xref: /freebsd/usr.bin/fmt/fmt.c (revision 74bf4e164ba5851606a27d4feff27717452583e5)
1 /*	$OpenBSD: fmt.c,v 1.16 2000/06/25 15:35:42 pjanzen Exp $	*/
2 
3 /* Sensible version of fmt
4  *
5  * Syntax: fmt [ options ] [ goal [ max ] ] [ filename ... ]
6  *
7  * Since the documentation for the original fmt is so poor, here
8  * is an accurate description of what this one does. It's usually
9  * the same. The *mechanism* used may differ from that suggested
10  * here. Note that we are *not* entirely compatible with fmt,
11  * because fmt gets so many things wrong.
12  *
13  * 1. Tabs are expanded, assuming 8-space tab stops.
14  *    If the `-t <n>' option is given, we assume <n>-space
15  *    tab stops instead.
16  *    Trailing blanks are removed from all lines.
17  *    x\b == nothing, for any x other than \b.
18  *    Other control characters are simply stripped. This
19  *    includes \r.
20  * 2. Each line is split into leading whitespace and
21  *    everything else. Maximal consecutive sequences of
22  *    lines with the same leading whitespace are considered
23  *    to form paragraphs, except that a blank line is always
24  *    a paragraph to itself.
25  *    If the `-p' option is given then the first line of a
26  *    paragraph is permitted to have indentation different
27  *    from that of the other lines.
28  *    If the `-m' option is given then a line that looks
29  *    like a mail message header, if it is not immediately
30  *    preceded by a non-blank non-message-header line, is
31  *    taken to start a new paragraph, which also contains
32  *    any subsequent lines with non-empty leading whitespace.
33  *    Unless the `-n' option is given, lines beginning with
34  *    a . (dot) are not formatted.
35  * 3. The "everything else" is split into words; a word
36  *    includes its trailing whitespace, and a word at the
37  *    end of a line is deemed to be followed by a single
38  *    space, or two spaces if it ends with a sentence-end
39  *    character. (See the `-d' option for how to change that.)
40  *    If the `-s' option has been given, then a word's trailing
41  *    whitespace is replaced by what it would have had if it
42  *    had occurred at end of line.
43  * 4. Each paragraph is sent to standard output as follows.
44  *    We output the leading whitespace, and then enough words
45  *    to make the line length as near as possible to the goal
46  *    without exceeding the maximum. (If a single word would
47  *    exceed the maximum, we output that anyway.) Of course
48  *    the trailing whitespace of the last word is ignored.
49  *    We then emit a newline and start again if there are any
50  *    words left.
51  *    Note that for a blank line this translates as "We emit
52  *    a newline".
53  *    If the `-l <n>' option is given, then leading whitespace
54  *    is modified slightly: <n> spaces are replaced by a tab.
55  *    Indented paragraphs (see above under `-p') make matters
56  *    more complicated than this suggests. Actually every paragraph
57  *    has two `leading whitespace' values; the value for the first
58  *    line, and the value for the most recent line. (While processing
59  *    the first line, the two are equal. When `-p' has not been
60  *    given, they are always equal.) The leading whitespace
61  *    actually output is that of the first line (for the first
62  *    line of *output*) or that of the most recent line (for
63  *    all other lines of output).
64  *    When `-m' has been given, message header paragraphs are
65  *    taken as having first-leading-whitespace empty and
66  *    subsequent-leading-whitespace two spaces.
67  *
68  * Multiple input files are formatted one at a time, so that a file
69  * never ends in the middle of a line.
70  *
71  * There's an alternative mode of operation, invoked by giving
72  * the `-c' option. In that case we just center every line,
73  * and most of the other options are ignored. This should
74  * really be in a separate program, but we must stay compatible
75  * with old `fmt'.
76  *
77  * QUERY: Should `-m' also try to do the right thing with quoted text?
