[YoutubeDL] Fix format selection with filters (Closes #10083)

This commit is contained in:
Sergey M․ 2016-07-16 00:55:43 +07:00
parent 23495d6a39
commit 317f7ab634
No known key found for this signature in database
GPG key ID: 2C393E0F18A9236D
2 changed files with 81 additions and 16 deletions

View file

@ -5,6 +5,7 @@ from __future__ import absolute_import, unicode_literals
import collections
import contextlib
import copy
import datetime
import errno
import fileinput
@ -1051,9 +1052,9 @@ class YoutubeDL(object):
if isinstance(selector, list):
fs = [_build_selector_function(s) for s in selector]
def selector_function(formats):
def selector_function(ctx):
for f in fs:
for format in f(formats):
for format in f(ctx):
yield format
return selector_function
elif selector.type == GROUP:
@ -1061,17 +1062,17 @@ class YoutubeDL(object):
elif selector.type == PICKFIRST:
fs = [_build_selector_function(s) for s in selector.selector]
def selector_function(formats):
def selector_function(ctx):
for f in fs:
picked_formats = list(f(formats))
picked_formats = list(f(ctx))
if picked_formats:
return picked_formats
return []
elif selector.type == SINGLE:
format_spec = selector.selector
def selector_function(formats):
formats = list(formats)
def selector_function(ctx):
formats = list(ctx['formats'])
if not formats:
return
if format_spec == 'all':
@ -1084,9 +1085,10 @@ class YoutubeDL(object):
if f.get('vcodec') != 'none' and f.get('acodec') != 'none']
if audiovideo_formats:
yield audiovideo_formats[format_idx]
# for audio only (soundcloud) or video only (imgur) urls, select the best/worst audio format
elif (all(f.get('acodec') != 'none' for f in formats) or
all(f.get('vcodec') != 'none' for f in formats)):
# for extractors with incomplete formats (audio only (soundcloud)
# or video only (imgur)) we will fallback to best/worst
# {video,audio}-only format
elif ctx['incomplete_formats']:
yield formats[format_idx]
elif format_spec == 'bestaudio':
audio_formats = [
@ -1160,17 +1162,18 @@ class YoutubeDL(object):
}
video_selector, audio_selector = map(_build_selector_function, selector.selector)
def selector_function(formats):
formats = list(formats)
for pair in itertools.product(video_selector(formats), audio_selector(formats)):
def selector_function(ctx):
for pair in itertools.product(
video_selector(copy.deepcopy(ctx)), audio_selector(copy.deepcopy(ctx))):
yield _merge(pair)
filters = [self._build_format_filter(f) for f in selector.filters]
def final_selector(formats):
def final_selector(ctx):
ctx_copy = copy.deepcopy(ctx)
for _filter in filters:
formats = list(filter(_filter, formats))
return selector_function(formats)
ctx_copy['formats'] = list(filter(_filter, ctx_copy['formats']))
return selector_function(ctx_copy)
return final_selector
stream = io.BytesIO(format_spec.encode('utf-8'))
@ -1377,7 +1380,35 @@ class YoutubeDL(object):
req_format_list.append('best')
req_format = '/'.join(req_format_list)
format_selector = self.build_format_selector(req_format)
formats_to_download = list(format_selector(formats))
# While in format selection we may need to have an access to the original
# format set in order to calculate some metrics or do some processing.
# For now we need to be able to guess whether original formats provided
# by extractor are incomplete or not (i.e. whether extractor provides only
# video-only or audio-only formats) for proper formats selection for
# extractors with such incomplete formats (see
# https://github.com/rg3/youtube-dl/pull/5556).
# Since formats may be filtered during format selection and may not match
# the original formats the results may be incorrect. Thus original formats
# or pre-calculated metrics should be passed to format selection routines
# as well.
# We will pass a context object containing all necessary additional data
# instead of just formats.
# This fixes incorrect format selection issue (see
# https://github.com/rg3/youtube-dl/issues/10083).
incomplete_formats = all(
# All formats are video-only or
f.get('vcodec') != 'none' and f.get('acodec') == 'none' or
# all formats are audio-only
f.get('vcodec') == 'none' and f.get('acodec') != 'none'
for f in formats)
ctx = {
'formats': formats,
'incomplete_formats': incomplete_formats,
}
formats_to_download = list(format_selector(ctx))
if not formats_to_download:
raise ExtractorError('requested format not available',
expected=True)