1 /*

     2  * Copyright (c) 1995, 2013, 2019, Oracle and/or its affiliates. All rights reserved.

     3  * ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.

     4  *

     5  *

     6  *

     7  *

     8  *

     9  *

    10  *

    11  *

    12  *

    13  *

    14  *

    15  *

    16  *

    17  *

    18  *

    19  *

    20  *

    21  *

    22  *

    23  *

    24  */

    25 

    26 // package java.util;

    27 

    28 import java.io.*;

    29 import java.math.BigInteger;

    30 import java.util.concurrent.atomic.AtomicLong;

    31 import java.util.function.DoubleConsumer;

    32 import java.util.function.IntConsumer;

    33 import java.util.function.LongConsumer;

    34 import java.util.stream.DoubleStream;

    35 import java.util.stream.IntStream;

    36 import java.util.stream.LongStream;

    37 import java.util.stream.StreamSupport;

    38 

    39 import sun.misc.Unsafe;

    40 

    41 /**

    42  * An instance of this class is used to generate a stream of

    43  * pseudorandom numbers. The class uses a 48-bit seed, which is

    44  * modified using a linear congruential formula. (See Donald Knuth,

    45  * <i>The Art of Computer Programming, Volume 2</i>, Section 3.2.1.)

    46  * <p>

    47  * If two instances of {@code Random} are created with the same

    48  * seed, and the same sequence of method calls is made for each, they

    49  * will generate and return identical sequences of numbers. In order to

    50  * guarantee this property, particular algorithms are specified for the

    51  * class {@code Random}. Java implementations must use all the algorithms

    52  * shown here for the class {@code Random}, for the sake of absolute

    53  * portability of Java code. However, subclasses of class {@code Random}

    54  * are permitted to use other algorithms, so long as they adhere to the

    55  * general contracts for all the methods.

    56  * <p>

    57  * The algorithms implemented by class {@code Random} use a

    58  * {@code protected} utility method that on each invocation can supply

    59  * up to 32 pseudorandomly generated bits.

    60  * <p>

    61  * Many applications will find the method {@link Math#random} simpler to use.

    62  *

    63  * <p>Instances of {@code java.util.Random} are threadsafe.

    64  * However, the concurrent use of the same {@code java.util.Random}

    65  * instance across threads may encounter contention and consequent

    66  * poor performance. Consider instead using

    67  * {@link java.util.concurrent.ThreadLocalRandom} in multithreaded

    68  * designs.

    69  *

    70  * <p>Instances of {@code java.util.Random} are not cryptographically

    71  * secure.  Consider instead using {@link java.security.SecureRandom} to

    72  * get a cryptographically secure pseudo-random number generator for use

    73  * by security-sensitive applications.

    74  *

    75  * @author  Frank Yellin

    76  * @since   1.0

    77  */

    78 public

    79     class Random extends AbstractSharedRng implements java.io.Serializable {

    80     /** use serialVersionUID from JDK 1.1 for interoperability */

    81     static final long serialVersionUID = 3905348978240129619L;

    82 

    83     /**

    84      * The internal state associated with this pseudorandom number generator.

    85      * (The specs for the methods in this class describe the ongoing

    86      * computation of this value.)

    87      */

    88     private final AtomicLong seed;

    89 

    90     private static final long multiplier = 0x5DEECE66DL;

    91     private static final long addend = 0xBL;

    92     private static final long mask = (1L << 48) - 1;

    93 

    94     private static final double DOUBLE_UNIT = 0x1.0p-53; // 1.0 / (1L << 53)

    95 

    96     // IllegalArgumentException messages

    97     static final String BadBound = "bound must be positive";

    98     static final String BadRange = "bound must be greater than origin";

    99     static final String BadSize  = "size must be non-negative";

   100 

   101     /**

   102      * Creates a new random number generator. This constructor sets

   103      * the seed of the random number generator to a value very likely

   104      * to be distinct from any other invocation of this constructor.

