]> jfr.im git - yt-dlp.git/blobdiff - yt_dlp/postprocessor/ffmpeg.py
[FixupM3u8] Fixup MPEG-TS in MP4 container
[yt-dlp.git] / yt_dlp / postprocessor / ffmpeg.py
index bf6677239b930262cecb817ade609bfbe1a7e44a..e8b569d72887164242f378cb356611dce2f66254 100644 (file)
@@ -8,24 +8,27 @@
 import re
 import json
 
-
 from .common import AudioConversionError, PostProcessor
 
-from ..compat import compat_str, compat_numeric_types
+from ..compat import compat_str
 from ..utils import (
+    dfxp2srt,
     encodeArgument,
     encodeFilename,
-    get_exe_version,
+    float_or_none,
+    _get_exe_version_output,
+    detect_exe_version,
     is_outdated_version,
+    ISO639Utils,
+    orderedSet,
+    Popen,
     PostProcessingError,
     prepend_extension,
-    shell_quote,
-    dfxp2srt,
-    ISO639Utils,
-    process_communicate_or_kill,
     replace_extension,
+    shell_quote,
     traverse_obj,
     variadic,
+    write_json_file,
 )
 
 
@@ -40,6 +43,7 @@
     'ts': 'mpegts',
     'wma': 'asf',
     'wmv': 'asf',
+    'vtt': 'webvtt',
 }
 ACODECS = {
     'mp3': 'libmp3lame',
@@ -49,6 +53,7 @@
     'opus': 'libopus',
     'vorbis': 'libvorbis',
     'wav': None,
+    'alac': None,
 }
 
 
@@ -72,16 +77,21 @@ def check_version(self):
                 self.basename, self.basename, required_version)
             self.report_warning(warning)
 
+    @staticmethod
+    def get_versions_and_features(downloader=None):
+        pp = FFmpegPostProcessor(downloader)
+        return pp._versions, pp._features
+
     @staticmethod
     def get_versions(downloader=None):
-        return FFmpegPostProcessor(downloader)._versions
+        return FFmpegPostProcessor.get_version_and_features(downloader)[0]
 
     def _determine_executables(self):
         programs = ['avprobe', 'avconv', 'ffmpeg', 'ffprobe']
-        prefer_ffmpeg = True
 
-        def get_ffmpeg_version(path):
-            ver = get_exe_version(path, args=['-version'])
+        def get_ffmpeg_version(path, prog):
+            out = _get_exe_version_output(path, ['-bsfs'])
+            ver = detect_exe_version(out) if out else False
             if ver:
                 regexs = [
                     r'(?:\d+:)?([0-9.]+)-[0-9]+ubuntu[0-9.]+$',  # Ubuntu, see [1]
@@ -92,39 +102,52 @@ def get_ffmpeg_version(path):
                     mobj = re.match(regex, ver)
                     if mobj:
                         ver = mobj.group(1)
-            return ver
+            self._versions[prog] = ver
+            if prog != 'ffmpeg' or not out:
+                return
+
+            mobj = re.search(r'(?m)^\s+libavformat\s+(?:[0-9. ]+)\s+/\s+(?P<runtime>[0-9. ]+)', out)
+            lavf_runtime_version = mobj.group('runtime').replace(' ', '') if mobj else None
+            self._features = {
+                'fdk': '--enable-libfdk-aac' in out,
+                'setts': 'setts' in out.splitlines(),
+                'needs_adtstoasc': is_outdated_version(lavf_runtime_version, '57.56.100', False),
+            }
 
