This code removes a lot of complexity with very likely
positive overall impact on system performance and
negligible downside.
We already cache display recipients on a per-user
level, so there's no need for another cache layer on
top of that that keys them with recipient ids.
We avoid strange things where Alice/Bob and Bob/Charlie
get put into the top layer cache and then we still have
a cache miss on Alice/Charlie despite the lower level
cache being able to support per-user lookups.
This change does introduce an extra database round trip
if any of our messages have a huddle, but the query is
extremely cheap, and we can always try to cache that
function more directly or try to re-use some of our
other huddle-based caches.
As part of this, we clean up the names for the
lower-level per-user cache of display recipients, and
we simplify the cache keys.
We also stop passing in a full Recipient object to the
`bulk_get_huddle_user_ids` functions.
The local impact of this change should be easy to
measure (at least approximately), since we use this
function every time a user gets messages via the
/messages endpoint.
The only overlap between how we fetched streams and
users was to share some really complicated data
structures.
We can also short-circuit some logic if a message
batch is either all-stream or all-DM.
We restrict the columns, avoid quadratic looping,
and don't bother with order_by.
We also return the user ids (per recipient) as
sets, since that's how the only caller uses the
info (albeit implicitly via set.union accepting
a list).
It’s unclear what was supposed to be “safe” about this wrapper. The
hashlib API is fine without it, and we don’t want to encourage further
use of SHA-1.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Translators benefit from the extra information in the field names, and
need the reordering freedom that isn’t available with multiple
positional fields.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
- Adds `message_handle_match` function to handle new pattern for
relative help links to "Drafts" and "Scheduled messages" for logged-in
users: `{relative|message|drafts}` and `{relative|message|scheduled}`.
`remove_denormalized_recipient_column_from_data` removes the
`recipient` data from `zerver_userprofile`, but did not remove it from
`zerver_userprofile_mirrordummy`, which was later appended to the list
of `zerver_userprofile` objects. This led to failure when inserting,
as the mirrordummy objects still tried to reference their previous
`recipient_id`s.
Move the merging of the two sets earlier, before we call
`remove_denormalized_recipient_column_from_data`.
If there are two huddles, with users A + B + C + D and A + B + C, and
user D is deleted, it is replaced with a mirrordummy user. If
mirrordummy subscriptions are not included in exports, then the two
huddles have duplicate member sets, and will not be able to be
imported successfully.
Include huddle subscriptions for mirrordummy users in exports.
Fundamentally, we should take a write lock on the message, check its
validity for a change, and then make and commit that change.
Previously, `check_update_message` did not operate in a transaction,
but `do_update_message` did -- which led to the ordering:
- `check_update_message` reads Message, not in a transaction
- `check_update_message` verifies properties of the Message
- `do_update_message` starts a transaction
- `do_update_message` takes a read lock on UserMessage
- `do_update_message` writes on UserMessage
- `do_update_message` writes Message
- `do_update_message` commits
This leads to race conditions, where the `check_update_message` may
have verified based on stale data, and `do_update_message` may
improperly overwrite it; as well as deadlocks, where
other (properly-written) codepaths take a write lock on Message
_before_ updating UserMessage, and thus deadlock with
`do_update_message`.
Change `check_update_message` to open a transaction, and take the
write lock when first accessing the Message row. We update the
comment above `do_update_message` to clarify this expectation.
The new ordering is thus:
- `check_update_message` starts a transaction
- `check_update_message` takes a write lock on Message
- `check_update_message` verifies properties of the Message
- `do_update_message` writes on UserMessage
- `do_update_message` writes Message
- `check_update_message` commits
The long-term idle topic participants are soft-reactivated
after email/push notifications are sent due to @topic mention.
The reason being that, generally, @topic mentions are going to
reach a small set of users who have a decent chance of being
reactivated by the notifications.
This commit completes the notifications part of the @topic
wildcard mention feature.
Notifications are sent to the topic participants for the
@topic wildcard mention.
The previous function was poorly named, asked for a
Realm object when realm_id sufficed, and returned a
tuple of strings that had different semantics.
I also avoid calling it duplicate times in a couple
places, although it was probably rarely the case that
both invocations actually happened if upstream
validations were working.
