xref: /freebsd/usr.bin/fmt/fmt.c (revision 6990ffd8a95caaba6858ad44ff1b3157d1efba8f)
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  * 3. The "everything else" is split into words; a word
34  *    includes its trailing whitespace, and a word at the
35  *    end of a line is deemed to be followed by a single
36  *    space, or two spaces if it ends with a sentence-end
37  *    character. (See the `-d' option for how to change that.)
38  *    If the `-s' option has been given, then a word's trailing
39  *    whitespace is replaced by what it would have had if it
40  *    had occurred at end of line.
41  * 4. Each paragraph is sent to standard output as follows.
42  *    We output the leading whitespace, and then enough words
43  *    to make the line length as near as possible to the goal
44  *    without exceeding the maximum. (If a single word would
45  *    exceed the maximum, we output that anyway.) Of course
46  *    the trailing whitespace of the last word is ignored.
47  *    We then emit a newline and start again if there are any
48  *    words left.
49  *    Note that for a blank line this translates as "We emit
50  *    a newline".
51  *    If the `-l <n>' option is given, then leading whitespace
52  *    is modified slightly: <n> spaces are replaced by a tab.
53  *    Indented paragraphs (see above under `-p') make matters
54  *    more complicated than this suggests. Actually every paragraph
55  *    has two `leading whitespace' values; the value for the first
56  *    line, and the value for the most recent line. (While processing
57  *    the first line, the two are equal. When `-p' has not been
58  *    given, they are always equal.) The leading whitespace
59  *    actually output is that of the first line (for the first
60  *    line of *output*) or that of the most recent line (for
61  *    all other lines of output).
62  *    When `-m' has been given, message header paragraphs are
63  *    taken as having first-leading-whitespace empty and
64  *    subsequent-leading-whitespace two spaces.
65  *
66  * Multiple input files are formatted one at a time, so that a file
67  * never ends in the middle of a line.
68  *
69  * There's an alternative mode of operation, invoked by giving
70  * the `-c' option. In that case we just center every line,
71  * and most of the other options are ignored. This should
72  * really be in a separate program, but we must stay compatible
73  * with old `fmt'.
74  *
75  * QUERY: Should `-m' also try to do the right thing with quoted text?
76  * QUERY: `-b' to treat backslashed whitespace as old `fmt' does?
77  * QUERY: Option meaning `never join lines'?
78  * QUERY: Option meaning `split in mid-word to avoid overlong lines'?
79  * (Those last two might not be useful, since we have `fold'.)
80  *
81  * Differences from old `fmt':
82  *
83  *   - We have many more options. Options that aren't understood
84  *     generate a lengthy usage message, rather than being
85  *     treated as filenames.
86  *   - Even with `-m', our handling of message headers is
87  *     significantly different. (And much better.)
88  *   - We don't treat `\ ' as non-word-breaking.
89  *   - Downward changes of indentation start new paragraphs
90  *     for us, as well as upward. (I think old `fmt' behaves
91  *     in the way it does in order to allow indented paragraphs,
92  *     but this is a broken way of making indented paragraphs
93  *     behave right.)
94  *   - Given the choice of going over or under |goal_length|
95  *     by the same amount, we go over; old `fmt' goes under.
96  *   - We treat `?' as ending a sentence, and not `:'. Old `fmt'
97  *     does the reverse.
98  *   - We return approved return codes. Old `fmt' returns
99  *     1 for some errors, and *the number of unopenable files*
100  *     when that was all that went wrong.
101  *   - We have fewer crashes and more helpful error messages.
102  *   - We don't turn spaces into tabs at starts of lines unless
103  *     specifically requested.
104  *   - New `fmt' is somewhat smaller and slightly faster than
105  *     old `fmt'.
106  *
107  * Bugs:
108  *
109  *   None known. There probably are some, though.
110  *
111  * Portability:
112  *
113  *   I believe this code to be pretty portable. It does require
114  *   that you have `getopt'. If you need to include "getopt.h"
115  *   for this (e.g., if your system didn't come with `getopt'
116  *   and you installed it yourself) then you should arrange for
117  *   NEED_getopt_h to be #defined.
118  *
119  *   Everything here should work OK even on nasty 16-bit
120  *   machines and nice 64-bit ones. However, it's only really
121  *   been tested on my FreeBSD machine. Your mileage may vary.
122  */
123 
124 /* Copyright (c) 1997 Gareth McCaughan. All rights reserved.
125  *
126  * Redistribution and use of this code, in source or binary forms,
127  * with or without modification, are permitted subject to the following
128  * conditions:
129  *
130  *  - Redistribution of source code must retain the above copyright
131  *    notice, this list of conditions and the following disclaimer.
