diff options
author | Sergey M․ <dstftw@gmail.com> | 2017-04-22 07:01:00 +0700 |
---|---|---|
committer | Sergey M․ <dstftw@gmail.com> | 2017-04-22 07:01:00 +0700 |
commit | cb2520802d7b8efdc71d6d97aeca984b5f878716 (patch) | |
tree | 8263f79f9b536130b5faa89d9ee4184b3b2b453a /youtube_dl/extractor | |
parent | f7799582500ca2b00c371c287028298f187c0477 (diff) |
[extractor/common] Improve m3u8 extraction (closes #12211)
* Extract m3u8 parsing to separate method
* Improve rendition groups extraction
* Build stream name according stream GROUP-ID
* Ignore reference to AUDIO group without URI when stream has no CODECS
+ Add test coverage for parsing m3u8 from #11507, #11995, #12211 and twitch vod
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r-- | youtube_dl/extractor/common.py | 154 |
1 files changed, 96 insertions, 58 deletions
diff --git a/youtube_dl/extractor/common.py b/youtube_dl/extractor/common.py index 61d97ab72..359c549c5 100644 --- a/youtube_dl/extractor/common.py +++ b/youtube_dl/extractor/common.py @@ -1312,17 +1312,25 @@ class InfoExtractor(object): entry_protocol='m3u8', preference=None, m3u8_id=None, note=None, errnote=None, fatal=True, live=False): - res = self._download_webpage_handle( m3u8_url, video_id, note=note or 'Downloading m3u8 information', errnote=errnote or 'Failed to download m3u8 information', fatal=fatal) + if res is False: return [] + m3u8_doc, urlh = res m3u8_url = urlh.geturl() + return self._parse_m3u8_formats( + m3u8_doc, m3u8_url, ext=ext, entry_protocol=entry_protocol, + preference=preference, m3u8_id=m3u8_id, live=live) + + def _parse_m3u8_formats(self, m3u8_doc, m3u8_url, ext=None, + entry_protocol='m3u8', preference=None, + m3u8_id=None, live=False): if '#EXT-X-FAXS-CM:' in m3u8_doc: # Adobe Flash Access return [] @@ -1333,19 +1341,21 @@ class InfoExtractor(object): if re.match(r'^https?://', u) else compat_urlparse.urljoin(m3u8_url, u)) - # We should try extracting formats only from master playlists [1], i.e. - # playlists that describe available qualities. On the other hand media - # playlists [2] should be returned as is since they contain just the media - # without qualities renditions. + # References: + # 1. https://tools.ietf.org/html/draft-pantos-http-live-streaming-21 + # 2. https://github.com/rg3/youtube-dl/issues/12211 + + # We should try extracting formats only from master playlists [1, 4.3.4], + # i.e. playlists that describe available qualities. On the other hand + # media playlists [1, 4.3.3] should be returned as is since they contain + # just the media without qualities renditions. # Fortunately, master playlist can be easily distinguished from media - # playlist based on particular tags availability. As of [1, 2] master - # playlist tags MUST NOT appear in a media playist and vice versa. - # As of [3] #EXT-X-TARGETDURATION tag is REQUIRED for every media playlist - # and MUST NOT appear in master playlist thus we can clearly detect media - # playlist with this criterion. - # 1. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.4 - # 2. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.3 - # 3. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.3.1 + # playlist based on particular tags availability. As of [1, 4.3.3, 4.3.4] + # master playlist tags MUST NOT appear in a media playist and vice versa. + # As of [1, 4.3.3.1] #EXT-X-TARGETDURATION tag is REQUIRED for every + # media playlist and MUST NOT appear in master playlist thus we can + # clearly detect media playlist with this criterion. + if '#EXT-X-TARGETDURATION' in m3u8_doc: # media playlist, return as is return [{ 'url': m3u8_url, @@ -1354,52 +1364,67 @@ class InfoExtractor(object): 'protocol': entry_protocol, 'preference': preference, }] - audio_in_video_stream = {} - last_info = {} - last_media = {} + + groups = {} + last_stream_inf = {} + + def extract_media(x_media_line): + media = parse_m3u8_attributes(x_media_line) + # As per [1, 4.3.4.1] TYPE, GROUP-ID and NAME are REQUIRED + media_type, group_id, name = media.get('TYPE'), media.get('GROUP-ID'), media.get('NAME') + if not (media_type and group_id and name): + return + groups.setdefault(group_id, []).append(media) + if media_type not in ('VIDEO', 'AUDIO'): + return + media_url = media.get('URI') + if media_url: + format_id = [] + for v in (group_id, name): + if v: + format_id.append(v) + f = { + 