http://hackage.haskell.org/package/validation-0.3.4/docs/Data-Validation.htm... On Thu, May 22, 2014 at 7:46 AM, John Meacham wrote:

On Mon, May 19, 2014 at 8:31 AM, Corentin Dupont wrote:

...

However the instance in Control.Monad.Trans.Error.Error gets in the way... I cannot use this one, it works only with Strings. My use case is that the Either handle in the Right a final value, or else in the Left a list of things "left to do" to compute the value. How can I do?

You want to use a newtype to wrap your Either. It sounds like you would want to do this independently of the String instance because you are not using left as shortcutting failure anyway, rather you want to collect together the failure actions.

newtype NotDoneYet b a = NotDoneYet (Either [b] a)

instance Applicative (NotDoneYet b) where Left xs <*> Left ys = Left (xs ++ ys) ...

-- John Meacham - http://notanumber.net/ _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

I use a similar idiom newtype P b a = P ([b],Maybe a) Note that the [b] is _outside_ the maybe (or either), this means that the accumulated value can be propagated through failure. Notably, <*> and `ap` have very different meanings for this which is not true for other monads. <*> will collect [b] from both sides while `ap` will only do the left. John John On Wed, May 21, 2014 at 9:18 PM, Kim-Ee Yeoh wrote:

On Thu, May 22, 2014 at 6:58 AM, Tony Morris wrote:

...

http://hackage.haskell.org/package/validation-0.3.4/docs/Data-Validation.htm...

I don't understand how to use this package.

Is it because Either is an ugly name for the data type and Validation is preferable?

And AccValidation is to be used over Validation, like in Hungarian notation, to indicate (the possibility of?) a Semigroup instance on Left / Failure?

-- Kim-Ee

-- John Meacham - http://notanumber.net/

On Thu, May 22, 2014 at 12:21 AM, Tony Morris wrote:

On 22/05/14 14:28, John Meacham wrote:

...

Notably, <*> and `ap` have very different meanings for this which is not true for other monads. Please don't do this. Kittens die and all that.

In my particular case I did this, it is actually collecting a non-deterministic set of values so the difference between the two does not matter for correctness, but is practically useful to have a greater set of errors to choose from. The difference is hidden behind an abstract interface so it actually is externally a monad. Manual imprecise exceptions. I just prefer to think of `ap` as buggy and unrelated to <*> sometimes... John -- John Meacham - http://notanumber.net/

On Thu, May 22, 2014 at 05:10:24PM +1000, Tony Morris wrote:

AccValidation has an Applicative instance that accumulates on "Left" (Failure) as per the previous email. Therefore, it has no Monad.

This applicative is called Errors in the transformers package (in Control.Applicative.Lift, of which it's a special case).