home *** CD-ROM | disk | FTP | other *** search
- #!/usr/bin/env python
-
- __license__ = 'GPL v3'
- __copyright__ = '2009, Darko Miletic <darko.miletic at gmail.com>'
- '''
- www.thebudgetfashionista.com
- '''
-
- from calibre.web.feeds.recipes import BasicNewsRecipe
-
- class TheBudgetFashionista(BasicNewsRecipe):
- title = 'The Budget Fashionista'
- __author__ = 'Darko Miletic'
- description = 'Saving your money since 2003'
- oldest_article = 7
- max_articles_per_feed = 100
- no_stylesheets = True
- use_embedded_content = False
- encoding = 'utf-8'
- publisher = 'TBF GROUP, LLC.'
- category = 'news, fashion, comsetics, women'
- lang = 'en-US'
- language = 'en'
-
-
- conversion_options = {
- 'comment' : description
- , 'tags' : category
- , 'publisher' : publisher
- , 'language' : lang
- }
-
- keep_only_tags = [dict(name='div', attrs={'class':'columnLeft'})]
- remove_tags_after = dict(name='div', attrs={'class':'postDetails'})
- remove_tags = [dict(name=['object','link','script','iframe','form','login-button'])]
-
- feeds = [(u'Articles', u'http://www.thebudgetfashionista.com/feeds/atom/')]
-
- def preprocess_html(self, soup):
- for it in soup.findAll('img'):
- if it.parent.name == 'a':
- it.parent.name = 'div'
- return soup;
-
-