没有办法CSDN博客只能写20万字左右，这里接着上一篇博客继续分析 。
用在线对比工具比对比，发现class字节码中有2处增加了内容，有一处由01 修改 成了02 ， 先来看增加的16进制转化为字符串后是什么 。
先来看 01 00 14 53 6F 75 72 63 65 44 65 62 75 67 45 78 74 65 6E 73 69 6F 6E
先来看常量池中UTF8常量结构
CONSTANT_Utf8_info {
u1 tag;
u2 length;
u1 bytes[length];
}
在常量池中第一个字节表示当前常量所属类型，之前我写了一个项目，叫自己手写虚拟机，项目中有一段下面截图中的代码，如第一个字节是1 ，表示utf8类型，tag = 7 ，则表示class类型， 10 为方法引用等
自己手写虚拟机博客
自己动手写Java虚拟机 (Java核心技术系列)_java版
还有一篇
Java字节码文件结构剖析（一） 关于字节码解析的，如果不懂，可以去看看这两篇博客 。
再来看 01 00 14 53 6F 75 72 63 65 44 65 62 75 67 45 78 74 65 6E 73 69 6F 6E ， 01 表示当前常量池 utf8类型， u2 表示后面两个字节表示utf8字符串的长度，00 14 从16进制转化为10进制为1 * 16 + 4 = 20，也就是向后取20个字节为53 6F 75 72 63 65 44 65 62 75 67 45 78 74 65 6E 73 69 6F 6E，刚好转化为字符串类型为SourceDebugExtension。
其他的，自己去查吧，从字节码的修改来看，是在常量池中添加了一个utf8的常量，并且常量的内容为SourceDebugExtension。
接下来，看第二段增加的内容，00 FB 00 00 00 82 53 4D 41 50 0A 69 6E 64 65 78 5F 6A 73 70 2E 6A 61 76 61 0A 4A 53 50 0A 2A 53 20 4A 53 50 0A 2A 46 0A 2B 20 30 20 69 6E 64 65 78 2E 6A 73 70 0A 69 6E 64 65 78 2E 6A 73 70 0A 2A 4C 0A 31 2C 39 3A 39 33 0A 39 3A 31 30 31 0A 31 31 2C 31 30 3A 31 30 33 0A 32 30 3A 31 31 33 0A 32 31 2C 32 3A 31 31 34 2C 33 0A 32 33 2C 35 3A 31 32 30 0A 32 37 3A 31 35 30 2C 39 0A 32 37 3A 31 32 36 0A 2A 45 0A， 既然常量池中加了SourceDebugExtension常量，那推测上面这段字节码和SourceDebugExtension属性有关，查一下他的结构 。
00 FB表示指向常量池的索引，转化为10进制为251
00 00 00 82 后面4个字节表示attribute_length 转化为10进制为130 。
53 4D 41 50 0A 69 6E 64 65 78 5F 6A 73 70 2E 6A 61 76 61 0A 4A 53 50 0A 2A 53 20 4A 53 50 0A 2A 46 0A 2B 20 30 20 69 6E 64 65 78 2E 6A 73 70 0A 69 6E 64 65 78 2E 6A 73 70 0A 2A 4C 0A 31 2C 39 3A 39 33 0A 39 3A 31 30 31 0A 31 31 2C 31 30 3A 31 30 33 0A 32 30 3A 31 31 33 0A 32 31 2C 32 3A 31 31 34 2C 33 0A 32 33 2C 35 3A 31 32 30 0A 32 37 3A 31 35 30 2C 39 0A 32 37 3A 31 32 36 0A 2A 45 0A 其余的转化为字符串为
SMAP
index_jsp.java
JSP
*S JSP
*F
- 0 index.jsp
index.jsp
*L
1,9:93
9:101
11,10:103
20:113
21,2:114,3
23,5:120
27:150,9
27:126
*E
现在应该比较明白了吧。 那下面红框里的这一段00 F9 00 00 00 02 00 FA 字节码又表示的是什么呢？
之前我们见过SourceFile的属性结构如下。
00 F9 ：F9 指向常量池的索引，转化为10进制为249
00 00 00 02 ： 属性长度占4个字节 2 ，因此向后读取两个字节。
00 FA ：指向常量池中的索引 250
有人会说，你说得好听的，但是怎么来证明呢？ 我们用jclasslib bytecode viewer打印修改过的class文件 。
是不是和我们之前的分析一模一样，当然啦，上面是对字节码的添加做的事情，下面来分析对字节码从01 修改为02 又是什么意思呢？还是借助于之前手动写java虚拟机的代码来看。
因此进入readAttributes()方法 。
java字节码规定，在读取属性表之前，有2个字节标识当前属性表长度，也就是原文件的00 01 和 修改后00 02表示属性表长度，显然之前只有SourceFile属性，现在增加了SourceDebugExtension属性信息，所以需要将原来的长度由1 修改为2 。
到这里，我们来总结一下 static void install(File classFile, byte[] smap) 到底对字节码做了哪些事情 。
- 在常量池中增加Utf8的SourceDebugExtension常量。
- 修改属性表长度
- 在属性表中添加SourceDebugExtension属性。
- 当然常量池的个数也做了修改，这里就不分析了。 一样的道理 。
从结果来看，tomcat对原字节码做了以上修改，接下来看代码如何实现，不过在看这段代码之前，继续回顾一下class文件的字节码结构 。
ClassFile {
u4 magic;
u2 minor_version;
u2 major_version;
u2 constant_pool_count;
cp_info constant_pool[constant_pool_count-1];
u2 access_flags;
u2 this_class;
u2 super_class;
u2 interfaces_count;
u2 interfaces[interfaces_count];
u2 fields_count;
field_info fields[fields_count];
u2 methods_count;
method_info methods[methods_count];
u2 attributes_count;
attribute_info attributes[attributes_count];
}
有了字节码结构，再来看源码就清淅多了。
void addSDE() throws UnsupportedEncodingException, IOException { // 我们知道前面4个字节是魔数，2 个字节是次版本号，2个字节是主版本号， // 这里主要是复制前面8个字节 copy(4 + 2 + 2); int constantPoolCountPos = genPos; // 再读取两个字节，为常量池个数 int constantPoolCount = readU2(); if (log.isDebugEnabled()) log.debug("constant pool count: " + constantPoolCount); // 将常量池个数以字节码形式写到gen的byte数组中 writeU2(constantPoolCount); // copy old constant pool return index of SDE symbol, if found // 复制常量池，如果返回值sdeIndex > 0 ，则表示常量池中已经有SourceDebugExtension // 的UTF8常量 sdeIndex = copyConstantPool(constantPoolCount); if (sdeIndex < 0) { // if "SourceDebugExtension" symbol not there add it writeUtf8ForSDE(); // increment the constantPoolCount sdeIndex = constantPoolCount; ++constantPoolCount; // 修改常量池个数constantPoolCountPos = constantPoolCountPos + 1 randomAccessWriteU2(constantPoolCountPos, constantPoolCount); if (log.isDebugEnabled()) log.debug("SourceDebugExtension not found, installed at: " + sdeIndex); } else { if (log.isDebugEnabled()) log.debug("SourceDebugExtension found at: " + sdeIndex); } // access, this, super // access_flags 2 个字节指向常量池索引，当前类的访问标识 // this_class 2个字节指向当前类 // super_class 2 个字节指向父类，这些都不需要修改，直接复制即可 copy(2 + 2 + 2); //2 个字节，表示当前类实现接口数 int interfaceCount = readU2(); // 将接口数以2个字节写到新byte[]数组中 writeU2(interfaceCount); if (log.isDebugEnabled()) log.debug("interfaceCount: " + interfaceCount); // 因为接口也是指向常量池中的索引 // interfaces[interfaces_count] ，而指向常量池的索引占两个字节 // 因此最终需要复制的字节数 = 接口数 * 2 copy(interfaceCount * 2); copyMembers(); // 复制类中所有field属性 copyMembers(); // 复制类中所有的方法表 int attrCountPos = genPos; // 获取属性个数 attributes_count ，当然属性个数占两个字节 int attrCount = readU2(); writeU2(attrCount); if (log.isDebugEnabled()) log.debug("class attrCount: " + attrCount); // copy the class attributes, return true if SDE attr found (not copied) // 如果属性中没有 SourceDebugExtension，则增加个数 if (!copyAttrs(attrCount)) { // 修改attributes_count = attributes_count + 1 ++attrCount; // 将修改的值写到attributes_count 对应的字节码位置 randomAccessWriteU2(attrCountPos, attrCount); if (log.isDebugEnabled()) log.debug("class attrCount incremented"); } // sdeIndex 表示SourceDebugExtension的UTF8所在常量池中索引 ，将smap解析得到的行号对应关系写到字节码中 writeAttrForSDE(sdeIndex); }
下面来看复制常量池方法 ，不过在看常量池结构之前先来看常量池中每种类型及其含义，以及每种类型所点字节。
int copyConstantPool(int constantPoolCount) throws UnsupportedEncodingException, IOException { int sdeIndex = -1; // copy const pool index zero not in class file for (int i = 1; i < constantPoolCount; ++i) { int tag = readU1(); writeU1(tag); switch (tag) { case 7 : // Class case 8 : // String case 16 : // MethodType if (log.isDebugEnabled()) log.debug(i + " copying 2 bytes"); copy(2); break; case 15 : // MethodHandle if (log.isDebugEnabled()) log.debug(i + " copying 3 bytes"); copy(3); break; case 9 : // Field case 10 : // Method case 11 : // InterfaceMethod case 3 : // Integer case 4 : // Float case 12 : // NameAndType case 18 : // InvokeDynamic if (log.isDebugEnabled()) log.debug(i + " copying 4 bytes"); copy(4); break; case 5 : // Long case 6 : // Double if (log.isDebugEnabled()) log.debug(i + " copying 8 bytes"); copy(8); i++; break; case 1 : // Utf8 int len = readU2(); writeU2(len); byte[] utf8 = readBytes(len); String str = new String(utf8, "UTF-8"); if (log.isDebugEnabled()) log.debug(i + " read class attr -- '" + str + "'"); if (str.equals("SourceDebugExtension")) { sdeIndex = i; } writeBytes(utf8); break; default : throw new IOException("unexpected tag: " + tag); } } return sdeIndex; }
