Android热修复:Qfix方案的gradle实践
一、Android热修复方案的发展
Android热修复技术现在主流分为NativeHook,ClassLoader以及新出现的Instant-Run方案,在工作中因为团队需要为团队引入了QZone的ClassLoader插桩方案,当初开发的时候(15年底)市面上还只有AndFix和ClassLoader两种方案,而那时的AndFix兼容性也还比较差,就选择了ClassLoader方案,基于网上开源的Nuwa方案修改而来。
2016年的时候Android热修复方案如雨后春笋,ClassLoader方案新起之秀有Tinker,QFix等。对比几种新出的方案,调研之后认为QFix与团队之前使用的ClassLoader方案最为接近,同时无需插桩,避免了插桩性能损失问题,在16年下半年时候将团队的热修复方案升级到了QFix方案。
由于团队的热修复方案需要对原来的方案做不少兼容,所以会有不少雍余的代码。17年年初就想着自己写个QFix的开源gradle实现,现项目已完成发布到Github,地址:https://github.com/alexclin0188/QFixPatch。 这篇文章主要是对开发过程的一个记录,发出来供有兴趣的同学参考和自己以后复习使用。
二、QFix方案原理
QFix方案原理介绍原文。有兴趣的同学可以仔细研读下这篇原理介绍,这里就不再详细讲解,补丁修复的基本原来和ClassLoader类同,反射在DexPahList类中dex数组前插入补丁dex,区别在于QFix使用在native层调用dvmResolveClass方法来解决pre-verify的问题,不再需要插桩。
GitHub上有一个开源的简单的QFix方案Eclipse工程demo:https://github.com/lizhangqu/QFix, QFixPatch项目即是基于这个demo扩展而来的一个gradle实践。
三、Gradle插件开发
gradle插件生成补丁逻辑和Nuwa类似,只是增加了dexdump分析补丁类class-id的操作,简单的流程示意图如下:
首先是buildBase,构建基础apk和保存基础信息,在dexTask之前增加hook,将原始apk中的类的sha值保存到hash.txt中,以便生成补丁时使用。然后在buildPatch执行时比较找出修改的类(补丁类),并通过dexdump分析这些补丁类在基础apk的dex中的class-id,将分析得到的class-id信息和补丁类通过dex工具打包成一个补丁apk
3.1 创建插件工程
在一个AS工程内,如果有buildSrc目录,这个目录就会被当做gradle插件源码目录编译,并可以直接在工程模块中直接使用编写的gradle插件,下面是QFixPatch工程gradle插件源码目录截图
和Android模块不同的是其中有resources目录,下面有个META-INF.gradle-plugins文件夹,内有propertites文件,而这个文件是用来注册插件入口类的,下面是alexclin.qfix.properties的内容
//alexclin.qfix.properties文件,properties文件名称即为gradle插件名
implementation-class=alexclin.qfix.QFixPlugin
我们gradle插件的使用方式是在模块的build.gradle中增加apply,如下
//build.gradle文件
......
apply plugin: 'alexclin.qfix'
......
而这里的apply plugin的作用最终就是让项目在执行gradle编译时调用我们的实现类alexclin.qfix.QFixPlugin的apply函数,下面让我们看QFixPlugin类的代码
//QFixPlugin类
public class QFixPlugin implements Plugin<Project> {
static final String CLASS_ID_TXT = "class-ids.txt"
static final String PATCH_NAME = "patch"
static final String DEBUG = "debug"
static final String PLUGIN_NAME = "qfix"
def debugOn
@Override
void apply(Project project) {
if (project.getPlugins().hasPlugin(AppPlugin)) {
project.extensions.create(PLUGIN_NAME, QFixExtension, project)
applyPlugin(project)
}
}
//真正的applyplugin逻辑代码
private void applyPlugin(Project project) {
project.afterEvaluate {
def extension = project.extensions.findByName(PLUGIN_NAME) as QFixExtension
Creator creator = new Creator(project, extension);
debugOn = extension.debugOn
project.android.applicationVariants.each { variant ->
if (variant.name.contains(DEBUG) && !debugOn)
return;
configTasks(project, variant, creator);
}
//添加总的buildPatch Task
def releasePatch = project.tasks.findByName("assembleReleasePatch")
def debugPatch = project.tasks.findByName("assembleDebugPatch")
if (releasePatch != null || debugPatch != null) {
def buildPatchTask = project.task("buildPatch")
if (debugPatch) buildPatchTask.dependsOn debugPatch
if (releasePatch) buildPatchTask.dependsOn releasePatch
}
def assembleDebugBase = project.tasks.findByName("assembleDebugBase")
def assembleReleaseBase = project.tasks.findByName("assembleReleaseBase")
if (assembleDebugBase != null || assembleReleaseBase != null) {
def assembleBase = project.task("buildBase");
if (assembleDebugBase) assembleBase.dependsOn assembleDebugBase
if (assembleReleaseBase) assembleBase.dependsOn assembleReleaseBase
}
}
}
....
