|
@ -379,6 +379,7 @@ class FacebookIE(InfoExtractor): |
|
|
if not video_data: |
|
|
if not video_data: |
|
|
raise ExtractorError('Cannot parse data') |
|
|
raise ExtractorError('Cannot parse data') |
|
|
|
|
|
|
|
|
|
|
|
subtitles = {} |
|
|
formats = [] |
|
|
formats = [] |
|
|
for f in video_data: |
|
|
for f in video_data: |
|
|
format_id = f['stream_type'] |
|
|
format_id = f['stream_type'] |
|
@ -402,6 +403,9 @@ class FacebookIE(InfoExtractor): |
|
|
if dash_manifest: |
|
|
if dash_manifest: |
|
|
formats.extend(self._parse_mpd_formats( |
|
|
formats.extend(self._parse_mpd_formats( |
|
|
compat_etree_fromstring(compat_urllib_parse_unquote_plus(dash_manifest)))) |
|
|
compat_etree_fromstring(compat_urllib_parse_unquote_plus(dash_manifest)))) |
|
|
|
|
|
subtitles_src = f[0].get('subtitles_src') |
|
|
|
|
|
if subtitles_src: |
|
|
|
|
|
subtitles.setdefault('en', []).append({'url': subtitles_src}) |
|
|
if not formats: |
|
|
if not formats: |
|
|
raise ExtractorError('Cannot find video formats') |
|
|
raise ExtractorError('Cannot find video formats') |
|
|
|
|
|
|
|
@ -447,6 +451,7 @@ class FacebookIE(InfoExtractor): |
|
|
'timestamp': timestamp, |
|
|
'timestamp': timestamp, |
|
|
'thumbnail': thumbnail, |
|
|
'thumbnail': thumbnail, |
|
|
'view_count': view_count, |
|
|
'view_count': view_count, |
|
|
|
|
|
'subtitles': subtitles, |
|
|
} |
|
|
} |
|
|
|
|
|
|
|
|
return webpage, info_dict |
|
|
return webpage, info_dict |
|
|