mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-16 03:53:35 +01:00
5d51a883c2
The errors while getting the subtitles are reported as warnings, if no subtitles are found return and empty dict.
589 lines
26 KiB
Python
589 lines
26 KiB
Python
#!/usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
|
|
from __future__ import absolute_import
|
|
|
|
import io
|
|
import os
|
|
import re
|
|
import shutil
|
|
import socket
|
|
import sys
|
|
import time
|
|
import traceback
|
|
|
|
from .utils import *
|
|
from .extractor import get_info_extractor, gen_extractors
|
|
from .FileDownloader import FileDownloader
|
|
|
|
|
|
class YoutubeDL(object):
|
|
"""YoutubeDL class.
|
|
|
|
YoutubeDL objects are the ones responsible of downloading the
|
|
actual video file and writing it to disk if the user has requested
|
|
it, among some other tasks. In most cases there should be one per
|
|
program. As, given a video URL, the downloader doesn't know how to
|
|
extract all the needed information, task that InfoExtractors do, it
|
|
has to pass the URL to one of them.
|
|
|
|
For this, YoutubeDL objects have a method that allows
|
|
InfoExtractors to be registered in a given order. When it is passed
|
|
a URL, the YoutubeDL object handles it to the first InfoExtractor it
|
|
finds that reports being able to handle it. The InfoExtractor extracts
|
|
all the information about the video or videos the URL refers to, and
|
|
YoutubeDL process the extracted information, possibly using a File
|
|
Downloader to download the video.
|
|
|
|
YoutubeDL objects accept a lot of parameters. In order not to saturate
|
|
the object constructor with arguments, it receives a dictionary of
|
|
options instead. These options are available through the params
|
|
attribute for the InfoExtractors to use. The YoutubeDL also
|
|
registers itself as the downloader in charge for the InfoExtractors
|
|
that are added to it, so this is a "mutual registration".
|
|
|
|
Available options:
|
|
|
|
username: Username for authentication purposes.
|
|
password: Password for authentication purposes.
|
|
videopassword: Password for acces a video.
|
|
usenetrc: Use netrc for authentication instead.
|
|
verbose: Print additional info to stdout.
|
|
quiet: Do not print messages to stdout.
|
|
forceurl: Force printing final URL.
|
|
forcetitle: Force printing title.
|
|
forceid: Force printing ID.
|
|
forcethumbnail: Force printing thumbnail URL.
|
|
forcedescription: Force printing description.
|
|
forcefilename: Force printing final filename.
|
|
simulate: Do not download the video files.
|
|
format: Video format code.
|
|
format_limit: Highest quality format to try.
|
|
outtmpl: Template for output names.
|
|
restrictfilenames: Do not allow "&" and spaces in file names
|
|
ignoreerrors: Do not stop on download errors.
|
|
nooverwrites: Prevent overwriting files.
|
|
playliststart: Playlist item to start at.
|
|
playlistend: Playlist item to end at.
|
|
matchtitle: Download only matching titles.
|
|
rejecttitle: Reject downloads for matching titles.
|
|
logtostderr: Log messages to stderr instead of stdout.
|
|
writedescription: Write the video description to a .description file
|
|
writeinfojson: Write the video description to a .info.json file
|
|
writethumbnail: Write the thumbnail image to a file
|
|
writesubtitles: Write the video subtitles to a file
|
|
writeautomaticsub: Write the automatic subtitles to a file
|
|
allsubtitles: Downloads all the subtitles of the video
|
|
listsubtitles: Lists all available subtitles for the video
|
|
subtitlesformat: Subtitle format [srt/sbv/vtt] (default=srt)
|
|
subtitleslang: Language of the subtitles to download
|
|
keepvideo: Keep the video file after post-processing
|
|
daterange: A DateRange object, download only if the upload_date is in the range.
|
|
skip_download: Skip the actual download of the video file
|
|
|
|
The following parameters are not used by YoutubeDL itself, they are used by
|
|
the FileDownloader:
|
|
nopart, updatetime, buffersize, ratelimit, min_filesize, max_filesize, test,
|
|
noresizebuffer, retries, continuedl, noprogress, consoletitle
|
|
"""
|
|
|
|
params = None
|
|
_ies = []
|
|
_pps = []
|
|
_download_retcode = None
|
|
_num_downloads = None
|
|
_screen_file = None
|
|
|
|
def __init__(self, params):
|
|
"""Create a FileDownloader object with the given options."""
