Sebb, Can you try running with the JIT disabled (-Xint) and see if the test fails the same way?
Thanks, Tim On 05/Sep/2009 02:02, Sebb (JIRA) wrote: > [ > https://issues.apache.org/jira/browse/HARMONY-6316?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12751680#action_12751680 > ] > > Sebb commented on HARMONY-6316: > ------------------------------- > > I've been investigating the ArrayIndexOutOfBoundsException in the test > EigenDecompositionImplTest.testBigMatrix(). > > In the file linear/EigenDecompositionImpl.java at line 1586 there is the > following for loop: > > for (int i4 = 4 * end - 10 + pingPong; i4 >= 4 * start + 2 + pingPong; i4 > -= 4) { > > I found that changing this to: > > final int dummy = 1; > for (int i4 = 4 * end - 10 + pingPong; i4 >= 4 * start + 2+ pingPong; i4 > -= 4) { > > prevents the array bound error from occurring. > > This suggests to me that the JVM is perhaps re-using a work register without > saving/restoring it. > The additional assignment statement somehow causes the JVM to be stricter > about register usage. > > I tried catching the AIOBE and printing the index. > It was a large number (406591607) , but the other variables still had their > original values. > This again suggests the JVM may be using a stale cached value for the for > loop calculations. > > I suspect this may be a similar problem to the testAngles() failure. > >> Errors and hang when running Commons Math 2.0 test suite >> -------------------------------------------------------- >> >> Key: HARMONY-6316 >> URL: https://issues.apache.org/jira/browse/HARMONY-6316 >> Project: Harmony >> Issue Type: Bug >> Affects Versions: 5.0M11 >> Environment: Maven 2.1.0 >> Win XP/SP3 >> Harmony svn = r782693, (Jun 12 2009), Windows/ia32/msvc 1310, release build >> Reporter: Sebb >> Attachments: HarmonyDoubleBug.java, surefire-log.txt, >> surefire-reports.zip >> >> >> I tried testing Commons Math 2.0 against Harmony r782693, (Jun 12 2009) and >> found a few strange bugs where variables appeared to have "impossible" >> values, causing array index out of bounds errors. >> Also, at least one test hangs: >> Running org.apache.commons.math.geometry.RotationTest. >> See attached Surefire output. >
