[fzltrp]

> There is functionally no chance of that software bug ever getting patched; card vendors don't care about older versions of their technology,

Well, is that an OpenGL issue? Wouldn't you get that very same problem with D3D?

> reasonably-performant way

I don't understand. How can you objectively define such thing? Doesn't it depend on the workload? If you're pushing 3 tris per frame, any feature can be labeled as reasonably performing, but if you have 300M, can any card these days maintain reasonable framerate even on the most basic settings? I am exagerating on purpose; some apps will require a very small amount of work on each stage of rendering, and could reasonably afford any extra pass, even if software implemented. And in other cases (which might be the majority), it doesn't cut work. I don't see how there could be an objective way of deciding if a feature is sufficiently performant. Your example is telling: an application as complex as F.E.A.R. Should clearly make its benchmarks (or keep a database) to decide which feature can be included without hurting performances. And even then, players have also different perceptions of what constitutes playability.

I agree with you: multiple standards, multiple vendors, multiple products, the fallout is "struggling compatibilities" at worst, "varying performances" at best. But that's a common point between D3D and OpenGL, not a divergence. Am I missing something?