Testing for two different version numbers, one in the 5.8.x branch and one in the 5.9.x branch, seems just plain ridiculous.
Either the version numbering should be put on its head, or the main branch should be the version to test against. An argument for the latter would be that nobody should be using an old 5.9.x release, anyway (ahem!), or live with the consequences.
But now, everything would appear to be dandy for the advance developers, while the mainstream public (those using 5.8.x) end up with a broken test, and there's a hiccup in the results.