/* -*- Mode: C; c-file-style: "python" -*- */
#include
#include
/* 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. */
#ifndef PY_NO_SHORT_FLOAT_REPR
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 = _Py_dg_infinity(negate);
}
else if (case_insensitive_match(s, "nan")) {
s += 3;
retval = _Py_dg_stdnan(negate);
}
else {
s = p;
retval = -1.0;
}
*endptr = (char *)s;
return retval;
}
#else
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;
}
#endif
/**
* _PyOS_ascii_strtod:
* @nptr: the string to convert to a numeric value.
* @endptr: if non-%NULL, it returns the character after
* the last character used in the conversion.
*
* Converts a string to a #gdouble value.
* This function behaves like the standard strtod() function
* does in the C locale. It does this without actually
* changing the current locale, since that would not be
* thread-safe.
*
* This function is typically used when reading configuration
* files or other non-user input that should be locale independent.
* To handle input from the user you should normally use the
* locale-sensitive system strtod() function.
*
* If the correct value would cause overflow, plus or minus %HUGE_VAL
* is returned (according to the sign of the value), and %ERANGE is
* stored in %errno. If the correct value would cause underflow,
* zero is returned and %ERANGE is stored in %errno.
* If memory allocation fails, %ENOMEM is stored in %errno.
*
* This function resets %errno before calling strtod() so that
* you can reliably detect overflow and underflow.
*
* Return value: the #gdouble value.
**/
#ifndef PY_NO_SHORT_FLOAT_REPR
static 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 an inf or nan */
result = _Py_parse_inf_or_nan(nptr, endptr);
return result;
}
#else
/*
Use system strtod; since strtod is locale aware, we may
have to first fix the decimal separator.
Note that unlike _Py_dg_strtod, the system strtod may not always give
correctly rounded results.
*/
static double
_PyOS_ascii_strtod(const char *nptr, char **endptr)
{
char *fail_pos;
double val;
struct lconv *locale_data;
const char *decimal_point;
size_t decimal_point_len;
const char *p, *decimal_point_pos;
const char *end = NULL; /* Silence gcc */
const char *digits_pos = NULL;
int negate = 0;
assert(nptr != NULL);
fail_pos = NULL;
locale_data = localeconv();
decimal_point = locale_data->decimal_point;
decimal_point_len = strlen(decimal_point);
assert(decimal_point_len != 0);
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
and underflows */
errno = 0;
/* We process the optional sign manually, then pass the remainder to
the system strtod. This ensures that the result of an underflow
has the correct sign. (bug #1725) */
p = nptr;
/* Process leading sign, if present */
if (*p == '-') {
negate = 1;
p++;
}
else if (*p == '+') {
p++;
}
/* Some platform strtods accept hex floats; Python shouldn't (at the
moment), so we check explicitly for strings starting with '0x'. */
if (*p == '0' && (*(p+1) == 'x' || *(p+1) == 'X'))
goto invalid_string;
/* Check that what's left begins with a digit or decimal point */
if (!Py_ISDIGIT(*p) && *p != '.')
goto invalid_string;
digits_pos = p;
if (decimal_point[0] != '.' ||
decimal_point[1] != 0)
{
/* Look for a '.' in the input; if present, it'll need to be
swapped for the current locale's decimal point before we
call strtod. On the other hand, if we find the current
locale's decimal point then the input is invalid. */
while (Py_ISDIGIT(*p))
p++;
if (*p == '.')
