |
94 | #else |
95 | # define dbg(...) ((void)0) |
96 | #endif |
97 | |
98 | |
99 | enum { |
100 | OPT_in_place = 1 << 0, |
101 | }; |
102 | |
103 | /* Each sed command turns into one of these structures. */ |
104 | typedef struct sed_cmd_s { |
105 | /* Ordered by alignment requirements: currently 36 bytes on x86 */ |
106 | struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */ |
107 | |
108 | /* address storage */ |
109 | regex_t *beg_match; /* sed -e '/match/cmd' */ |
110 | regex_t *end_match; /* sed -e '/match/,/end_match/cmd' */ |
111 | regex_t *sub_match; /* For 's/sub_match/string/' */ |
112 | int beg_line; /* 'sed 1p' 0 == apply commands to all lines */ |
113 | int beg_line_orig; /* copy of the above, needed for -i */ |
114 | int end_line; /* 'sed 1,3p' 0 == one line only. -1 = last line ($). -2-N = +N */ |
115 | int end_line_orig; |
116 | |
117 | FILE *sw_file; /* File (sw) command writes to, NULL for none. */ |
118 | char *string; /* Data string for (saicytb) commands. */ |
119 | |
120 | unsigned which_match; /* (s) Which match to replace (0 for all) */ |
121 | |
122 | /* Bitfields (gcc won't group them if we don't) */ |
123 | unsigned invert:1; /* the '!' after the address */ |
124 | unsigned in_match:1; /* Next line also included in match? */ |
125 | unsigned sub_p:1; /* (s) print option */ |
126 | |
127 | char sw_last_char; /* Last line written by (sw) had no '\n' */ |
128 | |
129 | /* GENERAL FIELDS */ |
130 | char cmd; /* The command char: abcdDgGhHilnNpPqrstwxy:={} */ |
131 | } sed_cmd_t; |
132 | |
133 | static const char semicolon_whitespace[] ALIGN1 = "; \n\r\t\v"; |
134 | |
135 | struct globals { |
136 | /* options */ |
137 | int be_quiet, regex_type; |
138 | |
139 | FILE *nonstdout; |
140 | char *outname, *hold_space; |
141 | smallint exitcode; |
142 | |
143 | /* list of input files */ |
144 | int current_input_file, last_input_file; |
145 | char **input_file_list; |
146 | FILE *current_fp; |
147 | |
148 | regmatch_t regmatch[10]; |
149 | regex_t *previous_regex_ptr; |
150 | |
151 | /* linked list of sed commands */ |
152 | sed_cmd_t *sed_cmd_head, **sed_cmd_tail; |
153 | |
154 | /* linked list of append lines */ |
155 | llist_t *append_head; |
156 | |
157 | char *add_cmd_line; |
158 | |
159 | struct pipeline { |
160 | char *buf; /* Space to hold string */ |
161 | int idx; /* Space used */ |
162 | int len; /* Space allocated */ |
163 | } pipeline; |
164 | } FIX_ALIASING; |
165 | #define G (*(struct globals*)bb_common_bufsiz1) |
166 | #define INIT_G() do { \ |
167 | setup_common_bufsiz(); \ |
168 | BUILD_BUG_ON(sizeof(G) > COMMON_BUFSIZE); \ |
169 | G.sed_cmd_tail = &G.sed_cmd_head; \ |
170 | } while (0) |
171 | |
172 | |
173 | #if ENABLE_FEATURE_CLEAN_UP |
174 | static void sed_free_and_close_stuff(void) |
175 | { |
176 | sed_cmd_t *sed_cmd = G.sed_cmd_head; |
177 | |
178 | llist_free(G.append_head, free); |
179 | |
180 | while (sed_cmd) { |
181 | sed_cmd_t *sed_cmd_next = sed_cmd->next; |
182 | |
183 | if (sed_cmd->sw_file) |
184 | fclose(sed_cmd->sw_file); |
185 | |
186 | if (sed_cmd->beg_match) { |
187 | regfree(sed_cmd->beg_match); |
188 | free(sed_cmd->beg_match); |
189 | } |
190 | if (sed_cmd->end_match) { |
191 | regfree(sed_cmd->end_match); |
192 | free(sed_cmd->end_match); |
193 | } |
194 | if (sed_cmd->sub_match) { |
195 | regfree(sed_cmd->sub_match); |
196 | free(sed_cmd->sub_match); |
197 | } |
198 | free(sed_cmd->string); |
199 | free(sed_cmd); |
200 | sed_cmd = sed_cmd_next; |
201 | } |
202 | |
203 | free(G.hold_space); |
204 | |
205 | if (G.current_fp) |
206 | fclose(G.current_fp); |
207 | } |
208 | #else |
209 | void sed_free_and_close_stuff(void); |
210 | #endif |
211 | |
212 | /* If something bad happens during -i operation, delete temp file */ |
213 | |
214 | static void cleanup_outname(void) |
215 | { |
216 | if (G.outname) unlink(G.outname); |
217 | } |
218 | |
219 | /* strcpy, replacing "\from" with 'to'. If to is NUL, replacing "\any" with 'any' */ |
220 | |
221 | static unsigned parse_escapes(char *dest, const char *string, int len, char from, char to) |
222 | { |
223 | char *d = dest; |
224 | int i = 0; |
225 | |
226 | if (len == -1) |
227 | len = strlen(string); |
228 | |
229 | while (i < len) { |
230 | if (string[i] == '\\') { |
231 | if (!to || string[i+1] == from) { |
232 | if ((*d = to ? to : string[i+1]) == '\0') |
233 | return d - dest; |
234 | i += 2; |
235 | d++; |
236 | continue; |
237 | } |
238 | i++; /* skip backslash in string[] */ |
239 | *d++ = '\\'; |
240 | /* fall through: copy next char verbatim */ |
241 | } |
242 | if ((*d = string[i++]) == '\0') |
243 | return d - dest; |
244 | d++; |
245 | } |
246 | *d = '\0'; |
247 | return d - dest; |
248 | } |
249 | |
250 | static char *copy_parsing_escapes(const char *string, int len) |
251 | { |
252 | const char *s; |
253 | char *dest = xmalloc(len + 1); |
254 | |
255 | /* sed recognizes \n */ |
256 | /* GNU sed also recognizes \t and \r */ |
257 | for (s = "\nn\tt\rr"; *s; s += 2) { |
258 | len = parse_escapes(dest, string, len, s[1], s[0]); |
259 | string = dest; |
260 | } |
261 | return dest; |
262 | } |
263 | |
264 | |
265 | /* |
266 | * index_of_next_unescaped_regexp_delim - walks left to right through a string |
267 | * beginning at a specified index and returns the index of the next regular |
268 | * expression delimiter (typically a forward slash ('/')) not preceded by |
269 | * a backslash ('\'). A negative delimiter disables square bracket checking. |
270 | */ |
271 | static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str) |
272 | { |
273 | int bracket = -1; |
274 | int escaped = 0; |
275 | int idx = 0; |
276 | char ch; |
277 | |
278 | if (delimiter < 0) { |
279 | bracket--; |
280 | delimiter = -delimiter; |
281 | } |
282 | |
283 | for (; (ch = str[idx]) != '\0'; idx++) { |
284 | if (bracket >= 0) { |
285 | if (ch == ']' |
286 | && !(bracket == idx - 1 || (bracket == idx - 2 && str[idx - 1] == '^')) |
287 | ) { |
288 | bracket = -1; |
289 | } |
290 | } else if (escaped) |
291 | escaped = 0; |
292 | else if (ch == '\\') |
293 | escaped = 1; |
294 | else if (bracket == -1 && ch == '[') |
295 | bracket = idx; |
296 | else if (ch == delimiter) |
297 | return idx; |
