1 /*
2  * Copyright (c) 1983, 1995, 1996 Eric P. Allman
3  * Copyright (c) 1988, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  * Portions Copyright (c) 1996-2021, PostgreSQL Global Development Group
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *	  notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *	  notice, this list of conditions and the following disclaimer in the
14  *	  documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *	  may be used to endorse or promote products derived from this software
17  *	  without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * src/port/snprintf.c
32  */
33 
34 #include "c.h"
35 
36 #include <math.h>
37 
38 /*
39  * We used to use the platform's NL_ARGMAX here, but that's a bad idea,
40  * first because the point of this module is to remove platform dependencies
41  * not perpetuate them, and second because some platforms use ridiculously
42  * large values, leading to excessive stack consumption in dopr().
43  */
44 #define PG_NL_ARGMAX 31
45 
46 
47 /*
48  *	SNPRINTF, VSNPRINTF and friends
49  *
50  * These versions have been grabbed off the net.  They have been
51  * cleaned up to compile properly and support for most of the C99
52  * specification has been added.  Remaining unimplemented features are:
53  *
54  * 1. No locale support: the radix character is always '.' and the '
55  * (single quote) format flag is ignored.
56  *
57  * 2. No support for the "%n" format specification.
58  *
59  * 3. No support for wide characters ("lc" and "ls" formats).
60  *
61  * 4. No support for "long double" ("Lf" and related formats).
62  *
63  * 5. Space and '#' flags are not implemented.
64  *
65  * In addition, we support some extensions over C99:
66  *
67  * 1. Argument order control through "%n$" and "*n$", as required by POSIX.
68  *
69  * 2. "%m" expands to the value of strerror(errno), where errno is the
70  * value that variable had at the start of the call.  This is a glibc
71  * extension, but a very useful one.
72  *
73  *
74  * Historically the result values of sprintf/snprintf varied across platforms.
75  * This implementation now follows the C99 standard:
76  *
77  * 1. -1 is returned if an error is detected in the format string, or if
78  * a write to the target stream fails (as reported by fwrite).  Note that
79  * overrunning snprintf's target buffer is *not* an error.
80  *
81  * 2. For successful writes to streams, the actual number of bytes written
82  * to the stream is returned.
83  *
84  * 3. For successful sprintf/snprintf, the number of bytes that would have
85  * been written to an infinite-size buffer (excluding the trailing '\0')
86  * is returned.  snprintf will truncate its output to fit in the buffer
87  * (ensuring a trailing '\0' unless count == 0), but this is not reflected
88  * in the function result.
89  *
90  * snprintf buffer overrun can be detected by checking for function result
91  * greater than or equal to the supplied count.
92  */
93 
94 /**************************************************************
95  * Original:
96  * Patrick Powell Tue Apr 11 09:48:21 PDT 1995
97  * A bombproof version of doprnt (dopr) included.
98  * Sigh.  This sort of thing is always nasty do deal with.  Note that
99  * the version here does not include floating point. (now it does ... tgl)
100  **************************************************************/
101 
102 /* Prevent recursion */
103 #undef	vsnprintf
104 #undef	snprintf
105 #undef	vsprintf
106 #undef	sprintf
107 #undef	vfprintf
108 #undef	fprintf
109 #undef	vprintf
110 #undef	printf
111 
112 /*
113  * Info about where the formatted output is going.
114  *
115  * dopr and subroutines will not write at/past bufend, but snprintf
116  * reserves one byte, ensuring it may place the trailing '\0' there.
117  *
118  * In snprintf, we use nchars to count the number of bytes dropped on the
119  * floor due to buffer overrun.  The correct result of snprintf is thus
120  * (bufptr - bufstart) + nchars.  (This isn't as inconsistent as it might
121  * seem: nchars is the number of emitted bytes that are not in the buffer now,
122  * either because we sent them to the stream or because we couldn't fit them
123  * into the buffer to begin with.)
124  */
125 typedef struct
126 {
127 	char	   *bufptr;			/* next buffer output position */
128 	char	   *bufstart;		/* first buffer element */
129 	char	   *bufend;			/* last+1 buffer element, or NULL */
130 	/* bufend == NULL is for sprintf, where we assume buf is big enough */
131 	FILE	   *stream;			/* eventual output destination, or NULL */
132 	int			nchars;			/* # chars sent to stream, or dropped */
133 	bool		failed;			/* call is a failure; errno is set */
134 } PrintfTarget;
135 
136 /*
137  * Info about the type and value of a formatting parameter.  Note that we
138  * don't currently support "long double", "wint_t", or "wchar_t *" data,
139  * nor the '%n' formatting code; else we'd need more types.  Also, at this
140  * level we need not worry about signed vs unsigned values.
141  */
142 typedef enum
143 {
144 	ATYPE_NONE = 0,
145 	ATYPE_INT,
146 	ATYPE_LONG,
147 	ATYPE_LONGLONG,
148 	ATYPE_DOUBLE,
149 	ATYPE_CHARPTR
150 } PrintfArgType;
151 
152 typedef union
153 {
154 	int			i;
155 	long		l;
156 	long long	ll;
157 	double		d;
158 	char	   *cptr;
159 } PrintfArgValue;
160 
161 
162 static void flushbuffer(PrintfTarget *target);
163 static void dopr(PrintfTarget *target, const char *format, va_list args);
164 
165 
166 /*
167  * Externally visible entry points.
168  *
169  * All of these are just wrappers around dopr().  Note it's essential that
170  * they not change the value of "errno" before reaching dopr().
171  */
172 
173 int
pg_vsnprintf(char * str,size_t count,const char * fmt,va_list args)174 pg_vsnprintf(char *str, size_t count, const char *fmt, va_list args)
175 {
176 	PrintfTarget target;
177 	char		onebyte[1];
178 
179 	/*
180 	 * C99 allows the case str == NULL when count == 0.  Rather than
181 	 * special-casing this situation further down, we substitute a one-byte
182 	 * local buffer.  Callers cannot tell, since the function result doesn't
183 	 * depend on count.
