From aca19e6a740c424aec243a4721b18d12e9129aa7 Mon Sep 17 00:00:00 2001 From: Eric Smith Date: Wed, 22 Apr 2009 13:29:05 +0000 Subject: [PATCH] Backport of some of the work in r71665 to trunk. This reworks much of int, long, and float __format__(), and it keeps their implementation in sync with py3k. Also added PyOS_double_to_string. This is the "fallback" version that's also available in trunk, and should be kept in sync with that code. I'll add an issue to document PyOS_double_to_string in the C API. There are many internal cleanups. Externally visible changes include: - Implement PEP 378, Format Specifier for Thousands Separator, for floats, ints, and longs. - Issue #5515: 'n' formatting for ints, longs, and floats handles leading zero formatting poorly. - Issue #5772: For float.__format__, don't add a trailing ".0" if we're using no type code and we have an exponent. --- Include/pystrtod.h | 19 + Include/stringobject.h | 22 +- Lib/test/test_format.py | 4 + Lib/test/test_types.py | 54 ++- Misc/NEWS | 9 + Objects/stringlib/formatter.h | 807 +++++++++++++++++++-------------- Objects/stringlib/localeutil.h | 274 +++++++---- Objects/stringlib/stringdefs.h | 14 +- Python/pystrtod.c | 187 +++++--- 9 files changed, 894 insertions(+), 496 deletions(-) diff --git a/Include/pystrtod.h b/Include/pystrtod.h index c6921da1a5f..1caa7aedfb7 100644 --- a/Include/pystrtod.h +++ b/Include/pystrtod.h @@ -10,6 +10,25 @@ PyAPI_FUNC(double) PyOS_ascii_strtod(const char *str, char **ptr); PyAPI_FUNC(double) PyOS_ascii_atof(const char *str); PyAPI_FUNC(char *) PyOS_ascii_formatd(char *buffer, size_t buf_len, const char *format, double d); +/* The caller is responsible for calling PyMem_Free to free the buffer + that's is returned. */ +PyAPI_FUNC(char *) PyOS_double_to_string(double val, + char format_code, + int precision, + int flags, + int *type); + + +/* PyOS_double_to_string's "flags" parameter can be set to 0 or more of: */ +#define Py_DTSF_SIGN 0x01 /* always add the sign */ +#define Py_DTSF_ADD_DOT_0 0x02 /* if the result is an integer add ".0" */ +#define Py_DTSF_ALT 0x04 /* "alternate" formatting. it's format_code + specific */ + +/* PyOS_double_to_string's "type", if non-NULL, will be set to one of: */ +#define Py_DTST_FINITE 0 +#define Py_DTST_INFINITE 1 +#define Py_DTST_NAN 2 #ifdef __cplusplus } diff --git a/Include/stringobject.h b/Include/stringobject.h index d9f184072f2..18b5b411aca 100644 --- a/Include/stringobject.h +++ b/Include/stringobject.h @@ -177,16 +177,26 @@ PyAPI_FUNC(int) PyString_AsStringAndSize( strings) */ ); + /* Using the current locale, insert the thousands grouping into the string pointed to by buffer. For the argument descriptions, see Objects/stringlib/localeutil.h */ +PyAPI_FUNC(Py_ssize_t) _PyString_InsertThousandsGroupingLocale(char *buffer, + Py_ssize_t n_buffer, + char *digits, + Py_ssize_t n_digits, + Py_ssize_t min_width); -PyAPI_FUNC(int) _PyString_InsertThousandsGrouping(char *buffer, - Py_ssize_t n_buffer, - Py_ssize_t n_digits, - Py_ssize_t buf_size, - Py_ssize_t *count, - int append_zero_char); +/* Using explicit passed-in values, insert the thousands grouping + into the string pointed to by buffer. For the argument descriptions, + see Objects/stringlib/localeutil.h */ +PyAPI_FUNC(Py_ssize_t) _PyString_InsertThousandsGrouping(char *buffer, + Py_ssize_t n_buffer, + char *digits, + Py_ssize_t n_digits, + Py_ssize_t min_width, + const char *grouping, + const char *thousands_sep); /* Format the object based on the format_spec, as defined in PEP 3101 (Advanced String Formatting). */ diff --git a/Lib/test/test_format.py b/Lib/test/test_format.py index cd46bc21c79..06179de5f4d 100644 --- a/Lib/test/test_format.py +++ b/Lib/test/test_format.py @@ -232,6 +232,10 @@ class FormatTest(unittest.TestCase): testboth("%o", -042L, "-42") testboth("%o", float(042), "42") + # alternate float formatting + testformat('%g', 1.1, '1.1') + testformat('%#g', 1.1, '1.10000') + # Test exception for unknown format characters if verbose: print 'Testing exceptions' diff --git a/Lib/test/test_types.py b/Lib/test/test_types.py index 10ad6346c4e..6fff22a5f0e 100644 --- a/Lib/test/test_types.py +++ b/Lib/test/test_types.py @@ -113,6 +113,9 @@ class TypesTests(unittest.TestCase): self.assertEqual(1.5e-101.