1 /* $OpenBSD: fmt.c,v 1.16 2000/06/25 15:35:42 pjanzen Exp $ */
3 /* Sensible version of fmt
5 * Syntax: fmt [ options ] [ goal [ max ] ] [ filename ... ]
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.
13 * 1. Tabs are expanded, assuming 8-space tab stops.
14 * If the `-t <n>' option is given, we assume <n>-space
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
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
51 * Note that for a blank line this translates as "We emit
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.
68 * Multiple input files are formatted one at a time, so that a file
69 * never ends in the middle of a line.
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
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'.)
83 * Differences from old `fmt':
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
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'
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
111 * None known. There probably are some, though.
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.
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.
126 /* Copyright (c) 1997 Gareth McCaughan. All rights reserved.
128 * Redistribution and use of this code, in source or binary forms,
129 * with or without modification, are permitted subject to the following
132 * - Redistribution of source code must retain the above copyright
133 * notice, this list of conditions and the following disclaimer.
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.
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.
149 * Revision 1.5 1998/03/02 18:02:21 gjm11
150 * Minor changes for portability.
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.
159 * Revision 1.3 1997/09/30 16:24:47 gjm11
160 * Add copyright notice, rcsid string and log message.
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.
167 * Revision 1.1 1997/09/30 11:29:57 gjm11
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("$FreeBSD$");
184 #include <sysexits.h>
189 /* Something that, we hope, will never be a genuine line length,
192 #define SILLY ((size_t)-1)
194 /* I used to use |strtoul| for this, but (1) not all systems have it
195 * and (2) it's probably better to use |strtol| to detect negative
197 * If |fussyp==0| then we don't complain about non-numbers
198 * (returning 0 instead), but we do complain about bad numbers.
201 get_positive(const char *s, const char *err_mess, int fussyP) {
203 long result = strtol(s,&t,0);
204 if (*t) { if (fussyP) goto Lose; else return 0; }
205 if (result<=0) { Lose: errx(EX_USAGE, "%s", err_mess); }
206 return (size_t) result;
210 get_nonnegative(const char *s, const char *err_mess, int fussyP) {
212 long result = strtol(s,&t,0);
213 if (*t) { if (fussyP) goto Lose; else return 0; }
214 if (result<0) { Lose: errx(EX_USAGE, "%s", err_mess); }
215 return (size_t) result;
218 /* Global variables */
220 static int centerP=0; /* Try to center lines? */
221 static size_t goal_length=0; /* Target length for output lines */
222 static size_t max_length=0; /* Maximum length for output lines */
223 static int coalesce_spaces_P=0; /* Coalesce multiple whitespace -> ' ' ? */
224 static int allow_indented_paragraphs=0; /* Can first line have diff. ind.? */
225 static int tab_width=8; /* Number of spaces per tab stop */
226 static size_t output_tab_width=8; /* Ditto, when squashing leading spaces */
227 static const wchar_t *sentence_enders=L".?!"; /* Double-space after these */
228 static int grok_mail_headers=0; /* treat embedded mail headers magically? */
229 static int format_troff=0; /* Format troff? */
231 static int n_errors=0; /* Number of failed files. Return on exit. */
232 static wchar_t *output_buffer=0; /* Output line will be built here */
233 static size_t x; /* Horizontal position in output line */
234 static size_t x0; /* Ditto, ignoring leading whitespace */
235 static size_t output_buffer_length = 0;
236 static size_t pending_spaces; /* Spaces to add before next word */
237 static int output_in_paragraph=0; /* Any of current para written out yet? */
241 static void process_named_file (const char *);
242 static void process_stream (FILE *, const char *);
243 static size_t indent_length (const wchar_t *, size_t);
244 static int might_be_header (const wchar_t *);
245 static void new_paragraph (size_t, size_t);
246 static void output_word (size_t, size_t, const wchar_t *, size_t,
248 static void output_indent (size_t);
249 static void center_stream (FILE *, const char *);
250 static wchar_t * get_line (FILE *, size_t *);
251 static void * xrealloc (void *, size_t);
253 #define XMALLOC(x) xrealloc(0,x)
255 /* Here is perhaps the right place to mention that this code is
256 * all in top-down order. Hence, |main| comes first.
