X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/47d80ec0b18245caeb97018d4c1af18d0b5b972b..a5a28b1eabe0533f26d23c4157c97144215b22b7:/youtube_dl/downloader/common.py diff --git a/youtube_dl/downloader/common.py b/youtube_dl/downloader/common.py index 3dc144b..cc16bbb 100644 --- a/youtube_dl/downloader/common.py +++ b/youtube_dl/downloader/common.py @@ -8,10 +8,11 @@ import random from ..compat import compat_os_name from ..utils import ( + decodeArgument, encodeFilename, error_to_compat_str, - decodeArgument, format_bytes, + shell_quote, timeconvert, ) @@ -48,6 +49,9 @@ class FileDownloader(object): external_downloader_args: A list of additional command-line arguments for the external downloader. hls_use_mpegts: Use the mpegts container for HLS videos. + http_chunk_size: Size of a chunk for chunk-based HTTP downloading. May be + useful for bypassing bandwidth throttling imposed by + a webserver (experimental) Subclasses of this one must re-define the real_download method. """ @@ -187,6 +191,9 @@ class FileDownloader(object): return filename[:-len('.part')] return filename + def ytdl_filename(self, filename): + return filename + '.ytdl' + def try_rename(self, old_filename, new_filename): try: if old_filename == new_filename: @@ -300,11 +307,11 @@ class FileDownloader(object): """Report attempt to resume at given byte.""" self.to_screen('[download] Resuming download at byte %s' % resume_len) - def report_retry(self, count, retries): + def report_retry(self, err, count, retries): """Report retry in case of HTTP error 5xx""" self.to_screen( - '[download] Got server HTTP error. Retrying (attempt %d of %s)...' - % (count, self.format_retries(retries))) + '[download] Got server HTTP error: %s. Retrying (attempt %d of %s)...' + % (error_to_compat_str(err), count, self.format_retries(retries))) def report_file_already_downloaded(self, file_name): """Report file has already been fully downloaded.""" @@ -327,27 +334,31 @@ class FileDownloader(object): os.path.exists(encodeFilename(filename)) ) - continuedl_and_exists = ( - self.params.get('continuedl', True) and - os.path.isfile(encodeFilename(filename)) and - not self.params.get('nopart', False) - ) - - # Check file already present - if filename != '-' and (nooverwrites_and_exists or continuedl_and_exists): - self.report_file_already_downloaded(filename) - self._hook_progress({ - 'filename': filename, - 'status': 'finished', - 'total_bytes': os.path.getsize(encodeFilename(filename)), - }) - return True + if not hasattr(filename, 'write'): + continuedl_and_exists = ( + self.params.get('continuedl', True) and + os.path.isfile(encodeFilename(filename)) and + not self.params.get('nopart', False) + ) + + # Check file already present + if filename != '-' and (nooverwrites_and_exists or continuedl_and_exists): + self.report_file_already_downloaded(filename) + self._hook_progress({ + 'filename': filename, + 'status': 'finished', + 'total_bytes': os.path.getsize(encodeFilename(filename)), + }) + return True min_sleep_interval = self.params.get('sleep_interval') if min_sleep_interval: max_sleep_interval = self.params.get('max_sleep_interval', min_sleep_interval) sleep_interval = random.uniform(min_sleep_interval, max_sleep_interval) - self.to_screen('[download] Sleeping %s seconds...' % sleep_interval) + self.to_screen( + '[download] Sleeping %s seconds...' % ( + int(sleep_interval) if sleep_interval.is_integer() + else '%.2f' % sleep_interval)) time.sleep(sleep_interval) return self.real_download(filename, info_dict) @@ -374,10 +385,5 @@ class FileDownloader(object): if exe is None: exe = os.path.basename(str_args[0]) - try: - import pipes - shell_quote = lambda args: ' '.join(map(pipes.quote, str_args)) - except ImportError: - shell_quote = repr self.to_screen('[debug] %s command line: %s' % ( exe, shell_quote(str_args)))