|
|
@ -1,7 +1,8 @@ |
|
|
|
/* $OpenBSD: gcvt.c,v 1.11 2009/10/16 12:15:03 martynas Exp $ */ |
|
|
|
/* $OpenBSD: gcvt.c,v 1.12 2010/09/25 13:19:19 millert Exp $ */ |
|
|
|
|
|
|
|
/* |
|
|
|
* Copyright (c) 2002, 2003, 2006 Todd C. Miller <Todd.Miller@courtesan.com> |
|
|
|
* Copyright (c) 2002, 2003, 2006, 2010 |
|
|
|
* Todd C. Miller <Todd.Miller@courtesan.com> |
|
|
|
* |
|
|
|
* Permission to use, copy, modify, and distribute this software for any |
|
|
|
* purpose with or without fee is hereby granted, provided that the above |
|
|
@ -28,6 +29,8 @@ |
|
|
|
extern char *__dtoa(double, int, int, int *, int *, char **); |
|
|
|
extern void __freedtoa(char *); |
|
|
|
|
|
|
|
#define DEFPREC 6 |
|
|
|
|
|
|
|
char * |
|
|
|
gcvt(double value, int ndigit, char *buf) |
|
|
|
{ |
|
|
@ -36,9 +39,9 @@ gcvt(double value, int ndigit, char *buf) |
|
|
|
struct lconv *lconv; |
|
|
|
|
|
|
|
lconv = localeconv(); |
|
|
|
if (ndigit == 0) { |
|
|
|
buf[0] = '\0'; |
|
|
|
return (buf); |
|
|
|
if (ndigit <= 0) { |
|
|
|
/* Match printf(3) behavior. */ |
|
|
|
ndigit = ndigit ? DEFPREC : 1; |
|
|
|
} |
|
|
|
|
|
|
|
digits = __dtoa(value, 2, ndigit, &decpt, &sign, NULL); |
|
|
@ -47,9 +50,10 @@ gcvt(double value, int ndigit, char *buf) |
|
|
|
if (decpt == 9999) { |
|
|
|
/* |
|
|
|
* Infinity or NaN, convert to inf or nan with sign. |
|
|
|
* We assume the buffer is at least ndigit long. |
|
|
|
* We can't infer buffer size based on ndigit. |
|
|
|
* We have to assume it is at least 5 chars. |
|
|
|
*/ |
|
|
|
snprintf(buf, ndigit + 1, "%s%s", sign ? "-" : "", |
|
|
|
snprintf(buf, 5, "%s%s", sign ? "-" : "", |
|
|
|
*digits == 'I' ? "inf" : "nan"); |
|
|
|
__freedtoa(digits); |
|
|
|
return (buf); |
|
|
@ -59,7 +63,8 @@ gcvt(double value, int ndigit, char *buf) |
|
|
|
if (sign) |
|
|
|
*dst++ = '-'; |
|
|
|
|
|
|
|
if (decpt < 0 || decpt > ndigit) { |
|
|
|
/* Match printf(3) behavior for exponential vs. regular fomatting. */ |
|
|
|
if (decpt <= -4 || decpt > ndigit) { |
|
|
|
/* exponential format (e.g. 1.2345e+13) */ |
|
|
|
if (--decpt < 0) { |
|
|
|
sign = 1; |
|
|
@ -68,9 +73,12 @@ gcvt(double value, int ndigit, char *buf) |
|
|
|
sign = 0; |
|
|
|
src = digits; |
|
|
|
*dst++ = *src++; |
|
|
|
*dst++ = *lconv->decimal_point; |
|
|
|
while (*src != '\0') |
|
|
|
*dst++ = *src++; |
|
|
|
if (*src != '\0') { |
|
|
|
*dst++ = *lconv->decimal_point; |
|
|
|
do { |
|
|
|
*dst++ = *src++; |
|
|
|
} while (*src != '\0'); |
|
|
|
} |
|
|
|
*dst++ = 'e'; |
|
|
|
if (sign) |
|
|
|
*dst++ = '-'; |
|
|
@ -102,6 +110,10 @@ gcvt(double value, int ndigit, char *buf) |
|
|
|
if (src == digits) |
|
|
|
*dst++ = '0'; /* zero before decimal point */ |
|
|
|
*dst++ = *lconv->decimal_point; |
|
|
|
while (decpt < 0) { |
|
|
|
*dst++ = '0'; |
|
|
|
decpt++; |
|
|
|
} |
|
|
|
for (i = decpt; digits[i] != '\0'; i++) { |
|
|
|
*dst++ = digits[i]; |
|
|
|
} |
|
|
|