   105      */

   106     public Random() {

   107         this(seedUniquifier() ^ System.nanoTime());

   108     }

   109 

   110     private static long seedUniquifier() {

   111         // L'Ecuyer, "Tables of Linear Congruential Generators of

   112         // Different Sizes and Good Lattice Structure", 1999

   113         for (;;) {

   114             long current = seedUniquifier.get();

   115             long next = current * 181783497276652981L;

   116             if (seedUniquifier.compareAndSet(current, next))

   117                 return next;

   118         }

   119     }

   120 

   121     private static final AtomicLong seedUniquifier

   122         = new AtomicLong(8682522807148012L);

   123 

   124     /**

   125      * Creates a new random number generator using a single {@code long} seed.

   126      * The seed is the initial value of the internal state of the pseudorandom

   127      * number generator which is maintained by method {@link #next}.

   128      *

   129      * <p>The invocation {@code new Random(seed)} is equivalent to:

   130      *  <pre> {@code

   131      * Random rnd = new Random();

   132      * rnd.setSeed(seed);}</pre>

   133      *

   134      * @param seed the initial seed

   135      * @see   #setSeed(long)

   136      */

   137     public Random(long seed) {

   138         if (getClass() == Random.class)

   139             this.seed = new AtomicLong(initialScramble(seed));

   140         else {

   141             // subclass might have overriden setSeed

   142             this.seed = new AtomicLong();

   143             setSeed(seed);

   144         }

   145     }

   146 

   147     private static long initialScramble(long seed) {

   148         return (seed ^ multiplier) & mask;

   149     }

   150 

   151     /**

   152      * Sets the seed of this random number generator using a single

   153      * {@code long} seed. The general contract of {@code setSeed} is

   154      * that it alters the state of this random number generator object

   155      * so as to be in exactly the same state as if it had just been

   156      * created with the argument {@code seed} as a seed. The method

   157      * {@code setSeed} is implemented by class {@code Random} by

   158      * atomically updating the seed to

   159      *  <pre>{@code (seed ^ 0x5DEECE66DL) & ((1L << 48) - 1)}</pre>

   160      * and clearing the {@code haveNextNextGaussian} flag used by {@link

   161      * #nextGaussian}.

   162      *

   163      * <p>The implementation of {@code setSeed} by class {@code Random}

   164      * happens to use only 48 bits of the given seed. In general, however,

   165      * an overriding method may use all 64 bits of the {@code long}

   166      * argument as a seed value.

   167      *

   168      * @param seed the initial seed

   169      */

   170     synchronized public void setSeed(long seed) {

   171         this.seed.set(initialScramble(seed));

   172         haveNextNextGaussian = false;

   173     }

   174 

   175     /**

   176      * Generates the next pseudorandom number. Subclasses should

   177      * override this, as this is used by all other methods.

   178      *

   179      * <p>The general contract of {@code next} is that it returns an

   180      * {@code int} value and if the argument {@code bits} is between

   181      * {@code 1} and {@code 32} (inclusive), then that many low-order

   182      * bits of the returned value will be (approximately) independently

   183      * chosen bit values, each of which is (approximately) equally

   184      * likely to be {@code 0} or {@code 1}. The method {@code next} is

   185      * implemented by class {@code Random} by atomically updating the seed to

   186      *  <pre>{@code (seed * 0x5DEECE66DL + 0xBL) & ((1L << 48) - 1)}</pre>

   187      * and returning

   188      *  <pre>{@code (int)(seed >>> (48 - bits))}.</pre>

   189      *

   190      * This is a linear congruential pseudorandom number generator, as

   191      * defined by D. H. Lehmer and described by Donald E. Knuth in

   192      * <i>The Art of Computer Programming,</i> Volume 3:

   193      * <i>Seminumerical Algorithms</i>, section 3.2.1.

