You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
youtube-dl/youtube_dl/extractor/lecturio.py

244 lines
8.2 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
clean_html,
determine_ext,
ExtractorError,
float_or_none,
int_or_none,
str_or_none,
url_or_none,
urlencode_postdata,
urljoin,
)
class LecturioBaseIE(InfoExtractor):
_API_BASE_URL = 'https://app.lecturio.com/api/en/latest/html5/'
_LOGIN_URL = 'https://app.lecturio.com/en/login'
_NETRC_MACHINE = 'lecturio'
def _real_initialize(self):
self._login()
def _login(self):
username, password = self._get_login_info()
if username is None:
return
# Sets some cookies
_, urlh = self._download_webpage_handle(
self._LOGIN_URL, None, 'Downloading login popup')
def is_logged(url_handle):
return self._LOGIN_URL not in url_handle.geturl()
# Already logged in
if is_logged(urlh):
return
login_form = {
'signin[email]': username,
'signin[password]': password,
'signin[remember]': 'on',
}
response, urlh = self._download_webpage_handle(
self._LOGIN_URL, None, 'Logging in',
data=urlencode_postdata(login_form))
# Logged in successfully
if is_logged(urlh):
return
errors = self._html_search_regex(
r'(?s)<ul[^>]+class=["\']error_list[^>]+>(.+?)</ul>', response,
'errors', default=None)
if errors:
raise ExtractorError('Unable to login: %s' % errors, expected=True)
raise ExtractorError('Unable to log in')
class LecturioIE(LecturioBaseIE):
_VALID_URL = r'''(?x)
https://
(?:
app\.lecturio\.com/([^/]+/(?P<nt>[^/?#&]+)\.lecture|(?:\#/)?lecture/c/\d+/(?P<id>\d+))|
(?:www\.)?lecturio\.de/[^/]+/(?P<nt_de>[^/?#&]+)\.vortrag
)
'''
_TESTS = [{
'url': 'https://app.lecturio.com/medical-courses/important-concepts-and-terms-introduction-to-microbiology.lecture#tab/videos',
'md5': '9a42cf1d8282a6311bf7211bbde26fde',
'info_dict': {
'id': '39634',
'ext': 'mp4',
'title': 'Important Concepts and Terms — Introduction to Microbiology',
},
'skip': 'Requires lecturio account credentials',
}, {
'url': 'https://www.lecturio.de/jura/oeffentliches-recht-staatsexamen.vortrag',
'only_matching': True,
}, {
'url': 'https://app.lecturio.com/#/lecture/c/6434/39634',
'only_matching': True,
}]
_CC_LANGS = {
'Arabic': 'ar',
'Bulgarian': 'bg',
'German': 'de',
'English': 'en',
'Spanish': 'es',
'Persian': 'fa',
'French': 'fr',
'Japanese': 'ja',
'Polish': 'pl',
'Pashto': 'ps',
'Russian': 'ru',
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
nt = mobj.group('nt') or mobj.group('nt_de')
lecture_id = mobj.group('id')
display_id = nt or lecture_id
api_path = 'lectures/' + lecture_id if lecture_id else 'lecture/' + nt + '.json'
video = self._download_json(
self._API_BASE_URL + api_path, display_id)
title = video['title'].strip()
if not lecture_id:
pid = video.get('productId') or video.get('uid')
if pid:
spid = pid.split('_')
if spid and len(spid) == 2:
lecture_id = spid[1]
formats = []
for format_ in video['content']['media']:
if not isinstance(format_, dict):
continue
file_ = format_.get('file')
if not file_:
continue
ext = determine_ext(file_)
if ext == 'smil':
# smil contains only broken RTMP formats anyway
continue
file_url = url_or_none(file_)
if not file_url:
continue
label = str_or_none(format_.get('label'))
filesize = int_or_none(format_.get('fileSize'))
f = {
'url': file_url,
'format_id': label,
'filesize': float_or_none(filesize, invscale=1000)
}
if label:
mobj = re.match(r'(\d+)p\s*\(([^)]+)\)', label)
if mobj:
f.update({
'format_id': mobj.group(2),
'height': int(mobj.group(1)),
})
formats.append(f)
self._sort_formats(formats)
subtitles = {}
automatic_captions = {}
captions = video.get('captions') or []
for cc in captions:
cc_url = cc.get('url')
if not cc_url:
continue
cc_label = cc.get('translatedCode')
lang = cc.get('languageCode') or self._search_regex(
r'/([a-z]{2})_', cc_url, 'lang',
default=cc_label.split()[0] if cc_label else 'en')
original_lang = self._search_regex(
r'/[a-z]{2}_([a-z]{2})_', cc_url, 'original lang',
default=None)
sub_dict = (automatic_captions
if 'auto-translated' in cc_label or original_lang
else subtitles)
sub_dict.setdefault(self._CC_LANGS.get(lang, lang), []).append({
'url': cc_url,
})
return {
'id': lecture_id or nt,
'title': title,
'formats': formats,
'subtitles': subtitles,
'automatic_captions': automatic_captions,
}
class LecturioCourseIE(LecturioBaseIE):
_VALID_URL = r'https://app\.lecturio\.com/(?:[^/]+/(?P<nt>[^/?#&]+)\.course|(?:#/)?course/c/(?P<id>\d+))'
_TESTS = [{
'url': 'https://app.lecturio.com/medical-courses/microbiology-introduction.course#/',
'info_dict': {
'id': 'microbiology-introduction',
'title': 'Microbiology: Introduction',
'description': 'md5:13da8500c25880c6016ae1e6d78c386a',
},
'playlist_count': 45,
'skip': 'Requires lecturio account credentials',
}, {
'url': 'https://app.lecturio.com/#/course/c/6434',
'only_matching': True,
}]
def _real_extract(self, url):
nt, course_id = re.match(self._VALID_URL, url).groups()
display_id = nt or course_id
api_path = 'courses/' + course_id if course_id else 'course/content/' + nt + '.json'
course = self._download_json(
self._API_BASE_URL + api_path, display_id)
entries = []
for lecture in course.get('lectures', []):
lecture_id = str_or_none(lecture.get('id'))
lecture_url = lecture.get('url')
if lecture_url:
lecture_url = urljoin(url, lecture_url)
else:
lecture_url = 'https://app.lecturio.com/#/lecture/c/%s/%s' % (course_id, lecture_id)
entries.append(self.url_result(
lecture_url, ie=LecturioIE.ie_key(), video_id=lecture_id))
return self.playlist_result(
entries, display_id, course.get('title'),
clean_html(course.get('description')))
class LecturioDeCourseIE(LecturioBaseIE):
_VALID_URL = r'https://(?:www\.)?lecturio\.de/[^/]+/(?P<id>[^/?#&]+)\.kurs'
_TEST = {
'url': 'https://www.lecturio.de/jura/grundrechte.kurs',
'only_matching': True,
}
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
entries = []
for mobj in re.finditer(
r'(?s)<td[^>]+\bdata-lecture-id=["\'](?P<id>\d+).+?\bhref=(["\'])(?P<url>(?:(?!\2).)+\.vortrag)\b[^>]+>',
webpage):
lecture_url = urljoin(url, mobj.group('url'))
lecture_id = mobj.group('id')
entries.append(self.url_result(
lecture_url, ie=LecturioIE.ie_key(), video_id=lecture_id))
title = self._search_regex(
r'<h1[^>]*>([^<]+)', webpage, 'title', default=None)
return self.playlist_result(entries, display_id, title)