On 26/10/2007, Stefan Dösinger stefan@codeweavers.com wrote:
Am Freitag, 26. Oktober 2007 01:22:22 schrieb Reece Dunn:
Also, I suspect that it would be worth adding vmware in the test cases. That is, something like:
ok( ret == 129 /* Windows */ || ret == 133 /* VMware */, ... );
This would mean that the tests that are consistently (and predictably) failing under vmware will not show up on the test failures, so we can investigate actual failures.
Well, then we could add a ret == 666 /* Parallels */ too... I am not really fond of "fixing" our tests to succeed in vmware, I'd rather skip them entirely. Although the technically correct thing to do is to just let them fail. VMware has a broken d3d implementation, our tests detect that, mission accomplished.
This is true. The tests should only check for valid Windows results.
The problem is how to filter out the noise of the VMware bugs from actual failures in Windows for the Wine tests. This is so that we know which tests need fixing (so that the Wine behaviour can be corrected), and those that don't.
Thinking about this, my idea of updating the tests for the broken VMware implementation was wrong. However, I don't have a better solution.
Appart of that, with these specific tests I have seen one failure on a real windows box too, perhaps I should loosen the test precision on some of those tests(This was in the area of 0.000x, not a difference of 129 and 133).
Having a +/- 0.000x tolerance is sensible.
- Reece