Server IP : 66.29.132.122 / Your IP : 3.141.37.40 Web Server : LiteSpeed System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64 User : admazpex ( 531) PHP Version : 7.2.34 Disable Function : NONE MySQL : OFF | cURL : ON | WGET : ON | Perl : ON | Python : ON | Sudo : OFF | Pkexec : OFF Directory : /proc/self/root/opt/imunify360/venv/lib64/python3.11/site-packages/babel/ |
Upload File : |
""" babel.numbers ~~~~~~~~~~~~~ Locale dependent formatting and parsing of numeric data. The default locale for the functions in this module is determined by the following environment variables, in that order: * ``LC_NUMERIC``, * ``LC_ALL``, and * ``LANG`` :copyright: (c) 2013-2023 by the Babel Team. :license: BSD, see LICENSE for more details. """ # TODO: # Padding and rounding increments in pattern: # - https://www.unicode.org/reports/tr35/ (Appendix G.6) from __future__ import annotations import datetime import decimal import re import warnings from typing import TYPE_CHECKING, Any, cast, overload from babel.core import Locale, default_locale, get_global from babel.localedata import LocaleDataDict if TYPE_CHECKING: from typing_extensions import Literal LC_NUMERIC = default_locale('LC_NUMERIC') class UnknownCurrencyError(Exception): """Exception thrown when a currency is requested for which no data is available. """ def __init__(self, identifier: str) -> None: """Create the exception. :param identifier: the identifier string of the unsupported currency """ Exception.__init__(self, f"Unknown currency {identifier!r}.") #: The identifier of the locale that could not be found. self.identifier = identifier def list_currencies(locale: Locale | str | None = None) -> set[str]: """ Return a `set` of normalized currency codes. .. versionadded:: 2.5.0 :param locale: filters returned currency codes by the provided locale. Expected to be a locale instance or code. If no locale is provided, returns the list of all currencies from all locales. """ # Get locale-scoped currencies. if locale: currencies = Locale.parse(locale).currencies.keys() else: currencies = get_global('all_currencies') return set(currencies) def validate_currency(currency: str, locale: Locale | str | None = None) -> None: """ Check the currency code is recognized by Babel. Accepts a ``locale`` parameter for fined-grained validation, working as the one defined above in ``list_currencies()`` method. Raises a `UnknownCurrencyError` exception if the currency is unknown to Babel. """ if currency not in list_currencies(locale): raise UnknownCurrencyError(currency) def is_currency(currency: str, locale: Locale | str | None = None) -> bool: """ Returns `True` only if a currency is recognized by Babel. This method always return a Boolean and never raise. """ if not currency or not isinstance(currency, str): return False try: validate_currency(currency, locale) except UnknownCurrencyError: return False return True def normalize_currency(currency: str, locale: Locale | str | None = None) -> str | None: """Returns the normalized identifier of any currency code. Accepts a ``locale`` parameter for fined-grained validation, working as the one defined above in ``list_currencies()`` method. Returns None if the currency is unknown to Babel. """ if isinstance(currency, str): currency = currency.upper() if not is_currency(currency, locale): return return currency def get_currency_name( currency: str, count: float | decimal.Decimal | None = None, locale: Locale | str | None = LC_NUMERIC, ) -> str: """Return the name used by the locale for the specified currency. >>> get_currency_name('USD', locale='en_US') u'US Dollar' .. versionadded:: 0.9.4 :param currency: the currency code. :param count: the optional count. If provided the currency name will be pluralized to that number if possible. :param locale: the `Locale` object or locale identifier. """ loc = Locale.parse(locale) if count is not None: try: plural_form = loc.plural_form(count) except (OverflowError, ValueError): plural_form = 'other' plural_names = loc._data['currency_names_plural'] if currency in plural_names: currency_plural_names = plural_names[currency] if plural_form in currency_plural_names: return currency_plural_names[plural_form] if 'other' in currency_plural_names: return currency_plural_names['other'] return loc.currencies.get(currency, currency) def get_currency_symbol(currency: str, locale: Locale | str | None = LC_NUMERIC) -> str: """Return the symbol used by the locale for the specified currency. >>> get_currency_symbol('USD', locale='en_US') u'$' :param currency: the currency code. :param locale: the `Locale` object or locale identifier. """ return Locale.parse(locale).currency_symbols.get(currency, currency) def get_currency_precision(currency: str) -> int: """Return currency's precision. Precision is the number of decimals found after the decimal point in the currency's format pattern. .. versionadded:: 2.5.0 :param currency: the currency code. """ precisions = get_global('currency_fractions') return precisions.get(currency, precisions['DEFAULT'])[0] def get_currency_unit_pattern( currency: str, count: float | decimal.Decimal | None = None, locale: Locale | str | None = LC_NUMERIC, ) -> str: """ Return the unit pattern used for long display of