摘要：前言在上次的博客中我們提到了最終由以為參數執行測試樣例，但并沒有解釋到底測試方法是如何被運行起來的，一些諸如之類的特性又到底是如何實現的呢。這次我們就集中深入的運行機制來探究樣例是如何被運行的。使用拿到的直接運行方法。

在上次的博客中我們提到了最終由Runner以Notifier為參數執行測試樣例，但并沒有解釋到底測試方法是如何被運行起來的，一些諸如RunWith、RunAfter之類的特性又到底是如何實現的呢。這次我們就集中深入Runner的運行機制來探究樣例是如何被運行的。

首先我們需要把注解等用戶配置信息收集起來并attach到對應的方法、類和屬性上，為了在之后的代碼中能夠方便的取到這些信息，我們要包裝原有的類、方法和域，分別如下。

TestClass包含原有的clazz信息，并且維護了兩個Map來管理它所包含的方法與屬性，每個map的鍵是注解，而值是標上注解的FrameWorkMethod或FrameWorkField。同時TestClass還默認內置兩個Comparator來排序自己所包含的方法和屬性。

下面給出如何構造一個TestClass的代碼。

public TestClass(Class clazz) {
        this.clazz = clazz;
        if (clazz != null && clazz.getConstructors().length > 1) {
            throw new IllegalArgumentException(
                    "Test class can only have one constructor");
        }

        Map, List> methodsForAnnotations =
                new LinkedHashMap, List>();
        Map, List> fieldsForAnnotations =
                new LinkedHashMap, List>();

        scanAnnotatedMembers(methodsForAnnotations, fieldsForAnnotations);

        this.methodsForAnnotations = makeDeeplyUnmodifiable(methodsForAnnotations);
        this.fieldsForAnnotations = makeDeeplyUnmodifiable(fieldsForAnnotations);
    }

    protected void scanAnnotatedMembers(Map, List> methodsForAnnotations, Map, List> fieldsForAnnotations) {
        for (Class eachClass : getSuperClasses(clazz)) {
            for (Method eachMethod : MethodSorter.getDeclaredMethods(eachClass)) {
                addToAnnotationLists(new FrameworkMethod(eachMethod), methodsForAnnotations);
            }
            // ensuring fields are sorted to make sure that entries are inserted
            // and read from fieldForAnnotations in a deterministic order
            for (Field eachField : getSortedDeclaredFields(eachClass)) {
                addToAnnotationLists(new FrameworkField(eachField), fieldsForAnnotations);
            }
        }
    }

TestClass的主要功能就是向Runner提供clazz信息以及附帶的注解信息，上文的addToAnnotationLists將對應member加入該annotation映射的member列表。下面給一個TestClass的方法列表截圖，大家可以感受一下。

我們先給出它的父類FrameWorkMember的定義

public abstract class FrameworkMember> implements
        Annotatable {
    abstract boolean isShadowedBy(T otherMember);

    boolean isShadowedBy(List members) {
        for (T each : members) {
            if (isShadowedBy(each)) {
                return true;
            }
        }
        return false;
    }

    protected abstract int getModifiers();

    /**
     * Returns true if this member is static, false if not.
     */
    public boolean isStatic() {
        return Modifier.isStatic(getModifiers());
    }

    /**
     * Returns true if this member is public, false if not.
     */
    public boolean isPublic() {
        return Modifier.isPublic(getModifiers());
    }

    public abstract String getName();

    public abstract Class getType();

    public abstract Class getDeclaringClass();
}

FrameWorkMethod包裝了方法信息以及方法相關的注解以及一些基本的驗證方法比如validatePublicVoid和是否被其他FrameWorkMethod覆蓋的判斷方法，除父類要求外它主要提供的信息如下：

Annotations

Method

ReturnType

ParameterTypes

同FrameWorkMethod差不多，FrameWorkField和它繼承自同一父類，較為簡單，此處就不再詳細介紹了。

Statement是最小的執行單元，諸如RunAfter、RunWith等功能均是通過嵌套Statement來實現的，下面我們先給出Statement的定義，再給出一個嵌套的例子。

public abstract class Statement {
    /**
     * Run the action, throwing a {@code Throwable} if anything goes wrong.
     */
    public abstract void evaluate() throws Throwable;
}

