mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-25 03:32:05 +00:00
[youtube:tab] Add support for hashtag videos extraction(closes #28308)
This commit is contained in:
parent
ca304beb15
commit
3748863070
1 changed files with 84 additions and 53 deletions
|
@ -1959,7 +1959,7 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
invidio\.us
|
||||
)/
|
||||
(?:
|
||||
(?:channel|c|user|feed)/|
|
||||
(?:channel|c|user|feed|hashtag)/|
|
||||
(?:playlist|watch)\?.*?\blist=|
|
||||
(?!(?:watch|embed|v|e)\b)
|
||||
)
|
||||
|
@ -2245,6 +2245,13 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
}, {
|
||||
'url': 'https://www.youtube.com/TheYoungTurks/live',
|
||||
'only_matching': True,
|
||||
}, {
|
||||
'url': 'https://www.youtube.com/hashtag/cctv9',
|
||||
'info_dict': {
|
||||
'id': 'cctv9',
|
||||
'title': '#cctv9',
|
||||
},
|
||||
'playlist_mincount': 350,
|
||||
}]
|
||||
|
||||
@classmethod
|
||||
|
@ -2392,6 +2399,14 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
for entry in self._post_thread_entries(renderer):
|
||||
yield entry
|
||||
|
||||
def _rich_grid_entries(self, contents):
|
||||
for content in contents:
|
||||
video_renderer = try_get(content, lambda x: x['richItemRenderer']['content']['videoRenderer'], dict)
|
||||
if video_renderer:
|
||||
entry = self._video_entry(video_renderer)
|
||||
if entry:
|
||||
yield entry
|
||||
|
||||
@staticmethod
|
||||
def _build_continuation_query(continuation, ctp=None):
|
||||
query = {
|
||||
|
@ -2442,55 +2457,60 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
if not tab_content:
|
||||
return
|
||||
slr_renderer = try_get(tab_content, lambda x: x['sectionListRenderer'], dict)
|
||||
if not slr_renderer:
|
||||
return
|
||||
is_channels_tab = tab.get('title') == 'Channels'
|
||||
continuation = None
|
||||
slr_contents = try_get(slr_renderer, lambda x: x['contents'], list) or []
|
||||
for slr_content in slr_contents:
|
||||
if not isinstance(slr_content, dict):
|
||||
continue
|
||||
is_renderer = try_get(slr_content, lambda x: x['itemSectionRenderer'], dict)
|
||||
if not is_renderer:
|
||||
continue
|
||||
isr_contents = try_get(is_renderer, lambda x: x['contents'], list) or []
|
||||
for isr_content in isr_contents:
|
||||
if not isinstance(isr_content, dict):
|
||||
if slr_renderer:
|
||||
is_channels_tab = tab.get('title') == 'Channels'
|
||||
continuation = None
|
||||
slr_contents = try_get(slr_renderer, lambda x: x['contents'], list) or []
|
||||
for slr_content in slr_contents:
|
||||
if not isinstance(slr_content, dict):
|
||||
continue
|
||||
renderer = isr_content.get('playlistVideoListRenderer')
|
||||
if renderer:
|
||||
for entry in self._playlist_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
is_renderer = try_get(slr_content, lambda x: x['itemSectionRenderer'], dict)
|
||||
if not is_renderer:
|
||||
continue
|
||||
renderer = isr_content.get('gridRenderer')
|
||||
if renderer:
|
||||
for entry in self._grid_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
continue
|
||||
renderer = isr_content.get('shelfRenderer')
|
||||
if renderer:
|
||||
for entry in self._shelf_entries(renderer, not is_channels_tab):
|
||||
yield entry
|
||||
continue
|
||||
renderer = isr_content.get('backstagePostThreadRenderer')
|
||||
if renderer:
|
||||
for entry in self._post_thread_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
continue
|
||||
renderer = isr_content.get('videoRenderer')
|
||||
if renderer:
|
||||
entry = self._video_entry(renderer)
|
||||
if entry:
|
||||
yield entry
|
||||
isr_contents = try_get(is_renderer, lambda x: x['contents'], list) or []
