70b36095ff1d377bc6b15df687b7d469c1edfe1f
[oweals/busybox.git] / editors / sed.c
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 <bug1@iinet.net.au>
9  * Copyright (C) 2003,2004 by Rob Landley <rob@landley.net>
10  *
11  * MAINTAINER: Rob Landley <rob@landley.net>
12  *
13  * Licensed under GPL version 2, see file LICENSE in this tarball for details.
14  */
15
16 /* Code overview.
17
18   Files are laid out to avoid unnecessary function declarations.  So for
19   example, every function add_cmd calls occurs before add_cmd in this file.
20
21   add_cmd() is called on each line of sed command text (from a file or from
22   the command line).  It calls get_address() and parse_cmd_args().  The
23   resulting sed_cmd_t structures are appended to a linked list
24   (bbg.sed_cmd_head/bbg.sed_cmd_tail).
25
26   add_input_file() adds a FILE * to the list of input files.  We need to
27   know all input sources ahead of time to find the last line for the $ match.
28
29   process_files() does actual sedding, reading data lines from each input FILE *
30   (which could be stdin) and applying the sed command list (sed_cmd_head) to
31   each of the resulting lines.
32
33   sed_main() is where external code calls into this, with a command line.
34 */
35
36
37 /*
38         Supported features and commands in this version of sed:
39
40          - comments ('#')
41          - address matching: num|/matchstr/[,num|/matchstr/|$]command
42          - commands: (p)rint, (d)elete, (s)ubstitue (with g & I flags)
43          - edit commands: (a)ppend, (i)nsert, (c)hange
44          - file commands: (r)ead
45          - backreferences in substitution expressions (\0, \1, \2...\9)
46          - grouped commands: {cmd1;cmd2}
47          - transliteration (y/source-chars/dest-chars/)
48          - pattern space hold space storing / swapping (g, h, x)
49          - labels / branching (: label, b, t, T)
50
51          (Note: Specifying an address (range) to match is *optional*; commands
52          default to the whole pattern space if no specific address match was
53          requested.)
54
55         Todo:
56          - Create a wrapper around regex to make libc's regex conform with sed
57
58         Reference http://www.opengroup.org/onlinepubs/007904975/utilities/sed.html
59 */
60
61 #include "busybox.h"
62 #include "xregex.h"
63
64 /* Each sed command turns into one of these structures. */
65 typedef struct sed_cmd_s {
66         /* Ordered by alignment requirements: currently 36 bytes on x86 */
67
68         /* address storage */
69         regex_t *beg_match;     /* sed -e '/match/cmd' */
70         regex_t *end_match;     /* sed -e '/match/,/end_match/cmd' */
71         regex_t *sub_match;     /* For 's/sub_match/string/' */
72         int beg_line;           /* 'sed 1p'   0 == apply commands to all lines */
73         int end_line;           /* 'sed 1,3p' 0 == one line only. -1 = last line ($) */
74
75         FILE *sw_file;          /* File (sw) command writes to, -1 for none. */
76         char *string;           /* Data string for (saicytb) commands. */
77
78         unsigned short which_match;     /* (s) Which match to replace (0 for all) */
79
80         /* Bitfields (gcc won't group them if we don't) */
81         unsigned int invert:1;          /* the '!' after the address */
82         unsigned int in_match:1;        /* Next line also included in match? */
83         unsigned int sub_p:1;           /* (s) print option */
84
85         char sw_last_char;                  /* Last line written by (sw) had no '\n' */
86
87         /* GENERAL FIELDS */
88         char cmd;               /* The command char: abcdDgGhHilnNpPqrstwxy:={} */
89         struct sed_cmd_s *next; /* Next command (linked list, NULL terminated) */
90 } sed_cmd_t;
91
92 static const char *const semicolon_whitespace = "; \n\r\t\v";
93
94 struct sed_globals {
95         /* options */
96         int be_quiet, regex_type;
97         FILE *nonstdout;
98         char *outname, *hold_space;
99
100         /* List of input files */
101         int input_file_count, current_input_file;
102         FILE **input_file_list;
103
104         regmatch_t regmatch[10];
105         regex_t *previous_regex_ptr;
106
107         /* linked list of sed commands */
108         sed_cmd_t sed_cmd_head, *sed_cmd_tail;
109
110         /* Linked list of append lines */
111         llist_t *append_head;
112
113         char *add_cmd_line;
114
115         struct pipeline {
116                 char *buf;      /* Space to hold string */
117                 int idx;        /* Space used */
118                 int len;        /* Space allocated */
119         } pipeline;
120 } bbg;
121
122
123 #if ENABLE_FEATURE_CLEAN_UP
124 static void sed_free_and_close_stuff(void)
125 {
126         sed_cmd_t *sed_cmd = bbg.sed_cmd_head.next;
127
128         llist_free(bbg.append_head, free);
129
130         while (sed_cmd) {
131                 sed_cmd_t *sed_cmd_next = sed_cmd->next;
132
133                 if (sed_cmd->sw_file)
134                         xprint_and_close_file(sed_cmd->sw_file);
135
136                 if (sed_cmd->beg_match) {
137                         regfree(sed_cmd->beg_match);
138                         free(sed_cmd->beg_match);
139                 }
140                 if (sed_cmd->end_match) {
141                         regfree(sed_cmd->end_match);
142                         free(sed_cmd->end_match);
143                 }
144                 if (sed_cmd->sub_match) {
145                         regfree(sed_cmd->sub_match);
146                         free(sed_cmd->sub_match);
147                 }
148                 free(sed_cmd->string);
149                 free(sed_cmd);
150                 sed_cmd = sed_cmd_next;
151         }
152
153         if (bbg.hold_space) free(bbg.hold_space);
154
155         while (bbg.current_input_file < bbg.input_file_count)
156                 fclose(bbg.input_file_list[bbg.current_input_file++]);
157 }
158 #else
159 void sed_free_and_close_stuff(void);
160 #endif
161
162 /* If something bad happens during -i operation, delete temp file */
163
164 static void cleanup_outname(void)
165 {
166         if (bbg.outname) unlink(bbg.outname);
167 }
168
169 /* strdup, replacing "\n" with '\n', and "\delimiter" with 'delimiter' */
170
171 static void parse_escapes(char *dest, const char *string, int len, char from, char to)
172 {
173         int i = 0;
174
175         while (i < len) {
176                 if (string[i] == '\\') {
177                         if (!to || string[i+1] == from) {
178                                 *dest++ = to ? to : string[i+1];
179                                 i += 2;
180                                 continue;
181                         }
182                         *dest++ = string[i++];
183                 }
184                 *dest++ = string[i++];
185         }
186         *dest = 0;
187 }
188
189 static char *copy_parsing_escapes(const char *string, int len)
190 {
191         char *dest = xmalloc(len + 1);
192
193         parse_escapes(dest, string, len, 'n', '\n');
194         return dest;
195 }
196
197
198 /*
199  * index_of_next_unescaped_regexp_delim - walks left to right through a string
200  * beginning at a specified index and returns the index of the next regular
201  * expression delimiter (typically a forward * slash ('/')) not preceded by
202  * a backslash ('\').  A negative delimiter disables square bracket checking.
