X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/96ffe46ed194b842c5df711fd0135d51655b880d..41e79f5deff58b34f185c8c7426f505793c1d9c7:/apps/catalogue/models.py
diff --git a/apps/catalogue/models.py b/apps/catalogue/models.py
index 18e7e804d..2008f0970 100644
--- a/apps/catalogue/models.py
+++ b/apps/catalogue/models.py
@@ -1,4 +1,7 @@
# -*- 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.
+#
from django.db import models
from django.db.models import permalink, Q
from django.utils.translation import ugettext_lazy as _
@@ -6,12 +9,16 @@ from django.contrib.auth.models import User
from django.core.files import File
from django.template.loader import render_to_string
from django.utils.safestring import mark_safe
+from django.utils.translation import get_language
from django.core.urlresolvers import reverse
+from datetime import datetime
from newtagging.models import TagBase
from newtagging import managers
+from catalogue.fields import JSONField
from librarian import html, dcparser
+from mutagen import id3
TAG_CATEGORIES = (
@@ -21,6 +28,7 @@ TAG_CATEGORIES = (
('genre', _('genre')),
('theme', _('theme')),
('set', _('set')),
+ ('book', _('book')),
)
@@ -28,49 +36,108 @@ class TagSubcategoryManager(models.Manager):
def __init__(self, subcategory):
super(TagSubcategoryManager, self).__init__()
self.subcategory = subcategory
-
+
def get_query_set(self):
return super(TagSubcategoryManager, self).get_query_set().filter(category=self.subcategory)
class Tag(TagBase):
name = models.CharField(_('name'), max_length=50, db_index=True)
- slug = models.SlugField(_('slug'), max_length=120, unique=True, db_index=True)
+ slug = models.SlugField(_('slug'), max_length=120, db_index=True)
sort_key = models.SlugField(_('sort key'), max_length=120, db_index=True)
- category = models.CharField(_('category'), max_length=50, blank=False, null=False,
+ category = models.CharField(_('category'), max_length=50, blank=False, null=False,
db_index=True, choices=TAG_CATEGORIES)
description = models.TextField(_('description'), blank=True)
main_page = models.BooleanField(_('main page'), default=False, db_index=True, help_text=_('Show tag on main page'))
-
+
user = models.ForeignKey(User, blank=True, null=True)
book_count = models.IntegerField(_('book count'), default=0, blank=False, null=False)
+ death = models.IntegerField(_(u'year of death'), blank=True, null=True)
+ gazeta_link = models.CharField(blank=True, max_length=240)
+ wiki_link = models.CharField(blank=True, max_length=240)
- def has_description(self):
- return len(self.description) > 0
- has_description.short_description = _('description')
- has_description.boolean = True
+ categories_rev = {
+ 'autor': 'author',
+ 'epoka': 'epoch',
+ 'rodzaj': 'kind',
+ 'gatunek': 'genre',
+ 'motyw': 'theme',
+ 'polka': 'set',
+ }
+ categories_dict = dict((item[::-1] for item in categories_rev.iteritems()))
- @permalink
- def get_absolute_url(self):
- return ('catalogue.views.tagged_object_list', [self.slug])
-
class Meta:
ordering = ('sort_key',)
verbose_name = _('tag')
verbose_name_plural = _('tags')
-
+ unique_together = (("slug", "category"),)
+
def __unicode__(self):
return self.name
+ def __repr__(self):
+ return "Tag(slug=%r)" % self.slug
+
+ @permalink
+ def get_absolute_url(self):
+ return ('catalogue.views.tagged_object_list', [self.url_chunk])
+
+ def has_description(self):
+ return len(self.description) > 0
+ has_description.short_description = _('description')
+ has_description.boolean = True
+
+ def alive(self):
+ return self.death is None
+
+ def in_pd(self):
+ """ tests whether an author is in public domain """
+ return self.death is not None and self.goes_to_pd() <= datetime.now().year
+
+ def goes_to_pd(self):
+ """ calculates the year of public domain entry for an author """
+ return self.death + 71 if self.death is not None else None
+
@staticmethod
def get_tag_list(tags):
if isinstance(tags, basestring):
- tag_slugs = tags.split('/')
- return [Tag.objects.get(slug=slug) for slug in tag_slugs]
+ real_tags = []
+ ambiguous_slugs = []
+ category = None
+ tags_splitted = tags.split('/')
+ for index, name in enumerate(tags_splitted):
+ if name in Tag.categories_rev:
+ category = Tag.categories_rev[name]
+ else:
+ if category:
+ real_tags.append(Tag.objects.get(slug=name, category=category))
+ category = None
+ else:
+ try:
+ real_tags.append(Tag.objects.exclude(category='book').get(slug=name))
+ except Tag.MultipleObjectsReturned, e:
+ ambiguous_slugs.append(name)
+
+ if category:
+ # something strange left off
+ raise Tag.DoesNotExist()
+ if ambiguous_slugs:
+ # some tags should be qualified
+ e = Tag.MultipleObjectsReturned()
+ e.tags = real_tags
+ e.ambiguous_slugs = ambiguous_slugs
+ raise e
+ else:
+ return real_tags
else:
return TagBase.get_tag_list(tags)
+
+ @property
+ def url_chunk(self):
+ return '/'.join((Tag.categories_dict[self.category], self.slug))
+# TODO: why is this hard-coded ?
