Hi Michael,

I have been taking this approach of State Monads and I have hit upon 3 common patterns that I think may not be the idiomatic way of dealing with state. I would like to continue with the example we have to explain those scenarios. Any input on this would be great..

1.  I see that almost in every function I deal with state, I have e <- get , expression in the begining. I always ending up having to use the state to query for different values. I guess this is OK.

2. In deeply nested function, where I pass state, I also end up calling evalState a couple of times to get to some values.  Is that common. Here is one example, from our toy problem.

first_student_owes_more :: RowId Student -> RowId Student -> State Environment Bool
first_student_owes_more student_1 student_2 = do
  e <- get
  let fees_owed_by_student_1 = evalState (student_totalFeesOwed student_1) $ e
  let fees_owed_by_student_2 = evalState (student_totalFeesOwed student_2) $ e
  return $ fees_owed_by_student_1 > fees_owed_by_student_2

You see, I have to evalState twice to get to what I want. Is that a common way to use the State.

3.  I also end up performing evalState while mapping over a list of values. Say, I wanted to loop around a list of students to perform the function in (2), then invariable for each iteration of Map, I am calling evalState once.

This gets hairy, if the value in my State is a Data.Map structure.

Am I using the State Monad in a round about way?

Thanks
Guru







On Wed, Jul 6, 2016 at 8:39 PM, Guru Devanla <gurudev.devanla@gmail.com> wrote:
The State monad makes a lot of sense for this. I was initially hesitant to go down this path *fearing* monads. But, today I was able to change most of my code to work with the same pattern you provided. Also, my initial impression on State monads was that, it was not a good idea to carry a *big blob* of State around. That impression comes from the thought process influenced by imperative programming. After coding up this, it is a lot clear that State monad declares operations and it is not the `state` itself that is carried around. I am elated!

Thank you for the help. I may have more questions as I progress down this path.

Thanks
Guru


On Tue, Jul 5, 2016 at 10:30 PM, Michael Burge <michaelburge@pobox.com> wrote:
The implicit parameter approach is best if the environment never changes, or at least doesn't change during the computation You can rebind the variable in the middle of a computation, but it's not a good road to go down.

The easiest way to simulate a changing environment is to use the State monad. There are other techniques: lenses, nested patterns, rebinding an implicit parameter, ST monad, generating a list of changes and applying the changes to the original state, etc. But - despite having to change your syntax somewhat - I think you'll find it easiest to use a state monad to manage this.

Here's a somewhat verbose example of using State to track updates. You can make it less verbose, but I chose to keep it simple. In this example, it updates student_feesOwed as part of registering for a class. So we no longer need to calculate anything: It just grabs the value off of the Student.

import Control.Applicative
import Control.Monad.Trans.State.Strict
import Data.Monoid

import qualified Data.IntMap as M

newtype RowId a = RowId Int deriving (Eq)

data Classroom = Classroom { classroom_id :: RowId Classroom, classroom_extraFees :: Float, classroom_students :: [ RowId Student ] }
data Student = Student { student_id :: RowId Student, student_name::String, student_feesOwed::Float}

data Environment = Environment {
  environment_classroom :: Maybe Classroom,
  environment_students  :: M.IntMap Student
  }

student_totalFeesOwed :: RowId Student -> State Environment Float
student_totalFeesOwed (RowId studentId) = do
  (Environment mClassroom students) <- get
  case mClassroom of
    Nothing -> return 0.0
    Just classroom -> do
      let fees = student_feesOwed $ students M.! studentId
      return fees

student_addFee :: RowId Student -> Float -> State Environment ()
student_addFee studentId fee = do
  modify $ \e -> e { environment_students = M.map (addFee studentId fee) $ environment_students e }
  where
    addFee studentId fee student =
      if studentId == student_id student
      then student { student_feesOwed = student_feesOwed student + fee }
      else student

environment_addStudent :: Student -> State Environment ()
environment_addStudent student = do
  let (RowId key) = student_id student
      value = student
  modify $ \e -> e { environment_students = M.insert key value (environment_students e) }

classroom_addStudent :: Classroom -> RowId Student -> State Environment ()
classroom_addStudent classroom studentId = do
  modify $ \e -> e { environment_classroom = addStudent studentId <$> environment_classroom e }
  where
    addStudent :: RowId Student -> Classroom -> Classroom
    addStudent studentId classroom = classroom { classroom_students = studentId : (classroom_students classroom) }

student_registerClass :: RowId Student -> Classroom -> State Environment ()
student_registerClass studentId classroom = do
  student_addFee studentId (classroom_extraFees classroom)
  modify $ \e -> e { environment_classroom = Just classroom }
  classroom_addStudent classroom studentId
  
main = do
  let studentId = RowId 1
      student = Student studentId "Bob" 250.00
      classroom = Classroom (RowId 1) 500.00 []
      initialEnvironment = Environment Nothing mempty
  let totalFeesOwed = flip evalState initialEnvironment $ do
        environment_addStudent student
        student_registerClass studentId classroom
        totalFeesOwed <- student_totalFeesOwed studentId
        return totalFeesOwed
  putStrLn $ show totalFeesOwed