203  */
204 static int index_of_next_unescaped_regexp_delim(int delimiter, const char *str)
205 {
206         int bracket = -1;
207         int escaped = 0;
208         int idx = 0;
209         char ch;
210
211         if (delimiter < 0) {
212                 bracket--;
213                 delimiter = -delimiter;
214         }
215
216         for (; (ch = str[idx]); idx++) {
217                 if (bracket >= 0) {
218                         if (ch == ']' && !(bracket == idx - 1 || (bracket == idx - 2
219                                         && str[idx - 1] == '^')))
220                                 bracket = -1;
221                 } else if (escaped)
222                         escaped = 0;
223                 else if (ch == '\\')
224                         escaped = 1;
225                 else if (bracket == -1 && ch == '[')
226                         bracket = idx;
227                 else if (ch == delimiter)
228                         return idx;
229         }
230
231         /* if we make it to here, we've hit the end of the string */
232         bb_error_msg_and_die("unmatched '%c'", delimiter);
233 }
234
235 /*
236  *  Returns the index of the third delimiter
237  */
238 static int parse_regex_delim(char *cmdstr, char **match, char **replace)
239 {
240         char *cmdstr_ptr = cmdstr;
241         char delimiter;
242         int idx = 0;
243
244         /* verify that the 's' or 'y' is followed by something.  That something
245          * (typically a 'slash') is now our regexp delimiter... */
246         if (*cmdstr == '\0')
247                 bb_error_msg_and_die("bad format in substitution expression");
248         delimiter = *cmdstr_ptr++;
249
250         /* save the match string */
251         idx = index_of_next_unescaped_regexp_delim(delimiter, cmdstr_ptr);
252         *match = copy_parsing_escapes(cmdstr_ptr, idx);
253
254         /* save the replacement string */
255         cmdstr_ptr += idx + 1;
256         idx = index_of_next_unescaped_regexp_delim(-delimiter, cmdstr_ptr);
257         *replace = copy_parsing_escapes(cmdstr_ptr, idx);
258
259         return ((cmdstr_ptr - cmdstr) + idx);
260 }
261
262 /*
263  * returns the index in the string just past where the address ends.
264  */
265 static int get_address(const char *my_str, int *linenum, regex_t ** regex)
266 {
267         const char *pos = my_str;
268
269         if (isdigit(*my_str)) {
270                 *linenum = strtol(my_str, (char**)&pos, 10);
271                 /* endstr shouldnt ever equal NULL */
272         } else if (*my_str == '$') {
273                 *linenum = -1;
274                 pos++;
275         } else if (*my_str == '/' || *my_str == '\\') {
276                 int next;
277                 char delimiter;
278                 char *temp;
279
280                 delimiter = '/';
281                 if (*my_str == '\\') delimiter = *++pos;
282                 next = index_of_next_unescaped_regexp_delim(delimiter, ++pos);
283                 temp = copy_parsing_escapes(pos, next);
284                 *regex = xmalloc(sizeof(regex_t));
285                 xregcomp(*regex, temp, bbg.regex_type|REG_NEWLINE);
286                 free(temp);
287                 /* Move position to next character after last delimiter */
288                 pos += (next+1);
289         }
290         return pos - my_str;
291 }
292
293 /* Grab a filename.  Whitespace at start is skipped, then goes to EOL. */
294 static int parse_file_cmd(sed_cmd_t *sed_cmd, char *filecmdstr, char **retval)
295 {
296         int start = 0, idx, hack = 0;
297
298         /* Skip whitespace, then grab filename to end of line */
299         while (isspace(filecmdstr[start])) start++;
300         idx = start;
301         while (filecmdstr[idx] && filecmdstr[idx] != '\n') idx++;
302
303         /* If lines glued together, put backslash back. */
304         if (filecmdstr[idx] == '\n') hack = 1;
305         if (idx == start)
306                 bb_error_msg_and_die("empty filename");
307         *retval = xstrndup(filecmdstr+start, idx-start+hack+1);
308         if (hack) (*retval)[idx] = '\\';
309
310         return idx;
311 }
312
313 static int parse_subst_cmd(sed_cmd_t *sed_cmd, char *substr)
314 {
315         int cflags = bbg.regex_type;
316         char *match;
317         int idx;
318
319         /*
320          * A substitution command should look something like this:
321          *    s/match/replace/ #gIpw
322          *    ||     |        |||
323          *    mandatory       optional
324          */
325         idx = parse_regex_delim(substr, &match, &sed_cmd->string);
326
327         /* determine the number of back references in the match string */
328         /* Note: we compute this here rather than in the do_subst_command()
329          * function to save processor time, at the expense of a little more memory
330          * (4 bits) per sed_cmd */
331
332         /* process the flags */
333
334         sed_cmd->which_match = 1;
335         while (substr[++idx]) {
336                 /* Parse match number */
337                 if (isdigit(substr[idx])) {
338                         if (match[0] != '^') {
339                                 /* Match 0 treated as all, multiple matches we take the last one. */
