linux, the very weak system for gaming

[jrch2k8]

Wrong. If I'm making some program that uses the GPU, guess what? I need to interact with the GPU drivers. Now, someone goes and changes the Kernel API, the drivers need a new update, and boom, my program stops working, or I have a massive performance regression, because some feature I was using got broken somewhere between the Kernel and driver.

When you change the Kernel API, you necessitate a driver redesign. When you necessitate a driver redesign, you really tick off the people who interact with said driver. Additions, fine. But you should almost NEVER remove functionallity.



Windows is partly POSIX compliant.

POSIX is fundamentally broken for one simple reason: the pthread_create() method is fundamentally flawed because it allows no way to create a thread in a suspended state. (And no, manually suspending a pthread after creation is an example of a horrid, wasteful, programming practice that belongs in the dark ages).

I had to support POSIX once, and my hatred of it exceeds even my hatred for the Ada language (which I can assure you, I hate with a passion).

1.) if you are making a program that need to interact with a GPU driver you are doing it wrong [i think you try to say gpu blobs??? cuz this is insane]

2.) linux rarely if ever broke ABI in the same major release and revisions [3.2 3.2.1 3.2.3] and since some time ago LTS releases are ABI stable very long time

3.) if you mean you have a 3rd party BLOB driver like FGLRX or Nvidia's and you want to run it outside the supported kernel versions for that driver is your fault not linux but if you use a linux supported[in tree] driver this rarely even happens and this happens in almost any OS including windows [vista driver in windows 7 == missing features - crashes - pain] you just think windows kernel is ABI stable cuz it has longer release cycles and you think since windows vista and 7 are different OSes is 100% rational to download new drivers[but no NT kernel 6.0/6.1]. So in resume linux major release cycle is 6-10 weeks windows is 2-5 years that is all, so a 3rd party driver need to stick with an LTS kernel release or put resources in support every major version each release cycle[if linux releases a major every 2 years you won't even notice this <-- this is what LTS is for].

3.a) we can discuss that maybe the linux release cycle should be extended or maybe provide an isolation layer between drivers and kernel API or another solution but a 3rd party is not responsability of linux[as was not microsofts the vista nvidia drivers mess] so the risk of breakage is always bigger than an linux natives in-tree drivers

3.b) internal kernel api changes breaking userspace is a rare scenario too and if it does normally in the changelog they put a big fat warning that an userspace software requires an upgrade to X.x.x version [kernel ppl are not crazy ....]

4.) about POSIX please post some code example and some logic behind cuz well i really can't see it useselfuness and looking in msdn that feature is deprecated for windows too[at least under suspend]

5.) about ada can't comment since its outside my area of expertise

[jrch2k8]

Wrong. You are also affected if you want up to date drivers, even if they're open source. Let's say you have some new wifi adapter that's supported by linux 3.5 or later, but your distribution only has 3.0. If you had a stable driver interface you could just get the driver binary and install it. But instead you'll either have to figure out how to compile a new kernel, or you could pull the driver sources from a git tree and pray that they'll compile against your current kernel, or you could wait a few months for your distribution to have another release which will hopefully include the new kernel. Neither of these solutions are any good compared to just running an installer and having the hardware working right away as you would do on Windows or OSX. Especially for the average user that probably knows nothing about APIs and kernels and just want their newly bought hardware to work.

i agree here but this don't mean ABI breakeage or messy developers as i explained in previous post linux release cycle is just faster but i think this problem is generated cuz the drivers and the internal Kernel API are released as a unique piece of software so i think could be interesting to strip drivers from kernel tree [need deeper discussion tho] so they can be separated entities hence solving this problem.

another interesting idea could be to expose subsystems layers [bluetooth, wifi, usb, sata,etc] APIs isolated of the internal API[so you can adapt the inner layer to new internal kernel API but keeping the outside layer stays unchanged to the drivers]

[Thaodan]

So the standardized way of installing software is to just unpack it in your home dir? Right. But lets say you want to install the game so that all users can access it. How would you do that? Lets say you're an average use with average computer skills. First you should be somewhat familiar with the FHS to know where you should unpack it so that it's accessible to everyone. This isn't easy to figure out just by looking at the names of the folders (/dev, /mnt, /var etc) so you'll probably have to google it. The next problem is to actually move it in place since you don't have write access to most folders. So you'll have to resort to google again to learn how to use sudo to gain temporary root access to the file system. After that the game might be accessible to everyone but you have to open it from the file browser, it wont show up in the Unity dash or whatever you are using (even if you only unpacked it in your home dir). So what do you do? You google it again to find out that you have to create a .desktop file and put it in either /usr/share/applications/. Great! Now the game is installed!

