Gen.elements only returns first or last element.

[niklas]

I ran into a problem with Gen.elements. No matter how often I use it with a Random Generator, it always gives me either the first or the last element of the Foldable1 f.

My suspicion is the internal helper fromIndex:

edge cases seem to work

fromIndex 0 $ 10 :| [20,30,40,50,60] -- 10
fromIndex 10 $ 10 :| [20,30,40,50,60] -- 60

but indices within the bounds don't

fromIndex 5 $ 10 :| [20,30,40,50,60] -- 60
fromIndex 1 $ 10 :| [20,30,40,50,60] -- 60

given a negative index, the function crashes for me

fromIndex -10 $ 10 :| [20,30,40,50,60]
Error found:
in module $PSCI
at :1:12 - 1:14 (line 1, column 12 - line 1, column 14)

  Could not match type
       
    Int
       
  with type
                      
    Int -> t0 t1 -> t1
                      

while checking that type Int
  is at least as general as type Int -> t0 t1 -> t1
while checking that expression 10
  has type Int -> t0 t1 -> t1
in value declaration it

where t1 is an unknown type
      t0 is an unknown type

I am still new to Purescript, so I am not sure why a negative Int can cause that.

Luckily, it won't be called with negative numbers as long the provided chooseInt behaves correctly. Nonetheless, above behavior would explain the the uneven distribution of elements.

[niklas]

If I switch the order of elements g and f in the type class instance, the distribution is more even. The index lookup is done from the back (so calling it with 0 gets the last element). Negative numbers still cause the same crash.

instance semigroupAtIndex :: Semigroup (AtIndex a) where
  append (AtIndex g) (AtIndex f) = AtIndex \i -> if i <= 0 then f i else g (i - 1)

[garyb]

You need to use parens around the negative number, -10 is an expression rather than a literal.

What type are you using for your MonadGen instance, maybe the poor distribution is down to that?

[garyb]

Hmm, no, I see it too. That's extremely concerning that it's been like this for a long time!

[hdgarrood]

I think we could do with some distribution tests, maybe in quickcheck if not here. I've definitely seen a few cases (separate from this one) where quickcheck is giving an odd distribution of results.

[garyb]

Yeah, I reproduced this by adding a test case. I thought this library had better test coverage than it does.