New Linux Development Model
Hi list.
I tought long about writting this mail. I'm a linux use since many years. At the moment i'll getting more then frustrated about the actual develoment model of the kernel. In the latest releases things where broken from release to release.
For example take the ipw2200 driver.
From 2.6.12 -> 2.6.13 the header file ieee80211.h was incompatible with driver. Also transfer speed decreased dramaticaly.
From 2.6.13 -> 2.6.14 you included the ipw2200 driver. But in an too old version without WPA support. The external driver on ipw2200.sourceforge.net seems not to work with 2.6.14.
I had also several problems with some other not in kernel drivers.
I realy liked it to have the latest state of the art kernel, but at the moment i'm forced to use 2.6.12 ( ipw2200 -> WPA ).
I can't understand it why you have to break compatibility from kernel release to kernel release. Don't you think that this makes 3'rd party driver developers frustrated? It can't be an option for 3'rd party developers and users to check if external drivers still works with new kenrel releases.
From my point of view the actual linux kernel is far away from a stable development process.
cu
ED.
hostmaster <[email protected]> :
[...]
Quoted 2 lines
I can't understand it why you have to break compatibility from kernel release to kernel release.
Please read the archives of [email protected]. ipw2200 popped up again yesterday.
--
Ueimor
Hi, huh...
... don't you have a real name? Posting as "hostmaster" may impress teenagers, but over there we tend to prefer people with real names.
Quoted 5 lines
I tought long about writting this mail. I'm a linux use since many years. At the moment i'll getting more then frustrated about the actual develoment model of the kernel. In the latest releases things where broken from release to release.
For example take the ipw2200 driver.
Quoted 8 lines
From 2.6.12 -> 2.6.13 the header file
ieee80211.hwas incompatible with
driver.
Also transfer speed decreased dramaticaly.
From 2.6.13 -> 2.6.14 you included the ipw2200 driver. But in an too
old version without WPA support.
(...)
I realy liked it to have the latest state of the art kernel, but at the moment i'm forced to use 2.6.12 ( ipw2200 -> WPA ).
Your problems seem to be very specific to wireless networking, right? Blaming the whole development model because one area seems to have problems is a bit awkward, don't you think? Report to the persons responsible for that area and/or specific driver, tell them about the problem, they'll surely listen to you and improve the process if possible.
Also, the point you mention for the 2.6.13 -> 2.6.14 transition is irrelevant with regards to the development model. With a different development model, the driver wouldn't have been added at all. This wouldn't have made any difference for you as far as I can see.
Quoted 4 lines
The external driver on ipw2200.sourceforge.net seems not to work with 2.6.14.
(...)
I had also several problems with some other not in kernel drivers.
Third party drivers don't work, and you complain to us. What's the point?
Quoted 2 lines
I can't understand it why you have to break compatibility from kernel release to kernel release.
You should read this document:
http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/stable_api_nonsense.txt
Quoted 2 lines
Don't you think that this makes 3'rd party driver developers frustrated?
We don't care. It's their choice, not ours.
Quoted 2 lines
It can't be an option for 3'rd party developers and users to check if external drivers still works with new kenrel releases.
It is. If they are not happy with that, they simply get their code integrated into the main Linux tree, and get to work with us rather than apart on their own. It's easier for us, it's easier for them, and it's easier for the users. Everyone benefits.
Quoted 2 lines
From my point of view the actual linux kernel is far away from a stable development process.
Yet you don't propose anything to improve it?
Signature · 1 line
Jean Delvare
Hi.
Sorry for not posting my Name.
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header files, change structures, interfaces and so on. Which makes the kernel releases incompatible.
There are several reasons why modules are not in the mainline kernel and will never get there. So saying, bring modules to the kernel is wrong.
The right way would be to take care of defined interfaces, header files, and so on. Otherwise you could only say the kernel 2.6.14 is only compatible to 2.6.14.X and you there is no stable 2.6 mainline kernel.
I think it's also no task for the user, to search the net why external driver xyz not works with a new kernel ( because of incompatibilties ). Basicly in new kernel there could be a chance for the user a driver works better, because a bug was fixed in the kernel.
Hopefully this time it's more clear why i blame the development process and i'm a so frustrated linux user.
cu
ED.
Quoted 83 lines · Jean Delvare
Hi, huh...
... don't you have a real name? Posting as "hostmaster" may impress teenagers, but over there we tend to prefer people with real names.
