1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-25 08:10:31 +01:00

[bandcamp] move the album test to the album extractor and return a single track instead of a playlist

This commit is contained in:
Jaime Marquínez Ferrándiz 2013-11-22 21:19:31 +01:00
parent 50123be421
commit d35dc6d3b5

View File

@ -20,28 +20,6 @@ class BandcampIE(InfoExtractor):
u"title": u"youtube-dl test song \"'/\\\u00e4\u21ad"
},
u'skip': u'There is a limit of 200 free downloads / month for the test song'
}, {
u'url': u'http://blazo.bandcamp.com/album/jazz-format-mixtape-vol-1',
u'playlist': [
{
u'file': u'1353101989.mp3',
u'md5': u'39bc1eded3476e927c724321ddf116cf',
u'info_dict': {
u'title': u'Intro',
}
},
{
u'file': u'38097443.mp3',
u'md5': u'1a2c32e2691474643e912cc6cd4bffaa',
u'info_dict': {
u'title': u'Kero One - Keep It Alive (Blazo remix)',
}
},
],
u'params': {
u'playlistend': 2
},
u'skip': u'Bancamp imposes download limits. See test_playlists:test_bandcamp_album for the playlist test'
}]
def _real_extract(self, url):
@ -63,13 +41,11 @@ class BandcampIE(InfoExtractor):
'url': format_url,
'ext': format_id.partition('-')[0]
} for format_id, format_url in sorted(d['file'].items())]
entries.append({
return {
'id': compat_str(d['id']),
'title': d['title'],
'formats': formats,
})
return self.playlist_result(entries, title, title)
}
else:
raise ExtractorError(u'No free songs found')
@ -112,6 +88,30 @@ class BandcampAlbumIE(InfoExtractor):
IE_NAME = u'Bandcamp:album'
_VALID_URL = r'http://.*?\.bandcamp\.com/album/(?P<title>.*)'
_TEST = {
u'url': u'http://blazo.bandcamp.com/album/jazz-format-mixtape-vol-1',
u'playlist': [
{
u'file': u'1353101989.mp3',
u'md5': u'39bc1eded3476e927c724321ddf116cf',
u'info_dict': {
u'title': u'Intro',
}
},
{
u'file': u'38097443.mp3',
u'md5': u'1a2c32e2691474643e912cc6cd4bffaa',
u'info_dict': {
u'title': u'Kero One - Keep It Alive (Blazo remix)',
}
},
],
u'params': {
u'playlistend': 2
},
u'skip': u'Bancamp imposes download limits. See test_playlists:test_bandcamp_album for the playlist test'
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
title = mobj.group('title')