[Haskell-cafe] Declaring Functors for Type constrained data types

Erik Hesselink hesselink at gmail.com
Wed Mar 1 14:33:38 UTC 2017


Or even:

data Task a = UpdateAcctsTask
            | SendEmailTask EmailConfig
            | GeneralWriterTask a

process :: Task a -> Bool
can_run :: Task a -> Bool

In general, start with data types based on the structure of your data and
functions that operate on that data. Type classes are not usually the right
abstraction for domain specific code, in my experience.

Erik

On 1 March 2017 at 15:19, Patrick Chilton <chpatrick at gmail.com> wrote:

> But if you have for example a tree of heterogenous tasks as you said, you
> can no longer dispatch based on type because the implementation depends on
> the value of each task in the tree, so you're better off representing it
> like this:
>
> data Task = Task
>   { doTask :: IO ()
>   , taskName :: String
>   }
>
> class Taskable a where
>   toTask :: a -> Task
>
> instance Taskable EmailConfig where
>   toTask (EmailConfig ...) = Task { ... }
>
> type TaskTree = Tree Task -- easy
>
> On Wed, Mar 1, 2017 at 3:11 PM, Guru Devanla <gurudev.devanla at gmail.com>
> wrote:
>
>> Just to make sure I understand. In my initial example, the following are
>> all types and I wanted the dispatch to happen on this types once they are
>> instances of the Taskable type class.
>>
>> -- task 1
>> data UpdateAcctsTask = UpdateAccts
>>
>> -- task 2
>> data EmailConfig = EmaiConfig {someattrs::String}
>> data SendEmailTask = SendEmailsTask EmailConfig
>>
>> -- task 3
>> data GeneralWriterTask a = GeneralWriterTask a
>>
>>
>> Now, later on I expect users to create more types which are all say,
>> instances of Taskable typeclass.  So, here I do not expect to have any
>> dispatch based on value, but based on types.
>>
>>
>>
>>
>>
>>
>> On Wed, Mar 1, 2017 at 5:24 AM, Patrick Chilton <chpatrick at gmail.com>
>> wrote:
>>
>>>  Is that thinking usually an anti-pattern.
>>>
>>>
>>> Typeclasses choose between implementations of a method based on static
>>> *types* known at compile time. In OOP, which implementation of an
>>> interface method you get depends on the run-time object *instance*.
>>> They're not really related despite the similarities.
>>>
>>> For example, if you have a list of widgets and you want to do something
>>> different depending on what their run-time value is, a typeclass would be
>>> the wrong thing to use since it dispatches on types, not values. It's
>>> possible to make them work a little bit like OOP classes, but then you end
>>> up with the classic antipattern.
>>>
>>> On Wed, Mar 1, 2017 at 2:16 PM, Guru Devanla <gurudev.devanla at gmail.com>
>>> wrote:
>>>
>>>> This problem I was solving is more of a re-implementation of some code
>>>> I had in Python. In Python, I had a very class-based structure of this
>>>> design. By class based structure, I mean we have class-level methods for
>>>> each class (where each class  is a type of Task), and all of them implement
>>>> the same set of methods. This classes are never instantiated, but all
>>>> methods are invoked on the class.
>>>>
>>>> WIth that perspective, the record patter @Patrick recommended directly
>>>> maps to that design. The record data type here becomes a
>>>> abstract-representation of my class(abstract base class in OO terms)  and
>>>> each task provides its methods.  I see that relationship.  Is that the
>>>> approach I should be aiming for?
>>>>
>>>> Secondly, while choosing the type-class approach,  I imagined all these
>>>> required class methods to be an interface and therefore an interface could
>>>> directly map to a type class in Haskell.  Is that thinking usually an
>>>> anti-pattern.
>>>>
>>>> The existential antipattern
>>>> <https://lukepalmer.wordpress.com/2010/01/24/haskell-antipattern-existential-typeclass/>.
>>>> link was very useful and made me re-asses my inclination to defining type
>>>> classes and embedded types right away.
