X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/e5e8226c4d4bd610de5983e69b1b13684a17fe75..e9c8751701e3f5e00f9ffe76b616966537c77b56:/apps/catalogue/views.py diff --git a/apps/catalogue/views.py b/apps/catalogue/views.py index 9687db6fd..232b3a749 100644 --- a/apps/catalogue/views.py +++ b/apps/catalogue/views.py @@ -6,7 +6,9 @@ import re import itertools from django.conf import settings +from django.core.cache import get_cache from django.template import RequestContext +from django.template.loader import render_to_string from django.shortcuts import render_to_response, get_object_or_404, redirect from django.http import HttpResponse, HttpResponseRedirect, Http404, HttpResponsePermanentRedirect from django.core.urlresolvers import reverse @@ -15,67 +17,125 @@ from django.contrib.auth.decorators import login_required, user_passes_test from django.utils.datastructures import SortedDict from django.utils.http import urlquote_plus from django.utils import translation -from django.utils.translation import ugettext as _, ugettext_lazy -from django.views.decorators.cache import never_cache +from django.utils.translation import get_language, ugettext as _, ugettext_lazy +from django.views.decorators.vary import vary_on_headers from ajaxable.utils import JSONResponse, AjaxableFormView - from catalogue import models from catalogue import forms -from catalogue.utils import split_tags, MultiQuerySet +from catalogue.utils import split_tags, MultiQuerySet, SortedMultiQuerySet +from catalogue.templatetags.catalogue_tags import tag_list, collection_list from pdcounter import models as pdcounter_models from pdcounter import views as pdcounter_views from suggest.forms import PublishingSuggestForm -from picture.models import Picture - +from picture.models import Picture, PictureArea +from picture.views import picture_list_thumb +import logging staff_required = user_passes_test(lambda user: user.is_staff) +permanent_cache = get_cache('permanent') +@vary_on_headers('X-Requested-With') def catalogue(request): - tags = models.Tag.objects.exclude( - category__in=('set', 'book')).exclude(book_count=0) - tags = list(tags) - for tag in tags: - tag.count = tag.book_count - categories = split_tags(tags) - fragment_tags = categories.get('theme', []) - - return render_to_response('catalogue/catalogue.html', locals(), - context_instance=RequestContext(request)) + cache_key='catalogue.catalogue/' + get_language() + output = permanent_cache.get(cache_key) + + if output is None: + tags = models.Tag.objects.exclude( + category__in=('set', 'book')).exclude(book_count=0, picture_count=0) + tags = list(tags) + for tag in tags: + tag.count = tag.book_count + tag.picture_count + categories = split_tags(tags) + fragment_tags = categories.get('theme', []) + collections = models.Collection.objects.all() + + render_tag_list = lambda x: render_to_string( + 'catalogue/tag_list.html', tag_list(x)) + has_pictures = lambda x: filter(lambda y: y.picture_count>0, x) + has_books = lambda x: filter(lambda y: y.book_count>0, x) + def render_split(tags): + with_books = has_books(tags) + with_pictures = has_pictures(tags) + ctx = {} + if with_books: + ctx['books'] = render_tag_list(with_books) + if with_pictures: + ctx['pictures'] = render_tag_list(with_pictures) + return render_to_string('catalogue/tag_list_split.html', ctx) + + output = {'theme': {}} + output['theme'] = render_tag_list(fragment_tags) + for category, tags in categories.items(): + if category in ('author', 'theme'): + output[category] = render_tag_list(tags) + else: + output[category] = render_split(tags) + + + output['collections'] = render_to_string( + 'catalogue/collection_list.html', collection_list(collections)) + permanent_cache.set(cache_key, output) + if request.is_ajax(): + return JSONResponse(output) + else: + return render_to_response('catalogue/catalogue.html', locals(), + context_instance=RequestContext(request)) -def book_list(request, filter=None, template_name='catalogue/book_list.html', - context=None): +def book_list(request, filter=None, get_filter=None, + template_name='catalogue/book_list.html', + nav_template_name='catalogue/snippets/book_list_nav.html', + list_template_name='catalogue/snippets/book_list.html', + cache_key='catalogue.book_list', + context=None, + ): """ generates a listing of all books, optionally filtered with a test function """ - - books_by_author, orphans, books_by_parent = models.Book.book_list(filter) - books_nav = SortedDict() - for tag in books_by_author: - if books_by_author[tag]: - books_nav.setdefault(tag.sort_key[0], []).append(tag) - + cache_key = "%s/%s" % (cache_key, get_language()) + cached = permanent_cache.get(cache_key) + if cached is not None: + rendered_nav, rendered_book_list = cached + else: + if get_filter: + filter = get_filter() + books_by_author, orphans, books_by_parent = models.Book.book_list(filter) + books_nav = SortedDict() + for tag in books_by_author: + if books_by_author[tag]: + books_nav.setdefault(tag.sort_key[0], []).append(tag) + rendered_nav = render_to_string(nav_template_name, locals()) + rendered_book_list = render_to_string(list_template_name, locals()) + permanent_cache.set(cache_key, (rendered_nav, rendered_book_list)) return render_to_response(template_name, locals(), context_instance=RequestContext(request)) def audiobook_list(request): return book_list(request, Q(media__type='mp3') | Q(media__type='ogg'), - template_name='catalogue/audiobook_list.html') + template_name='catalogue/audiobook_list.html', + list_template_name='catalogue/snippets/audiobook_list.html', + cache_key='catalogue.audiobook_list') def daisy_list(request): return book_list(request, Q(media__type='daisy'), - template_name='catalogue/daisy_list.html') + template_name='catalogue/daisy_list.html', + cache_key='catalogue.daisy_list') def collection(request, slug): coll = get_object_or_404(models.Collection, slug=slug) - slugs = coll.book_slugs.split() - # allow URIs - slugs = [slug.rstrip('/').rsplit('/', 1)[-1] if '/' in slug else slug - for slug in slugs] - return book_list(request, Q(slug__in=slugs), - template_name='catalogue/collection.html', + if coll.kind == 'book': + view = book_list + tmpl = "catalogue/collection.html" + elif coll.kind == 'picture': + view = picture_list_thumb + tmpl = "picture/collection.html" + else: + raise ValueError('How do I show this kind of collection? %s' % coll.kind) + return view(request, get_filter=coll.get_query, + template_name=tmpl, + cache_key='catalogue.collection:%s' % coll.slug, context={'collection': coll}) @@ -93,17 +153,21 @@ def differentiate_tags(request, tags, ambiguous_slugs): context_instance=RequestContext(request)) -@never_cache +# TODO: Rewrite this hellish piece of code which tries to do everything def tagged_object_list(request, tags=''): + # preliminary tests and conditions try: tags = models.Tag.get_tag_list(tags) except models.Tag.DoesNotExist: + # Perhaps the user is asking about an author in Public Domain + # counter (they are not represented in tags) chunks = tags.split('/') if len(chunks) == 2 and chunks[0] == 'autor': return pdcounter_views.author_detail(request, chunks[1]) else: raise Http404 except models.Tag.MultipleObjectsReturned, e: + # Ask the user to disambiguate return differentiate_tags(request, e.tags, e.ambiguous_slugs) except models.Tag.UrlDeprecationWarning, e: return HttpResponsePermanentRedirect(reverse('tagged_object_list', args=['/'.join(tag.url_chunk for tag in e.tags)])) @@ -117,18 +181,22 @@ def tagged_object_list(request, tags=''): if len([tag for tag in tags if tag.category == 'book']): raise Http404 + # beginning of digestion theme_is_set = [tag for tag in tags if tag.category == 'theme'] shelf_is_set = [tag for tag in tags if tag.category == 'set'] only_shelf = shelf_is_set and len(tags) == 1 only_my_shelf = only_shelf and request.user.is_authenticated() and request.user == tags[0].user + objects = only_author = None categories = {} + object_queries = [] if theme_is_set: shelf_tags = [tag for tag in tags if tag.category == 'set'] fragment_tags = [tag for tag in tags if tag.category != 'set'] fragments = models.Fragment.tagged.with_all(fragment_tags) + areas = PictureArea.tagged.with_all(fragment_tags) if shelf_tags: books = models.Book.tagged.with_all(shelf_tags).order_by() @@ -139,44 +207,81 @@ def tagged_object_list(request, tags=''): # newtagging goes crazy if we just try: #related_tags = models.Tag.objects.usage_for_queryset(fragments, counts=True, # extra={'where': ["catalogue_tag.category != 'book'"]}) + + related_tags = [] + fragment_keys = [fragment.pk for fragment in fragments.iterator()] if fragment_keys: related_tags = models.Fragment.tags.usage(counts=True, filters={'pk__in': fragment_keys}, extra={'where': ["catalogue_tag.category != 'book'"]}) related_tags = (tag for tag in related_tags