X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/3f03cde7c013b3ba884e1968702d9bdf8d987f95..9d6a3ba754469c85b7f0f88714203f99ac69087a:/youtube_dl/extractor/common.py diff --git a/youtube_dl/extractor/common.py b/youtube_dl/extractor/common.py index 59b9d37..a2548db 100644 --- a/youtube_dl/extractor/common.py +++ b/youtube_dl/extractor/common.py @@ -339,15 +339,17 @@ class InfoExtractor(object): _GEO_BYPASS attribute may be set to False in order to disable geo restriction bypass mechanisms for a particular extractor. Though it won't disable explicit geo restriction bypass based on - country code provided with geo_bypass_country. (experimental) + country code provided with geo_bypass_country. _GEO_COUNTRIES attribute may contain a list of presumably geo unrestricted countries for this extractor. One of these countries will be used by geo restriction bypass mechanism right away in order to bypass - geo restriction, of course, if the mechanism is not disabled. (experimental) + geo restriction, of course, if the mechanism is not disabled. - NB: both these geo attributes are experimental and may change in future - or be completely removed. + _GEO_IP_BLOCKS attribute may contain a list of presumably geo unrestricted + IP blocks in CIDR notation for this extractor. One of these IP blocks + will be used by geo restriction bypass mechanism similarly + to _GEO_COUNTRIES. Finally, the _WORKING attribute should be set to False for broken IEs in order to warn the users and skip the tests. @@ -358,6 +360,7 @@ class InfoExtractor(object): _x_forwarded_for_ip = None _GEO_BYPASS = True _GEO_COUNTRIES = None + _GEO_IP_BLOCKS = None _WORKING = True def __init__(self, downloader=None): @@ -392,12 +395,15 @@ class InfoExtractor(object): def initialize(self): """Initializes an instance (authentication, etc).""" - self._initialize_geo_bypass(self._GEO_COUNTRIES) + self._initialize_geo_bypass({ + 'countries': self._GEO_COUNTRIES, + 'ip_blocks': self._GEO_IP_BLOCKS, + }) if not self._ready: self._real_initialize() self._ready = True - def _initialize_geo_bypass(self, countries): + def _initialize_geo_bypass(self, geo_bypass_context): """ Initialize geo restriction bypass mechanism. @@ -408,28 +414,82 @@ class InfoExtractor(object): HTTP requests. This method will be used for initial geo bypass mechanism initialization - during the instance initialization with _GEO_COUNTRIES. + during the instance initialization with _GEO_COUNTRIES and + _GEO_IP_BLOCKS. - You may also manually call it from extractor's code if geo countries + You may also manually call it from extractor's code if geo bypass information is not available beforehand (e.g. obtained during - extraction) or due to some another reason. + extraction) or due to some other reason. In this case you should pass + this information in geo bypass context passed as first argument. It may + contain following fields: + + countries: List of geo unrestricted countries (similar + to _GEO_COUNTRIES) + ip_blocks: List of geo unrestricted IP blocks in CIDR notation + (similar to _GEO_IP_BLOCKS) + """ if not self._x_forwarded_for_ip: - country_code = self._downloader.params.get('geo_bypass_country', None) - # If there is no explicit country for geo bypass specified and - # the extractor is known to be geo restricted let's fake IP - # as X-Forwarded-For right away. - if (not country_code and - self._GEO_BYPASS and - self._downloader.params.get('geo_bypass', True) and - countries): - country_code = random.choice(countries) - if country_code: - self._x_forwarded_for_ip = GeoUtils.random_ipv4(country_code) + + # Geo bypass mechanism is explicitly disabled by user + if not self._downloader.params.get('geo_bypass', True): + return + + if not geo_bypass_context: + geo_bypass_context = {} + + # Backward compatibility: previously _initialize_geo_bypass + # expected a list of countries, some 3rd party code may still use + # it this way + if isinstance(geo_bypass_context, (list, tuple)): + geo_bypass_context = { + 'countries': geo_bypass_context, + } + + # The whole point of geo bypass mechanism is to fake IP + # as X-Forwarded-For HTTP header based on some IP block or + # country code. + + # Path 1: bypassing based on IP block in CIDR notation + + # Explicit IP block specified by user, use it right away + # regardless of whether extractor is geo bypassable or not + ip_block = self._downloader.params.get('geo_bypass_ip_block', None) + + # Otherwise use random IP block from geo bypass context