mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-10-31 23:02:40 +00:00
149 lines
5.8 KiB
Python
149 lines
5.8 KiB
Python
|
# coding: utf-8
|
|||
|
from __future__ import unicode_literals
|
|||
|
|
|||
|
from .common import InfoExtractor
|
|||
|
from ..compat import compat_str
|
|||
|
from ..utils import (
|
|||
|
int_or_none,
|
|||
|
str_or_none,
|
|||
|
try_get,
|
|||
|
unified_timestamp,
|
|||
|
update_url_query,
|
|||
|
urljoin,
|
|||
|
)
|
|||
|
|
|||
|
# compat_range
|
|||
|
try:
|
|||
|
if callable(xrange):
|
|||
|
range = xrange
|
|||
|
except (NameError, TypeError):
|
|||
|
pass
|
|||
|
|
|||
|
|
|||
|
class CPACIE(InfoExtractor):
|
|||
|
IE_NAME = 'cpac'
|
|||
|
_VALID_URL = r'https?://(?:www\.)?cpac\.ca/(?P<fr>l-)?episode\?id=(?P<id>[\da-f]{8}(?:-[\da-f]{4}){3}-[\da-f]{12})'
|
|||
|
_TEST = {
|
|||
|
# 'url': 'http://www.cpac.ca/en/programs/primetime-politics/episodes/65490909',
|
|||
|
'url': 'https://www.cpac.ca/episode?id=fc7edcae-4660-47e1-ba61-5b7f29a9db0f',
|
|||
|
'md5': 'e46ad699caafd7aa6024279f2614e8fa',
|
|||
|
'info_dict': {
|
|||
|
'id': 'fc7edcae-4660-47e1-ba61-5b7f29a9db0f',
|
|||
|
'ext': 'mp4',
|
|||
|
'upload_date': '20220215',
|
|||
|
'title': 'News Conference to Celebrate National Kindness Week – February 15, 2022',
|
|||
|
'description': 'md5:466a206abd21f3a6f776cdef290c23fb',
|
|||
|
'timestamp': 1644901200,
|
|||
|
},
|
|||
|
'params': {
|
|||
|
'format': 'bestvideo',
|
|||
|
'hls_prefer_native': True,
|
|||
|
},
|
|||
|
}
|
|||
|
|
|||
|
def _real_extract(self, url):
|
|||
|
video_id = self._match_id(url)
|
|||
|
url_lang = 'fr' if '/l-episode?' in url else 'en'
|
|||
|
|
|||
|
content = self._download_json(
|
|||
|
'https://www.cpac.ca/api/1/services/contentModel.json?url=/site/website/episode/index.xml&crafterSite=cpacca&id=' + video_id,
|
|||
|
video_id)
|
|||
|
video_url = try_get(content, lambda x: x['page']['details']['videoUrl'], compat_str)
|
|||
|
formats = []
|
|||
|
if video_url:
|
|||
|
content = content['page']
|
|||
|
title = str_or_none(content['details']['title_%s_t' % (url_lang, )])
|
|||
|
formats = self._extract_m3u8_formats(video_url, video_id, m3u8_id='hls', ext='mp4')
|
|||
|
for fmt in formats:
|
|||
|
# prefer language to match URL
|
|||
|
fmt_lang = fmt.get('language')
|
|||
|
if fmt_lang == url_lang:
|
|||
|
fmt['language_preference'] = 10
|
|||
|
elif not fmt_lang:
|
|||
|
fmt['language_preference'] = -1
|
|||
|
else:
|
|||
|
fmt['language_preference'] = -10
|
|||
|
|
|||
|
self._sort_formats(formats)
|
|||
|
|
|||
|
category = str_or_none(content['details']['category_%s_t' % (url_lang, )])
|
|||
|
|
|||
|
def is_live(v_type):
|
|||
|
return (v_type == 'live') if v_type is not None else None
|
|||
|
|
|||
|
return {
|
|||
|
'id': video_id,
|
|||
|
'formats': formats,
|
|||
|
'title': title,
|
|||
|
'description': str_or_none(content['details'].get('description_%s_t' % (url_lang, ))),
|
|||
|
'timestamp': unified_timestamp(content['details'].get('liveDateTime')),
|
|||
|
