Decode can may fail to unform valid specs

[brokebroke]

I have a case where I want to use select-spec to narrow a map to a subset described by a multi-spec. The multi-spec is dispatched on an item in the larger collection but would be stripped from the final map. This worked fine in previous releases where select-spec was based on conform (and as a workaround I'm passing the strip-extra-keys-transformer to conform, but I suspect there's an issue with decode

[ikitommi]

Thanks for the report. Could you provide an example?

[brokebroke]

Sure, little contrived but my case boils down to something like this:

(spec/def :test/type #{:a})
(spec/def :test/a number?)

(defmulti blob :type)
(defmethod blob :a [_]
  (st/spec (spec/keys :req-un [:test/a])))

(spec/def :test/blob
  (st/spec (spec/multi-spec blob :type)))


;; OK, returns {:a 1}
(st/conform :test/blob {:type :a :a 1} st/strip-extra-keys-transformer)

;; ERROR - CompilerException java.lang.IllegalStateException: No method of: rdp.api.core/blob for dispatch value: ,
(st/select-spec :test/blob {:type :a :a 1})

[ikitommi]

select-spec does both s/confom and s/unform to effectively return the original value with the extra . The latter fails as the :type key has been removed by the st/strip-extra-keys-transformer. The spec error is not very helpful.

If you change the multi-spec into:

(defmethod blob :a [_]
  (st/spec (spec/keys :req-un [:test/a :test/type])))

it works. Without unform, the transformation is only partial:

(spec/def :test/a (spec/or :int int? :string string?))

;; not good:
(st/conform :test/blob {:type :a :a 1} st/strip-extra-keys-transformer)
; {:type :a, :a [:int 1]}

;; good
(st/select-spec :test/blob {:type :a :a 1})
; {:type :a, :a 1}

hope this helps.

[brokebroke]

I guess my issue is more that select-spec shouldn't obviously require the type to be contained in the spec to work, this requirement is introduced by the implementation in terms of conform and inform.

…

On Mon., 25 Jun. 2018, 05:00 Tommi Reiman, ***@***.***> wrote: select-spec does both s/confom and s/unform to effectively return the original value with the extra . The latter fails as the :type key has been removed by the st/strip-extra-keys-transformer. The spec error is not very helpful. If you change the multi-spec into: (defmethod blob :a [_] (st/spec (spec/keys :req-un [:test/a :test/type]))) it works. Without unform, the transformation is only partial: (spec/def :test/a (spec/or :int int? :string string?)) ;; not good: (st/conform :test/blob {:type :a :a 1} st/strip-extra-keys-transformer); {:type :a, :a [:int 1]} ;; good (st/select-spec :test/blob {:type :a :a 1}); {:type :a, :a 1} hope this helps. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#119 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AlvrjR900JzKxduqUPLa2F3kUO6Eq3mQks5t_-G6gaJpZM4ULivq> .

[ikitommi]

I understand, but not sure is there anything we can do for this. https://dev.clojure.org/jira/browse/CLJ-2251 would help as we could just say "coerce" instead of conform + unform. Alex just said that some deep-walking might apper to clojure.spec later-later.

While waiting - if there is a way to ask from a multi-spec what spec it will dispatch for this given data, we could unform against that spec directly, not for the multispec itself. Don't have time now to try that, but would be a welcome addition to the lib.