# Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
#
from collections import OrderedDict
-import re
import random
from django.conf import settings
+from django.http.response import HttpResponseForbidden
from django.template import RequestContext
from django.template.loader import render_to_string
from django.shortcuts import render_to_response, get_object_or_404, render, redirect
-from django.http import HttpResponse, HttpResponseRedirect, Http404, HttpResponsePermanentRedirect, JsonResponse
+from django.http import HttpResponse, HttpResponseRedirect, Http404, HttpResponsePermanentRedirect
from django.core.urlresolvers import reverse
-from django.db.models import Q
+from django.db.models import Q, QuerySet
from django.contrib.auth.decorators import login_required, user_passes_test
from django.utils.http import urlquote_plus
from django.utils import translation
from django.utils.translation import ugettext as _, ugettext_lazy
from ajaxable.utils import AjaxableFormView
-from pdcounter.models import BookStub, Author
from pdcounter import views as pdcounter_views
from picture.models import Picture, PictureArea
from ssify import ssi_included, ssi_expect, SsiVariable as Var
-from suggest.forms import PublishingSuggestForm
from catalogue import constants
from catalogue import forms
from catalogue.helpers import get_top_level_related_tags
from catalogue.models import Book, Collection, Tag, Fragment
from catalogue.utils import split_tags
+from catalogue.models.tag import prefetch_relations
+from wolnelektury.utils import is_crawler
staff_required = user_passes_test(lambda user: user.is_staff)
else:
fragments = Fragment.objects.filter(book__in=objects)
related_tag_lists.append(
- Tag.objects.usage_for_queryset(fragments, counts=True).filter(category='theme').exclude(pk__in=tag_ids))
+ Tag.objects.usage_for_queryset(fragments, counts=True).filter(category='theme').exclude(pk__in=tag_ids)
+ .only('name', 'sort_key', 'category', 'slug'))
+ if isinstance(objects, QuerySet):
+ objects = prefetch_relations(objects, 'author')
categories = split_tags(*related_tag_lists)
objects = list(objects)
+
+ if not objects and len(tags) == 1 and list_type == 'books':
+ if PictureArea.tagged.with_any(tags).exists() or Picture.tagged.with_any(tags).exists():
+ return redirect('tagged_object_list_gallery', '/'.join(tag.url_chunk for tag in tags))
+
if len(objects) > 3:
best = random.sample(objects, 3)
else:
def literature(request):
books = Book.objects.filter(parent=None)
- last_published = Book.objects.exclude(cover_thumb='').filter(parent=None).order_by('-created_at')[:20]
- most_popular = Book.objects.exclude(cover_thumb='')\
- .order_by('-popularity__count', 'sort_key_author', 'sort_key')[:20]
- return object_list(request, books, related_tags=get_top_level_related_tags([]), extra={
- 'last_published': last_published,
- 'most_popular': most_popular,
- })
+ # last_published = Book.objects.exclude(cover_thumb='').filter(parent=None).order_by('-created_at')[:20]
+ # most_popular = Book.objects.exclude(cover_thumb='')\
+ # .order_by('-popularity__count', 'sort_key_author', 'sort_key')[:20]
+ return object_list(request, books, related_tags=get_top_level_related_tags([]))
+ # extra={
+ # 'last_published': last_published,
+ # 'most_popular': most_popular,
+ # })
def gallery(request):
books = Book.tagged.with_all(shelf_tags).order_by()
fragments = fragments.filter(Q(book__in=books) | Q(book__ancestor__in=books))
- if not fragments and len(tags) == 1:
- tag = tags[0]
- if tag.category == 'theme' and (
