aboutsummaryrefslogtreecommitdiff
path: root/toys/posix/sed.c
blob: d9bb5c94a788efc6695926cd70e92f9b30330488 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
/* sed.c - stream editor. Thing that does s/// and other stuff.
 *
 * Copyright 2014 Rob Landley <rob@landley.net>
 *
 * See http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sed.html
 *
 * TODO: lines > 2G could wrap signed int length counters. Not just getline()
 * but N and s///
 * TODO: make y// handle unicode, unicode delimiters
 * TODO: handle error return from emit(), error_msg/exit consistently
 *       What's the right thing to do for -i when write fails? Skip to next?
 * test '//q' with no previous regex, also repeat previous regex?

USE_SED(NEWTOY(sed, "(help)(version)e*f*i:;nErz(null-data)[+Er]", TOYFLAG_BIN|TOYFLAG_LOCALE|TOYFLAG_NOHELP))

config SED
  bool "sed"
  default y
  help
    usage: sed [-inrzE] [-e SCRIPT]...|SCRIPT [-f SCRIPT_FILE]... [FILE...]

    Stream editor. Apply one or more editing SCRIPTs to each line of input
    (from FILE or stdin) producing output (by default to stdout).

    -e	Add SCRIPT to list
    -f	Add contents of SCRIPT_FILE to list
    -i	Edit each file in place (-iEXT keeps backup file with extension EXT)
    -n	No default output (use the p command to output matched lines)
    -r	Use extended regular expression syntax
    -E	POSIX alias for -r
    -s	Treat input files separately (implied by -i)
    -z	Use \0 rather than \n as the input line separator

    A SCRIPT is a series of one or more COMMANDs separated by newlines or
    semicolons. All -e SCRIPTs are concatenated together as if separated
    by newlines, followed by all lines from -f SCRIPT_FILEs, in order.
    If no -e or -f SCRIPTs are specified, the first argument is the SCRIPT.

    Each COMMAND may be preceded by an address which limits the command to
    apply only to the specified line(s). Commands without an address apply to
    every line. Addresses are of the form:

      [ADDRESS[,ADDRESS]]COMMAND

    The ADDRESS may be a decimal line number (starting at 1), a /regular
    expression/ within a pair of forward slashes, or the character "$" which
    matches the last line of input. (In -s or -i mode this matches the last
    line of each file, otherwise just the last line of the last file.) A single
    address matches one line, a pair of comma separated addresses match
    everything from the first address to the second address (inclusive). If
    both addresses are regular expressions, more than one range of lines in
    each file can match.

    REGULAR EXPRESSIONS in sed are started and ended by the same character
    (traditionally / but anything except a backslash or a newline works).
    Backslashes may be used to escape the delimiter if it occurs in the
    regex, and for the usual printf escapes (\abcefnrtv and octal, hex,
    and unicode). An empty regex repeats the previous one. ADDRESS regexes
    (above) require the first delimiter to be escaped with a backslash when
    it isn't a forward slash (to distinguish it from the COMMANDs below).

    Sed mostly operates on individual lines one at a time. It reads each line,
    processes it, and either writes it to the output or discards it before
    reading the next line. Sed can remember one additional line in a separate
    buffer (using the h, H, g, G, and x commands), and can read the next line
    of input early (using the n and N command), but other than that command
    scripts operate on individual lines of text.

    Each COMMAND starts with a single character. The following commands take
    no arguments:

      {  Start a new command block, continuing until a corresponding "}".
         Command blocks may nest. If the block has an address, commands within
         the block are only run for lines within the block's address range.

      }  End command block (this command cannot have an address)

      d  Delete this line and move on to the next one
         (ignores remaining COMMANDs)

      D  Delete one line of input and restart command SCRIPT (same as "d"
         unless you've glued lines together with "N" or similar)

      g  Get remembered line (overwriting current line)

      G  Get remembered line (appending to current line)

      h  Remember this line (overwriting remembered line)

      H  Remember this line (appending to remembered line, if any)

      l  Print line, escaping \abfrtv (but not newline), octal escaping other
         nonprintable characters, wrapping lines to terminal width with a
         backslash, and appending $ to actual end of line.

      n  Print default output and read next line, replacing current line
         (If no next line available, quit processing script)

