Joseph Garvin wrote:
...Assuming you're testing on cards with the same capabilities (you would have to record what extensions should correspond to what results) I don't see why the output shouldn't be the same.
I would expect alias/filtering to be driver-level dependent, with individual pixel outputs varying as optimizations are made. Since you'll never be using "exactly" the same driver in Linux as in Windows (and one that's not even close, if you only run open-source drivers), I think a conformance test would show a lot of inconsequential variances that aren't really things that need to be fixed.
You would see similar variances between different Windows machines, depending on the hardware/driver as well, so there's no single "correct" value for which to test.
-J