mirror of
https://github.com/python/cpython.git
synced 2025-08-01 07:33:08 +00:00
Issue #7117 (backport py3k float repr) continued:
Backport pystrtod.c from py3k.
This commit is contained in:
parent
b05d3be2f1
commit
975d7576ca
1 changed files with 698 additions and 92 deletions
|
@ -3,6 +3,60 @@
|
||||||
#include <Python.h>
|
#include <Python.h>
|
||||||
#include <locale.h>
|
#include <locale.h>
|
||||||
|
|
||||||
|
/* Case-insensitive string match used for nan and inf detection; t should be
|
||||||
|
lower-case. Returns 1 for a successful match, 0 otherwise. */
|
||||||
|
|
||||||
|
static int
|
||||||
|
case_insensitive_match(const char *s, const char *t)
|
||||||
|
{
|
||||||
|
while(*t && Py_TOLOWER(*s) == *t) {
|
||||||
|
s++;
|
||||||
|
t++;
|
||||||
|
}
|
||||||
|
return *t ? 0 : 1;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* _Py_parse_inf_or_nan: Attempt to parse a string of the form "nan", "inf" or
|
||||||
|
"infinity", with an optional leading sign of "+" or "-". On success,
|
||||||
|
return the NaN or Infinity as a double and set *endptr to point just beyond
|
||||||
|
the successfully parsed portion of the string. On failure, return -1.0 and
|
||||||
|
set *endptr to point to the start of the string. */
|
||||||
|
|
||||||
|
double
|
||||||
|
_Py_parse_inf_or_nan(const char *p, char **endptr)
|
||||||
|
{
|
||||||
|
double retval;
|
||||||
|
const char *s;
|
||||||
|
int negate = 0;
|
||||||
|
|
||||||
|
s = p;
|
||||||
|
if (*s == '-') {
|
||||||
|
negate = 1;
|
||||||
|
s++;
|
||||||
|
}
|
||||||
|
else if (*s == '+') {
|
||||||
|
s++;
|
||||||
|
}
|
||||||
|
if (case_insensitive_match(s, "inf")) {
|
||||||
|
s += 3;
|
||||||
|
if (case_insensitive_match(s, "inity"))
|
||||||
|
s += 5;
|
||||||
|
retval = negate ? -Py_HUGE_VAL : Py_HUGE_VAL;
|
||||||
|
}
|
||||||
|
#ifdef Py_NAN
|
||||||
|
else if (case_insensitive_match(s, "nan")) {
|
||||||
|
s += 3;
|
||||||
|
retval = negate ? -Py_NAN : Py_NAN;
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
else {
|
||||||
|
s = p;
|
||||||
|
retval = -1.0;
|
||||||
|
}
|
||||||
|
*endptr = (char *)s;
|
||||||
|
return retval;
|
||||||
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* PyOS_ascii_strtod:
|
* PyOS_ascii_strtod:
|
||||||
* @nptr: the string to convert to a numeric value.
|
* @nptr: the string to convert to a numeric value.
|
||||||
|
@ -32,6 +86,33 @@
|
||||||
* Return value: the #gdouble value.
|
* Return value: the #gdouble value.
|
||||||
**/
|
**/
|
||||||
|
|
||||||
|
#ifndef PY_NO_SHORT_FLOAT_REPR
|
||||||
|
|
||||||
|
double
|
||||||
|
_PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
|
{
|
||||||
|
double result;
|
||||||
|
_Py_SET_53BIT_PRECISION_HEADER;
|
||||||
|
|
||||||
|
assert(nptr != NULL);
|
||||||
|
/* Set errno to zero, so that we can distinguish zero results
|
||||||
|
and underflows */
|
||||||
|
errno = 0;
|
||||||
|
|
||||||
|
_Py_SET_53BIT_PRECISION_START;
|
||||||
|
result = _Py_dg_strtod(nptr, endptr);
|
||||||
|
_Py_SET_53BIT_PRECISION_END;
|
||||||
|
|
||||||
|
if (*endptr == nptr)
|
||||||
|
/* string might represent and inf or nan */
|
||||||
|
result = _Py_parse_inf_or_nan(nptr, endptr);
|
||||||
|
|
||||||
|
return result;
|
||||||
|
|
||||||
|
}
|
||||||
|
|
||||||
|
#else
|
||||||
|
|
||||||
/*
|
/*
|
||||||
Use system strtod; since strtod is locale aware, we may
|
Use system strtod; since strtod is locale aware, we may
|
||||||
have to first fix the decimal separator.
|
have to first fix the decimal separator.
|
||||||
|
@ -40,21 +121,8 @@
|
||||||
correctly rounded results.
|
correctly rounded results.
