xref: /netbsd-src/lib/libc/stdio/vfscanf.c (revision 95d875fb90b1458e4f1de6950286ddcd6644bc61)
1 /*	$NetBSD: vfscanf.c,v 1.24 1999/09/20 04:39:34 lukem Exp $	*/
2 
3 /*-
4  * Copyright (c) 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Chris Torek.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 #if defined(LIBC_SCCS) && !defined(lint)
41 #if 0
42 static char sccsid[] = "@(#)vfscanf.c	8.1 (Berkeley) 6/4/93";
43 #else
44 __RCSID("$NetBSD: vfscanf.c,v 1.24 1999/09/20 04:39:34 lukem Exp $");
45 #endif
46 #endif /* LIBC_SCCS and not lint */
47 
48 #include "namespace.h"
49 
50 #include <assert.h>
51 #include <errno.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <ctype.h>
55 #if __STDC__
56 #include <stdarg.h>
57 #else
58 #include <varargs.h>
59 #endif
60 
61 #include "local.h"
62 
63 #ifdef FLOATING_POINT
64 #include "floatio.h"
65 #endif
66 
67 #define	BUF		513	/* Maximum length of numeric string. */
68 
69 /*
70  * Flags used during conversion.
71  */
72 #define	LONG		0x01	/* l: long or double */
73 #define	LONGDBL		0x02	/* L: long double; unimplemented */
74 #define	SHORT		0x04	/* h: short */
75 #define QUAD		0x08	/* q: quad */
76 #define	SUPPRESS	0x10	/* suppress assignment */
77 #define	POINTER		0x20	/* weird %p pointer (`fake hex') */
78 #define	NOSKIP		0x40	/* do not skip blanks */
79 
80 /*
81  * The following are used in numeric conversions only:
82  * SIGNOK, NDIGITS, DPTOK, and EXPOK are for floating point;
83  * SIGNOK, NDIGITS, PFXOK, and NZDIGITS are for integral.
84  */
85 #define	SIGNOK		0x080	/* +/- is (still) legal */
86 #define	NDIGITS		0x100	/* no digits detected */
87 
88 #define	DPTOK		0x200	/* (float) decimal point is still legal */
89 #define	EXPOK		0x400	/* (float) exponent (e+3, etc) still legal */
90 
91 #define	PFXOK		0x200	/* 0x prefix is (still) legal */
92 #define	NZDIGITS	0x400	/* no zero digits detected */
93 
94 /*
95  * Conversion types.
96  */
97 #define	CT_CHAR		0	/* %c conversion */
98 #define	CT_CCL		1	/* %[...] conversion */
99 #define	CT_STRING	2	/* %s conversion */
100 #define	CT_INT		3	/* integer, i.e., strtoq or strtouq */
101 #define	CT_FLOAT	4	/* floating, i.e., strtod */
102 
103 #define u_char unsigned char
104 #define u_long unsigned long
105 
106 static const u_char *__sccl __P((char *, const u_char *));
107 
108 /*
109  * vfscanf
110  */
111 int
112 __svfscanf(fp, fmt0, ap)
113 	FILE *fp;
114 	const char *fmt0;
115 	_BSD_VA_LIST_ ap;
116 {
117 	const u_char *fmt = (const u_char *)fmt0;
118 	int c;		/* character from format, or conversion */
119 	size_t width;	/* field width, or 0 */
120 	char *p;	/* points into all kinds of strings */
121 	int n;		/* handy integer */
122 	int flags;	/* flags as defined above */
123 	char *p0;	/* saves original value of p when necessary */
124 	int nassigned;		/* number of fields assigned */
125 	int nread;		/* number of characters consumed from fp */
126 	int base;		/* base argument to strtoq/strtouq */
127 	u_quad_t (*ccfn) __P((const char *, char **, int));
128 				/* conversion function (strtoq/strtouq) */
129 	char ccltab[256];	/* character class table for %[...] */
130 	char buf[BUF];		/* buffer for numeric conversions */
131 
132 	/* `basefix' is used to avoid `if' tests in the integer scanner */
133 	static const short basefix[17] =
134 		{ 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 };
135 
136 	_DIAGASSERT(fp != NULL);
137 	_DIAGASSERT(fmt0 != NULL);
138 
139 	nassigned = 0;
140 	nread = 0;
141 	base = 0;		/* XXX just to keep gcc happy */
142 	ccfn = NULL;		/* XXX just to keep gcc happy */
143 	for (;;) {
144 		c = *fmt++;
145 		if (c == 0)
146 			return (nassigned);
147 		if (isspace(c)) {
148 			while ((fp->_r > 0 || __srefill(fp) == 0) &&
149 			    isspace(*fp->_p))
150 				nread++, fp->_r--, fp->_p++;
151 			continue;
152 		}
153 		if (c != '%')
154 			goto literal;
155 		width = 0;
156 		flags = 0;
157 		/*
158 		 * switch on the format.  continue if done;
159 		 * break once format type is derived.
