|Re: [jgit-dev] Strange failing tests under Linux which go through when debugging|
måndagen den 12 juli 2010 13.28.12 skrev Matthias Sohn: > 2010/7/12 Christian Halstrick <christian.halstrick@xxxxxxxxx> > > > Hi, > > > > Robin pointed me to a problem that through JGIT commit > > 4be88168b69465cc our ReadTreeTest don't work anymore. The situation is > > so strange that I ask whether somebody has faced/solved that problem. > > I am reading a Index through GitIndex and it > > - works on windows > > - works on Linux (Ubuntu), OpenJDK 6b18) when I am debugging the test > > - doesn't work on Linux when I running without debugging (or when > > hudson runs the test). The tests fail regardless whether I run the > > JUnit test from Eclipse or on cmdline through "mvn clean test" > > > > Some timing problems when reading with GitIndex, has anybody heard about > > it? > > > > Here the detail: > > WorkDirCheckout_ReadTreeTest#testDirectoryFileSimple() the last > > assertion fails. The last call to preScanTwoTrees finds no index entry > > for "DF" when I debug ... but it finds one when I am not debugging. > > Because of that the "updated" set is not filled and the assertion > > failed. > > > > Any ideas anybody? > > How about the racy git problem ? It's hurting more on Linux since > filesystem timestamps > come with a 1s resolution only. Depends on filesystem actually. My filesystem has nanosecond resolution, though probably not with nanoseconds accuracy. -- robin'
Back to the top