At this point, we’ve defined the error model and know how to use it. Until now, we focused on custom validations in the project. However, what if the errors originate in the libraries you might be using for some validation? Open-source libraries cannot possibly know about your own error model. They have their way of representing and returning errors. We hinted at the solution already - convert library errors into your own model. Let’s do that here.

I’ll use malli as an example, but the same idea can be applied to any other library you might be using.


For clarity, let’s reiterate what we want to achieve.

  • We want to have one error representation - our own error model. This makes dealing with errors simple.
  • We want errors to collect as much information as possible to make debugging easier. Remember that the error model is an in-memory model. It doesn’t necessarily mean that everything you collect is going to be used as an HTTP response.
  • We want to carry that information to the code that can decide what it’s going to do with it. Perhaps it’ll use a subset of that information to respond to an HTTP request, log the error, collect metrics, write to DB, etc.

How to do that for an external library? By wrapping. If some function returns that library’s error representation then convert it on the fly to our own error model.

For completeness, let’s just state the obvious. Wrapping every function in that library is not the goal. The library is not the problem. “Non-standard” error representation is. For example, malli.core/validate is not a good candidate for wrapping. It’s a reasonable function to call when you only want a yes or no answer, but the boolean result is not something that can be usefully converted to own error model. Keep those calls in your project as they are.

(malli/validate [:string] "1")
=> true
(malli/validate [:string] 1)
=> false

The perfect candidate is malli.core/explain.

(malli/explain [:map
                [:user [:map {:closed true}
                        [:first-name :string]
                        [:last-name :string]]]]
               {:user {:first-name "John"
                       :last-name "Doe"}})
=> nil

(malli/explain [:map
                [:user [:map {:closed true}
                        [:first-name :string]
                        [:last-name :string]]]]
               {:user {:name "John Doe"}})
{:schema [:map [:user [:map {:closed true} [:first-name :string] [:last-name :string]]]]
 :value {:user {:name "John Doe"}}
 :errors '({:path [:user :first-name]
            :in [:user :first-name]
            :schema [:map {:closed true} [:first-name :string] [:last-name :string]]
            :value nil
            :type :malli.core/missing-key}
           {:path [:user :last-name]
            :in [:user :last-name]
            :schema [:map {:closed true} [:first-name :string] [:last-name :string]]
            :value nil
            :type :malli.core/missing-key}
           {:path [:user :name]
            :in [:user :name]
            :schema [:map {:closed true} [:first-name :string] [:last-name :string]]
            :value "John Doe"
            :type :malli.core/extra-key})}

The hash-map that malli.core/explain returns is what we would like to convert to our own error model.

That seems simple enough. It’s just a transformation from one data type into another. We just need to keep in mind that malli comes with a lot of built-in schemas. Here is a snippet of what that transformation would look like.



(def ^:private malli-code->error-code
  {'nil? :malli/null
   'some? :malli/some
   'boolean? :malli/boolean
   'true? :malli/true
   'false? :malli/false
   'number? :malli/number
   ::malli/missing-key :malli/required
   ::malli/extra-key :malli/extra-key

(defn- malli->error [root-schema root-value malli-error]
   (merge {::root-schema root-schema
           ::root-value root-value
           ::schema error-schema
           ::value error-value
           ::path (:path malli-error)
           ::in (:in malli-error)}
          (if-let [type (:type malli-error)]
            {:code (get malli-code->error-code type)}
            {:code (get malli-code->error-code schema-type)})

(defn validate [schema value]
  (when-let [{:keys [schema value errors]} (malli/explain schema value)]
    (->> errors
         (map #(malli->error schema value %))

Given that we’re referring to the error message via :code, we also need the error templates.


{:malli/required "%s is required"
 :malli/extra-key "System doesn't recognize property %s"
 :malli/null "%s must be empty"
 :malli/some "%s must not be null"
 :malli/boolean "%s must be a boolean"
 :malli/true "%s must have a value of `true`"
 :malli/false "%s must have a value of `false`"
 :malli/number "%s must be a number"

That is the core idea. It’s relatively simple. To complete the above we’d need to take a look at the malli’s implementation because the documentation doesn’t describe everything we need to know to complete it.

Or you can just refer to the full implementation below.

Malli validator

The full implementation with tests and error templates is here:

Notes regarding the implementation:

  • The implementation is compatible with malli 0.9.2. Latest at the time of this writing.
  • malli-code->error-code might be sensitive to malli’s internal implementation (or breaking changes). E.g. new malli version might add or remove a schema. To guard against that, just call malli-validator/check-compatibility-with-malli! at the start of your integration tests (or system start). It’s used to check if malli-validator is still in sync after you’ve updated malli.
  • The implementation includes the automatic generation of args, human (developer) breadcrumbs, and error templates with certain wording. It’s geared towards both HTTP service response as well as responding to a web application with the intent of displaying it to the customer (mostly as a global toast element vs. below the form field). Feel free to modify malli-validator and templates to suit your need.
  • Refer to clojure.core extensions for some of the functions used in the implementation.

Alternative implementation

You might be using malli.error/humanize in your project.

 (malli/explain [:map
                 [:user [:map {:closed true}
                         [:first-name :string]
                         [:last-name :string]]]]
                {:user {:name "John Doe"}}))
{:user {:first-name ["missing required key"]
        :last-name ["missing required key"]
        :name ["disallowed key"]}}

Can that be used instead of malli.core/explain to generate own error model? Yes, it can! The only problem is that you’d be throwing a lot of information. For example, malli.core/explain gives you very nested schemas and values that failed, not just the root schema and value. Besides that, you’d be losing the programmatic access to why an int or string failed the validation: is it due to min/max constraints, is it null, type mismatch, etc? With humanize that information is embedded inside of the error message itself, while with explain it can be embedded both in the error message and as a specific :code (e.g. :malli/required, :malli/string, :malli/string-with-length-between, :malli/string-with-length-max, :malli/string-with-length-min). However, if you don’t care about that information (not even while debugging?) then you can use the result from humanize. Just be aware that the result is not a simple “property -> vector-of-messages” hash-map, but can be nested as indicated in the example above.

You could even combine the results of malli.core/explain and malli.error/humanize to collect all of the information from explain, but also reuse the error messages from humanize. This could get rid of duplicated templates between error-template.edn and malli.error/default-errors. Feel free to consider if this is worthwhile for your project.

Update 2023-02-10: Also consider using malli.error/with-error-messages instead of malli.error/humanize. with-error-messages adds :message to all of the errors returned by malli.core/explain.

What to do with the errors?

(let [errors (malli-validator/validate schema value)]

Now that we have the errors, what do we do with them? We are in the same situation as before. Please see the previous suggestion.