aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/adobetv.py
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor/adobetv.py')
-rw-r--r--youtube_dl/extractor/adobetv.py165
1 files changed, 114 insertions, 51 deletions
diff --git a/youtube_dl/extractor/adobetv.py b/youtube_dl/extractor/adobetv.py
index 5e43adc51..8753ee2cf 100644
--- a/youtube_dl/extractor/adobetv.py
+++ b/youtube_dl/extractor/adobetv.py
@@ -1,23 +1,32 @@
from __future__ import unicode_literals
+import re
+
from .common import InfoExtractor
+from ..compat import compat_str
from ..utils import (
parse_duration,
unified_strdate,
str_to_int,
+ int_or_none,
float_or_none,
ISO639Utils,
+ determine_ext,
)
-class AdobeTVIE(InfoExtractor):
- _VALID_URL = r'https?://tv\.adobe\.com/watch/[^/]+/(?P<id>[^/]+)'
+class AdobeTVBaseIE(InfoExtractor):
+ _API_BASE_URL = 'http://tv.adobe.com/api/v4/'
+
+
+class AdobeTVIE(AdobeTVBaseIE):
+ _VALID_URL = r'https?://tv\.adobe\.com/(?:(?P<language>fr|de|es|jp)/)?watch/(?P<show_urlname>[^/]+)/(?P<id>[^/]+)'
_TEST = {
'url': 'http://tv.adobe.com/watch/the-complete-picture-with-julieanne-kost/quick-tip-how-to-draw-a-circle-around-an-object-in-photoshop/',
'md5': '9bc5727bcdd55251f35ad311ca74fa1e',
'info_dict': {
- 'id': 'quick-tip-how-to-draw-a-circle-around-an-object-in-photoshop',
+ 'id': '10981',
'ext': 'mp4',
'title': 'Quick Tip - How to Draw a Circle Around an Object in Photoshop',
'description': 'md5:99ec318dc909d7ba2a1f2b038f7d2311',
@@ -29,50 +38,106 @@ class AdobeTVIE(InfoExtractor):
}
def _real_extract(self, url):
- video_id = self._match_id(url)
- webpage = self._download_webpage(url, video_id)
-
- player = self._parse_json(
- self._search_regex(r'html5player:\s*({.+?})\s*\n', webpage, 'player'),
- video_id)
+ language, show_urlname, urlname = re.match(self._VALID_URL, url).groups()
+ if not language:
+ language = 'en'
- title = player.get('title') or self._search_regex(
- r'data-title="([^"]+)"', webpage, 'title')
- description = self._og_search_description(webpage)
- thumbnail = self._og_search_thumbnail(webpage)
-
- upload_date = unified_strdate(
- self._html_search_meta('datepublished', webpage, 'upload date'))
-
- duration = parse_duration(
- self._html_search_meta('duration', webpage, 'duration') or
- self._search_regex(
- r'Runtime:\s*(\d{2}:\d{2}:\d{2})',
- webpage, 'duration', fatal=False))
-
- view_count = str_to_int(self._search_regex(
- r'<div class="views">\s*Views?:\s*([\d,.]+)\s*</div>',
- webpage, 'view count'))
+ video_data = self._download_json(
+ self._API_BASE_URL + 'episode/get/?language=%s&show_urlname=%s&urlname=%s&disclosure=standard' % (language, show_urlname, urlname),
+ urlname)['data'][0]
formats = [{
- 'url': source['src'],
- 'format_id': source.get('quality') or source['src'].split('-')[-1].split('.')[0] or None,
- 'tbr': source.get('bitrate'),
- } for source in player['sources']]
+ 'url': source['url'],
+ 'format_id': source.get('quality_level') or source['url'].split('-')[-1].split('.')[0] or None,
+ 'width': int_or_none(source.get('width')),
+ 'height': int_or_none(source.get('height')),
+ 'tbr': int_or_none(source.get('video_data_rate')),
+ } for source in video_data['videos']]
self._sort_formats(formats)
return {
- 'id': video_id,
- 'title': title,
- 'description': description,
- 'thumbnail': thumbnail,
- 'upload_date': upload_date,
- 'duration': duration,
- 'view_count': view_count,
+ 'id': compat_str(video_data['id']),
+ 'title': video_data['title'],
+ 'description': video_data.get('description'),
+ 'thumbnail': video_data.get('thumbnail'),
+ 'upload_date': unified_strdate(video_data.get('start_date')),
+ 'duration': parse_duration(video_data.get('duration')),
+ 'view_count': str_to_int(video_data.get('playcount')),
'formats': formats,
}
+class AdobeTVPlaylistBaseIE(AdobeTVBaseIE):
+ def _parse_page_data(self, page_data):
+ return [self.url_result(self._get_element_url(element_data)) for element_data in page_data]
+
+ def _extract_playlist_entries(self, url, display_id):
