Merge branch 'Tassatux-noco'

This commit is contained in:
Sergey M․ 2015-05-05 02:50:58 +06:00
commit bbe718c97f

View File

@ -14,6 +14,9 @@ from ..compat import (
from ..utils import (
clean_html,
ExtractorError,
int_or_none,
float_or_none,
parse_iso8601,
unified_strdate,
)
@ -25,21 +28,38 @@ class NocoIE(InfoExtractor):
_SUB_LANG_TEMPLATE = '&sub_lang=%s'
_NETRC_MACHINE = 'noco'
_TEST = {
'url': 'http://noco.tv/emission/11538/nolife/ami-ami-idol-hello-france/',
'md5': '0a993f0058ddbcd902630b2047ef710e',
'info_dict': {
'id': '11538',
'ext': 'mp4',
'title': 'Ami Ami Idol - Hello! France',
'description': 'md5:4eaab46ab68fa4197a317a88a53d3b86',
'upload_date': '20140412',
'uploader': 'Nolife',
'uploader_id': 'NOL',
'duration': 2851.2,
_TESTS = [
{
'url': 'http://noco.tv/emission/11538/nolife/ami-ami-idol-hello-france/',
'md5': '0a993f0058ddbcd902630b2047ef710e',
'info_dict': {
'id': '11538',
'ext': 'mp4',
'title': 'Ami Ami Idol - Hello! France',
'description': 'md5:4eaab46ab68fa4197a317a88a53d3b86',
'upload_date': '20140412',
'uploader': 'Nolife',
'uploader_id': 'NOL',
'duration': 2851.2,
},
'skip': 'Requires noco account',
},
'skip': 'Requires noco account',
}
{
'url': 'http://noco.tv/emission/12610/lbl42/the-guild/s01e01-wake-up-call',
'md5': 'c190f1f48e313c55838f1f412225934d',
'info_dict': {
'id': '12610',
'ext': 'mp4',
'title': 'The Guild #1 - Wake-Up Call',
'timestamp': 1403863200,
'upload_date': '20140627',
'uploader': 'LBL42',
'uploader_id': 'LBL',
'duration': 233.023,
},
'skip': 'Requires noco account',
}
]
def _real_initialize(self):
self._login()
@ -90,51 +110,66 @@ class NocoIE(InfoExtractor):
'shows/%s/medias' % video_id,
video_id, 'Downloading video JSON')
show = self._call_api(
'shows/by_id/%s' % video_id,
video_id, 'Downloading show JSON')[0]
options = self._call_api(
'users/init', video_id,
'Downloading user options JSON')['options']
audio_lang_pref = options.get('audio_language') or options.get('language', 'fr')
if audio_lang_pref == 'original':
audio_lang_pref = show['original_lang']
if len(medias) == 1:
audio_lang_pref = list(medias.keys())[0]
elif audio_lang_pref not in medias:
audio_lang_pref = 'fr'
qualities = self._call_api(
'qualities',
video_id, 'Downloading qualities JSON')
formats = []
for lang, lang_dict in medias['fr']['video_list'].items():
for format_id, fmt in lang_dict['quality_list'].items():
format_id_extended = '%s-%s' % (lang, format_id) if lang != 'none' else format_id
for audio_lang, audio_lang_dict in medias.items():
preference = 1 if audio_lang == audio_lang_pref else 0
for sub_lang, lang_dict in audio_lang_dict['video_list'].items():
for format_id, fmt in lang_dict['quality_list'].items():
format_id_extended = 'audio-%s_sub-%s_%s' % (audio_lang, sub_lang, format_id)
video = self._call_api(
'shows/%s/video/%s/fr' % (video_id, format_id.lower()),
video_id, 'Downloading %s video JSON' % format_id_extended,
lang if lang != 'none' else None)
video = self._call_api(
'shows/%s/video/%s/%s' % (video_id, format_id.lower(), audio_lang),
video_id, 'Downloading %s video JSON' % format_id_extended,
sub_lang if sub_lang != 'none' else None)
file_url = video['file']
if not file_url:
continue
file_url = video['file']
if not file_url:
continue
if file_url in ['forbidden', 'not found']:
popmessage = video['popmessage']
self._raise_error(popmessage['title'], popmessage['message'])
if file_url in ['forbidden', 'not found']:
popmessage = video['popmessage']
self._raise_error(popmessage['title'], popmessage['message'])
formats.append({
'url': file_url,
'format_id': format_id_extended,
'width': fmt['res_width'],
'height': fmt['res_lines'],
'abr': fmt['audiobitrate'],
'vbr': fmt['videobitrate'],
'filesize': fmt['filesize'],
'format_note': qualities[format_id]['quality_name'],
'preference': qualities[format_id]['priority'],
})
formats.append({
'url': file_url,
'format_id': format_id_extended,
'width': int_or_none(fmt.get('res_width')),
'height': int_or_none(fmt.get('res_lines')),
'abr': int_or_none(fmt.get('audiobitrate')),
'vbr': int_or_none(fmt.get('videobitrate')),
'filesize': int_or_none(fmt.get('filesize')),
'format_note': qualities[format_id].get('quality_name'),
'quality': qualities[format_id].get('priority'),
'preference': preference,
})
self._sort_formats(formats)
show = self._call_api(
'shows/by_id/%s' % video_id,
video_id, 'Downloading show JSON')[0]
upload_date = unified_strdate(show['online_date_start_utc'])
uploader = show['partner_name']
uploader_id = show['partner_key']
duration = show['duration_ms'] / 1000.0
timestamp = parse_iso8601(show.get('online_date_start_utc'), ' ')
uploader = show.get('partner_name')
uploader_id = show.get('partner_key')
duration = float_or_none(show.get('duration_ms'), 1000)
thumbnails = []
for thumbnail_key, thumbnail_url in show.items():
@ -166,7 +201,7 @@ class NocoIE(InfoExtractor):
'title': title,
'description': description,
'thumbnails': thumbnails,
'upload_date': upload_date,
'timestamp': timestamp,
'uploader': uploader,
'uploader_id': uploader_id,
'duration': duration,