BINGO!!!!! Was actually the LHS class (but once I narrowed it down to two classes, wasn't that hard to figure out). Nicely played sir. I owe you several beverages of your choice. ApacheCon? Justin
________________________________ From: Stephen Connolly [mailto:[EMAIL PROTECTED] Sent: Thu 9/25/2008 4:09 PM To: user@mojo.codehaus.org Subject: Re: [mojo-user] test pass normally, but fail under cobertura I wonder if you have duplicate classes on your classpath? It may be that one set of them is resolving in one order while the other set it resolving in the other. Also doesn't Spring do some sort of classloader isolation... It's looking like it's probably a problem (unrelated to coverage) that has been exposed by running tests with coverage.... now if only you can figure out what the problem is! I'd start be checking if multiple jars in your dependency classpath define the class on the RHS of the instanceof expression 2008/9/25 Edelson, Justin <[EMAIL PROTECTED]> Good idea. Emma shows the same behavior. I'll try Clover next. The documentation of the Emma plugin is incorrect. Posted a patch as MOJO-1234. Would highly recommend someone apply the patch and do a site-deploy. Justin ________________________________ From: Stephen Connolly [mailto:[EMAIL PROTECTED] Sent: Thu 9/25/2008 3:10 PM To: user@mojo.codehaus.org Subject: Re: [mojo-user] test pass normally, but fail under cobertura Any chance you could give it a poke with emma or clover (they've a 30 day eval license for free) It might reveal if it's the instrumentation per se that is the problem or if it's limited to cobertura 2008/9/25 Edelson, Justin <[EMAIL PROTECTED]> See my other email about the instanceof problem. I don't see how this is a concurrency problem. It's the test on this line: http://fisheye1.atlassian.com/browse/springframework/spring/src/org/springframework/beans/factory/support/AbstractAutowireCapableBeanFactory.java?r=1.182#l1363 behaving differently. Justin ________________________________ From: Stephen Connolly [mailto:[EMAIL PROTECTED] Sent: Thu 9/25/2008 12:21 PM To: user@mojo.codehaus.org Subject: Re: [mojo-user] test pass normally, but fail under cobertura Noooo have you got parallel threads in your code? bet you have even if you thing you have not! bet you have not read Java Concurrency In Practice and have an unintended synchronization side-effect that is currently letting your tests pass on Sun's JVM... but they may fail on JRocket or IBM or when instrumented with emma/clover/ or cobertura 2008/9/25 Edelson, Justin <[EMAIL PROTECTED]> Thanks Stephen. Pretty sure it's not a timeout. I agree that tests should be able to be run any number of times. And in fact "mvn clean test test test test" works fine. It's "mvn cobertura:cobertura" which doesn't - and that only runs tests once. Justin ________________________________ From: Stephen Connolly [mailto:[EMAIL PROTECTED] Sent: Thu 9/25/2008 3:09 AM To: user@mojo.codehaus.org Subject: Re: [mojo-user] test pass normally, but fail under cobertura I have seen tests fail under cobertura when: 1. Tests have timeouts 2. Tests make incorrect assumptions about how the JVM memory model and synchronization work (or don't) This second one is why I always say run your tests twice ;-) 2008/9/25 Edelson, Justin <[EMAIL PROTECTED]> I've been struggling for the past 2 days with a set of Spring-based integration tests that pass when run normally (i.e. mvn test), but fail when run using the cobertura plugin. I recognize this is a shot in the dark, but if anyone has any ideas on things I can try, I'd appreciate it. Specifically, I'd be interested in any differences in the classpath between a normal test run and the cobertura test run. Thanks, Justin