__format__('e'), '1.500000e-101') self.assertEqual('%e' % 1.5e-101, '1.500000e-101') + self.assertEqual('%g' % 1.0, '1') + self.assertEqual('%#g' % 1.0, '1.00000') + def test_normal_integers(self): # Ensure the first 256 integers are shared a = 256 @@ -412,6 +415,9 @@ class TypesTests(unittest.TestCase): self.assertRaises(TypeError, 3 .__format__, None) self.assertRaises(TypeError, 3 .__format__, 0) + # can't have ',' with 'c' + self.assertRaises(ValueError, 3 .__format__, ",c") + # ensure that only int and float type specifiers work for format_spec in ([chr(x) for x in range(ord('a'), ord('z')+1)] + [chr(x) for x in range(ord('A'), ord('Z')+1)]): @@ -609,11 +615,37 @@ class TypesTests(unittest.TestCase): # a totaly empty format specifier means something else. # So, just use a sign flag test(1e200, '+g', '+1e+200') - test(1e200, '+', '+1.0e+200') + test(1e200, '+', '+1e+200') test(1.1e200, '+g', '+1.1e+200') test(1.1e200, '+', '+1.1e+200') - # % formatting + test(1.1e200, '+g', '+1.1e+200') + test(1.1e200, '+', '+1.1e+200') + + # 0 padding + test(1234., '010f', '1234.000000') + test(1234., '011f', '1234.000000') + test(1234., '012f', '01234.000000') + test(-1234., '011f', '-1234.000000') + test(-1234., '012f', '-1234.000000') + test(-1234., '013f', '-01234.000000') + test(-1234.12341234, '013f', '-01234.123412') + test(-123456.12341234, '011.2f', '-0123456.12') + + # 0 padding with commas + test(1234., '011,f', '1,234.000000') + test(1234., '012,f', '1,234.000000') + test(1234., '013,f', '01,234.000000') + test(-1234., '012,f', '-1,234.000000') + test(-1234., '013,f', '-1,234.000000') + test(-1234., '014,f', '-01,234.000000') + test(-12345., '015,f', '-012,345.000000') + test(-123456., '016,f', '-0,123,456.000000') + test(-123456., '017,f', '-0,123,456.000000') + test(-123456.12341234, '017,f', '-0,123,456.123412') + test(-123456.12341234, '013,.2f', '-0,123,456.12') + + # % formatting test(-1.0, '%', '-100.000000%') # format spec must be string @@ -637,6 +669,24 @@ class TypesTests(unittest.TestCase): self.assertRaises(ValueError, format, 0.0, '#') self.assertRaises(ValueError, format, 0.0, '#20f') + def test_format_spec_errors(self): + # int, float, and string all share the same format spec + # mini-language parser. + + # Check that we can't ask for too many digits. This is + # probably a CPython specific test. It tries to put the width + # into a C long. + self.assertRaises(ValueError, format, 0, '1'*10000 + 'd') + + # Similar with the precision. + self.assertRaises(ValueError, format, 0, '.' + '1'*10000 + 'd') + + # And may as well test both. + self.assertRaises(ValueError, format, 0, '1'*1000 + '.' + '1'*10000 + 'd') + + # Make sure commas aren't allowed with various type codes + for code in 'xXobns': + self.assertRaises(ValueError, format, 0, ',' + code) def test_main(): run_unittest(TypesTests) diff --git a/Misc/NEWS b/Misc/NEWS index d4dc2dcc61e..74af1470974 100644 --- a/Misc/NEWS +++ b/Misc/NEWS @@ -12,6 +12,15 @@ What's New in Python 2.7 alpha 1 Core and Builtins ----------------- +- Implement PEP 378, Format Specifier for Thousands Separator, for + floats, ints, and longs. + +- Issue #5515: 'n' formatting for ints, longs, and floats handles + leading zero formatting poorly. + +- Issue #5772: For float.__format__, don't add a trailing ".0" if + we're using no type code and we have an exponent. + - Issue #3166: Make long -> float (and int -> float) conversions correctly rounded. diff --git a/Objects/stringlib/formatter.h b/Objects/stringlib/formatter.h index df291584e79..61ca12badf8 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 + /* 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; @@ -120,6 +117,7 @@ typedef struct { int alternate; STRINGLIB_CHAR sign; Py_ssize_t width; + int thousands_separators; Py_ssize_t precision; STRINGLIB_CHAR type; } InternalFormatSpec; @@ -132,7 +130,7 @@ typedef struct { */ static int parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len, + Py_ssize_t