build (or at least provide automatable instructions to) build the test data files
diffoscope test suite is comprised of many binary files; unfortunately, many of those files can be quite obscure to anybody that hasn't studied their binary format, and in case of need they are very hard to replicate. Also, the presence of binaries without source can be seen as a form of sourcelss distribution, not very open-source friendly.
I understand that building them at build/test time can be tricky, since at that point we can't be sure anymore of whether the builds are going to be reproducible (heh) with enough certainty to not cause spurious test failures.
I suggest to at least provide a build target to be able to manually rebuild all the binary test files, and regularly try it to verify the thing is still working.