diff options
Diffstat (limited to 'Objects')
| -rw-r--r-- | Objects/bytesobject.c | 1 | ||||
| -rw-r--r-- | Objects/complexobject.c | 167 | ||||
| -rw-r--r-- | Objects/floatobject.c | 145 | ||||
| -rw-r--r-- | Objects/stringlib/formatter.h | 718 | ||||
| -rw-r--r-- | Objects/stringlib/localeutil.h | 269 | ||||
| -rw-r--r-- | Objects/unicodeobject.c | 170 |
6 files changed, 735 insertions, 735 deletions
diff --git a/Objects/bytesobject.c b/Objects/bytesobject.c index 1239680..f306cb8 100644 --- a/Objects/bytesobject.c +++ b/Objects/bytesobject.c @@ -562,6 +562,7 @@ PyBytes_AsStringAndSize(register PyObject *obj, /* -------------------------------------------------------------------- */ /* Methods */ +#include "stringlib/stringdefs.h" #define STRINGLIB_CHAR char #define STRINGLIB_CMP memcmp diff --git a/Objects/complexobject.c b/Objects/complexobject.c index 207ecdd0..721db8f 100644 --- a/Objects/complexobject.c +++ b/Objects/complexobject.c @@ -14,22 +14,6 @@ #ifndef WITHOUT_COMPLEX -/* Precisions used by repr() and str(), respectively. - - 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. - - The str() precision is chosen so that in most cases, the rounding noise - created by various operations is suppressed, while giving plenty of - precision for practical use. -*/ - -#define PREC_REPR 17 -#define PREC_STR 12 - /* elementary operations on complex numbers */ static Py_complex c_1 = {1., 0.}; @@ -345,71 +329,114 @@ complex_dealloc(PyObject *op) } -static void -complex_to_buf(char *buf, int bufsz, PyComplexObject *v, int precision) +static PyObject * +complex_format(PyComplexObject *v, char format_code) { - char format[32]; - if (v->cval.real == 0.) { - if (!Py_IS_FINITE(v->cval.imag)) { - if (Py_IS_NAN(v->cval.imag)) - strncpy(buf, "nan*j", 6); - else if (copysign(1, v->cval.imag) == 1) - strncpy(buf, "inf*j", 6); - else - strncpy(buf, "-inf*j", 7); - } - else { - PyOS_snprintf(format, sizeof(format), "%%.%ig", precision); - PyOS_ascii_formatd(buf, bufsz - 1, format, v->cval.imag); - strncat(buf, "j", 1); - } - } else { - char re[64], im[64]; - /* Format imaginary part with sign, real part without */ - if (!Py_IS_FINITE(v->cval.real)) { - if (Py_IS_NAN(v->cval.real)) - strncpy(re, "nan", 4); - /* else if (copysign(1, v->cval.real) == 1) */ - else if (v->cval.real > 0) - strncpy(re, "inf", 4); - else - strncpy(re, "-inf", 5); - } - else { - PyOS_snprintf(format, sizeof(format), "%%.%ig", precision); - PyOS_ascii_formatd(re, sizeof(re), format, v->cval.real); - } - if (!Py_IS_FINITE(v->cval.imag)) { - if (Py_IS_NAN(v->cval.imag)) - strncpy(im, "+nan*", 6); - /* else if (copysign(1, v->cval.imag) == 1) */ - else if (v->cval.imag > 0) - strncpy(im, "+inf*", 6); - else - strncpy(im, "-inf*", 6); - } - else { - PyOS_snprintf(format, sizeof(format), "%%+.%ig", precision); - PyOS_ascii_formatd(im, sizeof(im), format, v->cval.imag); - } - PyOS_snprintf(buf, bufsz, "(%s%sj)", re, im); - } + PyObject *result = NULL; + Py_ssize_t len; + + /* If these are non-NULL, they'll need to be freed. */ + char *pre = NULL; + char *pim = NULL; + char *buf = NULL; + + /* These do not need to be freed. They're either aliases for pim + and pre, or pointers to constants. */ + char *re = NULL; + char *im = NULL; + char *lead = ""; + char *tail = ""; + + + if (v->cval.real == 0.) { + re = ""; + if (!Py_IS_FINITE(v->cval.imag)) { + if (Py_IS_NAN(v->cval.imag)) + im = "nan*"; + else if (copysign(1, v->cval.imag) == 1) + im = "inf*"; + else + im = "-inf*"; + } + else { + pim = PyOS_double_to_string(v->cval.imag, format_code, + 0, 0, NULL); + if (!pim) { + PyErr_NoMemory(); + goto done; + } + im = pim; + } + } else { + /* Format imaginary part with sign, real part without */ + if (!Py_IS_FINITE(v->cval.real)) { + if (Py_IS_NAN(v->cval.real)) + re = "nan"; + /* else if (copysign(1, v->cval.real) == 1) */ + else if (v->cval.real > 0) + re = "inf"; + else + re = "-inf"; + } + else { + pre = PyOS_double_to_string(v->cval.real, format_code, + 0, 0, NULL); + if (!pre) { + PyErr_NoMemory(); + goto done; + } + re = pre; + } + + if (!Py_IS_FINITE(v->cval.imag)) { + if (Py_IS_NAN(v->cval.imag)) + im = "+nan*"; + /* else if (copysign(1, v->cval.imag) == 1) */ + else if (v->cval.imag > 0) + im = "+inf*"; + else + im = "-inf*"; + } + else { + pim = PyOS_double_to_string(v->cval.imag, format_code, + 0, Py_DTSF_SIGN, NULL); + if (!pim) { + PyErr_NoMemory(); + goto done; + } + im = pim; + } + lead = "("; + tail = ")"; + } + /* Alloc the final buffer. Add one for the "j" in the format string, and + one for the trailing zero. */ + len = strlen(lead) + strlen(re) + strlen(im) + strlen(tail) + 2; + buf = PyMem_Malloc(len); + if (!buf) { + PyErr_NoMemory(); + goto done; + } + PyOS_snprintf(buf, len, "%s%s%sj%s", lead, re, im, tail); + result = PyUnicode_FromString(buf); +done: + PyMem_Free(pim); + PyMem_Free(pre); + PyMem_Free(buf); + + return result; } static PyObject * complex_repr(PyComplexObject *v) { - char buf[100]; - complex_to_buf(buf, sizeof(buf), v, PREC_REPR); - return PyUnicode_FromString(buf); + return complex_format(v, 'r'); } static PyObject * complex_str(PyComplexObject *v) { - char buf[100]; - complex_to_buf(buf, sizeof(buf), v, PREC_STR); - return PyUnicode_FromString(buf); + return complex_format(v, 's'); } static long diff --git a/Objects/floatobject.c b/Objects/floatobject.c index 2ef4d1a..2fbe810 100644 --- a/Objects/floatobject.c +++ b/Objects/floatobject.c @@ -197,8 +197,7 @@ PyFloat_FromString(PyObject *v) sp = s; /* We don't care about overflow or underflow. If the platform supports * them, infinities and signed zeroes (on underflow) are fine. - * However, strtod can return 0 for denormalized numbers, where atof - * does not. So (alas!) we special-case a zero result. Note that + * However, strtod can return 0 for denormalized numbers. Note that * whether strtod sets errno on underflow is not defined, so we can't * key off errno. */ @@ -259,14 +258,6 @@ PyFloat_FromString(PyObject *v) "null byte in argument for float()"); goto error; } - if (x == 0.0) { - /* See above -- may have been strtod being anal - about denorms. */ - PyFPE_START_PROTECT("atof", goto error) - x = PyOS_ascii_atof(s); - PyFPE_END_PROTECT(x) - errno = 0; /* whether atof ever set errno is undefined */ - } result = PyFloat_FromDouble(x); error: if (s_buffer) @@ -320,72 +311,6 @@ PyFloat_AsDouble(PyObject *op) return val; } -/* Methods */ - -static void -format_double(char *buf, size_t buflen, double ob_fval, int precision) -{ - register char *cp; - char format[32]; - int i; - - /* Subroutine for float_repr, float_str and float_print. - We want float numbers to be recognizable as such, - i.e., they should contain a decimal point or an exponent. - However, %g may print the number as an integer; - in such cases, we append ".0" to the string. */ - - PyOS_snprintf(format, 32, "%%.