1
0
Fork 0
mirror of https://gitlab.com/dstftw/youtube-dl.git synced 2020-11-16 09:42:26 +00:00

[nhl] Add new extractor(closes #8419)(closes #8798)

This commit is contained in:
remitamine 2016-04-22 20:17:20 +01:00
parent eb01e97e10
commit 29a7e8f6f8
2 changed files with 96 additions and 7 deletions

View file

@ -490,9 +490,10 @@ from .nextmovie import NextMovieIE
from .nfb import NFBIE from .nfb import NFBIE
from .nfl import NFLIE from .nfl import NFLIE
from .nhl import ( from .nhl import (
NHLIE,
NHLNewsIE,
NHLVideocenterIE, NHLVideocenterIE,
NHLNewsIE,
NHLVideocenterCategoryIE,
NHLIE,
) )
from .nick import NickIE from .nick import NickIE
from .niconico import NiconicoIE, NiconicoPlaylistIE from .niconico import NiconicoIE, NiconicoPlaylistIE

View file

@ -8,10 +8,15 @@ from .common import InfoExtractor
from ..compat import ( from ..compat import (
compat_urlparse, compat_urlparse,
compat_urllib_parse_urlencode, compat_urllib_parse_urlencode,
compat_urllib_parse_urlparse compat_urllib_parse_urlparse,
compat_str,
) )
from ..utils import ( from ..utils import (
unified_strdate, unified_strdate,
determine_ext,
int_or_none,
parse_iso8601,
parse_duration,
) )
@ -70,8 +75,8 @@ class NHLBaseInfoExtractor(InfoExtractor):
return ret return ret
class NHLIE(NHLBaseInfoExtractor): class NHLVideocenterIE(NHLBaseInfoExtractor):
IE_NAME = 'nhl.com' IE_NAME = 'nhl.com:videocenter'
_VALID_URL = r'https?://video(?P<team>\.[^.]*)?\.nhl\.com/videocenter/(?:console|embed)?(?:\?(?:.*?[?&])?)(?:id|hlg|playlist)=(?P<id>[-0-9a-zA-Z,]+)' _VALID_URL = r'https?://video(?P<team>\.[^.]*)?\.nhl\.com/videocenter/(?:console|embed)?(?:\?(?:.*?[?&])?)(?:id|hlg|playlist)=(?P<id>[-0-9a-zA-Z,]+)'
_TESTS = [{ _TESTS = [{
@ -186,8 +191,8 @@ class NHLNewsIE(NHLBaseInfoExtractor):
return self._real_extract_video(video_id) return self._real_extract_video(video_id)
class NHLVideocenterIE(NHLBaseInfoExtractor): class NHLVideocenterCategoryIE(NHLBaseInfoExtractor):
IE_NAME = 'nhl.com:videocenter' IE_NAME = 'nhl.com:videocenter:category'
IE_DESC = 'NHL videocenter category' IE_DESC = 'NHL videocenter category'
_VALID_URL = r'https?://video\.(?P<team>[^.]*)\.nhl\.com/videocenter/(console\?[^(id=)]*catid=(?P<catid>[0-9]+)(?![&?]id=).*?)?$' _VALID_URL = r'https?://video\.(?P<team>[^.]*)\.nhl\.com/videocenter/(console\?[^(id=)]*catid=(?P<catid>[0-9]+)(?![&?]id=).*?)?$'
_TEST = { _TEST = {
@ -236,3 +241,86 @@ class NHLVideocenterIE(NHLBaseInfoExtractor):
'id': cat_id, 'id': cat_id,
'entries': [self._extract_video(v) for v in videos], 'entries': [self._extract_video(v) for v in videos],
} }
class NHLIE(InfoExtractor):
IE_NAME = 'nhl.com'
_VALID_URL = r'https?://(?:www\.)?nhl\.com/([^/]+/)*c-(?P<id>\d+)'
_TESTS = [{
# type=video
'url': 'https://www.nhl.com/video/anisimov-cleans-up-mess/t-277752844/c-43663503',
'md5': '0f7b9a8f986fb4b4eeeece9a56416eaf',
'info_dict': {
'id': '43663503',
'ext': 'mp4',
'title': 'Anisimov cleans up mess',
'description': 'md5:a02354acdfe900e940ce40706939ca63',
'timestamp': 1461288600,
'upload_date': '20160422',
},
}, {
# type=article
'url': 'https://www.nhl.com/news/dennis-wideman-suspended/c-278258934',
'md5': '1f39f4ea74c1394dea110699a25b366c',
'info_dict': {
'id': '40784403',
'ext': 'mp4',
'title': 'Wideman suspended by NHL',
'description': 'Flames defenseman Dennis Wideman was banned 20 games for violation of Rule 40 (Physical Abuse of Officials)',
'upload_date': '20160204',
'timestamp': 1454544904,
},
}]
def _real_extract(self, url):
tmp_id = self._match_id(url)
video_data = self._download_json(
'https://nhl.bamcontent.com/nhl/id/v1/%s/details/web-v1.json' % tmp_id,
tmp_id)
if video_data.get('type') == 'article':
video_data = video_data['media']
video_id = compat_str(video_data['id'])
title = video_data['title']
formats = []
for playback in video_data.get('playbacks', []):
playback_url = playback.get('url')
if not playback_url:
continue
ext = determine_ext(playback_url)
if ext == 'm3u8':
formats.extend(self._extract_m3u8_formats(
playback_url, video_id, 'mp4', 'm3u8_native',
m3u8_id=playback.get('name', 'hls'), fatal=False))
else:
height = int_or_none(playback.get('height'))
formats.append({
'format_id': playback.get('name', 'http' + ('-%dp' % height if height else '')),
'url': playback_url,
'width': int_or_none(playback.get('width')),
'height': height,
})
self._sort_formats(formats, ('preference', 'width', 'height', 'tbr', 'format_id'))
thumbnails = []
for thumbnail_id, thumbnail_data in video_data.get('image', {}).get('cuts', {}).items():
thumbnail_url = thumbnail_data.get('src')
if not thumbnail_url:
continue
thumbnails.append({
'id': thumbnail_id,
'url': thumbnail_url,
'width': int_or_none(thumbnail_data.get('width')),
'height': int_or_none(thumbnail_data.get('height')),
})
return {
'id': video_id,
'title': title,
'description': video_data.get('description'),
'timestamp': parse_iso8601(video_data.get('date')),
'duration': parse_duration(video_data.get('duration')),
'thumbnails': thumbnails,
'formats': formats,
}