The way to Wayland: Preparing for life After X

One sure way to stir up Linux users and developers is to propose replacing a tried-and-true technology with an up-and-coming technology. Especially replacing something as crucial as X, which is what Mark Shuttleworth suggested might happen on Ubuntu, with Wayland taking its place. The response to Shuttleworth's post, along comments and questions on development mailing lists since, show that Wayland is not well-understood in the larger Linux community. Moving to Wayland isn't as far-fetched as one might initially think.

So what is it? Wayland is not, as it was initially reported, a "new X Server" for Linux. Wayland is the name for the protocol and the MIT-licensed implementation of a compositor that can run as a standalone display server, or under X as a client. Most importantly, Wayland (when running on its own) removes a few layers of complexity.

As explained in the Wayland architecture document, X runs on top of the Linux kernel and, in conjunction with a compositor, it is in effect adding an extra layer between the kernel, hardware, and compositor. With X, windows and contents are sent to a separate buffer and then "composited" together into the frame buffer by Compiz, Kwin, or another compositing window manager. With Wayland, all of this will happen in one display manager.

Wayland works directly on top of the kernel, and lets the clients handle rendering directly without the intermediate layer. Wayland uses direct rendering through OpenGL, or OpenGL ES. X imposes additional layers, and that causes a performance hit. As Shuttleworth wrote when tapping Wayland as the future for Ubuntu and Unity, "we don't believe X is setup to deliver the user experience we want, with super-smooth graphics and effects."

The initial reactions and discussion about Wayland had more than a tinge of concern. Much of which based on fairly breezy reports about Wayland as a replacement for X for Ubuntu. As Andrew Haley points out on the Fedora devel list, there's no immediate cause for alarm:

Yes, the asylum remains in competent hands. Wayland is not a new idea. Wayland started as a "secret" project by Kristian Høgsberg in 2008. Høgsberg's creation was outed by Phoronix, caused a brief wave of excitement in Linux circles, and then went back to largely being ignored by most of the Linux world.

But X folks have been thinking about Wayland, at least occasionally, for some time. Last year at the Linux Foundation Collaboration Summit, Keith Packard talked about turning "the graphics stack upside down" by moving device configuration out of X and into the kernel, which would pave the way for other systems like Wayland. Packard also hinted that a post-X era may be in the offing while at this year's Linux Plumbers Conference, and mentioned Wayland as a possible replacement — with X running as a client.

Why not simply extend X, yet again? It's been extended to add all sorts of features never envisioned when it was first developed. Wayland is an option "of pushing X out of the hotpath between clients and the hardware and making it a compatibility option," as described in the FAQ. X running as a client is a particularly important feature. As Adam Jackson points out on the Fedora devel mailing list, X applications only need be ported to behave as native Wayland clients. Otherwise they can run within Wayland within a nested X server "and you wouldn't ever know the difference." Note that Wayland can also run as an X client, which allows for development and testing during the transition.

It may be beneficial to look at Wayland as an opportunity rather than a potential problem. For example, while many games now run well on X, it is not particularly friendly for fullscreen 3D games. Høgsberg indicates that thought has already gone into the specific problems of fullscreen games and how to address problems like modesetting and handling the pointer.

Wayland is also poised to support GPU hotswapping, something that X does not currently support. As more hardware ships with more than one GPU, which is intended to help with power savings, users will want Linux to support switching between the GPUs.

But we're not there yet. The big problem, of course, is that Wayland is not ready for prime time — or even early morning between infomercials. Wayland may see an influx of interest thanks to the attention it's getting, but there's a long way between vision and reality at the moment.

As Packard mentioned during his LPC talk, input is another problem for an alternative display system. Key mapping, accessibility work (using the keyboard for mouse movement, for instance) and handling more complex input devices like touchpads, all need to be addressed.

Aside from a general lack of readiness for Wayland itself, Wayland also lacks drivers. Nvidia has already explicitly said it has no interest in Wayland, though Nouveau may be able to take up the slack there. Wayland can use the open source KMS drivers for ATI, Intel, and Nvidia — but what about the new crop of video hardware coming with ARM-based devices? Here we have a new set of video hardware without open source drivers or existing efforts to create them.

There's also the question of who's going to do the work to ready Wayland. The work on Wayland up until now — and in the foreseeable future — has been on Red Hat and Intel's payroll. Høgsberg was a Red Hat employee when he first started Wayland, and is now working for Intel. Canonical doesn't have any resources currently assigned to work on Wayland. Canonical's Ted Gould has set up an import of Wayland's git tree into Launchpad to make it easier to build packages. But Gould says he's unaware of anyone directly working on Wayland who is on Canonical's payroll:

Users who are itching to get hands on bleeding-edge Wayland builds can look at the compile instructions, or add the "xorg crack pushers" PPAs for Ubuntu to install Wayland on Maverick (10.10) or Natty (11.04). Breakage is quite likely. Developers interested in pitching in are welcome to do so. Wayland is part of freedesktop.org, and the git repository is open.

But it will be some time before anyone needs to make the switch. With the renewed attention caused by Shuttleworth's post, Høgsberg has started working more actively on Wayland again. But while Høgsberg isn't quite going it solo, there's not a lot of commits from other developers yet. Shuttleworth indicated that it would be a year before Ubuntu could seriously consider switching to Wayland. Fedora will probably package Wayland for F15, but when it will be default is up in the air. Jackson says the "cabal" of Fedora graphics folks "don't even have a complete list of transition criteria yet, let alone a timeframe for switching the default." It would seem that replacing X has momentum, but we're a long way away from making a switch.