yt-dlp/yt_dlp/extractor/googlesearch.py
2022-02-01 00:12:34 +05:30

41 lines
1.2 KiB
Python

from __future__ import unicode_literals
import itertools
import re
from .common import SearchInfoExtractor
class GoogleSearchIE(SearchInfoExtractor):
IE_DESC = 'Google Video search'
IE_NAME = 'video.google:search'
_SEARCH_KEY = 'gvsearch'
_TESTS = [{
'url': 'gvsearch15:python language',
'info_dict': {
'id': 'python language',
'title': 'python language',
},
'playlist_count': 15,
}]
_PAGE_SIZE = 100
def _search_results(self, query):
for pagenum in itertools.count():
webpage = self._download_webpage(
'http://www.google.com/search', f'gvsearch:{query}',
note=f'Downloading result page {pagenum + 1}',
query={
'tbm': 'vid',
'q': query,
'start': pagenum * self._PAGE_SIZE,
'num': self._PAGE_SIZE,
'hl': 'en',
})
for url in re.findall(r'<div[^>]* class="dXiKIc"[^>]*><a href="([^"]+)"', webpage):
yield self.url_result(url)
if not re.search(r'id="pnnext"', webpage):
return