* Allow URLs that have a 'subdirectory' before the actual program name, e.g. 'xyz/xyz-episode-1'. * The author and broadcastDate fields in the XML file may be empty. * Add test case for the two problems above.
		
			
				
	
	
		
			100 lines
		
	
	
		
			4.1 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
			
		
		
	
	
			100 lines
		
	
	
		
			4.1 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
# coding: utf-8
 | 
						|
from __future__ import unicode_literals
 | 
						|
 | 
						|
import re
 | 
						|
 | 
						|
from .common import InfoExtractor
 | 
						|
from ..utils import ExtractorError
 | 
						|
 | 
						|
 | 
						|
class BRIE(InfoExtractor):
 | 
						|
    IE_DESC = "Bayerischer Rundfunk Mediathek"
 | 
						|
    _VALID_URL = r"^https?://(?:www\.)?br\.de/mediathek/video/(?:sendungen/)?(?:[a-z0-9\-/]+/)?(?P<id>[a-z0-9\-]+)\.html$"
 | 
						|
    _BASE_URL = "http://www.br.de"
 | 
						|
 | 
						|
    _TESTS = [
 | 
						|
        {
 | 
						|
            "url": "http://www.br.de/mediathek/video/anselm-gruen-114.html",
 | 
						|
            "md5": "c4f83cf0f023ba5875aba0bf46860df2",
 | 
						|
            "info_dict": {
 | 
						|
                "id": "2c8d81c5-6fb7-4a74-88d4-e768e5856532",
 | 
						|
                "ext": "mp4",
 | 
						|
                "title": "Feiern und Verzichten",
 | 
						|
                "description": "Anselm Grün: Feiern und Verzichten",
 | 
						|
                "uploader": "BR/Birgit Baier",
 | 
						|
                "upload_date": "20140301"
 | 
						|
            }
 | 
						|
        },
 | 
						|
        {
 | 
						|
            "url": "http://www.br.de/mediathek/video/sendungen/unter-unserem-himmel/unter-unserem-himmel-alpen-ueber-den-pass-100.html",
 | 
						|
            "md5": "ab451b09d861dbed7d7cc9ab0be19ebe",
 | 
						|
            "info_dict": {
 | 
						|
                "id": "2c060e69-3a27-4e13-b0f0-668fac17d812",
 | 
						|
                "ext": "mp4",
 | 
						|
                "title": "Über den Pass",
 | 
						|
                "description": "Die Eroberung der Alpen: Über den Pass",
 | 
						|
                "uploader": None,
 | 
						|
                "upload_date": None
 | 
						|
            }
 | 
						|
        }
 | 
						|
    ]
 | 
						|
 | 
						|
    def _real_extract(self, url):
 | 
						|
        mobj = re.match(self._VALID_URL, url)
 | 
						|
        display_id = mobj.group('id')
 | 
						|
        page = self._download_webpage(url, display_id)
 | 
						|
        xml_url = self._search_regex(
 | 
						|
            r"return BRavFramework\.register\(BRavFramework\('avPlayer_(?:[a-f0-9-]{36})'\)\.setup\({dataURL:'(/mediathek/video/[a-z0-9/~_.-]+)'}\)\);", page, "XMLURL")
 | 
						|
        xml = self._download_xml(self._BASE_URL + xml_url, None)
 | 
						|
 | 
						|
        videos = []
 | 
						|
        for xml_video in xml.findall("video"):
 | 
						|
            video = {
 | 
						|
                "id": xml_video.get("externalId"),
 | 
						|
                "title": xml_video.find("title").text,
 | 
						|
                "formats": self._extract_formats(xml_video.find("assets")),
 | 
						|
                "thumbnails": self._extract_thumbnails(xml_video.find("teaserImage/variants")),
 | 
						|
                "description": " ".join(xml_video.find("shareTitle").text.splitlines()),
 | 
						|
                "webpage_url": xml_video.find("permalink").text
 | 
						|
            }
 | 
						|
            if xml_video.find("author").text:
 | 
						|
                video["uploader"] = xml_video.find("author").text
 | 
						|
            if xml_video.find("broadcastDate").text:
 | 
						|
                video["upload_date"] =  "".join(reversed(xml_video.find("broadcastDate").text.split(".")))
 | 
						|
            videos.append(video)
 | 
						|
 | 
						|
        if len(videos) > 1:
 | 
						|
            self._downloader.report_warning(
 | 
						|
                'found multiple videos; please '
 | 
						|
                'report this with the video URL to http://yt-dl.org/bug')
 | 
						|
        if not videos:
 | 
						|
            raise ExtractorError('No video entries found')
 | 
						|
        return videos[0]
 | 
						|
 | 
						|
    def _extract_formats(self, assets):
 | 
						|
        formats = [{
 | 
						|
            "url": asset.find("downloadUrl").text,
 | 
						|
            "ext": asset.find("mediaType").text,
 | 
						|
            "format_id": asset.get("type"),
 | 
						|
            "width": int(asset.find("frameWidth").text),
 | 
						|
            "height": int(asset.find("frameHeight").text),
 | 
						|
            "tbr": int(asset.find("bitrateVideo").text),
 | 
						|
            "abr": int(asset.find("bitrateAudio").text),
 | 
						|
            "vcodec": asset.find("codecVideo").text,
 | 
						|
            "container": asset.find("mediaType").text,
 | 
						|
            "filesize": int(asset.find("size").text),
 | 
						|
        } for asset in assets.findall("asset")
 | 
						|
            if asset.find("downloadUrl") is not None]
 | 
						|
 | 
						|
        self._sort_formats(formats)
 | 
						|
        return formats
 | 
						|
 | 
						|
    def _extract_thumbnails(self, variants):
 | 
						|
        thumbnails = [{
 | 
						|
            "url": self._BASE_URL + variant.find("url").text,
 | 
						|
            "width": int(variant.find("width").text),
 | 
						|
            "height": int(variant.find("height").text),
 | 
						|
        } for variant in variants.findall("variant")]
 | 
						|
        thumbnails.sort(key=lambda x: x["width"] * x["height"], reverse=True)
 | 
						|
        return thumbnails
 |