mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-22 12:10:10 +09:00
[samplefocus] Add new extractor(closes #27763)
This commit is contained in:
parent
cf2dbec630
commit
21e872b19a
@ -1029,6 +1029,7 @@ from .safari import (
|
||||
SafariApiIE,
|
||||
SafariCourseIE,
|
||||
)
|
||||
from .samplefocus import SampleFocusIE
|
||||
from .sapo import SapoIE
|
||||
from .savefrom import SaveFromIE
|
||||
from .sbs import SBSIE
|
||||
|
100
youtube_dl/extractor/samplefocus.py
Normal file
100
youtube_dl/extractor/samplefocus.py
Normal file
@ -0,0 +1,100 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
extract_attributes,
|
||||
get_element_by_attribute,
|
||||
int_or_none,
|
||||
)
|
||||
|
||||
|
||||
class SampleFocusIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?samplefocus\.com/samples/(?P<id>[^/?&#]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://samplefocus.com/samples/lil-peep-sad-emo-guitar',
|
||||
'md5': '48c8d62d60be467293912e0e619a5120',
|
||||
'info_dict': {
|
||||
'id': '40316',
|
||||
'display_id': 'lil-peep-sad-emo-guitar',
|
||||
'ext': 'mp3',
|
||||
'title': 'Lil Peep Sad Emo Guitar',
|
||||
'thumbnail': r're:^https?://.+\.png',
|
||||
'license': 'Standard License',
|
||||
'uploader': 'CapsCtrl',
|
||||
'uploader_id': 'capsctrl',
|
||||
'like_count': int,
|
||||
'comment_count': int,
|
||||
'categories': ['Samples', 'Guitar', 'Electric guitar'],
|
||||
},
|
||||
}, {
|
||||
'url': 'https://samplefocus.com/samples/dababy-style-bass-808',
|
||||
'only_matching': True
|
||||
}, {
|
||||
'url': 'https://samplefocus.com/samples/young-chop-kick',
|
||||
'only_matching': True
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
display_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, display_id)
|
||||
|
||||
sample_id = self._search_regex(
|
||||
r'<input[^>]+id=(["\'])sample_id\1[^>]+value=(?:["\'])(?P<id>\d+)',
|
||||
webpage, 'sample id', group='id')
|
||||
|
||||
title = self._og_search_title(webpage, fatal=False) or self._html_search_regex(
|
||||
r'<h1>(.+?)</h1>', webpage, 'title')
|
||||
|
||||
mp3_url = self._search_regex(
|
||||
r'<input[^>]+id=(["\'])sample_mp3\1[^>]+value=(["\'])(?P<url>(?:(?!\2).)+)',
|
||||
webpage, 'mp3', fatal=False, group='url') or extract_attributes(self._search_regex(
|
||||
r'<meta[^>]+itemprop=(["\'])contentUrl\1[^>]*>',
|
||||
webpage, 'mp3 url', group=0))['content']
|
||||
|
||||
thumbnail = self._og_search_thumbnail(webpage) or self._html_search_regex(
|
||||
r'<img[^>]+class=(?:["\'])waveform responsive-img[^>]+src=(["\'])(?P<url>(?:(?!\1).)+)',
|
||||
webpage, 'mp3', fatal=False, group='url')
|
||||
|
||||
comments = []
|
||||
for author_id, author, body in re.findall(r'(?s)<p[^>]+class="comment-author"><a[^>]+href="/users/([^"]+)">([^"]+)</a>.+?<p[^>]+class="comment-body">([^>]+)</p>', webpage):
|
||||
comments.append({
|
||||
'author': author,
|
||||
'author_id': author_id,
|
||||
'text': body,
|
||||
})
|
||||
|
||||
uploader_id = uploader = None
|
||||
mobj = re.search(r'>By <a[^>]+href="/users/([^"]+)"[^>]*>([^<]+)', webpage)
|
||||
if mobj:
|
||||
uploader_id, uploader = mobj.groups()
|
||||
|
||||
breadcrumb = get_element_by_attribute('typeof', 'BreadcrumbList', webpage)
|
||||
categories = []
|
||||
if breadcrumb:
|
||||
for _, name in re.findall(r'<span[^>]+property=(["\'])name\1[^>]*>([^<]+)', breadcrumb):
|
||||
categories.append(name)
|
||||
|
||||
def extract_count(klass):
|
||||
return int_or_none(self._html_search_regex(
|
||||
r'<span[^>]+class=(?:["\'])?%s-count[^>]*>(\d+)' % klass,
|
||||
webpage, klass, fatal=False))
|
||||
|
||||
return {
|
||||
'id': sample_id,
|
||||
'title': title,
|
||||
'url': mp3_url,
|
||||
'display_id': display_id,
|
||||
'thumbnail': thumbnail,
|
||||
'uploader': uploader,
|
||||
'license': self._html_search_regex(
|
||||
r'<a[^>]+href=(["\'])/license\1[^>]*>(?P<license>[^<]+)<',
|
||||
webpage, 'license', fatal=False, group='license'),
|
||||
'uploader_id': uploader_id,
|
||||
'like_count': extract_count('sample-%s-favorites' % sample_id),
|
||||
'comment_count': extract_count('comments'),
|
||||
'comments': comments,
|
||||
'categories': categories,
|
||||
}
|
Loading…
Reference in New Issue
Block a user