Agent Skill
2/7/2026

tdd

This skill should be used when writing unit tests, TDD test cases, test coverage analysis, or test-driven development for NG component Pattern methods (e.g., OnModifyDone, OnDirtyLayoutWrapperSwap). Covers test principles, templates, common pitfalls, branch coverage, API verification, state reset, and mandatory self-checklist for ace_engine testing.

O
openharmony
7GitHub Stars
2Views
npx skills add openharmony/arkui_ace_engine

SKILL.md

Nametdd
DescriptionThis skill should be used when writing unit tests, TDD test cases, test coverage analysis, or test-driven development for NG component Pattern methods (e.g., OnModifyDone, OnDirtyLayoutWrapperSwap). Covers test principles, templates, common pitfalls, branch coverage, API verification, state reset, and mandatory self-checklist for ace_engine testing.

name: tdd description: This skill should be used when writing unit tests, TDD test cases, test coverage analysis, or test-driven development for NG component Pattern methods (e.g., OnModifyDone, OnDirtyLayoutWrapperSwap). Covers test principles, templates, common pitfalls, branch coverage, API verification, state reset, and mandatory self-checklist for ace_engine testing. version: 1.0.0

TDD Writing Guidelines for NG Component Pattern Methods

When writing TDD tests for NG component Pattern methods (e.g., OnModifyDone(), OnDirtyLayoutWrapperSwap()), follow these guidelines to avoid common pitfalls.

Core Principles

1. Direct Method Invocation

Call the target method directly, not through indirect triggers.

  • menuPattern->OnModifyDone()
  • menuPattern->FireBuilder() (doesn't trigger OnModifyDone())

2. State Reset

Always reset potentially interfering state before calling target method.

// Reset to clean state before testing
auto renderContext = menuNode->GetRenderContext();
renderContext->ResetOuterBorder();  // Reset entire property group

// Then call target method
menuPattern->OnModifyDone();

3. API Verification

Use Grep/Read tools to verify method names, never guess.

# Search for method definition
grep -rn "ResetOuterBorder\|Reset.*Outer" frameworks/core/components_ng/render/render_context.h

# Understand macro-generated methods
grep -A 5 "ACE_DEFINE_PROPERTY_GROUP" frameworks/core/components_ng/property/property.h
  • ACE_DEFINE_PROPERTY_GROUP(OuterBorder, OuterBorderProperty) generates:
    • GetOrCreateOuterBorder(), GetOuterBorder(), CloneOuterBorder(), ResetOuterBorder()
    • NOT ResetOuterBorderRadius() (common mistake)

4. Branch Coverage

Write paired tests for if/else branches.

// Branch 1: Condition is true
HWTEST_F(Xxx, Test_BranchTrue, TestSize.Level1) {
    borderRadiusVP.SetRadius(Dimension(10.0_vp));  // No PERCENT unit
    renderContext->ResetOuterBorder();
    pattern->OnModifyDone();
    EXPECT_TRUE(outerRadius.has_value());
}

// Branch 2: Condition is false
HWTEST_F(Xxx, Test_BranchFalse, TestSize.Level1) {
    borderRadiusPercent.SetRadius(Dimension(50.0f, DimensionUnit::PERCENT));
    renderContext->ResetOuterBorder();
    pattern->OnModifyDone();
    EXPECT_FALSE(outerRadius.has_value());
}

5. No Line Numbers in Comments

Avoid hardcoding line numbers in test comments.

  • OnModifyDone should skip UpdateBorderRadius at line 299
  • OnModifyDone should skip UpdateBorderRadius when borderRadius has percent unit
  • Reason: Line numbers change as code evolves, making comments outdated
  • Alternative: Describe the logical condition being tested rather than the physical line location

6. No Documentation-Only Tests

Every test case must have actual test logic.

  • ❌ Documentation-only test with only comments and EXPECT_TRUE(true)
  • ✅ Test with real assertions and verification logic
  • Reason: Tests without executable logic don't verify behavior and waste maintenance resources
  • Alternative: Put branch documentation in separate design documents or code comments, not in test cases
  • Example of what NOT to do:
    // ❌ BAD: Documentation-only test
    HWTEST_F(ComponentPatternTestNg, ComponentPattern_BranchDocumentation, TestSize.Level1) {
        /**
         * @tc.steps: step1. Document branches
         * Branch 1: condition is true
         * Branch 2: condition is false
         */
        EXPECT_TRUE(true);  // No actual testing!
    }
    
  • Correct approach: Write separate test cases for each branch with real logic

7. No Magic Numbers

Use named constants instead of hardcoded numbers.