160 		 */
161 again:		c = *fmt++;
162 		switch (c) {
163 		case '%':
164 literal:
165 			if (fp->_r <= 0 && __srefill(fp))
166 				goto input_failure;
167 			if (*fp->_p != c)
168 				goto match_failure;
169 			fp->_r--, fp->_p++;
170 			nread++;
171 			continue;
172 
173 		case '*':
174 			flags |= SUPPRESS;
175 			goto again;
176 		case 'L':
177 			flags |= LONGDBL;
178 			goto again;
179 		case 'h':
180 			flags |= SHORT;
181 			goto again;
182 		case 'l':
183 			if (*fmt == 'l') {
184 				fmt++;
185 				flags |= QUAD;
186 			} else {
187 				flags |= LONG;
188 			}
189 			goto again;
190 		case 'q':
191 			flags |= QUAD;
192 			goto again;
193 
194 		case '0': case '1': case '2': case '3': case '4':
195 		case '5': case '6': case '7': case '8': case '9':
196 			width = width * 10 + c - '0';
197 			goto again;
198 
199 		/*
200 		 * Conversions.
201 		 * Those marked `compat' are for 4.[123]BSD compatibility.
202 		 *
203 		 * (According to ANSI, E and X formats are supposed
204 		 * to the same as e and x.  Sorry about that.)
205 		 */
206 		case 'D':	/* compat */
207 			flags |= LONG;
208 			/* FALLTHROUGH */
209 		case 'd':
210 			c = CT_INT;
211 			ccfn = (u_quad_t (*) __P((const char *, char **, int)))strtoq;
212 			base = 10;
213 			break;
214 
215 		case 'i':
216 			c = CT_INT;
217 			ccfn = (u_quad_t (*) __P((const char *, char **, int)))strtoq;
218 			base = 0;
219 			break;
220 
221 		case 'O':	/* compat */
222 			flags |= LONG;
223 			/* FALLTHROUGH */
224 		case 'o':
225 			c = CT_INT;
226 			ccfn = strtouq;
227 			base = 8;
228 			break;
229 
230 		case 'u':
231 			c = CT_INT;
232 			ccfn = strtouq;
233 			base = 10;
234 			break;
235 
236 		case 'X':
237 		case 'x':
238 			flags |= PFXOK;	/* enable 0x prefixing */
239 			c = CT_INT;
240 			ccfn = strtouq;
241 			base = 16;
242 			break;
243 
244 #ifdef FLOATING_POINT
245 		case 'E':
246 		case 'G':
247 		case 'e':
248 		case 'f':
249 		case 'g':
250 			c = CT_FLOAT;
251 			break;
252 #endif
253 
254 		case 's':
255 			c = CT_STRING;
256 			break;
257 
258 		case '[':
259 			fmt = __sccl(ccltab, fmt);
260 			flags |= NOSKIP;
261 			c = CT_CCL;
262 			break;
263 
264 		case 'c':
265 			flags |= NOSKIP;
266 			c = CT_CHAR;
267 			break;
268 
269 		case 'p':	/* pointer format is like hex */
270 			flags |= POINTER | PFXOK;
271 			c = CT_INT;
272 			ccfn = strtouq;
273 			base = 16;
274 			break;
275 
276 		case 'n':
277 			if (flags & SUPPRESS)	/* ??? */
278 				continue;
279 			if (flags & SHORT)
280 				*va_arg(ap, short *) = nread;
281 			else if (flags & LONG)
282 				*va_arg(ap, long *) = nread;
283 			else
284 				*va_arg(ap, int *) = nread;
285 			continue;
286 
287 		/*
288 		 * Disgusting backwards compatibility hacks.	XXX
289 		 */
290 		case '\0':	/* compat */
291 			return (EOF);
292 
293 		default:	/* compat */
294 			if (isupper(c))
295 				flags |= LONG;
296 			c = CT_INT;
297 			ccfn = (u_quad_t (*) __P((const char *, char **, int)))strtoq;
298 			base = 10;
299 			break;
300 		}
301 
302 		/*
303 		 * We have a conversion that requires input.