   194      *

   195      * @param  bits random bits

   196      * @return the next pseudorandom value from this random number

   197      *         generator's sequence

   198      * @since  1.1

   199      */

   200     protected int next(int bits) {

   201         long oldseed, nextseed;

   202         AtomicLong seed = this.seed;

   203         do {

   204             oldseed = seed.get();

   205             nextseed = (oldseed * multiplier + addend) & mask;

   206         } while (!seed.compareAndSet(oldseed, nextseed));

   207         return (int)(nextseed >>> (48 - bits));

   208     }

   209 

   210     static final BigInteger thePeriod = BigInteger.valueOf(1L<<48);  // Period is 2**48

   211 

   212     /**

   213      * Returns the period of this random number generator.

   214      *

   215      * @return the period of this random number generator.

   216      */

   217     public BigInteger period() {

   218 	// Here we also take care of checking for instances of class SecureRandom,

   219 	// just so as not to bother the implementors of that class.

   220 	// (Any specific instance of SecureRandom can of course override this method.)

   221 	// The cast to (Object) is of course needed only during development.

   222 	return ((Object)this instanceof java.security.SecureRandom) ? Rng.HUGE_PERIOD : thePeriod;

   223     }

   224 

   225     /**

   226      * Generates random bytes and places them into a user-supplied

   227      * byte array.  The number of random bytes produced is equal to

   228      * the length of the byte array.

   229      *

   230      * <p>The method {@code nextBytes} is implemented by class {@code Random}

   231      * as if by:

   232      *  <pre> {@code

   233      * public void nextBytes(byte[] bytes) {

   234      *   for (int i = 0; i < bytes.length; )

   235      *     for (int rnd = nextInt(), n = Math.min(bytes.length - i, 4);

   236      *          n-- > 0; rnd >>= 8)

   237      *       bytes[i++] = (byte)rnd;

   238      * }}</pre>

   239      *

   240      * @param  bytes the byte array to fill with random bytes

   241      * @throws NullPointerException if the byte array is null

   242      * @since  1.1

   243      */

   244     public void nextBytes(byte[] bytes) {

   245         for (int i = 0, len = bytes.length; i < len; )

   246             for (int rnd = nextInt(),

   247                      n = Math.min(len - i, Integer.SIZE/Byte.SIZE);

   248                  n-- > 0; rnd >>= Byte.SIZE)

   249                 bytes[i++] = (byte)rnd;

   250     }

   251 

   252     /**

   253      * Returns the next pseudorandom, uniformly distributed {@code int}

   254      * value from this random number generator's sequence. The general

   255      * contract of {@code nextInt} is that one {@code int} value is

   256      * pseudorandomly generated and returned. All 2<sup>32</sup> possible

   257      * {@code int} values are produced with (approximately) equal probability.

   258      *

   259      * <p>The method {@code nextInt} is implemented by class {@code Random}

   260      * as if by:

   261      *  <pre> {@code

   262      * public int nextInt() {

   263      *   return next(32);

   264      * }}</pre>

   265      *

   266      * @return the next pseudorandom, uniformly distributed {@code int}

   267      *         value from this random number generator's sequence

   268      */

   269     public int nextInt() {

   270         return next(32);

   271     }

   272 

   273     /**

   274      * Returns a pseudorandom {@code int} value between zero (inclusive)

   275      * and the specified bound (exclusive).

   276      *

   277      * @param bound the upper bound (exclusive).  Must be positive.

   278      * @return a pseudorandom {@code int} value between zero

   279      *         (inclusive) and the bound (exclusive)

   280      * @throws IllegalArgumentException if {@code bound} is not positive

   281      */

   282     public int nextInt(int bound) {

   283         if (bound <= 0)

   284             throw new IllegalArgumentException(BadBound);

   285         // Specialize internalNextInt for origin 0

   286         int r = nextInt();

   287         int m = bound - 1;

   288         if ((bound & m) == 0) // power of two

   289             r &= m;

   290         else { // reject over-represented candidates

   291             for (int u = r >>> 1;

   292                  u + m - (r = u % bound) < 0;

   293                  u = nextInt() >>> 1)

   294                 ;

   295         }

   296         return r;

   297     }

   298 

   299     /**

   300      * Returns the next pseudorandom, uniformly distributed {@code long}

   301      * value from this random number generator's sequence. The general

   302      * contract of {@code nextLong} is that one {@code long} value is

   303      * pseudorandomly generated and returned.

