-from lxml.etree import XMLSyntaxError, XSLTApplyError
-
-import os
-import re
-from StringIO import StringIO
-
-class WLDocument(object):
- LINE_SWAP_EXPR = re.compile(r'/\s', re.MULTILINE | re.UNICODE)
- provider = None
-
- def __init__(self, edoc, parse_dublincore=True, provider=None, strict=False):
- self.edoc = edoc
- self.provider = provider
-
- root_elem = edoc.getroot()
-
- dc_path = './/' + RDFNS('RDF')
-
- if root_elem.tag != 'utwor':
- raise ValidationError("Invalid root element. Found '%s', should be 'utwor'" % root_elem.tag)
-
- if parse_dublincore:
- self.rdf_elem = root_elem.find(dc_path)
-
- if self.rdf_elem is None:
- raise NoDublinCore('Document has no DublinCore - which is required.')
-
- self.book_info = dcparser.BookInfo.from_element(
- self.rdf_elem, strict=strict)
- else:
- self.book_info = None
-
- @classmethod
- def from_string(cls, xml, *args, **kwargs):
- return cls.from_file(StringIO(xml), *args, **kwargs)
-
- @classmethod
- def from_file(cls, xmlfile, parse_dublincore=True, provider=None):
-
- # first, prepare for parsing
- if isinstance(xmlfile, basestring):
- file = open(xmlfile, 'rb')
- try:
- data = file.read()
- finally:
- file.close()
- else:
- data = xmlfile.read()
-
- if not isinstance(data, unicode):
- data = data.decode('utf-8')
-
- data = data.replace(u'\ufeff', '')
-
- try:
- parser = etree.XMLParser(remove_blank_text=False)
- tree = etree.parse(StringIO(data.encode('utf-8')), parser)
-
- return cls(tree, parse_dublincore=parse_dublincore, provider=provider)
- except (ExpatError, XMLSyntaxError, XSLTApplyError), e:
- raise ParseError(e)
-
- def swap_endlines(self):
- """Converts line breaks in stanzas into <br/> tags."""
- # only swap inside stanzas
- for elem in self.edoc.iter('strofa'):
- for child in list(elem):
- if child.tail:
- chunks = self.LINE_SWAP_EXPR.split(child.tail)
- ins_index = elem.index(child) + 1
- while len(chunks) > 1:
- ins = etree.Element('br')
- ins.tail = chunks.pop()
- elem.insert(ins_index, ins)
- child.tail = chunks.pop(0)
- if elem.text:
- chunks = self.LINE_SWAP_EXPR.split(elem.text)
- while len(chunks) > 1:
- ins = etree.Element('br')
- ins.tail = chunks.pop()
- elem.insert(0, ins)
- elem.text = chunks.pop(0)
-
- def parts(self):
- if self.provider is None:
- raise NoProvider('No document provider supplied.')
- if self.book_info is None:
- raise NoDublinCore('No Dublin Core in document.')
- for part_uri in self.book_info.parts:
- yield self.from_file(self.provider.by_uri(part_uri),
- provider=self.provider)
-
- def chunk(self, path):
- # convert the path to XPath
- expr = self.path_to_xpath(path)
- elems = self.edoc.xpath(expr)
-
- if len(elems) == 0:
- return None
- else:
- return elems[0]
-
- def path_to_xpath(self, path):
- parts = []
-
- for part in path.split('/'):
- match = re.match(r'([^\[]+)\[(\d+)\]', part)
- if not match:
- parts.append(part)
- else:
- tag, n = match.groups()
- parts.append("*[%d][name() = '%s']" % (int(n)+1, tag) )
-
- if parts[0] == '.':
- parts[0] = ''
-
- return '/'.join(parts)
-
- def transform(self, stylesheet, **options):
- return self.edoc.xslt(stylesheet, **options)
-
- def update_dc(self):
- if self.book_info:
- parent = self.rdf_elem.getparent()
- parent.replace( self.rdf_elem, self.book_info.to_etree(parent) )
-
- def serialize(self):
- self.update_dc()
- return etree.tostring(self.edoc, encoding=unicode, pretty_print=True)
-
- def merge_chunks(self, chunk_dict):
- unmerged = []
-
- for key, data in chunk_dict.iteritems():
- try:
- xpath = self.path_to_xpath(key)
- node = self.edoc.xpath(xpath)[0]
- repl = etree.fromstring(u"<%s>%s</%s>" %(node.tag, data, node.tag) )
- node.getparent().replace(node, repl);
- except Exception, e:
- unmerged.append( repr( (key, xpath, e) ) )
-
- return unmerged
-
- def clean_ed_note(self):
- """ deletes forbidden tags from nota_red """
-
- for node in self.edoc.xpath('|'.join('//nota_red//%s' % tag for tag in
- ('pa', 'pe', 'pr', 'pt', 'begin', 'end', 'motyw'))):
- tail = node.tail
- node.clear()
- node.tag = 'span'
- node.tail = tail
-
- # Converters
-
- def as_html(self, *args, **kwargs):
- from librarian import html
- return html.transform(self, *args, **kwargs)
-
- def as_text(self, *args, **kwargs):
- from librarian import text
- return text.transform(self, *args, **kwargs)
-
- def as_epub(self, *args, **kwargs):
- from librarian import epub
- return epub.transform(self, *args, **kwargs)
-
- def as_pdf(self, *args, **kwargs):
- from librarian import pdf
- return pdf.transform(self, *args, **kwargs)
-
- def as_mobi(self, *args, **kwargs):
- from librarian import mobi
- return mobi.transform(self, *args, **kwargs)
-
- def as_fb2(self, *args, **kwargs):
- from librarian import fb2
- return fb2.transform(self, *args, **kwargs)
-
- def save_output_file(self, output_file, output_path=None,
- output_dir_path=None, make_author_dir=False, ext=None):
- if output_dir_path:
- save_path = output_dir_path
- if make_author_dir:
- save_path = os.path.join(save_path,
- unicode(self.book_info.author).encode('utf-8'))
- save_path = os.path.join(save_path,
- self.book_info.uri.slug)
- if ext:
- save_path += '.%s' % ext
- else:
- save_path = output_path
-
- output_file.save_as(save_path)