forked from minhngoc25a/yt-dlc
bbc.py: correct syntax
This commit is contained in:
parent
36da48798a
commit
a3bfddfa5e
|
@ -397,14 +397,14 @@ class BBCNewsIE(BBCCoUkIE):
|
||||||
'title': 'Russia stages massive WW2 parade despite Western boycott',
|
'title': 'Russia stages massive WW2 parade despite Western boycott',
|
||||||
},
|
},
|
||||||
'playlist_count': 2,
|
'playlist_count': 2,
|
||||||
},{
|
}, {
|
||||||
'url': 'http://www.bbc.com/news/business-28299555',
|
'url': 'http://www.bbc.com/news/business-28299555',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': 'business-28299555',
|
'id': 'business-28299555',
|
||||||
'title': 'Farnborough Airshow: Video highlights',
|
'title': 'Farnborough Airshow: Video highlights',
|
||||||
},
|
},
|
||||||
'playlist_count': 9,
|
'playlist_count': 9,
|
||||||
},{
|
}, {
|
||||||
'url': 'http://www.bbc.com/news/world-europe-32041533',
|
'url': 'http://www.bbc.com/news/world-europe-32041533',
|
||||||
'note': 'Video',
|
'note': 'Video',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
|
@ -419,7 +419,7 @@ class BBCNewsIE(BBCCoUkIE):
|
||||||
'params': {
|
'params': {
|
||||||
'skip_download': True,
|
'skip_download': True,
|
||||||
}
|
}
|
||||||
},{
|
}, {
|
||||||
'url': 'http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu',
|
'url': 'http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu',
|
||||||
'note': 'Video',
|
'note': 'Video',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
|
@ -434,7 +434,7 @@ class BBCNewsIE(BBCCoUkIE):
|
||||||
'params': {
|
'params': {
|
||||||
'skip_download': True,
|
'skip_download': True,
|
||||||
}
|
}
|
||||||
},{
|
}, {
|
||||||
'url': 'http://www.bbc.com/mundo/video_fotos/2015/06/150619_video_honduras_militares_hospitales_corrupcion_aw',
|
'url': 'http://www.bbc.com/mundo/video_fotos/2015/06/150619_video_honduras_militares_hospitales_corrupcion_aw',
|
||||||
'note': 'Video',
|
'note': 'Video',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
|
@ -459,88 +459,88 @@ class BBCNewsIE(BBCCoUkIE):
|
||||||
|
|
||||||
pubdate = self._html_search_regex(r'"datePublished":\s*"(\d+-\d+-\d+)', webpage, 'date', default=None)
|
pubdate = self._html_search_regex(r'"datePublished":\s*"(\d+-\d+-\d+)', webpage, 'date', default=None)
|
||||||
if pubdate:
|
if pubdate:
|
||||||
pubdate = pubdate.replace('-','')
|
pubdate = pubdate.replace('-', '')
|
||||||
|
|
||||||
ret = []
|
ret = []
|
||||||
jsent = []
|
jsent = []
|
||||||
|
|
||||||
# works with bbc.com/news/something-something-123456 articles
|
# works with bbc.com/news/something-something-123456 articles
|
||||||
jsent = map(
|
jsent = map(
|
||||||
lambda m: self._parse_json(m,list_id),
|
lambda m: self._parse_json(m, list_id),
|
||||||
re.findall(r"data-media-meta='({[^']+})'", webpage)
|
re.findall(r"data-media-meta='({[^']+})'", webpage)
|
||||||
)
|
)
|
||||||
|
|
||||||
if len(jsent) == 0:
|
if len(jsent) == 0:
|
||||||
# http://www.bbc.com/news/video_and_audio/international
|
# http://www.bbc.com/news/video_and_audio/international
|
||||||
# and single-video articles
|
# and single-video articles
|
||||||
masset = self._html_search_regex(r'mediaAssetPage\.init\(\s*({.+?}), "/', webpage, 'mediaassets', default=None)
|
masset = self._html_search_regex(r'mediaAssetPage\.init\(\s*({.+?}), "/', webpage, 'mediaassets', default=None)
|
||||||
if masset:
|
if masset:
|
||||||
jmasset = self._parse_json(masset,list_id)
|
jmasset = self._parse_json(masset, list_id)
|
||||||
for key, val in jmasset.get('videos',{}).items():
|
for key, val in jmasset.get('videos', {}).items():
|
||||||
for skey, sval in val.items():
|
for skey, sval in val.items():
|
||||||
sval['id'] = skey
|
sval['id'] = skey
|
||||||
jsent.append(sval)
|
jsent.append(sval)
|
||||||
|
|
||||||
if len(jsent) == 0:
|
if len(jsent) == 0:
|
||||||
# stubbornly generic extractor for {json with "image":{allvideoshavethis},etc}
|
# stubbornly generic extractor for {json with "image":{allvideoshavethis},etc}
|
||||||
# in http://www.bbc.com/news/video_and_audio/international
|
# in http://www.bbc.com/news/video_and_audio/international
|
||||||
# prone to breaking if entries have sourceFiles list
|
# prone to breaking if entries have sourceFiles list
|
||||||
jsent = map(
|
jsent = map(
|
||||||
lambda m: self._parse_json(m,list_id),
|
lambda m: self._parse_json(m, list_id),
|