The standart way installing software is the package manager, even installing software via an installer the installer must handle operatingsystem directory's too in every operating system.
For example:

Windows	OS X	Linux
idk	/dev	/dev
not used	/Volumes	/mnt (manuly mounted) /media (automatic)
many, idk	/etc, /Library and /Users/$USER/Library	/etc and /home/$USER/.config
C:/Users/Pupblic/Startmenu	not used the .app itself is link	/usr/share/applications

You see that many things are common, so you see that what you said is not fully true, this are only a few examples if you google more you find more.

Com spare this to what you would do in windows (run a installer and click next a few times) or OSX (just drag the app bundle to the applications dir). App installations on linux aren't an issue you say?

The way of Windows with the installer is bad see the issues with them, the bundle system is an other concept, the linux concept is the package manager.

[ChrisXY]

Wrong. You are also affected if you want up to date drivers, even if they're open source. Let's say you have some new wifi adapter that's supported by linux 3.5 or later, but your distribution only has 3.0.

Are there any big distributions that won't have a repository/PPA/??? with bleeding edge software?

If you had a stable driver interface you could just get the driver binary and install it.

Yes. Do you really want to split that little developer time there is so they keep legacy abstraction layers up to date instead of actually progressing the ecosystem?

But instead you'll either have to figure out how to compile a new kernel, or you could pull the driver sources from a git tree and pray that they'll compile against your current kernel, or you could wait a few months for your distribution to have another release which will hopefully include the new kernel. Neither of these solutions are any good compared to

Why would you not just install a precompiled kernel? Again, are there big distributions where you can't get an updated kernel easily?

On archlinux you put an unofficial repository in /etc/pacman.conf

[miffe]
Server = http://arch.miffe.org/$arch/

pacman -Syu && pacman -S linux-mainline
Put it in your bootloader. Congratulations, you now run 3.6-rc3.

just running an installer and having the hardware working right away as you would do on Windows or OSX. Especially for the average user that probably knows nothing about APIs and kernels and just want their newly bought hardware to work.

That's a nice idea and all but just today I have seen an up to date windows 7 prof where somebody tried to install some dongle and while it installed the driver directly from windows update it got a bluescreen and never booted again without bluescreening or freezing at the login screen. And what about all that hardware that doesn't work with windows vista/7 anymore and everybody is just okay with that? If you use windows 7 you can throw away your HP Scanjet 3300C even though it works perfectly fine (today it still works with sane). Even Microsoft, the world market leader of legacy, just abandons their old compatibility layers for hardware.

I don't think you have a very convincing argument there.

[nej_simon]

The standart way installing software is the package manager, even installing software via an installer the installer must handle operatingsystem directory's too in every operating system.
For example:

Windows	OS X	Linux
idk	/dev	/dev
not used	/Volumes	/mnt (manuly mounted) /media (automatic)
many, idk	/etc, /Library and /Users/$USER/Library	/etc and /home/$USER/.config
C:/Users/Pupblic/Startmenu	not used the .app itself is link	/usr/share/applications

You see that many things are common, so you see that what you said is not fully true, this are only a few examples if you google more you find more.

The way of Windows with the installer is bad see the issues with them, the bundle system is an other concept, the linux concept is the package manager.

My point is that it's easy to create an installable package for OSX or windows while it's more difficult to do the same on Linux due to lack of standardization. If you download firefox on windows you get an installer, if you download it on OSX you get an app bundle but on Linux you get a tarball that you have to "install" yourself. The same is true for a lot of other project, most will just provide a source tarball. My guess is that they simply don't think it's worth the effort to maintain a range of packages for Linux distros. I'm not saying that Linux distros should standardize installers or app bundles but this is a problem that needs to be solved in some way.

[D0pamine]

I don't see the point of this thread, windows and osx are not even in competition with GNU and linux

'games don't work because of X/Y/Z' funny that other programs both free and non-free seem to keep on working....

[russofris]

The issue that is being described is accurate.

It takes too long for fixes and features to make their way into distributions in a consumer friendly manner.

That said, the root cause is not an unstable API/ABI. Worse, when people say this, I always begin to doubt that they know what an API/ABI is. It's really not even a problem. People see a fix/feature hit the mesa dev mailing list, and for some reason decide that the fix/feature should be available in the current 12.04 release of Ubuntu immediately. The cause of this type of thinking comes as a carryover from the old MS ecosystem when all releases were fairly secret until the release announcement.

The other thing about this is that the people complaining 'are not wrong'. A legitimate fix in Mesa should be available in Ubuntu-current "tomorrow", it is just that the Linux/gnu ecosystem (or any other ecosystem for that matter)has not yet figured out how to accomplish this gracefully. Even security fixes take 3-14 days and require near super-human efforts on the part of the distribution maintainers.