简单分析一下，如CONSTANT_Long_info，CONSTANT_Double_info类型结构为
{
tag u1
bytes u8
}
为什么复制字节码时是copy(8);而不是copy(9)呢，大家注意
int tag = readU1();
writeU1(tag);
这两行代码，tag = readU1(); 读取到tag标识当前常量属于常量池中哪种类型，并且将tag写到新的byte[] 数组中 ，因此后面只需要拷贝8个字节，并且有一行代码 if (str.equals(“SourceDebugExtension”)) 判断，如果当前UTF8类型是SourceDebugExtension，则不需要向常量池中写入UTF8的SourceDebugExtension类型了。 并将UTF8类型是SourceDebugExtension 常量的索引返回，方便写入attributeInfo时引用此索引 。
如果常量池中没有UTF8类型是SourceDebugExtension 常量，则调用writeUtf8ForSDE()向常量池中写入。
void writeUtf8ForSDE() { int len = "SourceDebugExtension".length(); writeU1(1); // 1 表示是UTF8常量 writeU2(len); // 2个字节表示常量字符串的长度 // 将 "SourceDebugExtension"字符串以16进制写入新byte[]数组中 for (int i = 0; i < len; ++i) { writeU1("SourceDebugExtension".charAt(i)); } }
修改常量池个数方法。从pos位置开始，将val转化为两个字节，从pos位置写入，再恢复之前的写入位置。
void randomAccessWriteU2(int pos, int val) { // 保存之前的位置 int savePos = genPos; genPos = pos; // 修改pos位置2个字节的内容 writeU2(val); // 恢复成之前的位置 genPos = savePos; }
将smap解析得到的jsp行号与生成java代码的行号对应关系写到字节码中。
void writeAttrForSDE(int index) { // 2 个字节指向常量池中SourceDebugExtension writeU2(index); // 用4个字节存储smap信息 writeU4(sdeAttr.length); // 将smap中信息存储到新的byte[]数组中 for (int i = 0; i < sdeAttr.length; ++i) { writeU1(sdeAttr[i]); } }
下面这个方法中需要注意一点，当attribute_info中有SourceDebugExtension时，是不会复制之前的SourceDebugExtension信息的，因此attrCount是不需要++，直接将新的SourceDebugExtension替代掉之前的SourceDebugExtension即可，这也是attrCount 不需要++，直接调用writeAttrForSDE()就可以的原因。
boolean copyAttrs(int attrCount) { boolean sdeFound = false; for (int i = 0; i < attrCount; ++i) { int nameIndex = readU2(); // don't write old SDE if (nameIndex == sdeIndex) { sdeFound = true; if (log.isDebugEnabled()) log.debug("SDE attr found"); } else { writeU2(nameIndex); // name int len = readU4(); writeU4(len); copy(len); if (log.isDebugEnabled()) log.debug("attr len: " + len); } } return sdeFound; }
我相信到这里，大家肯定已经明白smap是怎样被加到字节码中去了，那SourceDebugExtension加到字节码中有什么用呢？看了
《Java虚拟机规范.Java SE 8版》的第101页，也没有看到如何使用。
难道是在代码抛出异常时，定位异常用的不？来看一个例子。
生成的index_jsp.java的代码是
再看_jspx_page_context是个什么对象。
public void _jspService(final javax.servlet.http.HttpServletRequest request, final javax.servlet.http.HttpServletResponse response)
throws java.io.IOException, javax.servlet.ServletException {
final javax.servlet.jsp.PageContext pageContext;
javax.servlet.http.HttpSession session = null;
final javax.servlet.ServletContext application;
final javax.servlet.ServletConfig config;
javax.servlet.jsp.JspWriter out = null;
final java.lang.Object page = this;
javax.servlet.jsp.JspWriter _jspx_out = null;
javax.servlet.jsp.PageContext _jspx_page_context = null;
try {
response.setContentType("text/html;charset=UTF-8");
pageContext = _jspxFactory.getPageContext(this, request, response,
null, true, 8192, true);
_jspx_page_context = pageContext;
application = pageContext.getServletContext();
config = pageContext.getServletConfig();
session = pageContext.getSession();
out = pageContext.getOut();
_jspx_out = out;
...
从生成的jsp代码中得知，_jspx_page_context = _jspxFactory.getPageContext(this, request, response,
null, true, 8192, true);，那_jspxFactory又是怎样来的呢？从生成的jsp文件中得知。
private static final javax.servlet.jsp.JspFactory _jspxFactory =
javax.servlet.jsp.JspFactory.getDefaultFactory();
进入JspFactory类
public static void setDefaultFactory(JspFactory deflt) { JspFactory.deflt = deflt; } public static JspFactory getDefaultFactory() { return deflt; }
发现_jspxFactory最终又来源于setDefaultFactory()方法 ，查找一下，发现setDefaultFactory()方法又在JspRuntimeContext的静态初始化方法中调用了。
static {
JspFactoryImpl factory = new JspFactoryImpl();
SecurityClassLoad.securityClassLoad(factory.getClass().getClassLoader());
if( System.getSecurityManager() != null ) {
String basePackage = "org.apache.jasper.";
try {
...
} catch (ClassNotFoundException ex) {
throw new IllegalStateException(ex);
}
}
JspFactory.setDefaultFactory(factory);
}
最终得到_jspxFactory实际上是JspFactoryImpl，而进入其getPageContext()方法 。
public PageContext getPageContext(Servlet servlet, ServletRequest request, ServletResponse response, String errorPageURL, boolean needsSession, int bufferSize, boolean autoflush) { if( Constants.IS_SECURITY_ENABLED ) { PrivilegedGetPageContext dp = new PrivilegedGetPageContext( this, servlet, request, response, errorPageURL, needsSession, bufferSize, autoflush); return AccessController.doPrivileged(dp); } else { return internalGetPageContext(servlet, request, response, errorPageURL, needsSession, bufferSize, autoflush); } } private PageContext internalGetPageContext(Servlet servlet, ServletRequest request, ServletResponse response, String errorPageURL, boolean needsSession, int bufferSize, boolean autoflush) { try { PageContext pc; if (USE_POOL) { PageContextPool pool = localPool.get(); if (pool == null) { pool = new PageContextPool(); localPool.set(pool); } pc = pool.get(); if (pc == null) { pc = new PageContextImpl(); } } else { pc = new PageContextImpl(); } pc.initialize(servlet, request, response, errorPageURL, needsSession, bufferSize, autoflush); return pc; } catch (Throwable ex) { ExceptionUtils.handleThrowable(ex); if (ex instanceof RuntimeException) { throw (RuntimeException) ex; } log.fatal("Exception initializing page context", ex); return null; } }
最终得到PageContext实际上是PageContextImpl对象，而其private void doHandlePageException(Throwable t)方法如下 。
private void doHandlePageException(Throwable t) throws IOException,
ServletException {
if (errorPageURL != null && !errorPageURL.equals("")) {
/*
* Set request attributes. Do not set the
* javax.servlet.error.exception attribute here (instead, set in the
* generated servlet code for the error page) in order to prevent
* the ErrorReportValve, which is invoked as part of forwarding the
* request to the error page, from throwing it if the response has
* not been committed (the response will have been committed if the
* error page is a JSP page).
