[americastestkitchen] fix extraction

This commit is contained in:
Remita Amine 2020-01-15 14:16:58 +01:00
parent a9866c0366
commit 2c482bff7c
1 changed files with 16 additions and 26 deletions

View File

@ -5,6 +5,7 @@ from .common import InfoExtractor
from ..utils import ( from ..utils import (
clean_html, clean_html,
int_or_none, int_or_none,
js_to_json,
try_get, try_get,
unified_strdate, unified_strdate,
) )
@ -13,22 +14,21 @@ from ..utils import (
class AmericasTestKitchenIE(InfoExtractor): class AmericasTestKitchenIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?americastestkitchen\.com/(?:episode|videos)/(?P<id>\d+)' _VALID_URL = r'https?://(?:www\.)?americastestkitchen\.com/(?:episode|videos)/(?P<id>\d+)'
_TESTS = [{ _TESTS = [{
'url': 'https://www.americastestkitchen.com/episode/548-summer-dinner-party', 'url': 'https://www.americastestkitchen.com/episode/582-weeknight-japanese-suppers',
'md5': 'b861c3e365ac38ad319cfd509c30577f', 'md5': 'b861c3e365ac38ad319cfd509c30577f',
'info_dict': { 'info_dict': {
'id': '1_5g5zua6e', 'id': '5b400b9ee338f922cb06450c',
'title': 'Summer Dinner Party', 'title': 'Weeknight Japanese Suppers',
'ext': 'mp4', 'ext': 'mp4',
'description': 'md5:858d986e73a4826979b6a5d9f8f6a1ec', 'description': 'md5:3d0c1a44bb3b27607ce82652db25b4a8',
'thumbnail': r're:^https?://.*\.jpg', 'thumbnail': r're:^https?://',
'timestamp': 1497285541, 'timestamp': 1523664000,
'upload_date': '20170612', 'upload_date': '20180414',
'uploader_id': 'roger.metcalf@americastestkitchen.com', 'release_date': '20180414',
'release_date': '20170617',
'series': "America's Test Kitchen", 'series': "America's Test Kitchen",
'season_number': 17, 'season_number': 18,
'episode': 'Summer Dinner Party', 'episode': 'Weeknight Japanese Suppers',
'episode_number': 24, 'episode_number': 15,
}, },
'params': { 'params': {
'skip_download': True, 'skip_download': True,
@ -47,7 +47,7 @@ class AmericasTestKitchenIE(InfoExtractor):
self._search_regex( self._search_regex(
r'window\.__INITIAL_STATE__\s*=\s*({.+?})\s*;\s*</script>', r'window\.__INITIAL_STATE__\s*=\s*({.+?})\s*;\s*</script>',
webpage, 'initial context'), webpage, 'initial context'),
video_id) video_id, js_to_json)
ep_data = try_get( ep_data = try_get(
video_data, video_data,
@ -55,17 +55,7 @@ class AmericasTestKitchenIE(InfoExtractor):
lambda x: x['videoDetail']['content']['data']), dict) lambda x: x['videoDetail']['content']['data']), dict)
ep_meta = ep_data.get('full_video', {}) ep_meta = ep_data.get('full_video', {})
zype_id = ep_meta.get('zype_id') zype_id = ep_data.get('zype_id') or ep_meta['zype_id']
if zype_id:
embed_url = 'https://player.zype.com/embed/%s.js?api_key=jZ9GUhRmxcPvX7M3SlfejB6Hle9jyHTdk2jVxG7wOHPLODgncEKVdPYBhuz9iWXQ' % zype_id
ie_key = 'Zype'
else:
partner_id = self._search_regex(
r'src=["\'](?:https?:)?//(?:[^/]+\.)kaltura\.com/(?:[^/]+/)*(?:p|partner_id)/(\d+)',
webpage, 'kaltura partner id')
external_id = ep_data.get('external_id') or ep_meta['external_id']
embed_url = 'kaltura:%s:%s' % (partner_id, external_id)
ie_key = 'Kaltura'
title = ep_data.get('title') or ep_meta.get('title') title = ep_data.get('title') or ep_meta.get('title')
description = clean_html(ep_meta.get('episode_description') or ep_data.get( description = clean_html(ep_meta.get('episode_description') or ep_data.get(
@ -79,8 +69,8 @@ class AmericasTestKitchenIE(InfoExtractor):
return { return {
'_type': 'url_transparent', '_type': 'url_transparent',
'url': embed_url, 'url': 'https://player.zype.com/embed/%s.js?api_key=jZ9GUhRmxcPvX7M3SlfejB6Hle9jyHTdk2jVxG7wOHPLODgncEKVdPYBhuz9iWXQ' % zype_id,
'ie_key': ie_key, 'ie_key': 'Zype',
'title': title, 'title': title,
'description': description, 'description': description,
'thumbnail': thumbnail, 'thumbnail': thumbnail,