mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-05 01:44:33 +01:00
[bandcamp] Use our API to get more stable error messages (#1194)
This commit is contained in:
parent
28f1272870
commit
99c2398bc6
@ -72,26 +72,29 @@ class BandcampIE(InfoExtractor):
|
|||||||
|
|
||||||
download_link = m_download.group(1)
|
download_link = m_download.group(1)
|
||||||
video_id = self._search_regex(
|
video_id = self._search_regex(
|
||||||
r'var TralbumData = {.*?id: (?P<id>\d+),?$',
|
r'(?ms)var TralbumData = {.*?id: (?P<id>\d+),?$',
|
||||||
webpage, 'video id', flags=re.MULTILINE | re.DOTALL)
|
webpage, 'video id')
|
||||||
|
|
||||||
download_webpage = self._download_webpage(download_link, video_id, 'Downloading free downloads page')
|
download_webpage = self._download_webpage(download_link, video_id, 'Downloading free downloads page')
|
||||||
# We get the dictionary of the track from some javascript code
|
# We get the dictionary of the track from some javascript code
|
||||||
info = re.search(r'items: (.*?),$', download_webpage, re.MULTILINE).group(1)
|
all_info = self._parse_json(self._search_regex(
|
||||||
info = json.loads(info)[0]
|
r'(?sm)items: (.*?),$', download_webpage, 'items'), video_id)
|
||||||
|
info = All_info[0]
|
||||||
# We pick mp3-320 for now, until format selection can be easily implemented.
|
# We pick mp3-320 for now, until format selection can be easily implemented.
|
||||||
mp3_info = info['downloads']['mp3-320']
|
mp3_info = info['downloads']['mp3-320']
|
||||||
# If we try to use this url it says the link has expired
|
# If we try to use this url it says the link has expired
|
||||||
initial_url = mp3_info['url']
|
initial_url = mp3_info['url']
|
||||||
re_url = r'(?P<server>http://(.*?)\.bandcamp\.com)/download/track\?enc=mp3-320&fsig=(?P<fsig>.*?)&id=(?P<id>.*?)&ts=(?P<ts>.*)$'
|
m_url = re.match(
|
||||||
m_url = re.match(re_url, initial_url)
|
r'(?P<server>http://(.*?)\.bandcamp\.com)/download/track\?enc=mp3-320&fsig=(?P<fsig>.*?)&id=(?P<id>.*?)&ts=(?P<ts>.*)$',
|
||||||
|
initial_url)
|
||||||
# We build the url we will use to get the final track url
|
# We build the url we will use to get the final track url
|
||||||
# This url is build in Bandcamp in the script download_bunde_*.js
|
# This url is build in Bandcamp in the script download_bunde_*.js
|
||||||
request_url = '%s/statdownload/track?enc=mp3-320&fsig=%s&id=%s&ts=%s&.rand=665028774616&.vrs=1' % (m_url.group('server'), m_url.group('fsig'), video_id, m_url.group('ts'))
|
request_url = '%s/statdownload/track?enc=mp3-320&fsig=%s&id=%s&ts=%s&.rand=665028774616&.vrs=1' % (m_url.group('server'), m_url.group('fsig'), video_id, m_url.group('ts'))
|
||||||
final_url_webpage = self._download_webpage(request_url, video_id, 'Requesting download url')
|
final_url_webpage = self._download_webpage(request_url, video_id, 'Requesting download url')
|
||||||
# If we could correctly generate the .rand field the url would be
|
# If we could correctly generate the .rand field the url would be
|
||||||
# in the "download_url" key
|
# in the "download_url" key
|
||||||
final_url = re.search(r'"retry_url":"(.*?)"', final_url_webpage).group(1)
|
final_url = self._search_regex(
|
||||||
|
r'"retry_url":"(.*?)"', final_url_webpage, 'final video URL')
|
||||||
|
|
||||||
return {
|
return {
|
||||||
'id': video_id,
|
'id': video_id,
|
||||||
|
Loading…
Reference in New Issue
Block a user