29 KiB
Writing a new application feature
The changes needed to add a new feature will vary, of course, but this document provides a general outline of what you may need to do, as well as an example of the specific steps needed to add a new feature: adding a new option to the application that is dynamically synced through the data system in real-time to all browsers the user may have open.
As you read this, you may find you need to learn about Zulip's real-time push system; the real-time push and events documentation has a detailed explanation of how everything works. You may also find it beneficial to read Zulip's architectural overview. Zulip is a web application built using the Django framework, and some of the processes listed in this tutorial, such as database migrations and tests, use Django's tooling.
Zulip's directory structure will also be helpful to review when creating a new feature. Many aspects of the structure will be familiar to Django developers. Visit Django's documentation for more information about how Django projects are typically organized. And finally, the message sending documentation on the additional complexity involved in sending messages.
General process
Files impacted
This tutorial will walk through adding a new feature to a Realm (an organization in Zulip). The following files are involved in the process:
Backend
zerver/models.py
: Defines the database model.zerver/views/realm.py
: The view function that implements the API endpoint for editing realm objects.zerver/actions/realm_settings.py
: Contains code for updating and interacting with the database.zerver/lib/events.py
: Ensures that the state Zulip sends to clients is always consistent and correct.
Frontend
web/templates/settings/organization_permissions_admin.hbs
: defines the structure of the admin permissions page (checkboxes for each organization permission setting).web/src/settings_org.js
: handles organization setting form submission.web/src/server_events_dispatch.js
: handles events coming from the server (ex: pushing an organization change to other open browsers and updating the application's state).
Backend testing
zerver/tests/test_realm.py
: end-to-end API tests for updating realm settings.zerver/tests/test_events.py
: tests for possible race bugs in the zerver/lib/events.py implementation.
Frontend testing
web/e2e-tests/admin.test.ts
: end-to-end tests for the organization admin settings pages.web/tests/dispatch.test.js
Documentation
zerver/openapi/zulip.yaml
: OpenAPI definitions for the Zulip REST API.api_docs/changelog.md
: documentation listing all changes to the Zulip Server API.help/...
: end user facing documentation (Help Center) for the application.
Adding a field to the database
Update the model: The server accesses the underlying database in
zerver/models.py
. Add a new field in the appropriate class.
Create and run the migration: To create and apply a migration, run the following commands:
./manage.py makemigrations
./manage.py migrate
You can read our database migration documentation to learn more about creating and applying database migrations.
Test your changes: Once you've run the migration, flush memcached
on your development server (./scripts/setup/flush-memcached
) and then
restart the development server
to avoid interacting with cached objects.
Backend changes
We have a framework that automatically handles many of the steps for the
most common types of UserProfile and Realm settings. We refer to this as the
property_types
framework. However, it is valuable to understand
the flow of events even if the property_types
framework means you don't
have to write much code for a new setting.
Database interaction: Add any necessary code for updating and
interacting with the database in zerver/actions/realm_settings.py
. It should
update the database and send an event announcing the change.
Application state: Modify the fetch_initial_state_data
and
apply_event
functions in zerver/lib/events.py
to update the state
based on the event you just created.
Backend implementation: Make any other modifications to the backend required for your feature to do what it's supposed to do (this will be unique to the feature you're implementing).
New views: Add any new application views to zproject/urls.py
, or
update the appropriate existing view in zerver/views/
. This
includes both views that serve HTML (new pages on Zulip) as well as new
API endpoints that serve JSON-formatted data.
Testing: At the very least, add a test of your event data flowing
through the system in test_events.py
and an API test (e.g. for a
Realm setting, in test_realm.py
).
Frontend changes
JavaScript/TypeScript: Zulip's JavaScript and TypeScript sources are
located in the directory web/src/
. The exact files you may need to change
depend on your feature. If you've added a new event that is sent to clients,
be sure to add a handler for it in web/src/server_events_dispatch.js
.
CSS: The primary CSS file is web/styles/zulip.css
. If your new
feature requires UI changes, you may need to add additional CSS to this
file.
Templates: The initial page structure is rendered via Jinja2
templates located in templates/zerver/app
. For JavaScript, Zulip uses
Handlebars templates located in web/templates
. Templates are
precompiled as part of the build/deploy process.
Zulip is fully internationalized, so when writing both HTML templates or JavaScript/TypeScript/Python code that generates user-facing strings, be sure to tag those strings for translation.
