added more info on function schema explain:

(def MyFunctionSchema
  (m/schema
    [:function
     [:=> [:cat :int] :int]
     [:=> [:cat :int :int [:* :int]] :int]]
    {::m/function-checker mg/function-checker}))

(m/explain
  MyFunctionSchema
  (fn
    ([x] x)
    ([x y & zs] (apply max x y zs))))
; => nil

(m/explain
  MyFunctionSchema
  (fn
    ([x] x)
    ([x y & zs] (str (apply max x y zs)))))
;{:schema [:function 
;          [:=> [:cat :int] :int] 
;          [:=> [:cat :int :int [:* :int]] :int]],
; :value #object[],
; :errors (#Error{:path [],
;                 :in [],
;                 :schema [:function 
;                          [:=> [:cat :int] :int] 
;                          [:=> [:cat :int :int [:* :int]] :int]],
;                 :value #object[],
;                 :check ({:total-nodes-visited 2,
;                          :depth 1,
;                          :pass? false,
;                          :result false,
;                          :result-data nil,
;                          :time-shrinking-ms 0,
;                          :smallest [(0 0)],
;                          :malli.generator/explain-output {:schema :int,
;                                                           :value "0",
;                                                           :errors (#Error{:path []
;                                                                           :in []
;                                                                           :schema :int
;                                                                           :value "0"})}})})}

really happy how good the function schemas are now, but still - it’s just elegant (and slow) way to verify things at runtime, e.g. not a type system.

in master

compared to clojure.spec, how slow?

the sequence schemas are at least an order of magnitude faster than spec in initial tests. Full function verification uses test.check, I think that dominates, and quess they are about the same.

but, having arity info available for fns at runtime (in clojure.core) would help to see that fns are correct on arity, without testing it and getting arityexceptions - costs a lot.

Maybe you can combine that with static analysis (e.g. from clj-kondo)?

I guess it could be done using a clj-kondo hook? with defaults, the test.check runs 19ms for a function on my machine. that’s 500 funcs per sec (on 1 core?). errors are faster, the given example fails fast on invalid return, under 1ms.

i believe it could be made faster, less iterations on arguments etc.

[metosin/malli "0.3.0-SNAPSHOT"] has the stuff in.

with 10 iterations, it drops to 0.3ms per fn.'

> e.g. not a type system.

i mean

for me, it would serve much the same purpose i'd think

or rather - the same workflow

maybe i'm missing a worldview though

Btw, this sounds interesting, but I haven't heard anything about it yet: https://www.fulcrologic.com/copilot