reactos/sdk/tools/mkisofs/schilytools/libschily/fconv.c

417 lines
8.6 KiB
C
Raw Normal View History

/* @(#)fconv.c 1.45 10/11/06 Copyright 1985, 1995-2010 J. Schilling */
/*
* Convert floating point numbers to strings for format.c
* Should rather use the MT-safe routines [efg]convert()
*
* Copyright (c) 1985, 1995-2010 J. Schilling
*/
/*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License, Version 1.0 only
* (the "License"). You may not use this file except in compliance
* with the License.
*
* See the file CDDL.Schily.txt in this distribution for details.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file CDDL.Schily.txt from this distribution.
*/
#include <schily/mconfig.h> /* <- may define NO_FLOATINGPOINT */
#ifndef NO_FLOATINGPOINT
#ifndef __DO_LONG_DOUBLE__
#include <schily/stdlib.h>
#include <schily/standard.h>
#include <schily/string.h>
#include <schily/schily.h>
#include <schily/math.h> /* The default place for isinf()/isnan() */
#include <schily/nlsdefs.h>
#if !defined(HAVE_STDLIB_H) || defined(HAVE_DTOA)
extern char *ecvt __PR((double, int, int *, int *));
extern char *fcvt __PR((double, int, int *, int *));
#endif
#if defined(HAVE_ISNAN) && defined(HAVE_ISINF)
/*
* *BSD alike libc
*/
#define FOUND_ISNAN
#define FOUND_ISINF
#define FOUND_ISXX
#endif
#if defined(HAVE_C99_ISNAN) && defined(HAVE_C99_ISINF)
#ifndef FOUND_ISXX
#define FOUND_ISXX
#endif
#define FOUND_C99_ISNAN
#define FOUND_C99_ISINF
#define FOUND_C99_ISXX
#endif
#if defined(HAVE_FP_H) && !defined(FOUND_ISXX)
/*
* WAS:
* #if defined(__osf__) || defined(_IBMR2) || defined(_AIX)
*/
/*
* Moved before HAVE_IEEEFP_H for True64 due to a hint
* from Bert De Knuydt <Bert.Deknuydt@esat.kuleuven.ac.be>
*
* True64 has both fp.h & ieeefp.h but the functions
* isnand() & finite() seem to be erreneously not implemented
* as a macro and the function lives in libm.
* Let's hope that we will not get problems with the new order.
*/
#include <fp.h>
#if !defined(isnan) && defined(IS_NAN)
#define isnan IS_NAN
#define FOUND_ISNAN
#endif
#if !defined(isinf) && defined(FINITE)
#define isinf !FINITE
#define FOUND_ISINF
#endif
#if defined(FOUND_ISNAN) && defined(FOUND_ISINF)
#define FOUND_ISXX
#endif
#endif
#if defined(HAVE_IEEEFP_H) && !defined(FOUND_ISXX) && \
!defined(FOUND_C99_ISXX)
/*
* SVR4
*/
#include <ieeefp.h>
#ifdef HAVE_ISNAND
#ifndef isnan
#define isnan isnand
#define FOUND_ISNAN
#endif
#endif
#ifdef HAVE_FINITE
#ifndef isinf
#define isinf !finite
#define FOUND_ISINF
#endif
#endif
#if defined(FOUND_ISNAN) && defined(FOUND_ISINF)
#define FOUND_ISXX
#endif
#endif
/*
* WAS:
* #if defined(__hpux) || defined(VMS) || defined(_SCO_DS) || defined(__QNX__)
*/
#ifdef __nneded__
#if defined(__hpux) || defined(__QNX__) || defined(__DJGPP__)
#ifndef FOUND_C99_ISXX
#undef isnan
#undef isinf
#endif
#endif
#endif /* __needed__ */
/*
* As we no longer check for defined(isnan)/defined(isinf), the next block
* should also handle the problems with DJGPP, HP-UX, QNX and VMS.
