X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/83559fba0846ad77721b31231d1d807a91e52157..eb11e65717bdbddbc3ce08fa437739fd7bf5c279:/lib/markupstring.py?ds=sidebyside diff --git a/lib/markupstring.py b/lib/markupstring.py index d3ee955b2..2ecf4cf97 100644 --- a/lib/markupstring.py +++ b/lib/markupstring.py @@ -5,21 +5,21 @@ import xml.sax -class simpleHandler (xml.sax.ContentHandler): +class simpleHandler(xml.sax.ContentHandler): """A simple handler that provides us with indices of marked up content.""" - def __init__ (self): + def __init__(self): self.elements = [] #this will contain a list of elements and their start/end indices self.open_elements = [] #this holds info on open elements while we wait for their close self.content = "" - def startElement (self,name,attrs): + def startElement(self,name,attrs): if name=='foobar': return # we require an outer wrapper, which we promptly ignore. self.open_elements.append({'name':name, 'attrs':attrs.copy(), 'start':len(self.content), }) - def endElement (self, name): + def endElement(self, name): if name=='foobar': return # we require an outer wrapper, which we promptly ignore. for i in range(len(self.open_elements)): e = self.open_elements[i] @@ -32,23 +32,23 @@ class simpleHandler (xml.sax.ContentHandler): del self.open_elements[i] return - def characters (self, chunk): + def characters(self, chunk): self.content += chunk -class MarkupString (unicode): +class MarkupString(unicode): """A simple class for dealing with marked up strings. When we are sliced, we return valid marked up strings, preserving markup.""" - def __init__ (self, string): + def __init__(self, string): unicode.__init__(self, string) self.handler = simpleHandler() xml.sax.parseString((u"%s" % string).encode('utf-8'), self.handler) self.raw = self.handler.content - def __getitem__ (self, n): + def __getitem__(self, n): return self.__getslice__(n,n+1) - def __getslice__ (self, s, e): + def __getslice__(self, s, e): # only include relevant elements if not e or e > len(self.raw): e = len(self.raw) elements = filter(lambda tp: (tp[0][1] >= s and # end after the start...