298 | } |
299 | |
300 | /* if we make it to here, we've hit the end of the string */ |
301 | bb_error_msg_and_die("unmatched '%c'", delimiter); |
302 | } |
303 | |
304 | /* |
305 | * Returns the index of the third delimiter |
306 | */ |
307 | static int parse_regex_delim(const char *cmdstr, char **match, char **replace) |
308 | { |
309 | const char *cmdstr_ptr = cmdstr; |
310 | unsigned char delimiter; |
311 | int idx = 0; |
312 | |
313 | /* verify that the 's' or 'y' is followed by something. That something |
314 | * (typically a 'slash') is now our regexp delimiter... */ |
315 | if (*cmdstr == '\0') |
316 | bb_error_msg_and_die("bad format in substitution expression"); |
317 | delimiter = *cmdstr_ptr++; |
318 | |
319 | /* save the match string */ |
320 | idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr); |
321 | *match = copy_parsing_escapes(cmdstr_ptr, idx); |
322 | |
323 | /* save the replacement string */ |
324 | cmdstr_ptr += idx + 1; |
325 | idx = index_of_next_unescaped_regexp_delim(- (int)delimiter, cmdstr_ptr); |
326 | *replace = copy_parsing_escapes(cmdstr_ptr, idx); |
327 | |
328 | return ((cmdstr_ptr - cmdstr) + idx); |
329 | } |
330 | |
331 | /* |
332 | * returns the index in the string just past where the address ends. |
333 | */ |
334 | static int get_address(const char *my_str, int *linenum, regex_t ** regex) |
335 | { |
336 | const char *pos = my_str; |
337 | |
338 | if (isdigit(*my_str)) { |
339 | *linenum = strtol(my_str, (char**)&pos, 10); |
340 | /* endstr shouldnt ever equal NULL */ |
341 | } else if (*my_str == '$') { |
342 | *linenum = -1; |
343 | pos++; |
344 | } else if (*my_str == '/' || *my_str == '\\') { |
345 | int next; |
346 | char delimiter; |
347 | char *temp; |
348 | |
349 | delimiter = '/'; |
350 | if (*my_str == '\\') |
351 | delimiter = *++pos; |
352 | next = index_of_next_unescaped_regexp_delim(delimiter, ++pos); |
353 | temp = copy_parsing_escapes(pos, next); |
354 | *regex = xzalloc(sizeof(regex_t)); |
355 | xregcomp(*regex, temp, G.regex_type); |
356 | free(temp); |
357 | /* Move position to next character after last delimiter */ |
358 | pos += (next+1); |
359 | } |
360 | return pos - my_str; |
361 | } |
362 | |
363 | /* Grab a filename. Whitespace at start is skipped, then goes to EOL. */ |
364 | static int parse_file_cmd(/*sed_cmd_t *sed_cmd,*/ const char *filecmdstr, char **retval) |
365 | { |
366 | int start = 0, idx, hack = 0; |
367 | |
368 | /* Skip whitespace, then grab filename to end of line */ |
369 | while (isspace(filecmdstr[start])) |
370 | start++; |
371 | idx = start; |
372 | while (filecmdstr[idx] && filecmdstr[idx] != '\n') |
373 | idx++; |
374 | |
375 | /* If lines glued together, put backslash back. */ |
376 | if (filecmdstr[idx] == '\n') |
377 | hack = 1; |
378 | if (idx == start) |
379 | bb_error_msg_and_die("empty filename"); |
380 | *retval = xstrndup(filecmdstr+start, idx-start+hack+1); |
381 | if (hack) |
382 | (*retval)[idx] = '\\'; |
383 | |
384 | return idx; |
385 | } |
386 | |
387 | static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr) |
388 | { |
389 | int cflags = G.regex_type; |
390 | char *match; |
391 | int idx; |
392 | |
393 | /* |
394 | * A substitution command should look something like this: |
395 | * s/match/replace/ #giIpw |
396 | * || | ||| |
397 | * mandatory optional |
398 | */ |
399 | idx = parse_regex_delim(substr, &match, &sed_cmd->string); |
400 | |
401 | /* determine the number of back references in the match string */ |
402 | /* Note: we compute this here rather than in the do_subst_command() |
403 | * function to save processor time, at the expense of a little more memory |
404 | * (4 bits) per sed_cmd */ |
405 | |
406 | /* process the flags */ |
407 | |
408 | sed_cmd->which_match = 1; |
409 | dbg("s flags:'%s'", substr + idx + 1); |
410 | while (substr[++idx]) { |
411 | dbg("s flag:'%c'", substr[idx]); |
412 | /* Parse match number */ |
413 | if (isdigit(substr[idx])) { |
414 | if (match[0] != '^') { |
415 | /* Match 0 treated as all, multiple matches we take the last one. */ |
416 | const char *pos = substr + idx; |
417 | /* FIXME: error check? */ |
418 | sed_cmd->which_match = (unsigned)strtol(substr+idx, (char**) &pos, 10); |
419 | idx = pos - substr - 1; |
420 | } |
421 | continue; |
422 | } |
423 | /* Skip spaces */ |
424 | if (isspace(substr[idx])) |
425 | continue; |
426 | |
427 | switch (substr[idx]) { |
428 | /* Replace all occurrences */ |
429 | case 'g': |
430 | if (match[0] != '^') |
431 | sed_cmd->which_match = 0; |
432 | break; |
433 | /* Print pattern space */ |
434 | case 'p': |
435 | sed_cmd->sub_p = 1; |
436 | break; |
437 | /* Write to file */ |
438 | case 'w': |
439 | { |
440 | char *fname; |
441 | idx += parse_file_cmd(/*sed_cmd,*/ substr+idx+1, &fname); |
442 | sed_cmd->sw_file = xfopen_for_write(fname); |
443 | sed_cmd->sw_last_char = '\n'; |
444 | free(fname); |
445 | break; |
446 | } |
447 | /* Ignore case (gnu exension) */ |
448 | case 'i': |
449 | case 'I': |
450 | cflags |= REG_ICASE; |
451 | break; |
452 | /* Comment */ |
453 | case '#': |
454 | // while (substr[++idx]) continue; |
455 | idx += strlen(substr + idx); // same |
456 | /* Fall through */ |
457 | /* End of command */ |
458 | case ';': |
459 | case '}': |
460 | goto out; |
461 | default: |
462 | dbg("s bad flags:'%s'", substr + idx); |
463 | bb_error_msg_and_die("bad option in substitution expression"); |
464 | } |
465 | } |
466 | out: |
467 | /* compile the match string into a regex */ |
468 | if (*match != '\0') { |
469 | /* If match is empty, we use last regex used at runtime */ |
470 | sed_cmd->sub_match = xzalloc(sizeof(regex_t)); |
471 | dbg("xregcomp('%s',%x)", match, cflags); |
472 | xregcomp(sed_cmd->sub_match, match, cflags); |
473 | dbg("regcomp ok"); |
474 | } |
475 | free(match); |
476 | |
477 | return idx; |
478 | } |
479 | |
480 | /* |
481 | * Process the commands arguments |
482 | */ |
483 | static const char *parse_cmd_args(sed_cmd_t *sed_cmd, const char *cmdstr) |
484 | { |
485 | static const char cmd_letters[] ALIGN1 = "saicrw:btTydDgGhHlnNpPqx={}"; |
486 | enum { |
487 | IDX_s = 0, |
488 | IDX_a, |
489 | IDX_i, |
490 | IDX_c, |
491 | IDX_r, |
492 | IDX_w, |
493 | IDX_colon, |
