XFree86 dust-up questions X11 model (Register)
Key Linux kernel developer Alan Cox agreed that the project needed a wake-up call, but didn't think a splinter project by Packard could cause too much harm: "X has to evolve, X has to do cool stuff, X has to let people break stuff, X has to delegate trust to driver maintainers far more," he wrote. "To me it doesn't matter if Keith and friends spin off an "Xperimental" or XFree itself changes, but that change is vital to the future of X11.""
Posted Mar 22, 2003 0:03 UTC (Sat)
by rknop (guest, #66)
[Link] (6 responses)
Now, I'm sure there are reasons to be cautious. But the Open Source community cannot afford too much of this for two reasons. First, if the X releases do not support even one generation old hardware, Linux gets beat up for not being able to run with those cards. Second, when you have a manufacturer submitting open source drivers for their products, we need to do everything we can to encourage and support that. Not applying the patches is a slap in the face. The last thing we want to do is slap ATI in the face for giving us open source drivers! We want to encourage and support them, as we'd much rather have that than the binary-only closed-source drivers that NVidia provides. (And, regardless of the mess that X is in, everybody with a Linux system, please buy ATI rather than Nvidia, precisely for this reason.0 -Rob
Posted Mar 22, 2003 4:50 UTC (Sat)
by Peter (guest, #1127)
[Link]
You're thinking of Mike Harris of Red Hat. There are two sides to that one, it seems. Mike claims that ATI's patches take months to get in - and then the next release doesn't come out for a few months more - for a grand total of lots of latency. Others don't think ATI's code quality is high enough, and apparently Mike doesn't do much code review on his own, so what he forwards on to X still needs to be cleaned up. Allegedly, the latency of availability runs more or less in inverse proportion to the quality of the submission. Far be it from me to take sides, I'm just repeatin' what I hear....
Posted Mar 22, 2003 13:34 UTC (Sat)
by davej (subscriber, #354)
[Link] (4 responses)
Posted Mar 22, 2003 13:54 UTC (Sat)
by rknop (guest, #66)
[Link] (3 responses)
You may not be aware of it, but the newer radeons (those with R300 core) also require part binary modules if you want accelerated 3D. Damn! I wasn't aware of that, and that is extremely sad if true. Is there any in the latest generation of cards that lets one do 3-D in an open source fashion? Or are we moving to a world where video cards are going to require binary only drivers for 3-d performance? If so, then I think it's time to buy up the back stock of Radeon 7500's and Matrox G400's, and all the other cards which do have open source supported 3D. Performance may not be as good as on newer cards, but at least one can get some 3-d acceleration with free software. If there is a vendor out there who understands that supporting free software means putting out programming information for their cards-- binary drivers are fine, and that's support, but true support of Linux systems requires them allowing people to write free drivers for their cards-- then we as the Linux community need to support and encourage them as much as possible. I had thought that vendor, today, was ATI, but if it's not, then who is? And if it's nobody... well, life sucks if you ever try to be some sort of purist. -Rob
Posted Mar 22, 2003 15:01 UTC (Sat)
by mattdm (subscriber, #18)
[Link] (2 responses)
Posted Mar 22, 2003 20:47 UTC (Sat)
by Peter (guest, #1127)
[Link] (1 responses)
That's the R200 chip, right? Anyone know if that driver also supports the FireGL 8700/8800?
Posted Mar 23, 2003 2:33 UTC (Sun)
by mattdm (subscriber, #18)
[Link]
Posted Mar 23, 2003 18:30 UTC (Sun)
by HuskyDog (guest, #10294)
[Link] (8 responses)
We now seem to be being told that because Joe Sixpack doesn't want client-server X, folks like us have to do without it in the future. Sure, someone could dream up some sort of clumsy screen-scraper compatibility layer. But I've played with things like VNC (and the improved versions like TightVNC) and although very clever and often useful they are never going to be as good as a proper client-server graphics system like X. I'm not saying that the increased popularity is a bad thing, but pointing out that it can have costs for Linux's more traditional users. I just hope that some sort of compromise can be achieved. For example, perhaps 'client-server' or 'direct render' could be a compile time option for future versions of XFree86.