if tag not in fragment_tags) - categories = split_tags(related_tags) + categories = split_tags(related_tags, categories) - objects = fragments + object_queries.insert(0, fragments) + + area_keys = [area.pk for area in areas.iterator()] + if area_keys: + related_tags = PictureArea.tags.usage(counts=True, + filters={'pk__in': area_keys}) + related_tags = (tag for tag in related_tags if tag not in fragment_tags) + + categories = split_tags(related_tags, categories) + + # we want the Pictures to go first + object_queries.insert(0, areas) + objects = MultiQuerySet(*object_queries) else: if shelf_is_set: - objects = models.Book.tagged.with_all(tags) + books = models.Book.tagged.with_all(tags).order_by('sort_key_author') else: - objects = models.Book.tagged_top_level(tags) - - # get related tags from `tag_counter` and `theme_counter` - related_counts = {} - tags_pks = [tag.pk for tag in tags] - for book in objects.iterator(): - for tag_pk, value in itertools.chain(book.tag_counter.iteritems(), book.theme_counter.iteritems()): - if tag_pk in tags_pks: - continue - related_counts[tag_pk] = related_counts.get(tag_pk, 0) + value - related_tags = models.Tag.objects.filter(pk__in=related_counts.keys()) - related_tags = [tag for tag in related_tags if tag not in tags] - for tag in related_tags: - tag.count = related_counts[tag.pk] - - categories = split_tags(related_tags) - del related_tags + books = models.Book.tagged_top_level(tags).order_by('sort_key_author') + + pictures = Picture.tagged.with_all(tags).order_by('sort_key_author') + + if books.count() > 0: + # get related tags from `tag_counter` and `theme_counter` + related_counts = {} + tags_pks = [tag.pk for tag in tags] + for book in books: + for tag_pk, value in itertools.chain(book.tag_counter.iteritems(), book.theme_counter.iteritems()): + if tag_pk in tags_pks: + continue + related_counts[tag_pk] = related_counts.get(tag_pk, 0) + value + related_tags = models.Tag.objects.filter(pk__in=related_counts.keys()) + related_tags = [tag for tag in related_tags if tag not in tags] + for tag in related_tags: + tag.count = related_counts[tag.pk] + + categories = split_tags(related_tags) + del related_tags + + if pictures.count() > 0: + related_counts = {} + tags_pks = [tag.pk for tag in tags] + for picture in pictures: + for tag_pk, value in itertools.chain(picture.tag_counter.iteritems(), picture.theme_counter.iteritems()): + if tag_pk in tags_pks: + continue + logging.info("counting tag not in tags_pks: %d", tag_pk) + related_counts[tag_pk] = related_counts.get(tag_pk, 0) + value + related_tags = models.Tag.objects.filter(pk__in=related_counts.keys()) + related_tags = [tag for tag in related_tags if tag not in tags] + for tag in related_tags: + tag.count = related_counts[tag.pk] + + categories = split_tags(related_tags) + del related_tags + + logging.info("Returning %d picutres and %d books" % (pictures.count(), books.count())) + objects = SortedMultiQuerySet(pictures, books, order_by='sort_key_author') + + if not objects: only_author = len(tags) == 1 and tags[0].category == 'author' objects = models.Book.objects.none() - # Add pictures - objects = MultiQuerySet(Picture.tagged.with_all(tags), objects) - return render_to_response('catalogue/tagged_object_list.html', { 'object_list': objects, @@ -202,7 +307,6 @@ def book_fragments(request, slug, theme_slug): context_instance=RequestContext(request)) -@never_cache def book_detail(request, slug): try: book = models.Book.objects.get(slug=slug) @@ -258,13 +362,7 @@ def book_text(request, slug): if not book.has_html_file(): raise Http404 - book_themes = {} - for fragment in book.fragments.all().iterator(): - for theme in fragment.tags.filter(category='theme').iterator(): - book_themes.setdefault(theme, []).append(fragment) - - book_themes = book_themes.items() - book_themes.sort(key=lambda s: s[0].sort_key) + related = book.related_info() return render_to_response('catalogue/book_text.html', locals(), context_instance=RequestContext(request)) @@ -535,6 +633,11 @@ class CustomPDFFormView(AjaxableFormView): submit = ugettext_lazy('Download') honeypot = True + def __call__(self, *args, **kwargs): + if settings.NO_CUSTOM_PDF: + raise Http404('Custom PDF is disabled') + return super(CustomPDFFormView, self).__call__(*args, **kwargs) + def form_args(self, request, obj): """Override to parse view args and give additional args to the form.""" return (obj,), {}