LWN: Comments on "The Wayland Situation: Facts About X vs. Wayland (Phoronix)"

Works for me

paulj — Wed, 26 Jun 2013 06:21:27 +0000

Ditto.

Indeed, for me X works better than VNC, over a wifi connection that has OK latency but unreliable bandwidth. VNC is annoying to use over this, X is fine - you wouldn't know the window was a remote one using it!

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 25 Jun 2013 20:57:10 +0000

Aaah. Sorry, I completely misread it and had everything turned around. You're right, of course. (And that's a nifty feature: how long's it been there, I wonder...)

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Tue, 25 Jun 2013 20:52:01 +0000

Right, it'd be putting words in your mouth. Just like you saying I think text scrolling is as pointless as xeyes, is putting words in my mouth. EOT I hope.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 25 Jun 2013 20:38:21 +0000

I've used a couple, or seen others use them. Mostly big simulation apps with lots and lots of state. These days, one would probably write a server and a frontend with less state... but that happened not to be how those were written.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 25 Jun 2013 20:37:24 +0000

Good design, then. Unlike certain other programs I could mention (yes, mongodb, of *course* you should wake up a hundred times a second when idle because select() is too hard to use properly, sigh.)

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 25 Jun 2013 20:34:48 +0000

What? I never said any such thing. If you want to invent words and put them in my mouth, you are welcome to, but that still doesn't mean I actually said them.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 25 Jun 2013 20:32:47 +0000

Note that my example was from back when I was doing that sort of thing, which was 2010-era and older, and god only knows how old the RDP backend I was talking to was. They probably have improved in the meantime.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 25 Jun 2013 20:31:30 +0000

That's what happens when you implement impossible magic. People don't realise it could possibly work, so assume it doesn't. :)

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daglwn — Thu, 20 Jun 2013 18:28:46 +0000

My emacs 24 doesn't pop up frames when using TRAMP to invoke shells.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

raven667 — Thu, 20 Jun 2013 15:34:45 +0000

This is all worth highlighting and is a gem of this discussion.

> technology is wheeled out at the barely usable prototype stage, claimed to be the one true way with all that went before being completely broken and to be abandoned

Some of that is just a consequence of open development, by definition you have the software before it is in a "finished" state, and see the dark corners as it is being made. Often the software isn't really "done" or "ready" until after several iterations of release, sometimes spanning years, and everyone has to suffer through the iterations until it works the way it should. See GNOME 2 and GNOME 3, they really didn't get into their stride with feature completeness until after the .8 or .10 release.

> completely broken and to be abandoned... while the replacement doesn't actually DO everything that its predecessor did

Some times things are replaced without knowing why the old thing did the things it did, google "mjg59 lighdm" for a good example of that. Unfortunately there sometimes isn't enough manpower for parallel development and maintenance.

> Remote X display has never, ever been a "nightmare" for me, or for anyone I know - frankly I've always considered it verging on the miraculous! If X has problems, which I'm not disputing, they're at a level where _users_ don't see them

And that is due to the very hard work of people like Keith Packard and Daniel Stone and all the X.org team. They have busted their butts over the years trying to make X not suck so that you and I could be blissfully unaware of the contortions and brokenness behind the scenes. It's a testament to their success and hard work that people don't believe them when they say that X11 is fundamentally broken.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

ajmacleod — Thu, 20 Jun 2013 13:54:33 +0000

I didn't accuse anyone of "not trying"; I've said repeatedly that I hope Wayland does do everything X is actually used for today and does it better. I'm just a bit weary of a particular type of approach which seems quite common at the moment where a replacement for some piece of technology is wheeled out at the barely usable prototype stage, claimed to be the one true way with all that went before being completely broken and to be abandoned... while the replacement doesn't actually DO everything that its predecessor did.

