[lnkgo] Add new extractor
authorNaglis Jonaitis <njonaitis@gmail.com>
Tue, 13 Jan 2015 22:14:59 +0000 (00:14 +0200)
committerNaglis Jonaitis <njonaitis@gmail.com>
Tue, 13 Jan 2015 22:14:59 +0000 (00:14 +0200)
youtube_dl/extractor/__init__.py
youtube_dl/extractor/lnkgo.py [new file with mode: 0644]

index 2d31c298fbb20653df75443394bc7c792b785711..0902eb4374caf943e1af4e23aeb7c74a8386a794 100644 (file)
@@ -226,6 +226,7 @@ from .livestream import (
     LivestreamOriginalIE,
     LivestreamShortenerIE,
 )
+from .lnkgo import LnkGoIE
 from .lrt import LRTIE
 from .lynda import (
     LyndaIE,
diff --git a/youtube_dl/extractor/lnkgo.py b/youtube_dl/extractor/lnkgo.py
new file mode 100644 (file)
index 0000000..cfae601
--- /dev/null
@@ -0,0 +1,122 @@
+# coding: utf-8
+from __future__ import unicode_literals
+
+import re
+
+from .common import InfoExtractor
+from ..utils import (
+    int_or_none,
+    js_to_json,
+    unified_strdate,
+)
+
+
+class LnkGoIE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?lnkgo\.alfa\.lt/visi\-video/(?P<show>[^/]+)/ziurek\-(?P<display_id>[A-Za-z0-9\-]+)'
+    _TESTS = [{
+        'url': 'http://lnkgo.alfa.lt/visi-video/yra-kaip-yra/ziurek-yra-kaip-yra-162',
+        'info_dict': {
+            'id': '46712',
+            'ext': 'mp4',
+            'title': 'Yra kaip yra',
+            'upload_date': '20150107',
+            'description': 'md5:d82a5e36b775b7048617f263a0e3475e',
+            'age_limit': 7,
+            'duration': 3019,
+            'thumbnail': 're:^https?://.*\.jpg$'
+        },
+        'params': {
+            'skip_download': True,  # HLS download
+        },
+    }, {
+        'url': 'http://lnkgo.alfa.lt/visi-video/aktualai-pratesimas/ziurek-nerdas-taiso-kompiuteri-2',
+        'info_dict': {
+            'id': '47289',
+            'ext': 'mp4',
+            'title': 'Nėrdas: Kompiuterio Valymas',
+            'upload_date': '20150113',
+            'description': 'md5:7352d113a242a808676ff17e69db6a69',
+            'age_limit': 18,
+            'duration': 346,
+            'thumbnail': 're:^https?://.*\.jpg$'
+        },
+        'params': {
+            'skip_download': True,  # HLS download
+        },
+    }]
+    _AGE_LIMITS = {
+        'N-7': 7,
+        'N-14': 14,
+        'S': 18,
+    }
+
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        display_id = mobj.group('display_id')
+
+        webpage = self._download_webpage(
+            url, display_id, 'Downloading player webpage')
+
+        video_id = self._search_regex(
+            r'data-ep="([^"]+)"', webpage, 'video ID')
+        title = self._og_search_title(webpage)
+        description = self._og_search_description(webpage)
+
+        thumbnail = self._og_search_thumbnail(webpage)
+        thumbnail_w = int_or_none(
+            self._og_search_property('image:width', webpage, 'thumbnail width', fatal=False))
+        thumbnail_h = int_or_none(
+            self._og_search_property('image:height', webpage, 'thumbnail height', fatal=False))
+        thumbnails = [{
+            'url': thumbnail,
+            'width': thumbnail_w,
+            'height': thumbnail_h,
+        }]
+
+        upload_date = unified_strdate(self._search_regex(
+            r'class="meta-item\sair-time">.*?<strong>([^<]+)</strong>', webpage, 'upload date', fatal=False))
+        duration = int_or_none(self._search_regex(
+            r'VideoDuration = "([^"]+)"', webpage, 'duration', fatal=False))
+
+        pg_rating = self._search_regex(
+            r'pgrating="([^"]+)"', webpage, 'PG rating', fatal=False, default='')
+        age_limit = self._AGE_LIMITS.get(pg_rating, 0)
+
+        sources_js = self._search_regex(
+            r'(?s)sources:\s(\[.*?\]),', webpage, 'sources')
+        sources = self._parse_json(
+            sources_js, video_id, transform_source=js_to_json)
+
+        formats = []
+        for source in sources:
+            if source.get('provider') == 'rtmp':
+                m = re.search(r'^(?P<url>rtmp://[^/]+/(?P<app>[^/]+))/(?P<play_path>.+)$', source['file'])
+                if not m:
+                    continue
+                formats.append({
+                    'format_id': 'rtmp',
+                    'ext': 'flv',
+                    'url': m.group('url'),
+                    'play_path': m.group('play_path'),
+                    'page_url': url,
+                })
+            elif source.get('file').endswith('.m3u8'):
+                formats.append({
+                    'format_id': 'hls',
+                    'ext': source.get('type', 'mp4'),
+                    'url': source['file'],
+                })
+
+        self._sort_formats(formats)
+
+        return {
+            'id': video_id,
+            'display_id': display_id,
+            'title': title,
+            'formats': formats,
+            'thumbnails': thumbnails,
+            'duration': duration,
+            'description': description,
+            'age_limit': age_limit,
+            'upload_date': upload_date,
+        }