On 02/08/07, Rafael Schloming <[EMAIL PROTECTED]> wrote: > Martin Ritchie wrote: > > On 02/08/07, Gordon Sim <[EMAIL PROTECTED]> wrote: > >> Martin Ritchie wrote: > >>> On 01/08/07, Rajith Attapattu <[EMAIL PROTECTED]> wrote: > >>>> Martin, > >>>> > >>>> A python test execution fails on the broker in trunk. > >>>> As hack I get around by skipping them. > >>>> Can you please look into it. > >>>> > >>>> Regards, > >>>> > >>>> Rajith > >>> I hope to have time to look at trunk again next week. Did the test > >>> just start failing? Have the python tests changed recently? > >> The python test target was failing entirely when run from maven due to > >> an incorrect filename for the exempted tests. I've checked in a trivial > >> change for that. > >> > >> Two of the actual tests fail: test_rollback and test_auto_rollback. The > >> tests themselves have not changed, but the python client has had some > >> modifications. However using the client from trunk, those tests pass > >> against the M2 java broker so I suspect that it is not a python issue. > >> > >> I'll be happy to investigate further if someone has reason to believe > >> that it is indeed the python client that is causing these failures. > > > > I don't have any cause to blame the python tests I just don't know of > > any changes to the trunk broker. Perhaps merging the M2 changes to > > trunk should be done first as perhaps there was a fix on the branch > > that hasn't made it to trunk yet. Though I thought all the tests were > > passing before we branched. > > > > If it can wait till next week I'll have time to look at it then. > > I believe there is another failure with the ant script that starts the > python tests. If you do an mvn clean, followed by an mvn install, the > ant script that runs the python tests fails with a > ClassNotFoundException when trying to find RunBrokerWithCommand. You can > work around it by doing an mvn install -Dskip-python-tests, and it does > work on subsequent builds if you do an mvn install without cleaning. > > I suspect this means that there is a cyclic dependency in the tests, > i.e. the install phase needs to have ocurred for the classpath used in > the ant script to be valid. This probably means that even when the tests > work on subsequent builds they are actually running against whatever the > last copy of the broker is that you successfully installed rather than > running against an up-to-date build of the broker. > > I tried to figure out what the various maven properties mean and if > there was one more appropriate to use than maven.test.classpath, but > unfortunately I have had very little luck finding any decent level of > detailed documentation on the maven site. > > --Rafael
How odd. I've currently been building the M2 broker with a fresh repository (So no previous broker version to pick up) and never seen this problem. There must be something funky going on with the trunk java broker. As you point out the -Dskip.python.tests will allow you to get pass this problem for now but I'll have more time next week to look at the problems. -- Martin Ritchie
