[MATH-720] RandomDataImpl uses both JDKRandomGenerator and Well19937c, while contract/javadoc only specifies Well19937c.

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0
• Labels: random seed
• Assignee: Unassigned
• Reporter: Dennis Hendriks
• Created: 07/Dec/11 14:04
• Updated: 23/Jul/12 23:24
• Resolved: 12/Dec/11 01:35

Description

See attached unit test. I create a distribution, sample it (not printed), set the seed to 0, and then print the next sample. I also create the same distribution again, set the seed to 0, and then print the next sample. I expect the same sample, as in both cases the seed was set to 0, just before sampling. I however get this output:

5
4

The problem is in the org.apache.commons.math.random.RandomDataImpl class:

• The RandomDataImpl(RandomGenerator rand) constructor states in javadoc: "@param rand the source of (non-secure) random data (may be null, resulting in default JDK-supplied generator)"
• reSeed(long seed) method does: if (rand == null) rand = new JDKRandomGenerator();
• reSeed() method does: if (rand == null) rand = new JDKRandomGenerator();
• class javadoc states: "If no <code>RandomGenerator</code> is provided in the constructor, the default is to use a Well19937c generator."
• getRan() does: if (rand == null) rand = new Well19937c(System.currentTimeMillis() + System.identityHashCode(this));
• getRan() states in javadoc: "Creates and initializes a default generator if null. Uses a Well19937c generator with System.currentTimeMillis() + System.identityHashCode(this)) as the default seed."

It seems that only Well19937c should be used, but the constructor javadoc, and the implementation of the reSeed methods was not updated to match this. I think the partial changes were done in MATH-701, more specifically, in commit [1197626] (and related commit [1197716]).

Issue Links

• is part of: MATH-701 Seeding a default RNG