WIP: Start small ILIAS crawler

This commit is contained in:
I-Al-Istannen 2020-04-21 13:32:03 +02:00
parent b2a7af2e3e
commit 70d63e3e90

203
PFERD/ilias/crawler.py Normal file
View File

@ -0,0 +1,203 @@
"""
Contains an ILIAS crawler alongside helper functions.
"""
import datetime
import logging
import re
from pathlib import Path
from typing import Any, Dict, List, Optional
from urllib.parse import (parse_qs, urlencode, urljoin, urlparse, urlsplit,
urlunsplit)
import bs4
import requests
from ..cookie_jar import CookieJar
from ..utils import soupify
from .authenticators import IliasAuthenticator
from .downloader import IliasDownloadInfo
LOGGER = logging.getLogger(__name__)
class IliasCrawler:
# pylint: disable=too-few-public-methods
"""
A crawler for ILIAS.
"""
def __init__(self, authenticator: IliasAuthenticator, base_url: str, course_id: str):
"""
Create a new ILIAS crawler.
"""
self._cookie_jar = CookieJar(Path("/tmp/test/cookies"))
self._cookie_jar.load_cookies()
self._base_url = base_url
self._course_id = course_id
self._session = self._cookie_jar.create_session()
self._authenticator = authenticator
def _abs_url_from_link(self, link_tag: bs4.Tag) -> str:
"""
Create an absolute url from an <a> tag.
"""
return urljoin(self._base_url, link_tag.get("href"))
@staticmethod
def _url_set_query_param(url: str, param: str, value: str) -> str:
scheme, netloc, path, query, fragment = urlsplit(url)
query_parameters = parse_qs(query)
query_parameters[param] = [value]
new_query_string = urlencode(query_parameters, doseq=True)
return urlunsplit((scheme, netloc, path, new_query_string, fragment))
def crawl(self) -> List[IliasDownloadInfo]:
"""
Starts the crawl process, yielding a list of elements to (potentially) download.
"""
root_url = self._url_set_query_param(
self._base_url + "/goto.php", "target", f"crs_{self._course_id}"
)
return self._crawl_folder(Path(""), root_url)
def _switch_on_crawled_type(
self,
path: Path,
link_element: bs4.Tag,
url: str
) -> List[IliasDownloadInfo]:
"""
Decides which sub crawler to use for a given top level element.
"""
parsed_url = urlparse(url)
LOGGER.debug("Parsed url: %s", repr(parsed_url))
if "target=file_" in parsed_url.query:
return self._crawl_file(path, link_element, url)
# Skip forums
if "cmd=showThreads" in parsed_url.query:
LOGGER.debug("Skipping forum %s", repr(url))
return []
if "ref_id=" in parsed_url.query:
LOGGER.debug("Processing folder-like...")
return self._switch_on_folder_like(path, link_element, url)
LOGGER.warning("Got unknown type, %s, %s, %s", repr(path), repr(link_element), repr(url))
# TODO: Other types
raise Exception("Implement me!")
@staticmethod
def _crawl_file(path: Path, link_element: bs4.Tag, url: str) -> List[IliasDownloadInfo]:
"""
Crawls a file.
"""
properties_parent: bs4.Tag = link_element.findParent(
"div", {"class": lambda x: "il_ContainerListItem" in x}
).select_one(".il_ItemProperties")
file_type = properties_parent.select_one("span.il_ItemProperty").getText().strip()
modifcation_date = datetime.datetime.now()
all_properties_text = properties_parent.getText().strip()
print("Property text is", all_properties_text)
# todo demangle date from text above
name = link_element.getText()
full_path = Path(path, name + "." + file_type)
match_result = re.match(r".+target=file_(\d+).+", url)
if match_result is None:
LOGGER.warning("Could not download file %s", repr(url))
return []
return [IliasDownloadInfo(full_path, url, modifcation_date)]
def _switch_on_folder_like(
self,
path: Path,
link_element: bs4.Tag,
url: str
) -> List[IliasDownloadInfo]:
found_parent: Optional[bs4.Tag] = None
for parent in link_element.parents:
if "ilContainerListItemOuter" in parent["class"]:
found_parent = parent
break
if found_parent is None:
LOGGER.warning("Could not find element icon for %s", repr(url))
return []
img_tag: Optional[bs4.Tag] = found_parent.select_one("img.ilListItemIcon")
if img_tag is None:
LOGGER.warning("Could not find image tag for %s", repr(url))
return []
if str(img_tag["src"]).endswith("frm.svg"):
LOGGER.debug("Skipping forum at %s", repr(url))
return []
# Assume it is a folder
folder_name = link_element.getText()
folder_path = Path(path, folder_name)
return self._crawl_folder(folder_path, self._abs_url_from_link(link_element))
def _crawl_folder(self, path: Path, url: str) -> List[IliasDownloadInfo]:
soup = self._get_page(url, {})
result: List[IliasDownloadInfo] = []
links: List[bs4.Tag] = soup.select("a.il_ContainerItemTitle")
for link in links:
abs_url = self._abs_url_from_link(link)
result += self._switch_on_crawled_type(path, link, abs_url)
return result
def _get_page(self, url: str, params: Dict[str, Any]) -> bs4.BeautifulSoup:
"""
Fetches a page from ILIAS, authenticating when needed.
"""
print("Fetching", url)
response = self._session.get(url, params=params)
content_type = response.headers["content-type"]
if not content_type.startswith("text/html"):
# TODO: Correct exception type
raise Exception(f"Invalid content type {content_type}")
soup = soupify(response)
if self._is_logged_in(soup):
return soup
LOGGER.info("Not authenticated, changing that...")
self._authenticator.authenticate(self._session)
self._cookie_jar.save_cookies("Authed")
return self._get_page(url, params)
@staticmethod
def _is_logged_in(soup: bs4.BeautifulSoup) -> bool:
userlog = soup.find("li", {"id": "userlog"})
return userlog is not None
def run_as_test(ilias_url: str, course_id: int) -> List[IliasDownloadInfo]:
from ..organizer import Organizer
from .authenticators import KitShibbolethAuthenticator
organizer = Organizer(Path("/tmp/test/inner"))
crawler = IliasCrawler(KitShibbolethAuthenticator(), ilias_url, str(course_id))
return crawler.crawl()