[Haskell-cafe] Conditional properties in QuickCheck alter test data generation?

[Kevin Van Horn]

I'm learning QuickCheck, and I'm puzzled by the behavior I'm seeing  
with conditional properties.  After writing and loading a simple qsort  
function from a separate file, I typed these into ghci:

let prop_minimum xs = (length xs > 0) ==> head (qsort xs) == minimum xs
   where types = xs :: [Integer]

let prop_minimum1 xs = (length xs > 3) ==> head (qsort xs) == minimum xs
   where types = xs :: [Integer]

These differ only in that prop_minimum1 uses a more restrictive  
condition.

Then quickCheck prop_minimum outputs

  "OK, passed 100 tests"

but quickCheck prop_minimum1 outputs

  "Arguments exhausted after 0 tests"

Investigating with verboseCheck yields puzzling results.  For example,  
verboseCheck prop_minimum says "OK, passed 100 tests", but some of the  
test instances it shows have length 0, contrary to the condition...  
and a significant fraction of the test instances have length > 3, or  
even length > 10!  So why the problem when running quickCheck on  
prop_minimum1?

Running verboseCheck prop_minimum1 just deepens the mystery.  Now we  
see most of the test instances generated having 0 length, and none of  
them having length > 3.  What's going on here?  Why is the  
distribution of generated test instances so drastically altered?

BTW, I'm using ghc version 6.10.1, running on Intel Mac OS 10.5.5.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.haskell.org/pipermail/haskell-cafe/attachments/20081217/587ce466/attachment.htm