{
decimal_point_pos = p++;
/* locate end of number */
while (Py_ISDIGIT(*p))
p++;
if (*p == 'e' || *p == 'E')
p++;
if (*p == '+' || *p == '-')
p++;
while (Py_ISDIGIT(*p))
p++;
end = p;
}
else if (strncmp(p, decimal_point, decimal_point_len) == 0)
/* Python bug #1417699 */
goto invalid_string;
/* For the other cases, we need not convert the decimal
point */
}
if (decimal_point_pos) {
char *copy, *c;
/* Create a copy of the input, with the '.' converted to the
locale-specific decimal point */
copy = (char *)PyMem_MALLOC(end - digits_pos +
1 + decimal_point_len);
if (copy == NULL) {
*endptr = (char *)nptr;
errno = ENOMEM;
return val;
}
c = copy;
memcpy(c, digits_pos, decimal_point_pos - digits_pos);
c += decimal_point_pos - digits_pos;
memcpy(c, decimal_point, decimal_point_len);
c += decimal_point_len;
memcpy(c, decimal_point_pos + 1,
end - (decimal_point_pos + 1));
c += end - (decimal_point_pos + 1);
*c = 0;
val = strtod(copy, &fail_pos);
if (fail_pos)
{
if (fail_pos > decimal_point_pos)
fail_pos = (char *)digits_pos +
(fail_pos - copy) -
(decimal_point_len - 1);
else
fail_pos = (char *)digits_pos +
(fail_pos - copy);
}
PyMem_FREE(copy);
}
else {
val = strtod(digits_pos, &fail_pos);
}
if (fail_pos == digits_pos)
goto invalid_string;
if (negate && fail_pos != nptr)
val = -val;
*endptr = fail_pos;
return val;
invalid_string:
*endptr = (char*)nptr;
errno = EINVAL;
return -1.0;
}
#endif
/* PyOS_string_to_double 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. 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 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;
}
#ifdef PY_NO_SHORT_FLOAT_REPR
/* Given a string that may have a decimal point in the current
locale, change it back to a dot. Since the string cannot get
longer, no need for a maximum buffer size parameter. */
Py_LOCAL_INLINE(void)
change_decimal_from_locale_to_dot(char* buffer)
{
struct lconv *locale_data = localeconv();
const char *decimal_point = locale_data->decimal_point;
if (decimal_point[0] != '.' || decimal_point[1] != 0) {
size_t decimal_point_len = strlen(decimal_point);
if (*buffer == '+' || *buffer == '-')
buffer++;
while (Py_ISDIGIT(*buffer))
buffer++;
if (strncmp(buffer, decimal_point, decimal_point_len) == 0) {
*buffer = '.';
buffer++;
if (decimal_point_len > 1) {
/* buffer needs to get smaller */
size_t rest_len = strlen(buffer +
(decimal_point_len - 1));
memmove(buffer,
buffer + (decimal_point_len - 1),
rest_len);
buffer[rest_len] = 0;
}
}
}
}
/* From the C99 standard, section 7.19.6:
The exponent always contains at least two digits, and only as many more digits
as necessary to represent the exponent.
*/
#define MIN_EXPONENT_DIGITS 2
/* Ensure that any exponent, if present, is at least MIN_EXPONENT_DIGITS
in length. */
Py_LOCAL_INLINE(void)
ensure_minimum_exponent_length(char* buffer, size_t buf_size)
{
char *p = strpbrk(buffer, "eE");
if (p && (*(p + 1) == '-' || *(p + 1) == '+')) {
char *start = p + 2;
int exponent_digit_cnt = 0;
int leading_zero_cnt = 0;
int in_leading_zeros = 1;
int significant_digit_cnt;
/* Skip over the exponent and the sign. */
p += 2;
/* Find the end of the exponent, keeping track of leading
zeros. */
while (*p && Py_ISDIGIT(*p)) {
if (in_leading_zeros && *p == '0')
++leading_zero_cnt;
if (*p != '0')
in_leading_zeros = 0;
++p;
++exponent_digit_cnt;
}
significant_digit_cnt = exponent_digit_cnt - leading_zero_cnt;
if (exponent_digit_cnt == MIN_EXPONENT_DIGITS) {
/* If there are 2 exactly digits, we're done,
regardless of what they contain */
}
else if (exponent_digit_cnt > MIN_EXPONENT_DIGITS) {
int extra_zeros_cnt;
/* There are more than 2 digits in the exponent. See
if we can delete some of the leading zeros */
if (significant_digit_cnt < MIN_EXPONENT_DIGITS)
significant_digit_cnt = MIN_EXPONENT_DIGITS;
extra_zeros_cnt = exponent_digit_cnt -
significant_digit_cnt;
/* Delete extra_zeros_cnt worth of characters from the
front of the exponent */
assert(extra_zeros_cnt >= 0);
/* Add one to significant_digit_cnt to copy the
trailing 0 byte, thus setting the length */
memmove(start,
start + extra_zeros_cnt,
significant_digit_cnt + 1);
}
else {
/* If there are fewer than 2 digits, add zeros
until there are 2, if there's enough room */
int zeros = MIN_EXPONENT_DIGITS - exponent_digit_cnt;
if (start + zeros + exponent_digit_cnt + 1
< buffer + buf_size) {
memmove(start + zeros, start,
exponent_digit_cnt + 1);
memset(start, '0', zeros);
}
}
}
}
/* Remove trailing zeros after the decimal point from a numeric string; also
remove the decimal point if all digits following it are zero. The numeric
string must end in '\0', and should not have any leading or trailing
whitespace. Assumes that the decimal point is '.'. */
Py_LOCAL_INLINE(void)
remove_trailing_zeros(char *buffer)
{
char *old_fraction_end, *new_fraction_end, *end, *p;
p = buffer;
if (*p == '-' || *p == '+')
/* Skip leading sign, if present */
++p;
while (Py_ISDIGIT(*p))
++p;
/* if there's no decimal point there's nothing to do */
if (*p++ != '.')
