Clojurians Log v2

what exactly are you hoping to accomplish with a supposed generic?

Well for example, I'd like to write some code like this:

(m/=> my-map-fn
        (m/all [a b]
               [:-> [:-> a b] [:sequential a] [:sequential b]]))

(m/=> my-map-fn
        (m/all [a b]
               [:-> [:-> a b] [:sequential a] [:sequential b]]))

  (defn my-map-fn [f xs] (map f xs))

  (def thing-schema
    [:map
     [:tag :string]
     [:val :int]])

  (m/=> do-stuff
        [:-> [:sequential thing-schema] [:sequential :string]])

  (defn do-stuff [things]
    (my-map-fn
     (fn [thing]
       (:tag thing) ;; here clj-kondo would know that this complies with thing-schema
       )
     things))

  (do-stuff [{:tag "label"
              :val 0}])

oh interesting. i odn't have a lot of experience with clj-kondo's type system, but i believe it's much simpler than that

yup yup, at the moment m/all does not exist yet in malli, but it's proposed in the PR https://github.com/metosin/malli/pull/1053/files. Though, I'm not sure how it would integrate with clj-kondo yet. From what I've noticed, malli will emit a file into the .clj-kondo directory (relative to project root), and from there clj-kondo will lint based on whatever type mismatches are reported. So hopefully that means malli could report a mismatch after evaluating the generics

right, i'm talking about clj-kondo. i've reviewed ambrose's PR, i think it's cool. but if you're looking for clj-kondo to throw errors, i don't know that it can handle that. i believe it works purely with concrete types (x is a seq or a long)

yeah i agree that clj-kondo wouldn't support generic types (at least atm). But i was thinking it wouldn't need to support them if malli emits the concrete type as the mismatch, So the generics are resolved before clj-kondo attempts to lint

hmmmmm compelling idea

might run into issues if the generic is used in multiple places with different types, so the concrete type needs to change per invocation

oh yeah good point, hmm not sure how to overcome that atm

we'd almost need some way of emitting the resolved schema and linking it back to the invocation, but i don't see anything that can do that in the clj-kondo data

yeah, at that point we're into typed-clojure land

well i suppose the next simplest thing would be to teach clj-kondo (or a similar tool) how to evaluate the type mismatches with the generic arguments. so at least then we don't need to worry about resolving+linking the instantiated schemas ahead-of-time

@hagstromsg I need to gather my notes and I'll get back to you.

awesome, looking forward to it, thank you 🙏