aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSergey M <dstftw@gmail.com>2016-01-04 03:33:22 +0500
committerSergey M <dstftw@gmail.com>2016-01-04 03:33:22 +0500
commit2098aee7d6facb9de2253f9b71a2dfa9b932b4cb (patch)
treed84cf52cccc4328e0a4a7824095554260de7fd4f
parentfb588f6a5663811da83cde68d17ff4eb041022ab (diff)
parent896c7a23cd9fbfc589f26e9a7bbada38a325e2ad (diff)
Merge pull request #8063 from bpfoley/rteradio
[rte:radio] Add support for RTE radio player
-rw-r--r--youtube_dl/extractor/__init__.py2
-rw-r--r--youtube_dl/extractor/rte.py77
2 files changed, 70 insertions, 9 deletions
diff --git a/youtube_dl/extractor/__init__.py b/youtube_dl/extractor/__init__.py
index 625b0bf16..19cea5e99 100644
--- a/youtube_dl/extractor/__init__.py
+++ b/youtube_dl/extractor/__init__.py
@@ -563,7 +563,7 @@ from .ro220 import Ro220IE
from .rottentomatoes import RottenTomatoesIE
from .roxwel import RoxwelIE
from .rtbf import RTBFIE
-from .rte import RteIE
+from .rte import RteIE, RteRadioIE
from .rtlnl import RtlNlIE
from .rtl2 import RTL2IE
from .rtp import RTPIE
diff --git a/youtube_dl/extractor/rte.py b/youtube_dl/extractor/rte.py
index d9cfbf180..25abcee92 100644
--- a/youtube_dl/extractor/rte.py
+++ b/youtube_dl/extractor/rte.py
@@ -5,16 +5,19 @@ from .common import InfoExtractor
from ..utils import (
float_or_none,
+ unescapeHTML,
)
class RteIE(InfoExtractor):
+ IE_NAME = 'rte'
+ IE_DESC = 'Raidió Teilifís Éireann TV'
_VALID_URL = r'https?://(?:www\.)?rte\.ie/player/[^/]{2,3}/show/[^/]+/(?P<id>[0-9]+)'
_TEST = {
'url': 'http://www.rte.ie/player/ie/show/iwitness-862/10478715/',
'info_dict': {
'id': '10478715',
- 'ext': 'mp4',
+ 'ext': 'flv',
'title': 'Watch iWitness online',
'thumbnail': 're:^https?://.*\.jpg$',
'description': 'iWitness : The spirit of Ireland, one voice and one minute at a time.',
@@ -44,13 +47,6 @@ class RteIE(InfoExtractor):
# f4m_url = server + relative_url
f4m_url = json_string['shows'][0]['media:group'][0]['rte:server'] + json_string['shows'][0]['media:group'][0]['url']
f4m_formats = self._extract_f4m_formats(f4m_url, video_id)
- f4m_formats = [{
- 'format_id': f['format_id'],
- 'url': f['url'],
- 'ext': 'mp4',
- 'width': f['width'],
- 'height': f['height'],
- } for f in f4m_formats]
return {
'id': video_id,
@@ -60,3 +56,68 @@ class RteIE(InfoExtractor):
'thumbnail': thumbnail,
'duration': duration,
}
+
+
+
+class RteRadioIE(InfoExtractor):
+ IE_NAME = 'rte:radio'
+ IE_DESC = 'Raidió Teilifís Éireann radio'
+ # Radioplayer URLs have the specifier #!rii=<channel_id>:<id>:<playable_item_id>:<date>:
+ # where the IDs are int/empty, the date is DD-MM-YYYY, and the specifier may be truncated.
+ # An <id> uniquely defines an individual recording, and is the only part we require.
+ _VALID_URL = r'https?://(?:www\.)?rte\.ie/radio/utils/radioplayer/rteradioweb\.html#!rii=(?:[0-9]*)(?:%3A|:)(?P<id>[0-9]+)'
+
+ _TEST = {
+ 'url': 'http://www.rte.ie/radio/utils/radioplayer/rteradioweb.html#!rii=16:10507902:2414:27-12-2015:',
+ 'info_dict': {
+ 'id': '10507902',
+ 'ext': 'flv',
+ 'title': 'Gloria',
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ 'description': 'Tim Thurston guides you through a millennium of sacred music featuring Gregorian chant, pure solo voices and choral masterpieces, framed around the glorious music of J.S. Bach.',
+ 'duration': 7230.0,
+ },
+ 'params': {
+ 'skip_download': 'f4m fails with --test atm'
+ }
+ }
+
+ def _real_extract(self, url):
+ item_id = self._match_id(url)
+ feeds_url = 'http://www.rte.ie/rteavgen/getplaylist/?type=web&format=json&id=' + item_id
+ json_string = self._download_json(feeds_url, item_id)
+
+ # NB the string values in the JSON are stored using XML escaping(!)
+ show = json_string['shows'][0]
+ title = unescapeHTML(show['title'])
+ description = unescapeHTML(show.get('description'))
+ thumbnail = show.get('thumbnail')
+ duration = float_or_none(show.get('duration'), 1000)
+
+ mg = show['media:group'][0]
+
+ formats = []
+
+ if mg.get('url') and not mg['url'].startswith('rtmpe:'):
+ formats.append({'url': mg.get('url')})
+
+ if mg.get('hls_server') and mg.get('hls_url'):
+ hls_url = mg['hls_server'] + mg['hls_url']
+ hls_formats = self._extract_m3u8_formats(
+ hls_url, item_id, 'mp4', m3u8_id='hls', fatal=False)
+ formats.extend(hls_formats)
+
+ if mg.get('hds_server') and mg.get('hds_url'):
+ f4m_url = mg['hds_server'] + mg['hds_url']
+ f4m_formats = self._extract_f4m_formats(
+ f4m_url, item_id, f4m_id='hds', fatal=False)
+ formats.extend(f4m_formats)
+
+ return {
+ 'id': item_id,
+ 'title': title,
+ 'formats': formats,
+ 'description': description,
+ 'thumbnail': thumbnail,
+ 'duration': duration,
+ }