184 	 */
185 	if (count == 0)
186 	{
187 		str = onebyte;
188 		count = 1;
189 	}
190 	target.bufstart = target.bufptr = str;
191 	target.bufend = str + count - 1;
192 	target.stream = NULL;
193 	target.nchars = 0;
194 	target.failed = false;
195 	dopr(&target, fmt, args);
196 	*(target.bufptr) = '\0';
197 	return target.failed ? -1 : (target.bufptr - target.bufstart
198 								 + target.nchars);
199 }
200 
201 int
pg_snprintf(char * str,size_t count,const char * fmt,...)202 pg_snprintf(char *str, size_t count, const char *fmt,...)
203 {
204 	int			len;
205 	va_list		args;
206 
207 	va_start(args, fmt);
208 	len = pg_vsnprintf(str, count, fmt, args);
209 	va_end(args);
210 	return len;
211 }
212 
213 int
pg_vsprintf(char * str,const char * fmt,va_list args)214 pg_vsprintf(char *str, const char *fmt, va_list args)
215 {
216 	PrintfTarget target;
217 
218 	target.bufstart = target.bufptr = str;
219 	target.bufend = NULL;
220 	target.stream = NULL;
221 	target.nchars = 0;			/* not really used in this case */
222 	target.failed = false;
223 	dopr(&target, fmt, args);
224 	*(target.bufptr) = '\0';
225 	return target.failed ? -1 : (target.bufptr - target.bufstart
226 								 + target.nchars);
227 }
228 
229 int
pg_sprintf(char * str,const char * fmt,...)230 pg_sprintf(char *str, const char *fmt,...)
231 {
232 	int			len;
233 	va_list		args;
234 
235 	va_start(args, fmt);
236 	len = pg_vsprintf(str, fmt, args);
237 	va_end(args);
238 	return len;
239 }
240 
241 int
pg_vfprintf(FILE * stream,const char * fmt,va_list args)242 pg_vfprintf(FILE *stream, const char *fmt, va_list args)
243 {
244 	PrintfTarget target;
245 	char		buffer[1024];	/* size is arbitrary */
246 
247 	if (stream == NULL)
248 	{
249 		errno = EINVAL;
250 		return -1;
251 	}
252 	target.bufstart = target.bufptr = buffer;
253 	target.bufend = buffer + sizeof(buffer);	/* use the whole buffer */
254 	target.stream = stream;
255 	target.nchars = 0;
256 	target.failed = false;
257 	dopr(&target, fmt, args);
258 	/* dump any remaining buffer contents */
259 	flushbuffer(&target);
260 	return target.failed ? -1 : target.nchars;
261 }
262 
263 int
pg_fprintf(FILE * stream,const char * fmt,...)264 pg_fprintf(FILE *stream, const char *fmt,...)
265 {
266 	int			len;
267 	va_list		args;
268 
269 	va_start(args, fmt);
270 	len = pg_vfprintf(stream, fmt, args);
271 	va_end(args);
272 	return len;
273 }
274 
275 int
pg_vprintf(const char * fmt,va_list args)276 pg_vprintf(const char *fmt, va_list args)
277 {
278 	return pg_vfprintf(stdout, fmt, args);
279 }
280 
281 int
pg_printf(const char * fmt,...)282 pg_printf(const char *fmt,...)
283 {
284 	int			len;
285 	va_list		args;
286 
287 	va_start(args, fmt);
288 	len = pg_vfprintf(stdout, fmt, args);
289 	va_end(args);
290 	return len;
291 }
292 
293 /*
294  * Attempt to write the entire buffer to target->stream; discard the entire
295  * buffer in any case.  Call this only when target->stream is defined.
296  */
297 static void
flushbuffer(PrintfTarget * target)298 flushbuffer(PrintfTarget *target)
299 {
300 	size_t		nc = target->bufptr - target->bufstart;
301 
302 	/*
303 	 * Don't write anything if we already failed; this is to ensure we
304 	 * preserve the original failure's errno.
305 	 */
306 	if (!target->failed && nc > 0)
307 	{
308 		size_t		written;
309 
310 		written = fwrite(target->bufstart, 1, nc, target->stream);
311 		target->nchars += written;
312 		if (written != nc)
313 			target->failed = true;
314 	}
315 	target->bufptr = target->bufstart;
316 }
317 
318 
319 static bool find_arguments(const char *format, va_list args,
320 						   PrintfArgValue *argvalues);
321 static void fmtstr(const char *value, int leftjust, int minlen, int maxwidth,
322 				   int pointflag, PrintfTarget *target);
323 static void fmtptr(const void *value, PrintfTarget *target);
324 static void fmtint(long long value, char type, int forcesign,
325 				   int leftjust, int minlen, int zpad, int precision, int pointflag,
326 				   PrintfTarget *target);
327 static void fmtchar(int value, int leftjust, int minlen, PrintfTarget *target);
328 static void fmtfloat(double value, char type, int forcesign,
329 					 int leftjust, int minlen, int zpad, int precision, int pointflag,
330 					 PrintfTarget *target);
331 static void dostr(const char *str, int slen, PrintfTarget *target);
332 static void dopr_outch(int c, PrintfTarget *target);
333 static void dopr_outchmulti(int c, int slen, PrintfTarget *target);
334 static int	adjust_sign(int is_negative, int forcesign, int *signvalue);
335 static int	compute_padlen(int minlen, int vallen, int leftjust);
336 static void leading_pad(int zpad, int signvalue, int *padlen,
337 						PrintfTarget *target);
338 static void trailing_pad(int padlen, PrintfTarget *target);
339 
340 /*
341  * If strchrnul exists (it's a glibc-ism), it's a good bit faster than the
342  * equivalent manual loop.  If it doesn't exist, provide a replacement.
343  *
344  * Note: glibc declares this as returning "char *", but that would require
345  * casting away const internally, so we don't follow that detail.
346  */
347 #ifndef HAVE_STRCHRNUL
348 
349 static inline const char *
strchrnul(const char * s,int c)350 strchrnul(const char *s, int c)
351 {
352 	while (*s != '\0' && *s != c)
353 		s++;
354 	return s;
355 }
356 
357 #else
358 
359 /*
360  * glibc's <string.h> declares strchrnul only if _GNU_SOURCE is defined.
