[EMAIL PROTECTED] schrieb: > Niall Pemberton schrieb: > >> On Mon, Mar 3, 2008 at 3:26 AM, James Carman <[EMAIL PROTECTED]> wrote: >> >> >>> On 3/2/08, Niall Pemberton <[EMAIL PROTECTED]> wrote: >>> > It may seem like I broke a number of builds with my changes today, but >>> > half of them [codec, discovery, JCI, VFS] were already failing before, >>> > for the rest... >>> > >>> > 1) Validator - now fixed (caused by upgrade of surefire plugin from >>> > 2.3 to 2.4.1 in commons-parent). >>> > 2) IO - very strange, I upgraded the build profile from Java 1.4 to >>> > Java 5 a few weeks ago (see previous build[1]) - so either Continuum >>> > *lost* my change or someone reverted it. Anyway I've set it again to >>> > Java 5, so it should build OK after the next change to IO >>> > 3) Lang - not sure why it failed - works OK for me locally >>> > 4) Logging - I get this failure, reverting to surefire 2.3 fixes this >>> > - but not sure how to fix this properly >>> > >>> >>> We've had some build failures when we upgraded from 2.3 also. The >>> problem we saw was that the tests weren't run in the same order >>> anymore for some reason. Could that be what you're seeing? >>> >>> >> I don't know the internals of commons logging - hopefully one of the >> logging devs will have more of a clue. When I run mvn integration-test >> I see the following error >> >> org.apache.maven.surefire.testset.TestSetFailedException: >> org.apache.commons.logging.logkit.StandardTestCase; nested exception >> is java.lang.UnknownError: Logical lib [logkit] is not defined as a >> System property. >> >> Don't know if its related but logkit.StandardTestCase' s suite() >> method is called twice using Surefire 2.3 - but only once with >> Surefire 2.4.2 >> >> > > I'll try to find time to look into it, but it won't be for at least a > couple of weeks. > > Commons-logging does some *very* unusual things with classpaths during > integration-tests, because the tests need to check how things work in > all sorts of odd classloader setups. The same test is sometimes called > multiple times, for example, with different classpaths set up on each > invocation. > > Logging's test cases should be reasonably well documented though; I put > some effort into this as it is so unusual. > Oh, and by the way the easiest solution is just to lock commons-logging's version of maven-surefire-plugin to 2.3.1 (or whatever works).
Locking plugin versions is good practice anyway; there is no need to upgrade to a later version of a plugin if an earlier version does everything that is currently needed, so logging can just stick with 2.3.1 until we find that we need something newer - and fix the problem then. Feel free to update the commons-logging pom in trunk, or I will do that in the next few days. Cheers, Simon --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
