[instagram] Add base extractor for playlists and tag extractor
This commit is contained in:
parent
15870747f0
commit
31fbedc06a
2 changed files with 105 additions and 37 deletions
|
@ -494,7 +494,11 @@ from .ina import InaIE
|
||||||
from .inc import IncIE
|
from .inc import IncIE
|
||||||
from .indavideo import IndavideoEmbedIE
|
from .indavideo import IndavideoEmbedIE
|
||||||
from .infoq import InfoQIE
|
from .infoq import InfoQIE
|
||||||
from .instagram import InstagramIE, InstagramUserIE
|
from .instagram import (
|
||||||
|
InstagramIE,
|
||||||
|
InstagramUserIE,
|
||||||
|
InstagramTagIE,
|
||||||
|
)
|
||||||
from .internazionale import InternazionaleIE
|
from .internazionale import InternazionaleIE
|
||||||
from .internetvideoarchive import InternetVideoArchiveIE
|
from .internetvideoarchive import InternetVideoArchiveIE
|
||||||
from .iprima import IPrimaIE
|
from .iprima import IPrimaIE
|
||||||
|
|
|
@ -227,44 +227,37 @@ class InstagramIE(InfoExtractor):
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
class InstagramUserIE(InfoExtractor):
|
class InstagramPlaylistIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://(?:www\.)?instagram\.com/(?P<id>[^/]{2,})/?(?:$|[?#])'
|
# A superclass for handling any kind of query based on GraphQL which
|
||||||
IE_DESC = 'Instagram user profile'
|
# results in a playlist.
|
||||||
IE_NAME = 'instagram:user'
|
|
||||||
_TEST = {
|
|
||||||
'url': 'https://instagram.com/porsche',
|
|
||||||
'info_dict': {
|
|
||||||
'id': 'porsche',
|
|
||||||
'title': 'porsche',
|
|
||||||
},
|
|
||||||
'playlist_count': 5,
|
|
||||||
'params': {
|
|
||||||
'extract_flat': True,
|
|
||||||
'skip_download': True,
|
|
||||||
'playlistend': 5,
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
_gis_tmpl = None
|
_gis_tmpl = None # used to cache GIS request type
|
||||||
|
|
||||||
def _entries(self, data):
|
def _parse_graphql(self, webpage, item_id):
|
||||||
|
# Reads a webpage and returns its GraphQL data.
|
||||||
|
return self._parse_json(
|
||||||
|
self._search_regex(
|
||||||
|
r'sharedData\s*=\s*({.+?})\s*;\s*[<\n]', webpage, 'data'),
|
||||||
|
item_id)
|
||||||
|
|
||||||
|
def _extract_graphql(self, data, url):
|
||||||
|
# Parses GraphQL queries containing videos and generates a playlist.
|
||||||
def get_count(suffix):
|
def get_count(suffix):
|
||||||
return int_or_none(try_get(
|
return int_or_none(try_get(
|
||||||
node, lambda x: x['edge_media_' + suffix]['count']))
|
node, lambda x: x['edge_media_' + suffix]['count']))
|
||||||
|
|
||||||
uploader_id = data['entry_data']['ProfilePage'][0]['graphql']['user']['id']
|
uploader_id = self._match_id(url)
|
||||||
csrf_token = data['config']['csrf_token']
|
csrf_token = data['config']['csrf_token']
|
||||||
rhx_gis = data.get('rhx_gis') or '3c7ca9dcefcf966d11dacf1f151335e8'
|
rhx_gis = data.get('rhx_gis') or '3c7ca9dcefcf966d11dacf1f151335e8'
|
||||||
|
|
||||||
self._set_cookie('instagram.com', 'ig_pr', '1')
|
|
||||||
|
|
||||||
cursor = ''
|
cursor = ''
|
||||||
for page_num in itertools.count(1):
|
for page_num in itertools.count(1):
|
||||||
variables = json.dumps({
|
variables = {
|
||||||
'id': uploader_id,
|
|
||||||
'first': 12,
|
'first': 12,
|
||||||
'after': cursor,
|
'after': cursor,
|
||||||
})
|
}
|
||||||
|
variables.update(self._query_vars_for(data))
|
||||||
|
variables = json.dumps(variables)
|
||||||
|
|
||||||
if self._gis_tmpl:
|
if self._gis_tmpl:
|
||||||
gis_tmpls = [self._gis_tmpl]
|
gis_tmpls = [self._gis_tmpl]
|
||||||
|
@ -276,21 +269,26 @@ class InstagramUserIE(InfoExtractor):
|
||||||
'%s:%s:%s' % (rhx_gis, csrf_token, std_headers['User-Agent']),
|
'%s:%s:%s' % (rhx_gis, csrf_token, std_headers['User-Agent']),
|
||||||
]
|
]
|
||||||
|
|
||||||
|
# try all of the ways to generate a GIS query, and not only use the
|
||||||
|
# first one that works, but cache it for future requests
|
||||||
for gis_tmpl in gis_tmpls:
|
for gis_tmpl in gis_tmpls:
|
||||||
try:
|
try:
|
||||||
media = self._download_json(
|
json_data = self._download_json(
|
||||||
'https://www.instagram.com/graphql/query/', uploader_id,
|
'https://www.instagram.com/graphql/query/', uploader_id,
|
||||||
'Downloading JSON page %d' % page_num, headers={
|
'Downloading JSON page %d' % page_num, headers={
|
||||||
'X-Requested-With': 'XMLHttpRequest',
|
'X-Requested-With': 'XMLHttpRequest',
|
||||||