but only + # if extractor is known as geo bypassable + if not ip_block: + ip_blocks = geo_bypass_context.get('ip_blocks') + if self._GEO_BYPASS and ip_blocks: + ip_block = random.choice(ip_blocks) + + if ip_block: + self._x_forwarded_for_ip = GeoUtils.random_ipv4(ip_block) + if self._downloader.params.get('verbose', False): + self._downloader.to_screen( + '[debug] Using fake IP %s as X-Forwarded-For.' + % self._x_forwarded_for_ip) + return + + # Path 2: bypassing based on country code + + # Explicit country code specified by user, use it right away + # regardless of whether extractor is geo bypassable or not + country = self._downloader.params.get('geo_bypass_country', None) + + # Otherwise use random country code from geo bypass context but + # only if extractor is known as geo bypassable + if not country: + countries = geo_bypass_context.get('countries') + if self._GEO_BYPASS and countries: + country = random.choice(countries) + + if country: + self._x_forwarded_for_ip = GeoUtils.random_ipv4(country) if self._downloader.params.get('verbose', False): self._downloader.to_screen( '[debug] Using fake IP %s (%s) as X-Forwarded-For.' - % (self._x_forwarded_for_ip, country_code.upper())) + % (self._x_forwarded_for_ip, country.upper())) def extract(self, url): """Extracts URL information and returns it in list of dicts.""" @@ -682,18 +742,30 @@ class InfoExtractor(object): else: self.report_warning(errmsg + str(ve)) - def _download_json(self, url_or_request, video_id, - note='Downloading JSON metadata', - errnote='Unable to download JSON metadata', - transform_source=None, - fatal=True, encoding=None, data=None, headers={}, query={}): - json_string = self._download_webpage( + def _download_json_handle( + self, url_or_request, video_id, note='Downloading JSON metadata', + errnote='Unable to download JSON metadata', transform_source=None, + fatal=True, encoding=None, data=None, headers={}, query={}): + """Return a tuple (JSON object, URL handle)""" + res = self._download_webpage_handle( url_or_request, video_id, note, errnote, fatal=fatal, encoding=encoding, data=data, headers=headers, query=query) - if (not fatal) and json_string is False: - return None + if res is False: + return res + json_string, urlh = res return self._parse_json( - json_string, video_id, transform_source=transform_source, fatal=fatal) + json_string, video_id, transform_source=transform_source, + fatal=fatal), urlh + + def _download_json( + self, url_or_request, video_id, note='Downloading JSON metadata', + errnote='Unable to download JSON metadata', transform_source=None, + fatal=True, encoding=None, data=None, headers={}, query={}): + res = self._download_json_handle( + url_or_request, video_id, note=note, errnote=errnote, + transform_source=transform_source, fatal=fatal, encoding=encoding, + data=data, headers=headers, query=query) + return res if res is False else res[0] def _parse_json(self, json_string, video_id, transform_source=None, fatal=True): if transform_source: @@ -1008,6 +1080,40 @@ class InfoExtractor(object): if isinstance(json_ld, dict): json_ld = [json_ld] + INTERACTION_TYPE_MAP = { + 'CommentAction': 'comment', + 'AgreeAction': 'like', + 'DisagreeAction': 'dislike', + 'LikeAction': 'like', + 'DislikeAction': 'dislike', + 'ListenAction': 'view', + 'WatchAction': 'view', + 'ViewAction': 'view', + } + + def extract_interaction_statistic(e): + interaction_statistic = e.get('interactionStatistic') + if not isinstance(interaction_statistic, list): + return + for is_e in interaction_statistic: + if not isinstance(is_e, dict): + continue + if is_e.get('@type') != 'InteractionCounter': + continue + interaction_type = is_e.get('interactionType') + if not isinstance(interaction_type, compat_str): + continue + interaction_count = int_or_none(is_e.get('userInteractionCount')) + if interaction_count is None: + continue + count_kind = INTERACTION_TYPE_MAP.get(interaction_type.split('/')[-1]) + if not count_kind: + continue + count_key = '%s_count' % count_kind + if info.get(count_key) is not None: + continue + info[count_key] = interaction_count + def extract_video_object(e): assert e['@type'] == 'VideoObject' info.update({ @@ -1023,6 +1129,7 @@ class InfoExtractor(object): 'height': int_or_none(e.get('height')), 'view_count': int_or_none(e.get('interactionCount')), }) + extract_interaction_statistic(e) for e in json_ld: if isinstance(e.get('@context'), compat_str) and re.match(r'^https?://schema.org/?$', e.get('@context')):