mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 00:47:39 +00:00
836e06d246
This also adds the following test runners: - `3.12-dev` on `ubuntu-latest` - `3.12-dev` on `windows-latest` - `pypy-3.10` on `ubuntu-latest` Authored by: Grub4K
600 lines
25 KiB
Python
600 lines
25 KiB
Python
import calendar
|
|
import json
|
|
import functools
|
|
from datetime import datetime, timezone
|
|
from random import random
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_urllib_parse_urlparse,
|
|
compat_urlparse
|
|
)
|
|
|
|
from ..utils import (
|
|
bug_reports_message,
|
|
ExtractorError,
|
|
get_first,
|
|
int_or_none,
|
|
OnDemandPagedList,
|
|
parse_qs,
|
|
srt_subtitles_timecode,
|
|
traverse_obj,
|
|
)
|
|
|
|
|
|
class PanoptoBaseIE(InfoExtractor):
|
|
BASE_URL_RE = r'(?P<base_url>https?://[\w.-]+\.panopto.(?:com|eu)/Panopto)'
|
|
|
|
# see panopto core.js
|
|
_SUB_LANG_MAPPING = {
|
|
0: 'en-US',
|
|
1: 'en-GB',
|
|
2: 'es-MX',
|
|
3: 'es-ES',
|
|
4: 'de-DE',
|
|
5: 'fr-FR',
|
|
6: 'nl-NL',
|
|
7: 'th-TH',
|
|
8: 'zh-CN',
|
|
9: 'zh-TW',
|
|
10: 'ko-KR',
|
|
11: 'ja-JP',
|
|
12: 'ru-RU',
|
|
13: 'pt-PT',
|
|
14: 'pl-PL',
|
|
15: 'en-AU',
|
|
16: 'da-DK',
|
|
17: 'fi-FI',
|
|
18: 'hu-HU',
|
|
19: 'nb-NO',
|
|
20: 'sv-SE',
|
|
21: 'it-IT'
|
|
}
|
|
|
|
def _call_api(self, base_url, path, video_id, data=None, fatal=True, **kwargs):
|
|
response = self._download_json(
|
|
base_url + path, video_id, data=json.dumps(data).encode('utf8') if data else None,
|
|
fatal=fatal, headers={'accept': 'application/json', 'content-type': 'application/json'}, **kwargs)
|
|
if not response:
|
|
return
|
|
error_code = traverse_obj(response, 'ErrorCode')
|
|
if error_code == 2:
|
|
self.raise_login_required(method='cookies')
|
|
elif error_code is not None:
|
|
msg = f'Panopto said: {response.get("ErrorMessage")}'
|
|
if fatal:
|
|
raise ExtractorError(msg, video_id=video_id, expected=True)
|
|
else:
|
|
self.report_warning(msg, video_id=video_id)
|
|
return response
|
|
|
|
@staticmethod
|
|
def _parse_fragment(url):
|
|
return {k: json.loads(v[0]) for k, v in compat_urlparse.parse_qs(compat_urllib_parse_urlparse(url).fragment).items()}
|
|
|
|
|
|
class PanoptoIE(PanoptoBaseIE):
|
|
_VALID_URL = PanoptoBaseIE.BASE_URL_RE + r'/Pages/(Viewer|Embed)\.aspx.*(?:\?|&)id=(?P<id>[a-f0-9-]+)'
|
|
_EMBED_REGEX = [rf'<iframe[^>]+src=["\'](?P<url>{PanoptoBaseIE.BASE_URL_RE}/Pages/(Viewer|Embed|Sessions/List)\.aspx[^"\']+)']
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://demo.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=26b3ae9e-4a48-4dcc-96ba-0befba08a0fb',
|
|
'info_dict': {
|
|
'id': '26b3ae9e-4a48-4dcc-96ba-0befba08a0fb',
|
|
'title': 'Panopto for Business - Use Cases',
|
|
'timestamp': 1459184200,
|
|
'thumbnail': r're:https://demo\.hosted\.panopto\.com/.+',
|
|
'upload_date': '20160328',
|
|
'ext': 'mp4',
|
|
'cast': [],
|
|
'chapters': [],
|
|
'duration': 88.17099999999999,
|
|
'average_rating': int,
|
|
'uploader_id': '2db6b718-47a0-4b0b-9e17-ab0b00f42b1e',
|
|
'channel_id': 'e4c6a2fc-1214-4ca0-8fb7-aef2e29ff63a',
|
|
'channel': 'Showcase Videos'
|
|
},
|
|
},
|
|
{
|
|