Quoted 12 lines
I tought long about writting this mail. I'm a linux use since many years.
At the moment i'll getting more then frustrated about the actual develoment
model of the kernel. In the latest releases things where broken from release
to release.
For example take the ipw2200 driver.
From 2.6.12 -> 2.6.13 the header fileieee80211.hwas incompatible with driver.
Also transfer speed decreased dramaticaly.
From 2.6.13 -> 2.6.14 you included the ipw2200 driver. But in an too old version without WPA support. (...)
I realy liked it to have the latest state of the art kernel, but at the moment i'm forced to use 2.6.12 ( ipw2200 -> WPA ).Your problems seem to be very specific to wireless networking, right? Blaming the whole development model because one area seems to have problems is a bit awkward, don't you think? Report to the persons responsible for that area and/or specific driver, tell them about the problem, they'll surely listen to you and improve the process if possible.
Also, the point you mention for the 2.6.13 -> 2.6.14 transition is irrelevant with regards to the development model. With a different development model, the driver wouldn't have been added at all. This wouldn't have made any difference for you as far as I can see.
Quoted 6 lines
The external driver on ipw2200.sourceforge.net seems not to work with 2.6.14.
(...)
I had also several problems with some other not in kernel drivers.Third party drivers don't work, and you complain to us. What's the point?
Quoted 3 lines
I can't understand it why you have to break compatibility from kernel release to kernel release.
You should read this document:
http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/stable_api_nonsense.txtQuoted 4 lines
Don't you think that this makes 3'rd party driver developers frustrated?
We don't care. It's their choice, not ours.
Quoted 3 lines
It can't be an option for 3'rd party developers and users to check if external drivers still works with new kenrel releases.
It is. If they are not happy with that, they simply get their code integrated into the main Linux tree, and get to work with us rather than apart on their own. It's easier for us, it's easier for them, and it's easier for the users. Everyone benefits.
Quoted 3 lines
From my point of view the actual linux kernel is far away from a stable development process.
Yet you don't propose anything to improve it?
Quoted 1 line · Edgar Hucek
Hi.
Hi Edgar,
Quoted 10 lines
Sorry for not posting my Name.
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header
files, change structures, interfaces and so on. Which makes the kernel releases
incompatible.
you've already been given a pointer to the Documentation/stable_api_nonsense.txt document in the kernel sources that explains these issues.
Quoted 3 lines
There are several reasons why modules are not in the mainline kernel and will never
get there. So saying, bring modules to the kernel is wrong.
It's not wrong.
Every vendor of any kind of software will tell you A is supported and B is not supported.
It's a consensus among the Linux kernel developers that the Linux kernel does not support a stable API for external modules.
You don't have to like this decision, but stable_api_nonsense.txt explains it.
If you dislike this decision, there are several other operating systems whose vendors do offer a stable API for external modules.
Quoted 5 lines
The right way would be to take care of defined interfaces, header files, and so on.
Otherwise you could only say the kernel 2.6.14 is only compatible to 2.6.14.X and
you there is no stable 2.6 mainline kernel.
The 2.6 is a stable kernel series in the sense that it doesn't crash very often.
2.6.14 is not API-compatible with 2.6.15.
But this has always been this way, the new development model only brings more API changes than the previous development models.
Quoted 8 lines
I think it's also no task for the user, to search the net why external driver xyz not
works with a new kernel ( because of incompatibilties ). Basicly in new kernel there
could be a chance for the user a driver works better, because a bug was fixed in the
kernel.
...
I do agree, this is not a task for the user.
It's a task for the vendors of the external modules.
Quoted 3 lines
cu
ED.
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 7 lines · Edgar Hucek
Hi.
Sorry for not posting my Name.
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
Read the document linked to by Jean. It explains why a volatile API is ultimately beneficial to the Linux development process.
It's frustrating for vendors and maintainers to keep their code up to date with the latest kernel, but if it's done incrementally after Linus releases (or during the very lengthy RC process), it's usually very easy.
Also, you're gelling together many projects which have completely different reasons for breaking across kernel revisions.
ipw2200 "works" in 2.6.14, it's just the maintainer has opted to use a "stable version" which lacks experimental features while ieee80211 gets up to scratch.
OTOH, unionfs is poorly maintained and is a particularly sore example of how the linux development model is detrimental to external module development.
