1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-09 20:52:30 +01:00
yt-dlp/youtube_dl/extractor/gamersyde.py
snipem 3d24d997ae Fixed intendation of test cases
Leaded to error on Linux machine
2015-04-04 12:42:14 +02:00

94 lines
3.2 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
import time
from .common import InfoExtractor
class GamersydeIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?gamersyde\.com/hqstream_'
_TESTS = [{
'url': 'http://www.gamersyde.com/hqstream_bloodborne_birth_of_a_hero-34371_en.html',
'md5': 'f38d400d32f19724570040d5ce3a505f',
'info_dict': {
'id': '34371',
'ext': 'mp4',
'duration': 372,
'title': 'Bloodborne - Birth of a hero',
'thumbnail': 're:^https?://.*\.jpg$',
}
}, {
'url': 'http://www.gamersyde.com/hqstream_dark_souls_ii_scholar_of_the_first_sin_gameplay_part_1-34417_en.html',
'md5': '94bd7c3feff3275576cf5cb6c8a3a720',
'info_dict': {
'id': '34417',
'ext': 'mp4',
'duration': 270,
'title': 'Dark Souls II: Scholar of the First Sin - Gameplay - Part 1',
'thumbnail': 're:^https?://.*\.jpg$',
}
}, {
'url': 'http://www.gamersyde.com/hqstream_grand_theft_auto_v_heists_trailer-33786_en.html',
'md5': '65e442f5f340d571ece8c80d50700369',
'info_dict': {
'id': '33786',
'ext': 'mp4',
'duration': 59,
'title': 'Grand Theft Auto V - Heists Trailer',
'thumbnail': 're:^https?://.*\.jpg$',
}
}
]
def _calculateDuration(self, durationString):
if (durationString.find("minutes") > -1):
duration = time.strptime(durationString, "%M minutes %S seconds")
else:
duration = time.strptime(durationString, "%S seconds")
return duration.tm_min * 60 + duration.tm_sec
def _fixJsonSyntax(self, json):
json = re.sub(r",\s*}", "}", json, flags=re.DOTALL)
json = re.sub(r",\s*]", "]", json, flags=re.DOTALL)
json = json.replace('file: "', '"file": "')
json = json.replace('title: "', '"title": "')
json = json.replace('label: "', '"label": "')
json = json.replace('image: "', '"image": "')
json = json.replace('sources: [', '"sources": [')
return json
def _real_extract(self, url):
video_id = self._search_regex(r'-(.*?)_[a-z]{2}.html$', url, 'video_id')
webpage = self._download_webpage(url, video_id)
filesJson = self._search_regex(r'playlist: (.*?)\}\);', webpage, 'files', flags=re.DOTALL)
data = self._parse_json(filesJson,video_id, transform_source=self._fixJsonSyntax)
playlist = data[0]
formats = []
title = re.sub(r"[0-9]+ - ", "", playlist['title'])
length = self._search_regex(r'(([0-9]{1,2} minutes ){0,1}[0-9]{1,2} seconds)', webpage, 'length')
duration = self._calculateDuration(length)
for playlistEntry in playlist['sources']:
format = {
'url': playlistEntry['file'],
'format_id': playlistEntry['label']
}
formats.append(format)
return {
'id': video_id,
'title': title,
'formats': formats,
'duration': duration,
'thumbnail': playlist['image']
}