majiayu000 FP Domain Modeling
Domain modeling with types. Make illegal states unrepresentable, workflows as pipelines, error handling at the type level.
Cross-references: fp-principles | fp-hexagonal-architecture | fp-algebra-driven-design
1. The Two Building Blocks
[STARTER]
All domain types compose from two operations:
- AND (Record Types): Value has ALL of these fields. Order requires CustomerInfo AND ShippingAddress AND OrderLines.
- OR (Choice Types): Value is ONE OF these alternatives. ProductCode is either WidgetCode OR GizmoCode.
Combined recursively, these express virtually any domain structure.
2. Domain Wrappers for Primitives
[STARTER]
Never use primitives directly in the domain model. Each domain concept gets its own wrapper type.
What: Wrap primitives so the compiler distinguishes CustomerId from OrderId. When: Every primitive with domain meaning. Why: Prevents accidental mixing (compiler rejects comparing CustomerId with OrderId). Each wrapper carries its own validation rules. The type name IS the documentation.
3. Validated Construction (Smart Constructors)
[STARTER]
Raw constructor is private. A create function validates input and returns a Result type, making validation failure explicit.
Pattern: UnitQuantity must be between 1 and 1000. Its create function rejects values outside that range. A companion value function provides read access to the inner primitive.
When: Every domain wrapper with validation rules. Why: Once constructed, a value is guaranteed valid. No defensive checks deeper in the code.
4. Making Illegal States Unrepresentable
[STARTER]
The central design guideline. Instead of flags and runtime checks, model the domain so invalid states cannot be constructed.
Replace Flags with Distinct Types
Instead of { EmailAddress; IsVerified: bool }, create separate VerifiedEmailAddress and UnverifiedEmailAddress types. Functions requiring verification take VerifiedEmailAddress, making misuse a compile error.
Replace Optional Fields with Choice Types
Instead of { Email: option; Address: option } (where both could be None), create: EmailOnly | AddressOnly | EmailAndAddress. The "at least one required" rule becomes structurally enforced.
NonEmptyList for "At Least One" Rules
Define a type guaranteeing at least one element. Order with OrderLines: NonEmptyList<OrderLine> cannot have zero lines.
5. Workflows as Functions
[STARTER]
Every business workflow is a single function: Command in, Events out.
PlaceOrderWorkflow : PlaceOrderCommand -> AsyncResult<PlaceOrderEvent list>
Pipeline Composition
Workflows decompose into steps, each transforming one document type into the next:
UnvalidatedOrder -> ValidatedOrder -> PricedOrder -> Events
Each step is stateless, pure, has single input/output type, and is independently testable. The workflow assembles by piping steps together.
Why: Pipeline makes the business process visible. Each step name is a domain concept.
6. Document Lifecycle as State Types
[INTERMEDIATE]
Rather than one Order type with flags, create separate types for each lifecycle stage:
UnvalidatedOrder(raw input, all fields strings)ValidatedOrder(all fields checked)PricedOrder(prices calculated)
A top-level Order choice type unifies all states. New states (e.g., Refunded) added without breaking existing code.
When: Domain entities with distinct lifecycle stages where different data is available at each stage.
7. State Machines with Types
[INTERMEDIATE]
When an entity has distinct states with different data and different allowed operations, model each state as a separate type.
ShoppingCart = EmptyCart | ActiveCart of ActiveCartData | PaidCart of PaidCartData
Transition functions take the choice type, pattern-match on current state, return new state.
Benefits: All states explicit | each state has own data | invalid transitions prevented by types | pattern matching warnings reveal unhandled edge cases.