78  * QUERY: `-b' to treat backslashed whitespace as old `fmt' does?
79  * QUERY: Option meaning `never join lines'?
80  * QUERY: Option meaning `split in mid-word to avoid overlong lines'?
81  * (Those last two might not be useful, since we have `fold'.)
82  *
83  * Differences from old `fmt':
84  *
85  *   - We have many more options. Options that aren't understood
86  *     generate a lengthy usage message, rather than being
87  *     treated as filenames.
88  *   - Even with `-m', our handling of message headers is
89  *     significantly different. (And much better.)
90  *   - We don't treat `\ ' as non-word-breaking.
91  *   - Downward changes of indentation start new paragraphs
92  *     for us, as well as upward. (I think old `fmt' behaves
93  *     in the way it does in order to allow indented paragraphs,
94  *     but this is a broken way of making indented paragraphs
95  *     behave right.)
96  *   - Given the choice of going over or under |goal_length|
97  *     by the same amount, we go over; old `fmt' goes under.
98  *   - We treat `?' as ending a sentence, and not `:'. Old `fmt'
99  *     does the reverse.
100  *   - We return approved return codes. Old `fmt' returns
101  *     1 for some errors, and *the number of unopenable files*
102  *     when that was all that went wrong.
103  *   - We have fewer crashes and more helpful error messages.
104  *   - We don't turn spaces into tabs at starts of lines unless
105  *     specifically requested.
106  *   - New `fmt' is somewhat smaller and slightly faster than
107  *     old `fmt'.
108  *
109  * Bugs:
110  *
111  *   None known. There probably are some, though.
112  *
113  * Portability:
114  *
115  *   I believe this code to be pretty portable. It does require
116  *   that you have `getopt'. If you need to include "getopt.h"
117  *   for this (e.g., if your system didn't come with `getopt'
118  *   and you installed it yourself) then you should arrange for
119  *   NEED_getopt_h to be #defined.
120  *
121  *   Everything here should work OK even on nasty 16-bit
122  *   machines and nice 64-bit ones. However, it's only really
123  *   been tested on my FreeBSD machine. Your mileage may vary.
124  */
125 
126 /* Copyright (c) 1997 Gareth McCaughan. All rights reserved.
127  *
128  * Redistribution and use of this code, in source or binary forms,
129  * with or without modification, are permitted subject to the following
130  * conditions:
131  *
132  *  - Redistribution of source code must retain the above copyright
133  *    notice, this list of conditions and the following disclaimer.
134  *
135  *  - If you distribute modified source code it must also include
136  *    a notice saying that it has been modified, and giving a brief
137  *    description of what changes have been made.
138  *
139  * Disclaimer: I am not responsible for the results of using this code.
140  *             If it formats your hard disc, sends obscene messages to
141  *             your boss and kills your children then that's your problem
142  *             not mine. I give absolutely no warranty of any sort as to
143  *             what the program will do, and absolutely refuse to be held
144  *             liable for any consequences of your using it.
145  *             Thank you. Have a nice day.
146  */
147 
148 /* RCS change log:
149  * Revision 1.5  1998/03/02 18:02:21  gjm11
150  * Minor changes for portability.
151  *
152  * Revision 1.4  1997/10/01 11:51:28  gjm11
153  * Repair broken indented-paragraph handling.
154  * Add mail message header stuff.
155  * Improve comments and layout.
156  * Make usable with non-BSD systems.
157  * Add revision display to usage message.
158  *
159  * Revision 1.3  1997/09/30 16:24:47  gjm11
160  * Add copyright notice, rcsid string and log message.
161  *
162  * Revision 1.2  1997/09/30 16:13:39  gjm11
163  * Add options: -d <chars>, -l <width>, -p, -s, -t <width>, -h .
164  * Parse options with `getopt'. Clean up code generally.
165  * Make comments more accurate.