         self.basename = None
         self.probe_basename = None
-
         self._paths = None
         self._versions = None
-        if self._downloader:
-            prefer_ffmpeg = self.get_param('prefer_ffmpeg', True)
-            location = self.get_param('ffmpeg_location')
-            if location is not None:
-                if not os.path.exists(location):
-                    self.report_warning(
-                        'ffmpeg-location %s does not exist! '
-                        'Continuing without ffmpeg.' % (location))
-                    self._versions = {}
-                    return
-                elif not os.path.isdir(location):
-                    basename = os.path.splitext(os.path.basename(location))[0]
-                    basename = next((p for p in programs if basename.startswith(p)), 'ffmpeg')
-                    dirname = os.path.dirname(os.path.abspath(location))
-                    if basename in ('ffmpeg', 'ffprobe'):
-                        prefer_ffmpeg = True
-
-                self._paths = dict(
-                    (p, os.path.join(dirname, p)) for p in programs)
+        self._features = {}
+
+        prefer_ffmpeg = self.get_param('prefer_ffmpeg', True)
+        location = self.get_param('ffmpeg_location')
+        if location is None:
+            self._paths = {p: p for p in programs}
+        else:
+            if not os.path.exists(location):
+                self.report_warning(
+                    'ffmpeg-location %s does not exist! '
+                    'Continuing without ffmpeg.' % (location))
+                self._versions = {}
+                return
+            elif os.path.isdir(location):
+                dirname, basename = location, None
+            else:
+                basename = os.path.splitext(os.path.basename(location))[0]
+                basename = next((p for p in programs if basename.startswith(p)), 'ffmpeg')
+                dirname = os.path.dirname(os.path.abspath(location))
+                if basename in ('ffmpeg', 'ffprobe'):
+                    prefer_ffmpeg = True
+
+            self._paths = dict(
+                (p, os.path.join(dirname, p)) for p in programs)
+            if basename:
                 self._paths[basename] = location
-                self._versions = dict(
-                    (p, get_ffmpeg_version(self._paths[p])) for p in programs)
-        if self._versions is None:
-            self._versions = dict(
-                (p, get_ffmpeg_version(p)) for p in programs)
-            self._paths = dict((p, p) for p in programs)
+
+        self._versions = {}
+        for p in programs:
+            get_ffmpeg_version(self._paths[p], p)
 
         if prefer_ffmpeg is False:
             prefs = ('avconv', 'ffmpeg')
@@ -174,10 +197,8 @@ def get_audio_codec(self, path):
                     encodeArgument('-i')]
             cmd.append(encodeFilename(self._ffmpeg_filename_argument(path), True))
             self.write_debug('%s command line: %s' % (self.basename, shell_quote(cmd)))
-            handle = subprocess.Popen(
-                cmd, stderr=subprocess.PIPE,
-                stdout=subprocess.PIPE, stdin=subprocess.PIPE)
-            stdout_data, stderr_data = process_communicate_or_kill(handle)
+            handle = Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+            stdout_data, stderr_data = handle.communicate_or_kill()
             expected_ret = 0 if self.probe_available else 1
             if handle.wait() != expected_ret:
                 return None
@@ -219,7 +240,7 @@ def get_metadata_object(self, path, opts=[]):
         cmd += opts
         cmd.append(encodeFilename(self._ffmpeg_filename_argument(path), True))
         self.write_debug('ffprobe command line: %s' % shell_quote(cmd))
-        p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE)
+        p = Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE)
         stdout, stderr = p.communicate()
         return json.loads(stdout.decode('utf-8', 'replace'))
 
@@ -230,6 +251,23 @@ def get_stream_number(self, path, keys, value):
             None)
         return num, len(streams)
 
+    def _get_real_video_duration(self, info, fatal=True):
+        try:
+            if '_real_duration' not in info:
+                info['_real_duration'] = float_or_none(
+                    traverse_obj(self.get_metadata_object(info['filepath']), ('format', 'duration')))
+            if not info['_real_duration']:
+                raise PostProcessingError('ffprobe returned empty duration')
+        except PostProcessingError as e:
+            if fatal:
+                raise PostProcessingError(f'Unable to determine video duration; {e}')
+        return info.setdefault('_real_duration', None)
+
+    def _duration_mismatch(self, d1, d2):
+        if not d1 or not d2:
+            return None
+        return abs(d1 - d2) > 1
+
     def run_ffmpeg_multiple_files(self, input_paths, out_path, opts, **kwargs):
         return self.real_run_ffmpeg(
             [(path, []) for path in input_paths],
@@ -263,12 +301,11 @@ def make_args(file, args, name, number):
                 for i, (path, opts) in enumerate(path_opts) if path)
 
         self.write_debug('ffmpeg command line: %s' % shell_quote(cmd))
-        p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE)
-        stdout, stderr = process_communicate_or_kill(p)
+        p = Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE)
+        stdout, stderr = p.communicate_or_kill()
         if p.returncode not in variadic(expected_retcodes):
             stderr = stderr.decode('utf-8', 'replace').strip()
-            if self.get_param('verbose', False):
-                self.report_error(stderr)
+            self.write_debug(stderr)
             raise FFmpegPostProcessorError(stderr.split('\n')[-1])
         for out_path, _ in output_path_opts:
             if out_path:
@@ -278,7 +315,8 @@ def make_args(file, args, name, number):
     def run_ffmpeg(self, path, out_path, opts, **kwargs):
         return self.run_ffmpeg_multiple_files([path], out_path, opts, **kwargs)
 