}
可以看到上面这段代码里面,主要是在真正的applyplugin函数对项目进行了Task设置,并对buildBase,buildPatch,assembleDebugBase,assembleDebugPacth,assembleReleaseBase,assembleReleasePacth的依赖关系做了设置。
实际上只有assemble<Viriant>Base,assemble<Viriant>Patch的区别,前者是构建基础apk包和保存基础信息,后者是构建补丁
3.2 插件设置QFixExtension
gradle插件支持属性设置,如我们的QFix插件在build.gradle中属性设置,如下
//qfix插件在app模块build.gradle中的属性设置
qfix{
debugOn true //debug模式是否开启补丁构建task,可选
outputDir 'qfix_output'//补丁基础信息默认输入目录和构建输出目录,必须
excludeClass = ["App.class"]//需要排除的类,会自动排除Application及其父类,此处只是示例,可选
//includePackage = []//需要包含的package,可选
strictMode true //严格模式,用于解决ART下dex激进内联问题,详情参看README,可选
}
而上面这个设置实际对应的是一个Extension类,在qfix-gradle插件中对应的就是QFixExtension类
package alexclin.qfix
import org.gradle.api.Project
class QFixExtension {
HashSet<String> includePackage = []
HashSet<String> excludeClass = []
//补丁构建在debug版是否开启,默认开启
boolean debugOn = false
//补丁基础信息保存目录和补丁输出目录
String outputDir
//严格模式,启用则打包所有引用补丁类的class到补丁中,应对ART激进内联引起的问题
boolean strictMode = false;
QFixExtension(Project project) {
}
}
3.3 设置Task和hook dexTask
设置Task和hook dexTask的逻辑主要都在QFixPlugin的configTasks函数中,以下是简化的函数代码。
static void configTasks(Project project, BaseVariant variant, Creator creator) {
Map hashMap
//获取dexTask和proguardTask
def dexTask = ...
def proguardTask = ...
if (creator.patchTaskEnable()) {
//创建Task,寻找被修改的类,只有assemble<Virant>Patch时才会被调用
def diffClassBeforeDex = "diffClassBeforeDex${variant.name.capitalize()}"
def diffClassBeforeDexTask = ...
diffClassBeforeDexTask.dependsOn dexTask.taskDependencies.getDependencies(dexTask)
//创建Task,将改变的类打成一个dex
def hotfixPatch = "assemble${variant.name.capitalize()}Patch"
def hotfixPatchTask = ...
hotfixPatchTask.dependsOn diffClassBeforeDexTask
}
//创建Hook Task, 在dexTask执行之前 保存所有类的sha值到对应目录的hash.txt
def shaClassBeforeDex = "shaClassBeforeDex${variant.name.capitalize()}"
def shaClassBeforeDexTask = ...
//备份构建过程中的mapping.txt,如果有
if (proguardTask) {
def mapFile = new File("${project.buildDir}/outputs/mapping/${variant.dirName}/mapping.txt")
def newMapFile = creator.getMappingOutFile(variant);
Utils.copyFile(mapFile, newMapFile)
}
}
shaClassBeforeDexTask.dependsOn dexTask.taskDependencies.getDependencies(dexTask)
//对assembleRelease或assembleDebug添加Hook,保存apk
def assembleTaskName = "assemble${variant.name.capitalize()}Base";
def assembleTask = project.task(assembleTaskName);
Closure saveAssembleClosure = ...