|
|
self._ies = []
|
|
self._pps = []
|
|
self._progress_hooks = []
|
|
self._download_retcode = 0
|
|
self._num_downloads = 0
|
|
self._screen_file = [sys.stdout, sys.stderr][params.get('logtostderr', False)]
|
|
self.params = params
|
|
self.fd = FileDownloader(self, self.params)
|
|
|
|
if '%(stitle)s' in self.params['outtmpl']:
|
|
self.report_warning(u'%(stitle)s is deprecated. Use the %(title)s and the --restrict-filenames flag(which also secures %(uploader)s et al) instead.')
|
|
|
|
def add_info_extractor(self, ie):
|
|
"""Add an InfoExtractor object to the end of the list."""
|
|
self._ies.append(ie)
|
|
ie.set_downloader(self)
|
|
|
|
def add_default_info_extractors(self):
|
|
"""
|
|
Add the InfoExtractors returned by gen_extractors to the end of the list
|
|
"""
|
|
for ie in gen_extractors():
|
|
self.add_info_extractor(ie)
|
|
|
|
def add_post_processor(self, pp):
|
|
"""Add a PostProcessor object to the end of the chain."""
|
|
self._pps.append(pp)
|
|
pp.set_downloader(self)
|
|
|
|
def to_screen(self, message, skip_eol=False):
|
|
"""Print message to stdout if not in quiet mode."""
|
|
assert type(message) == type(u'')
|
|
if not self.params.get('quiet', False):
|
|
terminator = [u'\n', u''][skip_eol]
|
|
output = message + terminator
|
|
if 'b' in getattr(self._screen_file, 'mode', '') or sys.version_info[0] < 3: # Python 2 lies about the mode of sys.stdout/sys.stderr
|
|
output = output.encode(preferredencoding(), 'ignore')
|
|
self._screen_file.write(output)
|
|
self._screen_file.flush()
|
|
|
|
def to_stderr(self, message):
|
|
"""Print message to stderr."""
|
|
assert type(message) == type(u'')
|
|
output = message + u'\n'
|
|
if 'b' in getattr(self._screen_file, 'mode', '') or sys.version_info[0] < 3: # Python 2 lies about the mode of sys.stdout/sys.stderr
|
|
output = output.encode(preferredencoding())
|
|
sys.stderr.write(output)
|
|
|
|
def fixed_template(self):
|
|
"""Checks if the output template is fixed."""
|
|
return (re.search(u'(?u)%\\(.+?\\)s', self.params['outtmpl']) is None)
|
|
|
|
def trouble(self, message=None, tb=None):
|
|
"""Determine action to take when a download problem appears.
|
|
|
|
Depending on if the downloader has been configured to ignore
|
|
download errors or not, this method may throw an exception or
|
|
not when errors are found, after printing the message.
|
|
|
|
tb, if given, is additional traceback information.
|
|
"""
|
|
if message is not None:
|
|
self.to_stderr(message)
|
|
if self.params.get('verbose'):
|
|
if tb is None:
|
|
if sys.exc_info()[0]: # if .trouble has been called from an except block
|
|
tb = u''
|
|
if hasattr(sys.exc_info()[1], 'exc_info') and sys.exc_info()[1].exc_info[0]:
|
|
tb += u''.join(traceback.format_exception(*sys.exc_info()[1].exc_info))
|
|
tb += compat_str(traceback.format_exc())
|
|
else:
|
|
tb_data = traceback.format_list(traceback.extract_stack())
|
|
tb = u''.join(tb_data)
|
|
self.to_stderr(tb)
|
|
if not self.params.get('ignoreerrors', False):
|
|
if sys.exc_info()[0] and hasattr(sys.exc_info()[1], 'exc_info') and sys.exc_info()[1].exc_info[0]:
|
|
exc_info = sys.exc_info()[1].exc_info
|
|
else:
|
|
exc_info = sys.exc_info()
|
|
raise DownloadError(message, exc_info)
|
|
self._download_retcode = 1
|
|
|
|
def report_warning(self, message):
|
|
'''
|
|
Print the message to stderr, it will be prefixed with 'WARNING:'
|
|
If stderr is a tty file the 'WARNING:' will be colored
|
|
'''
|
|
if sys.stderr.isatty() and os.name != 'nt':
|
|
_msg_header=u'\033[0;33mWARNING:\033[0m'
|
|
else:
|
|
_msg_header=u'WARNING:'
|
|
warning_message=u'%s %s' % (_msg_header,message)
|
|
self.to_stderr(warning_message)
|
|
|
|
def report_error(self, message, tb=None):
|
|
'''
|
|
Do the same as trouble, but prefixes the message with 'ERROR:', colored
|
|
in red if stderr is a tty file.