Nobody has claimed that X is perfect, but I do claim that it does absolutely everything I want it to do, does it without any fuss and has done so for a decade and a half (over which time the things I've been actually using it for have changed quite a bit.) Remote X display has never, ever been a "nightmare" for me, or for anyone I know - frankly I've always considered it verging on the miraculous! If X has problems, which I'm not disputing, they're at a level where _users_ don't see them. I wish the Wayland developers all the best, and if they can make it all work as well and transparently as X currently does I'll be very happy!

Wayland explanations are STILL confusing and worrying users

Jonno — Wed, 19 Jun 2013 15:28:55 +0000

> Perhaps I'm reading too much into your choice of the word 'most', but does that mean you know of VNC servers which are better in this regard?
Most VNC servers make some use of the protocol feature (i.e. for window movement on the virtual desktop), but to my knowledge no X11-based VNC server is capable of automatically using the feature when scrolling part of a window.

That said, libVNCServer makes it available to custom applications, and x11vnc has an experimental command line option to try to detect scrolling in the frames it scrapes from the X11 server.

> And do they require a matching client?
The "CopyRect" image encoding is one of the five original encodings of the specification, but only the "Raw" image encoding (uncompressed bitmap data) is mandated, everything else is negotiated per session. So while most clients support it, some might not.

Wayland explanations are STILL confusing and worrying users

raven667 — Wed, 19 Jun 2013 14:47:15 +0000

I wonder how the OSX client/server rate, in my experience the pair is very good but VNC performance using another client is poor. It may be using these optimizations because performance is fairly good. Scrolling terminals isn't slow.

Wayland explanations are STILL confusing and worrying users

nye — Wed, 19 Jun 2013 13:56:41 +0000

>Actually, the VNC *protocol* are fully capable of efficient scrolling, but most VNC server *implementations* work by doing screen scraping and don't have any heuristics to detect scrolling, thus resulting in terrible performance.

Now *that* is interesting information. Perhaps I'm reading too much into your choice of the word 'most', but does that mean you know of VNC servers which are better in this regard? And do they require a matching client?

Wayland explanations are STILL confusing and worrying users

Jonno — Wed, 19 Jun 2013 13:05:58 +0000

> There is a middle ground in between basic line-drawing primitives and bitmap scraping. RDP does typically work at a higher level than screen scraping
> The point is that VNC is not comparable to this. It very clearly has to resend large amounts of data for even single line scrolls; it doesn't work in the same way at all.

Actually, the VNC *protocol* are fully capable of efficient scrolling, but most VNC server *implementations* work by doing screen scraping and don't have any heuristics to detect scrolling, thus resulting in terrible performance. The Wayland reference implementation is a lot better in this regard, and performs much better than what is typical for VNC, even though it uses the same design principles.

Also worth noting is that the Weston reference implementation actually uses the RDP protocol, which btw also uses the same design principles as VNC.

So if you compare a typical VNC implementation and the Windows RDP implementation, the Weston implementation should be closer to the later, as the server sits at the same position in the graphics stack *and* uses the same wire protocol (though implementation differences will of course make some impact).

Wayland explanations are STILL confusing and worrying users

nye — Wed, 19 Jun 2013 11:52:05 +0000

>The experience of RDP *is* vastly better than VNC, but that's not because it's using primitive drawing commands -- that strategy stopped working out very well years ago.

True, applications aren't requesting that the windowing system draw a line segment from A to B, and so on, but in practice X11 applications aren't doing that either.

There is a middle ground in between basic line-drawing primitives and bitmap scraping. RDP does typically work at a higher level than screen scraping, which is why nix's example of scrolling a text editor sounds highly suspect - I've just tried connecting to a Windows desktop (via whatever mechanism Remmina uses for rdp; I guess it wraps rdesktop), opening a large-ish file in gVim, and scrolling both casually line-by-line and madly by mashing page down. It's just as fast and responsive as running vim in an xterm over the same link[0] (ping ~45ms), and in fact seems to have slightly less input lag. Notably, looking at the data transferred, RDP is using a fair bit *less* bandwidth.