||||||
re.findall(r"({[^{}]+image\":{[^}]+}[^}]+})", webpage)
|
re.findall(r"({[^{}]+image\":{[^}]+}[^}]+})", webpage)
|
||||||
)
|
)
|
||||||
|
|
||||||
if len(jsent) == 0:
|
if len(jsent) == 0:
|
||||||
raise ExtractorError('No video found', expected=True)
|
raise ExtractorError('No video found', expected=True)
|
||||||
|
|
||||||
for jent in jsent:
|
for jent in jsent:
|
||||||
programme_id = jent.get('externalId')
|
programme_id = jent.get('externalId')
|
||||||
xml_url = jent.get('href')
|
xml_url = jent.get('href')
|
||||||
|
|
||||||
title = jent.get('caption','')
|
title = jent.get('caption', '')
|
||||||
if title == '':
|
if title == '':
|
||||||
title = list_title
|
title = list_title
|
||||||
|
|
||||||
duration = parse_duration(jent.get('duration'))
|
duration = parse_duration(jent.get('duration'))
|
||||||
description = list_title
|
description = list_title
|
||||||
if jent.get('caption', '') != '':
|
if jent.get('caption', '') != '':
|
||||||
description += ' - ' + jent.get('caption')
|
description += ' - ' + jent.get('caption')
|
||||||
thumbnail = None
|
thumbnail = None
|
||||||
if jent.has_key('image'):
|
if jent.get('image') is not None:
|
||||||
thumbnail=jent['image'].get('href')
|
thumbnail = jent['image'].get('href')
|
||||||
|
|
||||||
formats = []
|
formats = []
|
||||||
subtitles = []
|
subtitles = []
|
||||||
|
|
||||||
if programme_id:
|
if programme_id:
|
||||||
formats, subtitles = self._download_media_selector(programme_id)
|
formats, subtitles = self._download_media_selector(programme_id)
|
||||||
elif jent.has_key('sourceFiles'):
|
elif jent.get('sourceFiles') is not None:
|
||||||
# mediaselector not used at
|
# mediaselector not used at
|
||||||
# http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu
|
# http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu
|
||||||
for key, val in jent['sourceFiles'].items():
|
for key, val in jent['sourceFiles'].items():
|
||||||
formats.append( {
|
formats.append({
|
||||||
'ext': val.get('encoding'),
|
'ext': val.get('encoding'),
|
||||||
'url': val.get('url'),
|
'url': val.get('url'),
|
||||||
'filesize': int(val.get('filesize')),
|
'filesize': int(val.get('filesize')),
|
||||||
'format_id': key
|
'format_id': key
|
||||||
} )
|
})
|
||||||
elif xml_url:
|
elif xml_url:
|
||||||
# Cheap fallback
|
# Cheap fallback
|
||||||
# http://playlists.bbc.co.uk/news/(list_id)[ABC..]/playlist.sxml
|
# http://playlists.bbc.co.uk/news/(list_id)[ABC..]/playlist.sxml
|
||||||
xml = self._download_webpage(xml_url, programme_id, 'Downloading playlist.sxml for externalId (fallback)')
|
xml = self._download_webpage(xml_url, programme_id, 'Downloading playlist.sxml for externalId (fallback)')
|
||||||
programme_id = self._search_regex(r'<mediator [^>]*identifier="(.+?)"', xml, 'playlist.sxml (externalId fallback)')
|
programme_id = self._search_regex(r'<mediator [^>]*identifier="(.+?)"', xml, 'playlist.sxml (externalId fallback)')
|
||||||
formats, subtitles = self._download_media_selector(programme_id)
|
formats, subtitles = self._download_media_selector(programme_id)
|
||||||
|
|
||||||
if len(formats) == 0:
|
if len(formats) == 0:
|
||||||
raise ExtractorError('unsupported json media entry.\n '+str(jent)+'\n')
|
raise ExtractorError('unsupported json media entry.\n ' + str(jent) + '\n')
|
||||||
|
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
|
|
||||||
id = jent.get('id') if programme_id == None else programme_id
|
id = jent.get('id') if programme_id is None else programme_id
|
||||||
if id == None:
|
if id is None:
|
||||||
id = 'NA'
|
id = 'NA'
|
||||||
|
|
||||||
ret.append( {
|
ret.append({
|
||||||
'id': id,
|
'id': id,
|
||||||
'uploader': 'BBC News',
|
'uploader': 'BBC News',
|
||||||
'upload_date': pubdate,
|
'upload_date': pubdate,
|
||||||
|
@ -550,8 +550,8 @@ class BBCNewsIE(BBCCoUkIE):
|
||||||
'duration': duration,
|
'duration': duration,
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
'subtitles': subtitles,
|
'subtitles': subtitles,
|
||||||
} )
|
})
|
||||||
|
|
||||||
if len(ret) > 0:
|
if len(ret) > 0:
|
||||||
return self.playlist_result(ret, list_id, list_title)
|
return self.playlist_result(ret, list_id, list_title)
|
||||||
raise ExtractorError('No video found', expected=True)
|
raise ExtractorError('No video found', expected=True)
|
||||||
|
|
Loading…
Reference in New Issue