Merge 670bafe148 into f2a4983df7

Closes #11526
Authored by: jshumphrey

test/test_utils.py

@@ -456,10 +456,15 @@ class TestUtil(unittest.TestCase):
         self.assertEqual(unified_timestamp('2018-03-14T08:32:43.1493874+00:00'), 1521016363)
         self.assertEqual(unified_timestamp('Sunday, 26 Nov 2006, 19:00'), 1164567600)
         self.assertEqual(unified_timestamp('wed, aug 16, 2008, 12:00pm'), 1218931200)
+        self.assertEqual(unified_timestamp('2022-10-13T02:37:47.831Z'), 1665628667)
 
         self.assertEqual(unified_timestamp('December 31 1969 20:00:01 EDT'), 1)
         self.assertEqual(unified_timestamp('Wednesday 31 December 1969 18:01:26 MDT'), 86)
         self.assertEqual(unified_timestamp('12/31/1969 20:01:18 EDT', False), 78)
+        self.assertEqual(unified_timestamp('2023-03-09T18:01:33.646Z', with_milliseconds=True), 1678384893.646)
+        # ISO8601 spec says that if no timezone is specified, we should use local timezone;
+        # but yt-dlp uses UTC to keep things consistent
+        self.assertEqual(unified_timestamp('2023-03-11T06:48:34.008'), 1678517314)
 
     def test_determine_ext(self):
         self.assertEqual(determine_ext('http://example.com/foo/bar.mp4/?download'), 'mp4')

yt_dlp/YoutubeDL.py

@@ -28,7 +28,12 @@ from .cache import Cache
 from .compat import urllib  # isort: split
 from .compat import compat_os_name, urllib_req_to_req
 from .cookies import CookieLoadError, LenientSimpleCookie, load_cookies
-from .downloader import FFmpegFD, get_suitable_downloader, shorten_protocol_name
+from .downloader import (
+    DashSegmentsFD,
+    FFmpegFD,
+    get_suitable_downloader,
+    shorten_protocol_name,
+)
 from .downloader.rtmp import rtmpdump_version
 from .extractor import gen_extractor_classes, get_info_extractor
 from .extractor.common import UnsupportedURLIE
@@ -3373,7 +3378,7 @@ class YoutubeDL:
                 fd, success = None, True
                 if info_dict.get('protocol') or info_dict.get('url'):
                     fd = get_suitable_downloader(info_dict, self.params, to_stdout=temp_filename == '-')
-                    if fd != FFmpegFD and 'no-direct-merge' not in self.params['compat_opts'] and (
+                    if fd not in [FFmpegFD, DashSegmentsFD] and 'no-direct-merge' not in self.params['compat_opts'] and (
                             info_dict.get('section_start') or info_dict.get('section_end')):
                         msg = ('This format cannot be partially downloaded' if FFmpegFD.available()
                                else 'You have requested downloading the video partially, but ffmpeg is not installed')

yt_dlp/__init__.py

@@ -12,6 +12,7 @@ import itertools
 import optparse
 import os
 import re
+import time
 import traceback
 
 from .compat import compat_os_name
@@ -342,12 +343,13 @@ def validate_options(opts):
             (?P<end_sign>-?)(?P<end>[^-]+)
         )?'''
 
+        current_time = time.time()
         chapters, ranges, from_url = [], [], False
         for regex in value or []:
             if advanced and regex == '*from-url':
                 from_url = True
                 continue
-            elif not regex.startswith('*'):
+            elif not regex.startswith('*') and not regex.startswith('#'):
                 try:
                     chapters.append(re.compile(regex))
                 except re.error as err:
@@ -364,11 +366,16 @@ def validate_options(opts):
                     err = 'Must be of the form "*start-end"'
                 elif not advanced and any(signs):
                     err = 'Negative timestamps are not allowed'
-                else:
+                elif regex.startswith('*'):
                     dur[0] *= -1 if signs[0] else 1
                     dur[1] *= -1 if signs[1] else 1
                     if dur[1] == float('-inf'):
                         err = '"-inf" is not a valid end'
+                elif regex.startswith('#'):
+                    dur[0] = dur[0] * (-1 if signs[0] else 1) + current_time
+                    dur[1] = dur[1] * (-1 if signs[1] else 1) + current_time
+                    if dur[1] == float('-inf'):
+                        err = '"-inf" is not a valid end'
                 if err:
                     raise ValueError(f'invalid {name} time range "{regex}". {err}')
                 ranges.append(dur)

