Convert Clojure Roc

Clojure ↔ Roc Conversion

Bidirectional conversion between Clojure and Roc. This skill extends meta-convert-dev with Clojure↔Roc specific type mappings, idiom translations, and tooling for translating from dynamically-typed REPL-driven development to statically-typed platform-based architecture.

This Skill Extends

• meta-convert-dev - Foundational conversion patterns (APTV workflow, testing strategies)

For general concepts like the Analyze → Plan → Transform → Validate workflow, testing strategies, and common pitfalls, see the meta-skill first.

This Skill Adds

• Type mappings: Clojure's dynamic types → Roc's static types
• Idiom translations: REPL-driven patterns → compile-time verified code
• Error handling: Exception-based → Result type with pattern matching
• Concurrency patterns: Atoms/refs/agents → platform-managed tasks
• Platform architecture: JVM-based → platform/application separation
• Paradigm shift: Dynamic functional → static functional with structural types

This Skill Does NOT Cover

• General conversion methodology - see meta-convert-dev
• Clojure language fundamentals - see lang-clojure-dev
• Roc language fundamentals - see lang-roc-dev
• ClojureScript frontend patterns - focus is on Clojure backend to Roc applications

Quick Reference

When Converting Code

1. Analyze source thoroughly before writing target - understand dynamic behavior
2. Map types first - convert dynamic runtime checks to static types
3. Preserve semantics over syntax similarity - embrace Roc's type system
4. Adopt target idioms - don't write "Clojure code in Roc syntax"
5. Handle edge cases - nil → explicit Maybe, exceptions → Result
6. Leverage platform model - separate pure logic from I/O
7. Test equivalence - same inputs → same outputs
8. Add compile-time validation - use Roc's type system to replace runtime checks

Type System Mapping

Primitive Types

Important numeric differences:

• Clojure: Arbitrary precision integers (BigInt) available, automatic promotion
• Roc: Fixed-size integers, explicit overflow behavior (wrapping vs saturating vs checking)
• Clojure: (/ 1 3) returns exact ratio; (/ 1.0 3) returns float
• Roc: Must choose F64 or Dec (decimal) for division

Collection Types

Key difference:

• Clojure: Maps are the primary composite data structure, keys can be anything
• Roc: Records are typed with known fields at compile time; Dict is for dynamic keys

Composite Types

Function Types

Idiom Translation

Pattern: nil Handling → Tag Unions

Clojure uses nil idiomatically. Roc requires explicit handling through tag unions.

Clojure:

(defn find-user [id]
  (when (= id 1)
    {:name "Alice" :age 30}))

(defn display-name [user]
  (if user
    (:name user)
    "Anonymous"))

;; Using some->
(def name
  (some-> (find-user 1)
          :name
          (or "Anonymous")))

Roc:

# Explicit Maybe pattern
findUser : I64 -> [Some { name : Str, age : U32 }, None]
findUser = \id ->
    if id == 1 then
        Some({ name: "Alice", age: 30 })
    else
        None

displayName : [Some { name : Str, age : U32 }, None] -> Str
displayName = \maybeUser ->
    when maybeUser is
        Some(user) -> user.name
        None -> "Anonymous"

# Using Result for error context
name =
    when findUser(1) is
        Some(user) -> user.name
        None -> "Anonymous"

Why this translation:

• Roc eliminates null pointer errors at compile time
• Pattern matching on tag unions is exhaustive (compiler checks all cases)
• More explicit but prevents entire classes of runtime errors
• [Some a, None] or Result a err replace nil idioms

Pattern: Keywords and Maps → Records and Tags

Clojure uses keywords and maps for both data and discriminated unions. Roc uses records for data and tags for unions.

Clojure:

;; Data with keyword keys
(def user {:name "Alice" :email "alice@example.com" :age 30})

;; Accessing fields
(:name user) ; => "Alice"
(get user :name) ; => "Alice"

;; Discriminated unions with keywords
(defn handle-message [msg]
  (case (:type msg)
    :increment (update-in model [:count] inc)
    :decrement (update-in model [:count] dec)
    :set-count (assoc model :count (:value msg))))

;; Usage
(handle-message {:type :increment})
(handle-message {:type :set-count :value 42})

Roc:

# Records for data (compile-time known fields)
user = { name: "Alice", email: "alice@example.com", age: 30 }

# Accessing fields
user.name # "Alice"

# Type alias for clarity
User : { name : Str, email : Str, age : U32 }

# Discriminated unions with tags
Message : [
    Increment,
    Decrement,
    SetCount(I64),
]

handleMessage : Message, Model -> Model
handleMessage = \msg, model ->
    when msg is
        Increment -> { model & count: model.count + 1 }
        Decrement -> { model & count: model.count - 1 }
        SetCount(value) -> { model & count: value }

# Usage
handleMessage(Increment, model)
handleMessage(SetCount(42), model)

Why this translation:

• Records provide compile-time field checking (no typos in field names)
• Tags are lightweight and type-safe for discriminated unions
• Pattern matching ensures all message types are handled
• More rigid but catches errors at compile time instead of runtime

Pattern: Dynamic Sequences → Typed Lists

Clojure's sequences are lazily evaluated and dynamically typed. Roc's lists are strictly typed.