a currency value for a given locale. This is a string containing ``{0}`` where the numeric part should be substituted and ``{1}`` where the currency long display name should be substituted. >>> get_currency_unit_pattern('USD', locale='en_US', count=10) u'{0} {1}' .. versionadded:: 2.7.0 :param currency: the currency code. :param count: the optional count. If provided the unit pattern for that number will be returned. :param locale: the `Locale` object or locale identifier. """ loc = Locale.parse(locale) if count is not None: plural_form = loc.plural_form(count) try: return loc._data['currency_unit_patterns'][plural_form] except LookupError: # Fall back to 'other' pass return loc._data['currency_unit_patterns']['other'] @overload def get_territory_currencies( territory: str, start_date: datetime.date | None = ..., end_date: datetime.date | None = ..., tender: bool = ..., non_tender: bool = ..., include_details: Literal[False] = ..., ) -> list[str]: ... # pragma: no cover @overload def get_territory_currencies( territory: str, start_date: datetime.date | None = ..., end_date: datetime.date | None = ..., tender: bool = ..., non_tender: bool = ..., include_details: Literal[True] = ..., ) -> list[dict[str, Any]]: ... # pragma: no cover def get_territory_currencies( territory: str, start_date: datetime.date | None = None, end_date: datetime.date | None = None, tender: bool = True, non_tender: bool = False, include_details: bool = False, ) -> list[str] | list[dict[str, Any]]: """Returns the list of currencies for the given territory that are valid for the given date range. In addition to that the currency database distinguishes between tender and non-tender currencies. By default only tender currencies are returned. The return value is a list of all currencies roughly ordered by the time of when the currency became active. The longer the currency is being in use the more to the left of the list it will be. The start date defaults to today. If no end date is given it will be the same as the start date. Otherwise a range can be defined. For instance this can be used to find the currencies in use in Austria between 1995 and 2011: >>> from datetime import date >>> get_territory_currencies('AT', date(1995, 1, 1), date(2011, 1, 1)) ['ATS', 'EUR'] Likewise it's also possible to find all the currencies in use on a single date: >>> get_territory_currencies('AT', date(1995, 1, 1)) ['ATS'] >>> get_territory_currencies('AT', date(2011, 1, 1)) ['EUR'] By default the return value only includes tender currencies. This however can be changed: >>> get_territory_currencies('US') ['USD'] >>> get_territory_currencies('US', tender=False, non_tender=True, ... start_date=date(2014, 1, 1)) ['USN', 'USS'] .. versionadded:: 2.0 :param territory: the name of the territory to find the currency for. :param start_date: the start date. If not given today is assumed. :param end_date: the end date. If not given the start date is assumed. :param tender: controls whether tender currencies should be included. :param non_tender: controls whether non-tender currencies should be included. :param include_details: if set to `True`, instead of returning currency codes the return value will be dictionaries with detail information. In that case each dictionary will have the keys ``'currency'``, ``'from'``, ``'to'``, and ``'tender'``. """ currencies = get_global('territory_currencies') if start_date is None: start_date = datetime.date.today() elif isinstance(start_date, datetime.datetime): start_date = start_date.date() if end_date is None: end_date = start_date elif isinstance(end_date, datetime.datetime): end_date = end_date.date() curs = currencies.get(territory.upper(), ()) # TODO: validate that the territory exists def _is_active(start, end): return (start is None or start <= end_date) and \ (end is None or end >= start_date) result = [] for currency_code, start, end, is_tender in curs: if start: start = datetime.date(*start) if end: end = datetime.date(*end) if ((is_tender and tender) or (not is_tender and non_tender)) and _is_active(start, end): if include_details: result.append({ 'currency': currency_code, 'from': start, 'to': end, 'tender': is_tender, }) else: result.append(currency_code) return result def get_decimal_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the symbol used by the locale to separate decimal fractions. >>> get_decimal_symbol('en_US') u'.' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('decimal', '.') def get_plus_sign_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the plus sign symbol used by the current locale. >>> get_plus_sign_symbol('en_US') u'+' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('plusSign', '+') def get_minus_sign_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the plus sign symbol used by the current locale. >>> get_minus_sign_symbol('en_US') u'-' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('minusSign', '-') def get_exponential_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the symbol used by the locale to separate mantissa and exponent. >>> get_exponential_symbol('en_US') u'E' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('exponential', 'E') def get_group_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the