Hi. Is there a way to use spec to generate maps with "dependent" properties. For example, suppose my map spec requires than both :a and 😛 contain strings that have to be the same length?

That should be - Is there a way to use spec to generate maps with "dependent" properties. For example, suppose my map spec requires than both :a and :b contain strings that have to be the same length?

@mjmeintjes: Well, you can use s/and to restrict what a spec generates but that might not be the easiest way to make the generator work...

Hmm, perhaps it's OK after all:

boot.user> (require '[clojure.spec :as s])
nil
boot.user> (s/def ::a string?)
:boot.user/a
boot.user> (s/def ::b string?)
:boot.user/b
boot.user> (s/def ::foo (s/and (s/keys :req [::a ::b]) #(= (count (::a %)) (count (::b %)))))
:boot.user/foo
boot.user> (s/exercise ::foo)
([#:boot.user{:a "", :b ""} #:boot.user{:a "", :b ""}] [#:boot.user{:a "z", :b "S"} #:boot.user{:a "z", :b "S"}] [#:boot.user{:a "GF", :b "y8"} #:boot.user{:a "GF", :b "y8"}] [#:boot.user{:a "c5kf9D7nt88PhP2zRiIj", :b "Ax9Ua1vd561LvuWZ0o5r"} #:boot.user{:a "c5kf9D7nt88PhP2zRiIj", :b "Ax9Ua1vd561LvuWZ0o5r"}] [#:boot.user{:a "", :b ""} #:boot.user{:a "", :b ""}] [#:boot.user{:a "yaLME3MI4uSh0", :b "FIxs99gp43jSe"} #:boot.user{:a "yaLME3MI4uSh0", :b "FIxs99gp43jSe"}] [#:boot.user{:a "9", :b "3"} #:boot.user{:a "9", :b "3"}] [#:boot.user{:a "f7nOvuwZ30U000f", :b "3enZpUsZXP9ZVX1"} #:boot.user{:a "f7nOvuwZ30U000f", :b "3enZpUsZXP9ZVX1"}] [#:boot.user{:a "8aDmY08S8Q", :b "rHFcc5rAN9"} #:boot.user{:a "8aDmY08S8Q", :b "rHFcc5rAN9"}] [#:boot.user{:a "Ds8zm6", :b "yM0N6p"} #:boot.user{:a "Ds8zm6", :b "yM0N6p"}])

@seancorfield: Great, thanks! Looks promising, I'll investigate further.

After converting some property-based tests to use spec’s generators, I noticed that some of the tests became much slower. Investigation reveals that the growth characteristics of spec’s generators can be quite different from those of test.check. Here’s an example:

In most cases I doubt the difference would be very noticeable, but in my case two such generated integer values were being used as the dimensions of a two-dimensional matrix structure, so the size really blew up quickly.

I can see a rationale for making the growth a little steeper than the test.check version, but this seems a bit severe. 🙂

Especially since the default for :num-tests is 1000. 😮

I would say, in this case where the amount of data you test is N-squared based on random generated values, you probably want a custom generator that heavily restricts those dimension values?

How big, realistically, is a matrix you would be handling?

Oh, sure … I’ve done exactly that, now. Just wondering whether this is an appropriate default.

Given the actual behavior over 1000 iterations, you’d run into problems even if the amount of data was linear wrt the integer value.

And given that the visible symptom is just “my test run seems to just stall” it’s a bit confusing.

(Oh, and … my first attempt at constraining the generator was just to slap a #(< % 25) predicate on there, and then I got sporadic "Couldn't satisfy such-that predicate after 100 tries” failures. Which was easy enough to fix, but still leads me to think that the generator growth is just too fast.)

what is the justification of having spec in separate namespaces than the actual functions?

@tmtwd: per my earlier comment https://clojurians.slack.com/archives/clojure-spec/p1468450981002667

What happened to instrument-ns in alpha10? Not possible anymore?

folded into instrument. (instrument ‘my.namespace)

Ah, awesome.

Ah, I’ve figured out what’s going on with the growth rate of (s/gen pos-int?) mentioned above: some of the generators for built-in predicates use test.check functions that ignore the size guidance. This means (I think) that they won’t shrink properly when an error is found: https://github.com/clojure/clojure/blob/d920ada9fab7e9b8342d28d8295a600a814c1d8a/src/clj/clojure/spec/gen.clj#L128-L185

@alexmiller: I’d love to hear your thinking on this when you get a chance. Discussion starts at https://clojurians.slack.com/archives/clojure-spec/p1468507432002694

I think he'll be back tomorrow Glenn

out

OK, thanks!