Remove the --max-quality option
[youtube-dl] / youtube_dl / YoutubeDL.py
index 915963d96a6b06fbc0c809d8c1afb73ba27770d0..977141881b07f7a81acae4887369bb6bd9aec09e 100755 (executable)
@@ -61,9 +61,9 @@ from .utils import (
     render_table,
     SameFileError,
     sanitize_filename,
+    sanitize_path,
     std_headers,
     subtitles_filename,
-    takewhile_inclusive,
     UnavailableVideoError,
     url_basename,
     version_tuple,
@@ -134,7 +134,6 @@ class YoutubeDL(object):
                        (or video) as a single JSON line.
     simulate:          Do not download the video files.
     format:            Video format code. See options.py for more information.
-    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.
@@ -322,8 +321,10 @@ class YoutubeDL(object):
                 'Set the LC_ALL environment variable to fix this.')
             self.params['restrictfilenames'] = True
 
-        if '%(stitle)s' in self.params.get('outtmpl', ''):
-            self.report_warning('%(stitle)s is deprecated. Use the %(title)s and the --restrict-filenames flag(which also secures %(uploader)s et al) instead.')
+        if isinstance(params.get('outtmpl'), bytes):
+            self.report_warning(
+                'Parameter outtmpl is bytes, but should be a unicode string. '
+                'Put  from __future__ import unicode_literals  at the top of your code file or consider switching to Python 3.x.')
 
         self._setup_opener()
 
@@ -562,7 +563,7 @@ class YoutubeDL(object):
                                  if v is not None)
             template_dict = collections.defaultdict(lambda: 'NA', template_dict)
 
-            outtmpl = self.params.get('outtmpl', DEFAULT_OUTTMPL)
+            outtmpl = sanitize_path(self.params.get('outtmpl', DEFAULT_OUTTMPL))
             tmpl = compat_expanduser(outtmpl)
             filename = tmpl % template_dict
             # Temporary fix for #4787
@@ -629,7 +630,7 @@ class YoutubeDL(object):
         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:
             ies = [self.get_info_extractor(ie_key)]
@@ -916,6 +917,11 @@ class YoutubeDL(object):
         if format_spec == 'best' or format_spec is None:
             return available_formats[-1]
         elif format_spec == 'worst':
+            audiovideo_formats = [
+                f for f in available_formats
+                if f.get('vcodec') != 'none' and f.get('acodec') != 'none']
+            if audiovideo_formats:
+                return audiovideo_formats[0]
             return available_formats[0]
         elif format_spec == 'bestaudio':
             audio_formats = [
@@ -1060,12 +1066,6 @@ class YoutubeDL(object):
             full_format_info.update(format)
             format['http_headers'] = self._calc_headers(full_format_info)
 
-        format_limit = self.params.get('format_limit', None)
-        if format_limit:
-            formats = list(takewhile_inclusive(
-                lambda f: f['format_id'] != format_limit, formats
-            ))
-
         # TODO Central sorting goes here
 
         if formats[0] is not info_dict:
@@ -1083,10 +1083,13 @@ class YoutubeDL(object):
 
         req_format = self.params.get('format')
         if req_format is None:
-            req_format = 'best'
+            req_format_list = []
+            if info_dict['extractor'] in ['youtube', 'ted'] and FFmpegMergerPP(self).available:
+                req_format_list.append('bestvideo+bestaudio')
+            req_format_list.append('best')
+            req_format = '/'.join(req_format_list)
         formats_to_download = []
-        # The -1 is for supporting YoutubeIE
-        if req_format in ('-1', 'all'):
+        if req_format == 'all':
             formats_to_download = formats
         else:
             for rfstr in req_format.split(','):
@@ -1213,9 +1216,6 @@ class YoutubeDL(object):
         if len(info_dict['title']) > 200:
             info_dict['title'] = info_dict['title'][:197] + '...'
 
-        # Keep for backwards compatibility
-        info_dict['stitle'] = info_dict['title']
-
         if 'format' not in info_dict:
             info_dict['format'] = info_dict['ext']
 
