Java中使用 Long 表示枚举类

Java中使用 Long 表示枚举类

在日常的开发过程中,很多时候我们需要枚举类(enum)来表示对象的各种状态,并且每个状态往往会关联到指定的数字,如:

    private enum Color {
RED(11), GREEN(21), YELLOW(31), BLACK(160);
...
};

或者用枚举类来表示一系列状态的转变关系:

    enum Week{
SUNDAY(1), MONDAY(2), TUESDAY(3), WEDNESDAY(4), THRUSDAY(5), FRIDAY(6), SATRUDAY7);
...
};

那么,如何用最少的存储来实现这类需求,答案很简单,位存储。如 1bit 表示 0,1 两种状态,2bit 表示 00,01,10,11 四种状态,所以我们可以用一个 long 类型(64bit)/int 类型(32bit)存储多种状态,如下图:

Java中使用 Long 表示枚举类
位存储示例

但是每新建一个枚举类都需要自己操作 bit

  1. 导致程序不易理解
  2. 容易出错,耗费精力

Hadoop hdfs 的实现中,也遇到类似的问题,它借助于 LongBitFormat.java 类封装了 bit 操作:

    public class LongBitFormat implements Serializable {
private static final long serialVersionUID = 1L; private final String NAME;
/** Bit offset */
private final int OFFSET;
/** Bit length */
private final int LENGTH;
/** Minimum value */
private final long MIN;
/** Maximum value */
private final long MAX;
/** Bit mask */
private final long MASK; public LongBitFormat(String name, LongBitFormat previous, int length,
long min) {
NAME = name;
OFFSET = previous == null ? 0 : previous.OFFSET + previous.LENGTH;
LENGTH = length;
MIN = min;
MAX = ((-1L) >>> (64 - LENGTH));//移动的位数,右移64-Leng位,相当于保留length位
MASK = MAX << OFFSET;
} /** Retrieve the value from the record. */
public long retrieve(long record) {
return (record & MASK) >>> OFFSET;
} /** Combine the value to the record. */
public long combine(long value, long record) {
if (value < MIN) {
throw new IllegalArgumentException(
"Illagal value: " + NAME + " = " + value + " < MIN = " + MIN);
}
if (value > MAX) {
throw new IllegalArgumentException(
"Illagal value: " + NAME + " = " + value + " > MAX = " + MAX);
}
return (record & ~MASK) | (value << OFFSET);
} public long getMin() {
return MIN;
}
}

当然,你也可以实现 IntBigFormat,ShortBitFormat

首先分析该类的构造方法:

        NAME = name;
OFFSET = previous == null? 0: previous.OFFSET + previous.LENGTH;
LENGTH = length;
MIN = min;
MAX = ((-1L) >>> (64 - LENGTH));//移动的位数,右移64-Leng位,相当于保留length位
MASK = MAX << OFFSET;

字段:
NAME:状态名,可自定义

OFFSET:该状态在 long 字节中的偏移

LENGTH:用多少位存储该状态关联的数字

MIN:该状态关联的最小值

MAX:该状态关联的最大值

MASK:掩码,(OFFSET~OFFSET+LENGTH - 1) == 1

类方法:
retrieve(long record):获得该状态关联的数字

combine(long value, long record):将一个 value 加到 record 中,例如:将 value 值对应的枚举类存储在 32-40,则先将 32-40bits 清零,再将value 对应的二进制加入到 32-40

那么如何使用该类:

public class LongFormatTest {

    static enum HeaderFormat {
PREFERRED_BLOCK_SIZE(null, 48, 1),
REPLICATION(PREFERRED_BLOCK_SIZE.BITS, 12, 1),
STORAGE_POLICY_ID(REPLICATION.BITS, 4, 0); private final LongBitFormat BITS; HeaderFormat(LongBitFormat previous, int length, long min) {
BITS = new LongBitFormat(name(), previous, length, min);
} static short getReplication(long header) {
return (short) REPLICATION.BITS.retrieve(header);
} static long getPreferredBlockSize(long header) {
return PREFERRED_BLOCK_SIZE.BITS.retrieve(header);
} static byte getStoragePolicyID(long header) {
return (byte) STORAGE_POLICY_ID.BITS.retrieve(header);
} static long toLong(long preferredBlockSize, long replication,
long storagePolicyID) {
long h = 0;
h = PREFERRED_BLOCK_SIZE.BITS.combine(preferredBlockSize, h);
h = REPLICATION.BITS.combine(replication, h);
h = STORAGE_POLICY_ID.BITS.combine(storagePolicyID, h);
return h;
}
} public static void main(String[] args) { long blockSize = 512;
long replication = 3L;
long storagePolicyID = 2L;
long combine = HeaderFormat.toLong(blockSize,replication,storagePolicyID);
System.out.println("block size: " + HeaderFormat.getPreferredBlockSize(combine));
System.out.println("replication: " + HeaderFormat.getReplication(combine));
System.out.println("storagePolicyID: " + HeaderFormat.getStoragePolicyID(combine));
}
}

tolong 方法的返回值也就是我们状态存储的封装

上一篇:初入水:vector


下一篇:the error code is 2203