## About the changes
We found a problem generating the Go SDK client for the tokens API that
makes use of `oneOf`, combined with `allOf`. The generator doesn't know
how to map this type and leaves it as an object. This PR simplifies the spec and therefore the code generated after it
This PR updates endpoints and schemas for the API tokens tag.
As part of that, they also handle oneOf openapi validation errors and improve the console output for the enforcer tests.
<!-- Thanks for creating a PR! To make it easier for reviewers and
everyone else to understand what your changes relate to, please add some
relevant content to the headings below. Feel free to ignore or delete
sections that you don't think are relevant. Thank you! ❤️ -->
## About the changes
<!-- Describe the changes introduced. What are they and why are they
being introduced? Feel free to also add screenshots or steps to view the
changes if they're visual. -->
This deprecates the `username` properties on api-token schemas, and adds
a `tokenName` property.
DB field `username` has been renamed to `token_name`, migration added
for the rename.
Both `username` and `tokenName` can be used when consuming the service,
but only one of them.
## Discussion points
<!-- Anything about the PR you'd like to discuss before it gets merged?
Got any questions or doubts? -->
There's a couple of things I'd like to get opinions on and discuss:
- Frontend still uses the deprecated `username` property
- ApiTokenSchema is used both for input and output of `Create`
controller endpoints and should be split out into separate schemas. I'll
set up a task for this
---------
Co-authored-by: Thomas Heartman <thomas@getunleash.ai>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
## What
This change reverts the spec change to create-api-token-schema that
was introduced in 97c2b3c089
That commit changed the endpoint from allowing any string to only
allowing strings that match the values of the ApiTokenType value.
## Why
Because we already validate this on the backend, I didn't think it
would break anything. However, because OpenAPI enums are case
sensitive (and do not currently support anything else), it _did_ break
the API. ([stack overflow post that explains how this
works](https://stackoverflow.com/questions/60772786/case-insensitive-string-parameter-in-schema-of-openapi))
## Of note
This should have been caught by the tests, but it didn't show up until
it was merged to the main branch. I think it might be worth looking
into how we do the PR testing.
## What
This PR adds an extra layer of OpenAPI validation testing to what we already have. It also fixes any issues that make the new tests fail.
## Why
While the current OpenAPI validation takes care of _some_ things, there's also things it misses, as shown by #2055. By adding the OpenAPI Enforcer package, we should hopefully be able to catch more of these errors in the future. The enforcer _does_ flag the issue in #2055 as an error.
## How
By adding the OpenAPI Enforcer package and making whatever changes it picks up on.
By adding location headers to all our 201 endpoints. I also had to change some signatures on `create` store methods so that they actually return something (a lot of them just returned `void`).
## Discussion points
### Code changes
Some of the code changes may not be necessary or we may want to change more code to align with what changes have been done. It may be worth standardizing on a pattern for `*store.create` methods, so that they always return an identifier or the stored object, for instance.
### On relative URIs
The 201 location headers use relative URIs to point to the created resources. This seemed to be the easiest way to me, as we don't need to worry about figuring out what the absolute URL of the instance is (though we could probably just concat it to the request URI?). The algorithm for determining relative URIs is described in [RFC 3986 section 5](https://www.rfc-editor.org/rfc/rfc3986#section-5).
There's also some places where I'm not sure we _can_ provide accurate location url. I think they're supposed to point _directly at_ whatever the resource is, but for some resources (such as api tokens), you can only get it as part of a collection. From [RFC 9110 on the location field](https://httpwg.org/specs/rfc9110.html#field.location) (emphasis mine):
> the Location header field in a [201 (Created)](https://httpwg.org/specs/rfc9110.html#status.201) response is supposed to provide a URI that is **specific** to the created resource.
A link to a collection is not specific. I'm not sure what best to do about this.
### Inline comments
I've added a number of inline PR comments that I'd love to get some feedback on too. Have a look and let me know what you think!
### Unfinished business
I've added some juicy comments to some of the files here. They contain non-blocking issues that I'm tracking (via github issues). We should resolve them in the future, but right now they can stay as they are.
## Commits
* Feat: add openapi-enforcer + tests; fix _some_ issues
* Test: allow non-standard string formats
* validation: fix _some_ 201 created location header endpoints
* #1391: fix remaining 201 location headers missing
* Refactor: use the ajv options object instead of add* methods
* #1391: flag validation errors and warnings as test failures
* #1391: modify patch schema to specify either object or array
We don't provide many patch endpoints, so we _could_ create separate
patch operation objects for each one. I think that makes sense to do
as part of the larger cleanup. For now, I think it's worth to simply
turn it into one of these. While it's not entirely accurate, it's
better than what we had before.
* Refactor: make tests easier to read
* #1391: use enum for valid token types
This was previously only a description. This may seem like a breaking
change because OpenAPI would previously accept any string. However,
Joi also performs validation on this, so invalid values wouldn't work
previously either.
* #1391: Comment out default parameter values for now
This isn't the _right_ way, but it's the pragmatic solution. It's not
a big deal and this works as a stopgap solution.
* #1391: add todo note for api token schema fixes
* #1391: update snapshot
* Revert "#1391: modify patch schema to specify either object or array"
This reverts commit 0dd5d0faa1.
Turns out we need to allow much more than just objects and arrays.
I'll leave this as is for now.
* Chore(#1391): update comment explaining api token schema TODO
* #1391: modify some test code and add comment
* #1391: update tests and spec to allow 'any' type in schema
* Chore: remove comment
* #1391: add tests for context field stores
* #1391: add location header for public signup links
* #1391: fix query parameter description.
There was indeed a bug in the package, and this has been addressed
now, so we can go back to describing the params properly.
