How do I handle cases where error doesn't match the expected format?

[haggs]

Hi, I couldn't figure out what to do from reading the docs so here I am.

First off, I love using this tool. I'm using it with TanStack Query and it's going very well so far.

However, sometimes, due to unknown circumstances, occasionally I get a response that doesn't adhere to my OpenAPI spec, and a client.GET call will throw.

Is there something I can configure to have a default response? I noticed default response mentioned in the docs but couldn't figure out where to set that up.

[drwpow]

default response is usually for errors, and you specify it rather than a status code in your OpenAPI schema, e.g.:

responses:
  200:
    # …
  default:
    # …

But openapi-fetch will always prefer status codes if they are specified, and will only use default for the error type if another error code (e.g. 500) isn’t given. So even if you added a default type, it would still probably give you the types that your OpenAPI schema is telling you it’ll get (which should never change).

I can’t say for sure, but it sounds like your endpoint is throwing errors while returning “OK” status codes, which openapi-fetch can’t really account for since it’s all static type analysis with zero runtime.

[haggs]

Got it. I'm actually already using that, but my default definition has a JSON response body, which isn't always true. I guess most people have a more generic default response than that.

[haggs]

It's not that my endpoint is returning 200, it's that it's not returning a JSON body that i've defined in default in very specific cases, i.e. if the backend is completely unreachable.

[haggs]

I think the answer here is to have my OpenAPI definition account for failures where you don't get a response body, so my openapi-fetch client won't throw.