X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/fe979149c83b5a935f7d28baf75848a9137316fd..4231113a7be907539052ce59df44e23c54d21d38:/youtube_dl/extractor/pornhd.py diff --git a/youtube_dl/extractor/pornhd.py b/youtube_dl/extractor/pornhd.py index 8df12ee..b52879c 100644 --- a/youtube_dl/extractor/pornhd.py +++ b/youtube_dl/extractor/pornhd.py @@ -21,7 +21,7 @@ class PornHdIE(InfoExtractor): 'ext': 'mp4', 'title': 'Restroom selfie masturbation', 'description': 'md5:3748420395e03e31ac96857a8f125b2b', - 'thumbnail': 're:^https?://.*\.jpg', + 'thumbnail': r're:^https?://.*\.jpg', 'view_count': int, 'age_limit': 18, } @@ -35,7 +35,7 @@ class PornHdIE(InfoExtractor): 'ext': 'mp4', 'title': 'Sierra loves doing laundry', 'description': 'md5:8ff0523848ac2b8f9b065ba781ccf294', - 'thumbnail': 're:^https?://.*\.jpg', + 'thumbnail': r're:^https?://.*\.jpg', 'view_count': int, 'age_limit': 18, }, @@ -54,7 +54,7 @@ class PornHdIE(InfoExtractor): r'(.+?) - .*?[Pp]ornHD.*?'], webpage, 'title') sources = self._parse_json(js_to_json(self._search_regex( - r"(?s)'sources'\s*:\s*(\{.+?\})\s*\}[;,)]", + r"(?s)sources'?\s*[:=]\s*(\{.+?\})", webpage, 'sources', default='{}')), video_id) if not sources: @@ -82,7 +82,8 @@ class PornHdIE(InfoExtractor): view_count = int_or_none(self._html_search_regex( r'(\d+) views\s*<', webpage, 'view count', fatal=False)) thumbnail = self._search_regex( - r"'poster'\s*:\s*'([^']+)'", webpage, 'thumbnail', fatal=False) + r"poster'?\s*:\s*([\"'])(?P(?:(?!\1).)+)\1", webpage, + 'thumbnail', fatal=False, group='url') return { 'id': video_id,