There are people constantly working to make binary modules like VMWare, NVIDIA and ATI work under the newest kernels. Just google it. Ultimately we can't allow ourselves to be held down by slow-moving binary vendors. It's too inefficient (not to mention of questionable ethics).
Signature · 6 lines
Cheers, Alistair. 'No sense being pessimistic, it probably wouldn't work anyway.' Third year Computer Science undergraduate. 1F2 55 South Clerk Street, Edinburgh, UK.
Quoted 22 lines · Edgar Hucek
Hi.
Sorry for not posting my Name.
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header
files, change structures, interfaces and so on. Which makes the kernel releases
incompatible.
There are several reasons why modules are not in the mainline kernel and will never
get there. So saying, bring modules to the kernel is wrong. The right way would be to take care of defined interfaces, header files, and so on.
Otherwise you could only say the kernel 2.6.14 is only compatible to 2.6.14.X and
you there is no stable 2.6 mainline kernel.
I think it's also no task for the user, to search the net why external driver xyz not
Quoted 9 lines
works with a new kernel ( because of incompatibilties ). Basicly in new kernel there
could be a chance for the user a driver works better, because a bug was fixed in the
kernel.
Hopefully this time it's more clear why i blame the development process and i'm a
so frustrated linux user.
There is a very simple solution to your problem. Use the kernels provided by your distribution and not the kernel.org kernels.
--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html
Am Samstag, 5. November 2005 15:34 schrieb Jesper Juhl:
Quoted 2 lines
There is a very simple solution to your problem. Use the kernels provided by your distribution and not the kernel.org kernels.
If enough people do so, testing will suffer.
Regards
Oliver
Quoted 6 lines · Oliver Neukum
Am Samstag, 5. November 2005 15:34 schrieb Jesper Juhl:
Quoted 2 lines
There is a very simple solution to your problem. Use the kernels provided by your distribution and not the kernel.org kernels.
If enough people do so, testing will suffer.
But people using Linux from user-point of view or for running applications/services and not for system development then the kernel with the distributions are the best choice for them because they are optimized by-default with the other applications/services provided by the distro !
--
Fawad Lateef
Quoted 5 lines · Oliver Neukum
Am Samstag, 5. November 2005 15:34 schrieb Jesper Juhl:
Quoted 2 lines
There is a very simple solution to your problem. Use the kernels provided by your distribution and not the kernel.org kernels.
If enough people do so, testing will suffer.
For a _user_, it is the best choice to use the kernel provided by the distribution.
There _are_ incompatible changes between 2.6 kernels, and if you e.g. try to run kernel 2.6.14 on a Debian stable that ships only 2.6.8 you can always run into problems here or there.
And it seems we already have many big guinea pigs running the development branch of Fedora.
Quoted 2 lines
Regards
Oliver
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 2 lines · Alistair John Strachan
ipw2200 "works" in 2.6.14, it's just the maintainer has opted to use a "stable version" which lacks experimental features while ieee80211 gets up to scratch.
I haven't been able to get it working. Nor does the external one seem to work with 2.6.14 -- still stuck on 2.6.13 here, which impacts my ability to help with testing of dev kernels.
-ml
Quoted 8 lines · Mark Lord
Quoted 3 lines · Alistair John Strachan
ipw2200 "works" in 2.6.14, it's just the maintainer has opted to use a "stable version" which lacks experimental features while ieee80211 gets up to scratch.
I haven't been able to get it working. Nor does the external one seem to work with 2.6.14 -- still stuck on 2.6.13 here, which impacts my ability to help with testing of dev kernels.
Well, the in-kernel 1.0.0 works fine here on 2.6.14 (no WPA of course). The only caveat is that you need to downgrade your firmware if you've been using 1.0.7 (or 1.0.8) prior, which I had and was initially confusing.
Otherwise, I suggest you file a bug report.
Signature · 6 lines
Cheers, Alistair. 'No sense being pessimistic, it probably wouldn't work anyway.' Third year Computer Science undergraduate. 1F2 55 South Clerk Street, Edinburgh, UK.
Quoted 4 lines · Edgar Hucek
Hi.
Sorry for not posting my Name.
Quoted 8 lines
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header
files, change structures, interfaces and so on. Which makes the kernel releases
incompatible.
I will ask you just one question: as a user, why did you want to upgrade your kernel?
