zulip/zerver/lib/bulk_create.py

108 lines
5.6 KiB
Python
Raw Normal View History

from typing import Any, Dict, List, Optional, Set, Tuple
from zerver.lib.initial_password import initial_password
from zerver.models import Realm, Stream, UserProfile, \
Subscription, Recipient, RealmAuditLog
from zerver.lib.create_user import create_user_profile
def bulk_create_users(realm: Realm,
users_raw: Set[Tuple[str, str, str, bool]],
bot_type: Optional[int]=None,
bot_owner: Optional[UserProfile]=None,
tos_version: Optional[str]=None,
timezone: str="") -> None:
"""
Creates and saves a UserProfile with the given email.
Has some code based off of UserManage.create_user, but doesn't .save()
"""
existing_users = frozenset(UserProfile.objects.filter(
realm=realm).values_list('email', flat=True))
users = sorted([user_raw for user_raw in users_raw if user_raw[0] not in existing_users])
# If we have a different email_address_visibility mode, the code
# below doesn't have the logic to set user_profile.email properly.
assert realm.email_address_visibility == Realm.EMAIL_ADDRESS_VISIBILITY_EVERYONE
# Now create user_profiles
profiles_to_create = [] # type: List[UserProfile]
for (email, full_name, short_name, active) in users:
profile = create_user_profile(realm, email,
initial_password(email), active, bot_type,
full_name, short_name, bot_owner, False, tos_version,
timezone, tutorial_status=UserProfile.TUTORIAL_FINISHED,
enter_sends=True)
profiles_to_create.append(profile)
UserProfile.objects.bulk_create(profiles_to_create)
RealmAuditLog.objects.bulk_create(
[RealmAuditLog(realm=realm, modified_user=profile_,
event_type=RealmAuditLog.USER_CREATED, event_time=profile_.date_joined)
for profile_ in profiles_to_create])
profiles_by_email = {} # type: Dict[str, UserProfile]
profiles_by_id = {} # type: Dict[int, UserProfile]
for profile in UserProfile.objects.select_related().filter(realm=realm):
profiles_by_email[profile.email] = profile
profiles_by_id[profile.id] = profile
recipients_to_create = [] # type: List[Recipient]
2016-06-04 00:52:29 +02:00
for (email, full_name, short_name, active) in users:
recipients_to_create.append(Recipient(type_id=profiles_by_email[email].id,
type=Recipient.PERSONAL))
Recipient.objects.bulk_create(recipients_to_create)
recipients_by_email = {} # type: Dict[str, Recipient]
for recipient in recipients_to_create:
recipients_by_email[profiles_by_id[recipient.type_id].email] = recipient
subscriptions_to_create = [] # type: List[Subscription]
2016-06-04 00:52:29 +02:00
for (email, full_name, short_name, active) in users:
subscriptions_to_create.append(
Subscription(user_profile_id=profiles_by_email[email].id,
recipient=recipients_by_email[email]))
Subscription.objects.bulk_create(subscriptions_to_create)
# This is only sed in populate_db, so doesn't realy need tests
def bulk_create_streams(realm: Realm,
stream_dict: Dict[str, Dict[str, Any]]) -> None: # nocoverage
existing_streams = frozenset([name.lower() for name in
Stream.objects.filter(realm=realm)
.values_list('name', flat=True)])
streams_to_create = [] # type: List[Stream]
for name, options in stream_dict.items():
if 'history_public_to_subscribers' not in options:
options['history_public_to_subscribers'] = (
not options.get("invite_only", False) and not realm.is_zephyr_mirror_realm)
if name.lower() not in existing_streams:
from zerver.lib.actions import render_stream_description
streams_to_create.append(
Stream(
realm=realm,
name=name,
description=options["description"],
rendered_description=render_stream_description(options["description"]),
invite_only=options.get("invite_only", False),
is_announcement_only=options.get("is_announcement_only", False),
history_public_to_subscribers=options["history_public_to_subscribers"],
is_web_public=options.get("is_web_public", False),
is_in_zephyr_realm=realm.is_zephyr_mirror_realm,
)
)
# Sort streams by name before creating them so that we can have a
# reliable ordering of `stream_id` across different python versions.
# This is required for test fixtures which contain `stream_id`. Prior
# to python 3.3 hashes were not randomized but after a security fix
# hash randomization was enabled in python 3.3 which made iteration
# of dictionaries and sets completely unpredictable. Here the order
# of elements while iterating `stream_dict` will be completely random
# for python 3.3 and later versions.
streams_to_create.sort(key=lambda x: x.name)
Stream.objects.bulk_create(streams_to_create)
recipients_to_create = [] # type: List[Recipient]
for stream in Stream.objects.filter(realm=realm).values('id', 'name'):
if stream['name'].lower() not in existing_streams:
recipients_to_create.append(Recipient(type_id=stream['id'],
type=Recipient.STREAM))
Recipient.objects.bulk_create(recipients_to_create)