animals running pre-REL_10 scripts will not report full information about failures in the pg_upgrade test. This is particularly problematic right now, because we're wondering what the heck is going on with intermittent failures in that test:
There's a patch in place to try to gather more info about that, but it's useless if the affected animal doesn't capture the regression.diffs file. Thus for example, mayfly's latest failure might have yielded useful information, except it's using REL_8 :-(. We're only seeing maybe one occurrence every couple of weeks, so missed opportunities are painful.