      N  Append next line of input to this line, separated by a newline
         (This advances the line counter for address matching and "=", if no
         next line available quit processing script without default output)

      p  Print this line

      P  Print this line up to first newline (from "N")

      q  Quit (print default output, no more commands processed or lines read)

      x  Exchange this line with remembered line (overwrite in both directions)

      =  Print the current line number (followed by a newline)

    The following commands (may) take an argument. The "text" arguments (to
    the "a", "b", and "c" commands) may end with an unescaped "\" to append
    the next line (for which leading whitespace is not skipped), and also
    treat ";" as a literal character (use "\;" instead).

      a [text]   Append text to output before attempting to read next line

      b [label]  Branch, jumps to :label (or with no label, to end of SCRIPT)

      c [text]   Delete line, output text at end of matching address range
                 (ignores remaining COMMANDs)

      i [text]   Print text

      r [file]   Append contents of file to output before attempting to read
                 next line.

      s/S/R/F    Search for regex S, replace matched text with R using flags F.
                 The first character after the "s" (anything but newline or
                 backslash) is the delimiter, escape with \ to use normally.

                 The replacement text may contain "&" to substitute the matched
                 text (escape it with backslash for a literal &), or \1 through
                 \9 to substitute a parenthetical subexpression in the regex.
                 You can also use the normal backslash escapes such as \n and
                 a backslash at the end of the line appends the next line.

                 The flags are:

                 [0-9]    A number, substitute only that occurrence of pattern
                 g        Global, substitute all occurrences of pattern
                 i        Ignore case when matching
                 p        Print the line if match was found and replaced
                 w [file] Write (append) line to file if match replaced

      t [label]  Test, jump to :label only if an "s" command found a match in
                 this line since last test (replacing with same text counts)

      T [label]  Test false, jump only if "s" hasn't found a match.

      w [file]   Write (append) line to file

      y/old/new/ Change each character in 'old' to corresponding character
                 in 'new' (with standard backslash escapes, delimiter can be
                 any repeated character except \ or \n)

      : [label]  Labeled target for jump commands

      #  Comment, ignore rest of this line of SCRIPT

    Deviations from POSIX: allow extended regular expressions with -r,
    editing in place with -i, separate with -s, NUL-separated input with -z,
    printf escapes in text, line continuations, semicolons after all commands,
    2-address anywhere an address is allowed, "T" command, multiline
    continuations for [abc], \; to end [abc] argument before end of line.
*/

#define FOR_sed
#include "toys.h"

GLOBALS(
  char *i;
  struct arg_list *f, *e;

  // processed pattern list
  struct double_list *pattern;

  char *nextline, *remember;
  void *restart, *lastregex;
  long nextlen, rememberlen, count;
  int fdout, noeol;
  unsigned xx;
  char delim;
)

// Linked list of parsed sed commands. Offset fields indicate location where
// regex or string starts, ala offset+(char *)struct, because we remalloc()
// these to expand them for multiline inputs, and pointers would have to be
// individually adjusted.

struct sedcmd {
  struct sedcmd *next, *prev;

  // Begin and end of each match
  long lmatch[2]; // line number of match
  int rmatch[2];  // offset of regex struct for prefix matches (/abc/,/def/p)
  int arg1, arg2, w; // offset of two arguments per command, plus s//w filename
  unsigned not, hit;
  unsigned sflags; // s///flag bits: i=1, g=2, p=4
  char c; // action
};

// Write out line with potential embedded NUL, handling eol/noeol
static int emit(char *line, long len, int eol)
{
  int l, old = line[len];

  if (TT.noeol && !writeall(TT.fdout, "\n", 1)) return 1;
  TT.noeol = !eol;
  if (eol) line[len++] = '\n';
  if (!len) return 0;
  l = writeall(TT.fdout, line, len);
  if (eol) line[len-1] = old;
  if (l != len) {
    perror_msg("short write");

    return 1;
  }

  return 0;
}

// Extend allocation to include new string, with newline between if newlen<0

static char *extend_string(char **old, char *new, int oldlen, int newlen)
{
  int newline = newlen < 0;
  char *s;

  if (newline) newlen = -newlen;
  s = *old = xrealloc(*old, oldlen+newlen+newline+1);
  if (newline) s[oldlen++] = '\n';
  memcpy(s+oldlen, new, newlen);
  s[oldlen+newlen] = 0;

  return s+oldlen+newlen+1;
}

// An empty regex repeats the previous one
static void *get_regex(void *trump, int offset)
{
  if (!offset) {
    if (!TT.lastregex) error_exit("no previous regex");
    return TT.lastregex;
  }

  return TT.lastregex = offset+(char *)trump;
}

// Apply pattern to line from input file
static void sed_line(char **pline, long plen)
{
  struct append {
    struct append *next, *prev;
    int file;
    char *str;
  } *append = 0;
  char *line = TT.nextline;
  long len = TT.nextlen;
  struct sedcmd *command;
  int eol = 0, tea = 0;

