mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-14 15:22:35 +00:00
[mdr] Simplify
This commit is contained in:
parent
b2ae513586
commit
09dacfa57f
1 changed files with 47 additions and 46 deletions
|
@ -5,35 +5,28 @@ from ..utils import (
|
||||||
ExtractorError,
|
ExtractorError,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
class MDRIE(InfoExtractor):
|
class MDRIE(InfoExtractor):
|
||||||
_VALID_URL = r'^(?P<domain>(?:https?://)?(?:www\.)?mdr\.de)/mediathek/(?:.*)/(?P<type>video|audio)(?P<video_id>[^/_]+)_.*'
|
_VALID_URL = r'^(?P<domain>(?:https?://)?(?:www\.)?mdr\.de)/mediathek/(?:.*)/(?P<type>video|audio)(?P<video_id>[^/_]+)_.*'
|
||||||
_TITLE = r'<h2>(?P<title1>[^<]+)<span>(?P<title2>[^<]+)</span></h2>'
|
|
||||||
|
|
||||||
_MEDIA_XML = r'(?P<xmlurl>/mediathek/(.+)/(video|audio)([0-9]+)-avCustom.xml)'
|
|
||||||
_MEDIA_STREAM_VIDEO = r'<asset>.*<frameWidth>(?P<frameWidth>[0-9]+)</frameWidth>.*<flashMediaServerApplicationURL>(?P<flashMediaServerApplicationURL>[^<]+)</flashMediaServerApplicationURL><flashMediaServerURL>(?P<flashMediaServerURL>[^<]+)</flashMediaServerURL>.*<progressiveDownloadUrl>(?P<progressiveDownloadUrl>[^<]+)</progressiveDownloadUrl></asset>'
|
|
||||||
_MEDIA_STREAM_AUDIO = r'<asset>.*<mediaType>(?P<mediaType>[A-Z0-9]+)</mediaType><bitrateAudio>(?P<bitrateAudio>[0-9]+)</bitrateAudio>.*<flashMediaServerApplicationURL>(?P<flashMediaServerApplicationURL>[^<]+)</flashMediaServerApplicationURL><flashMediaServerURL>(?P<flashMediaServerURL>[^<]+)</flashMediaServerURL>.*<progressiveDownloadUrl>(?P<progressiveDownloadUrl>[^<]+)</progressiveDownloadUrl></asset>'
|
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
u'url': u'http://www.mdr.de/mediathek/themen/nachrichten/video165624_zc-c5c7de76_zs-3795826d.html',
|
u'url': u'http://www.mdr.de/mediathek/themen/nachrichten/video165624_zc-c5c7de76_zs-3795826d.html',
|
||||||
u'file': u'165624.mp4',
|
u'file': u'165624.mp4',
|
||||||
u'md5': u'95165945756198b8fa2dea10f0b04614',
|
u'md5': u'ae785f36ecbf2f19b42edf1bc9c85815',
|
||||||
u'info_dict': {
|
u'info_dict': {
|
||||||
u"title": u"MDR aktuell Eins30 09.12.2013, 22:48 Uhr"
|
u"title": u"MDR aktuell Eins30 09.12.2013, 22:48 Uhr"
|
||||||
},
|
},
|
||||||
#u'skip': u'Requires rtmpdump' # rtmp is optional
|
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
u'url': u' http://www.mdr.de/mediathek/radio/mdr1-radio-sachsen/audio718370_zc-67b21197_zs-1b9b2483.html',
|
u'url': u'http://www.mdr.de/mediathek/radio/mdr1-radio-sachsen/audio718370_zc-67b21197_zs-1b9b2483.html',
|
||||||
u'file': u'718370.mp4',
|
u'file': u'718370.mp3',
|
||||||
u'md5': u'4a5b1fbb5519fb0d929c384b6ff7cb8b',
|
u'md5': u'a9d21345a234c7b45dee612f290fd8d7',
|
||||||
u'info_dict': {
|
u'info_dict': {
|
||||||
u"title": u"MDR 1 RADIO SACHSEN 10.12.2013, 05:00 Uhr"
|
u"title": u"MDR 1 RADIO SACHSEN 10.12.2013, 05:00 Uhr"
|
||||||
},
|
},
|
||||||
#u'skip': u'Requires rtmpdump' # rtmp is optional
|
|
||||||
}]
|
}]
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
|
|
||||||
# determine video id from url
|
|
||||||
m = re.match(self._VALID_URL, url)
|
m = re.match(self._VALID_URL, url)
|
||||||
video_id = m.group('video_id')
|
video_id = m.group('video_id')
|
||||||
domain = m.group('domain')
|
domain = m.group('domain')
|
||||||
|
@ -41,38 +34,46 @@ class MDRIE(InfoExtractor):
|
||||||
|
|
||||||
# determine title and media streams from webpage
|
# determine title and media streams from webpage
|
||||||