yt_dlp/downloader/dash.py

@@ -36,6 +36,8 @@ class DashSegmentsFD(FragmentFD):
                 'filename': fmt.get('filepath') or filename,
                 'live': 'is_from_start' if fmt.get('is_from_start') else fmt.get('is_live'),
                 'total_frags': fragment_count,
+                'section_start': info_dict.get('section_start'),
+                'section_end': info_dict.get('section_end'),
             }
 
             if real_downloader:

yt_dlp/extractor/archiveorg.py

@@ -205,6 +205,26 @@ class ArchiveOrgIE(InfoExtractor):
                 },
             },
         ],
+    }, {
+        # The reviewbody is None for one of the reviews; just need to extract data without crashing
+        'url': 'https://archive.org/details/gd95-04-02.sbd.11622.sbeok.shnf/gd95-04-02d1t04.shn',
+        'info_dict': {
+            'id': 'gd95-04-02.sbd.11622.sbeok.shnf/gd95-04-02d1t04.shn',
+            'ext': 'mp3',
+            'title': 'Stuck Inside of Mobile with the Memphis Blues Again',
+            'creators': ['Grateful Dead'],
+            'duration': 338.31,
+            'track': 'Stuck Inside of Mobile with the Memphis Blues Again',
+            'description': 'md5:764348a470b986f1217ffd38d6ac7b72',
+            'display_id': 'gd95-04-02d1t04.shn',
+            'location': 'Pyramid Arena',
+            'uploader': 'jon@archive.org',
+            'album': '1995-04-02 - Pyramid Arena',
+            'upload_date': '20040519',
+            'track_number': 4,
+            'release_date': '19950402',
+            'timestamp': 1084927901,
+        },
     }]
 
     @staticmethod
@@ -335,7 +355,7 @@ class ArchiveOrgIE(InfoExtractor):
                 info['comments'].append({
                     'id': review.get('review_id'),
                     'author': review.get('reviewer'),
-                    'text': str_or_none(review.get('reviewtitle'), '') + '\n\n' + review.get('reviewbody'),
+                    'text': join_nonempty('reviewtitle', 'reviewbody', from_dict=review, delim='\n\n'),
                     'timestamp': unified_timestamp(review.get('createdate')),
                     'parent': 'root'})
 

yt_dlp/extractor/common.py

@@ -2733,7 +2733,7 @@ class InfoExtractor:
                             r = int(s.get('r', 0))
                             ms_info['total_number'] += 1 + r
                             ms_info['s'].append({
-                                't': int(s.get('t', 0)),
+                                't': int_or_none(s.get('t')),
                                 # @d is mandatory (see [1, 5.3.9.6.2, Table 17, page 60])
                                 'd': int(s.attrib['d']),
                                 'r': r,
@@ -2775,8 +2775,14 @@ class InfoExtractor:
             return ms_info
 
