1
0
mirror of https://github.com/janeczku/calibre-web synced 2024-12-11 02:30:30 +00:00
calibre-web/vendor/babel/messages/plurals.py

255 lines
6.5 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
"""
babel.messages.plurals
~~~~~~~~~~~~~~~~~~~~~~
Plural form definitions.
:copyright: (c) 2013 by the Babel Team.
:license: BSD, see LICENSE for more details.
"""
from babel.core import default_locale, Locale
from operator import itemgetter
# XXX: remove this file, duplication with babel.plural
LC_CTYPE = default_locale('LC_CTYPE')
PLURALS = {
# Afar
# 'aa': (),
# Abkhazian
# 'ab': (),
# Avestan
# 'ae': (),
# Afrikaans - From Pootle's PO's
'af': (2, '(n != 1)'),
# Akan
# 'ak': (),
# Amharic
# 'am': (),
# Aragonese
# 'an': (),
# Arabic - From Pootle's PO's
'ar': (6, '(n==0 ? 0 : n==1 ? 1 : n==2 ? 2 : n>=3 && n<=10 ? 3 : n>=11 && n<=99 ? 4 : 5)'),
# Assamese
# 'as': (),
# Avaric
# 'av': (),
# Aymara
# 'ay': (),
# Azerbaijani
# 'az': (),
# Bashkir
# 'ba': (),
# Belarusian
# 'be': (),
# Bulgarian - From Pootle's PO's
'bg': (2, '(n != 1)'),
# Bihari
# 'bh': (),
# Bislama
# 'bi': (),
# Bambara
# 'bm': (),
# Bengali - From Pootle's PO's
'bn': (2, '(n != 1)'),
# Tibetan - as discussed in private with Andrew West
'bo': (1, '0'),
# Breton
# 'br': (),
# Bosnian
# 'bs': (),
# Catalan - From Pootle's PO's
'ca': (2, '(n != 1)'),
# Chechen
# 'ce': (),
# Chamorro
# 'ch': (),
# Corsican
# 'co': (),
# Cree
# 'cr': (),
# Czech
'cs': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Church Slavic
# 'cu': (),
# Chuvash
'cv': (1, '0'),
# Welsh
'cy': (5, '(n==1 ? 1 : n==2 ? 2 : n==3 ? 3 : n==6 ? 4 : 0)'),
# Danish
'da': (2, '(n != 1)'),
# German
'de': (2, '(n != 1)'),
# Divehi
# 'dv': (),
# Dzongkha
'dz': (1, '0'),
# Greek
'el': (2, '(n != 1)'),
# English
'en': (2, '(n != 1)'),
# Esperanto
'eo': (2, '(n != 1)'),
# Spanish
'es': (2, '(n != 1)'),
# Estonian
'et': (2, '(n != 1)'),
# Basque - From Pootle's PO's
'eu': (2, '(n != 1)'),
# Persian - From Pootle's PO's
'fa': (1, '0'),
# Finnish
'fi': (2, '(n != 1)'),
# French
'fr': (2, '(n > 1)'),
# Friulian - From Pootle's PO's
'fur': (2, '(n > 1)'),
# Irish
'ga': (3, '(n==1 ? 0 : n==2 ? 1 : 2)'),
# Galician - From Pootle's PO's
'gl': (2, '(n != 1)'),
# Hausa - From Pootle's PO's
'ha': (2, '(n != 1)'),
# Hebrew
'he': (2, '(n != 1)'),
# Hindi - From Pootle's PO's
'hi': (2, '(n != 1)'),
# Croatian
'hr': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Hungarian
'hu': (1, '0'),
# Armenian - From Pootle's PO's
'hy': (1, '0'),
# Icelandic - From Pootle's PO's
'is': (2, '(n != 1)'),
# Italian
'it': (2, '(n != 1)'),
# Japanese
'ja': (1, '0'),
# Georgian - From Pootle's PO's
'ka': (1, '0'),
# Kongo - From Pootle's PO's
'kg': (2, '(n != 1)'),
# Khmer - From Pootle's PO's
'km': (1, '0'),
# Korean
'ko': (1, '0'),
# Kurdish - From Pootle's PO's
'ku': (2, '(n != 1)'),
