Comment by gf000
5 hours ago
> The overwhelming majority of applications belong in the same category as xeyes.
Well, I'm not sure you are using that many xmotif apps. Most of the GUI programs are gtk/qt (and let's be honest, electron) - and they are mostly bitmaps to X's eyes (pun not intended). They don't use draw commands with such a small granularity that network transparency would benefit.
And Arcan is so many things at once I'm not convinced it is a good alternative to Wayland. It has good ideas, but they sort of require the whole package. Meanwhile Wayland is just a minimal API over the Linux kernel API for managing display buffers, that can be extended with additional protocols.
>And Arcan is so many things at once I'm not convinced it is a good alternative to Wayland. It has good ideas, but they sort of require the whole package.
That's a bit of a double edged sword, it's the exact reason why I don't think Wayland is a good alternative to X. Wayland's minimalist attitude towards responsibility is good for one thing, and that's implementing new compositors from scratch. The bare bones compositor will be a long way from usable, but it will be technically complete. The question is, does it matter to me that there are 30 different compositors? Each in various states with their own eclectic featureset with no guarantees given, a la USB-C? Not really. In effect, it did present me with the conundrum of choosing between a half-baked compositor (dwl) or a desktop experience I have literally zero interest in (Gnome/KDE) which left me with a sour taste in the mouth.
Moving beyond that, a real problem with Wayland's architectural minimalism is that a display server does more than simply abstract a single API. It provides a lot of rather complex features, from accessibility to input handling. Not every compositor is capable of handling that kind of complexity, especially if it's to work well. What we will find going forward are two possible futures:
1. The future resembles the present status quo of fragmentation, made worse by time. The compositor archipelago is here to stay, and deploying software in this environment has become excessively annoying. There are 10 different competing libraries for any given category of basic infrastructure, and each library in each category has their own ideas and idiosyncrasies that have to be worked around. Most of them are buggy and incomplete.
2. Smaller compositors effectively die off, and we are left with a single monolithic compositor, back to the modular Window Manager/Desktop Environment. This monster implements the defacto Extended Wayland protocol, where all of the different parts don't quite match each other very well because that's the cost gained from not having vertical integration of complicated components. There's no cohesive rhyme or reason to the design of anything. Thus, the exercise in minimalism has wrought an uglier and more complicated beast than Xorg itself.
I think it's clear Wayland is going to continue to sweep the Linux desktop given its massive corporate backing. But I'm not really compelled to run a bare Wayland compositor under any circumstances, because my Arcan server already works perfectly fine as a Wayland compositor. It works as any number of Wayland compositors running whatever extensions they implement. In effect, no matter which future we end up in, by using Arcan I'm in a much better position than someone running a normal compositor. This fact alone makes me favor Arcan, even before we get into its unique merits.