[ie] Support multi-period MPD streams (#6654)

yt_dlp/YoutubeDL.py

@@ -3483,7 +3483,8 @@ class YoutubeDL:
                                      or info_dict.get('is_live') and self.params.get('hls_use_mpegts') is None,
                                      'Possible MPEG-TS in MP4 container or malformed AAC timestamps',
                                      FFmpegFixupM3u8PP)
-                        ffmpeg_fixup(info_dict.get('is_live') and downloader == 'dashsegments',
+                        ffmpeg_fixup(downloader == 'dashsegments'
+                                     and (info_dict.get('is_live') or info_dict.get('is_dash_periods')),
                                      'Possible duplicate MOOV atoms', FFmpegFixupDuplicateMoovPP)
 
                     ffmpeg_fixup(downloader == 'web_socket_fragment', 'Malformed timestamps detected', FFmpegFixupTimestampPP)

yt_dlp/extractor/common.py

@@ -247,6 +247,8 @@ class InfoExtractor:
                                  (For internal use only)
                                  * http_chunk_size Chunk size for HTTP downloads
                                  * ffmpeg_args     Extra arguments for ffmpeg downloader
+                    * is_dash_periods  Whether the format is a result of merging
+                                 multiple DASH periods.
                     RTMP formats can also have the additional fields: page_url,
                     app, play_path, tc_url, flash_version, rtmp_live, rtmp_conn,
                     rtmp_protocol, rtmp_real_time
@@ -2530,7 +2532,11 @@ class InfoExtractor:
             self._report_ignoring_subs('DASH')
         return fmts
 
-    def _extract_mpd_formats_and_subtitles(
+    def _extract_mpd_formats_and_subtitles(self, *args, **kwargs):
+        periods = self._extract_mpd_periods(*args, **kwargs)
+        return self._merge_mpd_periods(periods)
+
+    def _extract_mpd_periods(
             self, mpd_url, video_id, mpd_id=None, note=None, errnote=None,
             fatal=True, data=None, headers={}, query={}):
 
@@ -2543,17 +2549,16 @@ class InfoExtractor:
             errnote='Failed to download MPD manifest' if errnote is None else errnote,
             fatal=fatal, data=data, headers=headers, query=query)
         if res is False:
-            return [], {}
+            return []
         mpd_doc, urlh = res
         if mpd_doc is None:
-            return [], {}
+            return []
 
         # We could have been redirected to a new url when we retrieved our mpd file.
         mpd_url = urlh.url
         mpd_base_url = base_url(mpd_url)
 
-        return self._parse_mpd_formats_and_subtitles(
+        return self._parse_mpd_periods(mpd_doc, mpd_id, mpd_base_url, mpd_url)
-            mpd_doc, mpd_id, mpd_base_url, mpd_url)
 
     def _parse_mpd_formats(self, *args, **kwargs):
         fmts, subs = self._parse_mpd_formats_and_subtitles(*args, **kwargs)
@@ -2561,8 +2566,39 @@ class InfoExtractor:
             self._report_ignoring_subs('DASH')
         return fmts
 
-    def _parse_mpd_formats_and_subtitles(
+    def _parse_mpd_formats_and_subtitles(self, *args, **kwargs):
-            self, mpd_doc, mpd_id=None, mpd_base_url='', mpd_url=None):
+        periods = self._parse_mpd_periods(*args, **kwargs)
+        return self._merge_mpd_periods(periods)
+
+    def _merge_mpd_periods(self, periods):
+        """
+        Combine all formats and subtitles from an MPD manifest into a single list,
+        by concatenate streams with similar formats.
+        """
+        formats, subtitles = {}, {}
+        for period in periods:
+            for f in period['formats']:
+                assert 'is_dash_periods' not in f, 'format already processed'
+                f['is_dash_periods'] = True
+                format_key = tuple(v for k, v in f.items() if k not in (
+                    ('format_id', 'fragments', 'manifest_stream_number')))
+                if format_key not in formats:
+                    formats[format_key] = f
+                elif 'fragments' in f:
+                    formats[format_key].setdefault('fragments', []).extend(f['fragments'])
+
+            if subtitles and period['subtitles']:
+                self.report_warning(bug_reports_message(
+                    'Found subtitles in multiple periods in the DASH manifest; '
+                    'if part of the subtitles are missing,'
+                ), only_once=True)
+
+            for sub_lang, sub_info in period['subtitles'].items():
+                subtitles.setdefault(sub_lang, []).extend(sub_info)
+
+        return list(formats.values()), subtitles
+
+    def _parse_mpd_periods(self, mpd_doc, mpd_id=None, mpd_base_url='', mpd_url=None):
         """
         Parse formats from MPD manifest.
         References:
@@ -2641,9 +2677,13 @@ class InfoExtractor:
             return ms_info
 
         mpd_duration = parse_duration(mpd_doc.get('mediaPresentationDuration'))
-        formats, subtitles = [], {}
         stream_numbers = collections.defaultdict(int)
-        for period in mpd_doc.findall(_add_ns('Period')):
+        for period_idx, period in enumerate(mpd_doc.findall(_add_ns('Period'))):
+            period_entry = {
+                'id': period.get('id', f'period-{period_idx}'),
+                'formats': [],
+                'subtitles': collections.defaultdict(list),
+            }
             period_duration = parse_duration(period.get('duration')) or mpd_duration
             period_ms_info = extract_multisegment_info(period, {
                 'start_number': 1,
@@ -2893,11 +2933,10 @@ class InfoExtractor:
                     if content_type in ('video', 'audio', 'image/jpeg'):
                         f['manifest_stream_number'] = stream_numbers[f['url']]
                         stream_numbers[f['url']] += 1
-                        formats.append(f)
+                        period_entry['formats'].append(f)
                     elif content_type == 'text':
-                        subtitles.setdefault(lang or 'und', []).append(f)
+                        period_entry['subtitles'][lang or 'und'].append(f)
-
+            yield period_entry
-        return formats, subtitles
 
     def _extract_ism_formats(self, *args, **kwargs):
         fmts, subs = self._extract_ism_formats_and_subtitles(*args, **kwargs)