* Disable Percy While visual regression testing is without doubt an essential part of quality assurance, the way we're currently using Percy brings little to no value while causing mainenance overhead. Therefore, we're disabling it until we have figure out https://github.com/vector-im/wat-internal/issues/36 and https://github.com/vector-im/wat-internal/issues/56. * Also disable nightly build * Comment out Percy step instead of fiddling with the output value * Appease the frigging linter
13 KiB
Cypress in Element Web
Contents
- How to run the tests
- How the tests work
- How to write great Cypress tests
- Visual testing
Running the Tests
Our Cypress tests run automatically as part of our CI along with our other tests, on every pull request and on every merge to develop & master.
However the Cypress tests are run, an element-web must be running on
http://localhost:8080 (this is configured in cypress.json
) - this is what will
be tested. When running Cypress tests yourself, the standard yarn start
from the
element-web project is fine: leave it running it a different terminal as you would
when developing.
The tests use Docker to launch Homeserver (Synapse or Dendrite) instances to test against, so you'll also need to have Docker installed and working in order to run the Cypress tests.
There are a few different ways to run the tests yourself. The simplest is to run:
docker pull matrixdotorg/synapse:develop
yarn run test:cypress
This will run the Cypress tests once, non-interactively.
Note: you don't need to run the docker pull
command every time, but you should
do it regularly to ensure you are running against an up-to-date Synapse.
You can also run individual tests this way too, as you'd expect:
yarn run test:cypress --spec cypress/e2e/1-register/register.spec.ts
Cypress also has its own UI that you can use to run and debug the tests. To launch it:
yarn run test:cypress:open
Matching the CI environment
In our Continuous Integration environment, we run the Cypress tests in the Chrome browser, and with the latest Synapse image from Docker Hub.
In some rare cases, tests behave differently between different browsers, so if you see CI failures for the Cypress tests, but those tests work OK on your local machine, try running them in Chrome like this:
yarn run test:cypress --browser=chrome
(Use --browser=chromium
if you'd prefer to use Chromium.)
If you launch the interactive UI you can choose the browser you want to use. To match the CI setup, choose Chrome.
Note that you will need to have Chrome installed on your system to run the tests inside those browsers, whereas the default is to use Electron, which is included within the Cypress dependency.
Another cause of inconsistency between local and CI is the Synapse version. The first time you run the tests, they automatically fetch the latest Docker image of Synapse, but this won't update again unless you do it explicitly. To update the Synapse you are using, run:
docker pull matrixdotorg/synapse:develop
and then run the tests as normal.
Running with Rust cryptography
matrix-js-sdk
is currently in the
process of being
updated to replace its end-to-end encryption implementation to use the Matrix
Rust SDK. This is not currently
enabled by default, but it is possible to have Cypress configure Element to use
the Rust crypto implementation by setting the environment variable
CYPRESS_RUST_CRYPTO=1
.
How the Tests Work
Everything Cypress-related lives in the cypress/
subdirectory of react-sdk
as is typical for Cypress tests. Likewise, tests live in cypress/e2e
.
cypress/plugins/synapsedocker
contains a Cypress plugin that starts instances
of Synapse in Docker containers. These synapses are what Element-web runs against
in the Cypress tests.
Synapse can be launched with different configurations in order to test element
in different configurations. cypress/plugins/synapsedocker/templates
contains
template configuration files for each different configuration.
Each test suite can then launch whatever Synapse instances it needs in whatever configurations.
Note that although tests should stop the Homeserver instances after running and the
plugin also stop any remaining instances after all tests have run, it is possible
to be left with some stray containers if, for example, you terminate a test such
that the after()
does not run and also exit Cypress uncleanly. All the containers
it starts are prefixed, so they are easy to recognise. They can be removed safely.
After each test run, logs from the Synapse instances are saved in cypress/synapselogs
with each instance in a separate directory named after its ID. These logs are removed
at the start of each test run.
Writing Tests
Mostly this is the same advice as for writing any other Cypress test: the Cypress docs are well worth a read if you're not already familiar with Cypress testing, eg. https://docs.cypress.io/guides/references/best-practices. To avoid your tests being flaky it is also recommended to give https://docs.cypress.io/guides/core-concepts/retry-ability a read.
Getting a Synapse
The key difference is in starting Synapse instances. Tests use this plugin via
cy.startHomeserver()
to provide a Homeserver instance to log into:
cy.startHomeserver("consent").then((result) => {
homeserver = result;
});
This returns an object with information about the Homeserver instance, including what port
it was started on and the ID that needs to be passed to shut it down again. It also
returns the registration shared secret (registrationSecret
) that can be used to
register users via the REST API. The Homeserver has been ensured ready to go by awaiting
its internal health-check.
Homeserver instances should be reasonably cheap to start (you may see the first one take a
while as it pulls the Docker image), so it's generally expected that tests will start a
Homeserver instance for each test suite, i.e. in before()
, and then tear it down in after()
.