下面以RunAfter的實現為例來說明：

public class RunAfters extends Statement {
   private final Statement next;

   private final Object target;

   private final List afters;

   public RunAfters(Statement next, List afters, Object target) {
       this.next = next;
       this.afters = afters;
       this.target = target;
   }

   @Override
   public void evaluate() throws Throwable {
       List errors = new ArrayList();
       try {
           next.evaluate();
       } catch (Throwable e) {
           errors.add(e);
       } finally {
           for (FrameworkMethod each : afters) {
               try {
                   each.invokeExplosively(target);
               } catch (Throwable e) {
                   errors.add(e);
               }
           }
       }
       MultipleFailureException.assertEmpty(errors);
   }
}

可以看出新的Statement執行時會先執行舊有的Statement，再將附加上的一系列方法以target為參數運行。

Junit使用虛類ParentRunner來管理復合的Runner，使用composite模式，而BlockJunitClassRunner是ParentRunner的一個子類，主要負責同一測試類多個方法的組合測試，也就是最常用的情形。我們首先還是聚焦在如何運行測試樣例上。

首先看ParentRunner如何實現run方法

@Override
    public void run(final RunNotifier notifier) {
        EachTestNotifier testNotifier = new EachTestNotifier(notifier,
                getDescription());
        try {
            Statement statement = classBlock(notifier);
            statement.evaluate();
        } catch (AssumptionViolatedException e) {
            testNotifier.addFailedAssumption(e);
        } catch (StoppedByUserException e) {
            throw e;
        } catch (Throwable e) {
            testNotifier.addFailure(e);
        }
    }

這里有個classBlock方法用來提供真正運行的Statement，下面我們看一看

protected Statement classBlock(final RunNotifier notifier) {
        Statement statement = childrenInvoker(notifier);
        if (!areAllChildrenIgnored()) {
            statement = withBeforeClasses(statement);
            statement = withAfterClasses(statement);
            statement = withClassRules(statement);
        }
        return statement;
    }

這個過程就是先通過反射獲得初始Statement，然后附加上RunBefore、RunAfter、用戶自定義Rule，我們來看一下初始Statement是如何生成的。

其過程是先取得所有通過過濾器的Childeren，再使用內置的調度器來分別按順序調用runChild方法，下面我們給出BlockJunit4ClassRunner的runChild方法

@Override
    protected void runChild(final FrameworkMethod method, RunNotifier notifier) {
        Description description = describeChild(method);
        if (isIgnored(method)) {
            notifier.fireTestIgnored(description);
        } else {
            Statement statement;
            try {
                statement = methodBlock(method);
            }
            catch (Throwable ex) {
                statement = new Fail(ex);
            }
            runLeaf(statement, description, notifier);
        }
    }

這里面最重要的就是RunLeaf也就是原子測試方法以及如何為單個方法生成的Statement——methodBlock，我們在下面分別給出。

protected final void runLeaf(Statement statement, Description description,
            RunNotifier notifier) {
        EachTestNotifier eachNotifier = new EachTestNotifier(notifier, description);
        eachNotifier.fireTestStarted();
        try {
            statement.evaluate();
        } catch (AssumptionViolatedException e) {
            eachNotifier.addFailedAssumption(e);
        } catch (Throwable e) {
            eachNotifier.addFailure(e);
        } finally {
            eachNotifier.fireTestFinished();
        }
    }

RunLeaf的邏輯并不難，先通知Notifier測試開始，再直接調用statement的evaluate方法，最后通知Notifier測試結束。我們再來看看statement是如何生成的。

protected Statement methodBlock(final FrameworkMethod method) {
        Object test;
        try {
            test = new ReflectiveCallable() {
                @Override
                protected Object runReflectiveCall() throws Throwable {
                    return createTest(method);
                }
            }.run();
        } catch (Throwable e) {
            return new Fail(e);
        }

        Statement statement = methodInvoker(method, test);
        statement = possiblyExpectingExceptions(method, test, statement);
        statement = withPotentialTimeout(method, test, statement);
        statement = withBefores(method, test, statement);
        statement = withAfters(method, test, statement);
        statement = withRules(method, test, statement);
        return statement;
    }

