From 71bd93b89c667b7ca852b3a536dd771da7b67f67 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Sergey=20M=E2=80=A4?= Date: Sat, 21 Nov 2015 08:08:34 +0600 Subject: [PATCH] [pluralsight] Do not rely on argument order in query (Closes #7583) --- youtube_dl/extractor/pluralsight.py | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-) diff --git a/youtube_dl/extractor/pluralsight.py b/youtube_dl/extractor/pluralsight.py index fd32836cc..6cef7c829 100644 --- a/youtube_dl/extractor/pluralsight.py +++ b/youtube_dl/extractor/pluralsight.py @@ -1,6 +1,5 @@ from __future__ import unicode_literals -import re import json from .common import InfoExtractor @@ -19,11 +18,11 @@ from ..utils import ( class PluralsightIE(InfoExtractor): IE_NAME = 'pluralsight' - _VALID_URL = r'https?://(?:www\.)?pluralsight\.com/training/player\?author=(?P[^&]+)&name=(?P[^&]+)(?:&mode=live)?&clip=(?P\d+)&course=(?P[^&]+)' + _VALID_URL = r'https?://(?:(?:www|app)\.)?pluralsight\.com/training/player\?' _LOGIN_URL = 'https://www.pluralsight.com/id/' _NETRC_MACHINE = 'pluralsight' - _TEST = { + _TESTS = [{ 'url': 'http://www.pluralsight.com/training/player?author=mike-mckeown&name=hosting-sql-server-windows-azure-iaas-m7-mgmt&mode=live&clip=3&course=hosting-sql-server-windows-azure-iaas', 'md5': '4d458cf5cf4c593788672419a8dd4cf8', 'info_dict': { @@ -33,7 +32,10 @@ class PluralsightIE(InfoExtractor): 'duration': 338, }, 'skip': 'Requires pluralsight account credentials', - } + }, { + 'url': 'https://app.pluralsight.com/training/player?course=angularjs-get-started&author=scott-allen&name=angularjs-get-started-m1-introduction&clip=0&mode=live', + 'only_matching': True, + }] def _real_initialize(self): self._login() @@ -74,11 +76,15 @@ class PluralsightIE(InfoExtractor): raise ExtractorError('Unable to login: %s' % error, expected=True) def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - author = mobj.group('author') - name = mobj.group('name') - clip_id = mobj.group('clip') - course = mobj.group('course') + qs = compat_urlparse.parse_qs(compat_urlparse.urlparse(url).query) + + author = qs.get('author', [None])[0] + name = qs.get('name', [None])[0] + clip_id = qs.get('clip', [None])[0] + course = qs.get('course', [None])[0] + + if any(not f for f in (author, name, clip_id, course,)): + raise ExtractorError('Invalid URL', expected=True) display_id = '%s-%s' % (name, clip_id) -- 2.22.2