RDT - REPL-Driven tests

RDT is a REPL-Driven Test library for Clojure. It enables you to write tests that look like REPL sessions. It makes use of pattern matching and other functionalities to make the tests more readable.

RDT is a library designed to create tests that are like REPL sessions, low structure, easy to debug when they fail because they are just REPL sessions. This library favours tests readability over design purity.

RDT is piggybacking on top of the great work of Midje adding a couple of features for readability. I don't know whether it will always depend on Midje, or whether in the future it might just be its own thing, but for the moment it does internally use it, so the tooling compatible with Midje will (generally) be compatible with RDT as well.

WORK IN PROGRESS: Some behaviours might change.

Motivation

I strongly believe that tests should be "obvious" to the reader. I've seen tests that were more complex than the code they were testing. I believe that libraries such as Midje make a good attempt to make tests more readable to the developer. I think Midje is largely underappreciated in our community.

Most of the testing libraries have a strong focus on unit-tests, which in a functional language like Clojure tend to be pure functions which are stateless.

So the general pattern is something like:

;; midje tests example
(fact "some statement"
  (function arg1 arg2 arg3) => expected-result)

However, when tests are stateful, such as most of the integration tests, the test starts to be more and more complex.

The following example is a test which verifies the correct functionality of a queueing system.

;; midje tests example
(fact "items sent to a queue can be retrieved from the same"

  (let [cfg    (start-test-cluster)
        q-name (str (java.util.UUID/randomUUID))
        queue  (create-queue cfg q-name)]

    (pop-items cfg queue) => empty?

    (let [item {:some :message :random (rand)}
          msg  (send cfg queue item)]

      msg => (contains {:message-id string?})

      (let [q-item (first (pop-items cfg queue))]

        q-item => (contains {:message-id (:message-id msg)
                             :item item})

        (delete-item cfg queue q-item) => (contains {:status :deleted})))))

This test is a simplified version of a real-world test, however it is already explicative of the problem faced.

Assume that that this test is failing, in order to debug you will need to start the REPL and evaluate each expression in order. However, because of the nested let expressions, you will need to unnest it first, which is basically rewriting the whole test in the repl.

Let's see the same test written in RDT:

;; with RDT
(repl-test "items sent to a queue can be retrieved from the same"

  (def cfg    (start-test-cluster))
  (def q-name (str (java.util.UUID/randomUUID)))
  (def queue  (create-queue cfg q-name))

  (pop-items cfg queue) => empty?

  (def item {:some :message :random (rand)})
  (def msg  (send cfg queue item))

  msg => {:message-id string?}

  (def q-item (first (pop-items cfg queue)))

  q-item => {:message-id (:message-id msg)
             :item item}

  (delete-item cfg queue q-item) => {:status :deleted})

I don't think there is any doubt about which one of the two tests it is easier to read and understand what is happening. Moreover, if the test fails, it is extremely easy to start the REPL and evaluate each expression to debug and understand where is the problem.

Behind the scenes, RDT converts the test from the second form to the first form, so the two tests are equivalent, all the def are valid within the scope of repl-test only.

For example, the following snippet illustrates this point.

(repl-test "scoping rules for def and defn"

  (def  a 1)
  (def  b 2)
  (defn sum [x y] (+ x y))

  (sum a b) => 3

  (println "a:" a ", b:" b ", sum:" sum)
  ;; prints:
  ;; a: 1 , b: 2 , sum: #function[user/eval36636/fn--36637/fn--36638/sum--36640]
)

;; `a`, `b` and `sum` do not exists outside of the `repl-test` scope
;; (println "a:" a ", b:" b ", sum:" sum)
;; Syntax error
;; Unable to resolve symbol: a in this context

Internally the def and defn are converted into let binding thus avoiding the concurrency problems with the use of def.

Usage

In order to use the library add the dependency to your project.clj

;; Leiningen project
[com.brunobonacci/rdt "0.2.0"]

;; deps.edn format
{:deps { com.brunobonacci/rdt {:mvn/version "0.2.0"}}}

Current version:

Then require the namespace:

(ns foo.bar
  (:require [com.brunobonacci.rdt :refer [repl-test]]))

Check the online documentation

Writing tests

The general form is:

(repl-test "description of what you want to test (optional)"

    (function1 arg1 arg2 arg3) => expected-value1

    (function2 arg1 arg2 arg3) => expected-value2

    (functionN arg1 arg2 arg3) => expected-valueN
 )

The left side of the arrow => is the function you want to test, the right side of the arrow is the value or pattern you expect.

Fuzzy matching arrow =>

The single arrow (=>) performs a fuzzy match of the left-hand-side of the arrow with the right-hand-side of the arrow. It works differently than Midje.