304 		 */
305 		if (fp->_r <= 0 && __srefill(fp))
306 			goto input_failure;
307 
308 		/*
309 		 * Consume leading white space, except for formats
310 		 * that suppress this.
311 		 */
312 		if ((flags & NOSKIP) == 0) {
313 			while (isspace(*fp->_p)) {
314 				nread++;
315 				if (--fp->_r > 0)
316 					fp->_p++;
317 				else if (__srefill(fp))
318 					goto input_failure;
319 			}
320 			/*
321 			 * Note that there is at least one character in
322 			 * the buffer, so conversions that do not set NOSKIP
323 			 * ca no longer result in an input failure.
324 			 */
325 		}
326 
327 		/*
328 		 * Do the conversion.
329 		 */
330 		switch (c) {
331 
332 		case CT_CHAR:
333 			/* scan arbitrary characters (sets NOSKIP) */
334 			if (width == 0)
335 				width = 1;
336 			if (flags & SUPPRESS) {
337 				size_t sum = 0;
338 				for (;;) {
339 					if ((n = fp->_r) < width) {
340 						sum += n;
341 						width -= n;
342 						fp->_p += n;
343 						if (__srefill(fp)) {
344 							if (sum == 0)
345 							    goto input_failure;
346 							break;
347 						}
348 					} else {
349 						sum += width;
350 						fp->_r -= width;
351 						fp->_p += width;
352 						break;
353 					}
354 				}
355 				nread += sum;
356 			} else {
357 				size_t r = fread((void *)va_arg(ap, char *), 1,
358 				    width, fp);
359 
360 				if (r == 0)
361 					goto input_failure;
362 				nread += r;
363 				nassigned++;
364 			}
365 			break;
366 
367 		case CT_CCL:
368 			/* scan a (nonempty) character class (sets NOSKIP) */
369 			if (width == 0)
370 				width = ~0U;	/* `infinity' */
371 			/* take only those things in the class */
372 			if (flags & SUPPRESS) {
373 				n = 0;
374 				while (ccltab[*fp->_p]) {
375 					n++, fp->_r--, fp->_p++;
376 					if (--width == 0)
377 						break;
378 					if (fp->_r <= 0 && __srefill(fp)) {
379 						if (n == 0)
380 							goto input_failure;
381 						break;
382 					}
383 				}
384 				if (n == 0)
385 					goto match_failure;
386 			} else {
387 				p0 = p = va_arg(ap, char *);
388 				while (ccltab[*fp->_p]) {
389 					fp->_r--;
390 					*p++ = *fp->_p++;
391 					if (--width == 0)
392 						break;
393 					if (fp->_r <= 0 && __srefill(fp)) {
394 						if (p == p0)
395 							goto input_failure;
396 						break;
397 					}
398 				}
399 				n = p - p0;
400 				if (n == 0)
401 					goto match_failure;
402 				*p = 0;
403 				nassigned++;
404 			}
405 			nread += n;
406 			break;
407 
408 		case CT_STRING:
409 			/* like CCL, but zero-length string OK, & no NOSKIP */
410 			if (width == 0)
411 				width = ~0U;
412 			if (flags & SUPPRESS) {
413 				n = 0;
414 				while (!isspace(*fp->_p)) {
415 					n++, fp->_r--, fp->_p++;
416 					if (--width == 0)
417 						break;
418 					if (fp->_r <= 0 && __srefill(fp))
419 						break;
420 				}
421 				nread += n;
422 			} else {
423 				p0 = p = va_arg(ap, char *);
424 				while (!isspace(*fp->_p)) {
425 					fp->_r--;
426 					*p++ = *fp->_p++;
427 					if (--width == 0)