def book_upload_path(ext):
def get_dynamic_path(book, filename):
return 'lektura/%s.%s' % (book.slug, ext)
@@ -84,123 +151,222 @@ class Book(models.Model):
created_at = models.DateTimeField(_('creation date'), auto_now=True)
_short_html = models.TextField(_('short HTML'), editable=False)
parent_number = models.IntegerField(_('parent number'), default=0)
-
+ extra_info = JSONField(_('extra information'))
+ gazeta_link = models.CharField(blank=True, max_length=240)
+ wiki_link = models.CharField(blank=True, max_length=240)
+
+
# Formats
xml_file = models.FileField(_('XML file'), upload_to=book_upload_path('xml'), blank=True)
html_file = models.FileField(_('HTML file'), upload_to=book_upload_path('html'), blank=True)
pdf_file = models.FileField(_('PDF file'), upload_to=book_upload_path('pdf'), blank=True)
+ epub_file = models.FileField(_('EPUB file'), upload_to=book_upload_path('epub'), blank=True)
odt_file = models.FileField(_('ODT file'), upload_to=book_upload_path('odt'), blank=True)
txt_file = models.FileField(_('TXT file'), upload_to=book_upload_path('txt'), blank=True)
-
+ mp3_file = models.FileField(_('MP3 file'), upload_to=book_upload_path('mp3'), blank=True)
+ ogg_file = models.FileField(_('OGG file'), upload_to=book_upload_path('ogg'), blank=True)
+
parent = models.ForeignKey('self', blank=True, null=True, related_name='children')
-
+
objects = models.Manager()
tagged = managers.ModelTaggedItemManager(Tag)
tags = managers.TagDescriptor(Tag)
+ _tag_counter = JSONField(null=True, editable=False)
+ _theme_counter = JSONField(null=True, editable=False)
+
+ class AlreadyExists(Exception):
+ pass
+
+ class Meta:
+ ordering = ('title',)
+ verbose_name = _('book')
+ verbose_name_plural = _('books')
+
+ def __unicode__(self):
+ return self.title
+
+ def save(self, force_insert=False, force_update=False, reset_short_html=True, refresh_mp3=True):
+ if reset_short_html:
+ # Reset _short_html during save
+ for key in filter(lambda x: x.startswith('_short_html'), self.__dict__):
+ self.__setattr__(key, '')
+
+ book = super(Book, self).save(force_insert, force_update)
+
+ if refresh_mp3 and self.mp3_file:
+ print self.mp3_file, self.mp3_file.path
+ extra_info = self.get_extra_info_value()
+ extra_info.update(self.get_mp3_info())
+ self.set_extra_info_value(extra_info)
+ book = super(Book, self).save(force_insert, force_update)
+
+ return book
+
+ @permalink
+ def get_absolute_url(self):
+ return ('catalogue.views.book_detail', [self.slug])
+
@property
def name(self):
return self.title
+ def book_tag(self):
+ slug = ('l-' + self.slug)[:120]
+ book_tag, created = Tag.objects.get_or_create(slug=slug, category='book')
+ if created:
+ book_tag.name = self.title[:50]
+ book_tag.sort_key = slug
+ book_tag.save()
+ return book_tag
+
def short_html(self):
- if len(self._short_html):
- return mark_safe(self._short_html)
+ key = '_short_html_%s' % get_language()
+ short_html = getattr(self, key)
+
+ if short_html and len(short_html):
+ return mark_safe(short_html)
else:
- tags = self.tags.filter(~Q(category__in=('set', 'theme')))
- tags = [u'%s' % (tag.get_absolute_url(), tag.name) for tag in tags]
+ tags = self.tags.filter(~Q(category__in=('set', 'theme', 'book')))
+ tags = [mark_safe(u'%s' % (tag.get_absolute_url(), tag.name)) for tag in tags]
formats = []
if self.html_file:
- formats.append(u'Czytaj online' % reverse('book_text', kwargs={'slug': self.slug}))
