FP in Haskell -- Functional Software Crafter Skill

Cross-references: fp-principles | fp-domain-modeling | pbt-haskell

When to Choose Haskell

• Best for: correctness-critical systems | compiler-enforced purity | maximum type safety | financial systems
• Not ideal for: teams needing fast onboarding | rapid prototyping | .NET/JVM platform requirements

[STARTER] Quick Setup

# Install GHCup (manages GHC, cabal, stack, HLS)
curl --proto '=https' --tlsv1.2 -sSf https://get-ghcup.haskell.org | sh

# Create project
mkdir order-service && cd order-service && cabal init --interactive
# Or: stack new order-service simple && stack build && stack test

Test runner: cabal test or stack test. Add hspec, QuickCheck, hedgehog to build-depends.

[STARTER] Type System for Domain Modeling

Choice Types (Sum Types)

data PaymentMethod
  = CreditCard CardNumber ExpiryDate
  | BankTransfer AccountNumber
  | Cash
  deriving (Eq, Show)

Record Types and Newtypes

data Customer = Customer
  { customerId    :: CustomerId
  , customerName  :: CustomerName
  , customerEmail :: EmailAddress
  } deriving (Eq, Show)

newtype OrderId = OrderId Int deriving (Eq, Ord, Show)
newtype EmailAddress = EmailAddress Text deriving (Eq, Show)

newtype is erased at compile time -- zero runtime overhead, full type safety.

[STARTER] Validated Construction (Smart Constructors)

module Domain.Email (EmailAddress, mkEmailAddress, emailToText) where

import Data.Text (Text)
import qualified Data.Text as T

newtype EmailAddress = EmailAddress Text deriving (Eq, Show)

mkEmailAddress :: Text -> Either ValidationError EmailAddress
mkEmailAddress raw
  | "@" `T.isInfixOf` raw = Right (EmailAddress raw)
  | otherwise              = Left (InvalidEmail raw)

Export the type but not the constructor. Only mkEmailAddress can create values.

[INTERMEDIATE] Composition Style

Function Composition (Right-to-Left)

-- (.) composes right-to-left
processOrder :: RawOrder -> Either OrderError Confirmation
processOrder = confirmOrder . priceOrder . validateOrder

Monadic Chaining with do-notation

placeOrder :: RawOrder -> Either OrderError Confirmation
placeOrder raw = do
  validated <- validateOrder raw
  priced    <- priceOrder validated
  confirmOrder priced

Applicative for Independent Validation

mkCustomer :: Text -> Text -> Either ValidationError Customer
mkCustomer rawName rawEmail =
  Customer
    <$> mkCustomerId 0
    <*> mkCustomerName rawName
    <*> mkEmailAddress rawEmail

Error-Accumulating Validation

import Data.Validation (Validation, failure, success)

mkCustomerV :: Text -> Text -> Validation [ValidationError] Customer
mkCustomerV rawName rawEmail =
  Customer
    <$> validateName rawName    -- Validation [ValidationError] CustomerName
    <*> validateEmail rawEmail  -- all errors collected, not short-circuited

Unlike Either which stops at first error, Validation accumulates all failures via its Applicative instance.

[INTERMEDIATE] Effect Management

Haskell enforces purity at the compiler level. IO in return type means side effects.

calculateTotal :: Order -> Money           -- Pure: compiler guarantees no side effects
calculateTotal order = sumOf (orderLines order)

saveOrder :: Order -> IO ()                -- Impure: IO in the type
saveOrder order = writeToDatabase order
-- calculateTotal CANNOT call saveOrder -- compiler error

[ADVANCED] Three Layers Pattern (Hexagonal Architecture)

-- Layer 1: Pure domain (no IO, no effects)
module Domain.Order (calculateDiscount, validateOrder) where
calculateDiscount :: Order -> Discount
calculateDiscount order
  | totalLines order > 10 = Discount 0.1
  | otherwise              = Discount 0.0

-- Layer 2: Effect interfaces (type classes as ports)
class Monad m => OrderRepo m where
  findOrder :: OrderId -> m (Maybe Order)
  saveOrder :: Order -> m ()

-- Layer 3: IO implementations (adapters)
instance OrderRepo IO where
  findOrder orderId = queryDatabase orderId
  saveOrder order   = insertDatabase order

Effect libraries: Effectful (recommended starting point, best performance) | mtl (existing codebases) | Polysemy (algebraic effect semantics).

[INTERMEDIATE] Testing

Frameworks: QuickCheck (original PBT) | Hedgehog (integrated shrinking) | Hspec (BDD) | tasty (composable test tree). See pbt-haskell for detailed PBT patterns.

Property Test Example

import Test.Hspec
import Test.QuickCheck

spec :: Spec
spec = describe "validateOrder" $ do
  it "round-trips through serialization" $
    property $ \order ->
      deserializeOrder (serializeOrder order) === Right order

  it "validated orders always have positive totals" $
    property $ \rawOrder ->
      case validateOrder rawOrder of
        Left _      -> discard
        Right valid -> orderTotal valid > Money 0

Custom Generator

import Data.Text (pack)
import Test.QuickCheck

genValidEmail :: Gen EmailAddress
genValidEmail = do
  user   <- listOf1 (elements ['a'..'z'])
  domain <- listOf1 (elements ['a'..'z'])
  pure (EmailAddress (pack (user ++ "@" ++ domain ++ ".com")))

[ADVANCED] Idiomatic Patterns

GADTs for State Machines

{-# LANGUAGE GADTs, DataKinds #-}

data OrderState = Unvalidated | Validated | Priced

data Order (s :: OrderState) where
  UnvalidatedOrder :: RawData -> Order 'Unvalidated
  ValidatedOrder   :: ValidData -> Order 'Validated
  PricedOrder      :: PricedData -> Order 'Priced

-- Type-safe transitions: only validated orders can be priced
priceOrder :: Order 'Validated -> Either PricingError (Order 'Priced)
priceOrder (ValidatedOrder d) = Right (PricedOrder (addPricing d))

Lazy Evaluation for Decoupled Pipelines

eligibleOrders :: [Order] -> [Order]
eligibleOrders = take 10 . filter isEligible . sortBy orderDate

Maturity and Adoption

• Steep learning curve: Monads, type classes, category-theory vocabulary create significant onboarding barrier. Budget extra ramp-up time.
• GHC extensions confusion: Over 100 language extensions; knowing which to enable requires experience. Start with GHC2021 defaults.
• Space leaks from laziness: Default lazy evaluation causes subtle memory issues. Requires profiling discipline and strict annotations.
• Smaller talent pool: Hiring Haskell developers harder than mainstream languages. Consider team sustainability before committing.

Common Pitfalls

1. Lazy space leaks: Use foldl' (strict) instead of foldl. Use BangPatterns for strict accumulators.
2. String vs Text: Never use String ([Char]) for real data. Use Data.Text / Data.ByteString.
3. Orphan instances: Define instances in the module of the type or class. Use newtype wrappers otherwise.
4. Over-abstracting with type-level programming: GADTs and type families increase compile times and error complexity. Use for genuine safety gains.