Implementation of basic conversion between Django forms and ExtJS forms.
[philo.git] / contrib / penfield / utils.py
index 0318e5e..7b5e946 100644 (file)
@@ -1,33 +1,81 @@
-from django.core.paginator import Paginator
+from django.utils.feedgenerator import Atom1Feed, Rss201rev2Feed
+from django.conf.urls.defaults import url, patterns
+from django.core.urlresolvers import reverse
+from django.http import HttpResponse
 
 
-def paginate(objects, per_page=None, page_number=1):
+class FeedMultiViewMixin(object):
        """
-       Given a list of objects, return a (page, obj_list) tuple.
+       This mixin provides common methods for adding feeds to multiviews. In order to use this mixin,
+       the multiview must define feed_title (probably as properties that return values
+       on related objects.) feed_description may also be defined; it defaults to an empty string.
        """
-       try:
-               per_page = int(per_page)
-       except (TypeError, ValueError):
-               # Then either it wasn't set or it was set to an invalid value
-               return None, objects
+       feed_suffix = 'feed'
+       feeds_enabled = True
+       atom_feed = Atom1Feed
+       rss_feed = Rss201rev2Feed
+       feed_description = ''
        
-       # There also shouldn't be pagination if the list is too short. Try count()
-       # first - good chance it's a queryset, where count is more efficient.
-       try:
-               if objects.count() <= per_page:
-                       return None, objects
-       except AttributeError:
-               if len(objects) <= per_page:
-                       return None, objects
+       def page_view(self, func, page, list_var='entries'):
+               """
+               Wraps an object-fetching function and renders the results as a page.
+               """
+               def inner(request, node=None, extra_context=None, **kwargs):
+                       objects, extra_context = func(request=request, node=node, extra_context=extra_context, **kwargs)
+
+                       context = self.get_context()
+                       context.update(extra_context or {})
+
+                       if 'page' in kwargs or 'page' in request.GET:
+                               page_num = kwargs.get('page', request.GET.get('page', 1))
+                               paginator, paginated_page, objects = paginate(objects, self.per_page, page_num)
+                               context.update({'paginator': paginator, 'paginated_page': paginated_page, list_var: objects})
+                       else:
+                               context.update({list_var: objects})
+
+                       return page.render_to_response(node, request, extra_context=context)
+
+               return inner
+       
+       def feed_view(self, func, reverse_name):
+               """
+               Wraps an object-fetching function and renders the results as a rss or atom feed.
+               """
+               def inner(request, node=None, extra_context=None, **kwargs):
+                       objects, extra_context = func(request=request, node=node, extra_context=extra_context, **kwargs)
+       
+                       if 'HTTP_ACCEPT' in request.META and 'rss' in request.META['HTTP_ACCEPT'] and 'atom' not in request.META['HTTP_ACCEPT']:
+                               feed_type = 'rss'
+                       else:
+                               feed_type = 'atom'
+                       
+                       feed = self.get_feed(feed_type, request, node, kwargs, reverse_name)
+                       
+                       for obj in objects:
+                               feed.add_item(obj.title, '/%s/%s/' % (node.get_absolute_url().strip('/'), self.get_subpath(obj).strip('/')), description=self.get_obj_description(obj))
+       
+                       response = HttpResponse(mimetype=feed.mime_type)
+                       feed.write(response, 'utf-8')
+                       return response
+
+               return inner
        
-       paginator = Paginator(objects, per_page)
-       try:
-               page_number = int(page_number)
-       except:
-               page_number = 1
+       def get_feed(self, feed_type, request, node, kwargs, reverse_name):
+               title = self.feed_title
+               link = '/%s/%s/' % (node.get_absolute_url().strip('/'), reverse(reverse_name, urlconf=self, kwargs=kwargs).strip('/'))
+               description = self.feed_description
+               if feed_type == 'rss':
+                       return self.rss_feed(title, link, description)
+               
+               return self.atom_feed(title, link, description, subtitle=description)
        
-       # This will raise an EmptyPage error if the page number is out of range.
-       # This error is intentionally left for the calling function to handle.
-       page = paginator.page(page_number)
+       def feed_patterns(self, object_fetcher, page, base_name):
+               feed_name = '%s_feed' % base_name
+               urlpatterns = patterns('',
+                       url(r'^%s/$' % self.feed_suffix, self.feed_view(object_fetcher, feed_name), name=feed_name),
+                       url(r'^$', self.page_view(object_fetcher, page), name=base_name)
+               )
+               return urlpatterns
        
-       return page, page.object_list
\ No newline at end of file
+       def get_obj_description(self, obj):
+               raise NotImplementedError
\ No newline at end of file