On a server you want stability. So you don't upgrade. On a desktop, there are probably a bunch of out of kernel modules that will need upgrading with each new kernel modules. Just on the laptop I am using right now, I will have to upgrade the vmware bridge, nvidia driver, madwifi wireless driver, etc. And that's normal. The new development model didn't change that.
I avoid touching my kernel on boxes I do real work with. I do build a new kernel for test purposes and to give feedback if there's an issue. But most of the time I skip 2-3 versions before finding a very compelling reason to upgrade. And I stick with my distribution kernel as much as I can.
As for kernel/drivers developers, it's another story.
If it ain't broken, don't fix it.
Jerome
Quoted 18 lines · jerome lacoste
Quoted 4 lines · Edgar Hucek
Hi.
Sorry for not posting my Name.
Quoted 9 lines
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header
files, change structures, interfaces and so on. Which makes the kernel releases
incompatible.
Quoted 4 lines
I will ask you just one question: as a user, why did you want to upgrade your kernel?
Depends on the user and what he wants to do. There are several reasons why a user wanna upgrade to new kernel. Maybe new supported hardware and so on. It's frustrating for the user, have on the one side the new hardware supported but on the other side, mybe broken support for the existing hardware.
Quoted 1 line
On a server you want stability. So you don't upgrade.
Sure, but what about securrity updates. When a new kernel release comes out the updates are stopped for older releases. And why should dirstribution makers always backport new security fixes ?
Quoted 7 lines
On a desktop, there are probably a bunch of out of kernel modules that will need upgrading with each new kernel modules. Just on the laptop I am using right now, I will have to upgrade the vmware bridge, nvidia driver, madwifi wireless driver, etc. And that's normal. The new development model didn't change that.
From my point of view, it makes a difference if i have to recompile a module or realy upgrade it.
Quoted 8 lines
I avoid touching my kernel on boxes I do real work with. I do build a new kernel for test purposes and to give feedback if there's an issue. But most of the time I skip 2-3 versions before finding a very compelling reason to upgrade. And I stick with my distribution kernel as much as I can.
So you wanna say a new "stable" kernel isn't a realy a stable one and i can't relay that it behaves like the older one ? If it's so, then something is completely wrong in kernel development.
Quoted 8 lines
As for kernel/drivers developers, it's another story.
If it ain't broken, don't fix it.
Jerome
cu
ED.
Quoted 8 lines · Edgar Hucek
jerome lacoste wrote:
...Quoted 1 line
On a server you want stability. So you don't upgrade.
Sure, but what about securrity updates. When a new kernel release comes out the updates are stopped for older releases. And why should dirstribution makers always backport new security fixes ?
...
Because it's their job.
As long as a vendor supports a release of his distribution, the absolute minimum he has to provide are regular security fixes for all the software the release offers - including the kernels shipped with the release.
E.g. the latest security update RedHat offered for the kernel 2.4.9 (sic) used in their version 2.1 server products was two months ago. And they will provide security updates until 2009.
Quoted 3 lines
cu
ED.
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedOn Sun, 2005-11-06 at 12:55 +0100, Edgar Hucek wrote:
[...]
Quoted 3 lines
Sure, but what about securrity updates. When a new kernel release comes out the updates are stopped for older releases. And why should dirstribution makers always backport new security fixes ?
Because they have no other choice.
Or are you here to do this for them?
Bernd
Signature · 3 lines
Firmix Software GmbH http://www.firmix.at/
mobil: +43 664 4416156 fax: +43 1 7890849-55
Embedded Linux Development and ServicesQuoted 1 line · Edgar Hucek
jerome lacoste wrote:
[...]
Quoted 9 lines
Quoted 4 lines
I will ask you just one question: as a user, why did you want to upgrade your kernel?
Depends on the user and what he wants to do. There are several reasons why a user wanna upgrade to new kernel. Maybe new supported hardware and so on. It's frustrating for the user, have on the one side the new hardware supported but on the other side, mybe broken support for the existing hardware.
New kernel feature and new supported hardware would be the only reason for me to upgrade. Personally that doesn't come that often. My hardware configurations don't change that much. I make sure it's well supported, not just recently. When one buys a non supported hardware, one should know the path chosen won't be the easiest.
Quoted 4 lines
Quoted 2 lines
On a server you want stability. So you don't upgrade.
Sure, but what about securrity updates. When a new kernel release comes out the updates are stopped for older releases.
For vanilla kernels, yes.
Quoted 1 line
And why should dirstribution makers always backport new security fixes ?
