上半部分请看上一篇文章。

下面是下半部分：

Application 的编译任务

我们继续查看createTasksForVariantData的最后一行，taskManager.createTasksForVariantData，发现 createTasksForVariantData 是抽象方法，这里的 taskManager 具体实现是 ApplicationTaskManager，查看 ApplicationTaskManager 的 createTasksForVariantData 方法

/**
 * Creates the tasks for a given BaseVariantData.
 */
 @Override
 public void createTasksForVariantData(
 @NonNull final TaskFactory tasks,
 @NonNull final BaseVariantData<? extends BaseVariantOutputData> variantData) {
 assert variantData instanceof ApplicationVariantData;
 final VariantScope variantScope = variantData.getScope();
 //create sourceGenTask, resGenTask, assetGenTask
 createAnchorTasks(tasks, variantScope);
 createCheckManifestTask(tasks, variantScope);
 handleMicroApp(tasks, variantScope);
 // Create all current streams (dependencies mostly at this point)
 createDependencyStreams(tasks, variantScope);
 // Add a task to process the manifest(s)
 // Add a task to create the res values
 // Add a task to compile renderscript files.
 // Add a task to merge the resource folders
 // Add a task to merge the asset folders
 // Add a task to create the BuildConfig class
 // Add a task to process the Android Resources and generate source files
 // Add a task to process the java resources 
 // Add a task to process this aidl file
 // Add a task to process shader source
 // Add NDK tasks
 // Add external native build tasks
 // Add a task to merge the jni libs folders
 // Add a compile task
 // Add data binding tasks if enabled
 // create packaging task 
 // create the lint tasks.
 ...
 }

代码实在太长了，我只留下了每段代码的注释，注释也已经非常清楚了，这个主要就是生成 variantData 的一系列像 compileXXX、generateXXX、processXXX、mergeXXX的任务，这一系列 task 就是构建一个可运行的完整APK的所需的所有task。下面介绍在编译dex中的过程，涉及的几个task。

Dex的编译过程

// Add a compile task
 recorder.record(
 ExecutionType.APP_TASK_MANAGER_CREATE_COMPILE_TASK,
 project.getPath(),
 variantScope.getFullVariantName(),
 () -> {
 CoreJackOptions jackOptions =
 variantData.getVariantConfiguration().getJackOptions();
 // create data binding merge task before the javac task so that it can
 // parse jars before any consumer
 createDataBindingMergeArtifactsTaskIfNecessary(tasks, variantScope);
 AndroidTask<? extends JavaCompile> javacTask =
 // 创建 javac 任务
 createJavacTask(tasks, variantScope);
 if (jackOptions.isEnabled()) {
 AndroidTask<TransformTask> jackTask =
 createJackTask(tasks, variantScope, true /*compileJavaSource*/);
 setJavaCompilerTask(jackTask, tasks, variantScope);
 } else {
 ...
 addJavacClassesStream(variantScope);
 setJavaCompilerTask(javacTask, tasks, variantScope);
 getAndroidTasks()
 .create(
 tasks,
 // 创建 AndroidJarTask ，生成classes.jar
 new AndroidJarTask.JarClassesConfigAction(variantScope));
 createPostCompilationTasks(tasks, variantScope);
 }
 });

我们直接查看 Add a compile task 注释下的代码，在执行 createPostCompilationTasks 之前，先创建了 javac 任务，任务名称为 compileXXXJavaWithJavac ，该任务是将 java 源文件编译成 class 文件，具体实现是在 JavaCompileConfigAction 类中。创建 javac 任务之后，接着创建了 AndroidJarTask 任务，该任务是将 class 文件整合输出 jar 包，具体实现就是在 AndroidJarTask 类中。