'url': 'https://demo.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=ed01b077-c9e5-4c7b-b8ff-15fa306d7a59',
|
|
'info_dict': {
|
|
'id': 'ed01b077-c9e5-4c7b-b8ff-15fa306d7a59',
|
|
'title': 'Overcoming Top 4 Challenges of Enterprise Video',
|
|
'uploader': 'Panopto Support',
|
|
'timestamp': 1449409251,
|
|
'thumbnail': r're:https://demo\.hosted\.panopto\.com/.+',
|
|
'upload_date': '20151206',
|
|
'ext': 'mp4',
|
|
'chapters': 'count:12',
|
|
'cast': ['Panopto Support'],
|
|
'uploader_id': 'a96d1a31-b4de-489b-9eee-b4a5b414372c',
|
|
'average_rating': int,
|
|
'description': 'md5:4391837802b3fc856dadf630c4b375d1',
|
|
'duration': 1088.2659999999998,
|
|
'channel_id': '9f3c1921-43bb-4bda-8b3a-b8d2f05a8546',
|
|
'channel': 'Webcasts',
|
|
},
|
|
},
|
|
{
|
|
# Extra params in URL
|
|
'url': 'https://howtovideos.hosted.panopto.com/Panopto/Pages/Viewer.aspx?randomparam=thisisnotreal&id=5fa74e93-3d87-4694-b60e-aaa4012214ed&advance=true',
|
|
'info_dict': {
|
|
'id': '5fa74e93-3d87-4694-b60e-aaa4012214ed',
|
|
'ext': 'mp4',
|
|
'duration': 129.513,
|
|
'cast': ['Kathryn Kelly'],
|
|
'uploader_id': '316a0a58-7fa2-4cd9-be1c-64270d284a56',
|
|
'timestamp': 1569845768,
|
|
'tags': ['Viewer', 'Enterprise'],
|
|
'chapters': [],
|
|
'upload_date': '20190930',
|
|
'thumbnail': r're:https://howtovideos\.hosted\.panopto\.com/.+',
|
|
'description': 'md5:2d844aaa1b1a14ad0e2601a0993b431f',
|
|
'title': 'Getting Started: View a Video',
|
|
'average_rating': int,
|
|
'uploader': 'Kathryn Kelly',
|
|
'channel_id': 'fb93bc3c-6750-4b80-a05b-a921013735d3',
|
|
'channel': 'Getting Started',
|
|
}
|
|
},
|
|
{
|
|
# Does not allow normal Viewer.aspx. AUDIO livestream has no url, so should be skipped and only give one stream.
|
|
'url': 'https://unisa.au.panopto.com/Panopto/Pages/Embed.aspx?id=9d9a0fa3-e99a-4ebd-a281-aac2017f4da4',
|
|
'info_dict': {
|
|
'id': '9d9a0fa3-e99a-4ebd-a281-aac2017f4da4',
|
|
'ext': 'mp4',
|
|
'cast': ['LTS CLI Script'],
|
|
'chapters': [],
|
|
'duration': 2178.45,
|
|
'description': 'md5:ee5cf653919f55b72bce2dbcf829c9fa',
|
|
'channel_id': 'b23e673f-c287-4cb1-8344-aae9005a69f8',
|
|
'average_rating': int,
|
|
'uploader_id': '38377323-6a23-41e2-9ff6-a8e8004bf6f7',
|
|
'uploader': 'LTS CLI Script',
|
|
'timestamp': 1572458134,
|
|
'title': 'WW2 Vets Interview 3 Ronald Stanley George',
|
|
'thumbnail': r're:https://unisa\.au\.panopto\.com/.+',
|
|
'channel': 'World War II Veteran Interviews',
|
|
'upload_date': '20191030',
|
|
},
|
|
},
|
|
{
|
|
# Slides/storyboard
|
|
'url': 'https://demo.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=a7f12f1d-3872-4310-84b0-f8d8ab15326b',
|
|
'info_dict': {
|
|
'id': 'a7f12f1d-3872-4310-84b0-f8d8ab15326b',
|
|
'ext': 'mhtml',
|
|
'timestamp': 1448798857,
|
|
'duration': 4712.681,
|
|
'title': 'Cache Memory - CompSci 15-213, Lecture 12',
|
|
'channel_id': 'e4c6a2fc-1214-4ca0-8fb7-aef2e29ff63a',
|
|
'uploader_id': 'a96d1a31-b4de-489b-9eee-b4a5b414372c',
|
|
'upload_date': '20151129',
|
|
'average_rating': 0,
|
|
'uploader': 'Panopto Support',
|
|
'channel': 'Showcase Videos',
|
|
'description': 'md5:55e51d54233ddb0e6c2ed388ca73822c',