132  *
133  *  - If you distribute modified source code it must also include
134  *    a notice saying that it has been modified, and giving a brief
135  *    description of what changes have been made.
136  *
137  * Disclaimer: I am not responsible for the results of using this code.
138  *             If it formats your hard disc, sends obscene messages to
139  *             your boss and kills your children then that's your problem
140  *             not mine. I give absolutely no warranty of any sort as to
141  *             what the program will do, and absolutely refuse to be held
142  *             liable for any consequences of your using it.
143  *             Thank you. Have a nice day.
144  */
145 
146 /* RCS change log:
147  * Revision 1.5  1998/03/02 18:02:21  gjm11
148  * Minor changes for portability.
149  *
150  * Revision 1.4  1997/10/01 11:51:28  gjm11
151  * Repair broken indented-paragraph handling.
152  * Add mail message header stuff.
153  * Improve comments and layout.
154  * Make usable with non-BSD systems.
155  * Add revision display to usage message.
156  *
157  * Revision 1.3  1997/09/30 16:24:47  gjm11
158  * Add copyright notice, rcsid string and log message.
159  *
160  * Revision 1.2  1997/09/30 16:13:39  gjm11
161  * Add options: -d <chars>, -l <width>, -p, -s, -t <width>, -h .
162  * Parse options with `getopt'. Clean up code generally.
163  * Make comments more accurate.
164  *
165  * Revision 1.1  1997/09/30 11:29:57  gjm11
166  * Initial revision
167  */
168 
169 #ifndef lint
170 static const char rcsid[] =
171   "$FreeBSD$";
172 static const char copyright[] =
173   "Copyright (c) 1997 Gareth McCaughan. All rights reserved.\n";
174 #endif /* not lint */
175 
176 #include <ctype.h>
177 #include <err.h>
178 #include <locale.h>
179 #include <stdio.h>
180 #include <stdlib.h>
181 #include <string.h>
182 #include <sysexits.h>
183 #include <unistd.h>
184 
185 /* Something that, we hope, will never be a genuine line length,
186  * indentation etc.
187  */
188 #define SILLY ((size_t)-1)
189 
190 /* I used to use |strtoul| for this, but (1) not all systems have it
191  * and (2) it's probably better to use |strtol| to detect negative
192  * numbers better.
193  * If |fussyp==0| then we don't complain about non-numbers
194  * (returning 0 instead), but we do complain about bad numbers.
195  */
196 static size_t
197 get_positive(const char *s, const char *err_mess, int fussyP) {
198   char *t;
199   long result = strtol(s,&t,0);
200   if (*t) { if (fussyP) goto Lose; else return 0; }
201   if (result<=0) { Lose: errx(EX_USAGE, "%s", err_mess); }
202   return (size_t) result;
203 }
204 
205 static size_t
206 get_nonnegative(const char *s, const char *err_mess, int fussyP) {
207   char *t;
208   long result = strtol(s,&t,0);
209   if (*t) { if (fussyP) goto Lose; else return 0; }
210   if (result<0) { Lose: errx(EX_USAGE, "%s", err_mess); }
211   return (size_t) result;
212 }
213 
214 /* Global variables */
215 
216 static int centerP=0;		/* Try to center lines? */
217 static size_t goal_length=0;	/* Target length for output lines */
218 static size_t max_length=0;	/* Maximum length for output lines */
219 static int coalesce_spaces_P=0;	/* Coalesce multiple whitespace -> ' ' ? */
220 static int allow_indented_paragraphs=0;	/* Can first line have diff. ind.? */
221 static int tab_width=8;		/* Number of spaces per tab stop */
222 static size_t output_tab_width=8;	/* Ditto, when squashing leading spaces */
223 static const char *sentence_enders=".?!";	/* Double-space after these */
224 static int grok_mail_headers=0;	/* treat embedded mail headers magically? */
225 
226 static int n_errors=0;		/* Number of failed files. Return on exit. */
227 static char *output_buffer=0;	/* 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 pending_spaces;	/* Spaces to add before next word */
231 static int output_in_paragraph=0;	/* Any of current para written out yet? */
232 
233 /* Prototypes */
234 
235 static void process_named_file (const char *);
236 static void     process_stream (FILE *, const char *);
237 static size_t    indent_length (const char *, size_t);
238 static int     might_be_header (const unsigned char *);
239 static void      new_paragraph (size_t, size_t);
240 static void        output_word (size_t, size_t, const char *, size_t, size_t);
241 static void      output_indent (size_t);
242 static void      center_stream (FILE *, const char *);
243 static char *         get_line (FILE *, size_t *);
244 static void *         xrealloc (void *, size_t);
245 
246 #define XMALLOC(x) xrealloc(0,x)
247 
248 /* Here is perhaps the right place to mention that this code is
249  * all in top-down order. Hence, |main| comes first.