259 main(int argc, char *argv[]) {
260 int ch; /* used for |getopt| processing */
265 (void) setlocale(LC_CTYPE, "");
267 /* 1. Grok parameters. */
269 while ((ch = getopt(argc, argv, "0123456789cd:hl:mnpst:w:")) != -1)
277 len = mbsrtowcs(NULL, &src, 0, NULL);
278 if (len == (size_t)-1)
279 err(EX_USAGE, "bad sentence-ending character set");
280 tmp = XMALLOC((len + 1) * sizeof(wchar_t));
281 mbsrtowcs(tmp, &src, len + 1, NULL);
282 sentence_enders = tmp;
286 = get_nonnegative(optarg, "output tab width must be non-negative", 1);
289 grok_mail_headers = 1;
295 allow_indented_paragraphs = 1;
298 coalesce_spaces_P = 1;
301 tab_width = get_positive(optarg, "tab width must be positive", 1);
304 goal_length = get_positive(optarg, "width must be positive", 1);
305 max_length = goal_length;
307 case '0': case '1': case '2': case '3': case '4': case '5':
308 case '6': case '7': case '8': case '9':
309 /* XXX this is not a stylistically approved use of getopt() */
310 if (goal_length==0) {
312 p = argv[optind - 1];
313 if (p[0] == '-' && p[1] == ch && !p[2])
314 goal_length = get_positive(++p, "width must be nonzero", 1);
316 goal_length = get_positive(argv[optind]+1,
317 "width must be nonzero", 1);
318 max_length = goal_length;
323 "usage: fmt [-cmps] [-d chars] [-l num] [-t num]\n"
324 " [-w width | -width | goal [maximum]] [file ...]\n"
325 "Options: -c center each line instead of formatting\n"
326 " -d <chars> double-space after <chars> at line end\n"
327 " -l <n> turn each <n> spaces at start of line into a tab\n"
328 " -m try to make sure mail header lines stay separate\n"
329 " -n format lines beginning with a dot\n"
330 " -p allow indented paragraphs\n"
331 " -s coalesce whitespace inside lines\n"
332 " -t <n> have tabs every <n> columns\n"
333 " -w <n> set maximum width to <n>\n"
334 " goal set target width to goal\n");
335 exit(ch=='h' ? 0 : EX_USAGE);
337 argc -= optind; argv += optind;
339 /* [ goal [ maximum ] ] */
341 if (argc>0 && goal_length==0
342 && (goal_length=get_positive(*argv,"goal length must be positive", 0))
346 && (max_length=get_positive(*argv,"max length must be positive", 0))
349 if (max_length<goal_length)
350 errx(EX_USAGE, "max length must be >= goal length");
353 if (goal_length==0) goal_length = 65;
354 if (max_length==0) max_length = goal_length+10;
355 if (max_length >= SIZE_T_MAX / sizeof (wchar_t)) errx(EX_USAGE, "max length too large");
356 /* really needn't be longer */
357 output_buffer = XMALLOC((max_length+1) * sizeof(wchar_t));
359 /* 2. Process files. */
362 while (argc-->0) process_named_file(*argv++);
365 process_stream(stdin, "standard input");
370 return n_errors ? EX_NOINPUT : 0;
374 /* Process a single file, given its name.
377 process_named_file(const char *name) {
378 FILE *f=fopen(name, "r");
379 if (!f) { warn("%s", name); ++n_errors; }
381 process_stream(f, name);
382 if (ferror(f)) { warn("%s", name); ++n_errors; }
387 /* Types of mail header continuation lines:
390 hdr_ParagraphStart = -1,
396 /* Process a stream. This is where the real work happens,
397 * except that centering is handled separately.