|
|
'''
|
|
if sys.stderr.isatty() and os.name != 'nt':
|
|
_msg_header = u'\033[0;31mERROR:\033[0m'
|
|
else:
|
|
_msg_header = u'ERROR:'
|
|
error_message = u'%s %s' % (_msg_header, message)
|
|
self.trouble(error_message, tb)
|
|
|
|
def slow_down(self, start_time, byte_counter):
|
|
"""Sleep if the download speed is over the rate limit."""
|
|
rate_limit = self.params.get('ratelimit', None)
|
|
if rate_limit is None or byte_counter == 0:
|
|
return
|
|
now = time.time()
|
|
elapsed = now - start_time
|
|
if elapsed <= 0.0:
|
|
return
|
|
speed = float(byte_counter) / elapsed
|
|
if speed > rate_limit:
|
|
time.sleep((byte_counter - rate_limit * (now - start_time)) / rate_limit)
|
|
|
|
def report_writedescription(self, descfn):
|
|
""" Report that the description file is being written """
|
|
self.to_screen(u'[info] Writing video description to: ' + descfn)
|
|
|
|
def report_writesubtitles(self, sub_filename):
|
|
""" Report that the subtitles file is being written """
|
|
self.to_screen(u'[info] Writing video subtitles to: ' + sub_filename)
|
|
|
|
def report_writeinfojson(self, infofn):
|
|
""" Report that the metadata file has been written """
|
|
self.to_screen(u'[info] Video description metadata as JSON to: ' + infofn)
|
|
|
|
def report_file_already_downloaded(self, file_name):
|
|
"""Report file has already been fully downloaded."""
|
|
try:
|
|
self.to_screen(u'[download] %s has already been downloaded' % file_name)
|
|
except (UnicodeEncodeError) as err:
|
|
self.to_screen(u'[download] The file has already been downloaded')
|
|
|
|
def increment_downloads(self):
|
|
"""Increment the ordinal that assigns a number to each file."""
|
|
self._num_downloads += 1
|
|
|
|
def prepare_filename(self, info_dict):
|
|
"""Generate the output filename."""
|
|
try:
|
|
template_dict = dict(info_dict)
|
|
|
|
template_dict['epoch'] = int(time.time())
|
|
autonumber_size = self.params.get('autonumber_size')
|
|
if autonumber_size is None:
|
|
autonumber_size = 5
|
|
autonumber_templ = u'%0' + str(autonumber_size) + u'd'
|
|
template_dict['autonumber'] = autonumber_templ % self._num_downloads
|
|
if template_dict['playlist_index'] is not None:
|
|
template_dict['playlist_index'] = u'%05d' % template_dict['playlist_index']
|
|
|
|
sanitize = lambda k,v: sanitize_filename(
|
|
u'NA' if v is None else compat_str(v),
|
|
restricted=self.params.get('restrictfilenames'),
|
|
is_id=(k==u'id'))
|
|
template_dict = dict((k, sanitize(k, v)) for k,v in template_dict.items())
|
|
|
|
filename = self.params['outtmpl'] % template_dict
|
|
return filename
|
|
except KeyError as err:
|
|
self.report_error(u'Erroneous output template')
|
|
return None
|
|
except ValueError as err:
|
|
self.report_error(u'Insufficient system charset ' + repr(preferredencoding()))
|
|
return None
|
|
|
|
def _match_entry(self, info_dict):
|
|
""" Returns None iff the file should be downloaded """
|
|
|
|
title = info_dict['title']
|
|
matchtitle = self.params.get('matchtitle', False)
|
|
if matchtitle:
|
|
if not re.search(matchtitle, title, re.IGNORECASE):
|
|
return u'[download] "' + title + '" title did not match pattern "' + matchtitle + '"'
|
|
rejecttitle = self.params.get('rejecttitle', False)
|
|
if rejecttitle:
|
|
if re.search(rejecttitle, title, re.IGNORECASE):
|
|
return u'"' + title + '" title matched reject pattern "' + rejecttitle + '"'
|
|
date = info_dict.get('upload_date', None)
|
|
if date is not None:
|
|
dateRange = self.params.get('daterange', DateRange())
|
|
if date not in dateRange:
|
|
return u'[download] %s upload date is not in range %s' % (date_from_str(date).isoformat(), dateRange)
|
|
return None
|
|
|
|
def extract_info(self, url, download=True, ie_key=None, extra_info={}):
|
|
'''
|
|
Returns a list with a dictionary for each video we find.
