Wanted: hotfixes for -mm kernels
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
Signature · 1 line
Chuck
Quoted 9 lines · Chuck Ebbert
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Lee
Quoted 11 lines · Lee Revell
Quoted 9 lines · Chuck Ebbert
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
Quoted 6 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
--
Dmitry
Quoted 19 lines · Dmitry Torokhov
Quoted 11 lines · Lee Revell
Quoted 9 lines · Chuck Ebbert
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
Quoted 6 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Quoted 5 lines
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
Yep. I think it's a good idea, although it does move fast, like Lee says. I'd be willing to help if e.g. there was some place where several of us could upload patches to.
Signature · 1 line
~Randy
Quoted 21 lines · Randy.Dunlap
Quoted 19 lines · Dmitry Torokhov
Quoted 17 lines · Lee Revell
Quoted 9 lines · Chuck Ebbert
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Quoted 8 lines
Quoted 5 lines
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
Yep. I think it's a good idea, although it does move fast, like Lee says.
One can argue that it doesn't move fast enough - I am pulling from Linus into my working tree pretty much daily whereas -mm is static for about a week. SO with Linu's tree I'm getting fixes to abvious problems/regressions much faster.
Quoted 3 lines
I'd be willing to help if e.g. there was some place
where several of us could upload patches to.
That's a nice idea too.
--
Dmitry
Quoted 19 lines · Dmitry Torokhov
Quoted 11 lines · Lee Revell
Quoted 9 lines · Chuck Ebbert
Most -mm kernels have small but critical bugs that are found shortly after release. Patches for these are posted on linux-kernel but they aren't made available on kernel.org until the next -mm release.
Would it be possible to create a hotfix/ directory for each -mm release and put those patches there? A README could explain that the fixes are untested. At least people reading the files could see an issue exists even if they're not brave enough to try the patch. :)
Quoted 6 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Quoted 5 lines
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
... that's just why I so often wish to have a -git tree, Andrew. ;)
Marc
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Quoted 7 lines
Quoted 5 lines
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
... that's just why I so often wish to have a -git tree, Andrew. ;)
Why do people always thing a source code control system is magically going to fix all bugs and wipe their ass for them?
You still have to work out which patches are relevant and merge them. If he's just merging a new set of changes constantly, it won't help you a damn.
What is needed is somebody to do the grunt work, not SCM magic. I'll try to help with Randy's suggestion of a shared pool of patches in a common dumping ground at least - I've needed pretty much the same thing in the past for the test.kernel.org stuff ... mainly fixups needed to get the tree to boot so we can test it.
M.
Quoted 14 lines · Martin J. Bligh
Quoted 14 lines
Quoted 13 lines
Quoted 7 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
Quoted 7 lines
Quoted 1 line
... that's just why I so often wish to have a -git tree, Andrew. ;)
Why do people always thing a source code control system is magically going to fix all bugs and wipe their ass for them?
You still have to work out which patches are relevant and merge them. If he's just merging a new set of changes constantly, it won't help you a damn.
We talked about hotfixes for -mm. So why not check these into the -mm-git tree then? This would make sense and would conform fully to my understanding of what the -mm-git tree should be. I don't want to select 23 patches from LKML to make the tree compile or work. I want to checkout. Why make it easy when you may get it difficult.
Besides testing the stuff we would get more far by being able to test stuff faster (because a patch is applied to -mm and we do a checkout) instead of waiting a week for this mega-patch to be applied.
What sense does an -mm tree make when there are people that cannot test it because of known bugs that lead to the -mm tree not being bootable or - even worse - destroying the system?
git is you friend. Not only for Linus' tree, but as well for Andrew's tree. It would just make debugging and testing -mm more convenient and less time consuming for the testers. Instead of 1000 people seeking patches Andrew would just check in and we all could pull it.
If you agree with me or not - that's what I think.
Marc
Quoted 15 lines
We talked about hotfixes for -mm. So why not check these into the -mm-git tree then? This would make sense and would conform fully to my understanding of what the -mm-git tree should be. I don't want to select 23 patches from LKML to make the tree compile or work. I want to checkout. Why make it easy when you may get it difficult.
