usr.bin/fmt/fmt.c

   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  * $FreeBSD: head/usr.bin/fmt/fmt.c 200462 2009-12-13 03:14:06Z delphij $
 126  */
 127
 128 /* Copyright (c) 1997 Gareth McCaughan. All rights reserved.
 129  *
 130  * Redistribution and use of this code, in source or binary forms,
 131  * with or without modification, are permitted subject to the following
 132  * conditions:
 133  *
 134  *  - Redistribution of source code must retain the above copyright
 135  *    notice, this list of conditions and the following disclaimer.
 136  *
 137  *  - If you distribute modified source code it must also include
 138  *    a notice saying that it has been modified, and giving a brief
 139  *    description of what changes have been made.
 140  *
 141  * Disclaimer: I am not responsible for the results of using this code.
 142  *             If it formats your hard disc, sends obscene messages to
 143  *             your boss and kills your children then that's your problem
 144  *             not mine. I give absolutely no warranty of any sort as to
 145  *             what the program will do, and absolutely refuse to be held
 146  *             liable for any consequences of your using it.
 147  *             Thank you. Have a nice day.
 148  */
 149
 150 /* RCS change log:
 151  * Revision 1.5  1998/03/02 18:02:21  gjm11
 152  * Minor changes for portability.
 153  *
 154  * Revision 1.4  1997/10/01 11:51:28  gjm11
 155  * Repair broken indented-paragraph handling.
 156  * Add mail message header stuff.
 157  * Improve comments and layout.
 158  * Make usable with non-BSD systems.
 159  * Add revision display to usage message.
 160  *
 161  * Revision 1.3  1997/09/30 16:24:47  gjm11
 162  * Add copyright notice, rcsid string and log message.
 163  *
 164  * Revision 1.2  1997/09/30 16:13:39  gjm11
 165  * Add options: -d <chars>, -l <width>, -p, -s, -t <width>, -h .
 166  * Parse options with `getopt'. Clean up code generally.
 167  * Make comments more accurate.
 168  *
 169  * Revision 1.1  1997/09/30 11:29:57  gjm11
 170  * Initial revision
 171  */
 172
 173 #include <err.h>
 174 #include <limits.h>
 175 #include <locale.h>
 176 #include <stdio.h>
 177 #include <stdlib.h>
 178 #include <string.h>
 179 #include <sysexits.h>
 180 #include <unistd.h>
 181 #include <wchar.h>
 182 #include <wctype.h>
 183
 184 /* Something that, we hope, will never be a genuine line length,
 185  * indentation etc.
 186  */
 187 #define SILLY ((size_t)-1)
 188
 189 /* I used to use |strtoul| for this, but (1) not all systems have it
 190  * and (2) it's probably better to use |strtol| to detect negative
 191  * numbers better.
 192  * If |fussyp==0| then we don't complain about non-numbers
 193  * (returning 0 instead), but we do complain about bad numbers.
 194  */
 195 static size_t
 196 get_positive(const char *s, const char *err_mess, int fussyP) {
 197   char *t;
 198   long result = strtol(s,&t,0);
 199   if (*t) { if (fussyP) goto Lose; else return 0; }
 200   if (result<=0) { Lose: errx(EX_USAGE, "%s", err_mess); }
 201   return (size_t) result;
 202 }
 203
 204 static size_t
 205 get_nonnegative(const char *s, const char *err_mess, int fussyP) {
 206   char *t;
 207   long result = strtol(s,&t,0);
 208   if (*t) { if (fussyP) goto Lose; else return 0; }
 209   if (result<0) { Lose: errx(EX_USAGE, "%s", err_mess); }
 210   return (size_t) result;
 211 }
 212
 213 /* Global variables */
 214
 215 static int centerP=0;           /* Try to center lines? */
 216 static size_t goal_length=0;    /* Target length for output lines */
 217 static size_t max_length=0;     /* Maximum length for output lines */
 218 static int coalesce_spaces_P=0; /* Coalesce multiple whitespace -> ' ' ? */
 219 static int allow_indented_paragraphs=0; /* Can first line have diff. ind.? */
 220 static int tab_width=8;         /* Number of spaces per tab stop */
 221 static size_t output_tab_width=8;       /* Ditto, when squashing leading spaces */
 222 static const wchar_t *sentence_enders=L".?!";   /* Double-space after these */
 223 static int grok_mail_headers=0; /* treat embedded mail headers magically? */
 224 static int format_troff=0;      /* Format troff? */
 225
 226 static int n_errors=0;          /* Number of failed files. Return on exit. */
 227 static wchar_t *output_buffer=NULL;     /* Output line will be built here */
 228 static size_t x;                /* Horizontal position in output line */
 229 static size_t x0;               /* Ditto, ignoring leading whitespace */
 230 static size_t output_buffer_length = 0;
 231 static size_t pending_spaces;   /* Spaces to add before next word */
 232 static int output_in_paragraph=0;       /* Any of current para written out yet? */
 233
 234 /* Prototypes */
 235
 236 static void process_named_file (const char *);
 237 static void     process_stream (FILE *, const char *);
 238 static size_t    indent_length (const wchar_t *, size_t);
 239 static int     might_be_header (const wchar_t *);
 240 static void      new_paragraph (size_t, size_t);
 241 static void        output_word (size_t, size_t, const wchar_t *, size_t,
 242                                 size_t);
 243 static void      output_indent (size_t);
 244 static void      center_stream (FILE *, const char *);
 245 static wchar_t *      get_line (FILE *, size_t *);
 246 static void *         xrealloc (void *, size_t);
 247
 248 #define XMALLOC(x) xrealloc(0,x)
 249
 250 /* Here is perhaps the right place to mention that this code is
 251  * all in top-down order. Hence, |main| comes first.