*/
request.setAttribute(PageContext.EXCEPTION, t);
request.setAttribute(RequestDispatcher.ERROR_STATUS_CODE,
Integer.valueOf(HttpServletResponse.SC_INTERNAL_SERVER_ERROR));
request.setAttribute(RequestDispatcher.ERROR_REQUEST_URI,
((HttpServletRequest) request).getRequestURI());
request.setAttribute(RequestDispatcher.ERROR_SERVLET_NAME,
config.getServletName());
try {
forward(errorPageURL);
} catch (IllegalStateException ise) {
include(errorPageURL);
}
// The error page could be inside an include.
Object newException =
request.getAttribute(RequestDispatcher.ERROR_EXCEPTION);
// t==null means the attribute was not set.
if ((newException != null) && (newException == t)) {
request.removeAttribute(RequestDispatcher.ERROR_EXCEPTION);
}
// now clear the error code - to prevent double handling.
request.removeAttribute(RequestDispatcher.ERROR_STATUS_CODE);
request.removeAttribute(RequestDispatcher.ERROR_REQUEST_URI);
request.removeAttribute(RequestDispatcher.ERROR_SERVLET_NAME);
request.removeAttribute(PageContext.EXCEPTION);
} else {
// Otherwise throw the exception wrapped inside a ServletException.
// Set the exception as the root cause in the ServletException
// to get a stack trace for the real problem
if (t instanceof IOException)
throw (IOException) t;
if (t instanceof ServletException)
throw (ServletException) t;
if (t instanceof RuntimeException)
throw (RuntimeException) t;
Throwable rootCause = null;
if (t instanceof JspException) {
rootCause = ((JspException) t).getCause();
} else if (t instanceof ELException) {
rootCause = ((ELException) t).getCause();
}
if (rootCause != null) {
throw new ServletException(t.getClass().getName() + ": "
+ t.getMessage(), rootCause);
}
throw new ServletException(t);
}
}
因为没有配置错误页面，因此最终进入上面的加粗代码 ，抛出java.lang.ArithmeticException: / by zero 异常。 
在servlet.service()方法调用时，抛出了java.lang.ArithmeticException: / by zero 异常，最终被handleJspException()方法捕获到。
进入 handleJspException(Exception ex)方法 。
protected JasperException handleJspException(Exception ex) {
try {
Throwable realException = ex;
if (ex instanceof ServletException) {
realException = ((ServletException) ex).getRootCause();
}
// 获取到异常堆栈
StackTraceElement[] frames = realException.getStackTrace();
StackTraceElement jspFrame = null;
for (int i=0; i<frames.length; ++i) {
// 遍历所有的异常堆栈，如果栈中的类名与servlet中类名一样，则所在的栈帧的行号就是生成_jsp.java文件抛出异常所在的java行号
if ( frames[i].getClassName().equals(this.getServlet().getClass().getName()) ) {
jspFrame = frames[i];
break;
}
}
if (jspFrame == null ||
this.ctxt.getCompiler().getPageNodes() == null) {
// If we couldn't find a frame in the stack trace corresponding
// to the generated servlet class or we don't have a copy of the
// parsed JSP to hand, we can't really add anything
return new JasperException(ex);
}
// 获取抛出异常的java 行号
int javaLineNumber = jspFrame.getLineNumber();
// 创建异常详细信息
JavacErrorDetail detail = ErrorDispatcher.createJavacError(
jspFrame.getMethodName(),
this.ctxt.getCompiler().getPageNodes(),
null,
javaLineNumber,
ctxt);
// If the line number is less than one we couldn't find out
// where in the JSP things went wrong
int jspLineNumber = detail.getJspBeginLineNumber();
if (jspLineNumber < 1) {
throw new JasperException(ex);
}
if (options.getDisplaySourceFragment()) {
return new JasperException(Localizer.getMessage
("jsp.exception", detail.getJspFileName(),
"" + jspLineNumber) + Constants.NEWLINE +
Constants.NEWLINE + detail.getJspExtract() +
Constants.NEWLINE + Constants.NEWLINE +
"Stacktrace:", ex);
}
return new JasperException(Localizer.getMessage
("jsp.exception", detail.getJspFileName(),
"" + jspLineNumber), ex);
} catch (Exception je) {
// If anything goes wrong, just revert to the original behaviour
if (ex instanceof JasperException) {
return (JasperException) ex;
}
return new JasperException(ex);
}
}
如何得到异常的详细信息呢？进入createJavacError()方法 。
public static JavacErrorDetail createJavacError(String fname, Node.Nodes page, StringBuilder errMsgBuf, int lineNum, JspCompilationContext ctxt) throws JasperException { JavacErrorDetail javacError; // Attempt to map javac error line number to line in JSP page ErrorVisitor errVisitor = new ErrorVisitor(lineNum); page.visit(errVisitor); // 得到抛出异常的Node Node errNode = errVisitor.getJspSourceNode(); if ((errNode != null) && (errNode.getStart() != null)) { // If this is a scriplet node then there is a one to one mapping // between JSP lines and Java lines if (errVisitor.getJspSourceNode() instanceof Node.Scriptlet || errVisitor.getJspSourceNode() instanceof Node.Declaration) { javacError = new JavacErrorDetail( fname, lineNum, errNode.getStart().getFile(), // 通过java代码抛出异常的行号计算出jsp抛出异常行号 errNode.getStart().getLineNumber() + lineNum - errVisitor.getJspSourceNode().getBeginJavaLine(), errMsgBuf, ctxt); } else { javacError = new JavacErrorDetail( fname, lineNum, errNode.getStart().getFile(), errNode.getStart().getLineNumber(), errMsgBuf, ctxt); } } else { /* * javac error line number cannot be mapped to JSP page * line number. For example, this is the case if a * scriptlet is missing a closing brace, which causes * havoc with the try-catch-finally block that the code * generator places around all generated code: As a result * of this, the javac error line numbers will be outside * the range of begin and end java line numbers that were * generated for the scriptlet, and therefore cannot be * mapped to the start line number of the scriptlet in the * JSP page. * Include just the javac error info in the error detail. */ javacError = new JavacErrorDetail( fname, lineNum, errMsgBuf); } return javacError; }
先通过调用ErrorVisitor的访问者模式，得到抛出异常的Node，我们之前不是分析过，jsp最终被解析成一个一个Node，接下来看ErrorVisitor内部做了哪些事情 。
static class ErrorVisitor extends Node.Visitor {
// Java source line number to be mapped
private int lineNum;
/*
* JSP node whose Java source code range in the generated servlet
* contains the Java source line number to be mapped
*/
Node found;
/*
* Constructor.
*
* @param lineNum Source line number in the generated servlet code
*/
public ErrorVisitor(int lineNum) {
this.lineNum = lineNum;
}
@Override
public void doVisit(Node n) throws JasperException {
if ((lineNum >= n.getBeginJavaLine())
&& (lineNum < n.getEndJavaLine())) {
found = n;
}
}
/*
* Gets the JSP node to which the source line number in the generated
* servlet code was mapped.
*
* @return JSP node to which the source line number in the generated
* servlet code was mapped
*/
public Node getJspSourceNode() {
return found;
}
}
上面代码重点就是加粗代码，如果抛出异常的行号在Node的java代码开始行和结束行之间，那么抛出异常的地方肯定在这个Node中。在JavacErrorDetail对象创建之前，有一个变量 errNode.getStart().getLineNumber() + lineNum -
errVisitor.getJspSourceNode().getBeginJavaLine()，这个是什么意思呢？ 如Node节点的java开始行为102 ， 对应的jsp 代码是第10行，而java 代码抛出异常的行号是109 ，那么jsp中抛出异常的行号是多少呢？ 不就是109 - 102 + 10 = 17 不？因此这个变量就是计算jsp抛出异常的代码行号。
接下来进入JavacErrorDetail的构造函数 。
public JavacErrorDetail(String javaFileName,
int javaLineNum,
String jspFileName,
int jspBeginLineNum,
StringBuilder errMsg,
JspCompilationContext ctxt) {
this(javaFileName, javaLineNum, errMsg);
this.jspFileName = jspFileName;
this.jspBeginLineNum = jspBeginLineNum;
if (jspBeginLineNum > 0 && ctxt != null) {
InputStream is = null;
FileInputStream fis = null;
try {
// Read both files in, so we can inspect them
JarFile jarFile = null;
JarResource tagJarResource = ctxt.getTagFileJarResource();
if (tagJarResource != null) {
jarFile = tagJarResource.getJarFile();
}
is = JspUtil.getInputStream(jspFileName, jarFile, ctxt, null);
// 读取jsp文件
String[] jspLines = readFile(is);
fis = new FileInputStream(ctxt.getServletJavaFileName());
// 读取生成的java文件
String[] javaLines = readFile(fis);
if (jspLines.length < jspBeginLineNum) {
// Avoid ArrayIndexOutOfBoundsException
// Probably bug 48498 but could be some other cause
jspExtract = Localizer.getMessage("jsp.error.bug48498");
return;
}
// If the line contains the opening of a multi-line scriptlet
// block, then the JSP line number we got back is probably
// faulty. Scan forward to match the java line...