Besides testing the stuff we would get more far by being able to test stuff faster (because a patch is applied to -mm and we do a checkout) instead of waiting a week for this mega-patch to be applied.
What sense does an -mm tree make when there are people that cannot test it because of known bugs that lead to the -mm tree not being bootable or - even worse - destroying the system?
Quoted 6 lines
git is you friend. Not only for Linus' tree, but as well for Andrew's tree. It would just make debugging and testing -mm more convenient and less time consuming for the testers. Instead of 1000 people seeking patches Andrew would just check in and we all could pull it.
If you agree with me or not - that's what I think.
SCMs don't fix anything. The real work is in selecting patches and merging them. Frankly, I test a lot of stuff myself, and the tarballs are a damned sight easier to work with, and have a simple chronological timeline to work from.
Yes, of course you don't want to pull 23 separate patches from a mailing list. But quilt+tarballs is a crapload simpler than git / bk / cvs / subversion, and works just as well, if not better. It just needs a script to roll up patches into a consolidated one, and it's not like Andrew doesn't have that already.
M.
Quoted 29 lines · Martin J. Bligh
Quoted 29 lines
We talked about hotfixes for -mm. So why not check these into the -mm-git tree
then? This would make sense and would conform fully to my understanding of what
the -mm-git tree should be. I don't want to select 23 patches from LKML to make
the tree compile or work. I want to checkout. Why make it easy when you may get
it difficult.Besides testing the stuff we would get more far by being able to test stuff faster
(because a patch is applied to -mm and we do a checkout) instead of waiting a
week for this mega-patch to be applied.What sense does an -mm tree make when there are people that cannot test it because of
known bugs that lead to the -mm tree not being bootable or - even worse - destroying
the system?git is you friend. Not only for Linus' tree, but as well for Andrew's tree. It would just make debugging and testing -mm more convenient and less time consuming for the testers. Instead of 1000 people seeking patches Andrew would
just check in and we all could pull it.
Quoted 6 lines
Quoted 1 line
If you agree with me or not - that's what I think.
SCMs don't fix anything. The real work is in selecting patches and merging them. Frankly, I test a lot of stuff myself, and the tarballs are a damned sight easier to work with, and have a simple chronological timeline to work from.
To me it would be easier if I could use a repository that get's updated every day (or several times a day) than wait a week for the patch to be available.
Moreover, a repository in fact _is_ easyier when it comes to debugging than is a tarball or such. What if one says 'Gimme that driver 6 weeks ago'? See ... with a repository I could just get it whereas with patches I have to read lots of mails to get a clue when stuff got applied (in the tree and over here on my hard disk).
I didn't mean to start a discussion here. I was just telling you what is easier for me. Anyone has it's own style of coding. Someone likes patches, others just live with them. My boss doesn't like vi either. So I may not use it? Puh!
Marc
Quoted 23 lines · Marc Koschewski
Quoted 14 lines · Martin J. Bligh
Quoted 14 lines
Quoted 13 lines
Quoted 7 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
Quoted 7 lines
Quoted 1 line
... that's just why I so often wish to have a -git tree, Andrew. ;)
Why do people always thing a source code control system is magically going to fix all bugs and wipe their ass for them?
You still have to work out which patches are relevant and merge them. If he's just merging a new set of changes constantly, it won't help you a damn.
Quoted 9 lines
We talked about hotfixes for -mm. So why not check these into the -mm-git tree then? This would make sense and would conform fully to my understanding of what the -mm-git tree should be. I don't want to select 23 patches from LKML to make the tree compile or work. I want to checkout. Why make it easy when you may get it difficult.
...
What sense does an -mm tree make when there are people that cannot test it because of known bugs that lead to the -mm tree not being bootable or - even worse - destroying the system?
That's exactly what Andrew does now implement through the hot-fixes directory [1].
Git doesn't help for the problem that it's currently empty - it's more important that people tell Andrew that this or that patch should be made available there.
Quoted 5 lines
git is you friend. Not only for Linus' tree, but as well for Andrew's tree. It would just make debugging and testing -mm more convenient and less time consuming for the testers. Instead of 1000 people seeking patches Andrew would just check in and we all could pull it.
