Code review comment for lp://staging/~compiz-team/compiz/compiz.performance_1027211.2

>
> +1
>
> >
> > 3. From a high-level perspective, even if the implementation turns out to be
> > right then I think maybe more analysis was required. This fix is an example
> of
> > where the medicine could be worse than the disease. It's only useful to
> nvidia
> > because nvidia is buggy IMHO. And the price to pay is more abstraction and
> > complication that hinders maintainability of the code. We might be better
> off
> > not trying to "fix" window movement any further. Though resizing needs work
> > still.
>
> It is a little more abstract and complex, yes, however having an unusable
> system on nvidia is not an acceptable alternative either.
>
> I don't think the code is "unmaintainable", I think it just works in a way
> that is a tad more complex and different to the way that you would expect. Its
> fully tested and encapsulated from the other code, so future maintainers don't
> need to worry about inadvertently breaking it (except for their usage of the
> client hooks).
>

On the same note.

Over my many years of developing compiz, I've often found that the little quirks between drivers, opengl, the way the xserver works, the way specs are implemented etc often means the issues aren't that simple, so its means that the implementation is almost never going to be simple.

If you have a look at any window manager, they are often large and complicated beasts, doing lots of things that you wouldn't even expect a window manager to have to do.

However, window managers often start small and grow to accomadate the quirks in their surrounding environment. The best approach, in my opinion, is to encapsulate complexity into clearly defined areas of the code, so that your handing of corner cases doesn't pollute your main buisness logic or get tangled in-between it. While doing that, ensure that the corner cases are tested codepaths that future maintainers can rely upon the tests when making changes.

This is something that I think compiz really succeeds at. While our test coverage is pretty absymal at best, no other non-niche window manager I can think of is as well tested as compiz is. And its starting to pay off - I've been able to rely on the test suite alone for making some changes now, which is something that I was never able to do.

That being said, I think you're right into looking into any more workarounds for the nvidia problems with ConfigureWindow requests and simultaneous redirected GL. We've found the main cause and root of the problem now, so lets stick with this implementation and work on other things. Window movement is 60FPS here with this applied, so I doubt we can make it any faster :)