- We're in the process of rolling out a new e2e-framework, and getting rid of the regression tests
- We'll eventually merge all the tests together and into using the same framework
refs https://github.com/TryGhost/Team/issues/1353
- Doing a renaming-only commit to avoid merge conflicts while I work on the new members.test.js test file
- The new `members.test.js` file will set the `multipleProducts` flag to true
no-issue
Because we were returning the call to `boot`, rather than awaiting it,
it meant that once a test was provided with an API Agent the Ghost
application hadn't necessarily started. This was noticed whilst working
on the Members API which requires the frontend to be loaded.
When the frontend is loaded we must also wait for the url service to
have finished initialising, so that we do not run into 503 maintenance
errors when hitting the frontend.
no-issue
The Members API is served on a different endpoint to the Admin API, and
also requires that the frontend is booted. This agent is to be used when
testing the Members API, e.g. Stripe webhooks or Members config.
- The admin API test agent doesn't boot the Ghost frontend, which is where members routes are currently located
- This means we can't test members properly as we don't have access to webhooks etc
- This change adds a members option to getAdminAPIAgent(), which in turn enables the frontend
- We do it this way so that we can easily change the implementation later, e.g. if we have the option to boot members directly
no-issue
The way we choose which email to send is too tightly coupled with the
magic link generation, which meant that when we added the new paid
signup email, the magic link looked like <url>?action=signup-paid rather
than <url>?action=signup - our welcome page logic was working off of
only using welcome pages for action=signup and so this broke.
- When starting ghost for e2e tests we create a content folder in the os tmp dir
- This means that the folder can change between suites as ghost is started and restarted
- For the most part this is fine, but theme storage caches the path to config (which makes sense, it's not meant to change whilst Ghost is in-memory)
- This is a quick-n-dirty fix that just makes it possible to update that path in the tests, so we know it's in sync
- Ideally we'd not cache the path, use a function to fetch it etc, or fully reset the theme storage layer, but this is the fix I have working today
and so it's going in to unblock things for now
refs https://github.com/TryGhost/Team/issues/1071
We switched to using tiers pivot table that stores list of tiers with access to post when visibility is set to `tiers`. For themes using v3 API while having posts restricted to specific tiers visibility, the post data will not include the list of tiers, which caused an unexpected error while trying to determine post access from tiers list. This change blocks access to post if specific tiers visibility is enabled without data available for list of tiers on post.
- This part of the framework can be used in isolation
- Using mockManager everywhere makes it more visible how to use it
- Aside: fixed .getAdminAPIAgent not needing a URL in site tests whilst cleaning up
- Can now easily mock labs flags as enabled or disabled using mockManager
- Updated some bits of code that directly mocked labs
- Aside: improved error thrown when things go wrong booting Ghost
refs daeb06e835
- This is an additive change, and the parameter is optional so it's better to ship it right away (testing was hard as it's a chicken and an egg problem to have a feature flag in the setup route).
refs: https://github.com/TryGhost/Toolbox/issues/158
- We only use chai in a handful of places now, and it seems totally unnecessary
- Use assert instead
- Made other minor changes with a view to this being a reference aka "perfect" test suite
refs: https://github.com/TryGhost/Toolbox/issues/158
- if we had to combine a large list of params then this would make sense
- we could/should also add a .query() method to our agent
- however, I don't think this is worthwhile/necessary for just 2 params right now :)
no-issue
If Stripe subscriptions have manually been modified to contain more than
one price, then cancellation would fail, as Ghost/Members relies on
subscriptions only containing one price. This adds some handling so that
we do not error when cancelling.
refs: https://github.com/TryGhost/Toolbox/issues/158
- continuing to turn this into a reference test suite for how to do things with the new framework
- use fixtureManager and mockManager properly
- testUtils should _not_ be required anymore!