|
|
If 'download', also downloads the videos.
|
|
extra_info is a dict containing the extra values to add to each result
|
|
'''
|
|
|
|
if ie_key:
|
|
ie = get_info_extractor(ie_key)()
|
|
ie.set_downloader(self)
|
|
ies = [ie]
|
|
else:
|
|
ies = self._ies
|
|
|
|
for ie in ies:
|
|
if not ie.suitable(url):
|
|
continue
|
|
|
|
if not ie.working():
|
|
self.report_warning(u'The program functionality for this site has been marked as broken, '
|
|
u'and will probably not work.')
|
|
|
|
try:
|
|
ie_result = ie.extract(url)
|
|
if ie_result is None: # Finished already (backwards compatibility; listformats and friends should be moved here)
|
|
break
|
|
if isinstance(ie_result, list):
|
|
# Backwards compatibility: old IE result format
|
|
for result in ie_result:
|
|
result.update(extra_info)
|
|
ie_result = {
|
|
'_type': 'compat_list',
|
|
'entries': ie_result,
|
|
}
|
|
else:
|
|
ie_result.update(extra_info)
|
|
if 'extractor' not in ie_result:
|
|
ie_result['extractor'] = ie.IE_NAME
|
|
return self.process_ie_result(ie_result, download=download)
|
|
except ExtractorError as de: # An error we somewhat expected
|
|
self.report_error(compat_str(de), de.format_traceback())
|
|
break
|
|
except Exception as e:
|
|
if self.params.get('ignoreerrors', False):
|
|
self.report_error(compat_str(e), tb=compat_str(traceback.format_exc()))
|
|
break
|
|
else:
|
|
raise
|
|
else:
|
|
self.report_error(u'no suitable InfoExtractor: %s' % url)
|
|
|
|
def process_ie_result(self, ie_result, download=True, extra_info={}):
|
|
"""
|
|
Take the result of the ie(may be modified) and resolve all unresolved
|
|
references (URLs, playlist items).
|
|
|
|
It will also download the videos if 'download'.
|
|
Returns the resolved ie_result.
|
|
"""
|
|
|
|
result_type = ie_result.get('_type', 'video') # If not given we suppose it's a video, support the default old system
|
|
if result_type == 'video':
|
|
ie_result.update(extra_info)
|
|
if 'playlist' not in ie_result:
|
|
# It isn't part of a playlist
|
|
ie_result['playlist'] = None
|
|
ie_result['playlist_index'] = None
|
|
if download:
|
|
self.process_info(ie_result)
|
|
return ie_result
|
|
elif result_type == 'url':
|
|
# We have to add extra_info to the results because it may be
|
|
# contained in a playlist
|
|
return self.extract_info(ie_result['url'],
|
|
download,
|
|
ie_key=ie_result.get('ie_key'),
|
|
extra_info=extra_info)
|
|
elif result_type == 'playlist':
|
|
# We process each entry in the playlist
|
|
playlist = ie_result.get('title', None) or ie_result.get('id', None)
|
|
self.to_screen(u'[download] Downloading playlist: %s' % playlist)
|
|
|
|
playlist_results = []
|
|
|
|
n_all_entries = len(ie_result['entries'])
|
|
playliststart = self.params.get('playliststart', 1) - 1
|
|
playlistend = self.params.get('playlistend', -1)
|
|
|
|
if playlistend == -1:
|
|
entries = ie_result['entries'][playliststart:]
|
|
else:
|
|
entries = ie_result['entries'][playliststart:playlistend]
|
|
|
|
n_entries = len(entries)
|
|
|
|
self.to_screen(u"[%s] playlist '%s': Collected %d video ids (downloading %d of them)" %
|
|
(ie_result['extractor'], playlist, n_all_entries, n_entries))
|
|
|
|
for i,entry in enumerate(entries,1):
|
|
self.to_screen(u'[download] Downloading video #%s of %s' %(i, n_entries))
|
|
extra = {
|
|
'playlist': playlist,
|
|
'playlist_index': i + playliststart,
|
|
}
|
|
if not 'extractor' in entry:
|
|
# We set the extractor, if it's an url it will be set then to
|
|
# the new extractor, but if it's already a video we must make
|
|
# sure it's present: see issue #877
|
|
entry['extractor'] = ie_result['extractor']
|
|
entry_result = self.process_ie_result(entry,
|
|
download=download,
|
|
extra_info=extra)
|
|
playlist_results.append(entry_result)
|
|
ie_result['entries'] = playlist_results
|
|
return ie_result
|
|
elif result_type == 'compat_list':
|
|
def _fixup(r):
|
|
r.setdefault('extractor', ie_result['extractor'])
|
|
return r
|
|
ie_result['entries'] = [
|
|
self.process_ie_result(_fixup(r), download=download)
|
|
for r in ie_result['entries']
|
|
]
|
|
return ie_result
|
|
else:
|
|
raise Exception('Invalid result type: %s' % result_type)
|
|
|
|
def process_info(self, info_dict):
|
|
"""Process a single resolved IE result."""