format_spec_len, InternalFormatSpec *format, char default_type) { @@ -142,13 +140,14 @@ 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'; format->alternate = 0; format->sign = '\0'; format->width = -1; + format->thousands_separators = 0; format->precision = -1; format->type = default_type; @@ -168,18 +167,13 @@ 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 applies to integers. */ if (end-ptr >= 1 && ptr[0] == '#') { - format->alternate = 1; - ++ptr; + format->alternate = 1; + ++ptr; } /* The special case for 0-padding (backwards compat) */ @@ -191,25 +185,35 @@ 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] == ',') { + format->thousands_separators = 1; + ++ptr; } /* Parse field precision */ 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; @@ -217,10 +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; } @@ -230,6 +234,29 @@ parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec, ++ptr; } + /* 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; } @@ -238,6 +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 { @@ -245,38 +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 (ptrn_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: - | | - | | - | | + | | + | | + | | - 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) @@ -291,108 +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, - prefix, - spec->n_prefix * sizeof(STRINGLIB_CHAR)); - p_buf += spec->n_prefix; + memmove(buf, + prefix, + spec->n_prefix * sizeof(STRINGLIB_CHAR)); + 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; } - return p_digits; } + +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); + } +} + #endif /* FORMAT_FLOAT || FORMAT_LONG */ /************************************************************************/ @@ -420,7 +590,7 @@ format_string_internal(PyObject *value, const InternalFormatSpec *format) if (format->alternate) { PyErr_SetString(PyExc_ValueError, "Alternate form (#) not allowed in string format " - "specifier"); + "specifier"); goto done; } @@ -504,25 +674,27 @@ typedef PyObject* static PyObject * format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, - IntOrLongToString tostring) + IntOrLongToString tostring) { PyObject *result = NULL; 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, @@ -530,7 +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 */ @@ -541,6 +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 */ @@ -562,31 +741,38 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, goto done; } #endif - numeric_char = (STRINGLIB_CHAR)x; - pnumeric_chars = &numeric_char; + 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; - int leading_chars_to_skip = 0; /* Number of characters added by - PyNumber_ToBase that we want to - skip over. */ + int leading_chars_to_skip = 0; /* Number of characters added by + PyNumber_ToBase that we want to + skip over. */ /* Compute the base and how many characters will be added by PyNumber_ToBase */ switch (format->type) { case 'b': base = 2; - leading_chars_to_skip = 2; /* 0b */ + leading_chars_to_skip = 2; /* 0b */ break; case 'o': base = 8; - leading_chars_to_skip = 2; /* 0o */ + leading_chars_to_skip = 2; /* 0o */ break; case 'x': case 'X': base = 16; - leading_chars_to_skip = 2; /* 0x */ + leading_chars_to_skip = 2; /* 0x */ break; default: /* shouldn't be needed, but stops a compiler warning */ case 'd': @@ -595,102 +781,58 @@ format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format, break; } - /* The number of prefix chars is the same as the leading - chars to skip */ - if (format->alternate) - n_prefix = leading_chars_to_skip; + /* The number of prefix chars is the same as the leading + chars to skip */ + if (format->alternate) + n_prefix = leading_chars_to_skip; /* Do the hard part, converting to a string in a given base */ - tmp = tostring(value, base); + tmp = tostring(value, base); if (tmp == NULL) goto done; - pnumeric_chars = STRINGLIB_STR(tmp); + pnumeric_chars = STRINGLIB_STR(tmp); n_digits = STRINGLIB_LEN(tmp); - prefix = pnumeric_chars; + prefix = pnumeric_chars; - /* Remember not to modify what pnumeric_chars points to. it - might be interned. Only