1
0
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-12-21 00:46:41 +00:00
youtube-dl/youtube_dl/extractor/gaskrank.py

124 lines
4.7 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
float_or_none,
int_or_none,
js_to_json,
unified_strdate,
)
class GaskrankIE(InfoExtractor):
"""InfoExtractor for gaskrank.tv"""
_VALID_URL = r'https?://(?:www\.)?gaskrank\.tv/tv/(?P<categories>[^/]+)/(?P<id>[^/]+)\.html?'
_TESTS = [
{
'url': 'http://www.gaskrank.tv/tv/motorrad-fun/strike-einparken-durch-anfaenger-crash-mit-groesserem-flurschaden.htm',
'md5': '1ae88dbac97887d85ebd1157a95fc4f9',
'info_dict': {
'id': '201601/26955',
'ext': 'mp4',
'title': 'Strike! Einparken können nur Männer - Flurschaden hält sich in Grenzen *lol*',
'thumbnail': r're:^https?://.*\.jpg$',
'categories': ['motorrad-fun'],
'display_id': 'strike-einparken-durch-anfaenger-crash-mit-groesserem-flurschaden',
'uploader_id': 'Bikefun',
'upload_date': '20170110',
'uploader_url': None,
}
},
{
'url': 'http://www.gaskrank.tv/tv/racing/isle-of-man-tt-2011-michael-du-15920.htm',
'md5': 'c33ee32c711bc6c8224bfcbe62b23095',
'info_dict': {
'id': '201106/15920',
'ext': 'mp4',
'title': 'Isle of Man - Michael Dunlop vs Guy Martin - schwindelig kucken',
'thumbnail': r're:^https?://.*\.jpg$',
'categories': ['racing'],
'display_id': 'isle-of-man-tt-2011-michael-du-15920',
'uploader_id': 'IOM',
'upload_date': '20160506',
'uploader_url': 'www.iomtt.com',
}
}
]
def _real_extract(self, url):
"""extract information from gaskrank.tv"""
def fix_json(code):
"""Removes trailing comma in json: {{},} --> {{}}"""
return re.sub(r',\s*}', r'}', js_to_json(code))
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
categories = [re.match(self._VALID_URL, url).group('categories')]
title = self._search_regex(
r'movieName\s*:\s*\'([^\']*)\'',
webpage, 'title')
thumbnail = self._search_regex(
r'poster\s*:\s*\'([^\']*)\'',
webpage, 'thumbnail', default=None)
mobj = re.search(
r'Video von:\s*(?P<uploader_id>[^|]*?)\s*\|\s*vom:\s*(?P<upload_date>[0-9][0-9]\.[0-9][0-9]\.[0-9][0-9][0-9][0-9])',
webpage)
if mobj is not None:
uploader_id = mobj.groupdict().get('uploader_id')
upload_date = unified_strdate(mobj.groupdict().get('upload_date'))
uploader_url = self._search_regex(
r'Homepage:\s*<[^>]*>(?P<uploader_url>[^<]*)',
webpage, 'uploader_url', default=None)
tags = re.findall(
r'/tv/tags/[^/]+/"\s*>(?P<tag>[^<]*?)<',
webpage)
view_count = self._search_regex(
r'class\s*=\s*"gkRight"(?:[^>]*>\s*<[^>]*)*icon-eye-open(?:[^>]*>\s*<[^>]*)*>\s*(?P<view_count>[0-9\.]*)',
webpage, 'view_count', default=None)
if view_count:
view_count = int_or_none(view_count.replace('.', ''))
average_rating = self._search_regex(
r'itemprop\s*=\s*"ratingValue"[^>]*>\s*(?P<average_rating>[0-9,]+)',
webpage, 'average_rating')
if average_rating:
average_rating = float_or_none(average_rating.replace(',', '.'))
playlist = self._parse_json(
self._search_regex(
r'playlist\s*:\s*\[([^\]]*)\]',
webpage, 'playlist', default='{}'),
display_id, transform_source=fix_json, fatal=False)
video_id = self._search_regex(
r'https?://movies\.gaskrank\.tv/([^-]*?)(-[^\.]*)?\.mp4',
playlist.get('0').get('src'), 'video id')
formats = []
for key in playlist:
formats.append({
'url': playlist[key]['src'],
'format_id': key,
'quality': playlist[key].get('quality')})
self._sort_formats(formats, field_preference=['format_id'])
return {
'id': video_id,
'title': title,
'formats': formats,
'thumbnail': thumbnail,
'categories': categories,
'display_id': display_id,
'uploader_id': uploader_id,
'upload_date': upload_date,
'uploader_url': uploader_url,
'tags': tags,
'view_count': view_count,
'average_rating': average_rating,
}