On Tue, Jul 5, 2016 at 9:44 PM, Guru Devanla <gurudev.devanla@gmail.com> wrote:
Hi Michael,

That is excellent. I read about Implicit parameters after reading your post. I like this approach better than Reader monad for my current use case. I wanted to stay away from Reader Monad given that this is my first experimental project and dealing with Reader Monads into levels of nested function calls involved lot more head-ache for me.

That said, I plan to try this approach and also see how I can enable this set up in my HUnit tests as well.

One other question, I have regarding this design is as follows:  Say, during the progress of the computation, the `student_feesOwed` changes, and therefore we have a new instance of classroom with new instance of student in it (with the updated feesOwed). I am guessing, this would mean, wrapping up this new instance into the environment from there on and calling the subsequent functions. Is that assumption, right. Nevertheless, I will play with approach tomorrow and report back!

Thanks
Guru











On Tue, Jul 5, 2016 at 7:18 PM, Michael Burge <michaelburge@pobox.com> wrote:
When I have functions that are pure but depend on some common state(say in a config file, or retrieved from a database at startup), I like to use implicit parameters to hide it. You can use a type alias to avoid it cluttering up most signatures. Below, a value of type 'Environmental Float' means 'A float value, dependent on some fixed environment containing all students and the single unique classroom'. If you have a deep chain of 'Environmental a' values, the implicit parameter will be automatically propagated to the deepest parts of the expression.

You could also use a Reader monad, but they seem to require more invasive syntactic changes: They are better if you later expect to need other monads like IO, but if you're just doing calculations they're overkill. You could also define a type alias 'Environmental a = Environment -> a', but then if you have multiple such states they don't compose well(they require you to apply the implicit state in the correct order, and it can be a little awkward to propagate the parameter).

Here's how I would start to structure your example in a larger project:

{-# LANGUAGE ImplicitParams,RankNTypes #-}

import qualified Data.IntMap as M

newtype RowId a = RowId Int

data Classroom = Classroom { classroom_id :: RowId Classroom, classroom_extraFees :: Float, classroom_students :: [ RowId Student ] }
data Student = Student { student_id :: RowId Student, student_name::String, student_feesOwed::Float}

data Environment = Environment {
  environment_classroom :: Classroom,
  environment_students  :: M.IntMap Student
  }

type Environmental a = (?e :: Environment) => a

classroom :: (?e :: Environment) => Classroom
classroom = environment_classroom ?e

students :: (?e :: Environment) => M.IntMap Student
students = environment_students ?e

student_totalFeesOwed :: RowId Student -> Environmental Float
student_totalFeesOwed (RowId studentId) = classroom_extraFees classroom + (student_feesOwed $ students M.! studentId)

main = do
  let student = Student (RowId 1) "Bob" 250.00
  let ?e = Environment {
        environment_classroom = Classroom (RowId 1) 500.00 [ RowId 1 ],
        environment_students = M.fromList [ (1, student) ]
        }
  putStrLn $ show $ student_totalFeesOwed $ RowId 1


On Tue, Jul 5, 2016 at 6:26 PM, Guru Devanla <gurudev.devanla@gmail.com> wrote:
Hello All,

I am just getting myself to code in Haskell and would like to design advice.  Below, I have a made up example:
                                                                                                                                                                                                                                                           
                                                                                                                                                                                                                                                                                 
data ClassRoom = ClassRoom { classRoomNo:: Integer, extra_fees::Float, students: Map StudentId Student}
data Student = Student {name::String, feesOwed::Float}
data StudentId = Integer                                                                                                                                                                                                                                                         
                                                                                                                                                                                                                                                                                 
get_fees_owed classroom student_id = extra_fees + feesOwed $ (students classroom) M.! studentid                                                                                                                                                                       

Here the `get_fees_owed`  needs information from the container 'classroom'. 

Here is my question/problem:
                                                                                                  
I believe I should model most of my code as expressions, rather than storing pre-computed values such as `fees_owed`.  But,
defining expressions involve passing the container objects all over. For example, deep down in a function that deals with just
one `student`, I might need the fees owed information. Without, having a reference to the container, I cannot call get_fees_owed.

Also, I think it hinders composing of functions that just deal with one student at a time, but end up with some dependency on
the container.

I have several questions related to this design hurdle, but I will start with the one above.

Thanks!
Guru



_______________________________________________
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.