Because they want to ensure maximum stability. That's what users are (sometimes) paying for.
And second 90% of the security issues will not affect the majority of the home users (because they are restricted to a particular area of the kernel not affecting the user, or because they already require access on the machine to be exploitable). You will have much more risks using a box with an unpatched php or apache than with an unpached kernel, or without a proper firewall configuration.
Quoted 9 lines
Quoted 7 lines
On a desktop, there are probably a bunch of out of kernel modules that will need upgrading with each new kernel modules. Just on the laptop I am using right now, I will have to upgrade the vmware bridge, nvidia driver, madwifi wireless driver, etc. And that's normal. The new development model didn't change that.
From my point of view, it makes a difference if i have to recompile
a module or realy upgrade it.
That only happens for out ot tree modules, which shouldn't be really out of tree in the first place. That's the issue. If they are out of tree, it's for a reason. Either they cannot be in tree, or they are not stable enough.
There you see the issue.
Quoted 4 lines
[...]
cuED.
Jerome
Quoted 2 lines · hostmaster
It can't be an option for 3'rd party developers and users to check if external drivers still works with new kenrel releases.
Who should check? It can't be the kernel developers since they don't know who the 3'rd party developers are.
- Jim
Signature · 2 lines
[email protected] SDF Public Access UNIX System - http://sdf.lonestar.org
Quoted 50 lines · --- jerome lacoste
Quoted 5 lines · Edgar Hucek
Hi.
Sorry for not posting my Name.
Maybe you don't understand what i wanted to say or
it's my bad english.
Quoted 1 line
The ipw2200 driver was only an example. I had also
problems with, vmware,
Quoted 2 lines
unionfs...
What i mean ist, that kernel developers makeincompatible changes to the
Quoted 2 lines
header
files, change structures, interfaces and so on.Which makes the kernel
Quoted 2 lines
releases
incompatible.I will ask you just one question: as a user, why did you want to
upgrade your kernel?On a server you want stability. So you don't upgrade. On a desktop,
there are probably a bunch of out of kernel modules that will need
upgrading with each new kernel modules. Just on the laptop I am using
right now, I will have to upgrade the vmware bridge, nvidia driver,
madwifi wireless driver, etc. And that's normal. The new development
model didn't change that.I avoid touching my kernel on boxes I do real work with. I do build a
new kernel for test purposes and to give feedback if there's an issue.
But most of the time I skip 2-3 versions before finding a very
compelling reason to upgrade. And I stick with my distribution kernel
as much as I can.As for kernel/drivers developers, it's another
story.If it ain't broken, don't fix it.
Quoted 9 lines
Jerome
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected] More majordomo info at
http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Another method, the one I use, is to keep a small drive with a minimum system on it to test/update the kernel. After updating the kernel and any modules required on the test driver, its then a simple matter to move them, if there are no problems, to my working system.
Jack
John Carlson
<[email protected]>
__________________________________
Yahoo! FareChase: Search multiple travel sites in one click.
http://farechase.yahoo.com
Quoted 5 lines · --- Jim Nance
Quoted 1 line · hostmaster
It can't be an option for 3'rd party developers
and users to
Quoted 21 lines
Quoted 1 line
check if external drivers still works with new
kenrel releases.
Who should check? It can't be the kernel developers since they
don't know who the 3'rd party developers are.- Jim
Signature · 10 lines
[email protected] SDF Public Access UNIX System - http://sdf.lonestar.org - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Nor do they always have the required hardware to test the fixes.
jack
John Carlson
<[email protected]>
__________________________________
Yahoo! FareChase: Search multiple travel sites in one click.
http://farechase.yahoo.com
Hi Edgar,
Quoted 7 lines
It's frustrating for the user, have on the one side the
new hardware supported but on the other side, mybe broken support for the existing hardware.
(...)
So you wanna say a new "stable" kernel isn't a realy a stable one and i can't relay that it behaves like the older one ? If it's so, then something is completely wrong in kernel development.
"stable" is a relative notion. It's a tradeoff between improvements and the lack of regression. What you are asking for is absolute stability. This can only be achieved by the absolute lack of change. As soon as any change is made to the kernel, there is a risk of regression. Blame it on the complexity and variety of computer hardware.
It is our collective belief, as the kernel developers, that the current development model represents the best tradeoff we can achieve. We know, because we are the ones who suffer the most when the development model is not efficient.
