Fix streamZ

This commit is contained in:
Alhaziel01
2020-11-07 18:48:15 +01:00
parent d90062f6d3
commit 19240c6ab4

View File

@@ -19,15 +19,17 @@ def test_video_exists(page_url):
def get_video_url(page_url, video_password):
logger.info("(page_url='%s')" % page_url)
video_urls = []
from core.support import match
matches = match(data, patron=r'(eval\(function\(p,a,c,k,e,d\).*?)\s+</script>').matches
unpacked = ''
for packed in matches:
unpacked += jsunpack.unpack(packed) + '\n'
packed = scrapertools.find_single_match(data, r'(eval\(function\(p,a,c,k,e,d\).*?)\s+</script>')
unpacked = jsunpack.unpack(packed)
urls = match(unpacked, patron=r"videojs\d+[^;]+[^']+'[^']+'[^']+'(https://streamz.*?/get.*?.dll)").matches
url = scrapertools.find_single_match(unpacked, '(https://streamz.*?/get.*?.dll)')
url = url.replace("getmp4", "getlink").replace("getIink", "getlink")
url += "|User-Agent=%s" % httptools.get_user_agent()
video_urls.append(["[streamZ]", url])
for url in urls:
url = url + "|User-Agent=%s" % httptools.get_user_agent()
if not video_urls or url not in video_urls[-1]:
video_urls.append(["[streamZ]", url])
return video_urls