//设置依赖关系,保证assembleXXXBase在android自身的assembleDebug/assembleRelease之后执行
assembleTask.doLast(saveAssembleClosure)
assembleTask.dependsOn shaClassBeforeDexTask
assembleTask.dependsOn project.tasks["assemble${variant.name.capitalize()}"]
}
上面创建的几个task中间最重要的是shaClassBeforeDexTask和diffClassBeforeDexTask,前者是保存基础apk中所有类的sha信息,后者则是对比基础apk和修改后的代码类的信息,找出改变的类。先来看shaClassBeforeDexTask
//shaClassBeforeDexTask 保存基础apk中所有类的sha信息
def shaClassBeforeDexTask = project.task(shaClassBeforeDex) << {
//准备工作...
...
//保存所有类的sha值到hashFile中
Set<File> inputFiles = AndroidUtils.getDexTaskInputFiles(project, variant, dexTask)
if (proguardTask) {
inputFiles.each {
inputFile ->
if (inputFile.path.endsWith(".jar")) {
shaJarInfo(inputFile, creator.patchSetting, hashFile)
}
}
} else if (AndroidUtils.compareVersionName(Version.ANDROID_GRADLE_PLUGIN_VERSION, "2.2.3") > -1) {
//没有混淆在2.2.3及以后插件上inputFiles不包含当前模块类
//合并所有jar包
Set<File> jarAndDir = new HashSet<>(inputFiles)
jarAndDir.add(new File(project.buildDir, "intermediates/classes/${variant.dirName}"))
File combinedJar = combineJarAndDir(project, jarAndDir)
shaJarInfo(combinedJar, creator.patchSetting, hashFile)
}
//备份构建过程中的mapping.txt
......
}
diffClassBeforeDexTask的作用是找出改变的类并dump分析改变的类在基础apk的dex中的class-id,简化的代码如下:
//diffClassBeforeDexTask 找出改变的类并dump分析改变的类在基础apk的dex中的class-id
def diffClassBeforeDexTask = project.task(diffClassBeforeDex) << {
//补丁准备工作, 读取原来apk的dex目录,proguard-mapping,hash.txt,以及初始化补丁输出目录
def baseDexDir = ...
File mappingFile = ...
hashMap = ...
File patchOutDir = ...
//获取dexTask的输入jar报参数
//比较所有类与之前保存的sha值是否有差异,有差异则保存到patchClassDir
Set<File> inputFiles = AndroidUtils.getDexTaskInputFiles(project, variant, dexTask)
if (proguardTask) {
inputFiles.each {
inputFile ->
if (inputFile.path.endsWith(".jar")) {
diffJar(inputFile, hashMap, creator, variant, finder)
}
}
} else if (AndroidUtils.compareVersionName(Version.ANDROID_GRADLE_PLUGIN_VERSION, "2.2.3") > -1) {
//没有混淆在2.2.3及以后插件上inputFiles不包含当前模块类
//合并所有jar包
Set<File> jarAndDir = new HashSet<>(inputFiles)
jarAndDir.add(new File(project.buildDir, "intermediates/classes/${variant.dirName}"))
File combinedJar = combineJarAndDir(project, jarAndDir)
diffJar(combinedJar, hashMap, creator, variant, finder)
}
def allRefPatchClasses = ...
def appName = ...