The fuzzy match work as follow:

Basic values are matched with equality (=)

(repl-test
  (reduce + (range 1000)) => 499500    ;; numbers
  (str "foo" "-" "bar")   => "foo-bar" ;; strings
  (number? 23)            => true      ;; boolean
  (keyword "foo")         => :foo      ;; keywords
  (quote foo)             => 'foo      ;; symbols
  (first [])              => nil       ;; nil
  )

Lists, Vectors and Primitive arrays are matched on the given prefix

(repl-test
  (vector)       => []    ;; true
  (vector 1)     => [1]   ;; true
  (vector 2)     => [1]   ;; false, test fails
  (vector 1 2)   => [1 3] ;; false, test fails, [1 3] is not a prefix of [1 2]
  (vector 1 2 3) => [1 2] ;; true, [1 2] is a prefix of [1 2 3]

  (list)         => '()   ;; true
  (list 1)       => '(1)  ;; true
  (list 2)       => '(1)  ;; false, test fails
  (list 1 2)     => '(1 3);; false, test fails, (1 3) is not a prefix of (1 2)
  (list 1 2 3)   => '(1 2);; true, (1 2) is a prefix of (1 2 3)

  (byte-array 0)       => []    ;; true
  (byte-array [1])     => [1]   ;; true
  (byte-array [2])     => [1]   ;; false, test fails
  (byte-array [1 2])   => [1 3] ;; false, test fails, [1 3] is not a prefix of [1 2]
  (byte-array [1 2 3]) => [1 2] ;; true, [1 2] is a prefix of [1 2 3]
)

Maps are matched only on the given keys

Maps are matched only on the subset of keys which are present on the right-hand-side, the actual map can have more keys.

(repl-test
  (conj  {} :s 2)            => {:s 2}      ;; true
  (merge {:s 2} {:b 3}       => {:s 2}      ;; true, the map contain the key :s with the value 2
  (conj  {:s 2 :b 3} [:x 4]) => {:s 2 :x 4} ;; true, all required keys and values are there
  (conj  {:s 2 :b 3} [:Z 4]) => {:s 2 :x 4} ;; false, key :x is missing
)

Sets are matched based on the given subset of values

(repl-test
  (set 1 2 3)  => #{1 3 2} ;; true, the sets are the same
  (set 1 2 3)  => #{1 3}   ;; true, #{1 3} is a subset of #{1 2 3}
  (set 1 2 3)  => #{1 4}   ;; false, #{1 4} is NOT a subset of #{1 2 3}
)

Match against functions

In the right-hand-side you can use functions in place of values, in that case the function will be applied to the value and check whether the result is true or false.

(repl-test
  (+ 1 2 3)                   => number?  ;; true
  {:num 1 :str "two"}         => map?     ;; true
  {:num 1 :str "two" :k :z}   => {:num odd? :str string?} ;; all true,
)

Regex to match strings

In the right-hand-side you can use regular expression patterns in place of values, in that case the pattern will be matched against the value and check whether the result is true or false.

(repl-test
  (str "foo" 123) => #"foo.*"       ;; true
  (str "foo" 123) => #"foo"         ;; false, not a full match
  (str "foo" 123) => #"(?i)^FO+\d+" ;; true
  )

Why the fuzzy matcher?

Integration tests often return big nested maps with many keys, many of which values represent system state. On the other hand, your tests are often only focused on verifying that only specific keys have a particular value. The fuzzy matching is very useful in that case, as it only will match against the pattern you provided.

For example assume you have a function which returns a map with many nested keys and you are interested on testing only a few specific keys on the nested map, then the fuzzy match helps very much with it and allows for a very readable test.

For example assume you have a function which returns the current product availability if your depots:

(product-availability "SKU123ABZ")
;;=> {:product-id  "SKU123ABZ"
;;    :description "Next generation 3D printer"
;;    :availability
;;    {:free-stock {"LON1" {:location-id "LON1"
;;                          :description "London depot"
;;                          :quantity    13}
;;                  "BER1" {:location-id "BER1"
;;                          :description "Berlin depot"
;;                          :quantity    3}}
;;     :reserved   {"LON1" 0
;;                  "BER1" 0}}}

Let's assume you want to test a function which reverses some stock, if available, then you can write a test as follow:

(repl-test "Stock reservation"

  (reserve-stock "SKU123ABZ" "LON1" 10)
  => {:product-id  "SKU123ABZ"
      :availability
      {:free-stock {"LON1" {:quantity 3}}
       :reserved   {"LON1" 10}}}
  )

As you see in the above test you can specify only the subset of the map which is relevant to your test and ignore the rest. So much better for readability.

What if I need more precise matching?

If you need an exact matching or something in between the exact matching and the fuzzy matching then you can use the double-equal arrow ==> which behaves like Midje => arrow.

;; same behaviour as Midje arrow =>
(repl-test "Stock reservation"

  (reserve-stock "SKU123ABZ" "LON1" 10)
  ==> (contains
        {:product-id  "SKU123ABZ"
         :availability
         (contains
           {:free-stock (contains {"LON1" (contains {:quantity 3})})
            :reserved   (contains {"LON1" 10})})})
  )

Test labels and test filters

If you want selectively run or skip tests you can give them labels and use the labels to filter or skip some test.

(repl-test {:labels [:core]} "important core tests"

    ;; things that you must always run
)

(repl-test {:labels [:slow]} "compute pi to 1 trillion decimal values"

    ;; compute pi
)


(repl-test {:labels [:core :integration]} "Check DB transaction isolation "

    ;; start DB container
    ;; connect to DB
    ;; run transactions
    ;; verify isonlation
    ;; shutdown services
)

Then you can selectively run the tests you want or skip them. For example:

# run only :core tests
lein midje :filter core

# skip :slow tests
lein midje :filter -slow


# run :core tests, but skip integration
lein midje :filter core -integration

License

Copyright © 2021 Bruno Bonacci - Distributed under the Apache License v2.0