 252  */
 253 int
 254 main(int argc, char *argv[]) {
 255   int ch;                       /* used for |getopt| processing */
 256   wchar_t *tmp;
 257   size_t len;
 258   const char *src;
 259
 260   (void) setlocale(LC_CTYPE, "");
 261
 262   /* 1. Grok parameters. */
 263
 264   while ((ch = getopt(argc, argv, "0123456789cd:hl:mnpst:w:")) != -1)
 265     switch(ch) {
 266     case 'c':
 267       centerP = 1;
 268       format_troff = 1;
 269       continue;
 270     case 'd':
 271       src = optarg;
 272       len = mbsrtowcs(NULL, &src, 0, NULL);
 273       if (len == (size_t)-1)
 274         err(EX_USAGE, "bad sentence-ending character set");
 275       tmp = XMALLOC((len + 1) * sizeof(wchar_t));
 276       mbsrtowcs(tmp, &src, len + 1, NULL);
 277       sentence_enders = tmp;
 278       continue;
 279     case 'l':
 280       output_tab_width
 281         = get_nonnegative(optarg, "output tab width must be non-negative", 1);
 282       continue;
 283     case 'm':
 284       grok_mail_headers = 1;
 285       continue;
 286     case 'n':
 287       format_troff = 1;
 288       continue;
 289     case 'p':
 290       allow_indented_paragraphs = 1;
 291       continue;
 292     case 's':
 293       coalesce_spaces_P = 1;
 294       continue;
 295     case 't':
 296       tab_width = get_positive(optarg, "tab width must be positive", 1);
 297       continue;
 298     case 'w':
 299       goal_length = get_positive(optarg, "width must be positive", 1);
 300       max_length = goal_length;
 301       continue;
 302     case '0': case '1': case '2': case '3': case '4': case '5':
 303     case '6': case '7': case '8': case '9':
 304     /* XXX  this is not a stylistically approved use of getopt() */
 305       if (goal_length==0) {
 306         char *p;
 307         p = argv[optind - 1];
 308         if (p[0] == '-' && p[1] == ch && !p[2])
 309              goal_length = get_positive(++p, "width must be nonzero", 1);
 310         else
 311              goal_length = get_positive(argv[optind]+1,
 312                  "width must be nonzero", 1);
 313         max_length = goal_length;
 314       }
 315       continue;
 316     case 'h': default:
 317       fprintf(stderr,
 318 "usage:   fmt [-cmps] [-d chars] [-l num] [-t num]\n"
 319 "             [-w width | -width | goal [maximum]] [file ...]\n"
 320 "Options: -c     center each line instead of formatting\n"
 321 "         -d <chars> double-space after <chars> at line end\n"
 322 "         -l <n> turn each <n> spaces at start of line into a tab\n"
 323 "         -m     try to make sure mail header lines stay separate\n"
 324 "         -n     format lines beginning with a dot\n"
 325 "         -p     allow indented paragraphs\n"
 326 "         -s     coalesce whitespace inside lines\n"
 327 "         -t <n> have tabs every <n> columns\n"
 328 "         -w <n> set maximum width to <n>\n"
 329 "         goal   set target width to goal\n");
 330       exit(ch=='h' ? 0 : EX_USAGE);
 331     }
 332   argc -= optind; argv += optind;
 333
 334   /* [ goal [ maximum ] ] */
 335
 336   if (argc>0 && goal_length==0
 337       && (goal_length=get_positive(*argv,"goal length must be positive", 0))
 338          != 0) {
 339     --argc; ++argv;
 340     if (argc>0
 341         && (max_length=get_positive(*argv,"max length must be positive", 0))
 342            != 0) {
 343       --argc; ++argv;
 344       if (max_length<goal_length)
 345         errx(EX_USAGE, "max length must be >= goal length");
 346     }
 347   }
 348   if (goal_length==0) goal_length = 65;
 349   if (max_length==0) max_length = goal_length+10;
 350   if (max_length >= SIZE_T_MAX / sizeof (wchar_t)) errx(EX_USAGE, "max length too large");
