adventures in making stuff with Daniel Higginbotham

Vern, a Library For Database Fixtures

04 December 2014

At work, I've been using yesql and wanted a way to compactly describe some seed data. My first attempt was an utter abomination, but I think I've worked out something pleasant that could be useful to other people. vern is a 54-line project that I'd love to get feedback on before publishing. It's named after one of my cats, this guy:



Here's an example of vern in use:

(def fixtures [:endpoints
                {:name "wallace" :id 1}

                {:name "gromit"  :id 2}]

               [;; these permissions belong to wallace
                [{:endpoint [:endpoints :wallace]}

                 {:name "read"  :id 10}

                 {:name "write" :id 11}]

                ;; these permissions belong to gromit
                [{:endpoint [:endpoints :gromit]}

                 {:name "read"  :id 12}

                 {:name "write" :id 13}]]])

(def entities (atom []))

(do-named (fn [processed group-name entity]
              (swap! entities #(conj % (:data entity)))
              (get-in entity [:data :id]))

; =>
[{:id 1 :name "wallace"}
 {:id 2 :name "gromit"}
 {:id 10 :endpoint 1 :name "read"}
 {:id 11 :endpoint 1 :name "write"}
 {:id 12 :endpoint 2 :name "read"}
 {:id 13 :endpoint 2 :name "write"}]

First, we assign the fixture the colorful name data, then create a "database", which in this case is just an atom holding an empty vector.

do-named is where the real action happens. processed is a map of all the named entities that have been processed so far. For example, after processing the first two entities, the value of processed will be

{:endpoints {:wallace 1
             :gromit 2}}

Most likely you won't need it but it's there in case you do. group-name will be :endpoints and then :permissions in this example. In real life, I use this to determine which table I need to insert the entity's data into. entity will be something like

{:name :wallace
 :data {:name "wallace" :id 1}}

In the example, you're just using conjing (:data entity) onto the entities atom's vector. Here's what I do in real life, though:

(defn load-fixtures
  [db fixtures]
  (d/do-named (fn [processed group-name entity]
                (:generated_key (q/insert! group-name db (:data entity))))

This uses yesql's functionality to insert a row in a database and return the value of the key that was generated for every entity in the fixture. The reason you want to return the generated key is so that you can refer to this record in later entities. If a subsequent entity is defined as

{:name "belongs-to-wallace"
 :endpoint-id [:endpoints :wallace]}

then do-named uses the corresponding value in the processed map. When do-named sends the first "permission" entity to its function, the entity looks like this:

{:name :wallace-read
 :data {:name "read" :id 10 :endpoint 1}}

The last feature is that you can group common attributes together in a vector. You can see this in the permissions:

[[{:endpoint [:endpoints :wallace]}

  {:name "read"  :id 10}

  {:name "write" :id 11}]]

You could have written these entities as follows:

 {:name "read"  :id 10
  :endpoint [:endpoints :wallace]}

 {:name "write" :id 11
  :endpoint [:endpoints :wallace]}]

But that kind of repetition can pretty old pretty quickly.

Here's the general form of the fixtures:

 [:entity-1-key ;; entity keys are optional
  {:name "zip"}

  ;; using a sequential structure allows you to reference
  ;; another entity
  {:parent-id [:entity-group-key :entity-1-key]
   :name "zip's child 1"}

  ;; you can define common associations by grouping entities in a
  ;; sequential; the first element contains the common associations
  [{:parent-id [:entity-group-key :entity-1-key]}
   {:name "zip's child 2"}
   {:name "zip's child 3"}]]

 [{:entity-group-1-id [:entity-group-key :entity-1-key]}]]

Request for Feedback

This is a tiny library, and I hope it might yield a couple fun rounds of code golf. In particular, I have the sneaking suspicion that I'm neglecting some useful functions from Clojure's standard library. I considered using zip but that seemed too heavyweight. Also, I think I could just pass the entity's data to the do-named function.

I'm also curious if there's a name for this kind of pattern, where you aren't solely processing one sequence item at a time but care about what you last processed as well (this is how names get assigned to entities).