Should requestCrawl be a "procedure" (POST) rather than a "query" (GET)?

[DavidBuchanan314]

HTTP GET requests are supposed to be idempotent, but I'm not sure the behavior of requestCrawl strictly matches that.

[bnewbold]

I think you are probably correct, and this should be a "procedure" not "query" endpoint, mapping to an HTTP POST request.

Curious what @dholms thinks.

[dholms]

Yup you're right 👍

Tbh not sure what was going through my head when I sketched out that schema a few months back

[whyrusleeping]

requestCrawl is idempotent though, multiple calls with the same params should result in a no-op

[snarfed]

...but it has side effects, which ideally GET shouldn't, right? Also it triggers a relatively expensive/heavyweight operation, so POST might be more prudent in practical terms, regardless of which method's theoretical/standards semantics it matches best?

[whyrusleeping]

Yeah, it should be a post regardless, agreed

[sugyan]

One suggestion related to this discussion, how about removing optional() from output in query?
https://github.com/bluesky-social/atproto/blob/main/packages/lexicon/src/types.ts#L231
There should be no method that has neither side-effects nor return value.

I thought this would prevent the confusion between query and procedure definitions, as in this case.