symbol used by the locale to separate groups of thousands. >>> get_group_symbol('en_US') u',' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('group', ',') def get_infinity_symbol(locale: Locale | str | None = LC_NUMERIC) -> str: """Return the symbol used by the locale to represent infinity. >>> get_infinity_symbol('en_US') u'∞' :param locale: the `Locale` object or locale identifier """ return Locale.parse(locale).number_symbols.get('infinity', '∞') def format_number(number: float | decimal.Decimal | str, locale: Locale | str | None = LC_NUMERIC) -> str: """Return the given number formatted for a specific locale. >>> format_number(1099, locale='en_US') # doctest: +SKIP u'1,099' >>> format_number(1099, locale='de_DE') # doctest: +SKIP u'1.099' .. deprecated:: 2.6.0 Use babel.numbers.format_decimal() instead. :param number: the number to format :param locale: the `Locale` object or locale identifier """ warnings.warn('Use babel.numbers.format_decimal() instead.', DeprecationWarning) return format_decimal(number, locale=locale) def get_decimal_precision(number: decimal.Decimal) -> int: """Return maximum precision of a decimal instance's fractional part. Precision is extracted from the fractional part only. """ # Copied from: https://github.com/mahmoud/boltons/pull/59 assert isinstance(number, decimal.Decimal) decimal_tuple = number.normalize().as_tuple() # Note: DecimalTuple.exponent can be 'n' (qNaN), 'N' (sNaN), or 'F' (Infinity) if not isinstance(decimal_tuple.exponent, int) or decimal_tuple.exponent >= 0: return 0 return abs(decimal_tuple.exponent) def get_decimal_quantum(precision: int | decimal.Decimal) -> decimal.Decimal: """Return minimal quantum of a number, as defined by precision.""" assert isinstance(precision, (int, decimal.Decimal)) return decimal.Decimal(10) ** (-precision) def format_decimal( number: float | decimal.Decimal | str, format: str | NumberPattern | None = None, locale: Locale | str | None = LC_NUMERIC, decimal_quantization: bool = True, group_separator: bool = True, ) -> str: """Return the given decimal number formatted for a specific locale. >>> format_decimal(1.2345, locale='en_US') u'1.234' >>> format_decimal(1.2346, locale='en_US') u'1.235' >>> format_decimal(-1.2346, locale='en_US') u'-1.235' >>> format_decimal(1.2345, locale='sv_SE') u'1,234' >>> format_decimal(1.2345, locale='de') u'1,234' The appropriate thousands grouping and the decimal separator are used for each locale: >>> format_decimal(12345.5, locale='en_US') u'12,345.5' By default the locale is allowed to truncate and round a high-precision number by forcing its format pattern onto the decimal part. You can bypass this behavior with the `decimal_quantization` parameter: >>> format_decimal(1.2346, locale='en_US') u'1.235' >>> format_decimal(1.2346, locale='en_US', decimal_quantization=False) u'1.2346' >>> format_decimal(12345.67, locale='fr_CA', group_separator=False) u'12345,67' >>> format_decimal(12345.67, locale='en_US', group_separator=True) u'12,345.67' :param number: the number to format :param format: :param locale: the `Locale` object or locale identifier :param decimal_quantization: Truncate and round high-precision numbers to the format pattern. Defaults to `True`. :param group_separator: Boolean to switch group separator on/off in a locale's number format. """ locale = Locale.parse(locale) if format is None: format = locale.decimal_formats[format] pattern = parse_pattern(format) return pattern.apply( number, locale, decimal_quantization=decimal_quantization, group_separator=group_separator) def format_compact_decimal( number: float | decimal.Decimal | str, *, format_type: Literal["short", "long"] = "short", locale: Locale | str | None = LC_NUMERIC, fraction_digits: int = 0, ) -> str: """Return the given decimal number formatted for a specific locale in compact form. >>> format_compact_decimal(12345, format_type="short", locale='en_US') u'12K' >>> format_compact_decimal(12345, format_type="long", locale='en_US') u'12 thousand' >>> format_compact_decimal(12345, format_type="short", locale='en_US', fraction_digits=2) u'12.34K' >>> format_compact_decimal(1234567, format_type="short", locale="ja_JP") u'123万' >>> format_compact_decimal(2345678, format_type="long", locale="mk") u'2 милиони' >>> format_compact_decimal(21000000, format_type="long", locale="mk") u'21 милион' :param number: the number to format :param format_type: Compact format to use ("short" or "long") :param locale: the `Locale` object or locale identifier :param fraction_digits: Number of digits after the decimal point to use. Defaults to `0`. """ locale = Locale.parse(locale) compact_format = locale.compact_decimal_formats[format_type] number, format = _get_compact_format(number, compact_format, locale, fraction_digits) # Did not find a format, fall back. if format is None: format = locale.decimal_formats[None] pattern = parse_pattern(format) return pattern.apply(number, locale, decimal_quantization=False) def _get_compact_format( number: float | decimal.Decimal | str, compact_format: LocaleDataDict, locale: Locale, fraction_digits: int, ) -> tuple[decimal.Decimal, NumberPattern | None]: """Returns the number after dividing by the unit and the format pattern