  // Ignore EOF for all files before last unless -i
  if (!pline && !FLAG(i)) return;

  // Grab next line for deferred processing (EOF detection: we get a NULL
  // pline at EOF to flush last line). Note that only end of _last_ input
  // file matches $ (unless we're doing -i).
  TT.nextline = 0;
  TT.nextlen = 0;
  if (pline) {
    TT.nextline = *pline;
    TT.nextlen = plen;
    *pline = 0;
  }

  if (!line || !len) return;
  if (line[len-1] == '\n') line[--len] = eol++;
  TT.count++;

  // The restart-1 is because we added one to make sure it wasn't NULL,
  // otherwise N as last command would restart script
  command = TT.restart ? ((struct sedcmd *)TT.restart)-1 : (void *)TT.pattern;
  TT.restart = 0;

  while (command) {
    char *str, c = command->c;

    // Have we got a line or regex matching range for this rule?
    if (*command->lmatch || *command->rmatch) {
      int miss = 0;
      long lm;

      // In a match that might end?
      if (command->hit) {
        if (!(lm = command->lmatch[1])) {
          if (!command->rmatch[1]) command->hit = 0;
          else {
            void *rm = get_regex(command, command->rmatch[1]);

            // regex match end includes matching line, so defer deactivation
            if (line && !regexec0(rm, line, len, 0, 0, 0)) miss = 1;
          }
        } else if (lm > 0 && lm < TT.count) command->hit = 0;

      // Start a new match?
      } else {
        if (!(lm = *command->lmatch)) {
          void *rm = get_regex(command, *command->rmatch);

          if (line && !regexec0(rm, line, len, 0, 0, 0)) command->hit++;
        } else if (lm == TT.count || (lm == -1 && !pline)) command->hit++;

        if (!command->lmatch[1] && !command->rmatch[1]) miss = 1;
      } 

      // Didn't match?
      lm = !(command->hit ^ command->not);

      // Deferred disable from regex end match
      if (miss || command->lmatch[1] == TT.count) command->hit = 0;

      if (lm) {
        // Handle skipping curly bracket command group
        if (c == '{') {
          int curly = 1;

          while (curly) {
            command = command->next;
            if (command->c == '{') curly++;
            if (command->c == '}') curly--;
          }
        }
        command = command->next;
        continue;
      }
    }

    // A deleted line can still update line match state for later commands
    if (!line) {
      command = command->next;
      continue;
    }

    // Process command

    if (c=='a' || c=='r') {
      struct append *a = xzalloc(sizeof(struct append));
      if (command->arg1) a->str = command->arg1+(char *)command;
      a->file = c=='r';
      dlist_add_nomalloc((void *)&append, (void *)a);
    } else if (c=='b' || c=='t' || c=='T') {
      int t = tea;

      if (c != 'b') tea = 0;
      if (c=='b' || t^(c=='T')) {
        if (!command->arg1) break;
        str = command->arg1+(char *)command;
        for (command = (void *)TT.pattern; command; command = command->next)
          if (command->c == ':' && !strcmp(command->arg1+(char *)command, str))
            break;
        if (!command) error_exit("no :%s", str);
      }
    } else if (c=='c') {
      str = command->arg1+(char *)command;
      if (!command->hit) emit(str, strlen(str), 1);
      free(line);
      line = 0;
      continue;
    } else if (c=='d') {
      free(line);
      line = 0;
      continue;
    } else if (c=='D') {
      // Delete up to \n or end of buffer
      str = line;
      while ((str-line)<len) if (*(str++) == '\n') break;
      len -= str - line;
      memmove(line, str, len);

