""" Models for the history. This history should be able to generate history entries, which looks like human-based browsing, according to a dedicated user interests, keywords... """ from collections import namedtuple import random from math import floor from queue import Queue from xml.etree import ElementTree as ET from datetime import datetime from django.db import models import profiles.models as profiles from crawl import crawl from pinocchio.settings import HISTORY_MIN from .tor_runner import TorInstance class InvalidXml(Exception): def __init__(self, what='unexpected XML data.'): super().__init__() self.what = what def __str__(self): return "Invalid XML: " + self.what class HistoryEntry(models.Model): """ A history entry, aka a url, and a timestamp. """ search = models.URLField(help_text="The url to be searched") timestamp = models.DateTimeField() history = models.ForeignKey( 'History', on_delete=models.CASCADE ) def __str__(self): """ Returns the string representation of a history entry. """ return "{} : {}".format(self.timestamp, self.search) def to_xml(self, xml_root): entry = ET.Element('history') entry_url = ET.Element('url') entry_url.text = self.search entry_ts = ET.Element('timestamp') entry_ts.text = self.timestamp.timestamp() entry.append(entry_url) entry.append(entry_ts) xml_root.append(entry) @staticmethod def from_xml(xml_root, in_history): if xml_root.tag != 'history': raise InvalidXml("expected tag here.") url, timestamp = None, None for child in xml_root: if child.tag == 'url': url = child.text elif child.tag == 'timestamp': try: timestamp = datetime.fromtimestamp(child.text) except TypeError: raise InvalidXml("invalid timestamp {}".format(child.text)) else: raise InvalidXml("unknown tag {} as child of ".format( child.tag)) output = HistoryEntry() output.search = url output.timestamp = timestamp output.history = in_history return output class History(models.Model): """ A history for a user, containing some web connections (http, https). Each history is timed, in a human-behaviour manner. """ start_ts = models.DateTimeField( help_text=('The starting timestamp of the history. Useful for ' 'cron-like structure.') ) played = models.BooleanField(default=False) user = models.ForeignKey( profiles.Profile, on_delete=models.CASCADE ) def return_history(self): """ Returns the history, sorted by increasing timestamps """ output_history = self.historyentry_set.order_by('timestamp') output_history = [(item.search, item.timestamp.date()) for item in output_history] return output_history def __str__(self): """ Returns the string representation of a history. """ history_set = self.historyentry_set.order_by('timestamp') header = "[History]:\n" return header + "\n".join(history_set) def play_histories(self): """ Actually plays the history. """ self.played = True runner = TorInstance(self.history) self.save() def to_xml(self, xml_root): ''' Exports the current history to xml ''' hist_node = ET.Element("history", attrib={ 'start-ts': self.start_ts, 'played': 1 if self.played else 0, 'user': self.user.pk, }) xml_root.append(hist_node) for entry in self.historyentry_set: entry.to_xml(hist_node) @staticmethod def from_xml(xml_root): ''' Loads an history from an XML file ''' REQUIRED_ATTR = ['start-ts', 'played', 'user'] if xml_root.tag != 'history': raise InvalidXml('unexpected node {} as root of an history'.format( xml_root.tag)) for attr in REQUIRED_ATTR: if attr not in xml_root.attrib: raise InvalidXml(('missing attribute "{}" for tag of type ' 'history').format(attr)) start_ts = xml_root.attrib['start-ts'] played = xml_root.attrib['played'] user_pk = xml_root.attrib['user'] users = History.objects.filter(pk=1) if len(users) != 1: raise InvalidXml('primary key for History {} is invalid'.format( user_pk)) output = History() output.start_ts = start_ts output.played = played > 0 output.user = users[0] for child in xml_root: HistoryEntry.from_xml(child, output) return output PartialHistoryEntry = namedtuple('PartialHistoryEntry', ['url', 'timestamp']) def generate_partial_history(user, t_start): """ Generate the part of the history resulting from the crawl starting at the given url. """ timestamp = t_start result = [] basis = generate_first_url(user) result.append(PartialHistoryEntry(basis, timestamp)) t_start += 5 * random.weibullvariate(1, 1.5) queue = Queue() crawler = crawl.CrawlingThread(basis, queue) crawler.start() crawler.join() urls = queue.get() for url in urls: t_start += 5 * random.weibullvariate(1, 1.5) result.append(PartialHistoryEntry(url, timestamp)) return result def generate_first_url(user): """ Generate the first url of a partial history, based on the user information. """ def nonempty(seq): out = [] for elt in seq: if elt: out.append(elt) return out all_keywords = profiles.Keyword.objects.filter( interest__profile__in=[user]) all_websites = profiles.Website.objects.filter( interest__profile__in=[user]) all_places = profiles.Place.objects.filter( interest__profile__in=[user]) all_events = profiles.Event.objects.filter( interest__profile__in=[user]) interest = random.choice(nonempty([ all_keywords, all_websites, all_places, all_events, ])) search_term = random.choice(interest) url = search_term.generate_url(user) return url def generate_history(user, start_time): """ Generate a new history for the user `user`, starting from timestamp `ts_start`. A few heuristics are used in order to give the impression that the history is actually played by a user. """ # let's define a new history object. history = History(start_ts=start_time, user=user) length = HISTORY_MIN + floor(10 * random.weibullvariate(1, 1.5)) history.full_clean() history.save() history_line = 0 current_timestamp = start_time.timestamp() while history_line < length: current_timestamp += 5 * random.weibullvariate(1, 2.8) history_list = generate_partial_history(user, current_timestamp) current_timestamp = \ history_list[-1].timestamp + 5 * random.weibullvariate(1, 5) for (url, timestamp) in history_list: new_line = HistoryEntry( search=url, timestamp=datetime.fromtimestamp(timestamp), history=history ) new_line.full_clean() new_line.save() return history