// 正常情况都是
// <%
// int a = 0;
// int b = 1 / 0 ;
// %>
// 但如果出现
//
// int a = 0 ;
// %>
// <% int b = 1/0;
// 时， 需要对Node对应的jsp代码行重新遍历，直到jsp和java行号相等，重新更新
// jspBeginLineNum 行号
if (jspLines[jspBeginLineNum-1].lastIndexOf("<%") >
jspLines[jspBeginLineNum-1].lastIndexOf("%>")) {
String javaLine = javaLines[javaLineNum-1].trim();
for (int i=jspBeginLineNum-1; i<jspLines.length; i++) {
if (jspLines[i].indexOf(javaLine) != -1) {
// Update jsp line number
this.jspBeginLineNum = i+1;
break;
}
}
}
// copy out a fragment of JSP to display to the user
StringBuilder fragment = new StringBuilder(1024);
// 打印出抛出异常行号 -4 到 抛出异常行号 + 2 之间的jsp 代码
// 更加有利于开发者定位问题
int startIndex = Math.max(0, this.jspBeginLineNum-1-3);
int endIndex = Math.min(
jspLines.length-1, this.jspBeginLineNum-1+3);
for (int i=startIndex;i<=endIndex; ++i) {
// jsp 对应的行号， 但这里 i + 1 ，实际上打印的是 抛出异常代码行号 -3 ~ + 3之间的jsp代码
fragment.append(i+1);
fragment.append(": ");
// jsp对应行号 jsp 代码
fragment.append(jspLines[i]);
fragment.append(Constants.NEWLINE);
}
jspExtract = fragment.toString();
} catch (JasperException je) {
// Exception is never thrown - ignore
} catch (IOException ioe) {
// Can't read files - ignore
} finally {
if (is != null) {
try {
is.close();
} catch (IOException ioe) {
// Ignore
}
}
if (fis != null) {
try {
fis.close();
} catch (IOException ioe) {
// Ignore
}
}
}
}
}
关于jsp抛出异常的细节部分都已经分析过了。 最终通过JasperException异常抛出到前端页面 。
return new JasperException(Localizer.getMessage
"jsp.exception", detail.getJspFileName(),
"" + jspLineNumber) + Constants.NEWLINE +
Constants.NEWLINE + detail.getJspExtract() +
Constants.NEWLINE + Constants.NEWLINE +
"Stacktrace:", ex);
我们分析了那么多，终于知道当jsp文件抛出异常时，tomcat是如何处理，但遗憾的是，依然不知道SourceDebugExtension怎样用。
从网上资料显示 SourceDebugExtension属性用于存储额外的代码调试信息。 典型的场景是在进行JSP文件调试时， 没法经过Java堆栈来定位到JSP文件的行号。 可能是用来做调试用的吧。关于smap的分析到这里了，接下来，使是生成的jsp代码编译的分析 。
16.2.2 JDT Compiler编译器
通过JSP编译器编译后，生成了对应的Java文件，接下来， 要把Java文件编译成Class文件，对于 这部分，完全没有必要重复造轮子，常见的优秀编译工具有Eclispe JDT Java编译器和Ant编译器，Tomcat 其实同时支持两个编译器， 通过配置可以选择，而默认使用Eclipse JDT 编译器。
通过调用这些现在的编译器的API ，就可以方便的实现对Java 文件的编译，由于两个编译器的功能一样，因此就选 默认的Eclipse JDT 编译器，下面看如何用Eclipse JDT 编译器编译java文件 。
Eclipse JDT 提供了Compiler 类用于编译，它的构造函数比较复杂，如下所示 ，其实，实现自定义的构造函数包含的参数即基本完成了编译工作 。
public Compiler {
INameEnvironment environment ;
IErrorHandlingPolicy policy ;
CompilerOptions options ;
final ICompilerRequestor requestor ,
IProblemFactory problemFactory ;
}
为了方便说明，直接给出一个简单的编译实现，如下图所示 。
/**
* JDT class compiler. This compiler will load source dependencies from the
* context classloader, reducing dramatically disk access during
* the compilation process.
*
* @author Cocoon2
* @author Remy Maucherat
*/
public class JDTCompiler extends org.apache.jasper.compiler.Compiler {
private static final String JDT_JAVA_9_VERSION;
static {
// The constant for Java 9 changed between 4.6 and 4.7 in a way that is
// not backwards compatible. Need to figure out which version is in use
// so the correct constant value is used.
String jdtJava9Version = null;
Class<?> clazz = CompilerOptions.class;
for (Field field : clazz.getFields()) {
if ("VERSION_9".equals(field.getName())) {
// 4.7 onwards: CompilerOptions.VERSION_9
jdtJava9Version = "9";
break;
}
}
if (jdtJava9Version == null) {
// 4.6 and earlier: CompilerOptions.VERSION_1_9
jdtJava9Version = "1.9";
}
JDT_JAVA_9_VERSION = jdtJava9Version;
}
private final Log log = LogFactory.getLog(JDTCompiler.class); // must not be static
/**
* Compile the servlet from .java file to .class file
*/
@Override
protected void generateClass(String[] smap)
throws FileNotFoundException, JasperException, Exception {
long t1 = 0;
if (log.isDebugEnabled()) {
t1 = System.currentTimeMillis();
}
final String sourceFile = ctxt.getServletJavaFileName();
final String outputDir = ctxt.getOptions().getScratchDir().getAbsolutePath();
String packageName = ctxt.getServletPackageName();
final String targetClassName =
((packageName.length() != 0) ? (packageName + ".") : "")
+ ctxt.getServletClassName();
final ClassLoader classLoader = ctxt.getJspLoader();
String[] fileNames = new String[] {sourceFile};
String[] classNames = new String[] {targetClassName};
final ArrayList<JavacErrorDetail> problemList =
new ArrayList<JavacErrorDetail>();
class CompilationUnit implements ICompilationUnit {
private final String className;
private final String sourceFile;
CompilationUnit(String sourceFile, String className) {
this.className = className;
this.sourceFile = sourceFile;
}
@Override
public char[] getFileName() {
return sourceFile.toCharArray();
}
@Override
public char[] getContents() {
char[] result = null;
FileInputStream is = null;
InputStreamReader isr = null;
Reader reader = null;
try {
is = new FileInputStream(sourceFile);
isr = new InputStreamReader(is,
ctxt.getOptions().getJavaEncoding());
reader = new BufferedReader(isr);
char[] chars = new char[8192];
StringBuilder buf = new StringBuilder();
int count;
while ((count = reader.read(chars, 0,
chars.length)) > 0) {
buf.append(chars, 0, count);
}
result = new char[buf.length()];
buf.getChars(0, result.length, result, 0);
} catch (IOException e) {
log.error("Compilation error", e);
} finally {
if (reader != null) {
try {
reader.close();
} catch (IOException ioe) {/*Ignore*/}
}
if (isr != null) {
try {
isr.close();
} catch (IOException ioe) {/*Ignore*/}
}
if (is != null) {
try {
is.close();
} catch (IOException exc) {/*Ignore*/}
}
}
return result;
}
@Override
public char[] getMainTypeName() {
int dot = className.lastIndexOf('.');
if (dot > 0) {
return className.substring(dot + 1).toCharArray();
}
return className.toCharArray();
}
@Override
public char[][] getPackageName() {
StringTokenizer izer =
new StringTokenizer(className, ".");
char[][] result = new char[izer.countTokens()-1][];
for (int i = 0; i < result.length; i++) {
String tok = izer.nextToken();
result[i] = tok.toCharArray();
}
return result;
}
@Override
public boolean ignoreOptionalProblems() {
return false;
}
}
final INameEnvironment env = new INameEnvironment() {
@Override
public NameEnvironmentAnswer
findType(char[][] compoundTypeName) {
StringBuilder result = new StringBuilder();
String sep = "";
for (int i = 0; i < compoundTypeName.length; i++) {
result.append(sep);
result.append(compoundTypeName[i]);
sep = ".";
}
return findType(result.toString());
}
@Override
public NameEnvironmentAnswer
findType(char[] typeName,
char[][] packageName) {
StringBuilder result = new StringBuilder();
String sep = "";
for (int i = 0; i < packageName.length; i++) {
result.append(sep);
result.append(packageName[i]);
sep = ".";
}
result.append(sep);
result.append(typeName);
return findType(result.toString());
}
private NameEnvironmentAnswer findType(String className) {
InputStream is = null;
try {
if (className.equals(targetClassName)) {
ICompilationUnit compilationUnit =