250  */
251 int
252 main(int argc, char *argv[]) {
253   int ch;			/* used for |getopt| processing */
254 
255 
256   (void) setlocale(LC_CTYPE, "");
257 
258   /* 1. Grok parameters. */
259 
260   while ((ch = getopt(argc, argv, "0123456789cd:hl:mpst:w:")) != -1)
261   switch(ch) {
262     case 'c':
263       centerP = 1;
264       continue;
265     case 'd':
266       sentence_enders = optarg;
267       continue;
268     case 'l':
269       output_tab_width
270         = get_nonnegative(optarg, "output tab width must be non-negative", 1);
271       continue;
272     case 'm':
273       grok_mail_headers = 1;
274       continue;
275     case 'p':
276       allow_indented_paragraphs = 1;
277       continue;
278     case 's':
279       coalesce_spaces_P = 1;
280       continue;
281     case 't':
282       tab_width = get_positive(optarg, "tab width must be positive", 1);
283       continue;
284     case 'w':
285       goal_length = get_positive(optarg, "width must be positive", 1);
286       max_length = goal_length;
287       continue;
288     case '0': case '1': case '2': case '3': case '4': case '5':
289     case '6': case '7': case '8': case '9':
290     /* XXX  this is not a stylistically approved use of getopt() */
291       if (goal_length==0) {
292         char *p;
293         p = argv[optind - 1];
294         if (p[0] == '-' && p[1] == ch && !p[2])
295              goal_length = get_positive(++p, "width must be nonzero", 1);
296         else
297              goal_length = get_positive(argv[optind]+1,
298                  "width must be nonzero", 1);
299         max_length = goal_length;
300       }
301       continue;
302     case 'h': default:
303       fprintf(stderr,
304 "Usage:   fmt [-cmps] [-d chars] [-l num] [-t num]\n"
305 "             [-w width | -width | goal [maximum]] [file ...]\n"
306 "Options: -c     center each line instead of formatting\n"
307 "         -d <chars> double-space after <chars> at line end\n"
308 "         -l <n> turn each <n> spaces at start of line into a tab\n"
309 "         -m     try to make sure mail header lines stay separate\n"
310 "         -p     allow indented paragraphs\n"
311 "         -s     coalesce whitespace inside lines\n"
312 "         -t <n> have tabs every <n> columns\n"
313 "         -w <n> set maximum width to <n>\n"
314 "         goal   set target width to goal\n");
315       exit(ch=='h' ? 0 : EX_USAGE);
316   }
317   argc -= optind; argv += optind;
318 
319   /* [ goal [ maximum ] ] */
320 
321   if (argc>0 && goal_length==0
322       && (goal_length=get_positive(*argv,"goal length must be positive", 0))
323          != 0) {
324     --argc; ++argv;
325     if (argc>0
326         && (max_length=get_positive(*argv,"max length must be positive", 0))
327            != 0) {
328       --argc; ++argv;
329       if (max_length<goal_length)
330         errx(EX_USAGE, "max length must be >= goal length");
331     }
332   }
333   if (goal_length==0) goal_length = 65;
334   if (max_length==0) max_length = goal_length+10;
335   output_buffer = XMALLOC(max_length+1);	/* really needn't be longer */
336 
337   /* 2. Process files. */
338 
339   if (argc>0) {
340     while (argc-->0) process_named_file(*argv++);
341   }
342   else {
343     process_stream(stdin, "standard input");
344   }
345 
346   /* We're done. */
347 
348   return n_errors ? EX_NOINPUT : 0;
349 
350 }
351 
352 /* Process a single file, given its name.
353  */
354 static void
355 process_named_file(const char *name) {
356   FILE *f=fopen(name, "r");
357   if (!f) { perror(name); ++n_errors; }
358   else {
359     process_stream(f, name);
360     fclose(f);
361   }
362 }
363 
364 /* Types of mail header continuation lines:
365  */
366 typedef enum {
367   hdr_ParagraphStart = -1,
368   hdr_NonHeader      = 0,
369   hdr_Header         = 1,
370   hdr_Continuation   = 2
371 } HdrType;
372 
373 /* Process a stream. This is where the real work happens,
374  * except that centering is handled separately.