494 | IDX_b, |
495 | IDX_t, |
496 | IDX_T, |
497 | IDX_y, |
498 | IDX_d, |
499 | IDX_D, |
500 | IDX_g, |
501 | IDX_G, |
502 | IDX_h, |
503 | IDX_H, |
504 | IDX_l, |
505 | IDX_n, |
506 | IDX_N, |
507 | IDX_p, |
508 | IDX_P, |
509 | IDX_q, |
510 | IDX_x, |
511 | IDX_equal, |
512 | IDX_lbrace, |
513 | IDX_rbrace, |
514 | IDX_nul |
515 | }; |
516 | unsigned idx; |
517 | |
518 | BUILD_BUG_ON(sizeof(cmd_letters)-1 != IDX_nul); |
519 | |
520 | idx = strchrnul(cmd_letters, sed_cmd->cmd) - cmd_letters; |
521 | |
522 | /* handle (s)ubstitution command */ |
523 | if (idx == IDX_s) { |
524 | cmdstr += parse_subst_cmd(sed_cmd, cmdstr); |
525 | } |
526 | /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */ |
527 | else if (idx <= IDX_c) { /* a,i,c */ |
528 | unsigned len; |
529 | |
530 | if (idx < IDX_c) { /* a,i */ |
531 | if (sed_cmd->end_line || sed_cmd->end_match) |
532 | bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd); |
533 | } |
534 | for (;;) { |
535 | if (*cmdstr == '\n' || *cmdstr == '\\') { |
536 | cmdstr++; |
537 | break; |
538 | } |
539 | if (!isspace(*cmdstr)) |
540 | break; |
541 | cmdstr++; |
542 | } |
543 | len = strlen(cmdstr); |
544 | sed_cmd->string = copy_parsing_escapes(cmdstr, len); |
545 | cmdstr += len; |
546 | /* "\anychar" -> "anychar" */ |
547 | parse_escapes(sed_cmd->string, sed_cmd->string, -1, '\0', '\0'); |
548 | } |
549 | /* handle file cmds: (r)ead */ |
550 | else if (idx <= IDX_w) { /* r,w */ |
551 | if (idx < IDX_w) { /* r */ |
552 | if (sed_cmd->end_line || sed_cmd->end_match) |
553 | bb_error_msg_and_die("command '%c' uses only one address", sed_cmd->cmd); |
554 | } |
555 | cmdstr += parse_file_cmd(/*sed_cmd,*/ cmdstr, &sed_cmd->string); |
556 | if (sed_cmd->cmd == 'w') { |
557 | sed_cmd->sw_file = xfopen_for_write(sed_cmd->string); |
558 | sed_cmd->sw_last_char = '\n'; |
559 | } |
560 | } |
561 | /* handle branch commands */ |
562 | else if (idx <= IDX_T) { /* :,b,t,T */ |
563 | int length; |
564 | |
565 | cmdstr = skip_whitespace(cmdstr); |
566 | length = strcspn(cmdstr, semicolon_whitespace); |
567 | if (length) { |
568 | sed_cmd->string = xstrndup(cmdstr, length); |
569 | cmdstr += length; |
570 | } |
571 | } |
572 | /* translation command */ |
573 | else if (idx == IDX_y) { |
574 | char *match, *replace; |
575 | int i = cmdstr[0]; |
576 | |
577 | cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1; |
578 | /* \n already parsed, but \delimiter needs unescaping. */ |
579 | parse_escapes(match, match, -1, i, i); |
580 | parse_escapes(replace, replace, -1, i, i); |
581 | |
582 | sed_cmd->string = xzalloc((strlen(match) + 1) * 2); |
583 | for (i = 0; match[i] && replace[i]; i++) { |
584 | sed_cmd->string[i*2] = match[i]; |
585 | sed_cmd->string[i*2+1] = replace[i]; |
586 | } |
587 | free(match); |
588 | free(replace); |
589 | } |
590 | /* if it wasnt a single-letter command that takes no arguments |
591 | * then it must be an invalid command. |
592 | */ |
593 | else if (idx >= IDX_nul) { /* not d,D,g,G,h,H,l,n,N,p,P,q,x,=,{,} */ |
594 | bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd); |
595 | } |
596 | |
597 | /* give back whatever's left over */ |
598 | return cmdstr; |
599 | } |
600 | |
601 | |
602 | /* Parse address+command sets, skipping comment lines. */ |
603 | |
604 | static void add_cmd(const char *cmdstr) |
605 | { |
606 | sed_cmd_t *sed_cmd; |
607 | unsigned len, n; |
608 | |
609 | /* Append this line to any unfinished line from last time. */ |
610 | if (G.add_cmd_line) { |
611 | char *tp = xasprintf("%s\n%s", G.add_cmd_line, cmdstr); |
612 | free(G.add_cmd_line); |
613 | cmdstr = G.add_cmd_line = tp; |
614 | } |
615 | |
616 | /* If this line ends with unescaped backslash, request next line. */ |
617 | n = len = strlen(cmdstr); |
618 | while (n && cmdstr[n-1] == '\\') |
619 | n--; |
620 | if ((len - n) & 1) { /* if odd number of trailing backslashes */ |
621 | if (!G.add_cmd_line) |
622 | G.add_cmd_line = xstrdup(cmdstr); |
623 | G.add_cmd_line[len-1] = '\0'; |
624 | return; |
625 | } |
626 | |
627 | /* Loop parsing all commands in this line. */ |
628 | while (*cmdstr) { |
629 | /* Skip leading whitespace and semicolons */ |
630 | cmdstr += strspn(cmdstr, semicolon_whitespace); |
631 | |
632 | /* If no more commands, exit. */ |
633 | if (!*cmdstr) break; |
634 | |
635 | /* if this is a comment, jump past it and keep going */ |
636 | if (*cmdstr == '#') { |
637 | /* "#n" is the same as using -n on the command line */ |
638 | if (cmdstr[1] == 'n') |
639 | G.be_quiet++; |
640 | cmdstr = strpbrk(cmdstr, "\n\r"); |
641 | if (!cmdstr) break; |
642 | continue; |
643 | } |
644 | |
645 | /* parse the command |
646 | * format is: [addr][,addr][!]cmd |
647 | * |----||-----||-| |
648 | * part1 part2 part3 |
649 | */ |
650 | |
651 | sed_cmd = xzalloc(sizeof(sed_cmd_t)); |
652 | |
653 | /* first part (if present) is an address: either a '$', a number or a /regex/ */ |
654 | cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match); |
655 | sed_cmd->beg_line_orig = sed_cmd->beg_line; |
656 | |
657 | /* second part (if present) will begin with a comma */ |
658 | if (*cmdstr == ',') { |
659 | int idx; |
660 | |
661 | cmdstr++; |
662 | if (*cmdstr == '+' && isdigit(cmdstr[1])) { |
663 | /* http://sed.sourceforge.net/sedfaq3.html#s3.3 |
664 | * Under GNU sed 3.02+, ssed, and sed15+, <address2> |
665 | * may also be a notation of the form +num, |
666 | * indicating the next num lines after <address1> is |
667 | * matched. |
668 | * GNU sed 4.2.1 accepts even "+" (meaning "+0"). |
669 | * We don't (we check for isdigit, see above), think |
670 | * about the "+-3" case. |
671 | */ |
672 | char *end; |
673 | /* code is smaller compared to using &cmdstr here: */ |
674 | idx = strtol(cmdstr+1, &end, 10); |
675 | sed_cmd->end_line = -2 - idx; |
676 | cmdstr = end; |
677 | } else { |
678 | idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match); |
679 | cmdstr += idx; |
680 | idx--; /* if 0, trigger error check below */ |
681 | } |
682 | if (idx < 0) |
683 | bb_error_msg_and_die("no address after comma"); |
684 | sed_cmd->end_line_orig = sed_cmd->end_line; |
685 | } |
686 | |
687 | /* skip whitespace before the command */ |