If one of you readers is a process guy or PM, can you take a look at the path a patch takes to go from mesa-dev to Ubuntu/Fedora-current? While I do not think that a 24-48h release cycle is feasible, I'm curious as to how we would improve the current 3-6 month cycle without interfering with the current progressive development methods employed today..

F

[Hamish Wilson]

My point is that it's easy to create an installable package for OSX or windows while it's more difficult to do the same on Linux due to lack of standardization. If you download firefox on windows you get an installer, if you download it on OSX you get an app bundle but on Linux you get a tarball that you have to "install" yourself. The same is true for a lot of other project, most will just provide a source tarball. My guess is that they simply don't think it's worth the effort to maintain a range of packages for Linux distros. I'm not saying that Linux distros should standardize installers or app bundles but this is a problem that needs to be solved in some way.

They could just use a MojoSetup installer, or one from Bitrock, or Nixinstaller, or a multitude of other installers available for Linux.

Personally, I would rather not have proprietary applications handled through my package manager (and do not ATM). I do not even really want my games managed through my package manager either, which is why I have OpenArena and other free titles handled by Desura instead. I prefer to keep my package manager strictly handling free system applications, drivers, and codecs. But maybe that is just me?

Regardless, such solutions are available. If you have actually gamed on Linux (which I thought this thread was supposed to be about?) you would have already known this.

[nej_simon]

Are there any bigdistributions that won't have a repository/PPA/??? with bleeding edgesoftware?

Perhaps in some cases, but should a user really have to resort to repo with bleeding edge software just to get a working driver? Wouldn't it be better to support third party drivers so that the hardware manufacturer can bundle a driver with the hardware?

Yes. Do youreally want to split that little developer time there is so they keeplegacy abstraction layers up to date instead of actually progressingthe ecosystem?

It would significantly enhance the usability of Linux so it would be worth it. At least if Linux is ever going to be viable alternative to Windows and OSX for consumer PCs.

Why would you notjust install a precompiled kernel? Again, are there big distributionswhere you can't get an updated kernel easily?

Well, you can if there any available for your distribution. For ubuntu 12.04 for example the are backported unsupported kernels from 12.10 but they are a lot less tested than the default kernel and might have regressions.

On archlinux youput an unofficial repository in /etc/pacman.conf

[miffe]
Server =http://arch.miffe.org/$arch/

pacman -Syu && pacman -Slinux-mainline
Put it in your bootloader.Congratulations, you now run 3.6-rc3.

That's great if you run arch, but arch is an advanced distribution and probably not something a novice user would run nor something that would come preinstalled on consumer hardware. Soulutions like this might work for hackers and enthusiasts but again, if Linux is ever going to be viable alternative to Windows and OSX for consumer PCs you can't expect users to play around with different kernels and repos to find a working driver.

That's a niceidea and all but just today I have seen an up to date windows 7 profwhere somebody tried to install some dongle and while it installedthe driver directly from windows update it got a bluescreen and neverbooted again without bluescreening or freezing at the login screen.And what about all that hardware that doesn't work with windowsvista/7 anymore and everybody is just okay with that? If you usewindows 7 you can throw away your HP Scanjet 3300C even though itworks perfectly fine (today it still works with sane). EvenMicrosoft, the world market leader of legacy, just abandons their oldcompatibility layers for hardware.

That's besides the point. Just because Windows 7 has a stable driver API doesn't mean that every driver is perfectly stable. Or that drivers will always be available for older hardware. However when you buy new hardware and install the bundled driver on Windows it just works in most cases. You don't have to resort to backported kernels from pre-release distributions, or repos with bleeding edge software etc.

EDIT:
This is getting really off topic. Perhaps it's time to just end this discussion about stable API now.

[D0pamine]

really am i missing something? most gnu/linux distributions are perfectly viable as gaming platforms and what does this thread have to do with fglrx/catalyst?

does every game work out of the box on windows xp/vista/7/8 or is there a certain amount of fiddling to play the latest AAA+++ title?

and why are people comparing windows with gnu/linux anyway? the point of a gnu/linux system is freedom to do whatever YOU want to do with it, the point of a windows or osx system is to have a company wipe your anus - they are not the same thing and not in competition and why a gpu binary blob is such an affront to those of us who understand this

A fresh install of ubuntu will run ut2k4 if set up correctly - true it has to be set up correctly .... but what do you want? you have to set up any OS correctly or have somebody wipe your anus

now stfu or i'll kick you up and down the server of your choice on an EEEPC

HOLY SHIT!!!
WICKED SICK!!!