-    def _ffmpeg_filename_argument(self, fn):
+    @staticmethod
+    def _ffmpeg_filename_argument(fn):
         # Always use 'file:' because the filename may contain ':' (ffmpeg
         # interprets that as a protocol) or can start with '-' (-- is broken in
         # ffmpeg, see https://ffmpeg.org/trac/ffmpeg/ticket/2127 for details)
@@ -287,17 +325,95 @@ def _ffmpeg_filename_argument(self, fn):
             return fn
         return 'file:' + fn if fn != '-' else fn
 
+    @staticmethod
+    def _quote_for_ffmpeg(string):
+        # See https://ffmpeg.org/ffmpeg-utils.html#toc-Quoting-and-escaping
+        # A sequence of '' produces '\'''\'';
+        # final replace removes the empty '' between \' \'.
+        string = string.replace("'", r"'\''").replace("'''", "'")
+        # Handle potential ' at string boundaries.
+        string = string[1:] if string[0] == "'" else "'" + string
+        return string[:-1] if string[-1] == "'" else string + "'"
+
+    def force_keyframes(self, filename, timestamps):
+        timestamps = orderedSet(timestamps)
+        if timestamps[0] == 0:
+            timestamps = timestamps[1:]
+        keyframe_file = prepend_extension(filename, 'keyframes.temp')
+        self.to_screen(f'Re-encoding "{filename}" with appropriate keyframes')
+        self.run_ffmpeg(filename, keyframe_file, ['-force_key_frames', ','.join(
+            f'{t:.6f}' for t in timestamps)])
+        return keyframe_file
+
+    def concat_files(self, in_files, out_file, concat_opts=None):
+        """
+        Use concat demuxer to concatenate multiple files having identical streams.
+
+        Only inpoint, outpoint, and duration concat options are supported.
+        See https://ffmpeg.org/ffmpeg-formats.html#concat-1 for details
+        """
+        concat_file = f'{out_file}.concat'
+        self.write_debug(f'Writing concat spec to {concat_file}')
+        with open(concat_file, 'wt', encoding='utf-8') as f:
+            f.writelines(self._concat_spec(in_files, concat_opts))
+
+        out_flags = ['-c', 'copy']
+        if out_file.rpartition('.')[-1] in ('mp4', 'mov'):
+            # For some reason, '-c copy' is not enough to copy subtitles
+            out_flags.extend(['-c:s', 'mov_text', '-movflags', '+faststart'])
+
+        try:
+            self.real_run_ffmpeg(
+                [(concat_file, ['-hide_banner', '-nostdin', '-f', 'concat', '-safe', '0'])],
+                [(out_file, out_flags)])
+        finally:
+            os.remove(concat_file)
+
+    @classmethod
+    def _concat_spec(cls, in_files, concat_opts=None):
+        if concat_opts is None:
+            concat_opts = [{}] * len(in_files)
+        yield 'ffconcat version 1.0\n'
+        for file, opts in zip(in_files, concat_opts):
+            yield f'file {cls._quote_for_ffmpeg(cls._ffmpeg_filename_argument(file))}\n'
+            # Iterate explicitly to yield the following directives in order, ignoring the rest.
+            for directive in 'inpoint', 'outpoint', 'duration':
+                if directive in opts:
+                    yield f'{directive} {opts[directive]}\n'
+
 
 class FFmpegExtractAudioPP(FFmpegPostProcessor):
     COMMON_AUDIO_EXTS = ('wav', 'flac', 'm4a', 'aiff', 'mp3', 'ogg', 'mka', 'opus', 'wma')
-    SUPPORTED_EXTS = ('best', 'aac', 'flac', 'mp3', 'm4a', 'opus', 'vorbis', 'wav')
+    SUPPORTED_EXTS = ('best', 'aac', 'flac', 'mp3', 'm4a', 'opus', 'vorbis', 'wav', 'alac')
 
     def __init__(self, downloader=None, preferredcodec=None, preferredquality=None, nopostoverwrites=False):
         FFmpegPostProcessor.__init__(self, downloader)
         self._preferredcodec = preferredcodec or 'best'
-        self._preferredquality = preferredquality
+        self._preferredquality = float_or_none(preferredquality)
         self._nopostoverwrites = nopostoverwrites
 