- PictureArea.tagged.with_any([tag]).exists() or
- Picture.tagged.with_any([tag]).exists()):
+ if not fragments and len(tags) == 1 and list_type == 'books':
+ if PictureArea.tagged.with_any(tags).exists() or Picture.tagged.with_any(tags).exists():
return redirect('tagged_object_list_gallery', '/'.join(tag.url_chunk for tag in tags))
return object_list(request, fragments, tags=tags, list_type=list_type, extra={
except ResponseInstead as e:
return e.response
+ if is_crawler(request) and len(tags) > 1:
+ return HttpResponseForbidden('address removed from crawling. check robots.txt')
+
if list_type == 'gallery' and any(tag.category == 'set' for tag in tags):
raise Http404
- if any(tag.category == 'theme' for tag in tags):
+ if any(tag.category in ('theme', 'thing') for tag in tags):
return theme_list(request, tags, list_type=list_type)
if list_type == 'books':
'book': book,
'theme': theme,
'fragments': fragments,
+ 'active_menu_item': 'books',
}, context_instance=RequestContext(request))
'book': book,
'tags': book.tags.exclude(category__in=('set', 'theme')),
'book_children': book.children.all().order_by('parent_number', 'sort_key'),
+ 'active_menu_item': 'books',
}, context_instance=RequestContext(request))
-def get_audiobooks(book):
- ogg_files = {}
- for m in book.media.filter(type='ogg').order_by().iterator():
- ogg_files[m.name] = m
-
- audiobooks = []
- have_oggs = True
- projects = set()
- for mp3 in book.media.filter(type='mp3').iterator():
- # ogg files are always from the same project
- meta = mp3.extra_info
- project = meta.get('project')
- if not project:
- # temporary fallback
- project = u'CzytamySłuchając'
-
- projects.add((project, meta.get('funded_by', '')))
-
- media = {'mp3': mp3}
-
- ogg = ogg_files.get(mp3.name)
- if ogg:
- media['ogg'] = ogg
- else:
- have_oggs = False
- audiobooks.append(media)
-
- projects = sorted(projects)
- return audiobooks, projects, have_oggs
-
-
-# używane tylko do audiobook_tree, które jest używane tylko w snippets/audiobook_list.html, które nie jest używane
+# używane w publicznym interfejsie
def player(request, slug):
book = get_object_or_404(Book, slug=slug)
if not book.has_media('mp3'):
raise Http404
- audiobooks, projects, have_oggs = get_audiobooks(book)
-
- # extra_info = book.extra_info
+ audiobooks, projects = book.get_audiobooks()
return render_to_response('catalogue/player.html', {
'book': book,
if not book.has_html_file():
raise Http404
- return render_to_response('catalogue/book_text.html', {'book': book,}, context_instance=RequestContext(request))
-
-
-# ==========
-# = Search =
-# ==========
-
-def _no_diacritics_regexp(query):
- """ returns a regexp for searching for a query without diacritics
-
- should be locale-aware """
- names = {
- u'a': u'aąĄ', u'c': u'cćĆ', u'e': u'eęĘ', u'l': u'lłŁ', u'n': u'nńŃ', u'o': u'oóÓ', u's': u'sśŚ',
- u'z': u'zźżŹŻ',
- u'ą': u'ąĄ', u'ć': u'ćĆ', u'ę': u'ęĘ', u'ł': u'łŁ', u'ń': u'ńŃ', u'ó': u'óÓ', u'ś': u'śŚ', u'ź': u'źŹ',
- u'ż': u'żŻ'
- }
-
- def repl(m):
- l = m.group()
- return u"(%s)" % '|'.join(names[l])
-
- return re.sub(u'[%s]' % (u''.join(names.keys())), repl, query)
-
-
-def unicode_re_escape(query):
- """ Unicode-friendly version of re.escape """
- return re.sub(r'(?u)(\W)', r'\\\1', query)
-
-
-def _word_starts_with(name, prefix):
- """returns a Q object getting models having `name` contain a word
- starting with `prefix`
-
- We define word characters as alphanumeric and underscore, like in JS.
-
- Works for MySQL, PostgreSQL, Oracle.
- For SQLite, _sqlite* version is substituted for this.
