Pbt Workflow Guide

PBT Workflow Guide

Overview

This skill provides a workflow for introducing, updating, and improving Property-Based Testing (PBT) in MoonBit repositories using moonbitlang/quickcheck. Use this guide to:

• Select appropriate PBT patterns based on function characteristics
• Design effective generators with proper distribution control
• Implement custom Shrink for complex types
• Write meaningful property-based tests

Installation

moon add moonbitlang/quickcheck
moon install

Add to your moon.pkg.json:

{
  "import": [{ "path": "moonbitlang/quickcheck", "alias": "qc" }]
}

Quick Start

Writing Your First Property

A property is a function that should hold for all inputs. Use quick_check_fn to test it:

test "reverse is involutive" {
  @qc.quick_check_fn!(fn(arr : Array[Int]) {
    arr.copy().reverse().reverse() == arr
  })
}

Output on success:

+++ [100/0/100] Ok, passed!

When Properties Fail

QuickCheck finds counterexamples and shrinks them to minimal failing cases:

*** [8/0/100] Failed! Falsified.
(0, [0, 0])

Core Traits

Testable

Types that can be tested. Bool implements Testable, so most properties return Bool.

Arbitrary

Types that can generate random values. Derive automatically or implement manually:

enum Nat {
  Zero
  Succ(Nat)
} derive(Arbitrary, Show)

Shrink

Types that can be simplified to find minimal counterexamples:

pub impl Shrink for MyType with shrink(self) -> Iter[MyType] {
  let mut shrunk : Array[MyType] = []
  shrunk.push(MyType::default())  // Try simplest first
  for field_shrunk in self.field.shrink() {
    shrunk.push(MyType::new(field_shrunk, self.other_field))
  }
  shrunk.iter()
}

Pattern Decision Tree

Use this flow to select the appropriate pattern:

Q1: What type of function?
│
├─ Transformation (A -> B)
│   └─ Q2: Inverse exists?
│       ├─ YES → Round-Trip
│       └─ NO → Q3: Measurable output properties?
│           ├─ YES → Invariant
│           └─ NO → Q4: Reference implementation exists?
│               ├─ YES → Oracle
│               └─ NO → Producer-Consumer or unit tests
│
├─ Normalization (A -> A)
│   └─ Idempotent
│
└─ Stateful system
    └─ State Machine

Pattern Summary

Generator Design

Distribution Strategy

Generator Combinators

Example: Custom Distribution

fn gen_my_int() -> @qc.Gen[Int] {
  @qc.frequency([
    (70, @qc.Gen::spawn()),  // Normal values
    (15, @qc.one_of([@qc.pure(0), @qc.pure(1), @qc.pure(-1)])),  // Edge cases
    (15, @qc.one_of([@qc.pure(@int.max_value), @qc.pure(@int.min_value)])),  // Boundaries
  ])
}

Recursive Types with sized

fn gen_tree[T](gen_value : @qc.Gen[T]) -> @qc.Gen[Tree[T]] {
  @qc.sized(fn(size) {
    if size <= 0 {
      gen_value.fmap(Leaf)
    } else {
      @qc.frequency([
        (1, gen_value.fmap(Leaf)),
        (3, @qc.resize(size / 2, gen_tree(gen_value)).bind(fn(left) {
          @qc.resize(size / 2, gen_tree(gen_value)).fmap(fn(right) {
            Node(left, right)
          })
        })),
      ])
    }
  })
}

Using forall for Custom Generators

Use forall for explicit quantification with custom generators:

test "custom generator" {
  @qc.quick_check!(
    @qc.forall(@qc.Gen::spawn(), fn(x : List[Int]) {
      x.rev().rev() == x
    })
  )
}

Nested forall with dependent generators:

test "element from array" {
  @qc.quick_check!(
    @qc.forall(@qc.Gen::spawn(), fn(arr : Array[Int]) {
      @qc.forall(@qc.one_of_array(arr), fn(elem : Int) {
        arr.contains(elem)
      }) |> @qc.filter(arr.length() != 0)
    })
  )
}

Classifying Test Data

classify

test "with classification" {
  @qc.quick_check_fn!(fn(xs : List[Int]) {
    @qc.Arrow(fn(_x) { true })
    |> @qc.classify(xs.length() > 5, "long list")
    |> @qc.classify(xs.length() <= 5, "short list")
  })
}

Output:

+++ [100/0/100] Ok, passed!
22% : short list
78% : long list

label and collect

// Label with string
|> @qc.label(if x.is_empty() { "trivial" } else { "non-trivial" })

// Collect for value distribution
|> @qc.collect(value, "category")

Workflow Steps

Step 1: Identify Target Functions

Categorize functions in your module:

• Transformation functions (input -> different output)
• Normalization functions (input -> same type, simplified)
• Stateful operations (side effects, state changes)

Step 2: Apply Pattern Decision Tree

For each function, walk through the decision tree to select a pattern.

Step 3: Design Generators

1. Identify edge cases specific to your domain
2. Choose distribution percentages
3. Implement using frequency, one_of, sized

Step 4: Write Properties

Implement the property tests using selected patterns.

Step 5: Add Statistics

Add classify calls to verify coverage, adjust generators if needed.

Step 6: Run and Fix

moon info && moon fmt
moon test

Migration from Aletheia

For repositories currently using f4ah6o/aletheia:

Step 1: Update Dependencies

moon remove f4ah6o/aletheia
moon add moonbitlang/quickcheck
moon install

Step 2: Update Imports

{
  "import": [
-   { "path": "f4ah6o/aletheia/quickcheck", "alias": "qc" }
+   { "path": "moonbitlang/quickcheck", "alias": "qc" }
  ]
}

Step 3: API Compatibility

Step 4: Clean Up

Delete .pbt.md files (Aletheia templates are no longer used).

Step 5: Verify

moon info && moon fmt
moon test

CI Workflow (Branch + Worktree + PR)

1. Create branch: git checkout -b pbt/<short-topic>
2. Create worktree: git worktree add ../<repo>-pbt pbt/<short-topic>
3. Run PBT workflow in worktree
4. Commit, push, and create PR
5. After merge: git worktree remove ../<repo>-pbt

Type-Specific Edge Cases

Converted and distributed by TomeVault — claim your Tome and manage your conversions.