|
|
'cast': ['ISR Videographer', 'Panopto Support'],
|
|
'chapters': 'count:28',
|
|
'thumbnail': r're:https://demo\.hosted\.panopto\.com/.+',
|
|
},
|
|
'params': {'format': 'mhtml', 'skip_download': True}
|
|
},
|
|
{
|
|
'url': 'https://na-training-1.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=8285224a-9a2b-4957-84f2-acb0000c4ea9',
|
|
'info_dict': {
|
|
'id': '8285224a-9a2b-4957-84f2-acb0000c4ea9',
|
|
'ext': 'mp4',
|
|
'chapters': [],
|
|
'title': 'Company Policy',
|
|
'average_rating': 0,
|
|
'timestamp': 1615058901,
|
|
'channel': 'Human Resources',
|
|
'tags': ['HumanResources'],
|
|
'duration': 1604.243,
|
|
'thumbnail': r're:https://na-training-1\.hosted\.panopto\.com/.+',
|
|
'uploader_id': '8e8ba0a3-424f-40df-a4f1-ab3a01375103',
|
|
'uploader': 'Cait M.',
|
|
'upload_date': '20210306',
|
|
'cast': ['Cait M.'],
|
|
'subtitles': {'en-US': [{'ext': 'srt', 'data': 'md5:a3f4d25963fdeace838f327097c13265'}],
|
|
'es-ES': [{'ext': 'srt', 'data': 'md5:57e9dad365fd0fbaf0468eac4949f189'}]},
|
|
},
|
|
'params': {'writesubtitles': True, 'skip_download': True}
|
|
}, {
|
|
# On Panopto there are two subs: "Default" and en-US. en-US is blank and should be skipped.
|
|
'url': 'https://na-training-1.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=940cbd41-f616-4a45-b13e-aaf1000c915b',
|
|
'info_dict': {
|
|
'id': '940cbd41-f616-4a45-b13e-aaf1000c915b',
|
|
'ext': 'mp4',
|
|
'subtitles': 'count:1',
|
|
'title': 'HR Benefits Review Meeting*',
|
|
'cast': ['Panopto Support'],
|
|
'chapters': [],
|
|
'timestamp': 1575024251,
|
|
'thumbnail': r're:https://na-training-1\.hosted\.panopto\.com/.+',
|
|
'channel': 'Zoom',
|
|
'description': 'md5:04f90a9c2c68b7828144abfb170f0106',
|
|
'uploader': 'Panopto Support',
|
|
'average_rating': 0,
|
|
'duration': 409.34499999999997,
|
|
'uploader_id': 'b6ac04ad-38b8-4724-a004-a851004ea3df',
|
|
'upload_date': '20191129',
|
|
|
|
},
|
|
'params': {'writesubtitles': True, 'skip_download': True}
|
|
},
|
|
{
|
|
'url': 'https://ucc.cloud.panopto.eu/Panopto/Pages/Viewer.aspx?id=0e8484a4-4ceb-4d98-a63f-ac0200b455cb',
|
|
'only_matching': True
|
|
},
|
|
{
|
|
'url': 'https://brown.hosted.panopto.com/Panopto/Pages/Embed.aspx?id=0b3ff73b-36a0-46c5-8455-aadf010a3638',
|
|
'only_matching': True
|
|
},
|
|
]
|
|
|
|
@classmethod
|
|
def suitable(cls, url):
|
|
return False if PanoptoPlaylistIE.suitable(url) else super().suitable(url)
|
|
|
|
def _mark_watched(self, base_url, video_id, delivery_info):
|
|
duration = traverse_obj(delivery_info, ('Delivery', 'Duration'), expected_type=float)
|
|
invocation_id = delivery_info.get('InvocationId')
|
|
stream_id = traverse_obj(delivery_info, ('Delivery', 'Streams', ..., 'PublicID'), get_all=False, expected_type=str)
|
|
if invocation_id and stream_id and duration:
|
|
timestamp_str = f'/Date({calendar.timegm(datetime.now(timezone.utc).timetuple())}000)/'
|
|
data = {
|
|
'streamRequests': [
|
|
{
|
|
'ClientTimeStamp': timestamp_str,
|
|
'ID': 0,
|
|
'InvocationID': invocation_id,
|
|
'PlaybackSpeed': 1,
|
|
'SecondsListened': duration - 1,
|
|
'SecondsRejected': 0,
|
|
'StartPosition': 0,
