从JVM视角分析try...catch...性能

2,950 阅读4分钟

try..catch..到底影响性能吗?

实验1:简单认识

随便写一个简单的程序

public class Test {
    public static void main(String[] args) {
        int a = 0,b=2;
        try {
            a = b/0;
        }catch (Exception e){
            a = 1;
        }
    }
}

看一下字节码指令过程:

 public static void main(java.lang.String[]);
    descriptor: ([Ljava/lang/String;)V
    flags: (0x0009) ACC_PUBLIC, ACC_STATIC
    Code:
      stack=2, locals=4, args_size=1
         0: iconst_0        push 0
         1: istore_1        pop并保存到局部变量1
         2: iconst_2       push 2
         3: istore_2       pop并保存到局部变量2
         4: iload_2        从局部变量里拿出并push
         5: iconst_0        push 0 
         6: idiv          栈顶两数相除
         7: istore_1
         8: goto          14
        11: astore_3
        12: iconst_1
        13: istore_1
        14: return
      Exception table:
         from    to  target type
             4     8    11   Class java/lang/Exception
      LineNumberTable:
        line 8: 0
        line 10: 4
        line 13: 8
        line 11: 11
        line 12: 12
        line 14: 14
      LocalVariableTable:
        Start  Length  Slot  Name   Signature
           12       2     3     e   Ljava/lang/Exception;
            0      15     0  args   [Ljava/lang/String;
            2      13     1     a   I
            4      11     2     b   I
      StackMapTable: number_of_entries = 2
        frame_type = 255 /* full_frame */
          offset_delta = 11
          locals = [ class "[Ljava/lang/String;", int, int ]
          stack = [ class java/lang/Exception ]
        frame_type = 2 /* same */
}

可以看到有个异常表:

      Exception table:
         from    to  target type
             4     8    11   Class java/lang/Exception

from表示try catch的开始地址 to表示try catch的结束地址 target表示异常的处理起始位 type表示异常类名称

代码运行时出错时,会先判断出错位置是否在from - to的范围,如果是,则从target标志位往下执行,如果没有出错,直接gotoreturn。可以看出,如果代码不出错的话,性能几乎是不受影响的,和正常的代码执行是一样的。

那异常处理耗时是什么个概念呢?

实验二:异常处理耗时测试

public class Test {
    public static void main(String[] args) {
        int a = 0,b=2;
        long startTime = System.nanoTime();
        for (int i = 10; i>0;i--){
            try {
                a = b/i;
            }catch (Exception e){
                a = 1;
            }finally {

            }
        }
        long runTime = System.nanoTime()-startTime;
        System.out.println(runTime);
    }
}

我只需要把i>0改成i>=0,程序遍会进行一次异常处理,因为除数不能为0.

我在修改之前(无异常运行),运行的结果是1133 修改之后(会出现除数为0异常),运行结果是44177

当然,这个结果和cpu的算力有关,多次运行结果相差无几。

所以,可以看出一旦程序进入到catch里,是非常耗资源的。

那try catch 在for循环外面或者里面,哪个更好呢?

实验三:for循环在try里面

public class Test {
    public static void main(String[] args) {
        int a = 0,b=2;
        long startTime = System.nanoTime();
        try {
            for (int i = 10; i>=0;i--){
                    a = b/i;
            }
        }catch (Exception e){
            a = 1;
        }finally {
            long runTime = System.nanoTime()-startTime;
            System.out.println(runTime);
        }
    }
}

运行多次的控制台输出:

46820     48708 54749  47953   46820  45310
  public static void main(java.lang.String[]);
    descriptor: ([Ljava/lang/String;)V
    flags: (0x0009) ACC_PUBLIC, ACC_STATIC
    Code:
      stack=2, locals=5, args_size=1
         0: iconst_0
         1: istore_1
         2: iconst_2
         3: istore_2
         4: bipush        10
         6: istore_3
         7: iload_3
         8: iflt          21
        11: iload_2
        12: iload_3
        13: idiv
        14: istore_1
        15: iinc          3, -1
        18: goto          7
        21: goto          35
        24: astore_3
        25: iconst_1
        26: istore_1
        27: goto          35
        30: astore        4
        32: aload         4
        34: athrow
        35: return
      Exception table:
         from    to  target type
             4    21    24   Class java/lang/Exception
             4    21    30   any
            24    27    30   any
            30    32    30   any

实验四:try在for循环外面

public class Test {
    public static void main(String[] args) {
        int a = 0,b=2;
        long startTime = System.nanoTime();
        for (int i = 10; i>=0;i--){
            try {
                a = b/i;
            }catch (Exception e){
                a = 1;
            }finally {

            }
        }
                long runTime = System.nanoTime()-startTime;
                System.out.println(runTime);
    }
}

控制台打印:

42289  47953  49463  45688  45310
  public static void main(java.lang.String[]);
    descriptor: ([Ljava/lang/String;)V
    flags: (0x0009) ACC_PUBLIC, ACC_STATIC
    Code:
      stack=2, locals=6, args_size=1
         0: iconst_0
         1: istore_1
         2: iconst_2
         3: istore_2
         4: bipush        10
         6: istore_3
         7: iload_3
         8: iflt          36
        11: iload_2
        12: iload_3
        13: idiv
        14: istore_1
        15: goto          30
        18: astore        4
        20: iconst_1
        21: istore_1
        22: goto          30
        25: astore        5
        27: aload         5
        29: athrow
        30: iinc          3, -1
        33: goto          7
        36: return
      Exception table:
         from    to  target type
            11    15    18   Class java/lang/Exception
            11    15    25   any
            18    22    25   any
            25    27    25   any

综合实验三和实验四,我们发现无论从运行时长还是从字节码指令的角度看,它两的性能可以说是一样的。并没有你感觉到的for循环里放try代码会冗余、资源消耗加倍的问题。

但是从运行逻辑来看,两个是有点不同的,实验三中,因为for在try catch里,所以jvm在编译的时候,把异常处理放在for循环后面才进行。即:第24-27行;实验四中,异常处理是在for循环内部的,即:第18-22行。大同小异。

以上仅个人测试观点,如果有误请在下方留言,谢谢!