string' test broken about megaparsec

[mrkkrp]

Initially logic of string' was to return string cased as argument of the parser. This later was changed. It's strange that it took so many tests (iterations) to reveal this defect in the test. I'll investigate/fix it tomorrow when I have clearer state of mind. Enough hacking for today.

[bkaestner]

It's strange that it took so many tests to revel this defect in the test.

I've noticed that there are only property based tests, but not any with static values. This behavioural change could have been noticed with static tests. Maybe you want to add some in the future for edge cases? Not for every property, but it might be reasonable for string'. You could reuse the already written properties for them.

Either way, take your time. The culprit has been found, and it was "just" in a test, not in string' itself.

[mrkkrp]

@bkaestner, another approach would be to modify existing test for string' so this sort of behavior would manifest itself with higher probability.

[bkaestner]

Hm. Something like this comes to mind:

-- Randomly change the case in the given string
fuzzyCase :: String -> Gen String
fuzzyCase s = do
    b <- vector (length s)
    return $ zipWith f s b
  where f k True  = if isLower k then toUpper k else toLower k
        f k False = k

prop_string'_fuzzied :: String -> Property
prop_string'_fuzzied a =
  forAll (fuzzyCase a) $ \s ->
   checkString' (string' a) a casei (showToken a) s

This would test string' with a string that only differs in case (and can be used together with the current test).

[mrkkrp]

@bkaestner, I've fixed that bug (it was trivial Right a' → Right s' in definition of checkString). I think I'll add your prop_string'_fuzzied to the test suite.