   304      *

   305      * <p>The method {@code nextLong} is implemented by class {@code Random}

   306      * as if by:

   307      *  <pre> {@code

   308      * public long nextLong() {

   309      *   return ((long)next(32) << 32) + next(32);

   310      * }}</pre>

   311      *

   312      * Because class {@code Random} uses a seed with only 48 bits,

   313      * this algorithm will not return all possible {@code long} values.

   314      *

   315      * @return the next pseudorandom, uniformly distributed {@code long}

   316      *         value from this random number generator's sequence

   317      */

   318     public long nextLong() {

   319         // it's okay that the bottom word remains signed.

   320         return ((long)(next(32)) << 32) + next(32);

   321     }

   322 

   323     /**

   324      * Returns the next pseudorandom, uniformly distributed

   325      * {@code boolean} value from this random number generator's

   326      * sequence. The general contract of {@code nextBoolean} is that one

   327      * {@code boolean} value is pseudorandomly generated and returned.  The

   328      * values {@code true} and {@code false} are produced with

   329      * (approximately) equal probability.

   330      *

   331      * <p>The method {@code nextBoolean} is implemented by class {@code Random}

   332      * as if by:

   333      *  <pre> {@code

   334      * public boolean nextBoolean() {

   335      *   return next(1) != 0;

   336      * }}</pre>

   337      *

   338      * @return the next pseudorandom, uniformly distributed

   339      *         {@code boolean} value from this random number generator's

   340      *         sequence

   341      * @since 1.2

   342      */

   343     public boolean nextBoolean() {

   344         return next(1) != 0;

   345     }

   346 

   347     /**

   348      * Returns the next pseudorandom, uniformly distributed {@code float}

   349      * value between {@code 0.0} and {@code 1.0} from this random

   350      * number generator's sequence.

   351      *

   352      * <p>The general contract of {@code nextFloat} is that one

   353      * {@code float} value, chosen (approximately) uniformly from the

   354      * range {@code 0.0f} (inclusive) to {@code 1.0f} (exclusive), is

   355      * pseudorandomly generated and returned. All 2<sup>24</sup> possible

   356      * {@code float} values of the form <i>m&nbsp;x&nbsp;</i>2<sup>-24</sup>,

   357      * where <i>m</i> is a positive integer less than 2<sup>24</sup>, are

   358      * produced with (approximately) equal probability.

   359      *

   360      * <p>The method {@code nextFloat} is implemented by class {@code Random}

   361      * as if by:

   362      *  <pre> {@code

   363      * public float nextFloat() {

   364      *   return next(24) / ((float)(1 << 24));

   365      * }}</pre>

   366      *

   367      * <p>The hedge "approximately" is used in the foregoing description only

   368      * because the next method is only approximately an unbiased source of

   369      * independently chosen bits. If it were a perfect source of randomly

   370      * chosen bits, then the algorithm shown would choose {@code float}

   371      * values from the stated range with perfect uniformity.<p>

   372      * [In early versions of Java, the result was incorrectly calculated as:

   373      *  <pre> {@code

   374      *   return next(30) / ((float)(1 << 30));}</pre>

   375      * This might seem to be equivalent, if not better, but in fact it

   376      * introduced a slight nonuniformity because of the bias in the rounding

   377      * of floating-point numbers: it was slightly more likely that the

   378      * low-order bit of the significand would be 0 than that it would be 1.]

   379      *

   380      * @return the next pseudorandom, uniformly distributed {@code float}

   381      *         value between {@code 0.0} and {@code 1.0} from this

   382      *         random number generator's sequence

   383      */

   384     public float nextFloat() {

   385         return next(24) / ((float)(1 << 24));

   386     }

   387 

   388     /**

   389      * Returns the next pseudorandom, uniformly distributed

   390      * {@code double} value between {@code 0.0} and

   391      * {@code 1.0} from this random number generator's sequence.