Testing: There are two types of frontend tests: node-based unit
tests and blackbox end-to-end tests. The blackbox tests are run in a
headless Chromium browser using Puppeteer and are located in
web/e2e-tests/
. The unit tests use Node's assert
module are located in web/tests/
. For more
information on writing and running tests, see the
testing documentation.
Documentation changes
After implementing the new feature, you should document it and update any existing documentation that might be relevant to the new feature. For detailed information on the kinds of documentation Zulip has, see Documentation.
Help center documentation: You will likely need to at least update,
extend and link to articles in the help/
directory that are related
to your new feature. Writing help center articles
provides more detailed information about writing and editing feature
help/
directory articles.
API documentation: A new feature will probably impact the REST API
documentation as well, which will mean updating zerver/openapi/zulip.yaml
and modifying api_docs/changelog.md
for a new feature
level. Documenting REST API endpoints
explains Zulip's API documentation system and provides a step by step
guide to adding or updating documentation for an API endpoint.
Example feature
This example describes the process of adding a new setting to Zulip: a flag that allows an admin to require topics on stream messages (the default behavior is that topics can have no subject). This flag is an actual Zulip feature. You can review the original commit in the Zulip repo. (This commit displays the work of setting up a checkbox for the feature on the admin settings page, communicating and saving updates to the setting to the database, and updating the state of the application after the setting is updated. For the code that accomplishes the underlying task of requiring messages to have a topic, you can view this commit.)
Update the model
First, update the database and model to store the new setting. Add a new
boolean field, mandatory_topics
, to the Realm model in
zerver/models.py
.
# zerver/models.py
class Realm(models.Model):
# ...
emails_restricted_to_domains = models.BooleanField(default=True)
invite_required = models.BooleanField(default=False)
+ mandatory_topics = models.BooleanField(default=False)
The Realm model also contains an attribute, property_types
, which
other backend functions use to handle most realm settings without any custom
code for the setting (more on this process below). The attribute is a
dictionary, where the key is the name of the realm field and the value
is the field's type. Add the new field to the property_types
dictionary.
# zerver/models.py
class Realm(models.Model)
# ...
# Define the types of the various automatically managed properties
property_types = dict(
add_custom_emoji_policy=int,
allow_edit_history=bool,
# ...
+ mandatory_topics=bool,
# ...
The majority of realm settings can be included in
property_types
. However, there are some properties that need custom
logic and thus cannot use this framework. For example:
- The realm
authentication_methods
attribute is a bitfield and needs additional code for validation and updating. - The
allow_message_editing
andmessage_content_edit_limit_seconds
fields depend on one another, so they are also handled separately and not included inproperty_types
.
When creating a realm property that is not a boolean, Text or
integer field, or when adding a field that is dependent on other fields,
do not add the field to the property_types
dictionary. The steps
below will point out where to write additional code for these cases.
Create the migration
Create the migration file using the Django makemigrations
command:
./manage.py makemigrations
. Make sure to commit the generated file to git:
git add zerver/migrations/NNNN_realm_mandatory_topics.py
(NNNN is a number that is equal to the number of migrations.)
If you run into problems, the Django migration documentation is helpful.
Test your migration changes
Apply the migration using Django's migrate
command, ./manage.py migrate
:
shell $ ./manage.py migrate
Operations to perform:
Synchronize unmigrated apps: staticfiles, analytics, pipeline
Apply all migrations: zilencer, confirmation, sessions, guardian, zerver, sites, auth, contenttypes
Synchronizing apps without migrations:
Creating tables...
Running deferred SQL...
Installing custom SQL...
Running migrations:
Rendering model states... DONE
Applying zerver.NNNN_realm_mandatory_topics... OK
Once you've run the migration, restart memcached on your development
server (/etc/init.d/memcached restart
) and then restart the development server
to avoid interacting with cached objects.
Handle database interactions
Next, we will implement the backend part of this feature. Like typical apps, we will need our backend to update the database and send some response to the client that made the request.
Beyond that, we need to orchestrate notifications about the setting change to other clients (or other users, if you will). Clients find out about settings through two closely related code paths. When a client first contacts the server, the server sends the client its initial state. Subsequently, clients subscribe to "events," which can (among other things) indicate that settings have changed.
For the backend piece, we will need our action to make a call to send_event
to send the event to clients that are active. We will also need to
modify fetch_initial_state_data
so that the new field is passed to
clients. See our event system docs for all the
gory details.