'X-Instagram-GIS': hashlib.md5(
|
'X-Instagram-GIS': hashlib.md5(
|
||||||
('%s:%s' % (gis_tmpl, variables)).encode('utf-8')).hexdigest(),
|
('%s:%s' % (gis_tmpl, variables)).encode('utf-8')).hexdigest(),
|
||||||
}, query={
|
}, query={
|
||||||
'query_hash': '42323d64886122307be10013ad2dcc44',
|
'query_hash': self._QUERY_HASH,
|
||||||
'variables': variables,
|
'variables': variables,
|
||||||
})['data']['user']['edge_owner_to_timeline_media']
|
})
|
||||||
|
media = self._parse_timeline_from(json_data)
|
||||||
self._gis_tmpl = gis_tmpl
|
self._gis_tmpl = gis_tmpl
|
||||||
break
|
break
|
||||||
except ExtractorError as e:
|
except ExtractorError as e:
|
||||||
|
# if it's an error caused by a bad query, and there are
|
||||||
|
# more GIS templates to try, ignore it and keep trying
|
||||||
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403:
|
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403:
|
||||||
if gis_tmpl != gis_tmpls[-1]:
|
if gis_tmpl != gis_tmpls[-1]:
|
||||||
continue
|
continue
|
||||||
|
@ -348,14 +346,80 @@ class InstagramUserIE(InfoExtractor):
|
||||||
break
|
break
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
username = self._match_id(url)
|
user_or_tag = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, user_or_tag)
|
||||||
|
data = self._parse_graphql(webpage, user_or_tag)
|
||||||
|
|
||||||
webpage = self._download_webpage(url, username)
|
self._set_cookie('instagram.com', 'ig_pr', '1')
|
||||||
|
|
||||||
data = self._parse_json(
|
|
||||||
self._search_regex(
|
|
||||||
r'sharedData\s*=\s*({.+?})\s*;\s*[<\n]', webpage, 'data'),
|
|
||||||
username)
|
|
||||||
|
|
||||||
return self.playlist_result(
|
return self.playlist_result(
|
||||||
self._entries(data), username, username)
|
self._extract_graphql(data, url), user_or_tag, user_or_tag)
|
||||||
|
|
||||||
|
|
||||||
|
class InstagramUserIE(InstagramPlaylistIE):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?instagram\.com/(?P<id>[^/]{2,})/?(?:$|[?#])'
|
||||||
|
IE_DESC = 'Instagram user profile'
|
||||||
|
IE_NAME = 'instagram:user'
|
||||||
|
_TEST = {
|
||||||
|
'url': 'https://instagram.com/porsche',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'porsche',
|
||||||
|
'title': 'porsche',
|
||||||
|
},
|
||||||
|
'playlist_count': 5,
|
||||||
|
'params': {
|
||||||
|
'extract_flat': True,
|
||||||
|
'skip_download': True,
|
||||||
|
'playlistend': 5,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
_QUERY_HASH = '42323d64886122307be10013ad2dcc44',
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def _parse_timeline_from(data):
|
||||||
|
# extracts the media timeline data from a GraphQL result
|
||||||
|
return data['data']['user']['edge_owner_to_timeline_media']
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def _query_vars_for(data):
|
||||||
|
# returns a dictionary of variables to add to the timeline query based
|
||||||
|
# on the GraphQL of the original page
|
||||||
|
return {
|
||||||
|
'id': data['entry_data']['ProfilePage'][0]['graphql']['user']['id']
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class InstagramTagIE(InstagramPlaylistIE):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?instagram\.com/explore/tags/(?P<id>[^/]+)'
|
||||||
|
IE_DESC = 'Instagram hashtag search'
|
||||||
|
IE_NAME = 'instagram:tag'
|
||||||
|
_TEST = {
|
||||||
|
'url': 'https://instagram.com/explore/tags/lolcats',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'lolcats',
|
||||||
|
'title': 'lolcats',
|
||||||
|
},
|
||||||
|
'playlist_count': 50,
|
||||||
|
'params': {
|
||||||
|
'extract_flat': True,
|
||||||
|
'skip_download': True,
|
||||||
|
'playlistend': 50,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
_QUERY_HASH = 'f92f56d47dc7a55b606908374b43a314',
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def _parse_timeline_from(data):
|
||||||
|
# extracts the media timeline data from a GraphQL result
|
||||||
|
return data['data']['hashtag']['edge_hashtag_to_media']
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def _query_vars_for(data):
|
||||||
|
# returns a dictionary of variables to add to the timeline query based
|
||||||
|
# on the GraphQL of the original page
|
||||||
|
return {
|
||||||
|
'tag_name':
|
||||||
|
data['entry_data']['TagPage'][0]['graphql']['hashtag']['name']
|
||||||
|
}
|
||||||
|
|
Loading…
Reference in a new issue