In the last few hours I pushed my local changes on the SELinux userland utilities towards the hardened-development overlay. The utilities not only include some bugfixes, but have now also seen a first set of tests towards Python 3.2. In the past, I've made a few attempts at making the tools support Python 3, but I failed miserably. Although chances are still high that I failed, at least I got quite a bit further.
To make testing a bit easier, I previously made quite a few scripts that did all sorts of things, in order to catch regressions. However, along the way, I've started noticing I had to put lots of effort in streamlining these tests (cleanups), introducing dependency information (test A before B, or cleanup before test, ...) and parallellism (after all, if you have many, many tests, lots of cores but run in single-thread/process mode, it'll take a while). So I started looking at a good way to handle this for me. I switched my tests into a Makefile-driven approach.
Why Makefiles? Well, first of all, Makefiles support dependencies. You can define a target and then say which other targets need to be ran first before this target can run. If you want to support these dependencies in a run-independent manner, you can use trigger files but I'm not going to do that yet. Another simple feature is that you can tell make to not show output (silent mode) when not necessary. And of course, with make, you can execute targets concurrently. By using a simple, yet manageable directory structure and traverse the Makefiles in them, I am able to easily add more tests and add them to the runqueue.
But I'd like to hear from you what infrastructural testing tools you use because I can imagine Makefiles aren't the best solution here.
In the mean time, I also pushed the 10th revision of our SELinux policies to the hardened-dev overlay. The most notable fix in it is to improve support for those running \~arch systems (some fix on the kdevtmpfs support).