Note that there is a TypedDict called EmojiInfo, so I
chose EmojiData here. Perhaps a better name would be
TinyEmojiData or something.
I also simplify the reaction tests with a verify
helper.
The active realm emoji are just a subset of all your
realm emoji, so just use a single cache entry per
realm.
Cache misses should be very infrequent per realm.
If a realm has lots of deactivated realm emoji, then
there's a minor expense to deserialize them, but that
is gonna be dwarfed by all the other more expensive
operations in message-send.
I also renamed the two related functions. I erred on
the side of using somewhat verbose names, as we don't
want folks to confuse the two use cases. Fortunately
there are somewhat natural affordances to use one or
the other, and mypy helps too.
Finally, I use realm_id instead of realm in places
where we don't need the full Realm object.
This migration is reasonably complex because of various anomalies in existing
data.
Note that there are cases when extra_data does not contain data that is
proper json with possibly single quotes. Thus we need to use
"ast.literal_eval" to cover that.
There is also a special case for "event_type == USER_FULL_NAME_CHANGED",
where extra_data is a plain str. This event_type is only used for
RealmAuditLog, so the zilencer migration script does not need to handle
it.
The migration does not handle "event_type == REALM_DISCOUNT_CHANGED"
because ast.literal_eval only allow Python literals. We expect the admin
to populate the jsonified extra_data for extra_data_json manually
beforehand.
This chunks the backfilling migration to reduce potential block time.
The migration for zilencer is mostly similar to the one for zerver; except that
the backfill helper is added in a wrapper and unrelated events are
removed.
**Logging and error recovery**
We print out a warning when the extra_data_json field of an entry
would have been overwritten by a value inconsistent with what we derived
from extra_data. Usually this only happens when the extra_data was
corrupted before this migration. This prevents data loss by backing up
possibly corrupted data in extra_data_json with the keys
"inconsistent_old_extra_data" and "inconsistent_old_extra_data_json".
More roundtrips to the database are needed for inconsistent data, which are
expected to be infrequent.
This also outputs messages when there are audit log entries with decimals,
indicating that such entries are not backfilled. Do note that audit log
entries with decimals are not populated with "inconsistent_old_extra_data_*"
in the JSONField, because they are not overwritten.
For such audit log entries with "extra_data_json" marked as inconsistent,
we skip them in the migration. Because when we have discovered anomalies in a
previous run, there is no need to overwrite them again nesting the extra keys
we added to it.
**Testing**
We create a migration test case utilizing the property of bulk_create
that it doesn't call our modified save method.
We extend ZulipTestCase to support verifying console output at the test
case level. The implementation is crude but the use case should be rare
enough that we don't need it to be too elaborate.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
We did not send the stream creation events when subscribing
guests to public streams while we do send them when subscribing
non-admin users to private streams.
This commit adds code to send the stream creation events when
subscribing guests to public streams, so the clients can know
that the stream exists and fixes the bug where client tries
to process a subscription add event for a stream which it does
not know about.
This commit updates description for stream creation event
to mention that the event is also sent when user gains
access to a stream either due to being subscribed to it
or if a private stream is made public.
This tracks user group membership changes when the realm is first set
up, either through an import or not. This happens when we add users to
the system user groups by their roles.
For an imported realm, we do extra handling when the data doesn't include
user groups. This gets audited as well.
Django seems to have an aggressive check on the type of a field when
setting it through an relation, requiring the argument to be a UserGroup in
our case.
Reference:
02966a30dd/django/db/models/base.py (L537-L546)
The MissedMessage queue worker is the single callsite of
`handle_missedmessage_emails`, which immediately transforms the list
of events into a dict keyed by message-id.
Skip the intermediate list step, and use defaultdict and a dataclass
to simplify and make explicit the pieces. This removes the unused
user_profile_id and message_id pieces of the data structure.
This commit adds a boolean field `mentions_topic_wildcard`
to the `MessageRenderingResult` dataclass.
The field is set to true only if message rendering determines
the message has an actual topic wildcard mention in it (and not,
e.g., topic wildcard mention syntax inside a code block).
The rendered content for topic wildcard mention is
'<span class="topic-mention">{wildcard}</span>'.
The 'topic-mention' class is the identifier for the wildcard
mention being a topic wildcard mention.