+    def _quality_args(self, codec):
+        if self._preferredquality is None:
+            return []
+        elif self._preferredquality > 10:
+            return ['-b:a', f'{self._preferredquality}k']
+
+        limits = {
+            'libmp3lame': (10, 0),
+            'libvorbis': (0, 10),
+            # FFmpeg's AAC encoder does not have an upper limit for the value of -q:a.
+            # Experimentally, with values over 4, bitrate changes were minimal or non-existent
+            'aac': (0.1, 4),
+            'libfdk_aac': (1, 5),
+        }.get(codec)
+        if not limits:
+            return []
+
+        q = limits[1] + (limits[0] - limits[1]) * (self._preferredquality / 10)
+        if codec == 'libfdk_aac':
+            return ['-vbr', f'{int(q)}']
+        return ['-q:a', f'{q}']
+
     def run_ffmpeg(self, path, out_path, codec, more_opts):
         if codec is None:
             acodec_opts = []
@@ -311,7 +427,7 @@ def run_ffmpeg(self, path, out_path, codec, more_opts):
 
     @PostProcessor._restrict_to(images=False)
     def run(self, information):
-        path = information['filepath']
+        orig_path = path = information['filepath']
         orig_ext = information['ext']
 
         if self._preferredcodec == 'best' and orig_ext in self.COMMON_AUDIO_EXTS:
@@ -337,71 +453,73 @@ def run(self, information):
                     more_opts = ['-f', 'adts']
                 if filecodec == 'vorbis':
                     extension = 'ogg'
+            elif filecodec == 'alac':
+                acodec = None
+                extension = 'm4a'
+                more_opts += ['-acodec', 'alac']
             else:
                 # MP3 otherwise.
                 acodec = 'libmp3lame'
                 extension = 'mp3'
-                more_opts = []
-                if self._preferredquality is not None:
-                    if int(self._preferredquality) < 10:
-                        more_opts += ['-q:a', self._preferredquality]
-                    else:
-                        more_opts += ['-b:a', self._preferredquality + 'k']
+                more_opts = self._quality_args(acodec)
         else:
             # We convert the audio (lossy if codec is lossy)
             acodec = ACODECS[self._preferredcodec]
+            if acodec == 'aac' and self._features.get('fdk'):
+                acodec = 'libfdk_aac'
             extension = self._preferredcodec
-            more_opts = []
-            if self._preferredquality is not None:
-                # The opus codec doesn't support the -aq option
-                if int(self._preferredquality) < 10 and extension != 'opus':
-                    more_opts += ['-q:a', self._preferredquality]
-                else:
-                    more_opts += ['-b:a', self._preferredquality + 'k']
+            more_opts = self._quality_args(acodec)
             if self._preferredcodec == 'aac':
                 more_opts += ['-f', 'adts']
-            if self._preferredcodec == 'm4a':
+            elif self._preferredcodec == 'm4a':
                 more_opts += ['-bsf:a', 'aac_adtstoasc']
-            if self._preferredcodec == 'vorbis':
+            elif self._preferredcodec == 'vorbis':
                 extension = 'ogg'
-            if self._preferredcodec == 'wav':
+            elif self._preferredcodec == 'wav':
                 extension = 'wav'
                 more_opts += ['-f', 'wav']
+            elif self._preferredcodec == 'alac':
+                extension = 'm4a'
+                more_opts += ['-acodec', 'alac']
 
         prefix, sep, ext = path.rpartition('.')  # not os.path.splitext, since the latter does not work on unicode in all setups
-        new_path = prefix + sep + extension
-
-        information['filepath'] = new_path
-        information['ext'] = extension
+        temp_path = new_path = prefix + sep + extension
 
-        # If we download foo.mp3 and convert it to... foo.mp3, then don't delete foo.mp3, silly.
-        if (new_path == path
-                or (self._nopostoverwrites and os.path.exists(encodeFilename(new_path)))):
+        if new_path == path:
+            orig_path = prepend_extension(path, 'orig')
+            temp_path = prepend_extension(path, 'temp')
+        if (self._nopostoverwrites and os.path.exists(encodeFilename(new_path))
+                and os.path.exists(encodeFilename(orig_path))):
             self.to_screen('Post-process file %s exists, skipping' % new_path)
             return [], information
 
         try:
-            self.to_screen('Destination: ' + new_path)
-            self.run_ffmpeg(path, new_path, acodec, more_opts)
+            self.to_screen(f'Destination: {new_path}')
+            self.run_ffmpeg(path, temp_path, acodec, more_opts)
         except AudioConversionError as e:
             raise PostProcessingError(
                 'audio conversion failed: ' + e.msg)
         except Exception:
             raise PostProcessingError('error running ' + self.basename)
 
+        os.replace(path, orig_path)
+        os.replace(temp_path, new_path)
+        information['filepath'] = new_path
+        information['ext'] = extension
+
         # Try to update the date time for extracted audio file.
         if information.get('filetime') is not None:
             self.try_utime(
                 new_path, time.time(), information['filetime'],
                 errnote='Cannot update utime of audio file')
 
