From a4c72acc462668a938827370bd77084a1cd4733b Mon Sep 17 00:00:00 2001 From: sepro Date: Wed, 19 Nov 2025 00:52:55 +0100 Subject: [PATCH] [ie/MedalTV] Rework extractor (#15103) Closes #15102 Authored by: seproDev --- yt_dlp/extractor/medaltv.py | 170 ++++++++++++++---------------------- 1 file changed, 67 insertions(+), 103 deletions(-) diff --git a/yt_dlp/extractor/medaltv.py b/yt_dlp/extractor/medaltv.py index 94c51ed0e7..d294d8d881 100644 --- a/yt_dlp/extractor/medaltv.py +++ b/yt_dlp/extractor/medaltv.py @@ -1,14 +1,9 @@ -import re - from .common import InfoExtractor from ..utils import ( - ExtractorError, - float_or_none, - format_field, int_or_none, - str_or_none, - traverse_obj, + url_or_none, ) +from ..utils.traversal import traverse_obj class MedalTVIE(InfoExtractor): @@ -30,25 +25,8 @@ class MedalTVIE(InfoExtractor): 'view_count': int, 'like_count': int, 'duration': 13, - }, - }, { - 'url': 'https://medal.tv/games/cod-cold-war/clips/2mA60jWAGQCBH', - 'md5': 'fc7a3e4552ae8993c1c4006db46be447', - 'info_dict': { - 'id': '2mA60jWAGQCBH', - 'ext': 'mp4', - 'title': 'Quad Cold', - 'description': 'Medal,https://medal.tv/desktop/', - 'uploader': 'MowgliSB', - 'timestamp': 1603165266, - 'upload_date': '20201020', - 'uploader_id': '10619174', - 'thumbnail': 'https://cdn.medal.tv/10619174/thumbnail-34934644-720p.jpg?t=1080p&c=202042&missing', - 'uploader_url': 'https://medal.tv/users/10619174', - 'comment_count': int, - 'view_count': int, - 'like_count': int, - 'duration': 23, + 'thumbnail': r're:https://cdn\.medal\.tv/ugcp/content-thumbnail/.*\.jpg', + 'tags': ['headshot', 'valorant', '4k', 'clutch', 'mornu'], }, }, { 'url': 'https://medal.tv/games/cod-cold-war/clips/2um24TWdty0NA', @@ -57,12 +35,12 @@ class MedalTVIE(InfoExtractor): 'id': '2um24TWdty0NA', 'ext': 'mp4', 'title': 'u tk me i tk u bigger', - 'description': 'Medal,https://medal.tv/desktop/', - 'uploader': 'Mimicc', + 'description': '', + 'uploader': 'zahl', 'timestamp': 1605580939, 'upload_date': '20201117', 'uploader_id': '5156321', - 'thumbnail': 'https://cdn.medal.tv/5156321/thumbnail-36787208-360p.jpg?t=1080p&c=202046&missing', + 'thumbnail': r're:https://cdn\.medal\.tv/source/.*\.png', 'uploader_url': 'https://medal.tv/users/5156321', 'comment_count': int, 'view_count': int, @@ -70,91 +48,77 @@ class MedalTVIE(InfoExtractor): 'duration': 9, }, }, { - 'url': 'https://medal.tv/games/valorant/clips/37rMeFpryCC-9', - 'only_matching': True, - }, { + # API requires auth 'url': 'https://medal.tv/games/valorant/clips/2WRj40tpY_EU9', + 'md5': '6c6bb6569777fd8b4ef7b33c09de8dcf', + 'info_dict': { + 'id': '2WRj40tpY_EU9', + 'ext': 'mp4', + 'title': '1v5 clutch', + 'description': '', + 'uploader': 'adny', + 'uploader_id': '6256941', + 'uploader_url': 'https://medal.tv/users/6256941', + 'comment_count': int, + 'view_count': int, + 'like_count': int, + 'duration': 25, + 'thumbnail': r're:https://cdn\.medal\.tv/source/.*\.jpg', + 'timestamp': 1612896680, + 'upload_date': '20210209', + }, + 'expected_warnings': ['Video formats are not available through API'], + }, { + 'url': 'https://medal.tv/games/valorant/clips/37rMeFpryCC-9', 'only_matching': True, }] def _real_extract(self, url): video_id = self._match_id(url) - webpage = self._download_webpage(url, video_id, query={'mobilebypass': 'true'}) - - hydration_data = self._search_json( - r']*>[^<]*\bhydrationData\s*=', webpage, - 'next data', video_id, end_pattern='', fatal=False) - - clip = traverse_obj(hydration_data, ('clips', ...), get_all=False) - if not clip: - raise ExtractorError( - 'Could not find video information.', video_id=video_id) - - title = clip['contentTitle'] - - source_width = int_or_none(clip.get('sourceWidth')) - source_height = int_or_none(clip.get('sourceHeight')) - - aspect_ratio = source_width / source_height if source_width and source_height else 16 / 9 - - def add_item(container, item_url, height, id_key='format_id', item_id=None): - item_id = item_id or '%dp' % height - if item_id not in item_url: - return - container.append({ - 'url': item_url, - id_key: item_id, - 'width': round(aspect_ratio * height), - 'height': height, - }) + content_data = self._download_json( + f'https://medal.tv/api/content/{video_id}', video_id, + headers={'Accept': 'application/json'}) formats = [] - thumbnails = [] - for k, v in clip.items(): - if not (v and isinstance(v, str)): - continue - mobj = re.match(r'(contentUrl|thumbnail)(?:(\d+)p)?$', k) - if not mobj: - continue - prefix = mobj.group(1) - height = int_or_none(mobj.group(2)) - if prefix == 'contentUrl': - add_item( - formats, v, height or source_height, - item_id=None if height else 'source') - elif prefix == 'thumbnail': - add_item(thumbnails, v, height, 'id') - - error = clip.get('error') - if not formats and error: - if error == 404: - self.raise_no_formats( - 'That clip does not exist.', - expected=True, video_id=video_id) - else: - self.raise_no_formats( - f'An unknown error occurred ({error}).', - video_id=video_id) - - # Necessary because the id of the author is not known in advance. - # Won't raise an issue if no profile can be found as this is optional. - author = traverse_obj(hydration_data, ('profiles', ...), get_all=False) or {} - author_id = str_or_none(author.get('userId')) - author_url = format_field(author_id, None, 'https://medal.tv/users/%s') + if m3u8_url := url_or_none(content_data.get('contentUrlHls')): + formats.extend(self._extract_m3u8_formats(m3u8_url, video_id, 'mp4', m3u8_id='hls')) + if http_url := url_or_none(content_data.get('contentUrl')): + formats.append({ + 'url': http_url, + 'format_id': 'http-source', + 'ext': 'mp4', + 'quality': 1, + }) + formats = [fmt for fmt in formats if 'video/privacy-protected-guest' not in fmt['url']] + if not formats: + # Fallback, does not require auth + self.report_warning('Video formats are not available through API, falling back to social video URL') + urlh = self._request_webpage( + f'https://medal.tv/api/content/{video_id}/socialVideoUrl', video_id, + note='Checking social video URL') + formats.append({ + 'url': urlh.url, + 'format_id': 'social-video', + 'ext': 'mp4', + 'quality': -1, + }) return { 'id': video_id, - 'title': title, 'formats': formats, - 'thumbnails': thumbnails, - 'description': clip.get('contentDescription'), - 'uploader': author.get('displayName'), - 'timestamp': float_or_none(clip.get('created'), 1000), - 'uploader_id': author_id, - 'uploader_url': author_url, - 'duration': int_or_none(clip.get('videoLengthSeconds')), - 'view_count': int_or_none(clip.get('views')), - 'like_count': int_or_none(clip.get('likes')), - 'comment_count': int_or_none(clip.get('comments')), + **traverse_obj(content_data, { + 'title': ('contentTitle', {str}), + 'description': ('contentDescription', {str}), + 'timestamp': ('created', {int_or_none(scale=1000)}), + 'duration': ('videoLengthSeconds', {int_or_none}), + 'view_count': ('views', {int_or_none}), + 'like_count': ('likes', {int_or_none}), + 'comment_count': ('comments', {int_or_none}), + 'uploader': ('poster', 'displayName', {str}), + 'uploader_id': ('poster', 'userId', {str}), + 'uploader_url': ('poster', 'userId', {str}, filter, {lambda x: x and f'https://medal.tv/users/{x}'}), + 'tags': ('tags', ..., {str}), + 'thumbnail': ('thumbnailUrl', {url_or_none}), + }), }