I currently have some WIP code that passes all tests with our default corpus, but fails with the smallest performance corpus. The simplest thing to do would be to add a small sample from our performance corpus as one for our standard (correctness) suite. I'm currently looking at 146 LKML messages. Unpacked these are about 1.3M; they bloat the source tarball by about 285K, which is large in relative terms (about 40%), but small in absolute terms for most modern systems. If we switch to xz compression, the resulting tarball is only 711K. So comments: 1) is it worth it to have a larger test corpus to blow up our source tarball size? 2) Should we (independently) switch to xz compression for our tarballs? I'm not very enthusiastic about complicating the test system with yet another kind of artifact to be downloaded. I think if we want to minimize the size of the test corpora, I'd probably just extract the troublesome thread, which will work for testing my current bug, but maybe not so good for finding future bugs. d