from lxml import etree import os import uploader def get_fb2_info(tmp_file_path, original_file_name, original_file_extension): ns = { 'fb':'http://www.gribuser.ru/xml/fictionbook/2.0', 'l':'http://www.w3.org/1999/xlink', } fb2_file = open(tmp_file_path) tree = etree.fromstring(fb2_file.read()) authors = tree.xpath('/fb:FictionBook/fb:description/fb:title-info/fb:author', namespaces=ns) def get_author(element): return element.xpath('fb:first-name/text()', namespaces=ns)[0] + ' ' + element.xpath('fb:middle-name/text()', namespaces=ns)[0] + ' ' + element.xpath('fb:last-name/text()', namespaces=ns)[0] author = ", ".join(map(get_author, authors)) title = unicode(tree.xpath('/fb:FictionBook/fb:description/fb:title-info/fb:book-title/text()', namespaces=ns)[0]) description = unicode(tree.xpath('/fb:FictionBook/fb:description/fb:publish-info/fb:book-name/text()', namespaces=ns)[0]) return uploader.BookMeta( file_path = tmp_file_path, extension = original_file_extension, title = title, author = author, cover = None, description = description, tags = "", series = "", series_id="")