361  * While we typically use that on glibc platforms, configure will set
362  * HAVE_STRCHRNUL whether it's used or not.  Fill in the missing declaration
363  * so that this file will compile cleanly with or without _GNU_SOURCE.
364  */
365 #ifndef _GNU_SOURCE
366 extern char *strchrnul(const char *s, int c);
367 #endif
368 
369 #endif							/* HAVE_STRCHRNUL */
370 
371 
372 /*
373  * dopr(): the guts of *printf for all cases.
374  */
375 static void
dopr(PrintfTarget * target,const char * format,va_list args)376 dopr(PrintfTarget *target, const char *format, va_list args)
377 {
378 	int			save_errno = errno;
379 	const char *first_pct = NULL;
380 	int			ch;
381 	bool		have_dollar;
382 	bool		have_star;
383 	bool		afterstar;
384 	int			accum;
385 	int			longlongflag;
386 	int			longflag;
387 	int			pointflag;
388 	int			leftjust;
389 	int			fieldwidth;
390 	int			precision;
391 	int			zpad;
392 	int			forcesign;
393 	int			fmtpos;
394 	int			cvalue;
395 	long long	numvalue;
396 	double		fvalue;
397 	const char *strvalue;
398 	PrintfArgValue argvalues[PG_NL_ARGMAX + 1];
399 
400 	/*
401 	 * Initially, we suppose the format string does not use %n$.  The first
402 	 * time we come to a conversion spec that has that, we'll call
403 	 * find_arguments() to check for consistent use of %n$ and fill the
404 	 * argvalues array with the argument values in the correct order.
405 	 */
406 	have_dollar = false;
407 
408 	while (*format != '\0')
409 	{
410 		/* Locate next conversion specifier */
411 		if (*format != '%')
412 		{
413 			/* Scan to next '%' or end of string */
414 			const char *next_pct = strchrnul(format + 1, '%');
415 
416 			/* Dump literal data we just scanned over */
417 			dostr(format, next_pct - format, target);
418 			if (target->failed)
419 				break;
420 
421 			if (*next_pct == '\0')
422 				break;
423 			format = next_pct;
424 		}
425 
426 		/*
427 		 * Remember start of first conversion spec; if we find %n$, then it's
428 		 * sufficient for find_arguments() to start here, without rescanning
429 		 * earlier literal text.
430 		 */
431 		if (first_pct == NULL)
432 			first_pct = format;
433 
434 		/* Process conversion spec starting at *format */
435 		format++;
436 
437 		/* Fast path for conversion spec that is exactly %s */
438 		if (*format == 's')
439 		{
440 			format++;
441 			strvalue = va_arg(args, char *);
442 			if (strvalue == NULL)
443 				strvalue = "(null)";
444 			dostr(strvalue, strlen(strvalue), target);
445 			if (target->failed)
446 				break;
447 			continue;
448 		}
449 
450 		fieldwidth = precision = zpad = leftjust = forcesign = 0;
451 		longflag = longlongflag = pointflag = 0;
452 		fmtpos = accum = 0;
453 		have_star = afterstar = false;
454 nextch2:
455 		ch = *format++;
456 		switch (ch)
457 		{
458 			case '-':
459 				leftjust = 1;
460 				goto nextch2;
461 			case '+':
462 				forcesign = 1;
463 				goto nextch2;
464 			case '0':
465 				/* set zero padding if no nonzero digits yet */
466 				if (accum == 0 && !pointflag)
467 					zpad = '0';
468 				/* FALL THRU */
469 			case '1':
470 			case '2':
471 			case '3':
472 			case '4':
473 			case '5':
474 			case '6':
475 			case '7':
476 			case '8':
477 			case '9':
478 				accum = accum * 10 + (ch - '0');
479 				goto nextch2;
480 			case '.':
481 				if (have_star)
482 					have_star = false;
483 				else
484 					fieldwidth = accum;
485 				pointflag = 1;
486 				accum = 0;
487 				goto nextch2;
488 			case '*':
489 				if (have_dollar)
490 				{
491 					/*
492 					 * We'll process value after reading n$.  Note it's OK to
493 					 * assume have_dollar is set correctly, because in a valid
494 					 * format string the initial % must have had n$ if * does.
495 					 */
496 					afterstar = true;
497 				}
498 				else
499 				{
500 					/* fetch and process value now */
501 					int			starval = va_arg(args, int);
502 
503 					if (pointflag)
504 					{
505 						precision = starval;
506 						if (precision < 0)
507 						{
508 							precision = 0;
509 							pointflag = 0;
510 						}
511 					}
512 					else
513 					{
514 						fieldwidth = starval;
515 						if (fieldwidth < 0)
516 						{
517 							leftjust = 1;
518 							fieldwidth = -fieldwidth;
519 						}
520 					}
521 				}
522 				have_star = true;
523 				accum = 0;
524 				goto nextch2;
525 			case '$':
526 				/* First dollar sign? */
527 				if (!have_dollar)
528 				{
529 					/* Yup, so examine all conversion specs in format */
530 					if (!find_arguments(first_pct, args, argvalues))
531 						goto bad_format;
532 					have_dollar = true;
533 				}
534 				if (afterstar)
535 				{
536 					/* fetch and process star value */
537 					int			starval = argvalues[accum].i;
538 
539 					if (pointflag)
540 					{
541 						precision = starval;
542 						if (precision < 0)
543 						{
544 							precision = 0;
545 							pointflag = 0;
546 						}
547 					}
548 					else
549 					{
550 						fieldwidth = starval;
551 						if (fieldwidth < 0)
552 						{
553 							leftjust = 1;
554 							fieldwidth = -fieldwidth;
555 						}
556 					}
557 					afterstar = false;
558 				}
559 				else
560 					fmtpos = accum;
561 				accum = 0;
562 				goto nextch2;
563 			case 'l':
564 				if (longflag)
565 					longlongflag = 1;
566 				else
567 					longflag = 1;
568 				goto nextch2;
569 			case 'z':
570 #if SIZEOF_SIZE_T == 8
571 #ifdef HAVE_LONG_INT_64
572 				longflag = 1;
573 #elif defined(HAVE_LONG_LONG_INT_64)
