How to generate UUIDs that can work with test.check in Clojure

Question

Generative testing seems interesting, but I needed to generate random UUIDs as part of the testing. java.util.UUID/newRandom doesn't play nice with test.check shrinking.

The java code looks like:

public static UUID randomUUID()
{
  long lsb = r.nextLong();
  long msb = r.nextLong();

  lsb &= 0x3FFFFFFFFFFFFFFFL;
  lsb |= 0x8000000000000000L; // set top two bits to variant 2

  msb &= 0xFFFFFFFFFFFF0FFFL;
  msb |= 0x4000; // Version 4;

  return new UUID( msb, lsb );
}

Which is trickier to translate to Clojure than it would seem.

How do I write a random UUID function in Clojure that can be successfully shrunk?

Answer 1

A fn that takes two longs and generates a proper type 4 UUID is:

(defn make-uuid [[msb lsb]]
  (java.util.UUID. (-> msb
                   (bit-clear 15)
                   (bit-set   14)
                   (bit-clear 13)
                   (bit-clear 12))
               (-> lsb
                   (bit-set   63)
                   (bit-clear 62))))

You can use a regex to check the result, (need to convert it to a string first).

(def uuid-v4-regex
     #"(?i)[a-f0-9]{8}-[a-f0-9]{4}-4[a-f0-9]{3}-[98ab][a-f0-9]{3}-[a-f0-9]{12}")

Then you can test it as following:

(def uuids (gen/fmap make-uuid (gen/tuple (gen/choose 0 Long/MAX_VALUE)
                                          (gen/choose 0 Long/MAX_VALUE))))

(defspec check-random-uuid 100000
  (for-all [uuid uuids]
       (re-find uuid-v4-regex (str uuid))))

And the tests look like:

(check-random-uuid)
=> {:result true, :num-tests 100000, :seed 1422050154338}

Just for fun, I removed one of the valid characters (9) for the second field, and this is what a failing test looks like, so you can see how shrinking from :fail to :smallest can help.

(pp/pprint (check-random-uuid))
{:result nil,
 :seed 1422050276824,
 :failing-size 2,
 :num-tests 3,
 :fail [#uuid "2c6d1442-eec3-4800-972e-02905c1b3c00"],
 :shrunk
 {:total-nodes-visited 932,
  :depth 29,
  :result nil,
  :smallest [#uuid "00000000-0000-4000-9000-000000000000"]}}

Which shows you how much noise shrinking can remove from your test case.

Answer 2

As of test.check version 0.9.0, there is a built-in gen/uuid which generates random uuids that do not shrink.

If you really want shrinking UUIDs, the generate-two-longs method is also easier, as the new gen/large-integer will generate the full range of longs, though not with a uniform distribution.

There is still not a builtin generator than can generate a uniform distribution over all longs, but I think that's something that should be added in the future.