Having a separate development branch is not a solution. It was discussed on this list and various other places several times already. Read the archives. The bottom line is that separate trees increase the workload on developers and divide the testing efforts, so we're losing on all fronts.
Let's imagine we have a separate development branch for a minute. This means we wouldn't be adding new stuff to the stable branch, right? Else it wouldn't be stable. So, you would complain that the stable branch doesn't have the latset ipw2200 driver. We would direct you to the development branch, which has it. You would complain that some drivers are broken for you in this branch, which is expected to happen in a development branch. See? Separate branches did not even solve your problem.
You blame your current problems on the development model, yet you failed to demonstrate how exactly this development model was responsible for them, as you failed to propose a different model which would solve them.
Signature · 1 line
Jean Delvare
On Sat, 2005-11-05 at 14:19 +0100, Edgar Hucek wrote:
[...]
Quoted 5 lines
Maybe you don't understand what i wanted to say or it's my bad english. The ipw2200 driver was only an example. I had also problems with, vmware, unionfs...
What i mean ist, that kernel developers make incompatible changes to the header
ACK. The reason is the improve the kernel and drivers.
Quoted 2 lines
files, change structures, interfaces and so on. Which makes the kernel releases incompatible.
ACK. It needs (a lot of) work to keep backwards compatibility and who wants can do it (probably).
At what to do at the point where you actually have to break it?
Quoted 3 lines
There are several reasons why modules are not in the mainline kernel and will never
get there. So saying, bring modules to the kernel is wrong.
The last conclusion doesn't hold. Especially not in this general form. You have to list the "various reaons" and then we can discuss each of them.
Quoted 2 lines
The right way would be to take care of defined interfaces, header files, and so on.
The only defined interface of the kernel can be found in POSIX, SUSv3 and similar documents.
Quoted 3 lines
Otherwise you could only say the kernel 2.6.14 is only compatible to 2.6.14.X and
you there is no stable 2.6 mainline kernel.
Compatible in what way? Source? Object? User-space binary interface?
Quoted 1 line
I think it's also no task for the user, to search the net why external
Then don't do it.
Quoted 3 lines
driver xyz not
works with a new kernel ( because of incompatibilties ). Basicly in new kernel there
Who is interested in the new driver/kernel/..?
The user. So guess who's job is to do it (or find someone to do it - paid or unpaid).
Quoted 4 lines
could be a chance for the user a driver works better, because a bug was fixed in the kernel.
Hopefully this time it's more clear why i blame the development process and i'm a so frustrated linux user.
Who do you mean with "user"?
A typical "user" just installs $DISTRIBUTION (be it a free or commercial one) and that's it. *If* the bug is severe enough (and the relevant maintainers does it) there will be a new rpm/deb/... with the newest kernel release or a backport.
If not, you can do it on your own anyways.
But then you are half a programmer and more like a sys-admin and no longer a "user". Voila.
Bernd
Signature · 3 lines
Firmix Software GmbH http://www.firmix.at/
mobil: +43 664 4416156 fax: +43 1 7890849-55
Embedded Linux Development and ServicesQuoted 13 lines · jerome lacoste
Quoted 1 line · Edgar Hucek
jerome lacoste wrote:
[...]
Quoted 9 lines
Quoted 4 lines
I will ask you just one question: as a user, why did you want to upgrade your kernel?
Depends on the user and what he wants to do. There are several reasons why a user wanna upgrade to new kernel. Maybe new supported hardware and so on. It's frustrating for the user, have on the one side the new hardware supported but on the other side, mybe broken support for the existing hardware.
Quoted 6 lines
New kernel feature and new supported hardware would be the only reason for me to upgrade. Personally that doesn't come that often. My hardware configurations don't change that much. I make sure it's well supported, not just recently. When one buys a non supported hardware, one should know the path chosen won't be the easiest.
There are other reasons for using a new kernel. One of them is interactivity. In the days of 2.4 one could achieve decent interactivity for the desktop using preempt and low latency patches. For 2.6 interactivity was a real issue (possibly because of the new development model).
Quoted 5 lines
Quoted 1 line
And why should dirstribution makers always backport new security fixes ?
Because they want to ensure maximum stability. That's what users are (sometimes) paying for.
Maximum stability of what ? If the distribution kernels are based on vanila kernel (i.e. are based on unstable kernel) how stable will they be ?