...
git is the SCM Linus developed to fit his workflow.
Andrew has a completely different workflow, and he has developed the tools he needs for his workflow.
As long as they are able to interact (which seems to work without problems), there's nothing forcing them to use the same tools.
Quoted 1 line
Marc
cu
Adrian
[1] ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.16-rc1/2.6.16-rc1-mm5/hot-fixes/
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 47 lines · Adrian Bunk
Quoted 23 lines · Marc Koschewski
Quoted 14 lines · Martin J. Bligh
Quoted 14 lines
Quoted 13 lines
Quoted 7 lines
I doubt it - mm is an experimental kernel, hotfixes only make sense for production stuff. It moves too fast.
A better question is what does -mm give you that mainline does not, that causes you to want to "stabilize" a specific -mm version?
Some people just run -mm so the hotfixes/* would help them to get their boxes running until the next -mm without having to hunt through LKML for bugs already reported/fixed. This will allow better testing coverage because most obvious bugs are caught almost immediately and then people can continue using -mm to find more stuff.
Quoted 7 lines
Quoted 1 line
... that's just why I so often wish to have a -git tree, Andrew. ;)
Why do people always thing a source code control system is magically going to fix all bugs and wipe their ass for them?
You still have to work out which patches are relevant and merge them. If he's just merging a new set of changes constantly, it won't help you a damn.
Quoted 9 lines
We talked about hotfixes for -mm. So why not check these into the -mm-git tree then? This would make sense and would conform fully to my understanding of what the -mm-git tree should be. I don't want to select 23 patches from LKML to make the tree compile or work. I want to checkout. Why make it easy when you may get it difficult.
...
What sense does an -mm tree make when there are people that cannot test it because of known bugs that lead to the -mm tree not being bootable or - even worse - destroying the system?That's exactly what Andrew does now implement through the hot-fixes directory [1].
Git doesn't help for the problem that it's currently empty - it's more important that people tell Andrew that this or that patch should be made available there.
Quoted 5 lines
git is you friend. Not only for Linus' tree, but as well for Andrew's tree. It would just make debugging and testing -mm more convenient and less time consuming for the testers. Instead of 1000 people seeking patches Andrew would just check in and we all could pull it.
...
Quoted 7 lines
git is the SCM Linus developed to fit his workflow.
Andrew has a completely different workflow, and he has developed the tools he needs for his workflow.
As long as they are able to interact (which seems to work without problems), there's nothing forcing them to use the same tools.
I really didn't mean to 'force' anyone to use a specific tools or even type of tool such as an 'SCM' generally. I didn't even suggest it. I just wish to have one as I'm used to it and IMHO it makes thing easier. Applying is easier (where 'patch -p1 < ../file.patch is not really difficult either), getting back to a specific date is easier, commenting is easier (check-in descriptions).
Moreover, just _one_ more thing which is most important at least to me:
If one can do a checkout on a daily basis with let's say 3 new patches checked in, one can test these. The other day there are maybe another 2 new patches applied which are to be tested. This would just speed up the testing cycle whereas a 3 MB patch with tons of fixes and testing is just like a kick-down with the hand-brake pulled. Sometimes I#M really overwhelmed by how many patches came in with the recent -mm. ACPI, ALSA, VM, various FS stuff, ... It's quite impossible to me to really test the patches in an effective manner due to having my 2 eyes focussed on too many things at one. I do test ALSA with various types of in/out operation while my reiserfs partition is on the way to hell. To me that doesn't sound like effective testing.
This is just about testing and debugging time effectiveness and not about some religious thing a la SCM vs. patches.
Marc
Quoted 28 lines · Marc Koschewski
Quoted 1 line · Adrian Bunk
On Sat, Feb 04, 2006 at 07:57:39PM +0100, Marc Koschewski wrote:
...
Quoted 10 lines
Quoted 9 lines
We talked about hotfixes for -mm. So why not check these into the -mm-git tree then? This would make sense and would conform fully to my understanding of what the -mm-git tree should be. I don't want to select 23 patches from LKML to make the tree compile or work. I want to checkout. Why make it easy when you may get it difficult.
