mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-02 09:12:40 +01:00
[extractor/Hidive] Fix subtitles and age-restriction (#5828)
Authored by: chexxor Closes #408
This commit is contained in:
parent
b85faf6ffb
commit
7708df8da0
@ -1,5 +1,3 @@
|
|||||||
import re
|
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
ExtractorError,
|
ExtractorError,
|
||||||
@ -39,15 +37,27 @@ def _perform_login(self, username, password):
|
|||||||
form = self._search_regex(
|
form = self._search_regex(
|
||||||
r'(?s)<form[^>]+action="/account/login"[^>]*>(.+?)</form>',
|
r'(?s)<form[^>]+action="/account/login"[^>]*>(.+?)</form>',
|
||||||
webpage, 'login form', default=None)
|
webpage, 'login form', default=None)
|
||||||
if not form: # logged in
|
if not form:
|
||||||
return
|
return
|
||||||
data = self._hidden_inputs(form)
|
data = self._hidden_inputs(form)
|
||||||
data.update({
|
data.update({
|
||||||
'Email': username,
|
'Email': username,
|
||||||
'Password': password,
|
'Password': password,
|
||||||
})
|
})
|
||||||
self._download_webpage(
|
login_webpage = self._download_webpage(
|
||||||
self._LOGIN_URL, None, 'Logging in', data=urlencode_postdata(data))
|
self._LOGIN_URL, None, 'Logging in', data=urlencode_postdata(data))
|
||||||
|
# If the user has multiple profiles on their account, select one. For now pick the first profile.
|
||||||
|
profile_id = self._search_regex(r'<button [^>]+?data-profile-id="(\w+)"', login_webpage, 'profile_id')
|
||||||
|
if profile_id is None:
|
||||||
|
return # If only one profile, Hidive auto-selects it
|
||||||
|
profile_id_hash = self._search_regex(r'\<button [^>]+?data-hash="(\w+)"', login_webpage, 'profile_id_hash')
|
||||||
|
self._request_webpage(
|
||||||
|
'https://www.hidive.com/ajax/chooseprofile', None,
|
||||||
|
data=urlencode_postdata({
|
||||||
|
'profileId': profile_id,
|
||||||
|
'hash': profile_id_hash,
|
||||||
|
'returnUrl': '/dashboard'
|
||||||
|
}))
|
||||||
|
|
||||||
def _call_api(self, video_id, title, key, data={}, **kwargs):
|
def _call_api(self, video_id, title, key, data={}, **kwargs):
|
||||||
data = {
|
data = {
|
||||||
@ -60,26 +70,6 @@ def _call_api(self, video_id, title, key, data={}, **kwargs):
|
|||||||
'https://www.hidive.com/play/settings', video_id,
|
'https://www.hidive.com/play/settings', video_id,
|
||||||
data=urlencode_postdata(data), **kwargs) or {}
|
data=urlencode_postdata(data), **kwargs) or {}
|
||||||
|
|
||||||
def _extract_subtitles_from_rendition(self, rendition, subtitles, parsed_urls):
|
|
||||||
for cc_file in rendition.get('ccFiles', []):
|
|
||||||
cc_url = url_or_none(try_get(cc_file, lambda x: x[2]))
|
|
||||||
# name is used since we cant distinguish subs with same language code
|
|
||||||
cc_lang = try_get(cc_file, (lambda x: x[1].replace(' ', '-').lower(), lambda x: x[0]), str)
|
|
||||||
if cc_url not in parsed_urls and cc_lang:
|
|
||||||
parsed_urls.add(cc_url)
|
|
||||||
subtitles.setdefault(cc_lang, []).append({'url': cc_url})
|
|
||||||
|
|
||||||
def _get_subtitles(self, url, video_id, title, key, parsed_urls):
|
|
||||||
webpage = self._download_webpage(url, video_id, fatal=False) or ''
|
|
||||||
subtitles = {}
|
|
||||||
for caption in set(re.findall(r'data-captions=\"([^\"]+)\"', webpage)):
|
|
||||||
renditions = self._call_api(
|
|
||||||
video_id, title, key, {'Captions': caption}, fatal=False,
|
|
||||||
note=f'Downloading {caption} subtitle information').get('renditions') or {}
|
|
||||||
for rendition_id, rendition in renditions.items():
|
|
||||||
self._extract_subtitles_from_rendition(rendition, subtitles, parsed_urls)
|
|
||||||
return subtitles
|
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
video_id, title, key = self._match_valid_url(url).group('id', 'title', 'key')
|
video_id, title, key = self._match_valid_url(url).group('id', 'title', 'key')
|
||||||
settings = self._call_api(video_id, title, key)
|
settings = self._call_api(video_id, title, key)
|
||||||
@ -104,10 +94,20 @@ def _real_extract(self, url):
|
|||||||
f['format_note'] = f'{version}, {extra}'
|
f['format_note'] = f'{version}, {extra}'
|
||||||
formats.extend(frmt)
|
formats.extend(frmt)
|
||||||
|
|
||||||
|
subtitles = {}
|
||||||
|
for rendition_id, rendition in settings['renditions'].items():
|
||||||
|
audio, version, extra = rendition_id.split('_')
|
||||||
|
for cc_file in rendition.get('ccFiles') or []:
|
||||||
|
cc_url = url_or_none(try_get(cc_file, lambda x: x[2]))
|
||||||
|
cc_lang = try_get(cc_file, (lambda x: x[1].replace(' ', '-').lower(), lambda x: x[0]), str)
|
||||||
|
if cc_url not in parsed_urls and cc_lang:
|
||||||
|
parsed_urls.add(cc_url)
|
||||||
|
subtitles.setdefault(cc_lang, []).append({'url': cc_url})
|
||||||
|
|
||||||
return {
|
return {
|
||||||
'id': video_id,
|
'id': video_id,
|
||||||
'title': video_id,
|
'title': video_id,
|
||||||
'subtitles': self.extract_subtitles(url, video_id, title, key, parsed_urls),
|
'subtitles': subtitles,
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
'series': title,
|
'series': title,
|
||||||
'season_number': int_or_none(
|
'season_number': int_or_none(
|
||||||
|
Loading…
Reference in New Issue
Block a user