688 | cmdstr = skip_whitespace(cmdstr); |
689 | |
690 | /* Check for inversion flag */ |
691 | if (*cmdstr == '!') { |
692 | sed_cmd->invert = 1; |
693 | cmdstr++; |
694 | |
695 | /* skip whitespace before the command */ |
696 | cmdstr = skip_whitespace(cmdstr); |
697 | } |
698 | |
699 | /* last part (mandatory) will be a command */ |
700 | if (!*cmdstr) |
701 | bb_error_msg_and_die("missing command"); |
702 | sed_cmd->cmd = *cmdstr++; |
703 | cmdstr = parse_cmd_args(sed_cmd, cmdstr); |
704 | |
705 | /* cmdstr now points past args. |
706 | * GNU sed requires a separator, if there are more commands, |
707 | * else it complains "char N: extra characters after command". |
708 | * Example: "sed 'p;d'". We also allow "sed 'pd'". |
709 | */ |
710 | |
711 | /* Add the command to the command array */ |
712 | *G.sed_cmd_tail = sed_cmd; |
713 | G.sed_cmd_tail = &sed_cmd->next; |
714 | } |
715 | |
716 | /* If we glued multiple lines together, free the memory. */ |
717 | free(G.add_cmd_line); |
718 | G.add_cmd_line = NULL; |
719 | } |
720 | |
721 | /* Append to a string, reallocating memory as necessary. */ |
722 | |
723 | #define PIPE_GROW 64 |
724 | |
725 | static void pipe_putc(char c) |
726 | { |
727 | if (G.pipeline.idx == G.pipeline.len) { |
728 | G.pipeline.buf = xrealloc(G.pipeline.buf, |
729 | G.pipeline.len + PIPE_GROW); |
730 | G.pipeline.len += PIPE_GROW; |
731 | } |
732 | G.pipeline.buf[G.pipeline.idx++] = c; |
733 | } |
734 | |
735 | static void do_subst_w_backrefs(char *line, char *replace) |
736 | { |
737 | int i, j; |
738 | |
739 | /* go through the replacement string */ |
740 | for (i = 0; replace[i]; i++) { |
741 | /* if we find a backreference (\1, \2, etc.) print the backref'ed text */ |
742 | if (replace[i] == '\\') { |
743 | unsigned backref = replace[++i] - '0'; |
744 | if (backref <= 9) { |
745 | /* print out the text held in G.regmatch[backref] */ |
746 | if (G.regmatch[backref].rm_so != -1) { |
747 | j = G.regmatch[backref].rm_so; |
748 | while (j < G.regmatch[backref].rm_eo) |
749 | pipe_putc(line[j++]); |
750 | } |
751 | continue; |
752 | } |
753 | /* I _think_ it is impossible to get '\' to be |
754 | * the last char in replace string. Thus we dont check |
755 | * for replace[i] == NUL. (counterexample anyone?) */ |
756 | /* if we find a backslash escaped character, print the character */ |
757 | pipe_putc(replace[i]); |
758 | continue; |
759 | } |
760 | /* if we find an unescaped '&' print out the whole matched text. */ |
761 | if (replace[i] == '&') { |
762 | j = G.regmatch[0].rm_so; |
763 | while (j < G.regmatch[0].rm_eo) |
764 | pipe_putc(line[j++]); |
765 | continue; |
766 | } |
767 | /* Otherwise just output the character. */ |
768 | pipe_putc(replace[i]); |
769 | } |
770 | } |
771 | |
772 | static int do_subst_command(sed_cmd_t *sed_cmd, char **line_p) |
773 | { |
774 | char *line = *line_p; |
775 | unsigned match_count = 0; |
776 | bool altered = 0; |
777 | bool prev_match_empty = 1; |
778 | bool tried_at_eol = 0; |
779 | regex_t *current_regex; |
780 | |
781 | current_regex = sed_cmd->sub_match; |
782 | /* Handle empty regex. */ |
783 | if (!current_regex) { |
784 | current_regex = G.previous_regex_ptr; |
785 | if (!current_regex) |
786 | bb_error_msg_and_die("no previous regexp"); |
787 | } |
788 | G.previous_regex_ptr = current_regex; |
789 | |
790 | /* Find the first match */ |
791 | dbg("matching '%s'", line); |
792 | if (REG_NOMATCH == regexec(current_regex, line, 10, G.regmatch, 0)) { |
793 | dbg("no match"); |
794 | return 0; |
795 | } |
796 | dbg("match"); |
797 | |
798 | /* Initialize temporary output buffer. */ |
799 | G.pipeline.buf = xmalloc(PIPE_GROW); |
800 | G.pipeline.len = PIPE_GROW; |
801 | G.pipeline.idx = 0; |
802 | |
803 | /* Now loop through, substituting for matches */ |
804 | do { |
805 | int start = G.regmatch[0].rm_so; |
806 | int end = G.regmatch[0].rm_eo; |
807 | int i; |
808 | |
809 | match_count++; |
810 | |
811 | /* If we aren't interested in this match, output old line to |
812 | * end of match and continue */ |
813 | if (sed_cmd->which_match |
814 | && (sed_cmd->which_match != match_count) |
815 | ) { |
816 | for (i = 0; i < end; i++) |
817 | pipe_putc(*line++); |
818 | /* Null match? Print one more char */ |
819 | if (start == end && *line) |
820 | pipe_putc(*line++); |
821 | goto next; |
822 | } |
823 | |
824 | /* Print everything before the match */ |
825 | for (i = 0; i < start; i++) |
826 | pipe_putc(line[i]); |
827 | |
828 | /* Then print the substitution string, |
829 | * unless we just matched empty string after non-empty one. |
830 | * Example: string "cccd", pattern "c*", repl "R": |
831 | * result is "RdR", not "RRdR": first match "ccc", |
832 | * second is "" before "d", third is "" after "d". |
833 | * Second match is NOT replaced! |
834 | */ |
835 | if (prev_match_empty || start != 0 || start != end) { |
836 | //dbg("%d %d %d", prev_match_empty, start, end); |
837 | dbg("inserting replacement at %d in '%s'", start, line); |
838 | do_subst_w_backrefs(line, sed_cmd->string); |
839 | /* Flag that something has changed */ |
840 | altered = 1; |
841 | } else { |
842 | dbg("NOT inserting replacement at %d in '%s'", start, line); |
843 | } |
844 | |
845 | /* If matched string is empty (f.e. "c*" pattern), |
846 | * copy verbatim one char after it before attempting more matches |
847 | */ |
848 | prev_match_empty = (start == end); |
849 | if (prev_match_empty) { |
850 | if (!line[end]) { |
851 | tried_at_eol = 1; |
852 | } else { |
853 | pipe_putc(line[end]); |
854 | end++; |
855 | } |
856 | } |
857 | |
858 | /* Advance past the match */ |
859 | dbg("line += %d", end); |
860 | line += end; |
861 | |
862 | /* if we're not doing this globally, get out now */ |
863 | if (sed_cmd->which_match != 0) |
864 | break; |
865 | next: |
866 | /* Exit if we are at EOL and already tried matching at it */ |
867 | if (*line == '\0') { |
868 | if (tried_at_eol) |
869 | break; |
870 | tried_at_eol = 1; |
871 | } |
872 | |
873 | //maybe (end ? REG_NOTBOL : 0) instead of unconditional REG_NOTBOL? |
874 | } while (regexec(current_regex, line, 10, G.regmatch, REG_NOTBOL) != REG_NOMATCH); |
875 | |
876 | /* Copy rest of string into output pipeline */ |
877 | while (1) { |