166  *
167  * Revision 1.1  1997/09/30 11:29:57  gjm11
168  * Initial revision
169  */
170 
171 #ifndef lint
172 static const char copyright[] =
173   "Copyright (c) 1997 Gareth McCaughan. All rights reserved.\n";
174 #endif /* not lint */
175 #include <sys/cdefs.h>
176 __FBSDID("$FreeBSD$");
177 
178 #include <err.h>
179 #include <locale.h>
180 #include <stdio.h>
181 #include <stdlib.h>
182 #include <string.h>
183 #include <sysexits.h>
184 #include <unistd.h>
185 #include <wchar.h>
186 #include <wctype.h>
187 
188 /* Something that, we hope, will never be a genuine line length,
189  * indentation etc.
190  */
191 #define SILLY ((size_t)-1)
192 
193 /* I used to use |strtoul| for this, but (1) not all systems have it
194  * and (2) it's probably better to use |strtol| to detect negative
195  * numbers better.
196  * If |fussyp==0| then we don't complain about non-numbers
197  * (returning 0 instead), but we do complain about bad numbers.
198  */
199 static size_t
200 get_positive(const char *s, const char *err_mess, int fussyP) {
201   char *t;
202   long result = strtol(s,&t,0);
203   if (*t) { if (fussyP) goto Lose; else return 0; }
204   if (result<=0) { Lose: errx(EX_USAGE, "%s", err_mess); }
205   return (size_t) result;
206 }
207 
208 static size_t
209 get_nonnegative(const char *s, const char *err_mess, int fussyP) {
210   char *t;
211   long result = strtol(s,&t,0);
212   if (*t) { if (fussyP) goto Lose; else return 0; }
213   if (result<0) { Lose: errx(EX_USAGE, "%s", err_mess); }
214   return (size_t) result;
215 }
216 
217 /* Global variables */
218 
219 static int centerP=0;		/* Try to center lines? */
220 static size_t goal_length=0;	/* Target length for output lines */
221 static size_t max_length=0;	/* Maximum length for output lines */
222 static int coalesce_spaces_P=0;	/* Coalesce multiple whitespace -> ' ' ? */
223 static int allow_indented_paragraphs=0;	/* Can first line have diff. ind.? */
224 static int tab_width=8;		/* Number of spaces per tab stop */
225 static size_t output_tab_width=8;	/* Ditto, when squashing leading spaces */
226 static const wchar_t *sentence_enders=L".?!";	/* Double-space after these */
227 static int grok_mail_headers=0;	/* treat embedded mail headers magically? */
228 static int format_troff=0;	/* Format troff? */
229 
230 static int n_errors=0;		/* Number of failed files. Return on exit. */
231 static wchar_t *output_buffer=0;	/* Output line will be built here */
232 static size_t x;		/* Horizontal position in output line */
233 static size_t x0;		/* Ditto, ignoring leading whitespace */
234 static size_t output_buffer_length = 0;
235 static size_t pending_spaces;	/* Spaces to add before next word */
236 static int output_in_paragraph=0;	/* Any of current para written out yet? */
237 
238 /* Prototypes */
239 
240 static void process_named_file (const char *);
241 static void     process_stream (FILE *, const char *);
242 static size_t    indent_length (const wchar_t *, size_t);
243 static int     might_be_header (const wchar_t *);
244 static void      new_paragraph (size_t, size_t);
245 static void        output_word (size_t, size_t, const wchar_t *, size_t,
246 				size_t);
247 static void      output_indent (size_t);
248 static void      center_stream (FILE *, const char *);
249 static wchar_t *      get_line (FILE *, size_t *);
250 static void *         xrealloc (void *, size_t);
251 
252 #define XMALLOC(x) xrealloc(0,x)
253 
254 /* Here is perhaps the right place to mention that this code is
255  * all in top-down order. Hence, |main| comes first.