340                                 char *pos = substr + idx;
341                                 /* FIXME: error check? */
342                                 sed_cmd->which_match = (unsigned short)strtol(substr+idx, &pos, 10);
343                                 idx = pos - substr;
344                         }
345                         continue;
346                 }
347                 /* Skip spaces */
348                 if (isspace(substr[idx])) continue;
349
350                 switch (substr[idx]) {
351                 /* Replace all occurrences */
352                 case 'g':
353                         if (match[0] != '^') sed_cmd->which_match = 0;
354                         break;
355                 /* Print pattern space */
356                 case 'p':
357                         sed_cmd->sub_p = 1;
358                         break;
359                 /* Write to file */
360                 case 'w':
361                 {
362                         char *temp;
363                         idx += parse_file_cmd(sed_cmd, substr+idx, &temp);
364
365                         break;
366                 }
367                 /* Ignore case (gnu exension) */
368                 case 'I':
369                         cflags |= REG_ICASE;
370                         break;
371                 /* Comment */
372                 case '#':
373                         while (substr[++idx]) /*skip all*/;
374                         /* Fall through */
375                 /* End of command */
376                 case ';':
377                 case '}':
378                         goto out;
379                 default:
380                         bb_error_msg_and_die("bad option in substitution expression");
381                 }
382         }
383 out:
384         /* compile the match string into a regex */
385         if (*match != '\0') {
386                 /* If match is empty, we use last regex used at runtime */
387                 sed_cmd->sub_match = xmalloc(sizeof(regex_t));
388                 xregcomp(sed_cmd->sub_match, match, cflags);
389         }
390         free(match);
391
392         return idx;
393 }
394
395 /*
396  *  Process the commands arguments
397  */
398 static char *parse_cmd_args(sed_cmd_t *sed_cmd, char *cmdstr)
399 {
400         /* handle (s)ubstitution command */
401         if (sed_cmd->cmd == 's')
402                 cmdstr += parse_subst_cmd(sed_cmd, cmdstr);
403         /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
404         else if (strchr("aic", sed_cmd->cmd)) {
405                 if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
406                         bb_error_msg_and_die
407                                 ("only a beginning address can be specified for edit commands");
408                 for (;;) {
409                         if (*cmdstr == '\n' || *cmdstr == '\\') {
410                                 cmdstr++;
411                                 break;
412                         } else if (isspace(*cmdstr))
413                                 cmdstr++;
414                         else
415                                 break;
416                 }
417                 sed_cmd->string = xstrdup(cmdstr);
418                 parse_escapes(sed_cmd->string, sed_cmd->string, strlen(cmdstr), 0, 0);
419                 cmdstr += strlen(cmdstr);
420         /* handle file cmds: (r)ead */
421         } else if (strchr("rw", sed_cmd->cmd)) {
422                 if (sed_cmd->end_line || sed_cmd->end_match)
423                         bb_error_msg_and_die("command only uses one address");
424                 cmdstr += parse_file_cmd(sed_cmd, cmdstr, &sed_cmd->string);
425                 if (sed_cmd->cmd == 'w') {
426                         sed_cmd->sw_file = xfopen(sed_cmd->string, "w");
427                         sed_cmd->sw_last_char = '\n';
428                 }
429         /* handle branch commands */
430         } else if (strchr(":btT", sed_cmd->cmd)) {
431                 int length;
432
433                 cmdstr = skip_whitespace(cmdstr);
434                 length = strcspn(cmdstr, semicolon_whitespace);
435                 if (length) {
436                         sed_cmd->string = xstrndup(cmdstr, length);
437                         cmdstr += length;
438                 }
439         }
440         /* translation command */
441         else if (sed_cmd->cmd == 'y') {
442                 char *match, *replace;
443                 int i = cmdstr[0];
444
445                 cmdstr += parse_regex_delim(cmdstr, &match, &replace)+1;
446                 /* \n already parsed, but \delimiter needs unescaping. */
447                 parse_escapes(match, match, strlen(match), i, i);
448                 parse_escapes(replace, replace, strlen(replace), i, i);
449
450                 sed_cmd->string = xzalloc((strlen(match) + 1) * 2);
451                 for (i = 0; match[i] && replace[i]; i++) {
452                         sed_cmd->string[i*2] = match[i];
453                         sed_cmd->string[i*2+1] = replace[i];
454                 }
455                 free(match);
456                 free(replace);
457         }
458         /* if it wasnt a single-letter command that takes no arguments
459          * then it must be an invalid command.