'format_id': '-'.join(format_id), + 'url': format_url(media_url), + 'language': media.get('LANGUAGE'), + 'ext': ext, + 'protocol': entry_protocol, + 'preference': preference, + } + if media_type == 'AUDIO': + f['vcodec'] = 'none' + formats.append(f) + + def build_stream_name(): + # Despite specification does not mention NAME attribute for + # EXT-X-STREAM-INF it still sometimes may be present + stream_name = last_stream_inf.get('NAME') + if stream_name: + return stream_name + # If there is no NAME in EXT-X-STREAM-INF it will be obtained + # from corresponding rendition group + stream_group_id = last_stream_inf.get('VIDEO') + if not stream_group_id: + return + stream_group = groups.get(stream_group_id) + if not stream_group: + return stream_group_id + rendition = stream_group[0] + return rendition.get('NAME') or stream_group_id + for line in m3u8_doc.splitlines(): if line.startswith('#EXT-X-STREAM-INF:'): - last_info = parse_m3u8_attributes(line) + last_stream_inf = parse_m3u8_attributes(line) elif line.startswith('#EXT-X-MEDIA:'): - media = parse_m3u8_attributes(line) - media_type = media.get('TYPE') - if media_type in ('VIDEO', 'AUDIO'): - group_id = media.get('GROUP-ID') - media_url = media.get('URI') - if media_url: - format_id = [] - for v in (group_id, media.get('NAME')): - if v: - format_id.append(v) - f = { - 'format_id': '-'.join(format_id), - 'url': format_url(media_url), - 'language': media.get('LANGUAGE'), - 'ext': ext, - 'protocol': entry_protocol, - 'preference': preference, - } - if media_type == 'AUDIO': - f['vcodec'] = 'none' - if group_id and not audio_in_video_stream.get(group_id): - audio_in_video_stream[group_id] = False - formats.append(f) - else: - # When there is no URI in EXT-X-MEDIA let this tag's - # data be used by regular URI lines below - last_media = media - if media_type == 'AUDIO' and group_id: - audio_in_video_stream[group_id] = True + extract_media(line) elif line.startswith('#') or not line.strip(): continue else: - tbr = int_or_none(last_info.get('AVERAGE-BANDWIDTH') or last_info.get('BANDWIDTH'), scale=1000) + tbr = int_or_none(last_stream_inf.get('AVERAGE-BANDWIDTH') or last_stream_inf.get('BANDWIDTH'), scale=1000) format_id = [] if m3u8_id: format_id.append(m3u8_id) - # Despite specification does not mention NAME attribute for - # EXT-X-STREAM-INF it still sometimes may be present - stream_name = last_info.get('NAME') or last_media.get('NAME') + stream_name = build_stream_name() # Bandwidth of live streams may differ over time thus making # format_id unpredictable. So it's better to keep provided # format_id intact. @@ -1412,11 +1437,11 @@ class InfoExtractor(object): 'manifest_url': manifest_url, 'tbr': tbr, 'ext': ext, - 'fps': float_or_none(last_info.get('FRAME-RATE')), + 'fps': float_or_none(last_stream_inf.get('FRAME-RATE')), 'protocol': entry_protocol, 'preference': preference, } - resolution = last_info.get('RESOLUTION') + resolution = last_stream_inf.get('RESOLUTION') if resolution: mobj = re.search(r'(?P<width>\d+)[xX](?P<height>\d+)', resolution) if mobj: @@ -1432,13 +1457,26 @@ class InfoExtractor(object): 'vbr': vbr, 'abr': abr, }) - f.update(parse_codecs(last_info.get('CODECS'))) - if audio_in_video_stream.get(last_info.get('AUDIO')) is False and f['vcodec'] != 'none': - # TODO: update acodec for audio only formats with the same GROUP-ID - f['acodec'] = 'none' + codecs = parse_codecs(last_stream_inf.get('CODECS')) + f.update(codecs) + audio_group_id = last_stream_inf.get('AUDIO') + # As per [1, 4.3.4.1.1] any EXT-X-STREAM-INF tag which + # references a rendition group MUST have a CODECS attribute. + # However, this is not always respected, for example, [2] + # contains EXT-X-STREAM-INF tag which references AUDIO + # rendition group but does not have CODECS and despite + # referencing audio group an audio group, it represents + # a complete (with audio and video) format. So, for such cases + # we will ignore references to rendition groups and treat them + # as complete formats. + if audio_group_id and codecs and f.get('vcodec') != 'none': + audio_group = groups.get(audio_group_id) + if audio_group and audio_group[0].get('URI'): + # TODO: update acodec for audio only formats with + # the same GROUP-ID + f['acodec'] = 'none' formats.append(f) - last_info = {} - last_media = {} + last_stream_inf = {} return formats @staticmethod |