[fd/hls] Do not fall back to ffmpeg when native is required (#13655)

Authored by: bashonly

yt_dlp/downloader/hls.py

@@ -94,12 +94,19 @@ class HlsFD(FragmentFD):
         can_download, message = self.can_download(s, info_dict, self.params.get('allow_unplayable_formats')), None
         if can_download:
             has_ffmpeg = FFmpegFD.available()
-            no_crypto = not Cryptodome.AES and '#EXT-X-KEY:METHOD=AES-128' in s
-            if no_crypto and has_ffmpeg:
-                can_download, message = False, 'The stream has AES-128 encryption and pycryptodomex is not available'
-            elif no_crypto:
-                message = ('The stream has AES-128 encryption and neither ffmpeg nor pycryptodomex are available; '
-                           'Decryption will be performed natively, but will be extremely slow')
+            if not Cryptodome.AES and '#EXT-X-KEY:METHOD=AES-128' in s:
+                # Even if pycryptodomex isn't available, force HlsFD for m3u8s that won't work with ffmpeg
+                ffmpeg_can_dl = not traverse_obj(info_dict, ((
+                    'extra_param_to_segment_url', 'extra_param_to_key_url',
+                    'hls_media_playlist_data', ('hls_aes', ('uri', 'key', 'iv')),
+                ), any))
+                message = 'The stream has AES-128 encryption and {} available'.format(
+                    'neither ffmpeg nor pycryptodomex are' if ffmpeg_can_dl and not has_ffmpeg else
+                    'pycryptodomex is not')
+                if has_ffmpeg and ffmpeg_can_dl:
+                    can_download = False
+                else:
+                    message += '; decryption will be performed natively, but will be extremely slow'
             elif info_dict.get('extractor_key') == 'Generic' and re.search(r'(?m)#EXT-X-MEDIA-SEQUENCE:(?!0$)', s):
                 install_ffmpeg = '' if has_ffmpeg else 'install ffmpeg and '
                 message = ('Live HLS streams are not supported by the native downloader. If this is a livestream, '

yt_dlp/extractor/twitch.py

@@ -6,6 +6,7 @@ import re
 import urllib.parse
 
 from .common import InfoExtractor
+from ..networking.exceptions import HTTPError
 from ..utils import (
     ExtractorError,
     UserNotLive,
@@ -188,19 +189,39 @@ class TwitchBaseIE(InfoExtractor):
         }] if thumbnail else None
 
     def _extract_twitch_m3u8_formats(self, path, video_id, token, signature, live_from_start=False):
-        formats = self._extract_m3u8_formats(
-            f'{self._USHER_BASE}/{path}/{video_id}.m3u8', video_id, 'mp4', query={
-                'allow_source': 'true',
-                'allow_audio_only': 'true',
-                'allow_spectre': 'true',
-                'p': random.randint(1000000, 10000000),
-                'platform': 'web',
-                'player': 'twitchweb',
-                'supported_codecs': 'av1,h265,h264',
-                'playlist_include_framerate': 'true',
-                'sig': signature,
-                'token': token,
-            })
+        try:
+            formats = self._extract_m3u8_formats(
+                f'{self._USHER_BASE}/{path}/{video_id}.m3u8', video_id, 'mp4', query={
+                    'allow_source': 'true',
+                    'allow_audio_only': 'true',
+                    'allow_spectre': 'true',
+                    'p': random.randint(1000000, 10000000),
+                    'platform': 'web',
+                    'player': 'twitchweb',
+                    'supported_codecs': 'av1,h265,h264',
+                    'playlist_include_framerate': 'true',
+                    'sig': signature,
+                    'token': token,
+                })
+        except ExtractorError as e:
+            if (
+                not isinstance(e.cause, HTTPError)
+                or e.cause.status != 403
+                or e.cause.response.get_header('content-type') != 'application/json'
+            ):
+                raise
+
+            error_info = traverse_obj(e.cause.response.read(), ({json.loads}, 0, {dict})) or {}
+            if error_info.get('error_code') in ('vod_manifest_restricted', 'unauthorized_entitlements'):
+                common_msg = 'access to this subscriber-only content'
+                if self._get_cookies('https://gql.twitch.tv').get('auth-token'):
+                    raise ExtractorError(f'Your account does not have {common_msg}', expected=True)
+                self.raise_login_required(f'You must be logged into an account that has {common_msg}')
+
+            if error_msg := join_nonempty('error_code', 'error', from_dict=error_info, delim=': '):
+                raise ExtractorError(error_msg, expected=True)
+            raise
+
         for fmt in formats:
             if fmt.get('vcodec') and fmt['vcodec'].startswith('av01'):
                 # mpegts does not yet have proper support for av1

yt_dlp/extractor/youtube/_video.py

@@ -3978,7 +3978,9 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         def process_language(container, base_url, lang_code, sub_name, client_name, query):
             lang_subs = container.setdefault(lang_code, [])
             for fmt in self._SUBTITLE_FORMATS:
-                query = {**query, 'fmt': fmt}
+                # xosf=1 results in undesirable text position data for vtt, json3 & srv* subtitles
+                # See: https://github.com/yt-dlp/yt-dlp/issues/13654
+                query = {**query, 'fmt': fmt, 'xosf': []}
                 lang_subs.append({
                     'ext': fmt,
                     'url': urljoin('https://www.youtube.com', update_url_query(base_url, query)),