mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-21 20:46:36 -05:00
[aenetworks] add support for History Vault(closes #18460)
This commit is contained in:
parent
5f47a60c5d
commit
3ad6dabd33
1 changed files with 10 additions and 6 deletions
|
@ -22,18 +22,19 @@ class AENetworksBaseIE(ThePlatformIE):
|
|||
|
||||
class AENetworksIE(AENetworksBaseIE):
|
||||
IE_NAME = 'aenetworks'
|
||||
IE_DESC = 'A+E Networks: A&E, Lifetime, History.com, FYI Network'
|
||||
IE_DESC = 'A+E Networks: A&E, Lifetime, History.com, FYI Network and History Vault'
|
||||
_VALID_URL = r'''(?x)
|
||||
https?://
|
||||
(?:www\.)?
|
||||
(?P<domain>
|
||||
(?:history|aetv|mylifetime|lifetimemovieclub)\.com|
|
||||
(?:history(?:vault)?|aetv|mylifetime|lifetimemovieclub)\.com|
|
||||
fyi\.tv
|
||||
)/
|
||||
(?:
|
||||
shows/(?P<show_path>[^/]+(?:/[^/]+){0,2})|
|
||||
movies/(?P<movie_display_id>[^/]+)(?:/full-movie)?|
|
||||
specials/(?P<special_display_id>[^/]+)/full-special
|
||||
specials/(?P<special_display_id>[^/]+)/full-special|
|
||||
collections/[^/]+/(?P<collection_display_id>[^/]+)
|
||||
)
|
||||
'''
|
||||
_TESTS = [{
|
||||
|
@ -80,6 +81,9 @@ class AENetworksIE(AENetworksBaseIE):
|
|||
}, {
|
||||
'url': 'http://www.history.com/specials/sniper-into-the-kill-zone/full-special',
|
||||
'only_matching': True
|
||||
}, {
|
||||
'url': 'https://www.historyvault.com/collections/america-the-story-of-us/westward',
|
||||
'only_matching': True
|
||||
}]
|
||||
_DOMAIN_TO_REQUESTOR_ID = {
|
||||
'history.com': 'HISTORY',
|
||||
|
@ -90,9 +94,9 @@ class AENetworksIE(AENetworksBaseIE):
|
|||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
domain, show_path, movie_display_id, special_display_id = re.match(self._VALID_URL, url).groups()
|
||||
display_id = show_path or movie_display_id or special_display_id
|
||||
webpage = self._download_webpage(url, display_id)
|
||||
domain, show_path, movie_display_id, special_display_id, collection_display_id = re.match(self._VALID_URL, url).groups()
|
||||
display_id = show_path or movie_display_id or special_display_id or collection_display_id
|
||||
webpage = self._download_webpage(url, display_id, headers=self.geo_verification_headers())
|
||||
if show_path:
|
||||
url_parts = show_path.split('/')
|
||||
url_parts_len = len(url_parts)
|
||||
|
|
Loading…
Reference in a new issue