878 | char c = *line++; |
879 | pipe_putc(c); |
880 | if (c == '\0') |
881 | break; |
882 | } |
883 | |
884 | free(*line_p); |
885 | *line_p = G.pipeline.buf; |
886 | return altered; |
887 | } |
888 | |
889 | /* Set command pointer to point to this label. (Does not handle null label.) */ |
890 | static sed_cmd_t *branch_to(char *label) |
891 | { |
892 | sed_cmd_t *sed_cmd; |
893 | |
894 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
895 | if (sed_cmd->cmd == ':' |
896 | && sed_cmd->string |
897 | && strcmp(sed_cmd->string, label) == 0 |
898 | ) { |
899 | return sed_cmd; |
900 | } |
901 | } |
902 | bb_error_msg_and_die("can't find label for jump to '%s'", label); |
903 | } |
904 | |
905 | static void append(char *s) |
906 | { |
907 | llist_add_to_end(&G.append_head, s); |
908 | } |
909 | |
910 | /* Output line of text. */ |
911 | /* Note: |
912 | * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed. |
913 | * Without them, we had this: |
914 | * echo -n thingy >z1 |
915 | * echo -n again >z2 |
916 | * >znull |
917 | * sed "s/i/z/" z1 z2 znull | hexdump -vC |
918 | * output: |
919 | * gnu sed 4.1.5: |
920 | * 00000000 74 68 7a 6e 67 79 0a 61 67 61 7a 6e |thzngy.agazn| |
921 | * bbox: |
922 | * 00000000 74 68 7a 6e 67 79 61 67 61 7a 6e |thzngyagazn| |
923 | */ |
924 | enum { |
925 | NO_EOL_CHAR = 1, |
926 | LAST_IS_NUL = 2, |
927 | }; |
928 | static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char) |
929 | { |
930 | char lpc = *last_puts_char; |
931 | |
932 | /* Need to insert a '\n' between two files because first file's |
933 | * last line wasn't terminated? */ |
934 | if (lpc != '\n' && lpc != '\0') { |
935 | fputc('\n', file); |
936 | lpc = '\n'; |
937 | } |
938 | fputs(s, file); |
939 | |
940 | /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */ |
941 | if (s[0]) |
942 | lpc = 'x'; |
943 | |
944 | /* had trailing '\0' and it was last char of file? */ |
945 | if (last_gets_char == LAST_IS_NUL) { |
946 | fputc('\0', file); |
947 | lpc = 'x'; /* */ |
948 | } else |
949 | /* had trailing '\n' or '\0'? */ |
950 | if (last_gets_char != NO_EOL_CHAR) { |
951 | fputc(last_gets_char, file); |
952 | lpc = last_gets_char; |
953 | } |
954 | |
955 | if (ferror(file)) { |
956 | xfunc_error_retval = 4; /* It's what gnu sed exits with... */ |
957 | bb_error_msg_and_die(bb_msg_write_error); |
958 | } |
959 | *last_puts_char = lpc; |
960 | } |
961 | |
962 | static void flush_append(char *last_puts_char) |
963 | { |
964 | char *data; |
965 | |
966 | /* Output appended lines. */ |
967 | while ((data = (char *)llist_pop(&G.append_head)) != NULL) { |
968 | /* Append command does not respect "nonterminated-ness" |
969 | * of last line. Try this: |
970 | * $ echo -n "woot" | sed -e '/woot/a woo' - |
971 | * woot |
972 | * woo |
973 | * (both lines are terminated with \n) |
974 | * Therefore we do not propagate "last_gets_char" here, |
975 | * pass '\n' instead: |
976 | */ |
977 | puts_maybe_newline(data, G.nonstdout, last_puts_char, '\n'); |
978 | free(data); |
979 | } |
980 | } |
981 | |
982 | /* Get next line of input from G.input_file_list, flushing append buffer and |
983 | * noting if we ran out of files without a newline on the last line we read. |
984 | */ |
985 | static char *get_next_line(char *gets_char, char *last_puts_char) |
986 | { |
987 | char *temp = NULL; |
988 | int len; |
989 | char gc; |
990 | |
991 | flush_append(last_puts_char); |
992 | |
993 | /* will be returned if last line in the file |
994 | * doesn't end with either '\n' or '\0' */ |
995 | gc = NO_EOL_CHAR; |
996 | for (; G.current_input_file <= G.last_input_file; G.current_input_file++) { |
997 | FILE *fp = G.current_fp; |
998 | if (!fp) { |
999 | const char *path = G.input_file_list[G.current_input_file]; |
1000 | fp = stdin; |
1001 | if (path != bb_msg_standard_input) { |
1002 | fp = fopen_or_warn(path, "r"); |
1003 | if (!fp) { |
1004 | G.exitcode = EXIT_FAILURE; |
1005 | continue; |
1006 | } |
1007 | } |
1008 | G.current_fp = fp; |
1009 | } |
1010 | /* Read line up to a newline or NUL byte, inclusive, |
1011 | * return malloc'ed char[]. length of the chunk read |
1012 | * is stored in len. NULL if EOF/error */ |
1013 | temp = bb_get_chunk_from_file(fp, &len); |
1014 | if (temp) { |
1015 | /* len > 0 here, it's ok to do temp[len-1] */ |
1016 | char c = temp[len-1]; |
1017 | if (c == '\n' || c == '\0') { |
1018 | temp[len-1] = '\0'; |
1019 | gc = c; |
1020 | if (c == '\0') { |
1021 | int ch = fgetc(fp); |
1022 | if (ch != EOF) |
1023 | ungetc(ch, fp); |
1024 | else |
1025 | gc = LAST_IS_NUL; |
1026 | } |
1027 | } |
1028 | /* else we put NO_EOL_CHAR into *gets_char */ |
1029 | break; |
1030 | |
1031 | /* NB: I had the idea of peeking next file(s) and returning |
1032 | * NO_EOL_CHAR only if it is the *last* non-empty |
1033 | * input file. But there is a case where this won't work: |
1034 | * file1: "a woo\nb woo" |
1035 | * file2: "c no\nd no" |
1036 | * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang" |
1037 | * (note: *no* newline after "b bang"!) */ |
1038 | } |
1039 | /* Close this file and advance to next one */ |
1040 | fclose_if_not_stdin(fp); |
1041 | G.current_fp = NULL; |
1042 | } |
1043 | *gets_char = gc; |
1044 | return temp; |
1045 | } |
1046 | |
1047 | #define sed_puts(s, n) (puts_maybe_newline(s, G.nonstdout, &last_puts_char, n)) |
1048 | |
1049 | static int beg_match(sed_cmd_t *sed_cmd, const char *pattern_space) |
1050 | { |
1051 | int retval = sed_cmd->beg_match && !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0); |
1052 | if (retval) |
1053 | G.previous_regex_ptr = sed_cmd->beg_match; |
1054 | return retval; |
1055 | } |
1056 | |
1057 | /* Process all the lines in all the files */ |
1058 | |
1059 | static void process_files(void) |
1060 | { |
1061 | char *pattern_space, *next_line; |
1062 | int linenum = 0; |
1063 | char last_puts_char = '\n'; |
1064 | char last_gets_char, next_gets_char; |
1065 | sed_cmd_t *sed_cmd; |
1066 | int substituted; |
1067 | |
1068 | /* Prime the pump */ |
1069 | next_line = get_next_line(&next_gets_char, &last_puts_char); |
1070 | |
1071 | /* Go through every line in each file */ |
1072 | again: |
1073 | substituted = 0; |
1074 | |
