mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-29 16:35:51 +01:00
104 lines
3.3 KiB
Python
104 lines
3.3 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_str
|
|
from ..utils import int_or_none
|
|
|
|
|
|
class BeatportProIE(InfoExtractor):
|
|
_VALID_URL = r'https?://pro\.beatport\.com/track/(?P<display_id>[^/]+)/(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'https://pro.beatport.com/track/synesthesia-original-mix/5379371',
|
|
'md5': 'b3c34d8639a2f6a7f734382358478887',
|
|
'info_dict': {
|
|
'id': '5379371',
|
|
'display_id': 'synesthesia-original-mix',
|
|
'ext': 'mp4',
|
|
'title': 'Froxic - Synesthesia (Original Mix)',
|
|
},
|
|
}, {
|
|
'url': 'https://pro.beatport.com/track/love-and-war-original-mix/3756896',
|
|
'md5': 'e44c3025dfa38c6577fbaeb43da43514',
|
|
'info_dict': {
|
|
'id': '3756896',
|
|
'display_id': 'love-and-war-original-mix',
|
|
'ext': 'mp3',
|
|
'title': 'Wolfgang Gartner - Love & War (Original Mix)',
|
|
},
|
|
}, {
|
|
'url': 'https://pro.beatport.com/track/birds-original-mix/4991738',
|
|
'md5': 'a1fd8e8046de3950fd039304c186c05f',
|
|
'info_dict': {
|
|
'id': '4991738',
|
|
'display_id': 'birds-original-mix',
|
|
'ext': 'mp4',
|
|
'title': "Tos, Middle Milk, Mumblin' Johnsson - Birds (Original Mix)",
|
|
}
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
track_id = mobj.group('id')
|
|
display_id = mobj.group('display_id')
|
|
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
playables = self._parse_json(
|
|
self._search_regex(
|
|
r'window\.Playables\s*=\s*({.+?});', webpage,
|
|
'playables info', flags=re.DOTALL),
|
|
track_id)
|
|
|
|
track = next(t for t in playables['tracks'] if t['id'] == int(track_id))
|
|
|
|
title = ', '.join((a['name'] for a in track['artists'])) + ' - ' + track['name']
|
|
if track['mix']:
|
|
title += ' (' + track['mix'] + ')'
|
|
|
|
formats = []
|
|
for ext, info in track['preview'].items():
|
|
if not info['url']:
|
|
continue
|
|
fmt = {
|
|
'url': info['url'],
|
|
'ext': ext,
|
|
'format_id': ext,
|
|
'vcodec': 'none',
|
|
}
|
|
if ext == 'mp3':
|
|
fmt['preference'] = 0
|
|
fmt['acodec'] = 'mp3'
|
|
fmt['abr'] = 96
|
|
fmt['asr'] = 44100
|
|
elif ext == 'mp4':
|
|
fmt['preference'] = 1
|
|
fmt['acodec'] = 'aac'
|
|
fmt['abr'] = 96
|
|
fmt['asr'] = 44100
|
|
formats.append(fmt)
|
|
self._sort_formats(formats)
|
|
|
|
images = []
|
|
for name, info in track['images'].items():
|
|
image_url = info.get('url')
|
|
if name == 'dynamic' or not image_url:
|
|
continue
|
|
image = {
|
|
'id': name,
|
|
'url': image_url,
|
|
'height': int_or_none(info.get('height')),
|
|
'width': int_or_none(info.get('width')),
|
|
}
|
|
images.append(image)
|
|
|
|
return {
|
|
'id': compat_str(track.get('id')) or track_id,
|
|
'display_id': track.get('slug') or display_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'thumbnails': images,
|
|
}
|