256  */
257 int
258 main(int argc, char *argv[]) {
259   int ch;			/* used for |getopt| processing */
260   wchar_t *tmp;
261   size_t len;
262   const char *src;
263 
264   (void) setlocale(LC_CTYPE, "");
265 
266   /* 1. Grok parameters. */
267 
268   while ((ch = getopt(argc, argv, "0123456789cd:hl:mnpst:w:")) != -1)
269   switch(ch) {
270     case 'c':
271       centerP = 1;
272       format_troff = 1;
273       continue;
274     case 'd':
275       src = optarg;
276       len = mbsrtowcs(NULL, &src, 0, NULL);
277       if (len == (size_t)-1)
278         err(EX_USAGE, "bad sentence-ending character set");
279       tmp = XMALLOC((len + 1) * sizeof(wchar_t));
280       mbsrtowcs(tmp, &src, len + 1, NULL);
281       sentence_enders = tmp;
282       continue;
283     case 'l':
284       output_tab_width
285         = get_nonnegative(optarg, "output tab width must be non-negative", 1);
286       continue;
287     case 'm':
288       grok_mail_headers = 1;
289       continue;
290     case 'n':
291       format_troff = 1;
292       continue;
293     case 'p':
294       allow_indented_paragraphs = 1;
295       continue;
296     case 's':
297       coalesce_spaces_P = 1;
298       continue;
299     case 't':
300       tab_width = get_positive(optarg, "tab width must be positive", 1);
301       continue;
302     case 'w':
303       goal_length = get_positive(optarg, "width must be positive", 1);
304       max_length = goal_length;
305       continue;
306     case '0': case '1': case '2': case '3': case '4': case '5':
307     case '6': case '7': case '8': case '9':
308     /* XXX  this is not a stylistically approved use of getopt() */
309       if (goal_length==0) {
310         char *p;
311         p = argv[optind - 1];
312         if (p[0] == '-' && p[1] == ch && !p[2])
313              goal_length = get_positive(++p, "width must be nonzero", 1);
314         else
315              goal_length = get_positive(argv[optind]+1,
316                  "width must be nonzero", 1);
317         max_length = goal_length;
318       }
319       continue;
320     case 'h': default:
321       fprintf(stderr,
322 "usage:   fmt [-cmps] [-d chars] [-l num] [-t num]\n"
323 "             [-w width | -width | goal [maximum]] [file ...]\n"
324 "Options: -c     center each line instead of formatting\n"
325 "         -d <chars> double-space after <chars> at line end\n"
326 "         -l <n> turn each <n> spaces at start of line into a tab\n"
327 "         -m     try to make sure mail header lines stay separate\n"
328 "         -n     format lines beginning with a dot\n"
329 "         -p     allow indented paragraphs\n"
330 "         -s     coalesce whitespace inside lines\n"
331 "         -t <n> have tabs every <n> columns\n"
332 "         -w <n> set maximum width to <n>\n"
333 "         goal   set target width to goal\n");
334       exit(ch=='h' ? 0 : EX_USAGE);
335   }
336   argc -= optind; argv += optind;
337 
338   /* [ goal [ maximum ] ] */
339 
340   if (argc>0 && goal_length==0
341       && (goal_length=get_positive(*argv,"goal length must be positive", 0))
342          != 0) {
343     --argc; ++argv;
344     if (argc>0
345         && (max_length=get_positive(*argv,"max length must be positive", 0))
346            != 0) {
347       --argc; ++argv;
348       if (max_length<goal_length)
349         errx(EX_USAGE, "max length must be >= goal length");
350     }
351   }
352   if (goal_length==0) goal_length = 65;
353   if (max_length==0) max_length = goal_length+10;
354   /* really needn't be longer */
355   output_buffer = XMALLOC((max_length+1) * sizeof(wchar_t));
356 
357   /* 2. Process files. */
358 
359   if (argc>0) {
360     while (argc-->0) process_named_file(*argv++);
361   }
362   else {
363     process_stream(stdin, "standard input");
364   }
365 
366   /* We're done. */
367 
368   return n_errors ? EX_NOINPUT : 0;
369 
370 }
371 
372 /* Process a single file, given its name.