460          */
461         else if (strchr("dDgGhHlnNpPqx={}", sed_cmd->cmd) == 0) {
462                 bb_error_msg_and_die("unsupported command %c", sed_cmd->cmd);
463         }
464
465         /* give back whatever's left over */
466         return cmdstr;
467 }
468
469
470 /* Parse address+command sets, skipping comment lines. */
471
472 static void add_cmd(const char *cmdstr)
473 {
474         sed_cmd_t *sed_cmd;
475         int temp;
476
477         /* Append this line to any unfinished line from last time. */
478         if (bbg.add_cmd_line) {
479                 char *tp = xasprintf("%s\n%s", bbg.add_cmd_line, cmdstr);
480                 free(bbg.add_cmd_line);
481                 bbg.add_cmd_line = tp;
482         }
483
484         /* If this line ends with backslash, request next line. */
485         temp = strlen(cmdstr);
486         if (temp && cmdstr[temp-1] == '\\') {
487                 if (!bbg.add_cmd_line)
488                         bbg.add_cmd_line = xstrdup(cmdstr);
489                 bbg.add_cmd_line[temp-1] = 0;
490                 return;
491         }
492
493         /* Loop parsing all commands in this line. */
494         while (*cmdstr) {
495                 /* Skip leading whitespace and semicolons */
496                 cmdstr += strspn(cmdstr, semicolon_whitespace);
497
498                 /* If no more commands, exit. */
499                 if (!*cmdstr) break;
500
501                 /* if this is a comment, jump past it and keep going */
502                 if (*cmdstr == '#') {
503                         /* "#n" is the same as using -n on the command line */
504                         if (cmdstr[1] == 'n')
505                                 bbg.be_quiet++;
506                         cmdstr = strpbrk(cmdstr, "\n\r");
507                         if (!cmdstr) break;
508                         continue;
509                 }
510
511                 /* parse the command
512                  * format is: [addr][,addr][!]cmd
513                  *            |----||-----||-|
514                  *            part1 part2  part3
515                  */
516
517                 sed_cmd = xzalloc(sizeof(sed_cmd_t));
518
519                 /* first part (if present) is an address: either a '$', a number or a /regex/ */
520                 cmdstr += get_address(cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
521
522                 /* second part (if present) will begin with a comma */
523                 if (*cmdstr == ',') {
524                         int idx;
525
526                         cmdstr++;
527                         idx = get_address(cmdstr, &sed_cmd->end_line, &sed_cmd->end_match);
528                         if (!idx)
529                                 bb_error_msg_and_die("no address after comma");
530                         cmdstr += idx;
531                 }
532
533                 /* skip whitespace before the command */
534                 cmdstr = skip_whitespace(cmdstr);
535
536                 /* Check for inversion flag */
537                 if (*cmdstr == '!') {
538                         sed_cmd->invert = 1;
539                         cmdstr++;
540
541                         /* skip whitespace before the command */
542                         cmdstr = skip_whitespace(cmdstr);
543                 }
544
545                 /* last part (mandatory) will be a command */
546                 if (!*cmdstr)
547                         bb_error_msg_and_die("missing command");
548                 sed_cmd->cmd = *(cmdstr++);
549                 cmdstr = parse_cmd_args(sed_cmd, cmdstr);
550
551                 /* Add the command to the command array */
552                 bbg.sed_cmd_tail->next = sed_cmd;
553                 bbg.sed_cmd_tail = bbg.sed_cmd_tail->next;
554         }
555
556         /* If we glued multiple lines together, free the memory. */
557         free(bbg.add_cmd_line);
558         bbg.add_cmd_line = NULL;
559 }
560
561 /* Append to a string, reallocating memory as necessary. */
562
563 #define PIPE_GROW 64
564
565 static void pipe_putc(char c)
566 {
567         if (bbg.pipeline.idx == bbg.pipeline.len) {
568                 bbg.pipeline.buf = xrealloc(bbg.pipeline.buf,
569                                 bbg.pipeline.len + PIPE_GROW);
570                 bbg.pipeline.len += PIPE_GROW;
571         }
572         bbg.pipeline.buf[bbg.pipeline.idx++] = c;
573 }
574
575 static void do_subst_w_backrefs(char *line, char *replace)
576 {
577         int i,j;
578
579         /* go through the replacement string */
580         for (i = 0; replace[i]; i++) {
581                 /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
582                 if (replace[i] == '\\') {
583                         unsigned backref = replace[++i] - '0';
584                         if (backref <= 9) {
585                                 /* print out the text held in bbg.regmatch[backref] */
586                                 if (bbg.regmatch[backref].rm_so != -1) {
587                                         j = bbg.regmatch[backref].rm_so;
588                                         while (j < bbg.regmatch[backref].rm_eo)
589                                                 pipe_putc(line[j++]);
590                                 }
591                                 continue;
592                         }
593                         /* I _think_ it is impossible to get '\' to be
594                          * the last char in replace string. Thus we dont check
595                          * for replace[i] == NUL. (counterexample anyone?) */
596                         /* if we find a backslash escaped character, print the character */
597                         pipe_putc(replace[i]);
598                         continue;
599                 }
600                 /* if we find an unescaped '&' print out the whole matched text. */
601                 if (replace[i] == '&') {
602                         j = bbg.regmatch[0].rm_so;
603                         while (j < bbg.regmatch[0].rm_eo)
604                                 pipe_putc(line[j++]);
605                         continue;
606                 }
607                 /* Otherwise just output the character. */
608                 pipe_putc(replace[i]);
609         }
610 }
611
612 static int do_subst_command(sed_cmd_t *sed_cmd, char **line)
613 {
614         char *oldline = *line;
615         int altered = 0;
616         int match_count = 0;
617         regex_t *current_regex;
618
619         /* Handle empty regex. */