to use. The algorithm is described here: https://www.unicode.org/reports/tr35/tr35-45/tr35-numbers.html#Compact_Number_Formats. """ if not isinstance(number, decimal.Decimal): number = decimal.Decimal(str(number)) if number.is_nan() or number.is_infinite(): return number, None format = None for magnitude in sorted([int(m) for m in compact_format["other"]], reverse=True): if abs(number) >= magnitude: # check the pattern using "other" as the amount format = compact_format["other"][str(magnitude)] pattern = parse_pattern(format).pattern # if the pattern is "0", we do not divide the number if pattern == "0": break # otherwise, we need to divide the number by the magnitude but remove zeros # equal to the number of 0's in the pattern minus 1 number = cast(decimal.Decimal, number / (magnitude // (10 ** (pattern.count("0") - 1)))) # round to the number of fraction digits requested rounded = round(number, fraction_digits) # if the remaining number is singular, use the singular format plural_form = locale.plural_form(abs(number)) if plural_form not in compact_format: plural_form = "other" if number == 1 and "1" in compact_format: plural_form = "1" format = compact_format[plural_form][str(magnitude)] number = rounded break return number, format class UnknownCurrencyFormatError(KeyError): """Exception raised when an unknown currency format is requested.""" def format_currency( number: float | decimal.Decimal | str, currency: str, format: str | NumberPattern | None = None, locale: Locale | str | None = LC_NUMERIC, currency_digits: bool = True, format_type: Literal["name", "standard", "accounting"] = "standard", decimal_quantization: bool = True, group_separator: bool = True, ) -> str: """Return formatted currency value. >>> format_currency(1099.98, 'USD', locale='en_US') '$1,099.98' >>> format_currency(1099.98, 'USD', locale='es_CO') u'US$1.099,98' >>> format_currency(1099.98, 'EUR', locale='de_DE') u'1.099,98\\xa0\\u20ac' The format can also be specified explicitly. The currency is placed with the '¤' sign. As the sign gets repeated the format expands (¤ being the symbol, ¤¤ is the currency abbreviation and ¤¤¤ is the full name of the currency): >>> format_currency(1099.98, 'EUR', u'\xa4\xa4 #,##0.00', locale='en_US') u'EUR 1,099.98' >>> format_currency(1099.98, 'EUR', u'#,##0.00 \xa4\xa4\xa4', locale='en_US') u'1,099.98 euros' Currencies usually have a specific number of decimal digits. This function favours that information over the given format: >>> format_currency(1099.98, 'JPY', locale='en_US') u'\\xa51,100' >>> format_currency(1099.98, 'COP', u'#,##0.00', locale='es_ES') u'1.099,98' However, the number of decimal digits can be overridden from the currency information, by setting the last parameter to ``False``: >>> format_currency(1099.98, 'JPY', locale='en_US', currency_digits=False) u'\\xa51,099.98' >>> format_currency(1099.98, 'COP', u'#,##0.00', locale='es_ES', currency_digits=False) u'1.099,98' If a format is not specified the type of currency format to use from the locale can be specified: >>> format_currency(1099.98, 'EUR', locale='en_US', format_type='standard') u'\\u20ac1,099.98' When the given currency format type is not available, an exception is raised: >>> format_currency('1099.98', 'EUR', locale='root', format_type='unknown') Traceback (most recent call last): ... UnknownCurrencyFormatError: "'unknown' is not a known currency format type" >>> format_currency(101299.98, 'USD', locale='en_US', group_separator=False) u'$101299.98' >>> format_currency(101299.98, 'USD', locale='en_US', group_separator=True) u'$101,299.98' You can also pass format_type='name' to use long display names. The order of the number and currency name, along with the correct localized plural form of the currency name, is chosen according to locale: >>> format_currency(1, 'USD', locale='en_US', format_type='name') u'1.00 US dollar' >>> format_currency(1099.98, 'USD', locale='en_US', format_type='name') u'1,099.98 US dollars' >>> format_currency(1099.98, 'USD', locale='ee', format_type='name') u'us ga dollar 1,099.98' By default the locale is allowed to truncate and round a high-precision number by forcing its format pattern onto the decimal part. You can bypass this behavior with the `decimal_quantization` parameter: >>> format_currency(1099.9876, 'USD', locale='en_US') u'$1,099.99' >>> format_currency(1099.9876, 'USD', locale='en_US', decimal_quantization=False) u'$1,099.9876' :param number: the number to format :param currency: the currency code :param format: the format string to use :param locale: the `Locale` object or locale identifier :param currency_digits: use the currency's natural number of decimal digits :param format_type: the currency format type to use :param decimal_quantization: Truncate and round high-precision numbers to the format pattern. Defaults to `True`. :param group_separator: Boolean to switch group separator on/off in a locale's number format. """ if format_type == 'name': return _format_currency_long_name(number, currency, format=format, locale=locale, currency_digits=currency_digits, decimal_quantization=decimal_quantization, group_separator=group_separator) locale = Locale.parse(locale) if format: pattern = parse_pattern(format) else: try: pattern = locale.currency_formats[format_type] except KeyError: raise UnknownCurrencyFormatError(f"{format_type!r} is not a known currency format type") from None return pattern.apply( number, locale, currency=currency, currency_digits=currency_digits, decimal_quantization=decimal_quantization, group_separator=group_separator) def _format_currency_long_name( number: float | decimal.Decimal | str, currency: str, format: str | NumberPattern | None = None, locale: Locale | str | None = LC_NUMERIC, currency_digits: bool = True, format_type: Literal["name", "standard", "accounting"] = "standard", decimal_quantization: bool = True, group_separator: bool = True, ) -> str: # Algorithm described here: # https://www.unicode.org/reports/tr35/tr35-numbers.html#Currencies locale = Locale.parse(locale) # Step 1. # There are no examples of items with explicit count (0 or 1) in current # locale data. So there is no point implementing that. # Step 2. # Correct number to numeric type, important for looking up plural rules: number_n = float(number) if isinstance(number, str) else number # Step 3. unit_pattern = get_currency_unit_pattern(currency, count=number_n, locale=locale) # Step 4. display_name = get_currency_name(currency, count=number_n, locale=locale) # Step 5. if not format: format = locale.decimal_formats[format] pattern = parse_pattern(format) number_part = pattern.apply( number, locale, currency=currency, currency_digits=currency_digits, decimal_quantization=decimal_quantization, group_separator=group_separator) return unit_pattern.format(number_part, display_name) def format_compact_currency( number: float | decimal.Decimal | str, currency: str, *, format_type: Literal["short"] = "short", locale: Locale | str | None = LC_NUMERIC, fraction_digits: int = 0 ) -> str: """Format a number as a currency value in compact form. >>> format_compact_currency(12345, 'USD', locale='en_US') u'$12K' >>> format_compact_currency(123456789, 'USD', locale='en_US', fraction_digits=2) u'$123.46M' >>> format_compact_currency(123456789, 'EUR', locale='de_DE', fraction_digits=1) '123,5\xa0Mio.\xa0€' :param number: the number to format :param currency: the currency code :param format_type: the compact format type to use. Defaults to "short". :param locale: the `Locale` object or locale identifier :param fraction_digits: Number of digits after the decimal point to use. Defaults to `0`. """ locale = Locale.parse(locale) try: compact_format = locale.compact_currency_formats[format_type] except KeyError as error: raise UnknownCurrencyFormatError(f"{format_type!r} is not a known compact currency format type") from error number, format = _get_compact_format(number, compact_format, locale, fraction_digits) # Did not find a format, fall back. if format is None or "¤" not in str(format): # find first format that has a currency symbol for magnitude in compact_format['other']: format = compact_format['other'][magnitude].pattern if '¤' not in format: continue # remove characters that are not the currency symbol, 0's or spaces format = re.sub(r'[^0\s\¤]', '', format) # compress adjacent spaces into one format = re.sub(r'(\s)\s+', r'\1', format).strip() break if format is None: raise ValueError('No compact currency format found for the given number and locale.') pattern = parse_pattern(format) return pattern.apply(number, locale, currency=currency, currency_digits=False, decimal_quantization=False) def format_percent( number: float | decimal.Decimal | str, format: str | NumberPattern | None = None, locale: Locale | str | None = LC_NUMERIC, decimal_quantization: bool = True, group_separator: bool = True, ) -> str: """Return formatted percent value for a specific locale. >>> format_percent(0.34, locale='en_US') u'34%' >>> format_percent(25.1234, locale='en_US') u'2,512%' >>> format_percent(25.1234, locale='sv_SE') u'2\\xa0512\\xa0%' The format pattern can also be specified explicitly: >>> format_percent(25.1234, u'#,##0\u2030', locale='en_US') u'25,123\u2030' By default the locale is allowed to truncate and round a high-precision number by forcing its format pattern onto the decimal part. You can bypass this behavior with the `decimal_quantization` parameter: >>> format_percent(23.9876, locale='en_US') u'2,399%' >>> format_percent(23.9876, locale='en_US', decimal_quantization=False) u'2,398.76%' >>> format_percent(229291.1234, locale='pt_BR', group_separator=False) u'22929112%' >>> format_percent(229291.1234, locale='pt_BR', group_separator=True) u'22.929.112%' :param number: the percent number to format :param format: :param locale: the `Locale` object or locale identifier :param decimal_quantization: Truncate and round high-precision numbers to the format pattern. Defaults to `True`. :param group_separator: Boolean to switch group separator on/off in a locale's number format. """ locale = Locale.parse(locale) if not format: format = locale.percent_formats[format] pattern = parse_pattern(format) return pattern.apply( number, locale, decimal_quantization=decimal_quantization, group_separator=group_separator) def format_scientific( number: float | decimal.Decimal | str, format: str | NumberPattern | None = None, locale: Locale | str | None = LC_NUMERIC, decimal_quantization: bool = True, ) -> str: """Return value formatted in scientific notation for a specific locale. >>> format_scientific(10000, locale='en_US') u'1E4' The format pattern can also be specified explicitly: >>> format_scientific(1234567, u'##0.