         mpd_duration = parse_duration(mpd_doc.get('mediaPresentationDuration'))
+        availability_start_time = unified_timestamp(
+            mpd_doc.get('availabilityStartTime'), with_milliseconds=True) or 0
         stream_numbers = collections.defaultdict(int)
         for period_idx, period in enumerate(mpd_doc.findall(_add_ns('Period'))):
+            # segmentIngestTime is completely out of spec, but YT Livestream do this
+            segment_ingest_time = period.get('{http://youtube.com/yt/2012/10/10}segmentIngestTime')
+            if segment_ingest_time:
+                availability_start_time = unified_timestamp(segment_ingest_time, with_milliseconds=True)
             period_entry = {
                 'id': period.get('id', f'period-{period_idx}'),
                 'formats': [],
@@ -2955,13 +2961,17 @@ class InfoExtractor:
                                     'Bandwidth': bandwidth,
                                     'Number': segment_number,
                                 }
+                                duration = float_or_none(segment_d, representation_ms_info['timescale'])
+                                start = float_or_none(segment_time, representation_ms_info['timescale'])
                                 representation_ms_info['fragments'].append({
                                     media_location_key: segment_url,
-                                    'duration': float_or_none(segment_d, representation_ms_info['timescale']),
+                                    'duration': duration,
+                                    'start': availability_start_time + start,
+                                    'end': availability_start_time + start + duration,
                                 })
 
                             for s in representation_ms_info['s']:
-                                segment_time = s.get('t') or segment_time
+                                segment_time = s['t'] if s.get('t') is not None else segment_time
                                 segment_d = s['d']
                                 add_segment_url()
                                 segment_number += 1
@@ -2977,6 +2987,7 @@ class InfoExtractor:
                         fragments = []
                         segment_index = 0
                         timescale = representation_ms_info['timescale']
+                        start = 0
                         for s in representation_ms_info['s']:
                             duration = float_or_none(s['d'], timescale)
                             for _ in range(s.get('r', 0) + 1):
@@ -2984,8 +2995,11 @@ class InfoExtractor:
                                 fragments.append({
                                     location_key(segment_uri): segment_uri,
                                     'duration': duration,
+                                    'start': availability_start_time + start,
+                                    'end': availability_start_time + start + duration,
                                 })
                                 segment_index += 1
+                                start += duration
                         representation_ms_info['fragments'] = fragments
                     elif 'segment_urls' in representation_ms_info:
                         # Segment URLs with no SegmentTimeline

yt_dlp/extractor/facebook.py

@@ -563,13 +563,13 @@ class FacebookIE(InfoExtractor):
                 return extract_video_data(try_get(
                     js_data, lambda x: x['jsmods']['instances'], list) or [])
 
-        def extract_dash_manifest(video, formats):
+        def extract_dash_manifest(vid_data, formats, mpd_url=None):
             dash_manifest = traverse_obj(
-                video, 'dash_manifest', 'playlist', 'dash_manifest_xml_string', expected_type=str)
+                vid_data, 'dash_manifest', 'playlist', 'dash_manifest_xml_string', 'manifest_xml', expected_type=str)
             if dash_manifest:
                 formats.extend(self._parse_mpd_formats(
                     compat_etree_fromstring(urllib.parse.unquote_plus(dash_manifest)),
-                    mpd_url=url_or_none(video.get('dash_manifest_url'))))
+                    mpd_url=url_or_none(video.get('dash_manifest_url')) or mpd_url))
 