Clojure:

;; Lazy sequence operations
(defn process-items [items]
  (->> items
       (map #(* % 2))
       (filter even?)
       (take 10)
       (reduce + 0)))

;; Infinite sequences
(def naturals (iterate inc 0))
(take 5 naturals) ; => (0 1 2 3 4)

;; Mixed type handling (not recommended but possible)
(map str [1 :two "three"]) ; => ("1" ":two" "three")

Roc:

# Strict typed list operations
processItems : List I64 -> I64
processItems = \items ->
    items
    |> List.map(\x -> x * 2)
    |> List.keepIf(\x -> x % 2 == 0)
    |> List.takeFirst(10)
    |> List.sum

# No lazy evaluation - compute eagerly
# For large data, use platform streaming

# Type safety - all elements must be same type
# This won't compile:
# mixedList = [1, "two", :three] # Error!

# Must use tag unions for heterogeneous data
Item : [Num I64, Text Str, Symbol Str]
mixedList : List Item
mixedList = [Num(1), Text("two"), Symbol("three")]

Why this translation:

• Roc evaluates strictly, avoiding lazy evaluation pitfalls
• Type safety prevents runtime type errors
• Explicit tag unions for heterogeneous data
• Performance is more predictable (no hidden thunks)

Pattern: Exception Handling → Result Type

Clojure uses exceptions for error handling (Java interop). Roc uses the Result type.

Clojure:

(defn parse-age [s]
  (try
    (let [age (Long/parseLong s)]
      (cond
        (neg? age) (throw (ex-info "Age must be non-negative" {:age age}))
        (>= age 150) (throw (ex-info "Age must be less than 150" {:age age}))
        :else age))
    (catch NumberFormatException e
      (throw (ex-info "Not a valid number" {:input s})))))

;; Calling code
(try
  (let [age (parse-age input)]
    (str "Age: " age))
  (catch Exception e
    (str "Error: " (.getMessage e))))

Roc:

# Result type for errors
parseAge : Str -> Result U32 Str
parseAge = \s ->
    when Str.toU32(s) is
        Ok(age) ->
            if age < 0 then
                Err("Age must be non-negative")
            else if age >= 150 then
                Err("Age must be less than 150")
            else
                Ok(age)
        Err(_) ->
            Err("Not a valid number")

# Calling code with pattern matching
result = parseAge(input)
message = when result is
    Ok(age) -> "Age: \(Num.toStr(age))"
    Err(error) -> "Error: \(error)"

# Chaining Results with try
validateAndDouble : Str -> Result U32 Str
validateAndDouble = \s ->
    age = try parseAge(s)
    if age > 50 then
        Ok(age * 2)
    else
        Err("Age must be greater than 50")

Why this translation:

• Result type makes errors explicit in function signatures
• Compiler enforces error handling (can't ignore Err case)
• No hidden control flow (exceptions can jump anywhere)
• try keyword unwraps Result or early-returns Err
• More verbose but impossible to forget error handling

Pattern: Atoms and State → Platform State Management

Clojure uses atoms for shared mutable state. Roc pushes state to the platform layer.

Clojure:

;; Atom for application state
(def app-state (atom {:count 0 :users #{}}))

;; Pure update function
(defn increment-count [state]
  (update state :count inc))

;; Apply update
(swap! app-state increment-count)

;; Read state
@app-state

;; Watch state changes
(add-watch app-state :logger
  (fn [key atom old-state new-state]
    (println "State changed:" old-state "->" new-state)))

Roc:

# Application code is pure - no mutable state
# State is managed by the platform

# Define state type
Model : { count : I64, users : Set Str }

# Pure update functions
incrementCount : Model -> Model
incrementCount = \model ->
    { model & count: model.count + 1 }

# Platform integration (example with basic-cli)
app =
    { init: { count: 0, users: Set.empty() }
    , update: update
    , subscriptions: subscriptions
    }

update : Msg, Model -> Model
update = \msg, model ->
    when msg is
        Increment -> incrementCount(model)
        # ...other messages

