import os import dateutil.parser import logging import subprocess import ujson from django.conf import settings from django.forms.models import model_to_dict from django.utils.timezone import now as timezone_now from typing import Any, Dict, List, Set, Tuple from zerver.models import UserProfile, Recipient from zerver.lib.export import MESSAGE_BATCH_CHUNK_SIZE from zerver.data_import.import_util import ZerverFieldsT, build_zerver_realm, \ build_avatar, build_subscription, build_recipient, build_usermessages, \ build_defaultstream, process_avatars, build_realm, build_stream, \ build_message, create_converted_data_files, make_subscriber_map # stubs GitterDataT = List[Dict[str, Any]] realm_id = 0 def gitter_workspace_to_realm(domain_name: str, gitter_data: GitterDataT, realm_subdomain: str) -> Tuple[ZerverFieldsT, List[ZerverFieldsT], Dict[str, int], Dict[str, int]]: """ Returns: 1. realm, Converted Realm data 2. avatars, which is list to map avatars to zulip avatar records.json 3. user_map, which is a dictionary to map from gitter user id to zulip user id 4. stream_map, which is a dictionary to map from gitter rooms to zulip stream id """ NOW = float(timezone_now().timestamp()) zerver_realm: List[ZerverFieldsT] = build_zerver_realm(realm_id, realm_subdomain, NOW, 'Gitter') realm = build_realm(zerver_realm, realm_id, domain_name) zerver_userprofile, avatars, user_map = build_userprofile(int(NOW), domain_name, gitter_data) zerver_stream, zerver_defaultstream, stream_map = build_stream_map(int(NOW), gitter_data) zerver_recipient, zerver_subscription = build_recipient_and_subscription( zerver_userprofile, zerver_stream) realm['zerver_userprofile'] = zerver_userprofile realm['zerver_stream'] = zerver_stream realm['zerver_defaultstream'] = zerver_defaultstream realm['zerver_recipient'] = zerver_recipient realm['zerver_subscription'] = zerver_subscription return realm, avatars, user_map, stream_map def build_userprofile(timestamp: Any, domain_name: str, gitter_data: GitterDataT) -> Tuple[List[ZerverFieldsT], List[ZerverFieldsT], Dict[str, int]]: """ Returns: 1. zerver_userprofile, which is a list of user profile 2. avatar_list, which is list to map avatars to zulip avatars records.json 3. added_users, which is a dictionary to map from gitter user id to zulip id """ logging.info('######### IMPORTING USERS STARTED #########\n') zerver_userprofile = [] avatar_list: List[ZerverFieldsT] = [] user_map: Dict[str, int] = {} user_id = 0 for data in gitter_data: if data['fromUser']['id'] not in user_map: user_data = data['fromUser'] user_map[user_data['id']] = user_id email = get_user_email(user_data, domain_name) build_avatar(user_id, realm_id, email, user_data['avatarUrl'], timestamp, avatar_list) # Build userprofile object userprofile = UserProfile( full_name=user_data['displayName'], short_name=user_data['username'], id=user_id, email=email, delivery_email=email, avatar_source='U', pointer=-1, date_joined=timestamp, last_login=timestamp) userprofile_dict = model_to_dict(userprofile) # Set realm id separately as the corresponding realm is not yet a Realm model # instance userprofile_dict['realm'] = realm_id zerver_userprofile.append(userprofile_dict) user_id += 1 logging.info('######### IMPORTING USERS FINISHED #########\n') return zerver_userprofile, avatar_list, user_map def get_user_email(user_data: ZerverFieldsT, domain_name: str) -> str: # TODO Get user email from github email = ("%s@users.noreply.github.com" % (user_data['username'],)) return email def build_stream_map(timestamp: Any, gitter_data: GitterDataT) -> Tuple[List[ZerverFieldsT], List[ZerverFieldsT], Dict[str, int]]: """ Returns: 1. stream, which is the list of streams 2. defaultstreams, which is the list of default streams 3. stream_map, which is a dictionary to map from gitter rooms to zulip stream id """ logging.info('######### IMPORTING STREAM STARTED #########\n') stream_id = 0 stream: List[ZerverFieldsT] = [] # Default stream when no "room" field is present stream.append(build_stream(timestamp, realm_id, 'from gitter', "Imported from gitter", stream_id)) defaultstream = build_defaultstream(realm_id=realm_id, stream_id=stream_id, defaultstream_id=0) stream_id += 1 # Gathering streams from gitter_data stream_map: Dict[str, int] = {} for data in gitter_data: if 'room' in data and data['room'] not in stream_map: stream.append(build_stream(timestamp, realm_id, data['room'], f'Gitter room {data["room"]}', stream_id)) stream_map[data['room']] = stream_id stream_id += 1 logging.info('######### IMPORTING STREAMS FINISHED #########\n') return stream, [defaultstream], stream_map def build_recipient_and_subscription( zerver_userprofile: List[ZerverFieldsT], zerver_stream: List[ZerverFieldsT]) -> Tuple[List[ZerverFieldsT], List[ZerverFieldsT]]: """ Assumes that there is at least one stream with 'stream_id' = 0, and that this stream is the only defaultstream, with 'defaultstream_id' = 0 Returns: 1. zerver_recipient, which is a list of mapped recipient 2. zerver_subscription, which is a list of mapped subscription """ zerver_recipient = [] zerver_subscription = [] recipient_id = subscription_id = 0 # For streams # Initial recipients correspond to intitial streams # We enumerate all streams, and build a recipient for each # Hence 'recipient_id'=n corresponds to 'stream_id'=n for stream in zerver_stream: zerver_recipient.append(build_recipient(recipient_id, recipient_id, Recipient.STREAM)) recipient_id += 1 # For users for user in