-        return [path], information
+        return [orig_path], information
 
 
 class FFmpegVideoConvertorPP(FFmpegPostProcessor):
     SUPPORTED_EXTS = ('mp4', 'mkv', 'flv', 'webm', 'mov', 'avi', 'mp3', 'mka', 'm4a', 'ogg', 'opus')
     FORMAT_RE = re.compile(r'{0}(?:/{0})*$'.format(r'(?:\w+>)?(?:%s)' % '|'.join(SUPPORTED_EXTS)))
-    _action = 'converting'
+    _ACTION = 'converting'
 
     def __init__(self, downloader=None, preferedformat=None):
         super(FFmpegVideoConvertorPP, self).__init__(downloader)
@@ -420,29 +538,28 @@ def _options(target_ext):
         return []
 
     @PostProcessor._restrict_to(images=False)
-    def run(self, information):
-        path, source_ext = information['filepath'], information['ext'].lower()
+    def run(self, info):
+        filename, source_ext = info['filepath'], info['ext'].lower()
         target_ext = self._target_ext(source_ext)
         _skip_msg = (
-            'could not find a mapping for %s' if not target_ext
-            else 'already is in target format %s' if source_ext == target_ext
+            f'could not find a mapping for {source_ext}' if not target_ext
+            else f'already is in target format {source_ext}' if source_ext == target_ext
             else None)
         if _skip_msg:
-            self.to_screen('Not %s media file "%s"; %s' % (self._action, path, _skip_msg % source_ext))
-            return [], information
+            self.to_screen(f'Not {self._ACTION} media file {filename!r}; {_skip_msg}')
+            return [], info
 
-        prefix, sep, oldext = path.rpartition('.')
-        outpath = prefix + sep + target_ext
-        self.to_screen('%s video from %s to %s; Destination: %s' % (self._action.title(), source_ext, target_ext, outpath))
-        self.run_ffmpeg(path, outpath, self._options(target_ext))
+        outpath = replace_extension(filename, target_ext, source_ext)
+        self.to_screen(f'{self._ACTION.title()} video from {source_ext} to {target_ext}; Destination: {outpath}')
+        self.run_ffmpeg(filename, outpath, self._options(target_ext))
 
-        information['filepath'] = outpath
-        information['format'] = information['ext'] = target_ext
-        return [path], information
+        info['filepath'] = outpath
+        info['format'] = info['ext'] = target_ext
+        return [filename], info
 
 
 class FFmpegVideoRemuxerPP(FFmpegVideoConvertorPP):
-    _action = 'remuxing'
+    _ACTION = 'remuxing'
 
     @staticmethod
     def _options(target_ext):
@@ -468,6 +585,10 @@ def run(self, information):
             return [], information
 
         filename = information['filepath']
+        if information.get('duration') and self._duration_mismatch(
+                self._get_real_video_duration(information, False), information['duration']):
+            self.to_screen(f'Skipping {self.pp_key()} since the real and expected durations mismatch')
+            return [], information
 
         ext = information['ext']
         sub_langs, sub_names, sub_filenames = [], [], []
@@ -475,6 +596,9 @@ def run(self, information):
         mp4_ass_warn = False
 
         for lang, sub_info in subtitles.items():
+            if not os.path.exists(sub_info.get('filepath', '')):
+                self.report_warning(f'Skipping embedding {lang} subtitle because the file is missing')
+                continue
             sub_ext = sub_info['ext']
             if sub_ext == 'json':
                 self.report_warning('JSON subtitles cannot be embedded')
@@ -517,8 +641,7 @@ def run(self, information):
         temp_filename = prepend_extension(filename, 'temp')
         self.to_screen('Embedding subtitles in "%s"' % filename)
         self.run_ffmpeg_multiple_files(input_files, temp_filename, opts)
-        os.remove(encodeFilename(filename))
-        os.rename(encodeFilename(temp_filename), encodeFilename(filename))
+        os.replace(temp_filename, filename)
 
         files_to_delete = [] if self._already_have_subtitle else sub_filenames
         return files_to_delete, information
@@ -526,6 +649,12 @@ def run(self, information):
 
 class FFmpegMetadataPP(FFmpegPostProcessor):
 
+    def __init__(self, downloader, add_metadata=True, add_chapters=True, add_infojson='if_exists'):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._add_metadata = add_metadata
+        self._add_chapters = add_chapters
+        self._add_infojson = add_infojson
+
     @staticmethod
     def _options(target_ext):
         yield from ('-map', '0', '-dn')
@@ -536,16 +665,65 @@ def _options(target_ext):
 