|
||||||
*/
|
*/
|
||||||
|
|
||||||
/* Case-insensitive string match used for nan and inf detection; t should be
|
|
||||||
lower-case. Returns 1 for a successful match, 0 otherwise. */
|
|
||||||
|
|
||||||
static int
|
|
||||||
case_insensitive_match(const char *s, const char *t)
|
|
||||||
{
|
|
||||||
while(*t && Py_TOLOWER(*s) == *t) {
|
|
||||||
s++;
|
|
||||||
t++;
|
|
||||||
}
|
|
||||||
return *t ? 0 : 1;
|
|
||||||
}
|
|
||||||
|
|
||||||
double
|
double
|
||||||
PyOS_ascii_strtod(const char *nptr, char **endptr)
|
_PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
{
|
{
|
||||||
char *fail_pos;
|
char *fail_pos;
|
||||||
double val = -1.0;
|
double val = -1.0;
|
||||||
|
@ -78,19 +146,19 @@ PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
|
|
||||||
decimal_point_pos = NULL;
|
decimal_point_pos = NULL;
|
||||||
|
|
||||||
|
/* Parse infinities and nans */
|
||||||
|
val = _Py_parse_inf_or_nan(nptr, endptr);
|
||||||
|
if (*endptr != nptr)
|
||||||
|
return val;
|
||||||
|
|
||||||
/* Set errno to zero, so that we can distinguish zero results
|
/* Set errno to zero, so that we can distinguish zero results
|
||||||
and underflows */
|
and underflows */
|
||||||
errno = 0;
|
errno = 0;
|
||||||
|
|
||||||
/* We process any leading whitespace and the optional sign manually,
|
/* We process the optional sign manually, then pass the remainder to
|
||||||
then pass the remainder to the system strtod. This ensures that
|
the system strtod. This ensures that the result of an underflow
|
||||||
the result of an underflow has the correct sign. (bug #1725) */
|
has the correct sign. (bug #1725) */
|
||||||
|
|
||||||
p = nptr;
|
p = nptr;
|
||||||
/* Skip leading space */
|
|
||||||
while (Py_ISSPACE(*p))
|
|
||||||
p++;
|
|
||||||
|
|
||||||
/* Process leading sign, if present */
|
/* Process leading sign, if present */
|
||||||
if (*p == '-') {
|
if (*p == '-') {
|
||||||
negate = 1;
|
negate = 1;
|
||||||
|
@ -100,31 +168,6 @@ PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
p++;
|
p++;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* Parse infinities and nans */
|
|
||||||
if (*p == 'i' || *p == 'I') {
|
|
||||||
if (case_insensitive_match(p+1, "nf")) {
|
|
||||||
val = Py_HUGE_VAL;
|
|
||||||
if (case_insensitive_match(p+3, "inity"))
|
|
||||||
fail_pos = (char *)p+8;
|
|
||||||
else
|
|
||||||
fail_pos = (char *)p+3;
|
|
||||||
goto got_val;
|
|
||||||
}
|
|
||||||
else
|
|
||||||
goto invalid_string;
|
|
||||||
}
|
|
||||||
#ifdef Py_NAN
|
|
||||||
if (*p == 'n' || *p == 'N') {
|
|
||||||
if (case_insensitive_match(p+1, "an")) {
|
|
||||||
val = Py_NAN;
|
|
||||||
fail_pos = (char *)p+3;
|
|
||||||
goto got_val;
|
|
||||||
}
|
|
||||||
else
|
|
||||||
goto invalid_string;
|
|
||||||
}
|
|
||||||
#endif
|
|
||||||
|
|
||||||
/* Some platform strtods accept hex floats; Python shouldn't (at the
|
/* Some platform strtods accept hex floats; Python shouldn't (at the
|
||||||
moment), so we check explicitly for strings starting with '0x'. */
|
moment), so we check explicitly for strings starting with '0x'. */
|
||||||
if (*p == '0' && (*(p+1) == 'x' || *(p+1) == 'X'))
|
if (*p == '0' && (*(p+1) == 'x' || *(p+1) == 'X'))
|
||||||
|
@ -175,7 +218,6 @@ PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
copy = (char *)PyMem_MALLOC(end - digits_pos +
|
copy = (char *)PyMem_MALLOC(end - digits_pos +
|
||||||
1 + decimal_point_len);
|
1 + decimal_point_len);
|
||||||
if (copy == NULL) {
|
if (copy == NULL) {
|
||||||
if (endptr)
|
|
||||||
*endptr = (char *)nptr;
|
*endptr = (char *)nptr;
|
||||||
errno = ENOMEM;
|
errno = ENOMEM;
|
||||||
return val;
|
return val;
|
||||||
|
@ -214,28 +256,108 @@ PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
if (fail_pos == digits_pos)
|
if (fail_pos == digits_pos)
|
||||||
goto invalid_string;
|
goto invalid_string;
|
||||||
|
|
||||||
got_val:
|
|
||||||
if (negate && fail_pos != nptr)
|
if (negate && fail_pos != nptr)
|
||||||
val = -val;
|
val = -val;
|
||||||
|
|
||||||
if (endptr)
|
|
||||||
*endptr = fail_pos;
|
*endptr = fail_pos;
|
||||||
|
|
||||||
return val;
|
return val;
|
||||||
|
|
||||||
invalid_string:
|
invalid_string:
|
||||||
if (endptr)
|
|
||||||
*endptr = (char*)nptr;
|
*endptr = (char*)nptr;
|
||||||
errno = EINVAL;
|
errno = EINVAL;
|
||||||
return -1.0;
|
return -1.0;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#endif
|
||||||
|
|
||||||
|
double
|
||||||
|
PyOS_ascii_strtod(const char *nptr, char **endptr)
|
||||||
|
{
|
||||||
|
char *fail_pos;
|
||||||
|
const char *p;
|
||||||
|
double x;
|
||||||
|
|
||||||
|
/* _PyOS_ascii_strtod already does everything that we want,
|
||||||
|
except that it doesn't parse leading whitespace */
|
||||||
|
p = nptr;
|
||||||
|
while (Py_ISSPACE(*p))
|
||||||
|
p++;
|
||||||
|
x = _PyOS_ascii_strtod(p, &fail_pos);
|
||||||
|
if (fail_pos == p)
|
||||||
|
fail_pos = (char *)nptr;
|
||||||
|
if (endptr)
|
||||||
|
*endptr = (char *)fail_pos;
|
||||||
|
return x;
|
||||||
|
}
|
||||||
|
|
||||||
double
|
double
|
||||||
PyOS_ascii_atof(const char *nptr)
|
PyOS_ascii_atof(const char *nptr)
|
||||||
{
|
{
|
||||||
return PyOS_ascii_strtod(nptr, NULL);
|
return PyOS_ascii_strtod(nptr, NULL);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/* PyOS_string_to_double is the recommended replacement for the
|
||||||
|
PyOS_ascii_strtod and PyOS_ascii_atof functions. It converts a
|
||||||
|
null-terminated byte string s (interpreted as a string of ASCII characters)
|
||||||
|
to a float. The string should not have leading or trailing whitespace (in
|
||||||
|
contrast, PyOS_ascii_strtod allows leading whitespace but not trailing
|
||||||
|
whitespace). The conversion is independent of the current locale.