      // if "delete" blanks line, disable further processing
      // otherwise trim and restart script
      if (!len) {
        free(line);
        line = 0;
      } else {
        line[len] = 0;
        command = (void *)TT.pattern;
      }
      continue;
    } else if (c=='g') {
      free(line);
      line = xstrdup(TT.remember);
      len = TT.rememberlen;
    } else if (c=='G') {
      line = xrealloc(line, len+TT.rememberlen+2);
      line[len++] = '\n';
      memcpy(line+len, TT.remember, TT.rememberlen);
      line[len += TT.rememberlen] = 0;
    } else if (c=='h') {
      free(TT.remember);
      TT.remember = xstrdup(line);
      TT.rememberlen = len;
    } else if (c=='H') {
      TT.remember = xrealloc(TT.remember, TT.rememberlen+len+2);
      TT.remember[TT.rememberlen++] = '\n';
      memcpy(TT.remember+TT.rememberlen, line, len);
      TT.remember[TT.rememberlen += len] = 0;
    } else if (c=='i') {
      str = command->arg1+(char *)command;
      emit(str, strlen(str), 1);
    } else if (c=='l') {
      int i, x, off;

      if (!TT.xx) {
        terminal_size(&TT.xx, 0);
        if (!TT.xx) TT.xx = 80;
        if (TT.xx > sizeof(toybuf)-10) TT.xx = sizeof(toybuf)-10;
        if (TT.xx > 4) TT.xx -= 4;
      }

      for (i = off = 0; i<len; i++) {
        if (off >= TT.xx) {
          toybuf[off++] = '\\';
          emit(toybuf, off, 1);
          off = 0;
        }
        x = stridx("\\\a\b\f\r\t\v", line[i]);
        if (x != -1) {
          toybuf[off++] = '\\';
          toybuf[off++] = "\\abfrtv"[x];
        } else if (line[i] >= ' ') toybuf[off++] = line[i];
        else off += sprintf(toybuf+off, "\\%03o", line[i]);
      }
      toybuf[off++] = '$';
      emit(toybuf, off, 1);
    } else if (c=='n') {
      TT.restart = command->next+1;

      break;
    } else if (c=='N') {
      // Can't just grab next line because we could have multiple N and
      // we need to actually read ahead to get N;$p EOF detection right.
      if (pline) {
        TT.restart = command->next+1;
        extend_string(&line, TT.nextline, len, -TT.nextlen);
        free(TT.nextline);
        TT.nextline = line;
        TT.nextlen += len + 1;
        line = 0;
      }

      // Pending append goes out right after N
      goto done; 
    } else if (c=='p' || c=='P') {
      char *l = (c=='P') ? strchr(line, '\n') : 0;

      if (emit(line, l ? l-line : len, eol)) break;
    } else if (c=='q') {
      if (pline) *pline = (void *)1;
      free(TT.nextline);
      TT.nextline = 0;
      TT.nextlen = 0;

      break;
    } else if (c=='s') {
      char *rline = line, *new = command->arg2 + (char *)command, *l2 = 0;
      regmatch_t *match = (void *)toybuf;
      regex_t *reg = get_regex(command, command->arg1);
      int mflags = 0, count = 0, l2used = 0, zmatch = 1, l2l = len,
        mlen, off, newlen;

      // Loop finding match in remaining line (up to remaining len)
      while (!regexec0(reg, rline, len-(rline-line), 10, match, mflags)) {
        mflags = REG_NOTBOL;

        // Zero length matches don't count immediately after a previous match
        mlen = match[0].rm_eo-match[0].rm_so;
        if (!mlen && !zmatch) {
          if (rline-line == len) break;
          l2[l2used++] = *rline++;
          zmatch++;
          continue;
        } else zmatch = 0;

        // If we're replacing only a specific match, skip if this isn't it
        off = command->sflags>>3;
        if (off && off != ++count) {
          memcpy(l2+l2used, rline, match[0].rm_eo);
          l2used += match[0].rm_eo;
          rline += match[0].rm_eo;

          continue;
        }
        // The fact getline() can allocate unbounded amounts of memory is
        // a bigger issue, but while we're here check for integer overflow
        if (match[0].rm_eo > INT_MAX) perror_exit(0);

        // newlen = strlen(new) but with \1 and & and printf escapes
        for (off = newlen = 0; new[off]; off++) {
          int cc = -1;

          if (new[off] == '&') cc = 0;
          else if (new[off] == '\\') cc = new[++off] - '0';
          if (cc < 0 || cc > 9) {
            newlen++;
            continue;
          }
          newlen += match[cc].rm_eo-match[cc].rm_so;
        }