# Platform handles state persistence, not application code

Why this translation:

• Roc applications are pure functions of (state, message) → new state
• Platform layer handles actual state mutation and effects
• Clearer separation between pure logic and side effects
• Makes testing trivial (just test pure functions)
• State changes are tracked by platform, not manual watching

Paradigm Translation

Mental Model Shift: Dynamic REPL-Driven → Static Platform-Based

Architectural Shift: JVM Application → Platform/Application

Error Handling

Clojure Error Model → Roc Error Model

Clojure's approach:

• Exceptions (from Java) for error conditions
• try/catch/finally blocks
• Custom exception types with ex-info
• Error data attached to exceptions

Roc's approach:

• Result a err type for operations that can fail
• Pattern matching on Ok and Err variants
• try keyword for early returns from Result
• Errors are just values (typically strings or custom tags)

Common Error Patterns

Example: Error Propagation

Clojure:

(defn process-user [user-id]
  (try
    (let [user (fetch-user user-id)
          validated (validate-user user)
          updated (update-permissions validated)]
      {:success updated})
    (catch Exception e
      {:error (.getMessage e)})))

Roc:

processUser : I64 -> Result User Str
processUser = \userId ->
    user = try fetchUser(userId)
    validated = try validateUser(user)
    updated = try updatePermissions(validated)
    Ok(updated)

# The 'try' keyword automatically propagates Err
# If any step returns Err, the whole function returns that Err

Concurrency Patterns

Clojure Concurrency → Roc Platform Tasks

Clojure's approach:

• Atoms, Refs, Agents for coordinated state
• core.async for CSP-style channels
• Java threads and futures
• STM (Software Transactional Memory) with refs

Roc's approach:

• Platform provides Task type for async operations
• Tasks are descriptions of work (not running computations)
• Platform handles scheduling and execution
• No shared mutable state in application code

Task Model Translation

Example: HTTP Request

Clojure:

(require '[clj-http.client :as http])

(defn fetch-data [url]
  (try
    (let [response (http/get url {:as :json})]
      (:body response))
    (catch Exception e
      (println "Error:" (.getMessage e))
      nil)))

;; Async version
(require '[clojure.core.async :refer [go <!]])

(defn fetch-data-async [url]
  (go
    (try
      (let [response (<! (http/get url {:as :json :async? true}))]
        (:body response))
      (catch Exception e
        (println "Error:" e)
        nil))))

Roc:

# Tasks are values describing work to be done
fetchData : Str -> Task (List User) [HttpErr Str]
fetchData = \url ->
    response = try Http.get(url) |> Task.mapErr(\_ -> HttpErr("Request failed"))
    bytes = response.body
    decoded = try Decode.fromBytes(bytes) |> Task.mapErr(\_ -> HttpErr("Decode failed"))
    Ok(decoded)

# Platform executes tasks, not application code
# Composing tasks
processUsers : Task (List Str) [HttpErr Str]
processUsers =
    users = try fetchData("https://api.example.com/users")
    names = List.map(users, \u -> u.name)
    Task.ok(names)

Why this translation:

• Tasks are descriptions, not running code (referentially transparent)
• Platform handles actual I/O execution
• Error handling is explicit in the type signature
• Easier to test (tasks are just data until executed by platform)

Platform Architecture

JVM Process → Platform/Application Separation

Key conceptual shift:

In Clojure, your application is a JVM process that directly performs I/O. In Roc, your application is a pure module that describes transformations, and the platform handles I/O.

Clojure:                    Roc:
┌─────────────────┐         ┌─────────────────┐
│   Application   │         │   Application   │
│   (impure)      │         │   (pure Roc)    │
│  ┌───────────┐  │         │                 │
│  │ Database  │  │         │  Pure functions │
│  │ HTTP      │  │         │  Type defs      │
│  │ File I/O  │  │         │  Logic only     │
│  └───────────┘  │         └────────┬────────┘
└─────────────────┘                  │ Pure interface
      JVM                            ▼
                              ┌─────────────────┐
                              │    Platform     │
                              │  (Roc + host)   │
                              │  ┌───────────┐  │
                              │  │ Database  │  │
                              │  │ HTTP      │  │
                              │  │ File I/O  │  │
                              │  └───────────┘  │
                              └─────────────────┘

Platform Selection

When converting a Clojure application, choose the appropriate Roc platform:

Common Pitfalls

1. Trying to use nil directly: Roc has no null. Use tag unions like [Some a, None] or Result a err.

   • Bad: Expecting nil to work as in Clojure
   • Good: when maybeValue is Some(v) -> ... None -> ...