*/
#if !defined(FOUND_ISNAN) && !defined(HAVE_C99_ISNAN)
#undef isnan
#define isnan(val) (0)
#define NO_ISNAN
#endif
#if !defined(FOUND_ISINF) && !defined(HAVE_C99_ISINF)
#undef isinf
#define isinf(val) (0)
#define NO_ISINF
#endif
#if defined(NO_ISNAN) || defined(NO_ISINF)
#include <schily/float.h> /* For values.h */
#if (_IEEE - 0) > 0 /* We know that there is IEEE FP */
/*
* Note that older HP-UX versions have different #defines for MAXINT in
* values.h and sys/param.h
*/
#include <schily/utypes.h>
#include <schily/btorder.h>
#ifdef WORDS_BIGENDIAN
#define fpw_high(x) ((UInt32_t *)&x)[0]
#define fpw_low(x) ((UInt32_t *)&x)[1]
#else
#define fpw_high(x) ((UInt32_t *)&x)[1]
#define fpw_low(x) ((UInt32_t *)&x)[0]
#endif
#define FP_EXP 0x7FF00000
#define fp_exp(x) (fpw_high(x) & FP_EXP)
#define fp_exc(x) (fp_exp(x) == FP_EXP)
#ifdef NO_ISNAN
#undef isnan
#define isnan(val) (fp_exc(val) && \
(fpw_low(val) != 0 || (fpw_high(val) & 0xFFFFF) != 0))
#endif
#ifdef NO_ISINF
#undef isinf
#define isinf(val) (fp_exc(val) && \
fpw_low(val) == 0 && (fpw_high(val) & 0xFFFFF) == 0)
#endif
#endif /* We know that there is IEEE FP */
#endif /* defined(NO_ISNAN) || defined(NO_ISINF) */
#if !defined(HAVE_ECVT) || !defined(HAVE_FCVT) || !defined(HAVE_GCVT)
#ifdef NO_USER_XCVT
/*
* We cannot define our own ecvt()/fcvt()/gcvt() so we need to use
* local names instead.
*/
#ifndef HAVE_ECVT
#define ecvt js_ecvt
#endif
#ifndef HAVE_FCVT
#define fcvt js_fcvt
#endif
#ifndef HAVE_GCVT
#define gcvt js_gcvt
#endif
#endif
#include "cvt.c"
#endif
static char _js_nan[] = "(NaN)";
static char _js_inf[] = "(Infinity)";
static int _ferr __PR((char *, double));
#endif /* __DO_LONG_DOUBLE__ */
#ifdef __DO_LONG_DOUBLE__
#undef MDOUBLE
#define MDOUBLE long double
#else
#undef MDOUBLE
#define MDOUBLE double
#endif
#ifdef abs
#undef abs
#endif
#define abs(i) ((i) < 0 ? -(i) : (i))
EXPORT int
ftoes(s, val, fieldwidth, ndigits)
register char *s;
MDOUBLE val;
register int fieldwidth;
register int ndigits;
{
register char *b;
register char *rs;
register int len;
register int rdecpt;
int decpt;
int sign;
#ifndef __DO_LONG_DOUBLE__
if ((len = _ferr(s, val)) > 0)
return (len);
#endif
rs = s;
#ifdef V7_FLOATSTYLE
b = ecvt(val, ndigits, &decpt, &sign);
rdecpt = decpt;
#else
b = ecvt(val, ndigits+1, &decpt, &sign);
rdecpt = decpt-1;
#endif
#ifdef __DO_LONG_DOUBLE__
len = *b;
if (len < '0' || len > '9') { /* Inf/NaN */
strcpy(s, b);
return (strlen(b));
}
#endif
len = ndigits + 6; /* Punkt e +/- nnn */
if (sign)
len++;
if (fieldwidth > len)
while (fieldwidth-- > len)
*rs++ = ' ';
if (sign)
*rs++ = '-';
#ifndef V7_FLOATSTYLE
if (*b)
*rs++ = *b++;
#endif
#if defined(HAVE_LOCALECONV) && defined(USE_LOCALE)
*rs++ = *(localeconv()->decimal_point);
#else
*rs++ = '.';
#endif
while (*b && ndigits-- > 0)
*rs++ = *b++;
*rs++ = 'e';
*rs++ = rdecpt >= 0 ? '+' : '-';
rdecpt = abs(rdecpt);
#ifdef __DO_LONG_DOUBLE__
if (rdecpt >= 1000) { /* Max-Exp is > 4000 */
*rs++ = rdecpt / 1000 + '0';
rdecpt %= 1000;
}
#endif
#ifndef V7_FLOATSTYLE
if (rdecpt >= 100)
#endif
{
*rs++ = rdecpt / 100 + '0';
rdecpt %= 100;
}
*rs++ = rdecpt / 10 + '0';
*rs++ = rdecpt % 10 + '0';
*rs = '\0';
return (rs - s);
}
/*
* fcvt() from Cygwin32 is buggy.