|
||||
for isr_content in isr_contents:
|
||||
if not isinstance(isr_content, dict):
|
||||
continue
|
||||
renderer = isr_content.get('playlistVideoListRenderer')
|
||||
if renderer:
|
||||
for entry in self._playlist_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
continue
|
||||
renderer = isr_content.get('gridRenderer')
|
||||
if renderer:
|
||||
for entry in self._grid_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
continue
|
||||
renderer = isr_content.get('shelfRenderer')
|
||||
if renderer:
|
||||
for entry in self._shelf_entries(renderer, not is_channels_tab):
|
||||
yield entry
|
||||
continue
|
||||
renderer = isr_content.get('backstagePostThreadRenderer')
|
||||
if renderer:
|
||||
for entry in self._post_thread_entries(renderer):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(renderer)
|
||||
continue
|
||||
renderer = isr_content.get('videoRenderer')
|
||||
if renderer:
|
||||
entry = self._video_entry(renderer)
|
||||
if entry:
|
||||
yield entry
|
||||
|
||||
if not continuation:
|
||||
continuation = self._extract_continuation(is_renderer)
|
||||
if not continuation:
|
||||
continuation = self._extract_continuation(is_renderer)
|
||||
|
||||
if not continuation:
|
||||
continuation = self._extract_continuation(slr_renderer)
|
||||
continuation = self._extract_continuation(slr_renderer)
|
||||
else:
|
||||
rich_grid_renderer = tab_content.get('richGridRenderer')
|
||||
if not rich_grid_renderer:
|
||||
return
|
||||
for entry in self._rich_grid_entries(rich_grid_renderer.get('contents') or []):
|
||||
yield entry
|
||||
continuation = self._extract_continuation(rich_grid_renderer)
|
||||
|
||||
headers = {
|
||||
'x-youtube-client-name': '1',
|
||||
|
@ -2586,6 +2606,12 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
yield entry
|
||||
continuation = self._extract_continuation(continuation_renderer)
|
||||
continue
|
||||
renderer = continuation_item.get('richItemRenderer')
|
||||
if renderer:
|
||||
for entry in self._rich_grid_entries(continuation_items):
|
||||
yield entry
|
||||
continuation = self._extract_continuation({'contents': continuation_items})
|
||||
continue
|
||||
|
||||
break
|
||||
|
||||
|
@ -2642,7 +2668,8 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
selected_tab = self._extract_selected_tab(tabs)
|
||||
renderer = try_get(
|
||||
data, lambda x: x['metadata']['channelMetadataRenderer'], dict)
|
||||
playlist_id = title = description = None
|
||||
playlist_id = item_id
|
||||
title = description = None
|
||||
if renderer:
|
||||
channel_title = renderer.get('title') or item_id
|
||||
tab_title = selected_tab.get('title')
|
||||
|
@ -2651,12 +2678,16 @@ class YoutubeTabIE(YoutubeBaseInfoExtractor):
|
|||
title += ' - %s' % tab_title
|
||||
description = renderer.get('description')
|
||||
playlist_id = renderer.get('externalId')
|
||||
renderer = try_get(
|
||||
data, lambda x: x['metadata']['playlistMetadataRenderer'], dict)
|
||||
if renderer:
|
||||
title = renderer.get('title')
|
||||
description = None
|
||||
playlist_id = item_id
|
||||
else:
|
||||
renderer = try_get(
|
||||
data, lambda x: x['metadata']['playlistMetadataRenderer'], dict)
|
||||
if renderer:
|
||||
title = renderer.get('title')
|
||||
else:
|
||||
renderer = try_get(
|
||||
data, lambda x: x['header']['hashtagHeaderRenderer'], dict)
|
||||
if renderer:
|
||||
title = try_get(renderer, lambda x: x['hashtag']['simpleText'])
|
||||
playlist = self.playlist_result(
|
||||
self._entries(selected_tab, identity_token),
|
||||
playlist_id=playlist_id, playlist_title=title,
|
||||
|
|
Loading…
Reference in a new issue