2. Using maps for everything: In Roc, use records when fields are known at compile time.

   • Bad: { "dynamicKey": value } (dynamic string keys)
   • Good: { knownField: value } (compile-time known) or Dict.fromList([("key", value)]) for truly dynamic keys

3. Forgetting to handle errors: Result forces you to handle both Ok and Err.

   • Bad: Only pattern matching on Ok case
   • Good: when result is Ok(v) -> ... Err(e) -> ... (exhaustive)

4. Mixing pure logic with effects: Keep application code pure.

   • Bad: Calling I/O functions directly in application logic
   • Good: Return Task values, let platform execute them

5. Over-using type annotations: Roc infers most types.

   • Bad: Annotating every single function when types are obvious
   • Good: Annotate public API boundaries and complex functions only

6. Expecting lazy evaluation: Roc evaluates strictly.

   • Bad: Assuming List.map won't compute until needed
   • Good: Use platform streaming for large data

7. Not leveraging pattern matching: Use when exhaustively instead of if/else chains.

   • Bad: if x == A then ... else if x == B then ...
   • Good: when x is A -> ... B -> ... C -> ...

8. Ignoring numeric overflow: Roc has explicit overflow behavior.

   • Bad: Assuming automatic BigInt promotion like Clojure
   • Good: Choose appropriate integer size (I64, I32) and handle overflow explicitly

Limitations

Coverage Gaps

Combined Score: 14/17 (Good with noted gaps)

Gaps:

• Roc serialization patterns not fully covered in lang-roc-dev
• Roc build tooling not covered in lang-roc-dev
• REPL workflow differences require paradigm shift

Mitigation:

• Reference patterns-serialization-dev for encoding/decoding patterns
• Consult official Roc platform documentation for build process
• This skill documents REPL → compile-time workflow shift

Known Limitations

1. Serialization: This skill has limited guidance on Roc's serialization patterns (Encode/Decode) because lang-roc-dev lacks comprehensive serialization coverage. For production serialization, consult patterns-serialization-dev and Roc platform documentation.

2. Build tooling: Conversion patterns for build scripts (Leiningen/deps.edn → Roc platform config) may be incomplete. Refer to specific platform documentation.

3. Advanced macros: Some complex Clojure macros may not have direct Roc equivalents. Consider whether the macro is solving a problem that Roc's type system already addresses.

External Resources Used

Tooling

No direct Clojure → Roc transpiler exists. Conversion is manual but type-guided.

Examples

Example 1: Simple - Function with Optional Return

Before (Clojure):

(defn find-first-even [numbers]
  (first (filter even? numbers)))

;; Usage
(find-first-even [1 3 5 6 7]) ; => 6
(find-first-even [1 3 5])     ; => nil

After (Roc):

# Explicit Maybe return type
findFirstEven : List I64 -> [Some I64, None]
findFirstEven = \numbers ->
    when List.findFirst(numbers, \n -> n % 2 == 0) is
        Ok(n) -> Some(n)
        Err(_) -> None

# Usage
result1 = findFirstEven([1, 3, 5, 6, 7]) # Some(6)
result2 = findFirstEven([1, 3, 5])       # None

# Pattern match on result
message = when result1 is
    Some(n) -> "Found: \(Num.toStr(n))"
    None -> "No even number found"

Key changes:

• nil → explicit [Some a, None] tag union
• Return type documents possibility of no result
• Compiler enforces handling both cases

Example 2: Medium - Error Handling and Validation

Before (Clojure):

(require '[clojure.spec.alpha :as s])

(s/def ::email (s/and string? #(re-matches #".+@.+\..+" %)))
(s/def ::age (s/and int? #(< 0 % 150)))
(s/def ::user (s/keys :req-un [::email ::age]))

(defn create-user [email age]
  (let [user {:email email :age age}]
    (if (s/valid? ::user user)
      {:ok user}
      {:error (s/explain-str ::user user)})))

;; Usage
(create-user "alice@example.com" 30)
;; => {:ok {:email "alice@example.com", :age 30}}

(create-user "invalid" 200)
;; => {:error "Spec validation failed..."}

After (Roc):

# Type-safe user record
User : { email : Str, age : U32 }

# Validation errors as tag union
ValidationErr : [InvalidEmail, AgeOutOfRange]

# Validation functions
validateEmail : Str -> Result Str ValidationErr
validateEmail = \email ->
    if Str.contains(email, "@") && Str.contains(email, ".") then
        Ok(email)
    else
        Err(InvalidEmail)

validateAge : U32 -> Result U32 ValidationErr
validateAge = \age ->
    if age > 0 && age < 150 then
        Ok(age)
    else
        Err(AgeOutOfRange)