620         if (sed_cmd->sub_match == NULL) {
621                 current_regex = bbg.previous_regex_ptr;
622                 if (!current_regex)
623                         bb_error_msg_and_die("no previous regexp");
624         } else
625                 bbg.previous_regex_ptr = current_regex = sed_cmd->sub_match;
626
627         /* Find the first match */
628         if (REG_NOMATCH == regexec(current_regex, oldline, 10, bbg.regmatch, 0))
629                 return 0;
630
631         /* Initialize temporary output buffer. */
632         bbg.pipeline.buf = xmalloc(PIPE_GROW);
633         bbg.pipeline.len = PIPE_GROW;
634         bbg.pipeline.idx = 0;
635
636         /* Now loop through, substituting for matches */
637         do {
638                 int i;
639
640                 /* Work around bug in glibc regexec, demonstrated by:
641                    echo " a.b" | busybox sed 's [^ .]* x g'
642                    The match_count check is so not to break
643                    echo "hi" | busybox sed 's/^/!/g' */
644                 if (!bbg.regmatch[0].rm_so && !bbg.regmatch[0].rm_eo && match_count) {
645                         pipe_putc(*oldline++);
646                         continue;
647                 }
648
649                 match_count++;
650
651                 /* If we aren't interested in this match, output old line to
652                    end of match and continue */
653                 if (sed_cmd->which_match && sed_cmd->which_match != match_count) {
654                         for (i = 0; i < bbg.regmatch[0].rm_eo; i++)
655                                 pipe_putc(*oldline++);
656                         continue;
657                 }
658
659                 /* print everything before the match */
660                 for (i = 0; i < bbg.regmatch[0].rm_so; i++)
661                         pipe_putc(oldline[i]);
662
663                 /* then print the substitution string */
664                 do_subst_w_backrefs(oldline, sed_cmd->string);
665
666                 /* advance past the match */
667                 oldline += bbg.regmatch[0].rm_eo;
668                 /* flag that something has changed */
669                 altered++;
670
671                 /* if we're not doing this globally, get out now */
672                 if (sed_cmd->which_match) break;
673         } while (*oldline && (regexec(current_regex, oldline, 10, bbg.regmatch, 0) != REG_NOMATCH));
674
675         /* Copy rest of string into output pipeline */
676
677         while (*oldline)
678                 pipe_putc(*oldline++);
679         pipe_putc(0);
680
681         free(*line);
682         *line = bbg.pipeline.buf;
683         return altered;
684 }
685
686 /* Set command pointer to point to this label.  (Does not handle null label.) */
687 static sed_cmd_t *branch_to(char *label)
688 {
689         sed_cmd_t *sed_cmd;
690
691         for (sed_cmd = bbg.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
692                 if (sed_cmd->cmd == ':' && sed_cmd->string && !strcmp(sed_cmd->string, label)) {
693                         return sed_cmd;
694                 }
695         }
696         bb_error_msg_and_die("can't find label for jump to '%s'", label);
697 }
698
699 static void append(char *s)
700 {
701         llist_add_to_end(&bbg.append_head, xstrdup(s));
702 }
703
704 static void flush_append(void)
705 {
706         char *data;
707
708         /* Output appended lines. */
709         while ((data = (char *)llist_pop(&bbg.append_head))) {
710                 fprintf(bbg.nonstdout, "%s\n", data);
711                 free(data);
712         }
713 }
714
715 static void add_input_file(FILE *file)
716 {
717         bbg.input_file_list = xrealloc(bbg.input_file_list,
718                         (bbg.input_file_count + 1) * sizeof(FILE *));
719         bbg.input_file_list[bbg.input_file_count++] = file;
720 }
721
722 /* Get next line of input from bbg.input_file_list, flushing append buffer and
723  * noting if we ran out of files without a newline on the last line we read.
724  */
725 enum {
726         NO_EOL_CHAR = 1,
727         LAST_IS_NUL = 2,
728 };
729 static char *get_next_line(char *gets_char)
730 {
731         char *temp = NULL;
732         int len;
733         char gc;
734
735         flush_append();
736
737         /* will be returned if last line in the file
738          * doesn't end with either '\n' or '\0' */
739         gc = NO_EOL_CHAR;
740         while (bbg.current_input_file < bbg.input_file_count) {
741                 FILE *fp = bbg.input_file_list[bbg.current_input_file];
742                 /* Read line up to a newline or NUL byte, inclusive,
743                  * return malloc'ed char[]. length of the chunk read
744                  * is stored in len. NULL if EOF/error */
745                 temp = bb_get_chunk_from_file(fp, &len);
746                 if (temp) {
747                         /* len > 0 here, it's ok to do temp[len-1] */
748                         char c = temp[len-1];
749                         if (c == '\n' || c == '\0') {
750                                 temp[len-1] = '\0';
751                                 gc = c;
752                                 if (c == '\0') {
753                                         int ch = fgetc(fp);
754                                         if (ch != EOF)
755                                                 ungetc(ch, fp);
756                                         else
757                                                 gc = LAST_IS_NUL;
758                                 }
759                         }
760                         /* else we put NO_EOL_CHAR into *gets_char */
761                         break;
762
763                 /* NB: I had the idea of peeking next file(s) and returning
764                  * NO_EOL_CHAR only if it is the *last* non-empty
765                  * input file. But there is a case where this won't work:
766                  * file1: "a woo\nb woo"
767                  * file2: "c no\nd no"
768                  * sed -ne 's/woo/bang/p' input1 input2 => "a bang\nb bang"
769                  * (note: *no* newline after "b bang"!) */
770                 }
771                 /* Close this file and advance to next one */
772                 fclose(fp);
773                 bbg.current_input_file++;
774         }
775         *gets_char = gc;
776         return temp;
777 }
778
779 /* Output line of text. */
780 /* Note:
781  * The tricks with NO_EOL_CHAR and last_puts_char are there to emulate gnu sed.