574 				longlongflag = 1;
575 #else
576 #error "Don't know how to print 64bit integers"
577 #endif
578 #else
579 				/* assume size_t is same size as int */
580 #endif
581 				goto nextch2;
582 			case 'h':
583 			case '\'':
584 				/* ignore these */
585 				goto nextch2;
586 			case 'd':
587 			case 'i':
588 				if (!have_star)
589 				{
590 					if (pointflag)
591 						precision = accum;
592 					else
593 						fieldwidth = accum;
594 				}
595 				if (have_dollar)
596 				{
597 					if (longlongflag)
598 						numvalue = argvalues[fmtpos].ll;
599 					else if (longflag)
600 						numvalue = argvalues[fmtpos].l;
601 					else
602 						numvalue = argvalues[fmtpos].i;
603 				}
604 				else
605 				{
606 					if (longlongflag)
607 						numvalue = va_arg(args, long long);
608 					else if (longflag)
609 						numvalue = va_arg(args, long);
610 					else
611 						numvalue = va_arg(args, int);
612 				}
613 				fmtint(numvalue, ch, forcesign, leftjust, fieldwidth, zpad,
614 					   precision, pointflag, target);
615 				break;
616 			case 'o':
617 			case 'u':
618 			case 'x':
619 			case 'X':
620 				if (!have_star)
621 				{
622 					if (pointflag)
623 						precision = accum;
624 					else
625 						fieldwidth = accum;
626 				}
627 				if (have_dollar)
628 				{
629 					if (longlongflag)
630 						numvalue = (unsigned long long) argvalues[fmtpos].ll;
631 					else if (longflag)
632 						numvalue = (unsigned long) argvalues[fmtpos].l;
633 					else
634 						numvalue = (unsigned int) argvalues[fmtpos].i;
635 				}
636 				else
637 				{
638 					if (longlongflag)
639 						numvalue = (unsigned long long) va_arg(args, long long);
640 					else if (longflag)
641 						numvalue = (unsigned long) va_arg(args, long);
642 					else
643 						numvalue = (unsigned int) va_arg(args, int);
644 				}
645 				fmtint(numvalue, ch, forcesign, leftjust, fieldwidth, zpad,
646 					   precision, pointflag, target);
647 				break;
648 			case 'c':
649 				if (!have_star)
650 				{
651 					if (pointflag)
652 						precision = accum;
653 					else
654 						fieldwidth = accum;
655 				}
656 				if (have_dollar)
657 					cvalue = (unsigned char) argvalues[fmtpos].i;
658 				else
659 					cvalue = (unsigned char) va_arg(args, int);
660 				fmtchar(cvalue, leftjust, fieldwidth, target);
661 				break;
662 			case 's':
663 				if (!have_star)
664 				{
665 					if (pointflag)
666 						precision = accum;
667 					else
668 						fieldwidth = accum;
669 				}
670 				if (have_dollar)
671 					strvalue = argvalues[fmtpos].cptr;
672 				else
673 					strvalue = va_arg(args, char *);
674 				/* If string is NULL, silently substitute "(null)" */
675 				if (strvalue == NULL)
676 					strvalue = "(null)";
677 				fmtstr(strvalue, leftjust, fieldwidth, precision, pointflag,
678 					   target);
679 				break;
680 			case 'p':
681 				/* fieldwidth/leftjust are ignored ... */
682 				if (have_dollar)
683 					strvalue = argvalues[fmtpos].cptr;
684 				else
685 					strvalue = va_arg(args, char *);
686 				fmtptr((const void *) strvalue, target);
687 				break;
688 			case 'e':
689 			case 'E':
690 			case 'f':
691 			case 'g':
692 			case 'G':
693 				if (!have_star)
694 				{
695 					if (pointflag)
696 						precision = accum;
697 					else
698 						fieldwidth = accum;
699 				}
700 				if (have_dollar)
701 					fvalue = argvalues[fmtpos].d;
702 				else
703 					fvalue = va_arg(args, double);
704 				fmtfloat(fvalue, ch, forcesign, leftjust,
705 						 fieldwidth, zpad,
706 						 precision, pointflag,
707 						 target);
708 				break;
709 			case 'm':
710 				{
711 					char		errbuf[PG_STRERROR_R_BUFLEN];
712 					const char *errm = strerror_r(save_errno,
713 												  errbuf, sizeof(errbuf));
714 
715 					dostr(errm, strlen(errm), target);
716 				}
717 				break;
718 			case '%':
719 				dopr_outch('%', target);
720 				break;
721 			default:
722 
723 				/*
724 				 * Anything else --- in particular, '\0' indicating end of
725 				 * format string --- is bogus.
726 				 */
727 				goto bad_format;
728 		}
729 
730 		/* Check for failure after each conversion spec */
731 		if (target->failed)
732 			break;
733 	}
734 
735 	return;
736 
737 bad_format:
738 	errno = EINVAL;
739 	target->failed = true;
740 }
741 
742 /*
743  * find_arguments(): sort out the arguments for a format spec with %n$
744  *
745  * If format is valid, return true and fill argvalues[i] with the value
746  * for the conversion spec that has %i$ or *i$.  Else return false.
747  */
748 static bool
find_arguments(const char * format,va_list args,PrintfArgValue * argvalues)749 find_arguments(const char *format, va_list args,
750 			   PrintfArgValue *argvalues)
751 {
752 	int			ch;
753 	bool		afterstar;
754 	int			accum;
755 	int			longlongflag;
756 	int			longflag;
757 	int			fmtpos;
758 	int			i;
759 	int			last_dollar;
760 	PrintfArgType argtypes[PG_NL_ARGMAX + 1];
761 
762 	/* Initialize to "no dollar arguments known" */
763 	last_dollar = 0;
764 	MemSet(argtypes, 0, sizeof(argtypes));
765 
766 	/*
767 	 * This loop must accept the same format strings as the one in dopr().
768 	 * However, we don't need to analyze them to the same level of detail.
769 	 *
770 	 * Since we're only called if there's a dollar-type spec somewhere, we can
771 	 * fail immediately if we find a non-dollar spec.  Per the C99 standard,
772 	 * all argument references in the format string must be one or the other.
