[MATH-354] compiler dependency in RandomDataImpl.getPoisson(double mean)

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0
• Fix Version/s: 2.1

Environment

jre1.5/jre1.6

• Assignee: Unassigned
• Reporter: Michael Sammeth
• Created: 18/Mar/10 21:08
• Updated: 03/Apr/10 03:05
• Resolved: 18/Mar/10 22:34

Description

Hi,

in RandomDataImpl.getPoisson(double mean) I got the following problem in the case mean>= 6.0:

in the branch if (u <= c1):

if (x < -mu)
w = Double.POSITIVE_INFINITY;

implicits that (int) (mu+ x) < 0

I found that for some compiler/run-time environments the subsequent update of the "accept" value then fails, as by the right hand side of the comparison leads to an Exception in MathUtils.factorialLog((int) (mu + x)). Some compiler/jre combinations, however, skip evaluating the right side as by isInfinity(w).

To ensure stability, I currently worked around by an explicit if(Double.isInfinity(w)) branch, however, I would like to ask whether there is a more elegant way to ensure general functionality of that method.

Thank you, micha.