aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor
diff options
context:
space:
mode:
authorSergey M․ <dstftw@gmail.com>2017-09-23 05:35:55 +0700
committerSergey M․ <dstftw@gmail.com>2017-09-23 05:37:03 +0700
commit095774e59130c999ed8ce132f80a7164c5ee39a7 (patch)
treedde3bb36e94ed3b9db7ff7181c9545264f09250a /youtube_dl/extractor
parent2384f5a64e501d7abb844e8d31fe340b34d8d4e7 (diff)
[mixcloud] Improve and simplify (closes #14132)
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r--youtube_dl/extractor/mixcloud.py71
1 files changed, 41 insertions, 30 deletions
diff --git a/youtube_dl/extractor/mixcloud.py b/youtube_dl/extractor/mixcloud.py
index 481182380..f331db890 100644
--- a/youtube_dl/extractor/mixcloud.py
+++ b/youtube_dl/extractor/mixcloud.py
@@ -9,6 +9,7 @@ from .common import InfoExtractor
from ..compat import (
compat_chr,
compat_ord,
+ compat_str,
compat_urllib_parse_unquote,
compat_urlparse,
compat_zip
@@ -16,9 +17,12 @@ from ..compat import (
from ..utils import (
clean_html,
ExtractorError,
+ int_or_none,
OnDemandPagedList,
str_to_int,
- try_get)
+ try_get,
+ urljoin,
+)
class MixcloudIE(InfoExtractor):
@@ -61,13 +65,6 @@ class MixcloudIE(InfoExtractor):
compat_chr(compat_ord(ch) ^ compat_ord(k))
for ch, k in compat_zip(ciphertext, itertools.cycle(key))])
- @staticmethod
- def _decrypt_and_extend(stream_info, url_key, getter, key, formats):
- maybe_url = stream_info.get(url_key)
- if maybe_url is not None:
- decrypted = MixcloudIE._decrypt_xor_cipher(key, base64.b64decode(maybe_url))
- formats.extend(getter(decrypted))
-
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
uploader = mobj.group(1)
@@ -86,9 +83,12 @@ class MixcloudIE(InfoExtractor):
else:
# New path
full_info_json = self._parse_json(self._html_search_regex(
- r'<script id="relay-data" type="text/x-mixcloud">([^<]+)</script>', webpage, 'play info'), 'play info')
+ r'<script id="relay-data" type="text/x-mixcloud">([^<]+)</script>',
+ webpage, 'play info'), 'play info')
for item in full_info_json:
- item_data = try_get(item, lambda x: x['cloudcast']['data']['cloudcastLookup'])
+ item_data = try_get(
+ item, lambda x: x['cloudcast']['data']['cloudcastLookup'],
+ dict)
if try_get(item_data, lambda x: x['streamInfo']['url']):
info_json = item_data
break
@@ -100,13 +100,9 @@ class MixcloudIE(InfoExtractor):
webpage, 'error message', default=None)
js_url = self._search_regex(
- r'<script[^>]+\bsrc=["\"](https://(?:www\.)?mixcloud\.com/media/js2/www_js_4\.[^>]+\.js)',
- webpage, 'js url', default=None)
- if js_url is None:
- js_url = self._search_regex(
- r'<script[^>]+\bsrc=["\"](https://(?:www\.)?mixcloud\.com/media/js/www\.[^>]+\.js)',
- webpage, 'js url')
- js = self._download_webpage(js_url, track_id)
+ r'<script[^>]+\bsrc=["\"](https://(?:www\.)?mixcloud\.com/media/(?:js2/www_js_4|js/www)\.[^>]+\.js)',
+ webpage, 'js url')
+ js = self._download_webpage(js_url, track_id, 'Downloading JS')
# Known plaintext attack
if encrypted_play_info:
kps = ['{"stream_url":']
@@ -117,8 +113,9 @@ class MixcloudIE(InfoExtractor):
for kp in kps:
partial_key = self._decrypt_xor_cipher(kpa_target, kp)
for quote in ["'", '"']:
- key = self._search_regex(r'{0}({1}[^{0}]*){0}'.format(quote, re.escape(partial_key)), js,
- "encryption key", default=None)
+ key = self._search_regex(
+ r'{0}({1}[^{0}]*){0}'.format(quote, re.escape(partial_key)),
+ js, 'encryption key', default=None)
if key is not None:
break
else:
@@ -153,23 +150,37 @@ class MixcloudIE(InfoExtractor):
else:
title = info_json['name']
- thumbnail = try_get(info_json,
- lambda x: 'https://thumbnailer.mixcloud.com/unsafe/600x600/' + x['picture']['urlRoot'])
+ thumbnail = urljoin(
+ 'https://thumbnailer.mixcloud.com/unsafe/600x600/',
+ try_get(info_json, lambda x: x['picture']['urlRoot'], compat_str))
uploader = try_get(info_json, lambda x: x['owner']['displayName'])
uploader_id = try_get(info_json, lambda x: x['owner']['username'])
description = try_get(info_json, lambda x: x['description'])
- view_count = try_get(info_json, lambda x: x['plays'])
+ view_count = int_or_none(try_get(info_json, lambda x: x['plays']))
stream_info = info_json['streamInfo']
formats = []
- self._decrypt_and_extend(stream_info, 'url', lambda x: [{
- 'format_id': 'normal',
- 'url': x
- }], key, formats)
- self._decrypt_and_extend(stream_info, 'hlsUrl', lambda x: self._extract_m3u8_formats(x, title), key,
- formats)
- self._decrypt_and_extend(stream_info, 'dashUrl', lambda x: self._extract_mpd_formats(x, title), key,
- formats)
+
+ for url_key in ('url', 'hlsUrl', 'dashUrl'):
+ format_url = stream_info.get(url_key)
+ if not format_url:
+ continue
+ decrypted = self._decrypt_xor_cipher(key, base64.b64decode(format_url))
+ if not decrypted:
+ continue
+ if url_key == 'hlsUrl':
+ formats.extend(self._extract_m3u8_formats(
+ decrypted, track_id, 'mp4', entry_protocol='m3u8_native',
+ m3u8_id='hls', fatal=False))
+ elif url_key == 'dashUrl':
+ formats.extend(self._extract_mpd_formats(
+ decrypted, track_id, mpd_id='dash', fatal=False))
+ else:
+ formats.append({
+ 'format_id': 'http',
+ 'url': decrypted,
+ })
+ self._sort_formats(formats)
return {
'id': track_id,