|
||||||
|
|
||||||
|
If endptr is NULL, try to convert the whole string. Raise ValueError and
|
||||||
|
return -1.0 if the string is not a valid representation of a floating-point
|
||||||
|
number.
|
||||||
|
|
||||||
|
If endptr is non-NULL, try to convert as much of the string as possible.
|
||||||
|
If no initial segment of the string is the valid representation of a
|
||||||
|
floating-point number then *endptr is set to point to the beginning of the
|
||||||
|
string, -1.0 is returned and again ValueError is raised.
|
||||||
|
|
||||||
|
On overflow (e.g., when trying to convert '1e500' on an IEEE 754 machine),
|
||||||
|
if overflow_exception is NULL then +-Py_HUGE_VAL is returned, and no Python
|
||||||
|
exception is raised. Otherwise, overflow_exception should point to a
|
||||||
|
a Python exception, this exception will be raised, -1.0 will be returned,
|
||||||
|
and *endptr will point just past the end of the converted value.
|
||||||
|
|
||||||
|
If any other failure occurs (for example lack of memory), -1.0 is returned
|
||||||
|
and the appropriate Python exception will have been set.
|
||||||
|
*/
|
||||||
|
|
||||||
|
double
|
||||||
|
PyOS_string_to_double(const char *s,
|
||||||
|
char **endptr,
|
||||||
|
PyObject *overflow_exception)
|
||||||
|
{
|
||||||
|
double x, result=-1.0;
|
||||||
|
char *fail_pos;
|
||||||
|
|
||||||
|
errno = 0;
|
||||||
|
PyFPE_START_PROTECT("PyOS_string_to_double", return -1.0)
|
||||||
|
x = PyOS_ascii_strtod(s, &fail_pos);
|
||||||
|
PyFPE_END_PROTECT(x)
|
||||||
|
|
||||||
|
if (errno == ENOMEM) {
|
||||||
|
PyErr_NoMemory();
|
||||||
|
fail_pos = (char *)s;
|
||||||
|
}
|
||||||
|
else if (!endptr && (fail_pos == s || *fail_pos != '\0'))
|
||||||
|
PyErr_Format(PyExc_ValueError,
|
||||||
|
"could not convert string to float: "
|
||||||
|
"%.200s", s);
|
||||||
|
else if (fail_pos == s)
|
||||||
|
PyErr_Format(PyExc_ValueError,
|
||||||
|
"could not convert string to float: "
|
||||||
|
"%.200s", s);
|
||||||
|
else if (errno == ERANGE && fabs(x) >= 1.0 && overflow_exception)
|
||||||
|
PyErr_Format(overflow_exception,
|
||||||
|
"value too large to convert to float: "
|
||||||
|
"%.200s", s);
|
||||||
|
else
|
||||||
|
result = x;
|
||||||
|
|
||||||
|
if (endptr != NULL)
|
||||||
|
*endptr = fail_pos;
|
||||||
|
return result;
|
||||||
|
}
|
||||||
|
|
||||||
/* Given a string that may have a decimal point in the current
|
/* Given a string that may have a decimal point in the current
|
||||||
locale, change it back to a dot. Since the string cannot get
|
locale, change it back to a dot. Since the string cannot get
|
||||||
|
@ -505,7 +627,7 @@ ensure_decimal_point(char* buffer, size_t buf_size, int precision)
|
||||||
#define FLOAT_FORMATBUFLEN 120
|
#define FLOAT_FORMATBUFLEN 120
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* _PyOS_ascii_formatd:
|
* PyOS_ascii_formatd:
|
||||||
* @buffer: A buffer to place the resulting string in
|
* @buffer: A buffer to place the resulting string in
|
||||||
* @buf_size: The length of the buffer.
|
* @buf_size: The length of the buffer.
|
||||||
* @format: The printf()-style format to use for the
|
* @format: The printf()-style format to use for the
|
||||||
|
@ -523,12 +645,12 @@ ensure_decimal_point(char* buffer, size_t buf_size, int precision)
|
||||||
* Return value: The pointer to the buffer with the converted string.
|
* Return value: The pointer to the buffer with the converted string.
|
||||||
* On failure returns NULL but does not set any Python exception.
|
* On failure returns NULL but does not set any Python exception.