1075 | /* Advance to next line. Stop if out of lines. */ |
1076 | pattern_space = next_line; |
1077 | if (!pattern_space) |
1078 | return; |
1079 | last_gets_char = next_gets_char; |
1080 | |
1081 | /* Read one line in advance so we can act on the last line, |
1082 | * the '$' address */ |
1083 | next_line = get_next_line(&next_gets_char, &last_puts_char); |
1084 | linenum++; |
1085 | |
1086 | /* For every line, go through all the commands */ |
1087 | restart: |
1088 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
1089 | int old_matched, matched; |
1090 | |
1091 | old_matched = sed_cmd->in_match; |
1092 | |
1093 | /* Determine if this command matches this line: */ |
1094 | |
1095 | dbg("match1:%d", sed_cmd->in_match); |
1096 | dbg("match2:%d", (!sed_cmd->beg_line && !sed_cmd->end_line |
1097 | && !sed_cmd->beg_match && !sed_cmd->end_match)); |
1098 | dbg("match3:%d", (sed_cmd->beg_line > 0 |
1099 | && (sed_cmd->end_line || sed_cmd->end_match |
1100 | ? (sed_cmd->beg_line <= linenum) |
1101 | : (sed_cmd->beg_line == linenum) |
1102 | ) |
1103 | )); |
1104 | dbg("match4:%d", (beg_match(sed_cmd, pattern_space))); |
1105 | dbg("match5:%d", (sed_cmd->beg_line == -1 && next_line == NULL)); |
1106 | |
1107 | /* Are we continuing a previous multi-line match? */ |
1108 | sed_cmd->in_match = sed_cmd->in_match |
1109 | /* Or is no range necessary? */ |
1110 | || (!sed_cmd->beg_line && !sed_cmd->end_line |
1111 | && !sed_cmd->beg_match && !sed_cmd->end_match) |
1112 | /* Or did we match the start of a numerical range? */ |
1113 | || (sed_cmd->beg_line > 0 |
1114 | && (sed_cmd->end_line || sed_cmd->end_match |
1115 | /* note: even if end is numeric and is < linenum too, |
1116 | * GNU sed matches! We match too, therefore we don't |
1117 | * check here that linenum <= end. |
1118 | * Example: |
1119 | * printf '1\n2\n3\n4\n' | sed -n '1{N;N;d};1p;2,3p;3p;4p' |
1120 | * first three input lines are deleted; |
1121 | * 4th line is matched and printed |
1122 | * by "2,3" (!) and by "4" ranges |
1123 | */ |
1124 | ? (sed_cmd->beg_line <= linenum) /* N,end */ |
1125 | : (sed_cmd->beg_line == linenum) /* N */ |
1126 | ) |
1127 | ) |
1128 | /* Or does this line match our begin address regex? */ |
1129 | || (beg_match(sed_cmd, pattern_space)) |
1130 | /* Or did we match last line of input? */ |
1131 | || (sed_cmd->beg_line == -1 && next_line == NULL); |
1132 | |
1133 | /* Snapshot the value */ |
1134 | matched = sed_cmd->in_match; |
1135 | |
1136 | dbg("cmd:'%c' matched:%d beg_line:%d end_line:%d linenum:%d", |
1137 | sed_cmd->cmd, matched, sed_cmd->beg_line, sed_cmd->end_line, linenum); |
1138 | |
1139 | /* Is this line the end of the current match? */ |
1140 | |
1141 | if (matched) { |
1142 | if (sed_cmd->end_line <= -2) { |
1143 | /* address2 is +N, i.e. N lines from beg_line */ |
1144 | sed_cmd->end_line = linenum + (-sed_cmd->end_line - 2); |
1145 | } |
1146 | /* once matched, "n,xxx" range is dead, disabling it */ |
1147 | if (sed_cmd->beg_line > 0) { |
1148 | sed_cmd->beg_line = -2; |
1149 | } |
1150 | dbg("end1:%d", sed_cmd->end_line ? sed_cmd->end_line == -1 |
1151 | ? !next_line : (sed_cmd->end_line <= linenum) |
1152 | : !sed_cmd->end_match); |
1153 | dbg("end2:%d", sed_cmd->end_match && old_matched |
1154 | && !regexec(sed_cmd->end_match,pattern_space, 0, NULL, 0)); |
1155 | sed_cmd->in_match = !( |
1156 | /* has the ending line come, or is this a single address command? */ |
1157 | (sed_cmd->end_line |
1158 | ? sed_cmd->end_line == -1 |
1159 | ? !next_line |
1160 | : (sed_cmd->end_line <= linenum) |
1161 | : !sed_cmd->end_match |
1162 | ) |
1163 | /* or does this line matches our last address regex */ |
1164 | || (sed_cmd->end_match && old_matched |
1165 | && (regexec(sed_cmd->end_match, |
1166 | pattern_space, 0, NULL, 0) == 0) |
1167 | ) |
1168 | ); |
1169 | } |
1170 | |
1171 | /* Skip blocks of commands we didn't match */ |
1172 | if (sed_cmd->cmd == '{') { |
1173 | if (sed_cmd->invert ? matched : !matched) { |
1174 | unsigned nest_cnt = 0; |
1175 | while (1) { |
1176 | if (sed_cmd->cmd == '{') |
1177 | nest_cnt++; |
1178 | if (sed_cmd->cmd == '}') { |
1179 | nest_cnt--; |
1180 | if (nest_cnt == 0) |
1181 | break; |
1182 | } |
1183 | sed_cmd = sed_cmd->next; |
1184 | if (!sed_cmd) |
1185 | bb_error_msg_and_die("unterminated {"); |
1186 | } |
1187 | } |
1188 | continue; |
1189 | } |
1190 | |
1191 | /* Okay, so did this line match? */ |
1192 | if (sed_cmd->invert ? matched : !matched) |
1193 | continue; /* no */ |
1194 | |
1195 | /* Update last used regex in case a blank substitute BRE is found */ |
1196 | if (sed_cmd->beg_match) { |
1197 | G.previous_regex_ptr = sed_cmd->beg_match; |
1198 | } |
1199 | |
1200 | /* actual sedding */ |
1201 | dbg("pattern_space:'%s' next_line:'%s' cmd:%c", |
1202 | pattern_space, next_line, sed_cmd->cmd); |
1203 | switch (sed_cmd->cmd) { |
1204 | |
1205 | /* Print line number */ |
1206 | case '=': |
1207 | fprintf(G.nonstdout, "%d\n", linenum); |
1208 | break; |
1209 | |
1210 | /* Write the current pattern space up to the first newline */ |
1211 | case 'P': |
1212 | { |
1213 | char *tmp = strchr(pattern_space, '\n'); |
1214 | if (tmp) { |
1215 | *tmp = '\0'; |
1216 | /* TODO: explain why '\n' below */ |
1217 | sed_puts(pattern_space, '\n'); |
1218 | *tmp = '\n'; |
1219 | break; |
1220 | } |
1221 | /* Fall Through */ |
1222 | } |
1223 | |
1224 | /* Write the current pattern space to output */ |
1225 | case 'p': |
1226 | /* NB: we print this _before_ the last line |
1227 | * (of current file) is printed. Even if |
1228 | * that line is nonterminated, we print |
1229 | * '\n' here (gnu sed does the same) */ |
1230 | sed_puts(pattern_space, '\n'); |
1231 | break; |
1232 | /* Delete up through first newline */ |
1233 | case 'D': |
1234 | { |
1235 | char *tmp = strchr(pattern_space, '\n'); |
1236 | if (tmp) { |
1237 | overlapping_strcpy(pattern_space, tmp + 1); |
1238 | goto restart; |
1239 | } |
1240 | } |
1241 | /* discard this line. */ |
1242 | case 'd': |
1243 | goto discard_line; |
1244 | |
1245 | /* Substitute with regex */ |
1246 | case 's': |
1247 | if (!do_subst_command(sed_cmd, &pattern_space)) |
1248 | break; |
1249 | dbg("do_subst_command succeeded:'%s'", pattern_space); |