375  */
376 static void
377 process_stream(FILE *stream, const char *name) {
378   size_t last_indent=SILLY;	/* how many spaces in last indent? */
379   size_t para_line_number=0;	/* how many lines already read in this para? */
380   size_t first_indent=SILLY;	/* indentation of line 0 of paragraph */
381   HdrType prev_header_type=hdr_ParagraphStart;
382 	/* ^-- header_type of previous line; -1 at para start */
383   char *line;
384   size_t length;
385 
386   if (centerP) { center_stream(stream, name); return; }
387   while ((line=get_line(stream,&length)) != NULL) {
388     size_t np=indent_length(line, length);
389     { HdrType header_type=hdr_NonHeader;
390       if (grok_mail_headers && prev_header_type!=hdr_NonHeader) {
391         if (np==0 && might_be_header(line))
392           header_type = hdr_Header;
393         else if (np>0 && prev_header_type>hdr_NonHeader)
394           header_type = hdr_Continuation;
395       }
396       /* We need a new paragraph if and only if:
397        *   this line is blank,
398        *   OR it's a mail header,
399        *   OR it's not a mail header AND the last line was one,
400        *   OR the indentation has changed
401        *      AND the line isn't a mail header continuation line
402        *      AND this isn't the second line of an indented paragraph.
403        */
404       if ( length==0
405            || header_type==hdr_Header
406            || (header_type==hdr_NonHeader && prev_header_type>hdr_NonHeader)
407            || (np!=last_indent
408                && header_type != hdr_Continuation
409                && (!allow_indented_paragraphs || para_line_number != 1)) ) {
410         new_paragraph(output_in_paragraph ? last_indent : first_indent, np);
411         para_line_number = 0;
412         first_indent = np;
413         last_indent = np;
414         if (header_type==hdr_Header) last_indent=2;	/* for cont. lines */
415         if (length==0) {
416           putchar('\n');
417           prev_header_type=hdr_ParagraphStart;
418           continue;
419         }
420       }
421       else {
422         /* If this is an indented paragraph other than a mail header
423          * continuation, set |last_indent|.
424          */
425         if (np != last_indent && header_type != hdr_Continuation)
426           last_indent=np;
427       }
428       prev_header_type = header_type;
429     }
430 
431     { size_t n=np;
432       while (n<length) {
433         /* Find word end and count spaces after it */
434         size_t word_length=0, space_length=0;
435         while (n+word_length < length && line[n+word_length] != ' ')
436           ++word_length;
437         space_length = word_length;
438         while (n+space_length < length && line[n+space_length] == ' ')
439           ++space_length;
440         /* Send the word to the output machinery. */
441         output_word(first_indent, last_indent,
442                     line+n, word_length, space_length-word_length);
443         n += space_length;
444       }
445     }
446     ++para_line_number;
447   }
448   new_paragraph(output_in_paragraph ? last_indent : first_indent, 0);
449   if (ferror(stream)) { perror(name); ++n_errors; }
450 }
451 
452 /* How long is the indent on this line?
453  */
454 static size_t
455 indent_length(const char *line, size_t length) {
456   size_t n=0;
457   while (n<length && *line++ == ' ') ++n;
458   return n;
459 }
460 
461 /* Might this line be a mail header?
462  * We deem a line to be a possible header if it matches the
463  * Perl regexp /^[A-Z][-A-Za-z0-9]*:\s/. This is *not* the same
464  * as in RFC whatever-number-it-is; we want to be gratuitously
465  * conservative to avoid mangling ordinary civilised text.
466  */
467 static int
468 might_be_header(const unsigned char *line) {
469   if (!isupper(*line++)) return 0;
470   while (*line && (isalnum(*line) || *line=='-')) ++line;
471   return (*line==':' && isspace(line[1]));
472 }
473 
474 /* Begin a new paragraph with an indent of |indent| spaces.
475  */
476 static void
477 new_paragraph(size_t old_indent, size_t indent) {
478   if (x0) {
479     if (old_indent>0) output_indent(old_indent);
480     fwrite(output_buffer, 1, x0, stdout);
481     putchar('\n');
482   }
483   x=indent; x0=0; pending_spaces=0;
484   output_in_paragraph = 0;
485 }
486 
487 /* Output spaces or tabs for leading indentation.
488  */
489 static void
490 output_indent(size_t n_spaces) {
491   if (output_tab_width) {
492     while (n_spaces >= output_tab_width) {
493       putchar('\t');
494       n_spaces -= output_tab_width;
495     }
496   }
497   while (n_spaces-- > 0) putchar(' ');
498 }
499 
500 /* Output a single word, or add it to the buffer.
501  * indent0 and indent1 are the indents to use on the first and subsequent
502  * lines of a paragraph. They'll often be the same, of course.
