|
|
@ -82,6 +82,7 @@ class BYUtvIE(InfoExtractor):
|
|
|
|
|
|
|
|
|
|
|
|
info = {}
|
|
|
|
info = {}
|
|
|
|
formats = []
|
|
|
|
formats = []
|
|
|
|
|
|
|
|
subtitles = {}
|
|
|
|
for format_id, ep in video.items():
|
|
|
|
for format_id, ep in video.items():
|
|
|
|
if not isinstance(ep, dict):
|
|
|
|
if not isinstance(ep, dict):
|
|
|
|
continue
|
|
|
|
continue
|
|
|
@ -90,12 +91,16 @@ class BYUtvIE(InfoExtractor):
|
|
|
|
continue
|
|
|
|
continue
|
|
|
|
ext = determine_ext(video_url)
|
|
|
|
ext = determine_ext(video_url)
|
|
|
|
if ext == 'm3u8':
|
|
|
|
if ext == 'm3u8':
|
|
|
|
formats.extend(self._extract_m3u8_formats(
|
|
|
|
m3u8_fmts, m3u8_subs = self._extract_m3u8_formats_and_subtitles(
|
|
|
|
video_url, video_id, 'mp4', entry_protocol='m3u8_native',
|
|
|
|
video_url, video_id, 'mp4', entry_protocol='m3u8_native',
|
|
|
|
m3u8_id='hls', fatal=False))
|
|
|
|
m3u8_id='hls', fatal=False)
|
|
|
|
|
|
|
|
formats.extend(m3u8_fmts)
|
|
|
|
|
|
|
|
subtitles = self._merge_subtitles(subtitles, m3u8_subs)
|
|
|
|
elif ext == 'mpd':
|
|
|
|
elif ext == 'mpd':
|
|
|
|
formats.extend(self._extract_mpd_formats(
|
|
|
|
mpd_fmts, mpd_subs = self._extract_mpd_formats_and_subtitles(
|
|
|
|
video_url, video_id, mpd_id='dash', fatal=False))
|
|
|
|
video_url, video_id, mpd_id='dash', fatal=False)
|
|
|
|
|
|
|
|
formats.extend(mpd_fmts)
|
|
|
|
|
|
|
|
subtitles = self._merge_subtitles(subtitles, mpd_subs)
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
formats.append({
|
|
|
|
formats.append({
|
|
|
|
'url': video_url,
|
|
|
|
'url': video_url,
|
|
|
@ -114,4 +119,5 @@ class BYUtvIE(InfoExtractor):
|
|
|
|
'display_id': display_id,
|
|
|
|
'display_id': display_id,
|
|
|
|
'title': display_id,
|
|
|
|
'title': display_id,
|
|
|
|
'formats': formats,
|
|
|
|
'formats': formats,
|
|
|
|
|
|
|
|
'subtitles': subtitles,
|
|
|
|
})
|
|
|
|
})
|
|
|
|