We don't use 'data-user-id="*"' and "user-mention" class for
topic wildcard mentions and eventually plan to remove them for
stream wildcard mentions too in a separate mini-project.
This prep commit merges separate tests for '**@all**',
'**@stream**' and '**@everyone**' stream wildcard mentions
into a single test named 'test_mention_stream_wildcard'.
Similarly, it merges separate tests for '@all', '@stream',
and '@everyone' stream wildcard mentions into a single test
named 'test_mention_at_stream_wildcard'.
The aim is to finally have two separate tests for stream and
topic wildcard mentions (when we introduce topic wildcards)
instead of having separate tests for each mention text
(i.e. all, everyone, stream, topic).
This commit adds the 'topic_wildcard_mention_user_ids' and
'topic_wildcard_mention_in_followed_topic_user_ids'
attributes to the 'RecipientInfoResult' dataclass.
Only topic participants are notified of @topic mentions.
Topic participants are anyone who sent a message to a topic
or reacted to a message on the topic.
'topic_wildcard_mention_in_followed_topic_user_ids' stores the
ids of the topic participants who follow the topic and have
enabled the wildcard mention notifications for followed topics.
'topic_wildcard_mention_user_ids' stores the ids of the topic
participants for whom 'user_allows_notifications_in_StreamTopic'
with setting 'wildcard_mentions_notify' returns True.
This commit adds a 'has_topic_wildcards' instance variable
to the 'MentionData' class for the detection of
- possible topic wildcards mentions.
Fixes part of #22829.
Co-authored-by: Prakhar Pratyush <prakhar841301@gmail.com>
Co-authored-by: orientor <aditya.verma@students.iiit.ac.in>
This commit updates the existing tests in 'test_email_notifications'
and 'test_push_notifications' to properly configure user settings
and visibility policies before running the actual tests.
Earlier, the tests were passing, but the corner case expected
to be covered wasn't covered.
This should have been included in
d80779435a.
These comments should not have been included in
a8fd9eb701.
We covered the case "Private message should soft reactivate
the user" earlier in the test. So the comment was rightly added
there.
During stream wildcard or group mention, no such personal mention
is involved; hence, the comments are not needed.
This is a prep commit to replace 'wildcard' with 'stream_wildcard'.
This wasn't included in 179d5cb because we didn't decide to
use a different rendered_content for topic wildcard mention,
i.e., ''<span class="user-mention topic-mention">{wildcard}</span>'.
Our intention was not to create separate tests for both stream
and topic wildcard mentions, as they were expected to have the
same rendered content format.
Previously this limit was 1 week, which was fine for busy
organizations, but for organizations that send a few messages a week,
or have occasional bursts of activity but the last one was a few weeks
ago, this should give a significantly better new user experience.
There are still caps like 1000 messages total and 20
unread, but we're a bit more flexible about time.
This includes changing the URL to #settings/preferences, with a
transparent redirect so that existing links, like the one from Welcome
Bot, continue to work.
Pass the HttpRequest explicitly through the two webhooks that log to
the webhook loggers.
get_current_request is now unused, so remove it (in the same commit
for test coverage reasons).
Signed-off-by: Anders Kaseorg <anders@zulip.com>
The initial followup_day1 email confirms that the new user account
has been successfully created and should be sent to the user
independently of an organization's setting for send_welcome_emails.
Here we separate out the followup_day1 email into a separate function
from enqueue_welcome_emails and create a helper function for setting
the shared welcome email sender information.
The followup_day1 email is still a scheduled email so that the initial
account creation and log-in process for the user remains unchanged.
Fixes#25268.
The followup_day2 email is scheduled with a delay as a welcome email
and is therefore more likely to exist as a scheduled email in these
deactivation cases.
Updates comment to not include the number of emails generated so
that it doesn't need to be updated every time a new email is added.
The current count in the comment is already out-of-date.
Because the third party might not be expecting a 400 from our
webhooks, we now instead use 200 status code for unknown events,
while sending back the error to Sentry. Because it is no longer an error
response, the response type should now be "success".
Fixes#24721.
This commit removes "@" from name of role-based system groups
since we have added a restricion on having user group names
starting with "@" in the previous commit as they look odd in
mention syntax.
