Search.
[wolnelektury.git] / src / api / handlers.py
index b8eec2f..5af5cfb 100644 (file)
@@ -1,29 +1,13 @@
-# -*- coding: utf-8 -*-
 # This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
 # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
 #
-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 catalogue.models.tag import prefetch_relations
-from picture.models import Picture
-from picture.forms import PictureImportForm
-
-from stats.utils import piwik_track
+from catalogue.models import Book, Tag
 
-from . import emitters  # Register our emitters
-
-API_BASE = WL_BASE = MEDIA_BASE = lazy(
-    lambda: u'http://' + Site.objects.get_current().domain, unicode)()
 
+WL_BASE = lazy(
+    lambda: 'https://' + Site.objects.get_current().domain, str)()
 
 category_singular = {
     'authors': 'author',
@@ -33,11 +17,6 @@ category_singular = {
     '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, request, allowed):
@@ -50,6 +29,7 @@ def read_tags(tags, request, allowed):
 
     def process(category, slug):
         if category == 'book':
+            # FIXME: Unused?
             try:
                 books.append(Book.objects.get(slug=slug))
             except Book.DoesNotExist:
@@ -85,484 +65,3 @@ def read_tags(tags, request, allowed):
                 for slug in request.GET.getlist(key):
                     process(category, slug)
     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.order_by('parent_number', 'sort_key')
-
-    @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 ''
-
-    @classmethod
-    def cover_thumb(cls, book):
-        return MEDIA_BASE + default.backend.get_thumbnail(
-                    book.cover, "139x193").url if book.cover else ''
-
-    @classmethod
-    def cover_source_image(cls, book):
-        url = book.cover_source()
-        return url.rstrip('/') + '/file/'
-
-
-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', 'fragment_data'] + [
-            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 = book_tag_categories + ['href', 'title', 'url', 'cover', 'cover_thumb', 'slug']
-
-    @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,
-             recommended=False, newest=False, books=None,
-             after=None, before=None, count=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, request, allowed=book_tag_categories)
-        except ValueError:
-            return rc.NOT_FOUND
-
-        if 'after' in request.GET:
-            after = request.GET['after']
-        if 'before' in request.GET:
-            before = request.GET['before']
-        if 'count' in request.GET:
-            count = request.GET['count']
-
-        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:
-            books = books if books is not None else Book.objects.all()
-        books = books.order_by('slug')
-
-        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 recommended:
-            books = books.filter(recommended=True)
-        if newest:
-            books = books.order_by('-created_at')
-
-        if after:
-            books = books.filter(slug__gt=after)
-        if before:
-            books = books.filter(slug__lt=before)
-
-        books = books.only('slug', 'title', 'cover', 'cover_thumb')
-        for category in book_tag_categories:
-            books = prefetch_relations(books, category)
-
-        if count:
-            if before:
-                books = list(reversed(books.order_by('-slug')[:count]))
-            else:
-                books = books[:count]
-
-        return books
-
-    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', 'slug']
-    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.NOT_FOUND
-
-
-class EBooksHandler(AnonymousBooksHandler):
-    fields = ('author', 'href', 'title', 'cover') + tuple(Book.ebook_formats) + ('slug',)
-
-
-class FilterBooksHandler(AnonymousBooksHandler):
-    fields = book_tag_categories + [
-        'href', 'title', 'url', 'cover', 'cover_thumb', 'slug', 'cover_source_image']
-
-    def read(self, request, title_part=None, author_part=None, is_lektura=None, is_audiobook=None,
-             after=None, before=None, count=None):
-        if 'title_part' in request.GET:
-            title_part = request.GET['title_part']
-        if 'author_part' in request.GET:
-            author_part = request.GET['author_part']
-        if 'is_lektura' in request.GET:
-            is_lektura = request.GET['is_lektura']
-        if 'is_audiobook' in request.GET:
-            is_audiobook = request.GET['is_audiobook']
-
-        if count is None:
-            count = 50
-        if is_lektura in ('true', 'false'):
-            is_lektura = is_lektura == 'true'
-        else:
-            is_lektura = None
-        if is_audiobook in ('true', 'false'):
-            is_audiobook = is_audiobook == 'true'
-        books = Book.objects.distinct()
-        if title_part:
-            books = books.filter(title__iregex='\m' + title_part)
-        if author_part is not None:
-            books = books.filter(cached_author__iregex='\m' + author_part)
-        if is_lektura is not None:
-            books = books.filter(has_audience=is_lektura)
-        if is_audiobook is not None:
-            if is_audiobook:
-                books = books.filter(media__type='mp3')
-            else:
-                books = books.exclude(media__type='mp3')
-        return super(FilterBooksHandler, self).read(request, books=books, after=after, before=before, count=count)
-
-
-# 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 book.tag_unicode(category)
-    return get_tag
-
-
-def add_tag_getters():
-    for plural, singular in category_singular.items():
-        setattr(BookDetails, plural, _tags_getter(singular))
-        setattr(BookDetails, singular, _tag_getter(singular))
-
-add_tag_getters()
-
-
-# add fields for files in Book
-def _file_getter(book_format):
-    field = "%s_file" % book_format
-
-    @classmethod
-    def get_file(cls, book):
-        f = getattr(book, field)
-        if f:
-            return MEDIA_BASE + f.url
-        else:
-            return ''
-    return get_file
-
-
-def add_file_getters():
-    for book_format in Book.formats:
-        setattr(BookDetails, book_format, _file_getter(book_format))
-
-add_file_getters()
-
-
-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:
-            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', 'slug']
-
-    @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:
-            return rc.NOT_FOUND
-
-        after = request.GET.get('after')
-        before = request.GET.get('before')
-        count = request.GET.get('count')
-
-        tags = Tag.objects.filter(category=category_sng).exclude(items=None).order_by('slug')
-
-        if after:
-            tags = tags.filter(slug__gt=after)
-        if before:
-            tags = tags.filter(slug__lt=before)
-
-        if count:
-            if before:
-                tags = list(reversed(tags.order_by('-slug')[:count]))
-            else:
-                tags = tags[:count]
-
-        return tags
-
-
-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 = {'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