@@ -1261,7 +1261,7 @@ class YoutubeDL(object):
             return
 
         try:
-            dn = os.path.dirname(encodeFilename(filename))
+            dn = os.path.dirname(sanitize_path(encodeFilename(filename)))
             if dn and not os.path.exists(dn):
                 os.makedirs(dn)
         except (OSError, IOError) as err:
@@ -1357,7 +1357,7 @@ class YoutubeDL(object):
                 if info_dict.get('requested_formats') is not None:
                     downloaded = []
                     success = True
-                    merger = FFmpegMergerPP(self, not self.params.get('keepvideo'))
+                    merger = FFmpegMergerPP(self)
                     if not merger.available:
                         postprocessors = []
                         self.report_warning('You have requested multiple '
@@ -1365,16 +1365,42 @@ class YoutubeDL(object):
                                             ' The formats won\'t be merged')
                     else:
                         postprocessors = [merger]
-                    for f in info_dict['requested_formats']:
-                        new_info = dict(info_dict)
-                        new_info.update(f)
-                        fname = self.prepare_filename(new_info)
-                        fname = prepend_extension(fname, 'f%s' % f['format_id'])
-                        downloaded.append(fname)
-                        partial_success = dl(fname, new_info)
-                        success = success and partial_success
-                    info_dict['__postprocessors'] = postprocessors
-                    info_dict['__files_to_merge'] = downloaded
+
+                    def compatible_formats(formats):
+                        video, audio = formats
+                        # Check extension
+                        video_ext, audio_ext = audio.get('ext'), video.get('ext')
+                        if video_ext and audio_ext:
+                            COMPATIBLE_EXTS = (
+                                ('mp3', 'mp4', 'm4a', 'm4p', 'm4b', 'm4r', 'm4v'),
+                                ('webm')
+                            )
+                            for exts in COMPATIBLE_EXTS:
+                                if video_ext in exts and audio_ext in exts:
+                                    return True
+                        # TODO: Check acodec/vcodec
+                        return False
+
+                    requested_formats = info_dict['requested_formats']
+                    if self.params.get('merge_output_format') is None and not compatible_formats(requested_formats):
+                        filename = os.path.splitext(filename)[0] + '.mkv'
+                        self.report_warning('You have requested formats uncompatible for merge. '
+                                            'The formats will be merged into mkv')
+                    if os.path.exists(encodeFilename(filename)):
+                        self.to_screen(
+                            '[download] %s has already been downloaded and '
+                            'merged' % filename)
+                    else:
+                        for f in requested_formats:
+                            new_info = dict(info_dict)
+                            new_info.update(f)
+                            fname = self.prepare_filename(new_info)
+                            fname = prepend_extension(fname, 'f%s' % f['format_id'])
+                            downloaded.append(fname)
+                            partial_success = dl(fname, new_info)
+                            success = success and partial_success
+                        info_dict['__postprocessors'] = postprocessors
+                        info_dict['__files_to_merge'] = downloaded
                 else:
                     # Just a single file
                     success = dl(filename, info_dict)
@@ -1482,24 +1508,17 @@ class YoutubeDL(object):
             pps_chain.extend(ie_info['__postprocessors'])
         pps_chain.extend(self._pps)
         for pp in pps_chain:
-            keep_video = None
-            old_filename = info['filepath']
             try:
-                keep_video_wish, 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
+                files_to_delete, info = pp.run(info)
             except PostProcessingError as e:
                 self.report_error(e.msg)
-            if keep_video is False and not self.params.get('keepvideo', False):
-                try:
+            if files_to_delete and not self.params.get('keepvideo', False):
+                for old_filename in files_to_delete:
                     self.to_screen('Deleting original file %s (pass -k to keep)' % old_filename)
-                    os.remove(encodeFilename(old_filename))
-                except (IOError, OSError):
-                    self.report_warning('Unable to remove downloaded video file')
+                    try:
+                        os.remove(encodeFilename(old_filename))
+                    except (IOError, OSError):
+                        self.report_warning('Unable to remove downloaded original file')
 
     def _make_archive_id(self, info_dict):
         # Future-proof against any change in case
@@ -1702,10 +1721,10 @@ class YoutubeDL(object):
             out = out.decode().strip()
             if re.match('[0-9a-f]+', out):
                 self._write_string('[debug] Git HEAD: ' + out + '\n')
-        except:
+        except Exception:
             try:
                 sys.exc_clear()
-            except:
+            except Exception:
                 pass
         self._write_string('[debug] Python version %s - %s\n' % (
             platform.python_version(), platform_name()))
@@ -1771,7 +1790,8 @@ class YoutubeDL(object):
         https_handler = make_HTTPS_handler(self.params, debuglevel=debuglevel)
         ydlh = YoutubeDLHandler(self.params, debuglevel=debuglevel)
         opener = compat_urllib_request.build_opener(
-            https_handler, proxy_handler, cookie_processor, ydlh)
+            proxy_handler, https_handler, cookie_processor, ydlh)
+
         # Delete the default user-agent header, which would otherwise apply in
         # cases where our custom HTTP handler doesn't come into play
         # (See https://github.com/rg3/youtube-dl/issues/1309 for details)