>>>>
>>>> On Tue, Feb 28, 2017 at 9:55 AM, Patrick Chilton <chpatrick at gmail.com>
>>>> wrote:
>>>>
>>>>> You could also consider representing tasks like this instead of using
>>>>> a typeclass:
>>>>>
>>>>> data Task = Task
>>>>>   { process :: m ()
>>>>>   , canRun :: m Bool
>>>>>   }
>>>>>
>>>>> The Taskable + existential GADT example seems like it could be an
>>>>> example of the existential antipattern
>>>>> <https://lukepalmer.wordpress.com/2010/01/24/haskell-antipattern-existential-typeclass/>
>>>>> .
>>>>>
>>>>> If your GADT really does have a as a type parameter, it would be more
>>>>> idiomatic to check for the typeclass when you use it:
>>>>>
>>>>> doStuffWithTasks :: Taskable a => Task a -> ...
>>>>>
>>>>> But then what's the point of the Task datatype?
>>>>>
>>>>> On Tue, Feb 28, 2017 at 1:48 AM, Guru Devanla <
>>>>> gurudev.devanla at gmail.com> wrote:
>>>>>
>>>>>> Hello All,
>>>>>>
>>>>>> I am working on a program that will define a bunch of tasks. Each task
>>>>>> will have to implement certain methods as part of a type class.
>>>>>>
>>>>>> -- task 1
>>>>>> data UpdateAcctsTask = UpdateAccts
>>>>>>
>>>>>> -- task 2
>>>>>> data EmailConfig = EmaiConfig {someattrs::String}
>>>>>> data SendEmailTask = SendEmailsTask EmailConfig
>>>>>>
>>>>>> -- task 3
>>>>>> data GeneralWriterTask a = GeneralWriterTask a
>>>>>>
>>>>>> Each of these tasks implement a class, Taskable. The return
>>>>>> values are simplified for this example.
>>>>>>
>>>>>> class Taskable a where
>>>>>>   process :: a -> Bool
>>>>>>   can_run :: a -> Bool
>>>>>>
>>>>>>
>>>>>> This works fine. I can expand on these tasks and execute them.
>>>>>>
>>>>>> Now, I wanted to be able to defined dependencies between these
>>>>>> (Taskable's). I decided
>>>>>> I could create a data type for this dependency and may be also get a
>>>>>> FreeMonad
>>>>>> around this structure for further processing using a graph of Tasks.
>>>>>> But, before that I wanted
>>>>>> to create an wrapper for these Taskables and create a functor for it
>>>>>> as follows
>>>>>>
>>>>>> The first thing I did was, define a Task, which generalizes over all
>>>>>> the above defined (and future Taskables)
>>>>>>
>>>>>> data Task a where
>>>>>>   Task :: (Taskable a) => a -> Task a
>>>>>>
>>>>>>
>>>>>> instance Functor Task where
>>>>>>   fmap:: (Taskable a, Taskable b) -> (a -> b) -> Task a  -> Task b
>>>>>> --- THIS DOES NOT WORK
>>>>>>   fmap f (Task a) = Task $ f a
>>>>>>
>>>>>>
>>>>>> But, I realized that I cannot define an fmap over a type constraint.
>>>>>>
>>>>>> My questions are:
>>>>>>
>>>>>> 1. Is there any way to do this. I see there is an answer of SO. I
>>>>>> wanted
>>>>>>    to make sure if there were any improvements to this since that
>>>>>> answer'
>>>>>>    was posted.
>>>>>>    http://stackoverflow.com/questions/17157579/functor-instance
>>>>>> -for-a-gadt-with-type-constraint
>>>>>>
>>>>>> 2. Secondly, I would like to know why this is not possible. Is it a
>>>>>> current
>>>>>>    limitation of GHC or if there is some fundamental category theory
>>>>>> concepts
>>>>>>    that dis-allows such declarations that I need to grok!
>>>>>>
>>>>>> Appreciate any help on this. Thank you!
>>>>>>
>>>>>> _______________________________________________
>>>>>> Haskell-Cafe mailing list
>>>>>> To (un)subscribe, modify options or view archives go to:
>>>>>> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
>>>>>> Only members subscribed via the mailman list are allowed to post.
>>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
>
> _______________________________________________
> Haskell-Cafe mailing list
> To (un)subscribe, modify options or view archives go to:
> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> Only members subscribed via the mailman list are allowed to post.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-cafe/attachments/20170301/92127e47/attachment.html>


More information about the Haskell-Cafe mailing list