To later destroy your Homeserver you should call stopHomeserver
, passing the HomeserverInstance
object you received when starting it.
cy.stopHomeserver(homeserver);
Synapse Config Templates
When a Synapse instance is started, it's given a config generated from one of the config
templates in cypress/plugins/synapsedocker/templates
. There are a couple of special files
in these templates:
homeserver.yaml
: Template substitution happens in this file. Template variables are:REGISTRATION_SECRET
: The secret used to register users via the REST API.MACAROON_SECRET_KEY
: Generated each time for securityFORM_SECRET
: Generated each time for securityPUBLIC_BASEURL
: The localhost url + port combination the synapse is accessible at
localhost.signing.key
: A signing key is auto-generated and saved to this file. Config templates should not contain a signing key and instead assume that one will exist in this file.
All other files in the template are copied recursively to /data/
, so the file foo.html
in a template can be referenced in the config as /data/foo.html
.
Logging In
There exists a basic utility to start the app with a random user already logged in:
cy.initTestUser(homeserver, "Jeff");
It takes the HomeserverInstance you received from startHomeserver
and a display name for your test user.
This custom command will register a random userId using the registrationSecret with a random password
and the given display name. The returned Chainable will contain details about the credentials for if
they are needed for User-Interactive Auth or similar but localStorage will already be seeded with them
and the app loaded (path /
).
The internals of how this custom command run may be swapped out later, but the signature can be maintained for simpler maintenance.
Joining a Room
Many tests will also want to start with the client in a room, ready to send & receive messages. Best
way to do this may be to get an access token for the user and use this to create a room with the REST
API before logging the user in. You can make use of cy.getBot(homeserver)
and cy.getClient()
to do this.
Convenience APIs
We should probably end up with convenience APIs that wrap the homeserver creation, logging in and room creation that can be called to set up tests.
Try to write tests from the users's perspective
Like for instance a user will not look for a button by querying a CSS selector. Instead you should work
with roles / labels etc.. You can make use of cy.findBy…
queries provided by
Cypress Testing Library.
Using matrix-js-sdk
Due to the way we run the Cypress tests in CI, at this time you can only use the matrix-js-sdk module
exposed on window.matrixcs
. This has the limitation that it is only accessible with the app loaded.
This may be revisited in the future.
Good Test Hygiene
This section mostly summarises general good Cypress testing practice, and should not be news to anyone already familiar with Cypress.
- Test a well-isolated unit of functionality. The more specific, the easier it will be to tell what's wrong when they fail.
- Don't depend on state from other tests: any given test should be able to run in isolation.
- Try to avoid driving the UI for anything other than the UI you're trying to test. e.g. if you're testing that the user can send a reaction to a message, it's best to send a message using a REST API, then react to it using the UI, rather than using the element-web UI to send the message.
- Avoid explicit waits.
cy.get()
will implicitly wait for the specified element to appear and all assertions are retired until they either pass or time out, so you should never need to manually wait for an element.- For example, for asserting about editing an already-edited message, you can't wait for the 'edited' element to appear as there was already one there, but you can assert that the body of the message is what is should be after the second edit and this assertion will pass once it becomes true. You can then assert that the 'edited' element is still in the DOM.
- You can also wait for other things like network requests in the browser to complete (https://docs.cypress.io/guides/guides/network-requests#Waiting). Needing to wait for things can also be because of race conditions in the app itself, which ideally shouldn't be there!
This is a small selection - the Cypress best practices guide, linked above, has more good advice, and we should generally try to adhere to them.
Screenshot testing with Percy
⚠️ Percy is disabled while we're figuring out https://github.com/vector-im/wat-internal/issues/36 and https://github.com/vector-im/wat-internal/issues/56. We're hoping to turn it back on or switch to an alternative in the future.
We also support visual testing via Percy. Within many of our
Cypress tests you can see lines calling cy.percySnapshot()
. This creates a
screenshot and uses Percy to check whether it has changed from the last time
this test was run.
It can help to pass percyCSS
in as the 2nd argument to percySnapshot
to hide
elements that vary (e.g. timestamps). See the existing code for examples of
this. (Note: it is also possible for team members to mark certain parts of a
screenshot to be ignored. This is done within the Percy UI.)
Percy screenshots are created using custom renderers based on Safari, Firefox,
Chrome and Edge. Each percySnapshot
actually creates 8 screenshots (4
browsers, 2 sizes). Since we have a limited budget for Percy screenshots, by
default we only run Percy once per day against the develop
branch, based on a
nightly build at approximately 04:00 UTC every day. (The schedule is defined in
element-web.yaml and the Percy tests are
enabled/disabled in cypress.yaml.)
If your pull request makes visual changes, you are encouraged to request Percy
to run by adding the label X-Needs-Percy
to the PR, these will only run in
the merge queue to save snapshots. This will help us find any
visual bugs or validate visual changes at the time they are made, instead of
having to figure it out later after the nightly build. If you don't have
permission to add a label, please ask your reviewer to do it. Note: it's best to
add this label when the change is nearly ready, because the screenshots will be
re-created every time you make a change to your PR.
Some UI elements render differently between test runs, such as BaseAvatar when
there is no avatar set, choosing a colour from the theme palette based on the
hash of the user/room's Matrix ID. To avoid this creating flaky tests we can use
the @media only percy
CSS query to override the variable colour into a fixed one
for tests where it is not feasible to fix the underlying identifiers issued by the
server. See https://docs.percy.io/docs/percy-specific-css#percy-css-media-query.