上述代碼的邏輯還是比較復雜的，這里簡單概述一下，首先構造測試類的實例，然后為對應method構造statement的子類InvokeMethod，然后調用FrameWorkMethod的反射運行方法，如下：

public class InvokeMethod extends Statement {
    private final FrameworkMethod testMethod;
    private final Object target;

    public InvokeMethod(FrameworkMethod testMethod, Object target) {
        this.testMethod = testMethod;
        this.target = target;
    }

    @Override
    public void evaluate() throws Throwable {
        testMethod.invokeExplosively(target);
    }
}

Suite是對于ParentRunner的另一子類實現，主要用于多個測試類的情形。Suite自己維護一個runner列表，實現了getChilderen方法，其層次是在上文中提到的runChildren里，這一部分需要取出children節點然后調用runChild方法。我們著重考察suite和BlockJunit4ClassRunner在getChildren和runChild方法上的區別。Suite通過用戶傳入的runnerBuilder為每個類多帶帶建立runner作為children返回，而后者則返回帶Test注解的FrameWorkMethod列表。使用getChildren拿到的runner直接運行run方法。下面我們給出RunnerBuilder是如何為一系列測試類提供一系列對應的Runner，說來也簡單，就是使用為單個類建立Runner的方法為每個測試類建立最后組成一個集合。但是此處需要防止遞歸——this builder will throw an exception if it is requested for another runner for {@code parent} before this call completes（說實話這段如何防止遞歸我也沒看懂，有看懂的兄弟求教）。對于Suite而言，一般就是它維護一個BlockJUnit4ClassRunner列表。

public abstract class RunnerBuilder {
    private final Set> parents = new HashSet>();

    /**
     * Override to calculate the correct runner for a test class at runtime.
     *
     * @param testClass class to be run
     * @return a Runner
     * @throws Throwable if a runner cannot be constructed
     */
    public abstract Runner runnerForClass(Class testClass) throws Throwable;

    /**
     * Always returns a runner, even if it is just one that prints an error instead of running tests.
     *
     * @param testClass class to be run
     * @return a Runner
     */
    public Runner safeRunnerForClass(Class testClass) {
        try {
            return runnerForClass(testClass);
        } catch (Throwable e) {
            return new ErrorReportingRunner(testClass, e);
        }
    }

    Class addParent(Class parent) throws InitializationError {
        if (!parents.add(parent)) {
            throw new InitializationError(String.format("class "%s" (possibly indirectly) contains itself as a SuiteClass", parent.getName()));
        }
        return parent;
    }

    void removeParent(Class klass) {
        parents.remove(klass);
    }

    /**
     * Constructs and returns a list of Runners, one for each child class in
     * {@code children}.  Care is taken to avoid infinite recursion:
     * this builder will throw an exception if it is requested for another
     * runner for {@code parent} before this call completes.
     */
    public List runners(Class parent, Class[] children)
            throws InitializationError {
        addParent(parent);

        try {
            return runners(children);
        } finally {
            removeParent(parent);
        }
    }

    public List runners(Class parent, List> children)
            throws InitializationError {
        return runners(parent, children.toArray(new Class[0]));
    }

    private List runners(Class[] children) {
        List runners = new ArrayList();
        for (Class each : children) {
            Runner childRunner = safeRunnerForClass(each);
            if (childRunner != null) {
                runners.add(childRunner);
            }
        }
        return runners;
    }
}

Junit使用BlockJUnit4ClassRunnerWithParameters繼承BlockJUnit4ClassRunner來完成對于組合方法的帶參數測試。它覆寫了createTest方法和對構造器和域的驗證方法。

@Override
    public Object createTest() throws Exception {
        InjectionType injectionType = getInjectionType();
        switch (injectionType) {
            case CONSTRUCTOR:
                return createTestUsingConstructorInjection();
            case FIELD:
                return createTestUsingFieldInjection();
            default:
                throw new IllegalStateException("The injection type "
                        + injectionType + " is not supported.");
        }
    }

Parameterized繼承了Suite，用來完成對多個類的組合測試的帶參數版本。它提供三大注解——Parameters、Parameter、UseParametersRunnerFactory，前兩者是用來指定參數的，后者用來指定對于每一個測試類如何生成Runner的工廠,默認工廠返回BlockJUnit4ClassRunnerWithParameters。我們下面給出內置的工廠類如何創建runner的代碼。

        private List createRunnersForParameters(
                Iterable