        // Copy changed data to new string

        // Adjust allocation size of new string, copy data we know we'll keep
        l2l += newlen-mlen;
        if (newlen>mlen || !l2) l2 = xrealloc(l2, l2l+1);
        if (match[0].rm_so) {
          memcpy(l2+l2used, rline, match[0].rm_so);
          l2used += match[0].rm_so;
        }

        // copy in new replacement text
        for (off = mlen = 0; new[off]; off++) {
          int cc = 0, ll;

          if (new[off] == '\\') {
            cc = new[++off] - '0';
            if (cc<0 || cc>9) {
              if (!(l2[l2used+mlen++] = unescape(new[off])))
                l2[l2used+mlen-1] = new[off];

              continue;
            } else if (cc > reg->re_nsub) error_exit("no s//\\%d/", cc);
          } else if (new[off] != '&') {
            l2[l2used+mlen++] = new[off];

            continue;
          }

          if (match[cc].rm_so != -1) {
            ll = match[cc].rm_eo-match[cc].rm_so;
            memcpy(l2+l2used+mlen, rline+match[cc].rm_so, ll);
            mlen += ll;
          }
        }
        l2used += newlen;
        rline += match[0].rm_eo;

        // Stop after first substitution unless we have flag g
        if (!(command->sflags & 2)) break;
      }

      // If we made any changes, finish off l2 and swap it for line
      if (l2) {
        // grab trailing unmatched data and null terminator, swap with original
        mlen = len-(rline-line);
        memcpy(l2+l2used, rline, mlen+1);
        len = l2used + mlen;
        free(line);
        line = l2;
      }

      if (mflags) {
        // flag p
        if (command->sflags & 4) emit(line, len, eol);

        tea = 1;
        if (command->w) goto writenow;
      }
    } else if (c=='w') {
      int fd, noeol;
      char *name;

writenow:
      // Swap out emit() context
      fd = TT.fdout;
      noeol = TT.noeol;

      // We save filehandle and newline status before filename
      name = command->w + (char *)command;
      memcpy(&TT.fdout, name, 4);
      name += 4;
      TT.noeol = *(name++);

      // write, then save/restore context
      if (emit(line, len, eol))
        perror_exit("w '%s'", command->arg1+(char *)command);
      *(--name) = TT.noeol;
      TT.noeol = noeol;
      TT.fdout = fd;
    } else if (c=='x') {
      long swap = TT.rememberlen;

      str = TT.remember;
      TT.remember = line;
      line = str;
      TT.rememberlen = len;
      len = swap;
    } else if (c=='y') {
      char *from, *to = (char *)command;
      int i, j;

      from = to+command->arg1;
      to += command->arg2;

      for (i = 0; i < len; i++) {
        j = stridx(from, line[i]);
        if (j != -1) line[i] = to[j];
      }
    } else if (c=='=') {
      sprintf(toybuf, "%ld", TT.count);
      if (emit(toybuf, strlen(toybuf), 1)) break;
    }

    command = command->next;
  }

  if (line && !FLAG(n)) emit(line, len, eol);

done:
  if (dlist_terminate(append)) while (append) {
    struct append *a = append->next;

    if (append->file) {
      int fd = open(append->str, O_RDONLY);

      // Force newline if noeol pending
      if (fd != -1) {
        if (TT.noeol) xwrite(TT.fdout, "\n", 1);
        TT.noeol = 0;
        xsendfile(fd, TT.fdout);
        close(fd);
      }
    } else if (append->str) emit(append->str, strlen(append->str), 1);
    else emit(line, 0, 0);
    free(append);
    append = a;
  }
  free(line);
}

// Callback called on each input file
static void do_sed_file(int fd, char *name)
{
  char *tmp;

  if (FLAG(i)) {
    struct sedcmd *command;

    if (!fd) return error_msg("-i on stdin");
    TT.fdout = copy_tempfile(fd, name, &tmp);
    TT.count = 0;
    for (command = (void *)TT.pattern; command; command = command->next)
      command->hit = 0;
  }
  do_lines(fd, TT.delim, sed_line);
  if (FLAG(i)) {
    if (TT.i && *TT.i) {
      char *s = xmprintf("%s%s", name, TT.i);

      xrename(name, s);
      free(s);
    }
    replace_tempfile(-1, TT.fdout, &tmp);
    TT.fdout = 1;
    TT.nextline = 0;
    TT.nextlen = TT.noeol = 0;
  }
}