|
|
|
|
assert info_dict.get('_type', 'video') == 'video'
|
|
#We increment the download the download count here to match the previous behaviour.
|
|
self.increment_downloads()
|
|
|
|
info_dict['fulltitle'] = info_dict['title']
|
|
if len(info_dict['title']) > 200:
|
|
info_dict['title'] = info_dict['title'][:197] + u'...'
|
|
|
|
# Keep for backwards compatibility
|
|
info_dict['stitle'] = info_dict['title']
|
|
|
|
if not 'format' in info_dict:
|
|
info_dict['format'] = info_dict['ext']
|
|
|
|
reason = self._match_entry(info_dict)
|
|
if reason is not None:
|
|
self.to_screen(u'[download] ' + reason)
|
|
return
|
|
|
|
max_downloads = self.params.get('max_downloads')
|
|
if max_downloads is not None:
|
|
if self._num_downloads > int(max_downloads):
|
|
raise MaxDownloadsReached()
|
|
|
|
filename = self.prepare_filename(info_dict)
|
|
|
|
# Forced printings
|
|
if self.params.get('forcetitle', False):
|
|
compat_print(info_dict['title'])
|
|
if self.params.get('forceid', False):
|
|
compat_print(info_dict['id'])
|
|
if self.params.get('forceurl', False):
|
|
compat_print(info_dict['url'])
|
|
if self.params.get('forcethumbnail', False) and 'thumbnail' in info_dict:
|
|
compat_print(info_dict['thumbnail'])
|
|
if self.params.get('forcedescription', False) and 'description' in info_dict:
|
|
compat_print(info_dict['description'])
|
|
if self.params.get('forcefilename', False) and filename is not None:
|
|
compat_print(filename)
|
|
if self.params.get('forceformat', False):
|
|
compat_print(info_dict['format'])
|
|
|
|
# Do nothing else if in simulate mode
|
|
if self.params.get('simulate', False):
|
|
return
|
|
|
|
if filename is None:
|
|
return
|
|
|
|
try:
|
|
dn = os.path.dirname(encodeFilename(filename))
|
|
if dn != '' and not os.path.exists(dn):
|
|
os.makedirs(dn)
|
|
except (OSError, IOError) as err:
|
|
self.report_error(u'unable to create directory ' + compat_str(err))
|
|
return
|
|
|
|
if self.params.get('writedescription', False):
|
|
try:
|
|
descfn = filename + u'.description'
|
|
self.report_writedescription(descfn)
|
|
with io.open(encodeFilename(descfn), 'w', encoding='utf-8') as descfile:
|
|
descfile.write(info_dict['description'])
|
|
except (OSError, IOError):
|
|
self.report_error(u'Cannot write description file ' + descfn)
|
|
return
|
|
|
|
subtitles_are_requested = any([self.params.get('writesubtitles', False),
|
|
self.params.get('writeautomaticsub'),
|
|
self.params.get('allsubtitles', False)])
|
|
|
|
if subtitles_are_requested and 'subtitles' in info_dict and info_dict['subtitles']:
|
|
# subtitles download errors are already managed as troubles in relevant IE
|
|
# that way it will silently go on when used with unsupporting IE
|
|
subtitles = info_dict['subtitles']
|
|
sub_format = self.params.get('subtitlesformat')
|
|
for sub_lang in subtitles.keys():
|
|
sub = subtitles[sub_lang]
|
|
try:
|
|
sub_filename = filename.rsplit('.', 1)[0] + u'.' + sub_lang + u'.' + sub_format
|
|
self.report_writesubtitles(sub_filename)
|
|
with io.open(encodeFilename(sub_filename), 'w', encoding='utf-8') as subfile:
|
|
subfile.write(sub)
|
|
except (OSError, IOError):
|
|
self.report_error(u'Cannot write subtitles file ' + descfn)
