mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-28 12:25:56 +01:00
[webofstories] Fix extraction
This commit is contained in:
parent
71d53ace2f
commit
3d54788495
@ -45,19 +45,17 @@ class WebOfStoriesIE(InfoExtractor):
|
||||
description = self._html_search_meta('description', webpage)
|
||||
thumbnail = self._og_search_thumbnail(webpage)
|
||||
|
||||
story_filename = self._search_regex(
|
||||
r'\.storyFileName\("([^"]+)"\)', webpage, 'story filename')
|
||||
speaker_id = self._search_regex(
|
||||
r'\.speakerId\("([^"]+)"\)', webpage, 'speaker ID')
|
||||
story_id = self._search_regex(
|
||||
r'\.storyId\((\d+)\)', webpage, 'story ID')
|
||||
speaker_type = self._search_regex(
|
||||
r'\.speakerType\("([^"]+)"\)', webpage, 'speaker type')
|
||||
great_life = self._search_regex(
|
||||
r'isGreatLifeStory\s*=\s*(true|false)', webpage, 'great life story')
|
||||
embed_params = [s.strip(" \r\n\t'") for s in self._search_regex(
|
||||
r'(?s)\$\("#embedCode"\).html\(getEmbedCode\((.*?)\)',
|
||||
webpage, 'embed params').split(',')]
|
||||
|
||||
(
|
||||
_, speaker_id, story_id, story_duration,
|
||||
speaker_type, great_life, _thumbnail, _has_subtitles,
|
||||
story_filename, _story_order) = embed_params
|
||||
|
||||
is_great_life_series = great_life == 'true'
|
||||
duration = int_or_none(self._search_regex(
|
||||
r'\.duration\((\d+)\)', webpage, 'duration', fatal=False))
|
||||
duration = int_or_none(story_duration)
|
||||
|
||||
# URL building, see: http://www.webofstories.com/scripts/player.js
|
||||
ms_prefix = ''
|
||||
|
Loading…
Reference in New Issue
Block a user