// Copy chunk of string between two delimiters, converting printf escapes.
// returns processed copy of string (0 if error), *pstr advances to next
// unused char. if delim (or *delim) is 0 uses/saves starting char as delimiter
// if regxex, ignore delimiter in [ranges]
static char *unescape_delimited_string(char **pstr, char *delim)
{
  char *to, *from, mode = 0, d;

  // Grab leading delimiter (if necessary), allocate space for new string
  from = *pstr;
  if (!delim || !*delim) {
    if (!(d = *(from++))) return 0;
    if (d == '\\') d = *(from++);
    if (!d || d == '\\') return 0;
    if (delim) *delim = d;
  } else d = *delim;
  to = delim = xmalloc(strlen(*pstr)+1);

  while (mode || *from != d) {
    if (!*from) return 0;

    // delimiter in regex character range doesn't count
    if (*from == '[') {
      if (!mode) {
        mode = ']';
        if (from[1]=='-' || from[1]==']') *(to++) = *(from++);
      } else if (mode == ']' && strchr(".=:", from[1])) {
        *(to++) = *(from++);
        mode = *from;
      }
    } else if (*from == mode) {
      if (mode == ']') mode = 0;
      else {
        *(to++) = *(from++);
        mode = ']';
      }
    // Length 1 range (X-X with same X) is "undefined" and makes regcomp err,
    // but the perl build does it, so we need to filter it out.
    } else if (mode && *from == '-' && from[-1] == from[1]) {
      from+=2;
      continue;
    } else if (*from == '\\') {
      if (!from[1]) return 0;

      // Check escaped end delimiter before printf style escapes.
      if (from[1] == d) from++;
      else if (from[1]=='\\') *(to++) = *(from++);
      else {
        char c = unescape(from[1]);

        if (c) {
          *(to++) = c;
          from+=2;
          continue;
        } else if (!mode) *(to++) = *(from++);
      }
    }
    *(to++) = *(from++);
  }
  *to = 0;
  *pstr = from+1;

  return delim;
}

// Translate pattern strings into command structures. Each command structure
// is a single allocation (which requires some math and remalloc at times).
static void parse_pattern(char **pline, long len)
{
  struct sedcmd *command = (void *)TT.pattern;
  char *line, *reg, c, *errstart;
  int i;

  line = errstart = pline ? *pline : "";
  if (len && line[len-1]=='\n') line[--len] = 0;

  // Append this line to previous multiline command? (hit indicates type.)
  // During parsing "hit" stores data about line continuations, but in
  // sed_line() it means the match range attached to this command
  // is active, so processing the continuation must zero it again.
  if (command && command->prev->hit) {
    // Remove half-finished entry from list so remalloc() doesn't confuse it
    TT.pattern = TT.pattern->prev;
    command = dlist_pop(&TT.pattern);
    c = command->c;
    reg = (char *)command;
    reg += command->arg1 + strlen(reg + command->arg1);

    // Resume parsing for 'a' or 's' command. (Only two that can do this.)
    // TODO: using 256 to indicate 'a' means our s/// delimiter can't be
    // a unicode character.
    if (command->hit < 256) goto resume_s;
    else goto resume_a;
  }

  // Loop through commands in this line.

  command = 0;
  for (;;) {
    if (command) dlist_add_nomalloc(&TT.pattern, (void *)command);

    // If there's no more data on this line, return.
    for (;;) {
      while (isspace(*line) || *line == ';') line++;
      if (*line == '#') while (*line && *line != '\n') line++;
      else break;
    }
    if (!*line) return;

    // We start by writing data into toybuf. Later we'll allocate the
    // ex

    errstart = line;
    memset(toybuf, 0, sizeof(struct sedcmd));
    command = (void *)toybuf;
    reg = toybuf + sizeof(struct sedcmd);