+ formats.append(u'%s' % (reverse('book_text', kwargs={'slug': self.slug}), _('Read online')))
if self.pdf_file:
- formats.append(u'Plik PDF' % self.pdf_file.url)
+ formats.append(u'PDF' % self.pdf_file.url)
+ if self.epub_file:
+ formats.append(u'EPUB' % self.epub_file.url)
if self.odt_file:
- formats.append(u'Plik ODT' % self.odt_file.url)
+ formats.append(u'ODT' % self.odt_file.url)
if self.txt_file:
- formats.append(u'Plik TXT' % self.txt_file.url)
-
- self._short_html = unicode(render_to_string('catalogue/book_short.html',
- {'book': self, 'tags': tags, 'formats': formats}))
- self.save()
- return mark_safe(self._short_html)
-
+ formats.append(u'TXT' % self.txt_file.url)
+ if self.mp3_file:
+ formats.append(u'MP3' % self.mp3_file.url)
+ if self.ogg_file:
+ formats.append(u'OGG' % self.ogg_file.url)
+
+ formats = [mark_safe(format) for format in formats]
+
+ setattr(self, key, unicode(render_to_string('catalogue/book_short.html',
+ {'book': self, 'tags': tags, 'formats': formats})))
+ self.save(reset_short_html=False)
+ return mark_safe(getattr(self, key))
+
+
+ def get_mp3_info(self):
+ """Retrieves artist and director names from audio ID3 tags."""
+ audio = id3.ID3(self.mp3_file.path)
+ artist_name = ', '.join(', '.join(tag.text) for tag in audio.getall('TPE1'))
+ director_name = ', '.join(', '.join(tag.text) for tag in audio.getall('TPE3'))
+ return {'artist_name': artist_name, 'director_name': director_name}
+
def has_description(self):
return len(self.description) > 0
has_description.short_description = _('description')
has_description.boolean = True
-
+
def has_pdf_file(self):
return bool(self.pdf_file)
has_pdf_file.short_description = 'PDF'
has_pdf_file.boolean = True
-
+
+ def has_epub_file(self):
+ return bool(self.epub_file)
+ has_epub_file.short_description = 'EPUB'
+ has_epub_file.boolean = True
+
def has_odt_file(self):
return bool(self.odt_file)
has_odt_file.short_description = 'ODT'
has_odt_file.boolean = True
-
+
def has_html_file(self):
return bool(self.html_file)
has_html_file.short_description = 'HTML'
has_html_file.boolean = True
- class AlreadyExists(Exception):
- pass
-
- @staticmethod
- def from_xml_file(xml_file, overwrite=False):
+ @classmethod
+ def from_xml_file(cls, xml_file, overwrite=False):
+ # use librarian to parse meta-data
+ book_info = dcparser.parse(xml_file)
+
+ if not isinstance(xml_file, File):
+ xml_file = File(xml_file)
+
+ try:
+ return cls.from_text_and_meta(xml_file, book_info, overwrite)
+ finally:
+ xml_file.close()
+
+ @classmethod
+ def from_text_and_meta(cls, raw_file, book_info, overwrite=False):
from tempfile import NamedTemporaryFile
from slughifi import slughifi
from markupstring import MarkupString
-
+
# Read book metadata
- book_info = dcparser.parse(xml_file)
book_base, book_slug = book_info.url.rsplit('/', 1)
book, created = Book.objects.get_or_create(slug=book_slug)
-
+
if created:
book_shelves = []
else:
if not overwrite:
- raise Book.AlreadyExists('Book %s already exists' % book_slug)
+ raise Book.AlreadyExists(_('Book %s already exists') % book_slug)
# Save shelves for this book
- book_shelves = book.tags.filter(category='set')
-
+ book_shelves = list(book.tags.filter(category='set'))
+
book.title = book_info.title
+ book.set_extra_info_value(book_info.to_dict())
book._short_html = ''
book.save()
-
+
book_tags = []
- for category in ('kind', 'genre', 'author', 'epoch'):
