From 162bcc68dc73706699b559fffdd8bed3db6643b9 Mon Sep 17 00:00:00 2001 From: Remita Amine Date: Thu, 24 Oct 2019 12:53:33 +0100 Subject: [PATCH] [puhutv] improve extraction - fix subtitles extraction - transform HLS URLs to http URLs - improve metadata extraction --- youtube_dl/extractor/puhutv.py | 90 ++++++++++++++++++---------------- 1 file changed, 49 insertions(+), 41 deletions(-) diff --git a/youtube_dl/extractor/puhutv.py b/youtube_dl/extractor/puhutv.py index 5465e8ab7..fb704a3c4 100644 --- a/youtube_dl/extractor/puhutv.py +++ b/youtube_dl/extractor/puhutv.py @@ -25,21 +25,21 @@ class PuhuTVIE(InfoExtractor): _TESTS = [{ # film 'url': 'https://puhutv.com/sut-kardesler-izle', - 'md5': 'fbd8f2d8e7681f8bcd51b592475a6ae7', + 'md5': 'a347470371d56e1585d1b2c8dab01c96', 'info_dict': { 'id': '5085', 'display_id': 'sut-kardesler', 'ext': 'mp4', 'title': 'Süt Kardeşler', - 'description': 'md5:405fd024df916ca16731114eb18e511a', + 'description': 'md5:ca09da25b7e57cbb5a9280d6e48d17aa', 'thumbnail': r're:^https?://.*\.jpg$', 'duration': 4832.44, 'creator': 'Arzu Film', - 'timestamp': 1469778212, - 'upload_date': '20160729', + 'timestamp': 1561062602, + 'upload_date': '20190620', 'release_year': 1976, 'view_count': int, - 'tags': ['Aile', 'Komedi', 'Klasikler'], + 'tags': list, }, }, { # episode, geo restricted, bypassable with --geo-verification-proxy @@ -64,9 +64,10 @@ class PuhuTVIE(InfoExtractor): display_id)['data'] video_id = compat_str(info['id']) - title = info.get('name') or info['title']['name'] + show = info.get('title') or {} + title = info.get('name') or show['name'] if info.get('display_name'): - title = '%s %s' % (title, info.get('display_name')) + title = '%s %s' % (title, info['display_name']) try: videos = self._download_json( @@ -78,17 +79,36 @@ class PuhuTVIE(InfoExtractor): self.raise_geo_restricted() raise + urls = [] formats = [] + + def add_http_from_hls(m3u8_f): + http_url = m3u8_f['url'].replace('/hls/', '/mp4/').replace('/chunklist.m3u8', '.mp4') + if http_url != m3u8_f['url']: + f = m3u8_f.copy() + f.update({ + 'format_id': f['format_id'].replace('hls', 'http'), + 'protocol': 'http', + 'url': http_url, + }) + formats.append(f) + for video in videos['data']['videos']: media_url = url_or_none(video.get('url')) - if not media_url: + if not media_url or media_url in urls: continue + urls.append(media_url) + playlist = video.get('is_playlist') - if video.get('stream_type') == 'hls' and playlist is True: - formats.extend(self._extract_m3u8_formats( + if (video.get('stream_type') == 'hls' and playlist is True) or 'playlist.m3u8' in media_url: + m3u8_formats = self._extract_m3u8_formats( media_url, video_id, 'mp4', entry_protocol='m3u8_native', - m3u8_id='hls', fatal=False)) + m3u8_id='hls', fatal=False) + for m3u8_f in m3u8_formats: + formats.append(m3u8_f) + add_http_from_hls(m3u8_f) continue + quality = int_or_none(video.get('quality')) f = { 'url': media_url, @@ -96,34 +116,29 @@ class PuhuTVIE(InfoExtractor): 'height': quality } video_format = video.get('video_format') - if video_format == 'hls' and playlist is False: + is_hls = (video_format == 'hls' or '/hls/' in media_url or '/chunklist.m3u8' in media_url) and playlist is False + if is_hls: format_id = 'hls' f['protocol'] = 'm3u8_native' elif video_format == 'mp4': format_id = 'http' - else: continue if quality: format_id += '-%sp' % quality f['format_id'] = format_id formats.append(f) + if is_hls: + add_http_from_hls(f) self._sort_formats(formats) - description = try_get( - info, lambda x: x['title']['description'], - compat_str) or info.get('description') - timestamp = unified_timestamp(info.get('created_at')) creator = try_get( - info, lambda x: x['title']['producer']['name'], compat_str) + show, lambda x: x['producer']['name'], compat_str) - duration = float_or_none( - try_get(info, lambda x: x['content']['duration_in_ms'], int), - scale=1000) - view_count = try_get(info, lambda x: x['content']['watch_count'], int) + content = info.get('content') or {} images = try_get( - info, lambda x: x['content']['images']['wide'], dict) or {} + content, lambda x: x['images']['wide'], dict) or {} thumbnails = [] for image_id, image_url in images.items(): if not isinstance(image_url, compat_str): @@ -137,14 +152,8 @@ class PuhuTVIE(InfoExtractor): }) thumbnails.append(t) - release_year = try_get(info, lambda x: x['title']['released_at'], int) - - season_number = int_or_none(info.get('season_number')) - season_id = str_or_none(info.get('season_id')) - episode_number = int_or_none(info.get('episode_number')) - tags = [] - for genre in try_get(info, lambda x: x['title']['genres'], list) or []: + for genre in show.get('genres') or []: if not isinstance(genre, dict): continue genre_name = genre.get('name') @@ -152,12 +161,11 @@ class PuhuTVIE(InfoExtractor): tags.append(genre_name) subtitles = {} - for subtitle in try_get( - info, lambda x: x['content']['subtitles'], list) or []: + for subtitle in content.get('subtitles') or []: if not isinstance(subtitle, dict): continue lang = subtitle.get('language') - sub_url = url_or_none(subtitle.get('url')) + sub_url = url_or_none(subtitle.get('url') or subtitle.get('file')) if not lang or not isinstance(lang, compat_str) or not sub_url: continue subtitles[self._SUBTITLE_LANGS.get(lang, lang)] = [{ @@ -168,15 +176,15 @@ class PuhuTVIE(InfoExtractor): 'id': video_id, 'display_id': display_id, 'title': title, - 'description': description, - 'season_id': season_id, - 'season_number': season_number, - 'episode_number': episode_number, - 'release_year': release_year, - 'timestamp': timestamp, + 'description': info.get('description') or show.get('description'), + 'season_id': str_or_none(info.get('season_id')), + 'season_number': int_or_none(info.get('season_number')), + 'episode_number': int_or_none(info.get('episode_number')), + 'release_year': int_or_none(show.get('released_at')), + 'timestamp': unified_timestamp(info.get('created_at')), 'creator': creator, - 'view_count': view_count, - 'duration': duration, + 'view_count': int_or_none(content.get('watch_count')), + 'duration': float_or_none(content.get('duration_in_ms'), 1000), 'tags': tags, 'subtitles': subtitles, 'thumbnails': thumbnails, -- 2.22.2