1250 | substituted |= 1; |
1251 | |
1252 | /* handle p option */ |
1253 | if (sed_cmd->sub_p) |
1254 | sed_puts(pattern_space, last_gets_char); |
1255 | /* handle w option */ |
1256 | if (sed_cmd->sw_file) |
1257 | puts_maybe_newline( |
1258 | pattern_space, sed_cmd->sw_file, |
1259 | &sed_cmd->sw_last_char, last_gets_char); |
1260 | break; |
1261 | |
1262 | /* Append line to linked list to be printed later */ |
1263 | case 'a': |
1264 | append(xstrdup(sed_cmd->string)); |
1265 | break; |
1266 | |
1267 | /* Insert text before this line */ |
1268 | case 'i': |
1269 | sed_puts(sed_cmd->string, '\n'); |
1270 | break; |
1271 | |
1272 | /* Cut and paste text (replace) */ |
1273 | case 'c': |
1274 | /* Only triggers on last line of a matching range. */ |
1275 | if (!sed_cmd->in_match) |
1276 | sed_puts(sed_cmd->string, '\n'); |
1277 | goto discard_line; |
1278 | |
1279 | /* Read file, append contents to output */ |
1280 | case 'r': |
1281 | { |
1282 | FILE *rfile; |
1283 | rfile = fopen_for_read(sed_cmd->string); |
1284 | if (rfile) { |
1285 | char *line; |
1286 | while ((line = xmalloc_fgetline(rfile)) |
1287 | != NULL) |
1288 | append(line); |
1289 | fclose(rfile); |
1290 | } |
1291 | |
1292 | break; |
1293 | } |
1294 | |
1295 | /* Write pattern space to file. */ |
1296 | case 'w': |
1297 | puts_maybe_newline( |
1298 | pattern_space, sed_cmd->sw_file, |
1299 | &sed_cmd->sw_last_char, last_gets_char); |
1300 | break; |
1301 | |
1302 | /* Read next line from input */ |
1303 | case 'n': |
1304 | if (!G.be_quiet) |
1305 | sed_puts(pattern_space, last_gets_char); |
1306 | if (next_line == NULL) { |
1307 | /* If no next line, jump to end of script and exit. */ |
1308 | goto discard_line; |
1309 | } |
1310 | free(pattern_space); |
1311 | pattern_space = next_line; |
1312 | last_gets_char = next_gets_char; |
1313 | next_line = get_next_line(&next_gets_char, &last_puts_char); |
1314 | substituted = 0; |
1315 | linenum++; |
1316 | break; |
1317 | |
1318 | /* Quit. End of script, end of input. */ |
1319 | case 'q': |
1320 | /* Exit the outer while loop */ |
1321 | free(next_line); |
1322 | next_line = NULL; |
1323 | goto discard_commands; |
1324 | |
1325 | /* Append the next line to the current line */ |
1326 | case 'N': |
1327 | { |
1328 | int len; |
1329 | /* If no next line, jump to end of script and exit. */ |
1330 | /* http://www.gnu.org/software/sed/manual/sed.html: |
1331 | * "Most versions of sed exit without printing anything |
1332 | * when the N command is issued on the last line of |
1333 | * a file. GNU sed prints pattern space before exiting |
1334 | * unless of course the -n command switch has been |
1335 | * specified. This choice is by design." |
1336 | */ |
1337 | if (next_line == NULL) { |
1338 | //goto discard_line; |
1339 | goto discard_commands; /* GNU behavior */ |
1340 | } |
1341 | /* Append next_line, read new next_line. */ |
1342 | len = strlen(pattern_space); |
1343 | pattern_space = xrealloc(pattern_space, len + strlen(next_line) + 2); |
1344 | pattern_space[len] = '\n'; |
1345 | strcpy(pattern_space + len+1, next_line); |
1346 | last_gets_char = next_gets_char; |
1347 | next_line = get_next_line(&next_gets_char, &last_puts_char); |
1348 | linenum++; |
1349 | break; |
1350 | } |
1351 | |
1352 | /* Test/branch if substitution occurred */ |
1353 | case 't': |
1354 | if (!substituted) break; |
1355 | substituted = 0; |
1356 | /* Fall through */ |
1357 | /* Test/branch if substitution didn't occur */ |
1358 | case 'T': |
1359 | if (substituted) break; |
1360 | /* Fall through */ |
1361 | /* Branch to label */ |
1362 | case 'b': |
1363 | if (!sed_cmd->string) goto discard_commands; |
1364 | else sed_cmd = branch_to(sed_cmd->string); |
1365 | break; |
1366 | /* Transliterate characters */ |
1367 | case 'y': |
1368 | { |
1369 | int i, j; |
1370 | for (i = 0; pattern_space[i]; i++) { |
1371 | for (j = 0; sed_cmd->string[j]; j += 2) { |
1372 | if (pattern_space[i] == sed_cmd->string[j]) { |
1373 | pattern_space[i] = sed_cmd->string[j + 1]; |
1374 | break; |
1375 | } |
1376 | } |
1377 | } |
1378 | |
1379 | break; |
1380 | } |
1381 | case 'g': /* Replace pattern space with hold space */ |
1382 | free(pattern_space); |
1383 | pattern_space = xstrdup(G.hold_space ? G.hold_space : ""); |
1384 | break; |
1385 | case 'G': /* Append newline and hold space to pattern space */ |
1386 | { |
1387 | int pattern_space_size = 2; |
1388 | int hold_space_size = 0; |
1389 | |
1390 | if (pattern_space) |
1391 | pattern_space_size += strlen(pattern_space); |
1392 | if (G.hold_space) |
1393 | hold_space_size = strlen(G.hold_space); |
1394 | pattern_space = xrealloc(pattern_space, |
1395 | pattern_space_size + hold_space_size); |
1396 | if (pattern_space_size == 2) |
1397 | pattern_space[0] = 0; |
1398 | strcat(pattern_space, "\n"); |
1399 | if (G.hold_space) |
1400 | strcat(pattern_space, G.hold_space); |
1401 | last_gets_char = '\n'; |
1402 | |
1403 | break; |
1404 | } |
1405 | case 'h': /* Replace hold space with pattern space */ |
1406 | free(G.hold_space); |
1407 | G.hold_space = xstrdup(pattern_space); |
1408 | break; |
1409 | case 'H': /* Append newline and pattern space to hold space */ |
1410 | { |
1411 | int hold_space_size = 2; |
1412 | int pattern_space_size = 0; |
1413 | |
1414 | if (G.hold_space) |
1415 | hold_space_size += strlen(G.hold_space); |
1416 | if (pattern_space) |
1417 | pattern_space_size = strlen(pattern_space); |
1418 | G.hold_space = xrealloc(G.hold_space, |
1419 | hold_space_size + pattern_space_size); |
1420 | |
1421 | if (hold_space_size == 2) |
1422 | *G.hold_space = 0; |
1423 | strcat(G.hold_space, "\n"); |
1424 | if (pattern_space) |
1425 | strcat(G.hold_space, pattern_space); |
1426 | |
1427 | break; |
1428 | } |
1429 | case 'x': /* Exchange hold and pattern space */ |
1430 | { |
1431 | char *tmp = pattern_space; |
1432 | pattern_space = G.hold_space ? G.hold_space : xzalloc(1); |
1433 | last_gets_char = '\n'; |
1434 | G.hold_space = tmp; |
1435 | break; |
1436 | } |
1437 | } /* switch */ |
1438 | } /* for each cmd */ |
1439 | |
1440 | /* |
1441 | * Exit point from sedding... |
1442 | */ |
1443 | discard_commands: |