503  */
504 static void
505 output_word(size_t indent0, size_t indent1, const char *word, size_t length, size_t spaces) {
506   size_t new_x = x+pending_spaces+length;
507   size_t indent = output_in_paragraph ? indent1 : indent0;
508 
509   /* If either |spaces==0| (at end of line) or |coalesce_spaces_P|
510    * (squashing internal whitespace), then add just one space;
511    * except that if the last character was a sentence-ender we
512    * actually add two spaces.
513    */
514   if (coalesce_spaces_P || spaces==0)
515     spaces = strchr(sentence_enders, word[length-1]) ? 2 : 1;
516 
517   if (new_x<=goal_length) {
518     /* After adding the word we still aren't at the goal length,
519      * so clearly we add it to the buffer rather than outputing it.
520      */
521     memset(output_buffer+x0, ' ', pending_spaces);
522     x0 += pending_spaces; x += pending_spaces;
523     memcpy(output_buffer+x0, word, length);
524     x0 += length; x += length;
525     pending_spaces = spaces;
526   }
527   else {
528     /* Adding the word takes us past the goal. Print the line-so-far,
529      * and the word too iff either (1) the lsf is empty or (2) that
530      * makes us nearer the goal but doesn't take us over the limit,
531      * or (3) the word on its own takes us over the limit.
532      * In case (3) we put a newline in between.
533      */
534     if (indent>0) output_indent(indent);
535     fwrite(output_buffer, 1, x0, stdout);
536     if (x0==0 || (new_x <= max_length && new_x-goal_length <= goal_length-x)) {
537       printf("%*s", (int)pending_spaces, "");
538       goto write_out_word;
539     }
540     else {
541       /* If the word takes us over the limit on its own, just
542        * spit it out and don't bother buffering it.
543        */
544       if (indent+length > max_length) {
545         putchar('\n');
546         if (indent>0) output_indent(indent);
547 write_out_word:
548         fwrite(word, 1, length, stdout);
549         x0 = 0; x = indent1; pending_spaces = 0;
550       }
551       else {
552         memcpy(output_buffer, word, length);
553         x0 = length; x = length+indent1; pending_spaces = spaces;
554       }
555     }
556     putchar('\n');
557     output_in_paragraph = 1;
558   }
559 }
560 
561 /* Process a stream, but just center its lines rather than trying to
562  * format them neatly.
563  */
564 static void
565 center_stream(FILE *stream, const char *name) {
566   char *line;
567   size_t length;
568   while ((line=get_line(stream, &length)) != 0) {
569     size_t l=length;
570     while (l>0 && isspace(*line)) { ++line; --l; }
571     length=l;
572     while (l<goal_length) { putchar(' '); l+=2; }
573     fwrite(line, 1, length, stdout);
574     putchar('\n');
575   }
576   if (ferror(stream)) { perror(name); ++n_errors; }
577 }
578 
579 /* Get a single line from a stream. Expand tabs, strip control
580  * characters and trailing whitespace, and handle backspaces.
581  * Return the address of the buffer containing the line, and
582  * put the length of the line in |lengthp|.
583  * This can cope with arbitrarily long lines, and with lines
584  * without terminating \n.
585  * If there are no characters left or an error happens, we
586  * return 0.
587  * Don't confuse |spaces_pending| here with the global
588  * |pending_spaces|.
589  */
590 static char *
591 get_line(FILE *stream, size_t *lengthp) {
592   static char *buf=NULL;
593   static size_t length=0;
594   size_t len=0;
595   int ch;
596   size_t spaces_pending=0;
597 
598   if (buf==NULL) { length=100; buf=XMALLOC(length); }
599   while ((ch=getc(stream)) != '\n' && ch != EOF) {
600     if (ch==' ') ++spaces_pending;
601     else if (isprint(ch)) {
602       while (len+spaces_pending >= length) {
603         length*=2; buf=xrealloc(buf, length);
604       }
605       while (spaces_pending > 0) { --spaces_pending; buf[len++]=' '; }
606       buf[len++] = ch;
607     }
608     else if (ch=='\t')
609       spaces_pending += tab_width - (len+spaces_pending)%tab_width;
610     else if (ch=='\b') { if (len) --len; }
611   }
612   *lengthp=len;
613   return (len>0 || ch!=EOF) ? buf : 0;
614 }
615 
616 /* (Re)allocate some memory, exiting with an error if we can't.
617  */
618 static void *
619 xrealloc(void *ptr, size_t nbytes) {
620   void *p = realloc(ptr, nbytes);
621   if (p == NULL) errx(EX_OSERR, "out of memory");
622   return p;
623 }
624