    // Parse address range (if any)
    for (i = 0; i < 2; i++) {
      if (*line == ',') line++;
      else if (i) break;

      if (isdigit(*line)) command->lmatch[i] = strtol(line, &line, 0);
      else if (*line == '$') {
        command->lmatch[i] = -1;
        line++;
      } else if (*line == '/' || *line == '\\') {
        char *s = line;

        if (!(s = unescape_delimited_string(&line, 0))) goto error;
        if (!*s) command->rmatch[i] = 0;
        else {
          xregcomp((void *)reg, s, REG_EXTENDED*!!FLAG(r));
          command->rmatch[i] = reg-toybuf;
          reg += sizeof(regex_t);
        }
        free(s);
      } else break;
    }

    while (isspace(*line)) line++;
    if (!*line) break;

    while (*line == '!') {
      command->not = 1;
      line++;
    }
    while (isspace(*line)) line++;

    c = command->c = *(line++);
    if (strchr("}:", c) && i) break;
    if (strchr("aiqr=", c) && i>1) break;

    // Add step to pattern
    command = xmemdup(toybuf, reg-toybuf);
    reg = (reg-toybuf) + (char *)command;

    // Parse arguments by command type
    if (c == '{') TT.nextlen++;
    else if (c == '}') {
      if (!TT.nextlen--) break;
    } else if (c == 's') {
      char *end, delim = 0;

      // s/pattern/replacement/flags

      // line continuations use arg1 (back at the start of the function),
      // so let's fill out arg2 first (since the regex part can't be multiple
      // lines) and swap them back later.

      // get pattern (just record, we parse it later)
      command->arg2 = reg - (char *)command;
      if (!(TT.remember = unescape_delimited_string(&line, &delim)))
        goto error;

      reg += sizeof(regex_t);
      command->arg1 = reg-(char *)command;
      command->hit = delim;
resume_s:
      // get replacement - don't replace escapes yet because \1 and \& need
      // processing later, after we replace \\ with \ we can't tell \\1 from \1
      end = line;
      while (*end != command->hit) {
        if (!*end) goto error;
        if (*end++ == '\\') {
          if (!*end || *end == '\n') {
            end[-1] = '\n';
            break;
          }
          end++;
        }
      }

      reg = extend_string((void *)&command, line, reg-(char *)command,end-line);
      line = end;
      // line continuation? (note: '\n' can't be a valid delim).
      if (*line == command->hit) command->hit = 0;
      else {
        if (!*line) continue;
        reg--;
        line++;
        goto resume_s;
      }

      // swap arg1/arg2 so they're back in order arguments occur.
      i = command->arg1;
      command->arg1 = command->arg2;
      command->arg2 = i;

      // get flags
      for (line++; *line; line++) {
        long l;

        if (isspace(*line) && *line != '\n') continue;

        if (0 <= (l = stridx("igp", *line))) command->sflags |= 1<<l;
        else if (!(command->sflags>>3) && 0<(l = strtol(line, &line, 10))) {
          command->sflags |= l << 3;
          line--;
        } else break;
      }

      // We deferred actually parsing the regex until we had the s///i flag
      // allocating the space was done by extend_string() above
      if (!*TT.remember) command->arg1 = 0;
      else xregcomp((void *)(command->arg1 + (char *)command), TT.remember,
        (REG_EXTENDED*!!FLAG(r))|((command->sflags&1)*REG_ICASE));
      free(TT.remember);
      TT.remember = 0;
      if (*line == 'w') {
        line++;
        goto writenow;
      }
    } else if (c == 'w') {
      int fd, delim;
      char *cc;

      // Since s/// uses arg1 and arg2, and w needs a persistent filehandle and
      // eol status, and to retain the filename for error messages, we'd need
      // to go up to arg5 just for this. Compromise: dynamically allocate the
      // filehandle and eol status.

writenow:
      while (isspace(*line)) line++;
      if (!*line) goto error;
      for (cc = line; *cc; cc++) if (*cc == '\\' && cc[1] == ';') break;
      delim = *cc;
      *cc = 0;
      fd = xcreate(line, O_WRONLY|O_CREAT|O_TRUNC, 0644);
      *cc = delim;

      command->w = reg - (char *)command;
      command = xrealloc(command, command->w+(cc-line)+6);
      reg = command->w + (char *)command;

      memcpy(reg, &fd, 4);
      reg += 4;
      *(reg++) = 0;
      memcpy(reg, line, delim);
      reg += delim;
      *(reg++) = 0;