%ig", precision); - PyOS_ascii_formatd(buf, buflen, format, ob_fval); - cp = buf; - if (*cp == '-') - cp++; - for (; *cp != '\0'; cp++) { - /* Any non-digit means it's not an integer; - this takes care of NAN and INF as well. */ - if (!isdigit(Py_CHARMASK(*cp))) - break; - } - if (*cp == '\0') { - *cp++ = '.'; - *cp++ = '0'; - *cp++ = '\0'; - return; - } - /* Checking the next three chars should be more than enough to - * detect inf or nan, even on Windows. We check for inf or nan - * at last because they are rare cases. - */ - for (i=0; *cp != '\0' && i<3; cp++, i++) { - if (isdigit(Py_CHARMASK(*cp)) || *cp == '.') - continue; - /* found something that is neither a digit nor point - * it might be a NaN or INF - */ -#ifdef Py_NAN - if (Py_IS_NAN(ob_fval)) { - strcpy(buf, "nan"); - } - else -#endif - if (Py_IS_INFINITY(ob_fval)) { - cp = buf; - if (*cp == '-') - cp++; - strcpy(cp, "inf"); - } - break; - } - -} - -static void -format_float(char *buf, size_t buflen, PyFloatObject *v, int precision) -{ - assert(PyFloat_Check(v)); - format_double(buf, buflen, PyFloat_AS_DOUBLE(v), precision); -} - /* Macro and helper that convert PyObject obj to a C double and store the value in dbl. If conversion to double raises an exception, obj is set to NULL, and the function invoking this macro returns NULL. If @@ -398,6 +323,8 @@ format_float(char *buf, size_t buflen, PyFloatObject *v, int precision) else if (convert_to_double(&(obj), &(dbl)) < 0) \ return obj; +/* Methods */ + static int convert_to_double(PyObject **v, double *dbl) { @@ -418,38 +345,30 @@ convert_to_double(PyObject **v, double *dbl) return 0; } -/* Precisions used by repr() and str(), respectively. - - 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. - - The str() precision is chosen so that in most cases, the rounding noise - created by various operations is suppressed, while giving plenty of - precision for practical use. - -*/ - -#define PREC_REPR 17 -#define PREC_STR 12 +static PyObject * +float_str_or_repr(PyFloatObject *v, char format_code) +{ + PyObject *result; + char *buf = PyOS_double_to_string(PyFloat_AS_DOUBLE(v), + format_code, 0, Py_DTSF_ADD_DOT_0, + NULL); + if (!buf) + return PyErr_NoMemory(); + result = PyUnicode_FromString(buf); + PyMem_Free(buf); + return result; +} static PyObject * float_repr(PyFloatObject *v) { - char buf[100]; - format_float(buf, sizeof(buf), v, PREC_REPR); - - return PyUnicode_FromString(buf); + return float_str_or_repr(v, 'r'); } static PyObject * float_str(PyFloatObject *v) { - char buf[100]; - format_float(buf, sizeof(buf), v, PREC_STR); - return PyUnicode_FromString(buf); + return float_str_or_repr(v, 's'); } /* Comparison is pretty much a nightmare. When comparing float to float, @@ -1980,15 +1899,21 @@ PyFloat_Fini(void) i++, p++) { if (PyFloat_CheckExact(p) && Py_REFCNT(p) != 0) { - char buf[100]; - format_float(buf, sizeof(buf), p, PREC_STR); - /* XXX(twouters) cast refcount to - long until %zd is universally - available - */ - fprintf(stderr, + char *buf = PyOS_double_to_string( + PyFloat_AS_DOUBLE(p), 'r', + 0, 0, NULL); + if (buf) { + /* XXX(twouters) cast + refcount to long + until %zd is + universally + available + */ + fprintf(stderr, "# <float at %p, refcnt=%ld, val=%s>\n", p, (long)Py_REFCNT(p), buf); + PyMem_Free(buf); + } } } list = list->next; @@ -2233,14 +2158,6 @@ _PyFloat_Pack8(double x, unsigned char *p, int le) } } -/* Should only be used by marshal. */ -int -_PyFloat_Repr(double x, char *p, size_t len) -{ - format_double(p, len, x, PREC_REPR); - return (int)strlen(p); -} - double _PyFloat_Unpack4(const unsigned char *p, int le) { diff --git a/Objects/stringlib/formatter.h b/Objects/stringlib/formatter.h index d936a67..61ca12b 100644 --- a/Objects/stringlib/formatter.h +++ b/Objects/stringlib/formatter.h @@ -1,6 +1,8 @@ /* implements the string, long, and float formatters. that is, string.__format__, etc. */ +#include <locale.h> + /* Before including this, you must include either: stringlib/unicodedefs.h stringlib/stringdefs.h @@ -13,8 +15,6 @@ be. These are the only non-static functions defined here. */ -#define ALLOW_PARENS_FOR_SIGN 0 - /* Raises an exception about an unknown presentation type for this * type. */ @@ -104,9 +104,6 @@ is_sign_element(STRINGLIB_CHAR c) { switch (c) { case ' ': case '+': case '-': -#if ALLOW_PARENS_FOR_SIGN - case '(': -#endif return 1; default: return 0; @@ -143,7 +140,7 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, /* end-ptr is used throughout this code to specify the length of the input string */ - Py_ssize_t specified_width; + Py_ssize_t consumed; format->fill_char = '\0'; format->align = '\0'; @@ -170,11 +167,6 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, if (end-ptr >= 1 && is_sign_element(ptr[0])) { format->sign = ptr[0]; ++ptr; -#if ALLOW_PARENS_FOR_SIGN - if (end-ptr >= 1 && ptr[0] == ')') { - ++ptr; - } -#endif } /* If the next character is #, we're in alternate mode. This only @@ -193,15 +185,17 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, ++ptr; } - /* XXX add error checking */ - specified_width = get_integer(&ptr, end, &format->width); + consumed = get_integer(&ptr, end, &format->width); + if (consumed == -1) + /* Overflow error. Exception already set. */ + return 0; - /* if specified_width is 0, we didn't consume any characters for - the width. in that case, reset the width to -1, because - get_integer() will have set it to zero */ - if (specified_width == 0) { + /* If consumed is 0, we didn't consume any characters for the + width. In that case, reset the width to -1, because + get_integer() will have set it to zero. -1 is how we record + that the width wasn't specified. */ + if (consumed == 0) format->width = -1; - } /* Comma signifies add thousands separators */ if (end-ptr && ptr[0] == ',') { @@ -213,11 +207,13 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, if (end-ptr && ptr[0] == '.') { ++ptr; - /* XXX add error checking */ - specified_width = get_integer(&ptr, end, &format->precision); + consumed = get_integer(&ptr, end, &format->precision); + if (consumed == -1) + /* Overflow error. Exception already set. */ + return 0; - /* not having a precision after a dot is an error */ - if (specified_width == 0) { + /* Not having a precision after a dot is an error. */ + if (consumed == 0) { PyErr_Format(PyExc_ValueError, "Format specifier missing precision"); return 0; @@ -225,10 +221,10 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, } - /* Finally, parse the type field */ + /* Finally, parse the type field. */ if (end-ptr > 1) { - /* invalid conversion spec */ + /* More than one char remain, invalid conversion spec. */ PyErr_Format(PyExc_ValueError, "Invalid conversion specification"); return 0; } @@ -238,9 +234,27 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, ++ptr; } - if (format->type == 'n' && format->thousands_separators) { - PyErr_Format(PyExc_ValueError, "Cannot specify ',' with 'n'."); - return 0; + /* Do as much validating as we can, just by looking at the format + specifier. Do not take into account what type of formatting + we're doing (int, float, string). */ + + if (format->thousands_separators) { + switch (format->type) { + case 'd': + case 'e': + case 'f': + case 'g': + case 'E': + case 'G': + case '%': + case 'F': + /* These are allowed. See PEP 378.