JDK 1.8(Java 8)里新增String.join()方法用於字符串連接。本文基於《Java實現String.join()和效率比較》一文,分析和比較四種自定義實現與String.join()方法的效率,並糾正原文的一些錯誤。
代碼示例如下:
public class Test { public static void main(String[] args) { String[] strOri = {"a","b","c","d","e","f","g","h"}; //同new string[]{"a","b","c","d","e","f","g","h"} String strRes = ""; int loops = 100000; Date date = new Date(); for(int i = 0; i < loops; i++) {strRes = join1(strOri, ":");} date = recordTime(date, 1); //1 for(int i = 0; i < loops; i++) {strRes = join2(strOri, ":");} date = recordTime(date, 2); for(int i = 0; i < loops; i++) {strRes = join3(strOri, ":");} date = recordTime(date, 3); for(int i = 0; i < loops; i++) {strRes = join4(strOri, ":");} date = recordTime(date, 4); for(int i = 0; i < loops; i++) {strRes = join5(strOri, ":");} date = recordTime(date, 5); long startTime = System.currentTimeMillis(); //2 for(int i = 0; i < loops; i++) {strRes = join5(strOri, ":");} long endTime = System.currentTimeMillis(); System.out.println("5c:{" + strRes + "} costs " + (endTime-startTime) + "ms"); startTime = System.nanoTime(); for(int i = 0; i < loops; i++) {strRes = join5(strOri, ":");} endTime = System.nanoTime(); System.out.println("5n:{" + strRes + "} costs " + (endTime-startTime) + "ns"); } private static void recordTime_Wrong(Date date, int no) { System.out.println(no + ": costs " + (new Date().getTime()-date.getTime()) + "ms"); date = new Date(); } private static Date recordTime(Date date, int no) { System.out.println(no + ": costs " + (new Date().getTime()-date.getTime()) + "ms"); return new Date(); } private static String join1(String[] strOri, String delimiter) { StringBuffer sb = new StringBuffer(); //3 for(String s : strOri) { sb.append(s+delimiter); //4 } return sb.toString().substring(0, sb.toString().length()-1); } private static String join2(String[] strOri, String delimiter) { StringBuffer sb = new StringBuffer(); for(String s : strOri) { sb.append(s+delimiter); } String s = sb.toString(); return s.substring(0, s.length()-1); } private static String join3(String[] strOri, String delimiter) { StringBuffer sb = new StringBuffer(); for(int i = 0; i < strOri.length; i++) { if (i != strOri.length-1) { sb.append(strOri[i]+delimiter); } else { sb.append(strOri[i]); } } return sb.toString(); } private static String join4(String[] strOri, String delimiter) { StringBuilder stringBuilder = new StringBuilder(); for (int i = 0; i < strOri.length-1; i++) { stringBuilder.append(strOri[i]).append(delimiter); } stringBuilder.append(strOri[strOri.length-1]); return stringBuilder.toString(); } private static String join5(String[] strOri, String delimiter) { return String.join(delimiter, strOri); //5 } }
選取三次運行輸出結果如下:
1: costs 930ms 2: costs 902ms 3: costs 637ms 4: costs 230ms 5: costs 364ms 5c:{a:b:c:d:e:f:g:h} costs 413ms 5n:{a:b:c:d:e:f:g:h} costs 286466296ns
1: costs 834ms 2: costs 788ms 3: costs 576ms 4: costs 248ms 5: costs 350ms 5c:{a:b:c:d:e:f:g:h} costs 384ms 5n:{a:b:c:d:e:f:g:h} costs 283256112ns
1: costs 774ms 2: costs 728ms 3: costs 605ms 4: costs 297ms 5: costs 417ms 5c:{a:b:c:d:e:f:g:h} costs 280ms 5n:{a:b:c:d:e:f:g:h} costs 279838638ns
可見,join4()執行最快,其次是join5()。join1()和join2()執行效率接近,前者調用兩次toString(),故效率略低。
總結如下:
1. 原文recordTime(即本文recordTime_Wrong)方法中,無法通過"date = new Date()"修改外部的date引用(根因詳見《java中的傳值與傳引用》)。這會導致每次調用recordTime()時,起始時間始終是"Date date = new Date()"獲得的對象(表現為join*耗時遞增)。
2. 查看Java源碼可知,new Date()其實就是調用System.currentTimeMillis():
1 public Date() { 2 this(System.currentTimeMillis()); //相當於Date(System.currentTimeMillis()) 3 }
可以使用new Date().getTime()獲取當前時間戳(毫秒)。注意,該毫秒數一般以1970-01-01 00:00:00為參考點,但東八區要加上時區,即以1970-01-01 08:00:00為參考時間。此外,通過getTime()獲取毫秒數效率不如System.currentTimeMillis(),后者返回自1970年1月1日0時起的毫秒數。
System.nanoTime()的計時精度不保證一定高於System.currentTimeMillis(),但可保證數值遞增(后者相減時可能產生負值)。
若要對代碼進行更准確的計時,可參考《How do I write a correct micro-benchmark in Java?》一文。
3. StringBuffer對象是線程安全的,其方法都是同步的(synchronized)。臨時變量應使用StringBuilder(效率更高),避用StringBuffer。
4. 在循環內部,不要使用append(a+b)的形式,而應改為append(a).append(b)。
5. String.join()內部使用StringBuilder實現,因此join5()性能接近join4()。當然,String.join()的功能比join4()更多。