     @PostProcessor._restrict_to(images=False)
     def run(self, info):
+        filename, metadata_filename = info['filepath'], None
+        files_to_delete, options = [], []
+        if self._add_chapters and info.get('chapters'):
+            metadata_filename = replace_extension(filename, 'meta')
+            options.extend(self._get_chapter_opts(info['chapters'], metadata_filename))
+            files_to_delete.append(metadata_filename)
+        if self._add_metadata:
+            options.extend(self._get_metadata_opts(info))
+
+        if self._add_infojson:
+            if info['ext'] in ('mkv', 'mka'):
+                infojson_filename = info.get('infojson_filename')
+                options.extend(self._get_infojson_opts(info, infojson_filename))
+                if not infojson_filename:
+                    files_to_delete.append(info.get('infojson_filename'))
+            elif self._add_infojson is True:
+                self.to_screen('The info-json can only be attached to mkv/mka files')
+
+        if not options:
+            self.to_screen('There isn\'t any metadata to add')
+            return [], info
+
+        temp_filename = prepend_extension(filename, 'temp')
+        self.to_screen('Adding metadata to "%s"' % filename)
+        self.run_ffmpeg_multiple_files(
+            (filename, metadata_filename), temp_filename,
+            itertools.chain(self._options(info['ext']), *options))
+        for file in filter(None, files_to_delete):
+            os.remove(file)  # Don't obey --keep-files
+        os.replace(temp_filename, filename)
+        return [], info
+
+    @staticmethod
+    def _get_chapter_opts(chapters, metadata_filename):
+        with io.open(metadata_filename, 'wt', encoding='utf-8') as f:
+            def ffmpeg_escape(text):
+                return re.sub(r'([\\=;#\n])', r'\\\1', text)
+
+            metadata_file_content = ';FFMETADATA1\n'
+            for chapter in chapters:
+                metadata_file_content += '[CHAPTER]\nTIMEBASE=1/1000\n'
+                metadata_file_content += 'START=%d\n' % (chapter['start_time'] * 1000)
+                metadata_file_content += 'END=%d\n' % (chapter['end_time'] * 1000)
+                chapter_title = chapter.get('title')
+                if chapter_title:
+                    metadata_file_content += 'title=%s\n' % ffmpeg_escape(chapter_title)
+            f.write(metadata_file_content)
+        yield ('-map_metadata', '1')
+
+    def _get_metadata_opts(self, info):
         metadata = {}
+        meta_prefix = 'meta_'
 
         def add(meta_list, info_list=None):
-            if not meta_list:
-                return
-            for info_f in variadic(info_list or meta_list):
-                if isinstance(info.get(info_f), (compat_str, compat_numeric_types)):
-                    for meta_f in variadic(meta_list):
-                        metadata[meta_f] = info[info_f]
-                    break
+            value = next((
+                str(info[key]) for key in [meta_prefix] + list(variadic(info_list or meta_list))
+                if info.get(key) is not None), None)
+            if value not in ('', None):
+                metadata.update({meta_f: value for meta_f in variadic(meta_list)})
 
         # See [1-4] for some info on media metadata/metadata supported
         # by ffmpeg.
@@ -567,67 +745,46 @@ def add(meta_list, info_list=None):
         add('season_number')
         add('episode_id', ('episode', 'episode_id'))
         add('episode_sort', 'episode_number')
+        if 'embed-metadata' in self.get_param('compat_opts', []):
+            add('comment', 'description')
+            metadata.pop('synopsis', None)
 
-        prefix = 'meta_'
-        for key in filter(lambda k: k.startswith(prefix), info.keys()):
-            add(key[len(prefix):], key)
+        for key, value in info.items():
+            if value is not None and key != meta_prefix and key.startswith(meta_prefix):
+                metadata[key[len(meta_prefix):]] = value
 
-        filename, metadata_filename = info['filepath'], None
-        options = [('-metadata', f'{name}={value}') for name, value in metadata.items()]
+        for name, value in metadata.items():
+            yield ('-metadata', f'{name}={value}')
 
         stream_idx = 0
         for fmt in info.get('requested_formats') or []:
             stream_count = 2 if 'none' not in (fmt.get('vcodec'), fmt.get('acodec')) else 1
             if fmt.get('language'):
                 lang = ISO639Utils.short2long(fmt['language']) or fmt['language']
-                options.extend(('-metadata:s:%d' % (stream_idx + i), 'language=%s' % lang)
-                               for i in range(stream_count))
+                for i in range(stream_count):
+                    yield ('-metadata:s:%d' % (stream_idx + i), 'language=%s' % lang)
             stream_idx += stream_count
 