  • ❌ Hardcoded numbers like 720, 1280, 0, 3.0, -1, 100000
  • ✅ Named constants with clear semantic meaning
  • Reason: Magic numbers make code hard to understand and maintain. Constants provide context and make changes easier
  • Where to define: Place constants in anonymous namespace at the top of the test file
  • Example:
    // ❌ BAD: Magic numbers
    HWTEST_F(ComponentPatternTestNg, ComponentPattern_TestBranch, TestSize.Level1) {
        SystemProperties::InitDeviceInfo(720, 1280, 0, 3.0, false);
        auto size = GetSubWindowSize(100000, 0);
        EXPECT_GT(size.Width(), 0);
    }
    
    // ✅ GOOD: Named constants
    namespace {
        const int32_t TEST_DEVICE_WIDTH = 720;
        const int32_t TEST_DEVICE_HEIGHT = 1280;
        const int32_t TEST_DEVICE_ORIENTATION = 0;
        const double TEST_DEVICE_RESOLUTION = 3.0;
        const bool TEST_DEVICE_IS_ROUND = false;
        const int32_t TEST_PARENT_CONTAINER_ID = 100000;
        const uint32_t DEFAULT_DISPLAY_ID = 0;
    }
    HWTEST_F(ComponentPatternTestNg, ComponentPattern_TestBranch, TestSize.Level1) {
        SystemProperties::InitDeviceInfo(
            TEST_DEVICE_WIDTH, TEST_DEVICE_HEIGHT, TEST_DEVICE_ORIENTATION,
            TEST_DEVICE_RESOLUTION, TEST_DEVICE_IS_ROUND);
        auto size = GetSubWindowSize(TEST_PARENT_CONTAINER_ID, DEFAULT_DISPLAY_ID);
        EXPECT_GT(size.Width(), 0);
    }
    

Test Template

/**
 * @tc.name: PatternMethod_TestBranchName
 * @tc.desc: Test Description covering specific behavior (e.g., "when X happens, Y should occur")
 * @tc.type: FUNC
 */
HWTEST_F(ComponentPatternTestNg, ComponentPattern_TestBranchName, TestSize.Level1)
{
    // 1. Environment setup
    MockPipelineContext::GetCurrent()->SetMinPlatformVersion(static_cast<int32_t>(PlatformVersion::VERSION_ELEVEN));
    ScreenSystemManager::GetInstance().dipScale_ = DIP_SCALE;

    // 2. Create nodes (using existing helper methods like GetPreviewMenuWrapper)
    auto menuWrapperNode = GetPreviewMenuWrapper();
    ASSERT_NE(menuWrapperNode, nullptr);
    auto menuNode = AceType::DynamicCast<FrameNode>(menuWrapperNode->GetChildAtIndex(0));
    ASSERT_NE(menuNode, nullptr);
    auto menuPattern = menuNode->GetPattern<MenuPattern>();
    ASSERT_NE(menuPattern, nullptr);
    auto menuLayoutProperty = menuNode->GetLayoutProperty<MenuLayoutProperty>();
    ASSERT_NE(menuLayoutProperty, nullptr);
    auto renderContext = menuNode->GetRenderContext();
    ASSERT_NE(renderContext, nullptr);

    // 3. Set test data to trigger target branch
    TestDataType testData;
    testData.SetValue(...);  // Configure to trigger branch
    menuLayoutProperty->UpdateProperty(testData);

    // 4. Reset interfering state (CRITICAL!)
    renderContext->ResetTargetProperty();

    // 5. Call target method directly
    menuPattern->TargetMethod();

    // 6. Verify results match branch expectation
    auto result = renderContext->GetTargetProperty();
    EXPECT_TRUE(result.has_value());  // Or false based on branch
}

Common Pitfalls

PitfallSymptomSolution
Indirect method callTest doesn't trigger target code pathCall target method directly
No state resetTest fails with unexpected valuesReset all potentially interfering properties before calling method
Guessed API nameCompilation error or method not foundUse Grep to verify exact method name from source
Unclean test dataTests interfere with each otherEach test should be independent, reset all state
Missing branch coverageCode coverage shows gapsWrite tests for both true and false branches
Line numbers in commentsComments become outdated when code changesDescribe logical conditions instead of physical locations
Documentation-only testsTest contains only comments with EXPECT_TRUE(true)Every test must have real assertions and verification logic
Magic numbersHardcoded numbers like 720, 1280, 100000 without meaningUse named constants with clear semantic meaning in anonymous namespace

Mandatory Self-Checklist

After generating test code, verify:

  • All called methods exist (verified via Grep/Read)
  • State is reset before calling target method
  • Test data triggers the correct branch (HasPercentUnit returns expected value)
  • Assertions match branch behavior
  • All API signatures match source definitions
  • Test follows existing test file structure and naming conventions
  • Comments describe logical behavior, not line numbers (e.g., use "when X happens" instead of "at line 299")
  • Every test case has real test logic (not just documentation with EXPECT_TRUE(true))
  • No magic numbers - all numeric literals replaced with named constants

Reference Implementation

See working example at: test/unittest/core/pattern/menu/menu_pattern_test_ng.cpp:1482-1574

  • MenuPatternTest_OnModifyDone_UpdateBorderRadius - Tests when border has no PERCENT unit
  • MenuPatternTest_OnModifyDone_PercentBorderRadius - Tests when border has PERCENT unit
Skills Info
Original Name:tddAuthor:openharmony