Mario Lang mlang at delysid.org
Mon Oct 5 23:10:10 UTC 2015

Benjamin Edwards <edwards.benj at gmail.com> writes:

> If you want to use monad transformers and have Either e [a] as the result
> type then you need Either to be the inner monad and List to be the outer
> monad. If you look at the types of EitherT (from the either package) and
> ListT from transformers this should hopefully make sense. Then you would
> keep the same impl as you have now, only you would need to "run" the ListT
> computation to yield Either e [a]. Anything that you would like to do
> inside of the inner Error monad will need to lifted inside of it using
Hmm, sort of, and also not :-)  I finally managed to rewrite vs,
however, I used either to handle Either:

vs :: Music.Dur -> AmbiguousVoice -> Either e [Voice]
vs _ []     = return [[]]
vs l (x:xs) = either Left f $pms l x where f pms = fmap concat$ sequence $pms >>= \pm -> return$ either Left (\pmss -> Right $(pm :) <$> pmss) (vs (l - dur pm) xs)

It doesn't particularily look pretty, but I am rather relieved I finally
managed to make that change.  Took me roughly 4 attempts, a lot of
asking, and a few hours of playing around.  On the positive side, I have
learnt a lot.  However, I wonder if there is a more idiomatic way of
doing what I do above.  Took me a long time to realize there is a concat
missing.  I guess this is the result of sequence being used to collapse
the Eithers.

Thanks for helping by providing input.

> Ben
>
> On Mon, 5 Oct 2015 at 11:06 Mario Lang <mlang at delysid.org> wrote:
>
>> Hi.
>>
>> Consider this structure:
>>
>> vs :: Rational -> [Input] -> [[Output]]
>> vs _ []     = return []
>> vs l (x:xs) = pms l x >>= \pm -> (pm :) <$> vs (l - dur pm) xs >> >> pms :: Rational -> Input -> [Output] >> pms l x = [x, x+1, x+2, ...] -- Just an example, not real code. >> -- in reality, l is used to determine >> -- the result of pms. >> >> This is basically traverse, but with a state (l) added to it. >> So without the state, vs could be written as >> >> vs = traverse pms >> >> Now, I want to add Either e to this, like: >> >> vs :: Rational -> [Input] -> Either e [[Output]] >> pms :: Rational -> Input -> Either e [Output] >> >> However, I have no idea how to implement vs. >> >> Interestingly, adding Either e to vs without changing the code lets it >> compile, but it gives me the wrong result: >> >> vs :: Rational -> [Input] -> Either e [[Output]] >> vs _ [] = return [] >> vs l (x:xs) = pms l x >>= \pm -> (pm :) <$> vs (l - pm) xs
>>
>> Since I am in the Either monad now, >>= does not do non-determinism, it
>> simply unwraps the Either from pms.  I have to admit, I dont fully
>> understand why this compiles, and what exactly it does wrong.  I only
>> see from testing that the results can't be right.
>>
>> On IRC, Gurkenglas suggested to use the State monad, like this:
>>
>> vs :: Rational -> [Input] -> Either e [[Output]]
>> vs l = evalStateT l . mapM v where
>>   v x = do l <- get
>>            pm <- lift \$ pms l x
>>            put (l - dur pm)
>>            return pm
>>
>> This compiles, but also yields unexpected results.
>>
>> I have invested several hours now trying to add Either around this
>> algorithm, so that I can emit hard failures.  I am sort of frustrated
>> and out of ideas.  Somehow, I can't figure out what these
>> transformations actually change in behaviour.  I am being told, by quite
>> experienced Haskell programmers, that this is supposed to be correct,
>> but my testing tells me otherwise.  So before I just give up on this,
>> could someone please have a look and let me know if I have missed
>> something obvious?
>>
>> --
>> CYa,
>>   ⡍⠁⠗⠊⠕
>> _______________________________________________
>> Beginners mailing list
>>
> _______________________________________________
> Beginners mailing list
. '
-      <URL:http://delysid.org/>  <URL:http://www.staff.tugraz.at/mlang/>