373  */
374 static void
375 process_named_file(const char *name) {
376   FILE *f=fopen(name, "r");
377   if (!f) { warn("%s", name); ++n_errors; }
378   else {
379     process_stream(f, name);
380     if (ferror(f)) { warn("%s", name); ++n_errors; }
381     fclose(f);
382   }
383 }
384 
385 /* Types of mail header continuation lines:
386  */
387 typedef enum {
388   hdr_ParagraphStart = -1,
389   hdr_NonHeader      = 0,
390   hdr_Header         = 1,
391   hdr_Continuation   = 2
392 } HdrType;
393 
394 /* Process a stream. This is where the real work happens,
395  * except that centering is handled separately.
396  */
397 static void
398 process_stream(FILE *stream, const char *name) {
399   size_t last_indent=SILLY;	/* how many spaces in last indent? */
400   size_t para_line_number=0;	/* how many lines already read in this para? */
401   size_t first_indent=SILLY;	/* indentation of line 0 of paragraph */
402   HdrType prev_header_type=hdr_ParagraphStart;
403 	/* ^-- header_type of previous line; -1 at para start */
404   wchar_t *line;
405   size_t length;
406 
407   if (centerP) { center_stream(stream, name); return; }
408   while ((line=get_line(stream,&length)) != NULL) {
409     size_t np=indent_length(line, length);
410     { HdrType header_type=hdr_NonHeader;
411       if (grok_mail_headers && prev_header_type!=hdr_NonHeader) {
412         if (np==0 && might_be_header(line))
413           header_type = hdr_Header;
414         else if (np>0 && prev_header_type>hdr_NonHeader)
415           header_type = hdr_Continuation;
416       }
417       /* We need a new paragraph if and only if:
418        *   this line is blank,
419        *   OR it's a troff request (and we don't format troff),
420        *   OR it's a mail header,
421        *   OR it's not a mail header AND the last line was one,
422        *   OR the indentation has changed
423        *      AND the line isn't a mail header continuation line
424        *      AND this isn't the second line of an indented paragraph.
425        */
426       if ( length==0
427            || (line[0]=='.' && !format_troff)
428            || header_type==hdr_Header
429            || (header_type==hdr_NonHeader && prev_header_type>hdr_NonHeader)
430            || (np!=last_indent
431                && header_type != hdr_Continuation
432                && (!allow_indented_paragraphs || para_line_number != 1)) ) {
433         new_paragraph(output_in_paragraph ? last_indent : first_indent, np);
434         para_line_number = 0;
435         first_indent = np;
436         last_indent = np;
437         if (header_type==hdr_Header) last_indent=2;	/* for cont. lines */
438         if (length==0 || (line[0]=='.' && !format_troff)) {
439           if (length==0)
440             putwchar('\n');
441           else
442             wprintf(L"%.*ls\n", (int)length, line);
443           prev_header_type=hdr_ParagraphStart;
444           continue;
445         }
446       }
447       else {
448         /* If this is an indented paragraph other than a mail header
449          * continuation, set |last_indent|.
450          */
451         if (np != last_indent && header_type != hdr_Continuation)
452           last_indent=np;
453       }
454       prev_header_type = header_type;
455     }
456 
457     { size_t n=np;
458       while (n<length) {
459         /* Find word end and count spaces after it */
460         size_t word_length=0, space_length=0;
461         while (n+word_length < length && line[n+word_length] != ' ')
462           ++word_length;
463         space_length = word_length;
464         while (n+space_length < length && line[n+space_length] == ' ')
465           ++space_length;
466         /* Send the word to the output machinery. */
467         output_word(first_indent, last_indent,
468                     line+n, word_length, space_length-word_length);
469         n += space_length;
470       }
471     }
472     ++para_line_number;
473   }
474   new_paragraph(output_in_paragraph ? last_indent : first_indent, 0);
475   if (ferror(stream)) { warn("%s", name); ++n_errors; }
476 }
477 
478 /* How long is the indent on this line?