      line = cc;
      if (delim) line += 2;
    } else if (c == 'y') {
      char *s, delim = 0;
      int len;

      if (!(s = unescape_delimited_string(&line, &delim))) goto error;
      command->arg1 = reg-(char *)command;
      len = strlen(s);
      reg = extend_string((void *)&command, s, reg-(char *)command, len);
      free(s);
      command->arg2 = reg-(char *)command;
      if (!(s = unescape_delimited_string(&line, &delim))) goto error;
      if (len != strlen(s)) goto error;
      reg = extend_string((void *)&command, s, reg-(char*)command, len);
      free(s);
    } else if (strchr("abcirtTw:", c)) {
      int end;

      // trim leading spaces
      while (isspace(*line) && *line != '\n') line++;

      // Resume logic differs from 's' case because we don't add a newline
      // unless it's after something, so we add it on return instead.
resume_a:
      command->hit = 0;

      // btT: end with space or semicolon, aicrw continue to newline.
      if (!(end = strcspn(line, strchr(":btT", c) ? "}; \t\r\n\v\f" : "\n"))) {
        // Argument's optional for btT
        if (strchr("btT", c)) continue;
        else if (!command->arg1) break;
      }

      // Extend allocation to include new string. We use offsets instead of
      // pointers so realloc() moving stuff doesn't break things. Ok to write
      // \n over NUL terminator because call to extend_string() adds it back.
      if (!command->arg1) command->arg1 = reg - (char*)command;
      else if (*(command->arg1+(char *)command)) *(reg++) = '\n';
      else if (!pline) {
        command->arg1 = 0;
        continue;
      }
      reg = extend_string((void *)&command, line, reg - (char *)command, end);

      // Recopy data to remove escape sequences and handle line continuation.
      if (strchr("aci", c)) {
        reg -= end+1;
        for (i = end; i; i--) {
          if ((*reg++ = *line++)=='\\') {

            // escape at end of line: resume if -e escaped literal newline,
            // else request callback and resume with next line
            if (!--i) {
              *--reg = 0;
              if (*line) {
                line++;
                goto resume_a;
              }
              command->hit = 256;
              break;
            }
            if (!(reg[-1] = unescape(*line))) reg[-1] = *line;
            line++;
          }
        }
        *reg = 0;
      } else line += end;

    // Commands that take no arguments
    } else if (!strchr("{dDgGhHlnNpPqx=", c)) break;
  }

error:
  error_exit("bad pattern '%s'@%ld (%c)", errstart, line-errstart+1L, *line);
}

void sed_main(void)
{
  struct arg_list *al;
  char **args = toys.optargs;

  if (!FLAG(z)) TT.delim = '\n';

  // Lie to autoconf when it asks stupid questions, so configure regexes
  // that look for "GNU sed version %f" greater than some old buggy number
  // don't fail us for not matching their narrow expectations.
  if (FLAG(version)) {
    xprintf("This is not GNU sed version 9.0\n");
    return;
  }

  // Handling our own --version means we handle our own --help too.
  if (FLAG(help)) help_exit(0);

  // Parse pattern into commands.

  // If no -e or -f, first argument is the pattern.
  if (!TT.e && !TT.f) {
    if (!*toys.optargs) error_exit("no pattern");
    (TT.e = xzalloc(sizeof(struct arg_list)))->arg = *(args++);
  }

  // Option parsing infrastructure can't interlace "-e blah -f blah -e blah"
  // so handle all -e, then all -f. (At least the behavior's consistent.)

  for (al = TT.e; al; al = al->next) parse_pattern(&al->arg, strlen(al->arg));
  parse_pattern(0, 0);
  for (al = TT.f; al; al = al->next)
    do_lines(xopenro(al->arg), TT.delim, parse_pattern);
  dlist_terminate(TT.pattern);
  if (TT.nextlen) error_exit("no }");  

  TT.fdout = 1;
  TT.remember = xstrdup("");

  // Inflict pattern upon input files. Long version because !O_CLOEXEC
  loopfiles_rw(args, O_RDONLY|WARN_ONLY, 0, do_sed_file);

  // Provide EOF flush at end of cumulative input for non-i mode.
  if (!FLAG(i)) {
    toys.optflags |= FLAG_i;
    sed_line(0, 0);
  }

  // todo: need to close fd when done for TOYBOX_FREE?
}