new CompilationUnit(sourceFile, className);
return
new NameEnvironmentAnswer(compilationUnit, null);
}
String resourceName =
className.replace('.', '/') + ".class";
is = classLoader.getResourceAsStream(resourceName);
if (is != null) {
byte[] classBytes;
byte[] buf = new byte[8192];
ByteArrayOutputStream baos =
new ByteArrayOutputStream(buf.length);
int count;
while ((count = is.read(buf, 0, buf.length)) > 0) {
baos.write(buf, 0, count);
}
baos.flush();
classBytes = baos.toByteArray();
char[] fileName = className.toCharArray();
ClassFileReader classFileReader =
new ClassFileReader(classBytes, fileName,
true);
return
new NameEnvironmentAnswer(classFileReader, null);
}
} catch (IOException exc) {
log.error("Compilation error", exc);
} catch (org.eclipse.jdt.internal.compiler.classfmt.ClassFormatException exc) {
log.error("Compilation error", exc);
} finally {
if (is != null) {
try {
is.close();
} catch (IOException exc) {
// Ignore
}
}
}
return null;
}
private boolean isPackage(String result) {
if (result.equals(targetClassName)) {
return false;
}
String resourceName = result.replace('.', '/') + ".class";
InputStream is = null;
try {
is = classLoader.getResourceAsStream(resourceName);
return is == null;
} finally {
if (is != null) {
try {
is.close();
} catch (IOException e) {
}
}
}
}
@Override
public boolean isPackage(char[][] parentPackageName,
char[] packageName) {
StringBuilder result = new StringBuilder();
String sep = "";
if (parentPackageName != null) {
for (int i = 0; i < parentPackageName.length; i++) {
result.append(sep);
result.append(parentPackageName[i]);
sep = ".";
}
}
if (Character.isUpperCase(packageName[0])) {
if (!isPackage(result.toString())) {
return false;
}
}
result.append(sep);
result.append(packageName);
return isPackage(result.toString());
}
@Override
public void cleanup() {
}
};
final IErrorHandlingPolicy policy =
DefaultErrorHandlingPolicies.proceedWithAllProblems();
final Map<String,String> settings = new HashMap<String,String>();
settings.put(CompilerOptions.OPTION_LineNumberAttribute,
CompilerOptions.GENERATE);
settings.put(CompilerOptions.OPTION_SourceFileAttribute,
CompilerOptions.GENERATE);
settings.put(CompilerOptions.OPTION_ReportDeprecation,
CompilerOptions.IGNORE);
if (ctxt.getOptions().getJavaEncoding() != null) {
settings.put(CompilerOptions.OPTION_Encoding,
ctxt.getOptions().getJavaEncoding());
}
if (ctxt.getOptions().getClassDebugInfo()) {
settings.put(CompilerOptions.OPTION_LocalVariableAttribute,
CompilerOptions.GENERATE);
}
// Source JVM
if(ctxt.getOptions().getCompilerSourceVM() != null) {
String opt = ctxt.getOptions().getCompilerSourceVM();
if(opt.equals("1.1")) {
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_1);
} else if(opt.equals("1.2")) {
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_2);
} else if(opt.equals("1.3")) {
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_3);
...
} else {
log.warn(Localizer.getMessage("jsp.warning.unknown.sourceVM", opt));
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_6);
}
} else {
// Default to 1.6
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_6);
}
// Target JVM
if(ctxt.getOptions().getCompilerTargetVM() != null) {
String opt = ctxt.getOptions().getCompilerTargetVM();
settings.put(CompilerOptions.OPTION_Compliance,
JDT_JAVA_9_VERSION);
} else if(opt.equals("10")) {
// Constant not available in latest ECJ version that runs on
// Java 6.
// This is checked against the actual version below.
settings.put(CompilerOptions.OPTION_TargetPlatform, "10");
settings.put(CompilerOptions.OPTION_Compliance, "10");
} else if(opt.equals("11")) {
// Constant not available in latest ECJ version that runs on
// Java 6.
// This is checked against the actual version below.
settings.put(CompilerOptions.OPTION_TargetPlatform, "11");
settings.put(CompilerOptions.OPTION_Compliance, "11");
} else if(opt.equals("12")) {
// Constant not available in latest available ECJ version.
// May be supported in a snapshot build.
// This is checked against the actual version below.
settings.put(CompilerOptions.OPTION_TargetPlatform, "12");
settings.put(CompilerOptions.OPTION_Compliance, "12");
} else if(opt.equals("13")) {
// Constant not available in latest available ECJ version.
// May be supported in a snapshot build.
// This is checked against the actual version below.
settings.put(CompilerOptions.OPTION_TargetPlatform, "13");
settings.put(CompilerOptions.OPTION_Compliance, "13");
} else {
log.warn(Localizer.getMessage("jsp.warning.unknown.targetVM", opt));
settings.put(CompilerOptions.OPTION_TargetPlatform,
CompilerOptions.VERSION_1_6);
}
} else {
// Default to 1.6
settings.put(CompilerOptions.OPTION_TargetPlatform,
...
CompilerOptions.VERSION_1_6);
settings.put(CompilerOptions.OPTION_Compliance,
CompilerOptions.VERSION_1_6);
}
final IProblemFactory problemFactory =
new DefaultProblemFactory(Locale.getDefault());
final ICompilerRequestor requestor = new ICompilerRequestor() {
@Override
public void acceptResult(CompilationResult result) {
try {
if (result.hasProblems()) {
IProblem[] problems = result.getProblems();
for (int i = 0; i < problems.length; i++) {
IProblem problem = problems[i];
if (problem.isError()) {
String name =
new String(problems[i].getOriginatingFileName());
try {
problemList.add(ErrorDispatcher.createJavacError
(name, pageNodes, new StringBuilder(problem.getMessage()),
problem.getSourceLineNumber(), ctxt));
} catch (JasperException e) {
log.error("Error visiting node", e);
}
}
}
}
if (problemList.isEmpty()) {
ClassFile[] classFiles = result.getClassFiles();
for (int i = 0; i < classFiles.length; i++) {
ClassFile classFile = classFiles[i];
char[][] compoundName =
classFile.getCompoundName();
StringBuilder classFileName = new StringBuilder(outputDir).append('/');
for (int j = 0;
j < compoundName.length; j++) {
if(j > 0) {
classFileName.append('/');
}
classFileName.append(compoundName[j]);
}
byte[] bytes = classFile.getBytes();
classFileName.append(".class");
FileOutputStream fout = null;
BufferedOutputStream bos = null;
try {
fout = new FileOutputStream(classFileName.toString());
bos = new BufferedOutputStream(fout);
bos.write(bytes);
} finally {
if (bos != null) {
try {
bos.close();
} catch (IOException e) {
}
}
}
}
}
} catch (IOException exc) {
log.error("Compilation error", exc);
}
}
};
ICompilationUnit[] compilationUnits =
new ICompilationUnit[classNames.length];
for (int i = 0; i < compilationUnits.length; i++) {
String className = classNames[i];
compilationUnits[i] = new CompilationUnit(fileNames[i], className);
}
CompilerOptions cOptions = new CompilerOptions(settings);
// Check source/target JDK versions as the newest versions are allowed
// in Tomcat configuration but may not be supported by the ECJ version
// being used.