|
|
'StartReason': 2,
|
|
'StopReason': None,
|
|
'StreamID': stream_id,
|
|
'TimeStamp': timestamp_str,
|
|
'UpdatesRejected': 0
|
|
},
|
|
]}
|
|
|
|
self._download_webpage(
|
|
base_url + '/Services/Analytics.svc/AddStreamRequests', video_id,
|
|
fatal=False, data=json.dumps(data).encode('utf8'), headers={'content-type': 'application/json'},
|
|
note='Marking watched', errnote='Unable to mark watched')
|
|
|
|
@staticmethod
|
|
def _extract_chapters(timestamps):
|
|
chapters = []
|
|
for timestamp in timestamps or []:
|
|
caption = timestamp.get('Caption')
|
|
start, duration = int_or_none(timestamp.get('Time')), int_or_none(timestamp.get('Duration'))
|
|
if not caption or start is None or duration is None:
|
|
continue
|
|
chapters.append({
|
|
'start_time': start,
|
|
'end_time': start + duration,
|
|
'title': caption
|
|
})
|
|
return chapters
|
|
|
|
@staticmethod
|
|
def _extract_mhtml_formats(base_url, timestamps):
|
|
image_frags = {}
|
|
for timestamp in timestamps or []:
|
|
duration = timestamp.get('Duration')
|
|
obj_id, obj_sn = timestamp.get('ObjectIdentifier'), timestamp.get('ObjectSequenceNumber'),
|
|
if timestamp.get('EventTargetType') == 'PowerPoint' and obj_id is not None and obj_sn is not None:
|
|
image_frags.setdefault('slides', []).append({
|
|
'url': base_url + f'/Pages/Viewer/Image.aspx?id={obj_id}&number={obj_sn}',
|
|
'duration': duration
|
|
})
|
|
|
|
obj_pid, session_id, abs_time = timestamp.get('ObjectPublicIdentifier'), timestamp.get('SessionID'), timestamp.get('AbsoluteTime')
|
|
if None not in (obj_pid, session_id, abs_time):
|
|
image_frags.setdefault('chapter', []).append({
|
|
'url': base_url + f'/Pages/Viewer/Thumb.aspx?eventTargetPID={obj_pid}&sessionPID={session_id}&number={obj_sn}&isPrimary=false&absoluteTime={abs_time}',
|
|
'duration': duration,
|
|
})
|
|
for name, fragments in image_frags.items():
|
|
yield {
|
|
'format_id': name,
|
|
'ext': 'mhtml',
|
|
'protocol': 'mhtml',
|
|
'acodec': 'none',
|
|
'vcodec': 'none',
|
|
'url': 'about:invalid',
|
|
'fragments': fragments
|
|
}
|
|
|
|
@staticmethod
|
|
def _json2srt(data, delivery):
|
|
def _gen_lines():
|
|
for i, line in enumerate(data):
|
|
start_time = line['Time']
|
|
duration = line.get('Duration')
|
|
if duration:
|
|
end_time = start_time + duration
|
|
else:
|
|
end_time = traverse_obj(data, (i + 1, 'Time')) or delivery['Duration']
|
|
yield f'{i + 1}\n{srt_subtitles_timecode(start_time)} --> {srt_subtitles_timecode(end_time)}\n{line["Caption"]}'
|
|
return '\n\n'.join(_gen_lines())
|
|
|
|
def _get_subtitles(self, base_url, video_id, delivery):
|
|
subtitles = {}
|
|
for lang in delivery.get('AvailableLanguages') or []:
|
|
response = self._call_api(
|
|
base_url, '/Pages/Viewer/DeliveryInfo.aspx', video_id, fatal=False,
|
|
note='Downloading captions JSON metadata', query={
|
|
'deliveryId': video_id,
|
|
'getCaptions': True,
|
|
'language': str(lang),
|
|
'responseType': 'json'
|
|
}
|
|
)
|
|
if not isinstance(response, list):
|
|
continue
|
|
subtitles.setdefault(self._SUB_LANG_MAPPING.get(lang) or 'default', []).append({
|
|
'ext': 'srt',
|
|
