mirror of https://github.com/yt-dlp/yt-dlp.git
[animalplanet] Add new extractor(closes #5303)
This commit is contained in:
parent
9787c5f4c8
commit
80b8b72cb8
|
@ -19,6 +19,7 @@ from .aftonbladet import AftonbladetIE
|
||||||
from .airmozilla import AirMozillaIE
|
from .airmozilla import AirMozillaIE
|
||||||
from .aljazeera import AlJazeeraIE
|
from .aljazeera import AlJazeeraIE
|
||||||
from .alphaporno import AlphaPornoIE
|
from .alphaporno import AlphaPornoIE
|
||||||
|
from .animalplanet import AnimalPlanetIE
|
||||||
from .anitube import AnitubeIE
|
from .anitube import AnitubeIE
|
||||||
from .anysex import AnySexIE
|
from .anysex import AnySexIE
|
||||||
from .aol import AolIE
|
from .aol import AolIE
|
||||||
|
|
|
@ -0,0 +1,53 @@
|
||||||
|
# coding: utf-8
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..compat import compat_str
|
||||||
|
from ..utils import (
|
||||||
|
parse_duration,
|
||||||
|
parse_iso8601,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
|
class AnimalPlanetIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?animalplanet\.com/([^/]+/)*(?P<id>[^/\?#]+)'
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'http://www.animalplanet.com/tv-shows/i-shouldnt-be-alive/videos/dog-saves-injured-owner/',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '10608',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Dog Saves Injured Owner',
|
||||||
|
'description': 'A world class athlete is put to the test when she falls into a canyon and breaks her hip. Her only companion is her dog, Taz, who is on a mission to save her!',
|
||||||
|
'upload_date': '20100410',
|
||||||
|
'timestamp': 1270857727,
|
||||||
|
'duration': 220,
|
||||||
|
},
|
||||||
|
'params': {
|
||||||
|
# m3u8 download
|
||||||
|
'skip_download': True,
|
||||||
|
}
|
||||||
|
}, {
|
||||||
|
'url': 'http://www.animalplanet.com/longfin-eels-maneaters/',
|
||||||
|
'only_matching': True,
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
display_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, display_id)
|
||||||
|
|
||||||
|
video_data = self._parse_json(self._search_regex(
|
||||||
|
r'initialVideoData\s*=\s*({.+?});',
|
||||||
|
webpage, 'initialVideoData'), display_id)['playlist'][0]
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': compat_str(video_data['id']),
|
||||||
|
'display_id': display_id,
|
||||||
|
'title': video_data['title'],
|
||||||
|
'description': video_data.get('description'),
|
||||||
|
'thumbnail': video_data.get('thumbnailURL'),
|
||||||
|
'duration': parse_duration(video_data.get('video_length')),
|
||||||
|
'timestamp': parse_iso8601(video_data.get('publishedDate')),
|
||||||
|
'formats': self._extract_m3u8_formats(
|
||||||
|
video_data['src'], display_id, 'mp4',
|
||||||
|
'm3u8_native', m3u8_id='hls')
|
||||||
|
}
|
Loading…
Reference in New Issue