zerver_userprofile: zerver_recipient.append(build_recipient(user['id'], recipient_id, Recipient.PERSONAL)) zerver_subscription.append(build_subscription(recipient_id, user['id'], subscription_id)) recipient_id += 1 subscription_id += 1 # As suggested in #14830, we subscribe every user to every stream. # We rely on the above invariant: 'recipient_id'=n corresponds to 'stream_id'=n # # TODO: For multi-stream imports, subscribe users to streams # based either on Gitter API data or who sent messages where. for user in zerver_userprofile: for stream in zerver_stream: zerver_subscription.append(build_subscription(stream['id'], user['id'], subscription_id)) subscription_id += 1 return zerver_recipient, zerver_subscription def convert_gitter_workspace_messages(gitter_data: GitterDataT, output_dir: str, subscriber_map: Dict[int, Set[int]], user_map: Dict[str, int], stream_map: Dict[str, int], user_short_name_to_full_name: Dict[str, str], chunk_size: int=MESSAGE_BATCH_CHUNK_SIZE) -> None: """ Messages are stored in batches """ logging.info('######### IMPORTING MESSAGES STARTED #########\n') message_id = 0 low_index = 0 upper_index = low_index + chunk_size dump_file_id = 1 while True: message_json = {} zerver_message = [] zerver_usermessage: List[ZerverFieldsT] = [] message_data = gitter_data[low_index: upper_index] if len(message_data) == 0: break for message in message_data: message_time = dateutil.parser.parse(message['sent']).timestamp() mentioned_user_ids = get_usermentions(message, user_map, user_short_name_to_full_name) rendered_content = None topic_name = 'imported from gitter' + (f' room {message["room"]}' if 'room' in message else '') user_id = user_map[message['fromUser']['id']] recipient_id = stream_map[message['room']] if 'room' in message else 0 zulip_message = build_message(topic_name, float(message_time), message_id, message['text'], rendered_content, user_id, recipient_id) zerver_message.append(zulip_message) build_usermessages( zerver_usermessage=zerver_usermessage, subscriber_map=subscriber_map, recipient_id=recipient_id, mentioned_user_ids=mentioned_user_ids, message_id=message_id, is_private=False, ) message_id += 1 message_json['zerver_message'] = zerver_message message_json['zerver_usermessage'] = zerver_usermessage message_filename = os.path.join(output_dir, "messages-%06d.json" % (dump_file_id,)) logging.info("Writing Messages to %s\n", message_filename) write_data_to_file(os.path.join(message_filename), message_json) low_index = upper_index upper_index = chunk_size + low_index dump_file_id += 1 logging.info('######### IMPORTING MESSAGES FINISHED #########\n') def get_usermentions(message: Dict[str, Any], user_map: Dict[str, int], user_short_name_to_full_name: Dict[str, str]) -> List[int]: mentioned_user_ids = [] if 'mentions' in message: for mention in message['mentions']: if mention.get('userId') in user_map: gitter_mention = '@%s' % (mention['screenName'],) if mention['screenName'] not in user_short_name_to_full_name: logging.info("Mentioned user %s never sent any messages, so has no full name data", mention['screenName']) full_name = mention['screenName'] else: full_name = user_short_name_to_full_name[mention['screenName']] zulip_mention = ('@**%s**' % (full_name,)) message['text'] = message['text'].replace(gitter_mention, zulip_mention) mentioned_user_ids.append(user_map[mention['userId']]) return mentioned_user_ids def do_convert_data(gitter_data_file: str, output_dir: str, threads: int=6) -> None: # Subdomain is set by the user while running the import commands realm_subdomain = "" domain_name = settings.EXTERNAL_HOST os.makedirs(output_dir, exist_ok=True) # output directory should be empty initially if os.listdir(output_dir): raise Exception("Output directory should be empty!") # Read data from the gitter file with open(gitter_data_file) as fp: gitter_data = ujson.load(fp) realm, avatar_list, user_map, stream_map = gitter_workspace_to_realm( domain_name, gitter_data, realm_subdomain) subscriber_map = make_subscriber_map( zerver_subscription=realm['zerver_subscription'], ) # For user mentions user_short_name_to_full_name = {} for userprofile in realm['zerver_userprofile']: user_short_name_to_full_name[userprofile['short_name']] = userprofile['full_name'] convert_gitter_workspace_messages( gitter_data, output_dir, subscriber_map, user_map, stream_map, user_short_name_to_full_name) avatar_folder = os.path.join(output_dir, 'avatars') avatar_realm_folder = os.path.join(avatar_folder, str(realm_id)) os.makedirs(avatar_realm_folder, exist_ok=True) avatar_records = process_avatars(avatar_list, avatar_folder, realm_id, threads) attachment: Dict[str, List[Any]] = {"zerver_attachment": []} # IO realm.json create_converted_data_files(realm, output_dir, '/realm.json') # IO emoji records create_converted_data_files([], output_dir, '/emoji/records.json') # IO avatar records create_converted_data_files(avatar_records, output_dir, '/avatars/records.json') # IO uploads records create_converted_data_files([], output_dir, '/uploads/records.json') # IO attachments records create_converted_data_files(attachment, output_dir, '/attachment.json') subprocess.check_call(["tar", "-czf", output_dir + '.tar.gz', output_dir, '-P']) logging.info('######### DATA CONVERSION FINISHED #########\n') logging.info("Zulip data dump created at %s", output_dir) def write_data_to_file(output_file: str, data: Any) -> None: with open(output_file, "w") as f: f.write(ujson.dumps(data, indent=4))