//增加dexDump处理,分析补丁类在基础apk的dex中class-id并保存
def dumpCmdPath = AndroidUtils.getDexDumpPath(project, creator.sdkDir);
File patchClassDir = creator.getClassOutDir(variant);
File classIdsFile = new File(patchClassDir, CLASS_ID_TXT);
DexClassIdResolve.dumpDexClassIds(dumpCmdPath, baseDexDir, patchClassDir, classIdsFile,appName,allRefPatchClasses)
}
可以看到重要的逻辑就是比较sha值和dexDump分析class-id, diffJar函数主要是读取jar包中的类并保存sha信息,代码如下
static void diffJar(File jarFile, HashMap hashMap, Creator builder, BaseVariant variant, SubClsFinder finder) {
File basePatchClassDir = builder.getClassOutDir(variant);
if (!basePatchClassDir.exists()) basePatchClassDir.mkdirs();
if (jarFile && jarFile.isFile()) {
def file = new JarFile(jarFile);
builder.patchSetting.addApplicationAndSuper(file, variant);
Enumeration enumeration = file.entries();
while (enumeration.hasMoreElements()) {
JarEntry jarEntry = (JarEntry) enumeration.nextElement();
String entryName = jarEntry.getName();
InputStream inputStream = file.getInputStream(jarEntry);
if (!builder.patchSetting.isExcluded(entryName)) {
def bytes = Utils.readAllBytesAndClose(inputStream);
def hash = DigestUtils.shaHex(bytes)
if (Utils.notSame(hashMap, entryName, hash)) {
finder.addAbsPatchClass(bytes, entryName)
Utils.copyBytesToFile(bytes, Utils.touchFile(basePatchClassDir, entryName))
} else {
finder.addOutPatchClass(bytes, entryName)
}
}
}
Collection<String> collections = finder.getRefClasses();
for (String entryName : collections) {
ZipEntry zipEntry = file.getJarEntry(entryName);
InputStream inputStream = file.getInputStream(zipEntry);
def bytes = Utils.readAllBytesAndClose(inputStream);
Utils.copyBytesToFile(bytes, Utils.touchFile(basePatchClassDir, entryName))
inputStream.close();
}
file.close();
}
}
3.4 调用dexDump分析补丁类在原始apk的dex中class-id
dexDump的分析逻辑在alexclin.qfix.DexClassIdResolve类中,主要代码是readClassIdMap函数,该函数根据传入的dex和补丁类信息,结合dexdump工具的输出,将补丁类class-id保存下来
private static String readClassIdMap(InputStream mInputStream, int mDexIndex, HashSet<String> mPatchSet,Set<String> patchRefSet,ArrayList<ClassIdMap> mClassIdMapList) {
InputStreamReader isReader = null;
BufferedReader reader = null;
String entrance = null;
try {
isReader = new InputStreamReader(mInputStream);
reader = new BufferedReader(isReader);
boolean findHead = false;
boolean findClass = false;
int classIndex = -1;
long classIdx = -1;
def line
while ((line = reader.readLine()) != null) {
if (line.startsWith("Class #") && line.endsWith(" header:") && !findHead && classIndex < 0) {
findHead = true;
classIndex = Integer.parseInt(line.substring("Class #".length(), line.indexOf(" header:")));
} else if (line.startsWith("class_idx") && findHead && classIndex >= 0 && classIdx < 0) {
classIdx = Integer.parseInt(line.substring(line.indexOf(": ") + 2));
} else if (line.startsWith("Class #") && findHead && classIndex >= 0
&& line.contains(String.valueOf(classIndex)) && classIdx > 0) {
findClass = true;
} else if (line.startsWith(" Class descriptor") && findHead && findClass && classIndex >= 0 && classIdx > 0) {
String className = line.substring(line.indexOf("'L") + 2, line.indexOf(";'"));
if (mPatchSet.contains(className)) {
ClassIdMap item = new ClassIdMap(className, mDexIndex, classIdx);
mClassIdMapList.add(item);
}else if(!patchRefSet.contains(className)&&entrance==null){
entrance = className;
}
System.out.println("className:"+className)
findHead = false;
findClass = false;
classIndex = -1;
classIdx = -1;
}
}
} catch (Exception e) {
//异常处理和关闭流
......
}
return entrance;
}
因为在APP注入补丁时调用dvmResolveClass需要传入一个引用类,这里使用当前dex中和补丁类无调用关系的一个类作为该dex中补丁类的入口类
3.5 生成补丁
经过上面的流程,已经有所有改变的类(补丁类)和补丁类在基础apk的Dex中的class-ids信息(class-ids.txt),生成补丁就直接调用dx工具类即可
def hotfixPatchTask = project.task(hotfixPatch) << {
//调用dx工具生成apk
......