String requestedSource = ctxt.getOptions().getCompilerSourceVM();
if (requestedSource != null) {
String actualSource = CompilerOptions.versionFromJdkLevel(cOptions.sourceLevel);
if (!requestedSource.equals(actualSource)) {
log.warn(Localizer.getMessage("jsp.warning.unsupported.sourceVM", requestedSource, actualSource));
}
}
String requestedTarget = ctxt.getOptions().getCompilerTargetVM();
if (requestedTarget != null) {
String actualTarget = CompilerOptions.versionFromJdkLevel(cOptions.targetJDK);
if (!requestedTarget.equals(actualTarget)) {
log.warn(Localizer.getMessage("jsp.warning.unsupported.targetVM", requestedTarget, actualTarget));
}
}
cOptions.parseLiteralExpressionsAsConstants = true;
Compiler compiler = new Compiler(env,
policy,
cOptions,
requestor,
problemFactory);
compiler.compile(compilationUnits);
if (!ctxt.keepGenerated()) {
File javaFile = new File(ctxt.getServletJavaFileName());
javaFile.delete();
}
if (!problemList.isEmpty()) {
JavacErrorDetail[] jeds =
problemList.toArray(new JavacErrorDetail[0]);
errDispatcher.javacError(jeds);
}
if( log.isDebugEnabled() ) {
long t2=System.currentTimeMillis();
log.debug("Compiled " + ctxt.getServletJavaFileName() + " "
+ (t2-t1) + "ms");
}
if (ctxt.isPrototypeMode()) {
return;
}
// JSR45 Support
if (! options.isSmapSuppressed()) {
SmapUtil.installSmap(smap);
}
}
}
为了有助于理解，我们根据构造函数的参数依次看它的作用。
- INameEnvironment 接口，它需要实现的主要方法是findType和isPackage ，FindType有助于JDT 找到相应的Java源文件或者Class字节码，根据传进来的包名和类名去寻找，例如，传入了java.lang.String或org.apache.jsp.HelloWorld_jsp ，则分别找到JDK自带的String字节码及Tomcat 中编译的HelloWorld_jsp.java文件，接着，按要求封装这些对象，返回JDT 规定的NameEnvironmentAnswer对象，而isPackage则提供是否是包的判断 。
- IErrorHandlingPolicy 接口， 用于描述错误策略， 可直接使用DefaultErrorHandlingPolicies.exitOnFirstError()，如果表示第一个错误，就退出编译。
- CompilerOptions对象，指定编译时的一些参数，例如，这里指定的编译的java版本为1.7
- ICompilerRequestor接口，它只有一个acceptResult方法，这个方法用于处理编译后的结果，如果包含了错误信息，则抛出异常， 否则，把编译成功的字节码写到指定的路径的HelloWorld_jsp.class文件中，即生成字节码 。
- IProblemFactory接口， 主要用于控制编译错误的信息格式 。
所有Compiler构造函数需要的参数对象都已经准备后， 传入这些参数后创建一个Compiler对象，然后调用compile方法即可对指定的Java文件进行编译，这里完成了HelloWorld_jsp.java的编译， 结果生成了HelloWorld_jsp.class字节码， 实际上，Tomcat中基本上也这样使用了JDT 实现Servlet的编译，但它使用的某些策略可能不同，例如，使用DefaultErrorHandlingPolicies.proceedWithAllProblems()作为错误策略。
至此，我们已经清楚Tomcat 对JSP编译处理的整个过程了。 它首先根据JSP语法解析生成类似xxx.java的Servlet ，然后再通过Eclipse JDT 对xxx.java编译，最后生成 JVM 能识别的Class字节码 。
首先，需要一个后台执行线程，Tomcat 中有专门的一条线程负责处理不同的容器的后台任务，要在不同的容器中执行某些后台任务 ， 只需要重写backgroudProcess方法即可实现， 由于JSPServlet对应于Wrapper 级别，因此要在StandardWrapper重写backgroundProcess，它会调用实现了PeriodicEventListener接口的Servlet ，其中，JspServlet就实现了PeriodicEventListener 接口，此接口只有一个periodicEvent方法，具体的检测逻辑在此方法中实现即可。
其次，判断重新编译的根据是什么？重新编译就是再次把JSP变成java再变成Class，而触发这个动作的条件就是，当我们修改JSP文件后， 或者JSP文件引入的资源被修改后， 所以最好判断的依据就是某个JSP或资源的最后修改时间-LastModified属性， 正常顺序是JSP经过编译后生成的Class文件，把此class文件的LastModified 属性值决定是否重新编译，重新编译后，JSP 与Class 文件的LastModified属性再次设置为相同的值，对于引入的资源，内存中维护了上次编译时引入的资源的LastModified属性再次设置相同的值，不断的获取引入的资源LastModified属性并与内存中对应的LastModified属性进行比较，同样可以很容易的判断是否需要重新编译 。
最后，对于本地和远程资源分别如何检测？ 对于本地资源来说，使用java.io.File类可以很方便的实现对某JSP文件或其他文件的LastModified属性读取，对于远程资源，比如jar包，为了方便处理Jar 包含的属性，使用java.net.URL可以很方便的操作，它包含了很多协议，它包含了很多的协议，例如 ，常见的Jar , File , Ftp 等协议，使用它相当的方便 。
URL includeUrl = new URL(“jar:http://hostname/third.jar!/”);
URLConnection iuc = includeUrl.openConnection();
long includeLastModified = ((JarURLConnection)iuc).getJarEntry().getTime();
如前所述， 只需要三步可以远程对远程Jar 包的读取出最后修改时间，当然，URL 还支持本地文件资源的读取，所以，它是很好的资源读取抽象对象， Tomcat 中对引入的资源的管理都使用URL作为操作对象 。
本节探讨了Jasper自动检测机制的实现， 自动检测机制给我们开发带来了很好的检验，我们不必修改了JSP 后自动执行编译操作，而由Tomcat 通过Jasper帮我们检测编译操作。
关于编译代码这一块，我不想说太多，代码我也不太懂，为了更方便理解编译这一块的原理，我们来看一个例子。
public class MyJDTCompiler {
private static final String JDT_JAVA_9_VERSION;
static {
// The constant for Java 9 changed between 4.6 and 4.7 in a way that is
// not backwards compatible. Need to figure out which version is in use
// so the correct constant value is used.
String jdtJava9Version = null;
Class<?> clazz = CompilerOptions.class;
for (Field field : clazz.getFields()) {
if ("VERSION_9".equals(field.getName())) {
// 4.7 onwards: CompilerOptions.VERSION_9
jdtJava9Version = "9";
break;
}
}
if (jdtJava9Version == null) {
// 4.6 and earlier: CompilerOptions.VERSION_1_9
jdtJava9Version = "1.9";
}
JDT_JAVA_9_VERSION = jdtJava9Version;
}
private final Log log = LogFactory.getLog(MyJDTCompiler.class); // must not be static
/**
* Compile the servlet from .java file to .class file
*/
public void generateClass( final String sourceFile ,final String outputDir, String packageName
, String outclassName , final ClassLoader classLoader )
throws FileNotFoundException, JasperException, Exception {
long t1 = 0;
if (log.isDebugEnabled()) {
t1 = System.currentTimeMillis();
}
final String targetClassName =
((packageName.length() != 0) ? (packageName + ".") : "")
+ outclassName;
String[] fileNames = new String[] {sourceFile};
String[] classNames = new String[] {targetClassName};
final ArrayList<JavacErrorDetail> problemList =
new ArrayList<JavacErrorDetail>();
class CompilationUnit implements ICompilationUnit {
private final String className;
private final String sourceFile;
CompilationUnit(String sourceFile, String className) {
this.className = className;
this.sourceFile = sourceFile;
}
@Override
public char[] getFileName() {
return sourceFile.toCharArray();
}
@Override
public char[] getContents() {
char[] result = null;
FileInputStream is = null;
InputStreamReader isr = null;
Reader reader = null;
try {
is = new FileInputStream(sourceFile);
isr = new InputStreamReader(is, "UTF-8");
reader = new BufferedReader(isr);
char[] chars = new char[8192];
StringBuilder buf = new StringBuilder();
int count;
while ((count = reader.read(chars, 0,
chars.length)) > 0) {
buf.append(chars, 0, count);
}
result = new char[buf.length()];
buf.getChars(0, result.length, result, 0);
} catch (IOException e) {
log.error("Compilation error", e);
} finally {
if (reader != null) {
try {
reader.close();
} catch (IOException ioe) {/*Ignore*/}
}
if (isr != null) {
try {
isr.close();
} catch (IOException ioe) {/*Ignore*/}
}
if (is != null) {
try {
is.close();
} catch (IOException exc) {/*Ignore*/}
}
}
return result;
}
@Override
public char[] getMainTypeName() {
int dot = className.lastIndexOf('.');
if (dot > 0) {
return className.substring(dot + 1).toCharArray();
}
return className.toCharArray();
}
@Override
public char[][] getPackageName() {
StringTokenizer izer =
new StringTokenizer(className, ".");
char[][] result = new char[izer.countTokens()-1][];
for (int i = 0; i < result.length; i++) {
String tok = izer.nextToken();
result[i] = tok.toCharArray();
}
return result;
}
@Override
public boolean ignoreOptionalProblems() {
return false;
}
}
final INameEnvironment env = new INameEnvironment() {
@Override
public NameEnvironmentAnswer
findType(char[][] compoundTypeName) {
StringBuilder result = new StringBuilder();
String sep = "";
for (int i = 0; i < compoundTypeName.length; i++) {
result.append(sep);
result.append(compoundTypeName[i]);
sep = ".";
}
return findType(result.toString());
}
@Override
public NameEnvironmentAnswer
findType(char[] typeName,
char[][] packageName) {
StringBuilder result = new StringBuilder();
String sep = "";
for (int i = 0; i < packageName.length; i++) {
result.append(sep);
result.append(packageName[i]);
sep = ".";
}
result.append(sep);
result.append(typeName);
return findType(result.toString());