# Create user with validation
createUser : Str, U32 -> Result User ValidationErr
createUser = \email, age ->
    validEmail = try validateEmail(email)
    validAge = try validateAge(age)
    Ok({ email: validEmail, age: validAge })

# Usage
result1 = createUser("alice@example.com", 30)
# Ok({ email: "alice@example.com", age: 30 })

result2 = createUser("invalid", 200)
# Err(InvalidEmail)

# Pattern match on result
message = when result1 is
    Ok(user) -> "Created user: \(user.email)"
    Err(InvalidEmail) -> "Invalid email format"
    Err(AgeOutOfRange) -> "Age must be between 0 and 150"

Key changes:

• Runtime spec validation → compile-time types + Result validation
• Error strings → typed error variants
• try keyword chains validations, short-circuits on first error
• Pattern matching provides exhaustive error handling

Example 3: Complex - State Management and Updates

Before (Clojure):

(defrecord TodoItem [id text completed])

(def app-state
  (atom {:todos []
         :next-id 0
         :filter :all}))

(defn add-todo [state text]
  (let [id (:next-id state)
        todo (->TodoItem id text false)]
    (-> state
        (update :todos conj todo)
        (update :next-id inc))))

(defn toggle-todo [state id]
  (update state :todos
    (fn [todos]
      (mapv #(if (= (:id %) id)
              (update % :completed not)
              %)
            todos))))

(defn set-filter [state filter]
  (assoc state :filter filter))

(defn visible-todos [state]
  (let [todos (:todos state)
        filter (:filter state)]
    (case filter
      :all todos
      :active (filterv (complement :completed) todos)
      :completed (filterv :completed todos))))

;; Usage with atom
(swap! app-state add-todo "Learn Roc")
(swap! app-state toggle-todo 0)
(swap! app-state set-filter :completed)
(visible-todos @app-state)

After (Roc):

# Types
TodoItem : { id : U64, text : Str, completed : Bool }

Filter : [All, Active, Completed]

Model : {
    todos : List TodoItem,
    nextId : U64,
    filter : Filter,
}

# Messages for updates
Msg : [
    AddTodo Str,
    ToggleTodo U64,
    SetFilter Filter,
]

# Pure update functions
addTodo : Model, Str -> Model
addTodo = \model, text ->
    newTodo = { id: model.nextId, text: text, completed: Bool.false }
    { model &
        todos: List.append(model.todos, newTodo),
        nextId: model.nextId + 1,
    }

toggleTodo : Model, U64 -> Model
toggleTodo = \model, id ->
    updatedTodos = List.map(model.todos, \todo ->
        if todo.id == id then
            { todo & completed: !todo.completed }
        else
            todo
    )
    { model & todos: updatedTodos }

setFilter : Model, Filter -> Model
setFilter = \model, filter ->
    { model & filter: filter }

# Query function
visibleTodos : Model -> List TodoItem
visibleTodos = \model ->
    when model.filter is
        All -> model.todos
        Active -> List.keepIf(model.todos, \t -> !t.completed)
        Completed -> List.keepIf(model.todos, \t -> t.completed)

# Main update dispatcher
update : Msg, Model -> Model
update = \msg, model ->
    when msg is
        AddTodo(text) -> addTodo(model, text)
        ToggleTodo(id) -> toggleTodo(model, id)
        SetFilter(filter) -> setFilter(model, filter)

# Initial model
init : Model
init = {
    todos: [],
    nextId: 0,
    filter: All,
}

# Usage (in platform context, not shown)
# model1 = update(AddTodo("Learn Roc"), init)
# model2 = update(ToggleTodo(0), model1)
# model3 = update(SetFilter(Completed), model2)
# visible = visibleTodos(model3)

Key changes:

• Atom with mutable state → immutable Model type
• Keywords for actions → typed Msg tag union
• swap! updates → pure update function
• State managed by platform, not application
• All updates are pure functions: (Msg, Model) → Model
• Easier to test (no atoms, just pure functions)
• Type system prevents invalid messages or state shapes

See Also

For more examples and patterns, see:

• meta-convert-dev - Foundational patterns with cross-language examples
• lang-clojure-dev - Clojure development patterns (REPL, macros, sequences)
• lang-roc-dev - Roc development patterns (platform model, records, abilities)
• convert-elm-roc - Similar functional to functional conversion (Elm → Roc)
• convert-haskell-roc - Another ML-family language conversion

Cross-cutting pattern skills (for areas not fully covered by lang-*-dev):

• patterns-serialization-dev - Encode/Decode patterns across languages
• patterns-concurrency-dev - Async, tasks, channels across languages