|
||||||
**/
|
**/
|
||||||
/* DEPRECATED, will be deleted in 2.8 and 3.2 */
|
char *
|
||||||
PyAPI_FUNC(char *)
|
_PyOS_ascii_formatd(char *buffer,
|
||||||
PyOS_ascii_formatd(char *buffer,
|
|
||||||
size_t buf_size,
|
size_t buf_size,
|
||||||
const char *format,
|
const char *format,
|
||||||
double d)
|
double d,
|
||||||
|
int precision)
|
||||||
{
|
{
|
||||||
char format_char;
|
char format_char;
|
||||||
size_t format_len = strlen(format);
|
size_t format_len = strlen(format);
|
||||||
|
@ -537,11 +659,6 @@ PyOS_ascii_formatd(char *buffer,
|
||||||
also with at least one character past the decimal. */
|
also with at least one character past the decimal. */
|
||||||
char tmp_format[FLOAT_FORMATBUFLEN];
|
char tmp_format[FLOAT_FORMATBUFLEN];
|
||||||
|
|
||||||
if (PyErr_WarnEx(PyExc_DeprecationWarning,
|
|
||||||
"PyOS_ascii_formatd is deprecated, "
|
|
||||||
"use PyOS_double_to_string instead", 1) < 0)
|
|
||||||
return NULL;
|
|
||||||
|
|
||||||
/* The last character in the format string must be the format char */
|
/* The last character in the format string must be the format char */
|
||||||
format_char = format[format_len - 1];
|
format_char = format[format_len - 1];
|
||||||
|
|
||||||
|
@ -603,24 +720,24 @@ PyOS_ascii_formatd(char *buffer,
|
||||||
extra character would produce more significant digits that we
|
extra character would produce more significant digits that we
|
||||||
really want. */
|
really want. */
|
||||||
if (format_char == 'Z')
|
if (format_char == 'Z')
|
||||||
buffer = ensure_decimal_point(buffer, buf_size, -1);
|
buffer = ensure_decimal_point(buffer, buf_size, precision);
|
||||||
|
|
||||||
return buffer;
|
return buffer;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* Precisions used by repr() and str(), respectively.
|
char *
|
||||||
|
PyOS_ascii_formatd(char *buffer,
|
||||||
|
size_t buf_size,
|
||||||
|
const char *format,
|
||||||
|
double d)
|
||||||
|
{
|
||||||
|
if (PyErr_WarnEx(PyExc_DeprecationWarning,
|
||||||
|
"PyOS_ascii_formatd is deprecated, "
|
||||||
|
"use PyOS_double_to_string instead", 1) < 0)
|
||||||
|
return NULL;
|
||||||
|
|
||||||
The repr() precision (17 significant decimal digits) is the minimal number
|
return _PyOS_ascii_formatd(buffer, buf_size, format, d, -1);
|
||||||
that is guaranteed to have enough precision so that if the number is read
|
}
|
||||||
back in the exact same binary value is recreated. This is true for IEEE
|
|
||||||
floating point by design, and also happens to work for all other modern
|
|
||||||
hardware.
|
|
||||||
|
|
||||||
The str() precision (12 significant decimal digits) is chosen so that in
|
|
||||||
most cases, the rounding noise created by various operations is suppressed,
|
|
||||||
while giving plenty of precision for practical use.
|
|
||||||
|
|
||||||
*/
|
|
||||||
|
|
||||||
PyAPI_FUNC(void)
|
PyAPI_FUNC(void)
|
||||||
_PyOS_double_to_string(char *buf, size_t buf_len, double val,
|
_PyOS_double_to_string(char *buf, size_t buf_len, double val,
|
||||||
|
@ -738,31 +855,520 @@ _PyOS_double_to_string(char *buf, size_t buf_len, double val,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
|
#ifdef PY_NO_SHORT_FLOAT_REPR
|
||||||
|
|
||||||
|
/* The fallback code to use if _Py_dg_dtoa is not available. */
|
||||||
|
|
||||||
PyAPI_FUNC(char *) PyOS_double_to_string(double val,
|
PyAPI_FUNC(char *) PyOS_double_to_string(double val,
|
||||||
char format_code,
|
char format_code,
|
||||||
int precision,
|
int precision,
|
||||||
int flags,
|
int flags,
|
||||||
int *ptype)
|
int *type)
|
||||||
{
|
{
|
||||||
char buf[128];
|
char format[32];
|
||||||
Py_ssize_t len;
|
Py_ssize_t bufsize;
|
||||||
char *result;
|
char *buf;
|
||||||
|
int t, exp;
|
||||||
|
int upper = 0;
|
||||||
|
|
||||||
_PyOS_double_to_string(buf, sizeof(buf), val, format_code, precision,
|
/* Validate format_code, and map upper and lower case */
|
||||||
flags, ptype);
|
switch (format_code) {
|
||||||
len = strlen(buf);
|
case 'e': /* exponent */
|
||||||
if (len == 0) {
|
case 'f': /* fixed */
|
||||||
|
case 'g': /* general */
|
||||||
|
break;
|
||||||
|
case 'E':
|
||||||
|
upper = 1;
|
||||||
|
format_code = 'e';
|
||||||
|
break;
|
||||||
|
case 'F':
|
||||||
|
upper = 1;
|
||||||
|
format_code = 'f';
|
||||||
|
break;
|
||||||
|
case 'G':
|
||||||
|
upper = 1;
|
||||||
|
format_code = 'g';
|
||||||
|
break;
|
||||||
|
case 'r': /* repr format */
|
||||||
|
/* Supplied precision is unused, must be 0. */
|
||||||
|
if (precision != 0) {
|
||||||
|
PyErr_BadInternalCall();
|
||||||
|
return NULL;
|
||||||
|
}
|
||||||
|
/* The repr() precision (17 significant decimal digits) is the
|
||||||
|
minimal number that is guaranteed to have enough precision
|
||||||
|
so that if the number is read back in the exact same binary
|
||||||
|
value is recreated. This is true for IEEE floating point
|
||||||
|
by design, and also happens to work for all other modern
|
||||||
|
hardware. */
|
||||||
|
precision = 17;
|
||||||
|
format_code = 'g';
|
||||||
|
break;
|
||||||
|
default:
|
||||||
PyErr_BadInternalCall();
|
PyErr_BadInternalCall();
|
||||||
return NULL;
|
return NULL;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* Add 1 for the trailing 0 byte. */
|
/* Here's a quick-and-dirty calculation to figure out how big a buffer
|
||||||
result = PyMem_Malloc(len + 1);
|
we need. In general, for a finite float we need:
|
||||||
if (result == NULL) {
|
|
||||||
|
1 byte for each digit of the decimal significand, and
|
||||||
|
|
||||||
|
1 for a possible sign
|
||||||
|
1 for a possible decimal point
|
||||||
|
2 for a possible [eE][+-]
|
||||||
|
1 for each digit of the exponent; if we allow 19 digits
|
||||||
|
total then we're safe up to exponents of 2**63.