         def process_formats(info):
             # Downloads with browser's User-Agent are rate limited. Working around
@@ -619,9 +619,12 @@ class FacebookIE(InfoExtractor):
                         video = video['creation_story']
                         video['owner'] = traverse_obj(video, ('short_form_video_context', 'video_owner'))
                         video.update(reel_info)
-                    fmt_data = traverse_obj(video, ('videoDeliveryLegacyFields', {dict})) or video
+
                     formats = []
                     q = qualities(['sd', 'hd'])
+
+                    # Legacy formats extraction
+                    fmt_data = traverse_obj(video, ('videoDeliveryLegacyFields', {dict})) or video
                     for key, format_id in (('playable_url', 'sd'), ('playable_url_quality_hd', 'hd'),
                                            ('playable_url_dash', ''), ('browser_native_hd_url', 'hd'),
                                            ('browser_native_sd_url', 'sd')):
@@ -629,7 +632,7 @@ class FacebookIE(InfoExtractor):
                         if not playable_url:
                             continue
                         if determine_ext(playable_url) == 'mpd':
-                            formats.extend(self._extract_mpd_formats(playable_url, video_id))
+                            formats.extend(self._extract_mpd_formats(playable_url, video_id, fatal=False))
                         else:
                             formats.append({
                                 'format_id': format_id,
@@ -638,6 +641,28 @@ class FacebookIE(InfoExtractor):
                                 'url': playable_url,
                             })
                     extract_dash_manifest(fmt_data, formats)
+
+                    # New videoDeliveryResponse formats extraction
+                    fmt_data = traverse_obj(video, ('videoDeliveryResponseFragment', 'videoDeliveryResponseResult'))
+                    mpd_urls = traverse_obj(fmt_data, ('dash_manifest_urls', ..., 'manifest_url', {url_or_none}))
+                    dash_manifests = traverse_obj(fmt_data, ('dash_manifests', lambda _, v: v['manifest_xml']))
+                    for idx, dash_manifest in enumerate(dash_manifests):
+                        extract_dash_manifest(dash_manifest, formats, mpd_url=traverse_obj(mpd_urls, idx))
+                    if not dash_manifests:
+                        # Only extract from MPD URLs if the manifests are not already provided
+                        for mpd_url in mpd_urls:
+                            formats.extend(self._extract_mpd_formats(mpd_url, video_id, fatal=False))
+                    for prog_fmt in traverse_obj(fmt_data, ('progressive_urls', lambda _, v: v['progressive_url'])):
+                        format_id = traverse_obj(prog_fmt, ('metadata', 'quality', {str.lower}))
+                        formats.append({
+                            'format_id': format_id,
+                            # sd, hd formats w/o resolution info should be deprioritized below DASH
+                            'quality': q(format_id) - 3,
+                            'url': prog_fmt['progressive_url'],
+                        })
+                    for m3u8_url in traverse_obj(fmt_data, ('hls_playlist_urls', ..., 'hls_playlist_url', {url_or_none})):
+                        formats.extend(self._extract_m3u8_formats(m3u8_url, video_id, 'mp4', fatal=False, m3u8_id='hls'))
+
                     if not formats:
                         # Do not append false positive entry w/o any formats
                         return

yt_dlp/extractor/youtube.py

@@ -3078,17 +3078,17 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             microformats = traverse_obj(
                 prs, (..., 'microformat', 'playerMicroformatRenderer'),
                 expected_type=dict)
-            _, live_status, _, formats, _ = self._list_formats(video_id, microformats, video_details, prs, player_url)
-            is_live = live_status == 'is_live'
-            start_time = time.time()
+            with lock:
+                _, live_status, _, formats, _ = self._list_formats(video_id, microformats, video_details, prs, player_url)
+                is_live = live_status == 'is_live'
+                start_time = time.time()
 
         def mpd_feed(format_id, delay):
             """
             @returns (manifest_url, manifest_stream_number, is_live) or None
             """
             for retry in self.RetryManager(fatal=False):
-                with lock:
-                    refetch_manifest(format_id, delay)
+                refetch_manifest(format_id, delay)
 
                 f = next((f for f in formats if f['format_id'] == format_id), None)
                 if not f:
@@ -3119,6 +3119,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         begin_index = 0
         download_start_time = ctx.get('start') or time.time()
 
+        section_start = ctx.get('section_start') or 0
+        section_end = ctx.get('section_end') or math.inf
+
+        self.write_debug(f'Selected section: {section_start} -> {section_end}')
+
         lack_early_segments = download_start_time - (live_start_time or download_start_time) > MAX_DURATION
         if lack_early_segments:
             self.report_warning(bug_reports_message(
@@ -3139,9 +3144,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                                                or (mpd_url, stream_number, False))
             if not refresh_sequence:
                 if expire_fast and not is_live:
-                    return False, last_seq
+                    return False
                 elif old_mpd_url == mpd_url:
-                    return True, last_seq
+                    return True
+
             if manifestless_orig_fmt:
                 fmt_info = manifestless_orig_fmt
             else:
@@ -3152,14 +3158,13 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     fmts = None
                 if not fmts:
                     no_fragment_score += 2
-                    return False, last_seq
+                    return False
                 fmt_info = next(x for x in fmts if x['manifest_stream_number'] == stream_number)
             fragments = fmt_info['fragments']
             fragment_base_url = fmt_info['fragment_base_url']
             assert fragment_base_url
 