Anyway, getting back to implementation details...
If you are working on a feature that is in the realm property_types
dictionary, you will not need to add code to zerver/actions/realm_settings.py
, but
we will describe what the process in that file does:
In zerver/actions/realm_settings.py
, the function do_set_realm_property
takes
in the name of a realm property to update and the value it should
have. This function updates the database and triggers an event to
notify clients about the change. It uses the field's type, specified
in the Realm.property_types
dictionary, to validate the type of the
value before updating the property; this is primarily an assertion to
help catch coding mistakes, not to check for bad user input.
After updating the given realm field, do_set_realm_property
creates
an 'update' event with the name of the property and the new value. It
then calls send_event
, passing the event and the list of users whose
browser sessions should be notified as the second argument. The latter
argument can be a single user (if the setting is a personal one, like
time display format), members in a particular stream only or all
active users in a realm.
# zerver/actions/realm_settings.py
def do_set_realm_property(
realm: Realm, name: str, value: Any, *, acting_user: Optional[UserProfile]
) -> None:
"""Takes in a realm object, the name of an attribute to update, the
value to update and and the user who initiated the update.
"""
property_type = Realm.property_types[name]
assert isinstance(value, property_type), (
'Cannot update %s: %s is not an instance of %s' % (
name, value, property_type,))
setattr(realm, name, value)
realm.save(update_fields=[name])
event = dict(
type='realm',
op='update',
property=name,
value=value,
)
send_event(realm, event, active_user_ids(realm))
If the new realm property being added does not fit into the
property_types
framework (such as the authentication_methods
field), you'll need to create a new function to explicitly update this
field and send an event. For example:
# zerver/actions/realm_settings.py
def do_set_realm_authentication_methods(
realm: Realm, authentication_methods: Dict[str, bool], *, acting_user: Optional[UserProfile]
) -> None:
for key, value in list(authentication_methods.items()):
index = getattr(realm.authentication_methods, key).number
realm.authentication_methods.set_bit(index, int(value))
realm.save(update_fields=['authentication_methods'])
event = dict(
type="realm",
op="update_dict",
property='default',
data=dict(authentication_methods=realm.authentication_methods_dict())
)
send_event(realm, event, active_user_ids(realm))
Update application state
zerver/lib/events.py
contains code to ensure that your new setting is included
in the data sent down to clients: both when a new client is loaded
and when changes happen. This file also automatically
handles realm settings in the property_types
dictionary, so you would
not need to change this file if your setting fits that framework.
The fetch_initial_state_data
function is responsible for sending data when
a client is loaded (data added to the state
here will be available both
in page_params
in the browser, as well as to API clients like the mobile
apps). The apply_event
function in zerver/lib/events.py
is important for
making sure the state
is always correct, even in the event of rare
race conditions.
# zerver/lib/events.py
def fetch_initial_state_data(user_profile, event_types, queue_id, include_subscribers=True):
# ...
if want('realm'):
for property_name in Realm.property_types:
state['realm_' + property_name] = getattr(user_profile.realm, property_name)
state['realm_authentication_methods'] = user_profile.realm.authentication_methods_dict()
state['realm_allow_message_editing'] = user_profile.realm.allow_message_editing
# ...
def apply_event
user_profile: UserProfile,
# ...
) -> None:
for event in events:
# ...
elif event['type'] == 'realm':
field = 'realm_' + event['property']
state[field] = event['value']
# ...
If your new realm property fits the property_types
framework, you don't need to change fetch_initial_state_data
or
apply_event
. However, if you are adding a
property that is handled separately, you will need to explicitly add
the property to the state
dictionary in the fetch_initial_state_data
function. E.g., for authentication_methods
:
# zerver/lib/events.py
def fetch_initial_state_data(user_profile, event_types, queue_id, include_subscribers=True):
# ...
if want('realm'):
# ...
state['realm_authentication_methods'] = user_profile.realm.authentication_methods_dict()
# ...
For this setting, one won't need to change apply_event
since its
default code for realm
event types handles this case correctly, but
for a totally new type of feature, a few lines in that function may be
needed.
Add a new view
You will need to add a view for clients to access that will call the
actions.py
code to update the database. This example feature
adds a new parameter that will be sent to clients when the
application loads and should be accessible via JavaScript. There is
already a view that does this for related flags: update_realm
in
zerver/views/realm.py
. So in this case, we can add our code to the
existing view instead of creating a new one.