782  * Without them, we had this:
783  * echo -n thingy >z1
784  * echo -n again >z2
785  * >znull
786  * sed "s/i/z/" z1 z2 znull | hexdump -vC
787  * output:
788  * gnu sed 4.1.5:
789  * 00000000  74 68 7a 6e 67 79 0a 61  67 61 7a 6e              |thzngy.agazn|
790  * bbox:
791  * 00000000  74 68 7a 6e 67 79 61 67  61 7a 6e                 |thzngyagazn|
792  */
793 static void puts_maybe_newline(char *s, FILE *file, char *last_puts_char, char last_gets_char)
794 {
795         char lpc = *last_puts_char;
796
797         /* Need to insert a '\n' between two files because first file's
798          * last line wasn't terminated? */
799         if (lpc != '\n' && lpc != '\0') {
800                 fputc('\n', file);
801                 lpc = '\n';
802         }
803         fputs(s, file);
804
805         /* 'x' - just something which is not '\n', '\0' or NO_EOL_CHAR */
806         if (s[0])
807                 lpc = 'x';
808
809         /* had trailing '\0' and it was last char of file? */
810         if (last_gets_char == LAST_IS_NUL) {
811                 fputc('\0', file);
812                 lpc = 'x'; /* */
813         } else
814         /* had trailing '\n' or '\0'? */
815         if (last_gets_char != NO_EOL_CHAR) {
816                 fputc(last_gets_char, file);
817                 lpc = last_gets_char;
818         }
819
820         if (ferror(file)) {
821                 xfunc_error_retval = 4;  /* It's what gnu sed exits with... */
822                 bb_error_msg_and_die(bb_msg_write_error);
823         }
824         *last_puts_char = lpc;
825 }
826
827 #define sed_puts(s, n) (puts_maybe_newline(s, bbg.nonstdout, &last_puts_char, n))
828
829 /* Process all the lines in all the files */
830
831 static void process_files(void)
832 {
833         char *pattern_space, *next_line;
834         int linenum = 0;
835         char last_puts_char = '\n';
836         char last_gets_char, next_gets_char;
837         sed_cmd_t *sed_cmd;
838         int substituted;
839
840         /* Prime the pump */
841         next_line = get_next_line(&next_gets_char);
842
843         /* go through every line in each file */
844 again:
845         substituted = 0;
846
847         /* Advance to next line.  Stop if out of lines. */
848         pattern_space = next_line;
849         if (!pattern_space) return;
850         last_gets_char = next_gets_char;
851
852         /* Read one line in advance so we can act on the last line,
853          * the '$' address */
854         next_line = get_next_line(&next_gets_char);
855         linenum++;
856 restart:
857         /* for every line, go through all the commands */
858         for (sed_cmd = bbg.sed_cmd_head.next; sed_cmd; sed_cmd = sed_cmd->next) {
859                 int old_matched, matched;
860
861                 old_matched = sed_cmd->in_match;
862
863                 /* Determine if this command matches this line: */
864
865                 /* Are we continuing a previous multi-line match? */
866                 sed_cmd->in_match = sed_cmd->in_match
867                         /* Or is no range necessary? */
868                         || (!sed_cmd->beg_line && !sed_cmd->end_line
869                                 && !sed_cmd->beg_match && !sed_cmd->end_match)
870                         /* Or did we match the start of a numerical range? */
871                         || (sed_cmd->beg_line > 0 && (sed_cmd->beg_line == linenum))
872                         /* Or does this line match our begin address regex? */
873                         || (sed_cmd->beg_match &&
874                             !regexec(sed_cmd->beg_match, pattern_space, 0, NULL, 0))
875                         /* Or did we match last line of input? */
876                         || (sed_cmd->beg_line == -1 && next_line == NULL);
877
878                 /* Snapshot the value */
879
880                 matched = sed_cmd->in_match;
881
882                 /* Is this line the end of the current match? */
883
884                 if (matched) {
885                         sed_cmd->in_match = !(
886                                 /* has the ending line come, or is this a single address command? */
887                                 (sed_cmd->end_line ?
888                                         sed_cmd->end_line == -1 ?
889                                                 !next_line
890                                                 : (sed_cmd->end_line <= linenum)
891                                         : !sed_cmd->end_match
892                                 )
893                                 /* or does this line matches our last address regex */
894                                 || (sed_cmd->end_match && old_matched
895                                      && (regexec(sed_cmd->end_match,
896                                                  pattern_space, 0, NULL, 0) == 0))
897                         );
898                 }
899
900                 /* Skip blocks of commands we didn't match. */
901                 if (sed_cmd->cmd == '{') {
902                         if (sed_cmd->invert ? matched : !matched) {
903                                 while (sed_cmd->cmd != '}') {
904                                         sed_cmd = sed_cmd->next;
905                                         if (!sed_cmd)
906                                                 bb_error_msg_and_die("unterminated {");
907                                 }
908                         }
909                         continue;
910                 }
911
912                 /* Okay, so did this line match? */
913                 if (sed_cmd->invert ? !matched : matched) {
914                         /* Update last used regex in case a blank substitute BRE is found */
915                         if (sed_cmd->beg_match) {
916                                 bbg.previous_regex_ptr = sed_cmd->beg_match;
917                         }
918
919                         /* actual sedding */
920                         switch (sed_cmd->cmd) {
921
922                         /* Print line number */
923                         case '=':
924                                 fprintf(bbg.nonstdout, "%d\n", linenum);
925                                 break;
926
927                         /* Write the current pattern space up to the first newline */
928                         case 'P':
929                         {
930                                 char *tmp = strchr(pattern_space, '\n');
931
932                                 if (tmp) {
933                                         *tmp = '\0';
934                                         /* TODO: explain why '\n' below */
935                                         sed_puts(pattern_space, '\n');
936                                         *tmp = '\n';
937                                         break;
938                                 }
939                                 /* Fall Through */
940                         }
941
942                         /* Write the current pattern space to output */
943                         case 'p':
944                                 /* NB: we print this _before_ the last line
945                                  * (of current file) is printed. Even if
946                                  * that line is nonterminated, we print
947                                  * '\n' here (gnu sed does the same) */
948                                 sed_puts(pattern_space, '\n');
949                                 break;
950                         /* Delete up through first newline */
951                         case 'D':
952                         {
953                                 char *tmp = strchr(pattern_space, '\n');
954
955                                 if (tmp) {
956                                         tmp = xstrdup(tmp+1);
957                                         free(pattern_space);
958                                         pattern_space = tmp;
959                                         goto restart;
960                                 }
961                         }
962                         /* discard this line. */
963                         case 'd':
964                                 goto discard_line;
965
966                         /* Substitute with regex */
967                         case 's':
968                                 if (!do_subst_command(sed_cmd, &pattern_space))
969                                         break;
970                                 substituted |= 1;
971
972                                 /* handle p option */
973                                 if (sed_cmd->sub_p)
974                                         sed_puts(pattern_space, last_gets_char);
975                                 /* handle w option */
976                                 if (sed_cmd->sw_file)
977                                         puts_maybe_newline(
978                                                 pattern_space, sed_cmd->sw_file,
979                                                 &sed_cmd->sw_last_char, last_gets_char);
980                                 break;
981
982                         /* Append line to linked list to be printed later */
983                         case 'a':
984                                 append(sed_cmd->string);
985                                 break;
986
