273
KDE's Nate Graham On X11 Being A Bad Platform & The Wayland Future
(www.phoronix.com)
From Wikipedia, the free encyclopedia
Linux is a family of open source Unix-like operating systems based on the Linux kernel, an operating system kernel first released on September 17, 1991 by Linus Torvalds. Linux is typically packaged in a Linux distribution (or distro for short).
Distributions include the Linux kernel and supporting system software and libraries, many of which are provided by the GNU Project. Many Linux distributions use the word "Linux" in their name, but the Free Software Foundation uses the name GNU/Linux to emphasize the importance of GNU software, causing some controversy.
Community icon by Alpár-Etele Méder, licensed under CC BY 3.0
Over on Nate's other blog entry he indicates this:
And I think this is something people tend to forget. X11 as a protocol is complex and writing an implementation of it is difficult to say the least. Because of this, we've all kind of relied on Xorg's implementation of it and things like KDE and GNOME piggyback on top of that. However, nothing (outside of the pure complexity) prevented KWin (just as an example) implementing it's own X server. KWin having it's own X server would give it specific things that would better handle the things KWin specifically needed.
Good parallel is how crazy insane the HTML5 spec has become and how now pretty much only Google can write a browser for that spec (with thankfully Firefox also keeping up) and everyone is just cloning that browser and putting their specific spin to it. But if a deep enough core change happens, that's likely to find its way into all of the spins. And that was some of the issue with X. Good example here, because of the specific way X works an "OK" button (as an example) is actually implemented by your toolkit as a child window. Menus those are windows too. In fact pretty much no toolkit uses primitives anymore. It's all windows with lots and lots of text attributes. And your toolkit Qt, Gtk, WINGs, EFL, etc handle all those attributes so that events like "clicking a mouse button" work like had you clicked a button and not a window that's drawn to look like a button.
That's all because these toolkits want to do things that X won't explicitly allow them to do. Now the various DEs can just write an X server that has their concept of what a button should do, how it should look, etc... And that would work except that, say you fire up GIMP that uses Gtk and Gtk has it's idea of how that widget should look and work and boom things break with the KDE X server. That's because of the way X11 is defined. There's this middle man that always sits there dictating how things work. Clients draw to you, not to the screen in X. And that's fundamentally how X and Wayland are different.
I think people think of Wayland in the same way of X11. That there's this Xorg that exists and we'll all be using it and configuring it. And that's not wholly true. In X we have the X server and in that department we had Xorg/XFree86 (and some other minor bit players). The analog for that in Wayland (roughly, because Wayland ≠ X) is the Compositor. Of which we have Mutter, Clayland, KWin, Weston, Enlightenment, and so on. Which that's more than just one that we're used to. That's because the Wayland protocol is simple enough for these multiple implementations.
The skinny is that a Compositor needs to at the very least provide these:
And that's about it. The specifics of how to interface with hardware and what not is mostly left to the kernel. In fact, pretty much compositors are just doing everything in EGL, that is KWin's wl_buffer (just random example here) is a
eglCreatePbufferSurface
with other stuff specific to what KWin needs and that's it. I would assume Mutter is pretty much the same case here. This gets a ton of the formality stuff that X11 required out of the way and allows Compositors more direct access to the underlying hardware. Which was pretty much the case for all of the Window Managers since 2010ish. All of them basically Window Manage in OpenGL because OpenGL allowed them to skip a lot of X, but of course there is GLX (that one bit where X and OpenGL cross) but that's so much better than dealing with Xlib and everything it requires that would routinely require "creative" workarounds.This is what's great about Wayland, it allows KWin to focus on what KWin needs, mutter to focus on what mutter needs, but provides enough generic interface that Qt applications will show up on mutter just fine. Wayland goes out of its way to get out of the way. BUT that means things we've enjoyed previously aren't there, like clipboards, screen recording, etc. Because X dictated those things and for Wayland, that's outside of scope.
That's my problem with this. It tries to be a desktop display server protocol without unifying all desktop requirements. Sure, X11 is old and have unnecessary things that aren't relevant anymore, however, as someone who builds their own DE, (e.g.: tiling window managers) I see it as the end of this masterrace. Unless everybody moves to wlroots. Flameshot, for example, is already dealing with this, having at least 5 implementations only for linux, and only wlroots and x11 are standards.
Also, imo, having windows in windows is useful when you want to use your favourite terminal in your favourite IDE. But as you said DEs can implement it simply. Let's say wlroots will implement this but others can decide otherwise. And for those the app won't run.
Another example, that affects my app personally, is the ability to query which monitor is the pointer at. Wayland doesn't care having these so I doesn't care supporting wayland. And I"m being sad about this because X is slowly fading away so new apps will not run on my desktop.
Moreover with X11 I could write my own hotkey daemon in my lanuage of choice, now I would have to fork the compositor.
Do I see it wrong?
The wayland way to do that is to have the application be a compositor, they made sure that nesting introduces only minimal overhead. And that ties in with the base protocol being so simple: If you only need to deal with talking to the compositor you're running on, and to the client that you want to embed, a wayland compositor is indeed very small and lean. Much of the codebase in the big compositors deals with kms, multiple monitor support, complex windowing logic that you don't need, etc.
Oh and just for the record that doesn't mean that you can't undock the terminal: Just ask the compositor you're running on for a second window and compose it there. You can in principle even reparent (client disconnecting from one compositor and connecting to the other) but I think that's only standardised for the crash case there's no standard protocol to ask a client to connect to another compositor. Just need to standardise the negotiation protocol, not the mechanism.
So there's a Wayland hope for systemd-afflicted boxes and their cults.
You anti-systemd folks are so insufferable.