X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/e0de41ae1e67c230ad1e67f34f6f4f76e53d7234..013ec8a416d4ff1ea3e1f2fa9ee2935623197ecd:/apps/api/handlers.py?ds=inline diff --git a/apps/api/handlers.py b/apps/api/handlers.py index 799b60827..5fe931e69 100644 --- a/apps/api/handlers.py +++ b/apps/api/handlers.py @@ -2,34 +2,468 @@ # This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later. # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information. # -from django.shortcuts import get_object_or_404 -from django.contrib.auth.decorators import login_required, user_passes_test -from piston.handler import BaseHandler -from piston.utils import rc, validate -from catalogue.models import Book +import json + +from django.contrib.sites.models import Site +from django.core.urlresolvers import reverse +from django.utils.functional import lazy +from piston.handler import AnonymousBaseHandler, BaseHandler +from piston.utils import rc +from sorl.thumbnail import default + from catalogue.forms import BookImportForm +from catalogue.models import Book, Tag, BookMedia, Fragment, Collection +from picture.models import Picture +from picture.forms import PictureImportForm + +from stats.utils import piwik_track + +from . import emitters # Register our emitters + +API_BASE = WL_BASE = MEDIA_BASE = lazy( + lambda: u'http://' + Site.objects.get_current().domain, unicode)() + + +category_singular = { + 'authors': 'author', + 'kinds': 'kind', + 'genres': 'genre', + 'epochs': 'epoch', + 'themes': 'theme', + 'books': 'book', +} +category_plural = {} +for k, v in category_singular.items(): + category_plural[v] = k + +book_tag_categories = ['author', 'epoch', 'kind', 'genre'] + + + +def read_tags(tags, allowed): + """ Reads a path of filtering tags. + + :param str tags: a path of category and slug pairs, like: authors/an-author/... + :returns: list of Tag objects + :raises: ValueError when tags can't be found + """ + if not tags: + return [], [] + + tags = tags.strip('/').split('/') + real_tags = [] + books = [] + while tags: + category = tags.pop(0) + slug = tags.pop(0) + + try: + category = category_singular[category] + except KeyError: + raise ValueError('Unknown category.') + + if not category in allowed: + raise ValueError('Category not allowed.') + + if category == 'book': + try: + books.append(Book.objects.get(slug=slug)) + except Book.DoesNotExist: + raise ValueError('Unknown book.') + + try: + real_tags.append(Tag.objects.get(category=category, slug=slug)) + except Tag.DoesNotExist: + raise ValueError('Tag not found') + return real_tags, books + + +# RESTful handlers + + +class BookMediaHandler(BaseHandler): + """ Responsible for representing media in Books. """ + + model = BookMedia + fields = ['name', 'type', 'url', 'artist', 'director'] + + @classmethod + def url(cls, media): + """ Link to media on site. """ + + return MEDIA_BASE + media.file.url + + @classmethod + def artist(cls, media): + return media.extra_info.get('artist_name', '') + + @classmethod + def director(cls, media): + return media.extra_info.get('director_name', '') + + +class BookDetails(object): + """Custom fields used for representing Books.""" + + @classmethod + def href(cls, book): + """ Returns an URI for a Book in the API. """ + return API_BASE + reverse("api_book", args=[book.slug]) + + @classmethod + def url(cls, book): + """ Returns Book's URL on the site. """ + + return WL_BASE + book.get_absolute_url() + + @classmethod + def children(cls, book): + """ Returns all children for a book. """ + return book.children.all() -staff_required = user_passes_test(lambda user: user.is_staff) + @classmethod + def media(cls, book): + """ Returns all media for a book. """ + return book.media.all() + @classmethod + def cover(cls, book): + return MEDIA_BASE + book.cover.url if book.cover else '' -class BookHandler(BaseHandler): + @classmethod + def cover_thumb(cls, book): + return MEDIA_BASE + default.backend.get_thumbnail( + book.cover, "139x193").url if book.cover else '' + + + +class BookDetailHandler(BaseHandler, BookDetails): + """ Main handler for Book objects. + + Responsible for single Book details. + """ + allowed_methods = ['GET'] + fields = ['title', 'parent', 'children'] + Book.formats + [ + 'media', 