1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-12-27 10:25:49 +01:00
youtube-dl/youtube_dl/extractor/teamtreehouse.py

141 lines
5.4 KiB
Python
Raw Normal View History

# 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,
get_element_by_class,
get_element_by_id,
parse_duration,
remove_end,
urlencode_postdata,
urljoin,
)
class TeamTreeHouseIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?teamtreehouse\.com/library/(?P<id>[^/]+)'
_TESTS = [{
# Course
'url': 'https://teamtreehouse.com/library/introduction-to-user-authentication-in-php',
'info_dict': {
'id': 'introduction-to-user-authentication-in-php',
'title': 'Introduction to User Authentication in PHP',
'description': 'md5:405d7b4287a159b27ddf30ca72b5b053',
},
'playlist_mincount': 24,
}, {
# WorkShop
'url': 'https://teamtreehouse.com/library/deploying-a-react-app',
'info_dict': {
'id': 'deploying-a-react-app',
'title': 'Deploying a React App',
'description': 'md5:10a82e3ddff18c14ac13581c9b8e5921',
},
'playlist_mincount': 4,
}, {
# Video
'url': 'https://teamtreehouse.com/library/application-overview-2',
'info_dict': {
'id': 'application-overview-2',
'ext': 'mp4',
'title': 'Application Overview',
'description': 'md5:4b0a234385c27140a4378de5f1e15127',
},
'expected_warnings': ['This is just a preview'],
}]
_NETRC_MACHINE = 'teamtreehouse'
def _real_initialize(self):
email, password = self._get_login_info()
if email is None:
return
signin_page = self._download_webpage(
'https://teamtreehouse.com/signin',
None, 'Downloading signin page')
data = self._form_hidden_inputs('new_user_session', signin_page)
data.update({
'user_session[email]': email,
'user_session[password]': password,
})
error_message = get_element_by_class('error-message', self._download_webpage(
'https://teamtreehouse.com/person_session',
None, 'Logging in', data=urlencode_postdata(data)))
if error_message:
raise ExtractorError(clean_html(error_message), expected=True)
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
title = self._html_search_meta(['og:title', 'twitter:title'], webpage)
description = self._html_search_meta(
['description', 'og:description', 'twitter:description'], webpage)
entries = self._parse_html5_media_entries(url, webpage, display_id)
if entries:
info = entries[0]
for subtitles in info.get('subtitles', {}).values():
for subtitle in subtitles:
subtitle['ext'] = determine_ext(subtitle['url'], 'srt')
is_preview = 'data-preview="true"' in webpage
if is_preview:
self.report_warning(
'This is just a preview. You need to be signed in with a Basic account to download the entire video.', display_id)
duration = 30
else:
duration = float_or_none(self._search_regex(
r'data-duration="(\d+)"', webpage, 'duration'), 1000)
if not duration:
duration = parse_duration(get_element_by_id(
'video-duration', webpage))
info.update({
'id': display_id,
'title': title,
'description': description,
'duration': duration,
})
return info
else:
def extract_urls(html, extract_info=None):
for path in re.findall(r'<a[^>]+href="([^"]+)"', html):
page_url = urljoin(url, path)
entry = {
'_type': 'url_transparent',
'id': self._match_id(page_url),
'url': page_url,
'id_key': self.ie_key(),
}
if extract_info:
entry.update(extract_info)
entries.append(entry)
workshop_videos = self._search_regex(
r'(?s)<ul[^>]+id="workshop-videos"[^>]*>(.+?)</ul>',
webpage, 'workshop videos', default=None)
if workshop_videos:
extract_urls(workshop_videos)
else:
stages_path = self._search_regex(
r'(?s)<div[^>]+id="syllabus-stages"[^>]+data-url="([^"]+)"',
webpage, 'stages path')
if stages_path:
stages_page = self._download_webpage(
urljoin(url, stages_path), display_id, 'Downloading stages page')
for chapter_number, (chapter, steps_list) in enumerate(re.findall(r'(?s)<h2[^>]*>\s*(.+?)\s*</h2>.+?<ul[^>]*>(.+?)</ul>', stages_page), 1):
extract_urls(steps_list, {
'chapter': chapter,
'chapter_number': chapter_number,
})
title = remove_end(title, ' Course')
return self.playlist_result(
entries, display_id, title, description)