1444 | /* we will print the line unless we were told to be quiet ('-n') |
1445 | or if the line was suppressed (ala 'd'elete) */ |
1446 | if (!G.be_quiet) |
1447 | sed_puts(pattern_space, last_gets_char); |
1448 | |
1449 | /* Delete and such jump here. */ |
1450 | discard_line: |
1451 | flush_append(&last_puts_char /*,last_gets_char*/); |
1452 | free(pattern_space); |
1453 | |
1454 | goto again; |
1455 | } |
1456 | |
1457 | /* It is possible to have a command line argument with embedded |
1458 | * newlines. This counts as multiple command lines. |
1459 | * However, newline can be escaped: 's/e/z\<newline>z/' |
1460 | * add_cmd() handles this. |
1461 | */ |
1462 | |
1463 | static void add_cmd_block(char *cmdstr) |
1464 | { |
1465 | char *sv, *eol; |
1466 | |
1467 | cmdstr = sv = xstrdup(cmdstr); |
1468 | do { |
1469 | eol = strchr(cmdstr, '\n'); |
1470 | if (eol) |
1471 | *eol = '\0'; |
1472 | add_cmd(cmdstr); |
1473 | cmdstr = eol + 1; |
1474 | } while (eol); |
1475 | free(sv); |
1476 | } |
1477 | |
1478 | int sed_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
1479 | int sed_main(int argc UNUSED_PARAM, char **argv) |
1480 | { |
1481 | unsigned opt; |
1482 | llist_t *opt_e, *opt_f; |
1483 | char *opt_i; |
1484 | |
1485 | #if ENABLE_LONG_OPTS |
1486 | static const char sed_longopts[] ALIGN1 = |
1487 | /* name has_arg short */ |
1488 | "in-place\0" Optional_argument "i" |
1489 | "regexp-extended\0" No_argument "r" |
1490 | "quiet\0" No_argument "n" |
1491 | "silent\0" No_argument "n" |
1492 | "expression\0" Required_argument "e" |
1493 | "file\0" Required_argument "f"; |
1494 | #endif |
1495 | |
1496 | INIT_G(); |
1497 | |
1498 | /* destroy command strings on exit */ |
1499 | if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff); |
1500 | |
1501 | /* Lie to autoconf when it starts asking stupid questions. */ |
1502 | if (argv[1] && strcmp(argv[1], "--version") == 0) { |
1503 | puts("This is not GNU sed version 4.0"); |
1504 | return 0; |
1505 | } |
1506 | |
1507 | /* do normal option parsing */ |
1508 | opt_e = opt_f = NULL; |
1509 | opt_i = NULL; |
1510 | opt_complementary = "nn"; /* count -n */ |
1511 | |
1512 | IF_LONG_OPTS(applet_long_options = sed_longopts); |
1513 | |
1514 | /* -i must be first, to match OPT_in_place definition */ |
1515 | /* -E is a synonym of -r: |
1516 | * GNU sed 4.2.1 mentions it in neither --help |
1517 | * nor manpage, but does recognize it. |
1518 | */ |
1519 | opt = getopt32(argv, "i::rEne:*f:*", &opt_i, &opt_e, &opt_f, |
1520 | &G.be_quiet); /* counter for -n */ |
1521 | //argc -= optind; |
1522 | argv += optind; |
1523 | if (opt & OPT_in_place) { // -i |
1524 | atexit(cleanup_outname); |
1525 | } |
1526 | if (opt & (2|4)) |
1527 | G.regex_type |= REG_EXTENDED; // -r or -E |
1528 | //if (opt & 8) |
1529 | // G.be_quiet++; // -n (implemented with a counter instead) |
1530 | while (opt_e) { // -e |
1531 | add_cmd_block(llist_pop(&opt_e)); |
1532 | } |
1533 | while (opt_f) { // -f |
1534 | char *line; |
1535 | FILE *cmdfile; |
1536 | cmdfile = xfopen_stdin(llist_pop(&opt_f)); |
1537 | while ((line = xmalloc_fgetline(cmdfile)) != NULL) { |
1538 | add_cmd(line); |
1539 | free(line); |
1540 | } |
1541 | fclose_if_not_stdin(cmdfile); |
1542 | } |
1543 | /* if we didn't get a pattern from -e or -f, use argv[0] */ |
1544 | if (!(opt & 0x30)) { |
1545 | if (!*argv) |
1546 | bb_show_usage(); |
1547 | add_cmd_block(*argv++); |
1548 | } |
1549 | /* Flush any unfinished commands. */ |
1550 | add_cmd(""); |
1551 | |
1552 | /* By default, we write to stdout */ |
1553 | G.nonstdout = stdout; |
1554 | |
1555 | /* argv[0..(argc-1)] should be names of file to process. If no |
1556 | * files were specified or '-' was specified, take input from stdin. |
1557 | * Otherwise, we process all the files specified. */ |
1558 | G.input_file_list = argv; |
1559 | if (!argv[0]) { |
1560 | if (opt & OPT_in_place) |
1561 | bb_error_msg_and_die(bb_msg_requires_arg, "-i"); |
1562 | argv[0] = (char*)bb_msg_standard_input; |
1563 | /* G.last_input_file = 0; - already is */ |
1564 | } else { |
1565 | goto start; |
1566 | |
1567 | for (; *argv; argv++) { |
1568 | struct stat statbuf; |
1569 | int nonstdoutfd; |
1570 | sed_cmd_t *sed_cmd; |
1571 | |
1572 | G.last_input_file++; |
1573 | start: |
1574 | if (!(opt & OPT_in_place)) { |
1575 | if (LONE_DASH(*argv)) { |
1576 | *argv = (char*)bb_msg_standard_input; |
1577 | process_files(); |
1578 | } |
1579 | continue; |
1580 | } |
1581 | |
1582 | /* -i: process each FILE separately: */ |
1583 | |
1584 | if (stat(*argv, &statbuf) != 0) { |
1585 | bb_simple_perror_msg(*argv); |
1586 | G.exitcode = EXIT_FAILURE; |
1587 | G.current_input_file++; |
1588 | continue; |
1589 | } |
1590 | G.outname = xasprintf("%sXXXXXX", *argv); |
1591 | nonstdoutfd = xmkstemp(G.outname); |
1592 | G.nonstdout = xfdopen_for_write(nonstdoutfd); |
1593 | /* Set permissions/owner of output file */ |
1594 | /* chmod'ing AFTER chown would preserve suid/sgid bits, |
1595 | * but GNU sed 4.2.1 does not preserve them either */ |
1596 | fchmod(nonstdoutfd, statbuf.st_mode); |
1597 | fchown(nonstdoutfd, statbuf.st_uid, statbuf.st_gid); |
1598 | |
1599 | process_files(); |
1600 | fclose(G.nonstdout); |
1601 | G.nonstdout = stdout; |
1602 | |
1603 | if (opt_i) { |
1604 | char *backupname = xasprintf("%s%s", *argv, opt_i); |
1605 | xrename(*argv, backupname); |
1606 | free(backupname); |
1607 | } |
1608 | /* else unlink(*argv); - rename below does this */ |
1609 | xrename(G.outname, *argv); //TODO: rollback backup on error? |
1610 | free(G.outname); |
1611 | G.outname = NULL; |
1612 | |
1613 | /* Fix disabled range matches and mangled ",+N" ranges */ |
1614 | for (sed_cmd = G.sed_cmd_head; sed_cmd; sed_cmd = sed_cmd->next) { |
1615 | sed_cmd->beg_line = sed_cmd->beg_line_orig; |
1616 | sed_cmd->end_line = sed_cmd->end_line_orig; |
1617 | } |
1618 | } |
1619 | /* Here, to handle "sed 'cmds' nonexistent_file" case we did: |
1620 | * if (G.current_input_file[G.current_input_file] == NULL) |
1621 | * return G.exitcode; |
1622 | * but it's not needed since process_files() works correctly |
1623 | * in this case too. */ |
1624 | } |
1625 | |
1626 | process_files(); |
1627 | |
1628 | return G.exitcode; |
1629 | } |
1630 |