We also add a migration in this commit to update the name of
role-based system groups in existing realms to remove "@"
from the name. This migration also updates the names of
non-system user groups by removing the invalid prefixes
from their names and if there is a group already with that
name, we insted name the group as "group:{group_id}".
Fixes#26148.
We do not allow user group names to start with "@", "role:",
"user:", "stream:" and "channel:".
Group names starting with "@" look odd in mentions and
"role:", "user:" and "stream:" prefixes are reserved for
system groups which will be used in the new groups-based
permission model. We do not allow "channel:" prefix for
now just to be safe in a case where we use it instead of
"stream:" prefix for stream based groups in future.
Fixes part of #26148.
Previously we had database level restriction on length of
user group names. Now we add the same restriction to API
level as well, so we can return a better error response.
We remove the cache functionality for the
get_realm_stream function, and we also change it to
return a thin Stream object (instead of calling
select_related with no arguments).
The main goal here is to remove code complexity, as we
have been prone to at least one caching validation bug
related to how Realm and UserGroup interact. That
particular bug was more theoretical than practical in
terms of its impact, to be clear.
Even if we were to be perfectly disciplined about only
caching thin stream objects and always making sure to
delete cache entries when stream data changed, we would
still be prone to ugly situations like having
transactions get rolled back before we delete the cache
entry. The do_deactivate_stream is a perfect example of
where we have to consider the best time to unset the
cache. If you unset it too early, then you are prone to
races where somebody else churns the cache right before
you update the database. If you set it too late, then
you can have an invalid entry after a rollback or
deadlock situation. If you just eliminate the cache as
a moving part, that whole debate is moot.
As the lack of test changes here indicates, we rarely
fetch streams by name any more in critical sections of
our code.
The one place where we fetch by name is in loading the
home page, but that is **only** when you specify a
stream name. And, of course, that only causes about an
extra millisecond of time.
This changes bulk_get_streams so that it just uses the
database all the time. Also, we avoid calling
select_related(), so that we just get back thin and
tidy Stream objects with simple queries.
About not caching any more:
It's actually pretty rare that we fetch streams by name
in the main application. It's usually API requests that
send in stream names to find more info about streams.
It also turns out that for large queries (>= ~30 rows
for my measurements) it's more efficent to hit the
database than memcached. The database is super fast at
scale; it's just the startup cost of having Django
construct the query, and then having the database do
query planning or whatever, that slows us down. I don't
know the exact bottleneck, but you can clearly measure
that one-row queries are slow (on the order of a full
millisecond or so) but the marginal cost of additional
rows is minimal assuming you have a decent index (20
microseconds per row on my droplet).
All the query-count changes in the tests revolve around
unsubscribing somebody from a stream, and that's a
particularly odd use case for bulk_get_streams, since
you generally unsubscribe from a single stream at a
time. If there are some use cases where you do want to
unsubscribe from multiple streams, we should move
toward passing in stream ids, at least from the
application. And even if we don't do that, our cost for
most queries is a couple milliseconds.
We want to avoid Django going back to the database to
get a realm object that the caller already has.
It's actually currently the case that we often
pre-fetch realm objects when we get stream objects
using get_stream (using a call to select_related() with
no arguments), but that is an expensive operation that
we want to avoid going forward.
This commit prepares us to just fetch slim objects.
This commit creates separate events for issue milestoned and
demilestoned notifications. This allows the end-users to choose
whether they want these notifications or not.
Fixes#25793.
This add audit log entries when any group based setting of a user group
is updated. We store both the old and new values in extra_data, along
with the name of that setting. Entries populated during user group creation
are hardcoded to track "can_mention_group".
Potentially we can adjust "set_defaults_for_group_settings" so that it
populates realm audit logs with it, but that is out of scope for this change.
We use an atomic transaction so that the audit logs are committed
together with the updates.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This add audit log entries when the name or description of a user group
is updated. We store both the old and new values in extra_data. We wrap
the functions inside an atomic transaction so that the audit logs and
the updates are committed together.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This is mostly the same as tracking subgroup changes, except that now
modified_user_group is the subgroup.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
It's worth noting that instead of adding another field to the
RealmAuditLog model, we store the modified subgroup ids in extra_data as
a JSON encoded dict with the key "subgroup_ids". We don't create audit
log entries for supergroup changes at this point.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This also add audit log entries during user creation and role change,
because we modify system group memberships there.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
We also create RealmAuditLog entries for the initial memberships that
get added along with the creation of a UserGroup. System user groups are
not created with members so no audit logs are populated for that.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This helps reduce the impact on busy uwsgi processes in case there are
slow timeout failures of Sentry servers. The p99 is less than 300ms,
and p99.9 per day peaks at around 1s, so this will not affect more
than .1% of requests in normal operation.
