[:set [:enum …]] results in error in some case for reitit + malli request coercion

[piotr-yuxuan]

In some code I have a schema that is as close as possible to this one (edited for confidentiality):

(def MySchema
  [:map
   [:my-field {:default #{:created-at :id}}
    [:set {:min 1 :decode {:string #(if (string? %) (str/split % #",") %)}}
     [:enum {:decode {:string keyword
                      :json keyword}}
      :created-at :id]]]])

I use reitit 0.5.18 with malli coercion. When a request body contains a slightly invalid value, like:

{:my-field #{:credated-at}}

then with a debugger I can catch humanize being called twice. First with an explanation value that essentially boils down to:

{:value {:my-field #{:credated-at :id}}
 :errors '({:path [:my-field 0]
            :in [:my-field :credated-at]
            :schema [:enum :created-at :id]
            :value :credated-at})}

and the second time with an explanation value that is:

(def explanation
  {:value {:my-field ["id" "credated-at"]} ; Come from an http request, decoded with malli.
   :errors (list {:path [:my-field 0]
                  :in [:my-field :credated-at]
                  :schema [:enum :created-at :id]
                  :value :credated-at})})

This latter value looks like it comes from the json-decoded, but malli-undecoded value. The issue is that it somehow breaks humanize:

(malli.error/humanize explanation)
; => (IllegalArgumentException. "Key must be integer")

However, I'm very puzzled, because I can't reproduce the bug above:

(me/humanize
  (m/explain
    MySchema
    {:my-field #{:credated-at}}))
=> {:my-field #{["should be either :created-at or :id"]}}

I have observed that the issue disappear when I use a :vector instead of a `:set:

(associative? []) ;  => true
(associative? #{}) ; => false

(def MySchema
  [:map
   [:my-field {:default #{:created-at :id}}
    [:vector {:min 1 :decode {:string #(if (string? %) (str/split % #",") %)}}
     [:enum {:decode {:string keyword
                      :json keyword}}
      :created-at :id]]]])

I tried to suggest a fix in #774 but obviously more work, and a better understanding, are needed. I tried to make this bug report as accurate as possible but I know it doesn't pinpoint the root cause – I haven't been able to find it.

[piotr-yuxuan]

In my use case, passing down to malli.error/humanize the decoded value instead of the input value removes the exception. Not sure if it is a good way to fix it though.