You'll need to add a parameter for the new field to the update_realm
function in zerver/views/realm.py
(and add the appropriate mypy type
annotation).
# zerver/views/realm.py
def update_realm(
request: HttpRequest,
user_profile: UserProfile,
name: Optional[str] = REQ(str_validator=check_string, default=None),
# ...
+ mandatory_topics: Optional[bool] = REQ(json_validator=check_bool, default=None),
# ...
):
# ...
If this feature fits the property_types
framework and does
not require additional validation, this is the only change to make
to zerver/views/realm.py
.
Text fields or other realm properties that need additional validation
can be handled at the beginning of update_realm
.
# zerver/views/realm.py
# Additional validation/error checking beyond types go here, so
# the entire request can succeed or fail atomically.
if default_language is not None and default_language not in get_available_language_codes():
raise JsonableError(_("Invalid language '%s'" % (default_language,)))
if description is not None and len(description) > 100:
raise JsonableError(_("Realm description cannot exceed 100 characters."))
# ...
The code in update_realm
loops through the property_types
dictionary
and calls do_set_realm_property
on any property to be updated from
the request.
If the new feature is not in property_types
, you will need to write code
to call the function you wrote in actions.py
that updates the database
with the new value. E.g., for authentication_methods
, we created
do_set_realm_authentication_methods
, which we will call here:
# zerver/views/realm.py
# import do_set_realm_authentication_methods from actions.py
from zerver.actions.realm_settings import (
do_reactivate_realm,
do_set_realm_authentication_methods,
# ...
)
# ...
# ...
if authentication_methods is not None and realm.authentication_methods_dict() != authentication_methods:
do_set_realm_authentication_methods(realm, authentication_methods, acting_user=user_profile)
data['authentication_methods'] = authentication_methods
# ...
This completes the backend implementation. A great next step is to write automated backend tests for your new feature.
Backend tests
To test the new setting syncs correctly with the property_types
framework, one usually just needs to add a line in each of
test_events.py
and test_realm.py
with a list of values to switch
between in the test. In the case of a boolean field, no action is
required, because those tests will correctly assume that the only
values to test are True
and False
.
In test_events.py
, the function that runs tests for the property_types
framework is do_set_realm_property_test
, and in test_realm.py
, it is
do_test_realm_update_api
.
One still needs to add a test for whether the setting actually controls the feature it is supposed to control, however (e.g. for this example feature, whether sending a message without a topic fails with the setting enabled).
Visit Zulip's Django testing documentation to learn more about the backend testing framework.
Also note that you may already need to update the API documentation for your new feature to pass new or existing backend tests at this point. The tutorial for writing REST API endpoints can be a helpful resource, especially the section on debugging schema validation errors.
Update the frontend
After completing the process of adding a new feature on the backend, you should make the required frontend changes: in this case, a checkbox needs to be added to the admin page (and its value added to the data sent back to server when a realm is updated) and the change event needs to be handled on the client.
To add the checkbox to the admin page, modify the relevant template in
web/templates/settings/
, which can be
organization_permissions_admin.hbs
or organization_settings_admin.hbs
(omitted here since it is relatively straightforward).
If you're adding a non-checkbox field, you'll need to specify the type
of the field via the data-setting-widget-type
attribute in the HTML
template.
Then add the new form control in web/src/admin.js
.
// web/src/admin.js
export function build_page() {
const options = {
custom_profile_field_types: page_params.custom_profile_field_types,
full_name: page_params.full_name,
realm_name: page_params.realm_name,
// ...
+ realm_mandatory_topics: page_params.mandatory_topics,
// ...
The JavaScript code for organization settings and permissions can be found in
web/src/settings_org.js
.
In frontend, we have split the property_types
into three objects:
-
org_profile
: This contains properties for the "organization profile" settings page. -
org_settings
: This contains properties for the "organization settings" page. Settings belonging to this section generally decide what features should be available to a user like deleting a message, message edit history etc. Ourmandatory_topics
feature belongs in this section. -
org_permissions
: This contains properties for the "organization permissions" section. These properties control security controls like who can join the organization and whether normal users can create streams or upload custom emoji.
Once you've determined whether the new setting belongs, the next step
is to find the right subsection of that page to put the setting
in. For example in this case of mandatory_topics
it will lie in
"Other settings" (other_settings
) subsection.