*/
#if !defined(HAVE_FCVT) && defined(HAVE_ECVT)
#define USE_ECVT
#endif
EXPORT int
ftofs(s, val, fieldwidth, ndigits)
register char *s;
MDOUBLE val;
register int fieldwidth;
register int ndigits;
{
register char *b;
register char *rs;
register int len;
register int rdecpt;
int decpt;
int sign;
#ifndef __DO_LONG_DOUBLE__
if ((len = _ferr(s, val)) > 0)
return (len);
#endif
rs = s;
#ifdef USE_ECVT
/*
* Needed on systems with broken fcvt() implementation
* (e.g. Cygwin32)
*/
b = ecvt(val, ndigits, &decpt, &sign);
/*
* The next call is needed to force higher precision.
*/
if (decpt > 0)
b = ecvt(val, ndigits+decpt, &decpt, &sign);
#else
b = fcvt(val, ndigits, &decpt, &sign);
#endif
#ifdef __DO_LONG_DOUBLE__
len = *b;
if (len < '0' || len > '9') { /* Inf/NaN */
strcpy(s, b);
return (strlen(b));
}
#endif
rdecpt = decpt;
len = rdecpt + ndigits + 1;
if (rdecpt < 0)
len -= rdecpt;
if (sign)
len++;
if (fieldwidth > len)
while (fieldwidth-- > len)
*rs++ = ' ';
if (sign)
*rs++ = '-';
if (rdecpt > 0) {
len = rdecpt;
while (*b && len-- > 0)
*rs++ = *b++;
#ifdef USE_ECVT
while (len-- > 0)
*rs++ = '0';
#endif
}
#ifndef V7_FLOATSTYLE
else {
*rs++ = '0';
}
#endif
#if defined(HAVE_LOCALECONV) && defined(USE_LOCALE)
*rs++ = *(localeconv()->decimal_point);
#else
*rs++ = '.';
#endif
if (rdecpt < 0) {
len = rdecpt;
while (len++ < 0 && ndigits-- > 0)
*rs++ = '0';
}
while (*b && ndigits-- > 0)
*rs++ = *b++;
#ifdef USE_ECVT
while (ndigits-- > 0)
*rs++ = '0';
#endif
*rs = '\0';
return (rs - s);
}
#ifndef __DO_LONG_DOUBLE__
#ifdef HAVE_LONGDOUBLE
#ifdef HAVE__LDECVT
#define qecvt(ld, n, dp, sp) _ldecvt(*(long_double *)&ld, n, dp, sp)
#endif
#ifdef HAVE__LDFCVT
#define qfcvt(ld, n, dp, sp) _ldfcvt(*(long_double *)&ld, n, dp, sp)
#endif
#if (defined(HAVE_QECVT) || defined(HAVE__LDECVT)) && \
(defined(HAVE_QFCVT) || defined(HAVE__LDFCVT))
#define __DO_LONG_DOUBLE__
#define ftoes qftoes
#define ftofs qftofs
#define ecvt qecvt
#define fcvt qfcvt
#include "fconv.c"
#undef __DO_LONG_DOUBLE__
#endif
#endif /* HAVE_LONGDOUBLE */
LOCAL int
_ferr(s, val)
char *s;
double val;
{
if (isnan(val)) {
strcpy(s, _js_nan);
return (sizeof (_js_nan) - 1);
}
/*
* Check first for NaN because finite() will return 1 on Nan too.
*/
if (isinf(val)) {
strcpy(s, _js_inf);
return (sizeof (_js_inf) - 1);
}
return (0);
}
#endif /* __DO_LONG_DOUBLE__ */
#endif /* NO_FLOATINGPOINT */