aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor
diff options
context:
space:
mode:
authorSergey M․ <dstftw@gmail.com>2015-01-13 20:54:03 +0600
committerSergey M․ <dstftw@gmail.com>2015-01-13 20:54:03 +0600
commit04e0bac233fd4f094983d612d6c76b5590393230 (patch)
tree5fe8c72b013ce1504087fdba7292094647cf470d /youtube_dl/extractor
parentfbef83f399b6a54380cecfce6d00964cc1e063d0 (diff)
[npo:live] Add extractor (Closes #4691)
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r--youtube_dl/extractor/__init__.py1
-rw-r--r--youtube_dl/extractor/npo.py99
2 files changed, 91 insertions, 9 deletions
diff --git a/youtube_dl/extractor/__init__.py b/youtube_dl/extractor/__init__.py
index 5a1c2f69b..2d31c298f 100644
--- a/youtube_dl/extractor/__init__.py
+++ b/youtube_dl/extractor/__init__.py
@@ -295,6 +295,7 @@ from .nowness import NownessIE
from .nowvideo import NowVideoIE
from .npo import (
NPOIE,
+ NPOLiveIE,
TegenlichtVproIE,
)
from .nrk import (
diff --git a/youtube_dl/extractor/npo.py b/youtube_dl/extractor/npo.py
index 6d20f755a..8d496488b 100644
--- a/youtube_dl/extractor/npo.py
+++ b/youtube_dl/extractor/npo.py
@@ -13,7 +13,16 @@ from ..utils import (
)
-class NPOIE(InfoExtractor):
+class NPOBaseIE(InfoExtractor):
+ def _get_token(self, video_id):
+ token_page = self._download_webpage(
+ 'http://ida.omroep.nl/npoplayer/i.js',
+ video_id, note='Downloading token')
+ return self._search_regex(
+ r'npoplayer\.token = "(.+?)"', token_page, 'token')
+
+
+class NPOIE(NPOBaseIE):
IE_NAME = 'npo.nl'
_VALID_URL = r'https?://www\.npo\.nl/[^/]+/[^/]+/(?P<id>[^/?]+)'
@@ -80,8 +89,7 @@ class NPOIE(InfoExtractor):
]
def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
+ video_id = self._match_id(url)
return self._get_info(video_id)
def _get_info(self, video_id):
@@ -91,12 +99,8 @@ class NPOIE(InfoExtractor):
# We have to remove the javascript callback
transform_source=strip_jsonp,
)
- token_page = self._download_webpage(
- 'http://ida.omroep.nl/npoplayer/i.js',
- video_id,
- note='Downloading token'
- )
- token = self._search_regex(r'npoplayer\.token = "(.+?)"', token_page, 'token')
+
+ token = self._get_token(video_id)
formats = []
@@ -170,6 +174,83 @@ class NPOIE(InfoExtractor):
}
+class NPOLiveIE(NPOBaseIE):
+ IE_NAME = 'npo.nl:live'
+ _VALID_URL = r'https?://www\.npo\.nl/live/(?P<id>.+)'
+
+ _TEST = {
+ 'url': 'http://www.npo.nl/live/npo-1',
+ 'info_dict': {
+ 'id': 'LI_NEDERLAND1_136692',
+ 'display_id': 'npo-1',
+ 'ext': 'mp4',
+ 'title': 're:^Nederland 1 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
+ 'description': 'Livestream',
+ 'is_live': True,
+ },
+ 'params': {
+ 'skip_download': True,
+ }
+ }
+
+ def _real_extract(self, url):
+ display_id = self._match_id(url)
+
+ webpage = self._download_webpage(url, display_id)
+
+ live_id = self._search_regex(
+ r'data-prid="([^"]+)"', webpage, 'live id')
+
+ metadata = self._download_json(
+ 'http://e.omroep.nl/metadata/%s' % live_id,
+ display_id, transform_source=strip_jsonp)
+
+ token = self._get_token(display_id)
+
+ formats = []
+
+ streams = metadata.get('streams')
+ if streams:
+ for stream in streams:
+ stream_type = stream.get('type').lower()
+ if stream_type == 'ss':
+ continue
+ stream_info = self._download_json(
+ 'http://ida.omroep.nl/aapi/?stream=%s&token=%s&type=jsonp'
+ % (stream.get('url'), token),
+ display_id, 'Downloading %s JSON' % stream_type)
+ if stream_info.get('error_code', 0) or stream_info.get('errorcode', 0):
+ continue
+ stream_url = self._download_json(
+ stream_info['stream'], display_id,
+ 'Downloading %s URL' % stream_type,
+ transform_source=strip_jsonp)
+ if stream_type == 'hds':
+ f4m_formats = self._extract_f4m_formats(stream_url, display_id)
+ # f4m downloader downloads only piece of live stream
+ for f4m_format in f4m_formats:
+ f4m_format['preference'] = -1
+ formats.extend(f4m_formats)
+ elif stream_type == 'hls':
+ formats.extend(self._extract_m3u8_formats(stream_url, display_id, 'mp4'))
+ else:
+ formats.append({
+ 'url': stream_url,
+ })
+
+ self._sort_formats(formats)
+
+ return {
+ 'id': live_id,
+ 'display_id': display_id,
+ 'title': self._live_title(metadata['titel']),
+ 'description': metadata['info'],
+ 'thumbnail': metadata.get('images', [{'url': None}])[-1]['url'],
+ 'formats': formats,
+ 'is_live': True,
+ }
+
+
class TegenlichtVproIE(NPOIE):
IE_NAME = 'tegenlicht.vpro.nl'
_VALID_URL = r'https?://tegenlicht\.vpro\.nl/afleveringen/.*?'