forked from github/pelican
94 lines
2.9 KiB
Python
94 lines
2.9 KiB
Python
# -*- coding: utf-8 -*-
|
|
from pelican.utils import slugify, truncate_html_words
|
|
from pelican.log import *
|
|
|
|
class Page(object):
|
|
"""Represents a page
|
|
Given a content, and metadatas, create an adequate object.
|
|
|
|
:param content: the string to parse, containing the original content.
|
|
"""
|
|
mandatory_properties = ('title',)
|
|
|
|
def __init__(self, content, metadatas={}, settings={}, filename=None):
|
|
self._content = content
|
|
self.translations = []
|
|
|
|
self.status = "published" # default value
|
|
for key, value in metadatas.items():
|
|
setattr(self, key.lower(), value)
|
|
|
|
if not hasattr(self, 'author'):
|
|
if 'AUTHOR' in settings:
|
|
self.author = settings['AUTHOR']
|
|
|
|
default_lang = settings.get('DEFAULT_LANG').lower()
|
|
if not hasattr(self, 'lang'):
|
|
self.lang = default_lang
|
|
|
|
self.in_default_lang = (self.lang == default_lang)
|
|
|
|
if not hasattr(self, 'slug'):
|
|
self.slug = slugify(self.title)
|
|
|
|
if not hasattr(self, 'save_as'):
|
|
if self.in_default_lang:
|
|
self.save_as = '%s.html' % self.slug
|
|
clean_url = '%s/' % self.slug
|
|
else:
|
|
self.save_as = '%s-%s.html' % (self.slug, self.lang)
|
|
clean_url = '%s-%s/' % (self.slug, self.lang)
|
|
|
|
if settings.get('CLEAN_URLS', False):
|
|
self.url = clean_url
|
|
else:
|
|
self.url = self.save_as
|
|
|
|
if filename:
|
|
self.filename = filename
|
|
|
|
if not hasattr(self, 'date_format'):
|
|
if self.lang in settings['DATE_FORMATS']:
|
|
self.date_format = settings['DATE_FORMATS'][self.lang]
|
|
else:
|
|
self.date_format = settings['DEFAULT_DATE_FORMAT']
|
|
|
|
if hasattr(self, 'date'):
|
|
self.locale_date = self.date.strftime(self.date_format.encode('ascii','xmlcharrefreplace')).decode('utf')
|
|
|
|
if not hasattr(self, 'summary'):
|
|
self.summary = property(lambda self: truncate_html_words(self.content, 50)).__get__(self, Page)
|
|
|
|
# store the settings ref.
|
|
self._settings = settings
|
|
|
|
def check_properties(self):
|
|
"""test that each mandatory property is set."""
|
|
for prop in self.mandatory_properties:
|
|
if not hasattr(self, prop):
|
|
raise NameError(prop)
|
|
|
|
@property
|
|
def content(self):
|
|
if hasattr(self, "_get_content"):
|
|
content = self._get_content()
|
|
else:
|
|
content = self._content
|
|
return content
|
|
|
|
|
|
class Article(Page):
|
|
mandatory_properties = ('title', 'date', 'category')
|
|
|
|
|
|
class Quote(Page):
|
|
base_properties = ('author', 'date')
|
|
|
|
|
|
def is_valid_content(content, f):
|
|
try:
|
|
content.check_properties()
|
|
return True
|
|
except NameError, e:
|
|
error(u"Skipping %s: impossible to find informations about '%s'" % (f, e))
|
|
return False
|