'data': self._json2srt(response, delivery),
|
|
})
|
|
return subtitles
|
|
|
|
def _extract_streams_formats_and_subtitles(self, video_id, streams, **fmt_kwargs):
|
|
formats = []
|
|
subtitles = {}
|
|
for stream in streams or []:
|
|
stream_formats = []
|
|
http_stream_url = stream.get('StreamHttpUrl')
|
|
stream_url = stream.get('StreamUrl')
|
|
|
|
if http_stream_url:
|
|
stream_formats.append({'url': http_stream_url})
|
|
|
|
if stream_url:
|
|
media_type = stream.get('ViewerMediaFileTypeName')
|
|
if media_type in ('hls', ):
|
|
m3u8_formats, stream_subtitles = self._extract_m3u8_formats_and_subtitles(stream_url, video_id)
|
|
stream_formats.extend(m3u8_formats)
|
|
subtitles = self._merge_subtitles(subtitles, stream_subtitles)
|
|
else:
|
|
stream_formats.append({
|
|
'url': stream_url
|
|
})
|
|
for fmt in stream_formats:
|
|
fmt.update({
|
|
'format_note': stream.get('Tag'),
|
|
**fmt_kwargs
|
|
})
|
|
formats.extend(stream_formats)
|
|
|
|
return formats, subtitles
|
|
|
|
def _real_extract(self, url):
|
|
base_url, video_id = self._match_valid_url(url).group('base_url', 'id')
|
|
delivery_info = self._call_api(
|
|
base_url, '/Pages/Viewer/DeliveryInfo.aspx', video_id,
|
|
query={
|
|
'deliveryId': video_id,
|
|
'invocationId': '',
|
|
'isLiveNotes': 'false',
|
|
'refreshAuthCookie': 'true',
|
|
'isActiveBroadcast': 'false',
|
|
'isEditing': 'false',
|
|
'isKollectiveAgentInstalled': 'false',
|
|
'isEmbed': 'false',
|
|
'responseType': 'json',
|
|
}
|
|
)
|
|
|
|
delivery = delivery_info['Delivery']
|
|
session_start_time = int_or_none(delivery.get('SessionStartTime'))
|
|
timestamps = delivery.get('Timestamps')
|
|
|
|
# Podcast stream is usually the combined streams. We will prefer that by default.
|
|
podcast_formats, podcast_subtitles = self._extract_streams_formats_and_subtitles(
|
|
video_id, delivery.get('PodcastStreams'), format_note='PODCAST')
|
|
|
|
streams_formats, streams_subtitles = self._extract_streams_formats_and_subtitles(
|
|
video_id, delivery.get('Streams'), preference=-10)
|
|
|
|
formats = podcast_formats + streams_formats
|
|
formats.extend(self._extract_mhtml_formats(base_url, timestamps))
|
|
subtitles = self._merge_subtitles(
|
|
podcast_subtitles, streams_subtitles, self.extract_subtitles(base_url, video_id, delivery))
|
|
|
|
self.mark_watched(base_url, video_id, delivery_info)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': delivery.get('SessionName'),
|
|
'cast': traverse_obj(delivery, ('Contributors', ..., 'DisplayName'), expected_type=lambda x: x or None),
|
|
'timestamp': session_start_time - 11640000000 if session_start_time else None,
|
|
'duration': delivery.get('Duration'),
|
|
'thumbnail': base_url + f'/Services/FrameGrabber.svc/FrameRedirect?objectId={video_id}&mode=Delivery&random={random()}',
|
|
'average_rating': delivery.get('AverageRating'),
|
|
'chapters': self._extract_chapters(timestamps),
|
|
'uploader': delivery.get('OwnerDisplayName') or None,
|
|
'uploader_id': delivery.get('OwnerId'),
|
|
'description': delivery.get('SessionAbstract'),
|
|
'tags': traverse_obj(delivery, ('Tags', ..., 'Content')),
|
|