987                         /* Insert text before this line */
988                         case 'i':
989                                 sed_puts(sed_cmd->string, '\n');
990                                 break;
991
992                         /* Cut and paste text (replace) */
993                         case 'c':
994                                 /* Only triggers on last line of a matching range. */
995                                 if (!sed_cmd->in_match)
996                                         sed_puts(sed_cmd->string, NO_EOL_CHAR);
997                                 goto discard_line;
998
999                         /* Read file, append contents to output */
1000                         case 'r':
1001                         {
1002                                 FILE *rfile;
1003
1004                                 rfile = fopen(sed_cmd->string, "r");
1005                                 if (rfile) {
1006                                         char *line;
1007
1008                                         while ((line = xmalloc_getline(rfile))
1009                                                         != NULL)
1010                                                 append(line);
1011                                         xprint_and_close_file(rfile);
1012                                 }
1013
1014                                 break;
1015                         }
1016
1017                         /* Write pattern space to file. */
1018                         case 'w':
1019                                 puts_maybe_newline(
1020                                         pattern_space, sed_cmd->sw_file,
1021                                         &sed_cmd->sw_last_char, last_gets_char);
1022                                 break;
1023
1024                         /* Read next line from input */
1025                         case 'n':
1026                                 if (!bbg.be_quiet)
1027                                         sed_puts(pattern_space, last_gets_char);
1028                                 if (next_line) {
1029                                         free(pattern_space);
1030                                         pattern_space = next_line;
1031                                         last_gets_char = next_gets_char;
1032                                         next_line = get_next_line(&next_gets_char);
1033                                         linenum++;
1034                                         break;
1035                                 }
1036                                 /* fall through */
1037
1038                         /* Quit.  End of script, end of input. */
1039                         case 'q':
1040                                 /* Exit the outer while loop */
1041                                 free(next_line);
1042                                 next_line = NULL;
1043                                 goto discard_commands;
1044
1045                         /* Append the next line to the current line */
1046                         case 'N':
1047                         {
1048                                 int len;
1049                                 /* If no next line, jump to end of script and exit. */
1050                                 if (next_line == NULL) {
1051                                         /* Jump to end of script and exit */
1052                                         free(next_line);
1053                                         next_line = NULL;
1054                                         goto discard_line;
1055                                 /* append next_line, read new next_line. */
1056                                 }
1057                                 len = strlen(pattern_space);
1058                                 pattern_space = realloc(pattern_space, len + strlen(next_line) + 2);
1059                                 pattern_space[len] = '\n';
1060                                 strcpy(pattern_space + len+1, next_line);
1061                                 last_gets_char = next_gets_char;
1062                                 next_line = get_next_line(&next_gets_char);
1063                                 linenum++;
1064                                 break;
1065                         }
1066
1067                         /* Test/branch if substitution occurred */
1068                         case 't':
1069                                 if (!substituted) break;
1070                                 substituted = 0;
1071                                 /* Fall through */
1072                         /* Test/branch if substitution didn't occur */
1073                         case 'T':
1074                                 if (substituted) break;
1075                                 /* Fall through */
1076                         /* Branch to label */
1077                         case 'b':
1078                                 if (!sed_cmd->string) goto discard_commands;
1079                                 else sed_cmd = branch_to(sed_cmd->string);
1080                                 break;
1081                         /* Transliterate characters */
1082                         case 'y':
1083                         {
1084                                 int i, j;
1085
1086                                 for (i = 0; pattern_space[i]; i++) {
1087                                         for (j = 0; sed_cmd->string[j]; j += 2) {
1088                                                 if (pattern_space[i] == sed_cmd->string[j]) {
1089                                                         pattern_space[i] = sed_cmd->string[j + 1];
1090                                                         break;
1091                                                 }
1092                                         }
1093                                 }
1094
1095                                 break;
1096                         }
1097                         case 'g':       /* Replace pattern space with hold space */
1098                                 free(pattern_space);
1099                                 pattern_space = xstrdup(bbg.hold_space ? bbg.hold_space : "");
1100                                 break;
1101                         case 'G':       /* Append newline and hold space to pattern space */
1102                         {
1103                                 int pattern_space_size = 2;
1104                                 int hold_space_size = 0;
1105
1106                                 if (pattern_space)
1107                                         pattern_space_size += strlen(pattern_space);
1108                                 if (bbg.hold_space)
1109                                         hold_space_size = strlen(bbg.hold_space);
1110                                 pattern_space = xrealloc(pattern_space,
1111                                                 pattern_space_size + hold_space_size);
1112                                 if (pattern_space_size == 2)
1113                                         pattern_space[0] = 0;
1114                                 strcat(pattern_space, "\n");
1115                                 if (bbg.hold_space)
1116                                         strcat(pattern_space, bbg.hold_space);
1117                                 last_gets_char = '\n';
1118
1119                                 break;
1120                         }
1121                         case 'h':       /* Replace hold space with pattern space */
1122                                 free(bbg.hold_space);
1123                                 bbg.hold_space = xstrdup(pattern_space);
1124                                 break;
1125                         case 'H':       /* Append newline and pattern space to hold space */
1126                         {
1127                                 int hold_space_size = 2;
1128                                 int pattern_space_size = 0;
1129
1130                                 if (bbg.hold_space)
1131                                         hold_space_size += strlen(bbg.hold_space);
1132                                 if (pattern_space)
1133                                         pattern_space_size = strlen(pattern_space);
1134                                 bbg.hold_space = xrealloc(bbg.hold_space,
1135                                                 hold_space_size + pattern_space_size);
1136
1137                                 if (hold_space_size == 2)
1138                                         *bbg.hold_space = 0;
1139                                 strcat(bbg.hold_space, "\n");
1140                                 if (pattern_space)
1141                                         strcat(bbg.hold_space, pattern_space);
1142
1143                                 break;
1144                         }
1145                         case 'x': /* Exchange hold and pattern space */
1146                         {
1147                                 char *tmp = pattern_space;
1148                                 pattern_space = bbg.hold_space ? : xzalloc(1);
1149                                 last_gets_char = '\n';
1150                                 bbg.hold_space = tmp;
1151                                 break;
1152                         }
1153                         }
1154                 }
1155         }
1156
1157         /*
1158          * exit point from sedding...