This is not a complete solution (see #26229); it is merely stop-gap
mitigation.
Various cleanups:
* clean up comments
* improve names for constants and variables
* express first ORM query as a single statement
* use set differences to simplify logic
* avoid all the reversing churn
* avoid early-exit idiom since this function is so small
Note that it's plausible that we should just combine the two
queries and let the database exclude the already-used ids,
but that felt a little risky for now. As I mentioned on
Zulip, I think the one-week window has dubious value, but
I am biased by having wasted time chasing down a test
flake related to the time window.
Basically, I eliminate the use of select_all() in a query
that still makes a single round trip. We have good test
enforcement that Django never needs to lazily fetch
objects off the Stream object. (It used to be common
to fetch stream.realm a while back, but we upgraded
bulk_add_subscription, in particular, a while back.)
We extract code from process_new_human_user with
no modifications.
This has all the best outcomes of extracting a function:
* better profile info
* easier to test for query counts (signup gets real noisy)
* simplifies a long, messy function
It has no real drawbacks, since the helper function doesn't need
to pass back any intermediate state to the parent for the rest
of what the parent does.
When you profile test_signup and test_invite, with a decent
sample size, the set_up_streams_for_new_human_user function
does about 20% of the work for process_new_human_user, which
is a lot considering that most tests don't create a ton of
pre-registered or default streams.
At least as measured by test_events.py, which has over 1000
calls to fetch initial data for page loads, this should
be about a 10% improvement in how much time the server
spends fetching data.
We mostly avoid a select_related() query that did this nastiness:
INNER JOIN "zerver_realm" ON ("zerver_stream"."realm_id" = "zerver_realm"."id")
INNER JOIN "zerver_usergroup" ON ("zerver_stream"."can_remove_subscribers_group_id" = "zerver_usergroup"."id")
INNER JOIN "zerver_realm" T4 ON ("zerver_usergroup"."realm_id" = T4."id")
INNER JOIN "zerver_usergroup" T5 ON ("zerver_usergroup"."can_mention_group_id" = T5."id")
INNER JOIN "zerver_realm" T6 ON (T5."realm_id" = T6."id")
INNER JOIN "zerver_usergroup" T7 ON (T5."can_mention_group_id" = T7."id")
INNER JOIN "zerver_realm" T8 ON (T7."realm_id" = T8."id")
INNER JOIN "zerver_usergroup" T9 ON (T7."can_mention_group_id" = T9."id")
INNER JOIN "zerver_realm" T10 ON (T9."realm_id" = T10."id")
INNER JOIN "zerver_usergroup" T11 ON (T9."can_mention_group_id" = T11."id")
WHERE "zerver_stream"."id" IN (SELECT U0."stream_id" FROM "zerver_defaultstream" U0 WHERE U0."realm_id" = 2
Future commits will address the codepath for creating users.
I created zerver/lib/default_streams.py, so that various
views and events.py don't have to awkwardly reach into
an "actions" file.
I copied over two functions verbatim from actions/default_streams.py:
get_default_streams_for_realm
streams_to_dicts_sorted
The latter only remains as an internal detail in the new library.
I also created two new helpers:
get_default_stream_ids_for_realm:
This is both faster and easier to use in all the places
where we only need to get a set of default stream ids.
get_default_streams_for_realm_as_dicts:
This just wraps the prior calls to
streams_to_dicts_sorted(get_default_streams_for_realm(...)),
and it doesn't yet address the slowness of the underlying
code.
All the "real" code should be functionally the same.
In a few tests I now use this wrapper instead of
calling get_default_streams_for_realm, just to get
slightly deeper coverage.
Updates find_proper_insertion_index to check for the inline image
classes as matching at least one of the classes in the element's
attrib["class"] so that cases where an inline preview image has
multiple classes, like YouTube video previews, will have the
correct insertion index.
