parent
45e87ea106
commit
4cbfa570a1
3 changed files with 90 additions and 0 deletions
|
@ -284,6 +284,10 @@ from .camdemy import (
|
|||
CamdemyIE,
|
||||
CamdemyFolderIE
|
||||
)
|
||||
from .camfm import (
|
||||
CamFMEpisodeIE,
|
||||
CamFMShowIE
|
||||
)
|
||||
from .cammodels import CamModelsIE
|
||||
from .camsoda import CamsodaIE
|
||||
from .camtasia import CamtasiaEmbedIE
|
||||
|
|
85
yt_dlp/extractor/camfm.py
Normal file
85
yt_dlp/extractor/camfm.py
Normal file
|
@ -0,0 +1,85 @@
|
|||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
clean_html,
|
||||
get_element_by_class,
|
||||
get_elements_by_class,
|
||||
join_nonempty,
|
||||
traverse_obj,
|
||||
unified_timestamp,
|
||||
urljoin,
|
||||
)
|
||||
|
||||
|
||||
class CamFMShowIE(InfoExtractor):
|
||||
_VALID_URL = r'https://(?:www\.)?camfm\.co\.uk/shows/(?P<id>[^/]+)'
|
||||
_TESTS = [{
|
||||
'playlist_mincount': 5,
|
||||
'url': 'https://camfm.co.uk/shows/soul-mining/',
|
||||
'info_dict': {
|
||||
'id': 'soul-mining',
|
||||
'thumbnail': 'md5:6a873091f92c936f23bdcce80f75e66a',
|
||||
'title': 'Soul Mining',
|
||||
'description': 'Telling the stories of jazz, funk and soul from all corners of the world.',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
show_id = self._match_id(url)
|
||||
page = self._download_webpage(url, show_id)
|
||||
|
||||
return {
|
||||
'_type': 'playlist',
|
||||
'id': show_id,
|
||||
'entries': [self.url_result(urljoin('https://camfm.co.uk', i), CamFMEpisodeIE)
|
||||
for i in re.findall(r"javascript:popup\('(/player/[^']+)', 'listen'", page)],
|
||||
'thumbnail': urljoin('https://camfm.co.uk', self._search_regex(
|
||||
r'<img[^>]+class="thumb-expand"[^>]+src="([^"]+)"', page, 'thumbnail', fatal=False)),
|
||||
'title': self._html_search_regex('<h1>([^<]+)</h1>', page, 'title', fatal=False),
|
||||
'description': clean_html(get_element_by_class('small-12 medium-8 cell', page))
|
||||
}
|
||||
|
||||
|
||||
class CamFMEpisodeIE(InfoExtractor):
|
||||
_VALID_URL = r'https://(?:www\.)?camfm\.co\.uk/player/(?P<id>[^/]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://camfm.co.uk/player/43336',
|
||||
'skip': 'Episode will expire - don\'t actually know when, but it will go eventually',
|
||||
'info_dict': {
|
||||
'id': '43336',
|
||||
'title': 'AITAA: Am I the Agony Aunt? - 19:00 Tue 16/05/2023',
|
||||
'ext': 'mp3',
|
||||
'upload_date': '20230516',
|
||||
'description': 'md5:f165144f94927c0f1bfa2ee6e6ab7bbf',
|
||||
'timestamp': 1684263600,
|
||||
'series': 'AITAA: Am I the Agony Aunt?',
|
||||
'thumbnail': 'md5:5980a831360d0744c3764551be3d09c1',
|
||||
'categories': ['Entertainment'],
|
||||
}
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
episode_id = self._match_id(url)
|
||||
page = self._download_webpage(url, episode_id)
|
||||
audios = self._parse_html5_media_entries('https://audio.camfm.co.uk', page, episode_id)
|
||||
|
||||
caption = get_element_by_class('caption', page)
|
||||
series = clean_html(re.sub(r'<span[^<]+<[^<]+>', '', caption))
|
||||
|
||||
card_section = get_element_by_class('card-section', page)
|
||||
date = self._html_search_regex('>Aired at ([^<]+)<', card_section, 'air date', fatal=False)
|
||||
|
||||
return {
|
||||
'id': episode_id,
|
||||
'title': join_nonempty(series, date, delim=' - '),
|
||||
'formats': traverse_obj(audios, (..., 'formats', ...)),
|
||||
'timestamp': unified_timestamp(date), # XXX: Does not account for UK's daylight savings
|
||||
'series': series,
|
||||
'description': clean_html(re.sub(r'<b>[^<]+</b><br[^>]+/>', '', card_section)),
|
||||
'thumbnail': urljoin('https://camfm.co.uk', self._search_regex(
|
||||
r'<div[^>]+class="cover-art"[^>]+style="[^"]+url\(\'([^\']+)',
|
||||
page, 'thumbnail', fatal=False)),
|
||||
'categories': get_elements_by_class('label', caption),
|
||||
'was_live': True,
|
||||
}
|
|
@ -223,6 +223,7 @@ DATE_FORMATS_DAY_FIRST.extend([
|
|||
'%d/%m/%y',
|
||||
'%d/%m/%Y %H:%M:%S',
|
||||
'%d-%m-%Y %H:%M',
|
||||
'%H:%M %d/%m/%Y',
|
||||
])
|
||||
|
||||
DATE_FORMATS_MONTH_FIRST = list(DATE_FORMATS)
|
||||
|
|
Loading…
Reference in a new issue