400 process_stream(FILE *stream, const char *name) {
401 size_t last_indent=SILLY; /* how many spaces in last indent? */
402 size_t para_line_number=0; /* how many lines already read in this para? */
403 size_t first_indent=SILLY; /* indentation of line 0 of paragraph */
404 HdrType prev_header_type=hdr_ParagraphStart;
405 /* ^-- header_type of previous line; -1 at para start */
409 if (centerP) { center_stream(stream, name); return; }
410 while ((line=get_line(stream,&length)) != NULL) {
411 size_t np=indent_length(line, length);
412 { HdrType header_type=hdr_NonHeader;
413 if (grok_mail_headers && prev_header_type!=hdr_NonHeader) {
414 if (np==0 && might_be_header(line))
415 header_type = hdr_Header;
416 else if (np>0 && prev_header_type>hdr_NonHeader)
417 header_type = hdr_Continuation;
419 /* We need a new paragraph if and only if:
420 * this line is blank,
421 * OR it's a troff request (and we don't format troff),
422 * OR it's a mail header,
423 * OR it's not a mail header AND the last line was one,
424 * OR the indentation has changed
425 * AND the line isn't a mail header continuation line
426 * AND this isn't the second line of an indented paragraph.
429 || (line[0]=='.' && !format_troff)
430 || header_type==hdr_Header
431 || (header_type==hdr_NonHeader && prev_header_type>hdr_NonHeader)
433 && header_type != hdr_Continuation
434 && (!allow_indented_paragraphs || para_line_number != 1)) ) {
435 new_paragraph(output_in_paragraph ? last_indent : first_indent, np);
436 para_line_number = 0;
439 if (header_type==hdr_Header) last_indent=2; /* for cont. lines */
440 if (length==0 || (line[0]=='.' && !format_troff)) {
444 wprintf(L"%.*ls\n", (int)length, line);
445 prev_header_type=hdr_ParagraphStart;
450 /* If this is an indented paragraph other than a mail header
451 * continuation, set |last_indent|.
453 if (np != last_indent && header_type != hdr_Continuation)
456 prev_header_type = header_type;
461 /* Find word end and count spaces after it */
462 size_t word_length=0, space_length=0;
463 while (n+word_length < length && line[n+word_length] != ' ')
465 space_length = word_length;
466 while (n+space_length < length && line[n+space_length] == ' ')
468 /* Send the word to the output machinery. */
469 output_word(first_indent, last_indent,
470 line+n, word_length, space_length-word_length);
476 new_paragraph(output_in_paragraph ? last_indent : first_indent, 0);
477 if (ferror(stream)) { warn("%s", name); ++n_errors; }
480 /* How long is the indent on this line?
483 indent_length(const wchar_t *line, size_t length) {
485 while (n<length && *line++ == ' ') ++n;
489 /* Might this line be a mail header?
490 * We deem a line to be a possible header if it matches the
491 * Perl regexp /^[A-Z][-A-Za-z0-9]*:\s/. This is *not* the same
492 * as in RFC whatever-number-it-is; we want to be gratuitously
493 * conservative to avoid mangling ordinary civilised text.
496 might_be_header(const wchar_t *line) {
497 if (!iswupper(*line++)) return 0;
498 while (*line && (iswalnum(*line) || *line=='-')) ++line;
499 return (*line==':' && iswspace(line[1]));
502 /* Begin a new paragraph with an indent of |indent| spaces.
505 new_paragraph(size_t old_indent, size_t indent) {
506 if (output_buffer_length) {
507 if (old_indent>0) output_indent(old_indent);
508 wprintf(L"%.*ls\n", (int)output_buffer_length, output_buffer);
510 x=indent; x0=0; output_buffer_length=0; pending_spaces=0;
511 output_in_paragraph = 0;
514 /* Output spaces or tabs for leading indentation.
517 output_indent(size_t n_spaces) {
518 if (output_tab_width) {
519 while (n_spaces >= output_tab_width) {
521 n_spaces -= output_tab_width;
524 while (n_spaces-- > 0) putwchar(' ');
527 /* Output a single word, or add it to the buffer.
528 * indent0 and indent1 are the indents to use on the first and subsequent
529 * lines of a paragraph. They'll often be the same, of course.
532 output_word(size_t indent0, size_t indent1, const wchar_t *word, size_t length, size_t spaces) {
534 size_t indent = output_in_paragraph ? indent1 : indent0;
539 for (p = word, width = 0; p < &word[length]; p++)
540 width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
542 new_x = x + pending_spaces + width;
544 /* If either |spaces==0| (at end of line) or |coalesce_spaces_P|
545 * (squashing internal whitespace), then add just one space;
546 * except that if the last character was a sentence-ender we
547 * actually add two spaces.