return;
/* scan any digits after the point */
while (Py_ISDIGIT(*p))
++p;
old_fraction_end = p;
/* scan up to ending '\0' */
while (*p != '\0')
p++;
/* +1 to make sure that we move the null byte as well */
end = p+1;
/* scan back from fraction_end, looking for removable zeros */
p = old_fraction_end;
while (*(p-1) == '0')
--p;
/* and remove point if we've got that far */
if (*(p-1) == '.')
--p;
new_fraction_end = p;
memmove(new_fraction_end, old_fraction_end, end-old_fraction_end);
}
/* Ensure that buffer has a decimal point in it. The decimal point will not
be in the current locale, it will always be '.'. Don't add a decimal point
if an exponent is present. Also, convert to exponential notation where
adding a '.0' would produce too many significant digits (see issue 5864).
Returns a pointer to the fixed buffer, or NULL on failure.
*/
Py_LOCAL_INLINE(char *)
ensure_decimal_point(char* buffer, size_t buf_size, int precision)
{
int digit_count, insert_count = 0, convert_to_exp = 0;
char *chars_to_insert, *digits_start;
/* search for the first non-digit character */
char *p = buffer;
if (*p == '-' || *p == '+')
/* Skip leading sign, if present. I think this could only
ever be '-', but it can't hurt to check for both. */
++p;
digits_start = p;
while (*p && Py_ISDIGIT(*p))
++p;
digit_count = Py_SAFE_DOWNCAST(p - digits_start, Py_ssize_t, int);
if (*p == '.') {
if (Py_ISDIGIT(*(p+1))) {
/* Nothing to do, we already have a decimal
point and a digit after it */
}
else {
/* We have a decimal point, but no following
digit. Insert a zero after the decimal. */
/* can't ever get here via PyOS_double_to_string */
assert(precision == -1);
++p;
chars_to_insert = "0";
insert_count = 1;
}
}
else if (!(*p == 'e' || *p == 'E')) {
/* Don't add ".0" if we have an exponent. */
if (digit_count == precision) {
/* issue 5864: don't add a trailing .0 in the case
where the '%g'-formatted result already has as many
significant digits as were requested. Switch to
exponential notation instead. */
convert_to_exp = 1;
/* no exponent, no point, and we shouldn't land here
for infs and nans, so we must be at the end of the
string. */
assert(*p == '\0');
}
else {
assert(precision == -1 || digit_count < precision);
chars_to_insert = ".0";
insert_count = 2;
}
}
if (insert_count) {
size_t buf_len = strlen(buffer);
if (buf_len + insert_count + 1 >= buf_size) {
/* If there is not enough room in the buffer
for the additional text, just skip it. It's
not worth generating an error over. */
}
else {
memmove(p + insert_count, p,
buffer + strlen(buffer) - p + 1);
memcpy(p, chars_to_insert, insert_count);
}
}
if (convert_to_exp) {
int written;
size_t buf_avail;
p = digits_start;
/* insert decimal point */
assert(digit_count >= 1);
memmove(p+2, p+1, digit_count); /* safe, but overwrites nul */
p[1] = '.';
p += digit_count+1;
assert(p = 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
[][]
where either of the pieces could be empty, and there's a
decimal point that could appear either in or in the
leading or trailing .
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
(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 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 = (int)decpt - 1;
decpt = 1;
}
/* ensure vdigits_start < decpt decpt ? vdigits_end : decpt;
/* double check inequalities */
assert(vdigits_start