The point is that VNC is not comparable to this. It very clearly has to resend large amounts of data for even single line scrolls; it doesn't work in the same way at all.

With any luck Wayland's remoting will sit at a similar point to RDP on the primitives<->scraping spectrum, and everything will be rainbows and unicorns.

[0] That is, 'ssh -X <host> xterm vim'. For a more direct comparison, I did try 'ssh -X <host> gvim', but that really wasn't a fair comparison since X is useless over WAN links. The bandwidth requirement was higher still, and the input latency was so high that nobody could seriously consider it usable in any but the most desperate of circumstances. If you are only ever moving line-by-line, then it would win over VNC, but if you ever want to scroll a whole screenful it takes a similar amount of time.

Wayland explanations are STILL confusing and worrying users

mathstuf — Wed, 19 Jun 2013 07:14:58 +0000

> complete powerdown of the local machine without breaking the remote client

Other than Emacs, what apps actually support this? For me, every time the SSH connection does, anything that used X over it dies.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Wed, 19 Jun 2013 00:24:08 +0000

> One hopes that it actually pings clients which should have an event directed to them when such an event turns up, which would add no overhead and have the same effect of detecting slow or unresponsive clients reasonably rapidly.

It does.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Wed, 19 Jun 2013 00:23:12 +0000

I think anyone would be hard pressed to call it 'throw[ing] the baby out with the bathwater'.

In the 26 years since X11's initial release, it's gained multi-output support (four times), multi-input-device support (three times), hotplugging for both, a new keyboard model, a compositing model, a new rendering model, direct OpenGL rendering support (twice), indirect OpenGL rendering, accelerated indirect OpenGL rendering, a new font rendering model including anti-aliasing (three-ish times), autoconfiguration support, an acceleration architecture (four-ish times), Display PostScript support (later removed), a print server (also later removed, thank god), and has been ported to everything from mobile phones to renderfarms.

So how you can essentially accuse us of not having trying, and done this frivolously, is beyond me.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Wed, 19 Jun 2013 00:14:58 +0000

> Except if you want to scroll windows full of text, since that means repainting the entire screen rather than erasing a line from the top, shuffling the rest up, and painting a line at the bottom (or vice versa), since it has no understanding of the semantics of scrolling.

Again, this is completely false.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Wed, 19 Jun 2013 00:12:35 +0000

I suggested to do the rendering locally. Very few GL-using apps (apparently I need to disclaim this now, so let me be clear that I don't think they're marginal or pointless; just not the majority) these days have heavier geometry usage than textures. It comes down to Blender/Maya/etc, and CAD apps really. At that point, bandwidth-wise, you end up sending more over the wire than you would just sending the final post-composed image.

Then the more significant hurdle is updating GLX beyond GL 1.5, which as others have noted, is a hell of a lot of work.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

daniels — Wed, 19 Jun 2013 00:07:53 +0000

Yes, and I make a living out of vim. What you're saying, effectively, is that X11 is an abomination and inferior to SSH, just because vim under SSH is much more efficient than under an X11 terminal emulator.

Wayland explanations are STILL confusing and worrying users

foom — Tue, 18 Jun 2013 20:52:55 +0000

Well, those parts of RDP are nearly dead now. These days windows does basically all rendering on the host side (including 3D), and sends bitmaps to the client. Even fonts are rendered on the host side, ever since ClearType has been used.

The main exception to that strategy is that desktop compositing does get done on the client via dedicated compositor commands.

The experience of RDP *is* vastly better than VNC, but that's not because it's using primitive drawing commands -- that strategy stopped working out very well years ago.

