1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-05 00:51:29 +00:00
youtube-dl/youtube_dl/extractor/gfycat.py
2015-07-07 14:22:13 +08:00

110 lines
3.5 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
int_or_none,
float_or_none,
qualities,
ExtractorError,
)
class GfycatIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?gfycat\.com/(?:ifr/)?(?P<id>[^/?#]+)'
_TESTS = [{
'url': 'http://gfycat.com/DeadlyDecisiveGermanpinscher',
'info_dict': {
'id': 'DeadlyDecisiveGermanpinscher',
'ext': 'mp4',
'title': 'Ghost in the Shell',
'timestamp': 1410656006,
'upload_date': '20140914',
'uploader': 'anonymous',
'duration': 10.4,
'view_count': int,
'like_count': int,
'dislike_count': int,
'categories': list,
'age_limit': 0,
}
}, {
'url': 'http://gfycat.com/ifr/JauntyTimelyAmazontreeboa',
'info_dict': {
'id': 'JauntyTimelyAmazontreeboa',
'ext': 'mp4',
'title': 'JauntyTimelyAmazontreeboa',
'timestamp': 1411720126,
'upload_date': '20140926',
'uploader': 'anonymous',
'duration': 3.52,
'view_count': int,
'like_count': int,
'dislike_count': int,
'categories': list,
'age_limit': 0,
}
}]
def _real_extract(self, url):
video_id = self._match_id(url)
gfy = self._download_json(
'http://gfycat.com/cajax/get/%s' % video_id,
video_id, 'Downloading video info')
if 'error' in gfy:
raise ExtractorError('Gfycat said: ' + gfy['error'], expected=True)
gfy = gfy['gfyItem']
title = gfy.get('title') or gfy['gfyName']
description = gfy.get('description')
timestamp = int_or_none(gfy.get('createDate'))
uploader = gfy.get('userName')
view_count = int_or_none(gfy.get('views'))
like_count = int_or_none(gfy.get('likes'))
dislike_count = int_or_none(gfy.get('dislikes'))
age_limit = 18 if gfy.get('nsfw') == '1' else 0
width = int_or_none(gfy.get('width'))
height = int_or_none(gfy.get('height'))
fps = int_or_none(gfy.get('frameRate'))
num_frames = int_or_none(gfy.get('numFrames'))
duration = float_or_none(num_frames, fps) if num_frames and fps else None
categories = gfy.get('tags') or gfy.get('extraLemmas') or []
FORMATS = ('gif', 'webm', 'mp4')
quality = qualities(FORMATS)
formats = []
for format_id in FORMATS:
video_url = gfy.get('%sUrl' % format_id)
if not video_url:
continue
filesize = gfy.get('%sSize' % format_id)
formats.append({
'url': video_url,
'format_id': format_id,
'width': width,
'height': height,
'fps': fps,
'filesize': filesize,
'quality': quality(format_id),
})
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'timestamp': timestamp,
'uploader': uploader,
'duration': duration,
'view_count': view_count,
'like_count': like_count,
'dislike_count': dislike_count,
'categories': categories,
'age_limit': age_limit,
'formats': formats,
}