mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-02 17:22:31 +01:00
[bigflix] Improve formats extraction
This commit is contained in:
parent
6e99d5762a
commit
a9bbd26f1d
@ -20,7 +20,7 @@ class BigflixIE(InfoExtractor):
|
||||
'description': 'md5:3d2ba5815f14911d5cc6a501ae0cf65d',
|
||||
}
|
||||
}, {
|
||||
# multiple formats
|
||||
# 2 formats
|
||||
'url': 'http://www.bigflix.com/Tamil-movies/Drama-movies/Madarasapatinam/16070',
|
||||
'info_dict': {
|
||||
'id': '16070',
|
||||
@ -32,6 +32,10 @@ class BigflixIE(InfoExtractor):
|
||||
'params': {
|
||||
'skip_download': True,
|
||||
}
|
||||
}, {
|
||||
# multiple formats
|
||||
'url': 'http://www.bigflix.com/Malayalam-movies/Drama-movies/Indian-Rupee/15967',
|
||||
'only_matching': True,
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
@ -45,21 +49,32 @@ def _real_extract(self, url):
|
||||
|
||||
def decode_url(quoted_b64_url):
|
||||
return base64.b64decode(compat_urllib_parse_unquote(
|
||||
quoted_b64_url)).encode('ascii').decode('utf-8')
|
||||
quoted_b64_url).encode('ascii')).decode('utf-8')
|
||||
|
||||
formats = [{
|
||||
'url': decode_url(encoded_url),
|
||||
formats = []
|
||||
for height, encoded_url in re.findall(
|
||||
r'ContentURL_(\d{3,4})[pP][^=]+=([^&]+)', webpage):
|
||||
video_url = decode_url(encoded_url)
|
||||
f = {
|
||||
'url': video_url,
|
||||
'format_id': '%sp' % height,
|
||||
'height': int(height),
|
||||
} for height, encoded_url in re.findall(
|
||||
r'ContentURL_(\d{3,4})[pP][^=]+=([^&]+)', webpage)]
|
||||
}
|
||||
if video_url.startswith('rtmp'):
|
||||
f['ext'] = 'flv'
|
||||
formats.append(f)
|
||||
|
||||
if not formats:
|
||||
file_url = self._search_regex(
|
||||
r'file=([^&]+)', webpage, 'video url', default=None)
|
||||
if file_url:
|
||||
video_url = decode_url(file_url)
|
||||
if all(f['url'] != video_url for f in formats):
|
||||
formats.append({
|
||||
'url': decode_url(self._search_regex(
|
||||
r'file=([^&]+)', webpage, 'video url')),
|
||||
'url': decode_url(file_url),
|
||||
})
|
||||
|
||||
self._sort_formats(formats)
|
||||
|
||||
description = self._html_search_meta('description', webpage)
|
||||
|
||||
return {
|
||||
|
Loading…
Reference in New Issue
Block a user