Nvidia installer overwrites existing files

[brebs]

Hi, the nvidia installer has a bad reputation for overwriting existing files (e.g. /usr/include/GL/gl.h and /usr/lib/xorg/modules/extensions/libglx.so), which has the undesirable side-effects of stopping xorg from compiling, mucking up the distro's package manager by not informing it of changed files, and generally confusing newbies with FUD.

So, why the need to overwrite *any* files? Surely it is possible to agree a convention with xorg, e.g. use nvidia-gl.h rather than clobbering the existing gl.h ("--no-opengl-headers" is not the solution).

[Grogan]

Standard answer #23: Use correct packages for your distribution to install the Nvidia driver. (I don't mean to offend... that's just what these Fedorabuntu type pricks always say)

... and just how is nvidia-gl.h going to be found by build configuration systems that look for gl.h? Are they all going to have to be rewritten? This is obviously why they replace it. Personally I'm glad that Nvidia provides us with a complete interface to their OpenGL libraries.

Why is --no-opengl-headers not the solution, if you don't want the opengl headers to get installed?

What the heck should the xorg build be doing going for the openGL headers installed on your system during the build anyways? Do you not point the x11-server build to the location of the MESA source tree so it can compile the back end? It doesn't even install the headers at that stage. You (or at least, I) compile and install MESA stuff in a subsequent step (e.g. make linux-dri).

Oh wait, I see. You were compiling Xorg using Fedora's SRCRPMs. I read a little further in your fedora forum link and I see what it's including in the compile command. They should not be doing that.

Distributions like Fedora and Ubuntu have their own (silly) ways of doing things that deviate from traditional practices. So yes, installing and uninstalling "third party drivers" can cause problems. I would say that it's them who are causing the problems.

That's why I don't eat that dog food.

As for FUD, that just comes from certain free software zealots. Others just get on with their work and lives, and realize that if you want good support for your graphics card, it's necessary to install proprietary drivers. (Yes, it certainly would be "nice" if everyone opened their source and hardware specs. Don't get me wrong, I'd be very happy. It just isn't going to happen.)

Saying "use our packages to install third party drivers" isn't really FUD though.

[brebs]

Use correct packages for your distribution to install the Nvidia driver.

Well, the whole point of my post is that it would be better for Nvidia's installer to not overwrite the existing files. Then *all* distros could just use and recommend Nvidia's installer, and save their individual distros' repackaging effort, and save the users a ton of confusion.

Originally Posted by Grogan

Why is --no-opengl-headers not the solution, if you don't want the opengl headers to get installed?

Because they *should* be installed. Only, they should not *overwrite* the existing files. There's plenty of room in a Linux filesystem for new files.

[xorbe]

(er, nevermind.)

[Grogan]

You've got a good idea, mind you, but there are a few things that make it impractical for Nvidia to do.

1) If you want to link against the library, the headers must be found.
2) If existing binaries (and there are a lot of games that can't be recompiled and linked) are to work with the Nvidia driver, they must be able to load libGl.so.1 and friends that point to the Nvidia opengl libraries. When ldconfig is run, it must not move the symlinks back to the mesa libraries either. This is why they are yanked out of the way.
3) Distributors aren't exactly sticking with xorg defaults for the locations of files, or even doing everything consistently. (e.g. "/usr/lib/dri")

Who better to arrange the tomfoolery so that Xorg DRI and Nvidia OpenGL can coexist, and the package managers are aware of the changes, but the distributors themselves, then? The same goes with the includes, if they want to keep the option of compiling against mesa. (though to me the solution to that is to NOT replace the gl includes. Stuff compiled against Mesa would usually still work with Nvidia's OpenGL at runtime, at least in my experiences.)

That's all I'm trying to say here.

[brebs]

Originally Posted by Grogan

must be able to load libGl.so.1 and friends that point to the Nvidia opengl libraries.

Surely there's a better method which doesn't require to *overwrite* files. How about a config file, containing e.g. "opengl=mesa" or "opengl=nvidia" or "opengl=ati". Or maybe apps would need to be able to specify themselves which/whose opengl libs they wanted/preferred?

Who better to arrange the tomfoolery so that Xorg DRI and Nvidia OpenGL can coexist, and the package managers are aware of the changes, but the distributors themselves, then?

I suppose it requires co-operation from both sides. But this is a good point. Are the Linux zealots spreading FUD that Nvidia's installer eats babies and corrupts files, when they haven't provided any method for Nvidia's installer to work *without* overwriting files?

[AaronP]

Originally Posted by brebs

Surely there's a better method which doesn't require to *overwrite* files. How about a config file, containing e.g. "opengl=mesa" or "opengl=nvidia" or "opengl=ati". Or maybe apps would need to be able to specify themselves which/whose opengl libs they wanted/preferred?

At one point, I think there was a discussion about trying to get distributions to standarize on something akin to Gentoo's eselect to provide a standard way of choosing between OpenGL implementations, but it didn't get any traction and I can't seem to find the thread. Without a standard, distribution-agnostic way of installing multiple vendors' libGLs, nvidia-installer is doomed to failure unless it installs /usr/lib/libGL.so.1 itself, as required by the OpenGL Linux ABI.

Originally Posted by brebs

I suppose it requires co-operation from both sides. But this is a good point. Are the Linux zealots spreading FUD that Nvidia's installer eats babies and corrupts files, when they haven't provided any method for Nvidia's installer to work *without* overwriting files?

There's nothing preventing distribution packages of the NVIDIA driver from installing things in whichever way works best for that distro. See, for example, Gentoo's eselect. We just can't rely on anything like that in nvidia-installer because no two distros have the same mechanism.

Edit: I don't think zealotry has anything to do with it, and I wouldn't accuse anyone of spreading FUD here.

[brebs]

Looks like nothing ("In particular, Andy's response about why they're uninterested in a common libGL is basically The Last Word on the subject.") came of this thread. Presumably a death by committee.

I suppose, at least, Windows is in pretty much the same situation - nvidia & ati drivers will conflict there (with uninstall & reboot cycles needed), but it's much less user-visible, because the vast majority of users don't compile a single thing.