|
||||||
|
1 for the trailing nul byte
|
||||||
|
|
||||||
|
This gives a total of 24 + the number of digits in the significand,
|
||||||
|
and the number of digits in the significand is:
|
||||||
|
|
||||||
|
for 'g' format: at most precision, except possibly
|
||||||
|
when precision == 0, when it's 1.
|
||||||
|
for 'e' format: precision+1
|
||||||
|
for 'f' format: precision digits after the point, at least 1
|
||||||
|
before. To figure out how many digits appear before the point
|
||||||
|
we have to examine the size of the number. If fabs(val) < 1.0
|
||||||
|
then there will be only one digit before the point. If
|
||||||
|
fabs(val) >= 1.0, then there are at most
|
||||||
|
|
||||||
|
1+floor(log10(ceiling(fabs(val))))
|
||||||
|
|
||||||
|
digits before the point (where the 'ceiling' allows for the
|
||||||
|
possibility that the rounding rounds the integer part of val
|
||||||
|
up). A safe upper bound for the above quantity is
|
||||||
|
1+floor(exp/3), where exp is the unique integer such that 0.5
|
||||||
|
<= fabs(val)/2**exp < 1.0. This exp can be obtained from
|
||||||
|
frexp.
|
||||||
|
|
||||||
|
So we allow room for precision+1 digits for all formats, plus an
|
||||||
|
extra floor(exp/3) digits for 'f' format.
|
||||||
|
|
||||||
|
*/
|
||||||
|
|
||||||
|
if (Py_IS_NAN(val) || Py_IS_INFINITY(val))
|
||||||
|
/* 3 for 'inf'/'nan', 1 for sign, 1 for '\0' */
|
||||||
|
bufsize = 5;
|
||||||
|
else {
|
||||||
|
bufsize = 25 + precision;
|
||||||
|
if (format_code == 'f' && fabs(val) >= 1.0) {
|
||||||
|
frexp(val, &exp);
|
||||||
|
bufsize += exp/3;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
buf = PyMem_Malloc(bufsize);
|
||||||
|
if (buf == NULL) {
|
||||||
PyErr_NoMemory();
|
PyErr_NoMemory();
|
||||||
return NULL;
|
return NULL;
|
||||||
}
|
}
|
||||||
strcpy(result, buf);
|
|
||||||
|
|
||||||
return result;
|
/* Handle nan and inf. */
|
||||||
|
if (Py_IS_NAN(val)) {
|
||||||
|
strcpy(buf, "nan");
|
||||||
|
t = Py_DTST_NAN;
|
||||||
|
} else if (Py_IS_INFINITY(val)) {
|
||||||
|
if (copysign(1., val) == 1.)
|
||||||
|
strcpy(buf, "inf");
|
||||||
|
else
|
||||||
|
strcpy(buf, "-inf");
|
||||||
|
t = Py_DTST_INFINITE;
|
||||||
|
} else {
|
||||||
|
t = Py_DTST_FINITE;
|
||||||
|
if (flags & Py_DTSF_ADD_DOT_0)
|
||||||
|
format_code = 'Z';
|
||||||
|
|
||||||
|
PyOS_snprintf(format, sizeof(format), "%%%s.%i%c",
|
||||||
|
(flags & Py_DTSF_ALT ? "#" : ""), precision,
|
||||||
|
format_code);
|
||||||
|
_PyOS_ascii_formatd(buf, bufsize, format, val, precision);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/* Add sign when requested. It's convenient (esp. when formatting
|
||||||
|
complex numbers) to include a sign even for inf and nan. */
|
||||||
|
if (flags & Py_DTSF_SIGN && buf[0] != '-') {
|
||||||
|
size_t len = strlen(buf);
|
||||||
|
/* the bufsize calculations above should ensure that we've got
|
||||||
|
space to add a sign */
|
||||||
|
assert((size_t)bufsize >= len+2);
|
||||||
|
memmove(buf+1, buf, len+1);
|
||||||
|
buf[0] = '+';
|
||||||
|
}
|
||||||
|
if (upper) {
|
||||||
|
/* Convert to upper case. */
|
||||||
|
char *p1;
|
||||||
|
for (p1 = buf; *p1; p1++)
|
||||||
|
*p1 = Py_TOUPPER(*p1);
|
||||||
|
}
|
||||||
|
|
||||||
|
if (type)
|
||||||
|
*type = t;
|
||||||
|
return buf;
|
||||||
|
}
|
||||||
|
|
||||||
|
#else
|
||||||
|
|
||||||
|
/* _Py_dg_dtoa is available. */
|
||||||
|
|
||||||
|
/* I'm using a lookup table here so that I don't have to invent a non-locale
|
||||||
|
specific way to convert to uppercase */
|
||||||
|
#define OFS_INF 0
|
||||||
|
#define OFS_NAN 1
|
||||||
|
#define OFS_E 2
|
||||||
|
|
||||||
|
/* The lengths of these are known to the code below, so don't change them */
|
||||||
|
static char *lc_float_strings[] = {
|
||||||
|
"inf",
|
||||||
|
"nan",
|
||||||
|
"e",
|
||||||
|
};
|
||||||
|
static char *uc_float_strings[] = {
|
||||||
|
"INF",
|
||||||
|
"NAN",
|
||||||
|
"E",
|
||||||
|
};
|
||||||
|
|
||||||
|
|
||||||
|
/* Convert a double d to a string, and return a PyMem_Malloc'd block of
|
||||||
|
memory contain the resulting string.
