mirror of
https://github.com/janeczku/calibre-web
synced 2024-11-16 14:54:55 +00:00
bbf6d9b026
Bugfix for feeds - removed categories related and up - load new books now working - category random now working login page is free of non accessible elements boolean custom column is vivible in UI books with only with certain languages can be shown book shelfs can be deleted from UI Anonymous user view is more resticted Added browse of series in sidebar Dependencys in vendor folder are updated to newer versions (licencs files are now present) Bugfix editing Authors names Made upload on windows working
266 lines
8.6 KiB
Python
266 lines
8.6 KiB
Python
# coding=utf-8
|
|
"""
|
|
Python library for ISO 639 standard
|
|
|
|
Copyright (c) 2014-2016 Mikael Karlsson (CSC - IT Center for Science Ltd.).
|
|
Licensed under AGPLv3.
|
|
"""
|
|
|
|
# Fix for Python 3.0 - 3.2
|
|
if not __package__:
|
|
__package__ = __name__.split('.')[0]
|
|
|
|
|
|
def _fabtabular():
|
|
"""
|
|
This function retrieves the ISO 639 and inverted names datasets as tsv files and returns them as lists.
|
|
"""
|
|
import csv
|
|
import sys
|
|
from pkg_resources import resource_filename
|
|
|
|
data = resource_filename(__package__, 'iso-639-3.tab')
|
|
inverted = resource_filename(__package__, 'iso-639-3_Name_Index.tab')
|
|
macro = resource_filename(__package__, 'iso-639-3-macrolanguages.tab')
|
|
part5 = resource_filename(__package__, 'iso639-5.tsv')
|
|
part2 = resource_filename(__package__, 'iso639-2.tsv')
|
|
part1 = resource_filename(__package__, 'iso639-1.tsv')
|
|
|
|
# if sys.version_info[0] == 2:
|
|
# from urllib2 import urlopen
|
|
# from contextlib import closing
|
|
# data_fo = closing(urlopen('http://www-01.sil.org/iso639-3/iso-639-3.tab'))
|
|
# inverted_fo = closing(urlopen('http://www-01.sil.org/iso639-3/iso-639-3_Name_Index.tab'))
|
|
# else:
|
|
# from urllib.request import urlopen
|
|
# import io
|
|
# data_fo = io.StringIO(urlopen('http://www-01.sil.org/iso639-3/iso-639-3.tab').read().decode())
|
|
# inverted_fo = io.StringIO(urlopen('http://www-01.sil.org/iso639-3/iso-639-3_Name_Index.tab').read().decode())
|
|
|
|
if sys.version_info[0] == 3:
|
|
from functools import partial
|
|
|
|
global open
|
|
open = partial(open, encoding='utf-8')
|
|
|
|
data_fo = open(data)
|
|
inverted_fo = open(inverted)
|
|
macro_fo = open(macro)
|
|
part5_fo = open(part5)
|
|
part2_fo = open(part2)
|
|
part1_fo = open(part1)
|
|
with data_fo as u:
|
|
with inverted_fo as i:
|
|
with macro_fo as m:
|
|
with part5_fo as p5:
|
|
with part2_fo as p2:
|
|
with part1_fo as p1:
|
|
return (list(csv.reader(u, delimiter='\t'))[1:],
|
|
list(csv.reader(i, delimiter='\t'))[1:],
|
|
list(csv.reader(m, delimiter='\t'))[1:],
|
|
list(csv.reader(p5, delimiter='\t'))[1:],
|
|
list(csv.reader(p2, delimiter='\t'))[1:],
|
|
list(csv.reader(p1, delimiter='\t'))[1:])
|
|
|
|
|
|
class _Language(object):
|
|
"""
|
|
This class represents a language. It provides pycountry language class compatibility.
|
|
"""
|
|
|
|
def __init__(self, part3, part2b, part2t, part1, name, inverted, macro, names, part5):
|
|
self.part3 = part3
|
|
self.part2b = part2b
|
|
self.part2t = part2t
|
|
self.part1 = part1
|
|
self.name = name
|
|
self.inverted = inverted
|
|
self.macro = macro
|
|
self.names = names
|
|
self.part5 = part5
|
|
|
|
def __getattr__(self, item):
|
|
compat = {
|
|
'alpha2': self.part1,
|
|
'bibliographic': self.part2b,
|
|
'terminology': self.part2t,
|
|
}
|
|
if item not in compat:
|
|
raise AttributeError("'{o}' object has no attribute '{a}'".format(o=type(self).__name__, a=item))
|
|
return compat[item]
|
|
|
|
|
|
class lazy_property(object):
|
|
"""
|
|
Implements a lazy property decorator, that overwrites itself/property with value
|
|
"""
|
|
|
|
def __init__(self, f):
|
|
self.f = f
|
|
self.name = f.__name__
|
|
|
|
def __get__(self, instance, owner=None):
|
|
if instance is None:
|
|
return self
|
|
val = self.f(instance)
|
|
setattr(instance, self.name, val)
|
|
return val
|
|
|
|
|
|
class Iso639(object):
|
|
"""
|
|
This class is a close to drop-in replacement for pycountry.languages.
|
|
But unlike pycountry.languages it also supports ISO 639-3.
|
|
|
|
It implements the Singleton design pattern for performance reasons.
|
|
Is uses lazy properties for faster import time.