+ for category in ('kind', 'genre', 'author', 'epoch'):
tag_name = getattr(book_info, category)
tag_sort_key = tag_name
if category == 'author':
tag_sort_key = tag_name.last_name
tag_name = ' '.join(tag_name.first_names) + ' ' + tag_name.last_name
- tag, created = Tag.objects.get_or_create(slug=slughifi(tag_name))
+ tag, created = Tag.objects.get_or_create(slug=slughifi(tag_name), category=category)
if created:
tag.name = tag_name
tag.sort_key = slughifi(tag_sort_key)
- tag.category = category
tag.save()
book_tags.append(tag)
+
book.tags = book_tags
-
+
+ book_tag = book.book_tag()
+
if hasattr(book_info, 'parts'):
for n, part_url in enumerate(book_info.parts):
base, slug = part_url.rsplit('/', 1)
- child_book = Book.objects.get(slug=slug)
- child_book.parent = book
- child_book.parent_number = n
- child_book.save()
-
+ try:
+ child_book = Book.objects.get(slug=slug)
+ child_book.parent = book
+ child_book.parent_number = n
+ child_book.save()
+ except Book.DoesNotExist, e:
+ raise Book.DoesNotExist(_('Book with slug = "%s" does not exist.') % slug)
+
+ book_descendants = list(book.children.all())
+ while len(book_descendants) > 0:
+ child_book = book_descendants.pop(0)
+ child_book.tags = list(child_book.tags) + [book_tag]
+ child_book.save()
+ for fragment in child_book.fragments.all():
+ fragment.tags = set(list(fragment.tags) + [book_tag])
+ book_descendants += list(child_book.children.all())
+
# Save XML and HTML files
- book.xml_file.save('%s.xml' % book.slug, File(file(xml_file)), save=False)
-
+ book.xml_file.save('%s.xml' % book.slug, raw_file, save=False)
+
html_file = NamedTemporaryFile()
- if html.transform(book.xml_file.path, html_file):
+ if html.transform(book.xml_file.path, html_file, parse_dublincore=False):
book.html_file.save('%s.html' % book.slug, File(html_file), save=False)
-
+
# Extract fragments
closed_fragments, open_fragments = html.extract_fragments(book.html_file.path)
book_themes = []
@@ -209,43 +375,64 @@ class Book(models.Model):
short_text = ''
if (len(MarkupString(text)) > 240):
short_text = unicode(MarkupString(text)[:160])
- new_fragment, created = Fragment.objects.get_or_create(anchor=fragment.id, book=book,
+ new_fragment, created = Fragment.objects.get_or_create(anchor=fragment.id, book=book,
defaults={'text': text, 'short_text': short_text})
-
+
try:
theme_names = [s.strip() for s in fragment.themes.split(',')]
except AttributeError:
continue
themes = []
for theme_name in theme_names:
- tag, created = Tag.objects.get_or_create(slug=slughifi(theme_name))
+ tag, created = Tag.objects.get_or_create(slug=slughifi(theme_name), category='theme')
if created:
tag.name = theme_name
tag.sort_key = slughifi(theme_name)
- tag.category = 'theme'
tag.save()
themes.append(tag)
new_fragment.save()
- new_fragment.tags = list(book.tags) + themes
+ new_fragment.tags = set(list(book.tags) + themes + [book_tag])
book_themes += themes
-
+
book_themes = set(book_themes)
- book.tags = list(book.tags) + list(book_themes) + list(book_shelves)
-
+ book.tags = list(book.tags) + list(book_themes) + book_shelves
+
book.save()
return book
- @permalink
- def get_absolute_url(self):
- return ('catalogue.views.book_detail', [self.slug])
-
- class Meta:
- ordering = ('title',)
- verbose_name = _('book')
- verbose_name_plural = _('books')
+
+ def refresh_tag_counter(self):
+ tags = {}
+ for child in self.children.all().order_by():