 351   /* really needn't be longer */
 352   output_buffer = XMALLOC((max_length+1) * sizeof(wchar_t));
 353
 354   /* 2. Process files. */
 355
 356   if (argc>0) {
 357     while (argc-->0) process_named_file(*argv++);
 358   }
 359   else {
 360     process_stream(stdin, "standard input");
 361   }
 362
 363   /* We're done. */
 364
 365   return n_errors ? EX_NOINPUT : 0;
 366
 367 }
 368
 369 /* Process a single file, given its name.
 370  */
 371 static void
 372 process_named_file(const char *name) {
 373   FILE *f=fopen(name, "r");
 374   if (!f) { warn("%s", name); ++n_errors; }
 375   else {
 376     process_stream(f, name);
 377     if (ferror(f)) { warn("%s", name); ++n_errors; }
 378     fclose(f);
 379   }
 380 }
 381
 382 /* Types of mail header continuation lines:
 383  */
 384 typedef enum {
 385   hdr_ParagraphStart = -1,
 386   hdr_NonHeader      = 0,
 387   hdr_Header         = 1,
 388   hdr_Continuation   = 2
 389 } HdrType;
 390
 391 /* Process a stream. This is where the real work happens,
 392  * except that centering is handled separately.
 393  */
 394 static void
 395 process_stream(FILE *stream, const char *name) {
 396   size_t last_indent=SILLY;     /* how many spaces in last indent? */
 397   size_t para_line_number=0;    /* how many lines already read in this para? */
 398   size_t first_indent=SILLY;    /* indentation of line 0 of paragraph */
 399   HdrType prev_header_type=hdr_ParagraphStart;
 400         /* ^-- header_type of previous line; -1 at para start */
 401   wchar_t *line;
 402   size_t length;
 403
 404   if (centerP) { center_stream(stream, name); return; }
 405   while ((line=get_line(stream,&length)) != NULL) {
 406     size_t np=indent_length(line, length);
 407     { HdrType header_type=hdr_NonHeader;
 408       if (grok_mail_headers && prev_header_type!=hdr_NonHeader) {
 409         if (np==0 && might_be_header(line))
 410           header_type = hdr_Header;
 411         else if (np>0 && prev_header_type>hdr_NonHeader)
 412           header_type = hdr_Continuation;
 413       }
 414       /* We need a new paragraph if and only if:
 415        *   this line is blank,
 416        *   OR it's a troff request (and we don't format troff),
 417        *   OR it's a mail header,
 418        *   OR it's not a mail header AND the last line was one,
 419        *   OR the indentation has changed
 420        *      AND the line isn't a mail header continuation line
 421        *      AND this isn't the second line of an indented paragraph.
 422        */
 423       if ( length==0
 424            || (line[0]=='.' && !format_troff)
 425            || header_type==hdr_Header
 426            || (header_type==hdr_NonHeader && prev_header_type>hdr_NonHeader)
 427            || (np!=last_indent
 428                && header_type != hdr_Continuation
 429                && (!allow_indented_paragraphs || para_line_number != 1)) ) {
 430         new_paragraph(output_in_paragraph ? last_indent : first_indent, np);
 431         para_line_number = 0;
 432         first_indent = np;
 433         last_indent = np;
 434         if (header_type==hdr_Header) last_indent=2;     /* for cont. lines */
 435         if (length==0 || (line[0]=='.' && !format_troff)) {
 436           if (length==0)
 437             putwchar('\n');
 438           else
 439             wprintf(L"%.*ls\n", (int)length, line);
 440           prev_header_type=hdr_ParagraphStart;
 441           continue;
 442         }
 443       }
 444       else {
 445         /* If this is an indented paragraph other than a mail header
 446          * continuation, set |last_indent|.