549 if (coalesce_spaces_P || spaces==0)
550 spaces = wcschr(sentence_enders, word[length-1]) ? 2 : 1;
552 if (new_x<=goal_length) {
553 /* After adding the word we still aren't at the goal length,
554 * so clearly we add it to the buffer rather than outputing it.
556 wmemset(output_buffer+output_buffer_length, L' ', pending_spaces);
557 x0 += pending_spaces; x += pending_spaces;
558 output_buffer_length += pending_spaces;
559 wmemcpy(output_buffer+output_buffer_length, word, length);
560 x0 += width; x += width; output_buffer_length += length;
561 pending_spaces = spaces;
564 /* Adding the word takes us past the goal. Print the line-so-far,
565 * and the word too iff either (1) the lsf is empty or (2) that
566 * makes us nearer the goal but doesn't take us over the limit,
567 * or (3) the word on its own takes us over the limit.
568 * In case (3) we put a newline in between.
570 if (indent>0) output_indent(indent);
571 wprintf(L"%.*ls", (int)output_buffer_length, output_buffer);
572 if (x0==0 || (new_x <= max_length && new_x-goal_length <= goal_length-x)) {
573 wprintf(L"%*ls", (int)pending_spaces, L"");
577 /* If the word takes us over the limit on its own, just
578 * spit it out and don't bother buffering it.
580 if (indent+width > max_length) {
582 if (indent>0) output_indent(indent);
584 wprintf(L"%.*ls", (int)length, word);
585 x0 = 0; x = indent1; pending_spaces = 0;
586 output_buffer_length = 0;
589 wmemcpy(output_buffer, word, length);
590 x0 = width; x = width+indent1; pending_spaces = spaces;
591 output_buffer_length = length;
595 output_in_paragraph = 1;
599 /* Process a stream, but just center its lines rather than trying to
600 * format them neatly.
603 center_stream(FILE *stream, const char *name) {
608 while ((line=get_line(stream, &length)) != 0) {
610 while (l>0 && iswspace(*line)) { ++line; --l; }
612 for (p = line, width = 0; p < &line[length]; p++)
613 width += (cwidth = wcwidth(*p)) > 0 ? cwidth : 1;
615 while (l<goal_length) { putwchar(' '); l+=2; }
616 wprintf(L"%.*ls\n", (int)length, line);
618 if (ferror(stream)) { warn("%s", name); ++n_errors; }
621 /* Get a single line from a stream. Expand tabs, strip control
622 * characters and trailing whitespace, and handle backspaces.
623 * Return the address of the buffer containing the line, and
624 * put the length of the line in |lengthp|.
625 * This can cope with arbitrarily long lines, and with lines
626 * without terminating \n.
627 * If there are no characters left or an error happens, we
629 * Don't confuse |spaces_pending| here with the global
633 get_line(FILE *stream, size_t *lengthp) {
634 static wchar_t *buf=NULL;
635 static size_t length=0;
638 size_t spaces_pending=0;
643 if (buf==NULL) { length=100; buf=XMALLOC(length * sizeof(wchar_t)); }
644 while ((ch=getwc(stream)) != '\n' && ch != WEOF) {
645 if (len+spaces_pending==0 && ch=='.' && !format_troff) troff=1;
646 if (ch==' ') ++spaces_pending;
647 else if (troff || iswprint(ch)) {
648 while (len+spaces_pending >= length) {
649 length*=2; buf=xrealloc(buf, length * sizeof(wchar_t));
651 while (spaces_pending > 0) { --spaces_pending; buf[len++]=' '; col++; }
653 col += (cwidth = wcwidth(ch)) > 0 ? cwidth : 1;
656 spaces_pending += tab_width - (col+spaces_pending)%tab_width;
657 else if (ch=='\b') { if (len) --len; if (col) --col; }
660 return (len>0 || ch!=WEOF) ? buf : 0;
663 /* (Re)allocate some memory, exiting with an error if we can't.
666 xrealloc(void *ptr, size_t nbytes) {
667 void *p = realloc(ptr, nbytes);
668 if (p == NULL) errx(EX_OSERR, "out of memory");