Add support to VMware Video Library (not Explore)

This commit is contained in:
kclauhk 2024-10-27 03:07:03 +08:00
parent e3309bc775
commit d4768b66ea
2 changed files with 68 additions and 24 deletions

View File

@ -2367,7 +2367,8 @@ from .vk import (
)
from .vmware import (
VMwareExploreIE,
VMwareExploreSearchIE,
VMwareIE,
VMwareSearchIE,
)
from .vocaroo import VocarooIE
from .vodpl import VODPlIE

View File

@ -3,6 +3,44 @@ import itertools
from .common import InfoExtractor, SearchInfoExtractor
class VMwareIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?vmware\.com/video/(?P<id>\d+)'
_TESTS = [{
'url': 'https://www.vmware.com/video/6362484671112',
'info_dict': {
'id': '6362484671112',
'ext': 'mp4',
'title': 'GCI Communications',
'description': '',
'thumbnail': r're:^https?://.*/image\.jpg',
'tags': [],
'timestamp': 1727345356,
'upload_date': '20240926',
'uploader_id': '6415665063001',
'duration': 106.283,
},
}, {
'url': 'https://www.vmware.com/video/6350300466112',
'info_dict': {
'id': '6350300466112',
'ext': 'mp4',
'title': 'VMware Private AI',
'description': r're:^Learn the significance of AI and Generative AI',
'thumbnail': r're:^https?://.*/image\.jpg',
'tags': 'count:8',
'timestamp': 1712293111,
'upload_date': '20240405',
'uploader_id': '6415665063001',
'duration': 3154.624,
},
}]
BRIGHTCOVE_URL_TEMPLATE = 'https://players.brightcove.net/6415665063001/83iWkhhmz_default/index.html?videoId=%s'
def _real_extract(self, url):
video_id = self._match_id(url)
return self.url_result(self.BRIGHTCOVE_URL_TEMPLATE % video_id, url_transparent=True)
class VMwareExploreIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?vmware\.com/explore/video-library/video/(?P<id>\d+)'
_TESTS = [{
@ -12,7 +50,7 @@ class VMwareExploreIE(InfoExtractor):
'ext': 'mp4',
'title': 'VCFB1440LV',
'description': r're:^All About vSphere 8: What\'s New in the Technology',
'thumbnail': 'https://cf-images.us-east-1.prod.boltdns.net/v1/static/6164421911001/cde65c5a-51ff-4a0c-905f-ed71e25c0f2c/80d7489b-7b65-47d9-b30c-8056b132892f/1920x1080/match/image.jpg',
'thumbnail': r're:^https?://.*/image\.jpg',
'tags': 'count:6',
'timestamp': 1724585610,
'upload_date': '20240825',
@ -26,7 +64,7 @@ class VMwareExploreIE(InfoExtractor):
'ext': 'mp4',
'title': 'AODB1676LV',
'description': r're:^Automation, Analytics and Intelligence: Our Quest for Operational Excellence',
'thumbnail': 'https://cf-images.us-east-1.prod.boltdns.net/v1/jit/6164421911001/56cc0c8e-9d51-4c25-9d97-4b7364989c47/main/1920x1080/14m18s858ms/match/image.jpg',
'thumbnail': r're:^https?://.*/image\.jpg',
'tags': 'count:6',
'timestamp': 1724585574,
'upload_date': '20240825',
@ -40,7 +78,7 @@ class VMwareExploreIE(InfoExtractor):
'ext': 'mp4',
'title': 'ANSB1976LV',
'description': r're:^The Conman of the Digital Era — Ransomware',
'thumbnail': 'https://cf-images.us-east-1.prod.boltdns.net/v1/jit/6164421911001/4ec22e41-7812-49d9-9fc8-5dbcf1ef4b3c/main/1920x1080/22m36s555ms/match/image.jpg',
'thumbnail': r're:^https?://.*/image\.jpg',
'tags': 'count:6',
'timestamp': 1724585612,
'upload_date': '20240825',
@ -55,46 +93,51 @@ class VMwareExploreIE(InfoExtractor):
return self.url_result(self.BRIGHTCOVE_URL_TEMPLATE % video_id, url_transparent=True)
class VMwareExploreSearchIE(SearchInfoExtractor):
IE_NAME = 'VMwareExplore:search'
class VMwareSearchIE(SearchInfoExtractor):
IE_NAME = 'VMware:search'
_SEARCH_KEY = 'vmwaresearch'
_TESTS = [{
'url': 'vmwaresearch10:*',
'playlist_count': 10,
'info_dict': {
'id': '*',
'title': '*',
},
'playlist_count': 10,
}, {
'url': 'vmwaresearchall:ransomware',
'playlist_count': 15,
'url': 'vmwaresearchall:uptime',
'info_dict': {
'id': 'ransomware',
'title': 'ransomware',
'id': 'uptime',
'title': 'uptime',
},
'playlist_mincount': 5,
}]
_URL_TEMPLATE = 'https://www.vmware.com/explore/video-library/video/%s'
_LIBRARY_MAP = {
'explore': ('VMware Explore Video Library', 'https://www.vmware.com/explore/video-library/video/%s'),
'vmware': ('VMware Video Library', 'https://www.vmware.com/video/%s'),
}
def _search_results(self, query):
def search_query(query, offset, limit, total_count):
def search_query(query, offset, limit, account):
# search api:
# https://www.vmware.com/api/nocache/tools/brightcove/search?q=%2B{query}%20%2Byear:2023:2024%20%20-vod_on_demand_publish:%22False%22%2Bcomplete:%22true%22%2Bstate:%22ACTIVE%22&limit=12&offset=0&sort=-updated_at&account=explore
return self._download_json(
'https://www.vmware.com/api/nocache/tools/brightcove/search', query,
note=f'Downloading result {offset + 1}-{min(offset + limit, total_count or 99999999)}', query={
note=f'Searching videos in {self._LIBRARY_MAP[account][0]}', query={
'q': f'+{query} -vod_on_demand_publish:"False"+complete:"true"+state:"ACTIVE"',
'limit': limit,
'offset': offset,
'sort': 'updated_at', # chronological ascending order. For descending order: '-updated_at'
'account': 'explore',
'account': account,
})
limit, total_count = 100, None # limit: maximum 100
for i in itertools.count():
search_results = search_query(query, i * limit, limit, total_count)
total_count = search_results.get('count', 0)
for video in search_results.get('videos', []):
if video_id := video.get('id'):
yield self.url_result(self._URL_TEMPLATE % video_id)
if (i + 1) * limit >= total_count:
break
for account in ['explore', 'vmware']:
limit, total_count = 100, None # limit: maximum 100
for i in itertools.count():
search_results = search_query(query, i * limit, limit, account)
total_count = search_results.get('count', 0)
for video in search_results.get('videos', []):
if video_id := video.get('id'):
yield self.url_result(self._LIBRARY_MAP[account][1] % video_id)
if (i + 1) * limit >= total_count:
self.to_screen(f'{query}: {total_count} video(s) found')
break