'channel_id': delivery.get('SessionGroupPublicID'),
|
|
'channel': traverse_obj(delivery, 'SessionGroupLongName', 'SessionGroupShortName', get_all=False),
|
|
'formats': formats,
|
|
'subtitles': subtitles
|
|
}
|
|
|
|
|
|
class PanoptoPlaylistIE(PanoptoBaseIE):
|
|
_VALID_URL = PanoptoBaseIE.BASE_URL_RE + r'/Pages/(Viewer|Embed)\.aspx.*(?:\?|&)pid=(?P<id>[a-f0-9-]+)'
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://howtovideos.hosted.panopto.com/Panopto/Pages/Viewer.aspx?pid=f3b39fcf-882f-4849-93d6-a9f401236d36&id=5fa74e93-3d87-4694-b60e-aaa4012214ed&advance=true',
|
|
'info_dict': {
|
|
'title': 'Featured Video Tutorials',
|
|
'id': 'f3b39fcf-882f-4849-93d6-a9f401236d36',
|
|
'description': '',
|
|
},
|
|
'playlist_mincount': 36
|
|
},
|
|
{
|
|
'url': 'https://utsa.hosted.panopto.com/Panopto/Pages/Viewer.aspx?pid=e2900555-3ad4-4bdb-854d-ad2401686190',
|
|
'info_dict': {
|
|
'title': 'Library Website Introduction Playlist',
|
|
'id': 'e2900555-3ad4-4bdb-854d-ad2401686190',
|
|
'description': 'md5:f958bca50a1cbda15fdc1e20d32b3ecb',
|
|
},
|
|
'playlist_mincount': 4
|
|
},
|
|
|
|
]
|
|
|
|
def _entries(self, base_url, playlist_id, session_list_id):
|
|
session_list_info = self._call_api(
|
|
base_url, f'/Api/SessionLists/{session_list_id}?collections[0].maxCount=500&collections[0].name=items', playlist_id)
|
|
|
|
items = session_list_info['Items']
|
|
for item in items:
|
|
if item.get('TypeName') != 'Session':
|
|
self.report_warning('Got an item in the playlist that is not a Session' + bug_reports_message(), only_once=True)
|
|
continue
|
|
yield {
|
|
'_type': 'url',
|
|
'id': item.get('Id'),
|
|
'url': item.get('ViewerUri'),
|
|
'title': item.get('Name'),
|
|
'description': item.get('Description'),
|
|
'duration': item.get('Duration'),
|
|
'channel': traverse_obj(item, ('Parent', 'Name')),
|
|
'channel_id': traverse_obj(item, ('Parent', 'Id'))
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
base_url, playlist_id = self._match_valid_url(url).group('base_url', 'id')
|
|
|
|
video_id = get_first(parse_qs(url), 'id')
|
|
if video_id:
|
|
if self.get_param('noplaylist'):
|
|
self.to_screen('Downloading just video %s because of --no-playlist' % video_id)
|
|
return self.url_result(base_url + f'/Pages/Viewer.aspx?id={video_id}', ie_key=PanoptoIE.ie_key(), video_id=video_id)
|
|
else:
|
|
self.to_screen(f'Downloading playlist {playlist_id}; add --no-playlist to just download video {video_id}')
|
|
|
|
playlist_info = self._call_api(base_url, f'/Api/Playlists/{playlist_id}', playlist_id)
|
|
return self.playlist_result(
|
|
self._entries(base_url, playlist_id, playlist_info['SessionListId']),
|
|
playlist_id=playlist_id, playlist_title=playlist_info.get('Name'),
|
|
playlist_description=playlist_info.get('Description'))
|
|
|
|
|
|
class PanoptoListIE(PanoptoBaseIE):
|
|
_VALID_URL = PanoptoBaseIE.BASE_URL_RE + r'/Pages/Sessions/List\.aspx'
|
|
_PAGE_SIZE = 250
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://demo.hosted.panopto.com/Panopto/Pages/Sessions/List.aspx#folderID=%22e4c6a2fc-1214-4ca0-8fb7-aef2e29ff63a%22',
|
|
'info_dict': {
|
|
'id': 'e4c6a2fc-1214-4ca0-8fb7-aef2e29ff63a',