-            _last_seq = int(re.search(r'(?:/|^)sq/(\d+)', fragments[-1]['path']).group(1))
-            return True, _last_seq
+            return True
 
         self.write_debug(f'[{video_id}] Generating fragments for format {format_id}')
         while is_live:
@@ -3179,11 +3184,19 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     last_segment_url = None
                     continue
             else:
-                should_continue, last_seq = _extract_sequence_from_mpd(True, no_fragment_score > 15)
+                should_continue = _extract_sequence_from_mpd(True, no_fragment_score > 15)
                 no_fragment_score += 2
                 if not should_continue:
                     continue
 
+            last_fragment = fragments[-1]
+            last_seq = int(re.search(r'(?:/|^)sq/(\d+)', fragments[-1]['path']).group(1))
+
+            known_fragment = next(
+                (fragment for fragment in fragments if f'sq/{known_idx}' in fragment['path']), None)
+            if known_fragment and known_fragment['end'] > section_end:
+                break
+
             if known_idx > last_seq:
                 last_segment_url = None
                 continue
@@ -3193,20 +3206,36 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             if begin_index < 0 and known_idx < 0:
                 # skip from the start when it's negative value
                 known_idx = last_seq + begin_index
+
             if lack_early_segments:
-                known_idx = max(known_idx, last_seq - int(MAX_DURATION // fragments[-1]['duration']))
+                known_idx = max(known_idx, last_seq - int(MAX_DURATION // last_fragment['duration']))
+
+            fragment_count = last_seq - known_idx if section_end == math.inf else int(
+                (section_end - section_start) // last_fragment['duration'])
+
             try:
                 for idx in range(known_idx, last_seq):
                     # do not update sequence here or you'll get skipped some part of it
-                    should_continue, _ = _extract_sequence_from_mpd(False, False)
+                    should_continue = _extract_sequence_from_mpd(False, False)
                     if not should_continue:
                         known_idx = idx - 1
                         raise ExtractorError('breaking out of outer loop')
-                    last_segment_url = urljoin(fragment_base_url, f'sq/{idx}')
-                    yield {
-                        'url': last_segment_url,
-                        'fragment_count': last_seq,
-                    }
+
+                    frag_duration = last_fragment['duration']
+                    frag_start = last_fragment['start'] - (last_seq - idx) * frag_duration
+                    frag_end = frag_start + frag_duration
+
+                    if frag_start >= section_start and frag_end <= section_end:
+                        last_segment_url = urljoin(fragment_base_url, f'sq/{idx}')
+
+                        yield {
+                            'url': last_segment_url,
+                            'fragment_count': fragment_count,
+                            'duration': frag_duration,
+                            'start': frag_start,
+                            'end': frag_end,
+                        }
+
                 if known_idx == last_seq:
                     no_fragment_score += 5
                 else:
@@ -4380,6 +4409,9 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 dct['downloader_options'] = {'http_chunk_size': CHUNK_SIZE}
                 yield dct
 
+        if live_status == 'is_live' and self.get_param('download_ranges') and not self.get_param('live_from_start'):
+            self.report_warning('For YT livestreams, --download-sections is only supported with --live-from-start')
+
         needs_live_processing = self._needs_live_processing(live_status, duration)
         skip_bad_formats = 'incomplete' not in format_types
         if self._configuration_arg('include_incomplete_formats'):