773 	 */
774 	while (*format != '\0')
775 	{
776 		/* Locate next conversion specifier */
777 		if (*format != '%')
778 		{
779 			/* Unlike dopr, we can just quit if there's no more specifiers */
780 			format = strchr(format + 1, '%');
781 			if (format == NULL)
782 				break;
783 		}
784 
785 		/* Process conversion spec starting at *format */
786 		format++;
787 		longflag = longlongflag = 0;
788 		fmtpos = accum = 0;
789 		afterstar = false;
790 nextch1:
791 		ch = *format++;
792 		switch (ch)
793 		{
794 			case '-':
795 			case '+':
796 				goto nextch1;
797 			case '0':
798 			case '1':
799 			case '2':
800 			case '3':
801 			case '4':
802 			case '5':
803 			case '6':
804 			case '7':
805 			case '8':
806 			case '9':
807 				accum = accum * 10 + (ch - '0');
808 				goto nextch1;
809 			case '.':
810 				accum = 0;
811 				goto nextch1;
812 			case '*':
813 				if (afterstar)
814 					return false;	/* previous star missing dollar */
815 				afterstar = true;
816 				accum = 0;
817 				goto nextch1;
818 			case '$':
819 				if (accum <= 0 || accum > PG_NL_ARGMAX)
820 					return false;
821 				if (afterstar)
822 				{
823 					if (argtypes[accum] &&
824 						argtypes[accum] != ATYPE_INT)
825 						return false;
826 					argtypes[accum] = ATYPE_INT;
827 					last_dollar = Max(last_dollar, accum);
828 					afterstar = false;
829 				}
830 				else
831 					fmtpos = accum;
832 				accum = 0;
833 				goto nextch1;
834 			case 'l':
835 				if (longflag)
836 					longlongflag = 1;
837 				else
838 					longflag = 1;
839 				goto nextch1;
840 			case 'z':
841 #if SIZEOF_SIZE_T == 8
842 #ifdef HAVE_LONG_INT_64
843 				longflag = 1;
844 #elif defined(HAVE_LONG_LONG_INT_64)
845 				longlongflag = 1;
846 #else
847 #error "Don't know how to print 64bit integers"
848 #endif
849 #else
850 				/* assume size_t is same size as int */
851 #endif
852 				goto nextch1;
853 			case 'h':
854 			case '\'':
855 				/* ignore these */
856 				goto nextch1;
857 			case 'd':
858 			case 'i':
859 			case 'o':
860 			case 'u':
861 			case 'x':
862 			case 'X':
863 				if (fmtpos)
864 				{
865 					PrintfArgType atype;
866 
867 					if (longlongflag)
868 						atype = ATYPE_LONGLONG;
869 					else if (longflag)
870 						atype = ATYPE_LONG;
871 					else
872 						atype = ATYPE_INT;
873 					if (argtypes[fmtpos] &&
874 						argtypes[fmtpos] != atype)
875 						return false;
876 					argtypes[fmtpos] = atype;
877 					last_dollar = Max(last_dollar, fmtpos);
878 				}
879 				else
880 					return false;	/* non-dollar conversion spec */
881 				break;
882 			case 'c':
883 				if (fmtpos)
884 				{
885 					if (argtypes[fmtpos] &&
886 						argtypes[fmtpos] != ATYPE_INT)
887 						return false;
888 					argtypes[fmtpos] = ATYPE_INT;
889 					last_dollar = Max(last_dollar, fmtpos);
890 				}
891 				else
892 					return false;	/* non-dollar conversion spec */
893 				break;
894 			case 's':
895 			case 'p':
896 				if (fmtpos)
897 				{
898 					if (argtypes[fmtpos] &&
899 						argtypes[fmtpos] != ATYPE_CHARPTR)
900 						return false;
901 					argtypes[fmtpos] = ATYPE_CHARPTR;
902 					last_dollar = Max(last_dollar, fmtpos);
903 				}
904 				else
905 					return false;	/* non-dollar conversion spec */
906 				break;
907 			case 'e':
908 			case 'E':
909 			case 'f':
910 			case 'g':
911 			case 'G':
912 				if (fmtpos)
913 				{
914 					if (argtypes[fmtpos] &&
915 						argtypes[fmtpos] != ATYPE_DOUBLE)
916 						return false;
917 					argtypes[fmtpos] = ATYPE_DOUBLE;
918 					last_dollar = Max(last_dollar, fmtpos);
919 				}
920 				else
921 					return false;	/* non-dollar conversion spec */
922 				break;
923 			case 'm':
924 			case '%':
925 				break;
926 			default:
927 				return false;	/* bogus format string */
928 		}
929 
930 		/*
931 		 * If we finish the spec with afterstar still set, there's a
932 		 * non-dollar star in there.
933 		 */
934 		if (afterstar)
935 			return false;		/* non-dollar conversion spec */
936 	}
937 
938 	/*
939 	 * Format appears valid so far, so collect the arguments in physical
940 	 * order.  (Since we rejected any non-dollar specs that would have
941 	 * collected arguments, we know that dopr() hasn't collected any yet.)
942 	 */
943 	for (i = 1; i <= last_dollar; i++)
944 	{
945 		switch (argtypes[i])
946 		{
947 			case ATYPE_NONE:
948 				return false;
949 			case ATYPE_INT:
950 				argvalues[i].i = va_arg(args, int);
951 				break;
952 			case ATYPE_LONG:
953 				argvalues[i].l = va_arg(args, long);
954 				break;
955 			case ATYPE_LONGLONG:
956 				argvalues[i].ll = va_arg(args, long long);
957 				break;
958 			case ATYPE_DOUBLE:
959 				argvalues[i].d = va_arg(args, double);
960 				break;
961 			case ATYPE_CHARPTR:
962 				argvalues[i].cptr = va_arg(args, char *);
963 				break;
964 		}
965 	}
966 
967 	return true;
968 }
969 
970 static void
fmtstr(const char * value,int leftjust,int minlen,int maxwidth,int pointflag,PrintfTarget * target)971 fmtstr(const char *value, int leftjust, int minlen, int maxwidth,
972 	   int pointflag, PrintfTarget *target)
973 {
974 	int			padlen,
975 				vallen;			/* amount to pad */
976 
977 	/*
978 	 * If a maxwidth (precision) is specified, we must not fetch more bytes
979 	 * than that.
