On Sun, Dec 25, 2011 at 11:57 AM, Stefan Fuhrmann <eq...@web.de> wrote: > On 25.12.2011 12:04, Bert Huijben wrote: >>> >>> -----Original Message----- > > >>> Log: >>> Minor xdelta optimization: find short matches at both end of the delta >>> window. >> >> All buildbots are red after this patch. >> >> At least one test FAILED, checking E:\w2k3\tests\tests.log >> FAIL: svnrdump_tests.py 10: dump: copy from previous version and modify >> FAIL: svnrdump_tests.py 11: load: copy from previous version and modify >> FAIL: svnrdump_tests.py 12: dump: modified in place >> FAIL: svnrdump_tests.py 13: load: modified in place >> FAIL: svnrdump_tests.py 26: dump: copy revprops other than svn:* >> FAIL: svnrdump_tests.py 27: load: copy revprops other than svn:* >> FAIL: svnrdump_tests.py 28: dump: subdirectory >> FAIL: svnrdump_tests.py 29: dump: subdirectory with changes on root >> FAIL: svnrdump_tests.py 32: dump: move parent& modify child file in same >> rev >> FAIL: svnrdump_tests.py 33: load: move parent& modify child file in same >> rev > > > Thanks for noticing! (I thought I had run the tests before committing :/ ) > > The problem here is that the test compares the actual dump files > produced by svnrdump with the expected ones stored in the test > data folder. With my patch, the text delta is now often smaller, i.e. > different from the expected value. Nevertheless the new diffs seem > to be correct. > > So, I will XFAIL the tests in question and $somebody needs to > update the reference dump files.
I ran across a similar problem in the svnsync tests while working on Ev2, and added what I think to be a solution in r1230676. Per Bert's suggestion, I think we can do the same thing in the svnrdump tests, and have implemented the same in r1230714. -Hyrum -- uberSVN: Apache Subversion Made Easy http://www.uberSVN.com/