Wayland explanations are STILL confusing and worrying users

daniels — Tue, 18 Jun 2013 13:57:15 +0000

I've said it over and over, repeatedly, time after time, in the comments here. But unfortunately people insist on ascribing motivations we don't have to us, while critiquing a design which doesn't exist. Gets pretty tiring after a while, to be told neither myself or Kristian care about text, which is why the hypothetical design the internet has collectively made up for Wayland remoting (despite pointers to the actual design) is deficient. Really, really tiring.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 18 Jun 2013 13:11:12 +0000

Kristian's rolling-hash implementation

Oof. So you're still analyzing bitmaps on every scroll to determine that they consist of scrolled text. Cache-ruinous compared to what is currently done. Oh well, at least you're not throwing them over the network.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 13:09:39 +0000

Yeah, that's what I expected. Of course prototypes turn into production implementations all the time :P

Wayland explanations are STILL confusing and worrying users

nix — Tue, 18 Jun 2013 13:09:02 +0000

Awesome!

Why on earth didn't you just say that to start with rather than going all sarcastic and contemptuous? All I said was that I couldn't think of an implementation -- that doesn't mean there isn't one.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

renox — Tue, 18 Jun 2013 12:43:53 +0000

Weston being a prototype, I'm not sure that they worried about power management in their implementation especially as the protocol is the same in both cases.

Wayland explanations are STILL confusing and worrying users

daniels — Tue, 18 Jun 2013 12:09:05 +0000

Kristian's rolling-hash implementation of remote Wayland is _specifically designed_ for scrolling content, be it text or images. Have a look at:
http://people.freedesktop.org/~krh/rolling-hash/f8.png
http://people.freedesktop.org/~krh/rolling-hash/f9.png
http://people.freedesktop.org/~krh/rolling-hash/f8-f9-deb...

And then come back and tell me again how no-one working on Wayland cares about terminals, text editors, or scrolling text.

How you've taken from me saying that the overwhelming majority of my day is spent in vim, irssi, Evolution, Evince, or Chrome scrolling huge chunks of text, that I don't care about scrolling text, is honestly beyond me. I'm not comparing scrolling text to xeyes either: I'm saying that even X11 isn't a very efficient protocol for scrolling text, because oddly enough the most efficient protocol for doing that is text-based. Like SSH. Just as X11 is the most efficient protocol for transferring X11 primitives.

I'm done arguing with people who have constructed strawman motives for a strawman protocol, neither of which bear any resemblance whatsoever to reality.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:56:03 +0000

as the compositor ping continuously the clients

Isn't that rather bad for power management?

One hopes that it actually pings clients which should have an event directed to them when such an event turns up, which would add no overhead and have the same effect of detecting slow or unresponsive clients reasonably rapidly.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 18 Jun 2013 11:51:28 +0000

Well, I just hope that text editors forever continue to target X rather than paying the least attention to Wayland, then, since you don't care about text editors.

FWIW, with the exception of a bunch of molecular modellers almost the only people I have ever seen use X in anger outside Linux distributors, and the only people I have ever seen use remote X routinely, have used it overwhelmingly for... text editing! In fact this is a majority use case for *all* computers used in work environments, if you include word processors and spreadsheets, which will suffer the same scroll-related problems as text editors if run remotely over anything implemented via bitmap-hurling.

But you don't care about that, as you have made abundantly clear by comparing those of us who care about remote use of such applications to people using xeyes. I'm not sure what use cases you *do* care about, if text editors and word processors and spreadsheets are all excluded. It doesn't leave much except for video playback and web browsers, and if you only want to use *those* you might as well use a tablet these days.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 18 Jun 2013 11:41:42 +0000

As you point out modern toolkits tend to just produce fully rendered pixmaps and not use the Core X11 drawing primitives, so applications are reduced to just shuffling pixmaps around, inefficiently, for remote use.

If they're using the render extension properly (which they are) this is an entirely inaccurate description of how textual regions are handled. The bitmaps comprising the textual glyphs are *not* sent repeatedly over and over again in repeated bitmap shuffles, but just once.

Wayland explanations are STILL confusing and worrying users

nix — Tue, 18 Jun 2013 11:39:28 +0000

Regarding, "Wayland should be BETTER than X at remoting" - that may be true. But I didn't see in this text, or any other text, a commitment to actually implement remoting in Wayland. I think most users want to hear a commitment that Wayland will support network transparency in some manner - and all we hear is a statement of possibility.