|
|
return
|
|
|
|
if self.params.get('writeinfojson', False):
|
|
infofn = filename + u'.info.json'
|
|
self.report_writeinfojson(infofn)
|
|
try:
|
|
json_info_dict = dict((k, v) for k,v in info_dict.items() if not k in ['urlhandle'])
|
|
write_json_file(json_info_dict, encodeFilename(infofn))
|
|
except (OSError, IOError):
|
|
self.report_error(u'Cannot write metadata to JSON file ' + infofn)
|
|
return
|
|
|
|
if self.params.get('writethumbnail', False):
|
|
if info_dict.get('thumbnail') is not None:
|
|
thumb_format = determine_ext(info_dict['thumbnail'], u'jpg')
|
|
thumb_filename = filename.rpartition('.')[0] + u'.' + thumb_format
|
|
self.to_screen(u'[%s] %s: Downloading thumbnail ...' %
|
|
(info_dict['extractor'], info_dict['id']))
|
|
uf = compat_urllib_request.urlopen(info_dict['thumbnail'])
|
|
with open(thumb_filename, 'wb') as thumbf:
|
|
shutil.copyfileobj(uf, thumbf)
|
|
self.to_screen(u'[%s] %s: Writing thumbnail to: %s' %
|
|
(info_dict['extractor'], info_dict['id'], thumb_filename))
|
|
|
|
if not self.params.get('skip_download', False):
|
|
if self.params.get('nooverwrites', False) and os.path.exists(encodeFilename(filename)):
|
|
success = True
|
|
else:
|
|
try:
|
|
success = self.fd._do_download(filename, info_dict)
|
|
except (OSError, IOError) as err:
|
|
raise UnavailableVideoError()
|
|
except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
|
|
self.report_error(u'unable to download video data: %s' % str(err))
|
|
return
|
|
except (ContentTooShortError, ) as err:
|
|
self.report_error(u'content too short (expected %s bytes and served %s)' % (err.expected, err.downloaded))
|
|
return
|
|
|
|
if success:
|
|
try:
|
|
self.post_process(filename, info_dict)
|
|
except (PostProcessingError) as err:
|
|
self.report_error(u'postprocessing: %s' % str(err))
|
|
return
|
|
|
|
def download(self, url_list):
|
|
"""Download a given list of URLs."""
|
|
if len(url_list) > 1 and self.fixed_template():
|
|
raise SameFileError(self.params['outtmpl'])
|
|
|
|
for url in url_list:
|
|
try:
|
|
#It also downloads the videos
|
|
videos = self.extract_info(url)
|
|
except UnavailableVideoError:
|
|
self.report_error(u'unable to download video')
|
|
except MaxDownloadsReached:
|
|
self.to_screen(u'[info] Maximum number of downloaded files reached.')
|
|
raise
|
|
|
|
return self._download_retcode
|
|
|
|
def post_process(self, filename, ie_info):
|
|
"""Run all the postprocessors on the given file."""
|
|
info = dict(ie_info)
|
|
info['filepath'] = filename
|
|
keep_video = None
|
|
for pp in self._pps:
|
|
try:
|
|
keep_video_wish,new_info = pp.run(info)
|
|
if keep_video_wish is not None:
|
|
if keep_video_wish:
|
|
keep_video = keep_video_wish
|
|
elif keep_video is None:
|
|
# No clear decision yet, let IE decide
|
|
keep_video = keep_video_wish
|
|
except PostProcessingError as e:
|
|
self.to_stderr(u'ERROR: ' + e.msg)
|
|
if keep_video is False and not self.params.get('keepvideo', False):
|
|
try:
|
|
self.to_screen(u'Deleting original file %s (pass -k to keep)' % filename)
|
|
os.remove(encodeFilename(filename))
|
|
except (IOError, OSError):
|
|
self.report_warning(u'Unable to remove downloaded video file')
|