# Lao - Another member of the Tai language family, like Thai.
'lo': (1, '0'),
# Lithuanian
'lt': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Latvian
'lv': (3, '(n%10==1 && n%100!=11 ? 0 : n != 0 ? 1 : 2)'),
# Maltese - From Pootle's PO's
'mt': (4, '(n==1 ? 0 : n==0 || ( n%100>1 && n%100<11) ? 1 : (n%100>10 && n%100<20 ) ? 2 : 3)'),
# Norwegian Bokmål
'nb': (2, '(n != 1)'),
# Dutch
'nl': (2, '(n != 1)'),
# Norwegian Nynorsk
'nn': (2, '(n != 1)'),
# Norwegian
'no': (2, '(n != 1)'),
# Punjabi - From Pootle's PO's
'pa': (2, '(n != 1)'),
# Polish
'pl': (3, '(n==1 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Portuguese
'pt': (2, '(n != 1)'),
# Brazilian
'pt_BR': (2, '(n > 1)'),
# Romanian - From Pootle's PO's
'ro': (3, '(n==1 ? 0 : (n==0 || (n%100 > 0 && n%100 < 20)) ? 1 : 2)'),
# Russian
'ru': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Slovak
'sk': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Slovenian
'sl': (4, '(n%100==1 ? 0 : n%100==2 ? 1 : n%100==3 || n%100==4 ? 2 : 3)'),
# Serbian - From Pootle's PO's
'sr': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Southern Sotho - From Pootle's PO's
'st': (2, '(n != 1)'),
# Swedish
'sv': (2, '(n != 1)'),
# Thai
'th': (1, '0'),
# Turkish
'tr': (1, '0'),
# Ukrainian
'uk': (3, '(n%10==1 && n%100!=11 ? 0 : n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2)'),
# Venda - From Pootle's PO's
've': (2, '(n != 1)'),
# Vietnamese - From Pootle's PO's
'vi': (1, '0'),
# Xhosa - From Pootle's PO's
'xh': (2, '(n != 1)'),
# Chinese - From Pootle's PO's
'zh_CN': (1, '0'),
'zh_HK': (1, '0'),
'zh_TW': (1, '0'),
}
DEFAULT_PLURAL = (2, '(n != 1)')
class _PluralTuple(tuple):
"""A tuple with plural information."""
__slots__ = ()
num_plurals = property(itemgetter(0), doc="""
The number of plurals used by the locale.""")
plural_expr = property(itemgetter(1), doc="""
The plural expression used by the locale.""")
plural_forms = property(lambda x: 'npurals=%s; plural=%s' % x, doc="""
The plural expression used by the catalog or locale.""")
def __str__(self):
return self.plural_forms
def get_plural(locale=LC_CTYPE):
"""A tuple with the information catalogs need to perform proper
pluralization. The first item of the tuple is the number of plural
forms, the second the plural expression.
>>> get_plural(locale='en')
(2, '(n != 1)')
>>> get_plural(locale='ga')
(3, '(n==1 ? 0 : n==2 ? 1 : 2)')
The object returned is a special tuple with additional members:
>>> tup = get_plural("ja")
>>> tup.num_plurals
1
>>> tup.plural_expr
'0'
>>> tup.plural_forms
'npurals=1; plural=0'
Converting the tuple into a string prints the plural forms for a
gettext catalog:
>>> str(tup)
'npurals=1; plural=0'
"""
locale = Locale.parse(locale)
try:
tup = PLURALS[str(locale)]
except KeyError:
try:
tup = PLURALS[locale.language]
except KeyError:
tup = DEFAULT_PLURAL
return _PluralTuple(tup)