modify it after we copy it into a - newly allocated output buffer. */ + /* Remember not to modify what pnumeric_chars points to. it + might be interned. Only modify it after we copy it into a + newly allocated output buffer. */ /* Is a sign character present in the output? If so, remember it and skip it */ - sign = pnumeric_chars[0]; - if (sign == '-') { - ++prefix; - ++leading_chars_to_skip; + if (pnumeric_chars[0] == '-') { + sign_char = pnumeric_chars[0]; + ++prefix; + ++leading_chars_to_skip; } - /* Skip over the leading chars (0x, 0b, etc.) */ - n_digits -= leading_chars_to_skip; - pnumeric_chars += leading_chars_to_skip; + /* Skip over the leading chars (0x, 0b, etc.) */ + n_digits -= leading_chars_to_skip; + 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(NULL, n_digits, n_digits, - 0, &n_grouping_chars, 0); + /* 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); - /* Calculate the widths of the various leading and trailing parts */ - calc_number_widths(&spec, sign, n_prefix, n_digits + n_grouping_chars, - format); + /* 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); - /* Allocate a new string to hold the result */ - result = STRINGLIB_NEW(NULL, spec.n_total); + /* Allocate the memory. */ + result = STRINGLIB_NEW(NULL, 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]); - } - - /* 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 - STRINGLIB_GROUPING(pstart, n_digits, n_digits, - spec.n_total+n_grouping_chars-n_leading_chars, - NULL, 0); - assert(r); - } - - /* 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]); - } + 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); @@ -704,149 +846,152 @@ 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) + 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 = %#.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 " - "specifier"); + "specifier"); 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); + /* 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); - /* do the actual formatting */ - PyOS_ascii_formatd(charbuf, sizeof(charbuf), fmt, x); + 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; + } - /* 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. */ + /* 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); - /* allocate a string with enough space */ - result = STRINGLIB_NEW(NULL, spec.n_total); + /* 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); + + /* 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 */ @@ -856,8 +1001,8 @@ done: /************************************************************************/ PyObject * FORMAT_STRING(PyObject *obj, - STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len) + STRINGLIB_CHAR *format_spec, + Py_ssize_t format_spec_len) { InternalFormatSpec format; PyObject *result = NULL; @@ -871,7 +1016,7 @@ FORMAT_STRING(PyObject *obj, /* parse the format_spec */ if (!parse_internal_render_format_spec(format_spec, format_spec_len, - &format, 's')) + &format, 's')) goto done; /* type conversion? */ @@ -893,9 +1038,9 @@ done: #if defined FORMAT_LONG || defined FORMAT_INT static PyObject* format_int_or_long(PyObject* obj, - STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len, - IntOrLongToString tostring) + STRINGLIB_CHAR *format_spec, + Py_ssize_t format_spec_len, + IntOrLongToString tostring) { PyObject *result = NULL; PyObject *tmp = NULL; @@ -910,8 +1055,8 @@ format_int_or_long(PyObject* obj, /* parse the format_spec */ if (!parse_internal_render_format_spec(format_spec, - format_spec_len, - &format, 'd')) + format_spec_len, + &format, 'd')) goto done; /* type conversion? */ @@ -924,8 +1069,8 @@ format_int_or_long(PyObject* obj, case 'X': case 'n': /* no type conversion needed, already an int (or long). do - the formatting */ - result = format_int_or_long_internal(obj, &format, tostring); + the formatting */ + result = format_int_or_long_internal(obj, &format, tostring); break; case 'e': @@ -974,11 +1119,11 @@ long_format(PyObject* value, int base) PyObject * FORMAT_LONG(PyObject *obj, - STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len) + STRINGLIB_CHAR *format_spec, + Py_ssize_t