- """
- kwargs = {}
-
- prefix = _no_diacritics_regexp(unicode_re_escape(prefix))
- # can't use [[:<:]] (word start),
- # but we want both `xy` and `(xy` to catch `(xyz)`
- kwargs['%s__iregex' % name] = u"(^|[^[:alnum:]_])%s" % prefix
-
- return Q(**kwargs)
-
-
-def _word_starts_with_regexp(prefix):
- prefix = _no_diacritics_regexp(unicode_re_escape(prefix))
- return ur"(^|(?<=[^\wąćęłńóśźżĄĆĘŁŃÓŚŹŻ]))%s" % prefix
-
-
-def _sqlite_word_starts_with(name, prefix):
- """ version of _word_starts_with for SQLite
-
- SQLite in Django uses Python re module
- """
- kwargs = {'%s__iregex' % name: _word_starts_with_regexp(prefix)}
- return Q(**kwargs)
-
-
-if hasattr(settings, 'DATABASES'):
- if settings.DATABASES['default']['ENGINE'] == 'django.db.backends.sqlite3':
- _word_starts_with = _sqlite_word_starts_with
-elif settings.DATABASE_ENGINE == 'sqlite3':
- _word_starts_with = _sqlite_word_starts_with
-
-
-class App:
- def __init__(self, name, view):
- self.name = name
- self._view = view
- self.lower = name.lower()
- self.category = 'application'
-
- def view(self):
- return reverse(*self._view)
-
-_apps = (
- App(u'Leśmianator', (u'lesmianator', )),
- )
-
-
-def _tags_starting_with(prefix, user=None):
- prefix = prefix.lower()
- # PD counter
- book_stubs = BookStub.objects.filter(_word_starts_with('title', prefix))
- authors = Author.objects.filter(_word_starts_with('name', prefix))
-
- books = Book.objects.filter(_word_starts_with('title', prefix))
- tags = Tag.objects.filter(_word_starts_with('name', prefix))
- if user and user.is_authenticated():
- tags = tags.filter(~Q(category='set') | Q(user=user))
- else:
- tags = tags.exclude(category='set')
-
- prefix_regexp = re.compile(_word_starts_with_regexp(prefix))
- return list(books) + list(tags) + [app for app in _apps if prefix_regexp.search(app.lower)] + list(book_stubs) + \
- list(authors)
-
-
-def _get_result_link(match, tag_list):
- if isinstance(match, Tag):
- return reverse('catalogue.views.tagged_object_list',
- kwargs={'tags': '/'.join(tag.url_chunk for tag in tag_list + [match])})
- elif isinstance(match, App):
- return match.view()
- else:
- return match.get_absolute_url()
-
-
-def _get_result_type(match):
- if isinstance(match, Book) or isinstance(match, BookStub):
- match_type = 'book'
- else:
- match_type = match.category
- return match_type
-
-
-def books_starting_with(prefix):
- prefix = prefix.lower()
- return Book.objects.filter(_word_starts_with('title', prefix))
-
-
-def find_best_matches(query, user=None):
- """ Finds a Book, Tag, BookStub or Author best matching a query.
-
- Returns a with:
- - zero elements when nothing is found,
- - one element when a best result is found,
- - more then one element on multiple exact matches
-
- Raises a ValueError on too short a query.
- """
-
- query = query.lower()
- if len(query) < 2:
- raise ValueError("query must have at least two characters")
-
- result = tuple(_tags_starting_with(query, user))
- # remove pdcounter stuff
- book_titles = set(match.pretty_title().lower() for match in result
- if isinstance(match, Book))
- authors = set(match.name.lower() for match in result
- if isinstance(match, Tag) and match.category == 'author')
- result = tuple(res for res in result if not (
- (isinstance(res, BookStub) and res.pretty_title().lower() in book_titles) or
- (isinstance(res, Author) and res.name.lower() in authors)
- ))
-
- exact_matches = tuple(res for res in result if res.name.lower() == query)
- if exact_matches:
- return exact_matches
- else:
- return tuple(result)[:1]
-
-
-def search(request):
- tags = request.GET.get('tags', '')
- prefix = request.GET.get('q', '')
-
- try:
- tag_list = Tag.get_tag_list(tags)
- except (Tag.DoesNotExist, Tag.MultipleObjectsReturned, Tag.UrlDeprecationWarning):
- tag_list = []
-
- try:
- result = find_best_matches(prefix, request.user)
- except ValueError:
- return render_to_response(
- 'catalogue/search_too_short.html', {'tags': tag_list, 'prefix': prefix},
- context_instance=RequestContext(request))
-
- if len(result) == 1:
- return HttpResponseRedirect(_get_result_link(result[0], tag_list))
- elif len(result) > 1:
- return render_to_response(
- 'catalogue/search_multiple_hits.html',
- {
- 'tags': tag_list, 'prefix': prefix,
- 'results': ((x, _get_result_link(x, tag_list), _get_result_type(x)) for x in result)
- },