}
private NameEnvironmentAnswer findType(String className) {
InputStream is = null;
try {
if (className.equals(targetClassName)) {
ICompilationUnit compilationUnit =
new CompilationUnit(sourceFile, className);
return
new NameEnvironmentAnswer(compilationUnit, null);
}
String resourceName =
className.replace('.', '/') + ".class";
is = classLoader.getResourceAsStream(resourceName);
if (is != null) {
byte[] classBytes;
byte[] buf = new byte[8192];
ByteArrayOutputStream baos =
new ByteArrayOutputStream(buf.length);
int count;
while ((count = is.read(buf, 0, buf.length)) > 0) {
baos.write(buf, 0, count);
}
baos.flush();
classBytes = baos.toByteArray();
char[] fileName = className.toCharArray();
ClassFileReader classFileReader =
new ClassFileReader(classBytes, fileName,
true);
return
new NameEnvironmentAnswer(classFileReader, null);
}
} catch (IOException exc) {
log.error("Compilation error", exc);
} catch (org.eclipse.jdt.internal.compiler.classfmt.ClassFormatException exc) {
log.error("Compilation error", exc);
} finally {
if (is != null) {
try {
is.close();
} catch (IOException exc) {
// Ignore
}
}
}
return null;
}
private boolean isPackage(String result) {
if (result.equals(targetClassName)) {
return false;
}
String resourceName = result.replace('.', '/') + ".class";
InputStream is = null;
try {
is = classLoader.getResourceAsStream(resourceName);
return is == null;
} finally {
if (is != null) {
try {
is.close();
} catch (IOException e) {
}
}
}
}
@Override
public boolean isPackage(char[][] parentPackageName,
char[] packageName) {
StringBuilder result = new StringBuilder();
String sep = "";
if (parentPackageName != null) {
for (int i = 0; i < parentPackageName.length; i++) {
result.append(sep);
result.append(parentPackageName[i]);
sep = ".";
}
}
if (Character.isUpperCase(packageName[0])) {
if (!isPackage(result.toString())) {
return false;
}
}
result.append(sep);
result.append(packageName);
return isPackage(result.toString());
}
@Override
public void cleanup() {
}
};
final IErrorHandlingPolicy policy =
DefaultErrorHandlingPolicies.proceedWithAllProblems();
final Map<String,String> settings = new HashMap<String,String>();
settings.put(CompilerOptions.OPTION_LineNumberAttribute,
CompilerOptions.GENERATE);
settings.put(CompilerOptions.OPTION_SourceFileAttribute,
CompilerOptions.GENERATE);
settings.put(CompilerOptions.OPTION_ReportDeprecation,
CompilerOptions.IGNORE);
settings.put(CompilerOptions.OPTION_Encoding, "UTF-8");
settings.put(CompilerOptions.OPTION_LocalVariableAttribute,
CompilerOptions.GENERATE);
settings.put(CompilerOptions.OPTION_Source,
CompilerOptions.VERSION_1_8);
settings.put(CompilerOptions.OPTION_TargetPlatform,
CompilerOptions.VERSION_1_8);
settings.put(CompilerOptions.OPTION_Compliance,
CompilerOptions.VERSION_1_8);
final IProblemFactory problemFactory =
new DefaultProblemFactory(Locale.getDefault());
final ICompilerRequestor requestor = new ICompilerRequestor() {
@Override
public void acceptResult(CompilationResult result) {
try {
if (result.hasProblems()) {
IProblem[] problems = result.getProblems();
for (int i = 0; i < problems.length; i++) {
IProblem problem = problems[i];
if (problem.isError()) {
String name =
new String(problems[i].getOriginatingFileName());
try {
problemList.add(null);
} catch (Exception e) {
log.error("Error visiting node", e);
}
}
}
}
if (problemList.isEmpty()) {
ClassFile[] classFiles = result.getClassFiles();
for (int i = 0; i < classFiles.length; i++) {
ClassFile classFile = classFiles[i];
char[][] compoundName =
classFile.getCompoundName();
StringBuilder classFileName = new StringBuilder(outputDir).append('/');
for (int j = 0;
j < compoundName.length; j++) {
if(j > 0) {
classFileName.append('/');
}
classFileName.append(compoundName[j]);
}
byte[] bytes = classFile.getBytes();
classFileName.append(".class");
FileOutputStream fout = null;
BufferedOutputStream bos = null;
try {
fout = new FileOutputStream(classFileName.toString());
bos = new BufferedOutputStream(fout);
bos.write(bytes);
} finally {
if (bos != null) {
try {
bos.close();
} catch (IOException e) {
}
}
}
}
}
} catch (IOException exc) {
log.error("Compilation error", exc);
}
}
};
ICompilationUnit[] compilationUnits =
new ICompilationUnit[classNames.length];
for (int i = 0; i < compilationUnits.length; i++) {
String className = classNames[i];
compilationUnits[i] = new CompilationUnit(fileNames[i], className);
}
CompilerOptions cOptions = new CompilerOptions(settings);
cOptions.parseLiteralExpressionsAsConstants = true;
Compiler compiler = new Compiler(env,
policy,
cOptions,
requestor,
problemFactory);
compiler.compile(compilationUnits);
}
}
上面加粗代码的调用，就能将.java文件编译成.class文件。 而修改了generateClass()方法传入了4个参数 ，public void generateClass( final String sourceFile ,final String outputDir, String packageName
, String outclassName , final ClassLoader classLoader )，这4个参数分别是
- sourceFile ：源.java文件名称及路径
- outputDir ：生成的class文件目录
- packageName ： .java文件包名
- ReflictInvoke ：源文件名称
- classLoader ：类加载器
接下来看一个例子。
public class MyCompilerTest {
public static void main(String[] args) throws Exception {
String content = "package com.luban.compilerx;\n" +
"\n" +
"\n" +
"\n" +
"\n" +
"public class ReflictInvoke extends BaseInvoke {\n" +
"\n" +
" public void invoke(){\n" +
" System.out.println(\"ReflictInvoke调用11111111\");\n" +
" }\n" +
"}\n";
// 如果文件不存在 ，则创建文件
File file = new File("/Users/quyixiao/gitlab/tomcat/output/production/tomcat/com/luban/compilerx");
if (!file.exists()) {
file.mkdirs();
}
String sorceFile = "/Users/quyixiao/gitlab/tomcat/output/production/tomcat/com/luban/compilerx/ReflictInvoke.java";
FileOutputStream fos = new FileOutputStream(sorceFile);
Writer out = new OutputStreamWriter(fos, "UTF-8");
out.write(content);
out.close();
fos.close();
String outputDir = "/Users/quyixiao/gitlab/tomcat/output/production/tomcat";
String packageName = "com.luban.compilerx";
String targetClassName = "ReflictInvoke";
MyJDTCompiler myJDTCompiler = new MyJDTCompiler();
ClassLoader classLoader = MyCompilerTest.class.getClassLoader();
myJDTCompiler.generateClass(sorceFile, outputDir, packageName, targetClassName, classLoader);
Class<?> clazz = classLoader.loadClass("com.luban.compilerx.ReflictInvoke");
BaseInvoke c = (BaseInvoke) clazz.newInstance();
c.invoke();
}
}
这个例子的原理很简单，就是想动态的创建一个类ReflictInvoke
package com.luban.compilerx;
public class ReflictInvoke extends BaseInvoke {
public void invoke(){
System.out.println("ReflictInvoke调用11111111");
}
}
调用compiler.compile(compilationUnits);将java文件编译为.class文件，再通过Class<?> clazz = classLoader.loadClass(“com.luban.compilerx.ReflictInvoke”); 加载到虚拟机，再实例化它，最终调用其invoke()方法，这门技术，我觉得很牛逼，这样，我们可以在运行时创建一个类，并调用它，很多的框架都用到了这项技术，如Spring的CGLIB 代理，不就是运行时动态的修改原来类的代码结构，再编译它，最终再加载到虚拟机，替换掉原来的类，话不多说，看一下执行效果 。
看到没有，简单的一个类，就实现了java文件动态生成，编译，及使用。
之前已经分析到jsp文件的编译了， 接下来，看jsp如何获取实例对象 。
public void service(HttpServletRequest request,
HttpServletResponse response,
boolean precompile)
throws ServletException, IOException, FileNotFoundException {
Servlet servlet;
try {
if (ctxt.isRemoved()) {
throw new FileNotFoundException(jspUri);
}
if ((available > 0L) && (available < Long.MAX_VALUE)) {
if (available > System.currentTimeMillis()) {
response.setDateHeader("Retry-After", available);
response.sendError
(HttpServletResponse.SC_SERVICE_UNAVAILABLE,
Localizer.getMessage("jsp.error.unavailable"));
return;
}
// Wait period has expired. Reset.