format_spec_len) { return format_int_or_long(obj, format_spec, format_spec_len, - long_format); + long_format); } #endif /* FORMAT_LONG */ @@ -995,19 +1140,19 @@ int_format(PyObject* value, int base) PyObject * FORMAT_INT(PyObject *obj, - STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len) + STRINGLIB_CHAR *format_spec, + Py_ssize_t format_spec_len) { return format_int_or_long(obj, format_spec, format_spec_len, - int_format); + int_format); } #endif /* FORMAT_INT */ #ifdef FORMAT_FLOAT PyObject * FORMAT_FLOAT(PyObject *obj, - STRINGLIB_CHAR *format_spec, - Py_ssize_t format_spec_len) + STRINGLIB_CHAR *format_spec, + Py_ssize_t format_spec_len) { PyObject *result = NULL; InternalFormatSpec format; @@ -1021,17 +1166,13 @@ FORMAT_FLOAT(PyObject *obj, /* parse the format_spec */ if (!parse_internal_render_format_spec(format_spec, - format_spec_len, - &format, '\0')) + format_spec_len, + &format, '\0')) goto done; /* 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 1105609ff32..f5481338757 100644 --- a/Objects/stringlib/localeutil.h +++ b/Objects/stringlib/localeutil.h @@ -5,126 +5,208 @@ #include +#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(). * - * Inserts thousand grouping characters (as defined in the current - * locale) 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). + * 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. * * 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, - Py_ssize_t n_digits, - Py_ssize_t buf_size, - Py_ssize_t *count, - int append_zero_char) + Py_ssize_t n_buffer, + STRINGLIB_CHAR *digits, + Py_ssize_t n_digits, + Py_ssize_t min_width, + const char *grouping, + const char *thousands_sep) { - struct lconv *locale_data = localeconv(); - const char *grouping = locale_data->grouping; - const char *thousands_sep = locale_data->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 */ + 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); - /* 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; - } + if (buffer) { + buffer_end = buffer + n_buffer; + digits_end = digits + n_digits; + } - /* Starting at the end and working right-to-left, keep track of - what grouping needs to be added and insert that. */ - current_grouping = *grouping++; + 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)); - /* If the first character is 0, perform no grouping at all. */ - if (current_grouping == 0) - return 1; + /* Use n_zero zero's and n_chars chars */ - while (remaining > current_grouping) { - /* Always leave buffer and pend valid at the end of this - loop, since we might leave with a return statement. */ + /* Count only, don't do anything. */ + count += (use_separator ? thousands_sep_len : 0) + n_zeros + n_chars; - remaining -= current_grouping; - if (count) { - /* We're only counting, not touching the memory. */ - *count += thousands_sep_len; - } - else { - /* Do the formatting. */ + if (buffer) { + /* Copy into the output buffer. */ + fill(&digits_end, &buffer_end, n_chars, n_zeros, + use_separator ? thousands_sep : NULL, thousands_sep_len); + } - STRINGLIB_CHAR *plast = buffer + remaining; + /* Use a separator next time. */ + use_separator = 1; - /* Is there room to insert thousands_sep_len chars? */ - if (pmax - pend < thousands_sep_len) - /* No room. */ - return 0; + remaining -= n_chars; + min_width -= l; - /* 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 - } + if (remaining <= 0 && min_width <= 0) { + loop_broken = 1; + break; + } + min_width -= thousands_sep_len; + } + if (!loop_broken) { + /* We left the loop without using a break statement. */ - /* Adjust end pointer. */ - pend += thousands_sep_len; + l = MAX(MAX(remaining, min_width), 1); + n_zeros = MAX(0, l - remaining); + n_chars = MAX(0, MIN(remaining, l)); - /* 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; - } - } - 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; - } - return 1; + /* 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 count; +} + +/** + * _Py_InsertThousandsGroupingLocale: + * @buffer: A pointer to the start of a string. + * @n_digits: The number of digits in the string, in which we want + * to put the