[extractor/generic] Use compat_urllib_parse_unquote for unquoting video_id and title from URL
parent
58bde34a23
commit
1ddb9456c4
|
@ -9,9 +9,10 @@ from .common import InfoExtractor
|
|||
from .youtube import YoutubeIE
|
||||
from ..compat import (
|
||||
compat_urllib_parse,
|
||||
compat_urllib_parse_unquote,
|
||||
compat_urllib_request,
|
||||
compat_urlparse,
|
||||
compat_xml_parse_error,
|
||||
compat_urllib_request,
|
||||
)
|
||||
from ..utils import (
|
||||
determine_ext,
|
||||
|
@ -895,7 +896,7 @@ class GenericIE(InfoExtractor):
|
|||
force_videoid = smuggled_data['force_videoid']
|
||||
video_id = force_videoid
|
||||
else:
|
||||
video_id = os.path.splitext(compat_urllib_parse.unquote(url.rstrip('/').split('/')[-1]))[0]
|
||||
video_id = compat_urllib_parse_unquote(os.path.splitext(url.rstrip('/').split('/')[-1])[0])
|
||||
|
||||
self.to_screen('%s: Requesting header' % video_id)
|
||||
|
||||
|
@ -930,7 +931,7 @@ class GenericIE(InfoExtractor):
|
|||
head_response.headers.get('Last-Modified'))
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': os.path.splitext(compat_urllib_parse.unquote(url_basename(url)))[0],
|
||||
'title': compat_urllib_parse_unquote(os.path.splitext(url_basename(url))[0]),
|
||||
'direct': True,
|
||||
'formats': [{
|
||||
'format_id': m.group('format_id'),
|
||||
|
@ -966,7 +967,7 @@ class GenericIE(InfoExtractor):
|
|||
head_response.headers.get('Last-Modified'))
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': os.path.splitext(compat_urllib_parse.unquote(url_basename(url)))[0],
|
||||
'title': compat_urllib_parse_unquote(os.path.splitext(url_basename(url))[0]),
|
||||
'direct': True,
|
||||
'url': url,
|
||||
'upload_date': upload_date,
|
||||
|
|
Loading…
Reference in New Issue