|
||||||
|
|
||||||
|
Arguments:
|
||||||
|
d is the double to be converted
|
||||||
|
format_code is one of 'e', 'f', 'g', 'r'. 'e', 'f' and 'g'
|
||||||
|
correspond to '%e', '%f' and '%g'; 'r' corresponds to repr.
|
||||||
|
mode is one of '0', '2' or '3', and is completely determined by
|
||||||
|
format_code: 'e' and 'g' use mode 2; 'f' mode 3, 'r' mode 0.
|
||||||
|
precision is the desired precision
|
||||||
|
always_add_sign is nonzero if a '+' sign should be included for positive
|
||||||
|
numbers
|
||||||
|
add_dot_0_if_integer is nonzero if integers in non-exponential form
|
||||||
|
should have ".0" added. Only applies to format codes 'r' and 'g'.
|
||||||
|
use_alt_formatting is nonzero if alternative formatting should be
|
||||||
|
used. Only applies to format codes 'e', 'f' and 'g'. For code 'g',
|
||||||
|
at most one of use_alt_formatting and add_dot_0_if_integer should
|
||||||
|
be nonzero.
|
||||||
|
type, if non-NULL, will be set to one of these constants to identify
|
||||||
|
the type of the 'd' argument:
|
||||||
|
Py_DTST_FINITE
|
||||||
|
Py_DTST_INFINITE
|
||||||
|
Py_DTST_NAN
|
||||||
|
|
||||||
|
Returns a PyMem_Malloc'd block of memory containing the resulting string,
|
||||||
|
or NULL on error. If NULL is returned, the Python error has been set.
|
||||||
|
*/
|
||||||
|
|
||||||
|
static char *
|
||||||
|
format_float_short(double d, char format_code,
|
||||||
|
int mode, Py_ssize_t precision,
|
||||||
|
int always_add_sign, int add_dot_0_if_integer,
|
||||||
|
int use_alt_formatting, char **float_strings, int *type)
|
||||||
|
{
|
||||||
|
char *buf = NULL;
|
||||||
|
char *p = NULL;
|
||||||
|
Py_ssize_t bufsize = 0;
|
||||||
|
char *digits, *digits_end;
|
||||||
|
int decpt_as_int, sign, exp_len, exp = 0, use_exp = 0;
|
||||||
|
Py_ssize_t decpt, digits_len, vdigits_start, vdigits_end;
|
||||||
|
_Py_SET_53BIT_PRECISION_HEADER;
|
||||||
|
|
||||||
|
/* _Py_dg_dtoa returns a digit string (no decimal point or exponent).
|
||||||
|
Must be matched by a call to _Py_dg_freedtoa. */
|
||||||
|
_Py_SET_53BIT_PRECISION_START;
|
||||||
|
digits = _Py_dg_dtoa(d, mode, precision, &decpt_as_int, &sign,
|
||||||
|
&digits_end);
|
||||||
|
_Py_SET_53BIT_PRECISION_END;
|
||||||
|
|
||||||
|
decpt = (Py_ssize_t)decpt_as_int;
|
||||||
|
if (digits == NULL) {
|
||||||
|
/* The only failure mode is no memory. */
|
||||||
|
PyErr_NoMemory();
|
||||||
|
goto exit;
|
||||||
|
}
|
||||||
|
assert(digits_end != NULL && digits_end >= digits);
|
||||||
|
digits_len = digits_end - digits;
|
||||||
|
|
||||||
|
if (digits_len && !Py_ISDIGIT(digits[0])) {
|
||||||
|
/* Infinities and nans here; adapt Gay's output,
|
||||||
|
so convert Infinity to inf and NaN to nan, and
|
||||||
|
ignore sign of nan. Then return. */
|
||||||
|
|
||||||
|
/* ignore the actual sign of a nan */
|
||||||
|
if (digits[0] == 'n' || digits[0] == 'N')
|
||||||
|
sign = 0;
|
||||||
|
|
||||||
|
/* We only need 5 bytes to hold the result "+inf\0" . */
|
||||||
|
bufsize = 5; /* Used later in an assert. */
|
||||||
|
buf = (char *)PyMem_Malloc(bufsize);
|
||||||
|
if (buf == NULL) {
|
||||||
|
PyErr_NoMemory();
|
||||||
|
goto exit;
|
||||||
|
}
|
||||||
|
p = buf;
|
||||||
|
|
||||||
|
if (sign == 1) {
|
||||||
|
*p++ = '-';
|
||||||
|
}
|
||||||
|
else if (always_add_sign) {
|
||||||
|
*p++ = '+';
|
||||||
|
}
|
||||||
|
if (digits[0] == 'i' || digits[0] == 'I') {
|
||||||
|
strncpy(p, float_strings[OFS_INF], 3);
|
||||||
|
p += 3;
|
||||||
|
|
||||||
|
if (type)
|
||||||
|
*type = Py_DTST_INFINITE;
|
||||||
|
}
|
||||||
|
else if (digits[0] == 'n' || digits[0] == 'N') {
|
||||||
|
strncpy(p, float_strings[OFS_NAN], 3);
|
||||||
|
p += 3;
|
||||||
|
|
||||||
|
if (type)
|
||||||
|
*type = Py_DTST_NAN;
|
||||||
|
}
|
||||||
|
else {
|
||||||
|
/* shouldn't get here: Gay's code should always return
|
||||||
|
something starting with a digit, an 'I', or 'N' */
|
||||||
|
strncpy(p, "ERR", 3);
|
||||||
|
p += 3;
|
||||||
|
assert(0);
|
||||||
|
}
|
||||||
|
goto exit;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* The result must be finite (not inf or nan). */
|
||||||
|
if (type)
|
||||||
|
*type = Py_DTST_FINITE;
|
||||||
|
|
||||||
|
|
||||||
|
/* We got digits back, format them. We may need to pad 'digits'
|
||||||
|
either on the left or right (or both) with extra zeros, so in
|
||||||
|
general the resulting string has the form
|
||||||
|
|
||||||
|
[<sign>]<zeros><digits><zeros>[<exponent>]
|
||||||
|
|
||||||
|
where either of the <zeros> pieces could be empty, and there's a
|
||||||
|
decimal point that could appear either in <digits> or in the
|
||||||
|
leading or trailing <zeros>.