grouping chars. + * + * Reads thee current locale and calls _Py_InsertThousandsGrouping(). + **/ +Py_ssize_t +_Py_InsertThousandsGroupingLocale(STRINGLIB_CHAR *buffer, + Py_ssize_t n_buffer, + STRINGLIB_CHAR *digits, + Py_ssize_t n_digits, + 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, digits, n_digits, + min_width, grouping, thousands_sep); } #endif /* STRINGLIB_LOCALEUTIL_H */ diff --git a/Objects/stringlib/stringdefs.h b/Objects/stringlib/stringdefs.h index daaa2e2b0f7..d08971a7280 100644 --- a/Objects/stringlib/stringdefs.h +++ b/Objects/stringlib/stringdefs.h @@ -6,6 +6,15 @@ compiled as unicode. */ #define STRINGLIB_IS_UNICODE 0 +/* _tolower and _toupper are defined by SUSv2, but they're not ISO C */ +/* This needs to be cleaned up. See issue 5793. */ +#ifndef _tolower +#define _tolower tolower +#endif +#ifndef _toupper +#define _toupper toupper +#endif + #define STRINGLIB_OBJECT PyStringObject #define STRINGLIB_CHAR char #define STRINGLIB_TYPE_NAME "string" @@ -13,8 +22,8 @@ #define STRINGLIB_EMPTY nullstring #define STRINGLIB_ISDECIMAL(x) ((x >= '0') && (x <= '9')) #define STRINGLIB_TODECIMAL(x) (STRINGLIB_ISDECIMAL(x) ? (x - '0') : -1) -#define STRINGLIB_TOUPPER toupper -#define STRINGLIB_TOLOWER tolower +#define STRINGLIB_TOUPPER(x) _toupper(Py_CHARMASK(x)) +#define STRINGLIB_TOLOWER(x) _tolower(Py_CHARMASK(x)) #define STRINGLIB_FILL memset #define STRINGLIB_STR PyString_AS_STRING #define STRINGLIB_LEN PyString_GET_SIZE @@ -24,5 +33,6 @@ #define STRINGLIB_CMP memcmp #define STRINGLIB_TOSTR PyObject_Str #define STRINGLIB_GROUPING _PyString_InsertThousandsGrouping +#define STRINGLIB_GROUPING_LOCALE _PyString_InsertThousandsGroupingLocale #endif /* !STRINGLIB_STRINGDEFS_H */ diff --git a/Python/pystrtod.c b/Python/pystrtod.c index 302e01298a4..ce2e3825a77 100644 --- a/Python/pystrtod.c +++ b/Python/pystrtod.c @@ -37,6 +37,15 @@ * * Return value: the #gdouble value. **/ + +/* + 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. +*/ + double PyOS_ascii_strtod(const char *nptr, char **endptr) { @@ -187,6 +196,13 @@ PyOS_ascii_strtod(const char *nptr, char **endptr) return val; } +double +PyOS_ascii_atof(const char *nptr) +{ + return PyOS_ascii_strtod(nptr, NULL); +} + + /* 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. */ @@ -292,8 +308,9 @@ ensure_minumim_exponent_length(char* buffer, size_t buf_size) } } -/* Ensure that buffer has a decimal point in it. The decimal point - will not be in the current locale, it will always be '.' */ +/* 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 if an + exponent is present. */ Py_LOCAL_INLINE(void) ensure_decimal_point(char* buffer, size_t buf_size) { @@ -322,7 +339,8 @@ ensure_decimal_point(char* buffer, size_t buf_size) insert_count = 1; } } - else { + else if (!(*p == 'e' || *p == 'E')) { + /* Don't add ".0" if we have an exponent. */ chars_to_insert = ".0"; insert_count = 2; } @@ -341,37 +359,6 @@ ensure_decimal_point(char* buffer, size_t buf_size) } } -/* Add the locale specific grouping characters to buffer. Note - that any decimal point (if it's present) in buffer is already - locale-specific. Return 0 on error, else 1. */ -Py_LOCAL_INLINE(int) -add_thousands_grouping(char* buffer, size_t buf_size) -{ - Py_ssize_t len = strlen(buffer); - struct lconv *locale_data = localeconv(); - const char *decimal_point = locale_data->decimal_point; - - /* Find the decimal point, if any. We're only concerned - about the characters to the left of the decimal when - adding grouping. */ - char *p = strstr(buffer, decimal_point); - if (!p) { - /* No decimal, use the entire string. */ - - /* If any exponent, adjust p. */ - p = strpbrk(buffer, "eE"); - if (!p) - /* No exponent and no decimal. Use the entire - string. */ - p = buffer + len; - } - /* At this point, p points just past the right-most character we - want to format. We need to add the grouping string for the - characters between buffer and p. */ - return _PyString_InsertThousandsGrouping(buffer, len, p-buffer, - buf_size, NULL, 1); -} - /* see FORMATBUFLEN in unicodeobject.c */ #define FLOAT_FORMATBUFLEN 120 @@ -386,9 +373,8 @@ add_thousands_grouping(char* buffer, size_t buf_size) * Converts a #gdouble to a string, using the '.' as * decimal point. To format the number you pass in * a printf()-style format string. Allowed conversion - * specifiers are 'e', 'E', 'f', 'F', 'g', 'G', and 'n'. + * specifiers are 'e', 'E', 'f', 'F', 'g', 'G', and 'Z'. * - * 'n' is the same as 'g', except it uses the current locale. * 'Z' is the same as 'g', except it always has a decimal and * at least one digit after the decimal. * @@ -403,11 +389,6 @@ PyOS_ascii_formatd(char *buffer, char format_char; size_t format_len = strlen(format); - /* For type 'n', we need to make a copy of the format string, because - we're going to modify 'n' -> 'g', and format is const char*, so we - can't modify it directly. FLOAT_FORMATBUFLEN should be longer than - we ever need this to be. There's an upcoming check to ensure it's - big enough. */ /* Issue 2264: code 'Z' requires copying the format. 'Z' is 'g', but also with at least one character past the decimal. */ char tmp_format[FLOAT_FORMATBUFLEN]; @@ -433,12 +414,12 @@ PyOS_ascii_formatd(char *buffer, if (!(format_char == 'e' || format_char == 'E' || format_char == 'f' || format_char == 'F' || format_char == 'g' || format_char == 'G' || - format_char == 'n' || format_char == 'Z')) + format_char == 'Z')) return NULL; - /* Map 'n' or 'Z' format_char to 'g', by copying the format string and + /* Map 'Z' format_char to 'g', by copying the format string and replacing the final char with a 'g' */ - if (format_char == 'n' || format_char == 'Z') { + if (format_char == 'Z') { if (format_len + 1 >= sizeof(tmp_format)) { /* The format won't fit in our copy. Error out. In practice, this will never happen and will be @@ -457,11 +438,8 @@ PyOS_ascii_formatd(char *buffer, /* Do various fixups on the return string */ /* Get the current locale, and find the decimal point string. - Convert that string back to a dot. Do not do this if using the - 'n' (number) format code, since we want to keep the localized - decimal point in that case. */ - if (format_char != 'n') - change_decimal_from_locale_to_dot(buffer); + Convert that string back to a dot. */ + change_decimal_from_locale_to_dot(buffer); /* If an exponent exists, ensure that the exponent is at least MIN_EXPONENT_DIGITS digits, providing the buffer is large enough @@ -475,16 +453,111 @@ PyOS_ascii_formatd(char *buffer, if (format_char == 'Z') ensure_decimal_point(buffer, buf_size); - /* If format_char is 'n', add the thousands grouping. */ - if (format_char == 'n') - if (!add_thousands_grouping(buffer, buf_size)) - return NULL; - return buffer; } -double -PyOS_ascii_atof(const char *nptr) +PyAPI_FUNC(char *) PyOS_double_to_string(double val, + char format_code, + int precision, + int flags, + int *type) { - return PyOS_ascii_strtod(nptr, NULL); + char buf[128]; + char format[32]; + Py_ssize_t len; + char *result; + char *p; + int t; + int upper = 0; + + /* Validate format_code, and map upper and lower case */ + switch (format_code) { + case 'e': /* exponent */ + 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; + } + precision = 17; + format_code = 'g'; + break; + case 's': /* str format */ + /* Supplied precision is unused, must be 0. */ + if (precision != 0) { + PyErr_BadInternalCall(); + return NULL; + } + precision = 12; + format_code = 'g'; + break; + default: + PyErr_BadInternalCall(); + return NULL; + } + + /* 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, 32, "%%%s.%i%c", (flags & Py_DTSF_ALT ? "#" : ""), precision, format_code); + PyOS_ascii_formatd(buf, sizeof(buf), format, val); + } + + len = strlen(buf); + + /* Add 1 for the trailing 0 byte. + Add 1 because we might need to make room for the sign. + */ + result = PyMem_Malloc(len + 2); + if (result == NULL) { + PyErr_NoMemory(); + return NULL; + } + p = result; + + /* Never add sign for nan/inf, even if asked. */ + if (flags & Py_DTSF_SIGN && buf[0] != '-' && t == Py_DTST_FINITE) + *p++ = '+'; + + strcpy(p, buf); + + if (upper) { + /* Convert to upper case. */ + char *p1; + for (p1 = p; *p1; p1++) + *p1 = toupper(*p1); + } + + if (type) + *type = t; + return result; }