-# _PCHARS_DICT = dict(zip((ord(x) for x in u"ĄĆĘŁŃÓŚŻŹąćęłńóśżź "), u"ACELNOSZZacelnoszz_"))
-_PCHARS_DICT = dict(zip((ord(x) for x in u" "), u"_"))
-
-# I know this is barbaric, but I didn't find a better solution ;(
-def split_name(name):
- parts = name.translate(_PCHARS_DICT).split('__')
- return parts
-
-def join_name(*parts, **kwargs):
- name = u'__'.join(p.translate(_PCHARS_DICT) for p in parts)
- logger.info("JOIN %r -> %r", parts, name)
- return name
-
-def normalize_name(name):
- """
- >>> normalize_name("gąska".decode('utf-8'))
- u'g\u0105ska'
- """
- return name.translate(_PCHARS_DICT).lower()
-
-STAGE_TAGS_RE = re.compile(r'^#stage-finished: (.*)$', re.MULTILINE)
-
-
-class DocumentStorage(object):
- def __init__(self, path):
- self.vstorage = vstorage.VersionedStorage(path)
-
- def get(self, name, revision=None):
- text, rev = self.vstorage.page_text(name, revision)
- return Document(self, name=name, text=text, revision=rev)
-
- def get_by_tag(self, name, tag):
- text, rev = self.vstorage.page_text_by_tag(name, tag)
- return Document(self, name=name, text=text, revision=rev)
-
- def revert(self, name, revision):
- text, rev = self.vstorage.revert(name, revision)
- return Document(self, name=name, text=text, revision=rev)
-
- def get_or_404(self, *args, **kwargs):
- try:
- return self.get(*args, **kwargs)
- except DocumentNotFound:
- raise Http404
-
- def put(self, document, author, comment, parent=None):
- self.vstorage.save_text(
- title=document.name,
- text=document.text,
- author=author,
- comment=comment,
- parent=parent)
-
- return document
-
- def create_document(self, text, name):
- title = u', '.join(p.title for p in split_name(name))
-
- if text is None:
- text = u''
-
- document = Document(self, name=name, text=text, title=title)
- return self.put(document, u"<wiki>", u"Document created.")
-
- def delete(self, name, author, comment):
- self.vstorage.delete_page(name, author, comment)
-
- def all(self):
- return list(self.vstorage.all_pages())
+class Book(models.Model):
+ """ A document edited on the wiki """
+
+ title = models.CharField(_('title'), max_length=255)
+ slug = models.SlugField(_('slug'), max_length=128, unique=True)
+ gallery = models.CharField(_('scan gallery name'), max_length=255, blank=True)
+
+ parent = models.ForeignKey('self', null=True, blank=True, verbose_name=_('parent'), related_name="children")
+ parent_number = models.IntegerField(_('parent number'), null=True, blank=True, db_index=True)
+
+ _list_html = models.TextField(editable=False, null=True)
+
+ class NoTextError(BaseException):
+ pass
+
+ class Meta:
+ ordering = ['parent_number', 'title']
+ verbose_name = _('book')
+ verbose_name_plural = _('books')
+
+ def __unicode__(self):
+ return self.title
+
+ def get_absolute_url(self):
+ return reverse("wiki_book", args=[self.slug])
+
+ def save(self, reset_list_html=True, *args, **kwargs):
+ if reset_list_html:
+ self._list_html = None
+ return super(Book, self).save(*args, **kwargs)
+
+ @classmethod
+ def create(cls, creator=None, text=u'', *args, **kwargs):
+ """
+ >>> Book.create(slug='x', text='abc').materialize()
+ 'abc'
+ """
+ instance = cls(*args, **kwargs)
+ instance.save()
+ instance[0].commit(author=creator, text=text)
+ return instance
+
+ def __iter__(self):
+ return iter(self.chunk_set.all())
+
+ def __getitem__(self, chunk):
+ return self.chunk_set.all()[chunk]
+
+ def __len__(self):
+ return self.chunk_set.count()
+
+ def list_html(self):
+ if self._list_html is None:
+ print 'rendering', self.title
+ self._list_html = render_to_string('wiki/document_list_item.html',
+ {'book': self})
+ self.save(reset_list_html=False)
+ return mark_safe(self._list_html)
+
+ @staticmethod
+ def publish_tag():
+ return dvcs_models.Tag.get('publish')
+
+ def materialize(self, tag=None):
+ """
+ Get full text of the document compiled from chunks.