Posted Mar 23, 2003 22:11 UTC (Sun)
by vivek (guest, #6962)
[Link] (7 responses)
It is perhaps unsurprising that someone working mostly in the windows If XFree86 _does_ follow that path, I have no doubt that the project
Posted Mar 24, 2003 1:24 UTC (Mon)
by link (guest, #916)
[Link] (2 responses)
In other words, the core design of X is actually best suited for a _minority_ of the users instead of a majority, and that should probably be taken into account when considering the future of X and XFree86. 1. "Traditional X", where the essence is the remote display of the GUI for an application program running on a remote computer. 2. "VNC - Remote Console", where what is actually showing on the physical console is displayed in a remote window. 3. "Terminal Server", where you provide a remote, _graphical_, login capability with detatch/reattach capability (which is subtly different from XDM+XDMCP). Of these I suspect #3 and #2 combined make up a significant majority of the user needs. And consequently I would speculate that the main point made was that perhaps it would be better to do the core design of a future XFree86 around those rather then #1 as is currently the case.
Posted Mar 24, 2003 2:13 UTC (Mon)
by Peter (guest, #1127)
[Link] (1 responses)
You make some good points, but in any case, that's now how I read dwexelblat's statement. It sounded to me like he was saying "nobody cares about client/server GUI" - implying that basically everybody is happy with a fat client nowadays. It didn't sound to me like he was thinking Terminal Server or VNC here. Regardless, a 'screen'-like detach / reattach layer would be quite useful if optimised properly. Actually that wouldn't be so different from session management - but more pervasive and transparent. Or, a hacked version of Xnest, if you don't want the apps to have to care. (:
Posted Mar 24, 2003 6:23 UTC (Mon)
by jamesh (guest, #1159)
[Link]
Sun also has something for their Sun Rays, although they use a virtual frame buffer mode similar to the VNC X server. And of course, you have VNC itself as an option ...
Posted Mar 24, 2003 22:35 UTC (Mon)
by Baylink (guest, #755)
[Link] (3 responses)
> "X is obsolescent," he wrote in a mailing list posting.
> "I've been working in the Windows world for years now, and client-server
> display systems are utterly irrelevant to the majority of real-world
> computer users. X needs to be replaced by a direct-rendered model, on which
> backwards-compatible X server can be reasonably trivially implemented".
and what that tells *me*, is that it is time for dwex to get off the xFree board, and leave the seat to someone who *does* want to do what X does, rather than doing something different.
If this is what dwex thinks these days, my vote's with Packard.
But who knows; maybe it's just me.
Posted Mar 25, 2003 9:23 UTC (Tue)
by amikins (guest, #451)
[Link] (2 responses)
Posted Mar 26, 2003 3:38 UTC (Wed)
by Baylink (guest, #755)
[Link] (1 responses)
Let's look at this, you're right, from a technical standpoint. The goals are: apps must still write to the X11 API -- having a separate one encourages the writing of apps which would *not* be location transparent; this is A Bad Thing. Secondly, any app should be able to connect to which ever server it finds the location of in it's environment or arguments -- see above. But, y'know? That's what we have *now*. It seems to *me* that the problem is what everyone *says* it is: poor server implementation. I *believe* the arguments that local connections over Unix-domain sockets are not a performance hit. And, if they're not, then what else is there to fix? The server on the other end of the socket, of course. This is beginning to sound like mountains and molehills to me... except, perhaps, for the pissing-off-vendors aspect of things...
Posted Mar 26, 2003 9:05 UTC (Wed)
by amikins (guest, #451)
[Link]
Of course, if it's worked for decades, it's obviously not a BAD way of doing something, either.. => My thing is that the 'masses', so to speak, have little use for the particular type of 'server/client' split that the current X11 model provides.. To me, it's more useful to have a 'detachable desktop' along the lines of the VNC/Windows Terminal Server architecture. I see the uses of X11 network operation, but it doesn't suit the particular needs I'm working with. As for performance.. I really have trouble believing that copying data -- as you must do for sockets, even within a single host -- is somehow magically just as fast as using a direct-write model. But, as you say, maybe that's just me. =>
I forget where I saw it, but one developer was commenting that it takes forever for patches to get into the X tree. He was referring to open source patches for Radeon cards sent in by ATI, which were festering and which probably will not come out in the next release of XFree86.What really convinced me there was a problem...