##E00', locale='en_US') u'1.23E06' By default the locale is allowed to truncate and round a high-precision number by forcing its format pattern onto the decimal part. You can bypass this behavior with the `decimal_quantization` parameter: >>> format_scientific(1234.9876, u'#.##E0', locale='en_US') u'1.23E3' >>> format_scientific(1234.9876, u'#.##E0', locale='en_US', decimal_quantization=False) u'1.2349876E3' :param number: the number to format :param format: :param locale: the `Locale` object or locale identifier :param decimal_quantization: Truncate and round high-precision numbers to the format pattern. Defaults to `True`. """ locale = Locale.parse(locale) if not format: format = locale.scientific_formats[format] pattern = parse_pattern(format) return pattern.apply( number, locale, decimal_quantization=decimal_quantization) class NumberFormatError(ValueError): """Exception raised when a string cannot be parsed into a number.""" def __init__(self, message: str, suggestions: list[str] | None = None) -> None: super().__init__(message) #: a list of properly formatted numbers derived from the invalid input self.suggestions = suggestions def parse_number(string: str, locale: Locale | str | None = LC_NUMERIC) -> int: """Parse localized number string into an integer. >>> parse_number('1,099', locale='en_US') 1099 >>> parse_number('1.099', locale='de_DE') 1099 When the given string cannot be parsed, an exception is raised: >>> parse_number('1.099,98', locale='de') Traceback (most recent call last): ... NumberFormatError: '1.099,98' is not a valid number :param string: the string to parse :param locale: the `Locale` object or locale identifier :return: the parsed number :raise `NumberFormatError`: if the string can not be converted to a number """ try: return int(string.replace(get_group_symbol(locale), '')) except ValueError as ve: raise NumberFormatError(f"{string!r} is not a valid number") from ve def parse_decimal(string: str, locale: Locale | str | None = LC_NUMERIC, strict: bool = False) -> decimal.Decimal: """Parse localized decimal string into a decimal. >>> parse_decimal('1,099.98', locale='en_US') Decimal('1099.98') >>> parse_decimal('1.099,98', locale='de') Decimal('1099.98') >>> parse_decimal('12 345,123', locale='ru') Decimal('12345.123') When the given string cannot be parsed, an exception is raised: >>> parse_decimal('2,109,998', locale='de') Traceback (most recent call last): ... NumberFormatError: '2,109,998' is not a valid decimal number If `strict` is set to `True` and the given string contains a number formatted in an irregular way, an exception is raised: >>> parse_decimal('30.00', locale='de', strict=True) Traceback (most recent call last): ... NumberFormatError: '30.00' is not a properly formatted decimal number. Did you mean '3.000'? Or maybe '30,00'? >>> parse_decimal('0.00', locale='de', strict=True) Traceback (most recent call last): ... NumberFormatError: '0.00' is not a properly formatted decimal number. Did you mean '0'? :param string: the string to parse :param locale: the `Locale` object or locale identifier :param strict: controls whether numbers formatted in a weird way are accepted or rejected :raise NumberFormatError: if the string can not be converted to a decimal number """ locale = Locale.parse(locale) group_symbol = get_group_symbol(locale) decimal_symbol = get_decimal_symbol(locale) if not strict and ( group_symbol == '\xa0' and # if the grouping symbol is U+00A0 NO-BREAK SPACE, group_symbol not in string and # and the string to be parsed does not contain it, ' ' in string # but it does contain a space instead, ): # ... it's reasonable to assume it is taking the place of the grouping symbol. string = string.replace(' ', group_symbol) try: parsed = decimal.Decimal(string.replace(group_symbol, '') .replace(decimal_symbol, '.')) except decimal.InvalidOperation as exc: raise NumberFormatError(f"{string!r} is not a valid decimal number") from exc if strict and group_symbol in string: proper = format_decimal(parsed, locale=locale, decimal_quantization=False) if string != proper and string.rstrip('0') != (proper + decimal_symbol): try: parsed_alt = decimal.Decimal(string.replace(decimal_symbol, '') .replace(group_symbol, '.')) except decimal.InvalidOperation as exc: raise NumberFormatError( f"{string!r} is not a properly formatted decimal number. " f"Did you mean {proper!r}?", suggestions=[proper], ) from exc else: proper_alt = format_decimal(parsed_alt, locale=locale, decimal_quantization=False) if proper_alt == proper: raise NumberFormatError( f"{string!r} is not a properly formatted decimal number. " f"Did you mean {proper!r}?", suggestions=[proper], ) else: raise NumberFormatError( f"{string!r} is not a properly formatted decimal number. " f"Did you mean {proper!r}? Or maybe {proper_alt!r}?", suggestions=[proper, proper_alt], ) return parsed PREFIX_END = r'[^0-9@#.,]' NUMBER_TOKEN = r'[0-9@#.,E+]' PREFIX_PATTERN = r"(?P<prefix>(?:'[^']*'|%s)*)" % PREFIX_END NUMBER_PATTERN = r"(?P<number>%s*)" % NUMBER_TOKEN SUFFIX_PATTERN = r"(?P<suffix>.