+ Takes the current versions of all texts
+ or versions most recently tagged by a given tag.
+ """
+ if tag:
+ changes = [chunk.last_tagged(tag) for chunk in self]
+ else:
+ changes = [chunk.head for chunk in self]
+ if None in changes:
+ raise self.NoTextError('Some chunks have no available text.')
+ return compile_text(change.materialize() for change in changes)
+
+ def publishable(self):
+ if not len(self):
+ return False
+ for chunk in self:
+ if not chunk.publishable():
+ return False
+ return True
+
+ def make_chunk_slug(self, proposed):
+ """
+ Finds a chunk slug not yet used in the book.
+ """
+ slugs = set(c.slug for c in self)
+ i = 1
+ new_slug = proposed
+ while new_slug in slugs:
+ new_slug = "%s-%d" % (proposed, i)
+ i += 1
+ return new_slug
+
+ def append(self, other):
+ number = self[len(self) - 1].number + 1
+ single = len(other) == 1
+ for chunk in other:
+ # move chunk to new book
+ chunk.book = self
+ chunk.number = number
+
+ # try some title guessing
+ if other.title.startswith(self.title):
+ other_title_part = other.title[len(self.title):].lstrip(' /')
+ else:
+ other_title_part = other.title
+
+ if single:
+ # special treatment for appending one-parters:
+ # just use the guessed title and original book slug
+ chunk.comment = other_title_part
+ if other.slug.startswith(self.slug):
+ chunk_slug = other.slug[len(self.slug):].lstrip('-_')
+ else:
+ chunk_slug = other.slug
+ chunk.slug = self.make_chunk_slug(chunk_slug)
+ else:
+ chunk.comment = "%s, %s" % (other_title_part, chunk.comment)
+ chunk.slug = self.make_chunk_slug(chunk.slug)
+ chunk.save()
+ number += 1
+ other.delete()
+
+ @staticmethod
+ def listener_create(sender, instance, created, **kwargs):
+ if created:
+ instance.chunk_set.create(number=1, slug='1')
+
+models.signals.post_save.connect(Book.listener_create, sender=Book)
+
+
+class Chunk(dvcs_models.Document):
+ """ An editable chunk of text. Every Book text is divided into chunks. """
+
+ book = models.ForeignKey(Book, editable=False)
+ number = models.IntegerField()
+ slug = models.SlugField()
+ comment = models.CharField(max_length=255)
+
+ class Meta:
+ unique_together = [['book', 'number'], ['book', 'slug']]
+ ordering = ['number']
+
+ def __unicode__(self):
+ return "%d-%d: %s" % (self.book_id, self.number, self.comment)
+
+ def get_absolute_url(self):
+ return reverse("wiki_editor", args=[self.book.slug, self.slug])
+
+ @classmethod
+ def get(cls, slug, chunk=None):
+ if chunk is None:
+ return cls.objects.get(book__slug=slug, number=1)
+ else:
+ return cls.objects.get(book__slug=slug, slug=chunk)
+
+ def pretty_name(self):
+ return "%s, %s (%d/%d)" % (self.book.title, self.comment,
+ self.number, len(self.book))
+
+ def publishable(self):
+ return self.last_tagged(Book.publish_tag())
+
+ def split(self, slug, comment='', creator=None):
+ """ Create an empty chunk after this one """
+ self.book.chunk_set.filter(number__gt=self.number).update(
+ number=models.F('number')+1)
+ new_chunk = self.book.chunk_set.create(number=self.number+1,
+ creator=creator, slug=slug, comment=comment)
+ return new_chunk
+
+ @staticmethod
+ def listener_saved(sender, instance, created, **kwargs):
+ if instance.book:
+ # save book so that its _list_html is reset
+ instance.book.save()
+
+models.signals.post_save.connect(Chunk.listener_saved, sender=Chunk)
+
+
+class Theme(models.Model):
+ name = models.CharField(_('name'), max_length=50, unique=True)
+
+ class Meta:
+ ordering = ('name',)
+ verbose_name = _('theme')
+ verbose_name_plural = _('themes')
+
+ def __unicode__(self):
+ return self.name
+
+ def __repr__(self):
+ return "Theme(name=%r)" % self.name