Is forking xorg enough?

[darkhog]

I was thinking (yeah, I know...) and it came to me that forking xorg might not be enough to fix it and ensure dominance of X11 for decades to come. What we might need is for someone to write the entire new xserver from scratch that follows X11 protocol and its extensions, but avoid all the mistakes xorg made.

Unlike whale-land people, I don't think we need a brand new protocol or that the way desktop environments or window managers are written has to change. Instead, I believe X11 is great as a protocol, it's only that its implementation (xorg xserver) sucks donkey bollocks and therefore should be rewritten from scratch instead of trying to polish the xorg turd.

[reaperx7]

The problem with rewriting the whole of x-server extends out to many projects beyond X-server. Xcb, pixman, the protocols, etc. roughly 150+ packages comprises the x-server and userland.

[darkhog]

Oh, I never said it would be easy, in fact it would be extremely difficult. I've only said it might be what's needed.

[CustardEddy]

I agree, in fact, I've just created a feature request to create a plugin interface that enables direct integration of window/display managers into the X server itself.

A ground-up rewrite may be the ultimate destination, but we get there incrementally; the reality of 'big change' is that it's a lot of small changes, done all at the same time.

We need specific ideas and objectives. As far as I can see, the server falls into 5 broad elements:

1. event processing (including queueing)
2. hardware interfacing/abstraction
3. composition/rendering
4. local resource management (memory, files, network connections)
5. security

Obviously the big ones right now seem to be security and the composition/rendering parts, but it's likely that other optimisations could be achieved. For example, maybe event queues could be split per-screen (if they're not already), which may mean better performance on multi-core systems.

The only thing to do is to dive into the code and start pulling at the various threads and see what's what.

One thing I'm absolutely sure about: X11 should remain true its network-transparent roots, if for no other reason that the world is becoming MORE networked, not less. I mean, wouldn't it be awesome if Internet-based applications were able to be directly rendered onto your desktop using the X11 protocol, instead of being restricted to a Web browser?

[darkhog]

I mean sure, we can Ship of Theseus Xorg/Xlibre over time, but I don't know if it's the best way to get there. Xorg (and now Xlibre) has a lot of cruft and technical debt that's completely unnecessary for a proper implementation of X11 protocol, in fact, some stuff in Xorg/Xlibre goes against the spec.

And yes, network transparency is a must, though I am not sure if internet apps being able to render on your machine is best for the security.

[CustardEddy]

I mean sure, we can Ship of Theseus Xorg/Xlibre over time, but I don't know if it's the best way to get there.

Oh, I'm sure it is. Have you heard of Perl 6? Showing my age a little, but Perl 6 became a running-joke /cautionary tale as to what happens when you do a 'ground-up re-write'.

It's so bad, that it's not even called Perl 6 anymore!

And yes, network transparency is a must, though I am not sure if internet apps being able to render on your machine is best for the security.

That depends, entirely, on how it's implemented. It's only a semi-serious suggestion; it was more to make a point about the nature of protocol and that there are potential, untapped uses out there.

(On a technical note, Web sites generally work statelessly, because Internet connections were, historically, considered 'unreliable'. X11 is a stateful protocol, assuming a window will be destroyed once the client loses the connection to the server. But, HTTP 3 assumes reliable networks, so never say never! :) )

[darkhog]

Good points. I will have to think more about it.