Number of QuickCheck test cases is inconsistent

[UnkindPartition]

@jtdaugherty reports that somewhere between QuickCheck 2.8.2 and 2.9.1 a bug (or a silent and undocumented in the Changelog change in semantics) was introduced: UnkindPartition/tasty#147

Could you look into it?

[phadej]

that's related to

Remove "exhaustive" from Testable typeclass;
      instead, combinators which are nonexhaustive (such as forAll)
      call "again", which should be more robust

With

-    , testProperty "merged marking" $ property $ do
+    , testProperty "merged marking" $ property $ again $ do

I get

  merged marking:     OK (0.01s)
    +++ OK, passed 100 tests.

I'm not sure whether

 instance Testable prop => Testable (Gen prop) where
-  property mp = MkProperty $ do p <- mp; unProperty (property p)
+  property mp = MkProperty $ do p <- mp; unProperty (again p)

is the right solution on the QuickCheck side

EDIT It's run once as text-markup the Testable is Gen Bool, and Testable Bool is tested once, which propagates to Gen Bool, but obviously is incorrect. I guess, it's safe to make that change.

ping @nick8325

[nick8325]

Thanks @phadej for the patch.

QuickCheck has a mechanism where the user can ask a property to be tested only once (useful e.g. when you just want to check that an IO Bool evaluates to true). In 2.9 I simplified the way this mechanism is implemented, with the side effect that a property will only be repeatedly tested if it uses forAll or similar. Unfortunately this was a bit too aggressive - @phadej's patch makes it so that the property will also be repeatedly tested if it uses the Gen monad.

[jtdaugherty]

Thanks! Can you mention here when this is released to Hackage?

[nick8325]

Should be fixed in 2.9.2 which is out now.

[jtdaugherty]

Thanks!