Class variable dictionary not saving with pickle.dump in python 2.7 -

i using pickle save object graph dumping root. when load root has instance variables , connected object nodes. saving nodes in class variable of type dictionary. class variable full before being saved after unpickle data empty.

here class using:

class page():      __crawled = {}      def __init__(self, title = '', link = '', relatedurls = []):         self.__title = title         self.__link = link         self.__relatedurls = relatedurls         self.__related = []       @property     def relatedurls(self):         return self.__relatedurls      @property     def title(self):         return self.__title      @property     def related(self):         return self.__related      @property     def crawled(self):         return self.__crawled      def crawl(self,url):         if url not in self.__crawled:             webpage = urlopen(url).read()             patfindertitle = re.compile('<title>(.*)</title>')             patfinderlink = re.compile('<link rel="canonical" href="([^"]*)" />')             patfinderrelated = re.compile('<li><a href="([^"]*)"')              findpattitle = re.findall(patfindertitle, webpage)             findpatlink = re.findall(patfinderlink, webpage)             findpatrelated = re.findall(patfinderrelated, webpage)             newpage = page(findpattitle,findpatlink,findpatrelated)             self.__related.append(newpage)             self.__crawled[url] = newpage         else:             self.__related.append(self.__crawled[url])      def crawlrelated(self):         link in self.__relatedurls:             self.crawl(link) 

i save such:

with open('medtwicegraph.dat','w') outf:     pickle.dump(root,outf) 

and load such:

def loadgraph(filename): #returns root     open(filename,'r') inf:         return pickle.load(inf)  root = loadgraph('medtwicegraph.dat') 

all data loads except class variable __crawled.

what doing wrong?

python doesn't pickle class objects. saves names , find them. documentation of pickle:

similarly, classes pickled named reference, same restrictions in unpickling environment apply. note none of class’s code or data pickled, in following example class attribute attr not restored in unpickling environment:

class foo:     attr = 'a class attr'  picklestring = pickle.dumps(foo) 

these restrictions why picklable functions , classes must defined in top level of module.

similarly, when class instances pickled, class’s code , data not pickled along them. instance data pickled. done on purpose, can fix bugs in class or add methods class , still load objects created earlier version of class. if plan have long-lived objects see many versions of class, may worthwhile put version number in objects suitable conversions can made class’s __setstate__() method.

in example fix problems changing __crawled instance attribute or global variable.