980 	 */
981 	if (pointflag)
982 		vallen = strnlen(value, maxwidth);
983 	else
984 		vallen = strlen(value);
985 
986 	padlen = compute_padlen(minlen, vallen, leftjust);
987 
988 	if (padlen > 0)
989 	{
990 		dopr_outchmulti(' ', padlen, target);
991 		padlen = 0;
992 	}
993 
994 	dostr(value, vallen, target);
995 
996 	trailing_pad(padlen, target);
997 }
998 
999 static void
fmtptr(const void * value,PrintfTarget * target)1000 fmtptr(const void *value, PrintfTarget *target)
1001 {
1002 	int			vallen;
1003 	char		convert[64];
1004 
1005 	/* we rely on regular C library's sprintf to do the basic conversion */
1006 	vallen = sprintf(convert, "%p", value);
1007 	if (vallen < 0)
1008 		target->failed = true;
1009 	else
1010 		dostr(convert, vallen, target);
1011 }
1012 
1013 static void
fmtint(long long value,char type,int forcesign,int leftjust,int minlen,int zpad,int precision,int pointflag,PrintfTarget * target)1014 fmtint(long long value, char type, int forcesign, int leftjust,
1015 	   int minlen, int zpad, int precision, int pointflag,
1016 	   PrintfTarget *target)
1017 {
1018 	unsigned long long base;
1019 	unsigned long long uvalue;
1020 	int			dosign;
1021 	const char *cvt = "0123456789abcdef";
1022 	int			signvalue = 0;
1023 	char		convert[64];
1024 	int			vallen = 0;
1025 	int			padlen;			/* amount to pad */
1026 	int			zeropad;		/* extra leading zeroes */
1027 
1028 	switch (type)
1029 	{
1030 		case 'd':
1031 		case 'i':
1032 			base = 10;
1033 			dosign = 1;
1034 			break;
1035 		case 'o':
1036 			base = 8;
1037 			dosign = 0;
1038 			break;
1039 		case 'u':
1040 			base = 10;
1041 			dosign = 0;
1042 			break;
1043 		case 'x':
1044 			base = 16;
1045 			dosign = 0;
1046 			break;
1047 		case 'X':
1048 			cvt = "0123456789ABCDEF";
1049 			base = 16;
1050 			dosign = 0;
1051 			break;
1052 		default:
1053 			return;				/* keep compiler quiet */
1054 	}
1055 
1056 	/* disable MSVC warning about applying unary minus to an unsigned value */
1057 #ifdef _MSC_VER
1058 #pragma warning(push)
1059 #pragma warning(disable: 4146)
1060 #endif
1061 	/* Handle +/- */
1062 	if (dosign && adjust_sign((value < 0), forcesign, &signvalue))
1063 		uvalue = -(unsigned long long) value;
1064 	else
1065 		uvalue = (unsigned long long) value;
1066 #ifdef _MSC_VER
1067 #pragma warning(pop)
1068 #endif
1069 
1070 	/*
1071 	 * SUS: the result of converting 0 with an explicit precision of 0 is no
1072 	 * characters
1073 	 */
1074 	if (value == 0 && pointflag && precision == 0)
1075 		vallen = 0;
1076 	else
1077 	{
1078 		/* make integer string */
1079 		do
1080 		{
1081 			convert[sizeof(convert) - (++vallen)] = cvt[uvalue % base];
1082 			uvalue = uvalue / base;
1083 		} while (uvalue);
1084 	}
1085 
1086 	zeropad = Max(0, precision - vallen);
1087 
1088 	padlen = compute_padlen(minlen, vallen + zeropad, leftjust);
1089 
1090 	leading_pad(zpad, signvalue, &padlen, target);
1091 
1092 	if (zeropad > 0)
1093 		dopr_outchmulti('0', zeropad, target);
1094 
1095 	dostr(convert + sizeof(convert) - vallen, vallen, target);
1096 
1097 	trailing_pad(padlen, target);
1098 }
1099 
1100 static void
fmtchar(int value,int leftjust,int minlen,PrintfTarget * target)1101 fmtchar(int value, int leftjust, int minlen, PrintfTarget *target)
1102 {
1103 	int			padlen;			/* amount to pad */
1104 
1105 	padlen = compute_padlen(minlen, 1, leftjust);
1106 
1107 	if (padlen > 0)
1108 	{
1109 		dopr_outchmulti(' ', padlen, target);
1110 		padlen = 0;
1111 	}
1112 
1113 	dopr_outch(value, target);
1114 
1115 	trailing_pad(padlen, target);
1116 }
1117 
1118 static void
fmtfloat(double value,char type,int forcesign,int leftjust,int minlen,int zpad,int precision,int pointflag,PrintfTarget * target)1119 fmtfloat(double value, char type, int forcesign, int leftjust,
1120 		 int minlen, int zpad, int precision, int pointflag,
1121 		 PrintfTarget *target)
1122 {
1123 	int			signvalue = 0;
1124 	int			prec;
1125 	int			vallen;
1126 	char		fmt[8];
1127 	char		convert[1024];
1128 	int			zeropadlen = 0; /* amount to pad with zeroes */
1129 	int			padlen;			/* amount to pad with spaces */
1130 
1131 	/*
1132 	 * We rely on the regular C library's sprintf to do the basic conversion,
1133 	 * then handle padding considerations here.
1134 	 *
1135 	 * The dynamic range of "double" is about 1E+-308 for IEEE math, and not
1136 	 * too wildly more than that with other hardware.  In "f" format, sprintf
1137 	 * could therefore generate at most 308 characters to the left of the
1138 	 * decimal point; while we need to allow the precision to get as high as
1139 	 * 308+17 to ensure that we don't truncate significant digits from very
1140 	 * small values.  To handle both these extremes, we use a buffer of 1024
1141 	 * bytes and limit requested precision to 350 digits; this should prevent
1142 	 * buffer overrun even with non-IEEE math.  If the original precision
1143 	 * request was more than 350, separately pad with zeroes.