*/ + break; + default: + PyErr_Format(PyExc_ValueError, + "Cannot specify ',' with '%c'.", format->type); + return 0; + } } return 1; @@ -251,6 +265,20 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, /*********** common routines for numeric formatting *********************/ /************************************************************************/ +/* Locale type codes. */ +#define LT_CURRENT_LOCALE 0 +#define LT_DEFAULT_LOCALE 1 +#define LT_NO_LOCALE 2 + +/* Locale info needed for formatting integers and the part of floats + before and including the decimal. Note that locales only support + 8-bit chars, not unicode. */ +typedef struct { + char *decimal_point; + char *thousands_sep; + char *grouping; +} LocaleInfo; + /* describes the layout for an integer, see the comment in calc_number_widths() for details */ typedef struct { @@ -258,38 +286,84 @@ typedef struct { Py_ssize_t n_prefix; Py_ssize_t n_spadding; Py_ssize_t n_rpadding; - char lsign; - Py_ssize_t n_lsign; - char rsign; - Py_ssize_t n_rsign; - Py_ssize_t n_total; /* just a convenience, it's derivable from the - other fields */ + char sign; + Py_ssize_t n_sign; /* number of digits needed for sign (0/1) */ + Py_ssize_t n_grouped_digits; /* Space taken up by the digits, including + any grouping chars. */ + Py_ssize_t n_decimal; /* 0 if only an integer */ + Py_ssize_t n_remainder; /* Digits in decimal and/or exponent part, + excluding the decimal itself, if + present. */ + + /* These 2 are not the widths of fields, but are needed by + STRINGLIB_GROUPING. */ + Py_ssize_t n_digits; /* The number of digits before a decimal + or exponent. */ + Py_ssize_t n_min_width; /* The min_width we used when we computed + the n_grouped_digits width. */ } NumberFieldWidths; +/* Given a number of the form: + digits[remainder] + where ptr points to the start and end points to the end, find where + the integer part ends. This could be a decimal, an exponent, both, + or neither. + If a decimal point is present, set *has_decimal and increment + remainder beyond it. + Results are undefined (but shouldn't crash) for improperly + formatted strings. +*/ +static void +parse_number(STRINGLIB_CHAR *ptr, Py_ssize_t len, + Py_ssize_t *n_remainder, int *has_decimal) +{ + STRINGLIB_CHAR *end = ptr + len; + STRINGLIB_CHAR *remainder; + + while (ptr<end && isdigit(*ptr)) + ++ptr; + remainder = ptr; + + /* Does remainder start with a decimal point? */ + *has_decimal = ptr<end && *remainder == '.'; + + /* Skip the decimal point. */ + if (*has_decimal) + remainder++; + + *n_remainder = end - remainder; +} + /* not all fields of format are used. for example, precision is unused. should this take discrete params in order to be more clear about what it does? or is passing a single format parameter easier and more efficient enough to justify a little obfuscation? */ -static void -calc_number_widths(NumberFieldWidths *spec, STRINGLIB_CHAR actual_sign, - Py_ssize_t n_prefix, Py_ssize_t n_digits, +static Py_ssize_t +calc_number_widths(NumberFieldWidths *spec, Py_ssize_t n_prefix, + STRINGLIB_CHAR sign_char, STRINGLIB_CHAR *number, + Py_ssize_t n_number, Py_ssize_t n_remainder, + int has_decimal, const LocaleInfo *locale, const InternalFormatSpec *format) { + Py_ssize_t n_non_digit_non_padding; + Py_ssize_t n_padding; + + spec->n_digits = n_number - n_remainder - (has_decimal?1:0); spec->n_lpadding = 0; - spec->n_prefix = 0; + spec->n_prefix = n_prefix; + spec->n_decimal = has_decimal ? strlen(locale->decimal_point) : 0; + spec->n_remainder = n_remainder; spec->n_spadding = 0; spec->n_rpadding = 0; - spec->lsign = '\0'; - spec->n_lsign = 0; - spec->rsign = '\0'; - spec->n_rsign = 0; + spec->sign = '\0'; + spec->n_sign = 0; /* the output will look like: - | | - | <lpadding> <lsign> <prefix> <spadding> <digits> <rsign> <rpadding> | - | | + | | + | <lpadding> <sign> <prefix> <spadding> <grouped_digits> <decimal> <remainder> <rpadding> | + | | - lsign and rsign are computed from format->sign and the actual + sign is computed from format->sign and the actual sign of the number prefix is given (it's for the '0x' prefix) @@ -304,108 +378,191 @@ calc_number_widths(NumberFieldWidths *spec, STRINGLIB_CHAR actual_sign, */ /* compute the various parts we're going to write */ - if (format->sign == '+') { + switch (format->sign) { + case '+': /* always put a + or - */ - spec->n_lsign = 1; - spec->lsign = (actual_sign == '-' ? '-' : '+'); - } -#if ALLOW_PARENS_FOR_SIGN - else if (format->sign == '(') { - if (actual_sign == '-') { - spec->n_lsign = 1; - spec->lsign = '('; - spec->n_rsign = 1; - spec->rsign = ')'; - } - } -#endif - else if (format->sign == ' ') { - spec->n_lsign = 1; - spec->lsign = (actual_sign == '-' ? '-' : ' '); - } - else { - /* non specified, or the default (-) */ - if (actual_sign == '-') { - spec->n_lsign = 1; - spec->lsign = '-'; + spec->n_sign = 1; + spec->sign = (sign_char == '-' ? '-' : '+'); + break; + case ' ': + spec->n_sign = 1; + spec->sign = (sign_char == '-' ? '-' : ' '); + break; + default: + /* Not specified, or the default (-) */ + if (sign_char == '-') { + spec->n_sign = 1; + spec->sign = '-'; } } - spec->n_prefix = n_prefix; + /* The number of chars used for non-digits and non-padding. */ + n_non_digit_non_padding = spec->n_sign + spec->n_prefix + spec->n_decimal + + spec->n_remainder; - /* now the number of padding characters */ - if (format->width == -1) { - /* no padding at all, nothing to do */ - } - else { - /* see if any padding is needed */ - if (spec->n_lsign + n_digits + spec->n_rsign + - spec->n_prefix >= format->width) { - /* no padding needed, we're already bigger than the - requested width */ - } - else { - /* determine which of left, space, or right padding is - needed */ - Py_ssize_t padding = format->width - - (spec->n_lsign + spec->n_prefix + - n_digits + spec->n_rsign); - if (format->align == '<') - spec->n_rpadding = padding; - else if (format->align == '>') - spec->n_lpadding = padding; - else if (format->align == '^') { - spec->n_lpadding = padding / 2; - spec->n_rpadding = padding - spec->n_lpadding; - } - else if (format->align == '=') - spec->n_spadding = padding; - else - spec->n_lpadding = padding; + /* min_width can go negative, that's okay. format->width == -1 means + we don't care. */ + if (format->fill_char == '0') + spec->n_min_width = format->width - n_non_digit_non_padding; + else + spec->n_min_width = 0; + + if (spec->n_digits == 0) + /* This case only occurs when using 'c' formatting, we