That's interesting, but I really would rather write something stripped down with X as the base, though. This might be a good intermediary step, though.
It just feels like the wrong approach to me. Ripping the buffer management/KMS/DRI part out of a Wayland compositor as a starting point is reasonable, but if you pull XWayland along you pull most of Xorg along, with most of the maintenance / legacy hassle that brings.
An X server selectively deprecating older functionality can be lean. There are lots of the protocol you can just ignore, and what's left is fairly easy to implement. Time consuming, but easy.
> An X server selectively deprecating older functionality can be lean.
Is that actually true?
The big problem is that Xlib uses a lot of the dusty corners and the people using X11 apps expect those dusty corners to work. And Xlib is very, very, very badly architected in many places and you still have to do something sane or the apps won't work.
If you get Xcb right, you at least pick up Qt and that gets you probably 80+% of the modern apps.
It really depends on your personal needs, so your mileage may wary, but having run xtruss on most of the X11 programs I use: It's true for me at least.
The key thing is that most modern clients (and Xlib vs. Xcb doesn't matter much in this respect) use exceedingly few features of the protocol.
E.g. you'll find very few modern clients that actually uses the X protocol to draw lines with any complex options. You'll also find very few attempts at using exotic visuals, because most X servers won't support them anyway.
To take a more extreme example: You could probably get away with just ripping out most or all server-side font-rendering support. Most modern X apps use Xrender, and render the glyphs client side and creat a glyph set via Xrender or just render the whole buffer client side.
Removing those will break apps, including some that some people care about (like xterm), but API surface is similar enough that it'd be reatively easy to fix the few apps people care about, or just providing an optional shim for Xlib as a temporary measure.
For own my own use, if I were to write an X server to meet my needs, I'd just not support it - I've checked with xtruss and none of the apps I care about rely on server-side fonts at all, and none of them use any drawing primitives but the most basic line drawing and rectangles either, and all of them are capable of handling an X server that refuses anything but truecolor visuals.
I could rip out large parts of the drawing code from Xorg and all the apps I run would still work just fine...
For other people, of course, this would be a different matter, and that's fine. But I think moving a bunch of that code to an Xlib-compatible library to handle client side as a means to carry over legacy apps would be an option as a means to let people evolve leaner X servers and move shift the hassle of managing legacy code to those who care about the legacy apps.
I mean unless you're going to commit to maintaining xlibre or something, wayback seems like the future for x-based desktops.
> [wayback] is intended to eventually replace the classic X.Org server, thus reducing maintenance burden of X11 applications, but a lot of work needs to be done first.