1159          */
1160  discard_commands:
1161         /* we will print the line unless we were told to be quiet ('-n')
1162            or if the line was suppressed (ala 'd'elete) */
1163         if (!bbg.be_quiet)
1164                 sed_puts(pattern_space, last_gets_char);
1165
1166         /* Delete and such jump here. */
1167  discard_line:
1168         flush_append();
1169         free(pattern_space);
1170
1171         goto again;
1172 }
1173
1174 /* It is possible to have a command line argument with embedded
1175    newlines.  This counts as multiple command lines. */
1176
1177 static void add_cmd_block(char *cmdstr)
1178 {
1179         int go = 1;
1180         char *temp = xstrdup(cmdstr), *temp2 = temp;
1181
1182         while (go) {
1183                 int len = strcspn(temp2, "\n");
1184                 if (!temp2[len]) go = 0;
1185                 else temp2[len] = 0;
1186                 add_cmd(temp2);
1187                 temp2 += len+1;
1188         }
1189         free(temp);
1190 }
1191
1192 static void add_cmds_link(llist_t *opt_e)
1193 {
1194         if (!opt_e) return;
1195         add_cmds_link(opt_e->link);
1196         add_cmd_block(opt_e->data);
1197         free(opt_e);
1198 }
1199
1200 static void add_files_link(llist_t *opt_f)
1201 {
1202         char *line;
1203         FILE *cmdfile;
1204         if (!opt_f) return;
1205         add_files_link(opt_f->link);
1206         cmdfile = xfopen(opt_f->data, "r");
1207         while ((line = xmalloc_getline(cmdfile)) != NULL) {
1208                 add_cmd(line);
1209                 free(line);
1210         }
1211         xprint_and_close_file(cmdfile);
1212         free(opt_f);
1213 }
1214
1215 int sed_main(int argc, char **argv)
1216 {
1217         enum {
1218                 OPT_in_place = 1 << 0,
1219         };
1220         unsigned opt;
1221         llist_t *opt_e, *opt_f;
1222         int status = EXIT_SUCCESS;
1223
1224         bbg.sed_cmd_tail = &bbg.sed_cmd_head;
1225
1226         /* destroy command strings on exit */
1227         if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);
1228
1229         /* Lie to autoconf when it starts asking stupid questions. */
1230         if (argc == 2 && !strcmp(argv[1], "--version")) {
1231                 puts("This is not GNU sed version 4.0");
1232                 return 0;
1233         }
1234
1235         /* do normal option parsing */
1236         opt_e = opt_f = NULL;
1237         opt_complementary = "e::f::" /* can occur multiple times */
1238                             "nn"; /* count -n */
1239         opt = getopt32(argc, argv, "irne:f:", &opt_e, &opt_f,
1240                             &bbg.be_quiet); /* counter for -n */
1241         argc -= optind;
1242         argv += optind;
1243         if (opt & OPT_in_place) { // -i
1244                 atexit(cleanup_outname);
1245         }
1246         if (opt & 0x2) bbg.regex_type |= REG_EXTENDED; // -r
1247         //if (opt & 0x4) bbg.be_quiet++; // -n
1248         if (opt & 0x8) { // -e
1249                 /* getopt32 reverses order of arguments, handle it */
1250                 add_cmds_link(opt_e);
1251         }
1252         if (opt & 0x10) { // -f
1253                 /* getopt32 reverses order of arguments, handle it */
1254                 add_files_link(opt_f);
1255         }
1256         /* if we didn't get a pattern from -e or -f, use argv[0] */
1257         if (!(opt & 0x18)) {
1258                 if (!argc)
1259                         bb_show_usage();
1260                 add_cmd_block(*argv++);
1261                 argc--;
1262         }
1263         /* Flush any unfinished commands. */
1264         add_cmd("");
1265
1266         /* By default, we write to stdout */
1267         bbg.nonstdout = stdout;
1268
1269         /* argv[0..(argc-1)] should be names of file to process. If no
1270          * files were specified or '-' was specified, take input from stdin.
1271          * Otherwise, we process all the files specified. */
1272         if (argv[0] == NULL) {
1273                 if (opt & OPT_in_place)
1274                         bb_error_msg_and_die(bb_msg_requires_arg, "-i");
1275                 add_input_file(stdin);
1276                 process_files();
1277         } else {
1278                 int i;
1279                 FILE *file;
1280
1281                 for (i = 0; i < argc; i++) {
1282                         struct stat statbuf;
1283                         int nonstdoutfd;
1284
1285                         if (LONE_DASH(argv[i]) && !(opt & OPT_in_place)) {
1286                                 add_input_file(stdin);
1287                                 process_files();
1288                                 continue;
1289                         }
1290                         file = fopen_or_warn(argv[i], "r");
1291                         if (!file) {
1292                                 status = EXIT_FAILURE;
1293                                 continue;
1294                         }
1295                         if (!(opt & OPT_in_place)) {
1296                                 add_input_file(file);
1297                                 continue;
1298                         }
1299
1300                         bbg.outname = xasprintf("%sXXXXXX", argv[i]);
1301                         nonstdoutfd = mkstemp(bbg.outname);
1302                         if (-1 == nonstdoutfd)
1303                                 bb_error_msg_and_die("no temp file");
1304                         bbg.nonstdout = fdopen(nonstdoutfd, "w");
1305
1306                         /* Set permissions of output file */
1307
1308                         fstat(fileno(file), &statbuf);
1309                         fchmod(nonstdoutfd, statbuf.st_mode);
1310                         add_input_file(file);
1311                         process_files();
1312                         fclose(bbg.nonstdout);
1313
1314                         bbg.nonstdout = stdout;
1315                         /* unlink(argv[i]); */
1316                         // FIXME: error check / message?
1317                         rename(bbg.outname, argv[i]);
1318                         free(bbg.outname);
1319                         bbg.outname = 0;
1320                 }
1321                 if (bbg.input_file_count > bbg.current_input_file)
1322                         process_files();
1323         }
1324
1325         return status;
1326 }