[colanderman]

How much of that is actually due to a server-side model, and how much is due to brokenness of X?

With a well-specified server model (say using temporal logic), there should be no problem attaining pixel- and frame-perfect rendering.

[fmoralesc]

That's sounds like a plan! I suppose you have a tech demo of something like that?

[colanderman]

You're an ass. Check out some of the latest research into functional reactive programming. They're pretty far along that track.

Edit: if you want a hint at an actual solution, one need only send logical local timestamps along with each request, and introduce a timestamp handshake. The server very simply then need not swap in the backing store until it and the client have acknowledged that the logical timestamp should increment. Your convo then looks like this:

U - user / S - server / C - client

U->S: resize to 800x600

S->C: resize to 800x600, timestamp 12345

C->S: GL commands, timestamp 12345

C->S: more GL commands, timestamp 12345

C->S: more GL commands, timestamp 12345

C->S: OK, I'm done with timestamp 12345

S: render window decorations on C's backing buffer

S: push C's backing buffer from 12345 to the front

S->C: OK, it's now timestamp 12346

[fmoralesc]

Well, I agree I was an ass there. When I read your comment I thought you were pushing something in the lines of "well, the problem is simple if you treat it with [insert hight-concept mechanism here]", which always gets a bit on my nerves. You know, like none has thought of the problem domain before? Anyway, I think I might have been mistaken.

[colanderman]

No hard feelings; I tend to come on strong.

Also, I totally see how this is difficult (or maybe even impossible) in X, where the window manager is a separate client and there's no notion of inter-client synchronization. I haven't looked into Wayland enough to see what problems it presents (last time I checked it was in a state of major flux) but here's hoping its design entails a much simpler display model.