On 12/18/06, Alexander Kleymenov <[EMAIL PROTECTED]> wrote:
> > please, don't forget that JVM can crash before creating of empty xml > > file. > > In this case any start of vm should leads to crash. It will be easy to > reproduce. I can't agree with you. While working on regression tests I saw the situations when the same VM crashed on one test and successfully passed on another. This issue of an approach can lead to serious problems to be unnoticed. So I conclude it can't be used for DRLVM testing.
If no report file was generated it is usually mean that VM can't start (like em64t build 2 weeks ago :)). Actually the report file creation code works before test so it failed always or passed always. I just want to say that in case vm crashes on some particular case the empty report file will be created. By the way, if the same approach will be applied to DRLVM testing empty file detection will works for it too.
> And what about the tests not using xml formatter? > > I think, we can generate xml-files by build as post-processing of test run. > thanks, Vladimir Now I see this approach is loosing its main feature - simplicity.
No, it is very simple and does not depend on external tasks (only ant-core tasks). I agree that extension of junit task also can solve this problem but let it be the next step if the first simple step does not solve it. Thanks, Vladimir
It's starting to look as a set of workarounds for more and more arising problems. While other proposed approach based on task extending looks quite natural and simple [1]. By using of standard Ant's means (with available well-documented API) it allows to implement all the things we need for JVM testing: crash handling, std/err test output collection, and crash reporting.I use it for DRLVM regression testing and it works well. However, there is an opinion that JVM crash-handling means should be common for each task we do. So let's decide on approach to choose for further use. Thanks, Alexander [1] http://issues.apache.org/jira/browse/HARMONY-2416