紧接着我们来看一下 createPostCompilationTasks 的方法

/**
 * Creates the post-compilation tasks for the given Variant.
 *
 * These tasks create the dex file from the .class files, plus optional intermediary steps like
 * proguard and jacoco
 *
 */
 public void createPostCompilationTasks(
 @NonNull TaskFactory tasks,
 @NonNull final VariantScope variantScope) {
 checkNotNull(variantScope.getJavacTask());
 variantScope.getInstantRunBuildContext().setInstantRunMode(
 getIncrementalMode(variantScope.getVariantConfiguration()) != IncrementalMode.NONE);
 final BaseVariantData<? extends BaseVariantOutputData> variantData = variantScope.getVariantData();
 final GradleVariantConfiguration config = variantData.getVariantConfiguration();
 TransformManager transformManager = variantScope.getTransformManager();
 ...
 boolean isMinifyEnabled = isMinifyEnabled(variantScope);
 boolean isMultiDexEnabled = config.isMultiDexEnabled();
 // Switch to native multidex if possible when using instant run.
 boolean isLegacyMultiDexMode = isLegacyMultidexMode(variantScope);
 AndroidConfig extension = variantScope.getGlobalScope().getExtension();
 // ----- External Transforms -----
 // apply all the external transforms.
 ...
 // ----- Minify next -----
 if (isMinifyEnabled) {
 boolean outputToJarFile = isMultiDexEnabled && isLegacyMultiDexMode;
 // 内部会判断是否使用 proguard 来创建 proguard 任务和 shrinkResources 任务
 createMinifyTransform(tasks, variantScope, outputToJarFile);
 }
 // ----- 10x support
 ...
 // ----- Multi-Dex support
 Optional<AndroidTask<TransformTask>> multiDexClassListTask;
 // non Library test are running as native multi-dex
 if (isMultiDexEnabled && isLegacyMultiDexMode) {
 ...
 } else {
 multiDexClassListTask = Optional.empty();
 }
 // create dex transform
 // 从 extension 中获取 dexOptions 项的配置
 DefaultDexOptions dexOptions = DefaultDexOptions.copyOf(extension.getDexOptions());
 ...
 // 创建 DexTransform
 DexTransform dexTransform = new DexTransform(
 dexOptions,
 config.getBuildType().isDebuggable(),
 isMultiDexEnabled,
 isMultiDexEnabled && isLegacyMultiDexMode ? variantScope.getMainDexListFile() : null,
 variantScope.getPreDexOutputDir(),
 variantScope.getGlobalScope().getAndroidBuilder(),
 getLogger(),
 variantScope.getInstantRunBuildContext(),
 AndroidGradleOptions.getBuildCache(variantScope.getGlobalScope().getProject()));
 // 创建 dexTask
 Optional<AndroidTask<TransformTask>> dexTask =
 transformManager.addTransform(tasks, variantScope, dexTransform);
 // need to manually make dex task depend on MultiDexTransform since there's no stream
 // consumption making this automatic
 dexTask.ifPresent(t -> {
 t.optionalDependsOn(tasks, multiDexClassListTask.orElse(null));
 variantScope.addColdSwapBuildTask(t);
 });
 ...
 }

为了讲述主流程，我将一些 mutiDex 和 instantRun 判断的源码省略了，这里我们关注非mutiDex和非instantRun的情况。我们看到，如果我们设置了 minifyEnabled 为 true，那么这里就会去创建 createMinifyTransform ，如果use proguard，这里会创建 progruad 的任务和 shrinkResources 的任务。后面将创建 dexTask，这个是 transfromTask 类型的任务，我们先来看下 transFromTask 类

/**
 * A task running a transform.
 */
@ParallelizableTask
public class TransformTask extends StreamBasedTask implements Context {
 private Transform transform;
 ...
 public Transform getTransform() {
 return transform;
 }
 ...
 @TaskAction
 void transform(final IncrementalTaskInputs incrementalTaskInputs)
 throws IOException, TransformException, InterruptedException {
 ...
 recorder.record(
 ExecutionType.TASK_TRANSFORM,
 executionInfo,
 getProject().getPath(),
 getVariantName(),
 new Recorder.Block<Void>() {
 @Override
 public Void call() throws Exception {
 transform.transform(
 new TransformInvocationBuilder(TransformTask.this)
 .addInputs(consumedInputs.getValue())
 .addReferencedInputs(referencedInputs.getValue())
 .addSecondaryInputs(changedSecondaryInputs.getValue())
 .addOutputProvider(
 outputStream != null
 ? outputStream.asOutput()
 : null)
 .setIncrementalMode(isIncremental.getValue())
 .build());
 return null;
 }
 });
 }
}

