Bump version bound for QuickCheck 2.13

[nick8325]

I just released QuickCheck 2.13. There are no major API changes and Hspec at least builds with it (I haven't run it) but at least the version bounds will need relaxing.

It also uses splitmix rather than tf-random for random number generation, which perhaps makes #406 obsolete.

[sol]

@nick8325 there seems to be something fishy.

Given

ghci> import Test.QuickCheck
ghci> import Test.QuickCheck.Random 
ghci> let p = (/= 1) :: Int -> Bool
ghci> let qc = quickCheckWith stdArgs {replay = Just (mkQCGen 0, 0)}

I observe that verbose changes the number of tests that are run:

ghci> qc p
*** Failed! Falsified (after 6 tests):                      
1

vs

ghci> qc (verbose p)
...
...
...
*** Failed! Falsified (after 3 tests):                      
1

Is this expected?

[nick8325]

This is really strange. I hadn't thought about making verbose p and p behave identically (only that they should have the same distribution of test cases), but of course they should if at all possible.

The weird thing is that this works with TFGen, but not with SMGen or StdGen! Just changing Test.QuickCheck.Random so that it calls newStdGen and mkStdGen instead of newTFGen and mkTFGen causes your example to break. So it seems that TFGen is doing something very strange here. I'll have to look into it more closely.

[nick8325]

Oops:

import Test.QuickCheck
import System.Random
import System.Random.TF

prop_rng :: Int -> Property
prop_rng n =
  expectFailure $
  genInt g === genInt (fst (split g))
  where
    genInt :: TFGen -> Int
    genInt g = fst (next g)

    g :: TFGen
    g = mkTFGen n

> quickCheck prop_rng 
*** Failed! Passed 100 tests (expected failure).

It looks like the problem only occurs with mkTFGen and not newTFGen.

[nick8325]

My initial impression is that this bug shouldn't worsen testing, at least, because a seed in QuickCheck is either split or applied to next, never both in the same property.

[sol]

@nick8325 there is more. Apparently failingTestCase means something different now. Consider:

ghci> let qc = quickCheckWithResult stdArgs {chatty = False, replay = Just (mkQCGen 0, 0)}

With 2.12.6.1:

> failingTestCase <$> qc ((==) :: Int -> Int -> Bool)
["0","1"]

vs 2.13

ghci> failingTestCase <$> qc ((==) :: Int -> Int -> Bool)
["1\n0"]

[sol]

@snoyberg @alexeyzab I can't comment on commercialhaskell/stackage#4444. hspec compatibility with QuickChekc-1.13 will be blocked until we have sorted this stuff out.

[alexeyzab]

@sol 👍

[nick8325]

@sol Both of those should be fixed now. Let me know if you're happy for me to make a new release.

[nick8325]

Ah wait, not sure that it's working yet. I'll report back...

[nick8325]

There, that should be actually fixed now.

[sol]

@nick8325 looks good now. Let me know once it's on Hackage._

[nick8325]

There, 2.13.1 is now on Hackage!

[sol]

On Hackage, closing.