It looks like the current operation is "catch-all" or "catch-nothing". This is sub-optimal because if I only wish to test a small util in my "meta-foo" layer, I would not like to wait for every other ptest packages to be built. It is waste of time, resource and potentially money. Currently, I am planning to wait for this feature to be available before adopting. I understand and appreciate that the workaround is to remove the "ptest" from everywhere else, but that would be too intrusive compared to the gain of "ptest" as of now. I can just build my own "foo-test" for the time being, install it and then run my test script or binary on the target. That looks like a more pleasant way of achieving my goal. Having said, I would give a try to ptest in the future, once this report is fixed.
Enabling ptest globally doesn't imply installing all of the ptest packages, but if you want to avoid the overhead of building the ptest support for all but a few packages then you can selectively enable it per-recipe. For example assuming DISTRO_FEATURES doesn't contain ptest, then you can add it to specific recipes (eg glib-2.0) with DISTRO_FEATURES_append_pn-glib-2.0 = " ptest" in your configuration (local.conf, etc).