AndroidUtils.dex(project, creator.getClassOutDir(variant), creator.sdkDir, pathFilePath)
//签名补丁
SigningConfig signingConfig = variant.signingConfig;
if(signingConfig!=null){
......
if (AndroidUtils.signApk(patchFile, patchSignedFile, signingConfig,compatible))
patchFile.delete();
}
}
四、补丁客户端应用代码开发
4.1 使用Application代理方式
因为Application类被调起之后我们才有机会去加载我们自己的补丁,而如果将Application调用了某些其他类,这些类的class也可能在我们加载补丁前就已经被加载到ClassLoader中,这样这些类就没有机会被替换了。
为了解决如上这个问题,参考了Tinker的方式,也才用Application代理来解决,将真正的Application逻辑写在代理类中。demo中的Application类如下
public class App extends PatchApplication {
private static final String DELEGATE_NAME = "alexclin.qfix.qfixgradle.AppDelegate";
public App() {
super(DELEGATE_NAME);
}
@Override
protected void attachBaseContext(Context base) {
super.attachBaseContext(base);
MultiDex.install(base);
File debugPatch = new File(Environment.getExternalStorageDirectory(),"debugPatch.apk");
PatchTool.installPatch(this,debugPatch);
}
}
下面是PatchApplication的代码,主要是反射调起Application代理类的对应方法
public abstract class PatchApplication extends Application {
private String delegateClassName;
private ApplicationLifeCycle delegate;
public PatchApplication(String delegateClassName) {
this.delegateClassName = delegateClassName;
}
@Override
public final void onCreate() {
super.onCreate();
ensureDelegate();
delegate.onCreate();
}
@Override
public final void onTerminate() {
super.onTerminate();
if(delegate!=null) delegate.onTerminate();
}
@Override
public final void onConfigurationChanged(Configuration newConfig) {
super.onConfigurationChanged(newConfig);
if(delegate!=null) delegate.onConfigurationChanged(newConfig);
}
@Override
public final void onLowMemory() {
super.onLowMemory();
if(delegate!=null) delegate.onLowMemory();
}
@Override
public final void onTrimMemory(int level) {
super.onTrimMemory(level);
if(delegate!=null) delegate.onTrimMemory(level);
}
private void ensureDelegate(){
if(delegate==null){
try {
Class<?> delegateClass = Class.forName(delegateClassName,false,getClassLoader());
Constructor<?> constructor = delegateClass.getConstructor(Application.class);
delegate = (ApplicationLifeCycle) constructor.newInstance(this);
} catch (Exception e) {
throw new IllegalArgumentException("create app delegate failed",e);
}
}
}
}
4.1 反射注入补丁dex到ClassLoader中的数组前面
反射注入补丁dex到ClassLoader中的数组前面的方式和其它ClassLoader方案的注入方式差别不大,入口函数是PatchTool.installPatch(Application application,File patchFile),在此函数中先通过InjectUtil提供的函数注入补丁dex,再使用解析出来的class-id信息调用native函数
注入函数这里只简单列出
//InjectUtil类代码
//注入补丁Dex
static boolean injectDex(Application context, File patchFile) {
ArrayList<File> files = new ArrayList<File>();
files.add(patchFile);
try {
checkApkFiles(files);
if(isAliyunOs()){ //阿里云OS的注入
for(File file:files){
injectLexFile(context,file);
}
return true;
}else if(isAndroid()){ //普通android-os的注入
installDex(context,InjectUtil.class.getClassLoader(),context.getDir("dex", 0),files);
return true;
}
throw new IllegalStateException("Current system is not support");
} catch (Exception e) {
e.printStackTrace();
return false;
}
}
4.2 解析补丁apk中的补丁class-id信息