428 						break;
429 					if (fp->_r <= 0 && __srefill(fp))
430 						break;
431 				}
432 				*p = 0;
433 				nread += p - p0;
434 				nassigned++;
435 			}
436 			continue;
437 
438 		case CT_INT:
439 			/* scan an integer as if by strtoq/strtouq */
440 #ifdef hardway
441 			if (width == 0 || width > sizeof(buf) - 1)
442 				width = sizeof(buf) - 1;
443 #else
444 			/* size_t is unsigned, hence this optimisation */
445 			if (--width > sizeof(buf) - 2)
446 				width = sizeof(buf) - 2;
447 			width++;
448 #endif
449 			flags |= SIGNOK | NDIGITS | NZDIGITS;
450 			for (p = buf; width; width--) {
451 				c = *fp->_p;
452 				/*
453 				 * Switch on the character; `goto ok'
454 				 * if we accept it as a part of number.
455 				 */
456 				switch (c) {
457 
458 				/*
459 				 * The digit 0 is always legal, but is
460 				 * special.  For %i conversions, if no
461 				 * digits (zero or nonzero) have been
462 				 * scanned (only signs), we will have
463 				 * base==0.  In that case, we should set
464 				 * it to 8 and enable 0x prefixing.
465 				 * Also, if we have not scanned zero digits
466 				 * before this, do not turn off prefixing
467 				 * (someone else will turn it off if we
468 				 * have scanned any nonzero digits).
469 				 */
470 				case '0':
471 					if (base == 0) {
472 						base = 8;
473 						flags |= PFXOK;
474 					}
475 					if (flags & NZDIGITS)
476 					    flags &= ~(SIGNOK|NZDIGITS|NDIGITS);
477 					else
478 					    flags &= ~(SIGNOK|PFXOK|NDIGITS);
479 					goto ok;
480 
481 				/* 1 through 7 always legal */
482 				case '1': case '2': case '3':
483 				case '4': case '5': case '6': case '7':
484 					base = basefix[base];
485 					flags &= ~(SIGNOK | PFXOK | NDIGITS);
486 					goto ok;
487 
488 				/* digits 8 and 9 ok iff decimal or hex */
489 				case '8': case '9':
490 					base = basefix[base];
491 					if (base <= 8)
492 						break;	/* not legal here */
493 					flags &= ~(SIGNOK | PFXOK | NDIGITS);
494 					goto ok;
495 
496 				/* letters ok iff hex */
497 				case 'A': case 'B': case 'C':
498 				case 'D': case 'E': case 'F':
499 				case 'a': case 'b': case 'c':
500 				case 'd': case 'e': case 'f':
501 					/* no need to fix base here */
502 					if (base <= 10)
503 						break;	/* not legal here */
504 					flags &= ~(SIGNOK | PFXOK | NDIGITS);
505 					goto ok;
506 
507 				/* sign ok only as first character */
508 				case '+': case '-':
509 					if (flags & SIGNOK) {
510 						flags &= ~SIGNOK;
511 						goto ok;
512 					}
513 					break;
514 
515 				/* x ok iff flag still set & 2nd char */
516 				case 'x': case 'X':
517 					if (flags & PFXOK && p == buf + 1) {
518 						base = 16;	/* if %i */
519 						flags &= ~PFXOK;
520 						goto ok;
521 					}
522 					break;
523 				}
524 
525 				/*
526 				 * If we got here, c is not a legal character
527 				 * for a number.  Stop accumulating digits.
528 				 */
529 				break;
530 		ok:
531 				/*
532 				 * c is legal: store it and look at the next.