8. Error-Track Pipelines (Railway Pattern)
[INTERMEDIATE]
Each function returns a Result type. Pipeline short-circuits on first failure.
rawInput
|> validateOrder -- Result<ValidOrder, Error>
|> bind calculateTotal -- Result<PricedOrder, Error>
|> bind checkInventory -- Result<ConfirmedOrder, Error>
|> bind chargePayment -- Result<PaidOrder, Error>
|> map generateReceipt -- Result<Receipt, Error>
Key combinators:
- map: Transform success value (one-track into two-track)
- bind: Chain a function that itself returns Result
- mapError: Transform error value
- tee: Perform side effect without changing value (logging)
Error Classification
| Category | Examples | Strategy |
|---|---|---|
| Domain Errors | Validation failure, out of stock | Model as types, return via Result |
| Panics | Out of memory, null reference | Throw exceptions, catch at top level |
| Infrastructure Errors | Network timeout, auth failure | Case-by-case |
Unifying Error Types
Each step may have its own error type. Define a common error choice type and use mapError to lift step errors before composing.
Collecting All Errors (Applicative Validation)
[ADVANCED]
Standard bind short-circuits on first error. For validation where you want ALL errors, use Applicative style -- runs all validations and accumulates errors into a list. See fp-principles section 5.
When: Form validation | batch input checking | any place user needs all errors at once.
9. Modeling Dependencies
[INTERMEDIATE]
Each workflow step declares exactly the functions it needs as parameters:
CheckProductCodeExists : ProductCode -> bool
GetProductPrice : ProductCode -> Price
Convention: Dependencies first in parameter list, primary input last. Enables partial application (functional DI).
Public vs. internal: Top-level workflow hides dependencies from callers. Internal steps make them explicit.
See fp-hexagonal-architecture section 5 for full DI decision tree.
10. Persistence Ignorance
[INTERMEDIATE]
Domain model has no awareness of databases. Two distinct type hierarchies:
- Domain types: Rich, nested, use choice types, validated wrappers. Not serialization-friendly.
- DTO types: Flat, primitives, nullable, arrays. Designed for serialization.
Conversion Pattern
- fromDomain: Domain -> DTO. Always succeeds.
- toDomain: DTO -> Result<Domain, Error>. May fail (validation happens here).
Serialization as Pipeline Steps
JSON -> deserialize -> DTO -> toDomain -> [WORKFLOW] -> fromDomain -> DTO -> serialize -> JSON
Domain workflow never sees JSON or DTOs directly.
11. Bounded Contexts
[ADVANCED]
Each context has its own dialect of domain language. Contexts communicate only through events and DTOs. Design for autonomy.
Trust boundaries: Input gate validates and converts incoming DTOs to domain types. Output gate converts domain types to DTOs, deliberately dropping private information.
Inter-context relationships: Shared Kernel (shared design) | Customer/Supplier (downstream defines contract) | Anti-Corruption Layer (translator preventing external model from corrupting internal domain).
Decision Tree: How to Model This Domain Concept?
Is it a simple value with validation rules?
YES --> Domain Wrapper + Smart Constructor
NO -->
Does it represent one of several alternatives?
YES --> Choice Type (sum type)
NO -->
Does it group several values together?
YES --> Record Type (product type)
NO -->
Does it have distinct lifecycle stages?
YES --> State Machine with Types (section 7)
NO -->
Is it an operation that transforms data through stages?
YES --> Workflow Pipeline (section 5)
NO --> Evaluate whether it needs modeling at all
12. Key Design Heuristics
- Start from events and workflows, not data structures
- Let the domain expert drive naming -- their vocabulary, not technical jargon
- Use types to enforce business rules at compile time -- every rule in the type system needs no unit test
- Document effects in signatures -- Result for errors | Async for I/O | Option for missing data
- Separate domain types from serialization types -- domain is pure; serialization is infrastructure
- Prefer explicit over implicit -- every input and dependency is a function parameter
Naming Patterns
- Types as nouns: Order | ProductCode | CustomerInfo
- Workflows as verbs: ValidateOrder | PriceOrder | PlaceOrder
- Events in past tense: OrderPlaced | OrderShipped
- Commands in imperative: PlaceOrder | ShipOrder | CancelOrder
- Lifecycle prefixes: UnvalidatedOrder | ValidatedOrder | PricedOrder