解析补丁apk中class-id信息的逻辑在PatchTool.readPatchClassIds函数中,代码如下
//class-ids.txt中的格式如下 第一行是入口类信息,以:分割,格式:Dex1入口类:Dex2入口类
//之后是class-id信息,格式:classname-dexIndex-classId
private static List<Pair<String,Long>> readPatchClassIds(File patchFile, String defaultEntranceClass){
List<Pair<String,Long>> classIds = new ArrayList<Pair<String,Long>>();
InputStream inputStream = null;
BufferedReader reader = null;
SparseArray<String> dexEntrances = new SparseArray<String>();
try {
JarFile jarFile = new JarFile(patchFile);
ZipEntry entry = jarFile.getEntry(CLASS_ID_TXT);
inputStream = jarFile.getInputStream(entry);
reader = new BufferedReader(new InputStreamReader(inputStream));
String line;
boolean isFirst = true;
while ((line=reader.readLine())!=null){
if(isFirst){
isFirst = false;
if(line.contains(":")){ //解析入口
String[] entrances = line.split(":");
for(int i=0;i<entrances.length;i++){
String entrance = entrances[i];
if(TextUtils.isEmpty(entrance))
continue;
dexEntrances.put(i+1,entrance);
}
boolean loadEntrance = loadEntranceClasses(dexEntrances);
if(!loadEntrance){
return null;
}
continue;
}
}
if (!TextUtils.isEmpty(line)) {
String[] infos = line.split("-");
if (infos.length == 3) {
long classId = Long.valueOf(infos[2]);
int dexIndex = Integer.valueOf(infos[1]);
String entrance = getEntranceClass(dexEntrances,dexIndex,defaultEntranceClass);
classIds.add(Pair.create(entrance,classId));
}
}
}
} catch (Exception e) {
//.....异常处理,关闭流
}
return classIds;
}
然后使用解析出来的class-id信息调用PatchTool.resolvePatchClass函数
private static boolean resolvePatchClass(Application app, String[] referrerClassList, long[] classIdxList, int size) {
if (!sIsLibLoaded) {
sIsLibLoaded = loadPatchToolLib();
}
if (!sIsLibLoaded) {
boolean unloadResult = InjectUtil.unloadPatchElement(app, 0);
Log.e(TAG, "load lib failed, unload patch result=" + unloadResult);
return false;
} else {
int resolveResult = nativeResolvePatchClass(referrerClassList, classIdxList, size);
if (resolveResult != CODE_RESOLVE_PATCH_ALL_SUCCESS) {
//resolve不成功从dex数组中卸载对应dex
boolean unloadResult = InjectUtil.unloadPatchElement(app, 0);
Log.e(TAG, String.format(Locale.ENGLISH,"resolve patch class failed, unload patch result= %b,refClass1:%s",
unloadResult,referrerClassList[0]));
return false;
} else {
Log.d(TAG, "resolve patch class success");
return true;
}
}
}
4.3 调用nativeResolveClass函数
上一节中最终会调用到nativeResolvePatchClass方法,这个方法是native方法,实现在qfixlib/src/main/jni/ResolvePatch.c中,这个C文件也是直接使用https://github.com/lizhangqu/QFix的中的C文件
jint Java_alexclin_patch_qfix_tool_PatchTool_nativeResolvePatchClass(JNIEnv* env,
jobject thiz, jobjectArray referrerClassList, jlongArray classIdxList, jint size) {
LOGI("enter nativeResolvePatchClass");
int referrerClassSize = (*env)->GetArrayLength(env, referrerClassList);
int classIdxSize = (*env)->GetArrayLength(env, classIdxList);
if (size <= 0 || referrerClassSize != size || classIdxSize != size) {
LOGE("CODE_NATIVE_INIT_PARAMETER_ERROR");
return CODE_NATIVE_INIT_PARAMETER_ERROR;
}
jlong* jClassIdxArray = (*env)->GetLongArrayElements(env, classIdxList, 0);