yt_dlp/options.py

@@ -427,7 +427,14 @@ def create_parser():
     general.add_option(
         '--live-from-start',
         action='store_true', dest='live_from_start',
-        help='Download livestreams from the start. Currently only supported for YouTube (Experimental)')
+        help=('Download livestreams from the start. Currently only supported for YouTube (Experimental). '
+              'Time ranges can be specified using --download-sections to download only a part of the stream. '
+              'Negative values are allowed for specifying a relative previous time, using the # syntax '
+              'e.g. --download-sections "#-24hours - 0" (download last 24 hours), '
+              'e.g. --download-sections "#-1h - 30m" (download from 1 hour ago until the next 30 minutes), '
+              'e.g. --download-sections "#-3days - -2days" (download from 3 days ago until 2 days ago). '
+              'It is also possible to specify an exact unix timestamp range, using the * syntax, '
+              'e.g. --download-sections "*1672531200 - 1672549200" (download between those two timestamps)'))
     general.add_option(
         '--no-live-from-start',
         action='store_false', dest='live_from_start',

yt_dlp/utils/_utils.py

@@ -1254,7 +1254,7 @@ def unified_strdate(date_str, day_first=True):
         return str(upload_date)
 
 
-def unified_timestamp(date_str, day_first=True):
+def unified_timestamp(date_str, day_first=True, with_milliseconds=False):
     if not isinstance(date_str, str):
         return None
 
@@ -1280,7 +1280,7 @@ def unified_timestamp(date_str, day_first=True):
     for expression in date_formats(day_first):
         with contextlib.suppress(ValueError):
             dt_ = dt.datetime.strptime(date_str, expression) - timezone + dt.timedelta(hours=pm_delta)
-            return calendar.timegm(dt_.timetuple())
+            return calendar.timegm(dt_.timetuple()) + (dt_.microsecond / 1e6 if with_milliseconds else 0)
 
     timetuple = email.utils.parsedate_tz(date_str)
     if timetuple:
@@ -2090,16 +2090,19 @@ def parse_duration(s):
 
     days, hours, mins, secs, ms = [None] * 5
     m = re.match(r'''(?x)
+            (?P<sign>[+-])?
             (?P<before_secs>
                 (?:(?:(?P<days>[0-9]+):)?(?P<hours>[0-9]+):)?(?P<mins>[0-9]+):)?
             (?P<secs>(?(before_secs)[0-9]{1,2}|[0-9]+))
             (?P<ms>[.:][0-9]+)?Z?$
         ''', s)
     if m:
-        days, hours, mins, secs, ms = m.group('days', 'hours', 'mins', 'secs', 'ms')
+        sign, days, hours, mins, secs, ms = m.group('sign', 'days', 'hours', 'mins', 'secs', 'ms')
     else:
         m = re.match(
-            r'''(?ix)(?:P?
+            r'''(?ix)(?:
+                (?P<sign>[+-])?
+                P?
                 (?:
                     [0-9]+\s*y(?:ears?)?,?\s*
                 )?
@@ -2123,17 +2126,19 @@ def parse_duration(s):
                     (?P<secs>[0-9]+)(?P<ms>\.[0-9]+)?\s*s(?:ec(?:ond)?s?)?\s*
                 )?Z?$''', s)
         if m:
-            days, hours, mins, secs, ms = m.groups()
+            sign, days, hours, mins, secs, ms = m.groups()
         else:
-            m = re.match(r'(?i)(?:(?P<hours>[0-9.]+)\s*(?:hours?)|(?P<mins>[0-9.]+)\s*(?:mins?\.?|minutes?)\s*)Z?$', s)
+            m = re.match(r'(?i)(?P<sign>[+-])?(?:(?P<days>[0-9.]+)\s*(?:days?)|(?P<hours>[0-9.]+)\s*(?:hours?)|(?P<mins>[0-9.]+)\s*(?:mins?\.?|minutes?)\s*)Z?$', s)
             if m:
-                hours, mins = m.groups()
+                sign, days, hours, mins = m.groups()
             else:
                 return None
 
+    sign = -1 if sign == '-' else 1
+
     if ms:
         ms = ms.replace(':', '.')
-    return sum(float(part or 0) * mult for part, mult in (
+    return sign * sum(float(part or 0) * mult for part, mult in (
         (days, 86400), (hours, 3600), (mins, 60), (secs, 1), (ms, 1)))