html = self._download_webpage(url, video_id)
|
html = self._download_webpage(url, video_id)
|
||||||
t = re.search(self._TITLE, html)
|
|
||||||
if not t:
|
|
||||||
raise ExtractorError(u'no title found')
|
|
||||||
title = t.group('title1') + t.group('title2')
|
|
||||||
m = re.search(self._MEDIA_XML, html)
|
|
||||||
if not m:
|
|
||||||
raise ExtractorError(u'no xml found')
|
|
||||||
xmlurl = m.group('xmlurl')
|
|
||||||
xml = self._download_webpage(domain+xmlurl, video_id, 'download XML').replace('\n','').replace('\r','').replace('<asset>','\n<asset>').replace('</asset>','</asset>\n')
|
|
||||||
if(mediatype == "video"):
|
|
||||||
streams = [mo.groupdict() for mo in re.finditer(self._MEDIA_STREAM_VIDEO, xml)]
|
|
||||||
if not streams:
|
|
||||||
raise ExtractorError(u'no media found')
|
|
||||||
# choose default media type and highest quality for now
|
|
||||||
stream = max([s for s in streams if s["progressiveDownloadUrl"].startswith("http://") ],
|
|
||||||
key=lambda s: int(s["frameWidth"]))
|
|
||||||
else:
|
|
||||||
streams = [mo.groupdict() for mo in re.finditer(self._MEDIA_STREAM_AUDIO, xml)]
|
|
||||||
if not streams:
|
|
||||||
raise ExtractorError(u'no media found')
|
|
||||||
# choose default media type (MP4) and highest quality for now
|
|
||||||
stream = max([s for s in streams if s["progressiveDownloadUrl"].startswith("http://") and s["mediaType"] == "MP4" ],
|
|
||||||
key=lambda s: int(s["bitrateAudio"]))
|
|
||||||
|
|
||||||
# there's two possibilities: RTMP stream or HTTP download
|
title = self._html_search_regex(r'<h2>(.*?)</h2>', html, u'title')
|
||||||
info = {'id': video_id, 'title': title, 'ext': 'mp4'}
|
xmlurl = self._search_regex(
|
||||||
if not stream["progressiveDownloadUrl"]:
|
r'(/mediathek/(?:.+)/(?:video|audio)[0-9]+-avCustom.xml)', html, u'XML URL')
|
||||||
self.to_screen(u'RTMP download detected')
|
|
||||||
assert stream['flashMediaServerURL'].startswith('mp4:')
|
doc = self._download_xml(domain + xmlurl, video_id)
|
||||||
info["url"] = stream["flashMediaServerApplicationURL"]
|
formats = []
|
||||||
info["play_path"] = stream['flashMediaServerURL']
|
for a in doc.findall('./assets/asset'):
|
||||||
else:
|
url_el = a.find('.//progressiveDownloadUrl')
|
||||||
assert stream["progressiveDownloadUrl"].endswith('.mp4')
|
if url_el is None:
|
||||||
info["url"] = stream["progressiveDownloadUrl"]
|
continue
|
||||||
return [info]
|
abr = int(a.find('bitrateAudio').text) // 1000
|
||||||
|
media_type = a.find('mediaType').text
|
||||||
|
format = {
|
||||||
|
'abr': abr,
|
||||||
|
'filesize': int(a.find('fileSize').text),
|
||||||
|
'url': url_el.text,
|
||||||
|
}
|
||||||
|
|
||||||
|
vbr_el = a.find('bitrateVideo')
|
||||||
|
if vbr_el is None:
|
||||||
|
format.update({
|
||||||
|
'vcodec': 'none',
|
||||||
|
'format_id': u'%s-%d' % (media_type, abr),
|
||||||
|
})
|
||||||
|
else:
|
||||||
|
vbr = int(vbr_el.text) // 1000
|
||||||
|
format.update({
|
||||||
|
'vbr': vbr,
|
||||||
|
'width': int(a.find('frameWidth').text),
|
||||||
|
'height': int(a.find('frameHeight').text),
|
||||||
|
'format_id': u'%s-%d' % (media_type, vbr),
|
||||||
|
})
|
||||||
|
formats.append(format)
|
||||||
|
formats.sort(key=lambda f: (f.get('vbr'), f['abr']))
|
||||||
|
if not formats:
|
||||||
|
raise ValueError('Could not find any valid formats')
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'formats': formats,
|
||||||
|
}
|
||||||
|
|
Loading…
Reference in a new issue