479  */
480 static size_t
481 indent_length(const wchar_t *line, size_t length) {
482   size_t n=0;
483   while (n<length && *line++ == ' ') ++n;
484   return n;
485 }
486 
487 /* Might this line be a mail header?
488  * We deem a line to be a possible header if it matches the
489  * Perl regexp /^[A-Z][-A-Za-z0-9]*:\s/. This is *not* the same
490  * as in RFC whatever-number-it-is; we want to be gratuitously
491  * conservative to avoid mangling ordinary civilised text.
492  */
493 static int
494 might_be_header(const wchar_t *line) {
495   if (!iswupper(*line++)) return 0;
496   while (*line && (iswalnum(*line) || *line=='-')) ++line;
497   return (*line==':' && iswspace(line[1]));
498 }
499 
500 /* Begin a new paragraph with an indent of |indent| spaces.
501  */
502 static void
503 new_paragraph(size_t old_indent, size_t indent) {
504   if (output_buffer_length) {
505     if (old_indent>0) output_indent(old_indent);
506     wprintf(L"%.*ls\n", (int)output_buffer_length, output_buffer);
507   }
508   x=indent; x0=0; output_buffer_length=0; pending_spaces=0;
509   output_in_paragraph = 0;
510 }
511 
512 /* Output spaces or tabs for leading indentation.
513  */
514 static void
515 output_indent(size_t n_spaces) {
516   if (output_tab_width) {
517     while (n_spaces >= output_tab_width) {
518       putwchar('\t');
519       n_spaces -= output_tab_width;
520     }
521   }
522   while (n_spaces-- > 0) putwchar(' ');
523 }
524 
525 /* Output a single word, or add it to the buffer.
526  * indent0 and indent1 are the indents to use on the first and subsequent
527  * lines of a paragraph. They'll often be the same, of course.
528  */
529 static void
530 output_word(size_t indent0, size_t indent1, const wchar_t *word, size_t length, size_t spaces) {
531   size_t new_x;
532   size_t indent = output_in_paragraph ? indent1 : indent0;
533   size_t width;
534   const wchar_t *p;
535   int cwidth;
536 
537   for (p = word, width = 0; p < &word[length]; p++)
538     width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
539 
540   new_x = x + pending_spaces + width;
541 
542   /* If either |spaces==0| (at end of line) or |coalesce_spaces_P|
543    * (squashing internal whitespace), then add just one space;
544    * except that if the last character was a sentence-ender we
545    * actually add two spaces.
546    */
547   if (coalesce_spaces_P || spaces==0)
548     spaces = wcschr(sentence_enders, word[length-1]) ? 2 : 1;
549 
550   if (new_x<=goal_length) {
551     /* After adding the word we still aren't at the goal length,
552      * so clearly we add it to the buffer rather than outputing it.
553      */
554     wmemset(output_buffer+output_buffer_length, L' ', pending_spaces);
555     x0 += pending_spaces; x += pending_spaces;
556     output_buffer_length += pending_spaces;
557     wmemcpy(output_buffer+output_buffer_length, word, length);
558     x0 += width; x += width; output_buffer_length += length;
559     pending_spaces = spaces;
560   }
561   else {
562     /* Adding the word takes us past the goal. Print the line-so-far,
563      * and the word too iff either (1) the lsf is empty or (2) that
564      * makes us nearer the goal but doesn't take us over the limit,
565      * or (3) the word on its own takes us over the limit.
566      * In case (3) we put a newline in between.