need + to special case it because the grouping code always wants + to have at least one character. */ + spec->n_grouped_digits = 0; + else + spec->n_grouped_digits = STRINGLIB_GROUPING(NULL, 0, NULL, + spec->n_digits, + spec->n_min_width, + locale->grouping, + locale->thousands_sep); + + /* Given the desired width and the total of digit and non-digit + space we consume, see if we need any padding. format->width can + be negative (meaning no padding), but this code still works in + that case. */ + n_padding = format->width - + (n_non_digit_non_padding + spec->n_grouped_digits); + if (n_padding > 0) { + /* Some padding is needed. Determine if it's left, space, or right. */ + switch (format->align) { + case '<': + spec->n_rpadding = n_padding; + break; + case '^': + spec->n_lpadding = n_padding / 2; + spec->n_rpadding = n_padding - spec->n_lpadding; + break; + case '=': + spec->n_spadding = n_padding; + break; + default: + /* Handles '>', plus catch-all just in case. */ + spec->n_lpadding = n_padding; + break; } } - spec->n_total = spec->n_lpadding + spec->n_lsign + spec->n_prefix + - spec->n_spadding + n_digits + spec->n_rsign + spec->n_rpadding; + return spec->n_lpadding + spec->n_sign + spec->n_prefix + + spec->n_spadding + spec->n_grouped_digits + spec->n_decimal + + spec->n_remainder + spec->n_rpadding; } -/* fill in the non-digit parts of a numbers's string representation, - as determined in calc_number_widths(). returns the pointer to - where the digits go. */ -static STRINGLIB_CHAR * -fill_non_digits(STRINGLIB_CHAR *p_buf, const NumberFieldWidths *spec, - STRINGLIB_CHAR *prefix, Py_ssize_t n_digits, - STRINGLIB_CHAR fill_char) +/* Fill in the digit parts of a numbers's string representation, + as determined in calc_number_widths(). + No error checking, since we know the buffer is the correct size. */ +static void +fill_number(STRINGLIB_CHAR *buf, const NumberFieldWidths *spec, + STRINGLIB_CHAR *digits, Py_ssize_t n_digits, + STRINGLIB_CHAR *prefix, STRINGLIB_CHAR fill_char, + LocaleInfo *locale, int toupper) { - STRINGLIB_CHAR *p_digits; + /* Used to keep track of digits, decimal, and remainder. */ + STRINGLIB_CHAR *p = digits; + +#ifndef NDEBUG + Py_ssize_t r; +#endif if (spec->n_lpadding) { - STRINGLIB_FILL(p_buf, fill_char, spec->n_lpadding); - p_buf += spec->n_lpadding; + STRINGLIB_FILL(buf, fill_char, spec->n_lpadding); + buf += spec->n_lpadding; } - if (spec->n_lsign == 1) { - *p_buf++ = spec->lsign; + if (spec->n_sign == 1) { + *buf++ = spec->sign; } if (spec->n_prefix) { - memmove(p_buf, + memmove(buf, prefix, spec->n_prefix * sizeof(STRINGLIB_CHAR)); - p_buf += spec->n_prefix; + if (toupper) { + Py_ssize_t t; + for (t = 0; t < spec->n_prefix; ++t) + buf[t] = STRINGLIB_TOUPPER(buf[t]); + } + buf += spec->n_prefix; } if (spec->n_spadding) { - STRINGLIB_FILL(p_buf, fill_char, spec->n_spadding); - p_buf += spec->n_spadding; + STRINGLIB_FILL(buf, fill_char, spec->n_spadding); + buf += spec->n_spadding; } - p_digits = p_buf; - p_buf += n_digits; - if (spec->n_rsign == 1) { - *p_buf++ = spec->rsign; + + /* Only for type 'c' special case, it has no digits. */ + if (spec->n_digits != 0) { + /* Fill the digits with InsertThousandsGrouping. */ +#ifndef NDEBUG + r = +#endif + STRINGLIB_GROUPING(buf, spec->n_grouped_digits, digits, + spec->n_digits, spec->n_min_width, + locale->grouping, locale->thousands_sep); +#ifndef NDEBUG + assert(r == spec->n_grouped_digits); +#endif + p += spec->n_digits; + } + if (toupper) { + Py_ssize_t t; + for (t = 0; t < spec->n_grouped_digits; ++t) + buf[t] = STRINGLIB_TOUPPER(buf[t]); + } + buf += spec->n_grouped_digits; + + if (spec->n_decimal) { + Py_ssize_t t; + for (t = 0; t < spec->n_decimal; ++t) + buf[t] = locale->decimal_point[t]; + buf += spec->n_decimal; + p += 1; + } + + if (spec->n_remainder) { + memcpy(buf, p, spec->n_remainder * sizeof(STRINGLIB_CHAR)); + buf += spec->n_remainder; + p += spec->n_remainder; } + if (spec->n_rpadding) { - STRINGLIB_FILL(p_buf, fill_char, spec->n_rpadding); - p_buf += spec->n_rpadding; + STRINGLIB_FILL(buf, fill_char, spec->n_rpadding); + buf += spec->n_rpadding; + } +} + +static char no_grouping[1] = {CHAR_MAX}; + +/* Find the decimal point character(s?), thousands_separator(s?), and + grouping description, either for the current locale if type is + LT_CURRENT_LOCALE, a hard-coded locale if LT_DEFAULT_LOCALE, or + none if LT_NO_LOCALE. */ +static void +get_locale_info(int type, LocaleInfo *locale_info) +{ + switch (type) { + case LT_CURRENT_LOCALE: { + struct lconv *locale_data = localeconv(); + locale_info->decimal_point = locale_data->decimal_point; + locale_info->thousands_sep = locale_data->thousands_sep; + locale_info->grouping = locale_data->grouping; + break; + } + case LT_DEFAULT_LOCALE: + locale_info->decimal_point = "."; + locale_info->thousands_sep = ","; + locale_info->grouping = "\3"; /* Group every 3 characters, + trailing 0 means repeat + infinitely. */ + break; + case LT_NO_LOCALE: + locale_info->decimal_point = "."; + locale_info->thousands_sep = ""; + locale_info->grouping = no_grouping; + break; + default: + assert(0); } - return p_digits; } + #endif /* FORMAT_FLOAT || FORMAT_LONG */ /************************************************************************/ @@ -523,19 +680,21 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, PyObject *tmp = NULL; STRINGLIB_CHAR *pnumeric_chars; STRINGLIB_CHAR numeric_char; - STRINGLIB_CHAR sign = '\0'; - STRINGLIB_CHAR *p; + STRINGLIB_CHAR sign_char = '\0'; Py_ssize_t n_digits; /* count of digits need from the computed string */ - Py_ssize_t n_leading_chars; - Py_ssize_t n_grouping_chars = 0; /* Count of additional chars to - allocate, used for 'n' - formatting. */ + Py_ssize_t n_remainder = 0; /* Used only for 'c' formatting, which + produces non-digits */ Py_ssize_t n_prefix = 0; /* Count of prefix chars, (e.g., '0x') */ + Py_ssize_t n_total; STRINGLIB_CHAR *prefix = NULL; NumberFieldWidths spec; long x; + /* Locale settings, either from the actual locale or + from a hard-code pseudo-locale */ + LocaleInfo locale; + /* no precision allowed on integers */ if (format->precision != -1) { PyErr_SetString(PyExc_ValueError, @@ -543,7 +702,6 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, goto done; } - /* special case for character formatting */ if (format->type == 'c') { /* error to specify a sign */ @@ -554,6 +712,14 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, goto done; } + /* Error to specify a comma. */ + if (format->thousands_separators) { + PyErr_SetString(PyExc_ValueError, + "Thousands separators not allowed with integer" + " format specifier 'c'"); + goto done; + } + /* taken from unicodeobject.c formatchar() */ /* Integer input truncated to a character */ /* XXX: won't work for int */ @@ -578,6 +744,13 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, numeric_char = (STRINGLIB_CHAR)x; pnumeric_chars = &numeric_char; n_digits = 1; + + /* As a sort-of hack, we tell calc_number_widths that we only + have "remainder" characters. calc_number_widths thinks + these are characters that don't get formatted, only copied + into the output string. We do this for 'c' formatting, + because the characters are likely to be non-digits. */ + n_remainder = 1; } else { int base; @@ -629,8 +802,8 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, /* Is a sign character present in the output? If so, remember it and skip it */ - sign = pnumeric_chars[0]; - if (sign == '-') { + if (pnumeric_chars[0] == '-') { + sign_char = pnumeric_chars[0]; ++prefix; ++leading_chars_to_skip; } @@ -640,86 +813,26 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, pnumeric_chars += leading_chars_to_skip; } - if (format->type == 'n') - /* Compute how many additional chars we need to allocate - to hold the thousands grouping. */ - STRINGLIB_GROUPING_LOCALE(NULL, n_digits, n_digits, - 0, &n_grouping_chars, 0); - if (format->thousands_separators) - /* Compute how many additional chars we need to allocate - to hold the thousands grouping. */ - STRINGLIB_GROUPING(NULL, n_digits, n_digits, - 0, &n_grouping_chars, 0, "\3", ","); - - /* Calculate the widths of the various leading and trailing parts */ - calc_number_widths(&spec, sign, n_prefix, n_digits + n_grouping_chars, - format); - - /* Allocate a new string to hold the result */ - result = STRINGLIB_NEW(NULL, spec.n_total); - if (!result) - goto done; - p = STRINGLIB_STR(result); - - /* XXX There is too much magic here regarding the internals of - spec and the location of the prefix and digits. It would be - better if calc_number_widths returned a number of logical - offsets into the buffer, and those were used. Maybe in a - future code cleanup. */ - - /* Fill in the digit parts */ - n_leading_chars = spec.n_lpadding + spec.n_lsign + - spec.n_prefix + spec.n_spadding; - memmove(p + n_leading_chars, - pnumeric_chars, - n_digits * sizeof(STRINGLIB_CHAR)); - - /* If type is 'X', convert the filled in digits to uppercase */ - if (format->type == 'X') { - Py_ssize_t t; - for (t = 0; t < n_digits; ++t) - p[t + n_leading_chars] = STRINGLIB_TOUPPER(p[t + n_leading_chars]); - } + /* Determine the grouping, separator, and decimal point, if any. */ + get_locale_info(format->type == 'n' ? LT_CURRENT_LOCALE : + (format->thousands_separators ? + LT_DEFAULT_LOCALE : + LT_NO_LOCALE), + &locale); - /* Insert the grouping, if any, after the uppercasing of the digits, so - we can ensure that grouping chars won't be affected. */ - if (n_grouping_chars) { - /* We know this can't fail, since we've already - reserved enough space. */ - STRINGLIB_CHAR *pstart = p + n_leading_chars; -#ifndef NDEBUG - int r; -#endif - if (format->type == 'n') -#ifndef NDEBUG - r = -#endif - STRINGLIB_GROUPING_LOCALE(pstart, n_digits, n_digits, - spec.n_total+n_grouping_chars-n_leading_chars, - NULL, 0); - else -#ifndef NDEBUG - r = -#endif - STRINGLIB_GROUPING(pstart, n_digits, n_digits, - spec.n_total+n_grouping_chars-n_leading_chars, - NULL, 0, "\3", ","); - assert(r); - } + /* Calculate how much memory we'll need. */ + n_total = calc_number_widths(&spec, n_prefix, sign_char, pnumeric_chars, + n_digits, n_remainder, 0, &locale, format); - /* Fill in the non-digit parts (padding, sign, etc.) */ - fill_non_digits(p, &spec, prefix, n_digits + n_grouping_chars, - format->fill_char == '\0' ? ' ' : format->fill_char); - - /* If type is 'X', uppercase the prefix. This has to be done after the - prefix is filled in by fill_non_digits */ - if (format->type == 'X') { - Py_ssize_t t; - for (t = 0; t < n_prefix; ++t) - p[t + spec.n_lpadding + spec.n_lsign] = - STRINGLIB_TOUPPER(p[t + spec.n_lpadding + spec.n_lsign]); - } + /* Allocate the memory. */ + result = STRINGLIB_NEW(NULL, n_total); + if (!result) + goto done; + /* Populate the memory. */ + fill_number(STRINGLIB_STR(result), &spec, pnumeric_chars, n_digits, + prefix, format->fill_char == '\0' ? ' ' : format->fill_char, + &locale, format->type == 'X'); done: Py_XDECREF(tmp); @@ -733,64 +846,45 @@ done: #ifdef FORMAT_FLOAT #if STRINGLIB_IS_UNICODE -/* taken from unicodeobject.c */ -static Py_ssize_t -strtounicode(Py_UNICODE *buffer, const char *charbuffer) +static void +strtounicode(Py_UNICODE *buffer, const char *charbuffer, Py_ssize_t len) { - register Py_ssize_t i; - Py_ssize_t len = strlen(charbuffer); - for (i = len - 1; i >= 0; --i) - buffer[i] = (Py_UNICODE) charbuffer[i]; - - return len; + Py_ssize_t i; + for (i = 0; i < len; ++i) + buffer[i] = (Py_UNICODE)charbuffer[i]; } #endif -/* see FORMATBUFLEN in unicodeobject.c */ -#define FLOAT_FORMATBUFLEN 120 - /* much of this is taken from unicodeobject.c */ static PyObject * format_float_internal(PyObject *value, const InternalFormatSpec *format) { - /* fmt = '%.' + `prec` + `type` + '%%' - worst case length = 2 + 10 (len of INT_MAX) + 1 + 2 = 15 (use 20)*/ - char fmt[20]; - - /* taken from unicodeobject.c */ - /* Worst case length calc to ensure no buffer overrun: - - 'g' formats: - fmt = %#.<prec>g - buf = '-' + [0-9]*prec + '.' + 'e+' + (longest exp - for any double rep.) - len = 1 + prec + 1 + 2 + 5 = 9 + prec - - 'f' formats: - buf = '-' + [0-9]*x + '.' + [0-9]*prec (with x < 50) - len = 1 + 50 + 1 + prec = 52 + prec - - If prec=0 the effective precision is 1 (the leading digit is - always given), therefore increase the length by one. - - */ - char charbuf[FLOAT_FORMATBUFLEN]; + char *buf = NULL; /* buffer returned from PyOS_double_to_string */ Py_ssize_t n_digits; - double x; + Py_ssize_t n_remainder; + Py_ssize_t n_total; + int has_decimal; + double val; Py_ssize_t precision = format->precision; - PyObject *result = NULL; - STRINGLIB_CHAR sign; - char* trailing = ""; + STRINGLIB_CHAR type = format->type; + int add_pct = 0; STRINGLIB_CHAR *p; NumberFieldWidths spec; - STRINGLIB_CHAR type = format->type; + int flags = 0; + PyObject *result = NULL; + STRINGLIB_CHAR sign_char = '\0'; + int float_type; /* Used to see if we have a nan, inf, or regular float. */ #if STRINGLIB_IS_UNICODE - Py_UNICODE unicodebuf[FLOAT_FORMATBUFLEN]; + Py_UNICODE *unicode_tmp = NULL; #endif - /* alternate is not allowed on floats. */ + /* Locale settings, either from the actual locale or + from a hard-code pseudo-locale */ + LocaleInfo locale; + + /* Alternate is not allowed on floats. */ if (format->alternate) { PyErr_SetString(PyExc_ValueError, "Alternate form (#) not allowed in float format " @@ -798,84 +892,106 @@ format_float_internal(PyObject *value, goto done; } - /* first, do the conversion as 8-bit chars, using the platform's - snprintf. then, if needed, convert to unicode. */ + if (type == '\0') { + /* Omitted type specifier. This is like 'g' but with at least + one digit after the decimal point. */ + type = 'g'; + flags |= Py_DTSF_ADD_DOT_0; + } + + if (type == 'n') + /* 'n' is the same as 'g', except for the locale used to + format the result. We take care of that later. */ + type = 'g'; /* 'F' is the same as 'f', per the PEP */ if (type == 'F') type = 'f'; - x = PyFloat_AsDouble(value); - - if (x == -1.0 && PyErr_Occurred()) + val = PyFloat_AsDouble(value); + if (val == -1.0 && PyErr_Occurred()) goto done; if (type == '%') { type = 'f'; - x *= 100; - trailing = "%"; + val *= 100; + add_pct = 1; } if (precision < 0) precision = 6; - if (type == 'f' && fabs(x) >= 1e50) + if ((type == 'f' || type == 'F') && fabs(val) >= 1e50) type = 'g'; - /* cast "type", because if we're in unicode we need to pass a - 8-bit char. this is safe, because we've restricted what "type" - can be */ - PyOS_snprintf(fmt, sizeof(fmt), "%%.