name: moonbit-refactoring
description: "Refactor MoonBit code to be idiomatic: shrink public APIs, convert functions to methods, use pattern matching with views, add loop invariants, and ensure test coverage without regressions. Use when updating MoonBit packages or refactoring MoonBit APIs, modules, or tests."

MoonBit Refactoring Skill

Intent

• Preserve behavior and public contracts unless explicitly changed.
• Minimize the public API to what callers require.
• Prefer declarative style and pattern matching over incidental mutation.
• Use view types (ArrayView/StringView/BytesView) to avoid copies.
• Add tests and docs alongside refactors.

Workflow

Start broad, then refine locally:

1. Architecture first: Review package structure, dependencies, and API boundaries.
2. Inventory public APIs and call sites (moon doc, moon ide find-references).
3. Pick one refactor theme (API minimization, package splits, pattern matching, loop style).
4. Apply the smallest safe change.
5. Update docs/tests in the same patch.
6. Run moon check, then moon test.
7. Use coverage to target missing branches.

Avoid local cleanups (renaming, pattern matching) until the high-level structure is sound.

Improve Package Architecture

• Keep packages focused: aim for <10k lines per package.
• Keep files manageable: aim for <2k lines per file.
• Keep functions focused: aim for <200 lines per function.

Splitting Files

Treat files in MoonBit as organizational units; move code freely within a package as long as each file stays focused on one concept.

Splitting Packages

When spinning off package A into A and B:

1. Create the new package and re-export temporarily:
   mbt // In package B using @A { ... } // re-export A's APIs
   Ensure moon check passes before proceeding.

2. Find and update all call sites:
   bash moon ide find-references <symbol>
   Replace bare f with @B.f.

3. Remove the use statement once all call sites are updated.

4. Audit and remove newly-unused pub APIs from both packages.

Guidelines

• Prefer acyclic dependencies: lower-level packages should not import higher-level ones.
• Only expose what downstream packages actually need.
• Consider an internal/ package for helpers that shouldn't leak.

Minimize Public API and Modularize

• Remove pub from helpers; keep only required exports.
• Move helpers into internal/ packages to block external imports.
• Split large files by feature; files do not define modules in MoonBit.

Local refactoring

Convert Free Functions to Methods + Chaining

• Move behavior onto the owning type for discoverability.
• Use .. for fluent, mutating chains when it reads clearly.

Example:
```mbt nocheck
// Before
fn reader_next(r : Reader) -> Char? { ... }
let ch = reader_next(r)

// After
fn Reader::next(self : Reader) -> Char? { ... }
let ch = r.next()

Example (chaining):
```mbt nocheck
buf..write_string("#\\")..write_char(ch)

Prefer Explicit Qualification

• Use @pkg.fn instead of using when clarity matters.
• Keep call sites explicit during wide refactors.

Example:
```mbt nocheck
let n = @parser.parse_number(token)

### Simplify Enum Constructors When Type Is Known

When the expected type is known from context, you can omit the full package path for enum constructors:

- **Pattern matching**: Annotate the matched value; constructors need no path.
- **Nested constructors**: Only the outermost needs the full path.
- **Return values**: The return type provides context for constructors in the body.
- **Collections**: Type-annotate the collection; elements inherit the type.

Examples:
```mbt
// Pattern matching - annotate the value being matched
let tree : @pkga.Tree = ...
match tree {
  Leaf(x) => x
  Node(left~, x, right~) => left.sum() + x + right.sum()
}

// Nested constructors - only outer needs full path
let x = @pkga.Tree::Node(left=Leaf(1), x=2, right=Leaf(3))

// Return type provides context
fn make_tree() -> @pkga.Tree {
  Node(left=Leaf(1), x=2, right=Leaf(3))
}

// Collections - type annotation on the array
let trees : Array[@pkga.Tree] = [Leaf(1), Node(left=Leaf(2), x=3, right=Leaf(4))]

Pattern Matching and Views

• Pattern match arrays directly; the compiler inserts ArrayView implicitly.
• Use .. in the middle to match prefix and suffix at once.
• Pattern match strings directly; avoid converting to Array[Char].
• String/StringView indexing yields UInt16 code units. Use for ch in s for Unicode-aware iteration.

we prefer pattern matching over small functions

For example,

 match gen_results.get(0) {
   Some(value) => Iter::singleton(value)
   None => Iter::empty()
 }

We can pattern match directly, it is more efficient and as readable:

 match gen_results {
   [value, ..] => Iter::singleton(value)
   [] => Iter::empty()
 }

MoonBit pattern matching is pretty expressive, here are some more examples:

match items {
  [] => ()
  [head, ..tail] => handle(head, tail)
  [..prefix, mid, ..suffix] => handle_mid(prefix, mid, suffix)
}

match s {
  "" => ()
  [.."let", ..rest] => handle_let(rest)
  _ => ()
}

Char literal matching

Use char literal overloading for Char, UInt16, and Int; the examples below rely on it. This is handy when matching String indexing results (UInt16) against a char range.

test {
  let a_int : Int = 'b'
  if (a_int is 'a'..<'z') { () } else { () }
  let a_u16 : UInt16 = 'b'
  if (a_u16 is 'a'..<'z') { () } else { () }
  let a_char : Char = 'b'
  if (a_char is 'a'..<'z') { () } else { () }
}

Use Nested Patterns and is

• Use is patterns inside if/guard to keep branches concise.

Example:

match token {
  Some(Ident([.."@", ..rest])) if process(rest) is Some(x) => handle_at(rest)
  Some(Ident(name)) => handle_ident(name)
  None => ()
}

Prefer Range Loops for Simple Indexing

• Use for i in start..<end { ... }, for i in start..<=end { ... }, for i in large>..small, or for i in large>=..small for simple index loops.
• Keep functional-state for loops for algorithms that update state.

Example:

// Before
for i = 0; i < len; {
  items.push(fill)
  continue i + 1
}

// After
for i in 0..<len {
  items.push(fill)
}

Loop Specs (Dafny-Style Comments)

• Add specs for functional-state loops.
• Skip invariants for simple for x in xs loops.
• Add TODO when a decreases clause is unclear (possible bug).

Example:

for i = 0, acc = 0; i < xs.length(); {
  acc = acc + xs[i]
  i = i + 1
} else { acc }
where {
  invariant: 0 <= i <= xs.length(),
  reasoning: (
    #| ... rigorous explanation ...
    #| ...
  )
}

Tests and Docs

• Prefer black-box tests in *_test.mbt or *.mbt.md.
• Add docstring tests with mbt check for public APIs.

Example:

///|
/// Return the last element of a non-empty array.
///
/// # Example
/// ```mbt check
/// test {
///   inspect(last([1, 2, 3]), content="3")
/// }
/// ```
pub fn last(xs : Array[Int]) -> Int { ... }

Coverage-Driven Refactors

• Use coverage to target missing branches through public APIs.
• Prefer small, focused tests over white-box checks.

Commands:

moon coverage analyze -- -f summary
moon coverage analyze -- -f caret -F path/to/file.mbt

Moon IDE Commands

moon doc "<query>"
moon ide outline <dir|file>
moon ide find-references <symbol>
moon ide peek-def <symbol>
moon ide rename <symbol> -new-name <new_name>
moon check
moon test
moon info

Use these commands for reliable refactoring.

Example: spinning off package_b from package_a.

Temporary import in package_b:

using @package_a { a, type B }

Steps:
1. Use moon ide find-references <symbol> to find all call sites of a and B.
2. Replace them with @package_a.a and @package_a.B.
3. Remove the using statement and run moon check.