if (jClassIdxArray == 0) {
LOGE("CODE_NATIVE_INIT_PARAMETER_ERROR");
return CODE_NATIVE_INIT_PARAMETER_ERROR;
}
void* handle = 0;
handle = dlopen("/system/lib/libdvm.so", RTLD_LAZY);
if (handle) {
void* findFunc = 0;
int i = 0;
while(i < ARRAY_SIZE_FIND_CLASS) {
findFunc = dlsym(handle, ARRAY_SYMBOL_FIND_LOADED_CLASS[i]);
if (findFunc) {
break;
}
i++;
}
if (findFunc) {
g_pDvmFindLoadedClass_Addr = findFunc;
void* resolveFunc = 0;
i = 0;
while(i < ARRAY_SIZE_RESOLVE_CLASS) {
resolveFunc = dlsym(handle, ARRAY_SYMBOL_RESOLVE_CLASS[i]);
if (resolveFunc) {
break;
}
i++;
}
if (resolveFunc) {
g_pDvmResolveClass_Addr = resolveFunc;
i = 0;
while(i < size) {
jstring jClassItem = (jstring)((*env)->GetObjectArrayElement(env, referrerClassList, i));
const char* classItem = (*env)->GetStringUTFChars(env, jClassItem, 0);
if (classItem == 0) {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
LOGE("CODE_NATIVE_ITEM_PARAMETER_ERROR=%d", i);
return NUM_FACTOR_PATCH * i + CODE_NATIVE_ITEM_PARAMETER_ERROR;
}
if (strlen(classItem) < 5 || jClassIdxArray[i] < 0) {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
(*env)->ReleaseStringUTFChars(env, jClassItem, classItem);
LOGE("CODE_NATIVE_ITEM_PARAMETER_ERROR=%d", i);
return NUM_FACTOR_PATCH * i + CODE_NATIVE_ITEM_PARAMETER_ERROR;
}
void* referrerClassObj = g_pDvmFindLoadedClass_Addr(classItem);
if (referrerClassObj) {
void* resClassObj = g_pDvmResolveClass_Addr(referrerClassObj, (unsigned int)jClassIdxArray[i], 1);
if (!resClassObj) {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
(*env)->ReleaseStringUTFChars(env, jClassItem, classItem);
LOGE("CODE_PATCH_CLASS_OBJECT_ERROR=%d", i);
return NUM_FACTOR_PATCH * i + CODE_PATCH_CLASS_OBJECT_ERROR;
}
} else {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
(*env)->ReleaseStringUTFChars(env, jClassItem, classItem);
LOGE("CODE_REFERRER_CLASS_OBJECT_ERROR=%d", i);
return NUM_FACTOR_PATCH * i + CODE_REFERRER_CLASS_OBJECT_ERROR;
}
(*env)->ReleaseStringUTFChars(env, jClassItem, classItem);
i++;
}
} else {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
LOGE("CODE_RESOLVE_CLASS_ERROR");
return CODE_RESOLVE_CLASS_ERROR;
}
} else {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
LOGE("CODE_FIND_LOADED_CLASS_ERROR");
return CODE_FIND_LOADED_CLASS_ERROR;
}
} else {
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
LOGE("CODE_LOAD_DALVIK_SO_ERROR");
return CODE_LOAD_DALVIK_SO_ERROR;
}
(*env)->ReleaseLongArrayElements(env, classIdxList, jClassIdxArray, 0);
LOGI("CODE_RESOLVE_PATCH_ALL_SUCCESS");
return CODE_RESOLVE_PATCH_ALL_SUCCESS;
}
到这里加载流程就结束了。
五、总结
QFix方案原理和实现相对Tinker方案来说都简单不少,比较轻量级,不过Tinker实现比较重的同时可以实现功能级更新,而QFix方案更多是应对bug的修复。
QFixPatch这个项目算是QFix方案的一个gradle实践,实际开发过程中原创东西不多,主要工作还是在gradle插件。对于Android热修复我也是因为工作中有需求所以了解多一些。项目中有不完善的地方也欢迎各位同学来github提issue.
热修复目前主流是NativeHook,ClassLoader,InstantRun三种方案,NativeHook,ClassLoader在android7.0版本上都会有一些兼容性问题,相对来说InstantRun方案兼容性会更好,目前我也在研究学习Robust和Aceso的路上,欢迎有兴趣的同学一起来探讨。
本人Github:https://github.com/alexclin0188 欢迎关注
Stashed changes