   392      *

   393      * <p>The general contract of {@code nextDouble} is that one

   394      * {@code double} value, chosen (approximately) uniformly from the

   395      * range {@code 0.0d} (inclusive) to {@code 1.0d} (exclusive), is

   396      * pseudorandomly generated and returned.

   397      *

   398      * <p>The method {@code nextDouble} is implemented by class {@code Random}

   399      * as if by:

   400      *  <pre> {@code

   401      * public double nextDouble() {

   402      *   return (((long)next(26) << 27) + next(27))

   403      *     / (double)(1L << 53);

   404      * }}</pre>

   405      *

   406      * <p>The hedge "approximately" is used in the foregoing description only

   407      * because the {@code next} method is only approximately an unbiased

   408      * source of independently chosen bits. If it were a perfect source of

   409      * randomly chosen bits, then the algorithm shown would choose

   410      * {@code double} values from the stated range with perfect uniformity.

   411      * <p>[In early versions of Java, the result was incorrectly calculated as:

   412      *  <pre> {@code

   413      *   return (((long)next(27) << 27) + next(27))

   414      *     / (double)(1L << 54);}</pre>

   415      * This might seem to be equivalent, if not better, but in fact it

   416      * introduced a large nonuniformity because of the bias in the rounding

   417      * of floating-point numbers: it was three times as likely that the

   418      * low-order bit of the significand would be 0 than that it would be 1!

   419      * This nonuniformity probably doesn't matter much in practice, but we

   420      * strive for perfection.]

   421      *

   422      * @return the next pseudorandom, uniformly distributed {@code double}

   423      *         value between {@code 0.0} and {@code 1.0} from this

   424      *         random number generator's sequence

   425      * @see Math#random

   426      */

   427     public double nextDouble() {

   428         return (((long)(next(26)) << 27) + next(27)) * DOUBLE_UNIT;

   429     }

   430 

   431     private double nextNextGaussian;

   432     private boolean haveNextNextGaussian = false;

   433 

   434     /**

   435      * Returns the next pseudorandom, Gaussian ("normally") distributed

   436      * {@code double} value with mean {@code 0.0} and standard

   437      * deviation {@code 1.0} from this random number generator's sequence.

   438      * <p>

   439      * The general contract of {@code nextGaussian} is that one

   440      * {@code double} value, chosen from (approximately) the usual

   441      * normal distribution with mean {@code 0.0} and standard deviation

   442      * {@code 1.0}, is pseudorandomly generated and returned.

   443      *

   444      * <p>The method {@code nextGaussian} is implemented by class

   445      * {@code Random} as if by a threadsafe version of the following:

   446      *  <pre> {@code

   447      * private double nextNextGaussian;

   448      * private boolean haveNextNextGaussian = false;

   449      *

   450      * public double nextGaussian() {

   451      *   if (haveNextNextGaussian) {

   452      *     haveNextNextGaussian = false;

   453      *     return nextNextGaussian;

   454      *   } else {

   455      *     double v1, v2, s;

   456      *     do {

   457      *       v1 = 2 * nextDouble() - 1;   // between -1.0 and 1.0

   458      *       v2 = 2 * nextDouble() - 1;   // between -1.0 and 1.0

   459      *       s = v1 * v1 + v2 * v2;

   460      *     } while (s >= 1 || s == 0);

   461      *     double multiplier = StrictMath.sqrt(-2 * StrictMath.log(s)/s);

   462      *     nextNextGaussian = v2 * multiplier;

   463      *     haveNextNextGaussian = true;

   464      *     return v1 * multiplier;

   465      *   }

   466      * }}</pre>

   467      * This uses the <i>polar method</i> of G. E. P. Box, M. E. Muller, and

   468      * G. Marsaglia, as described by Donald E. Knuth in <i>The Art of

   469      * Computer Programming</i>, Volume 3: <i>Seminumerical Algorithms</i>,

   470      * section 3.4.1, subsection C, algorithm P. Note that it generates two

   471      * independent values at the cost of only one call to {@code StrictMath.log}

   472      * and one call to {@code StrictMath.sqrt}.