available = 0;
}
/*
* (1) Compile
* 将jsp文件编译成servlet
*/
if (options.getDevelopment() || mustCompile) {
synchronized (this) {
if (options.getDevelopment() || mustCompile) {
// The following sets reload to true, if necessary
ctxt.compile();
mustCompile = false;
}
}
} else {
if (compileException != null) {
// Throw cached compilation exception
throw compileException;
}
}
/*
* (2) (Re)load servlet class file
* 生成Servlet文件对应的对象实例
*/
servlet = getServlet();
// If a page is to be precompiled only, return.
// 一个页面仅仅只需要编译
if (precompile) {
return;
}
} catch (ServletException ex) {
if (options.getDevelopment()) {
throw handleJspException(ex);
}
throw ex;
} catch (FileNotFoundException fnfe) {
// File has been removed. Let caller handle this.
throw fnfe;
} catch (IOException ex) {
if (options.getDevelopment()) {
throw handleJspException(ex);
}
throw ex;
} catch (IllegalStateException ex) {
if (options.getDevelopment()) {
throw handleJspException(ex);
}
throw ex;
} catch (Exception ex) {
if (options.getDevelopment()) {
throw handleJspException(ex);
}
throw new JasperException(ex);
}
try {
/*
* (3) Handle limitation of number of loaded Jsps
*/
if (unloadAllowed) { // 是否要卸载jsp-servlet
synchronized(this) {
if (unloadByCount) {
// 如果配置了限制的数量,则表示ctxt.getRuntimeContext()中只能容纳固定的jsw
// 那么如果超过了限制则将队尾jsw移除掉
// 当然，就算没有配置限制的数量，background线程会定时执行，将超过jspIdleTimeout时间的移除掉
if (unloadHandle == null) {
unloadHandle = ctxt.getRuntimeContext().push(this);
} else if (lastUsageTime < ctxt.getRuntimeContext().getLastJspQueueUpdate()) {
// lastUsageTime表示当前jsw上次使用时间
// ctxt.getRuntimeContext().getLastJspQueueUpdate()这个时间会由background线程定时更新一次
// 如果lastUsageTime 大于 ctxt.getRuntimeContext().getLastJspQueueUpdate()不需要做什么操作
// 第一种情况
// 1. jsw被访问
// 2. background线程执行
// 3. jsw再次被访问
// 4. 符合当前条件，jsw被移动至队首
// 第二种情况
// 1. background线程执行
// 2. jsw第一次被访问
// 3. 不符合条件，而此时应该符合unloadHandle == null
// 将最近访问的jsw移动至队首
ctxt.getRuntimeContext().makeYoungest(unloadHandle); // 将unloadHandle移到队首
lastUsageTime = System.currentTimeMillis();
}
} else {
// 更新最近使用的时间
if (lastUsageTime < ctxt.getRuntimeContext().getLastJspQueueUpdate()) {
lastUsageTime = System.currentTimeMillis();
}
}
}
}
/*
* (4) Service request
*/
if (servlet instanceof SingleThreadModel) {
// sync on the wrapper so that the freshness
// of the page is determined right before servicing
synchronized (this) {
servlet.service(request, response);
}
} else {
servlet.service(request, response);
}
} catch (UnavailableException ex) {
String includeRequestUri = (String)
request.getAttribute(RequestDispatcher.INCLUDE_REQUEST_URI);
if (includeRequestUri != null) {
// This file was included. Throw an exception as
// a response.sendError() will be ignored by the
// servlet engine.
throw ex;
}
int unavailableSeconds = ex.getUnavailableSeconds();
if (unavailableSeconds <= 0) {
unavailableSeconds = 60; // Arbitrary default
}
available = System.currentTimeMillis() +
(unavailableSeconds * 1000L);
response.sendError
(HttpServletResponse.SC_SERVICE_UNAVAILABLE,
ex.getMessage());
} catch (ServletException ex) {
if(options.getDevelopment()) {
throw handleJspException(ex);
}
throw ex;
} catch (IOException ex) {
if (options.getDevelopment()) {
throw new IOException(handleJspException(ex).getMessage(), ex);
}
throw ex;
} catch (IllegalStateException ex) {
if(options.getDevelopment()) {
throw handleJspException(ex);
}
throw ex;
} catch (Exception ex) {
if(options.getDevelopment()) {
throw handleJspException(ex);
}
throw new JasperException(ex);
}
}
生成的_jsp.java 又是如何调用及执行的呢？重点就是上面两行加粗代码 ，先来看servlet = getServlet(); 这一行代码 。
public Servlet getServlet() throws ServletException {
/*
* DCL on 'reload' requires that 'reload' be volatile
* (this also forces a read memory barrier, ensuring the new servlet
* object is read consistently).
*
* When running in non development mode with a checkInterval it is
* possible (see BZ 62603) for a race condition to cause failures
* if a Servlet or tag is reloaded while a compile check is running
*/
if (getReloadInternal() || theServlet == null) {
synchronized (this) {
// Synchronizing on jsw enables simultaneous loading
// of different pages, but not the same page.
if (getReloadInternal() || theServlet == null) {
// This is to maintain the original protocol.
destroy();
final Servlet servlet;
try {
InstanceManager instanceManager = InstanceManagerFactory.getInstanceManager(config);
// 实例化对象
servlet = (Servlet) instanceManager.newInstance(ctxt.getFQCN(), ctxt.getJspLoader());
} catch (Exception e) {
Throwable t = ExceptionUtils
.unwrapInvocationTargetException(e);
ExceptionUtils.handleThrowable(t);
throw new JasperException(t);
}
servlet.init(config);
if (theServlet != null) {
ctxt.getRuntimeContext().incrementJspReloadCount();
}
theServlet = servlet;
reload = false;
// Volatile 'reload' forces in order write of 'theServlet' and new servlet object
}
}
}
return theServlet;
}
public Object newInstance(final String className, final ClassLoader classLoader)
throws IllegalAccessException, NamingException, InvocationTargetException,
InstantiationException, ClassNotFoundException, IllegalArgumentException,
NoSuchMethodException, SecurityException {
// loadClass()方法加载类
Class<?> clazz = classLoader.loadClass(className);
return newInstance(clazz.getDeclaredConstructor().newInstance(), clazz);
}
上面代码原理还是很简单就，先加载我们生成的index_jsp.class类，先调用其init()方法。
public final void init(ServletConfig config) throws ServletException { super.init(config); jspInit(); _jspInit(); }
看到没有，调用_jspInit()方法，不就是生成index_jsp.java的_jspInit() 方法不？
同理，来看 servlet.service(request, response);方法调用，最终也是调用了_jspService(request,response);
不就对应index_jsp.java 的 _jspService()方法不？如下图所示 。
总结：
文章分享到这里也告一段落了， 因为篇幅过长，所以分为Tomcat 源码解析一JSP编译器Jasper-佛怒火莲（上） ， Tomcat 源码解析一JSP编译器Jasper-佛怒火莲（下）两部分，前部分主要讲jsp 解析相关内容，后部分主要讲tomcat如何修改字节码，编译，及获取生成类实例。 可以涉及到的内容及范围也比较广，当然，你觉得关于jsp相关源码就完了吗？肯定没有，如果你想了解EL表达式如何解析，表达式替换，还需要看Tomcat 源码解析一EL表达式源码解析 这篇博客 ，内部细节实现也是很复杂的，当然，我也尽量的将所见到的源码分析清楚，希望这对大家有所帮助。由于我比较喜欢斗破苍穹，因此后面的博客中，每一篇博客可能以斗破苍穹中的一种斗技作为后缀， 话不多说，就到这里了，下一篇博客见。
本文相关的源码地址
https://github.com/quyixiao/tomcat.git
https://github.com/quyixiao/JspReader.git
504
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