Yes. This.

(It's not as if my 'run on a remote box with graphical output in a local window, scroll text efficiently in that window, allow disconnection and complete powerdown of the local machine without breaking the remote client' use case should be incredibly difficult, either. Heck, X could do all of that in the mid-to-late 80s!)

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:35:03 +0000

I wish that was the case. But... if that was the case, Daniel Stone wouldn't be suggesting that people who use remote X are as important a use case as people who use xeyes.

With that sort of attitude, I have no confidence that remote Wayland will *ever* work properly. Clearly Daniel just has a laptop and does all his work on it, and damn everyone who uses more than one machine for anything.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:31:39 +0000

X11 isn't going anywhere --- but if people start targetting applications for Wayland, every single such application is an application I suddenly cannot run remotely with anything like the transport efficiency of X apps, particularly not if they are using anything for which GlyphSets would have been used (I don't really care if e.g. 3D games start to use Wayland).

And *that* is a loss of functionality. As mmarq has pointed out, Wayland only doesn't reduce functionality if *nobody targets it*. In which case its existence seems rather pointless.

xeyes

nix — Tue, 18 Jun 2013 11:27:31 +0000

He's being patronizing and sarcastic because, well, as far as I can tell because I've knocked a hole in his argument and he has no constructive response he can give. Depressing.

It would be less depressing if I hadn't been banging on about the same subject in virtually every thread on the same subject here for *years* and if Daniel hadn't been reading those threads as well and hadn't completely ignored them or been just as patronizing in all of those.

I wouldn't make such a fuss, except, again, this workflow is the one in which I have earned every penny I have ever earned. And it's not at all a subtle or unlikely workflow, unlike the workflow of someone making their living from a remote instance of fucking xeyes.

(Why yes, Daniel *has* just made me very angry. Way to piss off your users and make your users suspect that the Wayland developers don't give a damn about how their actual *users* are using the system they're trying to replace, mate. Well done. Very well done. Do I have any confidence in Wayland at all? Not after *that* comment.)

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:23:48 +0000

Oh, I'm sorry. Obviously the Emacs in which I have done *all my work for the last fifteen years* and have earned *every penny I have ever earned* is only as important as bloody xeyes.

Sheesh. Talk about patronizing.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:22:31 +0000

Actually there have been suggestions (from Daniel Stone, IIRC) that might well get indirect GL working much much better with much less development overhead (IIRC the idea was to fall back to bitmap shuffling in situations in which you currently abort: yes, it's slower than hardware acceleration but at least it *works* and lets you do simple 3D stuff remotely which is just using it to do analogues of 2D-analogous stuff).

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:20:01 +0000

The end result is pretty impressive.

Except if you want to scroll windows full of text, since that means repainting the entire screen rather than erasing a line from the top, shuffling the rest up, and painting a line at the bottom (or vice versa), since it has no understanding of the semantics of scrolling.

I am told that in order to get such a rare and obscure use case to work I have to wait for toolkit-level remoting, which is as far as I can tell a complete mirage: nobody is working on it, nobody is planning to work on it, if people do work on it their work for distinct toolkits will be totally uncoordinated (natch), why aren't you happy with VNC-style bitmap shuffling nobody needs to scroll windows full of text anyway.

The Wayland Situation: Facts About X vs. Wayland (Phoronix)

nix — Tue, 18 Jun 2013 11:15:20 +0000

I think of VNC as 'incredibly slow'. I have never used any implementation of VNC nor whatever the MS Terminal Server protocol is called that could scroll an Emacs window with text in it at more than about three to five screens per second, with a noticeable lag -- over a local network! And this applies to scrolls by one line as much as to scrolls by whole pages, which is a bit tough if you want to scroll up five or six lines, line-by-line, as I do quite often. Meanwhile under X, even remote X (over an Ethernet LAN) scrolling is latency-free: it scrolls so fast it is a blur.