%" PY_FORMAT_SIZE_T "d%c", precision, - (char)type); - - /* do the actual formatting */ - PyOS_ascii_formatd(charbuf, sizeof(charbuf), fmt, x); - - /* adding trailing to fmt with PyOS_snprintf doesn't work, not - sure why. we'll just concatentate it here, no harm done. we - know we can't have a buffer overflow from the fmt size - analysis */ - strcat(charbuf, trailing); - - /* rather than duplicate the code for snprintf for both unicode - and 8 bit strings, we just use the 8 bit version and then - convert to unicode in a separate code path. that's probably - the lesser of 2 evils. */ + /* Cast "type", because if we're in unicode we need to pass a + 8-bit char. This is safe, because we've restricted what "type" + can be. */ + buf = PyOS_double_to_string(val, (char)type, precision, flags, + &float_type); + if (buf == NULL) + goto done; + n_digits = strlen(buf); + + if (add_pct) { + /* We know that buf has a trailing zero (since we just called + strlen() on it), and we don't use that fact any more. So we + can just write over the trailing zero. */ + buf[n_digits] = '%'; + n_digits += 1; + } + + /* Since there is no unicode version of PyOS_double_to_string, + just use the 8 bit version and then convert to unicode. */ #if STRINGLIB_IS_UNICODE - n_digits = strtounicode(unicodebuf, charbuf); - p = unicodebuf; + unicode_tmp = (Py_UNICODE*)PyMem_Malloc((n_digits)*sizeof(Py_UNICODE)); + if (unicode_tmp == NULL) { + PyErr_NoMemory(); + goto done; + } + strtounicode(unicode_tmp, buf, n_digits); + p = unicode_tmp; #else - /* compute the length. I believe this is done because the return - value from snprintf above is unreliable */ - n_digits = strlen(charbuf); - p = charbuf; + p = buf; #endif - /* is a sign character present in the output? if so, remember it + /* Is a sign character present in the output? If so, remember it and skip it */ - sign = p[0]; - if (sign == '-') { + if (*p == '-') { + sign_char = *p; ++p; --n_digits; } - calc_number_widths(&spec, sign, 0, n_digits, format); + /* Determine if we have any "remainder" (after the digits, might include + decimal or exponent or both (or neither)) */ + parse_number(p, n_digits, &n_remainder, &has_decimal); + + /* Determine the grouping, separator, and decimal point, if any. */ + get_locale_info(format->type == 'n' ? LT_CURRENT_LOCALE : + (format->thousands_separators ? + LT_DEFAULT_LOCALE : + LT_NO_LOCALE), + &locale); - /* allocate a string with enough space */ - result = STRINGLIB_NEW(NULL, spec.n_total); + /* Calculate how much memory we'll need. */ + n_total = calc_number_widths(&spec, 0, sign_char, p, n_digits, + n_remainder, has_decimal, &locale, format); + + /* Allocate the memory. */ + result = STRINGLIB_NEW(NULL, n_total); if (result == NULL) goto done; - /* Fill in the non-digit parts (padding, sign, etc.) */ - fill_non_digits(STRINGLIB_STR(result), &spec, NULL, n_digits, - format->fill_char == '\0' ? ' ' : format->fill_char); - - /* fill in the digit parts */ - memmove(STRINGLIB_STR(result) + - (spec.n_lpadding + spec.n_lsign + spec.n_spadding), - p, - n_digits * sizeof(STRINGLIB_CHAR)); + /* Populate the memory. */ + fill_number(STRINGLIB_STR(result), &spec, p, n_digits, NULL, + format->fill_char == '\0' ? ' ' : format->fill_char, &locale, + 0); done: + PyMem_Free(buf); +#if STRINGLIB_IS_UNICODE + PyMem_Free(unicode_tmp); +#endif return result; } #endif /* FORMAT_FLOAT */ @@ -1056,11 +1172,7 @@ FORMAT_FLOAT(PyObject *obj, /* type conversion? */ switch (format.type) { - case '\0': - /* 'Z' means like 'g', but with at least one decimal. See - PyOS_ascii_formatd */ - format.type = 'Z'; - /* Deliberate fall through to the next case statement */ + case '\0': /* No format code: like 'g', but with at least one decimal. */ case 'e': case 'E': case 'f': diff --git a/Objects/stringlib/localeutil.h b/Objects/stringlib/localeutil.h index 9254c09..f548133 100644 --- a/Objects/stringlib/localeutil.h +++ b/Objects/stringlib/localeutil.h @@ -5,161 +5,208 @@ #include <locale.h> +#define MAX(x, y) ((x) < (y) ? (y) : (x)) +#define MIN(x, y) ((x) < (y) ? (x) : (y)) + +typedef struct { + const char *grouping; + char previous; + Py_ssize_t i; /* Where we're currently pointing in grouping. */ +} GroupGenerator; + +static void +_GroupGenerator_init(GroupGenerator *self, const char *grouping) +{ + self->grouping = grouping; + self->i = 0; + self->previous = 0; +} + +/* Returns the next grouping, or 0 to signify end. */ +static Py_ssize_t +_GroupGenerator_next(GroupGenerator *self) +{ + /* Note that we don't really do much error checking here. If a + grouping string contains just CHAR_MAX, for example, then just + terminate the generator. That shouldn't happen, but at least we + fail gracefully. */ + switch (self->grouping[self->i]) { + case 0: + return self->previous; + case CHAR_MAX: + /* Stop the generator. */ + return 0; + default: { + char ch = self->grouping[self->i]; + self->previous = ch; + self->i++; + return (Py_ssize_t)ch; + } + } +} + +/* Fill in some digits, leading zeros, and thousands separator. All + are optional, depending on when we're called. */ +static void +fill(STRINGLIB_CHAR **digits_end, STRINGLIB_CHAR **buffer_end, + Py_ssize_t n_chars, Py_ssize_t n_zeros, const char* thousands_sep, + Py_ssize_t thousands_sep_len) +{ +#if STRINGLIB_IS_UNICODE + Py_ssize_t i; +#endif + + if (thousands_sep) { + *buffer_end -= thousands_sep_len; + + /* Copy the thousands_sep chars into the buffer. */ +#if STRINGLIB_IS_UNICODE + /* Convert from the char's of the thousands_sep from + the locale into unicode. */ + for (i = 0; i < thousands_sep_len; ++i) + (*buffer_end)[i] = thousands_sep[i]; +#else + /* No conversion, just memcpy the thousands_sep. */ + memcpy(*buffer_end, thousands_sep, thousands_sep_len); +#endif + } + + *buffer_end -= n_chars; + *digits_end -= n_chars; + memcpy(*buffer_end, *digits_end, n_chars * sizeof(STRINGLIB_CHAR)); + + *buffer_end -= n_zeros; + STRINGLIB_FILL(*buffer_end, '0', n_zeros); +} + /** * _Py_InsertThousandsGrouping: * @buffer: A pointer to the start of a string. - * @n_buffer: The length of the string. + * @n_buffer: Number of characters in @buffer. + * @digits: A pointer to the digits we're reading from. If count + * is non-NULL, this is unused. * @n_digits: The number of digits in the string, in which we want * to put the grouping chars. - * @buf_size: The maximum size of the buffer pointed to by buffer. - * @count: If non-NULL, points to a variable that will receive the - * number of characters we need to insert (and no formatting - * will actually occur). - * @append_zero_char: If non-zero, put a trailing zero at the end of - * of the resulting string, if and only if we modified the - * string. + * @min_width: The minimum width of the digits in the output string. + * Output will be zero-padded on the left to fill. * @grouping: see definition in localeconv(). * @thousands_sep: see definition in localeconv(). * + * There are 2 modes: counting and filling. If @buffer is NULL, + * we are in counting mode, else filling mode. + * If counting, the required buffer size is returned. + * If filling, we know the buffer will be large enough, so we don't + * need to pass in the buffer size. * Inserts thousand grouping characters (as defined by grouping and * thousands_sep) into the string between buffer and buffer+n_digits. - * If count is non-NULL, don't do any formatting, just count the - * number of characters to insert. This is used by the caller to - * appropriately resize the buffer, if needed. If count is non-NULL, - * buffer can be NULL (it is not dereferenced at all in that case). * * Return value: 0 on error, else 1. Note that no error can occur if * count is non-NULL. * * This name won't be used, the includer of this file should define * it to be the actual function name, based on unicode or string. + * + * As closely as possible, this code mimics the logic in decimal.py's + _insert_thousands_sep(). **/ -int +Py_ssize_t _Py_InsertThousandsGrouping(STRINGLIB_CHAR *buffer, Py_ssize_t n_buffer, + STRINGLIB_CHAR *digits, Py_ssize_t n_digits, - Py_ssize_t buf_size, - Py_ssize_t *count, - int append_zero_char, + Py_ssize_t min_width, const char *grouping, const char *thousands_sep) { - Py_ssize_t thousands_sep_len = strlen(thousands_sep); - STRINGLIB_CHAR *pend = NULL; /* current end of buffer */ - STRINGLIB_CHAR *pmax = NULL; /* max of buffer */ - char current_grouping; - Py_ssize_t remaining = n_digits; /* Number of chars remaining to - be looked at */ - - /* Initialize the character count, if we're just counting. */ - if (count) - *count = 0; - else { - /* We're not just counting, we're modifying buffer */ - pend = buffer + n_buffer; - pmax = buffer + buf_size; + Py_ssize_t count = 0; + Py_ssize_t n_zeros; + int loop_broken = 0; + int use_separator = 0; /* First time through, don't append the + separator. They only go between + groups. */ + STRINGLIB_CHAR *buffer_end = NULL; + STRINGLIB_CHAR *digits_end = NULL; + Py_ssize_t l; + Py_ssize_t n_chars; + Py_ssize_t thousands_sep_len = strlen(thousands_sep); + Py_ssize_t remaining = n_digits; /* Number of chars remaining to + be looked at */ + /* A generator that returns all of the grouping widths, until it + returns 0. */ + GroupGenerator groupgen; + _GroupGenerator_init(&groupgen, grouping); + + if (buffer) { + buffer_end = buffer + n_buffer; + digits_end = digits + n_digits; + } + + while ((l = _GroupGenerator_next(&groupgen)) > 0) { + l = MIN(l, MAX(MAX(remaining, min_width), 1)); + n_zeros = MAX(0, l - remaining); + n_chars = MAX(0, MIN(remaining, l)); + + /* Use n_zero zero's and n_chars chars */ + + /* Count only, don't do anything. */ + count += (use_separator ? thousands_sep_len : 0) + n_zeros + n_chars; + + if (buffer) { + /* Copy into the output buffer. */ + fill(&digits_end, &buffer_end, n_chars, n_zeros, + use_separator ? thousands_sep : NULL, thousands_sep_len); } - /* Starting at the end and working right-to-left, keep track of - what grouping needs to be added and insert that. */ - current_grouping = *grouping++; - - /* If the first character is 0, perform no grouping at all. */ - if (current_grouping == 0) - return 1; - - while (remaining > current_grouping) { - /* Always leave buffer and pend valid at the end of this - loop, since we might leave with a return statement. */ - - remaining -= current_grouping; - if (count) { - /* We're only counting, not touching the memory. */ - *count += thousands_sep_len; - } - else { - /* Do the formatting. */ - - STRINGLIB_CHAR *plast = buffer + remaining; - - /* Is there room to insert thousands_sep_len chars? */ - if (pmax - pend < thousands_sep_len) - /* No room. */ - return 0; - - /* Move the rest of the string down. */ - memmove(plast + thousands_sep_len, - plast, - (pend - plast) * sizeof(STRINGLIB_CHAR)); - /* Copy the thousands_sep chars into the buffer. */ -#if STRINGLIB_IS_UNICODE - /* Convert from the char's of the thousands_sep from - the locale into unicode. */ - { - Py_ssize_t i; - for (i = 0; i < thousands_sep_len; ++i) - plast[i] = thousands_sep[i]; - } -#else - /* No conversion, just memcpy the thousands_sep. */ - memcpy(plast, thousands_sep, thousands_sep_len); -#endif - } - - /* Adjust end pointer. */ - pend += thousands_sep_len; - - /* Move to the next grouping character, unless we're - repeating (which is designated by a grouping of 0). */ - if (*grouping != 0) { - current_grouping = *grouping++; - if (current_grouping == CHAR_MAX) - /* We're done. */ - break; - } + /* Use a separator next time. */ + use_separator = 1; + + remaining -= n_chars; + min_width -= l; + + if (remaining <= 0 && min_width <= 0) { + loop_broken = 1; + break; } - if (append_zero_char) { - /* Append a zero character to mark the end of the string, - if there's room. */ - if (pend - (buffer + remaining) < 1) - /* No room, error. */ - return 0; - *pend = 0; + min_width -= thousands_sep_len; + } + if (!loop_broken) { + /* We left the loop without using a break statement. */ + + l = MAX(MAX(remaining, min_width), 1); + n_zeros = MAX(0, l - remaining); + n_chars = MAX(0, MIN(remaining, l)); + + /* Use n_zero zero's and n_chars chars */ + count += (use_separator ? thousands_sep_len : 0) + n_zeros + n_chars; + if (buffer) { + /* Copy into the output buffer. */ + fill(&digits_end, &buffer_end, n_chars, n_zeros, + use_separator ? thousands_sep : NULL, thousands_sep_len); } - return 1; + } + return count; } /** * _Py_InsertThousandsGroupingLocale: * @buffer: A pointer to the start of a string. - * @n_buffer: The length of the string. * @n_digits: The number of digits in the string, in which we want * to put the grouping chars. - * @buf_size: The maximum size of the buffer pointed to by buffer. - * @count: If non-NULL, points to a variable that will receive the - * number of characters we need to insert (and no formatting - * will actually occur). - * @append_zero_char: If non-zero, put a trailing zero at the end of - * of the resulting string, if and only if we modified the - * string. * * Reads thee current locale and calls _Py_InsertThousandsGrouping(). **/ -int +Py_ssize_t _Py_InsertThousandsGroupingLocale(STRINGLIB_CHAR *buffer, Py_ssize_t n_buffer, + STRINGLIB_CHAR *digits, Py_ssize_t n_digits, - Py_ssize_t buf_size, - Py_ssize_t *count, - int append_zero_char) + Py_ssize_t min_width) { struct lconv *locale_data = localeconv(); const char *grouping = locale_data->grouping; const char *thousands_sep = locale_data->thousands_sep; - return _Py_InsertThousandsGrouping(buffer, n_buffer, n_digits, - buf_size, count, - append_zero_char, grouping, - thousands_sep); + return _Py_InsertThousandsGrouping(buffer, n_buffer, digits, n_digits, + min_width, grouping, thousands_sep); } #endif /* STRINGLIB_LOCALEUTIL_H */ diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c index f52c435..3cea899 100644 --- a/Objects/unicodeobject.c +++ b/Objects/unicodeobject.c @@ -8792,42 +8792,13 @@ getnextarg(PyObject *args, Py_ssize_t arglen, Py_ssize_t *p_argidx) return NULL; } -static Py_ssize_t -strtounicode(Py_UNICODE *buffer, const char *charbuffer) +static void +strtounicode(Py_UNICODE *buffer, const char *charbuffer, Py_ssize_t len) { register Py_ssize_t i; - Py_ssize_t len = strlen(charbuffer); for (i = len - 1; i >= 0; i--) buffer[i] = (Py_UNICODE) charbuffer[i]; - - return len; -} - -static int -doubletounicode(Py_UNICODE *buffer, size_t len, const char *format, double x) -{ - Py_ssize_t result; - - PyOS_ascii_formatd((char *)buffer, len, format, x); - result = strtounicode(buffer, (char *)buffer); - return Py_SAFE_DOWNCAST(result, Py_ssize_t, int); -} - -#if 0 -static int -longtounicode(Py_UNICODE *buffer, size_t len, const char *format, long x) -{ - Py_ssize_t result; - - PyOS_snprintf((char *)buffer, len, format, x); - result = strtounicode(buffer, (char *)buffer); - return Py_SAFE_DOWNCAST(result, Py_ssize_t, int); } -#endif - -/* XXX To save some code duplication, formatfloat/long/int could have been - shared with stringobject.c, converting from 8-bit to Unicode after the - formatting is done. */ static int formatfloat(Py_UNICODE *buf, @@ -8837,54 +8808,59 @@ formatfloat(Py_UNICODE *buf, int type, PyObject *v) { - /* fmt = '%#.' + `prec` + `type` - worst case length = 3 + 10 (len of INT_MAX) + 1 = 14 (use 20)*/ - char fmt[20]; + /* eric.smith: To minimize disturbances in PyUnicode_Format (the + only caller of this routine), I'm going to keep the existing + API to this function. That means that we'll allocate memory and + then copy back into the supplied buffer. But that's better than + all of the changes that would be required in PyUnicode_Format + because it does lots of memory management tricks. */ + + char* p = NULL; + int result = -1; double x; + Py_ssize_t len; x = PyFloat_AsDouble(v); if (x == -1.0 && PyErr_Occurred()) - return -1; + goto done; if (prec < 0) prec = 6; + /* make sure that the decimal representation of precision really does need at most 10 digits: platforms with sizeof(int) == 8 exist! */ if (prec > 0x7fffffffL) { PyErr_SetString(PyExc_OverflowError, "outrageously large precision " "for formatted float"); - return -1; + goto done; } if (type == 'f' && fabs(x) >= 1e50) type = 'g'; - /* Worst case length calc to ensure no buffer overrun: - - 'g' formats: - fmt = %#.<prec>g - buf = '-' + [0-9]*prec + '.' + 'e+' + (longest exp - for any double rep.) - len = 1 + prec + 1 + 2 + 5 = 9 + prec - - 'f' formats: - buf = '-' + [0-9]*x + '.' + [0-9]*prec (with x < 50) - len = 1 + 50 + 1 + prec = 52 + prec - If prec=0 the effective precision is 1 (the leading digit is - always given), therefore increase the length by one. - - */ if (((type == 'g' || type == 'G') && buflen <= (size_t)10 + (size_t)prec) || - (type == 'f' && buflen <= (size_t)53 + (size_t)prec)) { + ((type == 'f' || type == 'F') && + buflen <= (size_t)53 + (size_t)prec)) { PyErr_SetString(PyExc_OverflowError, "formatted float is too long (precision too large?)"); - return -1; + goto done; + } + + p = PyOS_double_to_string(x, type, prec, + (flags & F_ALT) ? Py_DTSF_ALT : 0, NULL); + len = strlen(p); + if (len+1 >= buflen) { + /* Caller supplied buffer is not large enough. */ + PyErr_NoMemory(); + goto done; } - PyOS_snprintf(fmt, sizeof(fmt), "%%%s.%d%c", - (flags&F_ALT) ? "#" : "", - prec, type); - return doubletounicode(buf, buflen, fmt, x); + strtounicode(buf, p, len); + result = Py_SAFE_DOWNCAST(len, Py_ssize_t, int); + +done: + PyMem_Free(p); + return result; } static PyObject* @@ -8903,84 +8879,6 @@ formatlong(PyObject *val, int flags, int prec, int type) return result; } -#if 0 -static int -formatint(Py_UNICODE *buf, - size_t buflen, - int flags, - int prec, - int type, - PyObject *v) -{ - /* fmt = '%#.' + `prec` + 'l' + `type` - * worst case length = 3 + 19 (worst len of INT_MAX on 64-bit machine) - * + 1 + 1 - * = 24 - */ - char fmt[64]; /* plenty big enough! */ - char *sign; - long x; - - x = PyLong_AsLong(v); - if (x == -1 && PyErr_Occurred()) - return -1; - if (x < 0 && type == 'u') { - type = 'd'; - } - if (x < 0 && (type == 'x' || type == 'X' || type == 'o')) - sign = "-"; - else - sign = ""; - if (prec < 0) - prec = 1; - - /* buf = '+'/'-'/'' + '0'/'0x'/'' + '[0-9]'*max(prec, len(x in octal)) - * worst case buf = '-0x' + [0-9]*prec, where prec >= 11 - */ - if (buflen <= 14 || buflen <= (size_t)3 + (size_t)prec) { - PyErr_SetString(PyExc_OverflowError, - "formatted integer is too long (precision too large?)"); - return -1; - } - - if ((flags & F_ALT) && - (type == 'x' || type == 'X' || type == 'o')) { - /* When converting under %#o, %#x or %#X, there are a number - * of issues that cause pain: - * - for %#o, we want a different base marker than C - * - when 0 is being converted, the C standard leaves off - * the '0x' or '0X', which is inconsistent with other - * %#x/%#X conversions and inconsistent with Python's - * hex() function - * - there are platforms that violate the standard and - * convert 0 with the '0x' or '0X' - * (Metrowerks, Compaq Tru64) - * - there are platforms that give '0x' when converting - * under %#X, but convert 0 in accordance with the - * standard (OS/2 EMX) - * - * We can achieve the desired consistency by inserting our - * own '0x' or '0X' prefix, and substituting %x/%X in place - * of %#x/%#X. - * - * Note that this is the same approach as used in - * formatint() in stringobject.c - */ - PyOS_snprintf(fmt, sizeof(fmt), "%s0%c%%.%dl%c", - sign, type, prec, type); - } - else { - PyOS_snprintf(fmt, sizeof(fmt), "%s%%%s.%dl%c", - sign, (flags&F_ALT) ? "#" : "", - prec, type); - } - if (sign[0]) - return longtounicode(buf, buflen, fmt, -x); - else - return longtounicode(buf, buflen, fmt, x); -} -#endif - static int formatchar(Py_UNICODE *buf, size_t buflen, @@ -9359,8 +9257,6 @@ PyObject *PyUnicode_Format(PyObject *format, case 'F': case 'g': case 'G': - if (c == 'F') - c = 'f'; pbuf = formatbuf; len = formatfloat(pbuf, sizeof(formatbuf)/sizeof(Py_UNICODE), flags, prec, c, v); |