533 				 */
534 				*p++ = c;
535 				if (--fp->_r > 0)
536 					fp->_p++;
537 				else if (__srefill(fp))
538 					break;		/* EOF */
539 			}
540 			/*
541 			 * If we had only a sign, it is no good; push
542 			 * back the sign.  If the number ends in `x',
543 			 * it was [sign] '0' 'x', so push back the x
544 			 * and treat it as [sign] '0'.
545 			 */
546 			if (flags & NDIGITS) {
547 				if (p > buf)
548 					(void) ungetc(*(u_char *)--p, fp);
549 				goto match_failure;
550 			}
551 			c = ((u_char *)p)[-1];
552 			if (c == 'x' || c == 'X') {
553 				--p;
554 				(void) ungetc(c, fp);
555 			}
556 			if ((flags & SUPPRESS) == 0) {
557 				u_quad_t res;
558 
559 				*p = 0;
560 				res = (*ccfn)(buf, (char **)NULL, base);
561 				if (flags & POINTER)
562 					*va_arg(ap, void **) =
563 					    (void *)(long)res;
564 				else if (flags & QUAD)
565 					*va_arg(ap, quad_t *) = res;
566 				else if (flags & LONG)
567 					*va_arg(ap, long *) = (long)res;
568 				else if (flags & SHORT)
569 					*va_arg(ap, short *) = (short)res;
570 				else
571 					*va_arg(ap, int *) = (int)res;
572 				nassigned++;
573 			}
574 			nread += p - buf;
575 			break;
576 
577 #ifdef FLOATING_POINT
578 		case CT_FLOAT:
579 			/* scan a floating point number as if by strtod */
580 #ifdef hardway
581 			if (width == 0 || width > sizeof(buf) - 1)
582 				width = sizeof(buf) - 1;
583 #else
584 			/* size_t is unsigned, hence this optimisation */
585 			if (--width > sizeof(buf) - 2)
586 				width = sizeof(buf) - 2;
587 			width++;
588 #endif
589 			flags |= SIGNOK | NDIGITS | DPTOK | EXPOK;
590 			for (p = buf; width; width--) {
591 				c = *fp->_p;
592 				/*
593 				 * This code mimicks the integer conversion
594 				 * code, but is much simpler.
595 				 */
596 				switch (c) {
597 
598 				case '0': case '1': case '2': case '3':
599 				case '4': case '5': case '6': case '7':
600 				case '8': case '9':
601 					flags &= ~(SIGNOK | NDIGITS);
602 					goto fok;
603 
604 				case '+': case '-':
605 					if (flags & SIGNOK) {
606 						flags &= ~SIGNOK;
607 						goto fok;
608 					}
609 					break;
610 				case '.':
611 					if (flags & DPTOK) {
612 						flags &= ~(SIGNOK | DPTOK);
613 						goto fok;
614 					}
615 					break;
616 				case 'e': case 'E':
617 					/* no exponent without some digits */
618 					if ((flags&(NDIGITS|EXPOK)) == EXPOK) {
619 						flags =
620 						    (flags & ~(EXPOK|DPTOK)) |
621 						    SIGNOK | NDIGITS;
622 						goto fok;
623 					}
624 					break;
625 				}
626 				break;
627 		fok:
628 				*p++ = c;
629 				if (--fp->_r > 0)
630 					fp->_p++;
631 				else if (__srefill(fp))
632 					break;	/* EOF */
633 			}
634 			/*
635 			 * If no digits, might be missing exponent digits
636 			 * (just give back the exponent) or might be missing
637 			 * regular digits, but had sign and/or decimal point.
