1
0
mirror of https://github.com/gryf/ebook-converter.git synced 2026-01-15 16:24:13 +01:00
Files
ebook-converter/ebook_converter/utils/localization.py

154 lines
3.9 KiB
Python

import json
import pkg_resources
def get_lang():
return 'en_US'
def is_rtl():
return get_lang()[:2].lower() in {'he', 'ar'}
_lang_trans = None
lcdata = {'abday': ('Sun', 'Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat'),
'abmon': ('Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug',
'Sep', 'Oct', 'Nov', 'Dec'),
'd_fmt': '%m/%d/%Y',
'd_t_fmt': '%a %d %b %Y %r %Z',
'day': ('Sunday', 'Monday', 'Tuesday', 'Wednesday', 'Thursday',
'Friday', 'Saturday'),
'mon': ('January', 'February', 'March', 'April', 'May', 'June',
'July', 'August', 'September', 'October', 'November',
'December'),
'noexpr': '^[nN].*',
'radixchar': '.',
't_fmt': '%r',
't_fmt_ampm': '%I:%M:%S %p',
'thousep': ',',
'yesexpr': '^[yY].*'}
_iso639 = None
def _load_iso639():
global _iso639
# NOTE(gryf): msgpacked data was originally added for speed purposes. In
# my tests, I cannot see any speed gain either on python2 or python3. It
# is even slower (around 4-8 times), than just using code below (which is
# excerpt form Calibre transform code which is executed during Calibre
# build).
if _iso639 is None:
src = pkg_resources.resource_filename('ebook_converter',
'data/iso_639-3.json')
with open(src, 'rb') as f:
root = json.load(f)
entries = root['639-3']
by_2 = {}
by_3 = {}
m2to3 = {}
m3to2 = {}
nm = {}
codes2, codes3 = set(), set()
for x in entries:
two = x.get('alpha_2')
threeb = x.get('alpha_3')
if threeb is None:
continue
name = x.get('inverted_name') or x.get('name')
if not name or name[0] in '!~=/\'"':
continue
if two is not None:
by_2[two] = name
codes2.add(two)
m2to3[two] = threeb
m3to2[threeb] = two
codes3.add(threeb)
by_3[threeb] = name
base_name = name.lower()
nm[base_name] = threeb
_iso639 = {'by_2': by_2,
'by_3': by_3,
'codes2': codes2,
'codes3': codes3,
'2to3': m2to3,
'3to2': m3to2,
'name_map': nm}
return _iso639
def get_iso_language(lang_trans, lang):
iso639 = _load_iso639()
ans = lang
lang = lang.split('_')[0].lower()
if len(lang) == 2:
ans = iso639['by_2'].get(lang, ans)
elif len(lang) == 3:
if lang in iso639['by_3']:
ans = iso639['by_3'][lang]
return lang_trans(ans)
def langcode_to_name(lc, localize=True):
iso639 = _load_iso639()
try:
return iso639['by_3'][lc]
except Exception:
pass
return lc
def canonicalize_lang(raw):
if not raw:
return None
if not isinstance(raw, str):
raw = raw.decode('utf-8', 'ignore')
raw = raw.lower().strip()
if not raw:
return None
raw = raw.replace('_', '-').partition('-')[0].strip()
if not raw:
return None
iso639 = _load_iso639()
m2to3 = iso639['2to3']
if len(raw) == 2:
ans = m2to3.get(raw, None)
if ans is not None:
return ans
elif len(raw) == 3:
if raw in iso639['by_3']:
return raw
return iso639['name_map'].get(raw, None)
def lang_as_iso639_1(name_or_code):
code = canonicalize_lang(name_or_code)
if code is not None:
iso639 = _load_iso639()
return iso639['3to2'].get(code, None)
_udc = None
def get_udc():
global _udc
if _udc is None:
from ebook_converter.ebooks.unihandecode import Unihandecoder
_udc = Unihandecoder(lang=get_lang())
return _udc
def localize_user_manual_link(url):
return url