'category': [category] if category else None,
|
|||
|
'thumbnail': urljoin(url, str_or_none(content['details'].get('image_%s_s' % (url_lang, )))),
|
|||
|
'is_live': is_live(content['details'].get('type')),
|
|||
|
}
|
|||
|
|
|||
|
|
|||
|
class CPACPlaylistIE(InfoExtractor):
|
|||
|
IE_NAME = 'cpac:playlist'
|
|||
|
_VALID_URL = r'(?i)https?://(?:www\.)?cpac\.ca/(?:program|search|(?P<fr>emission|rechercher))\?(?:[^&]+&)*?(?P<id>(?:id=\d+|programId=\d+|key=[^&]+))'
|
|||
|
|
|||
|
_TESTS = [{
|
|||
|
'url': 'https://www.cpac.ca/program?id=6',
|
|||
|
'info_dict': {
|
|||
|
'id': 'id=6',
|
|||
|
'title': 'Headline Politics',
|
|||
|
'description': 'Watch CPAC’s signature long-form coverage of the day’s pressing political events as they unfold.',
|
|||
|
},
|
|||
|
'playlist_count': 10,
|
|||
|
}, {
|
|||
|
'url': 'https://www.cpac.ca/search?key=hudson&type=all&order=desc',
|
|||
|
'info_dict': {
|
|||
|
'id': 'key=hudson',
|
|||
|
'title': 'hudson',
|
|||
|
},
|
|||
|
'playlist_count': 22,
|
|||
|
}, {
|
|||
|
'url': 'https://www.cpac.ca/search?programId=50',
|
|||
|
'info_dict': {
|
|||
|
'id': 'programId=50',
|
|||
|
'title': '50',
|
|||
|
},
|
|||
|
'playlist_count': 9,
|
|||
|
}, {
|
|||
|
'url': 'https://www.cpac.ca/emission?id=6',
|
|||
|
'only_matching': True,
|
|||
|
}, {
|
|||
|
'url': 'https://www.cpac.ca/rechercher?key=hudson&type=all&order=desc',
|
|||
|
'only_matching': True,
|
|||
|
}]
|
|||
|
|
|||
|
def _real_extract(self, url):
|
|||
|
video_id = self._match_id(url)
|
|||
|
url_lang = 'fr' if any(x in url for x in ('/emission?', '/rechercher?')) else 'en'
|
|||
|
pl_type, list_type = ('program', 'itemList') if any(x in url for x in ('/program?', '/emission?')) else ('search', 'searchResult')
|
|||
|
api_url = (
|
|||
|
'https://www.cpac.ca/api/1/services/contentModel.json?url=/site/website/%s/index.xml&crafterSite=cpacca&%s'
|
|||
|
% (pl_type, video_id, ))
|
|||
|
content = self._download_json(api_url, video_id)
|
|||
|
entries = []
|
|||
|
total_pages = int_or_none(try_get(content, lambda x: x['page'][list_type]['totalPages']), default=1)
|
|||
|
for page in range(1, total_pages + 1):
|
|||
|
if page > 1:
|
|||
|
api_url = update_url_query(api_url, {'page': '%d' % (page, ), })
|
|||
|
content = self._download_json(
|
|||
|
api_url, video_id,
|
|||
|
note='Downloading continuation - %d' % (page, ),
|
|||
|
fatal=False)
|
|||
|
|
|||
|
for item in try_get(content, lambda x: x['page'][list_type]['item'], list) or []:
|
|||
|
episode_url = urljoin(url, try_get(item, lambda x: x['url_%s_s' % (url_lang, )]))
|
|||
|
if episode_url:
|
|||
|
entries.append(episode_url)
|
|||
|
|
|||
|
return self.playlist_result(
|
|||
|
(self.url_result(entry) for entry in entries),
|
|||
|
playlist_id=video_id,
|
|||
|
playlist_title=try_get(content, lambda x: x['page']['program']['title_%s_t' % (url_lang, )]) or video_id.split('=')[-1],
|
|||
|
playlist_description=try_get(content, lambda x: x['page']['program']['description_%s_t' % (url_lang, )]),
|
|||
|
)
|