浅谈slf4j,logger中的{}功能

slf4j有一个common logger没有的功能，字符串中的{}会被替换，如下：

logger.info("Hello {}","world");

这个功能看起来好像很厉害。那实质上slf4j的工程师到底做了什么？会比我们单纯的字符串拼接更快吗？
在slf4j-api:1.7.21这个版本的slf4j的jar中，找到MessageFormatter类，里面有一段代码。

 final public static FormattingTuple arrayFormat(final String messagePattern, final Object[] argArray, Throwable throwable) {

        if (messagePattern == null) {
            return new FormattingTuple(null, argArray, throwable);
        }

        if (argArray == null) {
            return new FormattingTuple(messagePattern);
        }

        int i = 0;
        int j;
        // use string builder for better multicore performance
        StringBuilder sbuf = new StringBuilder(messagePattern.length() + 50);

        int L;
        for (L = 0; L < argArray.length; L++) {

            j = messagePattern.indexOf(DELIM_STR, i);

            if (j == -1) {
                // no more variables
                if (i == 0) { // this is a simple string
                    return new FormattingTuple(messagePattern, argArray, throwable);
                } else { // add the tail string which contains no variables and return
                    // the result.
                    sbuf.append(messagePattern, i, messagePattern.length());
                    return new FormattingTuple(sbuf.toString(), argArray, throwable);
                }
            } else {
                if (isEscapedDelimeter(messagePattern, j)) {
                    if (!isDoubleEscaped(messagePattern, j)) {
                        L--; // DELIM_START was escaped, thus should not be incremented
                        sbuf.append(messagePattern, i, j - 1);
                        sbuf.append(DELIM_START);
                        i = j + 1;
                    } else {
                        // The escape character preceding the delimiter start is
                        // itself escaped: "abc x:\\{}"
                        // we have to consume one backward slash
                        sbuf.append(messagePattern, i, j - 1);
                        deeplyAppendParameter(sbuf, argArray[L], new HashMap());
                        i = j + 2;
                    }
                } else {
                    // normal case
                    sbuf.append(messagePattern, i, j);
                    deeplyAppendParameter(sbuf, argArray[L], new HashMap());
                    i = j + 2;
                }
            }
        }
        // append the characters following the last {} pair.
        sbuf.append(messagePattern, i, messagePattern.length());
        return new FormattingTuple(sbuf.toString(), argArray, throwable);
    }

在这个方法中slf4j对{}进行了解析，实质也不是很高效的方法，使用indexOf找到”{}”，再对这个位置处理。简化了以后就是如下这段代码：

final public static FormattingTuple arrayFormat(final String messagePattern, final Object[] argArray, Throwable throwable) {
        int i = 0;
        int j;
        StringBuilder sbuf = new StringBuilder(messagePattern.length() + 50);
        int L;
        for (L = 0; L < argArray.length; L++) {
            j = messagePattern.indexOf("{}", i);
            if (j != -1){
                sbuf.append(messagePattern,i,j);
                sbuf.append(argArray[L]);
                i = j+2;
            }else{
                break;
            }
        }
        sbuf.append(messagePattern, i, messagePattern.length());
        return new FormattingTuple(sbuf.toString(), argArray, throwable);
    }

而String的indexOf这个方法还是比较消耗性能的。用slf4j这个特性实质上是比我们单纯用logger.info(“Hello “+name);这个方式慢的。于是我做了下实验。用字符串拼接的方式以及StringBuilder的方式去和slf4j的{}进行比较

public String loggerFormat(){
        long beginTs = System.currentTimeMillis();
        logger.info("loggerFormat begin:"+String.valueOf(beginTs));
        for (int i = 0; i < 1000; i++) {
            String[] args = new String[]{
                    getString(),
                    getString(),
                    getString(),
                    getString(),
                    getString()
            };
            logger.info("测试{},{},{},{},{}",args);
        }
        String elapsedTime = "loggerFormat finish:takes "+String.valueOf(System.currentTimeMillis()-beginTs);
        logger.info(elapsedTime);
        return elapsedTime;
    }

    public String loggerString(){
        long beginTs = System.currentTimeMillis();
        logger.info("loggerString begin:"+String.valueOf(beginTs));
        for (int i = 0; i < 1000; i++) {
            String info = "测试"+getString()+","+getString()+","+getString()+","+getString()+","+getString();
            logger.info(info);
        }
        String elapsedTime = "loggerString finish:takes "+String.valueOf(System.currentTimeMillis()-beginTs);
        logger.info(elapsedTime);
        return elapsedTime;
    }

    public String loggerStringBulider(){
        long beginTs = System.currentTimeMillis();
        logger.info("loggerStringBulider begin:"+String.valueOf(beginTs));
        for (int i = 0; i < 1000; i++) {
            StringBuilder info = new StringBuilder();
            info.append("测试")
                    .append(getString()).append(",")
                    .append(getString()).append(",")
                    .append(getString()).append(",")
                    .append(getString()).append(",")
                    .append(getString());
            logger.info(info.toString());
        }
        String elapsedTime = "loggerStringBulider finish:takes "+String.valueOf(System.currentTimeMillis()-beginTs);
        logger.info(elapsedTime);
        return elapsedTime;
    }

public String getString(){
        return UUID.randomUUID().toString();
    }

得到的结果

2017-01-04 11:42:49.560 [172.27.35.1] [main] INFO  c.banger.ubip.authdata.common.LoggerTest - loggerFormat finish:takes 97
2017-01-04 11:42:49.560 [172.27.35.1] [main] INFO  c.banger.ubip.authdata.common.LoggerTest - loggerStringBulider finish:takes 66
2017-01-04 11:42:49.560 [172.27.35.1] [main] INFO  c.banger.ubip.authdata.common.LoggerTest - loggerString finish:takes 74

最快的是StringBuilder的方式，字符串拼接的方式也比slf4j快。但是，如果是非常多变量的字符串拼接是不可取的，会产生非常多的对象，这时候可以用StringBuilder或者说用slf4j的{}的方式，注意StringBuilder的初始化长度。