|
||||||
|
|
||||||
|
Imagine an infinite 'virtual' string vdigits, consisting of the
|
||||||
|
string 'digits' (starting at index 0) padded on both the left and
|
||||||
|
right with infinite strings of zeros. We want to output a slice
|
||||||
|
|
||||||
|
vdigits[vdigits_start : vdigits_end]
|
||||||
|
|
||||||
|
of this virtual string. Thus if vdigits_start < 0 then we'll end
|
||||||
|
up producing some leading zeros; if vdigits_end > digits_len there
|
||||||
|
will be trailing zeros in the output. The next section of code
|
||||||
|
determines whether to use an exponent or not, figures out the
|
||||||
|
position 'decpt' of the decimal point, and computes 'vdigits_start'
|
||||||
|
and 'vdigits_end'. */
|
||||||
|
vdigits_end = digits_len;
|
||||||
|
switch (format_code) {
|
||||||
|
case 'e':
|
||||||
|
use_exp = 1;
|
||||||
|
vdigits_end = precision;
|
||||||
|
break;
|
||||||
|
case 'f':
|
||||||
|
vdigits_end = decpt + precision;
|
||||||
|
break;
|
||||||
|
case 'g':
|
||||||
|
if (decpt <= -4 || decpt >
|
||||||
|
(add_dot_0_if_integer ? precision-1 : precision))
|
||||||
|
use_exp = 1;
|
||||||
|
if (use_alt_formatting)
|
||||||
|
vdigits_end = precision;
|
||||||
|
break;
|
||||||
|
case 'r':
|
||||||
|
/* convert to exponential format at 1e16. We used to convert
|
||||||
|
at 1e17, but that gives odd-looking results for some values
|
||||||
|
when a 16-digit 'shortest' repr is padded with bogus zeros.
|
||||||
|
For example, repr(2e16+8) would give 20000000000000010.0;
|
||||||
|
the true value is 20000000000000008.0. */
|
||||||
|
if (decpt <= -4 || decpt > 16)
|
||||||
|
use_exp = 1;
|
||||||
|
break;
|
||||||
|
default:
|
||||||
|
PyErr_BadInternalCall();
|
||||||
|
goto exit;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* if using an exponent, reset decimal point position to 1 and adjust
|
||||||
|
exponent accordingly.*/
|
||||||
|
if (use_exp) {
|
||||||
|
exp = decpt - 1;
|
||||||
|
decpt = 1;
|
||||||
|
}
|
||||||
|
/* ensure vdigits_start < decpt <= vdigits_end, or vdigits_start <
|
||||||
|
decpt < vdigits_end if add_dot_0_if_integer and no exponent */
|
||||||
|
vdigits_start = decpt <= 0 ? decpt-1 : 0;
|
||||||
|
if (!use_exp && add_dot_0_if_integer)
|
||||||
|
vdigits_end = vdigits_end > decpt ? vdigits_end : decpt + 1;
|
||||||
|
else
|
||||||
|
vdigits_end = vdigits_end > decpt ? vdigits_end : decpt;
|
||||||
|
|
||||||
|
/* double check inequalities */
|
||||||
|
assert(vdigits_start <= 0 &&
|
||||||
|
0 <= digits_len &&
|
||||||
|
digits_len <= vdigits_end);
|
||||||
|
/* decimal point should be in (vdigits_start, vdigits_end] */
|
||||||
|
assert(vdigits_start < decpt && decpt <= vdigits_end);
|
||||||
|
|
||||||
|
/* Compute an upper bound how much memory we need. This might be a few
|
||||||
|
chars too long, but no big deal. */
|
||||||
|
bufsize =
|
||||||
|
/* sign, decimal point and trailing 0 byte */
|
||||||
|
3 +
|
||||||
|
|
||||||
|
/* total digit count (including zero padding on both sides) */
|
||||||
|
(vdigits_end - vdigits_start) +
|
||||||
|
|
||||||
|
/* exponent "e+100", max 3 numerical digits */
|
||||||
|
(use_exp ? 5 : 0);
|
||||||
|
|
||||||
|
/* Now allocate the memory and initialize p to point to the start of
|
||||||
|
it. */
|
||||||
|
buf = (char *)PyMem_Malloc(bufsize);
|
||||||
|
if (buf == NULL) {
|
||||||
|
PyErr_NoMemory();
|
||||||
|
goto exit;
|
||||||
|
}
|
||||||
|
p = buf;
|
||||||
|
|
||||||
|
/* Add a negative sign if negative, and a plus sign if non-negative
|
||||||
|
and always_add_sign is true. */
|
||||||
|
if (sign == 1)
|
||||||
|
*p++ = '-';
|
||||||
|
else if (always_add_sign)
|
||||||
|
*p++ = '+';
|
||||||
|
|
||||||
|
/* note that exactly one of the three 'if' conditions is true,
|
||||||
|
so we include exactly one decimal point */
|
||||||
|
/* Zero padding on left of digit string */
|
||||||
|
if (decpt <= 0) {
|
||||||