 447          */
 448         if (np != last_indent && header_type != hdr_Continuation)
 449           last_indent=np;
 450       }
 451       prev_header_type = header_type;
 452     }
 453
 454     { size_t n=np;
 455       while (n<length) {
 456         /* Find word end and count spaces after it */
 457         size_t word_length=0, space_length=0;
 458         while (n+word_length < length && line[n+word_length] != ' ')
 459           ++word_length;
 460         space_length = word_length;
 461         while (n+space_length < length && line[n+space_length] == ' ')
 462           ++space_length;
 463         /* Send the word to the output machinery. */
 464         output_word(first_indent, last_indent,
 465                     line+n, word_length, space_length-word_length);
 466         n += space_length;
 467       }
 468     }
 469     ++para_line_number;
 470   }
 471   new_paragraph(output_in_paragraph ? last_indent : first_indent, 0);
 472   if (ferror(stream)) { warn("%s", name); ++n_errors; }
 473 }
 474
 475 /* How long is the indent on this line?
 476  */
 477 static size_t
 478 indent_length(const wchar_t *line, size_t length) {
 479   size_t n=0;
 480   while (n<length && *line++ == ' ') ++n;
 481   return n;
 482 }
 483
 484 /* Might this line be a mail header?
 485  * We deem a line to be a possible header if it matches the
 486  * Perl regexp /^[A-Z][-A-Za-z0-9]*:\s/. This is *not* the same
 487  * as in RFC whatever-number-it-is; we want to be gratuitously
 488  * conservative to avoid mangling ordinary civilised text.
 489  */
 490 static int
 491 might_be_header(const wchar_t *line) {
 492   if (!iswupper(*line++)) return 0;
 493   while (*line && (iswalnum(*line) || *line=='-')) ++line;
 494   return (*line==':' && iswspace(line[1]));
 495 }
 496
 497 /* Begin a new paragraph with an indent of |indent| spaces.
 498  */
 499 static void
 500 new_paragraph(size_t old_indent, size_t indent) {
 501   if (output_buffer_length) {
 502     if (old_indent>0) output_indent(old_indent);
 503     wprintf(L"%.*ls\n", (int)output_buffer_length, output_buffer);
 504   }
 505   x=indent; x0=0; output_buffer_length=0; pending_spaces=0;
 506   output_in_paragraph = 0;
 507 }
 508
 509 /* Output spaces or tabs for leading indentation.
 510  */
 511 static void
 512 output_indent(size_t n_spaces) {
 513   if (output_tab_width) {
 514     while (n_spaces >= output_tab_width) {
 515       putwchar('\t');
 516       n_spaces -= output_tab_width;
 517     }
 518   }
 519   while (n_spaces-- > 0) putwchar(' ');
 520 }
 521
 522 /* Output a single word, or add it to the buffer.
 523  * indent0 and indent1 are the indents to use on the first and subsequent
 524  * lines of a paragraph. They'll often be the same, of course.
 525  */
 526 static void
 527 output_word(size_t indent0, size_t indent1, const wchar_t *word, size_t length, size_t spaces) {
 528   size_t new_x;
 529   size_t indent = output_in_paragraph ? indent1 : indent0;
 530   size_t width;
 531   const wchar_t *p;
 532   int cwidth;
 533
 534   for (p = word, width = 0; p < &word[length]; p++)
 535     width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
 536
 537   new_x = x + pending_spaces + width;
 538
 539   /* If either |spaces==0| (at end of line) or |coalesce_spaces_P|
 540    * (squashing internal whitespace), then add just one space;
 541    * except that if the last character was a sentence-ender we
 542    * actually add two spaces.
 543    */
 544   if (coalesce_spaces_P || spaces==0)
 545     spaces = wcschr(sentence_enders, word[length-1]) ? 2 : 1;
 546
 547   if (new_x<=goal_length) {
 548     /* After adding the word we still aren't at the goal length,
 549      * so clearly we add it to the buffer rather than outputing it.
 550      */
 551     wmemset(output_buffer+output_buffer_length, L' ', pending_spaces);
 552     x0 += pending_spaces; x += pending_spaces;
 553     output_buffer_length += pending_spaces;
 554     wmemcpy(output_buffer+output_buffer_length, word, length);
 555     x0 += width; x += width; output_buffer_length += length;
 556     pending_spaces = spaces;
 557   }
 558   else {
 559     /* Adding the word takes us past the goal. Print the line-so-far,
 560      * and the word too iff either (1) the lsf is empty or (2) that
 561      * makes us nearer the goal but doesn't take us over the limit,
 562      * or (3) the word on its own takes us over the limit.