+ page = self._download_json(url, display_id)
+ entries = self._parse_page_data(page['data'])
+ for page_num in range(2, page['paging']['pages'] + 1):
+ entries.extend(self._parse_page_data(
+ self._download_json(url + '&page=%d' % page_num, display_id)['data']))
+ return entries
+
+
+class AdobeTVShowIE(AdobeTVPlaylistBaseIE):
+ _VALID_URL = r'https?://tv\.adobe\.com/(?:(?P<language>fr|de|es|jp)/)?show/(?P<id>[^/]+)'
+
+ _TEST = {
+ 'url': 'http://tv.adobe.com/show/the-complete-picture-with-julieanne-kost',
+ 'info_dict': {
+ 'id': '36',
+ 'title': 'The Complete Picture with Julieanne Kost',
+ 'description': 'md5:fa50867102dcd1aa0ddf2ab039311b27',
+ },
+ 'playlist_mincount': 136,
+ }
+
+ def _get_element_url(self, element_data):
+ return element_data['urls'][0]
+
+ def _real_extract(self, url):
+ language, show_urlname = re.match(self._VALID_URL, url).groups()
+ if not language:
+ language = 'en'
+ query = 'language=%s&show_urlname=%s' % (language, show_urlname)
+
+ show_data = self._download_json(self._API_BASE_URL + 'show/get/?%s' % query, show_urlname)['data'][0]
+
+ return self.playlist_result(
+ self._extract_playlist_entries(self._API_BASE_URL + 'episode/?%s' % query, show_urlname),
+ compat_str(show_data['id']),
+ show_data['show_name'],
+ show_data['show_description'])
+
+
+class AdobeTVChannelIE(AdobeTVPlaylistBaseIE):
+ _VALID_URL = r'https?://tv\.adobe\.com/(?:(?P<language>fr|de|es|jp)/)?channel/(?P<id>[^/]+)(?:/(?P<category_urlname>[^/]+))?'
+
+ _TEST = {
+ 'url': 'http://tv.adobe.com/channel/development',
+ 'info_dict': {
+ 'id': 'development',
+ },
+ 'playlist_mincount': 96,
+ }
+
+ def _get_element_url(self, element_data):
+ return element_data['url']
+
+ def _real_extract(self, url):
+ language, channel_urlname, category_urlname = re.match(self._VALID_URL, url).groups()
+ if not language:
+ language = 'en'
+ query = 'language=%s&channel_urlname=%s' % (language, channel_urlname)
+ if category_urlname:
+ query += '&category_urlname=%s' % category_urlname
+
+ return self.playlist_result(
+ self._extract_playlist_entries(self._API_BASE_URL + 'show/?%s' % query, channel_urlname),
+ channel_urlname)
+
+
class AdobeTVVideoIE(InfoExtractor):
_VALID_URL = r'https?://video\.tv\.adobe\.com/v/(?P<id>\d+)'
@@ -91,28 +156,25 @@ class AdobeTVVideoIE(InfoExtractor):
def _real_extract(self, url):
video_id = self._match_id(url)
-
- webpage = self._download_webpage(url, video_id)
-
- player_params = self._parse_json(self._search_regex(
- r'var\s+bridge\s*=\s*([^;]+);', webpage, 'player parameters'),
- video_id)
+ video_data = self._download_json(url + '?format=json', video_id)
formats = [{
+ 'format_id': '%s-%s' % (determine_ext(source['src']), source.get('height')),
'url': source['src'],
- 'width': source.get('width'),
- 'height': source.get('height'),
- 'tbr': source.get('bitrate'),
- } for source in player_params['sources']]
+ 'width': int_or_none(source.get('width')),
+ 'height': int_or_none(source.get('height')),
+ 'tbr': int_or_none(source.get('bitrate')),
+ } for source in video_data['sources']]
+ self._sort_formats(formats)
# For both metadata and downloaded files the duration varies among
# formats. I just pick the max one
duration = max(filter(None, [
float_or_none(source.get('duration'), scale=1000)
- for source in player_params['sources']]))
+ for source in video_data['sources']]))
subtitles = {}
- for translation in player_params.get('translations', []):
+ for translation in video_data.get('translations', []):
lang_id = translation.get('language_w3c') or ISO639Utils.long2short(translation['language_medium'])
if lang_id not in subtitles:
subtitles[lang_id] = []
@@ -124,8 +186,9 @@ class AdobeTVVideoIE(InfoExtractor):
return {
'id': video_id,
'formats': formats,
- 'title': player_params['title'],
- 'description': self._og_search_description(webpage),
+ 'title': video_data['title'],
+ 'description': video_data.get('description'),
+ 'thumbnail': video_data['video'].get('poster'),
'duration': duration,
'subtitles': subtitles,
}