1144 	 *
1145 	 * We handle infinities and NaNs specially to ensure platform-independent
1146 	 * output.
1147 	 */
1148 	if (precision < 0)			/* cover possible overflow of "accum" */
1149 		precision = 0;
1150 	prec = Min(precision, 350);
1151 
1152 	if (isnan(value))
1153 	{
1154 		strcpy(convert, "NaN");
1155 		vallen = 3;
1156 		/* no zero padding, regardless of precision spec */
1157 	}
1158 	else
1159 	{
1160 		/*
1161 		 * Handle sign (NaNs have no sign, so we don't do this in the case
1162 		 * above).  "value < 0.0" will not be true for IEEE minus zero, so we
1163 		 * detect that by looking for the case where value equals 0.0
1164 		 * according to == but not according to memcmp.
1165 		 */
1166 		static const double dzero = 0.0;
1167 
1168 		if (adjust_sign((value < 0.0 ||
1169 						 (value == 0.0 &&
1170 						  memcmp(&value, &dzero, sizeof(double)) != 0)),
1171 						forcesign, &signvalue))
1172 			value = -value;
1173 
1174 		if (isinf(value))
1175 		{
1176 			strcpy(convert, "Infinity");
1177 			vallen = 8;
1178 			/* no zero padding, regardless of precision spec */
1179 		}
1180 		else if (pointflag)
1181 		{
1182 			zeropadlen = precision - prec;
1183 			fmt[0] = '%';
1184 			fmt[1] = '.';
1185 			fmt[2] = '*';
1186 			fmt[3] = type;
1187 			fmt[4] = '\0';
1188 			vallen = sprintf(convert, fmt, prec, value);
1189 		}
1190 		else
1191 		{
1192 			fmt[0] = '%';
1193 			fmt[1] = type;
1194 			fmt[2] = '\0';
1195 			vallen = sprintf(convert, fmt, value);
1196 		}
1197 		if (vallen < 0)
1198 			goto fail;
1199 
1200 		/*
1201 		 * Windows, alone among our supported platforms, likes to emit
1202 		 * three-digit exponent fields even when two digits would do.  Hack
1203 		 * such results to look like the way everyone else does it.
1204 		 */
1205 #ifdef WIN32
1206 		if (vallen >= 6 &&
1207 			convert[vallen - 5] == 'e' &&
1208 			convert[vallen - 3] == '0')
1209 		{
1210 			convert[vallen - 3] = convert[vallen - 2];
1211 			convert[vallen - 2] = convert[vallen - 1];
1212 			vallen--;
1213 		}
1214 #endif
1215 	}
1216 
1217 	padlen = compute_padlen(minlen, vallen + zeropadlen, leftjust);
1218 
1219 	leading_pad(zpad, signvalue, &padlen, target);
1220 
1221 	if (zeropadlen > 0)
1222 	{
1223 		/* If 'e' or 'E' format, inject zeroes before the exponent */
1224 		char	   *epos = strrchr(convert, 'e');
1225 
1226 		if (!epos)
1227 			epos = strrchr(convert, 'E');
1228 		if (epos)
1229 		{
1230 			/* pad before exponent */
1231 			dostr(convert, epos - convert, target);
1232 			dopr_outchmulti('0', zeropadlen, target);
1233 			dostr(epos, vallen - (epos - convert), target);
1234 		}
1235 		else
1236 		{
1237 			/* no exponent, pad after the digits */
1238 			dostr(convert, vallen, target);
1239 			dopr_outchmulti('0', zeropadlen, target);
1240 		}
1241 	}
1242 	else
1243 	{
1244 		/* no zero padding, just emit the number as-is */
1245 		dostr(convert, vallen, target);
1246 	}
1247 
1248 	trailing_pad(padlen, target);
1249 	return;
1250 
1251 fail:
1252 	target->failed = true;
1253 }
1254 
1255 /*
1256  * Nonstandard entry point to print a double value efficiently.
1257  *
1258  * This is approximately equivalent to strfromd(), but has an API more
1259  * adapted to what float8out() wants.  The behavior is like snprintf()
1260  * with a format of "%.ng", where n is the specified precision.
1261  * However, the target buffer must be nonempty (i.e. count > 0), and
1262  * the precision is silently bounded to a sane range.
1263  */
1264 int
pg_strfromd(char * str,size_t count,int precision,double value)1265 pg_strfromd(char *str, size_t count, int precision, double value)
1266 {
1267 	PrintfTarget target;
1268 	int			signvalue = 0;
1269 	int			vallen;
1270 	char		fmt[8];
1271 	char		convert[64];
1272 
1273 	/* Set up the target like pg_snprintf, but require nonempty buffer */
1274 	Assert(count > 0);
1275 	target.bufstart = target.bufptr = str;
1276 	target.bufend = str + count - 1;
1277 	target.stream = NULL;
1278 	target.nchars = 0;
1279 	target.failed = false;
1280 
1281 	/*
1282 	 * We bound precision to a reasonable range; the combination of this and
1283 	 * the knowledge that we're using "g" format without padding allows the
1284 	 * convert[] buffer to be reasonably small.
1285 	 */
1286 	if (precision < 1)
1287 		precision = 1;
1288 	else if (precision > 32)
1289 		precision = 32;
1290 
1291 	/*
1292 	 * The rest is just an inlined version of the fmtfloat() logic above,
1293 	 * simplified using the knowledge that no padding is wanted.