Fixes#26186.
Added an additional test case to `test_submessages.py` for testing the
message object containing `submessages` meta data.
Previous to this commit we were never validating the `submessage` schema
in the `message` objects.
Fixes#25896.
By relocating helper methods into a mixin class, we can be more flexible
with managing transactions in test cases, without always forcing the
django.test.TestCase behavior of always putting the test case into an
atomic transaction.
We include a check for side effects in ZulipTransactionTestCase. It only
checks for the set of row ids in all tables before and after each test.
It is not a comprehensive check for side effects, but should be
sufficient for the basics without much performance overhead.
It replaces the "File not found." text with:
"This file does not exist or has been deleted."
At present when a file is deleted it results in a confusing
experience when looking at the "File not found." message.
In order to clarify the situation is not a bug, the message
has been replaced with a better alternative.
Fixes part of Issue #23739.
This prep commit replaces the 'wildcard' keyword in the codebase
with 'stream_wildcard' at some places for better readability, as
we plan to introduce 'topic_wildcards' as a part of the
'@topic mention' project.
Currently, 'wildcards = ["all", "everyone", "stream"]' which is an
alias to mention everyone in the stream, hence better renamed as
'stream_wildcards'.
Eventually, we will have:
'stream_wildcard' as an alias to mention everyone in the stream.
'topic_wildcard' as an alias to mention everyone in the topic.
'wildcard' refers to 'stream_wildcard' and 'topic_wildcard' as a whole.
The 'get_gcm_alert' and 'get_apns_alert_subtitle' functions
don't include the case when the trigger is
'NotificationTriggers.FOLLOWED_TOPIC_WILDCARD_MENTION'.
This commit updates the functions to include
'NotificationTriggers.FOLLOWED_TOPIC_WILDCARD_MENTION'.
The emails sent for missed messages have a text at the bottom
explaining the reason why the email was sent.
This commit reorders the conditional statements in the email
template to align with the trigger priority order defined
in the 'get_email_notification_trigger'.
This commit fixes the incorrect calculation of the
'senders' list.
The effect of 'followed_topic_wildcard_mention'
wasn't considered earlier.
The bug was introduced in b052c8980e.
This commit uses 'NotificationTriggers' class attributes
instead of directly using loose strings.
This should have been ideally included in the commit
c3319a5231.
Combine nginx and Django middlware to stop putting misleading warnings
about `CSRF_TRUSTED_ORIGINS` when the issue is untrusted proxies.
This attempts to, in the error logs, diagnose and suggest next steps
to fix common proxy misconfigurations.
See also #24599 and zulip/docker-zulip#403.
Having exactly 17 or 18 middlewares, on Python 3.11.0 and above,
causes python to segfault when running tests with coverage; see
https://github.com/python/cpython/issues/106092
Work around this by adding one or two no-op middlewares if we would
hit those unlucky numbers. We only add them in testing, since
coverage is a requirement to trigger it, and there is no reason to
burden production with additional wrapping.
It takes about 31ms per page on my box, but 191
help pages adds up quickly. I am not sure how to
optimize this test, but it will be a good litmus
test for a future better markdown processor.
This did not speed up the tests as much as I expected,
but it certainly makes the code easier to read, and
Tim is pretty confident that the zephyr logic is
fairly stable, so it's sufficient to test it on a
subset of representative urls.
dbe930394f changed the
"missing string" from "Log in" to "xyz" for some
unknown reason. The current code makes no sense.
Also, even the original test code here had the common
pitfall of only testing one side of the condition.
Presumably if you are testing that a certain string
is missing in a landing-page scenario, then you also
want to check that it **does** exist in other
scenarios. Otherwise, the flag would have been
named something more generic. Of course, I am mostly
guessing due to lack of comments.
If there is some test logic here that we need to
resurrect, then we should just write a custom test
for the /hello page rather than crufting up
all our helpers.
Anders Kaseorg
Steve Howell
Alex Vandiver
David Rosa
abdullahm1
Prakhar Pratyush
Zixuan James Li
Zixuan Li
Satyam Bansal
Sahil Batra
nimish
Lauryn Menard
Lalit
arghyadeep10