# This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
# Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
#
-from django.conf.urls import patterns, url
+from django.conf.urls import url, include
from django.views.decorators.csrf import csrf_exempt
from django.views.generic import TemplateView
-from piston.authentication import OAuthAuthentication, oauth_access_token
-from piston.resource import Resource
-from ssify import ssi_included
-from api import handlers
-from api.helpers import CsrfExemptResource
+import catalogue.views
+from stats.utils import piwik_track_view
+from . import views
-auth = OAuthAuthentication(realm="Wolne Lektury")
-book_list_resource = CsrfExemptResource(handler=handlers.BooksHandler, authentication=auth)
-ebook_list_resource = Resource(handler=handlers.EBooksHandler)
-# book_list_resource = Resource(handler=handlers.BooksHandler)
-book_resource = Resource(handler=handlers.BookDetailHandler)
-filter_book_resource = Resource(handler=handlers.FilterBooksHandler)
+urlpatterns = [
+ url(r'^oauth/request_token/$', csrf_exempt(views.OAuth1RequestTokenView.as_view())),
+ url(r'^oauth/authorize/$', views.oauth_user_auth, name='oauth_user_auth'),
+ url(r'^oauth/access_token/$', csrf_exempt(views.OAuth1AccessTokenView.as_view())),
-collection_resource = Resource(handler=handlers.CollectionDetailHandler)
-collection_list_resource = Resource(handler=handlers.CollectionsHandler)
-
-tag_list_resource = Resource(handler=handlers.TagsHandler)
-tag_resource = Resource(handler=handlers.TagDetailHandler)
-
-fragment_resource = Resource(handler=handlers.FragmentDetailHandler)
-fragment_list_resource = Resource(handler=handlers.FragmentsHandler)
-
-picture_resource = CsrfExemptResource(handler=handlers.PictureHandler, authentication=auth)
-
-
-tags_re = r'^(?P<tags>(?:(?:[a-z0-9-]+/){2}){0,6})'
-paginate_re = r'(?:before/(?P<before>[a-z0-9-]+)/)?(?:after/(?P<after>[a-z0-9-]+)/)?(?:count/(?P<count>[0-9]+)/)?$'
-
-
-@ssi_included
-def incl(request, model, pk, emitter_format):
- resource = {
- 'book': book_list_resource,
- 'fragment': fragment_list_resource,
- 'tag': tag_list_resource,
- }[model]
- request.piwik_track = False
- resp = resource(request, pk=pk, emitter_format=emitter_format)
- if emitter_format == 'xml':
- # Ugly, but quick way of stripping <?xml?> header and <response> tags.
- resp.content = resp.content[49:-11]
- return resp
-
-
-urlpatterns = patterns(
- 'piston.authentication',
- url(r'^oauth/request_token/$', 'oauth_request_token'),
- url(r'^oauth/authorize/$', 'oauth_user_auth'),
- url(r'^oauth/access_token/$', csrf_exempt(oauth_access_token)),
-
-) + patterns(
- '',
url(r'^$', TemplateView.as_view(template_name='api/main.html'), name='api'),
- url(r'^include/(?P<model>book|fragment|tag)/(?P<pk>\d+)\.(?P<lang>.+)\.(?P<emitter_format>xml|json)$',
- incl, name='api_include'),
# info boxes (used by mobile app)
- url(r'book/(?P<book_id>\d*?)/info\.html$', 'catalogue.views.book_info'),
- url(r'tag/(?P<tag_id>\d*?)/info\.html$', 'catalogue.views.tag_info'),
-
- # books by collections
- url(r'^collections/$', collection_list_resource, name="api_collections"),
- url(r'^collections/(?P<slug>[^/]+)/$', collection_resource, name="api_collection"),
-
- # objects details
- url(r'^books/(?P<book>[a-z0-9-]+)/$', book_resource, name="api_book"),
- url(r'^(?P<category>[a-z0-9-]+)/(?P<slug>[a-z0-9-]+)/$',
- tag_resource, name="api_tag"),
- url(r'^books/(?P<book>[a-z0-9-]+)/fragments/(?P<anchor>[a-z0-9-]+)/$',
- fragment_resource, name="api_fragment"),
-
- # books by tags
- url(tags_re + r'books/' + paginate_re,
- book_list_resource, name='api_book_list'),
- url(tags_re + r'ebooks/' + paginate_re,
- ebook_list_resource, name='api_ebook_list'),
- url(tags_re + r'parent_books/' + paginate_re,
- book_list_resource, {"top_level": True}, name='api_parent_book_list'),
- url(tags_re + r'parent_ebooks/' + paginate_re,
- ebook_list_resource, {"top_level": True}, name='api_parent_ebook_list'),
- url(tags_re + r'audiobooks/' + paginate_re,
- book_list_resource, {"audiobooks": True}, name='api_audiobook_list'),
- url(tags_re + r'daisy/' + paginate_re,
- book_list_resource, {"daisy": True}, name='api_daisy_list'),
-
- url(r'^recommended/' + paginate_re, book_list_resource, {"recommended": True}, name='api_recommended_list'),
- url(r'^newest/', book_list_resource, {"newest": True, "count": 20}, name='api_newest_list'),
- url(r'^filter-books/(?:title-part/(?P<title_part>[^/]+)/)?(?:author-part/(?P<author_part>[^/]+)/)?'
- r'(?:lektura/(?P<is_lektura>(?:true|false))/)?(?:audiobook/(?P<is_audiobook>(?:true|false))/)?'
- + paginate_re, filter_book_resource, name='api_filter_books'),
-
- url(r'^pictures/$', picture_resource),
-
- # fragments by book, tags, themes
- # this should be paged
- url(r'^(?P<tags>(?:(?:[a-z0-9-]+/){2}){1,6})fragments/$', fragment_list_resource),
-
- # tags by category
- url(r'^(?P<category>[a-z0-9-]+)/$', tag_list_resource, name='api_tag_list'),
-)
+ url(r'book/(?P<book_id>\d*?)/info\.html$', catalogue.views.book_info),
+ url(r'tag/(?P<tag_id>\d*?)/info\.html$', catalogue.views.tag_info),
+
+ # reading data
+ url(r'^reading/(?P<slug>[a-z0-9-]+)/$',
+ piwik_track_view(views.BookUserDataView.as_view()),
+ name='api_reading'),
+ url(r'^reading/(?P<slug>[a-z0-9-]+)/(?P<state>[a-z]+)/$',
+ piwik_track_view(views.BookUserDataView.as_view()),
+ name='api_reading'),
+ url(r'^username/$',
+ piwik_track_view(views.UserView.as_view()),
+ name='api_username'),
+
+ url(r'^blog$',
+ piwik_track_view(views.BlogView.as_view())),
+
+ url(r'^pictures/', include('picture.api.urls')),
+ url(r'^', include('social.api.urls')),
+ url(r'^', include('catalogue.api.urls')),
+]