What really convinced me there was a problem...
I forget where I saw it, but one developer was commenting that it takes forever for patches to get into the X tree. He was referring to open source patches for Radeon cards sent in by ATI, which were festering and which probably will not come out in the next release of XFree86.
You may not be aware of it, but the newer radeons (those with R300 core) also require part binary modules if you want accelerated 3D.
What really convinced me there was a problem...
What really convinced me there was a problem...
You can go a generation newer than that. The Radeon 8500 has open source 3d support, thanks to funding from The Weather Channel.
What really convinced me there was a problem...
What really convinced me there was a problem...
The Radeon 8500 has open source 3d support, thanks to funding from The Weather Channel.
Oh, now we're going in circles. :) See Mike Harris' comments about how he has a patch to do just this which he's been too discouraged to send upstream.
What really convinced me there was a problem...
Here we see one of the main problems with the increased popularity of Linux. At work we use remote X all the time. I have a dozen colleagues who need to access graphical scientific applications on a large Unix server from their desktops and other systems with diskless clients running LTSP.XFree86 dust-up questions X11 model (Register)
That comment was made by dwexelblat, who stated, in his full posting,XFree86 dust-up questions X11 model (Register)
(iirc) that he hadn't worked on Xfree recently, and was mostly working
in the Windows world (I paraphrase, I haven't got the article to hand:
corrections welcome).
world has become infected by the "every user sits at the physical
console" pov: I have no doubt people who don't have true network
transparency don't miss it - but I very much doubt that the XFree86
project will follow the path he suggests, especially as he isn't, by
his own admission, an active member any more.
will fork about .003 milliseconds later, and that every single major
distribution will follow the branch that maintains the (or at least a)
c/s model.
I think perhaps the background for dwexelblat's comments was more the positive experience of using the Windows Terminal Server and Remote Desktop Client than any VNC-like scheme. A very large portion of the remote display needs are better served by a Terminal Server style of remote display then the X client/server design, and another portion would be better helped with the "remote console" functionality of VNC on Windows/Mac.XFree86 dust-up questions X11 model (Register)
For reference, I see three distinct modes of operation for a remote display function:
Regardless of whether or not you agree with that asessment, I think it is a reasonable position to take and not merely the result of "working on Windows too long". :-)
XFree86 dust-up questions X11 model (Register)
I think perhaps the background for dwexelblat's comments was more the positive experience of using the Windows Terminal Server and Remote Desktop Client than any VNC-like scheme.
SCO had/has such a thing in their XVision server. Essentially you had a headless X11 proxy server, and you could connect and disconnect it from the real X server (kind of like Xnest, but not dependent on the X server it is hosted on).XFree86 dust-up questions X11 model (Register)
And indeed, dwex says this:
XFree86 dust-up questions X11 model (Register)
To be honest, I'm not sure what the problem is. If backwards compatibility can be maintained, then the functionality you want will remain available. Why then does the implementation matter so much?XFree86 dust-up questions X11 model (Register)
The objective is getting the most reliable and most efficient system for the majority of users, and while I enjoy working with an application from multiple locales, and have used X functionality, on my primary machine a direct rendered interface makes more sense.
As long as backwards compatibility is available -- even through an abstraction layer -- then it shouldn't impact anything for traditional X client/server usage.
Hmmm.XFree86 dust-up questions X11 model (Register)
The view I am starting to end up with is that if something has been in place for this long, and people are starting to get concerned about performance, maybe it's time to try to look at everything from an alternate perspective. Just because something has worked for decades doesn't mean it's the best way to do something.XFree86 dust-up questions X11 model (Register)