|
|
"""
|
|
|
|
def __new__(cls):
|
|
if not hasattr(cls, '__instance'):
|
|
setattr(cls, '__instance', super(cls, cls).__new__(cls))
|
|
return getattr(cls, '__instance')
|
|
|
|
def __len__(self):
|
|
return len(self.languages)
|
|
|
|
def __iter__(self):
|
|
return iter(self.languages)
|
|
|
|
def __getattr__(self, item):
|
|
compat = {
|
|
'alpha2': self.part1,
|
|
'bibliographic': self.part2b,
|
|
'terminology': self.part2t,
|
|
}
|
|
if item not in compat:
|
|
raise AttributeError("'{o}' object has no attribute '{a}'".format(o=type(self).__name__, a=item))
|
|
return compat[item]
|
|
|
|
@lazy_property
|
|
def languages(self):
|
|
def generate():
|
|
# All of part3 and matching part2
|
|
for a, b, c, d, _, _, e, _ in l:
|
|
inv = alt[a].pop(e)
|
|
yield _Language(a, b, c,
|
|
d if d in p1c else '', # Fixes 'sh'
|
|
e, inv,
|
|
m.get(a, [''])[0],
|
|
list(alt[a].items()),
|
|
'')
|
|
p2.pop(b, None)
|
|
p2.pop(c, None)
|
|
|
|
# All of part5 and matching part2
|
|
for _, a, b, _ in p5:
|
|
yield _Language('',
|
|
a if a in p2 else '',
|
|
a if a in p2 else '',
|
|
p1n.get(b, ['', ''])[1],
|
|
b, '', '', '', a)
|
|
p2.pop(a, None)
|
|
|
|
# Rest of part2
|
|
p2.pop('qaa-qtz', None) # Is not a real code, but a range
|
|
for _, a, b, _ in p2.values():
|
|
n = [x.strip() for x in b.split('|')]
|
|
yield _Language('', a, a,
|
|
p1n.get(b, ['', ''])[1],
|
|
n[0], '', '', zip(n[1:], n[1:]), '')
|
|
|
|
import collections
|
|
|
|
l, i, m, p5, p2, p1 = _fabtabular()
|
|
alt = collections.defaultdict(dict)
|
|
for x in i:
|
|
alt[x[0]][x[1]] = x[2]
|
|
m = dict((x[1], x) for x in m)
|
|
p2 = dict((x[1], x) for x in p2)
|
|
p1c = dict((x[1], x) for x in p1)
|
|
p1n = dict((x[2].split('|')[0].strip(), x) for x in p1)
|
|
return list(generate())
|
|
|
|
@lazy_property
|
|
def part3(self):
|
|
return dict((x.part3, x) for x in self.languages if x.part3)
|
|
|
|
@lazy_property
|
|
def part2b(self):
|
|
return dict((x.part2b, x) for x in self.languages if x.part2b)
|
|
|
|
@lazy_property
|
|
def part2t(self):
|
|
return dict((x.part2t, x) for x in self.languages if x.part2t)
|
|
|
|
@lazy_property
|
|
def part1(self):
|
|
return dict((x.part1, x) for x in self.languages if x.part1)
|
|
|
|
@lazy_property
|
|
def part5(self):
|
|
return dict((x.part5, x) for x in self.languages if x.part5)
|
|
|
|
@lazy_property
|
|
def name(self):
|
|
def gen():
|
|
for x in self.languages:
|
|
if x.name:
|
|
yield x.name, x
|
|
for n in x.names:
|
|
yield n[0], x
|
|
|
|
return dict(gen())
|
|
|
|
@lazy_property
|
|
def inverted(self):
|
|
return dict((x.inverted, x) for x in self.languages if x.inverted)
|
|
|
|
@lazy_property
|
|
def macro(self):
|
|
import collections
|
|
|
|
m = collections.defaultdict(list)
|
|
for x in self.languages:
|
|
if x.macro:
|
|
m[x.macro].append(x)
|
|
return dict(m)
|
|
|
|
@lazy_property
|
|
def retired(self):
|
|
"""
|
|
Function for generating retired languages. Returns a dict('code', (datetime, [language, ...], 'description')).
|
|
"""
|
|
|
|
def gen():
|
|
import csv
|
|
import re
|
|
from datetime import datetime
|
|
from pkg_resources import resource_filename
|
|
|
|
with open(resource_filename(__package__, 'iso-639-3_Retirements.tab')) as rf:
|
|
rtd = list(csv.reader(rf, delimiter='\t'))[1:]
|
|
rc = [r[0] for r in rtd]
|
|
for i, _, _, m, s, d in rtd:
|
|
d = datetime.strptime(d, '%Y-%m-%d')
|
|
if not m:
|
|
m = re.findall('\[([a-z]{3})\]', s)
|
|
if m:
|
|
m = [m] if isinstance(m, str) else m
|
|
yield i, (d, [self.get(part3=x) for x in m if x not in rc], s)
|
|
else:
|
|
yield i, (d, [], s)
|
|
|
|
yield 'sh', self.get(part3='hbs') # Add 'sh' as deprecated
|
|
|
|
return dict(gen())
|
|
|
|
def get(self, **kwargs):
|
|
"""
|
|
Simple getter function for languages. Takes 1 keyword/value and returns 1 language object.
|
|
"""
|
|
if not len(kwargs) == 1:
|
|
raise AttributeError('Only one keyword expected')
|
|
key, value = kwargs.popitem()
|
|
return getattr(self, key)[value]
|