+ for tag_pk, value in child.tag_counter.iteritems():
+ tags[tag_pk] = tags.get(tag_pk, 0) + value
+ for tag in self.tags.exclude(category__in=('book', 'theme', 'set')).order_by():
+ tags[tag.pk] = 1
+ self.set__tag_counter_value(tags)
+ self.save(reset_short_html=False, refresh_mp3=False)
+ return tags
+
+ @property
+ def tag_counter(self):
+ if self._tag_counter is None:
+ return self.refresh_tag_counter()
+ return dict((int(k), v) for k, v in self.get__tag_counter_value().iteritems())
- def __unicode__(self):
- return self.title
+ def refresh_theme_counter(self):
+ tags = {}
+ for fragment in Fragment.tagged.with_any([self.book_tag()]).order_by():
+ for tag in fragment.tags.filter(category='theme').order_by():
+ tags[tag.pk] = tags.get(tag.pk, 0) + 1
+ self.set__theme_counter_value(tags)
+ self.save(reset_short_html=False, refresh_mp3=False)
+ return tags
+
+ @property
+ def theme_counter(self):
+ if self._theme_counter is None:
+ return self.refresh_theme_counter()
+ return dict((int(k), v) for k, v in self.get__theme_counter_value().iteritems())
+
class Fragment(models.Model):
@@ -258,24 +445,55 @@ class Fragment(models.Model):
objects = models.Manager()
tagged = managers.ModelTaggedItemManager(Tag)
tags = managers.TagDescriptor(Tag)
-
+
+ class Meta:
+ ordering = ('book', 'anchor',)
+ verbose_name = _('fragment')
+ verbose_name_plural = _('fragments')
+
+ def get_absolute_url(self):
+ return '%s#m%s' % (reverse('book_text', kwargs={'slug': self.book.slug}), self.anchor)
+
def short_html(self):
- if len(self._short_html):
- return mark_safe(self._short_html)
+ key = '_short_html_%s' % get_language()
+ short_html = getattr(self, key)
+ if short_html and len(short_html):
+ return mark_safe(short_html)
else:
- book_authors = [u'%s' % (tag.get_absolute_url(), tag.name)
+ book_authors = [mark_safe(u'%s' % (tag.get_absolute_url(), tag.name))
for tag in self.book.tags if tag.category == 'author']
-
- self._short_html = unicode(render_to_string('catalogue/fragment_short.html',
- {'fragment': self, 'book': self.book, 'book_authors': book_authors}))
+
+ setattr(self, key, unicode(render_to_string('catalogue/fragment_short.html',
+ {'fragment': self, 'book': self.book, 'book_authors': book_authors})))
self.save()
- return mark_safe(self._short_html)
-
- def get_absolute_url(self):
- return '%s#m%s' % (reverse('book_text', kwargs={'slug': self.book.slug}), self.anchor)
-
+ return mark_safe(getattr(self, key))
+
+
+class BookStub(models.Model):
+ title = models.CharField(_('title'), max_length=120)
+ author = models.CharField(_('author'), max_length=120)
+ pd = models.IntegerField(_('goes to public domain'), null=True, blank=True)
+ slug = models.SlugField(_('slug'), max_length=120, unique=True, db_index=True)
+ translator = models.TextField(_('translator'), blank=True)
+ translator_death = models.TextField(_('year of translator\'s death'), blank=True)
+
class Meta:
- ordering = ('book', 'anchor',)
- verbose_name = _('fragment')
- verbose_name_plural = _('fragments')
+ ordering = ('title',)
+ verbose_name = _('book stub')
+ verbose_name_plural = _('book stubs')
+
+ def __unicode__(self):
+ return self.title
+
+ @permalink
+ def get_absolute_url(self):
+ return ('catalogue.views.book_detail', [self.slug])
+
+ def in_pd(self):
+ return self.pd is not None and self.pd <= datetime.now().year
+
+ @property
+ def name(self):
+ return self.title
+