multisample 0 coverage depth writes?

[ccbrianf]

With alpha to coverage, alpha to one, and the depth test enabled, the depth buffer unmasked, and a depth test of less than or equal to, would you expect a 50 % alpha polygon rendered to a clear (color, depth, and stencil) frame buffer at z = x to cause the clear color to occlude a fully opaque polygon rendered at depth z = x + 1, where x and x + 1 are resolvable depth values?

It would appear that it does with current ATI drivers, but not with current nVidia ones. I believe some previous ATI drivers (I can't remember the exact versions) did not behave this way.

Which behavior is correct? Thanks.

[Brolingstanz]

What (if anything) does the spec say about it? (I haven't looked at that section lately.)

[ccbrianf]

Originally posted by Leghorn:
What (if anything) does the spec say about it? (I haven't looked at that section lately.) I can't figure that out. That's why I posted the question ;-).

It does appear to say that each sample should have an individual depth value, so that would imply to me that the samples should be treated separately in the depth test. But, I may not even have that much right.

If anyone knows positively, or has an interpretation of the spec on this point, I'd love to hear about it. Thanks.