[extractor/common] Extract HLS subtitle tracks

_extract_m3u8_formats is renamed to _extract_m3u8_formats_and_subtitles
and extended to handle subtitle tracks instead of skipping them;
a wrapper with the old name is provided for compatibility.

_parse_m3u8_formats is likewise renamed and extended, but without adding
the compatibility wrapper; the test suite is adjusted to test the enhanced
method instead.
This commit is contained in:
Felix S 2016-11-07 15:45:42 +01:00
parent 19bb39202d
commit a0c3b2d5cf
2 changed files with 43 additions and 18 deletions

View file

@ -684,17 +684,19 @@ jwplayer("mediaplayer").setup({"abouttext":"Visit Indie DB","aboutlink":"http:\/
'width': 1920,
'height': 1080,
'vcodec': 'avc1.64002a',
}]
}],
{}
),
]
for m3u8_file, m3u8_url, expected_formats in _TEST_CASES:
for m3u8_file, m3u8_url, expected_formats, expected_subs in _TEST_CASES:
with io.open('./test/testdata/m3u8/%s.m3u8' % m3u8_file,
mode='r', encoding='utf-8') as f:
formats = self.ie._parse_m3u8_formats(
formats, subs = self.ie._parse_m3u8_formats_and_subtitles(
f.read(), m3u8_url, ext='mp4')
self.ie._sort_formats(formats)
expect_value(self, formats, expected_formats, None)
expect_value(self, subs, expected_subs, None)
def test_parse_mpd_formats(self):
_TEST_CASES = [

View file

@ -1879,11 +1879,21 @@ class InfoExtractor(object):
'format_note': 'Quality selection URL',
}
def _extract_m3u8_formats(self, m3u8_url, video_id, ext=None,
entry_protocol='m3u8', preference=None, quality=None,
m3u8_id=None, note=None, errnote=None,
fatal=True, live=False, data=None, headers={},
def _extract_m3u8_formats(self, *args, **kwargs):
fmts, subs = self._extract_m3u8_formats_and_subtitles(*args, **kwargs)
if subs:
self.report_warning(bug_reports_message(
"Ignoring subtitle tracks found in the HLS manifest; "
"if any subtitle tracks are missing,"
))
return fmts
def _extract_m3u8_formats_and_subtitles(
self, m3u8_url, video_id, ext=None, entry_protocol='m3u8',
preference=None, quality=None, m3u8_id=None, note=None,
errnote=None, fatal=True, live=False, data=None, headers={},
query={}):
res = self._download_webpage_handle(
m3u8_url, video_id,
note=note or 'Downloading m3u8 information',
@ -1891,30 +1901,34 @@ class InfoExtractor(object):
fatal=fatal, data=data, headers=headers, query=query)
if res is False:
return []
return [], {}
m3u8_doc, urlh = res
m3u8_url = urlh.geturl()
return self._parse_m3u8_formats(
return self._parse_m3u8_formats_and_subtitles(
m3u8_doc, m3u8_url, ext=ext, entry_protocol=entry_protocol,
preference=preference, quality=quality, m3u8_id=m3u8_id,
note=note, errnote=errnote, fatal=fatal, live=live, data=data,
headers=headers, query=query, video_id=video_id)
def _parse_m3u8_formats(self, m3u8_doc, m3u8_url, ext=None,
entry_protocol='m3u8', preference=None, quality=None,
m3u8_id=None, live=False, note=None, errnote=None,
fatal=True, data=None, headers={}, query={}, video_id=None):
def _parse_m3u8_formats_and_subtitles(
self, m3u8_doc, m3u8_url, ext=None, entry_protocol='m3u8',
preference=None, quality=None, m3u8_id=None, live=False, note=None,
errnote=None, fatal=True, data=None, headers={}, query={},
video_id=None):
if '#EXT-X-FAXS-CM:' in m3u8_doc: # Adobe Flash Access
return []
return [], {}
if (not self._downloader.params.get('allow_unplayable_formats')
and re.search(r'#EXT-X-SESSION-KEY:.*?URI="skd://', m3u8_doc)): # Apple FairPlay
return []
return [], {}
formats = []
subtitles = {}
format_url = lambda u: (
u
if re.match(r'^https?://', u)
@ -2001,7 +2015,7 @@ class InfoExtractor(object):
}
formats.append(f)
return formats
return formats, subtitles
groups = {}
last_stream_inf = {}
@ -2013,6 +2027,15 @@ class InfoExtractor(object):
if not (media_type and group_id and name):
return
groups.setdefault(group_id, []).append(media)
# <https://tools.ietf.org/html/rfc8216#section-4.3.4.1>
if media_type == 'SUBTITLES':
lang = media['LANGUAGE'] # XXX: normalise?
url = format_url(media['URI'])
sub_info = {
'url': url,
'ext': determine_ext(url),
}
subtitles.setdefault(lang, []).append(sub_info)
if media_type not in ('VIDEO', 'AUDIO'):
return
media_url = media.get('URI')
@ -2160,7 +2183,7 @@ class InfoExtractor(object):
formats.append(http_f)
last_stream_inf = {}
return formats
return formats, subtitles
@staticmethod
def _xpath_ns(path, namespace=None):