If you're not sure in which section your feature belongs, it's better to discuss it in the Zulip development community before implementing it.
Note that some settings, like realm_msg_edit_limit_setting
,
require special treatment, because they don't match the common
pattern. We can't extract the property name and compare the value of
such input elements with those in page_params
, so we have to
manually handle such situations in a couple key functions:
-
settings_org.get_property_value
: This processes the property name when it doesn't match a corresponding key inpage_params
, and returns the current value of that property, which we can use to compare and set the values of corresponding DOM element. -
settings_org.update_dependent_subsettings
: This handles settings whose value and state depend on other elements. For example,realm_waiting_period_threshold_custom_input
is only shown for with the right state ofrealm_waiting_period_threshold
.
Finally, update server_events_dispatch.js
to handle related events coming from
the server. There is an object, realm_settings
, in the function
dispatch_normal_event
. The keys in this object are setting names and the
values are the UI updating functions to run when an event has occurred.
If there is no relevant UI change to make other than in settings page
itself, the value should be noop
(this is the case for
mandatory_topics
, since this setting only has an effect on the
backend, so no UI updates are required.).
However, if you had written a function to update the UI after a given
setting has changed, your function should be referenced in the
realm_settings
of server_events_dispatch.js
. See for example
settings_emoji.update_custom_emoji_ui
.
// web/src/server_events_dispatch.js
function dispatch_normal_event(event) {
switch (event.type) {
// ...
case 'realm':
var realm_settings = {
add_custom_emoji_policy: settings_emoji.update_custom_emoji_ui,
allow_edit_history: noop,
// ...
+ mandatory_topics: noop,
// ...
};
Checkboxes and other common input elements handle the UI updates
automatically through the logic in settings_org.sync_realm_settings
.
The rest of the dispatch_normal_events
function updates the state of the
application if an update event has occurred on a realm property and runs
the associated function to update the application's UI, if necessary.
Here are few important cases you should consider when testing your changes:
-
For organization settings where we have a "save/discard" model, make sure both the "Save" and "Discard changes" buttons are working properly.
-
If your setting is dependent on another setting, carefully check that both are properly synchronized. For example, the input element for
realm_waiting_period_threshold_custom_input
is shown only when we have selected the custom time limit option in therealm_waiting_period_threshold
dropdown. -
Do some manual testing for the real-time synchronization of input elements across the browsers and just like "Discard changes" button, check whether dependent settings are synchronized properly (this is easy to do by opening two browser windows to the settings page, and making changes in one while watching the other).
-
Each subsection has independent "Save" and "Discard changes" buttons, so changes and saving in one subsection shouldn't affect the others.
Frontend tests
A great next step is to write frontend tests. There are two types of frontend tests: node-based unit tests and Puppeteer end-to-end tests.
At the minimum, if you created a new function to update UI in
settings_org.js
, you will need to mock that function in
web/tests/dispatch.test.js
. Add the name of the UI
function you created to the following object with noop
as the value:
// web/tests/dispatch.test.js
set_global('settings_org', {
update_email_change_display: noop,
update_name_change_display: noop,
});
Beyond that, you should add any applicable tests that verify the behavior of the setting you just created.
Update documentation
Nice job! You've added a new feature to Zulip that will improve user and contributor experiences with the app, which is why it's really important to make sure that your new feature is well documented.
This example feature adds new functionality that requires messages to have topics if the setting is enabled. A recommended way to document this feature would be to update and/or augment Zulip's existing help center documentation to reflect your changes and additions.
At the very least, this will involve modifying (or adding) a Markdown
file documenting the feature in the help/
directory of the main Zulip
server repository, where the source for Zulip's end user documentation
is stored. Details about writing, editing and testing these Markdown
files can be found in:
Writing help center articles.
Also, new features will often impact Zulip's REST API documentation,
which is found in zerver/openapi/zulip.yaml
. You may have noticed
this during the testing process as the Zulip test suite should fail if
there is a change to the API without a corresponding update to the
documentation.
The best way to understand writing and updating Zulip's API documentation is to read more about Zulip's REST API documentation process and OpenAPI configuration.
In particular, if there is an API change, make sure you document
your new feature in api_docs/changelog.md
and bump the
API_FEATURE_LEVEL
in version.py
. The API feature level allows the
developers of mobile clients and other tools using the Zulip API to
programmatically determine whether the Zulip server they are
interacting with supports a given feature; see the
Zulip release lifecycle.