 563      * In case (3) we put a newline in between.
 564      */
 565     if (indent>0) output_indent(indent);
 566     wprintf(L"%.*ls", (int)output_buffer_length, output_buffer);
 567     if (x0==0 || (new_x <= max_length && new_x-goal_length <= goal_length-x)) {
 568       wprintf(L"%*ls", (int)pending_spaces, L"");
 569       goto write_out_word;
 570     }
 571     else {
 572       /* If the word takes us over the limit on its own, just
 573        * spit it out and don't bother buffering it.
 574        */
 575       if (indent+width > max_length) {
 576         putwchar('\n');
 577         if (indent>0) output_indent(indent);
 578 write_out_word:
 579         wprintf(L"%.*ls", (int)length, word);
 580         x0 = 0; x = indent1; pending_spaces = 0;
 581         output_buffer_length = 0;
 582       }
 583       else {
 584         wmemcpy(output_buffer, word, length);
 585         x0 = width; x = width+indent1; pending_spaces = spaces;
 586         output_buffer_length = length;
 587       }
 588     }
 589     putwchar('\n');
 590     output_in_paragraph = 1;
 591   }
 592 }
 593
 594 /* Process a stream, but just center its lines rather than trying to
 595  * format them neatly.
 596  */
 597 static void
 598 center_stream(FILE *stream, const char *name) {
 599   wchar_t *line, *p;
 600   size_t length;
 601   size_t width;
 602   int cwidth;
 603   while ((line=get_line(stream, &length)) != NULL) {
 604     size_t l=length;
 605     while (l>0 && iswspace(*line)) { ++line; --l; }
 606     length=l;
 607     for (p = line, width = 0; p < &line[length]; p++)
 608       width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
 609     l = width;
 610     while (l<goal_length) { putwchar(' '); l+=2; }
 611     wprintf(L"%.*ls\n", (int)length, line);
 612   }
 613   if (ferror(stream)) { warn("%s", name); ++n_errors; }
 614 }
 615
 616 /* Get a single line from a stream. Expand tabs, strip control
 617  * characters and trailing whitespace, and handle backspaces.
 618  * Return the address of the buffer containing the line, and
 619  * put the length of the line in |lengthp|.
 620  * This can cope with arbitrarily long lines, and with lines
 621  * without terminating \n.
 622  * If there are no characters left or an error happens, we
 623  * return 0.
 624  * Don't confuse |spaces_pending| here with the global
 625  * |pending_spaces|.
 626  */
 627 static wchar_t *
 628 get_line(FILE *stream, size_t *lengthp) {
 629   static wchar_t *buf=NULL;
 630   static size_t length=0;
 631   size_t len=0;
 632   wint_t ch;
 633   size_t spaces_pending=0;
 634   int troff=0;
 635   size_t col=0;
 636   int cwidth;
 637
 638   if (buf==NULL) { length=100; buf=XMALLOC(length * sizeof(wchar_t)); }
 639   while ((ch=getwc(stream)) != '\n' && ch != WEOF) {
 640     if (len+spaces_pending==0 && ch=='.' && !format_troff) troff=1;
 641     if (ch==' ') ++spaces_pending;
 642     else if (troff || iswprint(ch)) {
 643       while (len+spaces_pending >= length) {
 644         length*=2; buf=xrealloc(buf, length * sizeof(wchar_t));
 645       }
 646       while (spaces_pending > 0) { --spaces_pending; buf[len++]=' '; col++; }
 647       buf[len++] = ch;
 648       col += (cwidth = wcwidth(ch)) > 0 ? cwidth : 1;
 649     }
 650     else if (ch=='\t')
 651       spaces_pending += tab_width - (col+spaces_pending)%tab_width;
 652     else if (ch=='\b') { if (len) --len; if (col) --col; }
 653   }
 654   *lengthp=len;
 655   return (len>0 || ch!=WEOF) ? buf : 0;
 656 }
 657
 658 /* (Re)allocate some memory, exiting with an error if we can't.
 659  */
 660 static void *
 661 xrealloc(void *ptr, size_t nbytes) {
 662   void *p = realloc(ptr, nbytes);
 663   if (p == NULL) errx(EX_OSERR, "out of memory");
 664   return p;
 665 }