|
|
'title': 'Showcase Videos'
|
|
},
|
|
'playlist_mincount': 140
|
|
|
|
},
|
|
{
|
|
'url': 'https://demo.hosted.panopto.com/Panopto/Pages/Sessions/List.aspx#view=2&maxResults=250',
|
|
'info_dict': {
|
|
'id': 'panopto_list',
|
|
'title': 'panopto_list'
|
|
},
|
|
'playlist_mincount': 300
|
|
},
|
|
{
|
|
# Folder that contains 8 folders and a playlist
|
|
'url': 'https://howtovideos.hosted.panopto.com/Panopto/Pages/Sessions/List.aspx?noredirect=true#folderID=%224b9de7ae-0080-4158-8496-a9ba01692c2e%22',
|
|
'info_dict': {
|
|
'id': '4b9de7ae-0080-4158-8496-a9ba01692c2e',
|
|
'title': 'Video Tutorials'
|
|
},
|
|
'playlist_mincount': 9
|
|
}
|
|
|
|
]
|
|
|
|
def _fetch_page(self, base_url, query_params, display_id, page):
|
|
|
|
params = {
|
|
'sortColumn': 1,
|
|
'getFolderData': True,
|
|
'includePlaylists': True,
|
|
**query_params,
|
|
'page': page,
|
|
'maxResults': self._PAGE_SIZE,
|
|
}
|
|
|
|
response = self._call_api(
|
|
base_url, '/Services/Data.svc/GetSessions', f'{display_id} page {page+1}',
|
|
data={'queryParameters': params}, fatal=False)
|
|
|
|
for result in get_first(response, 'Results', default=[]):
|
|
# This could be a video, playlist (or maybe something else)
|
|
item_id = result.get('DeliveryID')
|
|
yield {
|
|
'_type': 'url',
|
|
'id': item_id,
|
|
'title': result.get('SessionName'),
|
|
'url': traverse_obj(result, 'ViewerUrl', 'EmbedUrl', get_all=False) or (base_url + f'/Pages/Viewer.aspx?id={item_id}'),
|
|
'duration': result.get('Duration'),
|
|
'channel': result.get('FolderName'),
|
|
'channel_id': result.get('FolderID'),
|
|
}
|
|
|
|
for folder in get_first(response, 'Subfolders', default=[]):
|
|
folder_id = folder.get('ID')
|
|
yield self.url_result(
|
|
base_url + f'/Pages/Sessions/List.aspx#folderID="{folder_id}"',
|
|
ie_key=PanoptoListIE.ie_key(), video_id=folder_id, title=folder.get('Name'))
|
|
|
|
def _extract_folder_metadata(self, base_url, folder_id):
|
|
response = self._call_api(
|
|
base_url, '/Services/Data.svc/GetFolderInfo', folder_id,
|
|
data={'folderID': folder_id}, fatal=False)
|
|
return {
|
|
'title': get_first(response, 'Name')
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = self._match_valid_url(url)
|
|
base_url = mobj.group('base_url')
|
|
|
|
query_params = self._parse_fragment(url)
|
|
folder_id, display_id = query_params.get('folderID'), 'panopto_list'
|
|
|
|
if query_params.get('isSubscriptionsPage'):
|
|
display_id = 'subscriptions'
|
|
if not query_params.get('subscribableTypes'):
|
|
query_params['subscribableTypes'] = [0, 1, 2]
|
|
elif query_params.get('isSharedWithMe'):
|
|
display_id = 'sharedwithme'
|
|
elif folder_id:
|
|
display_id = folder_id
|
|
|
|
query = query_params.get('query')
|
|
if query:
|
|
display_id += f': query "{query}"'
|
|
|
|
info = {
|
|
'_type': 'playlist',
|
|
'id': display_id,
|
|
'title': display_id,
|
|
}
|
|
if folder_id:
|
|
info.update(self._extract_folder_metadata(base_url, folder_id))
|
|
|
|
info['entries'] = OnDemandPagedList(
|
|
functools.partial(self._fetch_page, base_url, query_params, display_id), self._PAGE_SIZE)
|
|
|
|
return info
|