I can’t tell what the general flow of the API usage in general is. Am I supposed to login/authorize somehow first? Some common examples, especially in at least one programming language (whether that’s curl or python or whatever) I think would go a long way to help people understand what they’re supposed to do.

This could use indeed some work

How do I know if I need to authorize for a particular endpoint?

openapi spec allows one to specify the authentication, this has not be done already because there is an open issue to include auth in lemmy-js-client (I use this to generate the spec) + Need to figure the best approach to convey “elevated” auth endpoints. Endpoints like GetSite where authentication enhances the response.

Endpoint descriptions are often unhelpful.

They are actually summaries, they are fine imo. There is an open issue to add the descriptions though.

We have to guess what endpoint we might need for a lot of things. Example: /post/like is also for dislikes, but it doesn’t tell you that. It also never tells you HOW to like or dislike anything, the valid values of score do not appear to be documented. And you’re left to assume that’s the right field to even use for it.

Yes indeed, things like this should be documented in the description.

What is the content type of the request supposed to be? JSON is never mentioned anywhere.

This is explicitly mentioned on every request, visually and in the spec.

What are these named “parameters”? Is that a query parameter? Why does it say “object” and “(query)”? Does this parameter go in the request body instead?

Query is mentioned when they are query parameters, else it just a request body. This is pretty clear in the text (spec) or visually imo.

/user shows a parameter called “GetPersonDetails” except in reality this name is (I guess) supposed to be completely ignored, because no part of the request actually uses the string “GetPersonDetails”.

GetPersonDetails is the name of the object if you scroll all the way to the bottom you can inspect it. This more clear in the spec.

Schema is missing for many endpoints, like the request part of /user.

There is not a single endpoint missing nor its requests.

What are all these fields under “GetPersonDetails”? Are they all required? Only some? It doesn’t say anything about it.

This could be improved.

Many of the possible error codes are undocumented.

Status codes? or the 400 lemmy error? ex: { error: "report_reason_required" } Status codes are the same everywhere, 200/201 or 400 with Lemmy error as response. There is one exception that is 401 that can be thrown for every auth required endpoint where auth fails. But these are standard. Ig this should documented.

Now the “LemmyErrorTypes”. This could be improved, but it is hard to, not possible to be automated and tedious to add and frequently changes.

Thanks for the feedback.

Ah yes, all though I ve never really experienced HTTP days in their full glory

I’ll look into using it. If it has a github action or easy way to publish it from cli

I do not see how critisms for the JS API docs are relevant for my openapi documentation.

This documentation aims to solve all those problems in a language agnostic way. It descibes the endpoints, the request object, the status, the response object, the authentication needed in visual/text. It allows you test it right from the browser, allows you to copy a working curl command, search for endpoints based on keywords, allows you to import the entire spec into postman/alts.

I ve never had any problems with CF instances but I mostly test with voyager.lemmy.ml

Looks like the best openapi front. I looked into using it, but it didn’t seem free. Too bad this one is outdated. Lemmys API has changed quiet a bit since then.

(Author btw)

Can you expand on this? This was written for Kotlin usage. It automates generating API based on this spec.

I have no idea what you mean with ‘using other approaches to calling the endpoints’

Lemmy has only one API. It’s same API used for lemmy-ui.

This is just a “better” documentation for it.

The API works for for cloudfare instances too.

There is also a swagger ui variant

https://mv-gh.github.io/lemmy_openapi_spec/swagger_ui.html

Yes but now it must wait for acknowledgement of a request (activity), before sending the next one. If one request takes 333ms means you can do max 3 requests per second. Now big instances like lemmy.world have activity above that so instances too far will perpetual lag behind

Mostly because Lemmy has a hard time keeping up with the federation

https://github.com/LemmyNet/lemmy/issues/4529