-        chapters = info.get('chapters', [])
-        if chapters:
-            metadata_filename = replace_extension(filename, 'meta')
-            with io.open(metadata_filename, 'wt', encoding='utf-8') as f:
-                def ffmpeg_escape(text):
-                    return re.sub(r'(=|;|#|\\|\n)', r'\\\1', text)
-
-                metadata_file_content = ';FFMETADATA1\n'
-                for chapter in chapters:
-                    metadata_file_content += '[CHAPTER]\nTIMEBASE=1/1000\n'
-                    metadata_file_content += 'START=%d\n' % (chapter['start_time'] * 1000)
-                    metadata_file_content += 'END=%d\n' % (chapter['end_time'] * 1000)
-                    chapter_title = chapter.get('title')
-                    if chapter_title:
-                        metadata_file_content += 'title=%s\n' % ffmpeg_escape(chapter_title)
-                f.write(metadata_file_content)
-                options.append(('-map_metadata', '1'))
-
-        if ('no-attach-info-json' not in self.get_param('compat_opts', [])
-                and '__infojson_filename' in info and info['ext'] in ('mkv', 'mka')):
-            old_stream, new_stream = self.get_stream_number(filename, ('tags', 'mimetype'), 'application/json')
-            if old_stream is not None:
-                options.append(('-map', '-0:%d' % old_stream))
-                new_stream -= 1
-
-            options.append((
-                '-attach', info['__infojson_filename'],
-                '-metadata:s:%d' % new_stream, 'mimetype=application/json'
-            ))
+    def _get_infojson_opts(self, info, infofn):
+        if not infofn or not os.path.exists(infofn):
+            if self._add_infojson is not True:
+                return
+            infofn = infofn or '%s.temp' % (
+                self._downloader.prepare_filename(info, 'infojson')
+                or replace_extension(self._downloader.prepare_filename(info), 'info.json', info['ext']))
+            if not self._downloader._ensure_dir_exists(infofn):
+                return
+            self.write_debug(f'Writing info-json to: {infofn}')
+            write_json_file(self._downloader.sanitize_info(info, self.get_param('clean_infojson', True)), infofn)
+            info['infojson_filename'] = infofn
 
-        if not options:
-            self.to_screen('There isn\'t any metadata to add')
-            return [], info
+        old_stream, new_stream = self.get_stream_number(info['filepath'], ('tags', 'mimetype'), 'application/json')
+        if old_stream is not None:
+            yield ('-map', '-0:%d' % old_stream)
+            new_stream -= 1
 
-        temp_filename = prepend_extension(filename, 'temp')
-        self.to_screen('Adding metadata to "%s"' % filename)
-        self.run_ffmpeg_multiple_files(
-            (filename, metadata_filename), temp_filename,
-            itertools.chain(self._options(info['ext']), *options))
-        if chapters:
-            os.remove(metadata_filename)
-        os.remove(encodeFilename(filename))
-        os.rename(encodeFilename(temp_filename), encodeFilename(filename))
-        return [], info
+        yield ('-attach', infofn,
+               '-metadata:s:%d' % new_stream, 'mimetype=application/json')
 
 
 class FFmpegMergerPP(FFmpegPostProcessor):
@@ -636,9 +793,14 @@ def run(self, info):
         filename = info['filepath']
         temp_filename = prepend_extension(filename, 'temp')
         args = ['-c', 'copy']
+        audio_streams = 0
         for (i, fmt) in enumerate(info['requested_formats']):
             if fmt.get('acodec') != 'none':
-                args.extend(['-map', '%u:a:0' % (i)])
+                args.extend(['-map', f'{i}:a:0'])
+                aac_fixup = fmt['protocol'].startswith('m3u8') and self.get_audio_codec(fmt['filepath']) == 'aac'
+                if aac_fixup:
+                    args.extend([f'-bsf:a:{audio_streams}', 'aac_adtstoasc'])
+                audio_streams += 1
             if fmt.get('vcodec') != 'none':
                 args.extend(['-map', '%u:v:0' % (i)])
         self.to_screen('Merging formats into "%s"' % filename)
@@ -670,8 +832,7 @@ def _fixup(self, msg, filename, options):
         self.to_screen(f'{msg} of "{filename}"')
         self.run_ffmpeg(filename, temp_filename, options)
 
