1
mirror of https://github.com/yt-dlp/yt-dlp synced 2024-12-14 20:59:16 +01:00
yt-dlp/youtube_dl/extractor/adultswim.py
Naglis Jonaitis cc08b11d16 [adultswim] Improve video_info extraction (Fixes #5152)
Look for video_info inside `slugged_video`, if slug is not found among collections.
Also, simplify a bit.
2015-03-08 21:35:04 +02:00

192 lines
7.9 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
float_or_none,
xpath_text,
)
class AdultSwimIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?adultswim\.com/videos/(?P<is_playlist>playlists/)?(?P<show_path>[^/]+)/(?P<episode_path>[^/?#]+)/?'
_TESTS = [{
'url': 'http://adultswim.com/videos/rick-and-morty/pilot',
'playlist': [
{
'md5': '247572debc75c7652f253c8daa51a14d',
'info_dict': {
'id': 'rQxZvXQ4ROaSOqq-or2Mow-0',
'ext': 'flv',
'title': 'Rick and Morty - Pilot Part 1',
'description': "Rick moves in with his daughter's family and establishes himself as a bad influence on his grandson, Morty. "
},
},
{
'md5': '77b0e037a4b20ec6b98671c4c379f48d',
'info_dict': {
'id': 'rQxZvXQ4ROaSOqq-or2Mow-3',
'ext': 'flv',
'title': 'Rick and Morty - Pilot Part 4',
'description': "Rick moves in with his daughter's family and establishes himself as a bad influence on his grandson, Morty. "
},
},
],
'info_dict': {
'id': 'rQxZvXQ4ROaSOqq-or2Mow',
'title': 'Rick and Morty - Pilot',
'description': "Rick moves in with his daughter's family and establishes himself as a bad influence on his grandson, Morty. "
}
}, {
'url': 'http://www.adultswim.com/videos/playlists/american-parenting/putting-francine-out-of-business/',
'playlist': [
{
'md5': '2eb5c06d0f9a1539da3718d897f13ec5',
'info_dict': {
'id': '-t8CamQlQ2aYZ49ItZCFog-0',
'ext': 'flv',
'title': 'American Dad - Putting Francine Out of Business',
'description': 'Stan hatches a plan to get Francine out of the real estate business.Watch more American Dad on [adult swim].'
},
}
],
'info_dict': {
'id': '-t8CamQlQ2aYZ49ItZCFog',
'title': 'American Dad - Putting Francine Out of Business',
'description': 'Stan hatches a plan to get Francine out of the real estate business.Watch more American Dad on [adult swim].'
},
}, {
'url': 'http://www.adultswim.com/videos/tim-and-eric-awesome-show-great-job/dr-steve-brule-for-your-wine/',
'playlist': [
{
'md5': '3e346a2ab0087d687a05e1e7f3b3e529',
'info_dict': {
'id': 'sY3cMUR_TbuE4YmdjzbIcQ-0',
'ext': 'flv',
'title': 'Tim and Eric Awesome Show Great Job! - Dr. Steve Brule, For Your Wine',
'description': 'Dr. Brule reports live from Wine Country with a special report on wines. \r\nWatch Tim and Eric Awesome Show Great Job! episode #20, "Embarrassed" on Adult Swim.\r\n\r\n',
},
}
],
'info_dict': {
'id': 'sY3cMUR_TbuE4YmdjzbIcQ',
'title': 'Tim and Eric Awesome Show Great Job! - Dr. Steve Brule, For Your Wine',
'description': 'Dr. Brule reports live from Wine Country with a special report on wines. \r\nWatch Tim and Eric Awesome Show Great Job! episode #20, "Embarrassed" on Adult Swim.\r\n\r\n',
},
}]
@staticmethod
def find_video_info(collection, slug):
for video in collection.get('videos'):
if video.get('slug') == slug:
return video
@staticmethod
def find_collection_by_linkURL(collections, linkURL):
for collection in collections:
if collection.get('linkURL') == linkURL:
return collection
@staticmethod
def find_collection_containing_video(collections, slug):
for collection in collections:
for video in collection.get('videos'):
if video.get('slug') == slug:
return collection, video
return None, None
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
show_path = mobj.group('show_path')
episode_path = mobj.group('episode_path')
is_playlist = True if mobj.group('is_playlist') else False
webpage = self._download_webpage(url, episode_path)
# Extract the value of `bootstrappedData` from the Javascript in the page.
bootstrapped_data = self._parse_json(self._search_regex(
r'var bootstrappedData = ({.*});', webpage, 'bootstraped data'), episode_path)
# Downloading videos from a /videos/playlist/ URL needs to be handled differently.
# NOTE: We are only downloading one video (the current one) not the playlist
if is_playlist:
collections = bootstrapped_data['playlists']['collections']
collection = self.find_collection_by_linkURL(collections, show_path)
video_info = self.find_video_info(collection, episode_path)
show_title = video_info['showTitle']
segment_ids = [video_info['videoPlaybackID']]
else:
collections = bootstrapped_data['show']['collections']
collection, video_info = self.find_collection_containing_video(collections, episode_path)
# Video wasn't found in the collections, let's try `slugged_video`.
if video_info is None:
if bootstrapped_data.get('slugged_video', {}).get('slug') == episode_path:
video_info = bootstrapped_data['slugged_video']
else:
raise ExtractorError('Unable to find video info')
show = bootstrapped_data['show']
show_title = show['title']
segment_ids = [clip['videoPlaybackID'] for clip in video_info['clips']]
episode_id = video_info['id']
episode_title = video_info['title']
episode_description = video_info['description']
episode_duration = video_info.get('duration')
entries = []
for part_num, segment_id in enumerate(segment_ids):
segment_url = 'http://www.adultswim.com/videos/api/v0/assets?id=%s&platform=mobile' % segment_id
segment_title = '%s - %s' % (show_title, episode_title)
if len(segment_ids) > 1:
segment_title += ' Part %d' % (part_num + 1)
idoc = self._download_xml(
segment_url, segment_title,
'Downloading segment information', 'Unable to download segment information')
segment_duration = float_or_none(
xpath_text(idoc, './/trt', 'segment duration').strip())
formats = []
file_els = idoc.findall('.//files/file')
for file_el in file_els:
bitrate = file_el.attrib.get('bitrate')
ftype = file_el.attrib.get('type')
formats.append({
'format_id': '%s_%s' % (bitrate, ftype),
'url': file_el.text.strip(),
# The bitrate may not be a number (for example: 'iphone')
'tbr': int(bitrate) if bitrate.isdigit() else None,
'quality': 1 if ftype == 'hd' else -1
})
self._sort_formats(formats)
entries.append({
'id': segment_id,
'title': segment_title,
'formats': formats,
'duration': segment_duration,
'description': episode_description
})
return {
'_type': 'playlist',
'id': episode_id,
'display_id': episode_path,
'entries': entries,
'title': '%s - %s' % (show_title, episode_title),
'description': episode_description,
'duration': episode_duration
}