matlab-test-creator（matlab 测试创建器）

MATLAB Test Generator

Generate robust unit tests using the MATLAB Testing Framework. This skill covers:

• Test Creation: Writing test classes, methods, fixtures, assertions, and mocks

Must-Follow Rules

• Present a test plan if needed - Create a user-approved test plan before writing test code unless the scope is limited and straightforward.
• Show diff before updating - For existing test files, always show the user a diff and wait for approval before editing.
• Always use class-based tests - Every test file must define a class inheriting from matlab.unittest.TestCase. Never use function-based or script-based tests.
• Do not guess requirements - If scope or expected behaviors are unclear, ask

Test Creation

Test Class Template

Add properties, TestParameter, and setup/teardown blocks as needed.

classdef MyFunctionTest < matlab.unittest.TestCase

    methods (Test)
        % Individual test methods
    end
end

Critical Rules

File Naming

• Test files MUST end with Test.m (e.g., myFunctionTest.m)
• Class name must match filename

Test Method Naming

• Descriptive camelCase names starting with lowercase
• Example: testAdditionWithPositiveNumbers

Test location

Ideally, add all test files to a tests/ folder alongside the source

Properties

Use Mixed case unless it's a TestParameter. For a TestParameter, use camelCase. Only use properties if local variables won't suffice.

Assertions / Qualifications

• Prefer verify* methods (continue on failure) over assert* (stop on failure)
• Use verifyError(@() func(args), "errorID") for error testing
• Use verifyWarningFree(@() func(args)) for clean execution
• Prefer informal APIs over verifyThat calls
• Floating-point comparisons should use tolerance: testCase.verifyEqual(actual, expected, AbsTol=1e-10); testCase.verifyEqual(actual, expected, RelTol=1e-6);
• For advanced verification, constraint objects, and custom constraints see references/constraints.md

No Logic in Tests

• No if, switch, for, or try/catch in test methods. If a test needs conditionals, split into separate methods.
• Follow the Arrange-Act-Assert pattern: set up inputs, call the code under test, verify the result. Nothing else.

Using TestParameter

Parameterize only when assertion logic is identical across all cases — only the data varies. Use separate test methods when cases need different assertions, tolerances, setup, or when you'd need conditionals to distinguish them.

Test Scope

• Test public interfaces, not implementation. Never test private methods directly — verify correctness through the public API.
• If a private method seems complex enough to need its own tests, the user should refactor it into a separate, publicly testable function.

Determinism

For tests involving randomness, seed the RNG and restore it:

methods (TestMethodSetup)
    function resetRandomSeed(testCase)
        originalRng = rng;
        testCase.addTeardown(@() rng(originalRng));
        rng(42, "twister");
    end
end

Test Assumptions

Most tests do not need assumptions. Only add assume* when a test absolutely requires specific environment prerequisites that may not be present on all machines:

testCase.assumeTrue(canUseGPU(), "Requires GPU");

Test Tagging

Use TestTags attribute (e.g., 'Unit', 'Integration', 'Slow', 'GPU') on methods (Test) blocks for selective execution.

Test independence

Each test should be able to run independently and be compatible with running tests in parallel.

Adding path to source files

Use PathFixture to add paths so the tests have access to the source if needed. Use IncludingSubfolders when there are nested packages or subdirectories that also need to be on the path:

methods (TestClassSetup)
    function addSourceToPath(testCase)
        srcFolder = fullfile(fileparts(fileparts(mfilename('fullpath'))), 'src');
        testCase.applyFixture(matlab.unittest.fixtures.PathFixture(srcFolder, ...
            IncludingSubfolders=true));
    end
end

For more details, if necessary, see references/fixtures.md.

Diagnostics

Add additional diagnostics for clarity where the framework diagnostic may be insufficient.

Test Planning

Assess complexity first, then follow the appropriate path.

Simple tests (source code provided, clear behavior, no mocks/fixtures/parameterization)

1. Briefly state what you'll test (methods + key edge cases)
2. Write the test file after user confirms

Standard tests (Large codebase, multiple comprehensive test files) — 3-phase workflow: Gather → Plan → Implement

Phase 1: Gather Requirements

Checklist: Information needed (ask if unknown)

• Code to test - Provide path or content of the function/class to test
• Expected behaviors - What should the code do in normal cases?
• Error conditions - What inputs should cause errors/warnings?
• Test scope: Unit (isolated), Integration (with dependencies), or System?
• External dependencies: Files, databases, network, hardware?
• Determinism needs: Random numbers, timestamps, or other non-deterministic behavior?
• Deployment targets: MATLAB Coder or Compiler SDK? If yes, recommend equivalence testing via matlabtest.coder.TestCase / matlabtest.compiler.TestCase.

Phase 2: Present Test Plan for Approval

Present a test plan. Do NOT write any test files until the user confirms the plan. A plan may include: list of test methods with names, which behaviors each covers, parameterization strategy, fixtures needed, and edge cases selected

Edge Cases to Consider

• Empty inputs ([], '', {})
• Boundary values (0, 1, -1, max, min)
• Invalid types (string instead of number, etc.)
• Large inputs (performance/memory)
• Special values (NaN, Inf, -Inf)

Phase 3: Implement Approved Plan

Apply reference card patterns. Write new test files or show diffs for existing files (per Must-Follow Rules).

References

In many cases, what's present in this file should be sufficient. Do not read the references cards unless the conditions stated in the table are met.