X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/9d6a3ba754469c85b7f0f88714203f99ac69087a..e070b5f283f0eca1a445b8382dd7292803ba2c3f:/youtube_dl/extractor/pluralsight.py diff --git a/youtube_dl/extractor/pluralsight.py b/youtube_dl/extractor/pluralsight.py index a207ca9..abd08bc 100644 --- a/youtube_dl/extractor/pluralsight.py +++ b/youtube_dl/extractor/pluralsight.py @@ -4,6 +4,7 @@ import collections import json import os import random +import re from .common import InfoExtractor from ..compat import ( @@ -27,6 +28,60 @@ from ..utils import ( class PluralsightBaseIE(InfoExtractor): _API_BASE = 'https://app.pluralsight.com' + _GRAPHQL_EP = '%s/player/api/graphql' % _API_BASE + _GRAPHQL_HEADERS = { + 'Content-Type': 'application/json;charset=UTF-8', + } + _GRAPHQL_COURSE_TMPL = ''' +query BootstrapPlayer { + rpc { + bootstrapPlayer { + profile { + firstName + lastName + email + username + userHandle + authed + isAuthed + plan + } + course(courseId: "%s") { + name + title + courseHasCaptions + translationLanguages { + code + name + } + supportsWideScreenVideoFormats + timestamp + modules { + name + title + duration + formattedDuration + author + authorized + clips { + authorized + clipId + duration + formattedDuration + id + index + moduleIndex + moduleTitle + name + title + watched + } + } + } + } + } +}''' + def _download_course(self, course_id, url, display_id): try: return self._download_course_rpc(course_id, url, display_id) @@ -39,20 +94,14 @@ class PluralsightBaseIE(InfoExtractor): def _download_course_rpc(self, course_id, url, display_id): response = self._download_json( - '%s/player/functions/rpc' % self._API_BASE, display_id, - 'Downloading course JSON', - data=json.dumps({ - 'fn': 'bootstrapPlayer', - 'payload': { - 'courseId': course_id, - }, - }).encode('utf-8'), - headers={ - 'Content-Type': 'application/json;charset=utf-8', - 'Referer': url, - }) - - course = try_get(response, lambda x: x['payload']['course'], dict) + self._GRAPHQL_EP, display_id, data=json.dumps({ + 'query': self._GRAPHQL_COURSE_TMPL % course_id, + 'variables': {} + }).encode('utf-8'), headers=self._GRAPHQL_HEADERS) + + course = try_get( + response, lambda x: x['data']['rpc']['bootstrapPlayer']['course'], + dict) if course: return course @@ -90,6 +139,28 @@ class PluralsightIE(PluralsightBaseIE): 'only_matching': True, }] + GRAPHQL_VIEWCLIP_TMPL = ''' +query viewClip { + viewClip(input: { + author: "%(author)s", + clipIndex: %(clipIndex)d, + courseName: "%(courseName)s", + includeCaptions: %(includeCaptions)s, + locale: "%(locale)s", + mediaType: "%(mediaType)s", + moduleName: "%(moduleName)s", + quality: "%(quality)s" + }) { + urls { + url + cdn + rank + source + }, + status + } +}''' + def _real_initialize(self): self._login() @@ -126,7 +197,10 @@ class PluralsightIE(PluralsightBaseIE): if error: raise ExtractorError('Unable to login: %s' % error, expected=True) - if all(p not in response for p in ('__INITIAL_STATE__', '"currentUser"')): + if all(not re.search(p, response) for p in ( + r'__INITIAL_STATE__', r'["\']currentUser["\']', + # new layout? + r'>\s*Sign out\s*<')): BLOCKED = 'Your account has been blocked due to suspicious activity' if BLOCKED in response: raise ExtractorError( @@ -140,18 +214,26 @@ class PluralsightIE(PluralsightBaseIE): raise ExtractorError('Unable to log in') - def _get_subtitles(self, author, clip_idx, lang, name, duration, video_id): - captions_post = { - 'a': author, - 'cn': clip_idx, - 'lc': lang, - 'm': name, - } - captions = self._download_json( - '%s/player/retrieve-captions' % self._API_BASE, video_id, - 'Downloading captions JSON', 'Unable to download captions JSON', - fatal=False, data=json.dumps(captions_post).encode('utf-8'), - headers={'Content-Type': 'application/json;charset=utf-8'}) + def _get_subtitles(self, author, clip_idx, clip_id, lang, name, duration, video_id): + captions = None + if clip_id: + captions = self._download_json( + '%s/transcript/api/v1/caption/json/%s/%s' + % (self._API_BASE, clip_id, lang), video_id, + 'Downloading captions JSON', 'Unable to download captions JSON', + fatal=False) + if not captions: + captions_post = { + 'a': author, + 'cn': int(clip_idx), + 'lc': lang, + 'm': name, + } + captions = self._download_json( + '%s/player/retrieve-captions' % self._API_BASE, video_id, + 'Downloading captions JSON', 'Unable to download captions JSON', + fatal=False, data=json.dumps(captions_post).encode('utf-8'), + headers={'Content-Type': 'application/json;charset=utf-8'}) if captions: return { lang: [{ @@ -245,7 +327,7 @@ class PluralsightIE(PluralsightBaseIE): ) # Some courses also offer widescreen resolution for high quality (see - # https://github.com/rg3/youtube-dl/issues/7766) + # https://github.com/ytdl-org/youtube-dl/issues/7766) widescreen = course.get('supportsWideScreenVideoFormats') is True best_quality = 'high-widescreen' if widescreen else 'high' if widescreen: @@ -277,7 +359,7 @@ class PluralsightIE(PluralsightBaseIE): f = QUALITIES[quality].copy() clip_post = { 'author': author, - 'includeCaptions': False, + 'includeCaptions': 'false', 'clipIndex': int(clip_idx), 'courseName': course_name, 'locale': 'en', @@ -286,16 +368,28 @@ class PluralsightIE(PluralsightBaseIE): 'quality': '%dx%d' % (f['width'], f['height']), } format_id = '%s-%s' % (ext, quality) - viewclip = self._download_json( - '%s/video/clips/viewclip' % self._API_BASE, display_id, - 'Downloading %s viewclip JSON' % format_id, fatal=False, - data=json.dumps(clip_post).encode('utf-8'), - headers={'Content-Type': 'application/json;charset=utf-8'}) + + try: + viewclip = self._download_json( + self._GRAPHQL_EP, display_id, + 'Downloading %s viewclip graphql' % format_id, + data=json.dumps({ + 'query': self.GRAPHQL_VIEWCLIP_TMPL % clip_post, + 'variables': {} + }).encode('utf-8'), + headers=self._GRAPHQL_HEADERS)['data']['viewClip'] + except ExtractorError: + # Still works but most likely will go soon + viewclip = self._download_json( + '%s/video/clips/viewclip' % self._API_BASE, display_id, + 'Downloading %s viewclip JSON' % format_id, fatal=False, + data=json.dumps(clip_post).encode('utf-8'), + headers={'Content-Type': 'application/json;charset=utf-8'}) # Pluralsight tracks multiple sequential calls to ViewClip API and start # to return 429 HTTP errors after some time (see - # https://github.com/rg3/youtube-dl/pull/6989). Moreover it may even lead - # to account ban (see https://github.com/rg3/youtube-dl/issues/6842). + # https://github.com/ytdl-org/youtube-dl/pull/6989). Moreover it may even lead + # to account ban (see https://github.com/ytdl-org/youtube-dl/issues/6842). # To somewhat reduce the probability of these consequences # we will sleep random amount of time before each call to ViewClip. self._sleep( @@ -331,7 +425,7 @@ class PluralsightIE(PluralsightBaseIE): # TODO: other languages? subtitles = self.extract_subtitles( - author, clip_idx, 'en', name, duration, display_id) + author, clip_idx, clip.get('clipId'), 'en', name, duration, display_id) return { 'id': clip_id,