- context_instance=RequestContext(request))
- else:
- form = PublishingSuggestForm(initial={"books": prefix + ", "})
- return render_to_response(
- 'catalogue/search_no_hits.html',
- {'tags': tag_list, 'prefix': prefix, "pubsuggest_form": form},
- context_instance=RequestContext(request))
-
-
-def tags_starting_with(request):
- prefix = request.GET.get('q', '')
- # Prefix must have at least 2 characters
- if len(prefix) < 2:
- return HttpResponse('')
- tags_list = []
- result = ""
- for tag in _tags_starting_with(prefix, request.user):
- if tag.name not in tags_list:
- result += "\n" + tag.name
- tags_list.append(tag.name)
- return HttpResponse(result)
-
-
-def json_tags_starting_with(request, callback=None):
- # Callback for JSONP
- prefix = request.GET.get('q', '')
- callback = request.GET.get('callback', '')
- # Prefix must have at least 2 characters
- if len(prefix) < 2:
- return HttpResponse('')
- tags_list = []
- for tag in _tags_starting_with(prefix, request.user):
- if tag.name not in tags_list:
- tags_list.append(tag.name)
- if request.GET.get('mozhint', ''):
- result = [prefix, tags_list]
- else:
- result = {"matches": tags_list}
- response = JsonResponse(result, safe=False)
- if callback:
- response.content = callback + "(" + response.content + ");"
- return response
+ return render_to_response('catalogue/book_text.html', {'book': book}, context_instance=RequestContext(request))
# =========
form_class = forms.CustomPDFForm
title = ugettext_lazy('Download custom PDF')
submit = ugettext_lazy('Download')
+ template = 'catalogue/custom_pdf_form.html'
honeypot = True
def __call__(self, *args, **kwargs):
@ssi_included
def book_mini(request, pk, with_link=True):
- book = get_object_or_404(Book, pk=pk)
- author_str = ", ".join(tag.name for tag in book.tags.filter(category='author'))
+ # book = get_object_or_404(Book, pk=pk)
+ try:
+ book = Book.objects.only('cover_thumb', 'title', 'language', 'slug').get(pk=pk)
+ except Book.DoesNotExist:
+ raise Http404
return render(request, 'catalogue/book_mini_box.html', {
'book': book,
- 'author_str': author_str,
- 'with_link': with_link,
- 'show_lang': book.language_code() != settings.LANGUAGE_CODE,
+ 'no_link': not with_link,
})
))(ssi_expect(pk, int)))
def book_short(request, pk):
book = get_object_or_404(Book, pk=pk)
- stage_note, stage_note_url = book.stage_note()
- audiobooks, projects, have_oggs = get_audiobooks(book)
return render(request, 'catalogue/book_short.html', {
'book': book,
- 'has_audio': book.has_media('mp3'),
- 'main_link': book.get_absolute_url(),
- 'parents': book.parents(),
- 'tags': split_tags(book.tags.exclude(category__in=('set', 'theme'))),
- 'show_lang': book.language_code() != settings.LANGUAGE_CODE,
- 'stage_note': stage_note,
- 'stage_note_url': stage_note_url,
- 'audiobooks': audiobooks,
- 'have_oggs': have_oggs,
})
))(ssi_expect(pk, int)))
def book_wide(request, pk):
book = get_object_or_404(Book, pk=pk)
- stage_note, stage_note_url = book.stage_note()
extra_info = book.extra_info
- audiobooks, projects, have_oggs = get_audiobooks(book)
return render(request, 'catalogue/book_wide.html', {
'book': book,
- 'has_audio': book.has_media('mp3'),
'parents': book.parents(),
'tags': split_tags(book.tags.exclude(category__in=('set', 'theme'))),
'show_lang': book.language_code() != settings.LANGUAGE_CODE,
- 'stage_note': stage_note,
- 'stage_note_url': stage_note_url,
'main_link': reverse('book_text', args=[book.slug]) if book.html_file else None,
'extra_info': extra_info,
'hide_about': extra_info.get('about', '').startswith('http://wiki.wolnepodreczniki.pl'),
- 'audiobooks': audiobooks,
- 'have_oggs': have_oggs,
})
@ssi_included
def collection_box(request, pk):
- obj = get_object_or_404(Collection, pk=pk)
+ collection = get_object_or_404(Collection, pk=pk)
return render(request, 'catalogue/collection_box.html', {
- 'obj': obj,
+ 'collection': collection,
})
'objects': objects,
'best': best,
})
+
+
+def ridero_cover(request, slug):
+ from librarian.cover import make_cover
+ wldoc = Book.objects.get(slug=slug).wldocument()
+ cover = make_cover(wldoc.book_info, width=980, bleed=20, format='PNG')
+ response = HttpResponse(content_type="image/png")
+ cover.save(response)
+ return response
+
+
+def get_isbn(request, book_format, slug):
+ book = Book.objects.get(slug=slug)
+ return HttpResponse(book.extra_info.get('isbn_%s' % book_format))