   473      *

   474      * @return the next pseudorandom, Gaussian ("normally") distributed

   475      *         {@code double} value with mean {@code 0.0} and

   476      *         standard deviation {@code 1.0} from this random number

   477      *         generator's sequence

   478      */

   479     synchronized public double nextGaussian() {

   480         // See Knuth, ACP, Section 3.4.1 Algorithm C.

   481         if (haveNextNextGaussian) {

   482             haveNextNextGaussian = false;

   483             return nextNextGaussian;

   484         } else {

   485             double v1, v2, s;

   486             do {

   487                 v1 = 2 * nextDouble() - 1; // between -1 and 1

   488                 v2 = 2 * nextDouble() - 1; // between -1 and 1

   489                 s = v1 * v1 + v2 * v2;

   490             } while (s >= 1 || s == 0);

   491             double multiplier = StrictMath.sqrt(-2 * StrictMath.log(s)/s);

   492             nextNextGaussian = v2 * multiplier;

   493             haveNextNextGaussian = true;

   494             return v1 * multiplier;

   495         }

   496     }

   497 

   498     /**

   499      * Serializable fields for Random.

   500      *

   501      * @serialField    seed long

   502      *              seed for random computations

   503      * @serialField    nextNextGaussian double

   504      *              next Gaussian to be returned

   505      * @serialField      haveNextNextGaussian boolean

   506      *              nextNextGaussian is valid

   507      */

   508     private static final ObjectStreamField[] serialPersistentFields = {

   509         new ObjectStreamField("seed", Long.TYPE),

   510         new ObjectStreamField("nextNextGaussian", Double.TYPE),

   511         new ObjectStreamField("haveNextNextGaussian", Boolean.TYPE)

   512     };

   513 

   514     /**

   515      * Reconstitute the {@code Random} instance from a stream (that is,

   516      * deserialize it).

   517      */

   518     private void readObject(java.io.ObjectInputStream s)

   519         throws java.io.IOException, ClassNotFoundException {

   520 

   521         ObjectInputStream.GetField fields = s.readFields();

   522 

   523         // The seed is read in as {@code long} for

   524         // historical reasons, but it is converted to an AtomicLong.

   525         long seedVal = fields.get("seed", -1L);

   526         if (seedVal < 0)

   527           throw new java.io.StreamCorruptedException(

   528                               "Random: invalid seed");

   529         resetSeed(seedVal);

   530         nextNextGaussian = fields.get("nextNextGaussian", 0.0);

   531         haveNextNextGaussian = fields.get("haveNextNextGaussian", false);

   532     }

   533 

   534     /**

   535      * Save the {@code Random} instance to a stream.

   536      */

   537     synchronized private void writeObject(ObjectOutputStream s)

   538         throws IOException {

   539 

   540         // set the values of the Serializable fields

   541         ObjectOutputStream.PutField fields = s.putFields();

   542 

   543         // The seed is serialized as a long for historical reasons.

   544         fields.put("seed", seed.get());

   545         fields.put("nextNextGaussian", nextNextGaussian);

   546         fields.put("haveNextNextGaussian", haveNextNextGaussian);

   547 

   548         // save them

   549         s.writeFields();

   550     }

   551 

   552     // Support for resetting seed while deserializing

   553     private static final Unsafe unsafe = Unsafe.getUnsafe();

   554     private static final long seedOffset;

   555     static {

   556         try {

   557             seedOffset = unsafe.objectFieldOffset

   558                 (Random.class.getDeclaredField("seed"));

   559         } catch (Exception ex) { throw new Error(ex); }

   560     }

   561     private void resetSeed(long seedVal) {

   562         unsafe.putObjectVolatile(this, seedOffset, new AtomicLong(seedVal));

   563     }

   564 }