...
What sense does an -mm tree make when there are people that cannot test it because of known bugs that lead to the -mm tree not being bootable or - even worse - destroying the system?Quoted 2 lines
That's exactly what Andrew does now implement through the hot-fixes directory [1].
...
Moreover, just _one_ more thing which is most important at least to me:
If one can do a checkout on a daily basis with let's say 3 new patches checked in, one can test these. The other day there are maybe another 2 new patches applied which are to be tested. This would just speed up the testing cycle whereas a 3 MB patch with tons of fixes and testing is just like a kick-down with the hand-brake pulled. Sometimes I#M really overwhelmed by how many patches came in with the recent -mm. ACPI, ALSA, VM, various FS stuff, ... It's quite impossible to me to really test the patches in an effective manner due to having my 2 eyes focussed on too many things at one. I do test ALSA with various types of in/out operation while my reiserfs partition is on the way to hell. To me that doesn't sound like effective testing.
Quoted 2 lines
This is just about testing and debugging time effectiveness and not about some religious thing a la SCM vs. patches.
Hotfixes are something that can be done.
What you suggest is not really possible.
Your thinko is:
3 new patches a day don't sum up to 3 MB after one week.
As far as I understand it, the big work for Andrew is integrating two dozen git repositories and at about thousand patches in a way that:
- everything applies
- the kernel compiles on several architectures
- the kernel actually runs on some machines
Andrew does all this before he releases an -mm kernel (I know due to the feedback he sometimes gives for patches I sent), and this is the sole reason why most -mm kernels are half-way usable.
Some random snapshot from Andrew's repository wouldn't help anyone, either it's good enough that he releases an -mm kernel or it isn't worth other people testing it.
Quoted 1 line
Marc
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 SeedDo we need a place to put hotfix patches, or do we just need a list of links to lkml postings to said patches. Such a list has the advantage of pointing into the discussion surrounding each such fix, and such a list has the advantage of not holding so much redundant data (these patches will be redundant with what was posted on lkml). Redundant data out of context goes stale, and is less valuable.
I can imagine someone (not me ;) keeping a wiki web page, listing for each *-mm and Linus release the particular lkml patch postings that one needs to pick off to get a build and boot.
Just brainstorming ...
Signature · 3 lines
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.925.600.0401Quoted 7 lines · Paul Jackson
Do we need a place to put hotfix patches, or do we just need a list of links to lkml postings to said patches. Such a list has the advantage of pointing into the discussion surrounding each such fix, and such a list has the advantage of not holding so much redundant data (these patches will be redundant with what was posted on lkml). Redundant data out of context goes stale, and is less valuable.
Quoted 5 lines
I can imagine someone (not me ;) keeping a wiki web page, listing for each *-mm and Linus release the particular lkml patch postings that one needs to pick off to get a build and boot.
Just brainstorming ...
That would just come closer to a repositories check-in description. I vote for such a thing. If it is a wiki or not.
Marc
Quoted 14 lines · Marc Koschewski
Quoted 11 lines · Paul Jackson
Do we need a place to put hotfix patches, or do we just need a list of links to lkml postings to said patches. Such a list has the advantage of pointing into the discussion surrounding each such fix, and such a list has the advantage of not holding so much redundant data (these patches will be redundant with what was posted on lkml). Redundant data out of context goes stale, and is less valuable.
I can imagine someone (not me ;) keeping a wiki web page, listing for each *-mm and Linus release the particular lkml patch postings that one needs to pick off to get a build and boot.
Quoted 5 lines
Quoted 1 line
Just brainstorming ...
That would just come closer to a repositories check-in description. I vote for such a thing. If it is a wiki or not.
Why have a list of pointers to go fishing for things out of an email archive, rather than just dump the patchs straight into a directory? Seeing as Andrew seems to have already created a subdir to do this, the point seems somewhat moot by now ;-)
M.
Quoted 1 line
point seems somewhat moot by now ;-)
Sure ... whatever.
Signature · 3 lines
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.925.600.0401