Daniel Shahaf wrote on Wed, 01 Apr 2020 10:26 +0000: > Stefan Sperling wrote on Wed, 01 Apr 2020 12:18 +0200: > > On Wed, Apr 01, 2020 at 09:59:41AM +0000, Daniel Shahaf wrote: > > > Stefan Sperling wrote on Wed, 01 Apr 2020 11:13 +0200: > > > > https://ci.apache.org/builders/svn-x64-macosx-fsfs-v6/builds/2941/steps/Test%20ra_local%2Bfsfs-v6/logs/faillog > > > > > > > > The svnadmin build-repcache test runs into a checksum mismatch. > > > > > > > > Any ideas? > > > > > > Note the "actual" checksum is the checksum of the empty string. > > > > > > > Oh, interesting point :) > > > > I cannot see what's wrong. An empty stream is returned from > > svn_fs_fs__get_contents() if the passed in rep pointer is NULL, > > but I don't think that's where the bug is. > > > > Perhaps the failing test really read empty rep content for some reason? > > I'm now running this test in a loop, but I could not reproduce the > > failure yet. > > Are you running your local tests using the same configuration the > builder uses? (Even without looking up the build script, the builder's > name suggests it uses FSFS f6, i.e., --server-minor-version.) >
The failure is 100% reproducible with «./svnadmin_tests.py --fs-type=fsfs --fsfs-version=6 74» on Linux. Let's wait to hear from Denis. > If you haven't already, re-run that builder using subversion-bot to > check if the bug is reproducible. > > The error trace goes through cache.c; SVN_X_DOES_NOT_MARK_THE_SPOT in > notes/knobs can be used to rule out svn_cache__t as the cause.