On lkml someone said that "stable means it won't crash very often".
This sounds like Windows(TM)
Quoted 7 lines
And second 90% of the security issues will not affect the majority of the home users (because they are restricted to a particular area of the kernel not affecting the user, or because they already require access on the machine to be exploitable). You will have much more risks using a box with an unpatched php or apache than with an unpached kernel, or without a proper firewall configuration.
Some holes are remote ;-)
Quoted 15 lines
Quoted 9 lines
Quoted 7 lines
On a desktop, there are probably a bunch of out of kernel modules that will need upgrading with each new kernel modules. Just on the laptop I am using right now, I will have to upgrade the vmware bridge, nvidia driver, madwifi wireless driver, etc. And that's normal. The new development model didn't change that.
From my point of view, it makes a difference if i have to recompile
a module or realy upgrade it.That only happens for out ot tree modules, which shouldn't be really out of tree in the first place. That's the issue. If they are out of tree, it's for a reason. Either they cannot be in tree, or they are not stable enough.
Quoted 8 lines
There you see the issue.
Quoted 4 lines
[...]
cuED.
Jerome
Calin
--
Quoted 10 lines · Bernd Petrovitsch
Who do you mean with "user"?
A typical "user" just installs $DISTRIBUTION (be it a free or commercial one) and that's it. *If* the bug is severe enough (and the relevant maintainers does it) there will be a new rpm/deb/... with the newest kernel release or a backport.
If not, you can do it on your own anyways.
But then you are half a programmer and more like a sys-admin and no longer a "user". Voila.
So a user which is not typical user should not use a Linux distribution ?
--
Quoted 5 lines · On Wed, 9 Nov 2005 11:11 am, [email protected] wrote:
There are other reasons for using a new kernel. One of them is interactivity. In the days of 2.4 one could achieve decent interactivity for the desktop using preempt and low latency patches. For 2.6 interactivity was a real issue (possibly because of the new development model).
Eh? Where are the bug reports for this? I've only seen better interactivity and responsiveness with 2.6.
The only time I've seen bad behaviour repeatedly on forums and elsewhere has been due to misconfigured IDE drivers not using DMA and X servers still reniced to -10.
Cheers,
Con
Quoted 2 lines · Con Kolivas
Eh? Where are the bug reports for this? I've only seen better interactivity and responsiveness with 2.6.
It has certainly much better behaviour under load for me, to the point that sometimes I don't notice there's some computing-intensive process running in the background until I run top (and i thank you a lot for that ;)
Diego Calleja (who had lots of VM problems and suffered from filesystem corruption two times with 2.4 and its "stable" development model)
On 11/9/05, [email protected] <[email protected]> wrote:
[...]
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
reasons why a user wanna upgrade to new kernel. Maybe new supported hardware and so on. It's frustrating for the user, have on the one side the new hardware supported but on the other side, mybe broken support for the existing hardware.
New kernel feature and new supported hardware would be the only reason for me to upgrade. Personally that doesn't come that often. My hardware configurations don't change that much. I make sure it's well supported, not just recently. When one buys a non supported hardware, one should know the path chosen won't be the easiest.
Quoted 5 lines
There are other reasons for using a new kernel. One of them is interactivity. In the days of 2.4 one could achieve decent interactivity for the desktop using preempt and low latency patches. For 2.6 interactivity was a real issue (possibly because of the new development model).
I don't get it. You say that with 2.4 + patches you had good interactivity and with 2.6 you don't? Why did you switch then?
Quoted 9 lines
Quoted 5 lines
Quoted 1 line
And why should dirstribution makers always backport new security fixes ?
Because they want to ensure maximum stability. That's what users are (sometimes) paying for.
Maximum stability of what ? If the distribution kernels are based on vanila kernel (i.e. are based on unstable kernel) how stable will they be ?
Maximum stability of the kernel they deliver. When you fix a vulnerability, you fix a vulnerability. You don't just happen to add a new bunch of features, and a new bunch of bugs. Otherwise you are going to piss off your users a lot.
Everybody does that, including Microsoft (most of the time).
Quoted 2 lines
On lkml someone said that "stable means it won't crash very often".