|
memset(p, '0', decpt-vdigits_start);
|
||||||
|
p += decpt - vdigits_start;
|
||||||
|
*p++ = '.';
|
||||||
|
memset(p, '0', 0-decpt);
|
||||||
|
p += 0-decpt;
|
||||||
|
}
|
||||||
|
else {
|
||||||
|
memset(p, '0', 0-vdigits_start);
|
||||||
|
p += 0 - vdigits_start;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* Digits, with included decimal point */
|
||||||
|
if (0 < decpt && decpt <= digits_len) {
|
||||||
|
strncpy(p, digits, decpt-0);
|
||||||
|
p += decpt-0;
|
||||||
|
*p++ = '.';
|
||||||
|
strncpy(p, digits+decpt, digits_len-decpt);
|
||||||
|
p += digits_len-decpt;
|
||||||
|
}
|
||||||
|
else {
|
||||||
|
strncpy(p, digits, digits_len);
|
||||||
|
p += digits_len;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* And zeros on the right */
|
||||||
|
if (digits_len < decpt) {
|
||||||
|
memset(p, '0', decpt-digits_len);
|
||||||
|
p += decpt-digits_len;
|
||||||
|
*p++ = '.';
|
||||||
|
memset(p, '0', vdigits_end-decpt);
|
||||||
|
p += vdigits_end-decpt;
|
||||||
|
}
|
||||||
|
else {
|
||||||
|
memset(p, '0', vdigits_end-digits_len);
|
||||||
|
p += vdigits_end-digits_len;
|
||||||
|
}
|
||||||
|
|
||||||
|
/* Delete a trailing decimal pt unless using alternative formatting. */
|
||||||
|
if (p[-1] == '.' && !use_alt_formatting)
|
||||||
|
p--;
|
||||||
|
|
||||||
|
/* Now that we've done zero padding, add an exponent if needed. */
|
||||||
|
if (use_exp) {
|
||||||
|
*p++ = float_strings[OFS_E][0];
|
||||||
|
exp_len = sprintf(p, "%+.02d", exp);
|
||||||
|
p += exp_len;
|
||||||
|
}
|
||||||
|
exit:
|
||||||
|
if (buf) {
|
||||||
|
*p = '\0';
|
||||||
|
/* It's too late if this fails, as we've already stepped on
|
||||||
|
memory that isn't ours. But it's an okay debugging test. */
|
||||||
|
assert(p-buf < bufsize);
|
||||||
|
}
|
||||||
|
if (digits)
|
||||||
|
_Py_dg_freedtoa(digits);
|
||||||
|
|
||||||
|
return buf;
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
PyAPI_FUNC(char *) PyOS_double_to_string(double val,
|
||||||
|
char format_code,
|
||||||
|
int precision,
|
||||||
|
int flags,
|
||||||
|
int *type)
|
||||||
|
{
|
||||||
|
char **float_strings = lc_float_strings;
|
||||||
|
int mode;
|
||||||
|
|
||||||
|
/* Validate format_code, and map upper and lower case. Compute the
|
||||||
|
mode and make any adjustments as needed. */
|
||||||
|
switch (format_code) {
|
||||||
|
/* exponent */
|
||||||
|
case 'E':
|
||||||
|
float_strings = uc_float_strings;
|
||||||
|
format_code = 'e';
|
||||||
|
/* Fall through. */
|
||||||
|
case 'e':
|
||||||
|
mode = 2;
|
||||||
|
precision++;
|
||||||
|
break;
|
||||||
|
|
||||||
|
/* fixed */
|
||||||
|
case 'F':
|
||||||
|
float_strings = uc_float_strings;
|
||||||
|
format_code = 'f';
|
||||||
|
/* Fall through. */
|
||||||
|
case 'f':
|
||||||
|
mode = 3;
|
||||||
|
break;
|
||||||
|
|
||||||
|
/* general */
|
||||||
|
case 'G':
|
||||||
|
float_strings = uc_float_strings;
|
||||||
|
format_code = 'g';
|
||||||
|
/* Fall through. */
|
||||||
|
case 'g':
|
||||||
|
mode = 2;
|
||||||
|
/* precision 0 makes no sense for 'g' format; interpret as 1 */
|
||||||
|
if (precision == 0)
|
||||||
|
precision = 1;
|
||||||
|
break;
|
||||||
|
|
||||||
|
/* repr format */
|
||||||
|
case 'r':
|
||||||
|
mode = 0;
|
||||||
|
/* Supplied precision is unused, must be 0. */
|
||||||
|
if (precision != 0) {
|
||||||
|
PyErr_BadInternalCall();
|
||||||
|
return NULL;
|
||||||
|
}
|
||||||
|
break;
|
||||||
|
|
||||||
|
default:
|
||||||
|
PyErr_BadInternalCall();
|
||||||
|
return NULL;
|
||||||
|
}
|
||||||
|
|
||||||
|
return format_float_short(val, format_code, mode, precision,
|
||||||
|
flags & Py_DTSF_SIGN,
|
||||||
|
flags & Py_DTSF_ADD_DOT_0,
|
||||||
|
flags & Py_DTSF_ALT,
|
||||||
|
float_strings, type);
|
||||||
|
}
|
||||||
|
#endif /* ifdef PY_NO_SHORT_FLOAT_REPR */
|
||||||
|
|
Loading…
Add table
Add a link
Reference in a new issue