'url', 'cover', 'cover_thumb'] + [ + category_plural[c] for c in book_tag_categories] + + @piwik_track + def read(self, request, book): + """ Returns details of a book, identified by a slug and lang. """ + try: + return Book.objects.get(slug=book) + except Book.DoesNotExist: + return rc.NOT_FOUND + + +class AnonymousBooksHandler(AnonymousBaseHandler, BookDetails): + """ Main handler for Book objects. + + Responsible for lists of Book objects. + """ + allowed_methods = ('GET',) model = Book - fields = ('slug', 'title') - - @staff_required - def read(self, request, slug=None): - if slug: - return get_object_or_404(Book, slug=slug) + fields = book_tag_categories + ['href', 'title', 'url', 'cover', 'cover_thumb'] + + @classmethod + def genres(cls, book): + """ Returns all media for a book. """ + return book.tags.filter(category='genre') + + @piwik_track + def read(self, request, tags=None, top_level=False, + audiobooks=False, daisy=False, pk=None): + """ Lists all books with given tags. + + :param tags: filtering tags; should be a path of categories + and slugs, i.e.: authors/an-author/epoch/an-epoch/ + :param top_level: if True and a book is included in the results, + it's children are aren't. By default all books matching the tags + are returned. + """ + if pk is not None: + try: + return Book.objects.get(pk=pk) + except Book.DoesNotExist: + return rc.NOT_FOUND + + try: + tags, _ancestors = read_tags(tags, allowed=book_tag_categories) + except ValueError: + return rc.NOT_FOUND + + if tags: + if top_level: + books = Book.tagged_top_level(tags) + return books if books else rc.NOT_FOUND + else: + books = Book.tagged.with_all(tags) else: - return Book.objects.all() - - @staff_required - def create(self, request): - form = BookImportForm(request.POST, request.FILES) + books = Book.objects.all() + + if top_level: + books = books.filter(parent=None) + if audiobooks: + books = books.filter(media__type='mp3').distinct() + if daisy: + books = books.filter(media__type='daisy').distinct() + + if books.exists(): + return books + else: + return rc.NOT_FOUND + + def create(self, request, *args, **kwargs): + return rc.FORBIDDEN + + +class BooksHandler(BookDetailHandler): + allowed_methods = ('GET', 'POST') + model = Book + fields = book_tag_categories + ['href', 'title', 'url', 'cover', 'cover_thumb'] + anonymous = AnonymousBooksHandler + + def create(self, request, *args, **kwargs): + if not request.user.has_perm('catalogue.add_book'): + return rc.FORBIDDEN + + data = json.loads(request.POST.get('data')) + form = BookImportForm(data) if form.is_valid(): form.save() return rc.CREATED else: - return rc.BAD_REQUEST + return rc.NOT_FOUND + + +class EBooksHandler(AnonymousBooksHandler): + fields = ('author', 'href', 'title', 'cover') + tuple(Book.ebook_formats) + + +# add categorized tags fields for Book +def _tags_getter(category): + @classmethod + def get_tags(cls, book): + return book.tags.filter(category=category) + return get_tags +def _tag_getter(category): + @classmethod + def get_tag(cls, book): + return ', '.join(tag.name for tag in book.tags.filter(category=category)) + return get_tag +for plural, singular in category_singular.items(): + setattr(BookDetails, plural, _tags_getter(singular)) + setattr(BookDetails, singular, _tag_getter(singular)) + +# add fields for files in Book +def _file_getter(format): + field = "%s_file" % format + @classmethod + def get_file(cls, book): + f = getattr(book, field) + if f: + return MEDIA_BASE + f.url + else: + return '' + return get_file +for format in Book.formats: + setattr(BookDetails, format, _file_getter(format)) + +class CollectionDetails(object): + """Custom Collection fields.""" + + @classmethod + def href(cls, collection): + """ Returns URI in the API for the collection. """ + + return API_BASE + reverse("api_collection", args=[collection.slug]) + + @classmethod + def url(cls, collection): + """ Returns URL on the site. """ + + return WL_BASE + collection.get_absolute_url() + + @classmethod + def books(cls, collection): + return Book.objects.filter(collection.get_query()) + + + +class CollectionDetailHandler(BaseHandler, CollectionDetails): + allowed_methods = ('GET',) + fields = ['url', 'title', 'description', 'books'] + + @piwik_track + def read(self, request, slug): + """ Returns details of a collection, identified by slug. """ + try: + return Collection.objects.get(slug=slug) + except Collection.DoesNotExist: + return rc.NOT_FOUND + + +class CollectionsHandler(BaseHandler, CollectionDetails): + allowed_methods = ('GET',) + model = Collection + fields = ['url', 'href', 'title'] + + @piwik_track + def read(self, request): + """ Returns all collections. """ + return Collection.objects.all() + + +class TagDetails(object): + """Custom Tag fields.""" + + @classmethod + def href(cls, tag): + """ Returns URI in the API for the tag. """ + + return API_BASE + reverse("api_tag", args=[category_plural[tag.category], tag.slug]) + + @classmethod + def url(cls, tag): + """ Returns URL on the site. """ + + return WL_BASE + tag.get_absolute_url() + + +class TagDetailHandler(BaseHandler, TagDetails): + """ Responsible for details of a single Tag object. """ + + fields = ['name', 'url', 'sort_key', 'description'] + + @piwik_track + def read(self, request, category, slug): + """ Returns details of a tag, identified by category and slug. """ + + try: + category_sng = category_singular[category] + except KeyError, e: + return rc.NOT_FOUND + + try: + return Tag.objects.get(category=category_sng, slug=slug) + except Tag.DoesNotExist: + return rc.NOT_FOUND + + +class TagsHandler(BaseHandler, TagDetails): + """ Main handler for Tag objects. + + Responsible for lists of Tag objects + and fields used for representing Tags. + + """ + allowed_methods = ('GET',) + model = Tag + fields = ['name', 'href', 'url'] + + @piwik_track + def read(self, request, category=None, pk=None): + """ Lists all tags in the category (eg. all themes). """ + if pk is not None: + try: + return Tag.objects.exclude(category='set').get(pk=pk) + except Book.DoesNotExist: + return rc.NOT_FOUND + + try: + category_sng = category_singular[category] + except KeyError, e: + return rc.NOT_FOUND + + tags = Tag.objects.filter(category=category_sng).exclude(items=None) + if tags.exists(): + return tags + else: + return rc.NOT_FOUND + + +class FragmentDetails(object): + """Custom Fragment fields.""" + + @classmethod + def href(cls, fragment): + """ Returns URI in the API for the fragment. """ + + return API_BASE + reverse("api_fragment", + args=[fragment.book.slug, fragment.anchor]) + + @classmethod + def url(cls, fragment): + """ Returns URL on the site for the fragment. """ + + return WL_BASE + fragment.get_absolute_url() + + @classmethod + def themes(cls, fragment): + """ Returns a list of theme tags for the fragment. """ + + return fragment.tags.filter(category='theme') + + +class FragmentDetailHandler(BaseHandler, FragmentDetails): + fields = ['book', 'anchor', 'text', 'url', 'themes'] + + @piwik_track + def read(self, request, book, anchor): + """ Returns details of a fragment, identified by book slug and anchor. """ + try: + return Fragment.objects.get(book__slug=book, anchor=anchor) + except Fragment.DoesNotExist: + return rc.NOT_FOUND + + +class FragmentsHandler(BaseHandler, FragmentDetails): + """ Main handler for Fragments. + + Responsible for lists of Fragment objects + and fields used for representing Fragments. + + """ + model = Fragment + fields = ['book', 'url', 'anchor', 'href'] + allowed_methods = ('GET',) + + categories = set(['author', 'epoch', 'kind', 'genre', 'book', 'theme']) + + @piwik_track + def read(self, request, tags): + """ Lists all fragments with given book, tags, themes. + + :param tags: should be a path of categories and slugs, i.e.: + books/book-slug/authors/an-author/themes/a-theme/ + + """ + try: + tags, ancestors = read_tags(tags, allowed=self.categories) + except ValueError: + return rc.NOT_FOUND + fragments = Fragment.tagged.with_all(tags).select_related('book') + if fragments.exists(): + return fragments + else: + return rc.NOT_FOUND + + +class PictureHandler(BaseHandler): + model = Picture + fields = ('slug', 'title') + allowed_methods = ('POST',) + + def create(self, request): + if not request.user.has_perm('picture.add_picture'): + return rc.FORBIDDEN + + data = json.loads(request.POST.get('data')) + form = PictureImportForm(data) + if form.is_valid(): + form.save() + return rc.CREATED + else: + return rc.NOT_FOUND