def _real_main(argv=None):
# Compatibility fixes for Windows
if sys.platform == 'win32':
- # https://github.com/rg3/youtube-dl/issues/820
+ # https://github.com/ytdl-org/youtube-dl/issues/820
codecs.register(lambda name: codecs.lookup('utf-8') if name == 'cp65001' else None)
workaround_optparse_bug9161()
if opts.verbose:
write_string('[debug] Batch file urls: ' + repr(batch_urls) + '\n')
except IOError:
- sys.exit('ERROR: batch file could not be read')
+ sys.exit('ERROR: batch file %s could not be read' % opts.batchfile)
all_urls = batch_urls + [url.strip() for url in args] # batch_urls are already striped in read_batch_urls
_enc = preferredencoding()
all_urls = [url.decode(_enc, 'ignore') if isinstance(url, bytes) else url for url in all_urls]
if opts.max_sleep_interval is not None:
if opts.max_sleep_interval < 0:
parser.error('max sleep interval must be positive or 0')
+ if opts.sleep_interval is None:
+ parser.error('min sleep interval must be specified, use --min-sleep-interval')
if opts.max_sleep_interval < opts.sleep_interval:
parser.error('max sleep interval must be greater than or equal to min sleep interval')
else:
if numeric_buffersize is None:
parser.error('invalid buffer size specified')
opts.buffersize = numeric_buffersize
+ if opts.http_chunk_size is not None:
+ numeric_chunksize = FileDownloader.parse_bytes(opts.http_chunk_size)
+ if not numeric_chunksize:
+ parser.error('invalid http chunk size specified')
+ opts.http_chunk_size = numeric_chunksize
if opts.playliststart <= 0:
raise ValueError('Playlist start must be positive')
if opts.playlistend not in (-1, None) and opts.playlistend < opts.playliststart:
if opts.recodevideo not in ['mp4', 'flv', 'webm', 'ogg', 'mkv', 'avi']:
parser.error('invalid video recode format specified')
if opts.convertsubtitles is not None:
- if opts.convertsubtitles not in ['srt', 'vtt', 'ass']:
+ if opts.convertsubtitles not in ['srt', 'vtt', 'ass', 'lrc']:
parser.error('invalid subtitle format specified')
if opts.date is not None:
if opts.allsubtitles and not opts.writeautomaticsub:
opts.writesubtitles = True
- outtmpl = ((opts.outtmpl is not None and opts.outtmpl) or
- (opts.format == '-1' and opts.usetitle and '%(title)s-%(id)s-%(format)s.%(ext)s') or
- (opts.format == '-1' and '%(id)s-%(format)s.%(ext)s') or
- (opts.usetitle and opts.autonumber and '%(autonumber)s-%(title)s-%(id)s.%(ext)s') or
- (opts.usetitle and '%(title)s-%(id)s.%(ext)s') or
- (opts.useid and '%(id)s.%(ext)s') or
- (opts.autonumber and '%(autonumber)s-%(id)s.%(ext)s') or
- DEFAULT_OUTTMPL)
+ outtmpl = ((opts.outtmpl is not None and opts.outtmpl)
+ or (opts.format == '-1' and opts.usetitle and '%(title)s-%(id)s-%(format)s.%(ext)s')
+ or (opts.format == '-1' and '%(id)s-%(format)s.%(ext)s')
+ or (opts.usetitle and opts.autonumber and '%(autonumber)s-%(title)s-%(id)s.%(ext)s')
+ or (opts.usetitle and '%(title)s-%(id)s.%(ext)s')
+ or (opts.useid and '%(id)s.%(ext)s')
+ or (opts.autonumber and '%(autonumber)s-%(id)s.%(ext)s')
+ or DEFAULT_OUTTMPL)
if not os.path.splitext(outtmpl)[1] and opts.extractaudio:
parser.error('Cannot download a video and extract audio into the same'
' file! Use "{0}.%(ext)s" instead of "{0}" as the output'
'keep_fragments': opts.keep_fragments,
'buffersize': opts.buffersize,
'noresizebuffer': opts.noresizebuffer,
+ 'http_chunk_size': opts.http_chunk_size,
'continuedl': opts.continue_dl,
'noprogress': opts.noprogress,
'progress_with_newline': opts.progress_with_newline,
'config_location': opts.config_location,
'geo_bypass': opts.geo_bypass,
'geo_bypass_country': opts.geo_bypass_country,
+ 'geo_bypass_ip_block': opts.geo_bypass_ip_block,
# just for deprecation check
'autonumber': opts.autonumber if opts.autonumber is True else None,
'usetitle': opts.usetitle if opts.usetitle is True else None,