This sounds like Windows(TM)
I've had Linux running on various hardware, from Pentium I machines to AMD 64bits machines and various laptops. The most problematic hardware I've had to deal was a MythTV box running on a Via EPIA M motherboard (due to what seems to be DMA issues on the board), and PCMCIA hang a on a Dell laptop. Some installation issues on new hardware and a wireless pcmcia card whose drivers are not stable. Apart from that, Linux has always been running very well. I had less than 10 hangs and hard reboots in 8 year with 20 machines. I had more oopses but 95% of the time confined to a module and not critical. And I've had more hardware failures, drives, disks, keyboards, pcmcia&usb ports, mice, motherboard than kernel issues.
On the security side, I've been broken into twice, one because of a very dumb misconfiguration on my part on a desktop (guest/guest account, no firewall, ssh server enabled and not restricted....), one on a server, probably because of an external php program whose security was not handled by the distribution (haven't investigated the issue yet, server is down).
To me, the most problematic crashes on desktops are X failures, because I lose several minutes to set up my desktop again.
And firefox crashes (or becomes unusable) way too often (around twice a week).
Quoted 9 lines
Quoted 7 lines
And second 90% of the security issues will not affect the majority of the home users (because they are restricted to a particular area of the kernel not affecting the user, or because they already require access on the machine to be exploitable). You will have much more risks using a box with an unpatched php or apache than with an unpached kernel, or without a proper firewall configuration.
Some holes are remote ;-)
That's your 10% (probably even much lower than that). And if you have a firewall on your desktop, most services turned down, and are behind a private network with a router with its own firewall, the probability that you get a remote attack on your particular version of the kernel, for which they are very few exploits in the wild anyway, is much lower than having a hard drive failure or a X server crash because of a driver issue.
If you care about stability, don't touch what works.
J
Quoted 12 lines · jerome lacoste
On 11/9/05, [email protected] <[email protected]> wrote:
[...]Quoted 6 lines
There are other reasons for using a new kernel. One of them is interactivity. In the days of 2.4 one could achieve decent interactivity for the desktop using preempt and low latency patches. For 2.6 interactivity was a real issue (possibly because of the new development model).
I don't get it. You say that with 2.4 + patches you had good interactivity and with 2.6 you don't? Why did you switch then?
Because i like to test new kernels. On 2.4 I run the vanila kernel and a test kernel. When something went wrong on a test kernel was always a stable kernel to use. 2.6 looks a lot like 2.5. New features are added very quickly without much testing. Of course there is Andrew's -mm tree but this one sometimes is too broken. For me linux looks now like it has one unstable tree (2.6) which is something like -ac was in days of 2.4 and -mm was in the days of 2.4 -2.5 and -mm which looks like it became very unstable. This is what i saw ok lkml (maybe my view is distorted).
I'll stop ranting and try both of them because i have some bugs to report.
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 1 line
And why should dirstribution makers always backport new security fixes ?
Because they want to ensure maximum stability. That's what users are (sometimes) paying for.
Quoted 9 lines
Quoted 3 lines
Maximum stability of what ? If the distribution kernels are based on vanila kernel (i.e. are based on unstable kernel) how stable will they be ?
Maximum stability of the kernel they deliver. When you fix a vulnerability, you fix a vulnerability. You don't just happen to add a new bunch of features, and a new bunch of bugs. Otherwise you are going to piss off your users a lot.
That's what's happening on 2.6. Every 2.6.x release is different. The 2.6.x.y kernels sometimes are almost no different from 2.6.x
--
Quoted 4 lines · On Wed, 2005-11-09 at 16:03 +0200, [email protected] wrote:
On Wed, 9 Nov 2005, jerome lacoste wrote:
Because i like to test new kernels. On 2.4 I run the vanila kernel and a test kernel. When something went wrong on a test kernel was always a stable kernel to use.
Quoted 8 lines
2.6 looks a lot like 2.5. New features are added very quickly without much testing. Of course there is Andrew's -mm tree but this one sometimes is too broken.
For me linux looks now like it has one unstable tree (2.6) which is something like -ac was in days of 2.4 and -mm was in the days of 2.4 -2.5 and -mm which looks like it became very unstable.
This is what i saw ok lkml (maybe my view is distorted).
I'll stop ranting and try both of them because i have some bugs to report.
Man, -mm are unstable kernels, 2.6.x[.y] are the stable ones.
Quoted 1 line
The 2.6.x.y kernels sometimes are almost no different from 2.6.x
That's true, and good. the .y is the -stable tree, that is supposed to add only stability and security fixes.
Signature · 2 lines
Marcos Marado <[email protected]> Novis ISP