Java反射三种方式的效率对比

末蓝、 2022-07-16 02:21 257阅读 0赞

*  0 先放结论  
    org.apache.commons.beanutils.BeanUtils的性能比较差  
    spring自带的BeanUtils性能还可以（看了源码，也是通过反射来实现的）
 *  1 使用field

long start = System.nanoTime();
                Field[] fields = CallCount.class.getDeclaredFields();
                for (String str : dateList) { 
                    boolean exist = false;
                    for (Field field : fields){ 
                        field.setAccessible(true);
                        if (field.getName().endsWith(str)) { 
                            int value = 0;
                            for (CallCount callCount : callCountList) { 
                                value += (Integer)field.get(callCount);
                            }
                            resultList.add(value);
                            exist = true;
                            break;
                        }
                    }
                    if (!exist) { 
                        resultList.add(0);
                    }
                }
                long end = System.nanoTime();
                log.info("old call cost :" + (end-start));

*  2 使用 org.apache.commons.beanutils.BeanUtils来获取属性

long start = System.nanoTime();
                for (String str : dateList) { 
                    Integer value = getMinuteAccessCount(str, callCountList);
                    resultList.add(value);
                }
                long end = System.nanoTime();
                log.info("new call cost :" + (end-start));
    
    
            private Integer getMinuteAccessCount(String minute, List<CallCount> callCountList) { 
            Integer result = 0;
            String propertyName = "logMinute" + minute;
            for (CallCount callCount : callCountList) { 
                int value = 0;
                try { 
                    value = Integer.valueOf(BeanUtils.getProperty(callCount, propertyName));
                } catch (NumberFormatException | IllegalAccessException | InvocationTargetException | NoSuchMethodException e) { 
                    e.printStackTrace();
                }
                result += value;
            }
            return result;
        }

*  3 使用java.lang.reflect.Method获取值

long start = System.nanoTime();
                for (String str : dateList) { 
                    Integer value = getMinuteAccessCount(str, callCountList);
                    resultList.add(value);
                }
                long end = System.nanoTime();
                log.info("new call cost :" + (end-start));
    
    
            private Integer getMinuteAccessCount(String minute, List<CallCount> callCountList) { 
            Integer result = 0;
            for (CallCount callCount : callCountList) { 
                int value = 0;
                try { 
                    Method method = callCount.getClass().getDeclaredMethod("getLogMinute"+minute);
                    Object obj = method.invoke(callCount);
                    value = (Integer)obj;
                } catch (NumberFormatException | IllegalAccessException | InvocationTargetException | NoSuchMethodException e) { 
                    e.printStackTrace();
                }
                result += value;
            }
            return result;
        }

*  4 耗时对比，在使用相同的查询条件，相同的数据的情况下。通过实验得到以下两组数据（耗时/纳秒）

1 old call cost :517599  
1 old call cost :347916  
1 old call cost :337312  
1 old call cost :177893  
1 old call cost :131709  
1 old call cost :82789  
1 old call cost :63973

3 new call cost :925383  
3 new call cost :794016  
3 new call cost :912382

--------------------

1 old call cost :755016  
1 old call cost :365364  
1 old call cost :231944  
1 old call cost :123498  
1 old call cost :103315  
1 old call cost :92025  
1 old call cost :81762

2 new call cost :139741338  
2 new call cost :3387140  
2 new call cost :2230497  
2 new call cost :9215854  
2 new call cost :2313970  
2 new call cost :1549374  
2 new call cost :1884291  
2 new call cost :1100880  
2 new call cost :1488138

每组数据前的数字代表之前取值的方式。  
由数据对比可以看出，耗时最小的，始终是方式1，并且方式1在多次调用时，耗时是逐步减少的，可能是有缓存机制。  
耗时第二少的是方式3，耗时最多的是方式2.

*  5 解析

方式1，采用了最基本的java反射方式，使用Filed，循环bean的Field，得到Object，再转换为Integer类型。  
方式2，采用了看似最简洁的BeanUitls，根据属性名，获取属性值。这样最耗时。  
方式3，采用了获取bean的Method，然后invoke的方式来获取值。