567      */
568     if (indent>0) output_indent(indent);
569     wprintf(L"%.*ls", (int)output_buffer_length, output_buffer);
570     if (x0==0 || (new_x <= max_length && new_x-goal_length <= goal_length-x)) {
571       wprintf(L"%*ls", (int)pending_spaces, L"");
572       goto write_out_word;
573     }
574     else {
575       /* If the word takes us over the limit on its own, just
576        * spit it out and don't bother buffering it.
577        */
578       if (indent+width > max_length) {
579         putwchar('\n');
580         if (indent>0) output_indent(indent);
581 write_out_word:
582         wprintf(L"%.*ls", (int)length, word);
583         x0 = 0; x = indent1; pending_spaces = 0;
584         output_buffer_length = 0;
585       }
586       else {
587         wmemcpy(output_buffer, word, length);
588         x0 = width; x = width+indent1; pending_spaces = spaces;
589         output_buffer_length = length;
590       }
591     }
592     putwchar('\n');
593     output_in_paragraph = 1;
594   }
595 }
596 
597 /* Process a stream, but just center its lines rather than trying to
598  * format them neatly.
599  */
600 static void
601 center_stream(FILE *stream, const char *name) {
602   wchar_t *line, *p;
603   size_t length;
604   size_t width;
605   int cwidth;
606   while ((line=get_line(stream, &length)) != 0) {
607     size_t l=length;
608     while (l>0 && iswspace(*line)) { ++line; --l; }
609     length=l;
610     for (p = line, width = 0; p < &line[length]; p++)
611       width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
612     l = width;
613     while (l<goal_length) { putwchar(' '); l+=2; }
614     wprintf(L"%.*ls\n", (int)length, line);
615   }
616   if (ferror(stream)) { warn("%s", name); ++n_errors; }
617 }
618 
619 /* Get a single line from a stream. Expand tabs, strip control
620  * characters and trailing whitespace, and handle backspaces.
621  * Return the address of the buffer containing the line, and
622  * put the length of the line in |lengthp|.
623  * This can cope with arbitrarily long lines, and with lines
624  * without terminating \n.
625  * If there are no characters left or an error happens, we
626  * return 0.
627  * Don't confuse |spaces_pending| here with the global
628  * |pending_spaces|.
629  */
630 static wchar_t *
631 get_line(FILE *stream, size_t *lengthp) {
632   static wchar_t *buf=NULL;
633   static size_t length=0;
634   size_t len=0;
635   wint_t ch;
636   size_t spaces_pending=0;
637   int troff=0;
638   size_t col=0;
639   int cwidth;
640 
641   if (buf==NULL) { length=100; buf=XMALLOC(length * sizeof(wchar_t)); }
642   while ((ch=getwc(stream)) != '\n' && ch != WEOF) {
643     if (len+spaces_pending==0 && ch=='.' && !format_troff) troff=1;
644     if (ch==' ') ++spaces_pending;
645     else if (troff || iswprint(ch)) {
646       while (len+spaces_pending >= length) {
647         length*=2; buf=xrealloc(buf, length * sizeof(wchar_t));
648       }
649       while (spaces_pending > 0) { --spaces_pending; buf[len++]=' '; col++; }
650       buf[len++] = ch;
651       col += (cwidth = wcwidth(ch)) > 0 ? cwidth : 1;
652     }
653     else if (ch=='\t')
654       spaces_pending += tab_width - (col+spaces_pending)%tab_width;
655     else if (ch=='\b') { if (len) --len; if (col) --col; }
656   }
657   *lengthp=len;
658   return (len>0 || ch!=WEOF) ? buf : 0;
659 }
660 
661 /* (Re)allocate some memory, exiting with an error if we can't.
662  */
663 static void *
664 xrealloc(void *ptr, size_t nbytes) {
665   void *p = realloc(ptr, nbytes);
666   if (p == NULL) errx(EX_OSERR, "out of memory");
667   return p;
668 }
669