我们知道，自定义任务中，在任务执行阶段会去执行被 @TaskAction 注解的方法，这里也就是执行 transfrom 方法，而 transfrom 方法中最后又会调用到 transform 的 transfrom 方法，在我们 dexTask 中传入的 transfrom 是DexTransfrom，那我们就去看下 DexTransfrom 的 transfrom 具体实现

public class DexTransform extends Transform {
 @Override
 public void transform(@NonNull TransformInvocation transformInvocation)
 throws TransformException, IOException, InterruptedException {
 ...
 try {
 // if only one scope or no per-scope dexing, just do a single pass that
 // runs dx on everything.
 if ((jarInputs.size() + directoryInputs.size()) == 1
 || !dexOptions.getPreDexLibraries()) {
 // since there is only one dex file, we can merge all the scopes into the full
 // application one.
 File outputDir = outputProvider.getContentLocation("main",
 getOutputTypes(),
 TransformManager.SCOPE_FULL_PROJECT,
 Format.DIRECTORY);
 FileUtils.mkdirs(outputDir);
 // first delete the output folder where the final dex file(s) will be.
 FileUtils.cleanOutputDir(outputDir);
 // gather the inputs. This mode is always non incremental, so just
 // gather the top level folders/jars
 final List<File> inputFiles =
 Stream.concat(
 jarInputs.stream().map(JarInput::getFile),
 directoryInputs.stream().map(DirectoryInput::getFile))
 .collect(Collectors.toList());
 // 通过 AndroidBuilder 转化为 byte
 androidBuilder.convertByteCode(
 inputFiles,
 outputDir,
 multiDex,
 mainDexListFile,
 dexOptions,
 outputHandler);
 for (File file : Files.fileTreeTraverser().breadthFirstTraversal(outputDir)) {
 if (file.isFile()) {
 instantRunBuildContext.addChangedFile(FileType.DEX, file);
 }
 }
 } else {
 ...
 }

最后执行到androidBuilder.convertByteCode

/**
 * Converts the bytecode to Dalvik format
 * @param inputs the input files
 * @param outDexFolder the location of the output folder
 * @param dexOptions dex options
 * @throws IOException
 * @throws InterruptedException
 * @throws ProcessException
 */
 public void convertByteCode(
 @NonNull Collection<File> inputs,
 @NonNull File outDexFolder,
 boolean multidex,
 @Nullable File mainDexList,
 @NonNull DexOptions dexOptions,
 @NonNull ProcessOutputHandler processOutputHandler)
 throws IOException, InterruptedException, ProcessException {checkNotNull(inputs, "inputs cannot be null.");
 checkNotNull(outDexFolder, "outDexFolder cannot be null.");
 checkNotNull(dexOptions, "dexOptions cannot be null.");
 checkArgument(outDexFolder.isDirectory(), "outDexFolder must be a folder");
 checkState(mTargetInfo != null,
 "Cannot call convertByteCode() before setTargetInfo() is called.");
 ImmutableList.Builder<File> verifiedInputs = ImmutableList.builder();
 for (File input : inputs) {
 if (checkLibraryClassesJar(input)) {
 verifiedInputs.add(input);
 }
 }
 //创建 DexProcessBuilder
 DexProcessBuilder builder = new DexProcessBuilder(outDexFolder);
 builder.setVerbose(mVerboseExec)
 .setMultiDex(multidex)
 .setMainDexList(mainDexList)
 .addInputs(verifiedInputs.build());
 runDexer(builder, dexOptions, processOutputHandler);
 }

创建了 DexProcessBuilder ，随后执行到了 runDexer 方法中

public void runDexer(
 @NonNull final DexProcessBuilder builder,
 @NonNull final DexOptions dexOptions,
 @NonNull final ProcessOutputHandler processOutputHandler)
 throws ProcessException, IOException, InterruptedException {
 initDexExecutorService(dexOptions);
 if (dexOptions.getAdditionalParameters().contains("--no-optimize")) {
 mLogger.warning(DefaultDexOptions.OPTIMIZE_WARNING);
 }
 if (shouldDexInProcess(dexOptions)) {
 dexInProcess(builder, dexOptions, processOutputHandler);
 } else {
 dexOutOfProcess(builder, dexOptions, processOutputHandler);
 }
 }