*)" number_re = re.compile(f"{PREFIX_PATTERN}{NUMBER_PATTERN}{SUFFIX_PATTERN}") def parse_grouping(p: str) -> tuple[int, int]: """Parse primary and secondary digit grouping >>> parse_grouping('##') (1000, 1000) >>> parse_grouping('#,###') (3, 3) >>> parse_grouping('#,####,###') (3, 4) """ width = len(p) g1 = p.rfind(',') if g1 == -1: return 1000, 1000 g1 = width - g1 - 1 g2 = p[:-g1 - 1].rfind(',') if g2 == -1: return g1, g1 g2 = width - g1 - g2 - 2 return g1, g2 def parse_pattern(pattern: NumberPattern | str) -> NumberPattern: """Parse number format patterns""" if isinstance(pattern, NumberPattern): return pattern def _match_number(pattern): rv = number_re.search(pattern) if rv is None: raise ValueError(f"Invalid number pattern {pattern!r}") return rv.groups() pos_pattern = pattern # Do we have a negative subpattern? if ';' in pattern: pos_pattern, neg_pattern = pattern.split(';', 1) pos_prefix, number, pos_suffix = _match_number(pos_pattern) neg_prefix, _, neg_suffix = _match_number(neg_pattern) else: pos_prefix, number, pos_suffix = _match_number(pos_pattern) neg_prefix = f"-{pos_prefix}" neg_suffix = pos_suffix if 'E' in number: number, exp = number.split('E', 1) else: exp = None if '@' in number and '.' in number and '0' in number: raise ValueError('Significant digit patterns can not contain "@" or "0"') if '.' in number: integer, fraction = number.rsplit('.', 1) else: integer = number fraction = '' def parse_precision(p): """Calculate the min and max allowed digits""" min = max = 0 for c in p: if c in '@0': min += 1 max += 1 elif c == '#': max += 1 elif c == ',': continue else: break return min, max int_prec = parse_precision(integer) frac_prec = parse_precision(fraction) if exp: exp_plus = exp.startswith('+') exp = exp.lstrip('+') exp_prec = parse_precision(exp) else: exp_plus = None exp_prec = None grouping = parse_grouping(integer) return NumberPattern(pattern, (pos_prefix, neg_prefix), (pos_suffix, neg_suffix), grouping, int_prec, frac_prec, exp_prec, exp_plus, number) class NumberPattern: def __init__( self, pattern: str, prefix: tuple[str, str], suffix: tuple[str, str], grouping: tuple[int, int], int_prec: tuple[int, int], frac_prec: tuple[int, int], exp_prec: tuple[int, int] | None, exp_plus: bool | None, number_pattern: str | None = None, ) -> None: # Metadata of the decomposed parsed pattern. self.pattern = pattern self.prefix = prefix self.suffix = suffix self.number_pattern = number_pattern self.grouping = grouping self.int_prec = int_prec self.frac_prec = frac_prec self.exp_prec = exp_prec self.exp_plus = exp_plus self.scale = self.compute_scale() def __repr__(self) -> str: return f"<{type(self).__name__} {self.pattern!r}>" def compute_scale(self) -> Literal[0, 2, 3]: """Return the scaling factor to apply to the number before rendering. Auto-set to a factor of 2 or 3 if presence of a ``%`` or ``‰`` sign is detected in the prefix or suffix of the pattern. Default is to not mess with the scale at all and keep it to 0. """ scale = 0 if '%' in ''.join(self.prefix + self.suffix): scale = 2 elif '‰' in ''.join(self.prefix + self.suffix): scale = 3 return scale def scientific_notation_elements(self, value: decimal.Decimal, locale: Locale | str | None) -> tuple[decimal.Decimal, int, str]: """ Returns normalized scientific notation components of a value. """ # Normalize value to only have one lead digit. exp = value.adjusted() value = value * get_decimal_quantum(exp) assert value.adjusted() == 0 # Shift exponent and value by the minimum number of leading digits # imposed by the rendering pattern. And always make that number # greater or equal to 1. lead_shift = max([1, min(self.int_prec)]) - 1 exp = exp - lead_shift value = value * get_decimal_quantum(-lead_shift) # Get exponent sign symbol. exp_sign = '' if exp < 0: exp_sign = get_minus_sign_symbol(locale) elif self.exp_plus: exp_sign = get_plus_sign_symbol(locale) # Normalize exponent value now that we have the sign. exp = abs(exp) return value, exp, exp_sign def apply( self, value: float | decimal.Decimal | str, locale: Locale | str | None, currency: str | None = None, currency_digits: bool = True, decimal_quantization: bool = True, force_frac: tuple[int, int] | None = None, group_separator: bool = True, ): """Renders into a string a number following the defined pattern. Forced decimal quantization is active by default so we'll produce a number string that is strictly following CLDR pattern definitions. :param value: The value to format. If this is not a Decimal object, it will be cast to one. :type value: decimal.Decimal|float|int :param locale: The locale to use for formatting. :type locale: str|babel.core.Locale :param currency: Which currency, if any, to format as. :type currency: str|None :param currency_digits: Whether or not to use the currency's precision. If false, the pattern's precision is used. :type currency_digits: bool :param decimal_quantization: Whether decimal numbers should be forcibly quantized to produce a formatted output strictly matching the CLDR definition for the locale. :type decimal_quantization: bool :param force_frac: DEPRECATED - a forced override for `self.frac_prec` for a single formatting invocation. :return: Formatted decimal string. :rtype: str """ if not isinstance(value, decimal.Decimal): value = decimal.Decimal(str(value)) value = value.scaleb(self.scale) # Separate the absolute value from its sign. is_negative = int(value.is_signed()) value = abs(value).normalize() # Prepare scientific notation metadata. if self.exp_prec: value, exp, exp_sign = self.scientific_notation_elements(value, locale) # Adjust the precision of the fractional part and force it to the # currency's if necessary. if force_frac: # TODO (3.x?): Remove this parameter warnings.warn('The force_frac parameter to NumberPattern.apply() is deprecated.', DeprecationWarning) frac_prec = force_frac elif currency and currency_digits: frac_prec = (get_currency_precision(currency), ) * 2 else: frac_prec = self.frac_prec # Bump decimal precision to the natural precision of the number if it # exceeds the one we're about to use. This adaptative precision is only # triggered if the decimal quantization is disabled or if a scientific # notation pattern has a missing mandatory fractional part (as in the # default '#E0' pattern). This special case has been extensively # discussed at https://github.com/python-babel/babel/pull/494#issuecomment-307649969 . if not decimal_quantization or (self.exp_prec and frac_prec == (0, 0)): frac_prec = (frac_prec[0], max([frac_prec[1], get_decimal_precision(value)])) # Render scientific notation. if self.exp_prec: number = ''.join([ self._quantize_value(value, locale, frac_prec, group_separator), get_exponential_symbol(locale), exp_sign, # type: ignore # exp_sign is always defined here self._format_int(str(exp), self.exp_prec[0], self.exp_prec[1], locale) # type: ignore # exp is always defined here ]) # Is it a significant digits pattern? elif '@' in self.pattern: text = self._format_significant(value, self.int_prec[0], self.int_prec[1]) a, sep, b = text.partition(".") number = self._format_int(a, 0, 1000, locale) if sep: number += get_decimal_symbol(locale) + b # A normal number pattern. else: number = self._quantize_value(value, locale, frac_prec, group_separator) retval = ''.join([ self.prefix[is_negative], number if self.number_pattern != '' else '', self.suffix[is_negative]]) if '¤' in retval and currency is not None: retval = retval.replace('¤¤¤', get_currency_name(currency, value, locale)) retval = retval.replace('¤¤', currency.upper()) retval = retval.replace('¤', get_currency_symbol(currency, locale)) # remove single quotes around text, except for doubled single quotes # which are replaced with a single quote retval = re.sub(r"'([^']*)'", lambda m: m.group(1) or "'", retval) return retval # # This is one tricky piece of code. The idea is to rely as much as possible # on the decimal module to minimize the amount of code. # # Conceptually, the implementation of this method can be summarized in the # following steps: # # - Move or shift the decimal point (i.e. the exponent) so the maximum # amount of significant digits fall into the integer part (i.e. to the # left of the decimal point) # # - Round the number to the nearest integer, discarding all the fractional # part which contained extra digits to be eliminated # # - Convert the rounded integer to a string, that will contain the final # sequence of significant digits already trimmed to the maximum # # - Restore the original position of the decimal point, potentially # padding with zeroes on either side # def _format_significant(self, value: decimal.Decimal, minimum: int, maximum: int) -> str: exp = value.adjusted() scale = maximum - 1 - exp digits = str(value.scaleb(scale).quantize(decimal.Decimal(1))) if scale <= 0: result = digits + '0' * -scale else: intpart = digits[:-scale] i = len(intpart) j = i + max(minimum - i, 0) result = "{intpart}.{pad:0<{fill}}{fracpart}{fracextra}".format( intpart=intpart or '0', pad='', fill=-min(exp + 1, 0), fracpart=digits[i:j], fracextra=digits[j:].rstrip('0'), ).rstrip('.') return result def _format_int(self, value: str, min: int, max: int, locale: Locale | str | None) -> str: width = len(value) if width < min: value = '0' * (min - width) + value gsize = self.grouping[0] ret = '' symbol = get_group_symbol(locale) while len(value) > gsize: ret = symbol + value[-gsize:] + ret value = value[:-gsize] gsize = self.grouping[1] return value + ret def _quantize_value(self, value: decimal.Decimal, locale: Locale | str | None, frac_prec: tuple[int, int], group_separator: bool) -> str: # If the number is +/-Infinity, we can't quantize it if value.is_infinite(): return get_infinity_symbol(locale) quantum = get_decimal_quantum(frac_prec[1]) rounded = value.quantize(quantum) a, sep, b = f"{rounded:f}".partition(".") integer_part = a if group_separator: integer_part = self._format_int(a, self.int_prec[0], self.int_prec[1], locale) number = integer_part + self._format_frac(b or '0', locale, frac_prec) return number def _format_frac(self, value: str, locale: Locale | str | None, force_frac: tuple[int, int] | None = None) -> str: min, max = force_frac or self.frac_prec if len(value) < min: value += ('0' * (min - len(value))) if max == 0 or (min == 0 and int(value) == 0): return '' while len(value) > min and value[-1] == '0': value = value[:-1] return get_decimal_symbol(locale) + value