1294 	 */
1295 	if (isnan(value))
1296 	{
1297 		strcpy(convert, "NaN");
1298 		vallen = 3;
1299 	}
1300 	else
1301 	{
1302 		static const double dzero = 0.0;
1303 
1304 		if (value < 0.0 ||
1305 			(value == 0.0 &&
1306 			 memcmp(&value, &dzero, sizeof(double)) != 0))
1307 		{
1308 			signvalue = '-';
1309 			value = -value;
1310 		}
1311 
1312 		if (isinf(value))
1313 		{
1314 			strcpy(convert, "Infinity");
1315 			vallen = 8;
1316 		}
1317 		else
1318 		{
1319 			fmt[0] = '%';
1320 			fmt[1] = '.';
1321 			fmt[2] = '*';
1322 			fmt[3] = 'g';
1323 			fmt[4] = '\0';
1324 			vallen = sprintf(convert, fmt, precision, value);
1325 			if (vallen < 0)
1326 			{
1327 				target.failed = true;
1328 				goto fail;
1329 			}
1330 
1331 #ifdef WIN32
1332 			if (vallen >= 6 &&
1333 				convert[vallen - 5] == 'e' &&
1334 				convert[vallen - 3] == '0')
1335 			{
1336 				convert[vallen - 3] = convert[vallen - 2];
1337 				convert[vallen - 2] = convert[vallen - 1];
1338 				vallen--;
1339 			}
1340 #endif
1341 		}
1342 	}
1343 
1344 	if (signvalue)
1345 		dopr_outch(signvalue, &target);
1346 
1347 	dostr(convert, vallen, &target);
1348 
1349 fail:
1350 	*(target.bufptr) = '\0';
1351 	return target.failed ? -1 : (target.bufptr - target.bufstart
1352 								 + target.nchars);
1353 }
1354 
1355 
1356 static void
dostr(const char * str,int slen,PrintfTarget * target)1357 dostr(const char *str, int slen, PrintfTarget *target)
1358 {
1359 	/* fast path for common case of slen == 1 */
1360 	if (slen == 1)
1361 	{
1362 		dopr_outch(*str, target);
1363 		return;
1364 	}
1365 
1366 	while (slen > 0)
1367 	{
1368 		int			avail;
1369 
1370 		if (target->bufend != NULL)
1371 			avail = target->bufend - target->bufptr;
1372 		else
1373 			avail = slen;
1374 		if (avail <= 0)
1375 		{
1376 			/* buffer full, can we dump to stream? */
1377 			if (target->stream == NULL)
1378 			{
1379 				target->nchars += slen; /* no, lose the data */
1380 				return;
1381 			}
1382 			flushbuffer(target);
1383 			continue;
1384 		}
1385 		avail = Min(avail, slen);
1386 		memmove(target->bufptr, str, avail);
1387 		target->bufptr += avail;
1388 		str += avail;
1389 		slen -= avail;
1390 	}
1391 }
1392 
1393 static void
dopr_outch(int c,PrintfTarget * target)1394 dopr_outch(int c, PrintfTarget *target)
1395 {
1396 	if (target->bufend != NULL && target->bufptr >= target->bufend)
1397 	{
1398 		/* buffer full, can we dump to stream? */
1399 		if (target->stream == NULL)
1400 		{
1401 			target->nchars++;	/* no, lose the data */
1402 			return;
1403 		}
1404 		flushbuffer(target);
1405 	}
1406 	*(target->bufptr++) = c;
1407 }
1408 
1409 static void
dopr_outchmulti(int c,int slen,PrintfTarget * target)1410 dopr_outchmulti(int c, int slen, PrintfTarget *target)
1411 {
1412 	/* fast path for common case of slen == 1 */
1413 	if (slen == 1)
1414 	{
1415 		dopr_outch(c, target);
1416 		return;
1417 	}
1418 
1419 	while (slen > 0)
1420 	{
1421 		int			avail;
1422 
1423 		if (target->bufend != NULL)
1424 			avail = target->bufend - target->bufptr;
1425 		else
1426 			avail = slen;
1427 		if (avail <= 0)
1428 		{
1429 			/* buffer full, can we dump to stream? */
1430 			if (target->stream == NULL)
1431 			{
1432 				target->nchars += slen; /* no, lose the data */
1433 				return;
1434 			}
1435 			flushbuffer(target);
1436 			continue;
1437 		}
1438 		avail = Min(avail, slen);
1439 		memset(target->bufptr, c, avail);
1440 		target->bufptr += avail;
1441 		slen -= avail;
1442 	}
1443 }
1444 
1445 
1446 static int
adjust_sign(int is_negative,int forcesign,int * signvalue)1447 adjust_sign(int is_negative, int forcesign, int *signvalue)
1448 {
1449 	if (is_negative)
1450 	{
1451 		*signvalue = '-';
1452 		return true;
1453 	}
1454 	else if (forcesign)
1455 		*signvalue = '+';
1456 	return false;
1457 }
1458 
1459 
1460 static int
compute_padlen(int minlen,int vallen,int leftjust)1461 compute_padlen(int minlen, int vallen, int leftjust)
1462 {
1463 	int			padlen;
1464 
1465 	padlen = minlen - vallen;
1466 	if (padlen < 0)
1467 		padlen = 0;
1468 	if (leftjust)
1469 		padlen = -padlen;
1470 	return padlen;
1471 }
1472 
1473 
1474 static void
leading_pad(int zpad,int signvalue,int * padlen,PrintfTarget * target)1475 leading_pad(int zpad, int signvalue, int *padlen, PrintfTarget *target)
1476 {
1477 	int			maxpad;
1478 
1479 	if (*padlen > 0 && zpad)
1480 	{
1481 		if (signvalue)
1482 		{
1483 			dopr_outch(signvalue, target);
1484 			--(*padlen);
1485 			signvalue = 0;
1486 		}
1487 		if (*padlen > 0)
1488 		{
1489 			dopr_outchmulti(zpad, *padlen, target);
1490 			*padlen = 0;
1491 		}
1492 	}
1493 	maxpad = (signvalue != 0);
1494 	if (*padlen > maxpad)
1495 	{
1496 		dopr_outchmulti(' ', *padlen - maxpad, target);
1497 		*padlen = maxpad;
1498 	}
1499 	if (signvalue)
1500 	{
1501 		dopr_outch(signvalue, target);
1502 		if (*padlen > 0)
1503 			--(*padlen);
1504 		else if (*padlen < 0)
1505 			++(*padlen);
1506 	}
1507 }
1508 
1509 
1510 static void
trailing_pad(int padlen,PrintfTarget * target)1511 trailing_pad(int padlen, PrintfTarget *target)
1512 {
1513 	if (padlen < 0)
1514 		dopr_outchmulti(' ', -padlen, target);
1515 }
1516