aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--youtube_dl/extractor/yandexdisk.py144
1 files changed, 84 insertions, 60 deletions
diff --git a/youtube_dl/extractor/yandexdisk.py b/youtube_dl/extractor/yandexdisk.py
index e8f6ae10f..21f37c192 100644
--- a/youtube_dl/extractor/yandexdisk.py
+++ b/youtube_dl/extractor/yandexdisk.py
@@ -1,19 +1,40 @@
# coding: utf-8
from __future__ import unicode_literals
+import json
+
from .common import InfoExtractor
-from ..compat import compat_str
+from ..compat import compat_HTTPError
from ..utils import (
determine_ext,
+ ExtractorError,
float_or_none,
int_or_none,
- try_get,
- urlencode_postdata,
+ mimetype2ext,
+ parse_iso8601,
+ urljoin,
)
class YandexDiskIE(InfoExtractor):
- _VALID_URL = r'https?://yadi\.sk/[di]/(?P<id>[^/?#&]+)'
+ _VALID_URL = r'''(?x)https?://
+ (?:
+ (?:www\.)?yadi\.sk|
+ disk\.yandex\.
+ (?:
+ az|
+ by|
+ co(?:m(?:\.(?:am|ge|tr))?|\.il)|
+ ee|
+ fr|
+ k[gz]|
+ l[tv]|
+ md|
+ t[jm]|
+ u[az]|
+ ru
+ )
+ )/(?:[di]/|public.*?\bhash=)(?P<id>[^/?#&]+)'''
_TESTS = [{
'url': 'https://yadi.sk/i/VdOeDou8eZs6Y',
@@ -25,94 +46,97 @@ class YandexDiskIE(InfoExtractor):
'duration': 168.6,
'uploader': 'y.botova',
'uploader_id': '300043621',
+ 'timestamp': 1421396809,
+ 'upload_date': '20150116',
'view_count': int,
},
}, {
'url': 'https://yadi.sk/d/h3WAXvDS3Li3Ce',
'only_matching': True,
+ }, {
+ 'url': 'https://yadi.sk/public?hash=5DZ296JK9GWCLp02f6jrObjnctjRxMs8L6%2B%2FuhNqk38%3D',
+ 'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
- status = self._download_webpage(
- 'https://disk.yandex.com/auth/status', video_id, query={
- 'urlOrigin': url,
- 'source': 'public',
- 'md5': 'false',
- })
-
- sk = self._search_regex(
- r'(["\'])sk(?:External)?\1\s*:\s*(["\'])(?P<value>(?:(?!\2).)+)\2',
- status, 'sk', group='value')
-
- webpage = self._download_webpage(url, video_id)
-
- models = self._parse_json(
- self._search_regex(
- r'<script[^>]+id=["\']models-client[^>]+>\s*(\[.+?\])\s*</script',
- webpage, 'video JSON'),
- video_id)
-
- data = next(
- model['data'] for model in models
- if model.get('model') == 'resource')
-
- video_hash = data['id']
- title = data['name']
-
- models = self._download_json(
- 'https://disk.yandex.com/models/', video_id,
- data=urlencode_postdata({
- '_model.0': 'videoInfo',
- 'id.0': video_hash,
- '_model.1': 'do-get-resource-url',
- 'id.1': video_hash,
- 'version': '13.6',
- 'sk': sk,
- }), query={'_m': 'videoInfo'})['models']
-
- videos = try_get(models, lambda x: x[0]['data']['videos'], list) or []
- source_url = try_get(
- models, lambda x: x[1]['data']['file'], compat_str)
+ try:
+ resource = self._download_json(
+ 'https://cloud-api.yandex.net/v1/disk/public/resources',
+ video_id, query={'public_key': url})
+ except ExtractorError as e:
+ if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403:
+ error_description = self._parse_json(
+ e.cause.read().decode(), video_id)['description']
+ raise ExtractorError(error_description, expected=True)
+ raise
+
+ title = resource['name']
+ public_url = resource.get('public_url')
+ if public_url:
+ video_id = self._match_id(public_url)
+
+ self._set_cookie('yadi.sk', 'yandexuid', '0')
+
+ def call_api(action):
+ return (self._download_json(
+ urljoin(url, '/public/api/') + action, video_id, data=json.dumps({
+ 'hash': url,
+ # obtain sk if needed from call_api('check-auth') while
+ # the yandexuid cookie is set and sending an empty JSON object
+ 'sk': 'ya6b52f8c6b12abe91a66d22d3a31084b'
+ }).encode(), headers={
+ 'Content-Type': 'text/plain',
+ }, fatal=False) or {}).get('data') or {}
formats = []
+ source_url = resource.get('file')
+ if not source_url:
+ source_url = call_api('download-url').get('url')
if source_url:
formats.append({
'url': source_url,
'format_id': 'source',
- 'ext': determine_ext(title, 'mp4'),
+ 'ext': determine_ext(title, mimetype2ext(resource.get('mime_type')) or 'mp4'),
'quality': 1,
+ 'filesize': int_or_none(resource.get('size'))
})
- for video in videos:
+
+ video_streams = call_api('get-video-streams')
+ for video in (video_streams.get('videos') or []):
format_url = video.get('url')
if not format_url:
continue
- if determine_ext(format_url) == 'm3u8':
+ if video.get('dimension') == 'adaptive':
formats.extend(self._extract_m3u8_formats(
- format_url, video_id, 'mp4', entry_protocol='m3u8_native',
+ format_url, video_id, 'mp4', 'm3u8_native',
m3u8_id='hls', fatal=False))
else:
+ size = video.get('size') or {}
+ height = int_or_none(size.get('height'))
+ format_id = 'hls'
+ if height:
+ format_id += '-%dp' % height
formats.append({
+ 'ext': 'mp4',
+ 'format_id': format_id,
+ 'height': height,
+ 'protocol': 'm3u8_native',
'url': format_url,
+ 'width': int_or_none(size.get('width')),
})
self._sort_formats(formats)
- duration = float_or_none(try_get(
- models, lambda x: x[0]['data']['duration']), 1000)
- uploader = try_get(
- data, lambda x: x['user']['display_name'], compat_str)
- uploader_id = try_get(
- data, lambda x: x['user']['uid'], compat_str)
- view_count = int_or_none(try_get(
- data, lambda x: x['meta']['views_counter']))
+ owner = resource.get('owner') or {}
return {
'id': video_id,
'title': title,
- 'duration': duration,
- 'uploader': uploader,
- 'uploader_id': uploader_id,
- 'view_count': view_count,
+ 'duration': float_or_none(video_streams.get('duration'), 1000),
+ 'uploader': owner.get('display_name'),
+ 'uploader_id': owner.get('uid'),
+ 'view_count': int_or_none(resource.get('views_count')),
+ 'timestamp': parse_iso8601(resource.get('created')),
'formats': formats,
}