- TODO: cleanup labs mocking and side effect assertion using models
refs https://github.com/TryGhost/Team/issues/1321
- when updating the Email record after submitting all email batches we have a `catch` call but it was only logging the error
- added a call to Sentry so there's more visibility if those saves fail
refs: https://github.com/TryGhost/Toolbox/issues/158
- rather than just exposing any, anything and string matching, expose more specific matchers.
- this was triggered by `any(Date)` not working for dates in our API
- it seems poor to match `any(String)` for something we want to be a well formatted date
- establishes the pattern of using our defined matchers instead of requiring any/anything from jest
refs: https://github.com/TryGhost/Toolbox/issues/158
- Working up to having these be the defacto "reference" tests
- Changing this allows for the fact that there will be a getContentAPIAgent, and probably more in future
- Abstracting away the path to a single location will make updating API paths easier later too
refs TryGhost/Toolbox#158
refs TryGhost/Toolbox#129 (comment)
We now have @tryghost/jest-snapshot - a suite of utilities for using snapshots with mocha, should, etc
The @tryghost/express-test lib also uses it to provide chained snapshot assertions
This library was created because all the existing implementations I could find, including the in use @ethanresnick/chai-jest-snapshot
didn't properly support property matching. @ethanresnick/chai-jest-snapshot supposedly supported it, but the implementation was incorrect
and frequently lead to false test passes.
This library also has (after some back and forth) path resolution so that snapshots can live in a local __snapshots__ folder
refs https://github.com/TryGhost/Toolbox/issues/202
- this code suffers from two problems:
- when we don't have any new settings to insert, we still end up
fetching the columnInfo and owner info, even though we only need
them if we're inserting data. This results in 3 extra queries upon
boot
- secondly, we insert every setting with a separate query - MySQL and
SQLite both support batch inserts and knex has a utility to help us
that I've [used
before](38821c5242).
With 95 settings at the time of writing, this adds 94 extra queries
during the DB init
- this commit refactors the code so that we only fetch the columnInfo and
owner data if we've got new settings to insert, and batches the
inserts using knex's batchInsert util
- this query results in ~95 less queries during DB init and saves a
couple of queries during boot
refs a3cc66be50
- in the referenced commit, I made a util to speed up resetting the DB
for SQLite by copying the database file
- I inadvertently removed an optimization we had before - where we
truncate the tables and insert the fixtures instead of dropping the
entire database
- this would be missing on MySQL tests
- this seems to have a big difference so this commit re-adds the
optimization in
refs https://github.com/TryGhost/Team/issues/1316
Updates header card renderer so the `<h2>` and `<h3>` elements have slugified titles set on their `id` attributes to allow in-page and external #header-title anchor links.
- bumps `@tryghost/kg-default-cards` including the header card id fix
- bumps other `@tryghost/kg-*` packages that were released simultaneously - contains dependency updates and switch to `slugify()` from `@tryghost/kg-utils`
The idea here is to keep the concept of _what_ we're asserting (that an email was sent)
away from the implementation of how we check that assertion
This means that tests can have a consistent api like .sentEmail(), and if we change how that function works,
we only have to update the assertion function in the MockManager
Much more cleanup to come behind the scenes, but the aim is to make the tests as clean as possible
- encapsulated concerns within individual objects
- this will allow us to refactor these into classes or move them around later
- also makes it clearer how methods like restore relate to other methods
- e.g mocks.restore() restores mocks, whilst fixtureManager.restore() restores the database
- Ensure that the DB is always reset between tests
- We assume that the DB will be torn down and rebuilt between tests
- Without this, previous tests clobber the current tests, and it's not possible to tell when running files in isolation
which makes developing really tricky
loses TryGhost#13971
This fixes an issue with links containing # anchor. It makes sure the # part is at the end of the url like url?v=hash#anhor
Co-authored-by: Hannah Wolfe <github.erisds@gmail.com>
no issue
- the new eslint bump seems to flag this line up for using camelcase
- i'm not sure why this hasn't flagged before but it seems we want to
keep the naming because the function name has the same linting exemption