进入到dexInProcess方法

private void dexInProcess(
 @NonNull final DexProcessBuilder builder,
 @NonNull final DexOptions dexOptions,
 @NonNull final ProcessOutputHandler outputHandler)
 throws IOException, ProcessException {
 final String submission = Joiner.on(',').join(builder.getInputs());
 mLogger.verbose("Dexing in-process : %1$s", submission);
 try {
 sDexExecutorService.submit(() -> {
 Stopwatch stopwatch = Stopwatch.createStarted();
 ProcessResult result = DexWrapper.run(builder, dexOptions, outputHandler);
 result.assertNormalExitValue();
 mLogger.verbose("Dexing %1$s took %2$s.", submission, stopwatch.toString());
 return null;
 }).get();
 } catch (Exception e) {
 throw new ProcessException(e);
 }
 }
/**
 * Wrapper around the real dx classes.
 */
public class DexWrapper {
 /**
 * Runs the dex command.
 *
 * @return the integer return code of com.android.dx.command.dexer.Main.run()
 */
 public static ProcessResult run(
 @NonNull DexProcessBuilder processBuilder,
 @NonNull DexOptions dexOptions,
 @NonNull ProcessOutputHandler outputHandler) throws IOException, ProcessException {
 ProcessOutput output = outputHandler.createOutput();
 int res;
 try {
 DxContext dxContext = new DxContext(output.getStandardOutput(), output.getErrorOutput());
 // 构建 Main.Arguments 参数
 Main.Arguments args = buildArguments(processBuilder, dexOptions, dxContext);
 res = new Main(dxContext).run(args);
 } finally {
 output.close();
 }
 outputHandler.handleOutput(output);
 return new DexProcessResult(res);
 }
 ...
}

buildArguments方法通过传入的DexProcessBuilder、dexOptions、dxContext构建 arguments，后面使用的args的参数fileNames，outName，jarOutput都是从DexProcessBuilder来的，然后执行Main的run方法

package com.android.dx.command.dexer;
...
/**
 * Main class for the class file translator.
 */
public class Main {
 /**
 * Run and return a result code.
 * @param arguments the data + parameters for the conversion
 * @return 0 if success > 0 otherwise.
 */
 public int run(Arguments arguments) throws IOException {
 // Reset the error count to start fresh.
 errors.set(0);
 // empty the list, so that tools that load dx and keep it around
 // for multiple runs don't reuse older buffers.
 libraryDexBuffers.clear();
 args = arguments;
 args.makeOptionsObjects(context);
 OutputStream humanOutRaw = null;
 if (args.humanOutName != null) {
 humanOutRaw = openOutput(args.humanOutName);
 humanOutWriter = new OutputStreamWriter(humanOutRaw);
 }
 try {
 if (args.multiDex) {
 return runMultiDex();
 } else {
 return runMonoDex();
 }
 } finally {
 closeOutput(humanOutRaw);
 }
 }
}

这里我们关注非multiDex的情况，即执行了runMonoDex的方法

private int runMonoDex() throws IOException {
 ...
 // 内部会创建dexFile，并填充class
 if (!processAllFiles()) {
 return 1;
 }
 if (args.incremental && !anyFilesProcessed) {
 return 0; // this was a no-op incremental build
 }
 // this array is null if no classes were defined
 byte[] outArray = null;
 if (!outputDex.isEmpty() || (args.humanOutName != null)) {
 // 内部通过Dex类toDex 方法将 class 文件转化dex byte[]
 outArray = writeDex(outputDex);
 if (outArray == null) {
 return 2;
 }
 }
 if (args.incremental) {
 outArray = mergeIncremental(outArray, incrementalOutFile);
 }
 outArray = mergeLibraryDexBuffers(outArray);
 if (args.jarOutput) {
 // Effectively free up the (often massive) DexFile memory.
 outputDex = null;
 if (outArray != null) {
 // 输出的文件名为 classes.dex
 outputResources.put(DexFormat.DEX_IN_JAR_NAME, outArray);
 }
 if (!createJar(args.outName)) {
 return 3;
 }
 } else if (outArray != null && args.outName != null) {
 OutputStream out = openOutput(args.outName);
 out.write(outArray);
 closeOutput(out);
 }
 return 0;
 }

上面的代码中，填充class以及dex流转换，内部流程较为复杂，就不再继续深入，简单做下总结：

1.通过执行 processAllFiles ，内部创建 DexFile 也就是outputDex，并且填充 class 文件

2.通过 writeDex 方法，将 outputDex 传入，方法内部执行的是 outputDex.toDex 方法，将 outputDex 内部填充的 class 转化为 dex 的 byte[] 返回

3.最后将 byte[] 数组创建 classes.dex 输出

最后

Android Gradle Plugin源码繁多，以上文章只是对整体流程的简单梳理，其中简要介绍了构建变体任务的解析和添加，最后对编译dex流程做了简单分析。个人精力有限，这里的源码解析也只是九牛一毛，如有纰漏，欢迎大家拍砖，希望这篇文章能帮助到想了解 Android Gradle Plguin 原理的同学。

喜欢的话请帮忙转发让更多需要的人看到哦。更多Android进阶技术，面试资料整理分享，职业生涯规划，产品，思维，行业观察，谈天说地。可以加Android架构师群；701740775。备注csdn，免费获取

Android Gradle Plugin 源码解析（下）

最后