-        os.remove(encodeFilename(filename))
-        os.rename(encodeFilename(temp_filename), encodeFilename(filename))
+        os.replace(temp_filename, filename)
 
 
 class FFmpegFixupStretchedPP(FFmpegFixupPostProcessor):
@@ -694,10 +855,21 @@ def run(self, info):
 
 
 class FFmpegFixupM3u8PP(FFmpegFixupPostProcessor):
+    def _needs_fixup(self, info):
+        yield info['ext'] in ('mp4', 'm4a')
+        yield info['protocol'].startswith('m3u8')
+        try:
+            metadata = self.get_metadata_object(info['filepath'])
+        except PostProcessingError as e:
+            self.report_warning(f'Unable to extract metadata: {e.msg}')
+            yield True
+        else:
+            yield traverse_obj(metadata, ('format', 'format_name'), casesense=False) == 'mpegts'
+
     @PostProcessor._restrict_to(images=False)
     def run(self, info):
-        if self.get_audio_codec(info['filepath']) == 'aac':
-            self._fixup('Fixing malformed AAC bitstream', info['filepath'], [
+        if all(self._needs_fixup(info)):
+            self._fixup('Fixing MPEG-TS in MP4 container', info['filepath'], [
                 '-c', 'copy', '-map', '0', '-dn', '-f', 'mp4', '-bsf:a', 'aac_adtstoasc'])
         return [], info
 
@@ -712,11 +884,10 @@ def __init__(self, downloader=None, trim=0.001):
 
     @PostProcessor._restrict_to(images=False)
     def run(self, info):
-        required_version = '4.4'
-        if is_outdated_version(self._versions[self.basename], required_version):
+        if not self._features.get('setts'):
             self.report_warning(
                 'A re-encode is needed to fix timestamps in older versions of ffmpeg. '
-                f'Please install ffmpeg {required_version} or later to fixup without re-encoding')
+                'Please install ffmpeg 4.4 or later to fixup without re-encoding')
             opts = ['-vf', 'setpts=PTS-STARTPTS']
         else:
             opts = ['-c', 'copy', '-bsf', 'setts=ts=TS-STARTPTS']
@@ -750,6 +921,9 @@ def run(self, info):
         self.to_screen('Converting subtitles')
         sub_filenames = []
         for lang, sub in subs.items():
+            if not os.path.exists(sub.get('filepath', '')):
+                self.report_warning(f'Skipping embedding {lang} subtitle because the file is missing')
+                continue
             ext = sub['ext']
             if ext == new_ext:
                 self.to_screen('Subtitle file for %s is already in the requested format' % new_ext)
@@ -799,12 +973,15 @@ def run(self, info):
                 }
 
             info['__files_to_move'][new_file] = replace_extension(
-                info['__files_to_move'][old_file], new_ext)
+                info['__files_to_move'][sub['filepath']], new_ext)
 
         return sub_filenames, info
 
 
 class FFmpegSplitChaptersPP(FFmpegPostProcessor):
+    def __init__(self, downloader, force_keyframes=False):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._force_keyframes = force_keyframes
 
     def _prepare_filename(self, number, chapter, info):
         info = info.copy()
@@ -832,13 +1009,18 @@ def _ffmpeg_args_for_chapter(self, number, chapter, info):
     def run(self, info):
         chapters = info.get('chapters') or []
         if not chapters:
-            self.report_warning('Chapter information is unavailable')
+            self.to_screen('Chapter information is unavailable')
             return [], info
 
+        in_file = info['filepath']
+        if self._force_keyframes and len(chapters) > 1:
+            in_file = self.force_keyframes(in_file, (c['start_time'] for c in chapters))
         self.to_screen('Splitting video by chapters; %d chapters found' % len(chapters))
         for idx, chapter in enumerate(chapters):
             destination, opts = self._ffmpeg_args_for_chapter(idx + 1, chapter, info)
-            self.real_run_ffmpeg([(info['filepath'], opts)], [(destination, ['-c', 'copy'])])
+            self.real_run_ffmpeg([(in_file, opts)], [(destination, ['-c', 'copy'])])
+        if in_file != info['filepath']:
+            os.remove(in_file)
         return [], info
 
 
@@ -863,9 +1045,7 @@ def fixup_webp(self, info, idx=-1):
             if thumbnail_ext != 'webp' and self.is_webp(thumbnail_filename):
                 self.to_screen('Correcting thumbnail "%s" extension to webp' % thumbnail_filename)
                 webp_filename = replace_extension(thumbnail_filename, 'webp')
-                if os.path.exists(webp_filename):
-                    os.remove(webp_filename)
-                os.rename(encodeFilename(thumbnail_filename), encodeFilename(webp_filename))
+                os.replace(thumbnail_filename, webp_filename)
                 info['thumbnails'][idx]['filepath'] = webp_filename
                 info['__files_to_move'][webp_filename] = replace_extension(
                     info['__files_to_move'].pop(thumbnail_filename), 'webp')