638 			 */
639 			if (flags & NDIGITS) {
640 				if (flags & EXPOK) {
641 					/* no digits at all */
642 					while (p > buf)
643 						ungetc(*(u_char *)--p, fp);
644 					goto match_failure;
645 				}
646 				/* just a bad exponent (e and maybe sign) */
647 				c = *(u_char *)--p;
648 				if (c != 'e' && c != 'E') {
649 					(void) ungetc(c, fp);/* sign */
650 					c = *(u_char *)--p;
651 				}
652 				(void) ungetc(c, fp);
653 			}
654 			if ((flags & SUPPRESS) == 0) {
655 				double res;
656 
657 				*p = 0;
658 				res = strtod(buf, (char **) NULL);
659 				if (flags & LONGDBL)
660 					*va_arg(ap, long double *) = res;
661 				else if (flags & LONG)
662 					*va_arg(ap, double *) = res;
663 				else
664 					*va_arg(ap, float *) = res;
665 				nassigned++;
666 			}
667 			nread += p - buf;
668 			break;
669 #endif /* FLOATING_POINT */
670 		}
671 	}
672 input_failure:
673 	return (nassigned ? nassigned : EOF);
674 match_failure:
675 	return (nassigned);
676 }
677 
678 /*
679  * Fill in the given table from the scanset at the given format
680  * (just after `[').  Return a pointer to the character past the
681  * closing `]'.  The table has a 1 wherever characters should be
682  * considered part of the scanset.
683  */
684 static const u_char *
685 __sccl(tab, fmt)
686 	char *tab;
687 	const u_char *fmt;
688 {
689 	int c, n, v;
690 
691 	_DIAGASSERT(tab != NULL);
692 	_DIAGASSERT(fmt != NULL);
693 
694 	/* first `clear' the whole table */
695 	c = *fmt++;		/* first char hat => negated scanset */
696 	if (c == '^') {
697 		v = 1;		/* default => accept */
698 		c = *fmt++;	/* get new first char */
699 	} else
700 		v = 0;		/* default => reject */
701 	/* should probably use memset here */
702 	for (n = 0; n < 256; n++)
703 		tab[n] = v;
704 	if (c == 0)
705 		return (fmt - 1);/* format ended before closing ] */
706 
707 	/*
708 	 * Now set the entries corresponding to the actual scanset
709 	 * to the opposite of the above.
710 	 *
711 	 * The first character may be ']' (or '-') without being special;
712 	 * the last character may be '-'.
713 	 */
714 	v = 1 - v;
715 	for (;;) {
716 		tab[c] = v;		/* take character c */
717 doswitch:
718 		n = *fmt++;		/* and examine the next */
719 		switch (n) {
720 
721 		case 0:			/* format ended too soon */
722 			return (fmt - 1);
723 
724 		case '-':
725 			/*
726 			 * A scanset of the form
727 			 *	[01+-]
728 			 * is defined as `the digit 0, the digit 1,
729 			 * the character +, the character -', but
730 			 * the effect of a scanset such as
731 			 *	[a-zA-Z0-9]
732 			 * is implementation defined.  The V7 Unix
733 			 * scanf treats `a-z' as `the letters a through
734 			 * z', but treats `a-a' as `the letter a, the
735 			 * character -, and the letter a'.
736 			 *
737 			 * For compatibility, the `-' is not considerd
738 			 * to define a range if the character following
739 			 * it is either a close bracket (required by ANSI)
740 			 * or is not numerically greater than the character
741 			 * we just stored in the table (c).
742 			 */
743 			n = *fmt;
744 			if (n == ']' || n < c) {
745 				c = '-';
746 				break;	/* resume the for(;;) */
747 			}
748 			fmt++;
749 			do {		/* fill in the range */
750 				tab[++c] = v;
751 			} while (c < n);
752 #if 1	/* XXX another disgusting compatibility hack */
753 			/*
754 			 * Alas, the V7 Unix scanf also treats formats
755 			 * such as [a-c-e] as `the letters a through e'.
756 			 * This too is permitted by the standard....
757 			 */
758 			goto doswitch;
759 #else
760 			c = *fmt++;
761 			if (c == 0)
762 				return (fmt - 1);
763 			if (c == ']')
764 				return (fmt);
765 			break;
766 #endif
767 
768 		case ']':		/* end of scanset */
769 			return (fmt);
770 
771 		default:		/* just another character */
772 			c = n;
773 			break;
774 		}
775 	}
776 	/* NOTREACHED */
777 }
778