Re: FW: [GUMP] Test Failure

Mark,

One possibility is to have log4j tests output both the witness file
and the tested file upon failure by modifying Compare.compare() method.

Thus, instead of

     [junit] Files [output/plugins.PluginTestCase.test1.txt] and
[witness/plugins.PluginTestCase.test1.txt] differ on line 65
     [junit] One reads:  [DEBUG - plugin plugin1-id1 shutdown].
     [junit] Other reads:[DEBUG - plugin plugin2-id4 shutdown].
     [junit] ------------- ---------------- ---------------

we would see

    [junit] Files [output/plugins.PluginTestCase.test1.txt] and
[witness/plugins.PluginTestCase.test1.txt] differ on line 65
     [junit] One reads:  [DEBUG - plugin plugin1-id1 shutdown].
     [junit] Other reads:[DEBUG - plugin plugin2-id4 shutdown].
     [junit] ------------- ---------------- ---------------
     [junit] Dumping [output/plugins.PluginTestCase.test1.txt]
     ....

     [junit] ------------- ---------------- ---------------
     [junit] Dumping [witness/plugins.PluginTestCase.test1.txt]
     ....

The advantage is that you do not the intervention of gump
developers to send you the witness file or the tested file.


As for the reasons for the failure, my hunch is that Sam machine
orders the plugins differently. It might be simply because his JDK
implements HaspMap differently or because it orders threads
differently. Since I do not see multiple threads in the test case, it
can only be the former possibility.

Let us take continue this on log4j-dev.


 > Hi Sam,
 >
 > I'll look at the dependencies, but it is not obvious how a difference
 > there would affect this particular test.  Is there any way that I can
 > access the generated test output files to see if there is a clue
 > there?
 >
 > -Mark


--
Ceki