-# -*- coding: utf-8 -*-
+# This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
+# Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
+#
from datetime import date
from lxml import etree
from django.core.management.base import BaseCommand
from isbn.models import ISBNPool, ONIXRecord
+from librarian import XMLNamespace
-ONIXNS = '{http://ns.editeur.org/onix/3.0/reference}'
+
+ONIXNS = XMLNamespace('http://ns.editeur.org/onix/3.0/reference')
DIRECT_FIELDS = {
'product_form': 'ProductForm',
'imprint': 'ImprintName',
}
-UNNAMED = u'Autor nieznany'
+UNKNOWN = 'Autor nieznany'
def parse_date(date_str):
def get_descendants(element, tags):
- if isinstance(tags, basestring):
+ if isinstance(tags, str):
tags = [tags]
- return element.findall('.//' + '/'.join(ONIXNS + tag for tag in tags))
+ return element.findall('.//' + '/'.join(ONIXNS(tag) for tag in tags))
def get_field(element, tags, allow_multiple=False):
class Command(BaseCommand):
help = "Import data from ONIX."
- args = 'filename'
- def handle(self, filename, *args, **options):
+ def add_arguments(self, parser):
+ parser.add_argument('filename')
+
+ def handle(self, **options):
+ filename = options['filename']
tree = etree.parse(open(filename))
for product in get_descendants(tree, 'Product'):
isbn = get_field(product, ['ProductIdentifier', 'IDValue'])
assert len(isbn) == 13
- pool = ISBNPool.objects.get(prefix__in=[isbn[:i] for i in xrange(8, 11)])
+ pool = ISBNPool.objects.get(prefix__in=[isbn[:i] for i in range(8, 11)])
contributors = [
self.parse_contributor(contributor)
for contributor in get_descendants(product, 'Contributor')]
get_field(product, ['PublishingDate', 'Date'], allow_multiple=True)),
'contributors': contributors,
}
- for field, tag in DIRECT_FIELDS.iteritems():
+ for field, tag in DIRECT_FIELDS.items():
record_data[field] = get_field(product, tag) or ''
record = ONIXRecord.objects.create(**record_data)
ONIXRecord.objects.filter(pk=record.pk).update(datestamp=parse_date(product.attrib['datestamp']))
contributor_data = {
'role': get_field(contributor, 'ContributorRole'),
}
- for key, value in data.iteritems():
+ for key, value in data.items():
if value:
contributor_data[key] = value
- if contributor_data.get('name') == UNNAMED:
+ if contributor_data.get('name') == UNKNOWN:
del contributor_data['name']
contributor_data['unnamed'] = '01'
for date_elem in get_descendants(contributor, 'ContributorDate'):