On Sun, Apr 13, 2014 at 12:24 PM, Roman Cheplyaka <roma@ro-che.info> wrote:
> * Konstantine Rybnikov <k-bx@k-bx.com> [2014-04-13 12:21:44+0200]
>> Just FYI, this still gives a warning:
>> ...
>>     case m of
>>       [] -> putStrLn "empty"
>>       (M.toList -> (x:_)) -> putStrLn $ "ok, some random elem is: " ++ show x
>
> Does that surprise you? The compiler doesn't have any special knowledge about
> the M.toList function to infer that these two cases are exhaustive.

Roman is right, and I think it's clearer if you consider this without view patterns:

case m of
  [] -> ...
  m' -> case M.toList m' of
           (x : _) -> ...

Looking at this, it's clear that the patterns are not exhaustive - you didn't match the scenario that M.toList m' produced an empty list.

You know that M.toList (M.fromList []) == [], but GHC doesn't - and I think this is where the problem lies. With that information you might be able to have exhaustive pattern matches there.

- ocharles