We currently do a fair amount of redundant testing on multiple configs where the differences aren't important (e.g. running cross_compilation on host, host-cuda, host-opengl, etc). It would be useful to avoid this somehow.
Perhaps we could add some way to tag each test with the environment(s) it needs to be tested in? We currently do this in a pretty crude way in some cases, by sniffing HL_TARGET at the start of tests and doing a [SKIP] if inappropriate. It would be nice if this sort of info was available as metadata to the test harness, so that the test wouldn't need building or invoking for these cases.